{"id": "2512.00850", "title": "Smol-GS: Compact Representations for Abstract 3D Gaussian Splatting", "abstract": "We present Smol-GS, a novel method for learning compact representations for 3D Gaussian Splatting (3DGS). Our approach learns highly efficient splat-wise features to model 3D space, which capture abstracted cues, including color, opacity, transformation, and material properties. We propose octree-derived positional encoding, which explicitly models spatial locality and enhances representation efficiency. We further apply entropy-based compression to exploit feature redundancy and compress splat coordinates using a recursive voxel hierarchy. This design enables orders-of-magnitude reduction in storage while preserving representation flexibility. Smol-GS achieves state-of-the-art compression performance on standard benchmarks with high-level rendering quality.", "date": "2025-11-30", "categories": []}
{"id": "2605.00569", "title": "2D-SuGaR: Surface-Aware Gaussian Splatting for Geometrically Accurate Mesh Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a powerful technique for generating photorealistic renderings of a scene in real-time. However, the volumetric nature of 3DGS limits its ability to accurately capture surface geometry. To address this, 2D Gaussian Splatting (2DGS) was proposed to enable view-consistent and geometrically accurate surface reconstruction from multi-view images. However, 2DGS can be sensitive to the initialization of the Gaussian primitives. Reliance on Structure-from-Motion (SfM) initializations, which can produce poor estimates on challenging image sets, may lead to subpar results. In this work, we enhance 2DGS by incorporating monocular depth and normal priors to improve both geometric accuracy and robustness. We propose a depth-guided initialization strategy for Gaussians and introduce a clustering-based technique for pruning degenerate Gaussians. We evaluate our method on the DTU dataset, where it achieves state-of-the-art results in mesh reconstruction while preserving high-quality novel view synthesis.", "date": "2026-05-01", "categories": []}
{"id": "2606.18439", "title": "RegimeVGGT: Layer-Wise Spatially Preserving Redundancy Removal for Visual Geometry Grounded Transformer", "abstract": "Visual Geometry Grounded Transformer (VGGT) recovers dense 3D scene structure from multi-view images in one forward pass, but quadratic cross-frame attention limits its scalability. Existing training-free accelerators reduce computation uniformly along one axis, missing layer heterogeneity. Our spectral, probing, and causal analyses reveal three regimes: shallow layers lack cross-view structure, middle layers drive cross-view alignment, and deep layers are redundant for dense geometry yet their cross-frame attention remains essential for pose. RegimeVGGT applies layer-wise U-shaped compression along two axes: Saliency-Guided Banded Merging protects geometry- and edge-salient tokens, while Selectively Protected K/V Downsampling preserves cross-frame spatial coverage and the pose-critical path through a phase-shifted spatial grid, a reference-frame anchor, and uncompressed camera/register tokens. Training-free, RegimeVGGT achieves a 6.7x speedup over VGGT* at matched reconstruction quality.", "date": "2026-06-16", "categories": []}
{"id": "2606.18588", "title": "Splaxel: Efficient Distributed Training of 3D Gaussian Splatting for Large-scale Scene Reconstruction via Pixel-level Communication", "abstract": "3D Gaussian Splatting (3DGS) enables high-fidelity and real-time 3D scene reconstruction, but scaling training to large-scale scenes requires optimizing hundreds of millions of Gaussians across multiple GPUs. Existing distributed approaches either partition scenes into isolated regions, causing global inconsistency, or rely on global Gaussian-level exchanges, which lead to substantial growth in inter-GPU communication and quickly dominate iteration time.\n  We propose Splaxel, a communication-efficient distributed 3DGS training framework based on pixel-level local rendering and global composition. Instead of synchronizing Gaussians, each GPU renders its local subset and exchanges only partial pixel values, maintaining mathematical consistency while keeping communication cost stable as the scene size increases. Splaxel further reduces pixel-level redundancy through geometric and transmittance visibility prediction and improves GPU utilization via conflict-free camera-view consolidation. Evaluated on large-scale datasets with up to 120M Gaussians, Splaxel achieves up to 7.6$\\times$ speedup over the state-of-the-art distributed 3DGS framework while preserving high reconstruction quality.", "date": "2026-06-17", "categories": []}
{"id": "2606.18787", "title": "Learned Radius Estimation for UDF-Based Point Cloud Reconstruction", "abstract": "Surface reconstruction from point clouds is important for consumer-grade 3D capture, including AR/VR and indoor scanning. Local-patch Unsigned Distance Field (UDF) methods are lightweight and generalizable, but their accuracy depends on the support radius, traditionally fixed or selected by a one-dimensional curvature heuristic that cannot capture heterogeneous local geometry. We propose a learned per-query radius selector that predicts a continuous support radius and plugs into a frozen LoSF-UDF backbone. The selector is trained using off-grid target radii obtained by parabolic interpolation of cached UDF error curves. Experiments show improved fine-scale reconstruction accuracy.", "date": "2026-06-17", "categories": []}
{"id": "2606.19019", "title": "FlowObject: Flow Steering for Bridging Generative Priors and Reconstruction Fidelity", "abstract": "Recovering complete 3D representations of objects from few casual image captures remains a significant challenge. Recent 3D generative models, particularly those based on Flow-Matching (FM), can synthesize high-quality textured assets; however, they often suffer from ''synthetic bias'' where learned priors override observational evidence, alongside a lack of alignment with the observed instance. Conversely, optimization-based methods like 3D Gaussian Splatting (3DGS) provide high fidelity on visible surfaces but fail to reason about unobserved geometry. In this paper, we present FlowObject, a framework that reformulates sparse-view 3D reconstruction as a training-free, guided inverse problem. Our approach applies a dual-space guidance strategy to steer the Ordinary Differential Equation (ODE) trajectory of a flow-matching model, enabling the completion of unseen regions through learned generative priors while enforcing strict consistency with real-world observations. By integrating a 3DGS refinement stage, FlowObject further bridges the gap between ''synthetic-looking'' generative outputs and photorealistic reconstructions. Comprehensive benchmarks on synthetic and real-world datasets demonstrate that current state-of-the-art methods often struggle to achieve geometric completeness and observational consistency simultaneously, especially under severe occlusions. In contrast, our method significantly outperforms state-of-the-art generative models and optimization-based frameworks in both geometric completeness and view-dependent appearance fidelity.", "date": "2026-06-17", "categories": []}
{"id": "2506.07917", "title": "SpeeDe3DGS: Speedy Deformable 3D Gaussian Splatting with Temporal Pruning and Motion Grouping", "abstract": "Dynamic extensions of 3D Gaussian Splatting (3DGS) achieve high-quality reconstructions through neural motion fields, but per-Gaussian neural inference makes these models computationally expensive. Building on DeformableGS, we introduce Speedy Deformable 3D Gaussian Splatting (SpeeDe3DGS), which bridges this efficiency-fidelity gap through three complementary modules: Temporal Sensitivity Pruning (TSP) removes low-impact Gaussians via temporally aggregated sensitivity analysis, Temporal Sensitivity Sampling (TSS) perturbs timestamps to suppress floaters and improve temporal coherence, and GroupFlow distills the learned deformation field into shared SE(3) transformations for efficient groupwise motion. On the 50 dynamic scenes in MonoDyGauBench, integrating TSP and TSS into DeformableGS accelerates rendering by 6.78$\\times$ on average while maintaining neural-field fidelity and using 10$\\times$ fewer primitives. Adding GroupFlow culminates in 13.71$\\times$ faster rendering and 2.53$\\times$ shorter training, surpassing all baselines in speed while preserving superior image quality.", "date": "2025-06-09", "categories": []}
{"id": "2506.10981", "title": "SceneCompleter: Dense 3D Scene Completion for Generative Novel View Synthesis", "abstract": "Generative models have shown great promise for novel view synthesis (NVS) by leveraging strong image generation priors. However, existing approaches typically follow a 2D inpainting paradigm, first completing missing image regions and then performing 3D reconstruction. This strategy often causes geometry distortion and appearance drift, as 2D inpainting models cannot reliably infer the underlying 3D structure required for cross-view consistent generation. In this paper, we propose \\textbf{SceneCompleter}, a geometry-aware framework that reformulates generative NVS as dense 3D scene completion. Instead of hallucinating isolated 2D views, SceneCompleter jointly completes geometry and appearance through a geometry-appearance dual-stream diffusion model in a spatially aligned RGBD latent space. To provide holistic scene context, we further introduce a Scene Embedder that conditions generation on global semantic and stylistic information from reference images. The completed RGBD predictions are then aligned and integrated into an expandable 3D scene representation, enabling iterative and coherent scene completion. Extensive experiments on in-domain and out-of-distribution datasets demonstrate that SceneCompleter produces visually plausible and geometrically consistent novel views across diverse scenarios. Project Page: https://chen-wl20.github.io/SceneCompleter", "date": "2025-06-12", "categories": []}
{"id": "2507.05163", "title": "4DSloMo: 4D Reconstruction for High Speed Scene with Asynchronous Capture", "abstract": "Reconstructing fast-dynamic scenes from multi-view videos is crucial for high-speed motion analysis and realistic 4D reconstruction. However, the majority of 4D capture systems are limited to frame rates below 30 FPS (frames per second), and a direct 4D reconstruction of high-speed motion from low FPS input may lead to undesirable results. In this work, we propose a high-speed 4D capturing system only using low FPS cameras, through novel capturing and processing modules. On the capturing side, we propose an asynchronous capture scheme that increases the effective frame rate by staggering the start times of cameras. By grouping cameras and leveraging a base frame rate of 25 FPS, our method achieves an equivalent frame rate of 100-200 FPS without requiring specialized high-speed cameras. On processing side, we also propose a novel generative model to fix artifacts caused by 4D sparse-view reconstruction, as asynchrony reduces the number of viewpoints at each timestamp. Specifically, we propose to train a video-diffusion-based artifact-fix model for sparse 4D reconstruction, which refines missing details, maintains temporal consistency, and improves overall reconstruction quality. Experimental results demonstrate that our method significantly enhances high-speed 4D reconstruction compared to synchronous capture.", "date": "2025-07-07", "categories": []}
{"id": "2509.12742", "title": "Effective Gaussian Management for High-fidelity Object Reconstruction", "abstract": "This paper proposes an effective Gaussian management framework for high-fidelity scene reconstruction of both appearance and geometry. Unlike recent Gaussian Splatting (GS) pipelines that treat all primitives uniformly during optimization, our framework explicitly manages the attribute activation, representation and pruning of Gaussian. Specifically, our framework first introduces GauSep, a novel densification strategy that selectively activates Gaussian color or normal attributes to alleviate destructive gradient conflicts arising from dual supervision. We further propose GauRep, an adaptive Gaussian representation that dynamically adjusts spherical harmonics (SHs) orders and performs task-decoupled pruning to reduce redundancy at both the individual and global levels. To provide reliable geometric supervision for above mangement process, we additionally introduce CoRe, an regularized surface reconstruction module that distills robust normal fields from an SDF branch to the Gaussian representation through a confidence mechanism. Notably, the proposed Gaussian management is compatible with various reconstruction architectures and can be seamlessly integrated to improve performance while reducing size of the model. Extensive experiments demonstrate that our approach achieves superior or comparable performance in appearance and geometry reconstruction compared with state-of-the-art methods, while using significantly fewer parameters.", "date": "2025-09-16", "categories": []}
{"id": "2606.17520", "title": "GASE: Gaussian Splatting-Based Automated System for Reconstructing Embodied-Simulation Environments", "abstract": "Training embodied agents in the real world requires skilled operators and expensive hardware. Simulation environments offer a compelling alternative by enabling large-scale, cost-effective data augmentation. Consequently, rapidly constructing high-fidelity simulation scenes with a minimal sim-to-real gap has become a critical objective in robot learning. While reconstruction-based methods provide superior visual quality, current workflows are hindered by inefficient data acquisition and subpar foreground object extraction. We thus propose GASE, a highly automated system for simulation scene construction. GASE leverages multi-view video streams from panoramic camera arrays to enable rapid environment scanning. To ensure high-quality asset generation, our pipeline introduces a camera-pose-based strategy that robustly extracts objects across frames in the 2D domain, followed by high-fidelity scene inpainting. Foreground objects and the static background are then reconstructed independently and seamlessly imported into physics simulators for policy training. Extensive experiments demonstrate that GASE outperforms existing 3D Gaussian-based methods in segmentation accuracy by over 10\\% while achieving state-of-the-art inpainting quality. Furthermore, real-robot deployments across manipulation and navigation tasks maintains a performance gap of less than 10\\% compared to policies trained purely on real-world data. These results confirm that GASE provides an efficient and highly effective solution for bridging the sim-to-real gap. Code will be released.", "date": "2026-06-16", "categories": []}
{"id": "2606.17935", "title": "MoonSplat: Monocular Online Gaussian Splatting with Sim(3) Global Optimization", "abstract": "Online 3D reconstruction from monocular image sequences is a challenging and ongoing research topic. 3D Gaussian Splatting (3DGS), leveraging its high-quality real-time rendering capability, empowers online 3D reconstruction to represent dense scenes with enhanced expressiveness, and thus holds great promise for a wide range of applications such as robotics and AR/VR. However, existing online 3DGS methods still suffer from some key challenges: fragile camera pose estimation due to the lack of global optimization, and low optimization efficiency in large-scale or long-sequence scenarios. To address these issues, we propose a robust and efficient online voxelized 3DGS reconstruction framework integrated with global $\\text{Sim}(3)$ optimization, which enables reliable camera tracking and efficient global loop closure for both camera poses and voxelized 3DGS. To accelerate the convergence of the voxelized 3DGS, we further introduce a color residual learning strategy, which not only boosts optimization speed but also enhances rendering quality. Extensive experiments on diverse indoor and outdoor datasets demonstrate that our method achieves state-of-the-art performance in both camera pose estimation accuracy and rendering quality, while retaining real-time efficiency. Additionally, we develop and deploy a real-world UAV-based active reconstruction system grounded on our proposed method, validating its robustness and generalizability for practical online 3D reconstruction tasks. Our code and data are available at https://github.com/TrickyGo/MoonSplat.", "date": "2026-06-16", "categories": []}
{"id": "2606.18153", "title": "Neural Tree Reconstruction for the Open Forest Observatory", "abstract": "The Open Forest Observatory (OFO) is a collaboration across universities and other partners to make low-cost forest mapping accessible to ecologists, land managers, and the general public. The OFO is building both a database of geospatial forest data as well as open-source methods and tools for forest mapping by uncrewed aerial vehicle. Such data are useful for a variety of climate applications including prioritizing reforestation efforts, informing wildfire hazard reduction, and monitoring carbon sequestration. In the current iteration of the OFO's forest map database, 3D tree maps are created using classical structure-from-motion techniques. This approach is prone to artifacts, lacks detail, and has particular difficulty on the forest floor where the input data (overhead imagery) has limited visibility. These reconstruction errors can potentially propagate to the downstream scientific tasks (e.g. a wildfire simulation.) Advances in 3D reconstruction, including methods like Neural Radiance Fields (NeRF), produce higher quality results that are more robust to sparse views and support data-driven priors. We explore ways to incorporate NeRFs into the OFO dataset, outline future work to support even more state-of-the-art 3D vision models, and describe the importance of high-quality 3D reconstructions for forestry applications.", "date": "2026-06-16", "categories": []}
{"id": "2606.18250", "title": "Future Dynamic 3D Reconstruction: A 3D World Model with Disentangled Ego-Motion", "abstract": "Forecasting the evolution of dynamic environments is crucial for autonomous agents. While generative world models have recently achieved high photorealism in 2D video synthesis by mixing ego-motion and environmental dynamics within the image plane, they exhibit physical inconsistencies, such as morphing or vanishing objects, especially over long time horizons. In this paper, we propose FR3D, a world model that predicts a persistent 3D latent representation for future dynamic 3D reconstruction. Unlike prior works that treat the world as a sequence of image-based features, FR3D explicitly decouples the 3D evolution of the scene from the agent's trajectory, treating the inferred ego-motion as a latent proxy for action. This disentanglement resolves the ambiguities between self-motion and world-motion, ensuring geometric consistency into the future. Furthermore, we introduce a teacher-student distillation strategy that leverages the spatial \"common sense\" of off-the-shelf foundation models, leading to robust zero-shot generalization. Extensive experiments demonstrate FR3D's strong performance for future dynamic 3D reconstruction from monocular observations across multiple datasets, even 2 seconds into the future. Project page: https://fr3d-wm.github.io.", "date": "2026-06-16", "categories": []}
{"id": "2606.14811", "title": "S23DR 2026: End-to-End 3D Wireframe Prediction via DETR-Style Set Prediction with Contrastive Denoising", "abstract": "We present WireframeDETR, our submission to the Structured Semantic 3D Reconstruction (S23DR) 2026 Challenge, which requires predicting a 3D building wireframe from multi-view COLMAP point clouds. Our method applies DETR-style set prediction directly to 3D point clouds, producing wireframes as sets of edge coordinate pairs without any intermediate vertex detection stage. We introduce three technical contributions: (1) contrastive denoising training that stabilises noisy Hungarian matching in early epochs; (2) a multi-scale encoder that aggregates the last encoder layer outputs via learned scalar weights; and (3) progressive auxiliary loss weighting that concentrates gradient signal on the decoder layers that most benefit from it. Our model achieves a public test HSS of 0.575 (F1~=~0.664, IoU~=~0.516) and a best validation HSS of 0.534 on the cleaned val split.", "date": "2026-06-12", "categories": []}
{"id": "2606.15924", "title": "TurboGS: Accelerating 3D Gaussian Splatting via Error-Guided Sparse Pixel Sampling and Optimization", "abstract": "Consumer-level applications require fast optimization of 3D Gaussian Splatting (3DGS) with high-fidelity novel view rendering. However, existing 3DGS acceleration approaches still incur substantial computation on redundant pixels while sacrificing fine details. In this paper, we present TurboGS, an error-guided training framework that accelerates 3DGS by concentrating optimization on perceptually informative pixels. TurboGS is built upon four core components: (1) a tile-wise sparse pixel sampling, which, driven by multi-view reconstruction errors during training, prioritizes challenging regions and skips well-reconstructed ones to avoid redundant gradient computation; (2) a tile-wise structure-aware loss with sparse Normalized Cross-Correlation, which provides sparse yet effective supervision to preserve fine details and stabilize training; (3) an error-driven Gaussian density control strategy, which dynamically allocates model capacity and removes redundant primitives; and (4) a tailored hybrid optimizer that couples Hessian-informed updates with Adam moment damping to stabilize and improve convergence under sparse supervision. Experiments on standard benchmarks demonstrate that TurboGS can deliver on par or superior rendering quality within 100 seconds on a single RTX 5090 GPU card (up to 10x training speedup over vanilla 3DGS).", "date": "2026-06-14", "categories": []}
{"id": "2606.16048", "title": "PointDiffusion: Diffusion-Based Scene Completion in the Point Cloud Domain", "abstract": "Reconstructing dense 3D scenes from sparse LiDAR point clouds is a fundamental challenge in autonomous driving, where latent diffusion models offer a promising solution. However, existing approaches rely on object-level autoencoders that collapse into unstable global representations at outdoor scale and suffer from ground truth data corrupted by odometry drift that systematically degrades supervision quality. Furthermore, multi-step diffusion inference incurs prohibitive latency for real-time deployment. We propose a novel multi-token Gaussian VAE with cross-attention pooling for stable scene-scale LiDAR compression, combined with an anchor-based ICP ground truth refinement pipeline that eliminates drift-induced noise from training supervision. Together, these components enable a scaffold-free single-step diffusion completion model that achieves an approximately 16x reduction in squared Chamfer distance on SemanticKITTI seq. 08 (0.396 m^2 to 0.024 m^2), surpasses LiDiff and ScoreLiDAR by 17-19% and 10-11%, respectively, and operates at 25-143x lower inference latency. Our results demonstrate that data quality dominates model design in this regime and that multi-token latent spaces provide a stable first stage for latent diffusion-based scene completion.", "date": "2026-06-14", "categories": []}
{"id": "2606.16479", "title": "Uncertainty Quality of VGGT: An Analysis on the DTU Benchmark Dataset", "abstract": "Visual Geometry Grounded Transformer (VGGT) has already attracted a great deal of attention in a short period of time, not least due to the Best Paper Award at CVPR-2025. Similar to DUSt3R and MASt3R, VGGT aims to bring about a paradigm shift by replacing established methods like bundle adjustment and feature matching with a simple, unified, feed-forward neural network that predicts camera poses, depth maps, and dense 3D structure directly from multiple images of a scene in a few seconds. A key aspect is its ability to process an arbitrary number of views consistently in a single forward pass without any post-processing or iterative optimization. For photogrammetry, this opens new possibilities for real-time, scalable, and accessible 3D reconstruction. In this context, not only high reconstruction accuracy but also high-quality uncertainty estimates are crucial, as they foster trust and enable robust quality assurance. This paper therefore investigates the quality of VGGT's uncertainty predictions. The analysis identifies an effective confidence threshold for filtering VGGT's raw output and demonstrates that enhancing uncertainty quality holds strong potential for improving the accuracy of its 3D reconstructions.", "date": "2026-06-15", "categories": []}
{"id": "2606.05102", "title": "ZipSplat: Fewer Gaussians, Better Splats", "abstract": "Feed-forward 3D Gaussian Splatting methods reconstruct a scene from posed or pose-free images in a single forward pass, yet current approaches predict one Gaussian per input pixel, tying the representation budget to camera resolution rather than scene complexity. A flat wall and a richly textured object thus produce equally many Gaussians despite very different geometric needs. We propose ZipSplat, a token-based feed-forward model that decouples Gaussian placement from the pixel grid. A multi-view backbone extracts dense visual tokens, and k-means clustering compresses them into a compact set of scene tokens. Cross- and self-attention refine these tokens, and a lightweight MLP decodes each into a group of Gaussians with unconstrained 3D positions. Because clustering is applied at inference, a single trained model spans the quality-efficiency curve without retraining. ZipSplat operates without ground-truth poses or intrinsics, yet sets a new state of the art on DL3DV and RealEstate10K with ${\\sim}6{\\times}$ fewer Gaussians than pixel-aligned methods, surpassing the best pose-free baseline by 2.1dB and 1.2dB PSNR, respectively. It further generalizes zero-shot to Mip-NeRF360 and ScanNet++, outperforming all comparable baselines. Our project page is at https://veichta.com/zipsplat.", "date": "2026-06-03", "categories": []}
{"id": "2606.14307", "title": "Pano3D: Unified 3D Reconstruction and Panoptic Segmentation", "abstract": "Recent advances in 3D feedforward reconstruction neural networks have achieved remarkable success in dense reconstruction from images without any camera parameters. Yet, equipping these models with robust semantic understanding remains an open problem. Here we introduce an approach that performs 3D reconstruction and 3D panoptic segmentation in a unified framework. We build on existing 3D reconstruction models and augment them with a set-based mask decoder. The approach is jointly trained with a geometric and semantic loss, which are shown to be mutually beneficial. More precisely, the features are initialized from the geometric information and then finetuned to capture jointly geometry and semantics. We demonstrate the generality of our approach by successfully applying our framework both to online and all-to-all attention reconstruction backbones. Our method achieves state-of-the-art performance in 3D panoptic segmentation across ScanNet, ScanNet200, and ScanNet++ datasets. Ablation studies show that such joint training of a unified model equips 3D feedforward reconstruction neural networks with panoptic segmentation and yields mutually beneficial improvements.", "date": "2026-06-12", "categories": []}
{"id": "2505.01869", "title": "Visual enhancement and 3D representation for underwater scenes: a review", "abstract": "Underwater visual enhancement (UVE) and underwater 3D reconstruction pose significant challenges in computer vision and AI-based tasks due to complex imaging conditions in aquatic environments. Despite the development of numerous enhancement algorithms, a comprehensive and systematic review covering both UVE and underwater 3D reconstruction remains absent. To advance research in these areas, we present an in-depth review from multiple perspectives. First, we introduce the fundamental physical models, highlighting the peculiarities that challenge conventional techniques. We survey advanced methods for visual enhancement and 3D reconstruction specifically designed for underwater scenarios. The paper assesses various approaches from non-learning methods to advanced data-driven techniques, including Neural Radiance Fields and 3D Gaussian Splatting, discussing their effectiveness in handling underwater distortions. Finally, we conduct both quantitative and qualitative evaluations of state-of-the-art UVE and underwater 3D reconstruction algorithms across multiple benchmark datasets. Finally, we highlight key research directions for future advancements in underwater vision.", "date": "2025-05-03", "categories": []}
{"id": "2511.23030", "title": "DiskChunGS: Large-Scale 3D Gaussian SLAM Through Chunk-Based Memory Management", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated impressive results for novel view synthesis with real-time rendering capabilities. However, integrating 3DGS with SLAM systems faces a fundamental scalability limitation: methods are constrained by GPU memory capacity, restricting reconstruction to small-scale environments. We present DiskChunGS, a scalable 3DGS SLAM system that overcomes this bottleneck through an out-of-core approach that partitions scenes into spatial chunks and maintains only active regions in GPU memory while storing inactive areas on disk. Our architecture integrates seamlessly with existing SLAM frameworks for pose estimation and loop closure, enabling globally consistent reconstruction at scale. We validate DiskChunGS on indoor scenes (Replica, TUM-RGBD), urban driving scenarios (KITTI), and resource-constrained Nvidia Jetson platforms. Our method uniquely completes all 11 KITTI sequences without memory failures while achieving superior visual quality, demonstrating that algorithmic innovation can overcome the memory constraints that have limited previous 3DGS SLAM methods.", "date": "2025-11-28", "categories": []}
{"id": "2606.11894", "title": "Wild3R: Feed-Forward 3D Gaussian Splatting from Unconstrained Sparse Photo Collection", "abstract": "Feed-forward 3D Gaussian Splatting (3DGS) removes the need for time-consuming per-scene optimization required by traditional 3DGS. However, existing feed-forward approaches struggle with real-world photo collections that include diverse lighting conditions and transient objects. In this paper, we present Wild3R, a feed-forward approach for unconstrained sparse photo collections. The main bottleneck is the lack of training data that provides multiple viewpoints, a variety of illuminations, and transient variations necessary for learning robust scene representations. To address this, we introduce the WildCity dataset, which comprises 200 scenes, 170 lighting conditions, and transient objects, resulting in 337,500 images in total. By leveraging the dataset, our model learns appearance consistency across viewpoints conditioned on reference views, while removing transient content. Extensive experiments demonstrate that our method outperforms existing feed-forward approaches and achieves results competitive with prior per-scene optimization-based methods.", "date": "2026-06-10", "categories": []}
{"id": "2606.13644", "title": "Surflo: Consistent 3D Surface Flow Model with Global State", "abstract": "Geometry is invariant to viewpoint, which makes any collection of images a redundant encoding of a single 3D state. Existing feed-forward reconstruction models fail to exploit this: per-view methods emit overlapping, unaligned pointmaps that grow linearly with input count, while global-latent methods commit to a fixed, low-resolution output. We introduce Surflo, which compresses a variable number of unposed RGB views into K latent tokens-one global state-and decodes oriented 3D surface points by independently transporting them from noise onto the surface via flow matching. This frees the output from any fixed grid or token budget: the same latent yields from a few thousand to a million points in a single forward pass. To suppress the local inconsistencies inherent to independent per-point decoding, an inference-time guidance term correlates nearby points by injecting a photometric gradient during ODE integration. Surflo matches or surpasses feed-forward baselines on surface metrics, runs an order of magnitude faster than optimization-based methods that require hundreds of views, and is the only feed-forward approach to combine a global latent with arbitrary-resolution decoding.", "date": "2026-06-11", "categories": []}
{"id": "2606.13652", "title": "World Tracing: Generative Pixel-Aligned Geometry Beyond the Visible", "abstract": "Image-to-3D methods often trade off faithfulness and completeness: depth estimators are anchored to input pixels but stop at the visible surface, while image-to-3D models generate complete shapes that are often misaligned with the input. We introduce World Tracing, a generative pixel-aligned geometry representation that predicts 3D points aligned with observed pixels while completing geometry beyond the visible surface. For each input pixel, World Tracing predicts an ordered stack of camera-space 3D points, where the first layer represents the visible surface and subsequent layers represent front-to-back intersections with occluded surfaces. We instantiate this representation with a world-tracing diffusion transformer, WT-DiT, which treats multiple geometry layers as separate denoising tokens coupled through factorized and global attention. WT-DiT is trained with pixel-space flow matching and a mixed noise schedule that balances visible-surface reconstruction with occluded-geometry generation. World Tracing achieves strong performance on visible-surface reconstruction and complete geometry generation across object, scene, and dynamic benchmarks, outperforming both depth predictors and image-to-3D generators. It also preserves 2D-to-3D correspondence, enabling text-driven 3D scene editing, geometry-conditioned novel-view video synthesis, and training-free integration with textured-mesh generators.", "date": "2026-06-11", "categories": []}
{"id": "2606.11326", "title": "DarkVGGT: Seeing Through Darkness Using Thermal Geometry without Daylight Tax", "abstract": "Recent feed-forward 3D reconstruction methods have demonstrated strong performance and flexibility in efficient end-to-end scene geometry estimation from image streams. However, their reliance on visible-light appearance makes them vulnerable in dark and low-visibility environments, where RGB cues are severely degraded and geometric evidence becomes ambiguous. To address this challenge, we propose DarkVGGT, an RGB-T feed-forward geometry framework that uses physics-aware thermal modeling for robust 3D estimation in low-light scenes. DarkVGGT introduces two complementary modules. First, physics-inspired thermal factorization extracts emissive-dominant, geometry-consistent thermal cues while isolating sparse reflective residuals that may introduce geometric ambiguity. Second, geometry-shared thermal routing isolates modality-invariant geometric structures from thermal-specific patterns, selectively injecting reliability-aware structural guidance into the RGB stream. Together, these components enable accurate thermal-informed geometry estimation under degraded RGB conditions while largely preserving performance in well-lit environments. Experiments on low-visibility RGB-T benchmarks demonstrate consistent improvements in both depth and camera pose estimation over existing feed-forward geometry baselines.", "date": "2026-06-09", "categories": []}
{"id": "2606.11390", "title": "A Scalable PyTorch Abstraction for Multi-GPU Gaussian Splatting", "abstract": "Gaussian splatting methods have become increasingly popular for neural reconstruction of the real world. However, they are often limited in scale and resolution due to compute and memory constraints. We present a multi-GPU Gaussian splatting approach that scales reconstruction to higher resolutions and larger scenes while abstracting away the code complexity typically associated with distributing a model. To accomplish this, we propose a PyTorch backend that distributes the Gaussian parameters and splatting operators across GPUs via CUDA unified memory and NVLink. Because distribution occurs at the operator level, the model code requires no explicit cross-device communication. More broadly, the backend exposes multiple GPUs as an aggregate PyTorch device and supports other PyTorch operators. We demonstrate city-scale reconstructions with street-level detail consisting of over 1 billion Gaussian splats, more than 25 times as many as the current state of the art.", "date": "2026-06-09", "categories": []}
{"id": "2606.11841", "title": "Scene-Adaptive Nonlinear Tone Curves for Pseudo Ground-Truth Generation in Low-Light 3D Gaussian Splatting", "abstract": "Low-light novel view synthesis is challenging because dark multi-view images contain noise, weak structural detail, and compressed dynamic range. Recent 3D Gaussian Splatting (3DGS) methods address these challenges by generating pseudo ground-truth (pseudo-GT) images as supervision targets when paired normal-light references are unavailable. Existing pseudo-GT methods apply a uniform linear gain to all pixels, which clips bright regions while providing insufficient enhancement in dark regions, limiting reconstruction quality. We observe that nonlinear tone mappings, long established in 2D low-light enhancement, have not been explored for pseudo-GT generation in 3D reconstruction. Accordingly, we propose a scene-adaptive nonlinear tone-curve framework that replaces linear pseudo-GT with nonlinear alternatives. The framework introduces percentile-based normalisation for scene-agnostic curve application, a scene-adaptive offset for automatic black-level adjustment, and two complementary curves: Adaptive SoftExp (ASE), a bounded exponential curve, and Adaptive Poly3 (AP3), a data-driven cubic polynomial. The module changes only the pseudo-GT computation and leaves the 3DGS backbone unchanged. Experiments on three benchmarks covering 21 scenes show that both curves consistently outperform the linear baseline with PSNR improvements up to +4.34 dB on LOM and +3.25 dB on RealX3D. Both curves achieve similar performance despite their different mathematical forms, suggesting the improvement is curve-agnostic. Code is available at https://github.com/lvmingzhe/adaptiveToneCurve", "date": "2026-06-10", "categories": []}
{"id": "2504.18424", "title": "LaRI: Layered Ray Intersections for Single-view 3D Geometric Reasoning", "abstract": "We present Layered Ray Intersections (LaRI), a fully supervised method for occluded geometry reasoning from a single image. Unlike conventional depth estimation, which is limited to visible surfaces, LaRI predicts multiple surfaces intersected by the camera rays using layered point maps. Compared to the existing approaches that leverage neural implicit representations or iterative refinement, LaRI achieves complete scene reconstruction in one feed-forward pass, enabling efficient and view-aligned geometric reasoning to underpin both object-level and scene-level tasks. We further propose to predict the ray stopping index, which identifies valid intersecting pixels and layers from LaRI's output. To better underpin and evaluate this task, we build an annotation pipeline using rendering engines, construct annotations for five public datasets, including synthetic and real-world data covering 3D objects and scenes. As a generic method, LaRI's performance is validated in object-level and scene-level reconstruction tasks.", "date": "2025-04-25", "categories": []}
{"id": "2507.13595", "title": "NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision", "abstract": "Reconstructing accurate implicit surface representations from point clouds remains a challenging task, particularly when data is captured using low-quality scanning devices. These point clouds often contain substantial noise, leading to inaccurate surface reconstructions. Inspired by the Noise2Noise paradigm for 2D images, we introduce NoiseSDF2NoiseSDF, a novel method designed to extend this concept to 3D neural fields. Our approach enables learning clean neural SDFs from noisy point clouds through noisy supervision by minimizing the MSE loss between noisy SDF representations, allowing the network to implicitly denoise and refine surface estimations. We evaluate the effectiveness of NoiseSDF2NoiseSDF on benchmarks, including the ShapeNet, ABC, Famous, and Real datasets. Experimental results demonstrate that our framework significantly improves surface reconstruction quality from noisy inputs.", "date": "2025-07-18", "categories": []}
{"id": "2606.10364", "title": "Benchmarking stereo reconstruction for 3D printable Martian terrain models", "abstract": "Reconstructing printable 3D models from Mars rover imagery is challenging because Martian terrain is low-texture, irregular, and partially observed. We evaluate a pipeline that estimates stereo depth from NASA Curiosity images, completes geometry, and exports watertight OBJ meshes. On Middlebury, RAFT-Stereo outperforms semi-global block matching (SGBM), reducing disparity MAE from 3.22px to 0.73px and increasing valid prediction coverage from 76.3% to 100.0%. On Curiosity imagery, however, RAFT's denser disparities show weaker edge alignment and higher photometric reprojection error, suggesting that benchmark accuracy does not directly transfer to Martian terrain reconstruction. Geometry completion demonstrates a tradeoff between local fidelity and global connectivity. We find that alpha shapes preserve accurate but fragmented structure, Poisson reconstruction produces more coherent meshes but adds unsupported surfaces, and a deterministic diffusion-fill baseline is intermediate but sensitive to stereo quality. Overall, standard stereo and completion methods can produce printable approximations of Martian terrain, but reliable reconstruction requires stronger domain-specific validation.", "date": "2026-06-09", "categories": []}
{"id": "2606.10478", "title": "3D-CoS: A New 3D Reconstruction Paradigm Based on VLM Code Synthesis", "abstract": "Most recent 3D reconstruction and editing systems operate on implicit and explicit representations such as NeRF, point clouds, or meshes. While these representations enable high-fidelity rendering, they are fundamentally low-level and hard to control programmatically. In contrast, we propose and systematically evaluate a new 3D reconstruction paradigm, 3D Code Synthesis (3D-CoS), where 3D assets are constructed as executable Blender code, a programmatic and interpretable medium. To assess how well current VLMs can use code to represent 3D objects, we evaluate representative open-source and closed-source VLMs in code-based reconstruction under a unified protocol. We further introduce a suite of structured code-synthesis workflows, including blueprint-based planning, Retrieval-Augmented Generation (RAG) over Blender API documentation, few-shot geometric demonstrations, and a component-level Agent workflow for part-wise code generation. To demonstrate the unique advantages of this representation, we further evaluate localized text-driven modifications and compare our code-based edits with a point-cloud-based 3D editing baseline. Our study shows that code as a 3D representation offers strong controllability and locality, yielding stronger edit fidelity and better preservation of unedited regions in our targeted editing evaluation. Our work also analyzes the potential of this paradigm, delineates the current capability frontier of VLMs for programmatic 3D modeling, and highlights code synthesis as a promising direction for editable 3D reconstruction.", "date": "2026-06-09", "categories": []}
{"id": "2606.10645", "title": "ManiSplat: Manipulation Trajectory Synthesis from Monocular Video via Decoupled 3D Gaussian Splatting", "abstract": "Reconstructing dynamic and interactive 3D scenes from real-world observations remains a fundamental challenge in computer vision and robotics. While recent advances in 3D Gaussian Splatting have enabled high-fidelity static reconstruction, extending it to interactive environments with articulated robots and manipulable objects remains difficult due to complex contact interactions and abrupt pose changes. To address these challenges, we introduce ManiSplat, a unified framework that reconstructs controllable and decoupled Gaussian digital twins directly from monocular ego-view robotic videos. Our method introduces a Graph-Structured Disentangled Representation that separates the robot, objects, and background into independently optimizable Gaussian subfields organized within a scene graph. To ensure stability, we propose a Task-Oriented Spatio-Temporal Alignment module that leverages the inherent logic of manipulation tasks-alternating between Motion and Skill phases-to construct accurate pseudo-ground-truth trajectories. Finally, a joint photometric-geometric optimization ensures the reconstructed scenes are temporally coherent, physically consistent, and simulation-ready. Extensive experiments demonstrate that our approach reconstructs interaction-driven dynamic scenes with high fidelity and controllability, effectively supporting downstream robotic tasks and policy learning.", "date": "2026-06-09", "categories": []}
{"id": "2604.22482", "title": "Holo360D: A Large-Scale Real-World Dataset with Continuous Trajectories for Advancing Panoramic 3D Reconstruction and Beyond", "abstract": "While feed-forward 3D reconstruction models have advanced rapidly, they still exhibit degraded performance on panoramas due to spherical distortions. Moreover, existing panoramic 3D datasets are predominantly collected with 360 cameras fixed at discrete locations, resulting in discontinuous trajectories. These limitations critically hinder the development of panoramic feed-forward 3D reconstruction, especially for the multi-view setting. In this paper, we present Holo360D, a comprehensive dataset containing 109,495 panoramas paired with registered point clouds, meshes, and aligned camera poses. To our knowledge, Holo360D is the first large-scale dataset that provides continuous panoramic sequences with accurately aligned high-completeness depth maps. The raw data are initially collected using a 3D laser scanner coupled with a 360 camera. Subsequently, the raw data are processed with both online and offline SLAM systems. Furthermore, to enhance the 3D data quality, a post-processing pipeline tailored for the 360 dataset is proposed, including geometry denoising, mesh hole filling, and region-specific remeshing. Finally, we establish a new benchmark by fine-tuning 3D reconstruction models on Holo360D, providing key insights into effective fine-tuning strategies. Our results demonstrate that Holo360D delivers superior training signals and provides a comprehensive benchmark for advancing panoramic 3D reconstruction models. Datasets and Code will be made publicly available.", "date": "2026-04-24", "categories": []}
{"id": "2606.07932", "title": "LEGS: Laplacian-Enhanced Gaussian Splatting with a Nonlinear Weighted Loss", "abstract": "3D Gaussian Splatting (3DGS) has become an efficient explicit representation for radiance field reconstruction and real-time novel view synthesis. However, its standard photometric loss treats flat and structure-rich regions similarly, which may limit the recovery of sharp contours and fine details. Edge-Guided Gaussian Splatting (EGGS) improves structure awareness through edge-guided weighting, but mainly relies on first-order gradient responses and linear weighting. In this paper, we propose LEGS, a Laplacian-Enhanced Gaussian Splatting method with a nonlinearly weighted loss. LEGS replaces first-order gradient guidance with second-order Laplacian structural guidance and maps the normalized Laplacian response into pixel-wise weights through nonlinear response-to-weight functions. The proposed loss improves structure-aware Gaussian optimization while keeping the original 3DGS rendering pipeline unchanged. Experiments on the full Tanks\\&Temples and Mip-NeRF360 datasets show that LEGS improves peak signal-to-noise ratio (PSNR) by up to 1.68 dB over 3DGS and up to 0.52 dB over EGGS. Incorporating the proposed second-order nonlinear weighting strategy into FastGS and FasterGS further improves PSNR by up to 1.69 dB, demonstrating its effectiveness as a general loss-level extension for Gaussian Splatting pipelines with potential applications in AR/VR, immersive visualization, and real-time 3D content generation.", "date": "2026-06-06", "categories": []}
{"id": "2606.08205", "title": "Empowering Feed-Forward Reconstruction Models with Metric Scale via Satellite Images", "abstract": "Feed-forward 3D reconstruction models have recently shown strong generalization across diverse scenes, yet most of them recover geometry only up to an unknown global scale. This scale ambiguity limits their use in applications that require metric understanding of the environment. Existing metric reconstruction methods commonly rely on large-scale metric annotations or accurate camera calibration, both of which are costly or unreliable in many real-world settings. We propose a satellite-guided framework for resolving scale ambiguity in feed-forward 3D reconstruction. The key idea is to use readily available satellite imagery as a global metric reference. Given a coarse camera pose, our method retrieves a local satellite patch and integrates it with a feed-forward reconstruction backbone through bidirectional cross-view interaction. By enforcing consistency between the reconstructed scene and the satellite reference, the model infers absolute scale, refines scene geometry, and estimates camera pose in a metric coordinate frame. Experiments on KITTI, nuScenes, and Oxford RobotCar show consistent improvements in metric depth estimation, multi-view point-cloud reconstruction, and cross-view camera localization, while preserving strong generalization across datasets and geographic regions.", "date": "2026-06-06", "categories": []}
{"id": "2510.17568", "title": "PAGE-4D: VGGT-4D Perception via Disentangled Pose and Geometry Estimation", "abstract": "Recent 3D feed-forward models, such as the Visual Geometry Grounded Transformer (VGGT), have shown strong capability in inferring 3D attributes of static scenes. However, since they are typically trained on static datasets, these models often struggle in real-world scenarios involving complex dynamic elements, such as moving humans or deformable objects like umbrellas. To address this limitation, we introduce PAGE-4D, a feedforward model that extends VGGT to dynamic scenes, enabling camera pose estimation, depth prediction and point cloud reconstruction - all without post-processing. A central challenge in multitask 4D reconstruction is the inherent conflict between tasks: accurate camera pose estimation requires suppressing dynamic regions, while geometry reconstruction requires modeling them. To resolve this tension, we propose a dynamics aware aggregator that disentangles static and dynamic information by predicting a dynamics-aware mask - suppressing motion cues for pose estimation while amplifying them for geometry reconstruction. Extensive experiments show that PAGE-4D consistently outperforms the original VGGT in dynamic scenarios, achieving superior results in camera pose estimation, monocular and video depth estimation, and dense point map reconstruction. Necessary code and additional demos are available at Link: https://page4d.github.io/, including both the training-and-inference masking variant and the training-only masking variant (= VGGT architecture at inference). Keywords: VGGT-4D, 4D Perception, Dynamic Scene Reconstruction.", "date": "2025-10-20", "categories": []}
{"id": "2511.14019", "title": "RISE: Single Static Radar-based Indoor Scene Understanding", "abstract": "Robust and privacy-preserving indoor scene understanding remains a fundamental open problem. While optical sensors such as RGB and LiDAR offer high spatial fidelity, they suffer from severe occlusions and introduce privacy risks in indoor environments. In contrast, millimeter-wave (mmWave) radar preserves privacy and penetrates obstacles, but its inherently low spatial resolution makes reliable geometric reasoning difficult. We introduce RISE, the first benchmark and system for single-static-radar indoor scene understanding, jointly targeting layout reconstruction and object detection. RISE is built upon the key insight that multipath reflections-traditionally treated as noise-encode rich geometric cues. To exploit this, we propose a Bi-Angular Multipath Enhancement that explicitly models Angle-of-Arrival and Angle-of-Departure to recover secondary (ghost) reflections and reveal invisible structures. On top of these enhanced observations, a simulation-to-reality Hierarchical Diffusion framework transforms fragmented radar responses into complete layout reconstruction and object detection. Our benchmark contains 50,000 frames collected across 100 real indoor trajectories, forming the first large-scale dataset dedicated to single, static, radar-based indoor scene understanding. Extensive experiments show that RISE reduces the Chamfer Distance by 60% (down to 16 cm) compared to the state of the art in mmWave layout reconstruction, and delivers the first mmWave-based object detection, achieving 58% IoU. These results establish RISE as a new foundation for geometry-aware and privacy-preserving indoor scene understanding using a single static radar. Our website and code are available at https://rise-cvpr.github.io.", "date": "2025-11-18", "categories": []}
{"id": "2604.10578", "title": "Rein3D: Reinforced 3D Indoor Scene Generation with Panoramic Video Diffusion Models", "abstract": "The growing demand for Embodied AI and VR applications has highlighted the need for synthesizing high-quality 3D indoor scenes from sparse inputs. However, existing approaches struggle to infer massive amounts of missing geometry in large unseen areas while maintaining global consistency, often producing locally plausible but globally inconsistent reconstructions. We present Rein3D, a framework that reconstructs full 360-degree indoor environments by coupling explicit 3D Gaussian Splatting (3DGS) with temporally coherent priors from video diffusion models. Our approach follows a \"restore-and-refine\" paradigm: we employ a radial exploration strategy to render imperfect panoramic videos along trajectories starting from the origin, effectively uncovering occluded regions from a coarse 3DGS initialization. These sequences are restored by a panoramic video-to-video diffusion model and further enhanced via video super-resolution to synthesize high-fidelity geometry and textures. Finally, these refined videos serve as pseudo-ground truths to update the global 3D Gaussian field. To support this task, we construct PanoV2V-15K, a dataset of over 15K paired clean and degraded panoramic videos for diffusion-based scene restoration. Experiments demonstrate that Rein3D produces photorealistic and globally consistent 3D scenes and significantly improves long-range camera exploration compared with existing baselines.", "date": "2026-04-12", "categories": []}
{"id": "2606.06690", "title": "RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery", "abstract": "We present RPC-GS, the first Gaussian Splatting framework for satellite imagery that operates natively with Rational Polynomial Camera (RPC) models. The RPC model is the de facto standard for representing the complex imaging geometry of modern pushbroom satellite sensors. To simplify rendering, prior satellite Gaussian Splatting methods replace the RPC model with perspective or affine camera approximations, leading to geometric errors during reconstruction. RPC-GS avoids these approximations by projecting Gaussian means and covariances directly through the RPC model during the splatting process. We embed the RPC model in a chain of carefully selected geo-coordinate transformations representing a mapping from splatting-suitable scene coordinates to image coordinates. To map the Gaussian covariance matrices, we derive a numerically robust Jacobian-based covariance projection for the (partially nonlinear) coordinate transformations. Since RPCs lack an explicit notion of camera depth, we integrate a metric ray-based depth formulation. We benchmark RPC, perspective, and affine camera models in a unified framework, with our native RPC renderer consistently achieving the lowest reconstruction error on leading satellite benchmark datasets, improving mean altitude error over perspective and affine approximations by 29.6% and 63.8% on DFC2019, and by 9.9% and 37.9% on IARPA2016. We release our code to support future research of Gaussian Splatting in the satellite imaging domain.", "date": "2026-06-04", "categories": []}
{"id": "2606.06695", "title": "S23DR 2026 Winning Solution", "abstract": "This text presents the winning solution to the S23DR 2026 challenge for structured 3D wireframe reconstruction from sparse SfM, fitted depth, and semantic segmentations. The method treats vertices as a conditional set and denoises 64 vertex tokens with a flow-matching DiT conditioned on Perceiver-style scene tokens. A global pass predicts the coarse structure, a hull-cropped second pass refines it, and a small multi-sample consensus step keeps the stochastic sampler well behaved. The final system ranked first on the private leaderboard, achievingHSS = 0.654.", "date": "2026-06-04", "categories": []}
{"id": "2606.07288", "title": "ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation", "abstract": "Reconstructing surface meshes from multi-view images has remained a core challenge in recent years. Most existing methods, whether implicit or explicit, depend on intermediate representations and post-processing steps like Marching Cubes or TSDF fusion, often resulting in artifacts and fragmented geometry. Directly optimizing explicit meshes is a promising approach. However, it presents two critical challenges. The first is how to adaptively refine mesh topology to capture detail without introducing degenerate faces. The second is how to maintain consistent UV coordinates for high-fidelity texturing as the mesh structure evolves. To overcome these, we propose ExMesh, a novel framework that directly optimizes explicit meshes by integrating differentiable optimization with discrete topology updates. Specifically, we introduce an adaptive vertex splitting and merging strategy, along with real-time UV maintenance, to enable coarse-to-fine optimization while preserving geometric integrity. To our knowledge, ExMesh is the first framework to seamlessly integrate discrete topology operations into a continuous differentiable optimization pipeline. Extensive experiments demonstrate that ExMesh achieves a balance among accuracy, computational efficiency, and mesh conciseness.", "date": "2026-06-05", "categories": []}
{"id": "2606.01573", "title": "$\\text{VG}^2$GT: Voxel-Gaussian Splatting Visual Geometry Grounded Transformer", "abstract": "Gaussian splatting has shown strong potential for 3D reconstruction and novel view synthesis. However, most existing methods require accurate camera parameters and per-scene optimization, while feed-forward methods with pixel-aligned Gaussian primitives often suffer from artifacts and non-uniform primitives. In this paper, we propose $\\text{VG}^2$GT, a Voxel-Gaussian Splatting Visual Geometry-Grounded Transformer. $\\text{VG}^2$GT leverages a frozen pretrained visual foundation model (VFM), incorporates a multi-scale differentiable voxel module to enhance geometric understanding, and directly splits and regresses Gaussian primitive parameters from voxel features. During training, depth maps are supervised through stochastic solid volume rendering, enabling geometrically accurate Gaussian scene reconstruction while keeping the visual foundation model fully frozen. This design enables $\\text{VG}^2$GT to be seamlessly plugged into any patch-feature-based VFM, while substantially reducing the required training cost. $\\text{VG}^2$GT outperforms current state-of-the-art methods on widely used DTU, Replica, TAT, and ScanNet datasets.", "date": "2026-06-01", "categories": []}
{"id": "2606.04593", "title": "4D Reconstruction from Sparse Dynamic Cameras", "abstract": "Although dynamic 3D (i.e., 4D) reconstruction from a monocular dynamic camera has recently advanced, it remains fundamentally limited by depth ambiguity. In this paper, we focus on an alternative practical way, i.e., sparse dynamic camera setup, where a handful of independently moving cameras capture the same subjects. While keeping capture costs low, this setup introduces multi-view constraints and remains practical for real-world video production such as sports, concerts, and TV shows. Despite its potential, our experiments show that naive extensions of existing monocular or dense-fixed camera-based methods are insufficient since they fail to resolve the complex spatiotemporal inconsistencies across views and time. To fill this gap, we propose a simple yet effective 3D track initialization method designed to ensure spatiotemporal consistency by integrating inter-camera feature matching with intra-camera point tracking. Additionally, we incorporate a noise-robust depth-ordering regularization loss and a spatiotemporally diverse batch sampling strategy to enhance optimization stability and cross-view generalization. Furthermore, to address the lack of standardized benchmarks for this task, we introduce LetCamsGo, a new real-world video dataset with 5 sequences across 4 diverse environments, recorded by three independently moving cameras and one fixed camera. Comprehensive benchmarking on LetCamsGo demonstrated that our proposed framework improves 4D reconstruction quality in dynamic regions compared with baselines, paving the way for a low-cost 4D reconstruction paradigm in the wild.", "date": "2026-06-03", "categories": []}
{"id": "2606.04891", "title": "Hierarchical Space Partition for Surface Reconstruction", "abstract": "Generating compact polygonal models from point clouds is a key problem in 3D vision and computer graphics. However, due to inherent limitations of LiDAR scanning (e.g. range constraints and occlusions), critical scene information is often missing, leading to degraded reconstruction accuracy. To address this, we propose a plane assembling strategy that effectively recovers missing details while maintaining model compactness. We classify all the planes extracted from the scene into three categories: highly visible, barely visible, and invisible. The invisible planes, which are recovered by scene structure analysis, indicate the missing details. The three types of planes correspond to the three growth priorities. Each plane grows according to the priority level, and the space is partitioned progressively, namely, the hierarchical partition. Subsequently, we generate a watertight polygonal mesh from the partition via a min-cut-based optimization. Finally, comparisons on public datasets show the effectiveness and superiority of our method against mainstream approaches. The project page is available at https://hsr-3dv.github.io/.", "date": "2026-06-03", "categories": []}
{"id": "2606.05035", "title": "Anchor3R: Streaming 3D Reconstruction with Transient Anchors for Long-Horizon Visual Mapping", "abstract": "Long-horizon online visual mapping is a core capability for robot perception, requiring continuous camera-motion and scene-geometry estimation from visual streams under bounded memory and computation. Recent feed-forward 3D reconstruction models provide strong geometric priors, but their streaming variants often predict poses in a fixed coordinate system tied to the first frame or a persistent scene memory. This fixed-gauge design leads to train--test mismatch, attention bias toward early anchors, and accumulated drift on sequences much longer than those seen during training. We propose \\emph{Anchor3R}, a streaming 3D reconstruction framework that treats feed-forward reconstruction as current-centric local measurement prediction rather than persistent global-gauge regression. At each time step, Anchor3R predicts window-relative poses and a local pointmap in the current-frame coordinate system, turning streaming reconstruction into relative-pose measurement generation. These measurements support online pose updates, while loop-closure reinsertion and motion averaging align the trajectory and transform local pointmaps into a coherent global reconstruction. Experiments on indoor, outdoor, driving, and RGB-D benchmarks show that Anchor3R improves long-horizon pose accuracy and dense reconstruction quality over existing streaming baselines, while supporting bounded-memory online inference.", "date": "2026-06-03", "categories": []}
{"id": "2406.18544", "title": "GS-ROR$^2$: Bidirectional-guided 3DGS and SDF for Reflective Object Relighting and Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has shown a powerful capability for novel view synthesis due to its detailed expressive ability and highly efficient rendering speed. Unfortunately, creating relightable 3D assets and reconstructing faithful geometry with 3DGS is still problematic, particularly for reflective objects, as its discontinuous representation raises difficulties in constraining geometries. Volumetric signed distance field (SDF) methods provide robust geometry reconstruction, while the expensive ray marching hinders its real-time application and slows the training. Besides, these methods struggle to capture sharp geometric details. To this end, we propose to guide 3DGS and SDF bidirectionally in a complementary manner, including an SDF-aided Gaussian splatting for efficient optimization of the relighting model and a GS-guided SDF enhancement for high-quality geometry reconstruction. At the core of our SDF-aided Gaussian splatting is the mutual supervision of the depth and normal between blended Gaussians and SDF, which avoids the expensive volume rendering of SDF. Thanks to this mutual supervision, the learned blended Gaussians are well-constrained with a minimal time cost. As the Gaussians are rendered in a deferred shading mode, the alpha-blended Gaussians are smooth, while individual Gaussians may still be outliers, yielding floater artifacts. Therefore, we introduce an SDF-aware pruning strategy to remove Gaussian outliers located distant from the surface defined by SDF, avoiding floater issue. This way, our GS framework provides reasonable normal and achieves realistic relighting, while the mesh from depth is still problematic. Therefore, we design a GS-guided SDF refinement, which utilizes the blended normal from Gaussians to finetune SDF. With this enhancement, our method can further provide high-quality meshes for reflective objects at the cost of 17% extra training time.", "date": "2024-05-22", "categories": []}
{"id": "2512.18954", "title": "VOIC: Visible-Occluded Integrated Guidance for 3D Semantic Scene Completion", "abstract": "Camera-based 3D Semantic Scene Completion (SSC) is a critical task for autonomous driving and robotic scene understanding. It aims to infer a complete 3D volumetric representation of both semantics and geometry from a single image. Existing methods typically focus on end-to-end 2D-to-3D feature lifting and voxel completion. However, they often overlook the interference between high-confidence visible-region perception and low-confidence occluded-region reasoning caused by single-image input, which can lead to feature dilution and error propagation.\n  To address these challenges, we introduce an offline Visible Region Label Extraction (VRLE) strategy that explicitly separates and extracts voxel-level supervision for visible regions from dense 3D ground truth. This strategy purifies the supervisory space for two complementary sub-tasks: visible-region perception and occluded-region reasoning. Building on this idea, we propose the Visible-Occluded Interactive Completion Network (VOIC), a novel dual-decoder framework that explicitly decouples SSC into visible-region semantic perception and occluded-region scene completion. VOIC first constructs a base 3D voxel representation by fusing image features with depth-derived occupancy. The visible decoder focuses on generating high-fidelity geometric and semantic priors, while the occlusion decoder leverages these priors together with cross-modal interaction to perform coherent global scene reasoning.\n  Extensive experiments on the SemanticKITTI and SSCBench-KITTI360 benchmarks demonstrate that VOIC outperforms existing monocular SSC methods in both geometric completion and semantic segmentation accuracy, achieving state-of-the-art performance.", "date": "2025-12-22", "categories": []}
{"id": "2603.07664", "title": "Ref-DGS: Reflective Dual Gaussian Splatting", "abstract": "The reflective appearance, especially strong and typically near-field specular reflections, poses a fundamental challenge for accurate surface reconstruction and novel view synthesis. Existing Gaussian splatting methods either fail to model near-field specular reflections or rely on explicit ray tracing at substantial computational cost. We present \\textbf{Ref-DGS}, a reflective dual Gaussian splatting framework that addresses this trade-off by decoupling surface reconstruction from specular reflection within an efficient rasterization-based pipeline. Ref-DGS introduces a dual Gaussian scene representation consisting of geometry Gaussians and complementary local reflection Gaussians that capture near-field specular interactions without explicit ray tracing, along with a global environment reflection field for modeling far-field specular reflections. To predict specular radiance, we further propose a lightweight, physically-aware specular adaptive mixing shader that fuses global and local specular features. Experiments demonstrate that Ref-DGS achieves state-of-the-art performance on reflective scenes while training substantially faster than ray-based Gaussian methods.", "date": "2026-03-08", "categories": []}
{"id": "2603.27455", "title": "From None to All: Self-Supervised 3D Reconstruction via Novel View Synthesis", "abstract": "In this paper, we introduce NAS3R, a self-supervised feed-forward framework that jointly learns explicit 3D geometry and camera parameters with no ground-truth annotations and no pretrained priors. During training, NAS3R reconstructs 3D Gaussians from uncalibrated and unposed context views and renders target views using its self-predicted camera parameters, enabling self-supervised training from 2D photometric supervision. To ensure stable convergence, NAS3R integrates reconstruction and camera prediction within a shared transformer backbone regulated by masked attention, and adopts a depth-based Gaussian formulation that facilitates well-conditioned optimization. The framework is compatible with state-of-the-art supervised 3D reconstruction architectures and can incorporate pretrained priors or intrinsic information when available. Extensive experiments show that NAS3R achieves superior results to other self-supervised methods, establishing a scalable and geometry-aware paradigm for 3D reconstruction from unconstrained data. Code and models are publicly available at https://ranrhuang.github.io/nas3r/.", "date": "2026-03-29", "categories": []}
{"id": "2606.03287", "title": "BA-T: An Iterative Transformer for Two-View Bundle Adjustment", "abstract": "Feed-forward models for 3D reconstruction have achieved strong performance using deep cross-view attention to exchange information across images. However, these approaches often depend on heavy decoder stacks and lack a structured mechanism for geometry refinement, resulting in poor multi-view consistency. We address this by drawing inspiration from classical bundle adjustment (BA), which can be viewed as an iterative information propagation process between poses and local geometry. Inspired by BA, we propose BA-T, an iterative Transformer that implements BA-style structured updates as a repeatable layer in implicit token space. Instead of relying on deep attention stacks, BA-T refines predictions based on latent residual by a single lightweight layer. Experiments demonstrate that BA-T progressively improves pose and reconstruction accuracy across iterations, achieves stronger cross-view consistency than conventional decoders, and matches or surpasses substantially larger models while using only 16% of their decoder parameters. BA-T provides a compact, efficient, and structural alternative to depth-heavy attention, enabling accurate 3D reconstruction within a lightweight architecture. The code will be made publicly at https://github.com/zhangganlin/BA-T.", "date": "2026-06-02", "categories": []}
{"id": "2606.03479", "title": "PersistGS: Differentiable Physics for Object Permanence in 4D Gaussian Splatting", "abstract": "Dynamic 3D Gaussian Splatting (3DGS) methods reconstruct time-varying scenes from synchronized multi-camera video using photometric supervision. When a moving object becomes fully occluded from all training cameras, this supervision vanishes: the Gaussians representing it receive no gradient signal and degrade. Existing approaches to incomplete observations in neural reconstruction rely on learned generative priors that prioritize visual plausibility over physical correctness.\n  We propose $\\textbf{PersistGS}$, a method that restores object permanence during occlusion by coupling differentiable rigid body simulation with 3D Gaussian Splatting. Our approach decomposes the scene into per-object Gaussians and collision meshes, estimates friction and velocity from the observed pre-occlusion trajectory via differentiable simulation, and uses the resulting SE(3) trajectory to position object Gaussians throughout the occlusion period. Because the predicted trajectory satisfies the governing equations of rigid body dynamics, it faithfully captures contact events (bounces, friction-based deceleration, direction changes) that kinematic extrapolation cannot model. We introduce a centroid silhouette loss that isolates positional gradients from appearance noise, yielding 40% lower trajectory error than photometric supervision. We evaluate using cameras withheld from training that observe the object during its occlusion. Experiments on synthetic scenes show that PersistGS outperforms constant velocity extrapolation by +2.46dB PSNR and comes within 0.19dB of a ground-truth trajectory upper bound.", "date": "2026-06-02", "categories": []}
{"id": "2606.03915", "title": "PatchScene: Patch-based Voxel Diffusion for Large-Scale Scene Completion", "abstract": "We propose PatchScene, a novel diffusion-based framework for large-scale LiDAR scene completion. Unlike existing methods that rely on global latent representations or dense voxel grids, PatchScene adopts a patch-based voxel diffusion paradigm that explicitly generates fine-grained geometry within localized 3D regions. To ensure coherent reconstruction at both spatial and temporal scales, we introduce a confidence-guided spatio-temporal fusion mechanism that integrates overlapping patches and adjacent frames in a unified generative process. Furthermore, we design an Annular-Flow diffusion strategy that leverages the radial density pattern of LiDAR scans to progressively propagate high-fidelity information from near-range to far-range regions, enabling spatially unbounded scene completion. Extensive experiments on the SemanticKITTI benchmark demonstrate that PatchScene achieves state-of-the-art performance across all standard metrics, surpassing previous approaches in both geometric accuracy and temporal consistency. Remarkably, the model trained on 20 m LiDAR ranges generalizes effectively to 50 m scenes without retraining, highlighting its strong scalability and generalization capability for real-world autonomous driving applications.", "date": "2026-06-02", "categories": []}
{"id": "2606.03921", "title": "GARDEN: Gravity-Aligned Reconstruction of Disentangled ENvironments from RGB images", "abstract": "Converting multi-view RGB observations into simulation-ready 3D environments remains challenging because current reconstruction pipelines produce monolithic scene representations without explicit physical structure. They are typically defined up to an arbitrary global rotation and entangle rigid foreground objects with background geometry, which hinders stable physical interaction. Existing solutions often recover interactivity by replacing reconstructed objects with retrieved CAD assets, but this introduces a slow retrieval-and-replacement stage and weakens scene-specific geometric fidelity. We propose GARDEN, an RGB-only framework that reformulates reconstruction as physically-grounded scene factorization and outputs a structured hybrid scene representation. The key idea is to use gravity as a universal physical prior: we first align the reconstruction to a unified Gravity-View frame to resolve gauge ambiguity, then recover object-centric rigid meshes with accurate 6-DoF placement, and finally remove duplicate object geometry from the background through conditional 3D point classification. The resulting representation combines explicit rigid bodies with a decoupled background, enabling direct physics simulation while preserving visual realism. Experiments on both simulated and real multi-view scenes show that GARDEN improves object placement reliability, disentanglement quality, and rendering-simulation efficiency compared with retrieval-based baselines.", "date": "2026-06-02", "categories": []}
{"id": "2606.03994", "title": "SimuScene: Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image", "abstract": "Reconstructing interactive, simulation-ready 3D scenes from a single image is a critical bottleneck for robotic manipulation. While recent single-image lifters recover plausible per-object shapes, composing them yields scenes that collapse under physical simulation due to interpenetrating, hovering, or sinking objects. Existing physics-aware methods address this strictly as a post-hoc layout correction, leaving the underlying geometric errors unresolved. To address this, we introduce SimuScene, a compositional 3D reconstruction pipeline that puts physics in the loop of shape and layout estimation. Rather than using physics merely for layout cleanup, we utilize the physics engine as a diagnostic measurement tool during the generative process itself. By diagnostically simulating reconstructed objects under gravity, we convert penetration and support failures into quantitative correction signals that drive gravity-axis stretching and amodal shape resampling. This physics-informed feedback loop mitigates accumulated reconstruction errors and produces a stable, simulation-ready compositional 3D scene. Extensive experiments demonstrate state-of-the-art performance on physical stability and geometric alignment benchmarks. We further highlight SimuScene's utility by deploying reconstructed environments in humanoid control and robot-arm manipulation tasks.", "date": "2026-06-02", "categories": []}
{"id": "2505.08438", "title": "A Survey of 3D Reconstruction with Event Cameras", "abstract": "Event cameras are rapidly emerging as powerful vision sensors for 3D reconstruction, uniquely capable of asynchronously capturing per-pixel brightness changes. Compared to traditional frame-based cameras, event cameras produce sparse yet temporally dense data streams, enabling robust and accurate 3D reconstruction even under challenging conditions such as high-speed motion, low illumination, and extreme dynamic range scenarios. These capabilities offer substantial promise for transformative applications across various fields, including autonomous driving, robotics, aerial navigation, and immersive virtual reality. In this survey, we present the first comprehensive review exclusively dedicated to event-based 3D reconstruction. Existing approaches are systematically categorised based on input modality into stereo, monocular, and multimodal systems, and further classified according to reconstruction methodologies, including geometry-based techniques, deep learning approaches, and neural rendering techniques such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Within each category, methods are chronologically organised to highlight the evolution of key concepts and advancements. Furthermore, we provide a detailed summary of publicly available datasets specifically suited to event-based reconstruction tasks. Finally, we discuss significant open challenges in dataset availability, standardised evaluation, effective representation, and dynamic scene reconstruction, outlining insightful directions for future research. This survey aims to serve as an essential reference and provides a clear and motivating roadmap toward advancing the state of the art in event-driven 3D reconstruction.", "date": "2025-05-13", "categories": []}
{"id": "2507.23277", "title": "iLRM: An Iterative Large 3D Reconstruction Model", "abstract": "Feed-forward 3D modeling has emerged as a promising approach for rapid and high-quality 3D reconstruction. In particular, directly generating explicit 3D representations, such as 3D Gaussian splatting, has attracted significant attention due to its fast and high-quality rendering. However, many state-of-the-art methods, primarily based on transformer architectures, suffer from severe scalability issues because they rely on full attention across image tokens from multiple input views, resulting in prohibitive computational costs as the number of views or image resolution increases. Toward a scalable and efficient feed-forward 3D reconstruction, we introduce an iterative Large 3D Reconstruction Model (iLRM) that generates 3D Gaussian representations through an iterative refinement mechanism, guided by three core principles: (1) decoupling the scene representation from input images to enable compact 3D representations; (2) decomposing global multi-view interactions into a two-stage attention scheme to reduce computational costs; and (3) injecting high-resolution information at every layer to achieve high-fidelity reconstruction. Experimental results on widely used datasets, such as RE10K and DL3DV, demonstrate that iLRM outperforms existing methods in both reconstruction quality and speed.", "date": "2025-07-31", "categories": []}
{"id": "2512.07806", "title": "Multi-view Pyramid Transformer: Look Coarser to See Broader", "abstract": "We propose Multi-view Pyramid Transformer (MVP), a scalable multi-view transformer architecture that directly reconstructs large 3D scenes from tens to hundreds of images in a single forward pass. Drawing on the idea of ``looking broader to see the whole, looking finer to see the details,\" MVP is built on two core design principles: 1) a local-to-global inter-view hierarchy that gradually broadens the model's perspective from local views to groups and ultimately the full scene, and 2) a fine-to-coarse intra-view hierarchy that starts from detailed spatial representations and progressively aggregates them into compact, information-dense tokens. This dual hierarchy achieves both computational efficiency and representational richness, enabling fast reconstruction of large and complex scenes. We validate MVP on diverse datasets and show that, when coupled with 3D Gaussian Splatting as the underlying 3D representation, it achieves state-of-the-art generalizable reconstruction quality while maintaining high efficiency and scalability across a wide range of view configurations.", "date": "2025-12-08", "categories": []}
{"id": "2602.05293", "title": "Fast-SAM3D: 3Dfy Anything in Images but Faster", "abstract": "SAM3D enables scalable, open-world 3D reconstruction from complex scenes, yet its deployment is hindered by prohibitive inference latency. In this work, we conduct the \\textbf{first systematic investigation} into its inference dynamics, revealing that generic acceleration strategies are brittle in this context. We demonstrate that these failures stem from neglecting the pipeline's inherent multi-level \\textbf{heterogeneity}: the kinematic distinctiveness between shape and layout, the intrinsic sparsity of texture refinement, and the spectral variance across geometries. To address this, we present \\textbf{Fast-SAM3D}, a training-free framework that dynamically aligns computation with instantaneous generation complexity. Our approach integrates three heterogeneity-aware mechanisms: (1) \\textit{Modality-Aware Step Caching} to decouple structural evolution from sensitive layout updates; (2) \\textit{Joint Spatiotemporal Token Carving} to concentrate refinement on high-entropy regions; and (3) \\textit{Spectral-Aware Token Aggregation} to adapt decoding resolution. Extensive experiments demonstrate that Fast-SAM3D delivers up to \\textbf{2.67$\\times$} end-to-end speedup with negligible fidelity loss, establishing a new Pareto frontier for efficient single-view 3D generation. Our code is released in https://github.com/wlfeng0509/Fast-SAM3D.", "date": "2026-02-05", "categories": []}
{"id": "2602.08058", "title": "Picasso: Holistic Scene Reconstruction with Physics-Constrained Sampling", "abstract": "In the presence of occlusions and measurement noise, geometrically accurate scene reconstructions -- which fit the sensor data -- can still be physically incorrect. For instance, when estimating the poses and shapes of objects in the scene and importing the resulting estimates into a simulator, small errors might translate to implausible configurations including object interpenetration or unstable equilibrium. This makes it difficult to predict the dynamic behavior of the scene using a digital twin, an important step in simulation-based planning and control of contact-rich behaviors. In this paper, we posit that object pose and shape estimation requires reasoning holistically over the scene (instead of reasoning about each object in isolation), accounting for object interactions and physical plausibility. Towards this goal, our first contribution is Picasso, a physics-constrained reconstruction pipeline that builds multi-object scene reconstructions by considering geometry, non-penetration, and physics. Picasso relies on a fast rejection sampling method that reasons over multi-object interactions, leveraging an inferred object contact graph to guide samples. Second, we propose the Picasso dataset, a collection of 10 contact-rich real-world scenes with ground truth annotations, as well as a metric to quantify physical plausibility, which we open-source as part of our benchmark. Finally, we provide an extensive evaluation of Picasso on our newly introduced dataset and on the YCB-V dataset, and show it largely outperforms the state of the art while providing reconstructions that are both physically plausible and more aligned with human intuition.", "date": "2026-02-08", "categories": []}
{"id": "2604.09877", "title": "Genie 4D: Semantic-Prior-Guided 4D Dynamic Scene Reconstruction", "abstract": "At the intersection of computer vision and robotic perception, 4D reconstruction of dynamic scenes connects low-level geometric sensing with high-level semantic understanding. We present Genie 4D, a framework that turns hand-held phone capture into a semantically grounded, action-controllable 4D world model. Genie 4D couples a real-time visual-inertial Gaussian splatting front-end for metric geometry with a feed-forward 4D backbone regularized by frozen DINOv3 features acting as structural priors. The semantic priors suppress identity drift during dynamic tracking, while a short conditional diffusion refiner recovers high-frequency surface detail that regression backbones smooth away. Finally, a lightweight latent-action head exposes the reconstructed 4D state to a Genie-style world model trained with a JEPA-style next-embedding objective, so that the scene can be rolled forward under user actions. On the Point Odyssey and TUM-Dynamics benchmarks, Genie 4D retains the linear time complexity O(T) of feed-forward baselines while improving 3D tracking accuracy (APD) and reconstruction completeness, and it runs interactively on a single consumer GPU (RTX 5090) from iPhone, Mac, Windows, and Linux capture clients. Genie 4D offers a practical, semantic-prior-guided path toward physically grounded world models.", "date": "2026-04-10", "categories": []}
{"id": "2606.00444", "title": "Real-Time Physics Simulation with Dynamic Mesh-Gaussian Reconstructions", "abstract": "Integrating dynamic 3D reconstructions into physics simulation requires fixed mesh topology for efficient collision detection, but state-of-the-art methods like DG-Mesh produce varying topology optimized for geometric quality. We investigate whether topology conversion can enable physics integration while preserving reconstruction fidelity. We propose a dual-representation framework combining fixed-topology meshes for physics with Gaussian splatting for rendering, achieving 4.65$\\times$ speedup over varying-topology baselines through runtime vertex buffer updates. We evaluate two conversion strategies, temporal correspondence tracking and template-based projection, against native fixed-topology methods (MaGS) on the DG-Mesh dataset. Our evaluation reveals that both conversion approaches incur 65-80% geometric degradation, producing results inferior to MaGS despite DG-Mesh's superior initial quality. This demonstrates that high-quality reconstruction and physics-compatible topology represent fundamentally distinct objectives that cannot be reconciled through post-processing. Our findings inform future development of physics-aware reconstruction methods and our framework enables real-time simulation with any fixed-topology approach.", "date": "2026-05-30", "categories": []}
{"id": "2606.00450", "title": "Optimizing 3D Gaussian Splatting via Point Cloud Upsampling", "abstract": "3D Gaussian Splatting (3DGS) is a technique for creating and rendering 3D scenes, however its performance depends heavily on the quality of initial seed points. To improve 3DGS initialization, this study presents and evaluates several point cloud upsampling approaches: linear interpolation, triangular interpolation, spline-based surface reconstruction, moving least squares surface fitting, and Voronoi-based point generation. Additionally, this research introduces a depth-guided point lifting method that leverages depth maps to maintain geometric consistency with Structure-from-Motion (SfM) reconstructions. Through extensive experiments on the Mip-NeRF360 and Replica datasets, the proposed methods demonstrate improvements in reconstruction quality across diverse scene types. Results indicate that different upsampling strategies excel in different scenarios: surface reconstruction methods perform better with organic, detailed scenes, while simpler interpolation approaches are more suited for scenes dominated by piecewise-smooth geometries. In comparison, the depth-guided approach shows promise for adding geometry-aware points across the entire scene, importantly in texture-less regions. These findings, which provide preliminary practical guidelines for selecting appropriate upsampling methods based on scene characteristics and computational constraints, advances the understanding of how point cloud initialization affects 3DGS quality.", "date": "2026-05-30", "categories": []}
{"id": "2606.00452", "title": "Beyond Static Gaussians: An Empirical Investigation of Architectural Paradigms for Dynamic 3D Scene Reconstruction", "abstract": "Dynamic scene reconstruction via 3D Gaussian Splatting (3DGS) has emerged as a compelling approach for representing evolving environments, yet understanding trade-offs between methodologies remains crucial. This paper presents a comprehensive analysis of dynamic 3DGS methods, categorizing them into two paradigms: structure-guided methods employing auxiliary representations (deformation fields, canonical spaces, grids) to model temporal changes, and gaussian-centric methods encoding dynamics directly into primitives via continuous functions or 4D representations. We evaluate representative methods from both paradigms on the D-NeRF benchmark. Our findings reveal that structure-guided methods achieve superior reconstruction fidelity and compact model sizes, while gaussian-centric approaches demonstrate significantly higher rendering speeds enabling real-time performance, though with greater quality variability and potentially substantial storage overhead. This analysis highlights a fundamental trade-off between reconstruction quality/compactness versus rendering speed, providing insights to guide future research and application development in dynamic scene reconstruction.", "date": "2026-05-30", "categories": []}
{"id": "2606.01367", "title": "ActMVS: Active Scene Reconstruction with Monocular Multi-View Stereo", "abstract": "Active scene reconstruction enables robots/UAVs to autonomously plan trajectories and reconstruct environments without costly manual data acquisition. Unlike passive methods, active reconstruction requires real-time construction of high-confidence occupancy maps for collision-free navigation. Existing approaches rely on depth sensors for occupancy map updates, increasing platform cost and weight. To advance spatial intelligence, we aim for a vision-only monocular solution. However, current monocular scene reconstruction methods operate offline and fail to deliver globally consistent dense depth at the frame rates required for robots/UAVs navigation. To bridge this gap, we introduce ActMVS, the first framework for monocular active reconstruction. Our framework integrates a view factor graph construction for informed Multi-View Stereo depth prediction, along with a global depth optimization, to enable the online generation of high-quality, globally consistent dense depth maps. This enables monocular robots/UAVs to maintain reliable occupancy maps for safe trajectory planning during reconstruction. Experiments on Replica datasets demonstrate performance competitive with RGB-D methods. Our code and data are available at https://github.com/TrickyGo/ActMVS.", "date": "2026-05-31", "categories": []}
{"id": "2606.01939", "title": "SAVMap: Structure-Aided Visual Mapping of Large-Scale 2.5D Manhattan Wireframes from Panoramic Video", "abstract": "Precise 3D representations of industrial environments enable tasks such as robot localization and digital twin generation. We propose SAVMap, a method for generating a semantic wireframe map of warehouse shelf and light structures using only a panoramic video camera as the sensor input. Sequences of rectified images with shelf and ceiling-facing views are extracted from a panoramic video captured along the warehouse aisles. Using a semantic segmentation network front end, a set of sparse, semantic structure feature points (e.g., corners of shelf structures, centers of lights) are extracted from each image and tracked across the sequences. By accounting for real-world geometric relationships among the points such as Manhattan grids, a constrained structure-from-motion algorithm yields the 3D points that form a wireframe map. We demonstrate the scalability and accuracy of our proposal in a warehouse with 46 shelving rows, each with faces spanning 55\\,m by 7\\,m. From an hour of panoramic video content, we create wireframe maps for over 5000 shelf elements across the rows, achieving an aggregate mean absolute error of 4.8\\,cm with respect to ground-truth.", "date": "2026-06-01", "categories": []}
{"id": "2606.02096", "title": "WebSpline: Structure-Informed Splines for Real-Time 3D Gaussians from Monocular Videos", "abstract": "Dynamic scene reconstruction from monocular videos remains highly challenging, as existing methods often struggle to balance global structural coherence and local fine-grained details under limited multi-view cues. To address this challenge, we propose WebSpline, a novel dynamic 3D Gaussian framework that enables structurally coherent and high-fidelity reconstruction from monocular videos with fast rendering. The core of WebSpline is the Structure-Informed Spline (SIS) representation, which models each dynamic Gaussian trajectory using a learnable cubic Hermite spline whose motion is structurally organized with an auxiliary Structural Proxy Graph (SPG). The proposed framework is optimized in two stages: (i) in the first stage, the SPG is initialized from 2D point tracks and refined with temporal rigidity regularization to establish structural coherence for moving objects across the sequence; and (ii) in the second stage, the SIS representation is initialized from the refined SPG and optimized under both spatial and structural neighborhood constraints. At inference, Gaussian motion is obtained solely by evaluating the learned SIS, enabling fast rendering. Extensive experiments on the challenging monocular dynamic scene benchmarks, iPhone and NVIDIA, demonstrate that our WebSpline achieves state-of-the-art rendering quality while rendering over 10 times faster than WorldTree, the second-best method on the iPhone dataset.", "date": "2026-06-01", "categories": []}
{"id": "2606.02350", "title": "TROPHIES: Temporal Reconstruction of Places, Humans, and Cameras from Multi-view Videos", "abstract": "Reconstructing humans and their surrounding environments in a globally consistent 4D space is essential for comprehensive perception. However, prior works typically assume single-view inputs or decouple humans, scenes, and cameras, making them unable to recover coherent geometry, stable motion, and physically aligned trajectories. These limitations motivate us to introduce a new task: unified human-scene-camera reconstruction from multi-view videos, which aims to jointly estimate dynamic humans, static scenes, and camera poses in one global coordinate frame. We propose TROPHIES--Temporal Reconstruction of Places, Humans, and Cameras from Multi-view Videos-a unified framework tailored for this task. TROPHIES features a Human Branch that models humans through temporal and spatial reasoning, and a Scene Branch that reconstructs static geometry with human-aware attention. A global alignment and optimization module couples both branches by enforcing scale consistency, contact priors, and cross-view temporal coherence. Experiments on EgoHuman and EgoExo4D demonstrate that TROPHIES achieves globally aligned, physically plausible 4D reconstructions and consistently outperforms existing paradigms in both global fidelity and human-scene consistency.", "date": "2026-06-01", "categories": []}
{"id": "2606.02406", "title": "Edge Prediction for Roof Wireframe Reconstruction with Transformers", "abstract": "This paper presents a competitive solution to the S23DR Challenge 2026, which aims to reconstruct 3D house roof wireframe models from sparse SfM point clouds and ground-level semantic segmentations and depth maps. Our proposed method utilizes an end-to-end Transformer encoder-decoder architecture inspired by DETR. To effectively process the geometric and semantic data, the sparse SfM point cloud input is dynamically subsampled based on semantic priority and augmented with Gestalt and ADE20k class features. To further increase segmentation context, we fuse the point features with additional Gestalt feature encodings which are obtained by projecting the points into latent feature maps produced by a frozen autoencoder. Learned query embeddings are then decoded directly into 3D wireframe edges via cross-attention mechanisms. Evaluated on the \"HoHo 22k\" dataset, our approach significantly outperforms both handcrafted and learned baselines, achieving a Hybrid Structure Score (HSS) of 0.6476 and securing the second-highest position on the challenge's private leaderboard.", "date": "2026-06-01", "categories": []}
{"id": "2509.18898", "title": "DeblurSplat: SfM-free 3D Gaussian Splatting with Event Camera for Robust Deblurring", "abstract": "In this paper, we propose the first Structure-from-Motion (SfM)-free deblurring 3D Gaussian Splatting method via event camera, dubbed DeblurSplat. We address the motion-deblurring problem in two ways. First, we leverage the pretrained capability of the dense stereo module (DUSt3R) to directly obtain accurate initial point clouds from blurred images. Without calculating camera poses as an intermediate result, we avoid the cumulative errors transfer from inaccurate camera poses to the initial point clouds' positions. Second, we introduce the event stream into the deblur pipeline for its high sensitivity to dynamic change. By decoding the latent sharp images from the event stream and blurred images, we can provide a fine-grained supervision signal for scene reconstruction optimization. Extensive experiments across a range of scenes demonstrate that DeblurSplat not only excels in generating high-fidelity novel views but also achieves significant rendering efficiency compared to the SOTAs in deblur 3D-GS.", "date": "2025-09-23", "categories": []}
{"id": "2605.03337", "title": "FreeTimeGS++: Secrets of Dynamic Gaussian Splatting and Their Principles", "abstract": "The recent surge in 4D Gaussian Splatting (4DGS) has achieved impressive dynamic scene reconstruction. While these methods demonstrate remarkable performance, the specific drivers behind such gains remain less explored, making a systematic understanding of the underlying principles challenging. In this paper, we perform a comprehensive analysis of these hidden factors to provide a clearer perspective on the 4DGS framework. We first establish a controlled baseline, FreeTimeGS_ours, by formalizing and reproducing the heuristics of the state-of-the-art FreeTimeGS. Using this framework, we dissect 4DGS along its fundamental axes and uncover key secrets, including the emergent temporal partitioning driven by Gaussian durations and the discrepancy between photometric fidelity and spatiotemporal consistency. Based on these insights, we propose FreeTimeGS++, a principled method that employs gated marginalization and neural velocity fields to achieve superior stability and robust dynamic representations. Our approach yields reproducible results with reduced run-to-run variance. We will release our implementation to provide a reliable foundation for future 4DGS research.", "date": "2026-05-05", "categories": []}
{"id": "2605.26519", "title": "$R^3$: 3D Reconstruction via Relative Regression", "abstract": "Recent feed-forward geometry foundation models have demonstrated impressive generalization by recovering depth and poses in a single forward pass. However, these models are typically constrained by a global coordinate frame assumption. This dependency becomes a significant bottleneck for long-context and streaming reconstruction, as it forces the network to maintain an arbitrary temporal origin and handle translation magnitudes that grow unbounded over time. Our solution, which we call $R^3$, employs relative regression. We employ a lightweight MLP to predict confidence-weighted relative constraints. These confidences serve as a unified anchor: weighting losses during training and guiding pose aggregation during inference. $R^3$ supports both full-context offline reconstruction and causal, bounded-memory streaming. Our evaluation in both offline and streaming settings validates the effectiveness of our relative mechanism. Project page: https://kevinxu02.github.io/r3-site", "date": "2026-05-26", "categories": []}
{"id": "2605.30215", "title": "D\\'ej\\`a View: Looping Transformers for Multi-View 3D Reconstruction", "abstract": "Recent feed-forward 3D reconstruction transformers have scaled to over a billion parameters, following the broader trend of increasing model capacity in computer vision. Yet emerging evidence suggests that contiguous transformer layers often behave like repeated applications of similar operations, and multi-view reconstruction transformers refine their predictions progressively across decoder depth. We posit that model depth partially buys iteration, paid for inefficiently in unique parameters, and instead make that iteration explicit in architecture. Our model, D\\'ej\\`aView, applies a single looped transformer block recurrently to per-view features for K refinement steps. Trained once, it exposes K as an inference-time compute knob, matching or outperforming substantially larger feed-forward baselines across five reconstruction benchmarks spanning indoor, outdoor, object-centric, and driving scenes, while using a fraction of their parameters and comparable or lower compute. Importantly, the same looped block formulation outperforms an otherwise identical variant with independent per-step parameters under matched training data and compute, suggesting that explicit iteration is not merely a compute-efficient substitute for capacity but a stronger inductive bias for multi-view 3D reconstruction.", "date": "2026-05-28", "categories": []}
{"id": "2605.30863", "title": "DSD-GS: Dynamic-Static Decomposition of Gaussian Splatting for Efficient and High-Fidelity Dynamic Scene Reconstruction", "abstract": "Dynamic scene reconstruction and novel view synthesis are fundamental to next-generation visual intelligence applications such as virtual reality, robotics, and digital twins. However, high-fidelity reconstruction of complex, time-varying scenes from arbitrary viewpoints remains a significant challenge. Existing dynamic 3DGS methods suffer from computational inefficiency, since they model all Gaussians as dynamic components. While recent decomposition-based approaches address this issue, they still struggle with degraded reconstruction quality and prolonged training time. To mitigate these limitations, we propose a novel dynamic reconstruction framework built upon an efficient static-dynamic decomposition strategy using a Feed-Forward Gaussian Splatting encoder and an optical flow model. By eliminating redundant computations on static regions, our method achieves state-of-the-art performance, outperforming existing baselines across rendering quality, training and rendering speed, and storage efficiency. Notably, on the Neural 3D dataset, our framework requires only 10 minutes for training and achieves a rendering speed of over 700 FPS on a single NVIDIA RTX 5090 GPU at resolution of 1352x1014. Furthermore, our decomposition strategy eliminates the need for COLMAP preprocessing and enables deterministic initialization, thereby enhancing both efficiency and reproducibility.", "date": "2026-05-29", "categories": []}
{"id": "2605.31466", "title": "VolFill: Single-View Amodal 3D Scene Reconstruction with Volumetric Flow Matching", "abstract": "Reconstructing the complete geometry of a scene from a single RGB image remains challenging - especially when inferring hidden structures where visual evidence is incomplete. We introduce VolFill, a generative framework that predicts the 3D structure of the complete scene rather than relying on traditional pixel-aligned regression. Our method utilizes a hybrid 3D VAE to compress sparse truncated unsigned distance function grids into a compact latent space, paired with a latent Diffusion Transformer that denoises this representation to recover the complete scene. We condition the generation on geometry foundation models, leveraging rich spatial priors for robust reasoning. Unlike existing methods limited by per-ray constraints or unstructured point-cloud queries, VolFill provides a structured representation that supports direct surface extraction and occupancy queries at scale. Extensive experiments on the SCRREAM and NRGB-D datasets demonstrate that our approach significantly outperforms current baselines, providing a robust foundation for holistic spatial understanding.", "date": "2026-05-29", "categories": []}
{"id": "2605.31534", "title": "Feature-Optimized Vision for Adaptive 3D Scene Reconstruction", "abstract": "Three-dimensional scene reconstruction depends on local image evidence that is both visually discriminative and geometrically useful. Fixed feature thresholds and uniform feature budgets are easy to deploy, but they can waste computation on repeated texture, low-parallax regions, or unstable points. This paper proposes an adaptive feature-optimized vision front end for 3D reconstruction. The method scores candidate features by texture, repeatability, distinctiveness, expected triangulation angle, and spatial coverage, then allocates a per-view feature budget to maximize useful tracks under a fixed reconstruction pipeline. A small synthetic multi-view prototype evaluates four selection policies across corridor, facade, object-table, and cluttered scenes. Compared with random, texture-only, and uniform-grid baselines, the adaptive policy obtains the best quality-aware completeness and the lowest aggregate reconstruction RMSE while preserving broad image coverage. The result is not a replacement for modern learned matching or neural reconstruction systems; it is a modular front-end policy that can make classical and learned 3D pipelines more deliberate about which visual evidence they spend compute on.", "date": "2026-05-29", "categories": []}
{"id": "2605.31577", "title": "SurGe: Improved Surface Geometry in Point Maps", "abstract": "Recent feedforward 3D reconstruction methods predict point maps and estimate global 3D geometry remarkably well. However, their predictions still exhibit inaccurate local surface geometry, which is clearly visible qualitatively but only weakly reflected in common metrics. To make these errors more explicit in evaluation, we introduce a point map normal metric that evaluates the local surface orientation induced by neighboring 3D predictions. To reduce these errors, we propose two complementary components: a point gradient matching loss that supervises depth-normalized 3D finite differences, and a Neighborhood Attention Decoder (NAD) that progressively upsamples features and uses Neighborhood Attention for local feature mixing. Across eight zero-shot monocular geometry benchmarks, our model, SurGe, achieves the best average rank for global point map AbsRel and consistently improves local point map and point map normal evaluations.", "date": "2026-05-29", "categories": []}
{"id": "2605.31595", "title": "Learning Global Motion with Compact Gaussians for Feed-Forward 4D Reconstruction", "abstract": "Dynamic scene reconstruction from monocular video remains a fundamental challenge in computer vision. Existing feed-forward methods predict 3D Gaussians pixel-wise for each frame, suffering from duplicated Gaussians and view-dependent biases that hinder effective learning of scene motion. We present C4G, a feed-forward 4D reconstruction framework built upon a compact set of timestamp-conditioned learnable Gaussian query tokens. Each token aggregates corresponding features across the full temporal context and decodes a 3D Gaussian whose position is modulated by the target timestamp, enabling globally coherent motion modeling without per-scene optimization. To capture fine-grained details, we further introduce a video diffusion model-based rendering enhancement module. Since our framework effectively aggregates features into Gaussians, we extend this capability to feature lifting, producing a 4D feature field that supports point tracking and dynamic scene understanding. C4G achieves strong novel-view synthesis performance using significantly fewer Gaussians and without requiring camera poses, while exhibiting stronger motion modeling and robustness to large temporal gaps.", "date": "2026-05-29", "categories": []}
{"id": "2605.15852", "title": "GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction", "abstract": "Streaming 3D reconstruction from long monocular video sequences requires maintaining a key-value (KV) cache that grows linearly with sequence length, creating a severe memory bottleneck. Existing approaches either truncate the cache to a fixed set of anchor frames, leading to reconstruction quality degradation, or rely on attention-score heuristics that are agnostic to 3D scene structure, failing to preserve geometrically valuable tokens. To address these problems, we present GHOST (Geometry-Hierarchical Online Streaming Token Eviction), a training-free KV cache management framework that exploits the model's own 3D geometry outputs to evict redundant tokens online. GHOST introduces three mutually reinforcing innovations: a hierarchical dual-level importance scoring scheme, a privilege mechanism that protects special tokens from eviction, and a cosine-similarity-guided layer-wise budget allocation. Experiments on various benchmarks show that GHOST preserves excellent reconstruction quality while cutting the KV cache by nearly half and delivering 1.75x faster inference compared to state-of-the-art methods. Our code is available at https://github.com/lokiniuniu/GHOST.", "date": "2026-05-15", "categories": []}
{"id": "2605.22069", "title": "TWINGS: Thin Plate Splines Warp-aligned Initialization for Sparse-View Gaussian Splatting", "abstract": "Novel view synthesis from sparse-view inputs poses a significant challenge in 3D computer vision, particularly for achieving high-quality scene reconstructions with limited viewpoints. We introduce TWINGS, a framework that enhances 3D Gaussian Splatting (3DGS) by directly addressing point sparsity. We employ Thin Plate Splines (TPS), a smooth non-rigid deformation model that minimizes bending energy to estimate a globally coherent warp from control-point correspondences, to align backprojected points from estimated depth with triangulated 3D control points, yielding calibrated backprojected points. By sampling these calibrated points near the control points, TWINGS provides a fast and geometrically accurate initialization for 3DGS, ultimately improving structural detail preservation and color fidelity in reconstructed scenes. Extensive experiments on DTU, LLFF, and Mip-NeRF360 demonstrate that TWINGS consistently outperforms existing methods, delivering detailed and accurate reconstructions under sparse-view scenarios.", "date": "2026-05-21", "categories": []}
{"id": "2605.25975", "title": "F-RNG: Feed-Forward Relightable Neural Gaussians", "abstract": "Capturing relightable 3D assets from real-world objects is a widely researched problem. Several per-scene optimization-based methods, based on 3D Gaussian splatting (3DGS), support relighting; however, they usually require dense input views, and their overfitting nature makes it difficult to generalize across scenes. Unlike per-scene optimization methods, generalized feed-forward models can directly reconstruct Gaussians from sparse input views. However, the resulting assets have baked-in illumination and cannot be easily used for relighting. In this paper, we present F-RNG, a feed-forward framework that directly generates relightable 3DGS assets from sparse-view inputs. Training such a model from scratch can require massive data and computing resources, and it is especially challenging to generate relightable assets in a feed-forward manner with acceptable cost. We develop F-RNG upon an existing large reconstruction model (LRM) to extract relightable representations, while also utilizing priors from an intrinsic decomposition model (IDM). Specifically, we first introduce a latent-interpolated fine-grained geometry synthesis to enhance the LRM's geometry representation. Second, we propose a prior-guided relightable appearance distillation to extract relightable neural representations by incorporating IDM priors. Finally, a universal neural renderer enables flexible and high-fidelity relighting. F-RNG requires neither re-training nor fine-tuning of the underlying LRMs, thus can automatically benefit from better LRMs and IDMs in the future. With only small networks that can be trained with affordable data and computational resources, F-RNG avoids the repetitive inference of large models under different light conditions. By comparison to the state-of-the-art LRM-based relighting method, F-RNG achieves ~25x faster relighting, as well as superior quality (~+2.0 dB).", "date": "2026-05-25", "categories": []}
{"id": "2605.29098", "title": "Seeing through boxes: Non-Line-of-Sight 3D Reconstruction from Radar Signals", "abstract": "Reconstructing object geometry from radio frequency (RF) signals is fundamentally challenging due to the lensless imaging nature of RF sensing, which leads to low spatial resolution and high noise. Unlike light signals, RF signals can penetrate occlusions and thus capture information about hidden scenes. Existing Non-Line-of-Sight (NLoS) 3D neural reconstruction methods can recover coarse surfaces inside enclosed environments but often suffer from unstable optimization, noisy surface geometry, and surface ambiguity, failing to produce accurate zero-level sets from the signed distance field (SDF). These limitations largely stem from neglecting the role of Line-of-Sight (LoS) geometry outside the enclosed region, which provides valuable physical constraints for modeling signal propagation. In this paper, we introduce a Unified LoS and NLoS neural geometry reconstruction framework GeRaF 2.0 that leverages the outside LoS geometry to model and guide RF propagation from the LoS region into the NLoS region. By integrating visual LoS priors into the neural field formulation, GeRaF 2.0 achieves stable training and physically consistent reconstruction of both visible and hidden geometry, setting a new state-of-the-art in RF-based geometry reconstruction.", "date": "2026-05-27", "categories": []}
{"id": "2605.29136", "title": "Eulerian Gaussian Splatting using Hashed Probability Pyramids", "abstract": "We introduce a probabilistic splat-based radiance field framework that retains the fast rasterization and test-time efficiency of 3D Gaussian Splatting (3DGS) while replacing heuristic primitive manipulation with gradient-based optimization of a volumetric probability density. Rather than relocating, splitting, or culling Gaussians via hand-tuned densification (e.g., ADC), we treat primitive locations as samples drawn from a persistent, learnable density. We instantiate this density using a novel, memory-efficient multi-scale hierarchical grid that enables end-to-end gradient-based optimization. To stabilize the optimization, we derive an unbiased gradient estimator with control variates that markedly reduces variance. By allowing probability mass to flow to where the loss demands, our framework eliminates brittle priors and naturally explores the volume, achieving state-of-the-art reconstruction quality on mip-NeRF 360 while preserving 3DGS-level rendering speed.", "date": "2026-05-27", "categories": []}
{"id": "2605.29761", "title": "S2MDF: A Plug-And-Play Layer for Intersection-Free Multi-Object Signed Distance Fields", "abstract": "Compositional implicit surface representations model scenes as collections of objects, each encoded by a Signed Distance Field (SDF). A fundamental limitation of this approach is that multiple SDFs can produce geometries that interpenetrate, violating physical plausibility. Existing mitigation strategies rely on soft penalty terms that reduce but do not eliminate intersections, and require careful loss weighting. To truly prevent interpenetration, we propose a hard constraint on vector-valued SDFs and introduce S2MDF, a lightweight plug-and-play module that enforces the constraint on any object-compositional SDF representation without architectural modifications. It introduces negligible computational overhead and is compatible with linearly-interpolated standard meshing algorithms such as Marching Cubes. It can be applied during training or as a post-processing step. Experiments on multiple state-of-the-art compositional methods show that S2MDF reduces intersections to numerical precision while preserving reconstruction quality, outperforming existing mitigation strategies.", "date": "2026-05-28", "categories": []}
{"id": "2605.29997", "title": "FRUC: Feedforward Dynamic Scene Reconstruction from Uncalibrated Collaborative Driving Views", "abstract": "We present FRUC, a feed-forward 3D Gaussian splatting framework for dynamic scene reconstruction from uncalibrated collaborative driving views. Existing multi-agent reconstruction frameworks are often hindered by rigid prerequisites, demanding precise spatial calibration and slow per-scene optimization. In this paper, we rethink this task by conceptualizing a distributed multi-vehicle network as a spatio-temporally unstructured ego-centric multi-camera system, where the core challenge lies in enhancing ego-centric occluded geometry through collaboration without degrading the ego's accurately observed visible geometry, while preserving reconstruction efficiency. For efficient reconstruction, FRUC is built upon a visual grounded geometric Transformer backbone to enable one-shot, calibration-free inference from a flexible number of multi-vehicle views. To achieve non-destructive geometric supplementation under uncalibrated cross-agent misalignment, FRUC first introduces an ego-centric causal occlusion field that explicitly derives occlusion evolution as latent priors by modeling agent-wise spatio-temporal correlations. Guided by these occlusion priors, it further formulates cross-agent integration as a deterministic residual denoising process via zero-initialized injection, turning challenging cross-agent fusion into bounded residual learning for robust collaborative blind-spot completion. Through extensive evaluations on the real-world V2XReal and UrbanIng-V2X datasets, FRUC is shown to be a new state-of-the-art for the scene reconstruction of dynamic collaborative driving environments, significantly outperforming existing methods in both rendering quality and efficiency.", "date": "2026-05-28", "categories": []}
{"id": "2605.30239", "title": "SAM3D-Phys: Towards Multi-Object Interactive Simulation in Real World", "abstract": "This work addresses the problem of recovering complete, simulatable object geometry from reconstructed real-world scenes, enabling physics-based interaction with objects embedded in the scene. While modern multi-view reconstruction methods can produce visually accurate environments, objects are often incomplete due to occlusions and limited observations, making them unsuitable for physics simulation. To address this limitation, we propose SAM3D-Phys, a framework that integrates scene reconstruction with generative 3D priors of SAM3D to recover physically simulatable objects. Our approach first reconstructs the scene from multi-view images to obtain scene geometry and partial observations of objects. We then leverage SAM3D to infer complete object geometry from these partial observations. To ensure that the recovered objects remain consistent with the reconstructed scene, we restore scene-consistent object states through two complementary strategies: a physics-constrained spatial optimization algorithm that iteratively aligns the recovered object to its original location, and a mask-guided appearance distillation module that refines texture fidelity based on the observed images. By recovering complete object geometry and restoring its pose and appearance within the scene, SAM3D-Phys produces clean object representations suitable for physics-based simulation, enabling simultaneous and physically consistent interactive simulation of multiple objects within a reconstructed scene. Project page: https://chnxindong.github.io/sam3d-phys/", "date": "2026-05-28", "categories": []}
{"id": "2605.30310", "title": "City-Mesh3R: Simulation-Ready City-Scale 3D Mesh Reconstruction from Multi-View Images", "abstract": "City-scale 3D surface reconstruction from multiview images for downstream 3D simulation, poses highly challenging problems due to the scale and complexity of urban scenes. Existing city-scale 3D reconstruction methods based on NeRF, Gaussian Splatting etc. often fail to recover 3D meshes ready for simulation due to incomplete/missing geometry and irregular, noisy surfaces. Scaling existing small-scale 3D reconstruction methods to arbitrarily large urban scenes is highly infeasible due to their computational complexity. We present City-Mesh3R, a scalable framework for reconstructing watertight surface meshes directly from large unordered image collections. Unlike recent methods which use global sparse SfM point-cloud initialization followed by a distributed 3D dense reconstruction of large-scale scenes, our method follows an end-to-end images-to-mesh 3D reconstruction approach using a divide-and-conquer strategy. The sparse city map is reconstructed via topological image clustering, cluster-wise independent sparse SfM and map merging, without need for exhaustive image feature matching. Then this map is partitioned spatially to perform geometry-aware camera selection, followed by dense surface reconstruction and surface refinement using curvature-aware adaptive vertex density remeshing. These partition meshes are then stitched together to produce the global mesh of the city. The proposed end-to-end framework is evaluated on city-scale reconstruction datasets. As demonstrated by our qualitative and quantitative results, our proposed method yields high-fidelity watertight 3D meshes with regular geometry, capturing fine surface details, and is suitable for scaling to arbitrarily large scenes owing to the end-to-end processing in a distributed setting.", "date": "2026-05-28", "categories": []}
{"id": "2605.30328", "title": "Supercharging Thermal Gaussian Splatting with Depth Estimation", "abstract": "Efficient and robust 3D scene representation is crucial in autonomous driving, robotics, and related fields. While RGB images provide valuable content for 3D reconstruction, other modalities like thermal or depth can enable additional information on the environment. Lately, novel view synthesis methods like 3D Gaussian Splatting have started using multiple modalities to further boost their performance. But fusing or combining multimodal data can make the process slower and can bring in additional challenges. Therefore, our project aims to use single modality based on thermal infrared domain, by removing the reliance on visible light as much as possible. This single modality can be expected to be faster as it does not rely on multimodal data. We propose a method, Thermal-to-Depth Gaussian Splatting (TDg), that uses only thermal images and depth estimation in its architecture to derive the radiance fields. Our TDg method outperforms the MSMG (Multiple Single-Modal Gaussians) baseline in most cases on our test datasets, RGBT-Scenes and ThermalMix. On average, the rendering quality metrics such as learned perceptual image patch similarity (LPIPS), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) of TDg are 1.12%, 0.034%, and 0.01% better than the baseline MSMG values. It also reduces the training time significantly, by 12 mins 47 secs (55% improvement). Overall, our method is successful in deriving these thermal radiance fields, which can ultimately have several applications, such as identifying heat sources critical in surveillance, search or rescue operations, and industrial inspections where temperature is widely used to monitor machines.", "date": "2026-05-28", "categories": []}
{"id": "2605.30338", "title": "REST3D: Reconstructing Physically Stable 3D Scenes from a Single Image", "abstract": "Reconstructing physically stable 3D scenes from a single RGB image enables casual images to be converted into simulation-ready digital assets for applications such as immersive interaction and content creation. However, existing single-image reconstruction methods fall short in capturing the physical structure of a scene. As a result, they often produce geometrically plausible but physically inconsistent results, including object floating and penetration, which lead to unstable behavior in physics simulations. Image-conditioned scene generation methods improve physical plausibility but often rely on strong scene priors, yielding plausible yet inaccurate object arrangements that fail to match the input image. We propose REST3D, a single-image reconstruction framework that can reconstruct physically stable 3D scenes by integrating physical scene understanding with physics-constrained refinement. We first introduce an agentic physical scene understanding technique that constructs a scene-tree representation capturing object physical states and inter-object relationships from a gravity-support perspective, providing a structural prior for reconstruction. Leveraging this structure, we initialize the scene using image-to-3D models, followed by scene-tree-guided alignment and physics-constrained optimization to resolve physical violations while preserving visual consistency with the input image. Experiments show that our method significantly reduces physical errors and improves simulation stability on both synthetic and real-world datasets while maintaining strong reconstruction quality. We further demonstrate the reconstructed scenes in VR-based human-object interaction, showing their potential for immersive applications.", "date": "2026-05-28", "categories": []}
{"id": "2605.30342", "title": "Uncertainty-driven 3D Gaussian Splatting Active Mapping via Anisotropic Visibility Field", "abstract": "We present Gaussian Splatting Anisotropic Visibility Field (GAVIS), a novel framework for uncertainty quantification and active mapping in 3DGS. Our key insight is that regions unseen from the training views yield unreliable predictions from the 3DGS. To address this, we introduce a principled and efficient method for quantifying the visibility field in 3DGS, defined as the anisotropic visibility of each particle with respect to the training views, and represented using spherical harmonics. The resulting visibility field is integrated into a Bayesian Network-based uncertainty-aware 3DGS rasterizer, enabling real-time (200 FPS) uncertainty quantification for synthesized views. Active mapping is further performed within a maximum information gain framework building on this formulation. Extensive experiments across diverse environments demonstrate that GAVIS consistently and significantly outperforms prior approaches in both accuracy and efficiency. Moreover, beyond standalone use, our method can be applied post-hoc to improve the performance of existing approaches.", "date": "2026-05-28", "categories": []}
{"id": "2601.17354", "title": "PocketGS: On-Device Training of 3D Gaussian Splatting for High Perceptual Modeling", "abstract": "While 3D Gaussian Splatting (3DGS) enables real-time rendering, its training demands workstation-level compute and memory, making mobile deployment impractical under minute-scale time budgets and limited peak memory. We present PocketGS, a mobile scene modeling paradigm that enables on-device 3DGS training under these tightly coupled constraints while preserving high-fidelity reconstruction. PocketGS resolves the fundamental tension between training efficiency, memory compactness, and modeling quality through three co-designed operators: $\\mathcal{G}$ builds geometry-faithful point-cloud priors; $\\mathcal{I}$ injects local surface statistics to seed anisotropic Gaussians, thereby reducing early conditioning gaps; and $\\mathcal{T}$ unrolls alpha compositing with cached intermediates and index-mapped gradient scattering for stable mobile backpropagation. Extensive experiments demonstrate that PocketGS outperforms the powerful mainstream workstation 3DGS baseline under mobile budgets, delivering high-quality reconstructions and enabling a fully on-device, practical capture-to-rendering workflow.", "date": "2026-01-24", "categories": []}
{"id": "2602.22096", "title": "WeatherCity: Urban Scene Reconstruction with Controllable Multi-Weather Transformation", "abstract": "Editable high-fidelity 4D scenes are crucial for autonomous driving, as they can be applied to end-to-end training and closed-loop simulation. However, existing reconstruction methods are primarily limited to replicating observed scenes and lack the capability for diverse weather simulation. While image-level weather editing methods tend to introduce scene artifacts and offer poor controllability over the weather effects. To address these limitations, we propose \\textbf{WeatherCity}, a novel framework for 4D urban scene reconstruction and weather editing. Specifically, we leverage a text-guided image editing model to achieve flexible editing of image weather backgrounds. To tackle the challenge of multi-weather modeling, we introduce a novel weather Gaussian representation based on shared scene features and dedicated weather-specific decoders. This representation is further enhanced with a content consistency optimization, ensuring coherent modeling across different weather conditions. Additionally, we design a physics-driven model that simulates dynamic weather effects through particles and motion patterns. Extensive experiments on multiple datasets and various scenes demonstrate that WeatherCity achieves flexible controllability, high fidelity, and temporal consistency in 4D reconstruction and weather editing. Our framework not only enables fine-grained control over weather conditions (e.g., light rain and heavy snow) but also supports object-level manipulation within the scene. Codes are released at https://github.com/IRMVLab/WeatherCity.", "date": "2026-02-25", "categories": []}
{"id": "2605.20150", "title": "TideGS: Scalable Training of Over One Billion 3D Gaussian Splatting Primitives via Out-of-Core Optimization", "abstract": "Training 3D Gaussian Splatting (3DGS) at billion-primitive scale is fundamentally memory-bound: each Gaussian primitive carries a large attribute vector, and the aggregate parameter table quickly exceeds GPU capacity, limiting prior systems to tens of millions of Gaussians on commodity single-GPU hardware. We observe that 3DGS training is inherently sparse and trajectory-conditioned: each iteration activates only the Gaussians visible from the current camera batch, so GPU memory can serve as a working-set cache rather than a persistent parameter store. Building on this insight, we introduce TideGS, an out-of-core training framework that manages parameters across an SSD-CPU-GPU hierarchy via three synergistic techniques: block-virtualized geometry for SSD-aligned spatial locality, a hierarchical asynchronous pipeline to overlap I/O with computation, and trajectory-adaptive differential streaming that transfers only incremental working-set deltas between iterations. Experiments show that TideGS enables training with over one billion Gaussians on a single 24 GB GPU while achieving the best reconstruction quality among evaluated single-GPU baselines on large-scale scenes, scaling beyond prior out-of-core baselines (e.g., approximately 100M Gaussians) and standard in-memory training (e.g., approximately 11M Gaussians).", "date": "2026-05-19", "categories": []}
{"id": "2605.26368", "title": "Unified Panoramic Geometry Estimation via Multi-View Foundation Models", "abstract": "Geometry estimation from perspective images has greatly advanced, maturing to the point where off-the-shelf foundation models are able to reconstruct 3D scene structure not only from multi-view imagery, but even from a single view. A natural extension is 3D reconstruction from panoramas, with the exciting prospect of recovering a full 360-degree scene from a single panoramic image. In this work, we introduce PaGeR (Panoramic Geometry Reconstruction), a framework to lift powerful 3D foundation models designed for perspective imagery to the panorama domain. Our strategy is to start from a pre-trained transformer for 3D reconstruction and turn it into a unified high-performance model that predicts scale-invariant depth, metric depth, surface normals, and sky masks from both perspective and omnidirectional images, in a single forward pass. By keeping architectural changes to a minimum and mixing perspective and panoramic images during training, PaGeR retains the rich 3D prior of the underlying foundation model while learning to also estimate geometrically consistent 360-degree scenes from single panoramas. We extensively test our method in both indoor and outdoor environments and find that it delivers state-of-the-art performance and excellent zero-shot performance across a wide range of scenes. Code, data and models are available $\\href{https://github.com/prs-eth/PaGeR}{\\text{here}}$.", "date": "2026-05-25", "categories": []}
{"id": "2605.28125", "title": "CLEAR-NeRF: Collinearity and Local-region Enhanced Accurate 3D Reconstruction in Unbounded Scenes", "abstract": "Many real-world 3D reconstruction applications demand photorealism and metric accuracy across unbounded, complex scenes with challenging lighting and imperfect captures that current Neural Radiance Field (NeRF) pipelines only partly satisfy. This study adapts NeRF-based 3D reconstruction to multi-region of interest unbounded scenes to improve robustness to lighting and pose variation while enforcing metric accuracy suitable for digital-twin applications. Our approach introduces (i) automated local region localization/detection and reconstruction to seamlessly prioritize areas of interest without proliferating submodules, (ii) collinearity-enforcing ray sampling to learn smooth planar and curved surfaces, (iii) depth-localized neighborhood point extraction to suppress surface artifacts, and (iv) geometry-relevant color aggregation to mitigate lighting- and pose-caused variations. Results indicate superior performance of the proposed pipeline over the baseline NeRF models and established Structure from Motion (SfM) - Multi-View Stereo (MVS) solutions.", "date": "2026-05-27", "categories": []}
{"id": "2503.14359", "title": "ImViD: Immersive Volumetric Videos for Enhanced VR Engagement", "abstract": "User engagement is greatly enhanced by fully immersive multi-modal experiences that combine visual and auditory stimuli. Consequently, the next frontier in VR/AR technologies lies in immersive volumetric videos with complete scene capture, large 6-DoF interaction space, multi-modal feedback, and high resolution & frame-rate contents. To stimulate the reconstruction of immersive volumetric videos, we introduce ImViD, a multi-view, multi-modal dataset featuring complete space-oriented data capture and various indoor/outdoor scenarios. Our capture rig supports multi-view video-audio capture while on the move, a capability absent in existing datasets, significantly enhancing the completeness, flexibility, and efficiency of data capture.\n  The captured multi-view videos (with synchronized audios) are in 5K resolution at 60FPS, lasting from 1-5 minutes, and include rich foreground-background elements, and complex dynamics. We benchmark existing methods using our dataset and establish a base pipeline for constructing immersive volumetric videos from multi-view audiovisual inputs for 6-DoF multi-modal immersive VR experiences. The benchmark and the reconstruction and interaction results demonstrate the effectiveness of our dataset and baseline method, which we believe will stimulate future research on immersive volumetric video production.", "date": "2025-03-18", "categories": []}
{"id": "2604.00648", "title": "DirectFisheye-GS: Enabling Native Fisheye Input in Gaussian Splatting with Cross-View Joint Optimization", "abstract": "3D Gaussian Splatting (3DGS) has enabled efficient 3D scene reconstruction from everyday images with real-time, high-fidelity rendering, greatly advancing VR/AR applications. Fisheye cameras, with their wider field of view (FOV), promise high-quality reconstructions from fewer inputs and have recently attracted much attention. However, since 3DGS relies on rasterization, most subsequent works involving fisheye camera inputs first undistort images before training, which introduces two problems: 1) Black borders at image edges cause information loss and negate the fisheye's large FOV advantage; 2) Undistortion's stretch-and-interpolate resampling spreads each pixel's value over a larger area, diluting detail density -- causes 3DGS overfitting these low-frequency zones, producing blur and floating artifacts. In this work, we integrate fisheye camera model into the original 3DGS framework, enabling native fisheye image input for training without preprocessing. Despite correct modeling, we observed that the reconstructed scenes still exhibit floaters at image edges: Distortion increases toward the periphery, and 3DGS's original per-iteration random-selecting-view optimization ignores the cross-view correlations of a Gaussian, leading to extreme shapes (e.g., oversized or elongated) that degrade reconstruction quality. To address this, we introduce a feature-overlap-driven cross-view joint optimization strategy that establishes consistent geometric and photometric constraints across views-a technique equally applicable to existing pinhole-camera-based pipelines. Our DirectFisheye-GS matches or surpasses state-of-the-art performance on public datasets. Project Page: https://yzxqh.github.io/DirectFisheye-GS/ .", "date": "2026-04-01", "categories": []}
{"id": "2605.26230", "title": "Geometry-Aware Representation Denoising for Robust Multi-view 3D Reconstruction", "abstract": "Multi-view 3D reconstruction has achieved remarkable progress with the advent of feed-forward 3D reconstruction models. However, these models are typically trained and evaluated under ideal, degradation-free imaging conditions, whereas real-world observations often contain degradations that differ significantly from such settings. Improving robustness for multi-view 3D reconstruction under degraded conditions therefore remains an important challenge. We present Geometry-Aware Representation Denoising (GARD), a novel framework that performs diffusion-based multi-view restoration directly in the feature space of a feed-forward 3D reconstruction model. This design exploits the geometry-aware feature representations of the 3D reconstructor to effectively recover accurate scene geometry. Furthermore, by employing an additional RGB image decoder, the refined representations can also be used to restore high-quality RGB images, thereby enabling the simultaneous recovery of 3D scene geometry and high-quality imagery. Comprehensive experiments on the Depth Anything 3 (DA3) benchmark demonstrate the effectiveness of the proposed GARD framework.", "date": "2026-05-25", "categories": []}
{"id": "2605.26447", "title": "Underwater360: Reconstructing Underwater Scenes from Panoramic Images with Omnidirectional Gaussian Splatting", "abstract": "Underwater scene reconstruction is essential for immersive exploration of aquatic environments, yet remains challenging due to complex participating-media effects such as absorption and scattering, as well as the limited field of view (FoV) of conventional cameras. Although combining panoramic imaging with 3D Gaussian Splatting (3DGS) offers a promising direction for photorealistic underwater rendering, traditional 3DGS struggles with both spherical projection distortion and underwater medium degradation. In this paper, we propose \\textbf{Underwater360}, a physics-informed omnidirectional 3DGS framework for underwater panoramic scene reconstruction. First, we introduce an Omnidirectional Gaussian Splatting module that performs ray casting directly in spherical camera space instead of relying on 2D projection approximations, thereby reducing geometric distortions under 360$^\\circ$ FoV. Second, we design a physics-based appearance-medium modeling architecture with pose-conditioned appearance embeddings to explicitly decouple intrinsic scene radiance from depth-dependent backscatter and attenuation, enabling physically grounded scene appearance restoration. Finally, we establish a new panoramic underwater benchmark dataset containing both synthetic and real-world scenes. Extensive experiments demonstrate that Underwater360 achieves superior performance in underwater novel view synthesis and scene appearance restoration, delivering improved rendering quality and cross-view consistency in complex underwater environments. The code and datasets are released at https://github.com/SwcK423/Underwater360", "date": "2026-05-26", "categories": []}
{"id": "2605.26616", "title": "Gaussian-Voxel Duet: A Dual-Scaffolding Hybrid Representation for Fast and Accurate Monocular Surface Reconstruction", "abstract": "While 3D Gaussian Splatting has achieved remarkable success in photorealistic novel view synthesis, its pursuit of fast and high-fidelity 3D reconstruction has long been constrained by a trade-off between geometric accuracy and optimization efficiency. Methods specialized in image rendering converge quickly at the cost of imperfect geometry caused by superfluous primitives overfitting training views, while methods integrating neural signed-distance field (SDF) for better geometry incur prohibitive training costs. In this paper, we attempt to strike a better trade-off by tethering scaffold-anchored Gaussians to a jointly optimized sparse voxel scaffold. This hybrid Gaussian-Voxel representation explicitly confines anchored Gaussians to a narrow band around surfaces defined by voxelized SDFs, which effectively improves representation efficiency and condenses floating Gaussians without sacrificing geometry quality. An implicit surface tethering loss further pulls individual Gaussian primitives closer to SDF-induced surfaces in a mutually regularized manner for improved reconstruction accuracy. Extensive experiments on diverse real-world indoor scenes from ScanNet++, ScanNetv2, and DeepBlending datasets demonstrate that our method achieves state-of-the-art surface reconstruction quality as well as superior novel view synthesis against leading baselines, while maintaining fast training convergence and real-time rendering. Code will be available at https://github.com/duzh11/VoxelGS.", "date": "2026-05-26", "categories": []}
{"id": "2605.27372", "title": "G3T Up! Gravity Aligned Coordinate Frames Simplify Pointmap Processing", "abstract": "Modern feed-forward 3D reconstruction methods like VGGT predict pixel-aligned pointmaps in camera-centric coordinate frames. However, this choice of coordinate frame is not always optimal. We propose instead to predict pointmaps in upright, gravity-aligned frames that exploit strong structural cues present in many real-world scenes. Unlike camera-centric frames, gravity-aligned frames share a common vertical axis across viewpoints, reducing the rotational degrees of freedom needed to relate pointmaps to one another. To this end, we introduce the Gravity Grounded Geometry Transformer (G3T), fine-tuned from existing models on gravity-aligned 3D data. G3T produces highly accurate gravity-aware predictions, including upright pointmaps and camera-to-gravity poses. We further introduce G3T-Long, a submap-based incremental 3D reconstruction pipeline that leverages the reduced rotational degrees of freedom afforded by upright frames to achieve significantly improved reconstruction accuracy.", "date": "2026-05-26", "categories": []}
{"id": "2503.23670", "title": "Learning Bijective Surface Parameterization for Inferring Signed Distance Functions from Sparse Point Clouds with Grid Deformation", "abstract": "Inferring signed distance functions (SDFs) from sparse point clouds remains a challenge in surface reconstruction. The key lies in the lack of detailed geometric information in sparse point clouds, which is essential for learning a continuous field. To resolve this issue, we present a novel approach that learns a dynamic deformation network to predict SDFs in an end-to-end manner. To parameterize a continuous surface from sparse points, we propose a bijective surface parameterization (BSP) that learns the global shape from local patches. Specifically, we construct a bijective mapping for sparse points from the parametric domain to 3D local patches, integrating patches into the global surface. Meanwhile, we introduce grid deformation optimization (GDO) into the surface approximation to optimize the deformation of grid points and further refine the parametric surfaces. Experimental results on synthetic and real scanned datasets demonstrate that our method significantly outperforms the current state-of-the-art methods. Project page: https://takeshie.github.io/Bijective-SDF", "date": "2025-03-31", "categories": []}
{"id": "2511.18794", "title": "ChronoGS: Disentangling Invariants and Changes in Multi-Period Scenes", "abstract": "Multi-period image collections are common in real-world applications. Cities are re-scanned for mapping, construction sites are revisited for progress tracking, and natural regions are monitored for environmental change. Such data form multi-period scenes, where geometry and appearance evolve. Reconstructing such scenes is an important yet underexplored problem. Existing pipelines rely on incompatible assumptions: static and in-the-wild methods enforce a single geometry, while dynamic ones assume smooth motion, both failing under long-term, discontinuous changes. To solve this problem, we introduce ChronoGS, a temporally modulated Gaussian representation that reconstructs all periods within a unified anchor scaffold. It's also designed to disentangle stable and evolving components, achieving temporally consistent reconstruction of multi-period scenes. To catalyze relevant research, we release ChronoScene dataset, a benchmark of real and synthetic multi-period scenes, capturing geometric and appearance variation. Experiments demonstrate that ChronoGS consistently outperforms baselines in reconstruction quality and temporal consistency. Our code and the ChronoScene dataset are publicly available at https://github.com/ZhongtaoWang/ChronoGS.", "date": "2025-11-24", "categories": []}
{"id": "2605.24195", "title": "Single View Seafloor Recovery from Imaging Sonar via Differentiable Rendering", "abstract": "Sonar is often the only modality suitable for high-resolution imaging underwater due to light attenuation and turbidity. Forward-looking imaging sonar provides measurements over range and horizontal angle but collapses vertical structure into a flat image, creating ambiguities that make 3D recovery challenging. A common use case for imaging sonar is underwater terrain mapping (bathymetry), yet current methods require many views, expensive multi-sensor setups, or significant training data, which limits use and adaptability to new environments.\n  We present a training-free method that recovers bathymetry from a single sonar image in under 30 seconds via differentiable rendering, conditioned on a known seafloor tilt. To our knowledge, this is the first differentiable rendering approach for single-view height recovery in sonar. Our method implements differentiable sonar ray tracing and optimizes an explicit height field to reproduce the target image. On synthetic datasets, our approach outperforms a supervised CNN under distribution shift and remains close on rough terrain, while the CNN wins in-distribution. By modeling physically grounded priors of the sonar process, our method adapts across sensor configurations and environments without training data.", "date": "2026-05-22", "categories": []}
{"id": "2605.25563", "title": "CodecSplat: Ultra-Compact Latent Coding for Feed-Forward 3D Gaussian Splatting", "abstract": "While feed-forward 3D Gaussian splatting reconstructs renderable Gaussian primitives from sparse context views without per-scene optimization, existing pipelines do not provide a compact scene representation for storage or transmission. A natural solution is to apply existing 3DGS compression methods to the generated Gaussian primitives. However, this approach operates on the final irregular 3D representation and is decoupled from the internal feature-to-Gaussian generation process, which limits compression efficiency. To address this, we introduce CodecSplat, an ultra-compact latent coding framework for feed-forward 3D Gaussian splatting. CodecSplat first encodes an intermediate 2D Gaussian-generation feature into an entropy-coded scene bitstream. At the decoder, the latent feature is reconstructed and used to predict depth and Gaussian parameters, which are then mapped to 3D Gaussian primitives. Note that, by integrating compression into the feed-forward Gaussian generation pipeline, CodecSplat avoids inefficient compression over irregular 3D Gaussian primitives and allows the codec to exploit the structured intermediate feature representation. We instantiate CodecSplat on a feed-forward Gaussian splatting backbone with depth-guided multi-view feature refinement and a hierarchical learned feature codec. On DL3DV and RealEstate10K datasets, CodecSplat achieves 23.56-26.36 dB and 24.76-27.05 dB PSNR with only 20.00-107.77 KiB and 3.37-12.51 KiB per scene, respectively. This is roughly one order of magnitude smaller than compressing feed-forward generated Gaussian primitives, while preserving controllable rate-distortion behavior.", "date": "2026-05-25", "categories": []}
{"id": "2605.25909", "title": "R5DGS: Semantic-Aware 4D Gaussian Splatting with Rigid Body Constraints for Efficient Dynamic Scene Reconstruction", "abstract": "Reconstructing and predicting dynamic 3D scenes from multi-view videos is a foundational task for robotics, AR/VR, and digital twins. Recent physics-informed Gaussian Splatting methods achieve impressive future frame extrapolation but lack semantic awareness and suffer from large computational overhead. We introduce $\\textbf{R5DGS}$, a framework that augments a physics-driven 4D Gaussian representation with compact Identity Encoding vectors, enabling precise Gaussian-to-object association. By constructing an offline CLIP-based object lookup table, we support open-vocabulary text prompting to retrieve and render object-specific Gaussians across arbitrary timestamps and viewpoints. Furthermore, we propose a rigid-body inference constraint that predicts and integrates physical dynamics exclusively for object centroids, propagating motion to associated Gaussians via relative transformations. This optimization yields a 11 FPS speedup during extrapolation without compromising trajectories plausibility.", "date": "2026-05-25", "categories": []}
{"id": "2605.26115", "title": "TriSplat: Simulation-Ready Feed-Forward 3D Scene Reconstruction", "abstract": "Sparse-view 3D reconstruction is increasingly addressed with feed-forward splatting networks that predict explicit primitives directly from images. Yet most existing methods remain centered on Gaussian primitives and expose surfaces only indirectly: extracting a usable mesh for downstream simulation, physics reasoning, or embodied interaction still requires expensive post-hoc steps that break the feed-forward promise. This limitation is especially pronounced in pose-free settings, where scene structure and camera parameters must be estimated jointly from sparse observations. We present TriSplat, a feed-forward reconstruction network that represents scenes with oriented triangle primitives and directly exports simulation-ready mesh scenes from a single forward pass. Given input images, the network predicts local 3D point maps, triangle attributes, camera poses, and optional intrinsics. Rather than regressing triangle orientation as an unconstrained latent variable, our approach constructs geometry normals from the predicted point maps, refines them with an image-conditioned normal head, and converts them into stable local frames for triangle parameterization. A mono-normal bootstrap schedule further stabilizes early training, while opacity and blur scheduling progressively sharpens the learned surface representation for direct mesh extraction. Experiments on RealEstate10K and DL3DV show that this representation produces more geometry-faithful reconstructions than Gaussian feed-forward baselines while maintaining competitive novel-view rendering quality. Because the rendering primitives are themselves surface triangles, the output can be directly ingested by physics engines, collision detectors, and standard rendering pipelines without any conversion, making it a practical simulation-ready solution for feed-forward 3D scene reconstruction.", "date": "2026-05-25", "categories": []}
{"id": "2503.20066", "title": "Learning Scene-Level Signed Directional Distance Function with Ellipsoidal Priors and Neural Residuals", "abstract": "Dense reconstruction and differentiable rendering are fundamental tightly connected operations in 3D vision and computer graphics. Recent neural implicit representations demonstrate compelling advantages in reconstruction fidelity and differentiability over conventional discrete representations such as meshes, point clouds, and voxels. However, many neural implicit models, such as neural radiance fields (NeRF) and signed distance function (SDF) networks, are inefficient in rendering due to the need to perform multiple queries along each camera ray. Moreover, NeRF and Gaussian Splatting methods offer impressive photometric reconstruction but often require careful supervision to achieve accurate geometric reconstruction. To address these challenges, we propose a novel representation called signed directional distance function (SDDF). Unlike SDF and similar to NeRF, SDDF has a position and viewing direction as input. Like SDF and unlike NeRF, SDDF directly provides distance to the observed surface rather than integrating along the view ray. As a result, SDDF achieves accurate geometric reconstruction and efficient differentiable directional distance prediction. To learn and predict scene-level SDDF efficiently, we develop a differentiable hybrid representation that combines explicit ellipsoid priors and implicit neural residuals. This allows the model to handle distance discontinuities around obstacle boundaries effectively while preserving the ability for dense high-fidelity distance prediction. Through extensive evaluation against state-of-the-art representations, we show that SDDF achieves (i) competitive SDDF prediction accuracy, (ii) faster prediction speed than SDF and NeRF, and (iii) superior geometric consistency compared to NeRF and Gaussian Splatting.", "date": "2025-03-25", "categories": []}
{"id": "2605.22020", "title": "ForeSplat: Optimization-Aware Foresight for Feed-Forward 3D Gaussian Splatting", "abstract": "Feed-forward 3D Gaussian Splatting models offer fast single-pass reconstruction,but scaling them to match per-scene optimization quality is fundamentally hindered by the scarcity of large-scale 3D annotations. A practical compromise is predict-then-refine,where post-prediction optimization compensates for the limited capacity of the feed-forward network. However,standard feed-forward 3DGS is trained solely for zero-step rendering error,ignoring whether its output constitutes a good initialization for the downstream optimizer. We present ForeSplat,an optimization-aware training framework that equips feed-forward 3DGS models to produce initializations explicitly designed for rapid,effective refinement. By offloading part of the scene-modeling burden to the optimizer,ForeSplat substantially reduces the capacity pressure on the feed-forward model,making high-quality reconstruction feasible even with compact networks. At its core is MetaGrad,a lightweight multi-anchor meta-gradient training rule that bypasses costly higher-order differentiation through the 3DGS optimizer. MetaGrad unrolls a short inner-loop refinement trajectory,samples anchor states,and back-propagates aggregated first-order gradients to the prediction head as a surrogate optimization-aware signal. This fine-tuning adds no inference cost and enables high-quality reconstruction within seconds after a few refinement steps. We instantiate ForeSplat on diverse backbones,including AnySplat,Pi3X,and a distilled variant tailored for edge deployment. Across all tested architectures,a ForeSplat-trained initialization converges in fewer refinement steps and reaches a higher peak reconstruction quality than its vanilla counterpart,even fully converged. The framework consistently bridges the gap between amortized prediction and per-scene optimization,establishing a practical path toward lightweight,high-fidelity 3D reconstruction.", "date": "2026-05-21", "categories": []}
{"id": "2605.23287", "title": "LangFlash: Feed-forward 3D Language Gaussian Splatting from Sparse Unposed Images", "abstract": "We present LangFlash, a feed-forward framework for 3D Language Gaussian Splatting that reconstructs 3D scenes parameterized by Gaussian primitives enriched with language-aligned semantic features from sparse unposed multi-view images. Unlike optimization-based 3D methods, LangFlash directly predicts the geometry and semantics in a single forward pass, enabling low-latency 3D reconstruction and language-consistent scene understanding. To support large-scale training, we enriched the RealEstate10k dataset with coherent and dense semantic information for 3D semantic supervision. Furthermore, we propose a sparse semantic encoding scheme that combines a global semantic dictionary with locally varying per-primitive weights, preserving high-level linguistic information, while reducing representation complexity. Experimental results show that LangFlash achieves superior novel view synthesis and semantic consistency compared with previous methods. This study establishes a new paradigm for pose-free, language-grounded 3D scene reconstruction, advancing generalizable 3D vision and multimodal scene understanding. Demo is available at https://liylo.github.io/langflash.github.io/.", "date": "2026-05-22", "categories": []}
{"id": "2605.23602", "title": "GlowGS: Generative Semantic Feature Learning for 3D Gaussian Splatting in Nighttime Glow Scenes", "abstract": "Existing 3DGS methods effectively render high-quality novel views in clear-day scenes. However, they struggle with night scenes, particularly in glow regions, due to the lack of structural features such as textures and edges, which are key cues for splatting-based reconstruction. To address this problem, we leverage a diffusion model and a Vision Foundation Model (VFM) to compensate for missing structural cues. Our method consists of two key novel ideas: semantic feature generation and novel-view semantic learning. First, semantic feature generation produces high-quality semantic features as implicit structural cues for novel views. Specifically, a diffusion model synthesizes novel views with unknown camera poses from training views, while a VFM evaluates their quality. Once high-quality novel views are identified, the VFM extracts robust features to construct the semantic feature bank. Second, novel-view semantic learning enables 3DGS to optimize rendered novel views without requiring ground truth. It achieves this by extracting semantic features from a rendered novel view, searching the feature bank for the most similar features, and minimizing their distance. This process enforces implicit structural constraints, ensuring semantically coherent, artifact-free rendered views. Extensive experiments demonstrate the effectiveness of our GlowGS in generating semantically accurate 3D views, showing significant improvements over existing methods.", "date": "2026-05-22", "categories": []}
{"id": "2605.23888", "title": "GenRecon: Bridging Generative Priors for Multi-View 3D Scene Reconstruction", "abstract": "We introduce a new approach to high-fidelity 3D scene reconstruction from multi-view RGB images that tightly couples reconstruction with a strong generative 3D prior. We cast scene reconstruction as conditional 3D generation over a set of spatially-localized, overlapping chunks that together tile the scene, scaling generation to large scene extents. Crucially, we inherit the fidelity and completeness of state-of-the-art generative shape models -- we use Trellis.2 as an example -- which we generalize to the scene level. To this end, we propose a projection-based conditioning mechanism that lifts posed multi-view image features into a coherent 3D representation aligned with the generative model, independent of view ordering and spatially anchored to the scene, yielding high-fidelity, multi-view consistent generated geometry. This enables lifting the strong object-level prior of Trellis.2 to multi-view, scene-scale generation, producing faithful, editable PBR mesh reconstructions of indoor environments. As a result, we obtain high-fidelity results that outperform cutting-edge reconstruction methods by 16%.", "date": "2026-05-22", "categories": []}
{"id": "2605.23889", "title": "HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction", "abstract": "Online 3D reconstruction requires estimating camera pose and scene geometry under strict causal and bounded-memory constraints. Existing methods often suffer from drift, jitter, or collapse on long sequences. We trace these failures to a fundamental mismatch. Streaming geometry is inherently temporally heterogeneous, with evidence ranging from short-lived correspondences to persistent global scale. However, current architectures impose uniform and pathological influence patterns. For example, sliding windows enforce hard cutoffs, while ungated recurrence and causal attention cause cache saturation and spike-like attention sinks. To resolve this, we formalize geometric propagation as an \\emph{evidence influence kernel} and propose HorizonStream, a long-horizon Transformer that explicitly factorizes this kernel. For the long-range temporal factor, Geometric Linear Attention learns channel-wise decay rates to enable bounded, multi-timescale propagation of geometric evidence. For the short-range spatial factor, Geometric Local Attention with Spatiotemporal RoPE performs reliable 3D matching while suppressing attention sinks. Finally, Metric Readout Tokens recover stable scale and rigid pose directly from the persistent geometric state. Extensive experiments show that HorizonStream, trained on only 48-frame clips, generalizes stably to sequences exceeding 10,000\\ frames with constant memory and linear time, achieving state-of-the-art streaming 3D reconstruction performance. Project Page: https://3dagentworld.github.io/horizonstream/", "date": "2026-05-22", "categories": []}
{"id": "2605.23892", "title": "Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers", "abstract": "Visual geometry transformers have become powerful architectures for multi-view 3D reconstruction, enabling joint prediction of multiple 3D attributes in a feed-forward manner. However, their computational cost grows quadratically with the input sequence length due to the global attention layers inside these models. This limits both their scalability and efficiency. In this work, we address this challenge with a simple yet general strategy: restricting the number of key/value tokens that each query interacts with during global attention. To achieve effective token selection, we introduce a two-stage framework. First, an inter-frame selection step operates at the frame level to identify frames that should be preserved. Second, an intra-frame selection step further discards more redundant tokens within the selected frames. Our analysis highlights the advantage of a diversity-based strategy for inter-frame selection, which ensures broad coverage of the scene. For intra-frame selection, we show that layer-aware sparsification is necessary, with the selection process guided by the entropy of the global attention pattern. Our approach offers a superior speed-accuracy trade-off compared to existing solutions. Extensive experiments show that it accelerates visual geometry transformers by over 85% for scenes with 500 images while maintaining, or even improving, baseline performance, which hints that how our token selection strategy can play a crucial role in future applications of visual geometry transformers. Our project website is available at https://zsh2000.github.io/good-token-hunting.github.io.", "date": "2026-05-22", "categories": []}
{"id": "2605.05749", "title": "Ray-Aware Pointer Memory with Adaptive Updates for Streaming 3D Reconstruction", "abstract": "Dense 3D reconstruction from continuous image streams requires both accurate geometric aggregation and stable long-term memory management. Recent feed-forward reconstruction frameworks integrate observations through persistent memory representations, yet most rely primarily on appearance-based similarity when updating memory. Such appearance-driven integration often leads to redundant accumulation of observations and unstable geometry when viewpoint changes occur. In this work, we propose a ray-aware pointer memory for streaming 3D reconstruction that explicitly models both spatial location and viewing direction within a unified memory representation. Each memory pointer stores its 3D position, associated ray direction, and feature embedding, allowing the system to reason jointly about geometric proximity and viewpoint consistency. Based on this representation, we introduce an adaptive pointer update strategy that replaces traditional fusion-based memory compression with a retain-or-replace mechanism. Instead of averaging nearby observations, the system selectively retains informative pointers while discarding redundant ones, preserving distinctive geometric structures while maintaining bounded memory growth. Furthermore, the joint reasoning over spatial distance and ray-direction discrepancy enables the system to distinguish between local redundancy, novel observations, and potential loop revisits in a unified manner. When loop candidates are detected, pose refinement is triggered to enforce global geometric consistency across the reconstruction. Extensive experiments demonstrate that the proposed ray-aware memory design significantly improves long-term reconstruction stability and camera pose accuracy while maintaining efficient streaming inference. Our approach provides a principled framework for scalable and drift-resistant online 3D reconstruction from image streams.", "date": "2026-05-07", "categories": []}
{"id": "2605.22190", "title": "No Pose, No Problem in 4D: Feed-Forward Dynamic Gaussians from Unposed Multi-View Videos", "abstract": "Recent feed-forward 3D gaussian splatting methods have made dramatic progress on individual aspects of 3D scene reconstruction, but no existing method jointly addresses dynamic content, multi-view input, and unknown camera poses in a single feed-forward pass. Methods that handle dynamics either require accurate camera poses or accept only monocular input; pose-free multi-view methods address only static scenes; and per-scene optimization methods bridge some of these gaps but at minutes-to-hours cost per scene. We introduce NoPo4D, the first feed-forward system that addresses this empty quadrant. Building on a pretrained geometry backbone and recent 4D Gaussian frameworks, NoPo4D introduces a velocity decomposition that splits Gaussian motion into per-pixel image-plane shifts and depth changes, allowing direct supervision from pseudo ground-truth optical flow on the 2D component. This sidesteps both the differentiable rendering that couples prior posed methods to pose accuracy and the 3D motion ground truth that prior pose-free methods require. The system is rounded out by a bidirectional motion encoder for cross-view and cross-frame feature aggregation, and view-dependent opacity that mitigates cross-view and cross-timestep Gaussian misalignments. On four multi-view dynamic benchmarks, NoPo4D consistently outperforms prior feed-forward baselines, and with an optional post-optimization stage surpasses per-scene optimization methods, while running orders of magnitude faster.", "date": "2026-05-21", "categories": []}
{"id": "2605.22420", "title": "Diffusion-guided Generalizable Enhancer for Urban Scene Reconstruction", "abstract": "Urban scene reconstruction from real-world observations has emerged as a powerful tool for self-driving development and testing. While current neural rendering approaches achieve high-fidelity rendering along the recorded trajectories, their quality degrades significantly under large viewpoint shifts, limiting the applicability for closed-loop simulation. Recent works have shown promising results in using diffusion models to enhance quality at these challenging viewpoints and distill improvements back into 3D representations. However, they often require costly per-scene optimization, and the distilled representations remain fragile and fail to generalize beyond limited synthesized views. To address these limitations, we propose GenRe, a novel diffusion-guided generalizable enhancer for urban scene reconstruction. GenRe takes as input any pretrained 3D Gaussian representation and fixes the deficiencies within a few minutes. By learning to distill generative priors across diverse scenes, GenRe produces robust and high-fidelity representation efficiently that generalizes reliably to challenging unseen viewpoints (e.g., lane change). Experiments show that GenRe outperforms existing methods in both quality and efficiency and benefits various downstream tasks, enabling robust and scalable sensor simulation for autonomous driving.", "date": "2026-05-21", "categories": []}
{"id": "2406.14978", "title": "E2GS: Event Enhanced Gaussian Splatting", "abstract": "Event cameras, known for their high dynamic range, absence of motion blur, and low energy usage, have recently found a wide range of applications thanks to these attributes. In the past few years, the field of event-based 3D reconstruction saw remarkable progress, with the Neural Radiance Field (NeRF) based approach demonstrating photorealistic view synthesis results. However, the volume rendering paradigm of NeRF necessitates extensive training and rendering times. In this paper, we introduce Event Enhanced Gaussian Splatting (E2GS), a novel method that incorporates event data into Gaussian Splatting, which has recently made significant advances in the field of novel view synthesis. Our E2GS effectively utilizes both blurry images and event data, significantly improving image deblurring and producing high-quality novel view synthesis. Our comprehensive experiments on both synthetic and real-world datasets demonstrate our E2GS can generate visually appealing renderings while offering faster training and rendering speed (140 FPS). Our code is available at https://github.com/deguchihiroyuki/E2GS.", "date": "2024-06-21", "categories": []}
{"id": "2605.10830", "title": "Predicting 3D structure by latent posterior sampling", "abstract": "The remarkable achievements of both generative models of 2D images and neural field representations for 3D scenes present a compelling opportunity to integrate the strengths of both approaches. In this work, we propose a methodology that combines a NeRF-based representation of 3D scenes with probabilistic modeling and reasoning using diffusion models. We view 3D reconstruction as a perception problem with inherent uncertainty that can thereby benefit from probabilistic inference methods. The core idea is to represent the 3D scene as a stochastic latent variable for which we can learn a prior and use it to perform posterior inference given a set of observations. We formulate posterior sampling using the score-based inference method of diffusion models in conjunction with a likelihood term computed from a reconstruction model that includes volumetric rendering. We train the model using a two-stage process: first we train the reconstruction model while auto-decoding the latent representations for a dataset of 3D scenes, and then we train the prior over the latents using a diffusion model. By using the model to generate samples from the posterior we demonstrate that various 3D reconstruction tasks can be performed, differing by the type of observation used as inputs. We showcase reconstruction from single-view, multi-view, noisy images, sparse pixels, and sparse depth data. These observations vary in the amount of information they provide for the scene and we show that our method can model the varying levels of inherent uncertainty associated with each task. Our experiments illustrate that this approach yields a comprehensive method capable of accurately predicting 3D structure from diverse types of observations.", "date": "2026-05-11", "categories": []}
{"id": "2605.18447", "title": "NeRF-based Spacecraft Reconstruction from Monocular Imagery Under Illumination Variability and Pose Uncertainty", "abstract": "Autonomous rendezvous and proximity operations around uncooperative, unknown spacecraft are critical for active debris removal and on-orbit servicing missions. A key component of such operations is the offline reconstruction of a 3D model of the target from a set of 2D images. This task is challenging due to two main factors. First, in-orbit illumination conditions exhibit considerable variability, and change rapidly over time. Second, the inaccuracy of pose information in the images, results in 3D reconstruction uncertainty. To overcome these challenges, we propose to extend Neural Radiance Fields with per-image degrees of freedom: a learnable appearance embedding that captures the illumination conditions specific to each image, and an image-specific pose correction term that refines its noisy pose label to increase 3D consistency across images. These parameters add minimal complexity, as they are learned jointly with the NeRF, yet they substantially improve robustness to illumination variability and pose inaccuracies. We validate our approach on three image sets representative of in-orbit operations, demonstrating its effectiveness for offline reconstruction and highlighting its suitability for online reconstruction, an open problem in the field.", "date": "2026-05-18", "categories": []}
{"id": "2605.21032", "title": "Towards Physically Consistent 4D Scene Reconstruction for Closed-loop Autonomous Driving Simulation", "abstract": "High-fidelity street scene reconstruction is pivotal for end-to-end autonomous driving simulation, where novel-view synthesis (NVS) and time-varying information modeling are two fundamental capabilities to facilitate closed-loop training. However, existing 3DGS methods and their 4D extensions fail to simultaneously achieve both. To bridge this gap, we establish an information-geometric diagnostic framework, revealing that this limitation stems from a credit assignment dilemma between spatial and temporal parameters. Specifically, the deterministic coupling between viewpoint and time in single-source observation creates a low-rank structure that induces massive null-space ambiguity between static view-dependent and dynamic time-varying components. Temporal information overshadows spatial cues, causing the estimation variance of spatial parameters to diverge. To address this issue, we propose Orthogonal Projected Gradient (OPG), a hierarchical training method designed to restore spatial identifiability. OPG prioritizes the integrity of spatial representations by securing them in an initial stage, then restricts temporal updates to the spatial null space, enabling proactive credit assignment. While OPG isolates temporal updates algebraically, Temporal Regularization Strategy is proposed to further refine the temporal solution space by imposing a smoothness constraint based on the physical prior of consistent appearance evolution, ensuring that the reconstructed scene remains physically consistent in closed-loop simulation. Extensive experiments demonstrate that our method not only maintains stable NVS capabilities but also demonstrates superior performance in traditional observation-reproducing metrics, which indirectly reflect the capability of modeling temporal dynamics.", "date": "2026-05-20", "categories": []}
{"id": "2605.21131", "title": "UniT: Unified Geometry Learning with Group Autoregressive Transformer", "abstract": "Recent feed-forward models have significantly advanced geometry perception for inferring dense 3D structure from sensor observations. However, its essential capabilities remain fragmented across multiple incompatible paradigms, including online perception, offline reconstruction, multi-modal integration, long-horizon scalability, and metric-scale estimation. We present UniT, a unified model built upon a novel Group Autoregressive Transformer, which reformulates these seemingly disparate capabilities within a single framework. The key idea is to treat groups of sensor observations as the basic autoregressive units and predict the corresponding point maps in an anchor-free and scale-adaptive manner. More specifically, diverse view configurations in both online and offline settings are naturally unified within a single group autoregression process. By varying the group size, online mode operates over multiple autoregressive steps with single-frame groups, whereas offline mode aggregates a multi-frame group in a single forward pass. Meanwhile, a queue-style KV caching mechanism ensures bounded autoregressive memory over long horizons. This is enabled by reducing long-range dependencies on early frames through anchor-free relational modeling, thereby allowing outdated memory to be discarded on the fly. To improve metric-scale generalization across scenes, a scale-adaptive geometry loss is further introduced within this framework. It couples relative geometric constraints with a partial absolute scale term, implicitly regularizing global scale and inducing a progressive transition from scale-invariant geometry to metric-scale solutions. Together with a dedicated modal attention module for integrating auxiliary modalities, UniT achieves state-of-the-art performance in unified geometry perception, as validated on ten benchmarks spanning seven representative tasks.", "date": "2026-05-20", "categories": []}
{"id": "2605.06270", "title": "Spark3R: Asymmetric Token Reduction Makes Fast Feed-Forward 3D Reconstruction", "abstract": "Feed-forward 3D reconstruction models based on Vision Transformers can directly estimate scene geometry and camera poses from a small set of input images, but scaling them to video inputs with hundreds or thousands of frames remains challenging due to the quadratic cost of global attention layers. Recent token-merging methods accelerate these models by compressing the token sequence within the global attention layers, but they apply a uniform reduction to query tokens and key-value tokens, ignoring their functionally distinct roles in 3D reconstruction. In this work, we identify a key property of feed-forward 3D reconstruction models: query tokens encode view-specific geometric requests and are sensitive to compression, while key-value tokens represent shared scene context and tolerate aggressive compression. Guided by this insight, we propose Spark3R, a training-free acceleration framework that decouples the compression of query tokens and key-value tokens by assigning distinct reduction factors, with intra-group token merging applied to query tokens and lightweight token pruning to key-value tokens. Additionally, Spark3R adaptively adjusts the key-value reduction factor across layers, further improving the quality-efficiency trade-off. As a plug-and-play framework requiring no retraining, Spark3R integrates directly into multiple pretrained feed-forward 3D reconstruction models, including VGGT, $\\pi^3$, Depth-Anything-3, and VGGT-$\\Omega$, and achieves up to $28\\times$ speedup on 1,000-frame inputs while maintaining competitive reconstruction quality.", "date": "2026-05-07", "categories": []}
{"id": "2605.17942", "title": "UAVFF3D: A Geometry-Aware Benchmark for Feed-Forward UAV 3D Reconstruction", "abstract": "Feed-forward 3D reconstruction has advanced rapidly, but current models remain unreliable in UAV photogrammetric acquisition. We argue that this failure is caused not only by appearance-domain shift, but also by UAV-specific camera-geometry variations, especially oblique views and HFOV-height ambiguity. Existing UAV datasets mainly emphasize scene diversity and provide limited coverage of camera configurations, which restricts robustness evaluation and UAV-domain adaptation. To address this gap, we introduce UAVFF3D, a geometry-aware real-synthetic benchmark for feed-forward UAV 3D reconstruction. UAVFF3D contains more than 170k real UAV images and more than 370k synthetic images rendered from high-quality textured 3D models, covering diverse HFOVs, flight altitudes, viewing directions, and acquisition patterns. It also includes a controlled HFOV-height test subset for diagnosing projection-geometry ambiguity. We further propose an evaluation protocol that jointly assesses camera-geometry estimation and dense scene reconstruction under a shared global alignment, avoiding the bias caused by separate camera and geometry alignments. Experiments on representative feed-forward reconstruction models show that UAVFF3D-based domain adaptation consistently improves camera and geometry estimation, reducing Ray Error by up to 84.2%, Pose ATE by up to 76.0%, and Chamfer Distance by up to 41.1%. In oblique scenes, adaptation reduces the oblique-nadir rotation gap by up to 90.7%. Under HFOV-height ambiguity, it improves robustness across HFOV-height configurations and yields more stable performance across HFOV settings. Incorporating camera priors further improves reconstruction under UAV-specific acquisition geometries. The dataset and evaluation code are available at https://github.com/yanxian-ll/UAVFF3D .", "date": "2026-05-18", "categories": []}
{"id": "2605.19304", "title": "MMGS: 10$\\times$ Compressed 3DGS through Optimal Transport Aggregation based on Multi-view Ranking", "abstract": "While 3D Gaussian Splatting (3DGS) has revolutionized 3D reconstruction, it suffers from significant overhead due to massive redundant primitives. Existing compression methods typically rely on local sampling or fixed pruning thresholds, which often struggle to balance redundancy reduction with high-fidelity rendering. To address this, we propose a novel framework that formulates Gaussian optimization as a global geometric distribution matching problem. Specifically, our approach integrates three components: (1) we introduce a multi-view 3D Gaussian contribution ranking mechanism that filters primitives using geometric consistency instead of local heuristics; (2) we propose a global Optimal Transport (OT)-based aggregation algorithm that merges redundant primitives while preserving the underlying geometry; and (3) we design an OT-based densification operator that maintains the Gaussian's distributional properties for stable optimization. Our approach achieves state-of-the-art rendering quality with only \\textbf{10$\\%$} primitives and \\textbf{10$\\times$} accelerated training speeds compared to vanilla 3DGS.", "date": "2026-05-19", "categories": []}
{"id": "2605.19539", "title": "Trust It or Not: Evidential Uncertainty for Feed-Forward 3D Reconstruction with Trust3R", "abstract": "Geometric foundation models hold promise for unconstrained dense geometry prediction from uncalibrated images. However, in current feed-forward designs, their predicted confidence scores are heuristic, lack probabilistic interpretation, and often fail to indicate where and how much the predicted geometry can be trusted. To address this gap, we present Trust3R, a lightweight evidential uncertainty framework for feed-forward 3D reconstruction. Trust3R combines gated residual mean refinement with a Normal-Inverse-Wishart evidential head, yielding a closed-form multivariate Student-t distribution for per-point geometric uncertainty. This design provides probabilistically grounded pointmap uncertainty estimates while adding moderate inference overhead. We evaluate on diverse indoor and outdoor benchmarks and compare against MASt3R's built-in confidence map as well as common uncertainty-aware baselines spanning single-pass heteroscedastic regression and sampling-based methods such as MC dropout and deep ensembles. Experimental results show that Trust3R consistently improves risk-coverage and sparsification, and generally improves geometric accuracy. These gains are reflected in stronger uncertainty ranking across benchmarks, with 25% lower AURC and 41% lower AUSE on ScanNet++, providing a practical reliability signal for uncertainty-aware weighting in downstream geometry pipelines. The project page and code are available at https://trust3r-z.github.io/.", "date": "2026-05-19", "categories": []}
{"id": "2605.19656", "title": "Cross-View Splatter: Feed-Forward View Synthesis with Georeferenced Images", "abstract": "We present Cross-View Splatter, a feed-forward method that predicts pixel-aligned Gaussian splats for outdoor scenes captured at ground level AND by satellite. Faithful reconstructions require good camera coverage, but ground imagery is time-consuming and hard to capture at scale for large outdoor scenes. Fortunately, satellite imagery can provide a global geometric prior that is easy to access via public APIs. Cross-View Splatter fuses orthorectified satellite views with GPS-tagged ground photos to predict Gaussian splats in a unified 3D coordinate frame. By aligning ground and bird's-eye feature representations, our model improves scene coverage and novel-view synthesis, compared to ground imagery alone. We train on curated georeferenced datasets and paired satellite-terrain data, mined from open mapping services. We evaluate our method on a new benchmark for novel-view synthesis with georeferenced imagery allowing comparison to prior state-of-the-art methods. Our code and data preparation will be available at https://nianticspatial.github.io/cross-view-splatter/.", "date": "2026-05-19", "categories": []}
{"id": "2605.20044", "title": "OP2GS: Object-Aware 3D Gaussian Splatting with Dual-Opacity Primitives", "abstract": "3D Gaussian Splatting (3DGS) provides an explicit and efficient scene representation, but its primitives lack inherent object-level identity, hindering downstream tasks such as open-vocabulary scene understanding. Existing methods typically address this by either distilling high-dimensional feature embeddings into Gaussians or by lifting 2D mask labels into 3D via heuristic refinement. However, feature-based approaches incur heavy storage and decoding overhead, while lifting-based pipelines remain vulnerable to label contamination: Gaussians necessary for appearance reconstruction often receive incorrect object labels during 2D-to-3D projection. We propose OP2GS, an object-aware Gaussian representation that augments each primitive with an explicit instance identity and a dedicated instance opacity $\\sigma^{*}$ for object-mask rendering. The original opacity $\\sigma$ remains responsible for visual reconstruction, while $\\sigma^{*}$ models whether a Gaussian should contribute to a particular object mask. This dual-opacity formulation decouples visual existence from instance occupancy: mislabeled Gaussians can remain available for image rendering while becoming transparent in the object-mask branch. To learn this representation, we introduce a random object loss that optimizes the 1D instance occupancy field using the standard transmittance-based visibility of 3DGS. Semantic descriptors are then attached at the object level through multi-view aggregation, eliminating per-Gaussian feature storage. Compared with feature-training approaches, OP2GS achieves competitive open-vocabulary performance while significantly reducing computational overhead. Compared with training-free pipelines, it leverages physically consistent occupancy learning to resolve visibility ambiguities.", "date": "2026-05-19", "categories": []}
{"id": "2511.00392", "title": "SonarSweep: Fusing Sonar and Vision for Robust 3D Reconstruction via Plane Sweeping", "abstract": "Accurate 3D reconstruction in visually-degraded underwater environments remains a formidable challenge. Single-modality approaches are insufficient: vision-based methods fail due to poor visibility and geometric constraints, while sonar is crippled by inherent elevation ambiguity and low resolution. Consequently, prior fusion technique relies on heuristics and flawed geometric assumptions, leading to significant artifacts and an inability to model complex scenes. In this paper, we introduce SonarSweep, a novel, end-to-end deep learning framework that overcomes these limitations by adapting the principled plane sweep algorithm for cross-modal fusion between sonar and visual data. Extensive experiments in both high-fidelity simulation and real-world environments demonstrate that SonarSweep consistently generates dense and accurate depth maps, significantly outperforming state-of-the-art methods across challenging conditions, particularly in high turbidity. To foster further research, we will publicly release our code and a novel dataset featuring synchronized stereo-camera and sonar data, the first of its kind.", "date": "2025-11-01", "categories": []}
{"id": "2603.00952", "title": "Decoupling Motion and Geometry in 4D Gaussian Splatting", "abstract": "High-fidelity reconstruction of dynamic scenes is an important yet challenging problem. While recent 4D Gaussian Splatting (4DGS) has demonstrated the ability to model temporal dynamics, it couples Gaussian motion and geometric attributes within a single covariance formulation, which limits its expressiveness for complex motions and often leads to visual artifacts. To address this, we propose VeGaS, a novel velocity-based 4D Gaussian Splatting framework that decouples Gaussian motion and geometry. Specifically, we introduce a Galilean shearing matrix that explicitly incorporates time-varying velocity to flexibly model complex non-linear motions, while strictly isolating the effects of Gaussian motion from the geometry-related conditional Gaussian covariance. Furthermore, a Geometric Deformation Network is introduced to refine Gaussian shapes and orientations using spatio-temporal context and velocity cues, enhancing temporal geometric modeling. Extensive experiments on public datasets demonstrate that VeGaS achieves state-of-the-art performance.", "date": "2026-03-01", "categories": []}
{"id": "2604.20155", "title": "GSCompleter: A Distillation-Free Plugin for Metric-Aware 3D Gaussian Splatting Completion in Seconds", "abstract": "3D Gaussian Splatting (3DGS) has revolutionized high-fidelity neural rendering with its explicit representation and efficiency. However, reconstructing scenes from sparse viewpoints suffers from severe geometric voids and floaters due to limited coverage. Current scene completion methods typically rely on an iterative \"Repair-then-Distill\" paradigm, which is computationally intensive, prone to unstable optimization, and susceptible to overfitting. To address these limitations, we propose GSCompleter, a distillation-free plugin that shifts scene completion to a stable \"Generate-then-Register\" workflow. Specifically, GSCompleter synthesizes visually plausible 2D reference images and explicitly lifts them into 3D Gaussian primitives with a consistent metric scale via a robust Stereo-Anchor View Selection mechanism. These newly generated primitives are then seamlessly integrated into the global scene using a novel Ray-Constrained Registration strategy. By replacing unstable distillation with rapid geometric registration, GSCompleter exhibits superior 3DGS completion performance across three benchmarks, enhancing both quality and efficiency over various baselines and achieving new state-of-the-art (SOTA) results.", "date": "2026-04-22", "categories": []}
{"id": "2605.10239", "title": "AdaptSplat: Adapting Vision Foundation Models for Feed-Forward 3D Gaussian Splatting", "abstract": "This work explores a simple yet powerful lightweight adapter design for feed-forward 3D Gaussian Splatting (3DGS). Existing methods typically apply complex, architecture-specific designs on top of the generic pipeline of image feature extraction $\\rightarrow$ multi-view interaction $\\rightarrow$ feature decoding. However, constrained by the scale bottleneck of 3D training data and the low-pass filtering effect of deep networks, these methods still fall short in cross-domain generalization and high-frequency geometric fidelity. To address these problems, we propose AdaptSplat, which demonstrates that without complex component engineering, introducing a single adapter of only 1.5M parameters into the generic architecture is sufficient to achieve superior performance. Specifically, we design a lightweight Frequency-Preserving Adapter (FPA) that extracts direction-aware high-frequency structural priors from the shallow features of a powerful vision foundation model backbone, and seamlessly integrates them into the generic pipeline via high-frequency positional encodings and adaptive residual modulation. This effectively compensates for the high-frequency attenuation caused by over-smoothing in deep features, improving the fitting accuracy of Gaussian primitives on complex surfaces and sharp boundaries. Extensive experiments demonstrate that AdaptSplat achieves state-of-the-art feed-forward reconstruction performance on multiple standard benchmarks, with stable generalization across domains. Code available at: https://github.com/xmw666/AdaptSplat.", "date": "2026-05-11", "categories": []}
{"id": "2605.16807", "title": "DecoRec: Decomposed 3D Scene Reconstruction from Single-View Images via Object-Level Diffusion", "abstract": "In this paper, we introduce \\textit{DecoRec}, a novel system designed to elevate single-view 2D images to a decomposed 3D scene mesh. Current methods for single-view scene reconstruction typically rely on object retrieval or the regression of coarse 3D voxels or surfaces, leading to inaccuracies in capturing the appearance and geometry of the input image. The lack of high-quality large-scale scene-level datasets further complicates direct 3D scene generation from single-view images. To achieve high-quality 3D scene generation from a single-view image, DecoRec takes advantage of recent diffusion-based single-view object reconstruction methods to reconstruct individual objects separately. Subsequently, a refinement pipeline is proposed to effectively merge these reconstructed objects, enhancing appearance and geometry through a differentiable rendering technique and diffusion-guided refinement. Our results demonstrate that DecoRec facilitates high-quality single-view scene reconstruction in both geometry and novel synthesis, offering significant benefits for downstream applications like room interior design.", "date": "2026-05-16", "categories": []}
{"id": "2605.16873", "title": "HAD: Hallucination-Aware Diffusion Priors for 3D Reconstruction", "abstract": "Diffusion priors have recently demonstrated strong capability in enhancing the quality of sparse-view 3D reconstruction by augmenting training views at novel viewpoints, but they inevitably introduce hallucinated content -- artifacts inconsistent with the input views -- into the final 3D model. To address this challenge, we propose Hallucination-Aware Diffusion prior (HAD), which estimates pixel-wise hallucination score maps for augmented images by leveraging multi-view reasoning capabilities from a feedforward novel view synthesis (NVS) network pre-trained on large-scale 3D data. These hallucination scores enable selective masking of unreliable pixels during the progressive 3D reconstruction procedure, preventing the introduction of non-existent artifacts into the 3D model. To further enhance performance, we create multiple versions of augmented images at each novel view by conditioning the diffusion prior on different input views, which are then fused into a final image that leverages the broader context across all input views. We show that our method substantially reduces hallucination artifacts in diffusion-assisted 3D reconstruction, thereby achieving state-of-the-art performance across multiple benchmarks on novel view synthesis. Our project are publicly available at \\href{https://xiliu8006.github.io/HAD-Project-website/}{project website}.", "date": "2026-05-16", "categories": []}
{"id": "2605.16925", "title": "P2GS: Physical Prior-guided Gaussian Splatting for Photometrically Consistent Urban Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as a powerful explicit representation enabling fast, high-fidelity rendering, making it a promising foundation for closed-loop simulators and perception models in autonomous driving. However, conventional 3DGS implicitly assumes consistent exposure and tone mapping across views. Real driving data violates this assumption due to heterogeneous camera pipelines and dynamic outdoor illumination, baking exposure discrepancies and sensor noise into the radiance field and producing artifacts and inconsistent illumination especially in static backgrounds crucial for realistic simulation. These issues are amplified in autonomous driving, where sparse viewpoints, varying exposures, and outdoor lighting interact, while prior work mainly targets dynamic-object reconstruction and overlooks cross-view photometric consistency. To address this limitation, we introduce P2GS, a physically consistent Gaussian Splatting framework that jointly decomposes a view-invariant linear HDR radiance field, per-view exposure scales, and tone-mapping functions from only LDR images without HDR supervision. P2GS employs a unified optimization strategy grounded in the physical image-formation process, enforcing relative-exposure consistency and HDR-domain radiance regularization. This yields a radiance field robust to inter-camera illumination differences while preserving the real-time efficiency of standard 3DGS. Experiments across real and simulated driving environments show that P2GS matches or surpasses prior methods in LDR reconstruction while providing substantially improved photometric consistency, reliable exposure normalization, and physically coherent illumination across diverse scenes.", "date": "2026-05-16", "categories": []}
{"id": "2605.16981", "title": "Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction", "abstract": "Streaming 3D reconstruction under a strict constant-memory budget hinges on how the recurrent state is updated as the stream evolves. We profile TTT3R-style per-token gates across five benchmarks and discover a structural bottleneck: the gate is intrinsically bounded in magnitude (median $0.31$; never exceeding $0.6$) and nearly frame-invariant, yielding an effective memory horizon of only $\\sim$3 frames per state token, which serves as the structural origin of long-sequence drift. We trace this to a missing axis: existing inference-time methods modulate updates only at the per-token, intra-frame level, while the orthogonal frame-level question of \\emph{how strongly each frame should contribute to the state} has been treated as content-independent. We close this gap with a scalar frame-level gate $\\alpha_t \\in (0, 1]$ derived in closed form from frame-to-frame changes of internal features -- a continuous relaxation of classical Simultaneous Localization and Mapping (SLAM) keyframe selection that requires no parameters, no training, and no extra forward pass. Across six benchmarks spanning camera pose, video depth, and 3D reconstruction at sequence lengths up to $4,541$ frames, our gate cuts ATE by $51\\%$ on long TUM-RGBD pose sequences, reduces AbsRel by $12.8\\%$ on Bonn video depth, and on KITTI long-sequence pose estimation surpasses both LongStream and Keyframe-VO, while retaining strictly constant memory at zero training cost.", "date": "2026-05-16", "categories": []}
{"id": "2605.17014", "title": "RHINO: Reconstructing Human Interactions with Novel Objects from Monocular Videos", "abstract": "Reconstructing people, objects, and their interactions in 3D is a long-standing goal for intelligent systems. Often the input is RGB video from a moving camera, making the task ill-posed; depth is ambiguous, humans and objects occlude each other, and camera and object motion entangle to create apparent motion. Most prior work addresses humans or objects in isolation, ignoring their interplay, or assumes known 3D shapes or cameras, which is impractical for real-world applications. We develop RHINO (Reconstructing Human Interactions with Novel Objects), a three-step framework that recovers in 3D a human, novel (unseen) manipulated object, and static scene in a common world frame from a monocular RGB video. First, we leverage 3D-aware foundation models to obtain cues that stabilize Structure-from-Motion (SfM) even for low-texture regions; this yields a coarse shape and apparent motion of a manipulated object from foreground pixels, and a coarse scene shape and camera motion from background pixels. Second, we estimate a human in the camera frame via an off-the-shelf method, and subtract the camera motion from apparent motion to extract the object motion; this registers the human, object, and coarse scene shapes into a common world frame. Third, we refine shapes using a compositional neural field with per-component signed-distance fields. The latter further enables differentiable contact priors that attract surfaces while penalizing interpenetration, improving the physical plausibility of the final reconstruction. For evaluation, we capture a new dataset of handheld monocular videos synchronized with a volumetric 4D capture stage, providing ground-truth shape and camera motion. RHINO outperforms state-of-the-art baselines on novel-view synthesis and 4D reconstruction. Ablations show that each stage contributes substantially. Code and data are available at https://lxxue.github.io/RHINO.", "date": "2026-05-16", "categories": []}
{"id": "2605.17303", "title": "LongDPM: Overlap-Aware 4D Reconstruction from Long Monocular Videos", "abstract": "Recovering a dynamic 3D scene from a long monocular video is crucial for dense geometry, camera motion, and temporal correspondence to remain consistent in a shared coordinate system. Existing methods face two key challenges: (1) feed-forward reconstruction models provide accurate local predictions but are limited to short clips, and (2) long-range trackers preserve correspondences without producing dense sequence-level reconstruction. This paper presents LongDPM, a novel overlap-aware framework for scalable long-range monocular dynamic reconstruction. First, LongDPM processes long videos in overlapping chunks, keeping inference memory bounded by the chunk length. Second, it connects chunk-local coordinate systems through confidence-weighted registration with static-aware overlap abstraction. Third, it associates dynamic identities across chunk boundaries and fuses matched trajectories to recover coherent long-range 3D motion. Experimental results demonstrate that LongDPM achieves superior long-range reconstruction and tracking performance, reducing dense tracking EPE over V-DPM on PointOdyssey, Kubric-F, and Kubric-G, while obtaining the best TUM-dynamics ATE for camera pose estimation.", "date": "2026-05-17", "categories": []}
{"id": "2605.17478", "title": "Mamba-VGGT: Persistent Long-Sequence Video Geometry Grounded Transformer via External Sliding Window Mamba Memory", "abstract": "Visual Geometry Grounded Transformers (VGGT) have set new benchmarks in high-fidelity 3D scene reconstruction. However, as the sequence length increases, these models suffer from catastrophic geometric forgetting and accumulation drift, primarily due to the quadratic complexity of global attention which necessitates truncated temporal windows. To overcome the resulting geometric drift, we present Mamba-VGGT, an enhanced VGGT framework capable of persistent long-range reasoning. Our key contribution is a Sliding Window Mamba (SWM) memory module that maintains an explicit external memory token across temporal windows. This module leverages selective state-space modeling to distill and propagate global geometric priors, effectively bypassing the memory constraints of traditional transformers. To integrate these long-term temporal cues without disrupting the highly optimized spatial features of the pre-trained VGGT, we propose a Zero-Init Spatial Memory Injector. Utilizing zero-convolutional layers, this injector adaptively fuses persistent memory into the patch token stream, ensuring structural stability and seamless feature alignment. Extensive experiments demonstrate that our approach significantly outperforms existing VGGT-based methods in maintaining spatial consistency and reducing trajectory accumulation errors. Our work provides a scalable, linear-complexity solution for geometry-grounded world modeling in extensive 3D environments.", "date": "2026-05-17", "categories": []}
{"id": "2605.18052", "title": "Efficient 3D Content Reconstruction and Generation", "abstract": "Automatic 3D content creation seeks to replace labor-intensive modeling and scanning pipelines with systems that can synthesize or recover 3D assets directly from text or images. Its applications span video games, virtual reality, robotics, and simulation, enabling rapid asset prototyping, diverse interactive world generation, and efficient 3D data collection for training foundation models. Contemporary solutions largely follow two complementary paradigms: (i) text- or image-to-3D generation, which learns priors over 3D geometry and appearance to create novel assets from natural language or a single view image; and (ii) 3D reconstruction, which estimates camera poses and geometry from RGB images. This thesis advances both directions. On the generation side, I introduce Instant3D, which combines multi-view diffusion with feed-forward sparse-view 3D reconstruction to produce high-quality assets in 5-20 seconds. On the reconstruction side, I develop FastMap, a structure-from-motion pipeline that achieves up to 10x speedup over prior state-of-the-art by using first-order optimization with fused GPU kernels extensively, while maintaining comparable pose accuracy and downstream novel view synthesis quality.", "date": "2026-05-18", "categories": []}
{"id": "2605.18252", "title": "GaussianZoom: Progressive Zoom-in Generative 3D Gaussian Splatting with Geometric and Semantic Guidance", "abstract": "We introduce GaussianZoom, a generative zoom-in 3D reconstruction system with an iterative progressive framework that combines geometry-consistent scene modeling and multi-scale semantic reasoning to enable high-fidelity extreme zoom-in rendering from low-resolution inputs. To achieve this, we develop a novel multi-view consistent super-resolution module with depth-based feature warping and VLM-driven detail synthesis, ensuring accurate multi-view correspondence while enriching fine-scale appearance beyond the observed resolution. To support zooming across large magnification ranges, we further introduce a new expandable continuous Level-of-Detail hierarchy that dynamically modulates Gaussian visibility for smooth, alias-free cross-scale rendering. Experiments on Mip-NeRF360 and Tanks\\&Temples demonstrate that GaussianZoom achieves superior perceptual quality, multi-view consistency, and robustness under extreme magnification, establishing a strong baseline for generative zoom-in 3D scene reconstruction.", "date": "2026-05-18", "categories": []}
{"id": "2605.18334", "title": "3D Skew Gaussian Splatting with Any Camera Trajectory Visualization Engine", "abstract": "While 3D Gaussian Splatting (3DGS) has revolutionized real-time photorealistic view synthesis, its fundamental reliance on symmetric Gaussian distributions introduces visual artifacts that hinder accurate spatial data exploration. Specifically, symmetric kernels struggle to capture shape and color discontinuities , which cause blurriness and primitive redundancy that mislead human perception during visual analysis. To address these visualization barriers, we introduce 3D Skew Gaussian Splatting (3DSGS), a novel framework that significantly enhances the structural fidelity and compactness of explicit scene representations. Our key insight lies in extending the standard primitive to a general Skew Gaussian counterpart. This generalized primitive inherits the highly efficient rasterization properties of standard Gaussians while gaining intrinsic asymmetric modeling capabilities. We couple this with an enhanced opacity representation to better handle complex transparency, alongside a depth-aware densification strategy that intelligently manages primitive allocation. Furthermore, to make these advancements actionable for real-world visual analytics, we re-derive the CUDA rasterization pipeline to universally support both symmetric and skew Gaussians, integrating it into a decoupled, free-camera interactive visualization engine. Extensive experiments demonstrate that 3DSGS achieves superior rendering quality and structural compactness, particularly in regions with intricate details, while maintaining the real-time frame rates necessary for fluid interactive exploration. Supplementary derivations and visual results are available at \\textbf{\\textit{https://3d-skew-gs.github.io/}}.", "date": "2026-05-18", "categories": []}
{"id": "2508.01014", "title": "Hestia: Voxel-Face-Aware Hierarchical Next-Best-View Acquisition for Efficient 3D Reconstruction", "abstract": "Advances in 3D reconstruction and novel view synthesis have enabled efficient and photorealistic rendering. However, images for reconstruction are still either largely manual or constrained by simple preplanned trajectories. To address this issue, recent works propose generalizable next-best-view planners that do not require online learning. Nevertheless, robustness and performance remain limited across various shapes. Hence, this study introduces Voxel-Face-Aware Hierarchical Next-Best-View Acquisition for Efficient 3D Reconstruction (Hestia), which addresses the shortcomings of the reinforcement learning-based generalizable approaches for five-degree-of-freedom viewpoint prediction. Hestia systematically improves the planners through four components: a more diverse dataset to promote robustness, a hierarchical structure to manage the high-dimensional continuous action search space, a close-greedy strategy to mitigate spurious correlations, and a face-aware design to avoid overlooking geometry. Experimental results show that Hestia achieves non-marginal improvements, with at least a 4% gain in coverage ratio, while reducing Chamfer Distance by 50% and maintaining real-time inference. In addition, Hestia outperforms prior methods by at least 12% in coverage ratio with a 5-image budget and remains robust to object placement variations. Finally, we demonstrate that Hestia, as a next-best-view planner, is feasible for the real-world application. Our project page is https://johnnylu305.github.io/hestia web.", "date": "2025-08-01", "categories": []}
{"id": "2605.13073", "title": "HarmoGS: Robust 3D Gaussian Splatting in the Wild via Conflict-Aware Gradient Harmonization", "abstract": "In-the-wild 3D Gaussian Splatting remains challenging due to transient distractors and illumination-induced cross-view appearance inconsistencies. Existing methods mainly rely on image-level masking to suppress unreliable supervision, but masking alone cannot fully eliminate residual occlusions or resolve illumination-induced inconsistencies, both of which can introduce conflicting cross-view gradients. These unresolved conflicts may destabilize Gaussian optimization and lead to visible reconstruction artifacts. We propose a conflict-aware 3DGS framework that addresses this problem from both image-space supervision and gradient-level optimization. Semantic Consistency-Guided Masking learns pixel-wise consistency scores to adaptively refine prior masks and suppress unreliable supervision before gradient formation. A dual-view Conflict-Aware Gradient Harmonization strategy further reconciles view-specific gradients by mutually rotating them into an orthogonal configuration, reducing negative directional interference across views. We also introduce conflict-aware densification and pruning to stabilize Gaussian growth and remove persistently conflicting primitives. Extensive experiments on standard in-the-wild benchmarks demonstrate that our method achieves state-of-the-art rendering quality under complex transient distractors and cross-view inconsistencies.", "date": "2026-05-13", "categories": []}
{"id": "2605.15010", "title": "3D Skew-Normal Splatting", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a leading representation for real-time novel view synthesis and has been widely adopted in various downstream applications. The core strength of 3DGS lies in its efficient kernel-based scene representation, where Gaussian primitives provide favorable mathematical and computational properties. However, under a finite primitive budget, the symmetric shape of each primitive directly affects representation compactness, especially near asymmetric structures such as object boundaries and one-sided surfaces. Recent works have explored more complex kernel distributions; however, they either remain within the elliptical family or rely on hard truncation, which limits continuous shape control and introduces distributional discontinuities. In this paper, we propose Skew-Normal Splatting (SNS), which adopts the Azzalini Skew-Normal distribution as the fundamental primitive. By introducing a learnable and bounded skewness parameter, SNS can continuously interpolate between symmetric Gaussians and Half-Gaussian-like shapes, enabling flexible modeling of both sharp boundaries and interior regions. Moreover, SNS preserves analytical tractability under affine transformations and marginalization. This property allows seamless integration into existing Gaussian Splatting rasterization pipelines. Furthermore, to address the strong coupling between scale, rotation, and skewness parameters, we introduce a decoupled parameterization and a block-wise optimization strategy to enhance training stability and accuracy. Extensive experiments on standard novel-view synthesis benchmarks show that SNS consistently improves reconstruction quality over Gaussian and recent non-Gaussian kernels, with clearer benefits on sharp boundaries and thin or one-sided structures.", "date": "2026-05-14", "categories": []}
{"id": "2605.15760", "title": "Learn2Splat: Extending the Horizon of Learned 3DGS Optimization", "abstract": "3D Gaussian Splatting (3DGS) optimization is most commonly performed using standard optimizers (Adam, SGD). While stable across diverse scenes, standard optimizers are general-purpose and not tailored to the structure of the problem. In particular, they produce independent parameter updates that do not capture the structural and spatial relationships within a scene, leading to inefficient optimization and slow convergence. Recent works introduced learned optimizers that predict correlated updates informed by inter-parameter and inter-Gaussian dependencies. However, these methods are trained for a fixed number of optimization iterations and rely on manually scheduled learning rates to avoid degradation. In this paper, we introduce a learned optimizer for 3DGS that avoids degradation over extended optimization horizons without auxiliary mechanisms. To enable this, we propose a meta-learning scheme that extends the optimization horizon via a checkpoint buffer and an optimizer rollout strategy, combined with an architecture that encodes gradient scale information in its latent states. Results show improved early novel view synthesis quality while remaining stable over long horizons, with zero-shot generalization to unseen reconstruction settings. To support our findings, we introduce the first unified framework for training and evaluating both learned and conventional optimizers across sparse and dense view settings. Code and models will be released publicly. Our project page is available at https://naamapearl.github.io/learn2splat .", "date": "2026-05-15", "categories": []}
{"id": "2605.13855", "title": "SparseOIT: Improving Order-Independent Transparency 3DGS via Active Set Method", "abstract": "3D Gaussian Splatting (3DGS) has received tremendous popularity over the past few years due to its photorealistic visual appearance. However, 3DGS uses volumetric rendering that is not suitable for objects with non-lambertian or transparent materials. To remedy this issue, a family of Order-Independent Transparency (OIT) rendering methods propose to remove or modify the depth sorting step in the 3DGS rendering equation. However, the potential of OIT-based method is still underexplored. In this paper, we observe that the OIT modifications to the rendering equation significantly reduce the inter-independence among individual gaussian splats, resulting in very sparse variable dependencies that can be harnessed by specific optimization techniques such as active set method. To this end, we propose SparseOIT, an OIT-based 3DGS reconstruction algorithm that maintains an active set of gaussian splats and enjoys an acceleration ratio that is proportional to the potential sparsity. SparseOIT is designed by jointly considering the OIT rendering equation, the reconstruction algorithm and the geometric regularization. Through extensive experiments, we demonstrate that SparseOIT outperforms existing methods in the OIT-family by a large margin and also achieves comparable performance to the state-of-the-art 3DGS reconstruction methods based on volumetric rendering. Project page:", "date": "2026-04-07", "categories": []}
{"id": "2605.14315", "title": "TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention", "abstract": "Recent feed-forward 3D reconstruction methods, such as visual geometry transformers, have substantially advanced the traditional per-scene optimization paradigm by enabling effective multi-view reconstruction in a single forward pass. However, most existing methods struggle to achieve a balance between reconstruction quality and computational efficiency, which limits their scalability and efficiency. Although some efficient visual geometry transformers have recently emerged, they typically use the same sparsity ratio across layers and frames and lack mechanisms to adaptively learn representative tokens to capture global relationships, leading to suboptimal performance. In this work, we propose TurboVGGT, a novel approach that employs an efficient visual geometry transformer with adaptive alternating attention for fast multi-view 3D reconstruction. Specifically, TurboVGGT employs an end-to-end trainable framework with adaptive sparse global attention guided by adaptive sparsity selection to capture global relationships across frames and frame attention to aggregate local details within each frame. In the adaptive sparse global attention, TurboVGGT adaptively learns representative tokens with varying sparsity levels for global geometry modeling, considering that token importance varies across frames, attention layers operate tokens at different levels of abstraction, and global dependencies rely on structurally informative regions. Extensive experiments on multiple 3D reconstruction benchmarks demonstrate that TurboVGGT achieves fast multi-view reconstruction while maintaining competitive reconstruction quality compared with state-of-the-art methods. Project page: https://turbovggt.github.io/.", "date": "2026-05-14", "categories": []}
{"id": "2605.14629", "title": "Efficient Dense Matching for Enhanced Gaussian Splatting Using AV1 Motion Vectors", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a prominent framework for real-time, photorealistic scene reconstruction, offering significant speed-ups over Neural Radiance Fields (NeRF). However, the fidelity of 3DGS representations remains heavily dependent on the quality of the initial point cloud. While standard Structure-from-Motion (SfM) pipelines using COLMAP provide adequate initialisation, they often suffer from high computational costs and sparsity in textureless regions, which degrades subsequent reconstruction accuracy and convergence speed. In this work, we introduce an AV1-based feature detection and matching pipeline that significantly reduces SfM processing overhead. By leveraging motion vectors inherent to the AV1 video codec, we bypass computationally expensive exhaustive matching while maintaining geometric robustness. Our pipeline produces substantially denser point clouds, with up to eight times as many points as classical SfM. We demonstrate that this enhanced initialisation directly improves 3DGS performance, yielding an 9-point increase in VMAF and a 63% average reduction in training time required to reach baseline quality. The project page: https://sigmedia.tv/AV1-3DGS.github.io/", "date": "2026-05-14", "categories": []}
{"id": "2605.15195", "title": "VGGT-$\\Omega$", "abstract": "Recent feed-forward reconstruction models, such as VGGT, have proven competitive with traditional optimization-based reconstructors while also providing geometry-aware features useful for other tasks. Here, we show that the quality of these models scales predictably with model and data size. We do so by introducing VGGT-$\\Omega$, which substantially improves reconstruction accuracy, efficiency, and capabilities for both static and dynamic scenes. To enable training this model at an unprecedented scale, we introduce architectural changes that improve training efficiency, a high-quality data annotation pipeline that supports dynamic scenes, and a self-supervised learning protocol. We simplify VGGT's architecture by using a single dense prediction head with multi-task supervision and removing the expensive high-resolution convolutional layers. We also use registers to aggregate scene information into a compact representation and introduce register attention, which restricts inter-frame information exchange to these registers, in part replacing global attention. In this way, during training, VGGT-$\\Omega$ uses only about 30% of the GPU memory of its predecessor, allowing us to train with 15x more supervised data than prior work and to leverage vast amounts of unlabeled video data. VGGT-$\\Omega$ achieves strong results for reconstruction of static and dynamic scenes across multiple benchmarks, for example, improving over the previous best camera estimation accuracy on Sintel by 77%. We also show that the learned registers can improve vision-language-action models and support alignment with language, suggesting that reconstruction can be a powerful and scalable proxy task for spatial understanding. Project Page: http://vggt-omega.github.io/", "date": "2026-05-14", "categories": []}
{"id": "2505.21238", "title": "3D-UIR: 3D Gaussian for Underwater 3D Scene Reconstruction via Physics Based Appearance-Medium Decoupling", "abstract": "Novel view synthesis for underwater scene reconstruction presents unique challenges due to complex light-media interactions. Optical scattering and absorption in water body bring inhomogeneous medium attenuation interference that disrupts conventional volume rendering assumptions of uniform propagation medium. While 3D Gaussian Splatting (3DGS) offers real-time rendering capabilities, it struggles with underwater inhomogeneous environments where scattering media introduces artifacts and inconsistent appearance. In this study, we propose a physics-based framework that disentangles object appearance from water medium effects through tailored Gaussian modeling. Our approach introduces appearance embeddings, which are explicit medium representations for backscatter and attenuation, enhancing scene consistency. In addition, we propose a depth-guided optimization strategy that leverages pseudo-depth maps as supervision with depth regularization and scale penalty terms to improve geometric fidelity. By integrating the proposed appearance and medium modeling components via an underwater imaging model, our approach achieves both high-quality novel view synthesis and physically accurate scene restoration. Experiments demonstrate our significant improvements in rendering quality and restoration accuracy over existing methods. The project page is available at https://bilityniu.github.io/3D-UIR.", "date": "2025-05-27", "categories": []}
{"id": "2508.09479", "title": "SkySplat: Generalizable 3D Gaussian Splatting from Multi-Temporal Sparse Satellite Images", "abstract": "Three-dimensional scene reconstruction from sparse-view satellite images is a long-standing and challenging task. While 3D Gaussian Splatting (3DGS) and its variants have recently attracted attention for its high efficiency, existing methods remain unsuitable for satellite images due to incompatibility with rational polynomial coefficient (RPC) models and limited generalization capability. Recent advances in generalizable 3DGS approaches show potential, but they perform poorly on multi-temporal sparse satellite images due to limited geometric constraints, transient objects, and radiometric inconsistencies. To address these limitations, we propose SkySplat, a novel self-supervised framework that integrates the RPC model into the generalizable 3DGS pipeline, enabling more effective use of sparse geometric cues for improved reconstruction. SkySplat relies only on RGB images and radiometric-robust relative height supervision, thereby eliminating the need for ground-truth height maps. Key components include a Cross-Self Consistency Module (CSCM), which mitigates transient object interference via consistency-based masking, and a multi-view consistency aggregation strategy that refines reconstruction results. Compared to per-scene optimization methods, SkySplat achieves an 86 times speedup over EOGS with higher accuracy. It also outperforms generalizable 3DGS baselines, reducing MAE from 13.18 m to 1.80 m on the DFC19 dataset significantly, and demonstrates strong cross-dataset generalization on the MVS3D benchmark. The is available at https://github.com/NanCheng2001/SkySplat-main", "date": "2025-08-13", "categories": []}
{"id": "2602.20150", "title": "Simulation-Ready Cluttered Scene Estimation via Physics-aware Joint Shape and Pose Optimization", "abstract": "Estimating simulation-ready scenes from real-world observations is crucial for downstream planning and policy learning tasks. Regretfully, existing methods struggle in cluttered environments, often exhibiting prohibitive computational cost, poor robustness, and restricted generality when scaling to multiple interacting objects. We propose a unified optimization-based formulation for real-to-sim scene estimation that jointly recovers the shapes and poses of multiple rigid objects under physical constraints. Our method is built on two key technical innovations. First, we leverage the recently introduced shape-differentiable contact model, whose global differentiability permits joint optimization over object geometry and pose while modeling inter-object contacts. Second, we exploit the structured sparsity of the augmented Lagrangian Hessian to derive an efficient linear system solver whose computational cost scales favorably with scene complexity. Building on this formulation, we develop an end-to-end Simulation-ready Physics-Aware Reconstruction for Cluttered Scenes (SPARCS) pipeline, which integrates learning-based object initialization, physics-constrained joint shape-pose optimization, and differentiable texture refinement. Experiments on cluttered scenes with up to 5 objects and 22 convex hulls demonstrate that our approach robustly reconstructs physically valid, simulation-ready object shapes and poses. Project webpage: https://rory-weicheng.github.io/SPARCS/.", "date": "2026-02-23", "categories": []}
{"id": "2605.12072", "title": "PairDropGS: Paired Dropout-Induced Consistency Regularization for Sparse-View Gaussian Splatting", "abstract": "Dropout-based sparse-view 3D Gaussian Splatting (3DGS) methods alleviate overfitting by randomly suppressing Gaussian primitives during training. Existing methods mainly focus on designing increasingly sophisticated dropout strategies, while they overlook the resulting inconsistencies among different dropped Gaussian subsets. This oversight often leads to unstable reconstruction and suboptimal Gaussian representation learning.In this paper, we revisit dropout-based sparse-view 3DGS from a consistency regularization perspective and propose PairDropGS, a Paired Dropout-induced Consistency Regularization framework for sparse-view Gaussian splatting. Specifically, PairDropGS first constructs a pair of the dropped Gaussian subsets from a shared Gaussian field and designs a low-frequency consistency regularization to constrain their low-frequency rendered structures. This design encourages the shared Gaussian field to preserve stable scene layout and coarse geometry under different random dropouts, while avoiding excessive constraints on ambiguous high-frequency details. Moreover, we introduce a progressive consistency scheduling strategy to gradually strengthen the consistency regularization during training for stability and robustness of reconstruction. Extensive experiments on widely-used sparse-view benchmarks demonstrate that PairDropGS achieves superior training stability, significantly outperforms existing dropout-based 3DGS methods in reconstruction quality, while exhibiting the simplicity and plug-and-play nature for improving dropout-based optimization.", "date": "2026-05-12", "categories": []}
{"id": "2605.13018", "title": "OCH3R: Object-Centric Holistic 3D Reconstruction", "abstract": "Object-centric scene understanding is a fundamental challenge in computer vision. Existing approaches often rely on multi-stage pipelines that first apply pre-trained segmentors to extract individual objects, followed by per-object 3D reconstruction. Such methods are computationally expensive, fragile to segmentation errors, and scale poorly with scene complexity. We introduce OCH3R, a unified framework for Object-Centric Holistic 3D Reconstruction from a single RGB image. OCH3R performs one forward pass to simultaneously predict all object instances with their 6D poses and detailed 3D reconstructions. The key idea is a transformer architecture that predicts per-pixel attributes, including CLIP-based category embeddings, metric depth, normalized object coordinates (NOCS), and a fixed number of 3D Gaussians representing each object. To supervise these Gaussian reconstructions, we transform them into canonical space using the predicted 6D poses and align them with pre-rendered canonical ground truth, avoiding costly per-image Gaussian label generation. On standard indoor benchmarks, OCH3R achieves state-of-the-art performance across monocular depth estimation, open-vocabulary semantic segmentation, and RGB-only category-level 6D pose estimation, while producing high-fidelity, editable per-object reconstructions. Crucially, inference is fully feed-forward and scales independently of the number of objects, offering orders-of-magnitude speedups over conventional multi-stage pipelines in cluttered scenes.", "date": "2026-05-13", "categories": []}
{"id": "2605.13741", "title": "LEXI-SG: Monocular 3D Scene Graph Mapping with Room-Guided Feed-Forward Reconstruction", "abstract": "Scene graphs are becoming a standard representation for robot navigation, providing hierarchical geometric and semantic scene understanding. However, most scene graph mapping methods rely on depth cameras or LiDAR sensors. In this work, we present LEXI-SG, the first dense monocular visual mapping system for open-vocabulary 3D scene graphs using only RGB camera input. Our approach exploits the semantic priors of open-vocabulary foundation models to partition the scene into rooms, deferring feed-forward reconstruction to when each room is fully observed -- enabling scalable dense mapping without sliding-window scale inconsistencies. We propose a room-based factor graph formulation to globally align room reconstructions while preserving local map consistency and naturally imposing the semantic scene graph hierarchy. Within each room, we further support open-vocabulary object segmentation and tracking. We validate LEXI-SG on indoor scenes from the Habitat-Matterport 3D and self-collected egocentric office sequences. We evaluate its performance against existing feed-forward SLAM methods, as well as established scene graphs baselines. We demonstrate improved trajectory estimation and dense reconstruction, as well as, competitive performance in open-vocabulary segmentation. LEXI-SG shows that accurate, scalable, open-vocabulary 3D scene graphs can be achieved from monocular RGB alone. Our project page and office sequences are available here: https://ori-drs.github.io/lexisg-web/.", "date": "2026-05-13", "categories": []}
{"id": "2605.13794", "title": "BlitzGS: City-Scale Gaussian Splatting at Lightning Speed", "abstract": "We present BlitzGS, a distributed 3DGS framework that reduces active Gaussian workload for fast city-scale reconstruction. BlitzGS manages this workload at three coupled levels. At the system level, the framework shards Gaussians across GPUs by index parity rather than spatial blocks. This approach mitigates the cross-block visibility redundancy inherent in spatial partitioning. Furthermore, it distributes each rendering step through a single cross-GPU exchange that routes projected Gaussians to their tile owners. At the model level, scheduled importance-scoring passes shrink the global Gaussian population. During these passes, the framework generates a per-Gaussian visibility weight to bias density-control updates toward contributing primitives and a per-view importance mask for the view-level renderer. At the view level, BlitzGS trims each camera's active set with a distance-based LOD gate to exclude excessively fine primitives for the current frustum and the importance-based culling mask to skip Gaussians with negligible cross-view contribution.\n  On large-scale benchmarks, BlitzGS matches the rendering quality of recent large-scale baselines while delivering an order-of-magnitude speedup, training city-scale scenes in tens of minutes. Our code is available at https: //github.com/AkierRaee/BlitzGS.", "date": "2026-05-13", "categories": []}
{"id": "2605.10360", "title": "DySurface: Consistent 4D Surface Reconstruction via Bridging Explicit Gaussians and Implicit Functions", "abstract": "While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) offer powerful dynamic scene rendering capabilities; however, relying solely on photometric optimization often leads to geometric ambiguities. This results in discontinuous surfaces, severe artifacts, and broken surfaces over time. To address these limitations, we present DySurface, a novel framework that bridges the effectiveness of explicit Gaussians with the geometric fidelity of implicit Signed Distance Functions (SDFs) in dynamic scenes. Our approach tackles the structural discrepancy between the forward deformation of 3DGS ($canonical \\rightarrow dynamic$) and the backward deformation required for volumetric SDF rendering ($dynamic \\rightarrow canonical$). Specifically, we propose the VoxGS-DSDF branch that leverages deformed Gaussians to construct a dynamic sparse voxel grid, providing explicit geometric guidance to the implicit SDF field. This explicit anchoring effectively regularizes the volumetric rendering process, significantly improving surface reconstruction quality, with watertight boundaries and detailed representations. Quantitative and qualitative experiments demonstrate that DySurface significantly outperforms state-of-the-art baselines in geometric accuracy metrics while maintaining competitive rendering performance.", "date": "2026-05-11", "categories": []}
{"id": "2605.11354", "title": "Lite3R: A Model-Agnostic Framework for Efficient Feed-Forward 3D Reconstruction", "abstract": "Transformer-based 3D reconstruction has emerged as a powerful paradigm for recovering geometry and appearance from multi-view observations, offering strong performance across challenging visual conditions. As these models scale to larger backbones and higher-resolution inputs, improving their efficiency becomes increasingly important for practical deployment. However, modern 3D transformer pipelines face two coupled challenges: dense multi-view attention creates substantial token-mixing overhead, and low-precision execution can destabilize geometry-sensitive representations and degrade depth, pose, and 3D consistency. To address the first challenge, we propose Lite3R, a model-agnostic teacher-student framework that replaces dense attention with Sparse Linear Attention to preserve important geometric interactions while reducing attention cost. To address the second challenge, we introduce a parameter-efficient FP8-aware quantization-aware training (FP8-aware QAT) strategy with partial attention distillation, which freezes the vast majority of pretrained backbone parameters and trains only lightweight linear-branch projection layers, enabling stable low-precision deployment while retaining pretrained geometric priors. We further evaluate Lite3R on two representative backbones, VGGT and DA3-Large, over BlendedMVS and DTU64, showing that it substantially reduces latency (1.7-2.0x) and memory usage (1.9-2.4x) while preserving competitive reconstruction quality overall. These results demonstrate that Lite3R provides an effective algorithm-system co-design approach for practical transformer-based 3D reconstruction. Code: https://github.com/AIGeeksGroup/Lite3R. Website: https://aigeeksgroup.github.io/Lite3R.", "date": "2026-05-12", "categories": []}
{"id": "2605.11424", "title": "VidSplat: Gaussian Splatting Reconstruction with Geometry-Guided Video Diffusion Priors", "abstract": "Gaussian Splatting has achieved remarkable progress in multi-view surface reconstruction, yet it exhibits notable degradation when only few views are available. Although recent efforts alleviate this issue by enhancing multi-view consistency to produce plausible surfaces, they struggle to infer unseen, occluded, or weakly constrained regions beyond the input coverage. To address this limitation, we present VidSplat, a training-free generative reconstruction framework that leverages powerful video diffusion priors to iteratively synthesize novel views that compensate for missing input coverage, and thereby recover complete 3D scenes from sparse inputs. Specifically, we tackle two key challenges that enable the effective integration of generation and reconstruction. First, for 3D consistent generation, we elaborate a training-free, stage-wise denoising strategy that adaptively guides the denoising direction toward the underlying geometry using the rendered RGB and mask images. Second, to enhance the reconstruction, we develop an iterative mechanism that samples camera trajectories, explores unobserved regions, synthesizes novel views, and supplements training through confidence weighted refinement. VidSplat performs robustly to sparse input and even a single image. Extensive experiments on widely used benchmarks demonstrate our superior performance in sparse-view scene reconstruction.", "date": "2026-05-12", "categories": []}
{"id": "2605.11594", "title": "PointForward: Feedforward Driving Reconstruction through Point-Aligned Representations", "abstract": "High-fidelity reconstruction of driving scenes is crucial for autonomous driving. While recent feedforward 3D Gaussian Splatting (3DGS) methods enable fast reconstruction, their per-pixel Gaussian prediction paradigm often suffers from multi-view inconsistency and layering artifacts. Moreover, existing methods often model dynamic instances via dense flow prediction, which lacks explicit cross-view correspondence and instance-level consistency. In this paper, we propose PointForward, a feedforward driving reconstruction framework through point-aligned representations. Unlike pixel-aligned methods, we initialize sparse 3D queries in world space and aggregate multi-view image information via spatial-temporal fusion onto these queries, enforcing explicit cross-view consistency in a single feedforward pass. To handle scene dynamics, we introduce scene graphs that explicitly organize moving instances during reconstruction. By leveraging 3D bounding boxes, our method enables instance-level motion propagation and temporally consistent dynamic representations. Extensive experiments demonstrate that PointForward achieves state-of-the-art performance on large-scale driving benchmarks. The code will be available upon the publication of the paper.", "date": "2026-05-12", "categories": []}
{"id": "2605.12027", "title": "4DVGGT-D: 4D Visual Geometry Transformer with Improved Dynamic Depth Estimation", "abstract": "Reconstructing dynamic 4D scenes from monocular videos is a fundamental yet challenging task. While recent 3D foundation models provide strong geometric priors, their performance significantly degrades in dynamic environments. This degradation stems from a fundamental tension: the inherent coupling of camera ego-motion and object motion within global attention mechanisms. In this paper, we propose a novel, training-free progressive decoupling framework that disentangles dynamics from statics in a principled, coarse-to-fine manner. Our core insight is to resolve the tension by first stabilizing the camera pose, followed by geometric refinement. Specifically, our approach consists of three synergistic components: (1) a Dynamic-Mask-Guided Pose Decoupling module that isolates pose estimation from dynamic interference, yielding a stable motion-free reference frame; (2) a Topological Subspace Surgery mechanism that orthogonally decomposes the depth manifold, safely preserving dynamic objects while injecting refined, mask-aware geometry into static regions; and (3) an Information-Theoretic Confidence-Aware Fusion strategy that formulates depth integration as a heteroscedastic Bayesian inference problem, adaptively blending multi-pass predictions via inverse-variance weighting. Extensive experiments on standard 4D reconstruction benchmarks demonstrate that our method achieves consistent and substantial improvements across principal point-cloud metrics. Notably, our approach shows competitive performance in robust 4D scene reconstruction without requiring fine-tuning, suggesting the potential of mathematically grounded dynamic-static disentanglement.", "date": "2026-05-12", "categories": []}
{"id": "2605.12399", "title": "GeoQuery: Geometry-Query Diffusion for Sparse-View Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a prominent paradigm for 3D reconstruction and novel view synthesis. However, it remains vulnerable to severe artifacts when trained under sparse-view constraints. While recent methods attempt to rectify artifacts in rendered views using image diffusion models, they typically rely on multi-view self-attention to retrieve information from reference images. We observe that this mechanism often fails when the rendered novel views output by 3DGS are heavily corrupted: damaged query features lead to erroneous cross-view retrieval, resulting in inconsistent rendering refinement. To address this, we propose GeoQuery, a geometry-guided diffusion framework that integrates generative priors with explicit geometric cues via a novel Geometry-guided Cross-view Attention (GCA) mechanism. First, by leveraging predicted depth maps and camera poses, we construct a geometry-induced correspondence field to sample reference features, forming a geometry-aligned proxy query that replaces the corrupted rendering features. Furthermore, we design a new cross-view feature aggregation pipeline, in which we restrict the cross-view attention to a local window around each proxy query to effectively retrieve useful features while suppressing spurious matches. GeoQuery can be seamlessly integrated into existing diffusion-based pipelines, enabling robust reconstruction even under extreme view sparsity. Extensive experiments on sparse-view novel view synthesis and rendering artifact removal demonstrate the effectiveness of our approach.", "date": "2026-05-12", "categories": []}
{"id": "2605.12494", "title": "Revisiting Photometric Ambiguity for Accurate Gaussian-Splatting Surface Reconstruction", "abstract": "Surface reconstruction with differentiable rendering has achieved impressive performance in recent years, yet the pervasive photometric ambiguities have strictly bottlenecked existing approaches. This paper presents AmbiSuR, a framework that explores an intrinsic solution upon Gaussian Splatting for the photometric ambiguity-robust surface 3D reconstruction with high performance. Starting by revisiting the foundation, our investigation uncovers two built-in primitive-wise ambiguities in representation, while revealing an intrinsic potential for ambiguity self-indication in Gaussian Splatting. Stemming from these, a photometric disambiguation is first introduced, constraining ill-posed geometry solution for definite surface formation. Then, we propose an ambiguity indication module that unleashes the self-indication potential to identify and further guide correcting underconstrained reconstructions. Extensive experiments demonstrate our superior surface reconstructions compared to existing methods across various challenging scenarios, excelling in broad compatibility. Project: https://fictionarry.github.io/AmbiSuR-Proj/ .", "date": "2026-05-12", "categories": []}
{"id": "2603.11969", "title": "AstroSplat: Physics-Based Gaussian Splatting for Rendering and Reconstruction of Small Celestial Bodies", "abstract": "Image-based surface reconstruction and characterization are crucial for missions to small celestial bodies (e.g., asteroids), as it informs mission planning, navigation, and scientific analysis. Recent advances in Gaussian splatting enable high-fidelity neural scene representations but typically rely on a spherical harmonic intensity parameterization that is strictly appearance-based and does not explicitly model material properties or light-surface interactions. We introduce AstroSplat, a physics-based Gaussian splatting framework that integrates planetary reflectance models to improve the autonomous reconstruction and photometric characterization of small-body surfaces from in-situ imagery. The proposed framework is validated on real imagery taken by NASA's Dawn mission, where we demonstrate superior rendering performance and surface reconstruction accuracy compared to the typical spherical harmonic parameterization.", "date": "2026-03-12", "categories": []}
{"id": "2604.12592", "title": "ELoG-GS: Dual-Branch Gaussian Splatting with Luminance-Guided Enhancement for Extreme Low-light 3D Reconstruction", "abstract": "This paper presents our approach to the NTIRE 2026 3D Restoration and Reconstruction Challenge (Track 1), which focuses on reconstructing high-quality 3D representations from degraded multi-view inputs. The challenge involves recovering geometrically consistent and photorealistic 3D scenes in extreme low-light environments. To address this task, we propose Extreme Low-light Optimized Gaussian Splatting (ELoG-GS), a robust low-light 3D reconstruction pipeline that integrates learning-based point cloud initialization and luminance-guided color enhancement for stable and photorealistic Gaussian Splatting. Our method incorporates both geometry-aware initialization and photometric adaptation strategies to improve reconstruction fidelity under challenging conditions. Extensive experiments on the NTIRE Track 1 benchmark demonstrate that our approach significantly improves reconstruction quality over the baselines, achieving superior visual fidelity and geometric consistency. The proposed method provides a practical solution for robust 3D reconstruction in real-world degraded scenarios. In the final testing phase, our method achieved a PSNR of 18.6626 and an SSIM of 0.6855 on the official platform leaderboard. Code is available at https://github.com/lyh120/FSGS_EAPGS.", "date": "2026-04-14", "categories": []}
{"id": "2604.19923", "title": "UniCon3R: Unified Contact-aware 4D Human-Scene Reconstruction from Monocular Video", "abstract": "We introduce UniCon3R, a unified feed-forward framework for online human-scene 4D reconstruction from monocular video. Current feed-forward human-scene reconstruction methods suffer from artifacts, where bodies float above the ground or penetrate parts of the scene. A key reason is the lack of effective interaction modelling between the human and the environment. Our goal is to exploit contact between the human and the scene during inference to actively improve the human mesh reconstruction. To that end, we explicitly model interaction by inferring 4D contact from the human pose and scene geometry and use the contact as a corrective cue for generating the pose. This enables UniCon3R to jointly recover scene geometry and spatially aligned 4D humans within the scene. Experiments on standard human-centric video benchmarks show that UniCon3R outperforms state-of-the-art baselines on physical plausibility and global human motion estimation while preserving fast, feed-forward inference speeds. The results validate our central claim: contact serves as a powerful internal prior, thus establishing a new paradigm for physically grounded joint human-scene reconstruction. Project page is available at https://surtantheta.github.io/UniCon3R .", "date": "2026-04-21", "categories": []}
{"id": "2605.00408", "title": "Beyond Heuristics: Learnable Density Control for 3D Gaussian Splatting", "abstract": "While 3D Gaussian Splatting (3DGS) has demonstrated impressive real-time rendering performance, its efficacy remains constrained by a reliance on heuristic density control. Despite numerous refinements to these handcrafted rules, such methods inherently lack the flexibility to adapt to diverse scenes with complex geometries.\n  In this paper, we propose a paradigm shift for density control from rigid heuristics to fully learnable policies. Specifically, we introduce \\textbf{LeGS}, a framework that reformulates density control as a parameterized policy network optimized via Reinforcement Learning (RL). Central to our approach is the tailored effective reward function grounded in sensitivity analysis, which precisely quantifies the marginal contribution of individual Gaussians to reconstruction quality. To maintain computational tractability, we derive a closed-form solution that reduces the complexity of reward calculation from $O(N^2)$ to $O(N)$. Extensive experiments on the Mip-NeRF 360, Tanks \\& Temples, and Deep Blending datasets demonstrate that \\textbf{LeGS} significantly outperforms state-of-the-art methods, striking a superior balance between reconstruction quality and efficiency. The code will be released at https://github.com/AaronNZH/LeGS", "date": "2026-05-01", "categories": []}
{"id": "2605.07910", "title": "One World, Dual Timeline: Decoupled Spatio-Temporal Gaussian Scene Graph for 4D Cooperative Driving Reconstruction", "abstract": "Reconstructing dynamic scenes from Vehicle-to-Infrastructure Cooperative Autonomous Driving (VICAD) data is fundamentally complicated by temporal asynchrony: vehicle and infrastructure cameras operate on independent clocks, capturing the same dynamic agent such as cars and pedestrians at different physical times. Existing Gaussian Scene Graph methods implicitly assume synchronized observations and assign a single pose per agent per frame, which is an assumption that breaks in cooperative settings, where the resulting gradient conflicts cause severe ghosting on dynamic agents. We identify this as a representation-level failure, not an optimization artifact: we prove that any single-timeline formulation incurs an irreducible photometric loss scaling quadratically with agent velocity and cross-source time offset. To resolve this, we propose Dust (DecoUpled Spatio-Temporal) Gaussian Scene Graph for 4D Cooperative Driving Reconstruction. DUST Gaussian Scene Graph shares a canonical Gaussian set per agent for appearance consistency, while maintaining decouple pose trajectories aligned to each source's true capture timestamps. We prove that this decoupling enables the pose-gradient kernel block-diagonal, eliminating cross-source interference entirely. To make Dust practical, we further introduce a static anchor-based pose correction pipeline that corrects spatio misalignment between vehicle and infrastructure annotations, and a pose-regularized joint optimization scheme that prevents trajectory jitter and drift during early training. On 26 sequences from V2X-Seq, DUST achieves state-of-the-art performance, improving dynamic-area PSNR by 3.2 dB over the strongest baseline and reducing Fr\\'echet Video Distance by 37.7%, with keeping robustness under larger temporal asynchrony.", "date": "2026-05-08", "categories": []}
{"id": "2605.08635", "title": "Kinematics-Driven Gaussian Shape Deformation for Blurry Monocular Dynamic Scenes", "abstract": "Reconstructing dynamic 3D scenes from blurry monocular videos is challenging as motion-induced blur entangles object motion and geometry, hindering geometric consistency. We present Kinematics-GS, a kinematics-aware framework that models blur as motion-aligned deformation and introduces a kinematic prior to reparameterize Gaussian shapes along motion trajectories, thereby mitigating degenerate shape collapse without auxiliary motion supervision. To stabilize optimization, we decompose scenes into dynamic and static components using temporal deformation variance and employ a coarse-to-fine deformation strategy to capture both global motion and fine-grained details. We also introduce a challenging real-world dataset of deformable and elastic objects exhibiting non-rigid motion with spatially non-uniform motion blur that obscures geometric cues. Extensive experiments on real-world benchmarks with realistic motion blur demonstrate that Kinematics-GS outperforms prior methods by a clear margin in monocular dynamic scene reconstruction, highlighting its effectiveness in handling complex and non-rigid motion scenarios.", "date": "2026-05-09", "categories": []}
{"id": "2605.08739", "title": "ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting", "abstract": "A converged 3D Gaussian Splatting (3DGS) model may approximate the target scene while remaining poorly parameterized for further optimization. We identify this failure mode as \\emph{parameterization degeneration}: high-opacity floaters attenuate gradients to true surfaces through alpha compositing, and redundant overlapping clusters create strongly coupled parameter blocks with nearly collinear Jacobian responses. These effects explain why continued optimization can plateau even when the model still contains removable artifacts. We propose ReorgGS, an equivalent distribution reorganization method for converged 3DGS models. ReorgGS treats the existing Gaussian set as an empirical probability field, resamples centers from it, estimates local anisotropic covariances with kNN, initializes low opacity, and continues optimization with the original 3DGS renderer and loss. Unlike opacity reset, which only rescales opacity on the old overlap graph, ReorgGS rebuilds centers, covariances, and visibility structure, thereby changing the graph itself. Our analysis shows that distributional equivalence is not optimization equivalence. The reorganized model preserves scene support while improving gradient accessibility under alpha compositing and reducing opacity-weighted overlap, thereby weakening local parameter coupling during subsequent optimization. Under the same additional optimization budget, ReorgGS improves fitting quality at a fixed Gaussian count, suppresses persistent floaters, and reduces rendering overhead from redundant overlap.", "date": "2026-05-09", "categories": []}
{"id": "2605.09024", "title": "Relightable Gaussian Splatting for Virtual Production Using Image-Based Illumination", "abstract": "Virtual production (VP) use LED walls to provide both background imagery and image-based lighting. While this enables on-set compositing, it couples lighting to background and scene appearance, limiting flexibility for downstream editing. In addition, inverse rendering conventionally relies on physically-based rendering to estimates 3D geometry and lighting, using environment maps. However, these maps are typically low-resolution and assume far-field lighting. In VP, with near-field and high-resolution image-based lighting, this can lead to inaccuracies and introduce complexities when editing. Addressing this, we propose a VP-specific framework for 3D reconstruction and relighting using Gaussian Splatting. This uses the known background imagery to condition the relighting process. This avoids relying on environment maps and reduces compositing to a background-image editing task. To realize our framework, we introduce a process (and associated dataset) that captures real VP scenes under varying background content and illumination conditions. This data is used to decompose a 3D scene into fixed appearance and variable lighting components. The variable lighting process simulates light transport by parameterizing each primitive with a UV coordinate, intensity value and resolution modifier. Using mipmaps, these directly sample the background texture in image space - implicitly capturing reflections and refractions without physically-based rendering. Combined with the fixed appearance component, this allows us to render relit scenes using a Gaussian Splatting rasterizer. Compared to baselines, our approach achieves higher-quality 3D reconstruction and controllable relighting. The method is efficient (<3 GB RAM, <5 GB VRAM, <2 hours training, ~35 FPS) and supports rendering useful arbitrary output variables including depth, lighting intensity, lighting color, and unlit renders.", "date": "2026-05-09", "categories": []}
{"id": "2605.09039", "title": "SeasonScapes: Learning Large-scale Re-lightable 3D Landscapes with Seasonal Variation from Sparse Webcams", "abstract": "We introduce SeasonScapes framework and a the SeasonScapes dataset: Swiss Sparse-view Mountain Scenes with Seasonal Changes that covers over 50 km x 60 km, composed of more than 85,000 webcam images captured from 32 different locations across 13 timestamps throughout a full year. By projecting these timestamp-specific images onto a 3D mesh, we construct seasonal 3D landscapes that reflect natural appearance changes over time. To address occlusions and missing data, we leverage conditional diffusion models for image-guided inpainting directly on the mesh. The resulting completed meshes can be further relighted using standard physically-based renderer.", "date": "2026-05-09", "categories": []}
{"id": "2605.09644", "title": "Attention Itself Could Retrieve.RetrieveVGGT: Training-Free Long Context Streaming 3D Reconstruction via Query-Key Similarity Retrieval", "abstract": "Visual Geometry Grounded Transformer (VGGT) advances 3D reconstruction via scalable Transformer architecture, but the quadratic complexity of global attention prevents long context application. StreamVGGT enables streaming with causal attention, yet its KV cache grows linearly with frames, causing memory overflow and quality degradation. We present RetrieveVGGT, a training-free framework, which formulates context construction for VGGT as a retrieval problem. By retrieving a fixed number of relevant frames at each step, VGGT maintains a controllable memory budget, which is close to its training context length. Interestingly, we find that the similarity between current frame queries and cached history frame keys at the first global attention layer of VGGT is already a strong indicator of relevance, eliminating the need for additional learned scoring. To enhance information diversity similar to a recommender system, we propose Segment Sampling so that the retrieval spans distinct relevant segments rather than a single high-similarity region. We design a pose-aware spatial memory mechanism that organizes history frames according to their already estimated camera poses, enabling location-aware retrieval. Extensive experiments demonstrate that RetrieveVGGT achieves state-of-the-art performance, outperforming StreamVGGT, TTT3R, and InfiniteVGGT while maintaining constant memory usage regardless of sequence length. Code is available at https://github.com/zzctmd/RetrieveVGGT.", "date": "2026-05-10", "categories": []}
{"id": "2605.09688", "title": "ConFixGS: Learning to Fix Feedforward 3D Gaussian Splatting with Confidence-Aware Diffusion Priors in Driving Scenes", "abstract": "Feedforward 3D Gaussian Splatting (3DGS) often struggles in trajectory-based sparse-view driving scenes. Existing Gaussian repair methods mainly target optimization-based 3DGS, while diffusion-based repair is typically restricted to iterative refinement near observed viewpoints, leaving feedforward 3DGS repair underexplored. We propose ConFixGS, a plug-and-play method that learns to fix feedforward 3DGS with confidence-aware diffusion priors. Starting from a pretrained feedforward model, ConFixGS generates diffusion-enhanced local pseudo-targets and validates them through reprojection-based cross-checking against support views. The resulting dense confidence maps guide refinement, enhancing reliable details while suppressing hallucinated or inconsistent evidence. On Waymo, nuScenes, and KITTI, ConFixGS improves challenging novel view synthesis, with PSNR gains of up to 3.68 dB and FID reduced by nearly half. Our results highlight confidence-aware fusion of generative priors and support-view consistency as a key principle for robust feedforward 3D driving scene reconstruction.", "date": "2026-05-10", "categories": []}
{"id": "2605.10307", "title": "PaMoSplat: Part-Aware Motion-Guided Gaussian Splatting for Dynamic Scene Reconstruction", "abstract": "Dynamic scene reconstruction represents a fundamental yet demanding challenge in computer vision and robotics. While recent progress in 3DGS-based methods has advanced dynamic scene modeling, obtaining high-fidelity rendering and accurate tracking in scenarios with substantial, intricate motions remains significantly challenging. To address these challenges, we propose PaMoSplat, a novel dynamic Gaussian splatting framework incorporating part awareness and motion priors. Our approach is grounded in two key observations: 1) Parts serve as primitives for scene deformation, and 2) Motion cues from optical flow can effectively guide part motion. Specifically, PaMoSplat initializes by lifting multi-view segmentation masks into 3D space via graph clustering, establishing coherent Gaussian parts. For subsequent timestamps, we leverage a differential evolutionary algorithm to estimate the rigid motion of these parts using multi-view optical flow cues, providing a robust warm-start for further optimization. Additionally, PaMoSplat introduces an adaptive iteration count mechanism, internal learnable rigidity, and flow-supervised rendering loss to accelerate and optimize the training process. Comprehensive evaluations across diverse scenes, including real-world environments, demonstrate that PaMoSplat delivers superior rendering quality, improved tracking precision, and faster convergence compared to existing methods. Furthermore, it enables multiple part-level downstream applications, such as 4D scene editing.", "date": "2026-05-11", "categories": []}
{"id": "2605.10567", "title": "VeloGauss: Learning Physically Consistent Gaussian Velocity Fields from Videos", "abstract": "In this paper, we aim to jointly model the geometry, appearance, and physical information of 3D scenes solely from dynamic multi-view videos, without relying on any physical priors. Existing works typically employ physical losses merely as soft constraints or integrate physical simulations into neural networks; however, these approaches often fail to effectively learn complex motion physics. Although modeling velocity fields holds the potential to capture authentic physical information, due to the lack of appropriate physical constraints, current methods are unable to correctly learn the interaction mechanisms between rigid and non-rigid particles. To address this, we propose VeloGauss, designed to learn the physical properties of complex dynamic 3D scenes without physical priors. Our method learns the velocity field for each Gaussian particle by introducing a Physics Code and a Particle Dynamics System, and ultimately incorporates Global Physical Constraints to ensure the physical consistency of the scene. Extensive experiments on four public datasets demonstrate that our method outperforms achieves state-of-the-art performance in both Novel View Interpolation and Future Frame Extrapolation tasks.", "date": "2026-05-11", "categories": []}
{"id": "2605.10705", "title": "TransmissiveGS: Residual-Guided Disentangled Gaussian Splatting for Transmissive Scene Reconstruction and Rendering", "abstract": "Transmissive scenes are ubiquitous in daily life, yet reconstructing and rendering them remains highly challenging due to the inherent entanglement between near-field reflections from the surrounding environment on the transmissive surface, and the transmitted content of the scene behind it. This coupling gives rise to dual surface geometries and dual radiance components within each observation, posing ambiguities for standard methods. We present TransmissiveGS, a novel framework for disentangled reconstruction and rendering of transmissive scenes. Specifically, we model the scene with a dual-Gaussian representation and introduce a deferred shading function to jointly render the two Gaussian components. To separate reflection and transmission, we exploit the inherent multi-view inconsistency of reflections and leverage the residuals from reconstructing multi-view consistent content as cues for disentangled geometry and appearance modeling. We further propose a reflection light field that enables high-fidelity estimation of near-field reflections. During training, we introduce a high-frequency regularization to preserve fine details. We also contribute a new synthetic dataset for evaluating transmissive surface reconstruction. Experiments on both synthetic and real-world scenes demonstrate that TransmissiveGS consistently outperforms prior Gaussian Splatting-based methods in both reconstruction and rendering quality for transmissive scenes.", "date": "2026-05-11", "categories": []}
{"id": "2605.10715", "title": "UAV-Assisted Scan-to-Simulation for Landslides Using Physics-Informed Gaussian Splatting", "abstract": "Landslide monitoring and simulation play an important role in urban safety assessment and disaster prevention. Existing landslide simulation pipelines typically rely on digital elevation model and mesh-based representations, which are suitable for geometric analysis, but often lack visual realism. This limitation reduces their effectiveness in interactive applications, hazard communication, and public education. In this paper, we propose a UAV-based scan-to-simulation framework that bridges photorealistic scene capture and physics-based landslide simulation through 3DGS. Specifically, our pipeline includes four stages: (1) UAV-based acquisition of slope imagery, (2) reconstruction of a low-anisotropy 3DGS scene representation, (3) volumetric conversion of the target simulation region by filling the interior of the surface-based model, and (4) integration with the Material Point Method (MPM) for landslide simulation. We validate the proposed framework on a real landslide site in Hong Kong that experienced a severe landslide event. The results show that our method supports both realistic visual reconstruction and effective simulation.", "date": "2026-05-11", "categories": []}
{"id": "2605.06876", "title": "AdpSplit: Error-Driven Adaptive Splitting for Faster Geometry Discovery in 3D Gaussian Splatting", "abstract": "Adaptive density control in 3D Gaussian Splatting (3DGS) repeatedly grows the Gaussian population through fixed-cardinality random splitting to discover useful scene structure. However, in vanilla 3DGS, its binary split operator requires many densification rounds to expose fine details, making it a bottleneck for efficient training schedules with fewer iterations. We introduce AdpSplit, an error-driven adaptive split operator that determines the number of split children and initializes the child parameters from L1-pixel-error region statistics, enabling fewer densification iterations, thus reduced training time, while preserving the rendering quality of full-schedule training. Across the MipNeRF360, Deep-Blending, and Tanks&Temples datasets, AdpSplit reduces the training time of multiple accelerated 3DGS pipelines by 9.2%-22.3% as a simple drop-in replacement for the standard split operator. With FastGS, AdpSplit matches the full-schedule PSNR on MipNeRF360 while reducing training time by 16.4%, corresponding to a 12.6x acceleration over vanilla 3DGS.", "date": "2026-05-07", "categories": []}
{"id": "2605.07181", "title": "SatSurfGS: Generalizable 2D Gaussian Splatting for Sparse-View Satellite Surface Reconstruction", "abstract": "Sparse-view satellite image surface reconstruction remains highly challenging, fundamentally because the reliability of multi-view matching under satellite imaging conditions is strongly spatially heterogeneous. Affected by large photometric differences, weak textures, and repetitive textures, multi-view geometric constraints are often sparse, unevenly distributed, and locally unreliable. Although 2D Gaussian Splatting (2DGS) is more suitable than 3D Gaussian Splatting (3DGS) for the explicit representation of continuous surfaces, research on generalizable feed-forward 2DGS frameworks for sparse-view satellite surface reconstruction is still lacking. To address this issue, we propose SatSurfGS, a generalizable sparse-view surface reconstruction method for satellite imagery based on 2DGS. The proposed method builds a coarse-to-fine Gaussian attribute prediction framework and explicitly models local geometric reliability at three levels: feature learning, Gaussian parameter estimation, and training optimization. Specifically, we propose a confidence-aware monocular multi-view feature fusion module to adaptively integrate monocular priors and multi-view matching features according to local confidence; a cross-stage self-consistency residual guidance module to stabilize stage-wise Gaussian parameter refinement using the residual between the rendered height map from the previous stage and the current-stage MVS height map, together with confidence information; and a confidence bidirectional routing loss to achieve differentiated allocation of geometric and appearance supervision. Experiments on satellite datasets show that the proposed method achieves improved rendering quality, surface reconstruction accuracy, cross-dataset generalization, and inference efficiency compared with representative generalizable baselines and competitive per-scene optimization methods.", "date": "2026-05-08", "categories": []}
{"id": "2605.07192", "title": "AsyncEvGS: Asynchronous Event-Assisted Gaussian Splatting for Handheld Motion-Blurred Scenes", "abstract": "3D reconstruction methods such as 3D Gaussian Splatting (3DGS) and Neural Radiance Fields (NeRF) achieve impressive photorealism but fail when input images suffer from severe motion blur. While event cameras provide high-temporal-resolution motion cues, existing event-assisted approaches rely on low-resolution sensors and strict synchronization, limiting their practicality for handheld 3D capture on common devices, such as smartphones. We introduce a flexible, high-resolution asynchronous RGB-Event dual-camera system and a corresponding reconstruction framework. Our approach first reconstructs sharp images from the event data and then employs a cross-domain pose estimation module based on the Visual Geometry Transformer (VGGT) to obtain robust initialization for 3DGS. During optimization, we employ a structure-driven event loss and view-specific consistency regularizers to mitigate the ill-posed behavior of traditional event losses and deblurring losses, ensuring both stable and high-fidelity reconstruction. We further contribute AsyncEv-Deblur, a new high-resolution RGB-Event dataset captured with our asynchronous system. Experiments demonstrate that our method achieves state-of-the-art performance on both our challenging dataset and existing benchmarks, substantially improving reconstruction robustness under severe motion blur. Project page: https://openimaginglab.github.io/AsyncEvGS/", "date": "2026-05-08", "categories": []}
{"id": "2605.07254", "title": "High-Fidelity Surface Splatting-Based 3D Reconstruction from Multi-View Images", "abstract": "Multi-view mesh reconstruction remains a core challenge in computer graphics and vision, especially for recovering high-frequency geometry from sparse observations. Recent methods such as 3D Gaussian Splatting (3DGS) and Neural Radiance Fields (NeRF) rely on post-processing for mesh extraction, thereby limiting joint optimization of geometry and appearance. Implicit Moving Least Squares (IMLS) instead enables direct conversion of point clouds into signed distance and texture fields, supporting end-to-end reconstruction and rendering. However, existing IMLS formulations use exponential kernels that struggle with high-frequency detail. We introduce a compact polynomial kernel with local support and greater flexibility, allowing better control over frequency content and improved geometric fidelity. To further enhance fine details, we incorporate stochastic regularization with Laplacian filtering. Together, these improve the preservation of high-frequency structure while maintaining stable optimization. Experiments show state-of-the-art performance in both surface reconstruction and rendering, yielding more accurate geometry and sharper visuals from multi-view data.", "date": "2026-05-08", "categories": []}
{"id": "2605.07550", "title": "Mind the Gap: Geometrically Accurate Generative Reconstruction from Disjoint Views", "abstract": "3D vision systems are fundamentally constrained by their reliance on visual overlap: reconstruction methods require it for geometric alignment, while generative models use it to enforce multi-view consistency. This limitation is particularly acute in real-world scenarios such as distributed swarm robotics or crowd-sourced data collection, where capturing overlapping perspectives, both in terms of spatial and appearance overlap, is often impossible. We introduce Generative Reconstruction from Disjoint Views as a new paradigm, establish a comprehensive dataset, and propose specialized evaluation metrics for zero-overlap scenarios. Our benchmarking demonstrates that existing state-of-the-art methods fail catastrophically on this task, producing disconnected geometries or semantically incoherent reconstructions. To address these limitations, we propose GLADOS, a general, modular framework that operates through three stages: (1) Generative Bridging, where foundation models synthesize intermediate perspectives to connect disjoint inputs; (2) Robust Coarse 3D Reconstruction, that establish coarse geometric scaffold via global alignment which absorbs local contradictions from generative process; and (3) Iterative Context Expansion and Consistency Optimization to fill missing regions and unify the reconstruction. As an architectureagnostic framework, GLADOS enables seamless integration of future advances in generation, reconstruction, and inpainting. The source code is available at: https://github.com/gwilczynski95/GLADOS.", "date": "2026-05-08", "categories": []}
{"id": "2605.07978", "title": "Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images", "abstract": "Cross-view localization classically asks: where does this ground image lie on the satellite tile? Existing methods are typically limited to 3-DoF estimates -- an $(x,y)$ position and a yaw angle -- because nadir satellite imagery provides no direct cues for roll, pitch, or altitude, forcing a reliance on planar-motion and zero-tilt assumptions. These assumptions break on real terrain with slopes, ramps, and tilted camera mounts. To overcome this, we introduce a single UAV image as an intermediate viewpoint: it reveals the 3D structure invisible from nadir, supplies the cues for roll, pitch, and altitude that the satellite alone cannot provide, and needs only spatial overlap with the ground camera -- no known relative pose is required. Building on this insight, we propose **Cross3R**, a flexible feed-forward model that ingests a satellite tile together with a UAV image, a ground image, or both, and, in a single forward pass, recovers a cross-view 3D point cloud, the 6-DoF poses of every input camera, and the on-tile $(x,y)$ position and yaw of each perspective camera. For training and evaluation, we also construct **CrossGeo**, a 278K-image tri-view dataset spanning 85 scenes across every continent except Antarctica. On CrossGeo, Cross3R consistently outperforms feed-forward 3D baselines in point-cloud reconstruction, 6-DoF camera-pose estimation, and cross-view localization. On KITTI, it outperforms dedicated cross-view methods trained on KITTI on most metrics, despite having no KITTI training itself.", "date": "2026-05-08", "categories": []}
{"id": "2605.05664", "title": "Sparse-to-Complete: From Sparse Image Captures to Complete 3D Scenes", "abstract": "We introduce S2C-3D, a novel sparse-view 3D reconstruction framework for high-fidelity and complete scene reconstruction from as few as six to eight images. Our framework features three components: a specialized diffusion model for scene-specific image restoration, a training-free view-consistency conditioned sampling process in the diffusion model for refined Gaussian optimization, and a camera trajectory planning scheme to ensure comprehensive scene coverage. The specialized diffusion model is developed by finetuning a pretrained architecture on the input views and their corresponding degraded counterparts. The adaptation to the scene distribution allows the model to repair Gaussian renderings while effectively eliminating domain gaps. Meanwhile, the trajectory planning scheme optimizes scene coverage by connecting each newly sampled camera to its two nearest neighbors. By iteratively constructing paths and retaining only those that significantly enhance visibility, the scheme establishes a trajectory that covers the entire scene. To address multi-view conflicts, the view-consistency conditioned sampling process quantifies the consistency between neighboring repaired images. This information is injected as a condition into the sampling process of the frozen diffusion model, facilitating the generation of view-consistent images without additional training. Consequently, our approach produces high-fidelity 3D Gaussians that are robust to artifacts. Experimental results demonstrate that S2C-3D outperforms state-of-the-art methods, constructing high-quality scenes that are free from missing regions, blurring, or other artifacts with very sparse inputs. The source code and data are available at https://gapszju.github.io/S2C-3D.", "date": "2026-05-07", "categories": []}
{"id": "2510.09881", "title": "LTGS: Long-Term Gaussian Scene Chronology From Sparse View Updates", "abstract": "Recent advances in novel-view synthesis can create the photo-realistic visualization of real-world environments from conventional camera captures. However, the everyday environment experiences frequent scene changes, which require dense observations, both spatially and temporally, that an ordinary setup cannot cover. We propose long-term Gaussian scene chronology from sparse-view updates, coined LTGS, an efficient scene representation that can embrace everyday changes from highly under-constrained casual captures. Given an incomplete and unstructured 3D Gaussian Splatting (3DGS) representation obtained from an initial set of input images, we robustly model the long-term chronology of the scene despite abrupt movements and subtle environmental variations. We construct objects as template Gaussians, which serve as structural, reusable priors for shared object tracks. Then, the object templates undergo a further refinement pipeline that modulates the priors to adapt to temporally varying environments given few-shot observations. Once trained, our framework is generalizable across multiple time steps through simple transformations, significantly enhancing the scalability for a temporal evolution of 3D environments. As existing datasets do not explicitly represent the long-term real-world changes with a sparse capture setup, we collect real-world datasets to evaluate the practicality of our pipeline. Experiments demonstrate that our framework achieves superior reconstruction quality compared to other baselines while enabling fast and light-weight updates. Project page is available at: https://mkjjang3598.github.io/LTGS.", "date": "2025-10-10", "categories": []}
{"id": "2601.00285", "title": "SV-GS: Sparse View 4D Reconstruction with Skeleton-Driven Gaussian Splatting", "abstract": "Reconstructing a dynamic target moving over a large area is challenging. Standard approaches for dynamic object reconstruction require dense coverage in both the viewing space and the temporal dimension, typically relying on multi-view videos captured at each time step. However, such setups are only possible in constrained environments. In real-world scenarios, observations are often sparse over time and captured sparsely from diverse viewpoints (e.g., from security cameras), making dynamic reconstruction highly ill-posed. We present SV-GS, a framework that simultaneously estimates a deformation model and the object's motion over time under sparse observations. To initialize SV-GS, we leverage a rough skeleton graph and an initial static reconstruction as inputs to guide motion estimation. (Later, we show that this input requirement can be relaxed.) Our method optimizes a skeleton-driven deformation field composed of a coarse skeleton joint pose estimator and a module for fine-grained deformations. By making only the joint pose estimator time-dependent, our model enables smooth motion interpolation while preserving learned geometric details. Experiments on synthetic datasets show that our method outperforms existing approaches under sparse observations by up to 34% in PSNR, and achieves comparable performance to dense monocular video methods on real-world datasets despite using significantly fewer frames. Moreover, we demonstrate that the input initial static reconstruction can be replaced by a diffusion-based generative prior, making our method more practical for real-world scenarios.", "date": "2026-01-01", "categories": []}
{"id": "2603.00492", "title": "ArtiFixer: Enhancing and Extending 3D Reconstruction with Auto-Regressive Diffusion Models", "abstract": "Per-scene optimization methods such as 3D Gaussian Splatting provide state-of-the-art novel view synthesis quality but extrapolate poorly to under-observed areas. Methods that leverage generative priors to correct artifacts in these areas hold promise but currently suffer from two shortcomings. The first is scalability, as existing methods use image diffusion models or bidirectional video models that are limited in the number of views they can generate in a single pass (and thus require a costly iterative distillation process for consistency). The second is quality itself, as generators used in prior work tend to produce outputs that are inconsistent with existing scene content and fail entirely in completely unobserved regions. To solve these, we propose a two-stage pipeline that leverages two key insights. First, we train a powerful bidirectional generative model with a novel opacity mixing strategy that encourages consistency with existing observations while retaining the model's ability to extrapolate novel content in unseen areas. Second, we distill it into a causal auto-regressive model that generates hundreds of frames in a single pass. This model can directly produce novel views or serve as pseudo-supervision to improve the underlying 3D representation in a simple and highly efficient manner. We evaluate our method extensively and demonstrate that it can generate plausible reconstructions in scenarios where existing approaches fail completely. When measured on commonly benchmarked datasets, we outperform all existing baselines by a wide margin, exceeding prior state-of-the-art methods by 1-3 dB PSNR.", "date": "2026-02-28", "categories": []}
{"id": "2605.04435", "title": "Ground4D: Spatially-Grounded Feedforward 4D Reconstruction for Unstructured Off-Road Scenes", "abstract": "Feedforward Gaussian Splatting has recently emerged as an efficient paradigm for 4D reconstruction in autonomous driving. However, in unstructured off-road scenes, its performance degrades due to high-frequency geometry, ego-motion jitter, and increased non-rigid dynamics. These factors introduce conflicting Gaussian observations across timestamps, leading to either over-smoothed renderings or structural artifacts. To address this issue, we propose Ground4D, a spatially-grounded 4D feedforward framework for pose-free off-road reconstruction. The key idea is to resolve temporal conflicts through spatially localized conditioning. Specifically, we introduce voxel-grounded temporal Gaussian aggregation, which partitions the canonical Gaussian space into spatial voxels and performs query-conditioned temporal attention within each voxel. Intra-voxel softmax normalization ensures that temporal selectivity and spatial occupancy become mutually reinforcing rather than conflicting. We furthermore introduce surface normal cues as auxiliary geometric guidance to regularize the geometry of Gaussian primitives. Extensive experiments on ORAD-3D and RELLIS-3D demonstrate that Ground4D consistently outperforms existing feedforward methods in reconstruction quality and generalizes zero-shot to unseen off-road domains. Project page and code:https://github.com/wsnbws/Ground4D.", "date": "2026-05-06", "categories": []}
{"id": "2605.05095", "title": "A Bayesian Approach for Task-Specific Next-Best-View Selection with Uncertain Geometry", "abstract": "We develop a framework for task-specific active next-best-view selection in 3D reconstruction from point clouds, by casting the problem in the language of Bayesian decision theory. Our framework works by (a) placing a prior distribution over the space of implicit surfaces, (b) using recently-developed stochastic surface reconstruction methods to calculate the resulting posterior distribution, then (c) using the posterior distribution to carefully reason about which view to scan next. This enables us to perform camera selection in a manner that is directly optimized for the intended use of the reconstructed data - meaning, we reduce uncertainty only in those regions that make a difference in the task at hand, as opposed to prior approaches that reduce it uniformly across space. We evaluate our method across three distinct downstream tasks: semantic classification, segmentation, and PDE-guided physics simulation. Experimental results demonstrate that our framework achieves superior task performance with fewer views compared to commonly used baselines and prior general uncertainty-reduction techniques.", "date": "2026-05-06", "categories": []}
{"id": "2602.10094", "title": "4RC: 4D Reconstruction via Conditional Querying Anytime and Anywhere", "abstract": "We present 4RC, a unified feed-forward framework for 4D reconstruction from monocular videos. Unlike existing approaches that typically decouple motion from geometry or produce limited 4D attributes such as sparse trajectories or two-view scene flow, 4RC learns a holistic 4D representation that jointly captures dense scene geometry and motion dynamics. At its core, 4RC introduces a novel encode-once, query-anywhere and anytime paradigm: a transformer backbone encodes the entire video into a compact spatio-temporal latent space, from which a conditional decoder can efficiently query 3D geometry and motion for any query frame at any target timestamp. To facilitate learning, we represent per-view 4D attributes in a minimally factorized form by decomposing them into base geometry and time-dependent relative motion. Extensive experiments demonstrate that 4RC outperforms prior and concurrent methods across a wide range of 4D reconstruction tasks.", "date": "2026-02-10", "categories": []}
{"id": "2604.24311", "title": "BIMStruct3D: A Fully Automated Hybrid Learning Scan-to-BIM Pipeline with Integrated Topology Refinement", "abstract": "Automatic generation of Building Information Models (BIM) from building scans is a key challenge in architecture and construction. We present a modular pipeline for generating IFC-compliant BIM from 3D point clouds. The hybrid approach combines learning-based semantic segmentation with topology-aware geometric reconstruction to model structural elements accurately. We propose vIoU, adapting voxel-based overlap evaluation to Scan-to-BIM by enabling holistic, instance-matching-free comparison of reconstructed and ground-truth models. We release the German Hospital dataset (DeKH), including high-resolution point clouds, ground truth BIMs, and semantic annotations. Experiments on DeKH and CV4AEC datasets show significant improvements over a RANSAC-based baseline, demonstrating robustness and scalability.", "date": "2026-04-27", "categories": []}
{"id": "2605.03359", "title": "Mix3R: Mixing Feed-forward Reconstruction and Generative 3D Priors for Joint Multi-view Aligned 3D Reconstruction and Pose Estimation", "abstract": "Recent trends in sparse-view 3D reconstruction have taken two different paths: feed-forward reconstruction that predicts pixel-aligned point maps without a complete geometry, and generative 3D reconstruction that generates complete geometry but often with poor input-alignment. We present Mix3R, a novel generative 3D reconstruction method which mixes feed-forward reconstruction and 3D generation into a single framework in an aligned manner. Mix3R generates a 3D shape in two stages: a sparse voxel generation stage and a textured geometry generation stage. Unlike pure generative methods, our first-stage generation jointly produces a coarse 3D structure (sparse voxels), per-view point maps and camera parameters aligned to that 3D structure. This is made possible by introducing a Mixture-of-Transformers architecture that inserts global self-attentions to a feed-forward reconstruction model and a 3D generative model, both pretrained on large-scale data. This design effectively retains the pretrained priors but enables better 2D-3D alignment. Based on the initial aligned generations of sparse 3D voxels and point maps, we compute an overlap-based attention bias that is directly added to another pretrained textured geometry generation model, enabling it to correctly place input textures onto generated shapes in a training-free manner. Our design brings mutual benefits to both feed-forward reconstruction and 3D generation: The feed-forward branch learns to ground its predictions to a generative 3D prior, and conversely, the 3D generation branch is conditioned on geometrically informative features from the feed-forward branch. As a result, our method produces 3D shapes with better input alignment compared with pure 3D generative methods, together with camera pose estimations more accurate than previous feed-forward reconstruction methods. Our project page is at https://jsnln.github.io/mix3r/", "date": "2026-05-05", "categories": []}
{"id": "2605.03463", "title": "First Shape, Then Meaning: Efficient Geometry and Semantics Learning for Indoor Reconstruction", "abstract": "Neural Surface Reconstruction has become a standard methodology for indoor 3D reconstruction, with Signed Distance Functions (SDFs) proving particularly effective for representing scene geometry. A variety of applications require a detailed understanding of the scene context, driving the need for object-level semantic signals. While recent methods successfully integrate semantic labels, they often inherit the slow training time and limited scalability of multi-SDF learning. In this paper, we introduce FSTM, a unified approach for learning geometry and semantics through a two-step process: a geometry warm-up using RGB inputs and geometric cues, followed by semantic field estimation. By first optimising geometry without semantic supervision, we observe substantial improvements compared to the standard joint optimisation. Rather than relying on specialised modules or complex multi-SDF designs, FSTM shows that a streamlined formulation is sufficient to achieve strong geometric and semantic reconstructions. Experiments on both synthetic and real-world indoor datasets show that our method outperforms multi-SDF approaches. It trains 2.3x faster on Replica, improves robustness to real-world imperfections on ScanNet++, and achieves higher recall by recovering the surfaces of more objects in the scene. The code will be made available at https://remichierchia.github.io/FSTM.", "date": "2026-05-05", "categories": []}
{"id": "2411.18966", "title": "SVGS: Enhancing Gaussian Splatting Using Primitives with Spatially Varying Colors", "abstract": "Gaussian Splatting demonstrates impressive results in multi-view reconstruction based on Gaussian explicit representations. However, the current Gaussian primitives only have a single view-dependent color and an opacity to represent the appearance and geometry of the scene, resulting in a non-compact representation. In this paper, we introduce a new method called SVGS (Spatially Varying Gaussian Splatting) that utilizes spatially varying colors and opacity in a single Gaussian primitive to improve its representation ability. We have implemented bilinear interpolation, movable kernels, and tiny neural networks as spatially varying functions. SVGS employs 2D Gaussian surfels as primitives, which significantly enhances novel-view synthesis while maintaining high-quality geometric reconstruction. This approach is particularly effective in practical applications, as scenes combining complex textures with relatively simple geometry occur frequently in real-world environments. Quantitative and qualitative experimental results demonstrate that all three functions outperform the baseline, with the best movable kernels achieving superior novel view synthesis performance on multiple datasets, highlighting the strong potential of spatially varying functions. Project page: https://ruixu.me/html/SuperGaussians/index.html", "date": "2024-11-28", "categories": []}
{"id": "2503.12001", "title": "3D Gaussian Splatting against Moving Objects for High-Fidelity Street Scene Reconstruction", "abstract": "The accurate reconstruction of dynamic street scenes is critical for applications in autonomous driving, augmented reality, and virtual reality. Traditional methods relying on dense point clouds and triangular meshes struggle with moving objects, occlusions, and real-time processing constraints, limiting their effectiveness in complex urban environments. While multi-view stereo and neural radiance fields have advanced 3D reconstruction, they face challenges in computational efficiency and handling scene dynamics. This paper proposes a novel 3D Gaussian point distribution method for dynamic street scene reconstruction. Our approach introduces an adaptive transparency mechanism that eliminates moving objects while preserving high-fidelity static scene details. Additionally, iterative refinement of Gaussian point distribution enhances geometric accuracy and texture representation. We integrate directional encoding with spatial position optimization to optimize storage and rendering efficiency, reducing redundancy while maintaining scene integrity. Experimental results demonstrate that our method achieves high reconstruction quality, improved rendering performance, and adaptability in large-scale dynamic environments. These contributions establish a robust framework for real-time, high-precision 3D reconstruction, advancing the practicality of dynamic scene modeling across multiple applications. The source code for this work is available to the public at https://github.com/okic-ca/3dgs", "date": "2025-03-15", "categories": []}
{"id": "2503.19703", "title": "High-Quality Spatial Reconstruction and Orthoimage Generation Using Efficient 2D Gaussian Splatting", "abstract": "Highly accurate geometric precision and dense image features characterize True Digital Orthophoto Maps (TDOMs), which are in great demand for applications such as urban planning, infrastructure management, and environmental monitoring. Traditional TDOM generation methods need sophisticated processes, such as Digital Surface Models (DSM) and occlusion detection, which are computationally expensive and prone to errors. This work presents an alternative technique rooted in 2D Gaussian Splatting (2DGS), free of explicit DSM and occlusion detection. With depth map generation, spatial information for every pixel within the TDOM is retrieved and can reconstruct the scene with high precision. Divide-and-conquer strategy achieves excellent GS training and rendering with high-resolution TDOMs at a lower resource cost, which preserves higher quality of rendering on complex terrain and thin structure without a decrease in efficiency. Experimental results demonstrate the efficiency of large-scale scene reconstruction and high-precision terrain modeling. This approach provides accurate spatial data, which assists users in better planning and decision-making based on maps.", "date": "2025-03-25", "categories": []}
{"id": "2512.03210", "title": "Flux4D: Flow-based Unsupervised 4D Reconstruction", "abstract": "Reconstructing large-scale dynamic scenes from visual observations is a fundamental challenge in computer vision, with critical implications for robotics and autonomous systems. While recent differentiable rendering methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have achieved impressive photorealistic reconstruction, they suffer from scalability limitations and require annotations to decouple actor motion. Existing self-supervised methods attempt to eliminate explicit annotations by leveraging motion cues and geometric priors, yet they remain constrained by per-scene optimization and sensitivity to hyperparameter tuning. In this paper, we introduce Flux4D, a simple and scalable framework for 4D reconstruction of large-scale dynamic scenes. Flux4D directly predicts 3D Gaussians and their motion dynamics to reconstruct sensor observations in a fully unsupervised manner. By adopting only photometric losses and enforcing an \"as static as possible\" regularization, Flux4D learns to decompose dynamic elements directly from raw data without requiring pre-trained supervised models or foundational priors simply by training across many scenes. Our approach enables efficient reconstruction of dynamic scenes within seconds, scales effectively to large datasets, and generalizes well to unseen environments, including rare and unknown objects. Experiments on outdoor driving datasets show Flux4D significantly outperforms existing methods in scalability, generalization, and reconstruction quality.", "date": "2025-12-02", "categories": []}
{"id": "2601.17185", "title": "LGDWT-GS: Local and Global Discrete Wavelet-Regularized 3D Gaussian Splatting for Sparse-View Scene Reconstruction", "abstract": "We propose a new method for few-shot 3D reconstruction that integrates global and local frequency regularization to stabilize geometry and preserve fine details under sparse-view conditions, addressing a key limitation of existing 3D Gaussian Splatting (3DGS) models. We also introduce a new multispectral greenhouse dataset containing four spectral bands captured from diverse plant species under controlled conditions. Alongside the dataset, we release an open-source benchmarking package that defines standardized few-shot reconstruction protocols for evaluating 3DGS-based methods. Experiments on our multispectral dataset, as well as standard benchmarks, demonstrate that the proposed method achieves sharper, more stable, and spectrally consistent reconstructions than existing baselines. The dataset and code for this work are publicly available", "date": "2026-01-23", "categories": []}
{"id": "2605.01995", "title": "From Concept to Capability: Evaluating 3D Gaussian Splatting for Synthetic Scene Editing in Autonomous Driving", "abstract": "The perception of an Autonomous Driving System (ADS) critically depends on relevant, comprehensive, and diverse datasets to ensure its safety while operating in the environment. Field data collection lacks completeness with respect to the list of rare but still possible safety-related scenarios needed for the development, verification, and validation of the ADS. 3D Gaussian Splatting (3DGS) has shown promising capabilities for the reconstruction and editing of scenes based on data collected by cameras and LiDAR sensors. However, the industrial fidelity evaluation of reconstructions is underexplored, which is crucial when employing such methods in safety-related systems, especially for ADS. This becomes more challenging as ADS operates in a dynamic, uncontrolled environment with limited viewpoints and often partially occluded objects. This paper addresses this gap by proposing and implementing a framework (Fig. 1) to systematically analyze the capabilities and limitations of 3DGS for use in the reconstruction of safety-related scenes. It focuses on the quality of reconstruction for vehicles and pedestrians, which are the two most critical object classes for ADS. Our findings provide industry insights into the fidelity degradation of reconstructions from multiple novel viewpoints, both lateral and longitudinal, enabling the integration of these methods into real-world industrial AD software development and testing pipelines.", "date": "2026-05-03", "categories": []}
{"id": "2503.06740", "title": "Diffusion Models are Secretly Zero-Shot 3DGS Harmonizers", "abstract": "Gaussian Splatting has become a popular technique for various 3D Computer Vision tasks, including novel view synthesis, scene reconstruction, and dynamic scene rendering. However, the challenge of natural-looking object insertion, where the object's appearance seamlessly matches the scene, remains unsolved. In this work, we propose a method, dubbed D3DR, for inserting a 3DGS-parametrized object into a 3DGS scene while correcting its lighting, shadows, and other visual artifacts to ensure consistency. We reveal a hidden ability of diffusion models trained on large real-world datasets to implicitly understand correct scene lighting, and leverage it in our pipeline. After inserting the object, we optimize a diffusion-based Delta Denoising Score (DDS)-inspired objective to adjust its 3D Gaussian parameters for proper lighting correction. We introduce a novel diffusion personalization technique that preserves object geometry and texture across diverse lighting conditions, and utilize it to achieve consistent identity matching between original and inserted objects. Finally, we demonstrate the effectiveness of the method by comparing it to existing approaches, achieving 2.0 dB PSNR improvements in relighting quality.", "date": "2025-03-09", "categories": []}
{"id": "2511.12895", "title": "High Dynamic Range 3D Gaussian Splatting via Luminance-Chromaticity Decomposition", "abstract": "High Dynamic Range (HDR) 3D reconstruction is pivotal for professional content creation in filmmaking and virtual production. Existing methods typically rely on multi-exposure Low Dynamic Range (LDR) supervision to constrain the learning process within vast brightness spaces, resulting in complex, dual-branch architectures. This work explores the feasibility of learning HDR 3D models exclusively in the HDR data space to simplify model design. By analyzing 3D Gaussian Splatting (3DGS) for HDR imagery, we reveal that its failure stems from the limited capacity of Spherical Harmonics (SHs) to capture extreme radiance variations across views, often biasing towards high-radiance observations and underfitting. While increasing the maximum SH degree improves training fitting, it leads to severe overfitting and excessive parameter overhead. To address this, we propose \\textit{Luminance--Chromaticity Decomposition Gaussian Splatting} (LCD-GS). By decoupling luminance and chromaticity into independent parameters, LCD-GS significantly enhances learning flexibility with minimal parameter increase (\\textit{e.g.}, one extra scalar per primitive). Notably, LCD-GS maintains the original training and inference pipeline, requiring only a change in color representation. This explicit decomposition naturally enables primitive-level local and global luminance editing during inference. Extensive experiments on synthetic and real datasets demonstrate that LCD-GS consistently outperforms state-of-the-art methods in reconstruction fidelity and dynamic-range preservation even with a simpler, more efficient architecture, providing an elegant paradigm for professional-grade HDR 3D modeling. Code and datasets will be released.", "date": "2025-11-17", "categories": []}
{"id": "2605.00147", "title": "From Images2Mesh: A 3D Surface Reconstruction Pipeline for Non-Cooperative Space Objects", "abstract": "On-orbit inspection imagery is crucial as it enables characterization of non-cooperative resident space objects, providing the geometry and structural condition essential for active debris removal and on-orbit servicing mission planning. However, most existing neural implicit surface reconstruction methods have been confined to synthetic or hardware-in-the-loop data with known camera poses and controlled illumination. In this work, we present a pipeline for neural implicit surface reconstruction of non-cooperative space objects from monocular inspection imagery. We demonstrate it on publicly released ISS inspection footage from the STS-119 mission and publicly released on-orbit inspection footage of an H-IIA rocket upper stage. We find that segmentation-based background removal is essential for successful camera pose estimation from real on-orbit footage, where background variation between frames caused direct processing to fail entirely. We further incorporate photometric correction of per-frame exposure variations and analyze its behavior across datasets, finding that performance in shadowed regions varies with the illumination characteristics of the input footage.", "date": "2026-04-30", "categories": []}
{"id": "2512.10267", "title": "Long-LRM++: Preserving Fine Details in Feed-Forward Wide-Coverage Reconstruction", "abstract": "Recent advances in generalizable Gaussian splatting (GS) have enabled feed-forward reconstruction of scenes from tens of input views. Long-LRM notably scales this paradigm to 32 input images at $950\\times540$ resolution, achieving 360{\\deg} scene-level reconstruction in a single forward pass. However, directly predicting millions of Gaussian parameters at once remains highly error-sensitive: small inaccuracies in positions or other attributes lead to noticeable blurring, particularly in fine structures such as text. In parallel, implicit representation methods such as LVSM and LaCT have demonstrated significantly higher rendering fidelity by compressing scene information into model weights rather than explicit Gaussians, and decoding RGB frames using the full transformer or TTT backbone. However, this computationally intensive decompression process for every rendered frame makes real-time rendering infeasible. These observations raise key questions: Is the deep, sequential \"decompression\" process necessary? Can we retain the benefits of implicit representations while enabling real-time performance? We address these questions with Long-LRM++, a model that adopts a semi-explicit scene representation combined with a lightweight decoder. Long-LRM++ matches the rendering quality of LaCT on DL3DV while achieving real-time 14 FPS rendering on an A100 GPU, overcoming the speed limitations of prior implicit methods. Our design also scales to 64 input views at the $950\\times540$ resolution, demonstrating strong generalization to increased input lengths. Additionally, Long-LRM++ delivers superior novel-view depth prediction on ScanNetv2 compared to direct depth rendering from Gaussians. Extensive ablation studies validate the effectiveness of each component in the proposed framework.", "date": "2025-12-11", "categories": []}
{"id": "2601.23065", "title": "EAG-PT: Emission-Aware Gaussians and Path Tracing for Diffuse Indoor Scene Reconstruction and Editing", "abstract": "Recent radiance-field-based reconstruction methods, such as NeRF and 3DGS, achieve high visual fidelity for indoor scenes, but often break down under scene editing due to baked illumination and the lack of explicit light transport. In contrast, inverse path tracing methods based on mesh representations enforce correct light transport but require highly accurate geometry, making them difficult to apply robustly to real indoor scenes. We present Emission-Aware Gaussians and Path Tracing (EAG-PT), a method for physically based reconstruction and rendering of indoor scenes using a unified 2D Gaussian representation, targeting editable diffuse global illumination. Our approach consists of three key ideas: (1) representing indoor scenes with 2D Gaussians as a transport-friendly geometric proxy that avoids explicit mesh reconstruction; (2) explicitly separating emissive and non-emissive components during reconstruction to support editing; and (3) decoupling reconstruction from final rendering by using efficient single-bounce optimization and high-quality multi-bounce path tracing, respectively. Experiments on synthetic and real indoor scenes show that EAG-PT produces more natural and physically consistent edited renderings than radiance-field reconstructions, while preserving finer geometric detail and avoiding mesh-induced artifacts compared with mesh-based inverse path tracing. These results highlight the potential of our approach for applications such as interior design, XR content creation, and embodied AI.", "date": "2026-01-30", "categories": []}
{"id": "2604.27106", "title": "Reconstruction by Generation: 3D Multi-Object Scene Reconstruction from Sparse Observations", "abstract": "Accurately reconstructing complex full multi-object scenes from sparse observations remains a core challenge in computer vision and a key step toward scalable and reliable simulation for robotics. In this work, we introduce RecGen, a generative framework for probabilistic joint estimation of object and part shapes, as well as their pose under occlusion and partial visibility from one or multiple RGB-D images. By leveraging compositional synthetic scene generation and strong 3D shape priors, RecGen generalizes across diverse object types and real-world environments. RecGen achieves state-of-the-art performance on complex, heavily occluded datasets, robustly handling severe occlusions, symmetric objects, object parts, and intricate geometry and texture. Despite using nearly 80% fewer training meshes than the previous state of the art SAM3D, RecGen outperforms it by 30.1% in geometric shape quality, 9.1% in texture reconstruction, and 33.9% in pose estimation.", "date": "2026-04-29", "categories": []}
{"id": "2604.27437", "title": "Softmax-GS: Generalized Gaussians Learning When to Blend or Bound", "abstract": "3D Gaussian Splatting (3D GS) is widely adopted for novel view synthesis due to its high training and rendering efficiency. However, its efficiency relies on the key assumption that Gaussians do not overlap in the 3D space, which leads to noticeable artifacts and view inconsistencies. In addition, the inherently diffuse boundaries of Gaussians hinder accurate reconstruction of sharp object edges. We propose Softmax-GS, a unified solution that addresses both the view-inconsistency and the diffuse-boundary problem by enforcing a softmax-based competition in overlapping regions between two Gaussians. With learnable parameters controlling the strength of the competition, it enables a continuous spectrum from smooth color blending to crisp, well-defined boundaries. Our formulation explicitly preserves order invariance for any two overlapping Gaussians and ensures that the output transmittance remains unchanged irrespective of the extent of overlapping, preventing undesirable discontinuities in the rendered output. Ablation experiments on simple geometries demonstrate the effectiveness of each component of Softmax-GS, and evaluations on real-world benchmarks show that it achieves state-of-the-art performance, improving both reconstruction quality and parameter efficiency.", "date": "2026-04-30", "categories": []}
{"id": "2604.28016", "title": "Faster 3D Gaussian Splatting Convergence via Structure-Aware Densification", "abstract": "3D Gaussian Splatting has emerged as a powerful scene representation for real-time novel-view synthesis. However, its standard adaptive density control relies on screen-space positional gradients, which do not distinguish between geometric misplacement and frequency aliasing, often leading to either over-blurred high-frequency textures or inefficient over-densification. We present a structure-aware densification framework. Our key insight is that the decision to subdivide a Gaussian should be driven by an explicit comparison between its projected screen-space extent and the local structure of the texture it seeks to represent. We introduce a multi-scale frequency analysis combining structure tensors with Laplacian scale space analysis to estimate the dominant frequency at each pixel, enabling robust supervision across varying texture scales. Based on this analysis, we define $\\eta$, a per-Gaussian, per-axis frequency violation metric that indicates when a primitive may be under-resolving local texture details. Unlike methods that perform isotropic splitting (e.g., splitting each Gaussian into two smaller ones with uniform shape), our approach performs anisotropic splitting. For each axis with high $\\eta$, we compute a split factor to better resolve the local frequency content. We further introduce a multiview consistency criterion that aggregates $\\eta$ observations across multiple views. By performing densification early and faster, we skip the lengthy iterative densification phases required by baseline methods and achieve significantly faster convergence. Experiments on standard benchmarks demonstrate that our method also achieves superior reconstruction quality, particularly in high-frequency regions.", "date": "2026-04-30", "categories": []}
{"id": "2604.28064", "title": "3D Reconstruction Techniques in the Manufacturing Domain: Applications, Research Opportunities and Use Cases", "abstract": "This comprehensive review examines the evolution and the current state of the art in three-dimensional (3D) reconstruction techniques in manufacturing applications. The analysis covers both traditional approaches and emerging deep learning methods, showing a critical research gap in unified 3d reconstruction frameworks. Through systematic review of 106 recent publications, we classify reconstruction techniques into three primary categories: data acquisition, point cloud generation, post-processing and applications. Non-contact methods, particularly structured light scanning and stereo vision, have shown significant adoption in manufacturing, with 47% of surveyed applications focusing on quality inspection. The integration of deep learning has enhanced reconstruction accuracy and processing speed, particularly in feature extraction and matching. Key applications span design and development (13%), machining (8%), process (17%), assembly (22%), and quality inspection (40%). While current technologies achieve sub-millimeter accuracy in controlled environments, challenges persist in handling reflective surfaces and dynamic environments. Our findings indicate a trend toward hybrid systems combining multiple sensor types and processing methods to overcome individual limitations. This survey provides a structured framework for understanding current capabilities and future directions in manufacturing-focused 3D reconstruction.", "date": "2026-04-30", "categories": []}
{"id": "2604.28115", "title": "FreeOcc: Training-Free Embodied Open-Vocabulary Occupancy Prediction", "abstract": "Existing learning-based occupancy prediction methods rely on large-scale 3D annotations and generalize poorly across environments. We present FreeOcc, a training-free framework for open-vocabulary occupancy prediction from monocular or RGB-D sequences. Unlike prior approaches that require voxel-level supervision and ground-truth camera poses, FreeOcc operates without 3D annotations, pose ground truth, or any learning stage. FreeOcc incrementally builds a globally consistent occupancy map via a four-layer pipeline: a SLAM backbone estimates poses and sparse geometry; a geometrically consistent Gaussian update constructs dense 3D Gaussian maps; open-vocabulary semantics from off-the-shelf vision-language models are associated with Gaussian primitives; and a probabilistic Gaussian-to-occupancy projection produces dense voxel occupancy. Despite being entirely training-free and pose-agnostic, FreeOcc achieves over $2\\times$ improvements in IoU and mIoU on EmbodiedOcc-ScanNet compared to prior self-supervised methods. We further introduce ReplicaOcc, a benchmark for indoor open-vocabulary occupancy prediction, and show that FreeOcc transfers zero-shot to novel environments, substantially outperforming both supervised and self-supervised baselines. Project page: https://the-masses.github.io/freeocc-web/.", "date": "2026-04-30", "categories": []}
{"id": "2604.28122", "title": "Beyond Gaussian Bottlenecks: Topologically Aligned Encoding of Vision-Transformer Feature Spaces", "abstract": "Modern visual world modeling systems increasingly rely on high-capacity architectures and large-scale data to produce plausible motion, yet they often fail to preserve underlying 3D geometry or physically consistent camera dynamics. A key limitation lies not only in model capacity, but in the latent representations used to encode geometric structure. We propose S$^2$VAE, a geometry-first latent learning framework that focuses on compressing and representing the latent 3D state of a scene, including camera motion, depth, and point-level structure, rather than modeling appearance alone. Building on representations from a Visual Geometry Grounded Transformer (VGGT), we introduce a novel type of variational autoencoder using a product of Power Spherical latent distributions, explicitly enforcing hyperspherical structure in the bottleneck to preserve directional and geometric semantics under strong compression. Across depth estimation, camera pose recovery, and point cloud reconstruction, we show that geometry-aligned hyperspherical latents consistently outperform conventional Gaussian bottlenecks, particularly in high-compression regimes. Our results highlight latent geometry as a first-class design choice for physically grounded visual and world models.", "date": "2026-04-30", "categories": []}
{"id": "2604.28193", "title": "Generalizable Sparse-View 3D Reconstruction from Unconstrained Images", "abstract": "Reconstructing 3D scenes from sparse, unposed images remains challenging under real-world conditions with varying illumination and transient occlusions. Existing methods rely on scene-specific optimization using appearance embeddings or dynamic masks, which requires extensive per-scene training and fails under sparse views. Moreover, evaluations on limited scenes raise questions about generalization. We present GenWildSplat, a feed-forward framework for sparse-view outdoor reconstruction that requires no per-scene optimization. Given unposed internet images, GenWildSplat predicts depth, camera parameters, and 3D Gaussians in a canonical space using learned geometric priors. An appearance adapter modulates appearance for target lighting conditions, while semantic segmentation handles transient objects. Through curriculum learning on synthetic and real data, GenWildSplat generalizes across diverse illumination and occlusion patterns. Evaluations on PhotoTourism and MegaScenes benchmark demonstrate state-of-the-art feed-forward rendering quality, achieving real-time inference without test-time optimization", "date": "2026-04-30", "categories": []}
{"id": "2507.01110", "title": "A LoD of Gaussians: Unified Training and Rendering for Ultra-Large Scale Reconstruction with External Memory", "abstract": "Gaussian Splatting has emerged as a high-performance technique for novel view synthesis, enabling real-time rendering and high-quality reconstruction of small scenes. However, scaling to larger environments has so far relied on partitioning the scene into chunks -- a strategy that introduces artifacts at chunk boundaries, complicates training across varying scales, and is poorly suited to unstructured scenarios such as city-scale flyovers combined with street-level views. Moreover, rendering remains fundamentally limited by GPU memory, as all visible chunks must reside in VRAM simultaneously. We introduce A LoD of Gaussians, a framework for training and rendering ultra-large-scale Gaussian scenes on a single consumer-grade GPU -- without partitioning. Our method stores the full scene out-of-core (e.g., in CPU memory) and trains a Level-of-Detail (LoD) representation directly, dynamically streaming only the relevant Gaussians. A hybrid data structure combining Gaussian hierarchies with Sequential Point Trees enables efficient, view-dependent LoD selection, while a lightweight caching and view scheduling system exploits temporal coherence to support real-time streaming and rendering. Together, these innovations enable seamless multi-scale reconstruction and interactive visualization of complex scenes -- from broad aerial views to fine-grained ground-level details.", "date": "2025-07-01", "categories": []}
{"id": "2603.05959", "title": "OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer", "abstract": "Reconstructing 3D geometry from streaming video requires continuous inference under bounded resources. Recent geometric foundation models achieve impressive reconstruction quality through all-to-all attention, yet their quadratic cost confines them to short, offline sequences. Causal-attention variants such as StreamVGGT enable single-pass streaming but accumulate an ever-growing KV cache, exhausting GPU memory within hundreds of frames and precluding the long-horizon deployment that motivates streaming inference in the first place. We present OVGGT, a training-free framework that bounds both memory and compute to a fixed budget regardless of sequence length. Our approach combines Self-Selective Caching, which leverages FFN residual magnitudes to compress the KV cache while remaining fully compatible with FlashAttention, with Dynamic Anchor Protection, which shields coordinate-critical tokens from eviction to suppress geometric drift over extended trajectories. Extensive experiments on indoor, outdoor, and ultra-long sequence benchmarks demonstrate that OVGGT processes arbitrarily long videos within a constant VRAM envelope while achieving state-of-the-art 3D geometric accuracy. Project page: https://vaisr.github.io/OVGGT/ Code: https://github.com/VAISR/OVGGT", "date": "2026-03-06", "categories": []}
{"id": "2604.04135", "title": "NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Results", "abstract": "This paper presents a comprehensive review of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge, detailing the proposed methods and results. The challenge seeks to identify robust reconstruction pipelines that are robust under real-world adverse conditions, specifically extreme low-light and smoke-degraded environments, as captured by our RealX3D benchmark. A total of 279 participants registered for the competition, of whom 33 teams submitted valid results. We thoroughly evaluate the submitted approaches against state-of-the-art baselines, revealing significant progress in 3D reconstruction under adverse conditions. Our analysis highlights shared design principles among top-performing methods and provides insights into effective strategies for handling 3D scene degradation.", "date": "2026-04-05", "categories": []}
{"id": "2604.26238", "title": "EnerGS: Energy-Based Gaussian Splatting with Partial Geometric Priors", "abstract": "3D Gaussian Splatting (3DGS) has been widely adopted for scene reconstruction, where training inherently constitutes a highly coupled and non-convex optimization problem. Recent works commonly incorporate geometric priors, such as LiDAR measurements, either for initialization or as training constraints, with the goal of improving photometric reconstruction quality. However, in large-scale outdoor scenarios, such geometric supervision is often spatially incomplete and uneven, which limits its effectiveness as a reliable prior and can even be detrimental to the final reconstruction. To address this challenge, we model partially observable geometry as a continuous energy field induced by geometric evidence and propose EnerGS. Rather than enforcing geometry as a hard constraint, EnerGS provides a soft geometric guidance for the optimization of Gaussian primitives, allowing geometric information to steer the optimization process without directly restricting the solution space. Extensive experiments on large-scale outdoor scenes demonstrate that, under both sparse multi-view and monocular settings, EnerGS consistently improves photometric quality and geometric stability, while effectively mitigating overfitting during 3DGS training.", "date": "2026-04-29", "categories": []}
{"id": "2604.26920", "title": "Color-Encoded Illumination for High-Speed Volumetric Scene Reconstruction", "abstract": "The task of capturing and rendering 3D dynamic scenes from 2D images has become increasingly popular in recent years. However, most conventional cameras are bandwidth-limited to 30-60 FPS, restricting these methods to static or slowly evolving scenes. While overcoming bandwidth limitations is difficult for general scenes, recent years have seen a flurry of computational imaging methods that yield high-speed videos using conventional cameras for specific applications (e.g., motion capture and particle image velocimetry). However, most of these methods require modifications to a camera's optics or the addition of mechanically moving components, limiting them to a single-view high-speed capture. Consequently, these methods cannot be readily used to capture a 3D representation of rapid scene motion. In this paper, we propose a novel method to capture and reconstruct a volumetric representation of a high-speed scene using only unaugmented low-speed cameras. Instead of modifying the hardware or optics of each individual camera, we encode high-speed scene dynamics by illuminating the scene with a rapid, sequential color-coded sequence. This results in simultaneous multi-view capture of the scene, where high-speed temporal information is encoded in the spatial intensity and color variations of the captured images. To construct a high-speed volumetric representation of the dynamic scene, we develop a novel dynamic Gaussian Splatting-based approach that decodes the temporal information from the images. We evaluate our approach on simulated scenes and real-world experiments using a multi-camera imaging setup, showing first-of-a-kind high-speed volumetric scene reconstructions.", "date": "2026-04-29", "categories": []}
{"id": "2512.04021", "title": "C3G: Learning Compact 3D Representations with 2K Gaussians", "abstract": "Reconstructing and understanding 3D scenes from unposed sparse views in a feed-forward manner remains as a challenging task in 3D computer vision. Recent approaches use per-pixel 3D Gaussian Splatting for reconstruction, followed by a 2D-to-3D feature lifting stage for scene understanding. However, they generate excessive redundant Gaussians, causing high memory overhead and sub-optimal multi-view feature aggregation, leading to degraded novel view synthesis and scene understanding performance. We propose C3G, a novel feed-forward framework that estimates compact 3D Gaussians only at essential spatial locations, minimizing redundancy while enabling effective feature lifting. We introduce learnable tokens that aggregate multi-view features through self-attention to guide Gaussian generation, ensuring each Gaussian integrates relevant visual features across views. We then exploit the learned attention patterns for Gaussian decoding to efficiently lift features. Extensive experiments on pose-free novel view synthesis, 3D open-vocabulary segmentation, and view-invariant feature aggregation demonstrate our approach's effectiveness. Results show that a compact yet geometrically meaningful representation is sufficient for high-quality scene reconstruction and understanding, achieving superior memory efficiency and feature fidelity compared to existing methods.", "date": "2025-12-03", "categories": []}
{"id": "2512.09923", "title": "Splatent: Splatting Diffusion Latents for Novel View Synthesis", "abstract": "Radiance field representations have recently been explored in the latent space of VAEs that are commonly used by diffusion models. This direction offers efficient rendering and seamless integration with diffusion-based pipelines. However, these methods face a fundamental limitation: The VAE latent space lacks multi-view consistency, leading to blurred textures and missing details during 3D reconstruction. Existing approaches attempt to address this by fine-tuning the VAE, at the cost of reconstruction quality, or by relying on pre-trained diffusion models to recover fine-grained details, at the risk of some hallucinations. We present Splatent, a diffusion-based enhancement framework designed to operate on top of 3D Gaussian Splatting (3DGS) in the latent space of VAEs. Our key insight departs from the conventional 3D-centric view: rather than reconstructing fine-grained details in 3D space, we recover them in 2D from input views through multi-view attention mechanisms. This approach preserves the reconstruction quality of pretrained VAEs while achieving faithful detail recovery. Evaluated across multiple benchmarks, Splatent establishes a new state-of-the-art for VAE latent radiance field reconstruction. We further demonstrate that integrating our method with existing feed-forward frameworks, consistently improves detail preservation, opening new possibilities for high-quality sparse-view 3D reconstruction. Code is available on our project page: https://orhir.github.io/Splatent/", "date": "2025-12-10", "categories": []}
{"id": "2604.19133", "title": "BALTIC: A Benchmark and Cross-Domain Strategy for 3D Reconstruction Across Air and Underwater Domains Under Varying Illumination", "abstract": "Robust 3D reconstruction across varying environmental conditions remains a critical challenge for robotic perception, particularly when transitioning between air and water. To address this, we introduce BALTIC, a controlled benchmark designed to systematically evaluate modern 3D reconstruction methods under variations in medium and lighting. The benchmark comprises 13 datasets spanning two media (air and water) and three lighting conditions (ambient, artificial, and mixed), with additional variations in motion type, scanning pattern, and initialization trajectory, resulting in a diverse set of sequences. Our experimental setup features a custom water tank equipped with a monocular camera and an HTC Vive tracker, enabling accurate ground-truth pose estimation. We further investigate cross-domain reconstruction by augmenting underwater image sequences with a small number of in-air views captured under similar lighting conditions. We evaluate Structure-from-Motion reconstruction using COLMAP in terms of both trajectory accuracy and scene geometry, and use these reconstructions as input to Neural Radiance Fields and 3D Gaussian Splatting methods. The resulting models are assessed against ground-truth trajectories and in-air references, while rendered outputs are compared using perceptual and photometric metrics. Additionally, we perform a color restoration analysis to evaluate radiometric consistency across domains. Our results show that under controlled, texture-consistent conditions, Gaussian Splatting with simple preprocessing (e.g., white balance correction) can achieve performance comparable to specialized underwater methods, although its robustness decreases in more complex and heterogeneous real-world environments", "date": "2026-04-21", "categories": []}
{"id": "2506.05480", "title": "ODE-GS: Latent ODEs for Dynamic Scene Extrapolation with 3D Gaussian Splatting", "abstract": "We introduce ODE-GS, a novel approach that integrates 3D Gaussian Splatting with latent neural ordinary differential equations (ODEs) to enable future extrapolation of dynamic 3D scenes. Unlike existing dynamic scene reconstruction methods, which rely on time-conditioned deformation networks and are limited to interpolation within a fixed time window, ODE-GS eliminates timestamp dependency by modeling Gaussian parameter trajectories as continuous-time latent dynamics. Our approach first learns an interpolation model to generate accurate Gaussian trajectories within the observed window, then trains a Transformer encoder to aggregate past trajectories into a latent state evolved via a neural ODE. Finally, numerical integration produces smooth, physically plausible future Gaussian trajectories, enabling rendering at arbitrary future timestamps. On the D-NeRF, NVFi, and HyperNeRF benchmarks, ODE-GS achieves state-of-the-art extrapolation performance, improving metrics by 19.8% compared to leading baselines, demonstrating its ability to accurately represent and predict 3D scene dynamics.", "date": "2025-06-05", "categories": []}
{"id": "2603.03269", "title": "LoGeR: Long-Context Geometric Reconstruction with Hybrid Memory", "abstract": "Feedforward geometric foundation models achieve strong short-window reconstruction, yet scaling them to minutes-long videos is bottlenecked by quadratic attention complexity or limited effective memory in recurrent designs. We present LoGeR (Long-context Geometric Reconstruction), a novel architecture that scales dense 3D reconstruction to extremely long sequences without post-optimization. LoGeR processes video streams in chunks, leveraging strong bidirectional priors for high-fidelity intra-chunk reasoning. To manage the critical challenge of coherence across chunk boundaries, we propose a learning-based hybrid memory module. This dual-component system combines a parametric Test-Time Training (TTT) memory to anchor the global coordinate frame and prevent scale drift, alongside a non-parametric Sliding Window Attention (SWA) mechanism to preserve uncompressed context for high-precision adjacent alignment. Remarkably, this memory architecture enables LoGeR to be trained on sequences of 128 frames, and generalize up to thousands of frames during inference. Evaluated across standard benchmarks and a newly repurposed VBR dataset with sequences of up to 19k frames, LoGeR substantially outperforms prior state-of-the-art feedforward methods--reducing ATE on KITTI by over 74%--and achieves robust, globally consistent reconstruction over unprecedented horizons.", "date": "2026-03-03", "categories": []}
{"id": "2604.05621", "title": "FunRec: Reconstructing Functional 3D Scenes from Egocentric Interaction Videos", "abstract": "We present FunRec, a method for reconstructing functional 3D digital twins of indoor scenes directly from egocentric RGB-D interaction videos. Unlike existing methods on articulated reconstruction, which rely on controlled setups, multi-state captures, or CAD priors, FunRec operates directly on in-the-wild human interaction sequences to recover interactable 3D scenes. It automatically discovers articulated parts, estimates their kinematic parameters, tracks their 3D motion, and reconstructs static and moving geometry in canonical space, yielding simulation-compatible meshes. Across new real and simulated benchmarks, FunRec surpasses prior work by a large margin, achieving up to +50 mIoU improvement in part segmentation, 5-10 times lower articulation and pose errors, and significantly higher reconstruction accuracy. We further demonstrate applications on URDF/USD export for simulation, hand-guided affordance mapping and robot-scene interaction.", "date": "2026-04-07", "categories": []}
{"id": "2604.23551", "title": "Spatiotemporal Degradation-Aware 3D Gaussian Splatting for Realistic Underwater Scene Reconstruction", "abstract": "Reconstructing realistic underwater scenes from underwater video remains a meaningful yet challenging task in the multimedia domain. The inherent spatiotemporal degradations in underwater imaging, including caustics, flickering, attenuation, and backscattering, frequently result in inaccurate geometry and appearance in existing 3D reconstruction methods. While a few recent works have explored underwater degradation-aware reconstruction, they often address either spatial or temporal degradation alone, falling short in more real-world underwater scenarios where both types of degradation occur. We propose MarineSTD-GS, a novel 3D Gaussian Splatting-based framework that explicitly models both temporal and spatial degradations for realistic underwater scene reconstruction. Specifically, we introduce two paired Gaussian primitives: Intrinsic Gaussians represent the true scene, while Degraded Gaussians render the degraded observations. The color of each Degraded Gaussian is physically derived from its paired Intrinsic Gaussian via a Spatiotemporal Degradation Modeling (SDM) module, enabling self-supervised disentanglement of realistic appearance from degraded images. To ensure stable training and accurate geometry, we further propose a Depth-Guided Geometry Loss and a Multi-Stage Optimization strategy. We also construct a simulated benchmark with diverse spatial and temporal degradations and ground-truth appearances for comprehensive evaluation. Experiments on both simulated and real-world datasets show that MarineSTD-GS robustly handles spatiotemporal degradations and outperforms existing methods in novel view synthesis with realistic, water-free scene appearances.", "date": "2026-04-26", "categories": []}
{"id": "2604.23803", "title": "Bringing a Personal Point of View: Evaluating Dynamic 3D Gaussian Splatting for Egocentric Scene Reconstruction", "abstract": "Egocentric video provides a unique view into human perception and interaction, with growing relevance for augmented reality, robotics, and assistive technologies. However, rapid camera motion and complex scene dynamics pose major challenges for 3D reconstruction from this perspective. While 3D Gaussian Splatting (3DGS) has become a state-of-the-art method for efficient, high-quality novel view synthesis, variants, that focus on reconstructing dynamic scenes from monocular video are rarely evaluated on egocentric video. It remains unclear whether existing models generalize to this setting or if egocentric-specific solutions are needed. In this work, we evaluate dynamic monocular 3DGS models on egocentric and exocentric video using paired ego-exo recordings from the EgoExo4D dataset. We find that reconstruction quality is consistently lower in egocentric views. Analysis reveals that the difference in reconstruction quality, measured in peak signal-to-noise ratio, stems from the reconstruction of static, not dynamic, content. Our findings underscore current limitations and motivate the development of egocentric-specific approaches, while also highlighting the value of separately evaluating static and dynamic regions of a video.", "date": "2026-04-26", "categories": []}
{"id": "2604.24586", "title": "Point-MF: One-step Point Cloud Generation from a Single Image via Mean Flows", "abstract": "Single-image point cloud reconstruction must infer complete 3D geometry, including occluded parts, from a single RGB image. While diffusion-based reconstructors achieve high accuracy, they typically require many denoising iterations, resulting in slow and expensive inference. We propose Point-MF, a Mean-Flow-based framework for low-NFE single-image point cloud reconstruction that couples a Mean-Flow-compatible architecture with an auxiliary loss. Specifically, Point-MF operates directly in point-cloud space to learn the mean velocity field and enables one-step reconstruction with a single network function evaluation (1-NFE), without relying on VAE-based latent representations. To make Mean Flow effective under large interval jumps, Point-MF employs a Diffusion Transformer tailored to the Mean-Flow setting, conditioned on frozen DINOv3 image features via a lightweight token adapter and equipped with explicit interval/time conditioning. Moreover, we introduce Denoised Space Anchor, a set-distance auxiliary loss on the denoised-space estimate $x_\\theta$ induced by the predicted velocity field, to stabilize large-step generation and reduce outliers and density artifacts. On ShapeNet-R2N2 and Pix3D, Point-MF strikes a strong balance between reconstruction quality and inference speed compared to multi-step diffusion baselines and competitive feedforward models, while generating high-quality point clouds with millisecond-level latency.", "date": "2026-04-27", "categories": []}
{"id": "2510.18999", "title": "OREN: Octree Residual Network for Real-Time Euclidean Signed Distance Mapping", "abstract": "Reconstructing signed distance functions (SDFs) from point cloud data benefits many robot autonomy capabilities, including localization, mapping, motion planning, and control. Methods that support online and large-scale SDF reconstruction often rely on discrete volumetric data structures, which affects the continuity and differentiability of the SDF estimates. Neural network methods have demonstrated high-fidelity differentiable SDF reconstruction but they tend to be less efficient, experience catastrophic forgetting and memory limitations in large environments, and are often restricted to truncated SDF. This work proposes OREN, a hybrid method that combines an explicit prior from octree interpolation with an implicit residual from neural network regression. Our method achieves non-truncated (Euclidean) SDF reconstruction with computational and memory efficiency comparable to volumetric methods and differentiability and accuracy comparable to neural network methods. Extensive experiments demonstrate that OREN outperforms the state of the art in terms of accuracy and efficiency, providing a scalable solution for downstream tasks in robotics and computer vision.", "date": "2025-10-21", "categories": []}
{"id": "2604.21400", "title": "You Only Gaussian Once: Controllable 3D Gaussian Splatting for Ultra-Densely Sampled Scenes", "abstract": "3D Gaussian Splatting (3DGS) has revolutionized neural rendering, yet existing methods remain predominantly research prototypes ill-suited for production-level deployment. We identify a critical \"Industry-Academia Gap\" hindering real-world application: unpredictable resource consumption from heuristic Gaussian growth, the \"sparsity shield\" of current benchmarks that rewards hallucination over physical fidelity, and severe multi-sensor data pollution. To bridge this gap, we propose YOGO (You Only Gaussian Once), a system-level framework that reformulates the stochastic growth process into a deterministic, budget-aware equilibrium. YOGO integrates a novel budget controller for hardware-constrained resource allocation and an availability-registration protocol for robust multi-sensor fusion. To push the boundaries of reconstruction fidelity, we introduce Immersion v1.0, the first ultra-dense indoor dataset specifically designed to break the \"sparsity shield.\" By providing saturated viewpoint coverage, Immersion v1.0 forces algorithms to focus on extreme physical fidelity rather than viewpoint interpolation, and enables the community to focus on the upper limits of high-fidelity reconstruction. Extensive experiments demonstrate that YOGO achieves state-of-the-art visual quality while maintaining a strictly deterministic profile, establishing a new standard for production-grade 3DGS. To facilitate reproducibility, part scenes of Immersion v1.0 dataset and source code of YOGO has been publicly released. The project link is https://jjrcn.github.io/yogo-project-home/", "date": "2026-04-23", "categories": []}
{"id": "2604.22129", "title": "PAGaS: Pixel-Aligned 1DoF Gaussian Splatting for Depth Refinement", "abstract": "Gaussian Splatting (GS) has emerged as an efficient approach for high-quality novel view synthesis. While early GS variants struggled to accurately model the scene's geometry, recent advancements constraining the Gaussians' spread and shapes, such as 2D Gaussian Splatting, have significantly improved geometric fidelity. In this paper, we present Pixel-Aligned 1DoF Gaussian Splatting (PAGaS) that adapts the GS representation from novel view synthesis to the multi-view stereo depth task. Our key contribution is modeling a pixel's depth using one-degree-of-freedom (1DoF) Gaussians that remain tightly constrained during optimization. Unlike existing approaches, our Gaussians' positions and sizes are restricted by the back-projected pixel volumes, leaving depth as the sole degree of freedom to optimize. PAGaS produces highly detailed depths, as illustrated in Figure 1. We quantitatively validate these improvements on top of reference geometric and learning-based multi-view stereo baselines on challenging 3D reconstruction benchmarks. Code: davidrecasens.github.io/pagas", "date": "2026-04-24", "categories": []}
{"id": "2604.22183", "title": "EvFlow-GS: Event Enhanced Motion Deblurring with Optical Flow for 3D Gaussian Splatting", "abstract": "Achieving sharp 3D reconstruction from motion-blurred images alone becomes challenging, motivating recent methods to incorporate event cameras, benefiting from microsecond temporal resolution. However, they suffer from residual artifacts and blurry texture details due to misleading supervision from inaccurate event double integral priors and noisy, blurry events. In this study, we propose EvFlow-GS, a unified framework that leverages event streams and optical flow to optimize an end-to-end learnable double integral (LDI), camera poses, and 3D Gaussian Splatting (3DGS) jointly on-the-fly. Specifically, we first extract edge information from the events using optical flow and then formulate a novel event-based loss applied separately to different modules. Additionally, we exploit a novel event-residual prior to strengthen the supervision of intensity changes between images rendered from 3DGS. Finally, we integrate the outputs of both 3DGS and LDI into a joint loss, enabling their optimization to mutually facilitate each other. Experiments demonstrate the leading performance of our EvFlow-GS.", "date": "2026-04-24", "categories": []}
{"id": "2604.22714", "title": "Long-tail Internet photo reconstruction", "abstract": "Internet photo collections exhibit an extremely long-tailed distribution: a few famous landmarks are densely photographed and easily reconstructed in 3D, while most real-world sites are represented with sparse, noisy, uneven imagery beyond the capabilities of both classical and learned 3D methods. We believe that tackling this long-tail regime represents one of the next frontiers for 3D foundation models. Although reliable ground-truth 3D supervision from sparse scenes is challenging to acquire, we observe that it can be effectively simulated by sampling sparse subsets from well-reconstructed Internet landmarks. To this end, we introduce MegaDepth-X, a large dataset of 3D reconstructions with clean, dense depth, together with a strategy for sampling sets of training images that mimic camera distributions in long-tail scenes. Finetuning 3D foundation models with these components yields robust reconstructions under extreme sparsity, and also enables more reliable reconstruction in symmetric and repetitive scenes, while preserving generalization to standard, dense 3D benchmark datasets.", "date": "2026-04-24", "categories": []}
{"id": "2604.13589", "title": "Dehaze-then-Splat: Generative Dehazing with Physics-Informed 3D Gaussian Splatting for Smoke-Free Novel View Synthesis", "abstract": "We present Dehaze-then-Splat, a two-stage pipeline for multi-view smoke removal and novel view synthesis developed for Track~2 of the NTIRE 2026 3D Restoration and Reconstruction Challenge. In the first stage, we produce pseudo-clean training images via per-frame generative dehazing using Nano Banana Pro, followed by brightness normalization. In the second stage, we train 3D Gaussian Splatting (3DGS) with physics-informed auxiliary losses -- depth supervision via Pearson correlation with pseudo-depth, dark channel prior regularization, and dual-source gradient matching -- that compensate for cross-view inconsistencies inherent in frame-wise generative processing. We identify a fundamental tension in dehaze-then-reconstruct pipelines: per-image restoration quality does not guarantee multi-view consistency, and such inconsistency manifests as blurred renders and structural instability in downstream 3D reconstruction.Our analysis shows that MCMC-based densification with early stopping, combined with depth and haze-suppression priors, effectively mitigates these artifacts. On the Akikaze validation scene, our pipeline achieves 20.98\\,dB PSNR and 0.683 SSIM for novel view synthesis, a +1.50\\,dB improvement over the unregularized baseline.", "date": "2026-04-15", "categories": []}
{"id": "2603.20714", "title": "The Role and Relationship of Initialization and Densification in 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has become the method of choice for photo-realistic 3D reconstruction of scenes, due to being able to efficiently and accurately recover the scene appearance and geometry from images. 3DGS represents the scene through a set of 3D Gaussians, parameterized by their position, spatial extent, and view-dependent color. Starting from an initial point cloud, 3DGS refines the Gaussians' parameters as to reconstruct a set of training images as accurately as possible. Typically, a sparse Structure-from-Motion point cloud is used as initialization. In order to obtain dense Gaussian clouds, 3DGS methods thus rely on a densification stage. In this paper, we systematically study the relation between densification and initialization. Proposing a new benchmark, we study combinations of different types of initializations (dense laser scans, dense (multi-view) stereo point clouds, dense monocular depth estimates, sparse SfM point clouds) and different densification schemes. We show that current densification approaches are not able to take full advantage of dense initialization as they are often unable to (significantly) improve over sparse SfM-based initialization. We will make our benchmark publicly available.", "date": "2026-03-21", "categories": []}
{"id": "2603.24725", "title": "Confidence-Based Mesh Extraction from 3D Gaussians", "abstract": "Recently, 3D Gaussian Splatting (3DGS) greatly accelerated mesh extraction from posed images due to its explicit representation and fast software rasterization. While the addition of geometric losses and other priors has improved the accuracy of extracted surfaces, mesh extraction remains difficult in scenes with abundant view-dependent effects. To resolve the resulting ambiguities, prior works rely on multi-view techniques, iterative mesh extraction, or large pre-trained models, sacrificing the inherent efficiency of 3DGS. In this work, we present a simple and efficient alternative by introducing a self-supervised confidence framework to 3DGS: within this framework, learnable confidence values dynamically balance photometric and geometric supervision. Extending our confidence-driven formulation, we introduce losses which penalize per-primitive color and normal variance and demonstrate their benefits to surface extraction. Finally, we complement the above with an improved appearance model, by decoupling the individual terms of the D-SSIM loss. Our final approach delivers state-of-the-art results for unbounded meshes while remaining highly efficient.", "date": "2026-03-25", "categories": []}
{"id": "2604.05687", "title": "3D Smoke Scene Reconstruction Guided by Vision Priors from Multimodal Large Language Models", "abstract": "Reconstructing 3D scenes from smoke-degraded multi-view images is particularly difficult because smoke introduces strong scattering effects, view-dependent appearance changes, and severe degradation of cross-view consistency. To address these issues, we propose a framework that integrates visual priors with efficient 3D scene modeling. We employ Nano-Banana-Pro to enhance smoke-degraded images and provide clearer visual observations for reconstruction and develop Smoke-GS, a medium-aware 3D Gaussian Splatting framework for smoke scene reconstruction and restoration-oriented novel view synthesis. Smoke-GS models the scene using explicit 3D Gaussians and introduces a lightweight view-dependent medium branch to capture direction-dependent appearance variations caused by smoke. Our method preserves the rendering efficiency of 3D Gaussian Splatting while improving robustness to smoke-induced degradation. Results demonstrate the effectiveness of our method for generating consistent and visually clear novel views in challenging smoke environments.", "date": "2026-04-07", "categories": []}
{"id": "2604.20093", "title": "FurnSet: Exploiting Repeats for 3D Scene Reconstruction", "abstract": "Single-view 3D scene reconstruction involves inferring both object geometry and spatial layout. Existing methods typically reconstruct objects independently or rely on implicit scene context, failing to exploit the repeated instances commonly present in realworld scenes. We propose FurnSet, a framework that explicitly identifies and leverages repeated object instances to improve reconstruction. Our method introduces per-object CLS tokens and a set-aware self-attention mechanism that groups identical instances and aggregates complementary observations across them, enabling joint reconstruction. We further combine scene-level and object-level conditioning to guide object reconstruction, followed by layout optimization using object point clouds with 3D and 2D projection losses for scene alignment. Experiments on 3D-Future and 3D-Front demonstrate improved scene reconstruction quality, highlighting the effectiveness of exploiting repetition for robust 3D scene reconstruction.", "date": "2026-04-22", "categories": []}
{"id": "2604.20784", "title": "GeoRect4D: Geometry-Compatible Generative Rectification for Dynamic Sparse-View 3D Reconstruction", "abstract": "Reconstructing dynamic 3D scenes from sparse multi-view videos is highly ill-posed, often leading to geometric collapse, trajectory drift, and floating artifacts. Recent attempts introduce generative priors to hallucinate missing content, yet naive integration frequently causes structural drift and temporal inconsistency due to the mismatch between stochastic 2D generation and deterministic 3D geometry. In this paper, we propose GeoRect4D, a novel unified framework for sparse-view dynamic reconstruction that couples explicit 3D consistency with generative refinement via a closed-loop optimization process. Specifically, GeoRect4D introduces a degradation-aware feedback mechanism that incorporates a robust anchor-based dynamic 3DGS substrate with a single-step diffusion rectifier to hallucinate high-fidelity details. This rectifier utilizes a structural locking mechanism and spatiotemporal coordinated attention, effectively preserving physical plausibility while restoring missing content. Furthermore, we present a progressive optimization strategy that employs stochastic geometric purification to eliminate floaters and generative distillation to infuse texture details into the explicit representation. Extensive experiments demonstrate that GeoRect4D achieves state-of-the-art performance in reconstruction fidelity, perceptual quality, and spatiotemporal consistency across multiple datasets.", "date": "2026-04-22", "categories": []}
{"id": "2604.05301", "title": "SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration", "abstract": "Real-world smoke simultaneously attenuates scene radiance, adds airlight, and destabilizes multi-view appearance consistency, making robust 3D reconstruction particularly difficult. We present \\textbf{SmokeGS-R}, a practical pipeline developed for the NTIRE 2026 3D Restoration and Reconstruction Track 2 challenge. The key idea is to decouple geometry recovery from appearance correction: we generate physics-guided pseudo-clean supervision with a refined dark channel prior and guided filtering, train a sharp clean-only 3D Gaussian Splatting source model, and then harmonize its renderings with a donor ensemble using geometric-mean reference aggregation, LAB-space Reinhard transfer, and light Gaussian smoothing. On the official challenge testing leaderboard, the final submission achieved \\mbox{PSNR $=15.217$} and \\mbox{SSIM $=0.666$}. After the public release of RealX3D, we re-evaluated the same frozen result on the seven released challenge scenes without retraining and obtained \\mbox{PSNR $=15.209$}, \\mbox{SSIM $=0.644$}, and \\mbox{LPIPS $=0.551$}, outperforming the strongest official baseline average on the same scenes by $+3.68$ dB PSNR. These results suggest that a geometry-first reconstruction strategy combined with stable post-render appearance harmonization is an effective recipe for real-world multi-view smoke restoration. The code is available at https://github.com/windrise/3drr_Track2_SmokeGS-R.", "date": "2026-04-07", "categories": []}
{"id": "2604.19216", "title": "An Object-Centered Data Acquisition Method for 3D Gaussian Splatting using Mobile Phones", "abstract": "Data acquisition through mobile phones remains a challenge for 3D Gaussian Splatting (3DGS). In this work we target the object-centered scenario and enable reliable mobile acquisition by providing on-device capture guidance and recording onboard sensor signals for offline reconstruction. After the calibration step, the device orientations are aligned to a baseline frame to obtain relative poses, and the optical axis of the camera is mapped to an object-centered spherical grid for uniform viewpoint indexing. To curb polar sampling bias, we compute area-weighted spherical coverage in real-time and guide the user's motion accordingly. We compare the proposed method with RealityScan and the free-capture strategy. Our method achieves superior reconstruction quality using fewer input images compared to free capture and RealityScan. Further analysis shows that the proposed method is able to obtain more comprehensive and uniform viewpoint coverage during object-centered acquisition.", "date": "2026-04-21", "categories": []}
{"id": "2604.19257", "title": "Unposed-to-3D: Learning Simulation-Ready Vehicles from Real-World Images", "abstract": "Creating realistic and simulation-ready 3D assets is crucial for autonomous driving research and virtual environment construction. However, existing 3D vehicle generation methods are often trained on synthetic data with significant domain gaps from real-world distributions. The generated models often exhibit arbitrary poses and undefined scales, resulting in poor visual consistency when integrated into driving scenes. In this paper, we present Unposed-to-3D, a novel framework that learns to reconstruct 3D vehicles from real-world driving images using image-only supervision. Our approach consists of two stages. In the first stage, we train an image-to-3D reconstruction network using posed images with known camera parameters. In the second stage, we remove camera supervision and use a camera prediction head that directly estimates the camera parameters from unposed images. The predicted pose is then used for differentiable rendering to provide self-supervised photometric feedback, enabling the model to learn 3D geometry purely from unposed images. To ensure simulation readiness, we further introduce a scale-aware module to predict real-world size information, and a harmonization module that adapts the generated vehicles to the target driving scene with consistent lighting and appearance. Extensive experiments demonstrate that Unposed-to-3D effectively reconstructs realistic, pose-consistent, and harmonized 3D vehicle models from real-world images, providing a scalable path toward creating high-quality assets for driving scene simulation and digital twin environments.", "date": "2026-04-21", "categories": []}
{"id": "2604.19556", "title": "Paparazzo: Active Mapping of Moving 3D Objects", "abstract": "Current 3D mapping pipelines generally assume static environments, which limits their ability to accurately capture and reconstruct moving objects. To address this limitation, we introduce the novel task of active mapping of moving objects, in which a mapping agent must plan its trajectory while compensating for the object's motion. Our approach, Paparazzo, provides a learning-free solution that robustly predicts the target's trajectory and identifies the most informative viewpoints from which to observe it, to plan its own path. We also contribute a comprehensive benchmark designed for this new task. Through extensive experiments, we show that Paparazzo significantly improves 3D reconstruction completeness and accuracy compared to several strong baselines, marking an important step toward dynamic scene understanding. Project page: https://davidea97.github.io/paparazzo-page/", "date": "2026-04-21", "categories": []}
{"id": "2604.19624", "title": "GRAFT: Geometric Refinement and Fitting Transformer for Human Scene Reconstruction", "abstract": "Reconstructing physically plausible 3D human-scene interactions (HSI) from a single image currently presents a trade-off: optimization based methods offer accurate contact but are slow (~20s), while feed-forward approaches are fast yet lack explicit interaction reasoning, producing floating and interpenetration artifacts.\n  Our key insight is that geometry-based human--scene fitting can be amortized into fast feed-forward inference. We present GRAFT (Geometric Refinement And Fitting Transformer), a learned HSI prior that predicts Interaction Gradients: corrective parameter updates that iteratively refine human meshes by reasoning about their 3D relationship to the surrounding scene.\n  GRAFT encodes the interaction state into compact body-anchored tokens, each grounded in the scene geometry via Geometric Probes that capture spatial relationships with nearby surfaces.\n  A lightweight transformer recurrently updates human meshes and re-probes the scene, ensuring the final pose aligns with both learned priors and observed geometry.\n  GRAFT operates either as an end-to-end reconstructor using image features, or with geometry alone as a transferable plug-and-play HSI prior that improves feed-forward methods without retraining.\n  Experiments show GRAFT improves interaction quality by up to 113% over state-of-the-art feed-forward methods and matches optimization-based interaction quality at ${\\sim}50{\\times}$ lower runtime, while generalizing seamlessly to in-the-wild multi-person scenes and being preferred in 64.8% of three-way user study. Project page: https://pradyumnaym.github.io/graft .", "date": "2026-04-21", "categories": []}
{"id": "2604.19747", "title": "AnyRecon: Arbitrary-View 3D Reconstruction with Video Diffusion Model", "abstract": "Sparse-view 3D reconstruction is essential for modeling scenes from casual captures, but remain challenging for non-generative reconstruction. Existing diffusion-based approaches mitigates this issues by synthesizing novel views, but they often condition on only one or two capture frames, which restricts geometric consistency and limits scalability to large or diverse scenes. We propose AnyRecon, a scalable framework for reconstruction from arbitrary and unordered sparse inputs that preserves explicit geometric control while supporting flexible conditioning cardinality. To support long-range conditioning, our method constructs a persistent global scene memory via a prepended capture view cache, and removes temporal compression to maintain frame-level correspondence under large viewpoint changes. Beyond better generative model, we also find that the interplay between generation and reconstruction is crucial for large-scale 3D scenes. Thus, we introduce a geometry-aware conditioning strategy that couples generation and reconstruction through an explicit 3D geometric memory and geometry-driven capture-view retrieval. To ensure efficiency, we combine 4-step diffusion distillation with context-window sparse attention to reduce quadratic complexity. Extensive experiments demonstrate robust and scalable reconstruction across irregular inputs, large viewpoint gaps, and long trajectories.", "date": "2026-04-21", "categories": []}
{"id": "2511.05152", "title": "Splatography: Sparse multi-view dynamic Gaussian Splatting for filmmaking challenges", "abstract": "Deformable Gaussian Splatting (GS) accomplishes photorealistic dynamic 3-D reconstruction from dense multi-view video (MVV) by learning to deform a canonical GS representation. However, in filmmaking, tight budgets can result in sparse camera configurations, which limits state-of-the-art (SotA) methods when capturing complex dynamic features. To address this issue, we introduce an approach that splits the canonical Gaussians and deformation field into foreground and background components using a sparse set of masks for frames at t=0. Each representation is separately trained on different loss functions during canonical pre-training. Then, during dynamic training, different parameters are modeled for each deformation field following common filmmaking practices. The foreground stage contains diverse dynamic features so changes in color, position and rotation are learned. While, the background containing film-crew and equipment, is typically dimmer and less dynamic so only changes in point position are learned. Experiments on 3-D and 2.5-D entertainment datasets show that our method produces SotA qualitative and quantitative results; up to 3 PSNR higher with half the model size on 3-D scenes. Unlike the SotA and without the need for dense mask supervision, our method also produces segmented dynamic reconstructions including transparent and dynamic textures. Code and video comparisons are available online: https://azzarelli.github.io/splatographypage/index.html", "date": "2025-11-07", "categories": []}
{"id": "2511.09818", "title": "Lumos3D: A Single-Forward Framework for Low-Light 3D Scene Restoration", "abstract": "Restoring 3D scenes with low-light conditions is challenging, and most existing methods depend on precomputed camera poses and scene-specific optimization, which greatly restricts their application to real-world scenarios. To overcome these limitations, we propose Lumos3D, a pose-free single-forward framework for 3D low-light scene restoration. First, we develop a cross-illumination distillation scheme, where a frozen teacher network takes normal-light ground truth images as input to distill accurate geometric information to the student model. Second, we define a Lumos loss to improve the restoration quality of the reconstructed 3D Gaussian space. Trained on a single dataset, Lumos3D performs inference in a purely feed-forward manner, directly restoring illumination and structure from unposed, low-light multi-view images without any per-scene training or optimization. Experiments on real-world datasets demonstrate that Lumos3D achieves competitive restoration results compared to scene-specific methods. Our codes will be released soon.", "date": "2025-11-12", "categories": []}
{"id": "2604.16910", "title": "LAGS: Low-Altitude Gaussian Splatting with Groupwise Heterogeneous Graph Learning", "abstract": "Low-altitude Gaussian splatting (LAGS) facilitates 3D scene reconstruction by aggregating aerial images from distributed drones. However, as LAGS prioritizes maximizing reconstruction quality over communication throughput, existing low-altitude resource allocation schemes become inefficient. This inefficiency stems from their failure to account for image diversity introduced by varying viewpoints. To fill this gap, we propose a groupwise heterogeneous graph neural network (GW-HGNN) for LAGS resource allocation. GW-HGNN explicitly models the non-uniform contribution of different image groups to the reconstruction process, thus automatically balancing data fidelity and transmission cost. The key insight of GW-HGNN is to transform LAGS losses and communication constraints into graph learning costs for dual-level message passing. Experiments on real-world LAGS datasets demonstrate that GW-HGNN significantly outperforms state-of-the-art benchmarks across key rendering metrics, including PSNR, SSIM, and LPIPS. Furthermore, GW-HGNN reduces computational latency by approximately 100x compared to the widely-used MOSEK solver, achieving millisecond-level inference suitable for real-time deployment.", "date": "2026-04-18", "categories": []}
{"id": "2604.17155", "title": "Instant Colorization of Gaussian Splats", "abstract": "Gaussian Splatting has recently become one of the most popular frameworks for photorealistic 3D scene reconstruction and rendering. While current rasterizers allow for efficient mappings of 3D Gaussian splats onto 2D camera views, this work focuses on mapping 2D image information (e.g. color, neural features or segmentation masks) efficiently back onto an existing scene of Gaussian splats. This 'opposite' direction enables applications ranging from scene relighting and stylization to 3D semantic segmentation, but also introduces challenges, such as view-dependent colorization and occlusion handling.\n  Our approach tackles these challenges using the normal equation to solve a visibility-weighted least squares problem for every Gaussian and can be implemented efficiently with existing differentiable rasterizers. We demonstrate the effectiveness of our approach on scene relighting, feature enrichment and 3D semantic segmentation tasks, achieving up to an order of magnitude speedup compared to gradient descent-based baselines.", "date": "2026-04-18", "categories": []}
{"id": "2604.17965", "title": "MU-GeNeRF: Multi-view Uncertainty-guided Generalizable Neural Radiance Fields for Distractor-aware Scene", "abstract": "Generalizable Neural Radiance Fields (GeNeRFs) enable high-quality scene reconstruction from sparse views and can generalize to unseen scenes. However, in real-world settings, transient distractors break cross-view structural consistency, corrupting supervision and degrading reconstruction quality. Existing distractor-free NeRF methods rely on per-scene optimization and estimate uncertainty from per-view reconstruction errors, which are not reliable for GeNeRFs and often misjudge inconsistent static structures as distractors. To this end, we propose MU-GeNeRF, a Multi-view Uncertainty-guided distractor-aware GeNeRF framework designed to alleviate GeNeRF's robust modeling challenges in the presence of transient distractions. We decompose distractor awareness into two complementary uncertainty components: Source-view Uncertainty, which captures structural discrepancies across source views caused by viewpoint changes or dynamic factors; and Target-view Uncertainty, which detects observation anomalies in the target image induced by transient distractors.These two uncertainties address distinct error sources and are combined through a heteroscedastic reconstruction loss, which guides the model to adaptively modulate supervision, enabling more robust distractor suppression and geometric modeling.Extensive experiments show that our method not only surpasses existing GeNeRFs but also achieves performance comparable to scene-specific distractor-free NeRFs.", "date": "2026-04-20", "categories": []}
{"id": "2604.18468", "title": "Asset Harvester: Extracting 3D Assets from Autonomous Driving Logs for Simulation", "abstract": "Closed-loop simulation is a core component of autonomous vehicle (AV) development, enabling scalable testing, training, and safety validation before real-world deployment. Neural scene reconstruction converts driving logs into interactive 3D environments for simulation, but it does not produce complete 3D object assets required for agent manipulation and large-viewpoint novel-view synthesis. To address this challenge, we present Asset Harvester, an image-to-3D model and end-to-end pipeline that converts sparse, in-the-wild object observations from real driving logs into complete, simulation-ready assets. Rather than relying on a single model component, we developed a system-level design for real-world AV data that combines large-scale curation of object-centric training tuples, geometry-aware preprocessing across heterogeneous sensors, and a robust training recipe that couples sparse-view-conditioned multiview generation with 3D Gaussian lifting. Within this system, SparseViewDiT is explicitly designed to address limited-angle views and other real-world data challenges. Together with hybrid data curation, augmentation, and self-distillation, this system enables scalable conversion of sparse AV object observations into reusable 3D assets.", "date": "2026-04-20", "categories": []}
{"id": "2506.13215", "title": "DVP-MVS++: Synergize Depth-Normal-Edge and Harmonized Visibility Prior for Multi-View Stereo", "abstract": "Recently, patch deformation-based methods have demonstrated significant effectiveness in multi-view stereo due to their incorporation of deformable and expandable perception for reconstructing textureless areas. However, these methods generally focus on identifying reliable pixel correlations to mitigate matching ambiguity of patch deformation, while neglecting the deformation instability caused by edge-skipping and visibility occlusions, which may cause potential estimation deviations. To address these issues, we propose DVP-MVS++, an innovative approach that synergizes both depth-normal-edge aligned and harmonized cross-view priors for robust and visibility-aware patch deformation. Specifically, to avoid edge-skipping, we first apply DepthPro, Metric3Dv2 and Roberts operator to generate coarse depth maps, normal maps and edge maps, respectively. These maps are then aligned via an erosion-dilation strategy to produce fine-grained homogeneous boundaries for facilitating robust patch deformation. Moreover, we reformulate view selection weights as visibility maps, and then implement both an enhanced cross-view depth reprojection and an area-maximization strategy to help reliably restore visible areas and effectively balance deformed patch, thus acquiring harmonized cross-view priors for visibility-aware patch deformation. Additionally, we obtain geometry consistency by adopting both aggregated normals via view selection and projection depth differences via epipolar lines, and then employ SHIQ for highlight correction to enable geometry consistency with highlight-aware perception, thus improving reconstruction quality during propagation and refinement stage. Evaluation results on ETH3D, Tanks & Temples and Strecha datasets exhibit the state-of-the-art performance and robust generalization capability of our proposed method.", "date": "2025-06-16", "categories": []}
{"id": "2604.15237", "title": "StreamCacheVGGT: Streaming Visual Geometry Transformers with Robust Scoring and Hybrid Cache Compression", "abstract": "Reconstructing dense 3D geometry from continuous video streams requires stable inference under a constant memory budget. Existing $O(1)$ frameworks primarily rely on a ``pure eviction'' paradigm, which suffers from significant information destruction due to binary token deletion and evaluation noise from localized, single-layer scoring. To address these bottlenecks, we propose StreamCacheVGGT, a training-free framework that reimagines cache management through two synergistic modules: Cross-Layer Consistency-Enhanced Scoring (CLCES) and Hybrid Cache Compression (HCC). CLCES mitigates activation noise by tracking token importance trajectories across the Transformer hierarchy, employing order-statistical analysis to identify sustained geometric salience. Leveraging these robust scores, HCC transcends simple eviction by introducing a three-tier triage strategy that merges moderately important tokens into retained anchors via nearest-neighbor assignment on the key-vector manifold. This approach preserves essential geometric context that would otherwise be lost. Extensive evaluations on five benchmarks (7-Scenes, NRGBD, ETH3D, Bonn, and KITTI) demonstrate that StreamCacheVGGT sets a new state-of-the-art, delivering superior reconstruction accuracy and long-term stability while strictly adhering to constant-cost constraints.", "date": "2026-04-16", "categories": []}
{"id": "2604.15284", "title": "GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens", "abstract": "The efficient spatial allocation of primitives serves as the foundation of 3D Gaussian Splatting, as it directly dictates the synergy between representation compactness, reconstruction speed, and rendering fidelity. Previous solutions, whether based on iterative optimization or feed-forward inference, suffer from significant trade-offs between these goals, mainly due to the reliance on local, heuristic-driven allocation strategies that lack global scene awareness. Specifically, current feed-forward methods are largely pixel-aligned or voxel-aligned. By unprojecting pixels into dense, view-aligned primitives, they bake redundancy into the 3D asset. As more input views are added, the representation size increases and global consistency becomes fragile. To this end, we introduce GlobalSplat, a framework built on the principle of align first, decode later. Our approach learns a compact, global, latent scene representation that encodes multi-view input and resolves cross-view correspondences before decoding any explicit 3D geometry. Crucially, this formulation enables compact, globally consistent reconstructions without relying on pretrained pixel-prediction backbones or reusing latent features from dense baselines. Utilizing a coarse-to-fine training curriculum that gradually increases decoded capacity, GlobalSplat natively prevents representation bloat. On RealEstate10K and ACID, our model achieves competitive novel-view synthesis performance while utilizing as few as 16K Gaussians, significantly less than required by dense pipelines, obtaining a light 4MB footprint. Further, GlobalSplat enables significantly faster inference than the baselines, operating under 78 milliseconds in a single forward pass. Project page is available at https://r-itk.github.io/globalsplat/", "date": "2026-04-16", "categories": []}
{"id": "2604.12580", "title": "PDF-GS: Progressive Distractor Filtering for Robust 3D Gaussian Splatting", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have enabled impressive real-time photorealistic rendering. However, conventional training pipelines inherently assume full multi-view consistency among input images, which makes them sensitive to distractors that violate this assumption and cause visual artifacts. In this work, we revisit an underexplored aspect of 3DGS: its inherent ability to suppress inconsistent signals. Building on this insight, we propose PDF-GS (Progressive Distractor Filtering for Robust 3D Gaussian Splatting), a framework that amplifies this self-filtering property through a progressive multi-phase optimization. The progressive filtering phases gradually remove distractors by exploiting discrepancy cues, while the following reconstruction phase restores fine-grained, view-consistent details from the purified Gaussian representation. Through this iterative refinement, PDF-GS achieves robust, high-fidelity, and distractor-free reconstructions, consistently outperforming baselines across diverse datasets and challenging real-world conditions. Moreover, our approach is lightweight and easily adaptable to existing 3DGS frameworks, requiring no architectural changes or additional inference overhead, leading to a new state-of-the-art performance. The code is publicly available at https://github.com/kangrnin/PDF-GS.", "date": "2026-04-14", "categories": []}
{"id": "2604.14141", "title": "Geometric Context Transformer for Streaming 3D Reconstruction", "abstract": "Streaming 3D reconstruction aims to recover 3D information, such as camera poses and point clouds, from a video stream, which necessitates geometric accuracy, temporal consistency, and computational efficiency. Motivated by the principles of Simultaneous Localization and Mapping (SLAM), we introduce LingBot-Map, a feed-forward 3D foundation model for reconstructing scenes from streaming data, built upon a geometric context transformer (GCT) architecture. A defining aspect of LingBot-Map lies in its carefully designed attention mechanism, which integrates an anchor context, a pose-reference window, and a trajectory memory to address coordinate grounding, dense geometric cues, and long-range drift correction, respectively. This design keeps the streaming state compact while retaining rich geometric context, enabling stable efficient inference at around 20 FPS on 518 x 378 resolution inputs over long sequences exceeding 10,000 frames. Extensive evaluations across a variety of benchmarks demonstrate that our approach achieves superior performance compared to both existing streaming and iterative optimization-based approaches.", "date": "2026-04-15", "categories": []}
{"id": "2604.14268", "title": "HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds", "abstract": "We introduce HY-World 2.0, a multi-modal world model framework that advances our prior project HY-World 1.0. HY-World 2.0 accommodates diverse input modalities, including text prompts, single-view images, multi-view images, and videos, and produces 3D world representations. With text or single-view image inputs, the model performs world generation, synthesizing high-fidelity, navigable 3D Gaussian Splatting (3DGS) scenes. This is achieved through a four-stage method: a) Panorama Generation with HY-Pano 2.0, b) Trajectory Planning with WorldNav, c) World Expansion with WorldStereo 2.0, and d) World Composition with WorldMirror 2.0. Specifically, we introduce key innovations to enhance panorama fidelity, enable 3D scene understanding and planning, and upgrade WorldStereo, our keyframe-based view generation model with consistent memory. We also upgrade WorldMirror, a feed-forward model for universal 3D prediction, by refining model architecture and learning strategy, enabling world reconstruction from multi-view images or videos. Also, we introduce WorldLens, a high-performance 3DGS rendering platform featuring a flexible engine-agnostic architecture, automatic IBL lighting, efficient collision detection, and training-rendering co-design, enabling interactive exploration of 3D worlds with character support. Extensive experiments demonstrate that HY-World 2.0 achieves state-of-the-art performance on several benchmarks among open-source approaches, delivering results comparable to the closed-source model Marble. We release all model weights, code, and technical details to facilitate reproducibility and support further research on 3D world models.", "date": "2026-04-15", "categories": []}
{"id": "2602.16385", "title": "Adaptive Multi-Scale Channel-Spatial Attention Aggregation Framework for 3D Indoor Semantic Scene Completion Toward Assisting Visually Impaired", "abstract": "Independent indoor mobility remains a critical challenge for individuals with visual impairments, largely due to the limited capability of existing assistive systems in detecting fine-grained hazardous objects such as chairs, tables, and small obstacles. These perceptual blind zones substantially increase the risk of collision in unfamiliar environments. To bridge the gap between monocular 3D vision research and practical assistive deployment, this paper proposes an Adaptive Multi-scale Attention Aggregation (AMAA) framework for monocular 3D semantic scene completion using only a wearable RGB camera. The proposed framework addresses two major limitations in 2D-to-3D feature lifting: noise diffusion during back-projection and structural instability in multi-scale fusion. A parallel channel--spatial attention mechanism is introduced to recalibrate lifted features along semantic and geometric dimensions, while a hierarchical adaptive gating strategy regulates cross-scale information flow to preserve fine-grained structural details. Experiments on the NYUv2 benchmark demonstrate that AMAA achieves an overall mIoU of 27.88%. Crucially, it yields significant relative improvements of 16.9% for small objects and 10.4% for tables over the MonoScene baseline. Furthermore, a wearable prototype based on an NVIDIA Jetson Orin NX and a ZED~2i camera validates stable real-time performance in indoor environments, demonstrating the feasibility of deploying monocular 3D scene completion for assistive navigation.", "date": "2026-02-18", "categories": []}
{"id": "2604.11797", "title": "SyncFix: Fixing 3D Reconstructions via Multi-View Synchronization", "abstract": "We present SyncFix, a framework that enforces cross-view consistency during the diffusion-based refinement of reconstructed scenes. SyncFix formulates refinement as a joint latent bridge matching problem, synchronizing distorted and clean representations across multiple views to fix the semantic and geometric inconsistencies. This means SyncFix learns a joint conditional over multiple views to enforce consistency throughout the denoising trajectory. Our training is done only on image pairs, but it generalizes naturally to an arbitrary number of views during inference. Moreover, reconstruction quality improves with additional views, with diminishing returns at higher view counts. Qualitative and quantitative results demonstrate that SyncFix consistently generates high-quality reconstructions and surpasses current state-of-the-art baselines, even in the absence of clean reference images. SyncFix achieves even higher fidelity when sparse references are available.", "date": "2026-04-13", "categories": []}
{"id": "2604.13235", "title": "Neural 3D Reconstruction of Planetary Surfaces from Descent-Phase Wide-Angle Imagery", "abstract": "Digital elevation modeling of planetary surfaces is essential for studying past and ongoing geological processes. Wide-angle imagery acquired during spacecraft descent promises to offer a low-cost option for high-resolution terrain reconstruction. However, accurate 3D reconstruction from such imagery is challenging due to strong radial distortion and limited parallax from vertically descending, predominantly nadir-facing cameras. Conventional multi-view stereo exhibits limited depth range and reduced fidelity under these conditions and also lacks domain-specific priors. We present the first study of modern neural reconstruction methods for planetary descent imaging. We also develop a novel approach that incorporates an explicit neural height field representation, which provides a strong prior since planetary surfaces are generally continuous, smooth, solid, and free from floating objects. This study demonstrates that neural approaches offer a strong and competitive alternative to traditional multi-view stereo (MVS) methods. Experiments on simulated descent sequences over high-fidelity lunar and Mars terrains demonstrate that the proposed approach achieves increased spatial coverage while maintaining satisfactory estimation accuracy.", "date": "2026-04-14", "categories": []}
{"id": "2604.13746", "title": "ClipGStream: Clip-Stream Gaussian Splatting for Any Length and Any Motion Multi-View Dynamic Scene Reconstruction", "abstract": "Dynamic 3D scene reconstruction is essential for immersive media such as VR, MR, and XR, yet remains challenging for long multi-view sequences with large-scale motion. Existing dynamic Gaussian approaches are either Frame-Stream, offering scalability but poor temporal stability, or Clip, achieving local consistency at the cost of high memory and limited sequence length. We propose ClipGStream, a hybrid reconstruction framework that performs stream optimization at the clip level rather than the frame level. The sequence is divided into short clips, where dynamic motion is modeled using clip-independent spatio-temporal fields and residual anchor compensation to capture local variations efficiently, while inter-clip inherited anchors and decoders maintain structural consistency across clips. This Clip-Stream design enables scalable, flicker-free reconstruction of long dynamic videos with high temporal coherence and reduced memory overhead. Extensive experiments demonstrate that ClipGStream achieves state-of-the-art reconstruction quality and efficiency. The project page is available at: https://liangjie1999.github.io/ClipGStreamWeb/", "date": "2026-04-15", "categories": []}
{"id": "2604.14025", "title": "Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective", "abstract": "Reconstructing 3D representations from 2D inputs is a fundamental task in computer vision and graphics, serving as a cornerstone for understanding and interacting with the physical world. While traditional methods achieve high fidelity, they are limited by slow per-scene optimization or category-specific training, which hinders their practical deployment and scalability. Hence, generalizable feed-forward 3D reconstruction has witnessed rapid development in recent years. By learning a model that maps images directly to 3D representations in a single forward pass, these methods enable efficient reconstruction and robust cross-scene generalization. Our survey is motivated by a critical observation: despite the diverse geometric output representations, ranging from implicit fields to explicit primitives, existing feed-forward approaches share similar high-level architectural patterns, such as image feature extraction backbones, multi-view information fusion mechanisms, and geometry-aware design principles. Consequently, we abstract away from these representation differences and instead focus on model design, proposing a novel taxonomy centered on model design strategies that are agnostic to the output format. Our proposed taxonomy organizes the research directions into five key problems that drive recent research development: feature enhancement, geometry awareness, model efficiency, augmentation strategies and temporal-aware models. To support this taxonomy with empirical grounding and standardized evaluation, we further comprehensively review related benchmarks and datasets, and extensively discuss and categorize real-world applications based on feed-forward 3D models. Finally, we outline future directions to address open challenges such as scalability, evaluation standards, and world modeling.", "date": "2026-04-15", "categories": []}
{"id": "2604.14048", "title": "Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself", "abstract": "Feed-forward 3D reconstruction models are efficient but rigid: once trained, they perform inference in a zero-shot manner and cannot adapt to the test scene. As a result, visually plausible reconstructions often contain errors, particularly under occlusions, specularities, and ambiguous cues. To address this, we introduce Free Geometry, a framework that enables feed-forward 3D reconstruction models to self-evolve at test time without any 3D ground truth. Our key insight is that, when the model receives more views, it produces more reliable and view-consistent reconstructions. Leveraging this property, given a testing sequence, we mask a subset of frames to construct a self-supervised task. Free Geometry enforces cross-view feature consistency between representations from full and partial observations, while maintaining the pairwise relations implied by the held-out frames. This self-supervision allows for fast recalibration via lightweight LoRA updates, taking less than 2 minutes per dataset on a single GPU. Our approach consistently improves state-of-the-art foundation models, including Depth Anything 3 and VGGT, across 4 benchmark datasets, yielding an average improvement of 3.73% in camera pose accuracy and 2.88% in point map prediction. Code is available at https://github.com/hiteacherIamhumble/Free-Geometry .", "date": "2026-04-15", "categories": []}
{"id": "2602.15355", "title": "DAV-GSWT: Diffusion-Active-View Sampling for Data-Efficient Gaussian Splatting Wang Tiles", "abstract": "The emergence of 3D Gaussian Splatting has fundamentally redefined the capabilities of photorealistic neural rendering by enabling high-throughput synthesis of complex environments. While procedural methods like Wang Tiles have recently been integrated to facilitate the generation of expansive landscapes, these systems typically remain constrained by a reliance on densely sampled exemplar reconstructions. We present DAV-GSWT, a data-efficient framework that leverages diffusion priors and active view sampling to synthesize high-fidelity Gaussian Splatting Wang Tiles from minimal input observations. By integrating a hierarchical uncertainty quantification mechanism with generative diffusion models, our approach autonomously identifies the most informative viewpoints while hallucinating missing structural details to ensure seamless tile transitions. Experimental results indicate that our system significantly reduces the required data volume while maintaining the visual integrity and interactive performance necessary for large-scale virtual environments.", "date": "2026-02-17", "categories": []}
{"id": "2604.11992", "title": "ReefMapGS: Enabling Large-Scale Underwater Reconstruction by Closing the Loop Between Multimodal SLAM and Gaussian Splatting", "abstract": "3D Gaussian Splatting is a powerful visual representation, providing high-quality and efficient 3D scene reconstruction, but it is crucially dependent on accurate camera poses typically obtained from computationally intensive processes like structure-from-motion that are unsuitable for field robot applications. However, in these domains, multimodal sensor data from acoustic, inertial, pressure, and visual sensors are available and suitable for pose-graph optimization-based SLAM methods that can estimate the vehicle's trajectory and thus our needed camera poses while providing uncertainty. We propose a 3DGS-based incremental reconstruction framework, ReefMapGS, that builds an initial model from a high certainty region and progressively expands to incorporate the whole scene. We reconstruct the scene incrementally by interleaving local tracking of new image observations with optimization of the underlying 3DGS scene. These refined poses are integrated back into the pose-graph to globally optimize the whole trajectory. We show COLMAP-free 3D reconstruction of two underwater reef sites with complex geometry as well as more accurate global pose estimation of our AUV over survey trajectories spanning up to 700 m.", "date": "2026-04-13", "categories": []}
{"id": "2604.13036", "title": "Lyra 2.0: Explorable Generative 3D Worlds", "abstract": "Recent advances in video generation enable a new paradigm for 3D scene creation: generating camera-controlled videos that simulate scene walkthroughs, then lifting them to 3D via feed-forward reconstruction techniques. This generative reconstruction approach combines the visual fidelity and creative capacity of video models with 3D outputs ready for real-time rendering and simulation. Scaling to large, complex environments requires 3D-consistent video generation over long camera trajectories with large viewpoint changes and location revisits, a setting where current video models degrade quickly. Existing methods for long-horizon generation are fundamentally limited by two forms of degradation: spatial forgetting and temporal drifting. As exploration proceeds, previously observed regions fall outside the model's temporal context, forcing the model to hallucinate structures when revisited. Meanwhile, autoregressive generation accumulates small synthesis errors over time, gradually distorting scene appearance and geometry. We present Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale. To address spatial forgetting, we maintain per-frame 3D geometry and use it solely for information routing -- retrieving relevant past frames and establishing dense correspondences with the target viewpoints -- while relying on the generative prior for appearance synthesis. To address temporal drifting, we train with self-augmented histories that expose the model to its own degraded outputs, teaching it to correct drift rather than propagate it. Together, these enable substantially longer and 3D-consistent video trajectories, which we leverage to fine-tune feed-forward reconstruction models that reliably recover high-quality 3D scenes.", "date": "2026-04-14", "categories": []}
{"id": "2603.28287", "title": "TerraSky3D: Multi-View Reconstructions of European Landmarks in 4K", "abstract": "Despite the growing need for data of more and more sophisticated 3D reconstruction pipelines, we can still observe a scarcity of suitable public datasets. Existing 3D datasets are either low resolution, limited to a small amount of scenes, based on images of varying quality because retrieved from the internet, or limited to specific capturing scenarios.\n  Motivated by this lack of suitable 3D datasets, we captured TerraSky3D, a high-resolution large-scale 3D reconstruction dataset comprising 50,000 images divided into 150 ground, aerial, and mixed scenes. The dataset focuses on European landmarks and comes with curated calibration data, camera poses, and depth maps. TerraSky3D tries to answer the need for challenging dataset that can be used to train and evaluate 3D reconstruction-related pipelines.", "date": "2026-03-30", "categories": []}
{"id": "2604.02003", "title": "ProDiG: Progressive Diffusion-Guided Gaussian Splatting for Aerial to Ground Reconstruction", "abstract": "Generating ground-level views and coherent 3D site models from aerial-only imagery is challenging due to extreme viewpoint changes, missing intermediate observations, and large scale variations. Existing methods either refine renderings post-hoc, often producing geometrically inconsistent results, or rely on multi-altitude ground-truth, which is rarely available. Gaussian Splatting and diffusion-based refinements improve fidelity under small variations but fail under wide aerial-toground gaps. To address these limitations, we introduce ProDiG (Progressive Diffusion-Guided Gaussian Splatting for Aerial to Ground Reconstruction), a diffusionguided framework that progressively transforms aerial 3D representations toward ground-level fidelity. ProDiG synthesizes intermediate-altitude views and refines the Gaussian representation at each stage using a geometry-aware causal attention module that injects epipolar structure into reference-view diffusion. A distance-adaptive Gaussian module dynamically adjusts Gaussian scale and opacity based on camera distance, ensuring stable reconstruction across large viewpoint gaps. Together, these components enable progressive, geometrically grounded refinement without requiring additional ground-truth viewpoints. Extensive experiments on synthetic and real-world datasets demonstrate that ProDiG produces visually realistic ground-level renderings and coherent 3D geometry, significantly outperforming existing approaches in terms of visual quality, geometric consistency, and robustness to extreme viewpoint changes. Project Page: https://sirsh07.github.io/research/prodig", "date": "2026-04-02", "categories": []}
{"id": "2604.09862", "title": "FF3R: Feedforward Feature 3D Reconstruction from Unconstrained views", "abstract": "Recent advances in vision foundation models have revolutionized geometry reconstruction and semantic understanding. Yet, most of the existing approaches treat these capabilities in isolation, leading to redundant pipelines and compounded errors. This paper introduces FF3R, a fully annotation-free feed-forward framework that unifies geometric and semantic reasoning from unconstrained multi-view image sequences. Unlike previous methods, FF3R does not require camera poses, depth maps, or semantic labels, relying solely on rendering supervision for RGB and feature maps, establishing a scalable paradigm for unified 3D reasoning. In addition, we address two critical challenges in feedforward feature reconstruction pipelines, namely global semantic inconsistency and local structural inconsistency, through two key innovations: (i) a Token-wise Fusion Module that enriches geometry tokens with semantic context via cross-attention, and (ii) a Semantic-Geometry Mutual Boosting mechanism combining geometry-guided feature warping for global consistency with semantic-aware voxelization for local coherence. Extensive experiments on ScanNet and DL3DV-10K demonstrate FF3R's superior performance in novel-view synthesis, open-vocabulary semantic segmentation, and depth estimation, with strong generalization to in-the-wild scenarios, paving the way for embodied intelligence systems that demand both spatial and semantic understanding.", "date": "2026-04-10", "categories": []}
{"id": "2604.10246", "title": "A Comparison of Multi-View Stereo Methods for Photogrammetric 3D Reconstruction: From Traditional to Learning-Based Approaches", "abstract": "Photogrammetric 3D reconstruction has long relied on traditional Structure-from-Motion (SfM) and Multi-View Stereo (MVS) methods, which provide high accuracy but face challenges in speed and scalability. Recently, learning-based MVS methods have emerged, aiming for faster and more efficient reconstruction. This work presents a comparative evaluation between a representative traditional MVS pipeline (COLMAP) and state-of-the-art learning-based approaches, including geometry-guided methods (MVSNet, PatchmatchNet, MVSAnywhere, MVSFormer++) and end-to-end frameworks (Stereo4D, FoundationStereo, DUSt3R, MASt3R, Fast3R, VGGT). Two experiments were conducted on different aerial scenarios. The first experiment used the MARS-LVIG dataset, where ground-truth 3D reconstruction was provided by LiDAR point clouds. The second experiment used a public scene from the Pix4D official website, with ground truth generated by Pix4Dmapper. We evaluated accuracy, coverage, and runtime across all methods. Experimental results show that although COLMAP can provide reliable and geometrically consistent reconstruction results, it requires more computation time. In cases where traditional methods fail in image registration, learning-based approaches exhibit stronger feature-matching capability and greater robustness. Geometry-guided methods usually require careful dataset preparation and often depend on camera pose or depth priors generated by COLMAP. End-to-end methods such as DUSt3R and VGGT achieve competitive accuracy and reasonable coverage while offering substantially faster reconstruction. However, they exhibit relatively large residuals in 3D reconstruction, particularly in challenging scenarios.", "date": "2026-04-11", "categories": []}
{"id": "2604.10789", "title": "ReplicateAnyScene: Zero-Shot Video-to-3D Composition via Textual-Visual-Spatial Alignment", "abstract": "Humans exhibit an innate capacity to rapidly perceive and segment objects from video observations, and even mentally assemble them into structured 3D scenes. Replicating such capability, termed compositional 3D reconstruction, is pivotal for the advancement of Spatial Intelligence and Embodied AI. However, existing methods struggle to achieve practical deployment due to the insufficient integration of cross-modal information, leaving them dependent on manual object prompting, reliant on auxiliary visual inputs, and restricted to overly simplistic scenes by training biases. To address these limitations, we propose ReplicateAnyScene, a framework capable of fully automated and zero-shot transformation of casually captured videos into compositional 3D scenes. Specifically, our pipeline incorporates a five-stage cascade to extract and structurally align generic priors from vision foundation models across textual, visual, and spatial dimensions, grounding them into structured 3D representations and ensuring semantic coherence and physical plausibility of the constructed scenes. To facilitate a more comprehensive evaluation of this task, we further introduce the C3DR benchmark to assess reconstruction quality from diverse aspects. Extensive experiments demonstrate the superiority of our method over existing baselines in generating high-quality compositional 3D scenes.", "date": "2026-04-12", "categories": []}
{"id": "2604.11142", "title": "Naka-GS: A Bionics-inspired Dual-Branch Naka Correction and Progressive Point Pruning for Low-Light 3DGS", "abstract": "Low-light conditions severely hinder 3D restoration and reconstruction by degrading image visibility, introducing color distortions, and contaminating geometric priors for downstream optimization. We present NAKA-GS, a bionics-inspired framework for low-light 3D Gaussian Splatting that jointly improves photometric restoration and geometric initialization. Our method starts with a Naka-guided chroma-correction network, which combines physics-prior low-light enhancement, dual-branch input modeling, frequency-decoupled correction, and mask-guided optimization to suppress bright-region chromatic artifacts and edge-structure errors. The enhanced images are then fed into a feed-forward multi-view reconstruction model to produce dense scene priors. To further improve Gaussian initialization, we introduce a lightweight Point Preprocessing Module (PPM) that performs coordinate alignment, voxel pooling, and distance-adaptive progressive pruning to remove noisy and redundant points while preserving representative structures. Without introducing heavy inference overhead, NAKA-GS improves restoration quality, training stability, and optimization efficiency for low-light 3D reconstruction. The proposed method was presented in the NTIRE 3D Restoration and Reconstruction (3DRR) Challenge, and outperformed the baseline methods by a large margin. The code is available at https://github.com/RunyuZhu/Naka-GS", "date": "2026-04-13", "categories": []}
{"id": "2604.11401", "title": "GS4City: Hierarchical Semantic Gaussian Splatting via City-Model Priors", "abstract": "Recent semantic 3D Gaussian Splatting (3DGS) methods primarily rely on 2D foundation models, often yielding ambiguous boundaries and limited support for structured urban semantics. While city models such as CityGML encode hierarchically organized semantics together with building geometry, these labels cannot be directly mapped to Gaussian primitives. We present GS4City, a hierarchical semantic Gaussian Splatting method that incorporates city-model priors for urban scene understanding. GS4City derives reliable image-aligned masks from Level of Detail (LoD) 3 CityGML models via two-pass raycasting, explicitly using parent-child relations to validate and recover fine-grained facade elements. It then fuses these geometry-grounded masks with foundation-model predictions to establish scene-consistent instance correspondences, and learns a compact identity encoding for each Gaussian under joint 2D identity supervision and 3D spatial regularization. Experiments on the TUM2TWIN and Gold Coast datasets show that GS4City effectively incorporates structured building semantics into Gaussian scene representations, outperforming existing 2D-driven semantic 3DGS baselines, including LangSplat and Gaga, by up to 15.8 IoU points in coarse building segmentation and 14.2 mIoU points in fine-grained semantic segmentation. By bridging structured city models and photorealistic Gaussian scene representations, GS4City enables semantically queryable and structure-aware urban reconstruction. Code is available at https://github.com/Jinyzzz/GS4City.", "date": "2026-04-13", "categories": []}
{"id": "2510.01767", "title": "LoBE-GS: Load-Balanced and Efficient 3D Gaussian Splatting for Large-Scale Scene Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has established itself as an efficient representation for real-time, high-fidelity 3D scene reconstruction. However, scaling 3DGS to large and unbounded scenes such as city blocks remains difficult. Existing divide-and-conquer methods alleviate memory pressure by partitioning the scene into blocks and training on multiple, non-communicating GPUs, but introduce new bottlenecks: (i) partitions suffer from severe load imbalance since uniform or heuristic splits do not reflect actual computational demands, and (ii) coarse-to-fine pipelines fail to exploit the coarse stage efficiently, often reloading the entire model and incurring high overhead. In this work, we introduce LoBE-GS, a novel Load-Balanced and Efficient 3D Gaussian Splatting framework, that re-engineers the large-scale 3DGS pipeline. Specifically, LoBE-GS introduces a load-balanced KD-tree scene partitioning scheme with optimized cutlines that balance per-block camera counts. To accelerate preprocessing, it employs depth-based back-projection for fast camera assignment, reducing processing time from hours to minutes. It further reduces training cost through two lightweight techniques: visibility cropping and selective densification. Evaluations on large-scale urban and outdoor datasets show that LoBE-GS consistently achieves up to 2 times faster end-to-end training time than state-of-the-art baselines, while maintaining reconstruction quality and enabling scalability to scenes infeasible with vanilla 3DGS.", "date": "2025-10-02", "categories": []}
{"id": "2603.27222", "title": "HD-VGGT: High-Resolution Visual Geometry Transformer", "abstract": "High-resolution imagery is essential for accurate 3D reconstruction, as many geometric details only emerge at fine spatial scales. Recent feed-forward approaches, such as the Visual Geometry Grounded Transformer (VGGT), have demonstrated the ability to infer scene geometry from large collections of images in a single forward pass. However, scaling these models to high-resolution inputs remains challenging: the number of tokens in transformer architectures grows rapidly with both image resolution and the number of views, leading to prohibitive computational and memory costs. Moreover, we observe that visually ambiguous regions, such as repetitive patterns, weak textures, or specular surfaces, often produce unstable feature tokens that degrade geometric inference, especially at higher resolutions. We introduce HD-VGGT, a dual-branch architecture for efficient and robust high-resolution 3D reconstruction. A low-resolution branch predicts a coarse, globally consistent geometry, while a high-resolution branch refines details via a learned feature upsampling module. To handle unstable tokens, we propose Feature Modulation, which suppresses unreliable features early in the transformer. HD-VGGT leverages high-resolution images and supervision without full-resolution transformer costs, achieving state-of-the-art reconstruction quality.", "date": "2026-03-28", "categories": []}
{"id": "2604.09366", "title": "Robust 4D Visual Geometry Transformer with Uncertainty-Aware Priors", "abstract": "Reconstructing dynamic 4D scenes is an important yet challenging task. While 3D foundation models like VGGT excel in static settings, they often struggle with dynamic sequences where motion causes significant geometric ambiguity. To address this, we present a framework designed to disentangle dynamic and static components by modeling uncertainty across different stages of the reconstruction process. Our approach introduces three synergistic mechanisms: (1) Entropy-Guided Subspace Projection, which leverages information-theoretic weighting to adaptively aggregate multi-head attention distributions, effectively isolating dynamic motion cues from semantic noise; (2) Local-Consistency Driven Geometry Purification, which enforces spatial continuity via radius-based neighborhood constraints to eliminate structural outliers; and (3) Uncertainty-Aware Cross-View Consistency, which formulates multi-view projection refinement as a heteroscedastic maximum likelihood estimation problem, utilizing depth confidence as a probabilistic weight. Experiments on dynamic benchmarks show that our approach outperforms current state-of-the-art methods, reducing Mean Accuracy error by 13.43\\% and improving segmentation F-measure by 10.49\\%. Our framework maintains the efficiency of feed-forward inference and requires no task-specific fine-tuning or per-scene optimization.", "date": "2026-04-10", "categories": []}
{"id": "2604.09478", "title": "Incremental Semantics-Aided Meshing from LiDAR-Inertial Odometry and RGB Direct Label Transfer", "abstract": "Geometric high-fidelity mesh reconstruction from LiDAR-inertial scans remains challenging in large, complex indoor environments -- such as cultural buildings -- where point cloud sparsity, geometric drift, and fixed fusion parameters produce holes, over-smoothing, and spurious surfaces at structural boundaries. We propose a modular, incremental RGB+LiDAR pipeline that generates incremental semantics-aided high-quality meshes from indoor scans through scan frame-based direct label transfer. A vision foundation model labels each incoming RGB frame; labels are incrementally projected and fused onto a LiDAR-inertial odometry map; and an incremental semantics-aware Truncated Signed Distance Function (TSDF) fusion step produces the final mesh via marching cubes. This frame-level fusion strategy preserves the geometric fidelity of LiDAR while leveraging rich visual semantics to resolve geometric ambiguities at reconstruction boundaries caused by LiDAR point-cloud sparsity and geometric drift. We demonstrate that semantic guidance improves geometric reconstruction quality; quantitative evaluation is therefore performed using geometric metrics on the Oxford Spires dataset, while results from the NTU VIRAL dataset are analyzed qualitatively. The proposed method outperforms state-of-the-art geometric baselines ImMesh and Voxblox, demonstrating the benefit of semantics-aided fusion for geometric mesh quality. The resulting semantically labelled meshes are of value when reconstructing Universal Scene Description (USD) assets, offering a path from indoor LiDAR scanning to XR and digital modeling.", "date": "2026-04-10", "categories": []}
{"id": "2604.09480", "title": "Online3R: Online Learning for Consistent Sequential Reconstruction Based on Geometry Foundation Model", "abstract": "We present Online3R, a new sequential reconstruction framework that is capable of adapting to new scenes through online learning, effectively resolving inconsistency issues. Specifically, we introduce a set of learnable lightweight visual prompts into a pretrained, frozen geometry foundation model to capture the knowledge of new environments while preserving the fundamental capability of the foundation model for geometry prediction. To solve the problems of missing groundtruth and the requirement of high efficiency when updating these visual prompts at test time, we introduce a local-global self-supervised learning strategy by enforcing the local and global consistency constraints on predictions. The local consistency constraints are conducted on intermediate and previously local fused results, enabling the model to be trained with high-quality pseudo groundtruth signals; the global consistency constraints are operated on sparse keyframes spanning long distances rather than per frame, allowing the model to learn from a consistent prediction over a long trajectory in an efficient way. Our experiments demonstrate that Online3R outperforms previous state-of-the-art methods on various benchmarks. Project page: https://shunkaizhou.github.io/online3r-1.0/", "date": "2026-04-10", "categories": []}
{"id": "2401.03890", "title": "A Survey on 3D Gaussian Splatting", "abstract": "3D Gaussian splatting (GS) has emerged as a transformative technique in radiance fields. Unlike mainstream implicit neural models, 3D GS uses millions of learnable 3D Gaussians for an explicit scene representation. Paired with a differentiable rendering algorithm, this approach achieves real-time rendering and unprecedented editability, making it a potential game-changer for 3D reconstruction and representation. In the present paper, we provide the first systematic overview of the recent developments and critical contributions in 3D GS. We begin with a detailed exploration of the underlying principles and the driving forces behind the emergence of 3D GS, laying the groundwork for understanding its significance. A focal point of our discussion is the practical applicability of 3D GS. By enabling unprecedented rendering speed, 3D GS opens up a plethora of applications, ranging from virtual reality to interactive media and beyond. This is complemented by a comparative analysis of leading 3D GS models, evaluated across various benchmark tasks to highlight their performance and practical utility. The survey concludes by identifying current challenges and suggesting potential avenues for future research. Through this survey, we aim to provide a valuable resource for both newcomers and seasoned researchers, fostering further exploration and advancement in explicit radiance field.", "date": "2024-01-08", "categories": []}
{"id": "2603.04385", "title": "ZipMap: Linear-Time Stateful 3D Reconstruction via Test-Time Training", "abstract": "Feed-forward transformer models have driven rapid progress in 3D vision, but state-of-the-art methods such as VGGT and $\\pi^3$ have a computational cost that scales quadratically with the number of input images, making them inefficient when applied to large image collections. Sequential-reconstruction approaches reduce this cost but sacrifice reconstruction quality. We introduce ZipMap, a stateful feed-forward model that achieves linear-time, bidirectional 3D reconstruction while matching or surpassing the accuracy of quadratic-time methods. ZipMap employs test-time training layers to zip an entire image collection into a compact hidden scene state in a single forward pass, enabling reconstruction of over 700 frames in under 10 seconds on a single H100 GPU, more than $20\\times$ faster than state-of-the-art methods such as VGGT. Moreover, we demonstrate the benefits of having a stateful representation in real-time scene-state querying and its extension to sequential streaming reconstruction.", "date": "2026-03-04", "categories": []}
{"id": "2604.07053", "title": "AnchorSplat: Feed-Forward 3D Gaussian Splatting with 3D Geometric Priors", "abstract": "Recent feed-forward Gaussian reconstruction models adopt a pixel-aligned formulation that maps each 2D pixel to a 3D Gaussian, entangling Gaussian representations tightly with the input images. In this paper, we propose AnchorSplat, a novel feed-forward 3DGS framework for scene-level reconstruction that represents the scene directly in 3D space. AnchorSplat introduces an anchor-aligned Gaussian representation guided by 3D geometric priors (e.g., sparse point clouds, voxels, or RGB-D point clouds), enabling a more geometry-aware renderable 3D Gaussians that is independent of image resolution and number of views. This design substantially reduces the number of required Gaussians, improving computational efficiency while enhancing reconstruction fidelity. Beyond the anchor-aligned design, we utilize a Gaussian Refiner to adjust the intermediate Gaussiansy via merely a few forward passes. Experiments on the ScanNet++ v2 NVS benchmark demonstrate the SOTA performance, outperforming previous methods with more view-consistent and substantially fewer Gaussian primitives.", "date": "2026-04-08", "categories": []}
{"id": "2604.07882", "title": "ReconPhys: Reconstruct Appearance and Physical Attributes from Single Video", "abstract": "Reconstructing non-rigid objects with physical plausibility remains a significant challenge. Existing approaches leverage differentiable rendering for per-scene optimization, recovering geometry and dynamics but requiring expensive tuning or manual annotation, which limits practicality and generalizability. To address this, we propose ReconPhys, the first feedforward framework that jointly learns physical attribute estimation and 3D Gaussian Splatting reconstruction from a single monocular video. Our method employs a dual-branch architecture trained via a self-supervised strategy, eliminating the need for ground-truth physics labels. Given a video sequence, ReconPhys simultaneously infers geometry, appearance, and physical attributes. Experiments on a large-scale synthetic dataset demonstrate superior performance: our method achieves 21.64 PSNR in future prediction compared to 13.27 by state-of-the-art optimization baselines, while reducing Chamfer Distance from 0.349 to 0.004. Crucially, ReconPhys enables fast inference (<1 second) versus hours required by existing methods, facilitating rapid generation of simulation-ready assets for robotics and graphics.", "date": "2026-04-09", "categories": []}
{"id": "2604.07923", "title": "Stitch4D: Sparse Multi-Location 4D Urban Reconstruction via Spatio-Temporal Interpolation", "abstract": "Dynamic urban environments are often captured by cameras placed at spatially separated locations with little or no view overlap. However, most existing 4D reconstruction methods assume densely overlapping views. When applied to such sparse observations, these methods fail to reconstruct intermediate regions and often introduce temporal artifacts. To address this practical yet underexplored sparse multi-location setting, we propose Stitch4D, a unified 4D reconstruction framework that explicitly compensates for missing spatial coverage in sparse observations. Stitch4D (i) synthesizes intermediate bridge views to densify spatial constraints and improve spatial coverage, and (ii) jointly optimizes real and synthesized observations within a unified coordinate frame under explicit inter-location consistency constraints. By restoring intermediate coverage before optimization, Stitch4D prevents geometric collapse and reconstructs coherent geometry and smooth scene dynamics even in sparsely observed environments. To evaluate this setting, we introduce Urban Sparse 4D (U-S4D), a CARLA-based benchmark designed to assess spatiotemporal alignment under sparse multi-location configurations. Experimental results on U-S4D show that Stitch4D surpasses representative 4D reconstruction baselines and achieves superior visual quality. These results indicate that recovering intermediate spatial coverage is essential for stable 4D reconstruction in sparse urban environments.", "date": "2026-04-09", "categories": []}
{"id": "2604.07986", "title": "DP-DeGauss: Dynamic Probabilistic Gaussian Decomposition for Egocentric 4D Scene Reconstruction", "abstract": "Egocentric video is crucial for next-generation 4D scene reconstruction, with applications in AR/VR and embodied AI. However, reconstructing dynamic first-person scenes is challenging due to complex ego-motion, occlusions, and hand-object interactions. Existing decomposition methods are ill-suited, assuming fixed viewpoints or merging dynamics into a single foreground. To address these limitations, we introduce DP-DeGauss, a dynamic probabilistic Gaussian decomposition framework for egocentric 4D reconstruction. Our method initializes a unified 3D Gaussian set from COLMAP priors, augments each with a learnable category probability, and dynamically routes them into specialized deformation branches for background, hands, or object modeling. We employ category-specific masks for better disentanglement and introduce brightness and motion-flow control to improve static rendering and dynamic reconstruction. Extensive experiments show that DP-DeGauss outperforms baselines by +1.70dB in PSNR on average with SSIM and LPIPS gains. More importantly, our framework achieves the first and state-of-the-art disentanglement of background, hand, and object components, enabling explicit, fine-grained separation, paving the way for more intuitive ego scene understanding and editing.", "date": "2026-04-09", "categories": []}
{"id": "2604.08370", "title": "SurfelSplat: Learning Efficient and Generalizable Gaussian Surfel Representations for Sparse-View Surface Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated impressive performance in 3D scene reconstruction. Beyond novel view synthesis, it shows great potential for multi-view surface reconstruction. Existing methods employ optimization-based reconstruction pipelines that achieve precise and complete surface extractions. However, these approaches typically require dense input views and high time consumption for per-scene optimization. To address these limitations, we propose SurfelSplat, a feed-forward framework that generates efficient and generalizable pixel-aligned Gaussian surfel representations from sparse-view images. We observe that conventional feed-forward structures struggle to recover accurate geometric attributes of Gaussian surfels because the spatial frequency of pixel-aligned primitives exceeds Nyquist sampling rates. Therefore, we propose a cross-view feature aggregation module based on the Nyquist sampling theorem. Specifically, we first adapt the geometric forms of Gaussian surfels with spatial sampling rate-guided low-pass filters. We then project the filtered surfels across all input views to obtain cross-view feature correlations. By processing these correlations through a specially designed feature fusion network, we can finally regress Gaussian surfels with precise geometry. Extensive experiments on DTU reconstruction benchmarks demonstrate that our model achieves comparable results with state-of-the-art methods, and predict Gaussian surfels within 1 second, offering a 100x speedup without costly per-scene training.", "date": "2026-04-09", "categories": []}
{"id": "2604.08532", "title": "Self-Improving 4D Perception via Self-Distillation", "abstract": "Large-scale multi-view reconstruction models have made remarkable progress, but most existing approaches still rely on fully supervised training with ground-truth 3D/4D annotations. Such annotations are expensive and particularly scarce for dynamic scenes, limiting scalability. We propose SelfEvo, a self-improving framework that continually improves pretrained multi-view reconstruction models using unlabeled videos. SelfEvo introduces a self-distillation scheme using spatiotemporal context asymmetry, enabling self-improvement for learning-based 4D perception without external annotations. We systematically study design choices that make self-improvement effective, including loss signals, forms of asymmetry, and other training strategies. Across eight benchmarks spanning diverse datasets and domains, SelfEvo consistently improves pretrained baselines and generalizes across base models (e.g. VGGT and $\\pi^3$), with significant gains on dynamic scenes. Overall, SelfEvo achieves up to 36.5% relative improvement in video depth estimation and 20.1% in camera estimation, without using any labeled data. Project Page: https://self-evo.github.io/.", "date": "2026-04-09", "categories": []}
{"id": "2604.08542", "title": "Scal3R: Scalable Test-Time Training for Large-Scale 3D Reconstruction", "abstract": "This paper addresses the task of large-scale 3D scene reconstruction from long video sequences. Recent feed-forward reconstruction models have shown promising results by directly regressing 3D geometry from RGB images without explicit 3D priors or geometric constraints. However, these methods often struggle to maintain reconstruction accuracy and consistency over long sequences due to limited memory capacity and the inability to effectively capture global contextual cues. In contrast, humans can naturally exploit the global understanding of the scene to inform local perception. Motivated by this, we propose a novel neural global context representation that efficiently compresses and retains long-range scene information, enabling the model to leverage extensive contextual cues for enhanced reconstruction accuracy and consistency. The context representation is realized through a set of lightweight neural sub-networks that are rapidly adapted during test time via self-supervised objectives, which substantially increases memory capacity without incurring significant computational overhead. The experiments on multiple large-scale benchmarks, including the KITTI Odometry~\\cite{Geiger2012CVPR} and Oxford Spires~\\cite{tao2025spires} datasets, demonstrate the effectiveness of our approach in handling ultra-large scenes, achieving leading pose accuracy and state-of-the-art 3D reconstruction accuracy while maintaining efficiency. Code is available at https://zju3dv.github.io/scal3r.", "date": "2026-04-09", "categories": []}
{"id": "2512.07527", "title": "From Orbit to Ground: Generative City Photogrammetry from Extreme Off-Nadir Satellite Images", "abstract": "City-scale 3D reconstruction from satellite imagery presents the challenge of extreme viewpoint extrapolation, where our goal is to synthesize ground-level novel views from sparse orbital images with minimal parallax. This requires inferring nearly $90^\\circ$ viewpoint gaps from image sources with severely foreshortened facades and flawed textures, causing state-of-the-art reconstruction engines such as NeRF and 3DGS to fail. To address this problem, we propose two design choices tailored for city structures and satellite inputs. First, we model city geometry as a 2.5D height map, implemented as a Z-monotonic signed distance field (SDF) that matches urban building layouts from top-down viewpoints. This stabilizes geometry optimization under sparse, off-nadir satellite views and yields a watertight mesh with crisp roofs and clean, vertically extruded facades. Second, we paint the mesh appearance from satellite images via differentiable rendering techniques. While the satellite inputs may contain long-range, blurry captures, we further train a generative texture restoration network to enhance the appearance, recovering high-frequency, plausible texture details from degraded inputs. Our method's scalability and robustness are demonstrated through extensive experiments on large-scale urban reconstruction. For example, in our teaser figure, we reconstruct a $4\\,\\mathrm{km}^2$ real-world region from only a few satellite images, achieving state-of-the-art performance in synthesizing photorealistic ground views. The resulting models are not only visually compelling but also serve as high-fidelity, application-ready assets for downstream tasks like urban planning and simulation. Project page can be found at https://pku-vcl-geometry.github.io/Orbit2Ground/.", "date": "2025-12-08", "categories": []}
{"id": "2603.20284", "title": "STAC: Plug-and-Play Spatio-Temporal Aware Cache Compression for Streaming 3D Reconstruction", "abstract": "Online 3D reconstruction from streaming inputs requires both long-term temporal consistency and efficient memory usage. Although causal variants of VGGT address this challenge through a key-value (KV) cache mechanism, the cache grows linearly with the stream length, creating a major memory bottleneck. Under limited memory budgets, early cache eviction significantly degrades reconstruction quality and temporal consistency. In this work, we observe that attention in causal transformers for 3D reconstruction exhibits intrinsic spatio-temporal sparsity. Based on this insight, we propose STAC, a Spatio-Temporally Aware Cache Compression framework for streaming 3D reconstruction with large causal transformers. STAC consists of three key components: (1) a Working Temporal Token Caching mechanism that preserves long-term informative tokens using decayed cumulative attention scores; (2) a Long-term Spatial Token Caching scheme that compresses spatially redundant tokens into voxel-aligned representations for memory-efficient storage; and (3) a Chunk-based Multi-frame Optimization strategy that jointly processes consecutive frames to improve temporal coherence and GPU efficiency. Extensive experiments show that STAC achieves state-of-the-art reconstruction quality while reducing memory consumption by nearly 10x and accelerating inference by 4x, substantially improving the scalability of real-time 3D reconstruction in streaming settings.", "date": "2026-03-18", "categories": []}
{"id": "2604.02996", "title": "Rendering Multi-Human and Multi-Object with 3D Gaussian Splatting", "abstract": "Reconstructing dynamic scenes with multiple interacting humans and objects from sparse-view inputs is a critical yet challenging task, essential for creating high-fidelity digital twins for robotics and VR/AR. This problem, which we term Multi-Human Multi-Object (MHMO) rendering, presents two significant obstacles: achieving view-consistent representations for individual instances under severe mutual occlusion, and explicitly modeling the complex and combinatorial dependencies that arise from their interactions. To overcome these challenges, we propose MM-GS, a novel hierarchical framework built upon 3D Gaussian Splatting. Our method first employs a Per-Instance Multi-View Fusion module to establish a robust and consistent representation for each instance by aggregating visual information across all available views. Subsequently, a Scene-Level Instance Interaction module operates on a global scene graph to reason about relationships between all participants, refining their attributes to capture subtle interaction effects. Extensive experiments on challenging datasets demonstrate that our method significantly outperforms strong baselines, producing state-of-the-art results with high-fidelity details and plausible inter-instance contacts.", "date": "2026-04-03", "categories": []}
{"id": "2604.06739", "title": "DOC-GS: Dual-Domain Observation and Calibration for Reliable Sparse-View Gaussian Splatting", "abstract": "Sparse-view reconstruction with 3D Gaussian Splatting (3DGS) is fundamentally ill-posed due to insufficient geometric supervision, often leading to severe overfitting and the emergence of structural distortions and translucent haze-like artifacts. While existing approaches attempt to alleviate this issue via dropout-based regularization, they are largely heuristic and lack a unified understanding of artifact formation. In this paper, we revisit sparse-view 3DGS reconstruction from a new perspective and identify the core challenge as the unobservability of Gaussian primitive reliability. Unreliable Gaussians are insufficiently constrained during optimization and accumulate as haze-like degradations in rendered images. Motivated by this observation, we propose a unified Dual-domain Observation and Calibration (DOC-GS) framework that models and corrects Gaussian reliability through the synergy of optimization-domain inductive bias and observation-domain evidence. Specifically, in the optimization domain, we characterize Gaussian reliability by the degree to which each primitive is constrained during training, and instantiate this signal via a Continuous Depth-Guided Dropout (CDGD) strategy, where the dropout probability serves as an explicit proxy for primitive reliability. This imposes a smooth depth-aware inductive bias to suppress weakly constrained Gaussians and improve optimization stability. In the observation domain, we establish a connection between floater artifacts and atmospheric scattering, and leverage the Dark Channel Prior (DCP) as a structural consistency cue to identify and accumulate anomalous regions. Based on cross-view aggregated evidence, we further design a reliability-driven geometric pruning strategy to remove low-confidence Gaussians.", "date": "2026-04-08", "categories": []}
{"id": "2604.07010", "title": "Synthetic Dataset Generation for Partially Observed Indoor Objects", "abstract": "Learning-based methods for 3D scene reconstruction and object completion require large datasets containing partial scans paired with complete ground-truth geometry. However, acquiring such datasets using real-world scanning systems is costly and time-consuming, particularly when accurate ground truth for occluded regions is required.\n  In this work, we present a virtual scanning framework implemented in Unity for generating realistic synthetic 3D scan datasets. The proposed system simulates the behaviour of real-world scanners using configurable parameters such as scan resolution, measurement range, and distance-dependent noise. Instead of directly sampling mesh surfaces, the framework performs ray-based scanning from virtual viewpoints, enabling realistic modelling of sensor visibility and occlusion effects. In addition, panoramic images captured at the scanner location are used to assign colours to the resulting point clouds.\n  To support scalable dataset creation, the scanner is integrated with a procedural indoor scene generation pipeline that automatically produces diverse room layouts and furniture arrangements. Using this system, we introduce the \\textit{V-Scan} dataset, which contains synthetic indoor scans together with object-level partial point clouds, voxel-based occlusion grids, and complete ground-truth geometry. The resulting dataset provides valuable supervision for training and evaluating learning-based methods for scene reconstruction and object completion.", "date": "2026-04-08", "categories": []}
{"id": "2604.07279", "title": "Mem3R: Streaming 3D Reconstruction with Hybrid Memory via Test-Time Training", "abstract": "Streaming 3D perception is well suited to robotics and augmented reality, where long visual streams must be processed efficiently and consistently. Recent recurrent models offer a promising solution by maintaining fixed-size states and enabling linear-time inference, but they often suffer from drift accumulation and temporal forgetting over long sequences due to the limited capacity of compressed latent memories. We propose Mem3R, a streaming 3D reconstruction model with a hybrid memory design that decouples camera tracking from geometric mapping to improve temporal consistency over long sequences. For camera tracking, Mem3R employs an implicit fast-weight memory implemented as a lightweight Multi-Layer Perceptron updated via Test-Time Training. For geometric mapping, Mem3R maintains an explicit token-based fixed-size state. Compared with CUT3R, this design not only significantly improves long-sequence performance but also reduces the model size from 793M to 644M parameters. Mem3R supports existing improved plug-and-play state update strategies developed for CUT3R. Specifically, integrating it with TTT3R decreases Absolute Trajectory Error by up to 39% over the base implementation on 500 to 1000 frame sequences. The resulting improvements also extend to other downstream tasks, including video depth estimation and 3D reconstruction, while preserving constant GPU memory usage and comparable inference throughput. Project page: https://lck666666.github.io/Mem3R/", "date": "2026-04-08", "categories": []}
{"id": "2604.07350", "title": "Fast Spatial Memory with Elastic Test-Time Training", "abstract": "Large Chunk Test-Time Training (LaCT) has shown strong performance on long-context 3D reconstruction, but its fully plastic inference-time updates remain vulnerable to catastrophic forgetting and overfitting. As a result, LaCT is typically instantiated with a single large chunk spanning the full input sequence, falling short of the broader goal of handling arbitrarily long sequences in a single pass. We propose Elastic Test-Time Training inspired by elastic weight consolidation, that stabilizes LaCT fast-weight updates with a Fisher-weighted elastic prior around a maintained anchor state. The anchor evolves as an exponential moving average of past fast weights to balance stability and plasticity. Based on this updated architecture, we introduce Fast Spatial Memory (FSM), an efficient and scalable model for 4D reconstruction that learns spatiotemporal representations from long observation sequences and renders novel view-time combinations. We pre-trained FSM on large-scale curated 3D/4D data to capture the dynamics and semantics of complex spatial environments. Extensive experiments show that FSM supports fast adaptation over long sequences and delivers high-quality 3D/4D reconstruction with smaller chunks and mitigating the camera-interpolation shortcut. Overall, we hope to advance LaCT beyond the bounded single-chunk setting toward robust multi-chunk adaptation, a necessary step for generalization to genuinely longer sequences, while substantially alleviating the activation-memory bottleneck.", "date": "2026-04-08", "categories": []}
{"id": "2506.18601", "title": "BulletGen: Improving 4D Reconstruction with Bullet-Time Generation", "abstract": "Transforming casually captured, monocular videos into fully immersive dynamic experiences is a highly ill-posed task, and comes with significant challenges, e.g., reconstructing unseen regions, and dealing with the ambiguity in monocular depth estimation. In this work we introduce BulletGen, an approach that takes advantage of generative models to correct errors and complete missing information in a Gaussian-based dynamic scene representation. This is done by aligning the output of a diffusion-based video generation model with the 4D reconstruction at a single frozen \"bullet-time\" step. The generated frames are then used to supervise the optimization of the 4D Gaussian model. Our method seamlessly blends generative content with both static and dynamic scene components, achieving state-of-the-art results on both novel-view synthesis, and 2D/3D tracking tasks.", "date": "2025-06-23", "categories": []}
{"id": "2603.26481", "title": "SparseCam4D: Spatio-Temporally Consistent 4D Reconstruction from Sparse Cameras", "abstract": "High-quality 4D reconstruction enables photorealistic and immersive rendering of the dynamic real world. However, unlike static scenes that can be fully captured with a single camera, high-quality dynamic scenes typically require dense arrays of tens or even hundreds of synchronized cameras. Dependence on such costly lab setups severely limits practical scalability. To this end, we propose a sparse-camera dynamic reconstruction framework that exploits abundant yet inconsistent generative observations. Our key innovation is the Spatio-Temporal Distortion Field, which provides a unified mechanism for modeling inconsistencies in generative observations across both spatial and temporal dimensions. Building on this, we develop a complete pipeline that enables 4D reconstruction from sparse and uncalibrated camera inputs. We evaluate our method on multi-camera dynamic scene benchmarks, achieving spatio-temporally consistent high-fidelity renderings and significantly outperforming existing approaches. Project page available at https://inspatio.github.io/sparse-cam4d/", "date": "2026-03-27", "categories": []}
{"id": "2604.05715", "title": "In Depth We Trust: Reliable Monocular Depth Supervision for Gaussian Splatting", "abstract": "Using accurate depth priors in 3D Gaussian Splatting helps mitigate artifacts caused by sparse training data and textureless surfaces. However, acquiring accurate depth maps requires specialized acquisition systems. Foundation monocular depth estimation models offer a cost-effective alternative, but they suffer from scale ambiguity, multi-view inconsistency, and local geometric inaccuracies, which can degrade rendering performance when applied naively. This paper addresses the challenge of reliably leveraging monocular depth priors for Gaussian Splatting (GS) rendering enhancement. To this end, we introduce a training framework integrating scale-ambiguous and noisy depth priors into geometric supervision. We highlight the importance of learning from weakly aligned depth variations. We introduce a method to isolate ill-posed geometry for selective monocular depth regularization, restricting the propagation of depth inaccuracies into well-reconstructed 3D structures. Extensive experiments across diverse datasets show consistent improvements in geometric accuracy, leading to more faithful depth estimation and higher rendering quality across different GS variants and monocular depth backbones tested.", "date": "2026-04-07", "categories": []}
{"id": "2604.05780", "title": "Sparsity-Aware Voxel Attention and Foreground Modulation for 3D Semantic Scene Completion", "abstract": "Monocular Semantic Scene Completion (SSC) aims to reconstruct complete 3D semantic scenes from a single RGB image, offering a cost-effective solution for autonomous driving and robotics. However, the inherently imbalanced nature of voxel distributions, where over 93% of voxels are empty and foreground classes are rare, poses significant challenges. Existing methods often suffer from redundant emphasis on uninformative voxels and poor generalization to long-tailed categories. To address these issues, we propose VoxSAMNet (Voxel Sparsity-Aware Modulation Network), a unified framework that explicitly models voxel sparsity and semantic imbalance. Our approach introduces: (1) a Dummy Shortcut for Feature Refinement (DSFR) module that bypasses empty voxels via a shared dummy node while refining occupied ones with deformable attention; and (2) a Foreground Modulation Strategy combining Foreground Dropout (FD) and Text-Guided Image Filter (TGIF) to alleviate overfitting and enhance class-relevant features. Extensive experiments on the public benchmarks SemanticKITTI and SSCBench-KITTI-360 demonstrate that VoxSAMNet achieves state-of-the-art performance, surpassing prior monocular and stereo baselines with mIoU scores of 18.2% and 20.2%, respectively. Our results highlight the importance of sparsity-aware and semantics-guided design for efficient and accurate 3D scene completion, offering a promising direction for future research.", "date": "2026-04-07", "categories": []}
{"id": "2604.05908", "title": "Appearance Decomposition Gaussian Splatting for Multi-Traversal Reconstruction", "abstract": "Multi-traversal scene reconstruction is important for high-fidelity autonomous driving simulation and digital twin construction. This task involves integrating multiple sequences captured from the same geographical area at different times. In this context, a primary challenge is the significant appearance inconsistency across traversals caused by varying illumination and environmental conditions, despite the shared underlying geometry. This paper presents ADM-GS (Appearance Decomposition Gaussian Splatting for Multi-Traversal Reconstruction), a framework that applies an explicit appearance decomposition to the static background to alleviate appearance entanglement across traversals. For the static background, we decompose the appearance into traversal-invariant material, representing intrinsic material properties, and traversal-dependent illumination, capturing lighting variations. Specifically, we propose a neural light field that utilizes a frequency-separated hybrid encoding strategy. By incorporating surface normals and explicit reflection vectors, this design separately captures low-frequency diffuse illumination and high-frequency specular reflections. Quantitative evaluations on the Argoverse 2 and Waymo Open datasets demonstrate the effectiveness of ADM-GS. In multi-traversal experiments, our method achieves a +0.98 dB PSNR improvement over existing latent-based baselines while producing more consistent appearance across traversals. Code will be available at https://github.com/IRMVLab/ADM-GS.", "date": "2026-04-07", "categories": []}
{"id": "2512.17541", "title": "FLEG: Feed-Forward Language Embedded Gaussian Splatting from Any Views via Compact Semantic Representation", "abstract": "We present FLEG, a feed-forward network that reconstructs language-embedded 3D Gaussians from arbitrary views. Previous feed-forward language-embedded Gaussian reconstruction methods are restricted to a fixed number of input views and typically attach a language-aligned semantic embedding to each Gaussian, resulting in impractical input settings and semantic redundancy. In contrast, we introduce a geometry-semantic dual-branch distillation framework that enables flexible input from arbitrary multi-view images without camera parameters. We also propose a novel-view-based distillation strategy during training that mitigates overfitting to input views. In addition, we observe that semantic representations are significantly sparser than geometric ones, and per-Gaussian language embedding is unnecessary. To exploit this sparsity, we design a decoupled language embedding strategy that represents language information with a sparse set of semantic Gaussians, rather than attaching embeddings to every Gaussian. Compared with dense pixel-aligned per-Gaussian embedding schemes, our method uses only 5\\% of the language embeddings while maintaining comparable semantic fidelity, effectively reducing storage costs. Extensive experiments demonstrate that FLEG outperforms state-of-the-art feed-forward reconstruction and language-embedded Gaussian methods in both reconstruction quality and language-aligned semantic representation. Project page: https://fangzhou2000.github.io/projects/fleg.", "date": "2025-12-19", "categories": []}
{"id": "2512.25073", "title": "GaMO: Geometry-aware Multi-view Diffusion Outpainting for Sparse-View 3D Reconstruction", "abstract": "Recent 3D reconstruction methods achieve impressive results with dense multi-view imagery but struggle when only a few views are available. Various approaches, including regularization techniques, semantic priors, and geometric constraints, have been implemented to address this challenge. Recent diffusion-based approaches further improve performance by generating novel views to augment training data. Despite this progress, we identify three critical limitations in current state-of-the-art approaches: (i) inadequate coverage beyond known view peripheries, (ii) geometric inconsistencies across generated views, and (iii) computational inefficiency due to expensive pipelines. We introduce GaMO (Geometry-aware Multi-view Outpainter), a framework that reformulates sparse-view reconstruction through multi-view outpainting. Instead of generating new viewpoints, GaMO expands the field of view from existing camera poses, which inherently preserves geometric consistency while providing broader scene coverage. Our approach employs multi-view conditioning and geometry-aware denoising strategies in a zero-shot manner without training. Extensive experiments on Replica, ScanNet++, and Mip-NeRF 360 demonstrate strong reconstruction performance across sparse-view settings (3, 6, and 9 input views). Notably, our method is significantly more efficient than existing diffusion-based approaches, reducing the overall runtime to within 10 minutes. Project page: https://yichuanh.github.io/GaMO/", "date": "2025-12-31", "categories": []}
{"id": "2603.18634", "title": "SwiftGS: Episodic Priors for Immediate Satellite Surface Recovery", "abstract": "Rapid, large-scale 3D reconstruction from multi-date satellite imagery is vital for environmental monitoring, urban planning, and disaster response, yet remains difficult due to illumination changes, sensor heterogeneity, and the cost of per-scene optimization. We introduce SwiftGS, a meta-learned system that reconstructs 3D surfaces in a single forward pass by predicting geometry-radiation-decoupled Gaussian primitives together with a lightweight SDF, replacing expensive per-scene fitting with episodic training that captures transferable priors. The model couples a differentiable physics graph for projection, illumination, and sensor response with spatial gating that blends sparse Gaussian detail and global SDF structure, and incorporates semantic-geometric fusion, conditional lightweight task heads, and multi-view supervision from a frozen geometric teacher under an uncertainty-aware multi-task loss. At inference, SwiftGS operates zero-shot with optional compact calibration and achieves accurate DSM reconstruction and view-consistent rendering at significantly reduced computational cost, with ablations highlighting the benefits of the hybrid representation, physics-aware rendering, and episodic meta-training.", "date": "2026-03-19", "categories": []}
{"id": "2604.03797", "title": "Confidence-Driven Facade Refinement of 3D Building Models Using MLS Point Clouds", "abstract": "Digital twins require continuous maintenance to meet the increasing demand for high-precision geospatial data. However, traditional coarse CityGML building models, typically derived from Airborne Laser Scanning (ALS), often exhibit significant geometric deficiencies, particularly regarding facade accuracy due to the nadir perspective of airborne sensors. Integrating these coarse models with high-precision Mobile Laser Scanning (MLS) data is essential to recover detailed facade geometry. Unlike reconstruction-from-scratch approaches that discard existing semantic information and rely heavily on complete data coverage, this work presents an automated refinement framework that utilizes the coarse model as a geometric prior. This method enables targeted updates to facade geometry even in complex urban environments. It integrates surface matching to identify outdated surfaces and employs a binary integer optimization to select optimal faces from candidate data. Crucially, hard constraints are enforced within the optimization to ensure the topological validity of the refined output. Experimental results demonstrate that the proposed approach effectively corrects facade misalignments, reducing the Cloud-to-Mesh RMSE by approximately 36% and achieving centimeter-level alignment. Furthermore, the framework guarantees strictly watertight and manifold geometry, providing a robust solution for upgrading ALS-derived city models.", "date": "2026-04-04", "categories": []}
{"id": "2604.03878", "title": "Learning 3D Reconstruction with Priors in Test Time", "abstract": "We introduce a test-time framework for multiview Transformers (MVTs) that incorporates priors (e.g., camera poses, intrinsics, and depth) to improve 3D tasks without retraining or modifying pre-trained image-only networks. Rather than feeding priors into the architecture, we cast them as constraints on the predictions and optimize the network at inference time. The optimization loss consists of a self-supervised objective and prior penalty terms. The self-supervised objective captures the compatibility among multi-view predictions and is implemented using photometric or geometric loss between renderings from other views and each view itself. Any available priors are converted into penalty terms on the corresponding output modalities. Across a series of 3D vision benchmarks, including point map estimation and camera pose estimation, our method consistently improves performance over base MVTs by a large margin. On the ETH3D, 7-Scenes, and NRGBD datasets, our method reduces the point-map distance error by more than half compared with the base image-only models. Our method also outperforms retrained prior-aware feed-forward methods, demonstrating the effectiveness of our test-time constrained optimization (TCO) framework for incorporating priors into 3D vision tasks.", "date": "2026-04-04", "categories": []}
{"id": "2604.04016", "title": "HOIGS: Human-Object Interaction Gaussian Splatting", "abstract": "Reconstructing dynamic scenes with complex human-object interactions is a fundamental challenge in computer vision and graphics. Existing Gaussian Splatting methods either rely on human pose priors while neglecting dynamic objects, or approximate all motions within a single field, limiting their ability to capture interaction-rich dynamics. To address this gap, we propose Human-Object Interaction Gaussian Splatting (HOIGS), which explicitly models interaction-induced deformation between humans and objects through a cross-attention-based HOI module. Distinct deformation baselines are employed to extract features: HexPlane for humans and Cubic Hermite Spline (CHS) for objects. By integrating these heterogeneous features, HOIGS effectively captures interdependent motions and improves deformation estimation in scenarios involving occlusion, contact, and object manipulation. Comprehensive experiments on multiple datasets demonstrate that our method consistently outperforms state-of-the-art human-centric and 4D Gaussian approaches, highlighting the importance of explicitly modeling human-object interactions for high-fidelity reconstruction.", "date": "2026-04-05", "categories": []}
{"id": "2604.04331", "title": "GA-GS: Generation-Assisted Gaussian Splatting for Static Scene Reconstruction", "abstract": "Reconstructing static 3D scene from monocular video with dynamic objects is important for numerous applications such as virtual reality and autonomous driving. Current approaches typically rely on background for static scene reconstruction, limiting the ability to recover regions occluded by dynamic objects. In this paper, we propose GA-GS, a Generation-Assisted Gaussian Splatting method for Static Scene Reconstruction. The key innovation of our work lies in leveraging generation to assist in reconstructing occluded regions. We employ a motion-aware module to segment and remove dynamic regions, and thenuse a diffusion model to inpaint the occluded areas, providing pseudo-ground-truth supervision. To balance contributions from real background and generated region, we introduce a learnable authenticity scalar for each Gaussian primitive, which dynamically modulates opacity during splatting for authenticity-aware rendering and supervision. Since no existing dataset provides ground-truth static scene of video with dynamic objects, we construct a dataset named Trajectory-Match, using a fixed-path robot to record each scene with/without dynamic objects, enabling quantitative evaluation in reconstruction of occluded regions. Extensive experiments on both the DAVIS and our dataset show that GA-GS achieves state-of-the-art performance in static scene reconstruction, especially in challenging scenarios with large-scale, persistent occlusions.", "date": "2026-04-06", "categories": []}
{"id": "2604.04874", "title": "Free-Range Gaussians: Non-Grid-Aligned Generative 3D Gaussian Reconstruction", "abstract": "We present Free-Range Gaussians, a multi-view reconstruction method that predicts non-pixel, non-voxel-aligned 3D Gaussians from as few as four images. This is done through flow matching over Gaussian parameters. Our generative formulation of reconstruction allows the model to be supervised with non-grid-aligned 3D data, and enables it to synthesize plausible content in unobserved regions. Thus, it improves on prior methods that produce highly redundant grid-aligned Gaussians, and suffer from holes or blurry conditional means in unobserved regions. To handle the number of Gaussians needed for high-quality results, we introduce a hierarchical patching scheme to group spatially related Gaussians into joint transformer tokens, halving the sequence length while preserving structure. We further propose a timestep-weighted rendering loss during training, and photometric gradient guidance and classifier-free guidance at inference to improve fidelity. Experiments on Objaverse and Google Scanned Objects show consistent improvements over pixel and voxel-aligned methods while using significantly fewer Gaussians, with large gains when input views leave parts of the object unobserved.", "date": "2026-04-06", "categories": []}
{"id": "2512.13122", "title": "DePT3R: Joint Dense Point Tracking and 3D Reconstruction of Dynamic Scenes in a Single Forward Pass", "abstract": "Current methods for dense 3D point tracking in dynamic scenes typically rely on pairwise processing, require known camera poses, or assume temporal ordering of input frames, thereby constraining their flexibility and applicability. Additionally, recent advances have successfully enabled efficient 3D reconstruction from large-scale, unposed image collections, underscoring opportunities for unified approaches to dynamic scene understanding. Motivated by this, we propose DePT3R, a novel framework that simultaneously performs dense point tracking and 3D reconstruction of dynamic scenes from multiple images in a single forward pass. This multi-task learning is achieved by extracting deep spatio-temporal features with a powerful backbone and regressing pixel-wise maps with dense prediction heads. Crucially, DePT3R operates without requiring camera poses, substantially enhancing its adaptability and efficiency, especially important in dynamic environments with rapid changes. We validate DePT3R on several challenging benchmarks involving dynamic scenes, demonstrating strong performance and significant improvements in memory efficiency over existing state-of-the-art methods. Data and codes are available via the open repository: https://github.com/StructuresComp/DePT3R", "date": "2025-12-15", "categories": []}
{"id": "2603.26584", "title": "Scene Grounding In the Wild", "abstract": "Reconstructing accurate 3D models of large-scale real-world scenes from unstructured, in-the-wild imagery remains a core challenge in computer vision, especially when the input views have little or no overlap. In such cases, existing reconstruction pipelines often produce multiple disconnected partial reconstructions or erroneously merge non-overlapping regions into overlapping geometry. In this work, we propose a framework that grounds each partial reconstruction to a complete reference model of the scene, enabling globally consistent alignment even in the absence of visual overlap. We obtain reference models from dense, geospatially accurate pseudo-synthetic renderings derived from Google Earth Studio. These renderings provide full scene coverage but differ substantially in appearance from real-world photographs. Our key insight is that, despite this significant domain gap, both domains share the same underlying scene semantics. We represent the reference model using 3D Gaussian Splatting, augmenting each Gaussian with semantic features, and formulate alignment as an inverse feature-based optimization scheme that estimates a global 6DoF pose and scale while keeping the reference model fixed. Furthermore, we introduce the WikiEarth dataset, which registers existing partial 3D reconstructions with pseudo-synthetic reference models. We demonstrate that our approach consistently improves global alignment when initialized with various classical and learning-based pipelines, while mitigating failure modes of state-of-the-art end-to-end models.", "date": "2026-03-27", "categories": []}
{"id": "2604.01479", "title": "UniRecGen: Unifying Multi-View 3D Reconstruction and Generation", "abstract": "Sparse-view 3D modeling represents a fundamental tension between reconstruction fidelity and generative plausibility. While feed-forward reconstruction excels in efficiency and input alignment, it often lacks the global priors needed for structural completeness. Conversely, diffusion-based generation provides rich geometric details but struggles with multi-view consistency. We present UniRecGen, a unified framework that integrates these two paradigms into a single cooperative system. To overcome inherent conflicts in coordinate spaces, 3D representations, and training objectives, we align both models within a shared canonical space. We employ disentangled cooperative learning, which maintains stable training while enabling seamless collaboration during inference. Specifically, the reconstruction module is adapted to provide canonical geometric anchors, while the diffusion generator leverages latent-augmented conditioning to refine and complete the geometric structure. Experimental results demonstrate that UniRecGen achieves superior fidelity and robustness, outperforming existing methods in creating complete and consistent 3D models from sparse observations. Code is available at https://github.com/zsh523/UniRecGen.", "date": "2026-04-01", "categories": []}
{"id": "2604.02497", "title": "Delaunay Canopy: Building Wireframe Reconstruction from Airborne LiDAR Point Clouds via Delaunay Graph", "abstract": "Reconstructing building wireframe from airborne LiDAR point clouds yields a compact, topology-centric representation that enables structural understanding beyond dense meshes. Yet a key limitation persists: conventional methods have failed to achieve accurate wireframe reconstruction in regions afflicted by significant noise, sparsity, or internal corners. This failure stems from the inability to establish an adaptive search space to effectively leverage the rich 3D geometry of large, sparse building point clouds. In this work, we address this challenge with Delaunay Canopy, which utilizes the Delaunay graph as a geometric prior to define a geometrically adaptive search space. Central to our approach is Delaunay Graph Scoring, which not only reconstructs the underlying geometric manifold but also yields region-wise curvature signatures to robustly guide the reconstruction. Built on this foundation, our corner and wire selection modules leverage the Delaunay-induced prior to focus on highly probable elements, thereby shaping the search space and enabling accurate prediction even in previously intractable regions. Extensive experiments on the Building3D Tallinn city and entry-level datasets demonstrate state-of-the-art wireframe reconstruction, delivering accurate predictions across diverse and complex building geometries.", "date": "2026-04-02", "categories": []}
{"id": "2604.02828", "title": "NavCrafter: Exploring 3D Scenes from a Single Image", "abstract": "Creating flexible 3D scenes from a single image is vital when direct 3D data acquisition is costly or impractical. We introduce NavCrafter, a novel framework that explores 3D scenes from a single image by synthesizing novel-view video sequences with camera controllability and temporal-spatial consistency. NavCrafter leverages video diffusion models to capture rich 3D priors and adopts a geometry-aware expansion strategy to progressively extend scene coverage. To enable controllable multi-view synthesis, we introduce a multi-stage camera control mechanism that conditions diffusion models with diverse trajectories via dual-branch camera injection and attention modulation. We further propose a collision-aware camera trajectory planner and an enhanced 3D Gaussian Splatting (3DGS) pipeline with depth-aligned supervision, structural regularization and refinement. Extensive experiments demonstrate that NavCrafter achieves state-of-the-art novel-view synthesis under large viewpoint shifts and substantially improves 3D reconstruction fidelity.", "date": "2026-04-03", "categories": []}
{"id": "2604.03069", "title": "SparseSplat: Towards Applicable Feed-Forward 3D Gaussian Splatting with Pixel-Unaligned Prediction", "abstract": "Recent progress in feed-forward 3D Gaussian Splatting (3DGS) has notably improved rendering quality. However, the spatially uniform and highly redundant 3DGS map generated by previous feed-forward 3DGS methods limits their integration into downstream reconstruction tasks. We propose SparseSplat, the first feed-forward 3DGS model that adaptively adjusts Gaussian density according to scene structure and information richness of local regions, yielding highly compact 3DGS maps. To achieve this, we propose entropy-based probabilistic sampling, generating large, sparse Gaussians in textureless areas and assigning small, dense Gaussians to regions with rich information. Additionally, we designed a specialized point cloud network that efficiently encodes local context and decodes it into 3DGS attributes, addressing the receptive field mismatch between the general 3DGS optimization pipeline and feed-forward models. Extensive experimental results demonstrate that SparseSplat can achieve state-of-the-art rendering quality with only 22% of the Gaussians and maintain reasonable rendering quality with only 1.5% of the Gaussians. Project page: https://victkk.github.io/SparseSplat-page/.", "date": "2026-04-03", "categories": []}
{"id": "2601.20331", "title": "GVGS: Gaussian Visibility-Aware Multi-View Geometry for Accurate Surface Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) enables efficient rendering, yet accurate surface reconstruction remains challenging due to unreliable geometric supervision. Existing approaches predominantly rely on depth-based reprojection to infer visibility and enforce multi-view consistency, leading to a fundamental circular dependency: visibility estimation requires accurate depth, while depth supervision itself is conditioned on visibility. In this work, we revisit multi-view geometric supervision from the perspective of visibility modeling. Instead of inferring visibility from pixel-wise depth consistency, we explicitly model visibility at the level of Gaussian primitives. We introduce a Gaussian visibility-aware multi-view geometric consistency (GVMV) formulation, which aggregates cross-view visibility of shared Gaussians to construct reliable supervision over co-visible regions. To further incorporate monocular priors, we propose a progressive quadtree-calibrated depth alignment (QDC) strategy that performs block-wise affine calibration under visibility-aware guidance, effectively mitigating scale ambiguity while preserving local geometric structures. Extensive experiments on DTU and Tanks and Temples demonstrate that our method consistently improves reconstruction accuracy over prior Gaussian-based approaches. Our code is fully open-sourced and available at an anonymous repository: https://github.com/GVGScode/GVGS.", "date": "2026-01-28", "categories": []}
{"id": "2602.23205", "title": "EmbodMocap: In-the-Wild 4D Human-Scene Reconstruction for Embodied Agents", "abstract": "Human behaviors in the real world naturally encode rich, long-term contextual information that can be leveraged to train embodied agents for perception, understanding, and acting. However, existing capture systems typically rely on costly studio setups and wearable devices, limiting the large-scale collection of scene-conditioned human motion data in the wild. To address this, we propose EmbodMocap, a portable and affordable data collection pipeline using two moving iPhones. Our key idea is to jointly calibrate dual RGB-D sequences to reconstruct both humans and scenes within a unified metric world coordinate frame. The proposed method allows metric-scale and scene-consistent capture in everyday environments without static cameras or markers, bridging human motion and scene geometry seamlessly. Compared with optical capture ground truth, we demonstrate that the dual-view setting exhibits a remarkable ability to mitigate depth ambiguity, achieving superior alignment and reconstruction performance over single iphone or monocular models. Based on the collected data, we empower three embodied AI tasks: monocular human-scene-reconstruction, where we fine-tune on feedforward models that output metric-scale, world-space aligned humans and scenes; physics-based character animation, where we prove our data could be used to scale human-object interaction skills and scene-aware motion tracking; and robot motion control, where we train a humanoid robot via sim-to-real RL to replicate human motions depicted in videos. Experimental results validate the effectiveness of our pipeline and its contributions towards advancing embodied AI research.", "date": "2026-02-26", "categories": []}
{"id": "2604.01561", "title": "ReFlow: Self-correction Motion Learning for Dynamic Scene Reconstruction", "abstract": "We present ReFlow, a unified framework for monocular dynamic scene reconstruction that learns 3D motion in a novel self-correction manner from raw video. Existing methods often suffer from incomplete scene initialization for dynamic regions, leading to unstable reconstruction and motion estimation, which often resorts to external dense motion guidance such as pre-computed optical flow to further stabilize and constrain the reconstruction of dynamic components. However, this introduces additional complexity and potential error propagation. To address these issues, ReFlow integrates a Complete Canonical Space Construction module for enhanced initialization of both static and dynamic regions, and a Separation-Based Dynamic Scene Modeling module that decouples static and dynamic components for targeted motion supervision. The core of ReFlow is a novel self-correction flow matching mechanism, consisting of Full Flow Matching to align 3D scene flow with time-varying 2D observations, and Camera Flow Matching to enforce multi-view consistency for static objects. Together, these modules enable robust and accurate dynamic scene reconstruction. Extensive experiments across diverse scenarios demonstrate that ReFlow achieves superior reconstruction quality and robustness, establishing a novel self-correction paradigm for monocular 4D reconstruction.", "date": "2026-04-02", "categories": []}
{"id": "2604.01605", "title": "F3DGS: Federated 3D Gaussian Splatting for Decentralized Multi-Agent World Modeling", "abstract": "We present F3DGS, a federated 3D Gaussian Splatting framework for decentralized multi-agent 3D reconstruction. Existing 3DGS pipelines assume centralized access to all observations, which limits their applicability in distributed robotic settings where agents operate independently, and centralized data aggregation may be restricted. Directly extending centralized training to multi-agent systems introduces communication overhead and geometric inconsistency. F3DGS first constructs a shared geometric scaffold by registering locally merged LiDAR point clouds from multiple clients to initialize a global 3DGS model. During federated optimization, Gaussian positions are fixed to preserve geometric alignment, while each client updates only appearance-related attributes, including covariance, opacity, and spherical harmonic coefficients. The server aggregates these updates using visibility-aware aggregation, weighting each client's contribution by how frequently it observed each Gaussian, resolving the partial-observability challenge inherent to multi-agent exploration. To evaluate decentralized reconstruction, we collect a multi-sequence indoor dataset with synchronized LiDAR, RGB, and IMU measurements. Experiments show that F3DGS achieves reconstruction quality comparable to centralized training while enabling distributed optimization across agents. The dataset, development kit, and source code will be publicly released.", "date": "2026-04-02", "categories": []}
{"id": "2604.01678", "title": "Director: Instance-aware Gaussian Splatting for Dynamic Scene Modeling and Understanding", "abstract": "Volumetric video seeks to model dynamic scenes as temporally coherent 4D representations. While recent Gaussian-based approaches achieve impressive rendering fidelity, they primarily emphasize appearance but are largely agnostic to instance-level structure, limiting stable tracking and semantic reasoning in highly dynamic scenarios. In this paper, we present Director, a unified spatio-temporal Gaussian representation that jointly models human performance, high-fidelity rendering, and instance-level semantics. Our key insight is that embedding instance-consistent semantics naturally complements 4D modeling, enabling more accurate scene decomposition while supporting robust dynamic scene understanding. To this end, we leverage temporally aligned instance masks and sentence embeddings derived from Multimodal Large Language Models to supervise the learnable semantic features of each Gaussian via two MLP decoders, enabling language-aligned 4D representations and enforcing identity consistency over time. To enhance temporal stability, we bridge 2D optical flow with 4D Gaussians and finetune their motions, yielding reliable initialization and reducing drift. For the training, we further introduce a geometry-aware SDF constraints, along with regularization terms that enforces surface continuity, enhancing temporal coherence in dynamic foreground modeling. Experiments demonstrate that Director achieves temporally coherent 4D reconstructions while simultaneously enabling instance segmentation and open-vocabulary querying.", "date": "2026-04-02", "categories": []}
{"id": "2604.01884", "title": "GS^2: Graph-based Spatial Distribution Optimization for Compact 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated breakthrough performance in novel view synthesis and real-time rendering. Nevertheless, its practicality is constrained by the high memory cost due to a huge number of Gaussian points. Many pruning-based 3DGS variants have been proposed for memory saving, but often compromise spatial consistency and may lead to rendering artifacts. To address this issue, we propose graph-based spatial distribution optimization for compact 3D Gaussian Splatting (GS\\textasciicircum2), which enhances reconstruction quality by optimizing the spatial distribution of Gaussian points. Specifically, we introduce an evidence lower bound (ELBO)-based adaptive densification strategy that automatically controls the densification process. In addition, an opacity-aware progressive pruning strategy is proposed to further reduce memory consumption by dynamically removing low-opacity Gaussian points. Furthermore, we propose a graph-based feature encoding module to adjust the spatial distribution via feature-guided point shifting. Extensive experiments validate that GS\\textasciicircum2 achieves a compact Gaussian representation while delivering superior rendering quality. Compared with 3DGS, it achieves higher PSNR with only about 12.5\\% Gaussian points. Furthermore, it outperforms all compared baselines in both rendering quality and memory efficiency.", "date": "2026-04-02", "categories": []}
{"id": "2512.18640", "title": "Geometric-Photometric Event-based 3D Gaussian Ray Tracing", "abstract": "Event cameras offer a high temporal resolution over traditional frame-based cameras, which makes them suitable for motion and structure estimation. However, it has been unclear how event-based 3D Gaussian Splatting (3DGS) approaches could leverage fine-grained temporal information of sparse events. This work proposes GPERT, a framework to address the trade-off between accuracy and temporal resolution in event-based 3DGS. Our key idea is to decouple the rendering into two branches: event-by-event geometry (depth) rendering and snapshot-based radiance (intensity) rendering, by using ray-tracing and the image of warped events. The extensive evaluation shows that our method achieves state-of-the-art performance on the real-world datasets and competitive performance on the synthetic dataset. Also, the proposed method works without prior information (e.g., pretrained image reconstruction models) or COLMAP-based initialization, is more flexible in the event selection number, and achieves sharp reconstruction on scene edges with fast training time. We hope that this work deepens our understanding of the sparse nature of events for 3D reconstruction. https://github.com/e3ai/gpert", "date": "2025-12-21", "categories": []}
{"id": "2604.00538", "title": "TRiGS: Temporal Rigid-Body Motion for Scalable 4D Gaussian Splatting", "abstract": "Recent 4D Gaussian Splatting (4DGS) methods achieve impressive dynamic scene reconstruction but often rely on piecewise linear velocity approximations and short temporal windows. This disjointed modeling leads to severe temporal fragmentation, forcing primitives to be repeatedly eliminated and regenerated to track complex nonlinear dynamics. This makeshift approximation eliminates the long-term temporal identity of objects and causes an inevitable proliferation of Gaussians, hindering scalability to extended video sequences. To address this, we propose TRiGS, a novel 4D representation that utilizes unified, continuous geometric transformations. By integrating $SE(3)$ transformations, hierarchical Bezier residuals, and learnable local anchors, TRiGS models geometrically consistent rigid motions for individual primitives. This continuous formulation preserves temporal identity and effectively mitigates unbounded memory growth. Extensive experiments demonstrate that TRiGS achieves high fidelity rendering on standard benchmarks while uniquely scaling to extended video sequences (e.g., 600 to 1200 frames) without severe memory bottlenecks, significantly outperforming prior works in temporal stability.", "date": "2026-04-01", "categories": []}
{"id": "2604.00548", "title": "Reliev3R: Relieving Feed-forward Reconstruction from Multi-View Geometric Annotations", "abstract": "With recent advances, Feed-forward Reconstruction Models (FFRMs) have demonstrated great potential in reconstruction quality and adaptiveness to multiple downstream tasks. However, the excessive reliance on multi-view geometric annotations, e.g. 3D point maps and camera poses, makes the fully-supervised training scheme of FFRMs difficult to scale up. In this paper, we propose Reliev3R, a weakly-supervised paradigm for training FFRMs from scratch without cost-prohibitive multi-view geometric annotations. Relieving the reliance on geometric sensory data and compute-exhaustive structure-from-motion preprocessing, our method draws 3D knowledge directly from monocular relative depths and image sparse correspondences given by zero-shot predictions of pretrained models. At the core of Reliev3R, we design an ambiguity-aware relative depth loss and a trigonometry-based reprojection loss to facilitate supervision for multi-view geometric consistency. Training from scratch with the less data, Reliev3R catches up with its fully-supervised sibling models, taking a step towards low-cost 3D reconstruction supervisions and scalable FFRMs.", "date": "2026-04-01", "categories": []}
{"id": "2604.01030", "title": "Diff3R: Feed-forward 3D Gaussian Splatting with Uncertainty-aware Differentiable Optimization", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) present two main directions: feed-forward models offer fast inference in sparse-view settings, while per-scene optimization yields high-quality renderings but is computationally expensive. To combine the benefits of both, we introduce Diff3R, a novel framework that explicitly bridges feed-forward prediction and test-time optimization. By incorporating a differentiable 3DGS optimization layer directly into the training loop, our network learns to predict an optimal initialization for test-time optimization rather than a conventional zero-shot result. To overcome the computational cost of backpropagating through the optimization steps, we propose computing gradients via the Implicit Function Theorem and a scalable, matrix-free PCG solver tailored for 3DGS optimization. Additionally, we incorporate a data-driven uncertainty model into the optimization process by adaptively controlling how much the parameters are allowed to change during optimization. This approach effectively mitigates overfitting in under-constrained regions and increases robustness against input outliers. Since our proposed optimization layer is model-agnostic, we show that it can be seamlessly integrated into existing feed-forward 3DGS architectures for both pose-given and pose-free methods, providing improvements for test-time optimization.", "date": "2026-04-01", "categories": []}
{"id": "2505.08811", "title": "TUGS: Physics-based Compact Representation of Underwater Scenes by Tensorized Gaussian", "abstract": "Underwater 3D scene reconstruction is crucial for multimedia applications in adverse environments, such as underwater robotic perception and navigation. However, the complexity of interactions between light propagation, water medium, and object surfaces poses significant difficulties for existing methods in accurately simulating their interplay. Additionally, expensive training and rendering costs limit their practical application. Therefore, we propose Tensorized Underwater Gaussian Splatting (TUGS), a compact underwater 3D representation based on physical modeling of complex underwater light fields. TUGS includes a physics-based underwater Adaptive Medium Estimation (AME) module, enabling accurate simulation of both light attenuation and backscatter effects in underwater environments, and introduces Tensorized Densification Strategies (TDS) to efficiently refine the tensorized representation during optimization. TUGS is able to render high-quality underwater images with faster rendering speeds and less memory usage. Extensive experiments on real-world underwater datasets have demonstrated that TUGS can efficiently achieve superior reconstruction quality using a limited number of parameters. The code is available at https://liamlian0727.github.io/TUGS", "date": "2025-05-12", "categories": []}
{"id": "2507.11539", "title": "Streaming 4D Visual Geometry Transformer", "abstract": "Perceiving and reconstructing 3D geometry from videos is a fundamental yet challenging computer vision task. To facilitate interactive and low-latency applications, we propose a streaming visual geometry transformer that shares a similar philosophy with autoregressive large language models. We explore a simple and efficient design and employ a causal transformer architecture to process the input sequence in an online manner. We use temporal causal attention and cache the historical keys and values as implicit memory to enable efficient streaming long-term 3D reconstruction. This design can handle low-latency 3D reconstruction by incrementally integrating historical information while maintaining high-quality spatial consistency. For efficient training, we propose to distill knowledge from the dense bidirectional visual geometry grounded transformer (VGGT) to our causal model. For inference, our model supports the migration of optimized efficient attention operators (e.g., FlashAttention) from large language models. Extensive experiments on various 3D geometry perception benchmarks demonstrate that our model enhances inference speed in online scenarios while maintaining competitive performance, thereby facilitating scalable and interactive 3D vision systems. Code is available at: https://github.com/wzzheng/StreamVGGT.", "date": "2025-07-15", "categories": []}
{"id": "2603.27084", "title": "SceneExpander: Expanding 3D Scenes with Free-Form Inserted Views", "abstract": "World building with 3D scene representations is increasingly important for content creation, simulation, and interactive experiences, yet real workflows are inherently iterative: creators must repeatedly extend an existing scene under user control. Motivated by this research gap, we study 3D scene expansion in a user-centric workflow: starting from a real scene captured by multi-view images, we extend its coverage by inserting an additional view synthesized by a generative model. Unlike simple object editing or style transfer in a fixed scene, the inserted view is often 3D-misaligned with the original reconstruction, introducing geometry shifts, hallucinated content, or view-dependent artifacts that break global multi-view consistency. To address the challenge, we propose SceneExpander, which applies test-time adaptation to a parametric feed-forward 3D reconstruction model with two complementary distillation signals: anchor distillation stabilizes the original scene by distilling geometric cues from the captured views, while inserted-view self-distillation preserves observation-supported predictions yet adapts latent geometry and appearance to accommodate the misaligned inserted view. Experiments on ETH scenes and online data demonstrate improved expansion behavior and reconstruction quality under misalignment.", "date": "2026-03-28", "categories": []}
{"id": "2603.28896", "title": "Fisheye3R: Adapting Unified 3D Feed-Forward Foundation Models to Fisheye Lenses", "abstract": "Feed-forward foundation models for multi-view 3-dimensional (3D) reconstruction have been trained on large-scale datasets of perspective images; when tested on wide field-of-view images, e.g., from a fisheye camera, their performance degrades. Their error arises from changes in spatial positions of pixels due to a non-linear projection model that maps 3D points onto the 2D image plane. While one may surmise that training on fisheye images would resolve this problem, there are far fewer fisheye images with ground truth than perspective images, which limit generalization. To enable inference on imagery exhibiting high radial distortion, we propose Fisheye3R, a novel adaptation framework that extends these multi-view 3D reconstruction foundation models to natively accommodate fisheye inputs without performance regression on perspective images. To address the scarcity of fisheye images and ground truth, we introduce flexible learning schemes that support self-supervised adaptation using only unlabeled perspective images and supervised adaptation without any fisheye training data. Extensive experiments across three foundation models, including VGGT, $\\pi^3$, and MapAnything, demonstrate that our approach consistently improves camera pose, depth, point map, and field-of-view estimation on fisheye images.", "date": "2026-03-30", "categories": []}
{"id": "2603.29296", "title": "MotionScale: Reconstructing Appearance, Geometry, and Motion of Dynamic Scenes with Scalable 4D Gaussian Splatting", "abstract": "Realistic reconstruction of dynamic 4D scenes from monocular videos is essential for understanding the physical world. Despite recent progress in neural rendering, existing methods often struggle to recover accurate 3D geometry and temporally consistent motion in complex environments. To address these challenges, we propose MotionScale, a 4D Gaussian Splatting framework that scales efficiently to large scenes and extended sequences while maintaining high-fidelity structural and motion coherence. At the core of our approach is a scalable motion field parameterized by cluster-centric basis transformations that adaptively expand to capture diverse and evolving motion patterns. To ensure robust reconstruction over long durations, we introduce a progressive optimization strategy comprising two decoupled propagation stages: 1) A background extension stage that adapts to newly visible regions, refines camera poses, and explicitly models transient shadows; 2) A foreground propagation stage that enforces motion consistency through a specialized three-stage refinement process. Extensive experiments on challenging real-world benchmarks demonstrate that MotionScale significantly outperforms state-of-the-art methods in both reconstruction quality and temporal stability. Project page: https://hrzhou2.github.io/motion-scale-web/.", "date": "2026-03-31", "categories": []}
{"id": "2603.29394", "title": "AA-Splat: Anti-Aliased Feed-forward Gaussian Splatting", "abstract": "Feed-forward 3D Gaussian Splatting (FF-3DGS) emerges as a fast and robust solution for sparse-view 3D reconstruction and novel view synthesis (NVS). However, existing FF-3DGS methods are built on incorrect screen-space dilation filters, causing severe rendering artifacts when rendering at out-of-distribution sampling rates. We firstly propose an FF-3DGS model, called AA-Splat, to enable robust anti-aliased rendering at any resolution. AA-Splat utilizes an opacity-balanced band-limiting (OBBL) design, which combines two components: a 3D band-limiting post-filter integrates multi-view maximal frequency bounds into the feed-forward reconstruction pipeline, effectively band-limiting the resulting 3D scene representations and eliminating degenerate Gaussians; an Opacity Balancing (OB) to seamlessly integrate all pixel-aligned Gaussian primitives into the rendering process, compensating for the increased overlap between expanded Gaussian primitives. AA-Splat demonstrates drastic improvements with average 5.4$\\sim$7.5dB PSNR gains on NVS performance over a state-of-the-art (SOTA) baseline, DepthSplat, at all resolutions, between $4\\times$ and $1/4\\times$. Code will be made available.", "date": "2026-03-31", "categories": []}
{"id": "2511.22264", "title": "DriveVGGT: Calibration-Constrained Visual Geometry Transformers for Multi-Camera Autonomous Driving", "abstract": "Feed-forward reconstruction has been progressed rapidly, with the Visual Geometry Grounded Transformer (VGGT) being a notable baseline. However, directly applying VGGT to autonomous driving (AD) fails to capture three domain-specific priors: (i) Sparse Spatial Overlap: the overlap among mutli-view cameras is minimal due to $360^{\\circ}$ coverage requirements under budget control, which renders global attention among all images inefficient; (ii) Calibrated Geometric Constraints: the absolute distance among cameras is generally accessible for AD data with calibration process before driving. Standard VGGT is unable to directly utilize such information for absolute scale scene reconstruction; (iii) Rigid Extrinsic Constancy: relative poses of multi-view cameras are approximately static, i.e., the ego-motion is the same for all cameras.\n  To bridge these gaps, we propose DriveVGGT, a scale-aware reconstruction framework that explicitly integrates these priors through three targeted components. First, for the Sparse Spatial Overlap in (i), we introduce a Temporal Video Attention (TVA) module to process multi-camera videos independently. Second, for Calibrated Geometric Constraints in (ii), a Multi-camera Consistency Attention (MCA) module is designed to directly utilize the calibration information among cameras with a scale head for absolute scale scene reconstruction. Finally, to utilize Rigid Extrinsic Constancy in (iii), we reformulate the decoding process of VGGT into factorized sequential pose head and ego motion head. On AD datasets, experiments demonstrate that DriveVGGT reduces inference time by 49.3\\% while improving depth and pose estimation compared to vanilla VGGT in long-sequence scenarios. It consistently outperforms recent SOTA variants. Meanwhile, extensive ablation studies verify the effectiveness of each devised module.", "date": "2025-11-27", "categories": []}
{"id": "2512.10950", "title": "E-RayZer: Self-supervised 3D Reconstruction as Spatial Visual Pre-training", "abstract": "Self-supervised pre-training has driven rapid progress in foundation models for language, 2D images, and video, yet remains largely unexplored for learning 3D-aware representations from multi-view images. In this paper, we present E-RayZer, a self-supervised 3D vision model that learns geometrically grounded representations directly from unlabeled images. Unlike prior self-supervised methods such as RayZer, which infer 3D indirectly through latent-space view synthesis, E-RayZer operates directly in 3D space, performing self-supervised 3D reconstruction with Explicit geometry. This formulation eliminates shortcut solutions and yields representations that are 3D-aware. To ensure convergence and scalability, we introduce a fine-grained learning curriculum that organizes training from easy to hard samples and harmonizes heterogeneous data sources without any supervision. Experiments show that E-RayZer significantly outperforms RayZer on pose estimation and matches or sometimes surpasses fully supervised reconstruction models such as VGGT. Furthermore, its learned representations outperform leading visual pre-training models (e.g., DINOv3, CroCo v2, VideoMAE V2, and RayZer) on 3D downstream tasks, establishing E-RayZer as a promising paradigm for spatial visual pre-training.", "date": "2025-12-11", "categories": []}
{"id": "2512.13680", "title": "LASER: Layer-wise Scale Alignment for Training-Free Streaming 4D Reconstruction", "abstract": "Recent feed-forward reconstruction models like VGGT and $\\pi^3$ achieve impressive reconstruction quality but cannot process streaming videos due to quadratic memory complexity, limiting their practical deployment. While existing streaming methods address this through learned memory mechanisms or causal attention, they require extensive retraining and may not fully leverage the strong geometric priors of state-of-the-art offline models. We propose LASER, a training-free framework that converts an offline reconstruction model into a streaming system by aligning predictions across consecutive temporal windows. We observe that simple similarity transformation ($\\mathrm{Sim}(3)$) alignment fails due to layer depth misalignment: monocular scale ambiguity causes relative depth scales of different scene layers to vary inconsistently between windows. To address this, we introduce layer-wise scale alignment, which segments depth predictions into discrete layers, computes per-layer scale factors, and propagates them across both adjacent windows and timestamps. Extensive experiments show that LASER achieves state-of-the-art performance on camera pose estimation and point map reconstruction %quality with offline models while operating at 14 FPS with 6 GB peak memory on a RTX A6000 GPU, enabling practical deployment for kilometer-scale streaming videos. Project website: https://neu-vi.github.io/LASER/", "date": "2025-12-15", "categories": []}
{"id": "2512.15508", "title": "Off The Grid: Detection of Primitives for Feed-Forward 3D Gaussian Splatting", "abstract": "Feed-forward 3D Gaussian Splatting (3DGS) models enable real-time scene generation but are hindered by suboptimal pixel-aligned primitive placement, which relies on a dense, rigid grid that limits both quality and efficiency. We introduce a new feed-forward architecture that detects 3D Gaussian primitives at a sub-pixel level, replacing the pixel grid with an adaptive, ``Off-The-Grid\" distribution. Inspired by keypoint detection, our decoder learns to locally distribute primitives across image patches. We also provide an Adaptive Density mechanism by assigning varying number of primitives per patch based on Shannon entropy. We combine the proposed decoder with a pre-trained 3D reconstruction backbone and train them end-to-end using photometric supervision without any 3D annotation. The resulting pose-free model generates photorealistic 3DGS scenes in seconds, achieving state-of-the-art novel view synthesis for feed-forward models. It outperforms competitors while using far fewer primitives, demonstrating a more accurate and efficient allocation that captures fine details and reduces artifacts. Project page: https://arthurmoreau.github.io/OffTheGrid/.", "date": "2025-12-17", "categories": []}
{"id": "2602.08961", "title": "MotionCrafter: Dense Geometry and Motion Reconstruction with a 4D VAE", "abstract": "We present MotionCrafter, a framework that leverages video generators to jointly reconstruct 4D geometry and estimate dense motion from a monocular video. The key idea is a joint representation of dense 3D point maps and 3D scene flows in a shared coordinate system, together with a 4D VAE tailored to learn this representation effectively. Unlike prior work that strictly aligns 3D values and latents with RGB VAE latents-despite their fundamentally different distributions-we show that such alignment is unnecessary and can hurt performance. Instead, we propose a new data normalization and VAE training strategy that better transfers diffusion priors and greatly improves reconstruction quality. Extensive experiments on multiple datasets show that MotionCrafter achieves state-of-the-art performance in both geometry reconstruction and dense scene flow estimation, delivering 38.64% and 25.0% improvements in geometry and motion reconstruction, respectively, all without any post-optimization. Project page: https://ruijiezhu94.github.io/MotionCrafter_Page", "date": "2026-02-09", "categories": []}
{"id": "2603.17227", "title": "Adaptive Anchor Policies for Efficient 4D Gaussian Streaming", "abstract": "Dynamic scene reconstruction with Gaussian Splatting has enabled efficient streaming for real-time rendering and free-viewpoint video. However, most pipelines rely on fixed anchor selection such as Farthest Point Sampling (FPS), typically using 8,192 anchors regardless of scene complexity, which over-allocates computation under strict budgets. We propose Efficient Gaussian Streaming (EGS), a plug-in, budget-aware anchor sampler that replaces FPS with a reinforcement-learned policy while keeping the Gaussian streaming reconstruction backbone unchanged. The policy jointly selects an anchor budget and a subset of informative anchors under discrete constraints, balancing reconstruction quality and runtime using spatial features of the Gaussian representation. We evaluate EGS in two settings: fast rendering, which prioritizes runtime efficiency, and high-quality refinement, which enables additional optimization. Experiments on dynamic multi-view datasets show consistent improvements in the quality--efficiency trade-off over FPS sampling. On unseen data, in fast rendering at 256 anchors ($32\\times$ fewer than 8,192), EGS improves PSNR by $+0.52$--$0.61$\\,dB while running $1.29$--$1.35\\times$ faster than IGS@8192 (N3DV and MeetingRoom). In high-quality refinement, EGS remains competitive with the full-anchor baseline at substantially lower anchor budgets. \\emph{Code and pretrained checkpoints will be released upon acceptance.} \\keywords{4D Gaussian Splatting \\and 4D Gaussian Streaming \\and Reinforcement Learning}", "date": "2026-03-18", "categories": []}
{"id": "2603.21356", "title": "FluidGaussian: Propagating Simulation-Based Uncertainty Toward Functionally-Intelligent 3D Reconstruction", "abstract": "Real objects that inhabit the physical world follow physical laws and thus behave plausibly during interaction with other physical objects. However, current methods that perform 3D reconstructions of real-world scenes from multi-view 2D images optimize primarily for visual fidelity, i.e., they train with photometric losses and reason about uncertainty in the image or representation space. This appearance-centric view overlooks body contacts and couplings, conflates function-critical regions (e.g., aerodynamic or hydrodynamic surfaces) with ornamentation, and reconstructs structures suboptimally, even when physical regularizers are added. All these can lead to unphysical and implausible interactions. To address this, we consider the question: How can 3D reconstruction become aware of real-world interactions and underlying object functionality, beyond visual cues? To answer this question, we propose FluidGaussian, a plug-and-play method that tightly couples geometry reconstruction with ubiquitous fluid-structure interactions to assess surface quality at high granularity. We define a simulation-based uncertainty metric induced by fluid simulations and integrate it with active learning to prioritize views that improve both visual and physical fidelity. In an empirical evaluation on NeRF Synthetic (Blender), Mip-NeRF 360, and DrivAerNet++, our FluidGaussian method yields up to +8.6% visual PSNR (Peak Signal-to-Noise Ratio) and -62.3% velocity divergence during fluid simulations. Our code is available at https://github.com/delta-lab-ai/FluidGaussian.", "date": "2026-03-22", "categories": []}
{"id": "2603.25008", "title": "Few TensoRF: Enhance the Few-shot on Tensorial Radiance Fields", "abstract": "This paper presents Few TensoRF, a 3D reconstruction framework that combines TensorRF's efficient tensor based representation with FreeNeRF's frequency driven few shot regularization. Using TensorRF to significantly accelerate rendering speed and introducing frequency and occlusion masks, the method improves stability and reconstruction quality under sparse input views. Experiments on the Synthesis NeRF benchmark show that Few TensoRF method improves the average PSNR from 21.45 dB (TensorRF) to 23.70 dB, with the fine tuned version reaching 24.52 dB, while maintaining TensorRF's fast \\(\\approx10-15\\) minute training time. Experiments on the THuman 2.0 dataset further demonstrate competitive performance in human body reconstruction, achieving 27.37 - 34.00 dB with only eight input images. These results highlight Few TensoRF as an efficient and data effective solution for real-time 3D reconstruction across diverse scenes.", "date": "2026-03-26", "categories": []}
{"id": "2603.25053", "title": "GaussFusion: Improving 3D Reconstruction in the Wild with A Geometry-Informed Video Generator", "abstract": "We present GaussFusion, a novel approach for improving 3D Gaussian splatting (3DGS) reconstructions in the wild through geometry-informed video generation. GaussFusion mitigates common 3DGS artifacts, including floaters, flickering, and blur caused by camera pose errors, incomplete coverage, and noisy geometry initialization. Unlike prior RGB-based approaches limited to a single reconstruction pipeline, our method introduces a geometry-informed video-to-video generator that refines 3DGS renderings across both optimization-based and feed-forward methods. Given an existing reconstruction, we render a Gaussian primitive video buffer encoding depth, normals, opacity, and covariance, which the generator refines to produce temporally coherent, artifact-free frames. We further introduce an artifact synthesis pipeline that simulates diverse degradation patterns, ensuring robustness and generalization. GaussFusion achieves state-of-the-art performance on novel-view synthesis benchmarks, and an efficient variant runs in real time at 15 FPS while maintaining similar performance, enabling interactive 3D applications.", "date": "2026-03-26", "categories": []}
{"id": "2603.27228", "title": "NimbusGS: Unified 3D Scene Reconstruction under Hybrid Weather", "abstract": "We present NimbusGS, a unified framework for reconstructing high-quality 3D scenes from degraded multi-view inputs captured under diverse and mixed adverse weather conditions. Unlike existing methods that target specific weather types, NimbusGS addresses the broader challenge of generalization by modeling the dual nature of weather: a continuous, view-consistent medium that attenuates light, and dynamic, view-dependent particles that cause scattering and occlusion. To capture this structure, we decompose degradations into a global transmission field and per-view particulate residuals. The transmission field represents static atmospheric effects shared across views, while the residuals model transient disturbances unique to each input. To enable stable geometry learning under severe visibility degradation, we introduce a geometry-guided gradient scaling mechanism that mitigates gradient imbalance during the self-supervised optimization of 3D Gaussian representations. This physically grounded formulation allows NimbusGS to disentangle complex degradations while preserving scene structure, yielding superior geometry reconstruction and outperforming task-specific methods across diverse and challenging weather conditions. Code is available at https://github.com/lyy-ovo/NimbusGS.", "date": "2026-03-28", "categories": []}
{"id": "2603.27300", "title": "Complet4R: Geometric Complete 4D Reconstruction", "abstract": "We introduce Complet4R, a novel end-to-end framework for Geometric Complete 4D Reconstruction, which aims to recover temporally coherent and geometrically complete reconstruction for dynamic scenes. Our method formalizes the task of Geometric Complete 4D Reconstruction as a unified framework of reconstruction and completion, by directly accumulating full contexts onto each frame. Unlike previous approaches that rely on pairwise reconstruction or local motion estimation, Complet4R utilizes a decoder-only transformer to operate all context globally directly from sequential video input, reconstructing a complete geometry for every single timestamp, including occluded regions visible in other frames. Our method demonstrates the state-of-the-art performance on our proposed benchmark for Geometric Complete 4D Reconstruction and the 3D Point Tracking task. Code will be released to support future research.", "date": "2026-03-28", "categories": []}
{"id": "2603.27781", "title": "GS3LAM: Gaussian Semantic Splatting SLAM", "abstract": "Recently, the multi-modal fusion of RGB, depth, and semantics has shown great potential in dense Simultaneous Localization and Mapping (SLAM). However, a prerequisite for generating consistent semantic maps is the availability of dense, efficient, and scalable scene representations. Existing semantic SLAM systems based on explicit representations are often limited by resolution and an inability to predict unknown areas. Conversely, implicit representations typically rely on time-consuming ray tracing, failing to meet real-time requirements. Fortunately, 3D Gaussian Splatting (3DGS) has emerged as a promising representation that combines the efficiency of point-based methods with the continuity of geometric structures. To this end, we propose GS3LAM, a Gaussian Semantic Splatting SLAM framework that processes multimodal data to render consistent, dense semantic maps in real-time. GS3LAM models the scene as a Semantic Gaussian Field (SG-Field) and jointly optimizes camera poses and the field via multimodal error constraints. Furthermore, a Depth-adaptive Scale Regularization (DSR) scheme is introduced to resolve misalignments between scale-invariant Gaussians and geometric surfaces. To mitigate catastrophic forgetting, we propose a Random Sampling-based Keyframe Mapping (RSKM) strategy, which demonstrates superior performance over common local covisibility optimization methods. Extensive experiments on benchmark datasets show that GS3LAM achieves increased tracking robustness, superior rendering quality, and enhanced semantic precision compared to state-of-the-art methods. Source code is available at https://github.com/lif314/GS3LAM.", "date": "2026-03-29", "categories": []}
{"id": "2603.28064", "title": "\\textit{4DSurf}: High-Fidelity Dynamic Scene Surface Reconstruction", "abstract": "This paper addresses the problem of dynamic scene surface reconstruction using Gaussian Splatting (GS), aiming to recover temporally consistent geometry. While existing GS-based dynamic surface reconstruction methods can yield superior reconstruction, they are typically limited to either a single object or objects with only small deformations, struggling to maintain temporally consistent surface reconstruction of large deformations over time. We propose ``\\textit{4DSurf}'', a novel and unified framework for generic dynamic surface reconstruction that does not require specifying the number or types of objects in the scene, can handle large surface deformations and temporal inconsistency in reconstruction. The key innovation of our framework is the introduction of Gaussian deformations induced Signed Distance Function Flow Regularization that constrains the motion of Gaussians to align with the evolving surface. To handle large deformations, we introduce an Overlapping Segment Partitioning strategy that divides the sequence into overlapping segments with small deformations and incrementally passes geometric information across segments through the shared overlapping timestep. Experiments on two challenging dynamic scene datasets, Hi4D and CMU Panoptic, demonstrate that our method outperforms state-of-the-art surface reconstruction methods by 49\\% and 19\\% in Chamfer distance, respectively, and achieves superior temporal consistency under sparse-view settings.", "date": "2026-03-30", "categories": []}
{"id": "2603.28548", "title": "Seen2Scene: Completing Realistic 3D Scenes with Visibility-Guided Flow", "abstract": "We present Seen2Scene, the first flow matching-based approach that trains directly on incomplete, real-world 3D scans for scene completion and generation. Unlike prior methods that rely on complete and hence synthetic 3D data, our approach introduces visibility-guided flow matching, which explicitly masks out unknown regions in real scans, enabling effective learning from real-world, partial observations. We represent 3D scenes using truncated signed distance field (TSDF) volumes encoded in sparse grids and employ a sparse transformer to efficiently model complex scene structures while masking unknown regions. We employ 3D layout boxes as an input conditioning signal, and our approach is flexibly adapted to various other inputs such as text or partial scans. By learning directly from real-world, incomplete 3D scans, Seen2Scene enables realistic 3D scene completion for complex, cluttered real environments. Experiments demonstrate that our model produces coherent, complete, and realistic 3D scenes, outperforming baselines in completion accuracy and generation quality.", "date": "2026-03-30", "categories": []}
{"id": "2511.16542", "title": "EOGS++: Earth Observation Gaussian Splatting with Internal Camera Refinement and Direct Panchromatic Rendering", "abstract": "Recently, 3D Gaussian Splatting has been introduced as a compelling alternative to NeRF for Earth observation, offering competitive reconstruction quality with significantly reduced training times. In this work, we extend the Earth Observation Gaussian Splatting (EOGS) framework to propose EOGS++, a novel method tailored for satellite imagery that directly operates on raw high-resolution panchromatic data without requiring external preprocessing. Furthermore, leveraging optical flow techniques we embed bundle adjustment directly within the training process, avoiding reliance on external optimization tools while improving camera pose estimation. We also introduce several improvements to the original implementation, including early stopping and TSDF post-processing, all contributing to sharper reconstructions and better geometric accuracy. Experiments on the IARPA 2016 and DFC2019 datasets demonstrate that EOGS++ achieves state-of-the-art performance in terms of reconstruction quality and efficiency, outperforming the original EOGS method and other NeRF-based methods while maintaining the computational advantages of Gaussian Splatting. Our model demonstrates an improvement from 1.33 to 1.19 mean MAE errors on buildings compared to the original EOGS models", "date": "2025-11-20", "categories": []}
{"id": "2602.05321", "title": "Wid3R: Wide Field-of-View 3D Reconstruction via Camera Model Conditioning", "abstract": "We present Wid3R, a feed-forward neural network for multi-view visual geometry reconstruction that supports wide field-of-view camera models. Unlike existing methods that assume rectified or pinhole inputs, Wid3R directly models wide-angle imagery without explicit calibration or undistortion. Our approach leverages a ray-based representation with spherical harmonics and introduces a novel camera model token to enable distortion-aware reconstruction. To the best of our knowledge, Wid3R is the first multi-frame feed-forward 3D reconstruction method that supports 360 imagery. Moreover, we show that conditioning on diverse camera types improves generalization to 360 scenes and alleviates data sparsity issues. Wid3R achieves significant performance gains, improving AUC@30 by up to +33.67 on Zip-NeRF (fisheye) and +77.33 on Stanford2D3D (360).", "date": "2026-02-05", "categories": []}
{"id": "2602.21100", "title": "Skullptor: High Fidelity 3D Head Reconstruction in Seconds with Multi-View Normal Prediction", "abstract": "Reconstructing high-fidelity 3D head geometry from images is critical for a wide range of applications, yet existing methods face fundamental limitations. Traditional photogrammetry achieves exceptional detail but requires extensive camera arrays (25-200+ views), substantial computation, and manual cleanup in challenging areas like facial hair. Recent alternatives present a fundamental trade-off: foundation models enable efficient single-image reconstruction but lack fine geometric detail, while optimization-based methods achieve higher fidelity but require dense views and expensive computation. We bridge this gap with a hybrid approach that combines the strengths of both paradigms. Our method introduces a multi-view surface normal prediction model that extends monocular foundation models with cross-view attention to produce geometrically consistent normals in a feed-forward pass. We then leverage these predictions as strong geometric priors within an inverse rendering optimization framework to recover high-frequency surface details. Our approach outperforms state-of-the-art single-image and multi-view methods, achieving high-fidelity reconstruction on par with dense-view photogrammetry while reducing camera requirements and computational cost.", "date": "2026-02-24", "categories": []}
{"id": "2603.24994", "title": "Relaxed Rigidity with Ray-based Grouping for Dynamic Gaussian Splatting", "abstract": "The reconstruction of dynamic 3D scenes using 3D Gaussian Splatting has shown significant promise. A key challenge, however, remains in modeling realistic motion, as most methods fail to align the motion of Gaussians with real-world physical dynamics. This misalignment is particularly problematic for monocular video datasets, where failing to maintain coherent motion undermines local geometric structure, ultimately leading to degraded reconstruction quality. Consequently, many state-of-the-art approaches rely heavily on external priors, such as optical flow or 2D tracks, to enforce temporal coherence. In this work, we propose a novel method to explicitly preserve the local geometric structure of Gaussians across time in 4D scenes. Our core idea is to introduce a view-space ray grouping strategy that clusters Gaussians intersected by the same ray, considering only those whose $\\alpha$-blending weights exceed a threshold. We then apply constraints to these groups to maintain a consistent spatial distribution, effectively preserving their local geometry. This approach enforces a more physically plausible motion model by ensuring that local geometry remains stable over time, eliminating the reliance on external guidance. We demonstrate the efficacy of our method by integrating it into two distinct baseline models. Extensive experiments on challenging monocular datasets show that our approach significantly outperforms existing methods, achieving superior temporal consistency and reconstruction quality.", "date": "2026-03-26", "categories": []}
{"id": "2603.25827", "title": "Fus3D: Decoding Consolidated 3D Geometry from Feed-forward Geometry Transformer Latents", "abstract": "We propose a feed-forward method for dense Signed Distance Field (SDF) regression from unstructured image collections in less than three seconds, without camera calibration or post-hoc fusion. Our key insight is that the intermediate feature space of pretrained multi-view feed-forward geometry transformers already encodes a powerful joint world representation; yet, existing pipelines discard it, routing features through per-view prediction heads before assembling 3D geometry post-hoc, which discards valuable completeness information and accumulates inaccuracies.\n  We instead perform 3D extraction directly from geometry transformer features via learned volumetric extraction: voxelized canonical embeddings that progressively absorb multi-view geometry information through interleaved cross- and self-attention into a structured volumetric latent grid. A simple convolutional decoder then maps this grid to a dense SDF. We additionally propose a scalable, validity-aware supervision scheme directly using SDFs derived from depth maps or 3D assets, tackling practical issues like non-watertight meshes. Our approach yields complete and well-defined distance values across sparse- and dense-view settings and demonstrates geometrically plausible completions. Code and further material can be found at https://lorafib.github.io/fus3d.", "date": "2026-03-26", "categories": []}
{"id": "2603.26181", "title": "GLINT: Modeling Scene-Scale Transparency via Gaussian Radiance Transport", "abstract": "While 3D Gaussian splatting has emerged as a powerful paradigm, it fundamentally fails to model transparency such as glass panels. The core challenge lies in decoupling the intertwined radiance contributions from transparent interfaces and the transmitted geometry observed through the glass. We present GLINT, a framework that models scene-scale transparency through explicit decomposed Gaussian representation. GLINT reconstructs the primary interface and models reflected and transmitted radiance separately, enabling consistent radiance transport. During optimization, GLINT bootstraps transparency localization from geometry-separation cues induced by the decomposition, together with geometry and material priors from a pre-trained video relighting model. Extensive experiments demonstrate consistent improvements over prior methods for reconstructing complex transparent scenes.", "date": "2026-03-27", "categories": []}
{"id": "2603.26638", "title": "Drive-Through 3D Vehicle Exterior Reconstruction via Dynamic-Scene SfM and Distortion-Aware Gaussian Splatting", "abstract": "High-fidelity 3D reconstruction of vehicle exteriors improves buyer confidence in online automotive marketplaces, but generating these models in cluttered dealership drive-throughs presents severe technical challenges. Unlike static-scene photogrammetry, this setting features a dynamic vehicle moving against heavily cluttered, static backgrounds. This problem is further compounded by wide-angle lens distortion, specular automotive paint, and non-rigid wheel rotations that violate classical epipolar constraints. We propose an end-to-end pipeline utilizing a two-pillar camera rig. First, we resolve dynamic-scene ambiguities by coupling SAM 3 for instance segmentation with motion-gating to cleanly isolate the moving vehicle, explicitly masking out non-rigid wheels to enforce strict epipolar geometry. Second, we extract robust correspondences directly on raw, distorted 4K imagery using the RoMa v2 learned matcher guided by semantic confidence masks. Third, these matches are integrated into a rig-aware SfM optimization that utilizes CAD-derived relative pose priors to eliminate scale drift. Finally, we use a distortion-aware 3D Gaussian Splatting framework (3DGUT) coupled with a stochastic Markov Chain Monte Carlo (MCMC) densification strategy to render reflective surfaces. Evaluations on 25 real-world vehicles across 10 dealerships demonstrate that our full pipeline achieves a PSNR of 28.66 dB, an SSIM of 0.89, and an LPIPS of 0.21 on held-out views, representing a 3.85 dB improvement over standard 3D-GS, delivering inspection-grade interactive 3D models without controlled studio infrastructure.", "date": "2026-03-27", "categories": []}
{"id": "2503.07507", "title": "PE3R: Perception-Efficient 3D Reconstruction", "abstract": "Recent advances in 2D-to-3D perception have enabled the recovery of 3D scene semantics from unposed images. However, prevailing methods often suffer from limited generalization, reliance on per-scene optimization, and semantic inconsistencies across viewpoints. To address these limitations, we introduce PE3R, a tuning-free framework for efficient and generalizable 3D semantic reconstruction. By integrating multi-view geometry with 2D semantic priors in a feed-forward pipeline, PE3R achieves zero-shot generalization across diverse scenes and object categories without any scene-specific fine-tuning. Extensive evaluations on open-vocabulary segmentation and multi-view depth estimation show that PE3R not only achieves up to 9$\\times$ faster inference but also sets new state-of-the-art accuracy in both semantic and geometric metrics. Our approach paves the way for scalable, language-driven 3D scene understanding. Code is available at github.com/hujiecpp/PE3R.", "date": "2025-03-10", "categories": []}
{"id": "2512.05272", "title": "Inferring Compositional 4D Scenes without Ever Seeing One", "abstract": "Scenes in the real world are often composed of several static and dynamic objects. Capturing their 4-dimensional structures, composition and spatio-temporal configuration in-the-wild, though extremely interesting, is equally hard. Therefore, existing works often focus on one object at a time, while relying on some category-specific parametric shape model for dynamic objects. This can lead to inconsistent scene configurations, in addition to being limited to the modeled object categories. We propose COM4D (Compositional 4D), a method that consistently and jointly predicts the structure and spatio-temporal configuration of 4D/3D objects using only static multi-object or dynamic single object supervision. We achieve this by a carefully designed training of spatial and temporal attentions on 2D video input. The training is disentangled into learning from object compositions on the one hand, and single object dynamics throughout the video on the other, thus completely avoiding reliance on 4D compositional training data. At inference time, our proposed attention mixing mechanism combines these independently learned attentions, without requiring any 4D composition examples. By alternating between spatial and temporal reasoning, COM4D reconstructs complete and persistent 4D scenes with multiple interacting objects directly from monocular videos. Furthermore, COM4D provides state-of-the-art results in existing separate problems of 4D object and composed 3D reconstruction despite being purely data-driven.", "date": "2025-12-04", "categories": []}
{"id": "2601.00393", "title": "NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videos", "abstract": "In this paper, we propose NeoVerse, a versatile 4D world model that is capable of 4D reconstruction, novel-trajectory video generation, and rich downstream applications. We first identify a common limitation of scalability in current 4D world modeling methods, caused either by expensive and specialized multi-view 4D data or by cumbersome training pre-processing. In contrast, our NeoVerse is built upon a core philosophy that makes the full pipeline scalable to diverse in-the-wild monocular videos. Specifically, NeoVerse features pose-free feed-forward 4D reconstruction, online monocular degradation pattern simulation, and other well-aligned techniques. These designs empower NeoVerse with versatility and generalization to various domains. Meanwhile, NeoVerse achieves state-of-the-art performance in standard reconstruction and generation benchmarks. Our project page is available at https://neoverse-4d.github.io.", "date": "2026-01-01", "categories": []}
{"id": "2601.03824", "title": "IDESplat: Iterative Depth Probability Estimation for Generalizable 3D Gaussian Splatting", "abstract": "Generalizable 3D Gaussian Splatting aims to directly predict Gaussian parameters using a feed-forward network for scene reconstruction. Among these parameters, Gaussian means are particularly difficult to predict, so depth is usually estimated first and then unprojected to obtain the Gaussian sphere centers. Existing methods typically rely solely on a single warp to estimate depth probability, which hinders their ability to fully leverage cross-view geometric cues, resulting in unstable and coarse depth maps. To address this limitation, we propose IDESplat, which iteratively applies warp operations to boost depth probability estimation for accurate Gaussian mean prediction. First, to eliminate the inherent instability of a single warp, we introduce a Depth Probability Boosting Unit (DPBU) that integrates epipolar attention maps produced by cascading warp operations in a multiplicative manner. Next, we construct an iterative depth estimation process by stacking multiple DPBUs, progressively identifying potential depth candidates with high likelihood. As IDESplat iteratively boosts depth probability estimates and updates the depth candidates, the depth map is gradually refined, resulting in accurate Gaussian means. We conduct experiments on RealEstate10K, ACID, and DL3DV. IDESplat achieves outstanding reconstruction quality and state-of-the-art performance with real-time efficiency. On RE10K, it outperforms DepthSplat by 0.33 dB in PSNR, using only 10.7% of the parameters and 70% of the memory. Additionally, our IDESplat improves PSNR by 2.95 dB over DepthSplat on the DTU dataset in cross-dataset experiments, demonstrating its strong generalization ability.", "date": "2026-01-07", "categories": []}
{"id": "2602.07444", "title": "Perspective-aware fusion of incomplete depth maps and surface normals for accurate 3D reconstruction", "abstract": "We address the problem of reconstructing 3D surfaces from depth and surface normal maps acquired by a sensor system based on a single perspective camera. Depth and normal maps can be obtained through techniques such as structured-light scanning and photometric stereo, respectively. We propose a perspective-aware log-depth fusion approach that extends existing orthographic gradient-based depth-normals fusion methods by explicitly accounting for perspective projection, leading to metrically accurate 3D reconstructions. Additionally, the method handles missing depth measurements by leveraging available surface normal information to inpaint gaps. Experiments on the DiLiGenT-MV data set demonstrate the effectiveness of our approach and highlight the importance of perspective-aware depth-normals fusion.", "date": "2026-02-07", "categories": []}
{"id": "2603.19682", "title": "3D Gaussian Splatting with Self-Constrained Priors for High Fidelity Surface Reconstruction", "abstract": "Rendering 3D surfaces has been revolutionized within the modeling of radiance fields through either 3DGS or NeRF. Although 3DGS has shown advantages over NeRF in terms of rendering quality or speed, there is still room for improvement in recovering high fidelity surfaces through 3DGS. To resolve this issue, we propose a self-constrained prior to constrain the learning of 3D Gaussians, aiming for more accurate depth rendering. Our self-constrained prior is derived from a TSDF grid that is obtained by fusing the depth maps rendered with current 3D Gaussians. The prior measures a distance field around the estimated surface, offering a band centered at the surface for imposing more specific constraints on 3D Gaussians, such as removing Gaussians outside the band, moving Gaussians closer to the surface, and encouraging larger or smaller opacity in a geometry-aware manner. More importantly, our prior can be regularly updated by the most recent depth images which are usually more accurate and complete. In addition, the prior can also progressively narrow the band to tighten the imposed constraints. We justify our idea and report our superiority over the state-of-the-art methods in evaluations on widely used benchmarks.", "date": "2026-03-20", "categories": []}
{"id": "2603.22893", "title": "SLARM: Streaming and Language-Aligned Reconstruction Model for Dynamic Scenes", "abstract": "We propose SLARM, a feed-forward model that unifies dynamic scene reconstruction, semantic understanding, and real-time streaming inference. SLARM captures complex, non-uniform motion through higher-order motion modeling, trained solely on differentiable renderings without any flow supervision. Besides, SLARM distills semantic features from LSeg to obtain language-aligned representations. This design enables semantic querying via natural language, and the tight coupling between semantics and geometry further enhances the accuracy and robustness of dynamic reconstruction. Moreover, SLARM processes image sequences using window-based causal attention, achieving stable, low-latency streaming inference without accumulating memory cost. Within this unified framework, SLARM achieves state-of-the-art results in dynamic estimation, rendering quality, and scene parsing, improving motion accuracy by 21%, reconstruction PSNR by 1.6 dB, and segmentation mIoU by 20% over existing methods.", "date": "2026-03-24", "categories": []}
{"id": "2603.24684", "title": "KitchenTwin: Semantically and Geometrically Grounded 3D Kitchen Digital Twins", "abstract": "Embodied AI training and evaluation require object-centric digital twin environments with accurate metric geometry and semantic grounding. Recent transformer-based feedforward reconstruction methods can efficiently predict global point clouds from sparse monocular videos, yet these geometries suffer from inherent scale ambiguity and inconsistent coordinate conventions. This mismatch prevents the reliable fusion of these dimensionless point cloud predictions with locally reconstructed object meshes. We propose a novel scale-aware 3D fusion framework that registers visually grounded object meshes with transformer-predicted global point clouds to construct metrically consistent digital twins. Our method introduces a Vision-Language Model (VLM)-guided geometric anchor mechanism that resolves this fundamental coordinate mismatch by recovering an accurate real-world metric scale. To fuse these networks, we propose a geometry-aware registration pipeline that explicitly enforces physical plausibility through gravity-aligned vertical estimation, Manhattan-world structural constraints, and collision-free local refinement. Experiments on real indoor kitchen environments demonstrate improved cross-network object alignment and geometric consistency for downstream tasks, including multi-primitive fitting and metric measurement. We additionally introduce an open-source indoor digital twin dataset with metrically scaled scenes and semantically grounded and registered object-centric mesh annotations.", "date": "2026-03-25", "categories": []}
{"id": "2603.24770", "title": "DRoPS: Dynamic 3D Reconstruction of Pre-Scanned Objects", "abstract": "Dynamic scene reconstruction from casual videos has seen recent remarkable progress. Numerous approaches have attempted to overcome the ill-posedness of the task by distilling priors from 2D foundational models and by imposing hand-crafted regularization on the optimized motion. However, these methods struggle to reconstruct scenes from extreme novel viewpoints, especially when highly articulated motions are present. In this paper, we present DRoPS, a novel approach that leverages a static pre-scan of the dynamic object as an explicit geometric and appearance prior. While existing state-of-the-art methods fail to fully exploit the pre-scan, DRoPS leverages our novel setup to effectively constrain the solution space and ensure geometrical consistency throughout the sequence. The core of our novelty is twofold: first, we establish a grid-structured and surface-aligned model by organizing Gaussian primitives into pixel grids anchored to the object surface. Second, by leveraging the grid structure of our primitives, we parameterize motion using a CNN conditioned on those grids, injecting strong implicit regularization and correlating the motion of nearby points. Extensive experiments demonstrate that our method significantly outperforms the current state of the art in rendering quality and 3D tracking accuracy.", "date": "2026-03-25", "categories": []}
{"id": "2603.25042", "title": "MoRGS: Efficient Per-Gaussian Motion Reasoning for Streamable Dynamic 3D Scenes", "abstract": "Online reconstruction of dynamic scenes aims to learn from streaming multi-view inputs under low-latency constraints. The fast training and real-time rendering capabilities of 3D Gaussian Splatting have made on-the-fly reconstruction practically feasible, enabling online 4D reconstruction. However, existing online approaches, despite their efficiency and visual quality, fail to learn per-Gaussian motion that reflects true scene dynamics. Without explicit motion cues, appearance and motion are optimized solely under photometric loss, causing per-Gaussian motion to chase pixel residuals rather than true 3D motion. To address this, we propose MoRGS, an efficient online per-Gaussian motion reasoning framework that explicitly models per-Gaussian motion to improve 4D reconstruction quality. Specifically, we leverage optical flow on a sparse set of key views as lightweight motion cues that regularize per-Gaussian motion beyond photometric supervision. To compensate for the sparsity of flow supervision, we learn a per-Gaussian motion offset field that reconciles discrepancies between projected 3D motion and observed flow across views and time. In addition, we introduce a per-Gaussian motion confidence that separates dynamic from static Gaussians and weights Gaussian attribute residual updates, thereby suppressing redundant motion in static regions for better temporal consistency and accelerating the modeling of large motions. Extensive experiments demonstrate that MoRGS achieves state-of-the-art reconstruction quality and motion fidelity among online methods, while maintaining streamable performance.", "date": "2026-03-26", "categories": []}
{"id": "2603.25058", "title": "Learning Explicit Continuous Motion Representation for Dynamic Gaussian Splatting from Monocular Videos", "abstract": "We present an approach for high-quality dynamic Gaussian Splatting from monocular videos. To this end, we in this work go one step further beyond previous methods to explicitly model continuous position and orientation deformation of dynamic Gaussians, using an SE(3) B-spline motion bases with a compact set of control points. To improve computational efficiency while enhancing the ability to model complex motions, an adaptive control mechanism is devised to dynamically adjust the number of motion bases and control points. Besides, we develop a soft segment reconstruction strategy to mitigate long-interval motion interference, and employ a multi-view diffusion model to provide multi-view cues for avoiding overfitting to training views. Extensive experiments demonstrate that our method outperforms state-of-the-art methods in novel view synthesis. Our code is available at https://github.com/hhhddddddd/se3bsplinegs.", "date": "2026-03-26", "categories": []}
{"id": "2603.25265", "title": "ViewSplat: View-Adaptive Dynamic Gaussian Splatting for Feed-Forward Synthesis", "abstract": "We present ViewSplat, a view-adaptive 3D Gaussian splatting network for novel view synthesis from unposed images. While recent feed-forward 3D Gaussian splatting has significantly accelerated 3D scene reconstruction by bypassing per-scene optimization, a fundamental fidelity gap remains. We attribute this bottleneck to the limited capacity of single-step feed-forward networks to regress static Gaussian primitives that satisfy all viewpoints. To address this limitation, we shift the paradigm from static primitive regression to view-adaptive dynamic splatting. Instead of a rigid Gaussian representation, our pipeline learns a view-adaptable latent representation. Specifically, ViewSplat initially predicts base Gaussian primitives alongside the weights of dynamic MLPs. During rendering, these MLPs take target view coordinates as input and predict view-dependent residual updates for each Gaussian attribute (i.e., 3D position, scale, rotation, opacity, and color). This mechanism, which we term view-adaptive dynamic splatting, allows each primitive to rectify initial estimation errors, effectively capturing high-fidelity appearances. Extensive experiments demonstrate that ViewSplat achieves state-of-the-art fidelity while maintaining fast inference (17 FPS) and real-time rendering (154 FPS).", "date": "2026-03-26", "categories": []}
{"id": "2603.25494", "title": "AdaSFormer: Adaptive Serialized Transformers for Monocular Semantic Scene Completion from Indoor Environments", "abstract": "Indoor monocular semantic scene completion (MSSC) is notably more challenging than its outdoor counterpart due to complex spatial layouts and severe occlusions. While transformers are well suited for modeling global dependencies, their high memory cost and difficulty in reconstructing fine-grained details have limited their use in indoor MSSC. To address these limitations, we introduce AdaSFormer, a serialized transformer framework tailored for indoor MSSC. Our model features three key designs: (1) an Adaptive Serialized Transformer with learnable shifts that dynamically adjust receptive fields; (2) a Center-Relative Positional Encoding that captures spatial information richness; and (3) a Convolution-Modulated Layer Normalization that bridges heterogeneous representations between convolutional and transformer features. Extensive experiments on NYUv2 and Occ-ScanNet demonstrate that AdaSFormer achieves state-of-the-art performance. The code is publicly available at: https://github.com/alanWXZ/AdaSFormer.", "date": "2026-03-26", "categories": []}
{"id": "2512.01540", "title": "FlashVGGT: Efficient and Scalable Visual Geometry Transformers with Compressed Descriptor Attention", "abstract": "3D reconstruction from multi-view images is a core challenge in computer vision. Recently, feed-forward methods have emerged as efficient and robust alternatives to traditional per-scene optimization techniques. Among them, state-of-the-art models like the Visual Geometry Grounding Transformer (VGGT) leverage full self-attention over all image tokens to capture global relationships. However, this approach suffers from poor scalability due to the quadratic complexity of self-attention and the large number of tokens generated in long image sequences. In this work, we introduce FlashVGGT, an efficient alternative that addresses this bottleneck through a descriptor-based attention mechanism. Instead of applying dense global attention across all tokens, FlashVGGT compresses spatial information from each frame into a compact set of descriptor tokens. Global attention is then computed as cross-attention between the full set of image tokens and this smaller descriptor set, significantly reducing computational overhead. Moreover, the compactness of the descriptors enables online inference over long sequences via a chunk-recursive mechanism that reuses cached descriptors from previous chunks. Experimental results show that FlashVGGT achieves reconstruction accuracy competitive with VGGT while reducing inference time to just 9.3% of VGGT for 1,000 images, and scaling efficiently to sequences exceeding 3,000 images. Our project page is available at https://wzpscott.github.io/flashvggt_page/.", "date": "2025-12-01", "categories": []}
{"id": "2603.01601", "title": "Dehallu3D: Hallucination-Mitigated 3D Generation from Single Image via Cyclic View Consistency Refinement", "abstract": "Large 3D reconstruction models have revolutionized the 3D content generation field, enabling broad applications in virtual reality and gaming. Just like other large models, large 3D reconstruction models suffer from hallucinations as well, introducing structural outliers (e.g., odd holes or protrusions) that deviate from the input data. However, unlike other large models, hallucinations in large 3D reconstruction models remain severely underexplored, leading to malformed 3D-printed objects or insufficient immersion in virtual scenes. Such hallucinations majorly originate from that existing methods reconstruct 3D content from sparsely generated multi-view images which suffer from large viewpoint gaps and discontinuities. To mitigate hallucinations by eliminating the outliers, we propose Dehallu3D for 3D mesh generation. Our key idea is to design a balanced multi-view continuity constraint to enforce smooth transitions across dense intermediate viewpoints, while avoiding over-smoothing that could erase sharp geometric features. Therefore, Dehallu3D employs a plug-and-play optimization module with two key constraints: (i) adjacent consistency to ensure geometric continuity across views, and (ii) adaptive smoothness to retain fine details.We further propose the Outlier Risk Measure (ORM) metric to quantify geometric fidelity in 3D generation from the perspective of outliers. Extensive experiments show that Dehallu3D achieves high-fidelity 3D generation by effectively preserving structural details while removing hallucinated outliers.", "date": "2026-03-02", "categories": []}
{"id": "2603.21304", "title": "F4Splat: Feed-Forward Predictive Densification for Feed-Forward 3D Gaussian Splatting", "abstract": "Feed-forward 3D Gaussian Splatting methods enable single-pass reconstruction and real-time rendering. However, they typically adopt rigid pixel-to-Gaussian or voxel-to-Gaussian pipelines that uniformly allocate Gaussians, leading to redundant Gaussians across views. Moreover, they lack an effective mechanism to control the total number of Gaussians while maintaining reconstruction fidelity. To address these limitations, we present F4Splat, which performs Feed-Forward predictive densification for Feed-Forward 3D Gaussian Splatting, introducing a densification-score-guided allocation strategy that adaptively distributes Gaussians according to spatial complexity and multi-view overlap. Our model predicts per-region densification scores to estimate the required Gaussian density and allows explicit control over the final Gaussian budget without retraining. This spatially adaptive allocation reduces redundancy in simple regions and minimizes duplicate Gaussians across overlapping views, producing compact yet high-quality 3D representations. Extensive experiments demonstrate that our model achieves superior novel-view synthesis performance compared to prior uncalibrated feed-forward methods, while using significantly fewer Gaussians.", "date": "2026-03-22", "categories": []}
{"id": "2508.03643", "title": "Uni3R: Unified 3D Reconstruction and Semantic Understanding via Generalizable Gaussian Splatting from Unposed Multi-View Images", "abstract": "Reconstructing and semantically interpreting 3D scenes from sparse 2D views remains a fundamental challenge in computer vision. Conventional methods often decouple semantic understanding from reconstruction or necessitate costly per-scene optimization, thereby restricting their scalability and generalizability. In this paper, we introduce Uni3R, a novel feed-forward framework that jointly reconstructs a unified 3D scene representation enriched with open-vocabulary semantics, directly from unposed multi-view images. Our approach leverages a Cross-View Transformer to robustly integrate information across arbitrary multi-view inputs, which then regresses a set of 3D Gaussian primitives endowed with semantic feature fields. This unified representation facilitates high-fidelity novel view synthesis, open-vocabulary 3D semantic segmentation, and depth prediction, all within a single, feed-forward pass. Extensive experiments demonstrate that Uni3R establishes a new state-of-the-art across multiple benchmarks, including 25.07 PSNR on RE10K and 55.84 mIoU on ScanNet. Our work signifies a novel paradigm towards generalizable, unified 3D scene reconstruction and understanding. The code is available at https://github.com/HorizonRobotics/Uni3R.", "date": "2025-08-05", "categories": []}
{"id": "2603.21064", "title": "2Xplat: Two Experts Are Better Than One Generalist", "abstract": "Pose-free feed-forward 3D Gaussian Splatting (3DGS) has opened a new frontier for rapid 3D modeling, enabling high-quality Gaussian representations to be generated from uncalibrated multi-view images in a single forward pass. The dominant approach in this space adopts unified monolithic architectures, often built on geometry-centric 3D foundation models, to jointly estimate camera poses and synthesize 3DGS representations within a single network. While architecturally streamlined, such \"all-in-one\" designs may be suboptimal for high-fidelity 3DGS generation, as they entangle geometric reasoning and appearance modeling within a shared representation. In this work, we introduce 2Xplat, a pose-free feed-forward 3DGS framework based on a two-expert design that explicitly separates geometry estimation from Gaussian generation. A dedicated geometry expert first predicts camera poses, which are then explicitly passed to a powerful appearance expert that synthesizes 3D Gaussians. Despite its conceptual simplicity, being largely underexplored in prior works, the proposed approach proves highly effective. In fewer than 5K training iterations, the proposed two-experts pipeline substantially outperforms prior pose-free feed-forward 3DGS approaches and achieves performance on par with state-of-the-art posed methods. These results challenge the prevailing unified paradigm and suggest the potential advantages of modular design principles for complex 3D geometric estimation and appearance synthesis tasks.", "date": "2026-03-22", "categories": []}
{"id": "2603.22450", "title": "Static Scene Reconstruction from Dynamic Egocentric Videos", "abstract": "Egocentric videos present unique challenges for 3D reconstruction due to rapid camera motion and frequent dynamic interactions. State-of-the-art static reconstruction systems, such as MapAnything, often degrade in these settings, suffering from catastrophic trajectory drift and \"ghost\" geometry caused by moving hands. We bridge this gap by proposing a robust pipeline that adapts static reconstruction backbones to long-form egocentric video. Our approach introduces a mask-aware reconstruction mechanism that explicitly suppresses dynamic foreground in the attention layers, preventing hand artifacts from contaminating the static map. Furthermore, we employ a chunked reconstruction strategy with pose-graph stitching to ensure global consistency and eliminate long-term drift. Experiments on HD-EPIC and indoor drone datasets demonstrate that our pipeline significantly improves absolute trajectory error and yields visually clean static geometry compared to naive baselines, effectively extending the capability of foundation models to dynamic first-person scenes.", "date": "2026-03-23", "categories": []}
{"id": "2603.22572", "title": "FullCircle: Effortless 3D Reconstruction from Casual 360$^\\circ$ Captures", "abstract": "Radiance fields have emerged as powerful tools for 3D scene reconstruction. However, casual capture remains challenging due to the narrow field of view of perspective cameras, which limits viewpoint coverage and feature correspondences necessary for reliable camera calibration and reconstruction. While commercially available 360$^\\circ$ cameras offer significantly broader coverage than perspective cameras for the same capture effort, existing 360$^\\circ$ reconstruction methods require special capture protocols and pre-processing steps that undermine the promise of radiance fields: effortless workflows to capture and reconstruct 3D scenes. We propose a practical pipeline for reconstructing 3D scenes directly from raw 360$^\\circ$ camera captures. We require no special capture protocols or pre-processing, and exhibit robustness to a prevalent source of reconstruction errors: the human operator that is visible in all 360$^\\circ$ imagery. To facilitate evaluation, we introduce a multi-tiered dataset of scenes captured as raw dual-fisheye images, establishing a benchmark for robust casual 360$^\\circ$ reconstruction. Our method significantly outperforms not only vanilla 3DGS for 360$^\\circ$ cameras but also robust perspective baselines when perspective cameras are simulated from the same capture, demonstrating the advantages of 360$^\\circ$ capture for casual reconstruction. Additional results are available at: https://theialab.github.io/fullcircle", "date": "2026-03-23", "categories": []}
{"id": "2603.22631", "title": "CAM3R: Camera-Agnostic Model for 3D Reconstruction", "abstract": "Recovering dense 3D geometry from unposed images remains a foundational challenge in computer vision. Current state-of-the-art models are predominantly trained on perspective datasets, which implicitly constrains them to a standard pinhole camera geometry. As a result, these models suffer from significant geometric degradation when applied to wide-angle imagery captured via non-rectilinear optics, such as fisheye or panoramic sensors. To address this, we present CAM3R, a Camera-Agnostic, feed-forward Model for 3D Reconstruction capable of processing images from wide-angle camera models without prior calibration. Our framework consists of a two-view network which is bifurcated into a Ray Module (RM) to estimate per-pixel ray directions and a Cross-view Module (CVM) to infer radial distance with confidence maps, pointmaps, and relative poses. To unify these pairwise predictions into a consistent 3D scene, we introduce a Ray-Aware Global Alignment framework for pose refinement and scale optimization while strictly preserving the predicted local geometry. Extensive experiments on various camera model datasets, including panorama, fisheye and pinhole imagery, demonstrate that CAM3R establishes a new state-of-the-art in pose estimation and reconstruction.", "date": "2026-03-23", "categories": []}
{"id": "2603.22851", "title": "UniQueR: Unified Query-based Feedforward 3D Reconstruction", "abstract": "We present UniQueR, a unified query-based feedforward framework for efficient and accurate 3D reconstruction from unposed images. Existing feedforward models such as DUSt3R, VGGT, and AnySplat typically predict per-pixel point maps or pixel-aligned Gaussians, which remain fundamentally 2.5D and limited to visible surfaces. In contrast, UniQueR formulates reconstruction as a sparse 3D query inference problem. Our model learns a compact set of 3D anchor points that act as explicit geometric queries, enabling the network to infer scene structure, including geometry in occluded regions--in a single forward pass. Each query encodes spatial and appearance priors directly in global 3D space (instead of per-frame camera space) and spawns a set of 3D Gaussians for differentiable rendering. By leveraging unified query interactions across multi-view features and a decoupled cross-attention design, UniQueR achieves strong geometric expressiveness while substantially reducing memory and computational cost. Experiments on Mip-NeRF 360 and VR-NeRF demonstrate that UniQueR surpasses state-of-the-art feedforward methods in both rendering quality and geometric accuracy, using an order of magnitude fewer primitives than dense alternatives.", "date": "2026-03-24", "categories": []}
{"id": "2603.23297", "title": "Drop-In Perceptual Optimization for 3D Gaussian Splatting", "abstract": "Despite their output being ultimately consumed by human viewers, 3D Gaussian Splatting (3DGS) methods often rely on ad-hoc combinations of pixel-level losses, resulting in blurry renderings. To address this, we systematically explore perceptual optimization strategies for 3DGS by searching over a diverse set of distortion losses. We conduct the first-of-its-kind large-scale human subjective study on 3DGS, involving 39,320 pairwise ratings across several datasets and 3DGS frameworks. A regularized version of Wasserstein Distortion, which we call WD-R, emerges as the clear winner, excelling at recovering fine textures without incurring a higher splat count. WD-R is preferred by raters more than $2.3\\times$ over the original 3DGS loss, and $1.5\\times$ over current best method Perceptual-GS. WD-R also consistently achieves state-of-the-art LPIPS, DISTS, and FID scores across various datasets, and generalizes across recent frameworks, such as Mip-Splatting and Scaffold-GS, where replacing the original loss with WD-R consistently enhances perceptual quality within a similar resource budget (number of splats for Mip-Splatting, model size for Scaffold-GS), and leads to reconstructions being preferred by human raters $1.8\\times$ and $3.6\\times$, respectively. We also find that this carries over to the task of 3DGS scene compression, with $\\approx 50\\%$ bitrate savings for comparable perceptual metric performance.", "date": "2026-03-23", "categories": []}
{"id": "2511.19235", "title": "IDSplat: Instance-Decomposed 3D Gaussian Splatting for Driving Scenes", "abstract": "Reconstructing dynamic driving scenes is essential for developing autonomous systems through sensor-realistic simulation. Although recent methods achieve high-fidelity reconstructions, they either rely on costly human annotations for object trajectories or use time-varying representations without explicit object-level decomposition, leading to intertwined static and dynamic elements that hinder scene separation. We present IDSplat, a self-supervised 3D Gaussian Splatting framework that reconstructs dynamic scenes with explicit instance decomposition and learnable motion trajectories, without requiring human annotations. Our key insight is to model dynamic objects as coherent instances undergoing rigid transformations, rather than unstructured time-varying primitives. For instance decomposition, we employ zero-shot, language-grounded video tracking anchored to 3D using lidar, and estimate consistent poses via feature correspondences. We introduce a coordinated-turn smoothing scheme to obtain temporally and physically consistent motion trajectories, mitigating pose misalignments and tracking failures, followed by joint optimization of object poses and Gaussian parameters. Experiments on the Waymo Open Dataset demonstrate that our method achieves competitive reconstruction quality while maintaining instance-level decomposition and generalizes across diverse sequences and view densities without retraining, making it practical for large-scale autonomous driving applications. Code will be released.", "date": "2025-11-24", "categories": []}
{"id": "2511.21565", "title": "UAVLight: A Benchmark for Illumination-Robust 3D Reconstruction in Unmanned Aerial Vehicle (UAV) Scenes", "abstract": "Illumination inconsistency is a fundamental challenge in multi-view 3D reconstruction. Variations in sunlight direction, cloud cover, and shadows break the constant-lighting assumption underlying both classical multi-view stereo (MVS) and structure from motion (SfM) pipelines and recent neural rendering methods, leading to geometry drift, color inconsistency, and shadow imprinting. This issue is especially critical in UAV-based reconstruction, where long flight durations and outdoor environments make lighting changes unavoidable. However, existing datasets either restrict capture to short time windows, thus lacking meaningful illumination diversity, or span months and seasons, where geometric and semantic changes confound the isolated study of lighting robustness. We introduce UAVLight, a controlled-yet-real benchmark for illumination-robust 3D reconstruction. Each scene is captured along repeatable, geo-referenced flight paths at multiple fixed times of day, producing natural lighting variation under consistent geometry, calibration, and viewpoints. With standardized evaluation protocols across lighting conditions, UAVLight provides a reliable foundation for developing and benchmarking reconstruction methods that are consistent, faithful, and relightable in real outdoor environments.", "date": "2025-11-26", "categories": []}
{"id": "2512.13157", "title": "Intrinsic Image Fusion for Multi-View 3D Material Reconstruction", "abstract": "We introduce Intrinsic Image Fusion, a method that reconstructs high-quality physically based materials from multi-view images. Material reconstruction is highly underconstrained and typically relies on analysis-by-synthesis, which requires expensive and noisy path tracing. To better constrain the optimization, we incorporate single-view priors into the reconstruction process. We leverage a diffusion-based material estimator that produces multiple, but often inconsistent, candidate decompositions per view. To reduce the inconsistency, we fit an explicit low-dimensional parametric function to the predictions. We then propose a robust optimization framework using soft per-view prediction selection together with confidence-based soft multi-view inlier set to fuse the most consistent predictions of the most confident views into a consistent parametric material space. Finally, we use inverse path tracing to optimize for the low-dimensional parameters. Our results outperform state-of-the-art methods in material disentanglement on both synthetic and real scenes, producing sharp and clean reconstructions suitable for high-quality relighting.", "date": "2025-12-15", "categories": []}
{"id": "2601.07242", "title": "HERE: Hierarchical Active Exploration of Radiance Field with Epistemic Uncertainty Minimization", "abstract": "We present HERE, an active 3D scene reconstruction framework based on neural radiance fields, enabling high-fidelity implicit mapping. Our approach centers around an active learning strategy for camera trajectory generation, driven by accurate identification of unseen regions, which supports efficient data acquisition and precise scene reconstruction. The key to our approach is epistemic uncertainty quantification based on evidential deep learning, which directly captures data insufficiency and exhibits a strong correlation with reconstruction errors. This allows our framework to more reliably identify unexplored or poorly reconstructed regions compared to existing methods, leading to more informed and targeted exploration. Additionally, we design a hierarchical exploration strategy that leverages learned epistemic uncertainty, where local planning extracts target viewpoints from high-uncertainty voxels based on visibility for trajectory generation, and global planning uses uncertainty to guide large-scale coverage for efficient and comprehensive reconstruction. The effectiveness of the proposed method in active 3D reconstruction is demonstrated by achieving higher reconstruction completeness compared to previous approaches on photorealistic simulated scenes across varying scales, while a hardware demonstration further validates its real-world applicability. Project page: https://taekbum.github.io/here/", "date": "2026-01-12", "categories": []}
{"id": "2603.03744", "title": "DAGE: Dual-Stream Architecture for Efficient and Fine-Grained Geometry Estimation", "abstract": "Estimating accurate, view-consistent geometry and camera poses from uncalibrated multi-view/video inputs remains challenging - especially at high spatial resolutions and over long sequences. We present DAGE, a dual-stream transformer whose main novelty is to disentangle global coherence from fine detail. A low-resolution stream operates on aggressively downsampled frames with alternating frame/global attention to build a view-consistent representation and estimate cameras efficiently, while a high-resolution stream processes the original images per-frame to preserve sharp boundaries and small structures. A lightweight adapter fuses these streams via cross-attention, injecting global context without disturbing the pretrained single-frame pathway. This design scales resolution and clip length independently, supports inputs up to 2K, and maintains practical inference cost. DAGE delivers sharp depth/pointmaps, strong cross-view consistency, and accurate poses, establishing new state-of-the-art results for video geometry estimation and multi-view reconstruction.", "date": "2026-03-04", "categories": []}
{"id": "2603.20337", "title": "High-fidelity Multi-view Normal Integration with Scale-encoded Neural Surface Representation", "abstract": "Previous multi-view normal integration methods typically sample a single ray per pixel, without considering the spatial area covered by each pixel, which varies with camera intrinsics and the camera-to-object distance. Consequently, when the target object is captured at different distances, the normals at corresponding pixels may differ across views. This multi-view surface normal inconsistency results in the blurring of high-frequency details in the reconstructed surface. To address this issue, we propose a scale-encoded neural surface representation that incorporates the pixel coverage area into the neural representation. By associating each 3D point with a spatial scale and calculating its normal from a hybrid grid-based encoding, our method effectively represents multi-scale surface normals captured at varying distances. Furthermore, to enable scale-aware surface reconstruction, we introduce a mesh extraction module that assigns an optimal local scale to each vertex based on the training observations. Experimental results demonstrate that our approach consistently yields high-fidelity surface reconstruction from normals observed at varying distances, outperforming existing multi-view normal integration methods.", "date": "2026-03-20", "categories": []}
{"id": "2603.20588", "title": "RayMap3R: Inference-Time RayMap for Dynamic 3D Reconstruction", "abstract": "Streaming feed-forward 3D reconstruction enables real-time joint estimation of scene geometry and camera poses from RGB images. However, without explicit dynamic reasoning, streaming models can be affected by moving objects, causing artifacts and drift. In this work, we propose RayMap3R, a training-free streaming framework for dynamic scene reconstruction. We observe that RayMap-based predictions exhibit a static-scene bias, providing an internal cue for dynamic identification. Based on this observation, we construct a dual-branch inference scheme that identifies dynamic regions by contrasting RayMap and image predictions, suppressing their interference during memory updates. We further introduce reset metric alignment and state-aware smoothing to preserve metric consistency and stabilize predicted trajectories. Our method achieves state-of-the-art performance among streaming approaches on dynamic scene reconstruction across multiple benchmarks.", "date": "2026-03-21", "categories": []}
{"id": "2603.21166", "title": "Training-Free Instance-Aware 3D Scene Reconstruction and Diffusion-Based View Synthesis from Sparse Images", "abstract": "We introduce a novel, training-free system for reconstructing, understanding, and rendering 3D indoor scenes from a sparse set of unposed RGB images. Unlike traditional radiance field approaches that require dense views and per-scene optimization, our pipeline achieves high-fidelity results without any training or pose preprocessing. The system integrates three key innovations: (1) A robust point cloud reconstruction module that filters unreliable geometry using a warping-based anomaly removal strategy; (2) A warping-guided 2D-to-3D instance lifting mechanism that propagates 2D segmentation masks into a consistent, instance-aware 3D representation; and (3) A novel rendering approach that projects the point cloud into new views and refines the renderings with a 3D-aware diffusion model. Our method leverages the generative power of diffusion to compensate for missing geometry and enhances realism, especially under sparse input conditions. We further demonstrate that object-level scene editing such as instance removal can be naturally supported in our pipeline by modifying only the point cloud, enabling the synthesis of consistent, edited views without retraining. Our results establish a new direction for efficient, editable 3D content generation without relying on scene-specific optimization. Project page: https://jiatongxia.github.io/TID3R/", "date": "2026-03-22", "categories": []}
{"id": "2603.21436", "title": "PAS3R: Pose-Adaptive Streaming 3D Reconstruction for Long Video Sequences", "abstract": "Online monocular 3D reconstruction enables dense scene recovery from streaming video but remains fundamentally limited by the stability-adaptation dilemma: the reconstruction model must rapidly incorporate novel viewpoints while preserving previously accumulated scene structure. Existing streaming approaches rely on uniform or attention-based update mechanisms that often fail to account for abrupt viewpoint transitions, leading to trajectory drift and geometric inconsistencies over long sequences. We introduce PAS3R, a pose-adaptive streaming reconstruction framework that dynamically modulates state updates according to camera motion and scene structure. Our key insight is that frames contributing significant geometric novelty should exert stronger influence on the reconstruction state, while frames with minor viewpoint variation should prioritize preserving historical context. PAS3R operationalizes this principle through a motion-aware update mechanism that jointly leverages inter-frame pose variation and image frequency cues to estimate frame importance. To further stabilize long-horizon reconstruction, we introduce trajectory-consistent training objectives that incorporate relative pose constraints and acceleration regularization. A lightweight online stabilization module further suppresses high-frequency trajectory jitter and geometric artifacts without increasing memory consumption. Extensive experiments across multiple benchmarks demonstrate that PAS3R significantly improves trajectory accuracy, depth estimation, and point cloud reconstruction quality in long video sequences while maintaining competitive performance on shorter sequences.", "date": "2026-03-22", "categories": []}
{"id": "2603.21618", "title": "4DGS360: 360{\\deg} Gaussian Reconstruction of Dynamic Objects from a Single Video", "abstract": "We introduce 4DGS360, a diffusion-free framework for 360$^{\\circ}$ dynamic object reconstruction from casual monocular video. Existing methods often fail to reconstruct consistent 360$^{\\circ}$ geometry, as their heavy reliance on 2D-native priors causes initial points to overfit to visible surface in each training view. 4DGS360 addresses this challenge through a advanced 3D-native initialization that mitigates the geometric ambiguity of occluded regions. Our proposed 3D tracker, AnchorTAP3D, produces reinforced 3D point trajectories by leveraging confident 2D track points as anchors, suppressing drift and providing reliable initialization that preserves geometry in occluded regions. This initialization, combined with optimization, yields coherent 360$^{\\circ}$ 4D reconstructions. We further present iPhone360, a new benchmark where test cameras are placed up to 135$^{\\circ}$ apart from training views, enabling 360$^{\\circ}$ evaluation that existing datasets cannot provide. Experiments show that 4DGS360 achieves state-of-the-art performance on the iPhone360, iPhone, and DAVIS datasets, both qualitatively and quantitatively.", "date": "2026-03-23", "categories": []}
{"id": "2603.21931", "title": "SatGeo-NeRF: Geometrically Regularized NeRF for Satellite Imagery", "abstract": "We present SatGeo-NeRF, a geometrically regularized NeRF for satellite imagery that mitigates overfitting-induced geometric artifacts observed in current state-of-the-art models using three model-agnostic regularizers. Gravity-Aligned Planarity Regularization aligns depth-inferred, approximated surface normals with the gravity axis to promote local planarity, coupling adjacent rays via a corresponding surface approximation to facilitate cross-ray gradient flow. Granularity Regularization enforces a coarse-to-fine geometry-learning scheme, and Depth-Supervised Regularization stabilizes early training for improved geometric accuracy. On the DFC2019 satellite reconstruction benchmark, SatGeo-NeRF improves the Mean Altitude Error by 13.9% and 11.7% relative to state-of-the-art baselines such as EO-NeRF and EO-GS.", "date": "2026-03-23", "categories": []}
{"id": "2404.10318", "title": "SRGS: Super-Resolution 3D Gaussian Splatting", "abstract": "Low-resolution (LR) multi-view capture limits the fidelity of 3D Gaussian Splatting (3DGS). 3DGS super-resolution (SR) is therefore important, yet challenging because it must recover missing high-frequency details while enforcing cross-view geometric consistency. We revisit SRGS, a simple baseline that couples plug-in 2D SR priors with geometry-aware cross-view regularization, and observe that most subsequent advances follow the same paradigm, either strengthening prior injection, refining cross-view constraints, or modulating the objective. However, this shared structure is rarely formalized as a unified objective with explicit modules, limiting principled attribution of improvements and reusable design guidance. In this paper, we formalize SRGS as a unified modular framework that factorizes 3DGS SR into two components, prior injection and cross-view regularization, within a joint objective. This abstraction subsumes a broad family of recent methods as instantiations of the same recipe, enabling analysis beyond single-method innovation. Across five public benchmarks, we consolidate nine representative follow-up methods and trace reported improvements to specific modules and settings. Ablations disentangle the roles of priors and consistency, and stress tests under sparse-view input and challenging capture conditions characterize robustness. Overall, our study consolidates 3DGS SR into a coherent foundation and offers practical guidance for robust, comparable 3DGS SR methods.", "date": "2024-04-16", "categories": []}
{"id": "2603.19234", "title": "Matryoshka Gaussian Splatting", "abstract": "The ability to render scenes at adjustable fidelity from a single model, known as level of detail (LoD), is crucial for practical deployment of 3D Gaussian Splatting (3DGS). Existing discrete LoD methods expose only a limited set of operating points, while concurrent continuous LoD approaches enable smoother scaling but often suffer noticeable quality degradation at full capacity, making LoD a costly design decision. We introduce Matryoshka Gaussian Splatting (MGS), a training framework that enables continuous LoD for standard 3DGS pipelines without sacrificing full-capacity rendering quality. MGS learns a single ordered set of Gaussians such that rendering any prefix, the first k splats, produces a coherent reconstruction whose fidelity improves smoothly with increasing budget. Our key idea is stochastic budget training: each iteration samples a random splat budget and optimises both the corresponding prefix and the full set. This strategy requires only two forward passes and introduces no architectural modifications. Experiments across four benchmarks and six baselines show that MGS matches the full-capacity performance of its backbone while enabling a continuous speed-quality trade-off from a single model. Extensive ablations on ordering strategies, training objectives, and model capacity further validate the designs.", "date": "2026-03-19", "categories": []}
{"id": "2603.19552", "title": "StreetForward: Perceiving Dynamic Street with Feedforward Causal Attention", "abstract": "Feedforward reconstruction is crucial for autonomous driving applications, where rapid scene reconstruction enables efficient utilization of large-scale driving datasets in closed-loop simulation and other downstream tasks, eliminating the need for time-consuming per-scene optimization. We present StreetForward, a pose-free and tracker-free feedforward framework for dynamic street reconstruction. Building upon the alternating attention mechanism from Visual Geometry Grounded Transformer (VGGT), we propose a simple yet effective temporal mask attention module that captures dynamic motion information from image sequences and produces motion-aware latent representations. Static content and dynamic instances are represented uniformly with 3D Gaussian Splatting, and are optimized jointly by cross-frame rendering with spatio-temporal consistency, allowing the model to infer per-pixel velocities and produce high-fidelity novel views at new poses and times. We train and evaluate our model on the Waymo Open Dataset, demonstrating superior performance on novel view synthesis and depth estimation compared to existing methods. Furthermore, zero-shot inference on CARLA and other datasets validates the generalization capability of our approach. More visualizations are available on our project page: https://streetforward.github.io.", "date": "2026-03-20", "categories": []}
{"id": "2603.19753", "title": "ReLi3D: Relightable Multi-view 3D Reconstruction with Disentangled Illumination", "abstract": "Reconstructing 3D assets from images has long required separate pipelines for geometry reconstruction, material estimation, and illumination recovery, each with distinct limitations and computational overhead. We present ReLi3D, the first unified end-to-end pipeline that simultaneously reconstructs complete 3D geometry, spatially-varying physically-based materials, and environment illumination from sparse multi-view images in under one second. Our key insight is that multi-view constraints can dramatically improve material and illumination disentanglement, a problem that remains fundamentally ill-posed for single-image methods. Key to our approach is the fusion of the multi-view input via a transformer cross-conditioning architecture, followed by a novel unified two-path prediction strategy. The first path predicts the object's structure and appearance, while the second path predicts the environment illumination from image background or object reflections. This, combined with a differentiable Monte Carlo multiple importance sampling renderer, creates an optimal illumination disentanglement training pipeline. In addition, with our mixed domain training protocol, which combines synthetic PBR datasets with real-world RGB captures, we establish generalizable results in geometry, material accuracy, and illumination quality. By unifying previously separate reconstruction tasks into a single feed-forward pass, we enable near-instantaneous generation of complete, relightable 3D assets. Project Page: https://reli3d.jdihlmann.com/", "date": "2026-03-20", "categories": []}
{"id": "2405.19569", "title": "Improved Convex Decomposition with Ensembling and Negative Primitives", "abstract": "Describing a scene in terms of primitives -- geometrically simple shapes that offer a parsimonious but accurate abstraction of structure -- is an established and difficult fitting problem. Different scenes require different numbers of primitives, and these primitives interact strongly. Existing methods are evaluated by comparing predicted depth, normals, and segmentation against ground truth. The state of the art method involves a learned regression procedure to predict a start point consisting of a fixed number of primitives, followed by a descent method to refine the geometry and remove redundant primitives. CSG (Constructive Solid Geometry) representations are significantly enhanced by a set-differencing operation. Our representation incorporates negative primitives, which are differenced from the positive primitives. These notably enrich the geometry that the model can encode, while complicating the fitting problem. This paper presents a method that can (a) incorporate these negative primitives and (b) choose the overall number of positive and negative primitives by ensembling. Extensive experiments on the standard NYUv2 dataset confirm that (a) this approach results in substantial improvements in depth representation and segmentation over SOTA and (b) negative primitives improve fitting accuracy. Our method is robustly applicable across datasets: in a first, we evaluate primitive prediction for LAION images.", "date": "2024-05-29", "categories": []}
{"id": "2504.00992", "title": "SuperDec: 3D Scene Decomposition with Superquadric Primitives", "abstract": "We present SuperDec, an approach for creating compact 3D scene representations via decomposition into superquadric primitives. While most recent works leverage geometric primitives to obtain photorealistic 3D scene representations, we propose to leverage them to obtain a compact yet expressive representation. We propose to solve the problem locally on individual objects and leverage the capabilities of instance segmentation methods to scale our solution to full 3D scenes. In doing that, we design a new architecture which efficiently decompose point clouds of arbitrary objects in a compact set of superquadrics. We train our architecture on ShapeNet and we prove its generalization capabilities on object instances extracted from the ScanNet++ dataset as well as on full Replica scenes. Finally, we show how a compact representation based on superquadrics can be useful for a diverse range of downstream applications, including robotic tasks and controllable visual content generation and editing.", "date": "2025-04-01", "categories": []}
{"id": "2507.02861", "title": "LiteReality: Graphics-Ready 3D Scene Reconstruction from RGB-D Scans", "abstract": "We propose LiteReality, a novel pipeline that converts RGB-D scans of indoor environments into compact, realistic, and interactive 3D virtual replicas. LiteReality not only reconstructs scenes that visually resemble reality but also supports key features essential for graphics pipelines -- such as object individuality, articulation, high-quality physically based rendering materials, and physically based interaction. At its core, LiteReality first performs scene understanding and parses the results into a coherent 3D layout and objects with the help of a structured scene graph. It then reconstructs the scene by retrieving the most visually similar 3D artist-crafted models from a curated asset database. Next, the Material Painting module enhances realism by recovering high-quality, spatially varying materials. Finally, the reconstructed scene is integrated into a simulation engine with basic physical properties to enable interactive behavior. The resulting scenes are compact, editable, and fully compatible with standard graphics pipelines, making them suitable for applications in AR/VR, gaming, robotics, and digital twins. In addition, LiteReality introduces a training-free object retrieval module that achieves state-of-the-art similarity performance on the Scan2CAD benchmark, along with a robust material painting module capable of transferring appearances from images of any style to 3D assets -- even under severe misalignment, occlusion, and poor lighting. We demonstrate the effectiveness of LiteReality on both real-life scans and public datasets. Project page: https://litereality.github.io; Video: https://www.youtube.com/watch?v=ecK9m3LXg2c", "date": "2025-07-03", "categories": []}
{"id": "2508.15376", "title": "DriveSplat: Unified Neural Gaussian Reconstruction for Dynamic Driving Scenes", "abstract": "Reconstructing large-scale dynamic driving scenes remains challenging due to the coexistence of static environments with extreme depth variation and diverse dynamic actors exhibiting complex motions. Existing Gaussian Splatting based methods have primarily focused on limited-scale or object-centric settings, and their applicability to large-scale dynamic driving scenes remains underexplored, particularly in the presence of extreme scale variation and non-rigid motions. In this work, we propose DriveSplat, a unified neural Gaussian framework for reconstructing dynamic driving scenes within a unified Gaussian-based representation. For static backgrounds, we introduce a scene-aware learnable level-of-detail (LOD) modeling strategy that explicitly accounts for near, intermediate, and far depth ranges in driving environments, enabling adaptive multi-scale Gaussian allocation. For dynamic actors, we use an object-centric formulation with neural Gaussian primitives, modeling motion through a global rigid transformation and handling non-rigid dynamics via a two-stage deformation that first adjusts anchors and subsequently updates the Gaussians. To further regularize the optimization, we incorporate dense depth and surface normal priors from pre-trained models as auxiliary supervision. Extensive experiments on the Waymo and KITTI benchmarks demonstrate that DriveSplat achieves state-of-the-art performance in novel-view synthesis while producing temporally stable and geometrically consistent reconstructions of dynamic driving scenes. Project page: https://physwm.github.io/drivesplat.", "date": "2025-08-21", "categories": []}
{"id": "2510.02691", "title": "FSFSplatter: Build Surface and Novel Views with Sparse-Views within 2min", "abstract": "Gaussian Splatting has become a leading reconstruction technique, known for its high-quality novel view synthesis and detailed reconstruction. However, most existing methods require dense, calibrated views. Reconstructing from free sparse images often leads to poor surface due to limited overlap and overfitting. We introduce FSFSplatter, a new approach for fast surface reconstruction from free sparse images. Our method integrates end-to-end dense Gaussian initialization, camera parameter estimation, and geometry-enhanced scene optimization. Specifically, FSFSplatter employs a large Transformer to encode multi-view images and generates a dense and geometrically consistent Gaussian scene initialization via a self-splitting Gaussian head. It eliminates local floaters through contribution-based pruning and mitigates overfitting to limited views by leveraging depth and multi-view feature supervision with differentiable camera parameters during rapid optimization. FSFSplatter outperforms current state-of-the-art methods on widely used DTU, Replica, and BlendedMVS datasets.", "date": "2025-10-03", "categories": []}
{"id": "2603.18218", "title": "Semantic Segmentation and Depth Estimation for Real-Time Lunar Surface Mapping Using 3D Gaussian Splatting", "abstract": "Navigation and mapping on the lunar surface require robust perception under challenging conditions, including poorly textured environments, high-contrast lighting, and limited computational resources. This paper presents a real-time mapping framework that integrates dense perception models with a 3D Gaussian Splatting (3DGS) representation. We first benchmark several models on synthetic datasets generated with the LuPNT simulator, selecting a stereo dense depth estimation model based on Gated Recurrent Units for its balance of speed and accuracy in depth estimation, and a convolutional neural network for its superior performance in detecting semantic segments. Using ground truth poses to decouple the local scene understanding from the global state estimation, our pipeline reconstructs a 120-meter traverse with a geometric height accuracy of approximately 3 cm, outperforming a traditional point cloud baseline without LiDAR. The resulting 3DGS map enables novel view synthesis and serves as a foundation for a full SLAM system, where its capacity for joint map and pose optimization would offer significant advantages. Our results demonstrate that combining semantic segmentation and dense depth estimation with learned map representations is an effective approach for creating detailed, large-scale maps to support future lunar surface missions.", "date": "2026-03-18", "categories": []}
{"id": "2603.18493", "title": "FILT3R: Latent State Adaptive Kalman Filter for Streaming 3D Reconstruction", "abstract": "Streaming 3D reconstruction maintains a persistent latent state that is updated online from incoming frames, enabling constant-memory inference. A key failure mode is the state update rule: aggressive overwrites forget useful history, while conservative updates fail to track new evidence, and both behaviors become unstable beyond the training horizon. To address this challenge, we propose FILT3R, a training-free latent filtering layer that casts recurrent state updates as stochastic state estimation in token space. FILT3R maintains a per-token variance and computes a Kalman-style gain that adaptively balances memory retention against new observations. Process noise -- governing how much the latent state is expected to change between frames -- is estimated online from EMA-normalized temporal drift of candidate tokens. Using extensive experiments, we demonstrate that FILT3R yields an interpretable, plug-in update rule that generalizes common overwrite and gating policies as special cases. Specifically, we show that gains shrink in stable regimes as uncertainty contracts with accumulated evidence, and rise when genuine scene change increases process uncertainty, improving long-horizon stability for depth, pose, and 3D reconstruction, compared to the existing methods. Code will be released at https://github.com/jinotter3/FILT3R.", "date": "2026-03-19", "categories": []}
{"id": "2603.18707", "title": "From ex(p) to poly: Gaussian Splatting with Polynomial Kernels", "abstract": "Recent advancements in Gaussian Splatting (3DGS) have introduced various modifications to the original kernel, resulting in significant performance improvements. However, many of these kernel changes are incompatible with existing datasets optimized for the original Gaussian kernel, presenting a challenge for widespread adoption. In this work, we address this challenge by proposing an alternative kernel that maintains compatibility with existing datasets while improving computational efficiency. Specifically, we replace the original exponential kernel with a polynomial approximation combined with a ReLU function. This modification allows for more aggressive culling of Gaussians, leading to enhanced performance across different 3DGS implementations. Our results show a notable performance improvement of 4 to 15% with negligible impact on image quality. We also provide a detailed mathematical analysis of the new kernel and discuss its potential benefits for 3DGS implementations on NPU hardware.", "date": "2026-03-19", "categories": []}
{"id": "2603.18774", "title": "SEAR: Simple and Efficient Adaptation of Visual Geometric Transformers for RGB+Thermal 3D Reconstruction", "abstract": "Foundational feed-forward visual geometry models enable accurate and efficient camera pose estimation and scene reconstruction by learning strong scene priors from massive RGB datasets. However, their effectiveness drops when applied to mixed sensing modalities, such as RGB-thermal (RGB-T) images. We observe that while a visual geometry grounded transformer pretrained on RGB data generalizes well to thermal-only reconstruction, it struggles to align RGB and thermal modalities when processed jointly. To address this, we propose SEAR, a simple yet efficient fine-tuning strategy that adapts a pretrained geometry transformer to multimodal RGB-T inputs. Despite being trained on a relatively small RGB-T dataset, our approach significantly outperforms state-of-the-art methods for 3D reconstruction and camera pose estimation, achieving significant improvements over all metrics (e.g., over 29\\% in AUC@30) and delivering higher detail and consistency between modalities with negligible overhead in inference time compared to the original pretrained model. Notably, SEAR enables reliable multimodal pose estimation and reconstruction even under challenging conditions, such as low lighting and dense smoke. We validate our architecture through extensive ablation studies, demonstrating how the model aligns both modalities. Additionally, we introduce a new dataset featuring RGB and thermal sequences captured at different times, viewpoints, and illumination conditions, providing a robust benchmark for future work in multimodal 3D scene reconstruction. Code and models are publicly available at https://www.github.com/Schindler-EPFL-Lab/SEAR.", "date": "2026-03-19", "categories": []}
{"id": "2512.02341", "title": "TALO: Pushing 3D Vision Foundation Models Towards Globally Consistent Online Reconstruction", "abstract": "3D vision foundation models have shown strong generalization in reconstructing key 3D attributes from uncalibrated images through a single feed-forward pass. However, when deployed in online settings such as driving scenarios, predictions are made over temporal windows, making it non-trivial to maintain consistency across time. Recent strategies align consecutive predictions by solving global transformation, yet our analysis reveals their fundamental limitations in assumption validity, local alignment scope, and robustness under noisy geometry. In this work, we propose a higher-DOF and long-term alignment framework based on Thin Plate Spline, leveraging globally propagated control points to correct spatially varying inconsistencies. In addition, we adopt a point-agnostic submap registration design that is inherently robust to noisy geometry predictions. The proposed framework is fully plug-and-play, compatible with diverse 3D foundation models and camera configurations (e.g., monocular or surround-view). Extensive experiments demonstrate that our method consistently yields more coherent geometry and lower trajectory errors across multiple datasets, backbone models, and camera setups, highlighting its robustness and generality. Code is available at https://github.com/Xian-Bei/TALO.", "date": "2025-12-02", "categories": []}
{"id": "2603.11298", "title": "InstantHDR: Single-forward Gaussian Splatting for High Dynamic Range 3D Reconstruction", "abstract": "High dynamic range (HDR) novel view synthesis (NVS) aims to reconstruct HDR scenes from multi-exposure low dynamic range (LDR) images. Existing HDR pipelines heavily rely on known camera poses, well-initialized dense point clouds, and time-consuming per-scene optimization. Current feed-forward alternatives overlook the HDR problem by assuming exposure-invariant appearance. To bridge this gap, we propose InstantHDR, a feed-forward network that reconstructs 3D HDR scenes from uncalibrated multi-exposure LDR collections in a single forward pass. Specifically, we design a geometry-guided appearance modeling for multi-exposure fusion, and a meta-network for generalizable scene-specific tone mapping. Due to the lack of HDR scene data, we build a pre-training dataset, called HDR-Pretrain, for generalizable feed-forward HDR models, featuring 168 Blender-rendered scenes, diverse lighting types, and multiple camera response functions. Comprehensive experiments show that our InstantHDR delivers comparable synthesis performance to the state-of-the-art optimization-based HDR methods while enjoying $\\sim700\\times$ and $\\sim20\\times$ reconstruction speed improvement with our single-forward and post-optimization settings. All code, models, and datasets will be released after the review process.", "date": "2026-03-11", "categories": []}
{"id": "2603.12789", "title": "Coherent Human-Scene Reconstruction from Multi-Person Multi-View Video in a Single Pass", "abstract": "Recent advances in 3D foundation models have led to growing interest in reconstructing humans and their surrounding environments. However, most existing approaches focus on monocular inputs, and extending them to multi-view settings requires additional overhead modules or preprocessed data. To this end, we present CHROMM, a unified framework that jointly estimates cameras, scene point clouds, and human meshes from multi-person multi-view videos without relying on external modules or preprocessing. We integrate strong geometric and human priors from Pi3X and Multi-HMR into a single trainable neural network architecture, and introduce a scale adjustment module to solve the scale discrepancy between humans and the scene. We also introduce a multi-view fusion strategy to aggregate per-view estimates into a single representation at test-time. Finally, we propose a geometry-based multi-person association method, which is more robust than appearance-based approaches. Experiments on EMDB, RICH, EgoHumans, and EgoExo4D show that CHROMM achieves competitive performance in global human motion and multi-view pose estimation while running over 8x faster than prior optimization-based multi-view approaches. Project page: https://nstar1125.github.io/chromm.", "date": "2026-03-13", "categories": []}
{"id": "2603.16736", "title": "World Reconstruction From Inconsistent Views", "abstract": "Video diffusion models generate high-quality and diverse worlds; however, individual frames often lack 3D consistency across the output sequence, which makes the reconstruction of 3D worlds difficult. To this end, we propose a new method that handles these inconsistencies by non-rigidly aligning the video frames into a globally-consistent coordinate frame that produces sharp and detailed pointcloud reconstructions. First, a geometric foundation model lifts each frame into a pixel-wise 3D pointcloud, which contains unaligned surfaces due to these inconsistencies. We then propose a tailored non-rigid iterative frame-to-model ICP to obtain an initial alignment across all frames, followed by a global optimization that further sharpens the pointcloud. Finally, we leverage this pointcloud as initialization for 3D reconstruction and propose a novel inverse deformation rendering loss to create high quality and explorable 3D environments from inconsistent views. We demonstrate that our 3D scenes achieve higher quality than baselines, effectively turning video models into 3D-consistent world generators.", "date": "2026-03-17", "categories": []}
{"id": "2603.17358", "title": "A 3D Reconstruction Benchmark for Asset Inspection", "abstract": "Asset management requires accurate 3D models to inform the maintenance, repair, and assessment of buildings, maritime vessels, and other key structures as they age. These downstream applications rely on high-fidelity models produced from aerial surveys in close proximity to the asset, enabling operators to locate and characterise deterioration or damage and plan repairs. Captured images typically have high overlap between adjacent camera poses, sufficient detail at millimetre scale, and challenging visual appearances such as reflections and transparency. However, existing 3D reconstruction datasets lack examples of these conditions, making it difficult to benchmark methods for this task. We present a new dataset with ground truth depth maps, camera poses, and mesh models of three synthetic scenes with simulated inspection trajectories and varying levels of surface condition on non-Lambertian scene content. We evaluate state-of-the-art reconstruction methods on this dataset. Our results demonstrate that current approaches struggle significantly with the dense capture trajectories and complex surface conditions inherent to this domain, exposing a critical scalability gap and pointing toward new research directions for deployable 3D reconstruction in asset inspection. Project page: https://roboticimaging.org/Projects/asset-inspection-dataset/", "date": "2026-03-18", "categories": []}
{"id": "2603.17519", "title": "UniSem: Generalizable Semantic 3D Reconstruction from Sparse Unposed Images", "abstract": "Semantic-aware 3D reconstruction from sparse, unposed images remains challenging for feed-forward 3D Gaussian Splatting (3DGS). Existing methods often predict an over-complete set of Gaussian primitives under sparse-view supervision, leading to unstable geometry and inferior depth quality. Meanwhile, they rely solely on 2D segmenter features for semantic lifting, which provides weak 3D-level and limited generalizable supervision, resulting in incomplete 3D semantics in novel scenes. To address these issues, we propose UniSem, a unified framework that jointly improves depth accuracy and semantic generalization via two key components. First, Error-aware Gaussian Dropout (EGD) performs error-guided capacity control by suppressing redundancy-prone Gaussians using rendering error cues, producing meaningful, geometrically stable Gaussian representations for improved depth estimation. Second, we introduce a Mix-training Curriculum (MTC) that progressively blends 2D segmenter-lifted semantics with the model's own emergent 3D semantic priors, implemented with object-level prototype alignment to enhance semantic coherence and completeness. Extensive experiments on ScanNet and Replica show that UniSem achieves superior performance in depth prediction and open-vocabulary 3D segmentation across varying numbers of input views. Notably, with 16-view inputs, UniSem reduces depth Rel by 15.2% and improves open-vocabulary segmentation mAcc by 3.7% over strong baselines.", "date": "2026-03-18", "categories": []}
{"id": "2603.17571", "title": "PanoVGGT: Feed-Forward 3D Reconstruction from Panoramic Imagery", "abstract": "Panoramic imagery offers a full 360{\\deg} field of view and is increasingly common in consumer devices. However, it introduces non-pinhole distortions that challenge joint pose estimation and 3D reconstruction. Existing feed-forward models, built for perspective cameras, generalize poorly to this setting. We propose PanoVGGT, a permutation-equivariant Transformer framework that jointly predicts camera poses, depth maps, and 3D point clouds from one or multiple panoramas in a single forward pass. The model incorporates spherical-aware positional embeddings and a panorama-specific three-axis SO(3) rotation augmentation, enabling effective geometric reasoning in the spherical domain. To resolve inherent global-frame ambiguity, we further introduce a stochastic anchoring strategy during training. In addition, we contribute PanoCity, a large-scale outdoor panoramic dataset with dense depth and 6-DoF pose annotations. Extensive experiments on PanoCity and standard benchmarks demonstrate that PanoVGGT achieves competitive accuracy, strong robustness, and improved cross-domain generalization. Code and dataset will be released.", "date": "2026-03-18", "categories": []}
{"id": "2603.17625", "title": "S-VGGT: Structure-Aware Subscene Decomposition for Scalable 3D Foundation Models", "abstract": "Feed-forward 3D foundation models face a key challenge: the quadratic computational cost introduced by global attention, which severely limits scalability as input length increases. Concurrent acceleration methods, such as token merging, operate at the token level. While they offer local savings, the required nearest-neighbor searches introduce undesirable overhead. Consequently, these techniques fail to tackle the fundamental issue of structural redundancy dominant in dense capture data. In this work, we introduce \\textbf{S-VGGT}, a novel approach that addresses redundancy at the structural frame level, drastically shifting the optimization focus. We first leverage the initial features to build a dense scene graph, which characterizes structural scene redundancy and guides the subsequent scene partitioning. Using this graph, we softly assign frames to a small number of subscenes, guaranteeing balanced groups and smooth geometric transitions. The core innovation lies in designing the subscenes to share a common reference frame, establishing a parallel geometric bridge that enables independent and highly efficient processing without explicit geometric alignment. This structural reorganization provides strong intrinsic acceleration by cutting the global attention cost at its source. Crucially, S-VGGT is entirely orthogonal to token-level acceleration methods, allowing the two to be seamlessly combined for compounded speedups without compromising reconstruction fidelity. Code is available at https://github.com/Powertony102/S-VGGT.", "date": "2026-03-18", "categories": []}
{"id": "2502.05175", "title": "Fillerbuster: Unified Generative Scene Completion Model for Casual Captures", "abstract": "We present Fillerbuster, a unified model that completes unknown regions of a 3D scene with a multi-view latent diffusion transformer. Casual captures are often sparse and miss surrounding content behind objects or above the scene. Existing methods are not suitable for this challenge as they focus on making known pixels look good with sparse-view priors, or on creating missing sides of objects from just one or two photos. In reality, we often have hundreds of input frames and want to complete areas that are missing and unobserved from the input frames. Our solution is to train a generative model that can consume a large context of input frames while generating unknown target views and recovering image poses when camera parameters are unknown. We show results where we complete partial captures on two existing datasets. We also present an uncalibrated scene completion task where our unified model predicts both poses and creates new content. We open-source our framework for integration into popular reconstruction platforms like Nerfstudio or Gsplat. We present a flexible, unified inpainting framework to predict many images and poses together, where all inputs are jointly inpainted, and it could be extended to predict more modalities such as depth.", "date": "2025-02-07", "categories": []}
{"id": "2512.07107", "title": "COREA: Coupled Relightable 3D Gaussians and SDFs for Efficient Normal Alignment", "abstract": "We present COREA, the first unified three-tasks framework that couples an SDF and relightable 3D Gaussians (3DGS) to jointly support SH-based novel-view synthesis (NVS), surface reconstruction, and inverse physically-based rendering (inverse PBR). While recent relightable 3DGS methods have progressed, inverse PBR remains bottlenecked by normal estimation, as the discrete nature of 3DGS often yields oversmoothed and unstable normals. To address this limitation, COREA couples the complementary geometric properties of an SDF and relightable 3DGS on a shared underlying surface, where geometry-constrained relightable 3DGS provides reliable depth signals to anchor SDF geometry and the continuous SDF normal field provides spatially consistent supervision for Gaussian normal learning. We couple these signals through depth-guided alignment and normal supervision with normal-aware densification, and introduce Dual-Density Control to regulate densification by balancing photometric and geometric gradients for stable, memory-efficient training. Experiments on standard benchmarks show that COREA is the only framework that supports all three tasks, achieving competitive performance overall, with particularly superior results in inverse PBR.", "date": "2025-12-08", "categories": []}
{"id": "2512.11508", "title": "On Geometric Understanding and Learned Priors in Feed-forward 3D Reconstruction Models", "abstract": "Feed-forward 3D reconstruction models such as DUSt3R, VGGT, and Depth Anything 3 (DA3) are transformer-based foundation models that infer camera geometry and dense scene structure in a single forward pass. Trained at scale in a supervised fashion, they raise a central question: do these models build upon geometric principles akin to traditional multi-view pipelines, or do they primarily rely on learned priors arising from the large-scale training setup? We find that epipolar geometry emerges within the intermediate layers of all three models and is causally linked to correspondence patterns in attention heads. To study this, we perform a systematic analysis of their internal representations across three real-world datasets and a controlled synthetic dataset. We quantify geometric understanding by probing intermediate features, analyzing attention patterns to identify correspondence matching patterns, and performing targeted interventions at the attention level. Further, we assess the role of learned priors by applying challenging input-level perturbations, such as occlusions, scene ambiguities, and varying camera configurations, and compare them against classical multi-stage reconstruction pipelines.", "date": "2025-12-12", "categories": []}
{"id": "2603.16133", "title": "DualPrim: Compact 3D Reconstruction with Positive and Negative Primitives", "abstract": "Neural reconstructions often trade structure for fidelity, yielding dense and unstructured meshes with irregular topology and weak part boundaries that hinder editing, animation, and downstream asset reuse. We present DualPrim, a compact and structured 3D reconstruction framework. Unlike additive-only implicit or primitive methods, DualPrim represents shapes with positive and negative superquadrics: the former builds the bases while the latter carves local volumes through a differentiable operator, enabling topology-aware modeling of holes and concavities. This additive-subtractive design increases the representational power without sacrificing compactness or differentiability. We embed DualPrim in a volumetric differentiable renderer, enabling end-to-end learning from multi-view images and seamless mesh export via closed-form boolean difference. Empirically, DualPrim delivers state-of-the-art accuracy and produces compact, structured, and interpretable outputs that better satisfy downstream needs than additive-only alternatives.", "date": "2026-03-17", "categories": []}
{"id": "2603.16211", "title": "Leveling3D: Leveling Up 3D Reconstruction with Feed-Forward 3D Gaussian Splatting and Geometry-Aware Generation", "abstract": "Feed-forward 3D reconstruction has revolutionized 3D vision, providing a powerful baseline for downstream tasks such as novel-view synthesis with 3D Gaussian Splatting. Previous works explore fixing the corrupted rendering results with a diffusion model. However, they lack geometric concern and fail at filling the missing area on the extrapolated view. In this work, we introduce Leveling3D, a novel pipeline that integrates feed-forward 3D reconstruction with geometrical-consistent generation to enable holistic simultaneous reconstruction and generation. We propose a geometry-aware leveling adapter, a lightweight technique that aligns internal knowledge in the diffusion model with the geometry prior from the feed-forward model. The leveling adapter enables generation on the artifact area of the extrapolated novel views caused by underconstrained regions of the 3D representation. Specifically, to learn a more diverse distributed generation, we introduce the palette filtering strategy for training, and a test-time masking refinement to prevent messy boundaries along the fixing regions. More importantly, the enhanced extrapolated novel views from Leveling3D could be used as the inputs for feed-forward 3DGS, leveling up the 3D reconstruction. We achieve SOTA performance on public datasets, including tasks such as novel-view synthesis and depth estimation.", "date": "2026-03-17", "categories": []}
{"id": "2603.16306", "title": "DriveFix: Spatio-Temporally Coherent Driving Scene Restoration", "abstract": "Recent advancements in 4D scene reconstruction, particularly those leveraging diffusion priors, have shown promise for novel view synthesis in autonomous driving. However, these methods often process frames independently or in a view-by-view manner, leading to a critical lack of spatio-temporal synergy. This results in spatial misalignment across cameras and temporal drift in sequences. We propose DriveFix, a novel multi-view restoration framework that ensures spatio-temporal coherence for driving scenes. Our approach employs an interleaved diffusion transformer architecture with specialized blocks to explicitly model both temporal dependencies and cross-camera spatial consistency. By conditioning the generation on historical context and integrating geometry-aware training losses, DriveFix enforces that the restored views adhere to a unified 3D geometry. This enables the consistent propagation of high-fidelity textures and significantly reduces artifacts. Extensive evaluations on the Waymo, nuScenes, and PandaSet datasets demonstrate that DriveFix achieves state-of-the-art performance in both reconstruction and novel view synthesis, marking a substantial step toward robust 4D world modeling for real-world deployment.", "date": "2026-03-17", "categories": []}
{"id": "2603.16844", "title": "M^3: Dense Matching Meets Multi-View Foundation Models for Monocular Gaussian Splatting SLAM", "abstract": "Streaming reconstruction from uncalibrated monocular video remains challenging, as it requires both high-precision pose estimation and computationally efficient online refinement in dynamic environments. While coupling 3D foundation models with SLAM frameworks is a promising paradigm, a critical bottleneck persists: most multi-view foundation models estimate poses in a feed-forward manner, yielding pixel-level correspondences that lack the requisite precision for rigorous geometric optimization. To address this, we present M^3, which augments the Multi-view foundation model with a dedicated Matching head to facilitate fine-grained dense correspondences and integrates it into a robust Monocular Gaussian Splatting SLAM. M^3 further enhances tracking stability by incorporating dynamic area suppression and cross-inference intrinsic alignment. Extensive experiments on diverse indoor and outdoor benchmarks demonstrate state-of-the-art accuracy in both pose estimation and scene reconstruction. Notably, M^3 reduces ATE RMSE by 64.3% compared to VGGT-SLAM 2.0 and outperforms ARTDECO by 2.11 dB in PSNR on the ScanNet++ dataset.", "date": "2026-03-17", "categories": []}
{"id": "2603.16868", "title": "MessyKitchens: Contact-rich object-level 3D scene reconstruction", "abstract": "Monocular 3D scene reconstruction has recently seen significant progress. Powered by the modern neural architectures and large-scale data, recent methods achieve high performance in depth estimation from a single image. Meanwhile, reconstructing and decomposing common scenes into individual 3D objects remains a hard challenge due to the large variety of objects, frequent occlusions and complex object relations. Notably, beyond shape and pose estimation of individual objects, applications in robotics and animation require physically-plausible scene reconstruction where objects obey physical principles of non-penetration and realistic contacts. In this work we advance object-level scene reconstruction along two directions. First, we introduceMessyKitchens, a new dataset with real-world scenes featuring cluttered environments and providing high-fidelity object-level ground truth in terms of 3D object shapes, poses and accurate object contacts. Second, we build on the recent SAM 3D approach for single-object reconstruction and extend it with Multi-Object Decoder (MOD) for joint object-level scene reconstruction. To validate our contributions, we demonstrate MessyKitchens to significantly improve previous datasets in registration accuracy and inter-object penetration. We also compare our multi-object reconstruction approach on three datasets and demonstrate consistent and significant improvements of MOD over the state of the art. Our new benchmark, code and pre-trained models will become publicly available on our project website: https://messykitchens.github.io/.", "date": "2026-03-17", "categories": []}
{"id": "2507.16443", "title": "VGGT-Long: Chunk it, Loop it, Align it -- Pushing VGGT's Limits on Kilometer-scale Long RGB Sequences", "abstract": "Foundation models for 3D vision have recently demonstrated remarkable capabilities in 3D perception. However, extending these models to large-scale RGB stream 3D reconstruction remains challenging due to memory limitations. In this work, we propose VGGT-Long, a simple yet effective system that pushes the limits of monocular 3D reconstruction to kilometer-scale, unbounded outdoor environments. Our approach addresses the scalability bottlenecks of existing models through a chunk-based processing strategy combined with overlapping alignment and lightweight loop closure optimization. Without requiring camera calibration, depth supervision or model retraining, VGGT-Long achieves trajectory and reconstruction performance comparable to traditional methods. We evaluate our method on KITTI, Waymo, and Virtual KITTI datasets. VGGT-Long not only runs successfully on long RGB sequences where foundation models typically fail, but also produces accurate and consistent geometry across various conditions. Our results highlight the potential of leveraging foundation models for scalable monocular 3D scene in real-world settings, especially for autonomous driving scenarios. Code is available at https://github.com/DengKaiCQ/VGGT-Long.", "date": "2025-07-22", "categories": []}
{"id": "2509.00831", "title": "UPGS: Unified Pose-aware Gaussian Splatting for Dynamic Scene Deblurring", "abstract": "Reconstructing dynamic 3D scenes from monocular video has broad applications in AR/VR, robotics, and autonomous navigation, but often fails due to severe motion blur caused by camera and object motion. Existing methods commonly follow a two-step pipeline, where camera poses are first estimated and then 3D Gaussians are optimized. Since blurring artifacts usually undermine pose estimation, pose errors could be accumulated to produce inferior reconstruction results. To address this issue, we introduce a unified optimization framework by incorporating camera poses as learnable parameters complementary to 3DGS attributes for end-to-end optimization. Specifically, we recast camera and object motion as per-primitive SE(3) affine transformations on 3D Gaussians and formulate a unified optimization objective. For stable optimization, we introduce a three-stage training schedule that optimizes camera poses and Gaussians alternatively. Particularly, 3D Gaussians are first trained with poses being fixed, and then poses are optimized with 3D Gaussians being untouched. Finally, all learnable parameters are optimized together. Extensive experiments on the Stereo Blur dataset and challenging real-world sequences demonstrate that our method achieves significant gains in reconstruction quality and pose estimation accuracy over prior dynamic deblurring methods.", "date": "2025-08-31", "categories": []}
{"id": "2603.12064", "title": "Dense Dynamic Scene Reconstruction and Camera Pose Estimation from Multi-View Videos", "abstract": "We address the challenging problem of dense dynamic scene reconstruction and camera pose estimation from multiple freely moving cameras -- a setting that arises naturally when multiple observers capture a shared event. Prior approaches either handle only single-camera input or require rigidly mounted, pre-calibrated camera rigs, limiting their practical applicability. We propose a two-stage optimization framework that decouples the task into robust camera tracking and dense depth refinement. In the first stage, we extend single-camera visual SLAM to the multi-camera setting by constructing a spatiotemporal connection graph that exploits both intra-camera temporal continuity and inter-camera spatial overlap, enabling consistent scale and robust tracking. To ensure robustness under limited overlap, we introduce a wide-baseline initialization strategy using feed-forward reconstruction models. In the second stage, we refine depth and camera poses by optimizing dense inter- and intra-camera consistency using wide-baseline optical flow. Additionally, we introduce MultiCamRobolab, a new real-world dataset with ground-truth poses from a motion capture system. Finally, we demonstrate that our method significantly outperforms state-of-the-art feed-forward models on both synthetic and real-world benchmarks, while requiring less memory.", "date": "2026-03-12", "categories": []}
{"id": "2603.13556", "title": "Semantic Aware Feature Extraction for Enhanced 3D Reconstruction", "abstract": "Feature matching is a fundamental problem in computer vision with wide-ranging applications, including simultaneous localization and mapping (SLAM), image stitching, and 3D reconstruction. While recent advances in deep learning have improved keypoint detection and description, most approaches focus primarily on geometric attributes and often neglect higher-level semantic information. This work proposes a semantic-aware feature extraction framework that employs multi-task learning to jointly train keypoint detection, keypoint description, and semantic segmentation. The method is benchmarked against standard feature matching techniques and evaluated in the context of 3D reconstruction. To enhance feature correspondence, a deep matching module is integrated. The system is tested using input from a single monocular fisheye camera mounted on a vehicle and evaluated within a multi-floor parking structure. The proposed approach supports semantic 3D reconstruction with altitude estimation, capturing elevation changes and enabling multi-level mapping. Experimental results demonstrate that the method produces semantically annotated 3D point clouds with improved structural detail and elevation information, underscoring the effectiveness of joint training with semantic cues for more consistent feature matching and enhanced 3D reconstruction.", "date": "2026-03-13", "categories": []}
{"id": "2603.14150", "title": "CIPHER: Culvert Inspection through Pairwise Frame Selection and High-Efficiency Reconstruction", "abstract": "Automated culvert inspection systems can help increase the safety and efficiency of flood management operations. As a key step to this system, we present an efficient RGB-based 3D reconstruction pipeline for culvert-like structures in visually repetitive environments. Our approach first selects informative frame pairs to maximize viewpoint diversity while ensuring valid correspondence matching using a plug-and-play module, followed by a reconstruction model that simultaneously estimates RGB appearance, geometry, and semantics in real-time. Experiments demonstrate that our method effectively generates accurate 3D reconstructions and depth maps, enhancing culvert inspection efficiency with minimal human intervention.", "date": "2026-03-14", "categories": []}
{"id": "2603.14232", "title": "S2GS: Streaming Semantic Gaussian Splatting for Online Scene Understanding and Reconstruction", "abstract": "Existing offline feed-forward methods for joint scene understanding and reconstruction on long image streams often repeatedly perform global computation over an ever-growing set of past observations, causing runtime and GPU memory to increase rapidly with sequence length and limiting scalability. We propose Streaming Semantic Gaussian Splatting (S2GS), a strictly causal, incremental 3D Gaussian semantic field framework: it does not leverage future frames and continuously updates scene geometry, appearance, and instance-level semantics without reprocessing historical frames, enabling scalable online joint reconstruction and understanding. S2GS adopts a geometry-semantic decoupled dual-backbone design: the geometry branch performs causal modeling to drive incremental Gaussian updates, while the semantic branch leverages a 2D foundation vision model and a query-driven decoder to predict segmentation masks and identity embeddings, further stabilized by query-level contrastive alignment and lightweight online association with an instance memory. Experiments show that S2GS matches or outperforms strong offline baselines on joint reconstruction-and-understanding benchmarks, while significantly improving long-horizon scalability: it processes 1,000+ frames with much slower growth in runtime and GPU memory, whereas offline global-processing baselines typically run out of memory at around 80 frames under the same setting.", "date": "2026-03-15", "categories": []}
{"id": "2603.14301", "title": "4D Synchronized Fields: Motion-Language Gaussian Splatting for Temporal Scene Understanding", "abstract": "Current 4D representations decouple geometry, motion, and semantics: reconstruction methods discard interpretable motion structure; language-grounded methods attach semantics after motion is learned, blind to how objects move; and motion-aware methods encode dynamics as opaque per-point residuals without object-level organization. We propose 4D Synchronized Fields, a 4D Gaussian representation that learns object-factored motion in-loop during reconstruction and synchronizes language to the resulting kinematics through a per-object conditioned field. Each Gaussian trajectory is decomposed into shared object motion plus an implicit residual, and a kinematic-conditioned ridge map predicts temporal semantic variation, yielding a single representation in which reconstruction, motion, and semantics are structurally coupled and enabling open-vocabulary temporal queries that retrieve both objects and moments. On HyperNeRF, 4D Synchronized Fields achieves 28.52 dB mean PSNR, the highest among all language-grounded and motion-aware baselines, within 1.5 dB of reconstruction-only methods. On targeted temporal-state retrieval, the kinematic-conditioned field attains 0.884 mean accuracy, 0.815 mean vIoU, and 0.733 mean tIoU, surpassing 4D LangSplat (0.620, 0.433, and 0.439 respectively) and LangSplat (0.415, 0.304, and 0.262). Ablation confirms that kinematic conditioning is the primary driver, accounting for +0.45 tIoU over a static-embedding-only baseline. 4D Synchronized Fields is the only method that jointly exposes interpretable motion primitives and temporally grounded language fields from a single trained representation. Code will be released.", "date": "2026-03-15", "categories": []}
{"id": "2603.14316", "title": "Direct Object-Level Reconstruction via Probabilistic Gaussian Splatting", "abstract": "Object-level 3D reconstruction play important roles across domains such as cultural heritage digitization, industrial manufacturing, and virtual reality. However, existing Gaussian Splatting-based approaches generally rely on full-scene reconstruction, in which substantial redundant background information is introduced, leading to increased computational and storage overhead. To address this limitation, we propose an efficient single-object 3D reconstruction method based on 2D Gaussian Splatting. By directly integrating foreground-background probability cues into Gaussian primitives and dynamically pruning low-probability Gaussians during training, the proposed method fundamentally focuses on an object of interest and improves the memory and computational efficiency. Our pipeline leverages probability masks generated by YOLO and SAM to supervise probabilistic Gaussian attributes, replacing binary masks with continuous probability values to mitigate boundary ambiguity. Additionally, we propose a dual-stage filtering strategy for training's startup to suppress background Gaussians. And, during training, rendered probability masks are conversely employed to refine supervision and enhance boundary consistency across views. Experiments conducted on the MIP-360, T&T, and NVOS datasets demonstrate that our method exhibits strong self-correction capability in the presence of mask errors and achieves reconstruction quality comparable to standard 3DGS approaches, while requiring only approximately 1/10 of their Gaussian amount. These results validate the efficiency and robustness of our method for single-object reconstruction and highlight its potential for applications requiring both high fidelity and computational efficiency.", "date": "2026-03-15", "categories": []}
{"id": "2603.14684", "title": "E2EGS: Event-to-Edge Gaussian Splatting for Pose-Free 3D Reconstruction", "abstract": "The emergence of neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS) has advanced novel view synthesis (NVS). These methods, however, require high-quality RGB inputs and accurate corresponding poses, limiting robustness under real-world conditions such as fast camera motion or adverse lighting. Event cameras, which capture brightness changes at each pixel with high temporal resolution and wide dynamic range, enable precise sensing of dynamic scenes and offer a promising solution. However, existing event-based NVS methods either assume known poses or rely on depth estimation models that are bounded by their initial observations, failing to generalize as the camera traverses previously unseen regions. We present E2EGS, a pose-free framework operating solely on event streams. Our key insight is that edge information provides rich structural cues essential for accurate trajectory estimation and high-quality NVS. To extract edges from noisy event streams, we exploit the distinct spatio-temporal characteristics of edges and non-edge regions. The event camera's movement induces consistent events along edges, while non-edge regions produce sparse noise. We leverage this through a patch-based temporal coherence analysis that measures local variance to extract edges while robustly suppressing noise. The extracted edges guide structure-aware Gaussian initialization and enable edge-weighted losses throughout initialization, tracking, and bundle adjustment. Extensive experiments on both synthetic and real datasets demonstrate that E2EGS achieves superior reconstruction quality and trajectory accuracy, establishing a fully pose-free paradigm for event-based 3D reconstruction.", "date": "2026-03-16", "categories": []}
{"id": "2603.14765", "title": "SSR: A Training-Free Approach for Streaming 3D Reconstruction", "abstract": "Streaming 3D reconstruction demands long-horizon state updates under strict latency constraints, yet stateful recurrent models often suffer from geometric drift as errors accumulate over time. We revisit this problem from a Grassmannian manifold perspective: the latent persistent state can be viewed as a subspace representation, i.e., a point evolving on a Grassmannian manifold, where temporal coherence implies the state trajectory should remain on (or near) this manifold.Based on this view, we propose Self-expressive Sequence Regularization (SSR), a plug-and-play, training-free operator that enforces Grassmannian sequence regularity during inference.Given a window of historical states, SSR computes an analytical affinity matrix via the self-expressive property and uses it to regularize the current update, effectively pulling noisy predictions back toward the manifold-consistent trajectory with minimal overhead. Experiments on long-sequence benchmarks demonstrate that SSR consistently reduces drift and improves reconstruction quality across multiple streaming 3D reconstruction tasks.", "date": "2026-03-16", "categories": []}
{"id": "2603.15330", "title": "MeMix: Writing Less, Remembering More for Streaming 3D Reconstruction", "abstract": "Reconstruction is a fundamental task in 3D vision and a fundamental capability for spatial intelligence. Particularly, streaming 3D reconstruction is central to real-time spatial perception, yet existing recurrent online models often suffer from progressive degradation on long sequences due to state drift and forgetting, motivating inference-time remedies. We present MeMix, a training-free, plug-and-play module that improves streaming reconstruction by recasting the recurrent state into a Memory Mixture. MeMix partitions the state into multiple independent memory patches and updates only the least-aligned memory patches while exactly preserving others. This selective update mitigates catastrophic forgetting while retaining $O(1)$ inference memory, and requires no fine-tuning or additional learnable parameters, making it directly applicable to existing recurrent reconstruction models. Across standard benchmarks (ScanNet, 7-Scenes, KITTI, etc.), under identical backbones and inference settings, MeMix reduces reconstruction completeness error by 15.3% on average (up to 40.0%) across 300--500 frame streams on 7-Scenes. The code is available at https://dongjiacheng06.github.io/MeMix/", "date": "2026-03-16", "categories": []}
{"id": "2603.15612", "title": "HSImul3R: Physics-in-the-Loop Reconstruction of Simulation-Ready Human-Scene Interactions", "abstract": "We present HSImul3R, a unified framework for simulation-ready 3D reconstruction of human-scene interactions (HSI) from casual captures, including sparse-view images and monocular videos. Existing methods suffer from a perception-simulation gap: visually plausible reconstructions often violate physical constraints, leading to instability in physics engines and failure in embodied AI applications. To bridge this gap, we introduce a physically-grounded bi-directional optimization pipeline that treats the physics simulator as an active supervisor to jointly refine human dynamics and scene geometry. In the forward direction, we employ Scene-targeted Reinforcement Learning to optimize human motion under dual supervision of motion fidelity and contact stability. In the reverse direction, we propose Direct Simulation Reward Optimization, which leverages simulation feedback on gravitational stability and interaction success to refine scene geometry. We further present HSIBench, a new benchmark with diverse objects and interaction scenarios. Extensive experiments demonstrate that HSImul3R produces the first stable, simulation-ready HSI reconstructions and can be directly deployed to real-world humanoid robots.", "date": "2026-03-16", "categories": []}
{"id": "2602.13172", "title": "LongStream: Long-Sequence Streaming Autoregressive Visual Geometry", "abstract": "Long-sequence streaming 3D reconstruction remains a significant open challenge. Existing autoregressive models often fail when processing long sequences because they anchor poses to the first frame, leading to attention decay, scale drift, and extrapolation errors. We introduce LongStream, a novel gauge-decoupled streaming visual geometry model for metric-scale scene reconstruction across thousands of frames under a strictly online, future-invisible setting. Our approach is threefold. First, we discard the first-frame anchor and predict keyframe-relative poses. This reformulates long-range extrapolation into a constant-difficulty local task. Second, we introduce orthogonal scale learning. This method fully disentangles geometry from scale estimation to suppress drift. Finally, we identify attention bias issues in Transformers, including attention-sink reliance and long-term KV-cache saturation. We propose cache-consistent training combined with periodic cache refresh. This approach suppresses attention biases and contamination over ultra-long sequences and reduces the gap between training and inference. Experiments show that LongStream achieves state-of-the-art performance, enabling stable, metric-scale reconstruction over kilometer-scale sequences at 18 FPS. Project Page: https://3dagentworld.github.io/longstream/", "date": "2026-02-13", "categories": []}
{"id": "2603.12657", "title": "VFM-Recon: Unlocking Cross-Domain Scene-Level Neural Reconstruction with Scale-Aligned Foundation Priors", "abstract": "Scene-level neural volumetric reconstruction from monocular videos remains challenging, especially under severe domain shifts. Although recent advances in vision foundation models (VFMs) provide transferable generalized priors learned from large-scale data, their scaleambiguous predictions are incompatible with the scale consistency required by volumetric fusion. To address this gap, we present VFMRecon, the first attempt to bridge transferable VFM priors with scaleconsistent requirements in scene-level neural reconstruction. Specifically, we first introduce a lightweight scale alignment stage that restores multiview scale coherence. We then integrate pretrained VFM features into the neural volumetric reconstruction pipeline via lightweight task-specific adapters, which are trained for reconstruction while preserving the crossdomain robustness of pretrained representations. We train our model on ScanNet train split and evaluate on both in-distribution ScanNet test split and out-of-distribution TUM RGB-D and Tanks and Temples datasets. The results demonstrate that our model achieves state-of-theart performance across all datasets domains. In particular, on the challenging outdoor Tanks and Temples dataset, our model achieves an F1 score of 70.1 in reconstructed mesh evaluation, substantially outperforming the closest competitor, VGGT, which only attains 51.8.", "date": "2026-03-13", "categories": []}
{"id": "2509.20681", "title": "Efficient Construction of Implicit Surface Models From a Single Image for Motion Generation", "abstract": "Implicit representations have been widely applied in robotics for obstacle avoidance and path planning. In this paper, we explore the problem of constructing an implicit distance representation from a single image. Past methods for implicit surface reconstruction, such as NeuS and its variants generally require a large set of multi-view images as input, and require long training times. In this work, we propose Fast Image-to-Neural Surface (FINS), a lightweight framework that can reconstruct high-fidelity surfaces and SDF fields based on a single or a small set of images. FINS integrates a multi-resolution hash grid encoder with lightweight geometry and color heads, making the training via an approximate second-order optimizer highly efficient and capable of converging within a few seconds. Additionally, we achieve the construction of a neural surface requiring only a single RGB image, by leveraging pre-trained foundation models to estimate the geometry inherent in the image. Our experiments demonstrate that under the same conditions, our method outperforms state-of-the-art baselines in both convergence speed and accuracy on surface reconstruction and SDF field estimation. Moreover, we demonstrate the applicability of FINS for robot surface following tasks and show its scalability to a variety of benchmark datasets. Code is publicly available at https://github.com/waynechu1109/FINS.", "date": "2025-09-25", "categories": []}
{"id": "2510.08575", "title": "ReSplat: Learning Recurrent Gaussian Splatting", "abstract": "While existing feed-forward Gaussian splatting models offer computational efficiency and can generalize to sparse view settings, their performance is fundamentally constrained by relying on a single forward pass for inference. We propose ReSplat, a feed-forward recurrent Gaussian splatting model that iteratively refines 3D Gaussians without explicitly computing gradients. Our key insight is that the Gaussian splatting rendering error serves as a rich feedback signal, guiding the recurrent network to learn effective Gaussian updates. This feedback signal naturally adapts to unseen data distributions at test time, enabling robust generalization across datasets, view counts, and image resolutions. To initialize the recurrent process, we introduce a compact reconstruction model that operates in a $16 \\times$ subsampled space, producing $16 \\times$ fewer Gaussians than previous per-pixel Gaussian models. This substantially reduces computational overhead and allows for efficient Gaussian updates. Extensive experiments across varying number of input views (2, 8, 16, 32), resolutions ($256 \\times 256$ to $540 \\times 960$), and datasets (DL3DV, RealEstate10K, and ACID) demonstrate that our method achieves state-of-the-art performance while significantly reducing the number of Gaussians and improving the rendering speed. Our project page is at https://haofeixu.github.io/resplat/.", "date": "2025-10-09", "categories": []}
{"id": "2602.08958", "title": "Grow with the Flow: 4D Reconstruction of Growing Plants with Gaussian Flow Fields", "abstract": "Modeling the time-varying 3D appearance of plants during growth poses unique challenges: unlike most dynamic scenes, plants continuously generate new geometry as they expand, branch, and differentiate. Existing dynamic scene representations are ill-suited to this setting: deformation fields provide insufficient constraints to yield physically plausible scene dynamics, and 4D Gaussian splatting represents the same physical structures with different Gaussian primitives at different times, breaking temporal consistency. We introduce GrowFlow, a dynamic representation that couples 3D Gaussian primitives with a neural ordinary differential equation to model plant growth as a continuous flow field over geometric parameters (position, scale, and orientation). Our representation enables consistent appearance rendering and models nonlinear, continuous-time growth dynamics with full temporal correspondences for every primitive. To initialize a sufficient set of Gaussian primitives, we first reconstruct the mature plant and then learn a reverse-growth process, effectively simulating the plant's developmental history in reverse. GrowFlow achieves superior image quality and geometric coherence compared to prior methods on a new, multi-view timelapse dataset of plant growth, and provides the first temporally coherent representation for appearance modeling of growing 3D structures.", "date": "2026-02-09", "categories": []}
{"id": "2603.11174", "title": "GGPT: Geometry Grounded Point Transformer", "abstract": "Recent feed-forward networks have achieved remarkable progress in sparse-view 3D reconstruction by predicting dense point maps directly from RGB images. However, they often suffer from geometric inconsistencies and limited fine-grained accuracy due to the absence of explicit multi-view constraints. We introduce the Geometry-Grounded Point Transformer (GGPT), a framework that augments feed-forward reconstruction with reliable sparse geometric guidance. We first propose an improved Structure-from-Motion pipeline based on dense feature matching and lightweight geometric optimisation to efficiently estimate accurate camera poses and partial 3D point clouds from sparse input views. Building on this foundation, we propose a geometry-guided 3D point transformer that refines dense point maps under explicit partial-geometry supervision using an optimised guidance encoding. Extensive experiments demonstrate that our method provides a principled mechanism for integrating geometric priors with dense feed-forward predictions, producing reconstructions that are both geometrically consistent and spatially complete, recovering fine structures and filling gaps in textureless areas. Trained solely on ScanNet++ with VGGT predictions, GGPT generalises across architectures and datasets, substantially outperforming state-of-the-art feed-forward 3D reconstruction models in both in-domain and out-of-domain settings.", "date": "2026-03-11", "categories": []}
{"id": "2603.11543", "title": "Mango-GS: Enhancing Spatio-Temporal Consistency in Dynamic Scenes Reconstruction using Multi-Frame Node-Guided 4D Gaussian Splatting", "abstract": "Reconstructing dynamic 3D scenes with photorealistic detail and strong temporal coherence remains a significant challenge. Existing Gaussian splatting approaches for dynamic scene modeling often rely on per-frame optimization, which can overfit to instantaneous states instead of capturing underlying motion dynamics. To address this, we present Mango-GS, a multi-frame, node-guided framework for high-fidelity 4D reconstruction. Mango-GS leverages a temporal Transformer to model motion dependencies within a short window of frames, producing temporally consistent deformations. For efficiency, temporal modeling is confined to a sparse set of control nodes. Each node is represented by a decoupled canonical position and a latent code, providing a stable semantic anchor for motion propagation and preventing correspondence drift under large motion. Our framework is trained end-to-end, enhanced by an input masking strategy and two multi-frame losses to improve robustness. Extensive experiments demonstrate that Mango-GS achieves state-of-the-art reconstruction quality and real-time rendering speed, enabling high-fidelity reconstruction and interactive rendering of dynamic scenes.", "date": "2026-03-12", "categories": []}
{"id": "2603.11606", "title": "Articulat3D: Reconstructing Articulated Digital Twins From Monocular Videos with Geometric and Motion Constraints", "abstract": "Building high-fidelity digital twins of articulated objects from visual data remains a central challenge. Existing approaches depend on multi-view captures of the object in discrete, static states, which severely constrains their real-world scalability. In this paper, we introduce Articulat3D, a novel framework that constructs such digital twins from casually captured monocular videos by jointly enforcing explicit 3D geometric and motion constraints. We first propose Motion Prior-Driven Initialization, which leverages 3D point tracks to exploit the low-dimensional structure of articulated motion. By modeling scene dynamics with a compact set of motion bases, we facilitate soft decomposition of the scene into multiple rigidly-moving groups. Building on this initialization, we introduce Geometric and Motion Constraints Refinement, which enforces physically plausible articulation through learnable kinematic primitives parameterized by a joint axis, a pivot point, and per-frame motion scalars, yielding reconstructions that are both geometrically accurate and temporally coherent. Extensive experiments demonstrate that Articulat3D achieves state-of-the-art performance on synthetic benchmarks and real-world casually captured monocular videos, significantly advancing the feasibility of digital twin creation under uncontrolled real-world conditions. Our project page is at https://maxwell-zhao.github.io/Articulat3D.", "date": "2026-03-12", "categories": []}
{"id": "2601.22046", "title": "PLANING: A Loosely Coupled Triangle-Gaussian Framework for Streaming 3D Reconstruction", "abstract": "Streaming reconstruction from monocular image sequences remains challenging, as existing methods typically favor either high-quality rendering or accurate geometry, but rarely both. We present PLANING, an efficient on-the-fly reconstruction framework built on a hybrid representation that loosely couples explicit geometric primitives with neural Gaussians, enabling geometry and appearance to be modeled in a decoupled manner. This decoupling supports an online initialization and optimization strategy that separates geometry and appearance updates, yielding stable streaming reconstruction with substantially reduced structural redundancy. PLANING improves dense mesh Chamfer-L2 by 18.52% over PGSR, surpasses ARTDECO by 1.31 dB PSNR, and reconstructs ScanNetV2 scenes in under 100 seconds, over 5x faster than 2D Gaussian Splatting, while matching the quality of offline per-scene optimization. Beyond reconstruction quality, the structural clarity and computational efficiency of PLANING make it well suited for a broad range of downstream applications, such as enabling large-scale scene modeling and simulation-ready environments for embodied AI. Project page: https://city-super.github.io/PLANING/ .", "date": "2026-01-29", "categories": []}
{"id": "2603.10801", "title": "PolGS++: Physically-Guided Polarimetric Gaussian Splatting for Fast Reflective Surface Reconstruction", "abstract": "Accurate reconstruction of reflective surfaces remains a fundamental challenge in computer vision, with broad applications in real-time virtual reality and digital content creation. Although 3D Gaussian Splatting (3DGS) enables efficient novel-view rendering with explicit representations, its performance on reflective surfaces still lags behind implicit neural methods, especially in recovering fine geometry and surface normals. To address this gap, we propose PolGS++, a physically-guided polarimetric Gaussian Splatting framework for fast reflective surface reconstruction. Specifically, we integrate a polarized BRDF (pBRDF) model into 3DGS to explicitly decouple diffuse and specular components, providing physically grounded reflectance modeling and stronger geometric cues for reflective surface recovery. Furthermore, we introduce a depth-guided visibility mask acquisition mechanism that enables angle-of-polarization (AoP)-based tangent-space consistency constraints in Gaussian Splatting without costly ray-tracing intersections. This physically guided design improves reconstruction quality and efficiency, requiring only about 10 minutes of training. Extensive experiments on both synthetic and real-world datasets validate the effectiveness of our method.", "date": "2026-03-11", "categories": []}
{"id": "2603.10893", "title": "S2D: Sparse to Dense Lifting for 3D Reconstruction with Minimal Inputs", "abstract": "Explicit 3D representations have already become an essential medium for 3D simulation and understanding. However, the most commonly used point cloud and 3D Gaussian Splatting (3DGS) each suffer from non-photorealistic rendering and significant degradation under sparse inputs. In this paper, we introduce Sparse to Dense lifting (S2D), a novel pipeline that bridges the two representations and achieves high-quality 3DGS reconstruction with minimal inputs. Specifically, the S2D lifting is two-fold. We first present an efficient one-step diffusion model that lifts sparse point cloud for high-fidelity image artifact fixing. Meanwhile, to reconstruct 3D consistent scenes, we also design a corresponding reconstruction strategy with random sample drop and weighted gradient for robust model fitting from sparse input views to dense novel views. Extensive experiments show that S2D achieves the best consistency in generating novel view guidance and first-tier sparse view reconstruction quality under different input sparsity. By reconstructing stable scenes with the least possible captures among existing methods, S2D enables minimal input requirements for 3DGS applications.", "date": "2026-03-11", "categories": []}
{"id": "2509.04859", "title": "CoRe-GS: Coarse-to-Refined Gaussian Splatting with Semantic Object Focus", "abstract": "Fast and efficient 3D reconstruction is essential for time-critical robotic applications such as tele-guidance and disaster response, where operators must rapidly analyze specific points of interest (POIs). Existing semantic Gaussian Splatting (GS) approaches optimize the entire scene uniformly, incurring substantial computational cost even when only a small subset of the scene is operationally relevant. We propose CoRe-GS, a coarse-to-refine GS framework that enables task-driven POI-focused optimization. Our method first produces a segmentation-ready GS representation using a lightweight late-stage semantic refinement. Subsequently, only Gaussians associated with the selected POI are further optimized, reducing unnecessary background computation. To mitigate segmentation-induced outliers (floaters) during selective refinement, we introduce a color-based filtering mechanism that removes inconsistent Gaussians without requiring mask rasterization. We evaluate robustness multiple datasets. On LERF-Mask, our segmentation-ready representation achieves competitive mIoU using tremendously fewer optimization steps. Across synthetic and real-world datasets (NeRDS360, SCRREAM, Tanks and Temples), CoRe-GS drastically reduces training time compared to full semantic GS while improving POI reconstruction quality and mitigating floaters. These results demonstrate that task-aware selective refinement enables faster and higher-quality scene reconstruction tailored to robotic operational needs.", "date": "2025-09-05", "categories": []}
{"id": "2603.08997", "title": "SkipGS: Post-Densification Backward Skipping for Efficient 3DGS Training", "abstract": "3D Gaussian Splatting (3DGS) achieves real-time novel-view synthesis by optimizing millions of anisotropic Gaussians, yet its training remains expensive, with the backward pass dominating runtime in the post-densification refinement phase. We observe substantial update redundancy in this phase: many sampled views have near-plateaued losses and provide diminishing gradient benefits, but standard training still runs full backpropagation. We propose SkipGS with a novel view-adaptive backward gating mechanism for efficient post-densification training. SkipGS always performs the forward pass to update per-view loss statistics, and selectively skips backward passes when the sampled view's loss is consistent with its recent per-view baseline, while enforcing a minimum backward budget for stable optimization. On Mip-NeRF 360, compared to 3DGS, SkipGS reduces end-to-end training time by 23.1%, driven by a 42.0% reduction in post-densification time, with comparable reconstruction quality. Because it only changes when to backpropagate -- without modifying the renderer, representation, or loss -- SkipGS is plug-and-play and compatible with other complementary efficiency strategies for additive speedups.", "date": "2026-03-09", "categories": []}
{"id": "2603.09277", "title": "Speeding Up the Learning of 3D Gaussians with Much Shorter Gaussian Lists", "abstract": "3D Gaussian splatting (3DGS) has become a vital tool for learning a radiance field from multiple posed images. Although 3DGS shows great advantages over NeRF in terms of rendering quality and efficiency, it remains a research challenge to further improve the efficiency of learning 3D Gaussians. To overcome this challenge, we propose novel training strategies and losses to shorten each Gaussian list used to render a pixel, which speeds up the splatting by involving fewer Gaussians along a ray. Specifically, we shrink the size of each Gaussian by resetting their scales regularly, encouraging smaller Gaussians to cover fewer nearby pixels, which shortens the Gaussian lists of pixels. Additionally, we introduce an entropy constraint on the alpha blending procedure to sharpen the weight distribution of Gaussians along each ray, which drives dominant weights larger while making minor weights smaller. As a result, each Gaussian becomes more focused on the pixels where it is dominant, which reduces its impact on nearby pixels, leading to even shorter Gaussian lists. Eventually, we integrate our method into a rendering resolution scheduler which further improves efficiency through progressive resolution increase. We evaluate our method by comparing it with state-of-the-art methods on widely used benchmarks. Our results show significant advantages over others in efficiency without sacrificing rendering quality.", "date": "2026-03-10", "categories": []}
{"id": "2603.09291", "title": "DenoiseSplat: Feed-Forward Gaussian Splatting for Noisy 3D Scene Reconstruction", "abstract": "3D scene reconstruction and novel-view synthesis are fundamental for VR, robotics, and content creation. However, most NeRF and 3D Gaussian Splatting pipelines assume clean inputs and degrade under real noise and artifacts. We therefore propose DenoiseSplat, a feed-forward 3D Gaussian splatting method for noisy multi-view images. We build a large-scale, scene-consistent noisy--clean benchmark on RE10K by injecting Gaussian, Poisson, speckle, and salt-and-pepper noise with controlled intensities. With a lightweight MVSplat-style feed-forward backbone, we train end-to-end using only clean 2D renderings as supervision and no 3D ground truth. On noisy RE10K, DenoiseSplat outperforms vanilla MVSplat and a strong two-stage baseline (IDF + MVSplat) in PSNR/SSIM and LPIPS across noise types and levels.", "date": "2026-03-10", "categories": []}
{"id": "2507.13347", "title": "$\\pi^3$: Permutation-Equivariant Visual Geometry Learning", "abstract": "We introduce $\\pi^3$, a feed-forward neural network that offers a novel approach to visual geometry reconstruction, breaking the reliance on a conventional fixed reference view. Previous methods often anchor their reconstructions to a designated viewpoint, an inductive bias that can lead to instability and failures if the reference is suboptimal. In contrast, $\\pi^3$ employs a fully permutation-equivariant architecture to predict affine-invariant camera poses and scale-invariant local point maps without any reference frames. This design not only makes our model inherently robust to input ordering, but also leads to higher accuracy and performance. These advantages enable our simple and bias-free approach to achieve state-of-the-art performance on a wide range of tasks, including camera pose estimation, monocular/video depth estimation, and dense point map reconstruction. Code and models are available at https://github.com/yyfz/Pi3.", "date": "2025-07-17", "categories": []}
{"id": "2508.06968", "title": "3D Gaussian Splatting with Fisheye Images: Field of View Analysis and Depth-Based Initialization", "abstract": "We present the first evaluation of 3D Gaussian Splatting methods on real fisheye imagery with fields of view above 180\\textdegree{}. Our study evaluates Fisheye-GS \\cite{liao2024fisheyegslightweightextensiblegaussian} and 3DGUT \\cite{wu20253dgut} on indoor and outdoor scenes captured with 200\\textdegree{} fisheye cameras, with the aim of assessing the practicality of wide-angle reconstruction under severe distortion. By comparing reconstructions at 200\\textdegree{}, 160\\textdegree{}, and 120\\textdegree{} field-of-view, we show that both methods achieve their best results at 160\\textdegree{}, which balances scene coverage with image quality, while distortion at 200\\textdegree{} degrades performance. To address the common failure of Structure-from-Motion (SfM) initialization at such wide angles, we introduce a depth-based alternative using UniK3D (Universal Camera Monocular 3D Estimation) \\cite{piccinelli2025unik3d}. This represents the first application of UniK3D to fisheye imagery beyond 200\\textdegree{}, despite the model not being trained on such data. With the number of predicted points controlled to match SfM for fairness, UniK3D produces geometrically accurate reconstructions that rival or surpass SfM, even in challenging scenes with fog, glare, or open sky. These results demonstrate the feasibility of fisheye-based 3D Gaussian Splatting and provides a benchmark for future research on wide-angle reconstruction from sparse and distorted inputs.", "date": "2025-08-09", "categories": []}
{"id": "2509.22276", "title": "GS-2M: Material-aware Gaussian Splatting for High-fidelity Mesh Reconstruction", "abstract": "We propose a material-aware optimization framework for high-fidelity mesh reconstruction from multi-view images based on 3D Gaussian Splatting, referred to as GS-2M. Previous works handle these tasks separately and struggle to reconstruct highly reflective surfaces, often relying on priors from external models to enhance the decomposition results. Conversely, our method addresses these two problems by jointly optimizing attributes relevant to the quality of rendered depth and normals, maintaining geometric details while being resilient to reflective surfaces. Although contemporary works effectively solve these tasks together, they often employ sophisticated neural components to learn scene properties, which hinders their performance at scale. To further eliminate these neural components, we propose a novel roughness supervision strategy based on multi-view photometric variation. When combined with a carefully designed loss and optimization process, our unified framework produces reconstruction results comparable to state-of-the-art methods, delivering accurate triangle meshes even for reflective surfaces. We validate the effectiveness of our approach with widely used datasets from previous works and qualitative comparisons with state-of-the-art surface reconstruction methods. Project page: https://ndming.github.io/publications/gs2m/.", "date": "2025-09-26", "categories": []}
{"id": "2510.19210", "title": "MoE-GS: Mixture of Experts for Dynamic Gaussian Splatting", "abstract": "Recent advances in dynamic scene reconstruction have significantly benefited from 3D Gaussian Splatting, yet existing methods show inconsistent performance across diverse scenes, indicating no single approach effectively handles all dynamic challenges. To overcome these limitations, we propose Mixture of Experts for Dynamic Gaussian Splatting (MoE-GS), a unified framework integrating multiple specialized experts via a novel Volume-aware Pixel Router. Unlike sparsity-oriented MoE architectures in large language models, MoE-GS is designed to improve dynamic novel view synthesis quality by combining heterogeneous deformation priors, rather than to reduce training or inference-time FLOPs. Our router adaptively blends expert outputs by projecting volumetric Gaussian-level weights into pixel space through differentiable weight splatting, ensuring spatially and temporally coherent results. Although MoE-GS improves rendering quality, the increased model capacity and reduced FPS are inherent to the MoE architecture. To mitigate this, we explore two complementary directions: (1) single-pass multi-expert rendering and gate-aware Gaussian pruning, which improve efficiency within the MoE framework, and (2) a distillation strategy that transfers MoE performance to individual experts, enabling lightweight deployment without architectural changes. To the best of our knowledge, MoE-GS is the first approach incorporating Mixture-of-Experts techniques into dynamic Gaussian splatting. Extensive experiments on the N3V and Technicolor datasets demonstrate that MoE-GS consistently outperforms state-of-the-art methods with improved efficiency. Video demonstrations are available at cvsp-lab.github.io/MoE-GS.", "date": "2025-10-22", "categories": []}
{"id": "2603.06971", "title": "SurgCUT3R: Surgical Scene-Aware Continuous Understanding of Temporal 3D Representation", "abstract": "Reconstructing surgical scenes from monocular endoscopic video is critical for advancing robotic-assisted surgery. However, the application of state-of-the-art general-purpose reconstruction models is constrained by two key challenges: the lack of supervised training data and performance degradation over long video sequences. To overcome these limitations, we propose SurgCUT3R, a systematic framework that adapts unified 3D reconstruction models to the surgical domain. Our contributions are threefold. First, we develop a data generation pipeline that exploits public stereo surgical datasets to produce large-scale, metric-scale pseudo-ground-truth depth maps, effectively bridging the data gap. Second, we propose a hybrid supervision strategy that couples our pseudo-ground-truth with geometric self-correction to enhance robustness against inherent data imperfections. Third, we introduce a hierarchical inference framework that employs two specialized models to effectively mitigate accumulated pose drift over long surgical videos: one for global stability and one for local accuracy. Experiments on the SCARED and StereoMIS datasets demonstrate that our method achieves a competitive balance between accuracy and efficiency, delivering near state-of-the-art but substantially faster pose estimation and offering a practical and effective solution for robust reconstruction in surgical environments. Project page: https://chumo-xu.github.io/SurgCUT3R-ICRA26/.", "date": "2026-03-07", "categories": []}
{"id": "2603.07552", "title": "ReconDrive: Fast Feed-Forward 4D Gaussian Splatting for Autonomous Driving Scene Reconstruction", "abstract": "High-fidelity visual reconstruction and novel-view synthesis are essential for realistic closed-loop evaluation in autonomous driving. While 4D Gaussian Splatting (4DGS) offers a promising balance of accuracy and efficiency, existing per-scene optimization methods require costly iterative refinement, rendering them unscalable for extensive urban environments. Conversely, current feed-forward approaches often suffer from degraded photometric quality. To address these limitations, we propose ReconDrive, a feed-forward framework that leverages and extends the 3D foundation model VGGT for rapid, high-fidelity 4DGS generation. Our architecture introduces two core adaptations to tailor the foundation model to dynamic driving scenes: (1) Hybrid Gaussian Prediction Heads, which decouple the regression of spatial coordinates and appearance attributes to overcome the photometric deficiencies inherent in generalized foundation features; and (2) a Static-Dynamic 4D Composition strategy that explicitly captures temporal motion via velocity modeling to represent complex dynamic environments. Benchmarked on nuScenes, ReconDrive significantly outperforms existing feed-forward baselines in reconstruction, novel-view synthesis, and 3D perception. It achieves performance competitive with per-scene optimization while being orders of magnitude faster, providing a scalable and practical solution for realistic driving simulation.", "date": "2026-03-08", "categories": []}
{"id": "2603.07587", "title": "3DGS-HPC: Distractor-free 3D Gaussian Splatting with Hybrid Patch-wise Classification", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated remarkable performance in novel view synthesis and 3D scene reconstruction, yet its quality often degrades in real-world environments due to transient distractors, such as moving objects and varying shadows. Existing methods commonly rely on semantic cues extracted from pre-trained vision models to identify and suppress these distractors, but such semantics are misaligned with the binary distinction between static and transient regions and remain fragile under the appearance perturbations introduced during 3DGS optimization. We propose 3DGS-HPC, a framework that circumvents these limitations by combining two complementary principles: a patch-wise classification strategy that leverages local spatial consistency for robust region-level decisions, and a hybrid classification metric that adaptively integrates photometric and perceptual cues for more reliable separation. Extensive experiments demonstrate the superiority and robustness of our method in mitigating distractors to improve 3DGS-based novel view synthesis.", "date": "2026-03-08", "categories": []}
{"id": "2603.08055", "title": "Speed3R: Sparse Feed-forward 3D Reconstruction Models", "abstract": "While recent feed-forward 3D reconstruction models accelerate 3D reconstruction by jointly inferring dense geometry and camera poses in a single pass, their reliance on dense attention imposes a quadratic complexity, creating a prohibitive computational bottleneck that severely limits inference speed. To resolve this, we introduce Speed3R, an end-to-end trainable model inspired by the core principle of Structure-from-Motion: that a sparse set of keypoints is sufficient for robust pose estimation. Speed3R features a dual-branch attention mechanism where a compression branch creates a coarse contextual prior to guide a selection branch, which performs fine-grained attention only on the most informative image tokens. This strategy mimics the efficiency of traditional keypoint matching, achieving a remarkable 12.4x inference speedup on 1000-view sequences, while introducing a minimal, controlled trade-off in geometric accuracy. Validated on standard benchmarks with both VGGT and $\\pi^3$ backbones, our method delivers high-quality reconstructions at a fraction of computational cost, paving the way for efficient large-scale scene modeling.", "date": "2026-03-09", "categories": []}
{"id": "2603.08133", "title": "Fast Low-light Enhancement and Deblurring for 3D Dark Scenes", "abstract": "Novel view synthesis from low-light, noisy, and motion-blurred imagery remains a valuable and challenging task. Current volumetric rendering methods struggle with compound degradation, and sequential 2D preprocessing introduces artifacts due to interdependencies. In this work, we introduce FLED-GS, a fast low-light enhancement and deblurring framework that reformulates 3D scene restoration as an alternating cycle of enhancement and reconstruction. Specifically, FLED-GS inserts several intermediate brightness anchors to enable progressive recovery, preventing noise blow-up from harming deblurring or geometry. Each iteration sharpens inputs with an off-the-shelf 2D deblurrer and then performs noise-aware 3DGS reconstruction that estimates and suppresses noise while producing clean priors for the next level. Experiments show FLED-GS outperforms state-of-the-art LuSh-NeRF, achieving 21$\\times$ faster training and 11$\\times$ faster rendering.", "date": "2026-03-09", "categories": []}
{"id": "2603.08254", "title": "DynamicVGGT: Learning Dynamic Point Maps for 4D Scene Reconstruction in Autonomous Driving", "abstract": "Dynamic scene reconstruction in autonomous driving remains a fundamental challenge due to significant temporal variations, moving objects, and complex scene dynamics. Existing feed-forward 3D models have demonstrated strong performance in static reconstruction but still struggle to capture dynamic motion. To address these limitations, we propose DynamicVGGT, a unified feed-forward framework that extends VGGT from static 3D perception to dynamic 4D reconstruction. Our goal is to model point motion within feed-forward 3D models in a dynamic and temporally coherent manner. To this end, we jointly predict the current and future point maps within a shared reference coordinate system, allowing the model to implicitly learn dynamic point representations through temporal correspondence. To efficiently capture temporal dependencies, we introduce a Motion-aware Temporal Attention (MTA) module that learns motion continuity. Furthermore, we design a Dynamic 3D Gaussian Splatting Head that explicitly models point motion by predicting Gaussian velocities using learnable motion tokens under scene flow supervision. It refines dynamic geometry through continuous 3D Gaussian optimization. Extensive experiments on autonomous driving datasets demonstrate that DynamicVGGT significantly outperforms existing methods in reconstruction accuracy, achieving robust feed-forward 4D dynamic scene reconstruction under complex driving scenarios.", "date": "2026-03-09", "categories": []}
{"id": "2603.08499", "title": "Improving Continual Learning for Gaussian Splatting based Environments Reconstruction on Commercial Off-the-Shelf Edge Devices", "abstract": "Novel view synthesis (NVS) is increasingly relevant for edge robotics, where compact and incrementally updatable 3D scene models are needed for SLAM, navigation, and inspection under tight memory and latency budgets. Variational Bayesian Gaussian Splatting (VBGS) enables replay-free continual updates for the 3DGS algorithm by maintaining a probabilistic scene model, but its high-precision computations and large intermediate tensors make on-device training impractical. We present a precision-adaptive optimization framework that enables VBGS training on resource-constrained hardware without altering its variational formulation. We (i) profile VBGS to identify memory/latency hotspots, (ii) fuse memory-dominant kernels to reduce materialized intermediate tensors, and (iii) automatically assign operation-level precisions via a mixed-precision search with bounded relative error. Across the Blender, Habitat, and Replica datasets, our optimised pipeline reduces peak memory from 9.44 GB to 1.11 GB and training time from ~234 min to ~61 min on an A5000 GPU, while preserving (and in some cases improving) reconstruction quality of the state-of-the-art VBGS baseline. We also enable for the first time NVS training on a commercial embedded platform, the Jetson Orin Nano, reducing per-frame latency by 19x compared to 3DGS.", "date": "2026-03-09", "categories": []}
{"id": "2603.08503", "title": "Spherical-GOF: Geometry-Aware Panoramic Gaussian Opacity Fields for 3D Scene Reconstruction", "abstract": "Omnidirectional images are increasingly used in robotics and vision due to their wide field of view. However, extending 3D Gaussian Splatting (3DGS) to panoramic camera models remains challenging, as existing formulations are designed for perspective projections and naive adaptations often introduce distortion and geometric inconsistencies. We present Spherical-GOF, an omnidirectional Gaussian rendering framework built upon Gaussian Opacity Fields (GOF). Unlike projection-based rasterization, Spherical-GOF performs GOF ray sampling directly on the unit sphere in spherical ray space, enabling consistent ray-Gaussian interactions for panoramic rendering. To make the spherical ray casting efficient and robust, we derive a conservative spherical bounding rule for fast ray-Gaussian culling and introduce a spherical filtering scheme that adapts Gaussian footprints to distortion-varying panoramic pixel sampling. Extensive experiments on standard panoramic benchmarks (OmniBlender and OmniPhotos) demonstrate competitive photometric quality and substantially improved geometric consistency. Compared with the strongest baseline, Spherical-GOF reduces depth reprojection error by 57% and improves cycle inlier ratio by 21%. Qualitative results show cleaner depth and more coherent normal maps, with strong robustness to global panorama rotations. We further validate generalization on OmniRob, a real-world robotic omnidirectional dataset introduced in this work, featuring UAV and quadruped platforms. The source code and the OmniRob dataset will be released at https://github.com/1170632760/Spherical-GOF.", "date": "2026-03-09", "categories": []}
{"id": "2603.08661", "title": "ImprovedGS+: A High-Performance C++/CUDA Re-Implementation Strategy for 3D Gaussian Splatting", "abstract": "Recent advancements in 3D Gaussian Splatting (3DGS) have shifted the focus toward balancing reconstruction fidelity with computational efficiency. In this work, we propose ImprovedGS+, a high-performance, low-level reinvention of the ImprovedGS strategy, implemented natively within the LichtFeld-Studio framework. By transitioning from high-level Python logic to hardware-optimized C++/CUDA kernels, we achieve a significant reduction in host-device synchronization and training latency. Our implementation introduces a Long-Axis-Split (LAS) CUDA kernel, custom Laplacian-based importance kernels with Non-Maximum Suppression (NMS) for edge scores, and an adaptive Exponential Scale Scheduler. Experimental results on the Mip-NeRF360 dataset demonstrate that ImprovedGS+ establishes a new Pareto-optimal front for scene reconstruction. Our 1M-budget variant outperforms the state-of-the-art MCMC baseline by achieving a 26.8% reduction in training time (saving 17 minutes per session) and utilizing 13.3% fewer Gaussians while maintaining superior visual quality. Furthermore, our full variant demonstrates a 1.28 dB PSNR increase over the ADC baseline with a 38.4% reduction in parametric complexity. These results validate ImprovedGS+ as a scalable, high-speed solution that upholds the core pillars of Speed, Quality, and Usability within the LichtFeld-Studio ecosystem.", "date": "2026-03-09", "categories": []}
{"id": "2501.17655", "title": "FeatureGS: Eigenvalue-Feature Optimization in 3D Gaussian Splatting for Geometrically Accurate and Artifact-Reduced Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a powerful approach for 3D scene reconstruction using 3D Gaussians. However, neither the centers nor surfaces of the Gaussians are accurately aligned to the object surface, complicating their direct use in point cloud and mesh reconstruction. Additionally, 3DGS typically produces floater artifacts, increasing the number of Gaussians and storage requirements. To address these issues, we present FeatureGS, which incorporates an additional geometric loss term based on an eigenvalue-derived 3D shape feature into the optimization process of 3DGS. The goal is to improve geometric accuracy and enhance properties of planar surfaces with reduced structural entropy in local 3D neighborhoods.We present four alternative formulations for the geometric loss term based on 'planarity' of Gaussians, as well as 'planarity', 'omnivariance', and 'eigenentropy' of Gaussian neighborhoods. We provide quantitative and qualitative evaluations on 15 scenes of the DTU benchmark dataset focusing on following key aspects: Geometric accuracy and artifact-reduction, measured by the Chamfer distance, and memory efficiency, evaluated by the total number of Gaussians. Additionally, rendering quality is monitored by Peak Signal-to-Noise Ratio. FeatureGS achieves a 30 % improvement in geometric accuracy, reduces the number of Gaussians by 90 %, and suppresses floater artifacts, while maintaining comparable photometric rendering quality. The geometric loss with 'planarity' from Gaussians provides the highest geometric accuracy, while 'omnivariance' in Gaussian neighborhoods reduces floater artifacts and number of Gaussians the most. This makes FeatureGS a strong method for geometrically accurate, artifact-reduced and memory-efficient 3D scene reconstruction, enabling the direct use of Gaussian centers for geometric representation.", "date": "2025-01-29", "categories": []}
{"id": "2603.05078", "title": "MoRe: Motion-aware Feed-forward 4D Reconstruction Transformer", "abstract": "Reconstructing dynamic 4D scenes remains challenging due to the presence of moving objects that corrupt camera pose estimation. Existing optimization methods alleviate this issue with additional supervision, but they are mostly computationally expensive and impractical in real-time applications. To address these limitations, we propose MoRe, a feedforward 4D reconstruction network that efficiently recovers dynamic 3D scenes from monocular videos. Built upon a strong static reconstruction backbone, MoRe employs an attention-forcing strategy to disentangle dynamic motion from static structure. To further enhance robustness, we fine-tune the model on large-scale, diverse datasets encompassing both dynamic and static scenes. Moreover, our grouped causal attention captures temporal dependencies and adapts to varying token lengths across frames, ensuring temporally coherent geometry reconstruction. Extensive experiments on multiple benchmarks demonstrate that MoRe achieves high-quality dynamic reconstructions with exceptional efficiency.", "date": "2026-03-05", "categories": []}
{"id": "2603.05787", "title": "Spectral Probing of Feature Upsamplers in 2D-to-3D Scene Reconstruction", "abstract": "A typical 2D-to-3D pipeline takes multi-view images as input, where a Vision Foundation Model (VFM) extracts features that are spatially upsampled to dense representations for 3D reconstruction. If dense features across views preserve geometric consistency, differentiable rendering can recover an accurate 3D representation, making the feature upsampler a critical component. Recent learnable upsampling methods mainly aim to enhance spatial details, such as sharper geometry or richer textures, yet their impact on 3D awareness remains underexplored. To address this gap, we introduce a spectral diagnostic framework with six complementary metrics that characterize amplitude redistribution, structural spectral alignment, and directional stability. Across classical interpolation and learnable upsampling methods on CLIP and DINO backbones, we observe three key findings. First, structural spectral consistency (SSC/CSC) is the strongest predictor of NVS quality, whereas High-Frequency Spectral Slope Drift (HFSS) often correlates negatively with reconstruction performance, indicating that emphasizing high-frequency details alone does not necessarily improve 3D reconstruction. Second, geometry and texture respond to different spectral properties: Angular Energy Consistency (ADC) correlates more strongly with geometry-related metrics, while SSC/CSC influence texture fidelity slightly more than geometric accuracy. Third, although learnable upsamplers often produce sharper spatial features, they rarely outperform classical interpolation in reconstruction quality, and their effectiveness depends on the reconstruction model. Overall, our results indicate that reconstruction quality is more closely related to preserving spectral structure than to enhancing spatial detail, highlighting spectral consistency as an important principle for designing upsampling strategies in 2D-to-3D pipelines.", "date": "2026-03-06", "categories": []}
{"id": "2603.05888", "title": "PixARMesh: Autoregressive Mesh-Native Single-View Scene Reconstruction", "abstract": "We introduce PixARMesh, a method to autoregressively reconstruct complete 3D indoor scene meshes directly from a single RGB image. Unlike prior methods that rely on implicit signed distance fields and post-hoc layout optimization, PixARMesh jointly predicts object layout and geometry within a unified model, producing coherent and artist-ready meshes in a single forward pass. Building on recent advances in mesh generative models, we augment a point-cloud encoder with pixel-aligned image features and global scene context via cross-attention, enabling accurate spatial reasoning from a single image. Scenes are generated autoregressively from a unified token stream containing context, pose, and mesh, yielding compact meshes with high-fidelity geometry. Experiments on synthetic and real-world datasets show that PixARMesh achieves state-of-the-art reconstruction quality while producing lightweight, high-quality meshes ready for downstream applications.", "date": "2026-03-06", "categories": []}
{"id": "2603.05932", "title": "FTSplat: Feed-forward Triangle Splatting Network", "abstract": "High-fidelity three-dimensional (3D) reconstruction is essential for robotics and simulation. While Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) achieve impressive rendering quality, their reliance on time-consuming per-scene optimization limits real-time deployment. Emerging feed-forward Gaussian splatting methods improve efficiency but often lack explicit, manifold geometry required for direct simulation. To address these limitations, we propose a feed-forward framework for triangle primitive generation that directly predicts continuous triangle surfaces from calibrated multi-view images. Our method produces simulation-ready models in a single forward pass, obviating the need for per-scene optimization or post-processing. We introduce a pixel-aligned triangle generation module and incorporate relative 3D point cloud supervision to enhance geometric learning stability and consistency. Experiments demonstrate that our method achieves efficient reconstruction while maintaining seamless compatibility with standard graphics and robotic simulators.", "date": "2026-03-06", "categories": []}
{"id": "2603.06061", "title": "Transforming Omnidirectional RGB-LiDAR data into 3D Gaussian Splatting", "abstract": "The demand for large-scale digital twins is rapidly growing in robotics and autonomous driving. However, constructing these environments with 3D Gaussian Splatting (3DGS) usually requires expensive, purpose-built data collection. Meanwhile, deployed platforms routinely collect extensive omnidirectional RGB and LiDAR logs, but a significant portion of these sensor data is directly discarded or strictly underutilized due to transmission constraints and the lack of scalable reuse pipeline. In this paper, we present an omnidirectional RGB-LiDAR reuse pipeline that transforms these archived logs into robust initialization assets for 3DGS. Direct conversion of such raw logs introduces practical bottlenecks: inherent non-linear distortion leads to unreliable Structure-from-Motion (SfM) tracking, and dense, unorganized LiDAR clouds cause computational overhead during 3DGS optimization. To overcome these challenges, our pipeline strategically integrates an ERP-to-cubemap conversion module for deterministic spatial anchoring, alongside PRISM-a color stratified downsampling strategy. By bridging these multi-modal inputs via Fast Point Feature Histograms (FPFH) based global registration and Iterative Closest Point (ICP), our pipeline successfully repurposes a considerable fraction of discarded data into usable SfM geometry. Furthermore, our LiDAR-reinforced initialization consistently enhances the final 3DGS rendering fidelity in structurally complex scenes compared to vision-only baselines. Ultimately, this work provides a deterministic workflow for creating simulation-grade digital twins from standard archived sensor logs.", "date": "2026-03-06", "categories": []}
{"id": "2603.06216", "title": "EntON: Eigenentropy-Optimized Neighborhood Densification in 3D Gaussian Splatting", "abstract": "We present a novel Eigenentropy-optimized neighboorhood densification strategy EntON in 3D Gaussian Splatting (3DGS) for geometrically accurate and high-quality rendered 3D reconstruction. While standard 3DGS produces Gaussians whose centers and surfaces are poorly aligned with the underlying object geometry, surface-focused reconstruction methods frequently sacrifice photometric accuracy. In contrast to the conventional densification strategy, which relies on the magnitude of the view-space position gradient, our approach introduces a geometry-aware strategy to guide adaptive splitting and pruning. Specifically, we compute the 3D shape feature Eigenentropy from the eigenvalues of the covariance matrix in the k-nearest neighborhood of each Gaussian center, which quantifies the local structural order. These Eigenentropy values are integrated into an alternating optimization framework: During the optimization process, the algorithm alternates between (i) standard gradient-based densification, which refines regions via view-space gradients, and (ii) Eigenentropy-aware densification, which preferentially densifies Gaussians in low-Eigenentropy (ordered, flat) neighborhoods to better capture fine geometric details on the object surface, and prunes those in high-Eigenentropy (disordered, spherical) regions. We provide quantitative and qualitative evaluations on two benchmark datasets: small-scale DTU dataset and large-scale TUM2TWIN dataset, covering man-made objects and urban scenes. Experiments demonstrate that our Eigenentropy-aware alternating densification strategy improves geometric accuracy by up to 33% and rendering quality by up to 7%, while reducing the number of Gaussians by up to 50% and training time by up to 23%. Overall, EnTON achieves a favorable balance between geometric accuracy, rendering quality and efficiency by avoiding unnecessary scene expansion.", "date": "2026-03-06", "categories": []}
{"id": "2602.24290", "title": "UFO-4D: Unposed Feedforward 4D Reconstruction from Two Images", "abstract": "Dense 4D reconstruction from unposed images remains a critical challenge, with current methods relying on slow test-time optimization or fragmented, task-specific feedforward models. We introduce UFO-4D, a unified feedforward framework to reconstruct a dense, explicit 4D representation from just a pair of unposed images. UFO-4D directly estimates dynamic 3D Gaussian Splats, enabling the joint and consistent estimation of 3D geometry, 3D motion, and camera pose in a feedforward manner. Our core insight is that differentiably rendering multiple signals from a single Dynamic 3D Gaussian representation offers major training advantages. This approach enables a self-supervised image synthesis loss while tightly coupling appearance, depth, and motion. Since all modalities share the same geometric primitives, supervising one inherently regularizes and improves the others. This synergy overcomes data scarcity, allowing UFO-4D to outperform prior work by up to 3 times in joint geometry, motion, and camera pose estimation. Our representation also enables high-fidelity 4D interpolation across novel views and time. Please visit our project page for visual results: https://ufo-4d.github.io/", "date": "2026-02-27", "categories": []}
{"id": "2603.04179", "title": "NOVA3R: Non-pixel-aligned Visual Transformer for Amodal 3D Reconstruction", "abstract": "We present NOVA3R, an effective approach for non-pixel-aligned 3D reconstruction from a set of unposed images in a feed-forward manner. Unlike pixel-aligned methods that tie geometry to per-ray predictions, our formulation learns a global, view-agnostic scene representation that decouples reconstruction from pixel alignment. This addresses two key limitations in pixel-aligned 3D: (1) it recovers both visible and invisible points with a complete scene representation, and (2) it produces physically plausible geometry with fewer duplicated structures in overlapping regions. To achieve this, we introduce a scene-token mechanism that aggregates information across unposed images and a diffusion-based 3D decoder that reconstructs complete, non-pixel-aligned point clouds. Extensive experiments on both scene-level and object-level datasets demonstrate that NOVA3R outperforms state-of-the-art methods in terms of reconstruction accuracy and completeness.", "date": "2026-03-04", "categories": []}
{"id": "2603.04847", "title": "GloSplat: Joint Pose-Appearance Optimization for Faster and More Accurate 3D Reconstruction", "abstract": "Feature extraction, matching, structure from motion (SfM), and novel view synthesis (NVS) have traditionally been treated as separate problems with independent optimization objectives. We present GloSplat, a framework that performs \\emph{joint pose-appearance optimization} during 3D Gaussian Splatting training. Unlike prior joint optimization methods (BARF, NeRF--, 3RGS) that rely purely on photometric gradients for pose refinement, GloSplat preserves \\emph{explicit SfM feature tracks} as first-class entities throughout training: track 3D points are maintained as separate optimizable parameters from Gaussian primitives, providing persistent geometric anchors via a reprojection loss that operates alongside photometric supervision. This architectural choice prevents early-stage pose drift while enabling fine-grained refinement -- a capability absent in photometric-only approaches. We introduce two pipeline variants: (1) \\textbf{GloSplat-F}, a COLMAP-free variant using retrieval-based pair selection for efficient reconstruction, and (2) \\textbf{GloSplat-A}, an exhaustive matching variant for maximum quality. Both employ global SfM initialization followed by joint photometric-geometric optimization during 3DGS training. Experiments demonstrate that GloSplat-F achieves state-of-the-art among COLMAP-free methods while GloSplat-A surpasses all COLMAP-based baselines.", "date": "2026-03-05", "categories": []}
{"id": "2603.05152", "title": "SSR-GS: Separating Specular Reflection in Gaussian Splatting for Glossy Surface Reconstruction", "abstract": "In recent years, 3D Gaussian splatting (3DGS) has achieved remarkable progress in novel view synthesis. However, accurately reconstructing glossy surfaces under complex illumination remains challenging, particularly in scenes with strong specular reflections and multi-surface interreflections. To address this issue, we propose SSR-GS, a specular reflection modeling framework for glossy surface reconstruction. Specifically, we introduce a prefiltered Mip-Cubemap to model direct specular reflections efficiently, and propose an IndiASG module to capture indirect specular reflections.\n  Furthermore, we design Visual Geometry Priors (VGP) that couple a reflection-aware visual prior via a reflection score (RS) to downweight the photometric loss contribution of reflection-dominated regions, with geometry priors derived from VGGT, including progressively decayed depth supervision and transformed normal constraints. Extensive experiments on both synthetic and real-world datasets demonstrate that SSR-GS achieves state-of-the-art performance in glossy surface reconstruction.", "date": "2026-03-05", "categories": []}
{"id": "2603.05473", "title": "Towards 3D Scene Understanding of Gas Plumes in LWIR Hyperspectral Images Using Neural Radiance Fields", "abstract": "Hyperspectral images (HSI) have many applications, ranging from environmental monitoring to national security, and can be used for material detection and identification. Longwave infrared (LWIR) HSI can be used for gas plume detection and analysis. Oftentimes, only a few images of a scene of interest are available and are analyzed individually. The ability to combine information from multiple images into a single, cohesive representation could enhance analysis by providing more context on the scene's geometry and spectral properties. Neural radiance fields (NeRFs) create a latent neural representation of volumetric scene properties that enable novel-view rendering and geometry reconstruction, offering a promising avenue for hyperspectral 3D scene reconstruction. We explore the possibility of using NeRFs to create 3D scene reconstructions from LWIR HSI and demonstrate that the model can be used for the basic downstream analysis task of gas plume detection. The physics-based DIRSIG software suite was used to generate a synthetic multi-view LWIR HSI dataset of a simple facility with a strong sulfur hexafluoride gas plume. Our method, built on the standard Mip-NeRF architecture, combines state-of-the-art methods for hyperspectral NeRFs and sparse-view NeRFs, along with a novel adaptive weighted MSE loss. Our final NeRF method requires around 50% fewer training images than the standard Mip-NeRF and achieves an average PSNR of 39.8 dB with as few as 30 training images. Gas plume detection applied to NeRF-rendered test images using the adaptive coherence estimator achieves an average AUC of 0.821 when compared with detection masks generated from ground-truth test images.", "date": "2026-03-05", "categories": []}
{"id": "2511.03950", "title": "Improving Multi-View Reconstruction via Texture-Guided Gaussian-Mesh Joint Optimization", "abstract": "Reconstructing real-world objects from multi-view images is essential for applications in 3D editing, AR/VR, and digital content creation. Existing methods typically prioritize either geometric accuracy (Multi-View Stereo) or photorealistic rendering (Novel View Synthesis), often decoupling geometry and appearance optimization, which hinders downstream editing tasks. This paper advocates an unified treatment on geometry and appearance optimization for seamless Gaussian-mesh joint optimization. More specifically, we propose a novel framework that simultaneously optimizes mesh geometry (vertex positions and faces) and vertex colors via Gaussian-guided mesh differentiable rendering, leveraging photometric consistency from input images and geometric regularization from normal and depth maps. The obtained high-quality 3D reconstruction can be further exploit in down-stream editing tasks, such as relighting and shape deformation. Our code will be released in https://github.com/zhejia01/TexGuided-GS2Mesh", "date": "2025-11-06", "categories": []}
{"id": "2603.02887", "title": "Generalized non-exponential Gaussian splatting", "abstract": "In this work we generalize 3D Gaussian splatting (3DGS) to a wider family of physically-based alpha-blending operators. 3DGS has become the standard de-facto for radiance field rendering and reconstruction, given its flexibility and efficiency. At its core, it is based on alpha-blending sorted semitransparent primitives, which in the limit converges to the classic radiative transfer function with exponential transmittance. Inspired by recent research on non-exponential radiative transfer, we generalize the image formation model of 3DGS to non-exponential regimes. Based on this generalization, we use a quadratic transmittance to define sub-linear, linear, and super-linear versions of 3DGS, which exhibit faster-than-exponential decay. We demonstrate that these new non-exponential variants achieve similar quality than the original 3DGS but significantly reduce the number of overdraws, which result on speed-ups of up to $4\\times$ in complex real-world captures, on a ray-tracing-based renderer.", "date": "2026-03-03", "categories": []}
{"id": "2603.04254", "title": "EmbodiedSplat: Online Feed-Forward Semantic 3DGS for Open-Vocabulary 3D Scene Understanding", "abstract": "Understanding a 3D scene immediately with its exploration is essential for embodied tasks, where an agent must construct and comprehend the 3D scene in an online and nearly real-time manner. In this study, we propose EmbodiedSplat, an online feed-forward 3DGS for open-vocabulary scene understanding that enables simultaneous online 3D reconstruction and 3D semantic understanding from the streaming images. Unlike existing open-vocabulary 3DGS methods which are typically restricted to either offline or per-scene optimization setting, our objectives are two-fold: 1) Reconstructs the semantic-embedded 3DGS of the entire scene from over 300 streaming images in an online manner. 2) Highly generalizable to novel scenes with feed-forward design and supports nearly real-time 3D semantic reconstruction when combined with real-time 2D models. To achieve these objectives, we propose an Online Sparse Coefficients Field with a CLIP Global Codebook where it binds the 2D CLIP embeddings to each 3D Gaussian while minimizing memory consumption and preserving the full semantic generalizability of CLIP. Furthermore, we generate 3D geometric-aware CLIP features by aggregating the partial point cloud of 3DGS through 3D U-Net to compensate the 3D geometric prior to 2D-oriented language embeddings. Extensive experiments on diverse indoor datasets, including ScanNet, ScanNet++, and Replica, demonstrate both the effectiveness and efficiency of our method. Check out our project page in https://0nandon.github.io/EmbodiedSplat/.", "date": "2026-03-04", "categories": []}
{"id": "2506.08862", "title": "StreamSplat: Towards Online Dynamic 3D Reconstruction from Uncalibrated Video Streams", "abstract": "Real-time reconstruction of dynamic 3D scenes from uncalibrated video streams demands robust online methods that recover scene dynamics from sparse observations under strict latency and memory constraints. Yet most dynamic reconstruction methods rely on hours of per-scene optimization under full-sequence access, limiting practical deployment. In this work, we introduce StreamSplat, a fully feed-forward framework that instantly transforms uncalibrated video streams of arbitrary length into dynamic 3D Gaussian Splatting (3DGS) representations in an online manner. It is achieved via three key technical innovations: 1) a probabilistic sampling mechanism that robustly predicts 3D Gaussians from uncalibrated inputs; 2) a bidirectional deformation field that yields reliable associations across frames and mitigates long-term error accumulation; 3) an adaptive Gaussian fusion operation that propagates persistent Gaussians while handling emerging and vanishing ones. Extensive experiments on standard dynamic and static benchmarks demonstrate that StreamSplat achieves state-of-the-art reconstruction quality and dynamic scene modeling. Uniquely, our method supports the online reconstruction of arbitrarily long video streams with a 1200x speedup over optimization-based methods. Our code and models are available at https://streamsplat3d.github.io/.", "date": "2025-06-10", "categories": []}
{"id": "2509.24236", "title": "PROFusion: Robust and Accurate Dense Reconstruction via Camera Pose Regression and Optimization", "abstract": "Real-time dense scene reconstruction during unstable camera motions is crucial for robotics, yet current RGB-D SLAM systems fail when cameras experience large viewpoint changes, fast motions, or sudden shaking. Classical optimization-based methods deliver high accuracy but fail with poor initialization during large motions, while learning-based approaches provide robustness but lack sufficient accuracy for dense reconstruction. We address this challenge through a combination of learning-based initialization with optimization-based refinement. Our method employs a camera pose regression network to predict metric-aware relative poses from consecutive RGB-D frames, which serve as reliable starting points for a randomized optimization algorithm that further aligns depth images with the scene geometry. Extensive experiments demonstrate promising results: our approach outperforms the best competitor on challenging benchmarks, while maintaining comparable accuracy on stable motion sequences. The system operates in real-time, showcasing that combining simple and principled techniques can achieve both robustness for unstable motions and accuracy for dense reconstruction. Code released: https://github.com/siyandong/PROFusion.", "date": "2025-09-29", "categories": []}
{"id": "2509.26645", "title": "TTT3R: 3D Reconstruction as Test-Time Training", "abstract": "Modern Recurrent Neural Networks have become a competitive architecture for 3D reconstruction due to their linear-time complexity. However, their performance degrades significantly when applied beyond the training context length, revealing limited length generalization. In this work, we revisit the 3D reconstruction foundation models from a Test-Time Training perspective, framing their designs as an online learning problem. Building on this perspective, we leverage the alignment confidence between the memory state and incoming observations to derive a closed-form learning rate for memory updates, to balance between retaining historical information and adapting to new observations. This training-free intervention, termed TTT3R, substantially improves length generalization, achieving a $2\\times$ improvement in global pose estimation over baselines, while operating at 20 FPS with just 6 GB of GPU memory to process thousands of images. Code is available in https://rover-xingyu.github.io/TTT3R", "date": "2025-09-30", "categories": []}
{"id": "2510.06219", "title": "Human3R: Everyone Everywhere All at Once", "abstract": "We present Human3R, a unified, feed-forward framework for online 4D human-scene reconstruction, in the world frame, from casually captured monocular videos. Unlike previous approaches that rely on multi-stage pipelines, iterative contact-aware refinement between humans and scenes, and heavy dependencies, e.g., human detection, depth estimation, and SLAM pre-processing, Human3R jointly recovers global multi-person SMPL-X bodies (\"everyone\"), dense 3D scene (\"everywhere\"), and camera trajectories in a single forward pass (\"all-at-once\"). Our method builds upon the 4D online reconstruction model CUT3R, and uses parameter-efficient visual prompt tuning, to strive to preserve CUT3R's rich spatiotemporal priors, while enabling direct readout of multiple SMPL-X bodies. Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, in real-time (15 FPS) with a low memory footprint (8 GB). Extensive experiments demonstrate that Human3R delivers state-of-the-art or competitive performance across tasks, including global human motion estimation, local human mesh recovery, video depth estimation, and camera pose estimation, with a single unified model. We hope that Human3R will serve as a simple yet strong baseline, which can be easily adapted for downstream applications. Code, models and 4D interactive demos are available at https://fanegg.github.io/Human3R/.", "date": "2025-10-07", "categories": []}
{"id": "2603.01099", "title": "HeroGS: Hierarchical Guidance for Robust 3D Gaussian Splatting under Sparse Views", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as a promising approach in novel view synthesis, combining photorealistic rendering with real-time efficiency. However, its success heavily relies on dense camera coverage; under sparse-view conditions, insufficient supervision leads to irregular Gaussian distributions, characterized by globally sparse coverage, blurred background, and distorted high-frequency areas. To address this, we propose HeroGS, Hierarchical Guidance for Robust 3D Gaussian Splatting, a unified framework that establishes hierarchical guidance across the image, feature, and parameter levels. At the image level, sparse supervision is converted into pseudo-dense guidance, globally regularizing the Gaussian distributions and forming a consistent foundation for subsequent optimization. Building upon this, Feature-Adaptive Densification and Pruning (FADP) at the feature level leverages low-level features to refine high-frequency details and adaptively densifies Gaussians in background regions. The optimized distributions then support Co-Pruned Geometry Consistency (CPG) at parameter level, which guides geometric consistency through parameter freezing and co-pruning, effectively removing inconsistent splats. The hierarchical guidance strategy effectively constrains and optimizes the overall Gaussian distributions, thereby enhancing both structural fidelity and rendering quality. Extensive experiments demonstrate that HeroGS achieves high-fidelity reconstructions and consistently surpasses state-of-the-art baselines under sparse-view conditions.", "date": "2026-03-01", "categories": []}
{"id": "2603.02133", "title": "SimRecon: SimReady Compositional Scene Reconstruction from Real Videos", "abstract": "Compositional scene reconstruction seeks to create object-centric representations rather than holistic scenes from real-world videos, which is natively applicable for simulation and interaction. Conventional compositional reconstruction approaches primarily emphasize on visual appearance and show limited generalization ability to real-world scenarios. In this paper, we propose SimRecon, a framework that realizes a \"Perception-Generation-Simulation\" pipeline towards cluttered scene reconstruction, which first conducts scene-level semantic reconstruction from video input, then performs single-object generation, and finally assembles these assets in the simulator. However, naively combining these three stages leads to visual infidelity of generated assets and physical implausibility of the final scene, a problem particularly severe for complex scenes. Thus, we further propose two bridging modules between the three stages to address this problem. To be specific, for the transition from Perception to Generation, critical for visual fidelity, we introduce Active Viewpoint Optimization, which actively searches in 3D space to acquire optimal projected images as conditions for single-object completion. Moreover, for the transition from Generation to Simulation, essential for physical plausibility, we propose a Scene Graph Synthesizer, which guides the construction from scratch in 3D simulators, mirroring the native, constructive principle of the real world. Extensive experiments on the ScanNet dataset validate our method's superior performance over previous state-of-the-art approaches.", "date": "2026-03-02", "categories": []}
{"id": "2603.02134", "title": "OnlineX: Unified Online 3D Reconstruction and Understanding with Active-to-Stable State Evolution", "abstract": "Recent advances in generalizable 3D Gaussian Splatting (3DGS) have enabled rapid 3D scene reconstruction within seconds, eliminating the need for per-scene optimization. However, existing methods primarily follow an offline reconstruction paradigm, lacking the capacity for continuous reconstruction, which limits their applicability to online scenarios such as robotics and VR/AR. In this paper, we introduce OnlineX, a feed-forward framework that reconstructs both 3D visual appearance and language fields in an online manner using only streaming images. A key challenge in online formulation is the cumulative drift issue, which is rooted in the fundamental conflict between two opposing roles of the memory state: an active role that constantly refreshes to capture high-frequency local geometry, and a stable role that conservatively accumulates and preserves the long-term global structure. To address this, we introduce a decoupled active-to-stable state evolution paradigm. Our framework decouples the memory state into a dedicated active state and a persistent stable state, and then cohesively fuses the information from the former into the latter to achieve both fidelity and stability. Moreover, we jointly model visual appearance and language fields and incorporate an implicit Gaussian fusion module to enhance reconstruction quality. Experiments on mainstream datasets demonstrate that our method consistently outperforms prior work in novel view synthesis and semantic understanding, showcasing robust performance across input sequences of varying lengths with real-time inference speed.", "date": "2026-03-02", "categories": []}
{"id": "2603.02351", "title": "MERG3R: A Divide-and-Conquer Approach to Large-Scale Neural Visual Geometry", "abstract": "Recent advancements in neural visual geometry, including transformer-based models such as VGGT and Pi3, have achieved impressive accuracy on 3D reconstruction tasks. However, their reliance on full attention makes them fundamentally limited by GPU memory capacity, preventing them from scaling to large, unordered image collections. We introduce MERG3R, a training-free divide-and-conquer framework that enables geometric foundation models to operate far beyond their native memory limits. MERG3R first reorders and partitions unordered images into overlapping, geometrically diverse subsets that can be reconstructed independently. It then merges the resulting local reconstructions through an efficient global alignment and confidence-weighted bundle adjustment procedure, producing a globally consistent 3D model. Our framework is model-agnostic and can be paired with existing neural geometry models. Across large-scale datasets, including 7-Scenes, NRGBD, Tanks & Temples, and Cambridge Landmarks, MERG3R consistently improves reconstruction accuracy, memory efficiency, and scalability, enabling high-quality reconstruction when the dataset exceeds memory capacity limits.", "date": "2026-03-02", "categories": []}
{"id": "2603.02548", "title": "SemGS: Feed-Forward Semantic 3D Gaussian Splatting from Sparse Views for Generalizable Scene Understanding", "abstract": "Semantic understanding of 3D scenes is essential for robots to operate effectively and safely in complex environments. Existing methods for semantic scene reconstruction and semantic-aware novel view synthesis often rely on dense multi-view inputs and require scene-specific optimization, limiting their practicality and scalability in real-world applications. To address these challenges, we propose SemGS, a feed-forward framework for reconstructing generalizable semantic fields from sparse image inputs. SemGS uses a dual-branch architecture to extract color and semantic features, where the two branches share shallow CNN layers, allowing semantic reasoning to leverage textural and structural cues in color appearance. We also incorporate a camera-aware attention mechanism into the feature extractor to explicitly model geometric relationships between camera viewpoints. The extracted features are decoded into dual-Gaussians that share geometric consistency while preserving branch-specific attributes, and further rasterized to synthesize semantic maps under novel viewpoints. Additionally, we introduce a regional smoothness loss to enhance semantic coherence. Experiments show that SemGS achieves state-of-the-art performance on benchmark datasets, while providing rapid inference and strong generalization capabilities across diverse synthetic and real-world scenarios.", "date": "2026-03-03", "categories": []}
{"id": "2507.23782", "title": "MonoFusion: Sparse-View 4D Reconstruction via Monocular Fusion", "abstract": "We address the problem of dynamic scene reconstruction from sparse-view videos. Prior work often requires dense multi-view captures with hundreds of calibrated cameras (e.g. Panoptic Studio). Such multi-view setups are prohibitively expensive to build and cannot capture diverse scenes in-the-wild. In contrast, we aim to reconstruct dynamic human behaviors, such as repairing a bike or dancing, from a small set of sparse-view cameras with complete scene coverage (e.g. four equidistant inward-facing static cameras). We find that dense multi-view reconstruction methods struggle to adapt to this sparse-view setup due to limited overlap between viewpoints. To address these limitations, we carefully align independent monocular reconstructions of each camera to produce time- and view-consistent dynamic scene reconstructions. Extensive experiments on PanopticStudio and Ego-Exo4D demonstrate that our method achieves higher quality reconstructions than prior art, particularly when rendering novel views. Code, data, and data-processing scripts are available on https://github.com/Z1hanW/MonoFusion.", "date": "2025-07-31", "categories": []}
{"id": "2510.18489", "title": "Mono4DGS-HDR: High Dynamic Range 4D Gaussian Splatting from Alternating-exposure Monocular Videos", "abstract": "We introduce Mono4DGS-HDR, the first system for reconstructing renderable 4D high dynamic range (HDR) scenes from unposed monocular low dynamic range (LDR) videos captured with alternating exposures. To tackle such a challenging problem, we present a unified framework with two-stage optimization approach based on Gaussian Splatting. The first stage learns a video HDR Gaussian representation in orthographic camera coordinate space, eliminating the need for camera poses and enabling robust initial HDR video reconstruction. The second stage transforms video Gaussians into world space and jointly refines the world Gaussians with camera poses. Furthermore, we propose a temporal luminance regularization strategy to enhance the temporal consistency of the HDR appearance. Since our task has not been studied before, we construct a new evaluation benchmark using publicly available datasets for HDR video reconstruction. Extensive experiments demonstrate that Mono4DGS-HDR significantly outperforms alternative solutions adapted from state-of-the-art methods in both rendering quality and speed.", "date": "2025-10-21", "categories": []}
{"id": "2512.14696", "title": "CRISP: Contact-Guided Real2Sim from Monocular Video with Planar Scene Primitives", "abstract": "We introduce CRISP, a method that recovers simulatable human motion and scene geometry from monocular video. Prior work on joint human-scene reconstruction relies on data-driven priors and joint optimization with no physics in the loop, or recovers noisy geometry with artifacts that cause motion tracking policies with scene interactions to fail. In contrast, our key insight is to recover convex, clean, and simulation-ready geometry by fitting planar primitives to a point cloud reconstruction of the scene, via a simple clustering pipeline over depth, normals, and flow. To reconstruct scene geometry that might be occluded during interactions, we make use of human-scene contact modeling (e.g., we use human posture to reconstruct the occluded seat of a chair). Finally, we ensure that human and scene reconstructions are physically-plausible by using them to drive a humanoid controller via reinforcement learning. Our approach reduces motion tracking failure rates from 55.2\\% to 6.9\\% on human-centric video benchmarks (EMDB, PROX), while delivering a 43\\% faster RL simulation throughput. We further validate it on in-the-wild videos including casually-captured videos, Internet videos, and even Sora-generated videos. This demonstrates CRISP's ability to generate physically-valid human motion and interaction environments at scale, greatly advancing real-to-sim applications for robotics and AR/VR.", "date": "2025-12-16", "categories": []}
{"id": "2602.20160", "title": "tttLRM: Test-Time Training for Long Context and Autoregressive 3D Reconstruction", "abstract": "We propose tttLRM, a novel large 3D reconstruction model that leverages a Test-Time Training (TTT) layer to enable long-context, autoregressive 3D reconstruction with linear computational complexity, further scaling the model's capability. Our framework efficiently compresses multiple image observations into the fast weights of the TTT layer, forming an implicit 3D representation in the latent space that can be decoded into various explicit formats, such as Gaussian Splats (GS) for downstream applications. The online learning variant of our model supports progressive 3D reconstruction and refinement from streaming observations. We demonstrate that pretraining on novel view synthesis tasks effectively transfers to explicit 3D modeling, resulting in improved reconstruction quality and faster convergence. Extensive experiments show that our method achieves superior performance in feedforward 3D Gaussian reconstruction compared to state-of-the-art approaches on both objects and scenes.", "date": "2026-02-23", "categories": []}
{"id": "2603.00697", "title": "TokenSplat: Token-aligned 3D Gaussian Splatting for Feed-forward Pose-free Reconstruction", "abstract": "We present TokenSplat, a feed-forward framework for joint 3D Gaussian reconstruction and camera pose estimation from unposed multi-view images. At its core, TokenSplat introduces a Token-aligned Gaussian Prediction module that aligns semantically corresponding information across views directly in the feature space. Guided by coarse token positions and fusion confidence, it aggregates multi-scale contextual features to enable long-range cross-view reasoning and reduce redundancy from overlapping Gaussians. To further enhance pose robustness and disentangle viewpoint cues from scene semantics, TokenSplat employs learnable camera tokens and an Asymmetric Dual-Flow Decoder (ADF-Decoder) that enforces directionally constrained communication between camera and image tokens. This maintains clean factorization within a feed-forward architecture, enabling coherent reconstruction and stable pose estimation without iterative refinement. Extensive experiments demonstrate that TokenSplat achieves higher reconstruction fidelity and novel-view synthesis quality in pose-free settings, and significantly improves pose estimation accuracy compared to prior pose-free methods. Project page: https://kidleyh.github.io/tokensplat/.", "date": "2026-02-28", "categories": []}
{"id": "2603.01194", "title": "RnG: A Unified Transformer for Complete 3D Modeling from Partial Observations", "abstract": "Human perceive the 3D world through 2D observations from limited viewpoints. While recent feed-forward generalizable 3D reconstruction models excel at recovering 3D structures from sparse images, their representations are often confined to observed regions, leaving unseen geometry un-modeled. This raises a key, fundamental challenge: Can we infer a complete 3D structure from partial 2D observations? We present RnG (Reconstruction and Generation), a novel feed-forward Transformer that unifies these two tasks by predicting an implicit, complete 3D representation. At the core of RnG, we propose a reconstruction-guided causal attention mechanism that separates reconstruction and generation at the attention level, and treats the KV-cache as an implicit 3D representation. Then, arbitrary poses can efficiently query this cache to render high-fidelity, novel-view RGBD outputs. As a result, RnG not only accurately reconstructs visible geometry but also generates plausible, coherent unseen geometry and appearance. Our method achieves state-of-the-art performance in both generalizable 3D reconstruction and novel view generation, while operating efficiently enough for real-time interactive applications. Project page: https://npucvr.github.io/RnG", "date": "2026-03-01", "categories": []}
{"id": "2603.01603", "title": "Sparse View Distractor-Free Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) enables efficient training and fast novel view synthesis in static environments. To address challenges posed by transient objects, distractor-free 3DGS methods have emerged and shown promising results when dense image captures are available. However, their performance degrades significantly under sparse input conditions. This limitation primarily stems from the reliance on the color residual heuristics to guide the training, which becomes unreliable with limited observations. In this work, we propose a framework to enhance distractor-free 3DGS under sparse-view conditions by incorporating rich prior information. Specifically, we first adopt the geometry foundation model VGGT to estimate camera parameters and generate a dense set of initial 3D points. Then, we harness the attention maps from VGGT for efficient and accurate semantic entity matching. Additionally, we utilize Vision-Language Models (VLMs) to further identify and preserve the large static regions in the scene. We also demonstrate how these priors can be seamlessly integrated into existing distractor-free 3DGS methods. Extensive experiments confirm the effectiveness and robustness of our approach in mitigating transient distractors for sparse-view 3DGS training.", "date": "2026-03-02", "categories": []}
{"id": "2603.02049", "title": "WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories", "abstract": "Recent advances in foundational Video Diffusion Models (VDMs) have yielded significant progress. Yet, despite the remarkable visual quality of generated videos, reconstructing consistent 3D scenes from these outputs remains challenging, due to limited camera controllability and inconsistent generated content when viewed from distinct camera trajectories. In this paper, we propose WorldStereo, a novel framework that bridges camera-guided video generation and 3D reconstruction via two dedicated geometric memory modules. Formally, the global-geometric memory enables precise camera control while injecting coarse structural priors through incrementally updated point clouds. Moreover, the spatial-stereo memory constrains the model's attention receptive fields with 3D correspondence to focus on fine-grained details from the memory bank. These components enable WorldStereo to generate multi-view-consistent videos under precise camera control, facilitating high-quality 3D reconstruction. Furthermore, the flexible control branch-based WorldStereo shows impressive efficiency, benefiting from the distribution matching distilled VDM backbone without joint training. Extensive experiments across both camera-guided video generation and 3D reconstruction benchmarks demonstrate the effectiveness of our approach. Notably, we show that WorldStereo acts as a powerful world model, tackling diverse scene generation tasks (whether starting from perspective or panoramic images) with high-fidelity 3D results. Models will be released.", "date": "2026-03-02", "categories": []}
{"id": "2510.12768", "title": "Uncertainty Matters in Dynamic Gaussian Splatting for Monocular 4D Reconstruction", "abstract": "Reconstructing dynamic 3D scenes from monocular input is fundamentally under-constrained, with ambiguities arising from occlusion and extreme novel views. While dynamic Gaussian Splatting offers an efficient representation, vanilla models optimize all Gaussian primitives uniformly, ignoring whether they are well or poorly observed. This limitation leads to motion drifts under occlusion and degraded synthesis when extrapolating to unseen views. We argue that uncertainty matters: Gaussians with recurring observations across views and time act as reliable anchors to guide motion, whereas those with limited visibility are treated as less reliable. To this end, we introduce USplat4D, a novel Uncertainty-aware dynamic Gaussian Splatting framework that propagates reliable motion cues to enhance 4D reconstruction. Our approach estimates time-varying per-Gaussian uncertainty and leverages it to construct a spatio-temporal graph for uncertainty-aware optimization. Experiments on diverse real and synthetic datasets show that explicitly modeling uncertainty consistently improves dynamic Gaussian Splatting models, yielding more stable geometry under occlusion and high-quality synthesis at extreme viewpoints.", "date": "2025-10-14", "categories": []}
{"id": "2602.18882", "title": "SceneTok: A Compressed, Diffusable Token Space for 3D Scenes", "abstract": "We present SceneTok, a novel tokenizer for encoding view sets of scenes into a compressed and diffusable set of unstructured tokens. Existing approaches for 3D scene representation and generation commonly use 3D data structures or view-aligned fields. In contrast, we introduce the first method that encodes scene information into a small set of permutation-invariant tokens that is disentangled from the spatial grid. The scene tokens are predicted by a multi-view tokenizer given many context views and rendered into novel views by employing a light-weight rectified flow decoder. We show that the compression is 1-3 orders of magnitude stronger than for other representations while still reaching state-of-the-art reconstruction quality. Further, our representation can be rendered from novel trajectories, including ones deviating from the input trajectory, and we show that the decoder gracefully handles uncertainty. Finally, the highly-compressed set of unstructured latent scene tokens enables simple and efficient scene generation in 5 seconds, achieving a much better quality-speed trade-off than previous paradigms.", "date": "2026-02-21", "categories": []}
{"id": "2602.19766", "title": "One2Scene: Geometric Consistent Explorable 3D Scene Generation from a Single Image", "abstract": "Generating explorable 3D scenes from a single image is a highly challenging problem in 3D vision. Existing methods struggle to support free exploration, often producing severe geometric distortions and noisy artifacts when the viewpoint moves far from the original perspective. We introduce \\textbf{One2Scene}, an effective framework that decomposes this ill-posed problem into three tractable sub-tasks to enable immersive explorable scene generation. We first use a panorama generator to produce anchor views from a single input image as initialization. Then, we lift these 2D anchors into an explicit 3D geometric scaffold via a generalizable, feed-forward Gaussian Splatting network. Instead of treating the panorama as a single image for reconstruction, we project it into multiple sparse anchor views and reformulate the reconstruction task as multi-view stereo matching, which allows us to leverage robust geometric priors learned from large-scale multi-view datasets. A bidirectional feature fusion module is used to enforce cross-view consistency, yielding an efficient and geometrically reliable scaffold. Finally, the scaffold serves as a strong prior for a novel view generator to produce photorealistic and geometrically accurate views at arbitrary cameras. By explicitly conditioning on a 3D-consistent scaffold to perform reconstruction, One2Scene works stably under large camera motions, supporting immersive scene exploration. Extensive experiments show that One2Scene substantially outperforms state-of-the-art methods in panorama depth estimation, feed-forward 360{\\deg} reconstruction, and explorable 3D scene generation. Project page: https://one2scene5406.github.io/", "date": "2026-02-23", "categories": []}
{"id": "2602.23645", "title": "BuildAnyPoint: 3D Building Structured Abstraction from Diverse Point Clouds", "abstract": "We introduce BuildAnyPoint, a novel generative framework for structured 3D building reconstruction from point clouds with diverse distributions, such as those captured by airborne LiDAR and Structure-from-Motion. To recover artist-created building abstraction in this highly underconstrained setting, we capitalize on the role of explicit 3D generative priors in autoregressive mesh generation. Specifically, we design a Loosely Cascaded Diffusion Transformer (Loca-DiT) that initially recovers the underlying distribution from noisy or sparse points, followed by autoregressively encapsulating them into compact meshes. We first formulate distribution recovery as a conditional generation task by training latent diffusion models conditioned on input point clouds, and then tailor a decoder-only transformer for conditional autoregressive mesh generation based on the recovered point clouds. Our method delivers substantial qualitative and quantitative improvements over prior building abstraction methods. Furthermore, the effectiveness of our approach is evidenced by the strong performance of its recovered point clouds on building point cloud completion benchmarks, which exhibit improved surface accuracy and distribution uniformity.", "date": "2026-02-27", "categories": []}
{"id": "2602.24136", "title": "Prune Wisely, Reconstruct Sharply: Compact 3D Gaussian Splatting via Adaptive Pruning and Difference-of-Gaussian Primitives", "abstract": "Recent significant advances in 3D scene representation have been driven by 3D Gaussian Splatting (3DGS), which has enabled real-time rendering with photorealistic quality. 3DGS often requires a large number of primitives to achieve high fidelity, leading to redundant representations and high resource consumption, thereby limiting its scalability for complex or large-scale scenes. Consequently, effective pruning strategies and more expressive primitives that can reduce redundancy while preserving visual quality are crucial for practical deployment. We propose an efficient, integrated reconstruction-aware pruning strategy that adaptively determines pruning timing and refining intervals based on reconstruction quality, thus reducing model size while enhancing rendering quality. Moreover, we introduce a 3D Difference-of-Gaussians primitive that jointly models both positive and negative densities in a single primitive, improving the expressiveness of Gaussians under compact configurations. Our method significantly improves model compactness, achieving up to 90\\% reduction in Gaussian-count while delivering visual quality that is similar to, or in some cases better than, that produced by state-of-the-art methods. Code will be made publicly available.", "date": "2026-02-27", "categories": []}
{"id": "2411.17605", "title": "Distractor-free Generalizable 3D Gaussian Splatting", "abstract": "We present DGGS, a novel framework that addresses the previously unexplored challenge: $\\textbf{Distractor-free Generalizable 3D Gaussian Splatting}$ (3DGS). It mitigates 3D inconsistency and training instability caused by distractor data in the cross-scenes generalizable train setting while enabling feedforward inference for 3DGS and distractor masks from references in the unseen scenes. To achieve these objectives, DGGS proposes a scene-agnostic reference-based mask prediction and refinement module during the training phase, effectively eliminating the impact of distractor on training stability. Moreover, we combat distractor-induced artifacts and holes at inference time through a novel two-stage inference framework for references scoring and re-selection, complemented by a distractor pruning mechanism that further removes residual distractor 3DGS-primitive influences. Extensive feedforward experiments on the real and our synthetic data show DGGS's reconstruction capability when dealing with novel distractor scenes. Moreover, our generalizable mask prediction even achieves an accuracy superior to existing scene-specific training methods. Homepage is https://github.com/bbbbby-99/DGGS.", "date": "2024-11-26", "categories": []}
{"id": "2501.02158", "title": "Joint Optimization for 4D Human-Scene Reconstruction in the Wild", "abstract": "Reconstructing human motion and its surrounding environment is crucial for understanding human-scene interaction and predicting human movements in the scene. While much progress has been made in capturing human-scene interaction in constrained environments, those prior methods can hardly reconstruct the natural and diverse human motion and scene context from web videos. In this work, we propose JOSH, a novel optimization-based method for 4D human-scene reconstruction in the wild from monocular videos. JOSH uses techniques in both dense scene reconstruction and human mesh recovery as initialization, and then it leverages the human-scene contact constraints to jointly optimize the scene, the camera poses, and the human motion. Experiment results show JOSH achieves better results on both global human motion estimation and dense scene reconstruction by joint optimization of scene geometry and human motion. We further design a more efficient model, JOSH3R, and directly train it with pseudo-labels from web videos. JOSH3R outperforms other optimization-free methods by only training with labels predicted from JOSH, further demonstrating its accuracy and generalization ability.", "date": "2025-01-04", "categories": []}
{"id": "2510.12099", "title": "G4Splat: Geometry-Guided Gaussian Splatting with Generative Prior", "abstract": "Despite recent advances in leveraging generative prior from pre-trained diffusion models for 3D scene reconstruction, existing methods still face two critical limitations. First, due to the lack of reliable geometric supervision, they struggle to produce high-quality reconstructions even in observed regions, let alone in unobserved areas. Second, they lack effective mechanisms to mitigate multi-view inconsistencies in the generated images, leading to severe shape-appearance ambiguities and degraded scene geometry. In this paper, we identify accurate geometry as the fundamental prerequisite for effectively exploiting generative models to enhance 3D scene reconstruction. We first propose to leverage the prevalence of planar structures to derive accurate metric-scale depth maps, providing reliable supervision in both observed and unobserved regions. Furthermore, we incorporate this geometry guidance throughout the generative pipeline to improve visibility mask estimation, guide novel view selection, and enhance multi-view consistency when inpainting with video diffusion models, resulting in accurate and consistent scene completion. Extensive experiments on Replica, ScanNet++, DeepBlending and Mip-NeRF 360 show that our method consistently outperforms existing baselines in both geometry and appearance reconstruction, particularly for unobserved regions. Moreover, our method naturally supports single-view inputs and unposed videos, with strong generalizability in both indoor and outdoor scenarios with practical real-world applicability. The project page is available at https://dali-jack.github.io/g4splat-web/.", "date": "2025-10-14", "categories": []}
{"id": "2602.22376", "title": "AeroDGS: Physically Consistent Dynamic Gaussian Splatting for Single-Sequence Aerial 4D Reconstruction", "abstract": "Recent advances in 4D scene reconstruction have significantly improved dynamic modeling across various domains. However, existing approaches remain limited under aerial conditions with single-view capture, wide spatial range, and dynamic objects of limited spatial footprint and large motion disparity. These challenges cause severe depth ambiguity and unstable motion estimation, making monocular aerial reconstruction inherently ill-posed. To this end, we present AeroDGS, a physics-guided 4D Gaussian splatting framework for monocular UAV videos. AeroDGS introduces a Monocular Geometry Lifting module that reconstructs reliable static and dynamic geometry from a single aerial sequence, providing a robust basis for dynamic estimation. To further resolve monocular ambiguity, we propose a Physics-Guided Optimization module that incorporates differentiable ground-support, upright-stability, and trajectory-smoothness priors, transforming ambiguous image cues into physically consistent motion. The framework jointly refines static backgrounds and dynamic entities with stable geometry and coherent temporal evolution. We additionally build a real-world UAV dataset that spans various altitudes and motion conditions to evaluate dynamic aerial reconstruction. Experiments on synthetic and real UAV scenes demonstrate that AeroDGS outperforms state-of-the-art methods, achieving superior reconstruction fidelity in dynamic aerial environments.", "date": "2026-02-25", "categories": []}
{"id": "2602.22565", "title": "SwiftNDC: Fast Neural Depth Correction for High-Fidelity 3D Reconstruction", "abstract": "Depth-guided 3D reconstruction has gained popularity as a fast alternative to optimization-heavy approaches, yet existing methods still suffer from scale drift, multi-view inconsistencies, and the need for substantial refinement to achieve high-fidelity geometry. Here, we propose SwiftNDC, a fast and general framework built around a Neural Depth Correction field that produces cross-view consistent depth maps. From these refined depths, we generate a dense point cloud through back-projection and robust reprojection-error filtering, obtaining a clean and uniformly distributed geometric initialization for downstream reconstruction. This reliable dense geometry substantially accelerates 3D Gaussian Splatting (3DGS) for mesh reconstruction, enabling high-quality surfaces with significantly fewer optimization iterations. For novel-view synthesis, SwiftNDC can also improve 3DGS rendering quality, highlighting the benefits of strong geometric initialization. We conduct a comprehensive study across five datasets, including two for mesh reconstruction, as well as three for novel-view synthesis. SwiftNDC consistently reduces running time for accurate mesh reconstruction and boosts rendering fidelity for view synthesis, demonstrating the effectiveness of combining neural depth refinement with robust geometric initialization for high-fidelity and efficient 3D reconstruction.", "date": "2026-02-26", "categories": []}
{"id": "2602.22571", "title": "GIFSplat: Generative Prior-Guided Iterative Feed-Forward 3D Gaussian Splatting from Sparse Views", "abstract": "Feed-forward 3D reconstruction offers substantial runtime advantages over per-scene optimization, which remains slow at inference and often fragile under sparse views. However, existing feed-forward methods still have potential for further performance gains, especially for out-of-domain data, and struggle to retain second-level inference time once a generative prior is introduced. These limitations stem from the one-shot prediction paradigm in existing feed-forward pipeline: models are strictly bounded by capacity, lack inference-time refinement, and are ill-suited for continuously injecting generative priors. We introduce GIFSplat, a purely feed-forward iterative refinement framework for 3D Gaussian Splatting from sparse unposed views. A small number of forward-only residual updates progressively refine current 3D scene using rendering evidence, achieve favorable balance between efficiency and quality. Furthermore, we distill a frozen diffusion prior into Gaussian-level cues from enhanced novel renderings without gradient backpropagation or ever-increasing view-set expansion, thereby enabling per-scene adaptation with generative prior while preserving feed-forward efficiency. Across DL3DV, RealEstate10K, and DTU, GIFSplat consistently outperforms state-of-the-art feed-forward baselines, improving PSNR by up to +2.1 dB, and it maintains second-scale inference time without requiring camera poses or any test-time gradient optimization.", "date": "2026-02-26", "categories": []}
{"id": "2602.23224", "title": "UniScale: Unified Scale-Aware 3D Reconstruction for Multi-View Understanding via Prior Injection for Robotic Perception", "abstract": "We present UniScale, a unified, scale-aware multi-view 3D reconstruction framework for robotic applications that flexibly integrates geometric priors through a modular, semantically informed design. In vision-based robotic navigation, the accurate extraction of environmental structure from raw image sequences is critical for downstream tasks. UniScale addresses this challenge with a single feed-forward network that jointly estimates camera intrinsics and extrinsics, scale-invariant depth and point maps, and the metric scale of a scene from multi-view images, while optionally incorporating auxiliary geometric priors when available. By combining global contextual reasoning with camera-aware feature representations, UniScale is able to recover the metric-scale of the scene. In robotic settings where camera intrinsics are known, they can be easily incorporated to improve performance, with additional gains obtained when camera poses are also available. This co-design enables robust, metric-aware 3D reconstruction within a single unified model. Importantly, UniScale does not require training from scratch, and leverages world priors exhibited in pre-existing models without geometric encoding strategies, making it particularly suitable for resource-constrained robotic teams. We evaluate UniScale on multiple benchmarks, demonstrating strong generalization and consistent performance across diverse environments. We will release our implementation upon acceptance.", "date": "2026-02-26", "categories": []}
{"id": "2602.23361", "title": "VGG-T$^3$: Offline Feed-Forward 3D Reconstruction at Scale", "abstract": "We present a scalable 3D reconstruction model that addresses a critical limitation in offline feed-forward methods: their computational and memory requirements grow quadratically w.r.t. the number of input images. Our approach is built on the key insight that this bottleneck stems from the varying-length Key-Value (KV) space representation of scene geometry, which we distill into a fixed-size Multi-Layer Perceptron (MLP) via test-time training. VGG-T$^3$ (Visual Geometry Grounded Test Time Training) scales linearly w.r.t. the number of input views, similar to online models, and reconstructs a $1k$ image collection in just $54$ seconds, achieving a $11.6\\times$ speed-up over baselines that rely on softmax attention. Since our method retains global scene aggregation capability, our point map reconstruction error outperforming other linear-time methods by large margins. Finally, we demonstrate visual localization capabilities of our model by querying the scene representation with unseen images.", "date": "2026-02-26", "categories": []}
{"id": "2602.21535", "title": "Pseudo-View Enhancement via Confidence Fusion for Unposed Sparse-View Reconstruction", "abstract": "3D scene reconstruction under unposed sparse viewpoints is a highly challenging yet practically important problem, especially in outdoor scenes due to complex lighting and scale variation. With extremely limited input views, directly utilizing diffusion model to synthesize pseudo frames will introduce unreasonable geometry, which will harm the final reconstruction quality. To address these issues, we propose a novel framework for sparse-view outdoor reconstruction that achieves high-quality results through bidirectional pseudo frame restoration and scene perception Gaussian management. Specifically, we introduce a bidirectional pseudo frame restoration method that restores missing content by diffusion-based synthesis guided by adjacent frames with a lightweight pseudo-view deblur model and confidence mask inference algorithm. Then we propose a scene perception Gaussian management strategy that optimize Gaussians based on joint depth-density information. These designs significantly enhance reconstruction completeness, suppress floating artifacts and improve overall geometric consistency under extreme view sparsity. Experiments on outdoor benchmarks demonstrate substantial gains over existing methods in both fidelity and stability.", "date": "2026-02-25", "categories": []}
{"id": "2602.21645", "title": "Lie Flow: Video Dynamic Fields Modeling and Predicting with Lie Algebra as Geometric Physics Principle", "abstract": "Modeling 4D scenes requires capturing both spatial structure and temporal motion, which is challenging due to the need for physically consistent representations of complex rigid and non-rigid motions. Existing approaches mainly rely on translational displacements, which struggle to represent rotations, articulated transformations, often leading to spatial inconsistency and physically implausible motion. LieFlow, a dynamic radiance representation framework that explicitly models motion within the SE(3) Lie group, enabling coherent learning of translation and rotation in a unified geometric space. The SE(3) transformation field enforces physically inspired constraints to maintain motion continuity and geometric consistency. The evaluation includes a synthetic dataset with rigid-body trajectories and two real-world datasets capturing complex motion under natural lighting and occlusions. Across all datasets, LieFlow consistently improves view-synthesis fidelity, temporal coherence, and physical realism over NeRF-based baselines. These results confirm that SE(3)-based motion modeling offers a robust and physically grounded framework for representing dynamic 4D scenes.", "date": "2026-02-25", "categories": []}
{"id": "2602.21780", "title": "XStreamVGGT: Extremely Memory-Efficient Streaming Vision Geometry Grounded Transformer with KV Cache Compression", "abstract": "Learning-based 3D visual geometry models have significantly advanced with the advent of large-scale transformers. Among these, StreamVGGT leverages frame-wise causal attention to deliver robust and efficient streaming 3D reconstruction. However, it suffers from unbounded growth in the Key-Value (KV) cache due to the massive influx of vision tokens from multi-image and long-video inputs, leading to increased memory consumption and inference latency as input frames accumulate. This ultimately limits its scalability for long-horizon applications. To address this gap, we propose XStreamVGGT, a tuning-free approach that seamlessly integrates pruning and quantization to systematically compress the KV cache, enabling extremely memory-efficient streaming inference. Specifically, redundant KVs generated from multi-frame inputs are initially pruned to conform to a fixed KV memory budget using an efficient token-importance identification mechanism that maintains full compatibility with high-performance attention kernels (e.g., FlashAttention). Additionally, leveraging the inherent distribution patterns of KV tensors, we apply dimension-adaptive KV quantization within the pruning pipeline to further minimize memory overhead while preserving numerical accuracy. Extensive evaluations show that XStreamVGGT achieves mostly negligible performance degradation while substantially reducing memory usage by 4.42$\\times$ and accelerating inference by 5.48$\\times$, enabling practical and scalable streaming 3D applications. The code is available at https://github.com/ywh187/XStreamVGGT/.", "date": "2026-02-25", "categories": []}
{"id": "2506.14856", "title": "Peering into the Unknown: Active View Selection with Neural Uncertainty Maps for 3D Reconstruction", "abstract": "Some perspectives naturally provide more information than others. How can an AI system determine which viewpoint offers the most valuable insight for accurate and efficient 3D object reconstruction? Active view selection (AVS) for 3D reconstruction remains a fundamental challenge in computer vision. The aim is to identify the minimal set of views that yields the most accurate 3D reconstruction. Instead of learning radiance fields, like NeRF or 3D Gaussian Splatting, from a current observation and computing uncertainty for each candidate viewpoint, we introduce a novel AVS approach guided by neural uncertainty maps predicted by a lightweight feedforward deep neural network, named UPNet. UPNet takes a single input image of a 3D object and outputs a predicted uncertainty map, representing uncertainty values across all possible candidate viewpoints. By leveraging heuristics derived from observing many natural objects and their associated uncertainty patterns, we train UPNet to learn a direct mapping from viewpoint appearance to uncertainty in the underlying volumetric representations. Next, our approach aggregates all previously predicted neural uncertainty maps to suppress redundant candidate viewpoints and effectively select the most informative one. Using these selected viewpoints, we train 3D neural rendering models and evaluate the quality of novel view synthesis against other competitive AVS methods. Remarkably, despite using half of the viewpoints than the upper bound, our method achieves comparable reconstruction accuracy. In addition, it significantly reduces computational overhead during AVS, achieving up to a 400 times speedup along with over 50\\% reductions in CPU, RAM, and GPU usage compared to baseline methods. Notably, our approach generalizes effectively to AVS tasks involving novel object categories, without requiring any additional training.", "date": "2025-06-17", "categories": []}
{"id": "2511.16030", "title": "CuriGS: Curriculum-Guided Gaussian Splatting for Sparse View Synthesis", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as an efficient, high-fidelity representation for real-time scene reconstruction and rendering. However, extending 3DGS to sparse-view settings remains challenging because of supervision scarcity and overfitting caused by limited viewpoint coverage. In this paper, we present CuriGS, a curriculum-guided framework for sparse-view 3D reconstruction using 3DGS. CuriGS addresses the core challenge of sparse-view synthesis by introducing student views: pseudo-views sampled around ground-truth poses (teacher). For each teacher, we generate multiple groups of student views with different perturbation levels. During training, we follow a curriculum schedule that gradually unlocks higher perturbation level, randomly sampling candidate students from the active level to assist training. Each sampled student is regularized via depth-correlation and co-regularization, and evaluated using a multi-signal metric that combines SSIM, LPIPS, and an image-quality measure. For every teacher and perturbation level, we periodically retain the best-performing students and promote those that satisfy a predefined quality threshold to the training set, resulting in a stable augmentation of sparse training views. Experimental results show that CuriGS outperforms state-of-the-art baselines in both rendering fidelity and geometric consistency across various synthetic and real sparse-view scenes. Project page: https://zijian1026.github.io/CuriGS/", "date": "2025-11-20", "categories": []}
{"id": "2602.20342", "title": "Large-scale Photorealistic Outdoor 3D Scene Reconstruction from UAV Imagery Using Gaussian Splatting Techniques", "abstract": "In this study, we present an end-to-end pipeline capable of converting drone-captured video streams into high-fidelity 3D reconstructions with minimal latency. Unmanned aerial vehicles (UAVs) are extensively used in aerial real-time perception applications. Moreover, recent advances in 3D Gaussian Splatting (3DGS) have demonstrated significant potential for real-time neural rendering. However, their integration into end-to-end UAV-based reconstruction and visualization systems remains underexplored. Our goal is to propose an efficient architecture that combines live video acquisition via RTMP streaming, synchronized sensor fusion, camera pose estimation, and 3DGS optimization, achieving continuous model updates and low-latency deployment within interactive visualization environments that supports immersive augmented and virtual reality (AR/VR) applications. Experimental results demonstrate that the proposed method achieves competitive visual fidelity, while delivering significantly higher rendering performance and substantially reduced end-to-end latency, compared to NeRF-based approaches. Reconstruction quality remains within 4-7\\% of high-fidelity offline references, confirming the suitability of the proposed system for real-time, scalable augmented perception from aerial platforms.", "date": "2026-02-23", "categories": []}
{"id": "2602.20584", "title": "Long-Term Multi-Session 3D Reconstruction Under Substantial Appearance Change", "abstract": "Long-term environmental monitoring requires the ability to reconstruct and align 3D models across repeated site visits separated by months or years. However, existing Structure-from-Motion (SfM) pipelines implicitly assume near-simultaneous image capture and limited appearance change, and therefore fail when applied to long-term monitoring scenarios such as coral reef surveys, where substantial visual and structural change is common. In this paper, we show that the primary limitation of current approaches lies in their reliance on post-hoc alignment of independently reconstructed sessions, which is insufficient under large temporal appearance change. We address this limitation by enforcing cross-session correspondences directly within a joint SfM reconstruction. Our approach combines complementary handcrafted and learned visual features to robustly establish correspondences across large temporal gaps, enabling the reconstruction of a single coherent 3D model from imagery captured years apart, where standard independent and joint SfM pipelines break down. We evaluate our method on long-term coral reef datasets exhibiting significant real-world change, and demonstrate consistent joint reconstruction across sessions in cases where existing methods fail to produce coherent reconstructions. To ensure scalability to large datasets, we further restrict expensive learned feature matching to a small set of likely cross-session image pairs identified via visual place recognition, which reduces computational cost and improves alignment robustness.", "date": "2026-02-24", "categories": []}
{"id": "2602.20933", "title": "Dropping Anchor and Spherical Harmonics for Sparse-view Gaussian Splatting", "abstract": "Recent 3D Gaussian Splatting (3DGS) Dropout methods address overfitting under sparse-view conditions by randomly nullifying Gaussian opacities. However, we identify a neighbor compensation effect in these approaches: dropped Gaussians are often compensated by their neighbors, weakening the intended regularization. Moreover, these methods overlook the contribution of high-degree spherical harmonic coefficients (SH) to overfitting. To address these issues, we propose DropAnSH-GS, a novel anchor-based Dropout strategy. Rather than dropping Gaussians independently, our method randomly selects certain Gaussians as anchors and simultaneously removes their spatial neighbors. This effectively disrupts local redundancies near anchors and encourages the model to learn more robust, globally informed representations. Furthermore, we extend the Dropout to color attributes by randomly dropping higher-degree SH to concentrate appearance information in lower-degree SH. This strategy further mitigates overfitting and enables flexible post-training model compression via SH truncation. Experimental results demonstrate that DropAnSH-GS substantially outperforms existing Dropout methods with negligible computational overhead, and can be readily integrated into various 3DGS variants to enhance their performances. Project Website: https://sk-fun.fun/DropAnSH-GS", "date": "2026-02-24", "categories": []}
{"id": "2602.20943", "title": "UFO: Unifying Feed-Forward and Optimization-based Methods for Large Driving Scene Modeling", "abstract": "Dynamic driving scene reconstruction is critical for autonomous driving simulation and closed-loop learning. While recent feed-forward methods have shown promise for 3D reconstruction, they struggle with long-range driving sequences due to quadratic complexity in sequence length and challenges in modeling dynamic objects over extended durations. We propose UFO, a novel recurrent paradigm that combines the benefits of optimization-based and feed-forward methods for efficient long-range 4D reconstruction. Our approach maintains a 4D scene representation that is iteratively refined as new observations arrive, using a visibility-based filtering mechanism to select informative scene tokens and enable efficient processing of long sequences. For dynamic objects, we introduce an object pose-guided modeling approach that supports accurate long-range motion capture. Experiments on the Waymo Open Dataset demonstrate that our method significantly outperforms both per-scene optimization and existing feed-forward methods across various sequence lengths. Notably, our approach can reconstruct 16-second driving logs within 0.5 second while maintaining superior visual quality and geometric accuracy.", "date": "2026-02-24", "categories": []}
{"id": "2408.10135", "title": "$R^2$-Mesh: Reinforcement Learning Powered Mesh Reconstruction via Geometry and Appearance Refinement", "abstract": "Mesh reconstruction from Neural Radiance Fields (NeRF) is widely used in 3D reconstruction and has been applied across numerous domains. However, existing methods typically rely solely on the given training set images, which restricts supervision to limited observations and makes it difficult to fully constrain geometry and appearance. Moreover, the contribution of each viewpoint for training is not uniform and changes dynamically during the optimization process, which can result in suboptimal guidance for both geometric refinement and rendering quality. To address these limitations, we propose $R^2$-Mesh, a reinforcement learning framework that combines NeRF-rendered pseudo-supervision with online viewpoint selection. Our key insight is to exploit NeRF's rendering ability to synthesize additional high-quality images, enriching training with diverse viewpoint information. To ensure that supervision focuses on the most beneficial perspectives, we introduce a UCB-based strategy with a geometry-aware reward, which dynamically balances exploration and exploitation to identify informative viewpoints throughout training. Within this framework, we jointly optimize SDF geometry and view-dependent appearance under differentiable rendering, while periodically refining meshes to capture fine geometric details. Experiments demonstrate that our method achieves competitive results in both geometric accuracy and rendering quality.", "date": "2024-08-19", "categories": []}
{"id": "2602.19323", "title": "DefenseSplat: Enhancing the Robustness of 3D Gaussian Splatting via Frequency-Aware Filtering", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for real-time and high-fidelity 3D reconstruction from posed images. However, recent studies reveal its vulnerability to adversarial corruptions in input views, where imperceptible yet consistent perturbations can drastically degrade rendering quality, increase training and rendering time, and inflate memory usage, even leading to server denial-of-service. In our work, to mitigate this issue, we begin by analyzing the distinct behaviors of adversarial perturbations in the low- and high-frequency components of input images using wavelet transforms. Based on this observation, we design a simple yet effective frequency-aware defense strategy that reconstructs training views by filtering high-frequency noise while preserving low-frequency content. This approach effectively suppresses adversarial artifacts while maintaining the authenticity of the original scene. Notably, it does not significantly impair training on clean data, achieving a desirable trade-off between robustness and performance on clean inputs. Through extensive experiments under a wide range of attack intensities on multiple benchmarks, we demonstrate that our method substantially enhances the robustness of 3DGS without access to clean ground-truth supervision. By highlighting and addressing the overlooked vulnerabilities of 3D Gaussian Splatting, our work paves the way for more robust and secure 3D reconstructions.", "date": "2026-02-22", "categories": []}
{"id": "2602.19697", "title": "BayesFusion-SDF: Probabilistic Signed Distance Fusion with View Planning on CPU", "abstract": "Key part of robotics, augmented reality, and digital inspection is dense 3D reconstruction from depth observations. Traditional volumetric fusion techniques, including truncated signed distance functions (TSDF), enable efficient and deterministic geometry reconstruction; however, they depend on heuristic weighting and fail to transparently convey uncertainty in a systematic way. Recent neural implicit methods, on the other hand, get very high fidelity but usually need a lot of GPU power for optimization and aren't very easy to understand for making decisions later on. This work presents BayesFusion-SDF, a CPU-centric probabilistic signed distance fusion framework that conceptualizes geometry as a sparse Gaussian random field with a defined posterior distribution over voxel distances. First, a rough TSDF reconstruction is used to create an adaptive narrow-band domain. Then, depth observations are combined using a heteroscedastic Bayesian formulation that is solved using sparse linear algebra and preconditioned conjugate gradients. Randomized diagonal estimators are a quick way to get an idea of posterior uncertainty. This makes it possible to extract surfaces and plan the next best view while taking into account uncertainty. Tests on a controlled ablation scene and a CO3D object sequence show that the new method is more accurate geometrically than TSDF baselines and gives useful estimates of uncertainty for active sensing. The proposed formulation provides a clear and easy-to-use alternative to GPU-heavy neural reconstruction methods while still being able to be understood in a probabilistic way and acting in a predictable way. GitHub: https://mazumdarsoumya.github.io/BayesFusionSDF", "date": "2026-02-23", "categories": []}
{"id": "2602.20157", "title": "Flow3r: Factored Flow Prediction for Scalable Visual Geometry Learning", "abstract": "Current feed-forward 3D/4D reconstruction systems rely on dense geometry and pose supervision -- expensive to obtain at scale and particularly scarce for dynamic real-world scenes. We present Flow3r, a framework that augments visual geometry learning with dense 2D correspondences (`flow') as supervision, enabling scalable training from unlabeled monocular videos. Our key insight is that the flow prediction module should be factored: predicting flow between two images using geometry latents from one and pose latents from the other. This factorization directly guides the learning of both scene geometry and camera motion, and naturally extends to dynamic scenes. In controlled experiments, we show that factored flow prediction outperforms alternative designs and that performance scales consistently with unlabeled data. Integrating factored flow into existing visual geometry architectures and training with ${\\sim}800$K unlabeled videos, Flow3r achieves state-of-the-art results across eight benchmarks spanning static and dynamic scenes, with its largest gains on in-the-wild dynamic videos where labeled data is most scarce.", "date": "2026-02-23", "categories": []}
{"id": "2602.18258", "title": "RoEL: Robust Event-based 3D Line Reconstruction", "abstract": "Event cameras in motion tend to detect object boundaries or texture edges, which produce lines of brightness changes, especially in man-made environments. While lines can constitute a robust intermediate representation that is consistently observed, the sparse nature of lines may lead to drastic deterioration with minor estimation errors. Only a few previous works, often accompanied by additional sensors, utilize lines to compensate for the severe domain discrepancies of event sensors along with unpredictable noise characteristics. We propose a method that can stably extract tracks of varying appearances of lines using a clever algorithmic process that observes multiple representations from various time slices of events, compensating for potential adversaries within the event data. We then propose geometric cost functions that can refine the 3D line maps and camera poses, eliminating projective distortions and depth ambiguities. The 3D line maps are highly compact and can be equipped with our proposed cost function, which can be adapted for any observations that can detect and extract line structures or projections of them, including 3D point cloud maps or image observations. We demonstrate that our formulation is powerful enough to exhibit a significant performance boost in event-based mapping and pose refinement across diverse datasets, and can be flexibly applied to multimodal scenarios. Our results confirm that the proposed line-based formulation is a robust and effective approach for the practical deployment of event-based perceptual modules. Project page: https://gwangtak.github.io/roel/", "date": "2026-02-20", "categories": []}
{"id": "2602.18314", "title": "Diff2DGS: Reliable Reconstruction of Occluded Surgical Scenes via 2D Gaussian Splatting", "abstract": "Real-time reconstruction of deformable surgical scenes is vital for advancing robotic surgery, improving surgeon guidance, and enabling automation. Recent methods achieve dense reconstructions from da Vinci robotic surgery videos, with Gaussian Splatting (GS) offering real-time performance via graphics acceleration. However, reconstruction quality in occluded regions remains limited, and depth accuracy has not been fully assessed, as benchmarks like EndoNeRF and StereoMIS lack 3D ground truth. We propose Diff2DGS, a novel two-stage framework for reliable 3D reconstruction of occluded surgical scenes. In the first stage, a diffusion-based video module with temporal priors inpaints tissue occluded by instruments with high spatial-temporal consistency. In the second stage, we adapt 2D Gaussian Splatting (2DGS) with a Learnable Deformation Model (LDM) to capture dynamic tissue deformation and anatomical geometry. We also extend evaluation beyond prior image-quality metrics by performing quantitative depth accuracy analysis on the SCARED dataset. Diff2DGS outperforms state-of-the-art approaches in both appearance and geometry, reaching 38.02 dB PSNR on EndoNeRF and 34.40 dB on StereoMIS. Furthermore, our experiments demonstrate that optimizing for image quality alone does not necessarily translate into optimal 3D reconstruction accuracy. To address this, we further optimize the depth quality of the reconstructed 3D results, ensuring more faithful geometry in addition to high-fidelity appearance.", "date": "2026-02-20", "categories": []}
{"id": "2412.02039", "title": "Multi-View 3D Reconstruction using Knowledge Distillation", "abstract": "Large Foundation Models like Dust3r can produce high quality outputs such as pointmaps, camera intrinsics, and depth estimation, given stereo-image pairs as input. However, the application of these outputs on tasks like Visual Localization requires a large amount of inference time and compute resources. To address these limitations, in this paper, we propose the use of a knowledge distillation pipeline, where we aim to build a student-teacher model with Dust3r as the teacher and explore multiple architectures of student models that are trained using the 3D reconstructed points output by Dust3r. Our goal is to build student models that can learn scene-specific representations and output 3D points with replicable performance such as Dust3r. The data set we used to train our models is 12Scenes. We test two main architectures of models: a CNN-based architecture and a Vision Transformer based architecture. For each architecture, we also compare the use of pre-trained models against models built from scratch. We qualitatively compare the reconstructed 3D points output by the student model against Dust3r's and discuss the various features learned by the student model. We also perform ablation studies on the models through hyperparameter tuning. Overall, we observe that the Vision Transformer presents the best performance visually and quantitatively.", "date": "2024-12-02", "categories": []}
{"id": "2602.16713", "title": "Three-dimensional Damage Visualization of Civil Structures via Gaussian Splatting-enabled Digital Twins", "abstract": "Recent advancements in civil infrastructure inspections underscore the need for precise three-dimensional (3D) damage visualization on digital twins, transcending traditional 2D image-based damage identifications. Compared to conventional photogrammetric 3D reconstruction techniques, modern approaches such as Neural Radiance Field (NeRF) and Gaussian Splatting (GS) excel in scene representation, rendering quality, and handling featureless regions. Among them, GS stands out for its efficiency, leveraging discrete anisotropic 3D Gaussians to represent radiance fields, unlike NeRF's continuous implicit model. This study introduces a GS-enabled digital twin method tailored for effective 3D damage visualization. The method's key contributions include: 1) utilizing GS-based 3D reconstruction to visualize 2D damage segmentation results while reducing segmentation errors; 2) developing a multi-scale reconstruction strategy to balance efficiency and damage detail; 3) enabling digital twin updates as damage evolves over time. Demonstrated on an open-source synthetic dataset for post-earthquake inspections, the proposed approach offers a promising solution for comprehensive 3D damage visualization in civil infrastructure digital twins.", "date": "2026-01-23", "categories": []}
{"id": "2602.16950", "title": "HS-3D-NeRF: 3D Surface and Hyperspectral Reconstruction From Stationary Hyperspectral Images Using Multi-Channel NeRFs", "abstract": "Advances in hyperspectral imaging (HSI) and 3D reconstruction have enabled accurate, high-throughput characterization of agricultural produce quality and plant phenotypes, both essential for advancing agricultural sustainability and breeding programs. HSI captures detailed biochemical features of produce, while 3D geometric data substantially improves morphological analysis. However, integrating these two modalities at scale remains challenging, as conventional approaches involve complex hardware setups incompatible with automated phenotyping systems. Recent advances in neural radiance fields (NeRF) offer computationally efficient 3D reconstruction but typically require moving-camera setups, limiting throughput and reproducibility in standard indoor agricultural environments. To address these challenges, we introduce HSI-SC-NeRF, a stationary-camera multi-channel NeRF framework for high-throughput hyperspectral 3D reconstruction targeting postharvest inspection of agricultural produce. Multi-view hyperspectral data is captured using a stationary camera while the object rotates within a custom-built Teflon imaging chamber providing diffuse, uniform illumination. Object poses are estimated via ArUco calibration markers and transformed to the camera frame of reference through simulated pose transformations, enabling standard NeRF training on stationary-camera data. A multi-channel NeRF formulation optimizes reconstruction across all hyperspectral bands jointly using a composite spectral loss, supported by a two-stage training protocol that decouples geometric initialization from radiometric refinement. Experiments on three agricultural produce samples demonstrate high spatial reconstruction accuracy and strong spectral fidelity across the visible and near-infrared spectrum, confirming the suitability of HSI-SC-NeRF for integration into automated agricultural workflows.", "date": "2026-02-18", "categories": []}
{"id": "2602.17124", "title": "3D Scene Rendering with Multimodal Gaussian Splatting", "abstract": "3D scene reconstruction and rendering are core tasks in computer vision, with applications spanning industrial monitoring, robotics, and autonomous driving. Recent advances in 3D Gaussian Splatting (GS) and its variants have achieved impressive rendering fidelity while maintaining high computational and memory efficiency. However, conventional vision-based GS pipelines typically rely on a sufficient number of camera views to initialize the Gaussian primitives and train their parameters, typically incurring additional processing cost during initialization while falling short in conditions where visual cues are unreliable, such as adverse weather, low illumination, or partial occlusions. To cope with these challenges, and motivated by the robustness of radio-frequency (RF) signals to weather, lighting, and occlusions, we introduce a multimodal framework that integrates RF sensing, such as automotive radar, with GS-based rendering as a more efficient and robust alternative to vision-only GS rendering. The proposed approach enables efficient depth prediction from only sparse RF-based depth measurements, yielding a high-quality 3D point cloud for initializing Gaussian functions across diverse GS architectures. Numerical tests demonstrate the merits of judiciously incorporating RF sensing into GS pipelines, achieving high-fidelity 3D scene rendering driven by RF-informed structural accuracy.", "date": "2026-02-19", "categories": []}
{"id": "2412.11762", "title": "GS-ProCams: Gaussian Splatting-based Projector-Camera Systems", "abstract": "We present GS-ProCams, the first Gaussian Splatting-based framework for projector-camera systems (ProCams). GS-ProCams is not only view-agnostic but also significantly enhances the efficiency of projection mapping (PM) that requires establishing geometric and radiometric mappings between the projector and the camera. Previous CNN-based ProCams are constrained to a specific viewpoint, limiting their applicability to novel perspectives. In contrast, NeRF-based ProCams support view-agnostic projection mapping, however, they require an additional co-located light source and demand significant computational and memory resources. To address this issue, we propose GS-ProCams that employs 2D Gaussian for scene representations, and enables efficient view-agnostic ProCams applications. In particular, we explicitly model the complex geometric and photometric mappings of ProCams using projector responses, the projection surface's geometry and materials represented by Gaussians, and the global illumination component. Then, we employ differentiable physically-based rendering to jointly estimate them from captured multi-view projections. Compared to state-of-the-art NeRF-based methods, our GS-ProCams eliminates the need for additional devices, achieving superior ProCams simulation quality. It also uses only 1/10 of the GPU memory for training and is 900 times faster in inference speed. Please refer to our project page for the code and dataset: https://realqingyue.github.io/GS-ProCams/.", "date": "2024-12-16", "categories": []}
{"id": "2501.12369", "title": "DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions", "abstract": "Splatting-based 3D reconstruction methods have gained popularity with the advent of 3D Gaussian Splatting, efficiently synthesizing high-quality novel views. These methods commonly resort to using exponential family functions, such as the Gaussian function, as reconstruction kernels due to their anisotropic nature, ease of projection, and differentiability in rasterization. However, the field remains restricted to variations within the exponential family, leaving generalized reconstruction kernels largely underexplored, partly due to the lack of easy integrability in 3D to 2D projections. In this light, we show that a class of decaying anisotropic radial basis functions (DARBFs), which are non-negative functions of the Mahalanobis distance, supports splatting by approximating the Gaussian function's closed-form integration advantage. With this fresh perspective, we demonstrate varying performances across selected DARB reconstruction kernels, achieving comparable training convergence and memory footprints, with on-par PSNR, SSIM, and LPIPS results.", "date": "2025-01-21", "categories": []}
{"id": "2504.13159", "title": "Digital Twin Generation from Visual Data: A Survey", "abstract": "This survey examines recent advances in generating digital twins from visual data. These digital twins - virtual 3D replicas of physical assets - can be applied to robotics, media content creation, design or construction workflows. We analyze a range of approaches, including 3D Gaussian Splatting, generative inpainting, semantic segmentation, and foundation models, highlighting their respective advantages and limitations. In addition, we discuss key challenges such as occlusions, lighting variations, and scalability, as well as identify gaps, trends, and directions for future research. Overall, this survey aims to provide a comprehensive overview of state-of-the-art methodologies and their implications for real-world applications. Awesome Digital Twin: https://awesomedigitaltwin.github.io", "date": "2025-04-17", "categories": []}
{"id": "2507.14841", "title": "Towards Geometric and Textural Consistency 3D Scene Generation via Single Image-guided Model Generation and Layout Optimization", "abstract": "In recent years, 3D generation has made great strides in both academia and industry. However, generating 3D scenes from a single RGB image remains a significant challenge, as current approaches often struggle to ensure both object generation quality and scene coherence in multi-object scenarios. To overcome these limitations, we propose a novel three-stage framework for 3D scene generation with explicit geometric representations and high-quality textural details via single image-guided model generation and spatial layout optimization. Our method begins with an image instance segmentation and inpainting phase, which recovers missing details of occluded objects in the input images, thereby achieving complete generation of foreground 3D assets. Subsequently, our approach captures the spatial geometry of reference image by constructing pseudo-stereo viewpoint for camera parameter estimation and scene depth inference, while employing a model selection strategy to ensure optimal alignment between the 3D assets generated in the previous step and the input. Finally, through model parameterization and minimization of the Chamfer distance between point clouds in 3D and 2D space, our approach optimizes layout parameters to produce an explicit 3D scene representation that maintains precise alignment with input guidance image. Extensive experiments on multi-object scene image sets have demonstrated that our approach not only outperforms state-of-the-art methods in terms of geometric accuracy and texture fidelity of individual generated 3D models, but also has significant advantages in scene layout synthesis.", "date": "2025-07-20", "categories": []}
{"id": "2509.23607", "title": "ZeroScene: A Zero-Shot Framework for 3D Scene Generation from a Single Image and Controllable Texture Editing", "abstract": "In the field of 3D content generation, single image scene reconstruction methods still struggle to simultaneously ensure the quality of individual assets and the coherence of the overall scene in complex environments, while texture editing techniques often fail to maintain both local continuity and multi-view consistency. In this paper, we propose a novel system ZeroScene, which leverages the prior knowledge of large vision models to accomplish both single image-to-3D scene reconstruction and texture editing in a zero-shot manner. ZeroScene extracts object-level 2D segmentation and depth information from input images to infer spatial relationships within the scene. It then jointly optimizes 3D and 2D projection losses of the point cloud to update object poses for precise scene alignment, ultimately constructing a coherent and complete 3D scene that encompasses both foreground and background. Moreover, ZeroScene supports texture editing of objects in the scene. By imposing constraints on the diffusion model and introducing a mask-guided progressive image generation strategy, we effectively maintain texture consistency across multiple viewpoints and further enhance the realism of rendered results through Physically Based Rendering (PBR) material estimation. Experimental results demonstrate that our framework not only ensures the geometric and appearance accuracy of generated assets, but also faithfully reconstructs scene layouts and produces highly detailed textures that closely align with text prompts.", "date": "2025-09-28", "categories": []}
{"id": "2601.22615", "title": "TTSA3R: Training-Free Temporal-Spatial Adaptive Persistent State for Streaming 3D Reconstruction", "abstract": "Streaming recurrent models enable efficient 3D reconstruction by maintaining persistent state representations. However, they suffer from catastrophic forgetting over long sequences due to balancing historical information with new observations. Recent methods alleviate this by deriving adaptive signals from attention perspective, but they operate on single dimensions without considering temporal and spatial consistency. To this end, we propose a training-free framework termed TTSA3R that leverages both temporal state evolution and spatial observation quality for adaptive state updates in 3D reconstruction. In particular, we devise a Temporal Adaptive Update Module that regulates update magnitude by analyzing temporal state evolution patterns. Then, a Spatial Contextual Update Module is introduced to localize spatial regions that require updates through observation-state alignment and scene dynamics. These complementary signals are finally fused to determine the state updating strategies. Extensive experiments demonstrate the effectiveness of TTSA3R in diverse 3D tasks. Moreover, our method exhibits only 1.33x error increase compared to over 4x degradation in the baseline model on extended sequences of 3D reconstruction, significantly improving long-term reconstruction stability. Our codes are available at https://github.com/anonus2357/ttsa3r.", "date": "2026-01-30", "categories": []}
{"id": "2602.15516", "title": "Semantic-Guided 3D Gaussian Splatting for Transient Object Removal", "abstract": "Transient objects in casual multi-view captures cause ghosting artifacts in 3D Gaussian Splatting (3DGS) reconstruction. Existing solutions relied on scene decomposition at significant memory cost or on motion-based heuristics that were vulnerable to parallax ambiguity. A semantic filtering framework was proposed for category-aware transient removal using vision-language models. CLIP similarity scores between rendered views and distractor text prompts were accumulated per-Gaussian across training iterations. Gaussians exceeding a calibrated threshold underwent opacity regularization and periodic pruning. Unlike motion-based approaches, semantic classification resolved parallax ambiguity by identifying object categories independently of motion patterns. Experiments on the RobustNeRF benchmark demonstrated consistent improvement in reconstruction quality over vanilla 3DGS across four sequences, while maintaining minimal memory overhead and real-time rendering performance. Threshold calibration and comparisons with baselines validated semantic guidance as a practical strategy for transient removal in scenarios with predictable distractor categories.", "date": "2026-02-17", "categories": []}
{"id": "2602.15734", "title": "Language and Geometry Grounded Sparse Voxel Representations for Holistic Scene Understanding", "abstract": "Existing 3D open-vocabulary scene understanding methods mostly emphasize distilling language features from 2D foundation models into 3D feature fields, but largely overlook the synergy among scene appearance, semantics, and geometry. As a result, scene understanding often deviates from the underlying geometric structure of scenes and becomes decoupled from the reconstruction process. In this work, we propose a novel approach that leverages language and geometry grounded sparse voxel representations to comprehensively model appearance, semantics, and geometry within a unified framework. Specifically, we use 3D sparse voxels as primitives and employ an appearance field, a density field, a feature field, and a confidence field to holistically represent a 3D scene. To promote synergy among the appearance, density, and feature fields, we construct a feature modulation module and distill language features from a 2D foundation model into our 3D scene model. In addition, we integrate geometric distillation into feature field distillation to transfer geometric knowledge from a geometry foundation model to our 3D scene representations via depth correlation regularization and pattern consistency regularization. These components work together to synergistically model the appearance, semantics, and geometry of the 3D scene within a unified framework. Extensive experiments demonstrate that our approach achieves superior overall performance compared with state-of-the-art methods in holistic scene understanding and reconstruction.", "date": "2026-02-17", "categories": []}
{"id": "2602.13549", "title": "Nighttime Autonomous Driving Scene Reconstruction with Physically-Based Gaussian Splatting", "abstract": "This paper focuses on scene reconstruction under nighttime conditions in autonomous driving simulation. Recent methods based on Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) have achieved photorealistic modeling in autonomous driving scene reconstruction, but they primarily focus on normal-light conditions. Low-light driving scenes are more challenging to model due to their complex lighting and appearance conditions, which often causes performance degradation of existing methods. To address this problem, this work presents a novel approach that integrates physically based rendering into 3DGS to enhance nighttime scene reconstruction for autonomous driving. Specifically, our approach integrates physically based rendering into composite scene Gaussian representations and jointly optimizes Bidirectional Reflectance Distribution Function (BRDF) based material properties. We explicitly model diffuse components through a global illumination module and specular components by anisotropic spherical Gaussians. As a result, our approach improves reconstruction quality for outdoor nighttime driving scenes, while maintaining real-time rendering. Extensive experiments across diverse nighttime scenarios on two real-world autonomous driving datasets, including nuScenes and Waymo, demonstrate that our approach outperforms the state-of-the-art methods both quantitatively and qualitatively.", "date": "2026-02-14", "categories": []}
{"id": "2602.13801", "title": "Joint Orientation and Weight Optimization for Robust Watertight Surface Reconstruction via Dirichlet-Regularized Winding Fields", "abstract": "We propose Dirichlet Winding Reconstruction (DiWR), a robust method for reconstructing watertight surfaces from unoriented point clouds with non-uniform sampling, noise, and outliers. Our method uses the generalized winding number (GWN) field as the target implicit representation and jointly optimizes point orientations, per-point area weights, and confidence coefficients in a single pipeline. The optimization minimizes the Dirichlet energy of the induced winding field together with additional GWN-based constraints, allowing DiWR to compensate for non-uniform sampling, reduce the impact of noise, and downweight outliers during reconstruction, with no reliance on separate preprocessing. We evaluate DiWR on point clouds from 3D Gaussian Splatting, a computer-vision pipeline, and corrupted graphics benchmarks. Experiments show that DiWR produces plausible watertight surfaces on these challenging inputs and outperforms both traditional multi-stage pipelines and recent joint orientation-reconstruction methods.", "date": "2026-02-14", "categories": []}
{"id": "2602.13806", "title": "Gaussian Sequences with Multi-Scale Dynamics for 4D Reconstruction from Monocular Casual Videos", "abstract": "Understanding dynamic scenes from casual videos is critical for scalable robot learning, yet four-dimensional (4D) reconstruction under strictly monocular settings remains highly ill-posed. To address this challenge, our key insight is that real-world dynamics exhibits a multi-scale regularity from object to particle level. To this end, we design the multi-scale dynamics mechanism that factorizes complex motion fields. Within this formulation, we propose Gaussian sequences with multi-scale dynamics, a novel representation for dynamic 3D Gaussians derived through compositions of multi-level motion. This layered structure substantially alleviates ambiguity of reconstruction and promotes physically plausible dynamics. We further incorporate multi-modal priors from vision foundation models to establish complementary supervision, constraining the solution space and improving the reconstruction fidelity. Our approach enables accurate and globally consistent 4D reconstruction from monocular casual videos. Experiments of dynamic novel-view synthesis (NVS) on benchmark and real-world manipulation datasets demonstrate considerable improvements over existing methods.", "date": "2026-02-14", "categories": []}
{"id": "2602.13909", "title": "High-fidelity 3D reconstruction for planetary exploration", "abstract": "Planetary exploration increasingly relies on autonomous robotic systems capable of perceiving, interpreting, and reconstructing their surroundings in the absence of global positioning or real-time communication with Earth. Rovers operating on planetary surfaces must navigate under sever environmental constraints, limited visual redundancy, and communication delays, making onboard spatial awareness and visual localization key components for mission success. Traditional techniques based on Structure-from-Motion (SfM) and Simultaneous Localization and Mapping (SLAM) provide geometric consistency but struggle to capture radiometric detail or to scale efficiently in unstructured, low-texture terrains typical of extraterrestrial environments. This work explores the integration of radiance field-based methods - specifically Neural Radiance Fields (NeRF) and Gaussian Splatting - into a unified, automated environment reconstruction pipeline for planetary robotics. Our system combines the Nerfstudio and COLMAP frameworks with a ROS2-compatible workflow capable of processing raw rover data directly from rosbag recordings. This approach enables the generation of dense, photorealistic, and metrically consistent 3D representations from minimal visual input, supporting improved perception and planning for autonomous systems operating in planetary-like conditions. The resulting pipeline established a foundation for future research in radiance field-based mapping, bridging the gap between geometric and neural representations in planetary exploration.", "date": "2026-02-14", "categories": []}
{"id": "2602.14021", "title": "Flow4R: Unifying 4D Reconstruction and Tracking with Scene Flow", "abstract": "Reconstructing and tracking dynamic 3D scenes remains a fundamental challenge in computer vision. Existing approaches often decouple geometry from motion: multi-view reconstruction methods assume static scenes, while dynamic tracking frameworks rely on explicit camera pose estimation or separate motion models. We propose Flow4R, a unified framework that treats camera-space scene flow as the central representation linking 3D structure, object motion, and camera motion. Flow4R predicts a minimal per-pixel property set-3D point position, scene flow, pose weight, and confidence-from two-view inputs using a Vision Transformer. This flow-centric formulation allows local geometry and bidirectional motion to be inferred symmetrically with a shared decoder in a single forward pass, without requiring explicit pose regressors or bundle adjustment. Trained jointly on static and dynamic datasets, Flow4R achieves state-of-the-art performance on 4D reconstruction and tracking tasks, demonstrating the effectiveness of the flow-central representation for spatiotemporal scene understanding.", "date": "2026-02-15", "categories": []}
{"id": "2602.14119", "title": "GeoFusionLRM: Geometry-Aware Self-Correction for Consistent 3D Reconstruction", "abstract": "Single-image 3D reconstruction with large reconstruction models (LRMs) has advanced rapidly, yet reconstructions often exhibit geometric inconsistencies and misaligned details that limit fidelity. We introduce GeoFusionLRM, a geometry-aware self-correction framework that leverages the model's own normal and depth predictions to refine structural accuracy. Unlike prior approaches that rely solely on features extracted from the input image, GeoFusionLRM feeds back geometric cues through a dedicated transformer and fusion module, enabling the model to correct errors and enforce consistency with the conditioning image. This design improves the alignment between the reconstructed mesh and the input views without additional supervision or external signals. Extensive experiments demonstrate that GeoFusionLRM achieves sharper geometry, more consistent normals, and higher fidelity than state-of-the-art LRM baselines.", "date": "2026-02-15", "categories": []}
{"id": "2602.04053", "title": "Seeing Through Clutter: Structured 3D Scene Reconstruction via Iterative Object Removal", "abstract": "We present SeeingThroughClutter, a method for reconstructing structured 3D representations from single images by segmenting and modeling objects individually. Prior approaches rely on intermediate tasks such as semantic segmentation and depth estimation, which often underperform in complex scenes, particularly in the presence of occlusion and clutter. We address this by introducing an iterative object removal and reconstruction pipeline that decomposes complex scenes into a sequence of simpler subtasks. Using VLMs as orchestrators, foreground objects are removed one at a time via detection, segmentation, object removal, and 3D fitting. We show that removing objects allows for cleaner segmentations of subsequent objects, even in highly occluded scenes. Our method requires no task-specific training and benefits directly from ongoing advances in foundation models. We demonstrate stateof-the-art robustness on 3D-Front and ADE20K datasets. Project Page: https://rioak.github.io/seeingthroughclutter/", "date": "2026-02-03", "categories": []}
{"id": "2602.12508", "title": "Monocular Reconstruction of Neural Tactile Fields", "abstract": "Robots operating in the real world must plan through environments that deform, yield, and reconfigure under contact, requiring interaction-aware 3D representations that extend beyond static geometric occupancy. To address this, we introduce neural tactile fields, a novel 3D representation that maps spatial locations to the expected tactile response upon contact. Our model predicts these neural tactile fields from a single monocular RGB image -- the first method to do so. When integrated with off-the-shelf path planners, neural tactile fields enable robots to generate paths that avoid high-resistance objects while deliberately routing through low-resistance regions (e.g. foliage), rather than treating all occupied space as equally impassable. Empirically, our learning framework improves volumetric 3D reconstruction by $85.8\\%$ and surface reconstruction by $26.7\\%$ compared to state-of-the-art monocular 3D reconstruction methods (LRM and Direct3D).", "date": "2026-02-13", "categories": []}
{"id": "2602.13041", "title": "Implicit-Scale 3D Reconstruction for Multi-Food Volume Estimation from Monocular Images", "abstract": "We present Implicit-Scale 3D Reconstruction from Monocular Multi-Food Images, a benchmark dataset designed to advance geometry-based food portion estimation in realistic dining scenarios. Existing dietary assessment methods largely rely on single-image analysis or appearance-based inference, including recent vision-language models, which lack explicit geometric reasoning and are sensitive to scale ambiguity. This benchmark reframes food portion estimation as an implicit-scale 3D reconstruction problem under monocular observations. To reflect real-world conditions, explicit physical references and metric annotations are removed; instead, contextual objects such as plates and utensils are provided, requiring algorithms to infer scale from implicit cues and prior knowledge. The dataset emphasizes multi-food scenes with diverse object geometries, frequent occlusions, and complex spatial arrangements. The benchmark was adopted as a challenge at the MetaFood 2025 Workshop, where multiple teams proposed reconstruction-based solutions. Experimental results show that while strong vision--language baselines achieve competitive performance, geometry-based reconstruction methods provide both improved accuracy and greater robustness, with the top-performing approach achieving 0.21 MAPE in volume estimation and 5.7 L1 Chamfer Distance in geometric accuracy.", "date": "2026-02-13", "categories": []}
{"id": "2601.21421", "title": "From Implicit Ambiguity to Explicit Solidity: Diagnosing Interior Geometric Degradation in Neural Radiance Fields for Dense 3D Scene Understanding", "abstract": "Neural Radiance Fields (NeRFs) have emerged as a powerful paradigm for multi-view reconstruction, complementing classical photogrammetric pipelines based on Structure-from-Motion (SfM) and Multi-View Stereo (MVS). However, their reliability for quantitative 3D analysis in dense, self-occluding scenes remains poorly understood. In this study, we identify a fundamental failure mode of implicit density fields under heavy occlusion, which we term Interior Geometric Degradation (IGD). We show that transmittance-based volumetric optimization satisfies photometric supervision by reconstructing hollow or fragmented structures rather than solid interiors, leading to systematic instance undercounting. Through controlled experiments on synthetic datasets with increasing occlusion, we demonstrate that state-of-the-art mask-supervised NeRFs saturate at approximately 89% instance recovery in dense scenes, despite improved surface coherence and mask quality. To overcome this limitation, we introduce an explicit geometric pipeline based on Sparse Voxel Rasterization (SVRaster), initialized from SfM feature geometry. By projecting 2D instance masks onto an explicit voxel grid and enforcing geometric separation via recursive splitting, our approach preserves physical solidity and achieves a 95.8% recovery rate in dense clusters. A sensitivity analysis using degraded segmentation masks further shows that explicit SfM-based geometry is substantially more robust to supervision failure, recovering 43% more instances than implicit baselines. These results demonstrate that explicit geometric priors are a prerequisite for reliable quantitative analysis in highly self-occluding 3D scenes.", "date": "2026-01-29", "categories": []}
{"id": "2602.11845", "title": "WorldTree: Towards 4D Dynamic Worlds from Monocular Video using Tree-Chains", "abstract": "Dynamic reconstruction has achieved remarkable progress, but there remain challenges in monocular input for more practical applications. The prevailing works attempt to construct efficient motion representations, but lack a unified spatiotemporal decomposition framework, suffering from either holistic temporal optimization or coupled hierarchical spatial composition. To this end, we propose WorldTree, a unified framework comprising Temporal Partition Tree (TPT) that enables coarse-to-fine optimization based on the inheritance-based partition tree structure for hierarchical temporal decomposition, and Spatial Ancestral Chains (SAC) that recursively query ancestral hierarchical structure to provide complementary spatial dynamics while specializing motion representations across ancestral nodes. Experimental results on different datasets indicate that our proposed method achieves 8.26% improvement of LPIPS on NVIDIA-LS and 9.09% improvement of mLPIPS on DyCheck compared to the second-best method. Code: https://github.com/iCVTEAM/WorldTree.", "date": "2026-02-12", "categories": []}
{"id": "2504.00437", "title": "ADGaussian: Generalizable Gaussian Splatting for Autonomous Driving via Multi-modal Joint Learning", "abstract": "We present a novel approach, termed ADGaussian, for generalizable street scene reconstruction. The proposed method enables high-quality rendering from merely single-view input. Unlike prior Gaussian Splatting methods that primarily focus on geometry refinement, we emphasize the importance of joint optimization of image and depth features for accurate Gaussian prediction. To this end, we first incorporate sparse LiDAR depth as an additional input modality, formulating the Gaussian prediction process as a joint learning framework of visual information and geometric clue. Furthermore, we propose a Multi-modal Feature Matching strategy coupled with a Multi-scale Gaussian Decoding model to enhance the joint refinement of multi-modal features, thereby enabling efficient multi-modal Gaussian learning. Extensive experiments on Waymo and KITTI demonstrate that our ADGaussian achieves state-of-the-art performance and exhibits superior zero-shot generalization capabilities in novel-view shifting.", "date": "2025-04-01", "categories": []}
{"id": "2602.10278", "title": "ERGO: Excess-Risk-Guided Optimization for High-Fidelity Monocular 3D Gaussian Splatting", "abstract": "Generating 3D content from a single image remains a fundamentally challenging and ill-posed problem due to the inherent absence of geometric and textural information in occluded regions. While state-of-the-art generative models can synthesize auxiliary views to provide additional supervision, these views inevitably contain geometric inconsistencies and textural misalignments that propagate and amplify artifacts during 3D reconstruction. To effectively harness these imperfect supervisory signals, we propose an adaptive optimization framework guided by excess risk decomposition, termed ERGO. Specifically, ERGO decomposes the optimization losses in 3D Gaussian splatting into two components, i.e., excess risk that quantifies the suboptimality gap between current and optimal parameters, and Bayes error that models the irreducible noise inherent in synthesized views. This decomposition enables ERGO to dynamically estimate the view-specific excess risk and adaptively adjust loss weights during optimization. Furthermore, we introduce geometry-aware and texture-aware objectives that complement the excess-risk-derived weighting mechanism, establishing a synergistic global-local optimization paradigm. Consequently, ERGO demonstrates robustness against supervision noise while consistently enhancing both geometric fidelity and textural quality of the reconstructed 3D content. Extensive experiments on the Google Scanned Objects dataset and the OmniObject3D dataset demonstrate the superiority of ERGO over existing state-of-the-art methods.", "date": "2026-02-10", "categories": []}
{"id": "2602.10943", "title": "Towards Learning a Generalizable 3D Scene Representation from 2D Observations", "abstract": "We introduce a Generalizable Neural Radiance Field approach for predicting 3D workspace occupancy from egocentric robot observations. Unlike prior methods operating in camera-centric coordinates, our model constructs occupancy representations in a global workspace frame, making it directly applicable to robotic manipulation. The model integrates flexible source views and generalizes to unseen object arrangements without scene-specific finetuning. We demonstrate the approach on a humanoid robot and evaluate predicted geometry against 3D sensor ground truth. Trained on 40 real scenes, our model achieves 26mm reconstruction error, including occluded regions, validating its ability to infer complete 3D occupancy beyond traditional stereo vision methods.", "date": "2026-02-11", "categories": []}
{"id": "2602.09999", "title": "Faster-GS: Analyzing and Improving Gaussian Splatting Optimization", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have focused on accelerating optimization while preserving reconstruction quality. However, many proposed methods entangle implementation-level improvements with fundamental algorithmic modifications or trade performance for fidelity, leading to a fragmented research landscape that complicates fair comparison. In this work, we consolidate and evaluate the most effective and broadly applicable strategies from prior 3DGS research and augment them with several novel optimizations. We further investigate underexplored aspects of the framework, including numerical stability, Gaussian truncation, and gradient approximation. The resulting system, Faster-GS, provides a rigorously optimized algorithm that we evaluate across a comprehensive suite of benchmarks. Our experiments demonstrate that Faster-GS achieves up to 5$\\times$ faster training while maintaining visual quality, establishing a new cost-effective and resource efficient baseline for 3DGS optimization. Furthermore, we demonstrate that optimizations can be applied to 4D Gaussian reconstruction, leading to efficient non-rigid scene optimization.", "date": "2026-02-10", "categories": []}
{"id": "2602.06991", "title": "LangGS-SLAM: Real-Time Language-Feature Gaussian Splatting SLAM", "abstract": "In this paper, we propose a RGB-D SLAM system that reconstructs a language-aligned dense feature field while sustaining low-latency tracking and mapping. First, we introduce a Top-K Rendering pipeline, a high-throughput and semantic-distortion-free method for efficiently rendering high-dimensional feature maps. To address the resulting semantic-geometric discrepancy and mitigate the memory consumption, we further design a multi-criteria map management strategy that prunes redundant or inconsistent Gaussians while preserving scene integrity. Finally, a hybrid field optimization framework jointly refines the geometric and semantic fields under real-time constraints by decoupling their optimization frequencies according to field characteristics. The proposed system achieves superior geometric fidelity compared to geometric-only baselines and comparable semantic fidelity to offline approaches while operating at 15 FPS. Our results demonstrate that online SLAM with dense, uncompressed language-aligned feature fields is both feasible and effective, bridging the gap between 3D perception and language-based reasoning.", "date": "2026-01-28", "categories": []}
{"id": "2602.07016", "title": "Gaussian-Constrained LeJEPA Representations for Unsupervised Scene Discovery and Pose Consistency", "abstract": "Unsupervised 3D scene reconstruction from unstructured image collections remains a fundamental challenge in computer vision, particularly when images originate from multiple unrelated scenes and contain significant visual ambiguity. The Image Matching Challenge 2025 (IMC2025) highlights these difficulties by requiring both scene discovery and camera pose estimation under real-world conditions, including outliers and mixed content. This paper investigates the application of Gaussian-constrained representations inspired by LeJEPA (Joint Embedding Predictive Architecture) to address these challenges. We present three progressively refined pipelines, culminating in a LeJEPA-inspired approach that enforces isotropic Gaussian constraints on learned image embeddings. Rather than introducing new theoretical guarantees, our work empirically evaluates how these constraints influence clustering consistency and pose estimation robustness in practice. Experimental results on IMC2025 demonstrate that Gaussian-constrained embeddings can improve scene separation and pose plausibility compared to heuristic-driven baselines, particularly in visually ambiguous settings. These findings suggest that theoretically motivated representation constraints offer a promising direction for bridging self-supervised learning principles and practical structure-from-motion pipelines.", "date": "2026-01-31", "categories": []}
{"id": "2602.07891", "title": "Scalable Adaptation of 3D Geometric Foundation Models via Weak Supervision from Internet Video", "abstract": "Geometric foundation models show promise in 3D reconstruction, yet their progress is severely constrained by the scarcity of diverse, large-scale 3D annotations. While Internet videos offer virtually unlimited raw data, utilizing them as a scaling source for geometric learning is challenging due to the absence of ground-truth geometry and the presence of observational noise. To address this, we propose SAGE, a framework for Scalable Adaptation of GEometric foundation models from raw video streams. SAGE leverages a hierarchical mining pipeline to transform videos into training trajectories and hybrid supervision: (1) Informative training trajectory selection; (2) Sparse Geometric Anchoring via SfM point clouds for global structural guidance; and (3) Dense Differentiable Consistency via 3D Gaussian rendering for multi-view constraints. To prevent catastrophic forgetting, we introduce a regularization strategy using anchor data. Extensive experiments show that SAGE significantly enhances zero-shot generalization, reducing Chamfer Distance by 20-42% on unseen benchmarks (7Scenes, TUM-RGBD, Matterport3D) compared to state-of-the-art baselines. To our knowledge, SAGE pioneers the adaptation of geometric foundation models via Internet video, establishing a scalable paradigm for general-purpose 3D learning.", "date": "2026-02-08", "categories": []}
{"id": "2602.08266", "title": "Informative Object-centric Next Best View for Object-aware 3D Gaussian Splatting in Cluttered Scenes", "abstract": "In cluttered scenes with inevitable occlusions and incomplete observations, selecting informative viewpoints is essential for building a reliable representation. In this context, 3D Gaussian Splatting (3DGS) offers a distinct advantage, as it can explicitly guide the selection of subsequent viewpoints and then refine the representation with new observations. However, existing approaches rely solely on geometric cues, neglect manipulation-relevant semantics, and tend to prioritize exploitation over exploration. To tackle these limitations, we introduce an instance-aware Next Best View (NBV) policy that prioritizes underexplored regions by leveraging object features. Specifically, our object-aware 3DGS distills instancelevel information into one-hot object vectors, which are used to compute confidence-weighted information gain that guides the identification of regions associated with erroneous and uncertain Gaussians. Furthermore, our method can be easily adapted to an object-centric NBV, which focuses view selection on a target object, thereby improving reconstruction robustness to object placement. Experiments demonstrate that our NBV policy reduces depth error by up to 77.14% on the synthetic dataset and 34.10% on the real-world GraspNet dataset compared to baselines. Moreover, compared to targeting the entire scene, performing NBV on a specific object yields an additional reduction of 25.60% in depth error for that object. We further validate the effectiveness of our approach through real-world robotic manipulation tasks.", "date": "2026-02-09", "categories": []}
{"id": "2602.08558", "title": "FLAG-4D: Flow-Guided Local-Global Dual-Deformation Model for 4D Reconstruction", "abstract": "We introduce FLAG-4D, a novel framework for generating novel views of dynamic scenes by reconstructing how 3D Gaussian primitives evolve through space and time. Existing methods typically rely on a single Multilayer Perceptron (MLP) to model temporal deformations, and they often struggle to capture complex point motions and fine-grained dynamic details consistently over time, especially from sparse input views. Our approach, FLAG-4D, overcomes this by employing a dual-deformation network that dynamically warps a canonical set of 3D Gaussians over time into new positions and anisotropic shapes. This dual-deformation network consists of an Instantaneous Deformation Network (IDN) for modeling fine-grained, local deformations and a Global Motion Network (GMN) for capturing long-range dynamics, refined through mutual learning. To ensure these deformations are both accurate and temporally smooth, FLAG-4D incorporates dense motion features from a pretrained optical flow backbone. We fuse these motion cues from adjacent timeframes and use a deformation-guided attention mechanism to align this flow information with the current state of each evolving 3D Gaussian. Extensive experiments demonstrate that FLAG-4D achieves higher-fidelity and more temporally coherent reconstructions with finer detail preservation than state-of-the-art methods.", "date": "2026-02-09", "categories": []}
{"id": "2601.08414", "title": "SPARK: Scalable Real-Time Point Cloud Aggregation with Multi-View Self-Calibration", "abstract": "Real-time multi-camera 3D reconstruction is crucial for 3D perception, immersive interaction, and robotics. Existing methods struggle with multi-view fusion, camera extrinsic uncertainty, and scalability for large camera setups. We propose SPARK, a self-calibrating real-time multi-camera point cloud reconstruction framework that jointly handles point cloud fusion and extrinsic uncertainty. SPARK consists of: (1) a geometry-aware online extrinsic estimation module leveraging multi-view priors and enforcing cross-view and temporal consistency for stable self-calibration, and (2) a confidence-driven point cloud fusion strategy modeling depth reliability and visibility at pixel and point levels to suppress noise and view-dependent inconsistencies. By performing frame-wise fusion without accumulation, SPARK produces stable point clouds in dynamic scenes while scaling linearly with the number of cameras. Extensive experiments on real-world multi-camera systems show that SPARK outperforms existing approaches in extrinsic accuracy, geometric consistency, temporal stability, and real-time performance, demonstrating its effectiveness and scalability for large-scale multi-camera 3D reconstruction.", "date": "2026-01-13", "categories": []}
{"id": "2602.06044", "title": "COSMOS: Coherent Supergaussian Modeling with Spatial Priors for Sparse-View 3D Splatting", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as a promising approach for 3D reconstruction, providing explicit, point-based representations and enabling high-quality real time rendering. However, when trained with sparse input views, 3DGS suffers from overfitting and structural degradation, leading to poor generalization on novel views. This limitation arises from its optimization relying solely on photometric loss without incorporating any 3D structure priors. To address this issue, we propose Coherent supergaussian Modeling with Spatial Priors (COSMOS). Inspired by the concept of superpoints from 3D segmentation, COSMOS introduces 3D structure priors by newly defining supergaussian groupings of Gaussians based on local geometric cues and appearance features. To this end, COSMOS applies inter group global self-attention across supergaussian groups and sparse local attention among individual Gaussians, enabling the integration of global and local spatial information. These structure-aware features are then used for predicting Gaussian attributes, facilitating more consistent 3D reconstructions. Furthermore, by leveraging supergaussian-based grouping, COSMOS enforces an intra-group positional regularization to maintain structural coherence and suppress floaters, thereby enhancing training stability under sparse-view conditions. Our experiments on Blender and DTU show that COSMOS surpasses state-of-the-art methods in sparse-view settings without any external depth supervision.", "date": "2025-12-17", "categories": []}
{"id": "2602.06158", "title": "MGP-KAD: Multimodal Geometric Priors and Kolmogorov-Arnold Decoder for Single-View 3D Reconstruction in Complex Scenes", "abstract": "Single-view 3D reconstruction in complex real-world scenes is challenging due to noise, object diversity, and limited dataset availability. To address these challenges, we propose MGP-KAD, a novel multimodal feature fusion framework that integrates RGB and geometric prior to enhance reconstruction accuracy. The geometric prior is generated by sampling and clustering ground-truth object data, producing class-level features that dynamically adjust during training to improve geometric understanding. Additionally, we introduce a hybrid decoder based on Kolmogorov-Arnold Networks (KAN) to overcome the limitations of traditional linear decoders in processing complex multimodal inputs. Extensive experiments on the Pix3D dataset demonstrate that MGP-KAD achieves state-of-the-art (SOTA) performance, significantly improving geometric integrity, smoothness, and detail preservation. Our work provides a robust and effective solution for advancing single-view 3D reconstruction in complex scenes.", "date": "2026-02-05", "categories": []}
{"id": "2602.06163", "title": "MetaSSP: Enhancing Semi-supervised Implicit 3D Reconstruction through Meta-adaptive EMA and SDF-aware Pseudo-label Evaluation", "abstract": "Implicit SDF-based methods for single-view 3D reconstruction achieve high-quality surfaces but require large labeled datasets, limiting their scalability. We propose MetaSSP, a novel semi-supervised framework that exploits abundant unlabeled images. Our approach introduces gradient-based parameter importance estimation to regularize adaptive EMA updates and an SDF-aware pseudo-label weighting mechanism combining augmentation consistency with SDF variance. Beginning with a 10% supervised warm-up, the unified pipeline jointly refines labeled and unlabeled data. On the Pix3D benchmark, our method reduces Chamfer Distance by approximately 20.61% and increases IoU by around 24.09% compared to existing semi-supervised baselines, setting a new state of the art.", "date": "2026-02-05", "categories": []}
{"id": "2602.01035", "title": "FUSE-Flow: Scalable Real-Time Multi-View Point Cloud Reconstruction Using Confidence", "abstract": "Real-time multi-view point cloud reconstruction is a core problem in 3D vision and immersive perception, with wide applications in VR, AR, robotic navigation, digital twins, and computer interaction. Despite advances in multi-camera systems and high-resolution depth sensors, fusing large-scale multi-view depth observations into high-quality point clouds under strict real-time constraints remains challenging. Existing methods relying on voxel-based fusion, temporal accumulation, or global optimization suffer from high computational complexity, excessive memory usage, and limited scalability, failing to simultaneously achieve real-time performance, reconstruction quality, and multi-camera extensibility. We propose FUSE-Flow, a frame-wise, stateless, and linearly scalable point cloud streaming reconstruction framework. Each frame independently generates point cloud fragments, fused via two weights, measurement confidence and 3D distance consistency to suppress noise while preserving geometric details. For large-scale multi-camera efficiency, we introduce an adaptive spatial hashing-based weighted aggregation method: 3D space is adaptively partitioned by local point cloud density, representative points are selected per cell, and weighted fusion is performed to handle both sparse and dense regions. With GPU parallelization, FUSE-Flow achieves high-throughput, low-latency point cloud generation and fusion with linear complexity. Experiments demonstrate that the framework improves reconstruction stability and geometric fidelity in overlapping, depth-discontinuous, and dynamic scenes, while maintaining real-time frame rates on modern GPUs, verifying its effectiveness, robustness, and scalability.", "date": "2026-02-01", "categories": []}
{"id": "2602.02989", "title": "SharpTimeGS: Sharp and Stable Dynamic Gaussian Splatting via Lifespan Modulation", "abstract": "Novel view synthesis of dynamic scenes is fundamental to achieving photorealistic 4D reconstruction and immersive visual experiences. Recent progress in Gaussian-based representations has significantly improved real-time rendering quality, yet existing methods still struggle to maintain a balance between long-term static and short-term dynamic regions in both representation and optimization. To address this, we present SharpTimeGS, a lifespan-aware 4D Gaussian framework that achieves temporally adaptive modeling of both static and dynamic regions under a unified representation. Specifically, we introduce a learnable lifespan parameter that reformulates temporal visibility from a Gaussian-shaped decay into a flat-top profile, allowing primitives to remain consistently active over their intended duration and avoiding redundant densification. In addition, the learned lifespan modulates each primitives' motion, reducing drift in long-lived static points while retaining unrestricted motion for short-lived dynamic ones. This effectively decouples motion magnitude from temporal duration, improving long-term stability without compromising dynamic fidelity. Moreover, we design a lifespan-velocity-aware densification strategy that mitigates optimization imbalance between static and dynamic regions by allocating more capacity to regions with pronounced motion while keeping static areas compact and stable. Extensive experiments on multiple benchmarks demonstrate that our method achieves state-of-the-art performance while supporting real-time rendering up to 4K resolution at 100 FPS on one RTX 4090.", "date": "2026-02-03", "categories": []}
{"id": "2602.04439", "title": "TrajVG: 3D Trajectory-Coupled Visual Geometry Learning", "abstract": "Feed-forward multi-frame 3D reconstruction models often degrade on videos with object motion. Global-reference becomes ambiguous under multiple motions, while the local pointmap relies heavily on estimated relative poses and can drift, causing cross-frame misalignment and duplicated structures. We propose TrajVG, a reconstruction framework that makes cross-frame 3D correspondence an explicit prediction by estimating camera-coordinate 3D trajectories. We couple sparse trajectories, per-frame local point maps, and relative camera poses with geometric consistency objectives: (i) bidirectional trajectory-pointmap consistency with controlled gradient flow, and (ii) a pose consistency objective driven by static track anchors that suppresses gradients from dynamic regions. To scale training to in-the-wild videos where 3D trajectory labels are scarce, we reformulate the same coupling constraints into self-supervised objectives using only pseudo 2D tracks, enabling unified training with mixed supervision. Extensive experiments across 3D tracking, pose estimation, pointmap reconstruction, and video depth show that TrajVG surpasses the current feedforward performance baseline.", "date": "2026-02-04", "categories": []}
{"id": "2602.05029", "title": "Differentiable Inverse Graphics for Zero-shot Scene Reconstruction and Robot Grasping", "abstract": "Operating effectively in novel real-world environments requires robotic systems to estimate and interact with previously unseen objects. Current state-of-the-art models address this challenge by using large amounts of training data and test-time samples to build black-box scene representations. In this work, we introduce a differentiable neuro-graphics model that combines neural foundation models with physics-based differentiable rendering to perform zero-shot scene reconstruction and robot grasping without relying on any additional 3D data or test-time samples. Our model solves a series of constrained optimization problems to estimate physically consistent scene parameters, such as meshes, lighting conditions, material properties, and 6D poses of previously unseen objects from a single RGBD image and bounding boxes. We evaluated our approach on standard model-free few-shot benchmarks and demonstrated that it outperforms existing algorithms for model-free few-shot pose estimation. Furthermore, we validated the accuracy of our scene reconstructions by applying our algorithm to a zero-shot grasping task. By enabling zero-shot, physically-consistent scene reconstruction and grasping without reliance on extensive datasets or test-time sampling, our approach offers a pathway towards more data efficient, interpretable and generalizable robot autonomy in novel environments.", "date": "2026-02-04", "categories": []}
{"id": "2602.05423", "title": "NeVStereo: A NeRF-Driven NVS-Stereo Architecture for High-Fidelity 3D Tasks", "abstract": "In modern dense 3D reconstruction, feed-forward systems (e.g., VGGT, pi3) focus on end-to-end matching and geometry prediction but do not explicitly output the novel view synthesis (NVS). Neural rendering-based approaches offer high-fidelity NVS and detailed geometry from posed images, yet they typically assume fixed camera poses and can be sensitive to pose errors. As a result, it remains non-trivial to obtain a single framework that can offer accurate poses, reliable depth, high-quality rendering, and accurate 3D surfaces from casually captured views. We present NeVStereo, a NeRF-driven NVS-stereo architecture that aims to jointly deliver camera poses, multi-view depth, novel view synthesis, and surface reconstruction from multi-view RGB-only inputs. NeVStereo combines NeRF-based NVS for stereo-friendly renderings, confidence-guided multi-view depth estimation, NeRF-coupled bundle adjustment for pose refinement, and an iterative refinement stage that updates both depth and the radiance field to improve geometric consistency. This design mitigated the common NeRF-based issues such as surface stacking, artifacts, and pose-depth coupling. Across indoor, outdoor, tabletop, and aerial benchmarks, our experiments indicate that NeVStereo achieves consistently strong zero-shot performance, with up to 36% lower depth error, 10.4% improved pose accuracy, 4.5% higher NVS fidelity, and state-of-the-art mesh quality (F1 91.93%, Chamfer 4.35 mm) compared to existing prestigious methods.", "date": "2026-02-05", "categories": []}
{"id": "2602.05434", "title": "LD-SLRO: Latent Diffusion Structured Light for 3-D Reconstruction of Highly Reflective Objects", "abstract": "Fringe projection profilometry-based 3-D reconstruction of objects with high reflectivity and low surface roughness remains a significant challenge. When measuring such glossy surfaces, specular reflection and indirect illumination often lead to severe distortion or loss of the projected fringe patterns. To address these issues, we propose a latent diffusion-based structured light for reflective objects (LD-SLRO). Phase-shifted fringe images captured from highly reflective surfaces are first encoded to extract latent representations that capture surface reflectance characteristics. These latent features are then used as conditional inputs to a latent diffusion model, which probabilistically suppresses reflection-induced artifacts and recover lost fringe information, yielding high-quality fringe images. The proposed components, including the specular reflection encoder, time-variant channel affine layer, and attention modules, further improve fringe restoration quality. In addition, LD-SLRO provides high flexibility in configuring the input and output fringe sets. Experimental results demonstrate that the proposed method improves both fringe quality and 3-D reconstruction accuracy over state-of-the-art methods, reducing the average root-mean-squared error from 1.8176 mm to 0.9619 mm.", "date": "2026-02-05", "categories": []}
{"id": "2602.04517", "title": "S-MUSt3R: Sliding Multi-view 3D Reconstruction", "abstract": "The recent paradigm shift in 3D vision led to the rise of foundation models with remarkable capabilities in 3D perception from uncalibrated images. However, extending these models to large-scale RGB stream 3D reconstruction remains challenging due to memory limitations. This work proposes S-MUSt3R, a simple and efficient pipeline that extends the limits of foundation models for monocular 3D reconstruction. Our approach addresses the scalability bottleneck of foundation models through a simple strategy of sequence segmentation followed by segment alignment and lightweight loop closure optimization. Without model retraining, we benefit from remarkable 3D reconstruction capacities of MUSt3R model and achieve trajectory and reconstruction performance comparable to traditional methods with more complex architecture. We evaluate S-MUSt3R on TUM, 7-Scenes and proprietary robot navigation datasets and show that S-MUSt3R runs successfully on long RGB sequences and produces accurate and consistent 3D reconstruction. Our results highlight the potential of leveraging the MUSt3R model for scalable monocular 3D scene in real-world settings, with an important advantage of making predictions directly in the metric space.", "date": "2026-02-04", "categories": []}
{"id": "2505.17001", "title": "Seeing through Satellite Images at Street Views", "abstract": "This paper studies the task of SatStreet-view synthesis, which aims to render photorealistic street-view panorama images and videos given any satellite image and specified camera positions or trajectories. We formulate to learn neural radiance field from paired images captured from satellite and street viewpoints, which comes to be a challenging learning problem due to the sparse-view natural and the extremely-large viewpoint changes between satellite and street-view images. We tackle the challenges based on a task-specific observation that street-view specific elements, including the sky and illumination effects are only visible in street-view panoramas, and present a novel approach Sat2Density++ to accomplish the goal of photo-realistic street-view panoramas rendering by modeling these street-view specific in neural networks. In the experiments, our method is testified on both urban and suburban scene datasets, demonstrating that Sat2Density++ is capable of rendering photorealistic street-view panoramas that are consistent across multiple views and faithful to the satellite image.", "date": "2025-05-22", "categories": []}
{"id": "2511.20348", "title": "Material-informed Gaussian Splatting for 3D World Reconstruction in a Digital Twin", "abstract": "3D reconstruction for Digital Twins often relies on LiDAR-based methods, which provide accurate geometry but lack the semantics and textures naturally captured by cameras. Traditional LiDAR-camera fusion approaches require complex calibration and still struggle with certain materials like glass, which are visible in images but poorly represented in point clouds. We propose a camera-only pipeline that reconstructs scenes using 3D Gaussian Splatting from multi-view images, extracts semantic material masks via vision models, converts Gaussian representations to mesh surfaces with projected material labels, and assigns physics-based material properties for accurate sensor simulation in modern graphics engines and simulators. This approach combines photorealistic reconstruction with physics-based material assignment, providing sensor simulation fidelity comparable to LiDAR-camera fusion while eliminating hardware complexity and calibration requirements. We validate our camera-only method using an internal dataset from an instrumented test vehicle, leveraging LiDAR as ground truth for reflectivity validation alongside image similarity metrics.", "date": "2025-11-25", "categories": []}
{"id": "2602.02000", "title": "SurfSplat: Conquering Feedforward 2D Gaussian Splatting with Surface Continuity Priors", "abstract": "Reconstructing 3D scenes from sparse images remains a challenging task due to the difficulty of recovering accurate geometry and texture without optimization. Recent approaches leverage generalizable models to generate 3D scenes using 3D Gaussian Splatting (3DGS) primitive. However, they often fail to produce continuous surfaces and instead yield discrete, color-biased point clouds that appear plausible at normal resolution but reveal severe artifacts under close-up views. To address this issue, we present SurfSplat, a feedforward framework based on 2D Gaussian Splatting (2DGS) primitive, which provides stronger anisotropy and higher geometric precision. By incorporating a surface continuity prior and a forced alpha blending strategy, SurfSplat reconstructs coherent geometry together with faithful textures. Furthermore, we introduce High-Resolution Rendering Consistency (HRRC), a new evaluation metric designed to evaluate high-resolution reconstruction quality. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that SurfSplat consistently outperforms prior methods on both standard metrics and HRRC, establishing a robust solution for high-fidelity 3D reconstruction from sparse inputs. Project page: https://hebing-sjtu.github.io/SurfSplat-website/", "date": "2026-02-02", "categories": []}
{"id": "2602.03200", "title": "Hand3R: Online 4D Hand-Scene Reconstruction in the Wild", "abstract": "For Embodied AI, jointly reconstructing dynamic hands and the dense scene context is crucial for understanding physical interaction. However, most existing methods recover isolated hands in local coordinates, overlooking the surrounding 3D environment. To address this, we present Hand3R, the first online framework for joint 4D hand-scene reconstruction from monocular video. Hand3R synergizes a pre-trained hand expert with a 4D scene foundation model via a scene-aware visual prompting mechanism. By injecting high-fidelity hand priors into a persistent scene memory, our approach enables simultaneous reconstruction of accurate hand meshes and dense metric-scale scene geometry in a single forward pass. Experiments demonstrate that Hand3R bypasses the reliance on offline optimization and delivers competitive performance in both local hand reconstruction and global positioning.", "date": "2026-02-03", "categories": []}
{"id": "2602.03538", "title": "Constrained Dynamic Gaussian Splatting", "abstract": "While Dynamic Gaussian Splatting enables high-fidelity 4D reconstruction, its deployment is severely hindered by a fundamental dilemma: unconstrained densification leads to excessive memory consumption incompatible with edge devices, whereas heuristic pruning fails to achieve optimal rendering quality under preset Gaussian budgets. In this work, we propose Constrained Dynamic Gaussian Splatting (CDGS), a novel framework that formulates dynamic scene reconstruction as a budget-constrained optimization problem to enforce a strict, user-defined Gaussian budget during training. Our key insight is to introduce a differentiable budget controller as the core optimization driver. Guided by a multi-modal unified importance score, this controller fuses geometric, motion, and perceptual cues for precise capacity regulation. To maximize the utility of this fixed budget, we further decouple the optimization of static and dynamic elements, employing an adaptive allocation mechanism that dynamically distributes capacity based on motion complexity. Furthermore, we implement a three-phase training strategy to seamlessly integrate these constraints, ensuring precise adherence to the target count. Coupled with a dual-mode hybrid compression scheme, CDGS not only strictly adheres to hardware constraints (error < 2%}) but also pushes the Pareto frontier of rate-distortion performance. Extensive experiments demonstrate that CDGS delivers optimal rendering quality under varying capacity limits, achieving over 3x compression compared to state-of-the-art methods.", "date": "2026-02-03", "categories": []}
{"id": "2602.03809", "title": "Split&Splat: Zero-Shot Panoptic Segmentation via Explicit Instance Modeling and 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (GS) enables fast and high-quality scene reconstruction, but it lacks an object-consistent and semantically aware structure. We propose Split&Splat, a framework for panoptic scene reconstruction using 3DGS. Our approach explicitly models object instances. It first propagates instance masks across views using depth, thus producing view-consistent 2D masks. Each object is then reconstructed independently and merged back into the scene while refining its boundaries. Finally, instance-level semantic descriptors are embedded in the reconstructed objects, supporting various applications, including panoptic segmentation, object retrieval, and 3D editing. Unlike existing methods, Split&Splat tackles the problem by first segmenting the scene and then reconstructing each object individually. This design naturally supports downstream tasks and allows Split&Splat to achieve state-of-the-art performance on the ScanNetv2 segmentation benchmark.", "date": "2026-02-01", "categories": []}
{"id": "2602.03847", "title": "EventNeuS: 3D Mesh Reconstruction from a Single Event Camera", "abstract": "Event cameras offer a considerable alternative to RGB cameras in many scenarios. While there are recent works on event-based novel-view synthesis, dense 3D mesh reconstruction remains scarcely explored and existing event-based techniques are severely limited in their 3D reconstruction accuracy. To address this limitation, we present EventNeuS, a self-supervised neural model for learning 3D representations from monocular colour event streams. Our approach, for the first time, combines 3D signed distance function and density field learning with event-based supervision. Furthermore, we introduce spherical harmonics encodings into our model for enhanced handling of view-dependent effects. EventNeuS outperforms existing approaches by a significant margin, achieving 34% lower Chamfer distance and 31% lower mean absolute error on average compared to the best previous method.", "date": "2026-02-03", "categories": []}
{"id": "2601.22861", "title": "Under-Canopy Terrain Reconstruction in Dense Forests Using RGB Imaging and Neural 3D Reconstruction", "abstract": "Mapping the terrain and understory hidden beneath dense forest canopies is of great interest for numerous applications such as search and rescue, trail mapping, forest inventory tasks, and more. Existing solutions rely on specialized sensors: either heavy, costly airborne LiDAR, or Airborne Optical Sectioning (AOS), which uses thermal synthetic aperture photography and is tailored for person detection.\n  We introduce a novel approach for the reconstruction of canopy-free, photorealistic ground views using only conventional RGB images. Our solution is based on the celebrated Neural Radiance Fields (NeRF), a recent 3D reconstruction method. Additionally, we include specific image capture considerations, which dictate the needed illumination to successfully expose the scene beneath the canopy. To better cope with the poorly lit understory, we employ a low light loss. Finally, we propose two complementary approaches to remove occluding canopy elements by controlling per-ray integration procedure.\n  To validate the value of our approach, we present two possible downstream tasks. For the task of search and rescue (SAR), we demonstrate that our method enables person detection which achieves promising results compared to thermal AOS (using only RGB images). Additionally, we show the potential of our approach for forest inventory tasks like tree counting. These results position our approach as a cost-effective, high-resolution alternative to specialized sensors for SAR, trail mapping, and forest-inventory tasks.", "date": "2026-01-30", "categories": []}
{"id": "2602.00395", "title": "3DGS$^2$-TR: Scalable Second-Order Trust-Region Method for 3D Gaussian Splatting", "abstract": "We propose 3DGS$^2$-TR,a second-order optimizer for accelerating the scene training problem in 3D Gaussian Splatting (3DGS). Unlike existing second-order approaches that rely on explicit or dense curvature representations, such as 3DGS-LM (H\\\"ollein et al., 2025) or 3DGS2 (Lan et al., 2025), our method approximates curvature using only the diagonal of the Hessian matrix, efficiently via Hutchinson's method. Our approach is fully matrix-free and has the same complexity as ADAM (Kingma, 2024), $O(n)$ in both computation and memory costs. To ensure stable optimization in the presence of strong nonlinearity in the 3DGS rasterization process, we introduce a parameter-wise trust-region technique based on the squared Hellinger distance, regularizing updates to Gaussian parameters. Under identical parameter initialization and without densification, 3DGS$^2$-TR is able to achieve better reconstruction quality on standard datasets, using 50% fewer training iterations compared to ADAM, while incurring less than 1GB of peak GPU memory overhead (17% more than ADAM and 85% less than 3DGS-LM), enabling scalability to very large scenes and potentially to distributed training settings.", "date": "2026-01-30", "categories": []}
{"id": "2602.00865", "title": "Distill3R: A Pipeline for Democratizing 3D Foundation Models on Commodity Hardware", "abstract": "While multi-view 3D reconstruction has shifted toward large-scale foundation models capable of inferring globally consistent geometry, their reliance on massive computational clusters for training has created a significant barrier to entry for most academic laboratories. To bridge this compute divide, we introduce Distill3R, a framework designed to distill the geometric reasoning of 3D foundation models into compact students fully trainable on a single workstation. Our methodology centers on two primary innovations: (1) an offline caching pipeline that decouples heavy teacher inference from the training loop through compressed supervision signals, and (2) a confidence-aware distillation loss that leverages teacher uncertainty to enable training on commodity hardware. We propose a 72M-parameter student model which achieves a 9x reduction in parameters and a 5x inference speedup compared to its 650M-parameter teacher. The student is fully trainable in under 3 days on a single workstation, whereas its teacher requires massive GPU clusters for up to a week. We demonstrate that the student preserves the structural consistency and qualitative geometric understanding required for functional 3D awareness. By providing a reproducible, single-workstation training recipe, Distill3R serves as an exploratory entry point for democratized 3D vision research and efficient edge deployment. This work is not intended to compete with state-of-the-art foundation models, but to provide an accessible research baseline for laboratories without access to large-scale compute to train and specialize models on their own domain-specific data at minimal cost.", "date": "2026-01-31", "categories": []}
{"id": "2602.01296", "title": "Interacted Planes Reveal 3D Line Mapping", "abstract": "3D line mapping from multi-view RGB images provides a compact and structured visual representation of scenes. We study the problem from a physical and topological perspective: a 3D line most naturally emerges as the edge of a finite 3D planar patch. We present LiP-Map, a line-plane joint optimization framework that explicitly models learnable line and planar primitives. This coupling enables accurate and detailed 3D line mapping while maintaining strong efficiency (typically completing a reconstruction in 3 to 5 minutes per scene). LiP-Map pioneers the integration of planar topology into 3D line mapping, not by imposing pairwise coplanarity constraints but by explicitly constructing interactions between plane and line primitives, thus offering a principled route toward structured reconstruction in man-made environments. On more than 100 scenes from ScanNetV2, ScanNet++, Hypersim, 7Scenes, and Tanks\\&Temple, LiP-Map improves both accuracy and completeness over state-of-the-art methods. Beyond line mapping quality, LiP-Map significantly advances line-assisted visual localization, establishing strong performance on 7Scenes. Our code is released at https://github.com/calmke/LiPMAP for reproducible research.", "date": "2026-02-01", "categories": []}
{"id": "2512.18237", "title": "Joint Learning of Depth, Pose, and Local Radiance Field for Large Scale Monocular 3D Reconstruction", "abstract": "Photorealistic 3-D reconstruction from monocular video collapses in large-scale scenes when depth, pose, and radiance are solved in isolation: scale-ambiguous depth yields ghost geometry, long-horizon pose drift corrupts alignment, and a single global NeRF cannot model hundreds of metres of content. We introduce a joint learning framework that couples all three factors and demonstrably overcomes each failure case. Our system begins with a Vision-Transformer (ViT) depth network trained with metric-scale supervision, giving globally consistent depths despite wide field-of-view variations. A multi-scale feature bundle-adjustment (BA) layer refines camera poses directly in feature space--leveraging learned pyramidal descriptors instead of brittle keypoints--to suppress drift on unconstrained trajectories. For scene representation, we deploy an incremental local-radiance-field hierarchy: new hash-grid NeRFs are allocated and frozen on-the-fly when view overlap falls below a threshold, enabling city-block-scale coverage on a single GPU. Evaluated on the Tanks and Temples benchmark, our method reduces Absolute Trajectory Error to 0.001-0.021 m across eight indoor-outdoor sequences--up to 18x lower than BARF and 2x lower than NoPe-NeRF--while maintaining sub-pixel Relative Pose Error. These results demonstrate that metric-scale, drift-free 3-D reconstruction and high-fidelity novel-view synthesis are achievable from a single uncalibrated RGB camera.", "date": "2025-12-20", "categories": []}
{"id": "2601.16272", "title": "GR3EN: Generative Relighting for 3D Environments", "abstract": "We present a method for relighting 3D reconstructions of large room-scale environments. Existing solutions for 3D scene relighting often require solving under-determined or ill-conditioned inverse rendering problems, and are as such unable to produce high-quality results on complex real-world scenes. Though recent progress in using generative image and video diffusion models for relighting has been promising, these techniques are either limited to 2D image and video relighting or 3D relighting of individual objects. Our approach enables controllable 3D relighting of room-scale scenes by distilling the outputs of a video-to-video relighting diffusion model into a 3D reconstruction. This side-steps the need to solve a difficult inverse rendering problem, and results in a flexible system that can relight 3D reconstructions of complex real-world scenes. We validate our approach on both synthetic and real-world datasets to show that it can faithfully render novel views of scenes under new lighting conditions.", "date": "2026-01-22", "categories": []}
{"id": "2601.21786", "title": "Synthetic-to-Real Domain Bridging for Single-View 3D Reconstruction of Ships for Maritime Monitoring", "abstract": "Three-dimensional (3D) reconstruction of ships is an important part of maritime monitoring, allowing improved visualization, inspection, and decision-making in real-world monitoring environments. However, most state-ofthe-art 3D reconstruction methods require multi-view supervision, annotated 3D ground truth, or are computationally intensive, making them impractical for real-time maritime deployment. In this work, we present an efficient pipeline for single-view 3D reconstruction of real ships by training entirely on synthetic data and requiring only a single view at inference. Our approach uses the Splatter Image network, which represents objects as sparse sets of 3D Gaussians for rapid and accurate reconstruction from single images. The model is first fine-tuned on synthetic ShapeNet vessels and further refined with a diverse custom dataset of 3D ships, bridging the domain gap between synthetic and real-world imagery. We integrate a state-of-the-art segmentation module based on YOLOv8 and custom preprocessing to ensure compatibility with the reconstruction network. Postprocessing steps include real-world scaling, centering, and orientation alignment, followed by georeferenced placement on an interactive web map using AIS metadata and homography-based mapping. Quantitative evaluation on synthetic validation data demonstrates strong reconstruction fidelity, while qualitative results on real maritime images from the ShipSG dataset confirm the potential for transfer to operational maritime settings. The final system provides interactive 3D inspection of real ships without requiring real-world 3D annotations. This pipeline provides an efficient, scalable solution for maritime monitoring and highlights a path toward real-time 3D ship visualization in practical applications. Interactive demo: https://dlr-mi.github.io/ship3d-demo/.", "date": "2026-01-29", "categories": []}
{"id": "2601.22045", "title": "Urban Neural Surface Reconstruction from Constrained Sparse Aerial Imagery with 3D SAR Fusion", "abstract": "Neural surface reconstruction (NSR) has recently shown strong potential for urban 3D reconstruction from multi-view aerial imagery. However, existing NSR methods often suffer from geometric ambiguity and instability, particularly under sparse-view conditions. This issue is critical in large-scale urban remote sensing, where aerial image acquisition is limited by flight paths, terrain, and cost. To address this challenge, we present the first urban NSR framework that fuses 3D synthetic aperture radar (SAR) point clouds with aerial imagery for high-fidelity reconstruction under constrained, sparse-view settings. 3D SAR can efficiently capture large-scale geometry even from a single side-looking flight path, providing robust priors that complement photometric cues from images. Our framework integrates radar-derived spatial constraints into an SDF-based NSR backbone, guiding structure-aware ray selection and adaptive sampling for stable and efficient optimization. We also construct the first benchmark dataset with co-registered 3D SAR point clouds and aerial imagery, facilitating systematic evaluation of cross-modal 3D reconstruction. Extensive experiments show that incorporating 3D SAR markedly enhances reconstruction accuracy, completeness, and robustness compared with single-modality baselines under highly sparse and oblique-view conditions, highlighting a viable route toward scalable high-fidelity urban reconstruction with advanced airborne and spaceborne optical-SAR sensing.", "date": "2026-01-29", "categories": []}
{"id": "2501.14277", "title": "Dense-SfM: Structure from Motion with Dense Consistent Matching", "abstract": "We present Dense-SfM, a novel Structure from Motion (SfM) framework designed for dense and accurate 3D reconstruction from multi-view images. Sparse keypoint matching, which traditional SfM methods often rely on, limits both accuracy and point density, especially in texture-less areas. Dense-SfM addresses this limitation by integrating dense matching with a Gaussian Splatting (GS) based track extension which gives more consistent, longer feature tracks. To further improve reconstruction accuracy, Dense-SfM is equipped with a multi-view kernelized matching module leveraging transformer and Gaussian Process architectures, for robust track refinement across multi-views. Evaluations on the ETH3D and Texture-Poor SfM datasets show that Dense-SfM offers significant improvements in accuracy and density over state-of-the-art methods. Project page: https://icetea-cv.github.io/densesfm/.", "date": "2025-01-24", "categories": []}
{"id": "2503.22349", "title": "GCRayDiffusion: Pose-Free Surface Reconstruction via Geometric Consistent Ray Diffusion", "abstract": "Accurate surface reconstruction from unposed images is crucial for efficient 3D object or scene creation. However, it remains challenging, particularly for the joint camera pose estimation. Previous approaches have achieved impressive pose-free surface reconstruction results in dense-view settings, but could easily fail for sparse-view scenarios without sufficient visual overlap. In this paper, we propose a new technique for pose-free surface reconstruction, which follows triplane-based signed distance field (SDF) learning but regularizes the learning by explicit points sampled from ray-based diffusion of camera pose estimation. Our key contribution is a novel Geometric Consistent Ray Diffusion model (GCRayDiffusion), where we represent camera poses as neural bundle rays and regress the distribution of noisy rays via a diffusion model. More importantly, we further condition the denoising process of RGRayDiffusion using the triplane-based SDF of the entire scene, which provides effective 3D consistent regularization to achieve multi-view consistent camera pose estimation. Finally, we incorporate RGRayDiffusion into the triplane-based SDF learning by introducing on-surface geometric regularization from the sampling points of the neural bundle rays, which leads to highly accurate pose-free surface reconstruction results even for sparse-view inputs. Extensive evaluations on public datasets show that our GCRayDiffusion achieves more accurate camera pose estimation than previous approaches, with geometrically more consistent surface reconstruction results, especially given sparse-view inputs.", "date": "2025-03-28", "categories": []}
{"id": "2509.19073", "title": "WaveletGaussian: Wavelet-domain Diffusion for Sparse-view 3D Gaussian Object Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has become a powerful representation for image-based object reconstruction, yet its performance drops sharply in sparse-view settings. Prior works address this limitation by employing diffusion models to repair corrupted renders, subsequently using them as pseudo ground truths for later optimization. While effective, such approaches incur heavy computation from the diffusion fine-tuning and repair steps. We present WaveletGaussian, a framework for more efficient sparse-view 3D Gaussian object reconstruction. Our key idea is to shift diffusion into the wavelet domain: diffusion is applied only to the low-resolution LL subband, while high-frequency subbands are refined with a lightweight network. We further propose an efficient online random masking strategy to curate training pairs for diffusion fine-tuning, replacing the commonly used, but inefficient, leave-one-out strategy. Experiments across two benchmark datasets, Mip-NeRF 360 and OmniObject3D, show WaveletGaussian achieves competitive rendering quality while substantially reducing training time.", "date": "2025-09-23", "categories": []}
{"id": "2510.18714", "title": "PLANA3R: Zero-shot Metric Planar 3D Reconstruction via Feed-Forward Planar Splatting", "abstract": "This paper addresses metric 3D reconstruction of indoor scenes by exploiting their inherent geometric regularities with compact representations. Using planar 3D primitives - a well-suited representation for man-made environments - we introduce PLANA3R, a pose-free framework for metric Planar 3D Reconstruction from unposed two-view images. Our approach employs Vision Transformers to extract a set of sparse planar primitives, estimate relative camera poses, and supervise geometry learning via planar splatting, where gradients are propagated through high-resolution rendered depth and normal maps of primitives. Unlike prior feedforward methods that require 3D plane annotations during training, PLANA3R learns planar 3D structures without explicit plane supervision, enabling scalable training on large-scale stereo datasets using only depth and normal annotations. We validate PLANA3R on multiple indoor-scene datasets with metric supervision and demonstrate strong generalization to out-of-domain indoor environments across diverse tasks under metric evaluation protocols, including 3D surface reconstruction, depth estimation, and relative pose estimation. Furthermore, by formulating with planar 3D representation, our method emerges with the ability for accurate plane segmentation. The project page is available at https://lck666666.github.io/plana3r", "date": "2025-10-21", "categories": []}
{"id": "2601.19489", "title": "Fast Converging 3D Gaussian Splatting for 1-Minute Reconstruction", "abstract": "We present a fast 3DGS reconstruction pipeline designed to converge within one minute, developed for the SIGGRAPH Asia 3DGS Fast Reconstruction Challenge. The challenge consists of an initial round using SLAM-generated camera poses (with noisy trajectories) and a final round using COLMAP poses (highly accurate). To robustly handle these heterogeneous settings, we develop a two-stage solution. In the first round, we use reverse per-Gaussian parallel optimization and compact forward splatting based on Taming-GS and Speedy-splat, load-balanced tiling, an anchor-based Neural-Gaussian representation enabling rapid convergence with fewer learnable parameters, initialization from monocular depth and partially from feed-forward 3DGS models, and a global pose refinement module for noisy SLAM trajectories. In the final round, the accurate COLMAP poses change the optimization landscape; we disable pose refinement, revert from Neural-Gaussians back to standard 3DGS to eliminate MLP inference overhead, introduce multi-view consistency-guided Gaussian splitting inspired by Fast-GS, and introduce a depth estimator to supervise the rendered depth. Together, these techniques enable high-fidelity reconstruction under a strict one-minute budget. Our method achieved the top performance with a PSNR of 28.43 and ranked first in the competition.", "date": "2026-01-27", "categories": []}
{"id": "2504.11946", "title": "R-Meshfusion: Reinforcement Learning Powered Sparse-View Mesh Reconstruction with Diffusion Priors", "abstract": "Mesh reconstruction from multi-view images is a fundamental problem in computer vision, but its performance degrades significantly under sparse-view conditions, especially in unseen regions where no ground-truth observations are available. While recent advances in diffusion models have demonstrated strong capabilities in synthesizing novel views from limited inputs, their outputs often suffer from visual artifacts and lack 3D consistency, posing challenges for reliable mesh optimization. In this paper, we propose a novel framework that leverages diffusion models to enhance sparse-view mesh reconstruction in a principled and reliable manner. To address the instability of diffusion outputs, we propose a Consensus Diffusion Module that filters unreliable generations via interquartile range (IQR) analysis and performs variance-aware image fusion to produce robust pseudo-supervision. Building on this, we design an online reinforcement learning strategy based on the Upper Confidence Bound (UCB) to adaptively select the most informative viewpoints for enhancement, guided by diffusion loss. Finally, the fused images are used to jointly supervise a NeRF-based model alongside sparse-view ground truth, ensuring consistency across both geometry and appearance. Extensive experiments demonstrate that our method achieves significant improvements in both geometric quality and rendering quality.", "date": "2025-04-16", "categories": []}
{"id": "2508.09239", "title": "Gradient-Direction-Aware Density Control for 3D Gaussian Splatting", "abstract": "The emergence of 3D Gaussian Splatting (3DGS) has significantly advanced Novel View Synthesis (NVS) through explicit scene representation, enabling real-time photorealistic rendering. However, existing approaches manifest two critical limitations in complex scenarios: (1) Over-reconstruction occurs when persistent large Gaussians cannot meet adaptive splitting thresholds during density control. This is exacerbated by conflicting gradient directions that prevent effective splitting of these Gaussians; (2) Over-densification of Gaussians occurs in regions with aligned gradient aggregation, leading to redundant component proliferation. This redundancy significantly increases memory overhead due to unnecessary data retention. We present Gradient-Direction-Aware Gaussian Splatting (GDAGS) to address these challenges. Our key innovations: the Gradient Coherence Ratio (GCR), computed through normalized gradient vector norms, which explicitly discriminates Gaussians with concordant versus conflicting gradient directions; and a nonlinear dynamic weighting mechanism leverages the GCR to enable gradient-direction-aware density control. Specifically, GDAGS prioritizes conflicting-gradient Gaussians during splitting operations to enhance geometric details while suppressing redundant concordant-direction Gaussians. Conversely, in cloning processes, GDAGS promotes concordant-direction Gaussian densification for structural completion while preventing conflicting-direction Gaussian overpopulation. Comprehensive evaluations across diverse real-world benchmarks demonstrate that GDAGS achieves superior rendering quality while effectively mitigating over-reconstruction, suppressing over-densification, and constructing compact scene representations.", "date": "2025-08-12", "categories": []}
{"id": "2601.17835", "title": "Geometry-Grounded Gaussian Splatting", "abstract": "Gaussian Splatting (GS) has demonstrated impressive quality and efficiency in novel view synthesis. However, shape extraction from Gaussian primitives remains an open problem. Due to inadequate geometry parameterization and approximation, existing shape reconstruction methods suffer from poor multi-view consistency and are sensitive to floaters. In this paper, we present a rigorous theoretical derivation that establishes Gaussian primitives as a specific type of stochastic solids. This theoretical framework provides a principled foundation for Geometry-Grounded Gaussian Splatting by enabling the direct treatment of Gaussian primitives as explicit geometric representations. Using the volumetric nature of stochastic solids, our method efficiently renders high-quality depth maps for fine-grained geometry extraction. Experiments show that our method achieves the best shape reconstruction results among all Gaussian Splatting-based methods on public datasets.", "date": "2026-01-25", "categories": []}
{"id": "2601.19753", "title": "WaterClear-GS: Optical-Aware Gaussian Splatting for Underwater Reconstruction and Restoration", "abstract": "Underwater 3D reconstruction and appearance restoration are hindered by the complex optical properties of water, such as wavelength-dependent attenuation and scattering. Existing Neural Radiance Fields (NeRF)-based methods struggle with slow rendering speeds and suboptimal color restoration, while 3D Gaussian Splatting (3DGS) inherently lacks the capability to model complex volumetric scattering effects. To address these issues, we introduce WaterClear-GS, the first pure 3DGS-based framework that explicitly integrates underwater optical properties of local attenuation and scattering into Gaussian primitives, eliminating the need for an auxiliary medium network. Our method employs a dual-branch optimization strategy to ensure underwater photometric consistency while naturally recovering water-free appearances. This strategy is enhanced by depth-guided geometry regularization and perception-driven image loss, together with exposure constraints, spatially-adaptive regularization, and physically guided spectral regularization, which collectively enforce local 3D coherence and maintain natural visual perception. Experiments on standard benchmarks and our newly collected dataset demonstrate that WaterClear-GS achieves outstanding performance on both novel view synthesis (NVS) and underwater image restoration (UIR) tasks, while maintaining real-time rendering. The code will be available at https://buaaxrzhang.github.io/WaterClear-GS/.", "date": "2026-01-27", "categories": []}
{"id": "2407.09386", "title": "Radiance Fields from Photons", "abstract": "Neural radiance fields, or NeRFs, have become the de facto approach for high-quality view synthesis from a collection of images captured from multiple viewpoints. However, many issues remain when capturing images in-the-wild under challenging conditions, such as low light, high dynamic range, or rapid motion leading to smeared reconstructions with noticeable artifacts. In this work, we introduce quanta radiance fields, a novel class of neural radiance fields that are trained at the granularity of individual photons using single-photon cameras (SPCs). We develop theory and practical computational techniques for building radiance fields and estimating dense camera poses from unconventional, stochastic, and high-speed binary frame sequences captured by SPCs. We demonstrate, both via simulations and a SPC hardware prototype, high-fidelity reconstructions under high-speed motion, in low light, and for extreme dynamic range settings.", "date": "2024-07-12", "categories": []}
{"id": "2502.12138", "title": "FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views", "abstract": "We present FLARE, a feed-forward model designed to infer high-quality camera poses and 3D geometry from uncalibrated sparse-view images (i.e., as few as 2-8 inputs), which is a challenging yet practical setting in real-world applications. Our solution features a cascaded learning paradigm with camera pose serving as the critical bridge, recognizing its essential role in mapping 3D structures onto 2D image planes. Concretely, FLARE starts with camera pose estimation, whose results condition the subsequent learning of geometric structure and appearance, optimized through the objectives of geometry reconstruction and novel-view synthesis. Utilizing large-scale public datasets for training, our method delivers state-of-the-art performance in the tasks of pose estimation, geometry reconstruction, and novel view synthesis, while maintaining the inference efficiency (i.e., less than 0.5 seconds). The project page and code can be found at: https://zhanghe3z.github.io/FLARE/", "date": "2025-02-17", "categories": []}
{"id": "2506.20671", "title": "IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals", "abstract": "Semantic Scene Completion (SSC) has emerged as a pivotal approach for jointly learning scene geometry and semantics, enabling downstream applications such as navigation in mobile robotics. The recent generalization to Panoptic Scene Completion (PSC) advances the SSC domain by integrating instance-level information, thereby enhancing object-level sensitivity in scene understanding. While PSC was introduced using LiDAR modality, methods based on camera images remain largely unexplored. Moreover, recent Transformer-based approaches utilize a fixed set of learned queries to reconstruct objects within the scene volume. Although these queries are typically updated with image context during training, they remain static at test time, limiting their ability to dynamically adapt specifically to the observed scene. To overcome these limitations, we propose IPFormer, the first method that leverages context-adaptive instance proposals at train and test time to address vision-based 3D Panoptic Scene Completion. Specifically, IPFormer adaptively initializes these queries as panoptic instance proposals derived from image context and further refines them through attention-based encoding and decoding to reason about semantic instance-voxel relationships. Extensive experimental results show that our approach achieves state-of-the-art in-domain performance, exhibits superior zero-shot generalization on out-of-domain data, and achieves a runtime reduction exceeding 14x. These results highlight our introduction of context-adaptive instance proposals as a pioneering effort in addressing vision-based 3D Panoptic Scene Completion. Code available at https://github.com/markus-42/ipformer.", "date": "2025-06-25", "categories": []}
{"id": "2601.05208", "title": "MoE3D: A Mixture-of-Experts Module for 3D Reconstruction", "abstract": "We propose a simple yet effective approach to enhance the performance of feed-forward 3D reconstruction models. Existing methods often struggle near depth discontinuities, where standard regression losses encourage spatial averaging and thus blur sharp boundaries. To address this issue, we introduce a mixture-of-experts formulation that handles uncertainty at depth boundaries by combining multiple smooth depth predictions. A softmax weighting head dynamically selects among these hypotheses on a per-pixel basis. By integrating our mixture model into a pre-trained state-of-the-art 3D model, we achieve a substantial reduction of boundary artifacts and gains in overall reconstruction accuracy. Notably, our approach is highly compute efficient, delivering generalizable improvements even when fine-tuned on a small subset of training data while incurring only negligible additional inference computation, suggesting a promising direction for lightweight and accurate 3D reconstruction.", "date": "2026-01-08", "categories": []}
{"id": "2601.17720", "title": "Advancing Structured Priors for Sparse-Voxel Surface Reconstruction", "abstract": "Reconstructing accurate surfaces with radiance fields has progressed rapidly, yet two promising explicit representations, 3D Gaussian Splatting and sparse-voxel rasterization, exhibit complementary strengths and weaknesses. 3D Gaussian Splatting converges quickly and carries useful geometric priors, but surface fidelity is limited by its point-like parameterization. Sparse-voxel rasterization provides continuous opacity fields and crisp geometry, but its typical uniform dense-grid initialization slows convergence and underutilizes scene structure. We combine the advantages of both by introducing a voxel initialization method that places voxels at plausible locations and with appropriate levels of detail, yielding a strong starting point for per-scene optimization. To further enhance depth consistency without blurring edges, we propose refined depth geometry supervision that converts multi-view cues into direct per-ray depth regularization. Experiments on standard benchmarks demonstrate improvements over prior methods in geometric accuracy, better fine-structure recovery, and more complete surfaces, while maintaining fast convergence.", "date": "2026-01-25", "categories": []}
{"id": "2601.17900", "title": "Revisiting 3D Reconstruction Kernels as Low-Pass Filters", "abstract": "3D reconstruction is to recover 3D signals from the sampled discrete 2D pixels, with the goal to converge continuous 3D spaces. In this paper, we revisit 3D reconstruction from the perspective of signal processing, identifying the periodic spectral extension induced by discrete sampling as the fundamental challenge. Previous 3D reconstruction kernels, such as Gaussians, Exponential functions, and Student's t distributions, serve as the low pass filters to isolate the baseband spectrum. However, their unideal low-pass property results in the overlap of high-frequency components with low-frequency components in the discrete-time signal's spectrum. To this end, we introduce Jinc kernel with an instantaneous drop to zero magnitude exactly at the cutoff frequency, which is corresponding to the ideal low pass filters. As Jinc kernel suffers from low decay speed in the spatial domain, we further propose modulated kernels to strick an effective balance, and achieves superior rendering performance by reconciling spatial efficiency and frequency-domain fidelity. Experimental results have demonstrated the effectiveness of our Jinc and modulated kernels.", "date": "2026-01-25", "categories": []}
{"id": "2601.18475", "title": "LoD-Structured 3D Gaussian Splatting for Streaming Video Reconstruction", "abstract": "Free-Viewpoint Video (FVV) reconstruction enables photorealistic and interactive 3D scene visualization; however, real-time streaming is often bottlenecked by sparse-view inputs, prohibitive training costs, and bandwidth constraints. While recent 3D Gaussian Splatting (3DGS) has advanced FVV due to its superior rendering speed, Streaming Free-Viewpoint Video (SFVV) introduces additional demands for rapid optimization, high-fidelity reconstruction under sparse constraints, and minimal storage footprints. To bridge this gap, we propose StreamLoD-GS, an LoD-based Gaussian Splatting framework designed specifically for SFVV. Our approach integrates three core innovations: 1) an Anchor- and Octree-based LoD-structured 3DGS with a hierarchical Gaussian dropout technique to ensure efficient and stable optimization while maintaining high-quality rendering; 2) a GMM-based motion partitioning mechanism that separates dynamic and static content, refining dynamic regions while preserving background stability; and 3) a quantized residual refinement framework that significantly reduces storage requirements without compromising visual fidelity. Extensive experiments demonstrate that StreamLoD-GS achieves competitive or state-of-the-art performance in terms of quality, efficiency, and storage.", "date": "2026-01-26", "categories": []}
{"id": "2509.13414", "title": "MapAnything: Universal Feed-Forward Metric 3D Reconstruction", "abstract": "We introduce MapAnything, a unified transformer-based feed-forward model that ingests one or more images along with optional geometric inputs such as camera intrinsics, poses, depth, or partial reconstructions, and then directly regresses the metric 3D scene geometry and cameras. MapAnything leverages a factored representation of multi-view scene geometry, i.e., a collection of depth maps, local ray maps, camera poses, and a metric scale factor that effectively upgrades local reconstructions into a globally consistent metric frame. Standardizing the supervision and training across diverse datasets, along with flexible input augmentation, enables MapAnything to address a broad range of 3D vision tasks in a single feed-forward pass, including uncalibrated structure-from-motion, calibrated multi-view stereo, monocular depth estimation, camera localization, depth completion, and more. We provide extensive experimental analyses and model ablations demonstrating that MapAnything outperforms or matches specialist feed-forward models while offering more efficient joint training behavior, thus paving the way toward a universal 3D reconstruction backbone.", "date": "2025-09-16", "categories": []}
{"id": "2411.17392", "title": "NumGrad-Pull: Numerical Gradient Guided Tri-plane Representation for Surface Reconstruction from Point Clouds", "abstract": "Reconstructing continuous surfaces from unoriented and unordered 3D points is a fundamental challenge in computer vision and graphics. Recent advancements address this problem by training neural signed distance functions to pull 3D location queries to their closest points on a surface, following the predicted signed distances and the analytical gradients computed by the network. In this paper, we introduce NumGrad-Pull, leveraging the representation capability of tri-plane structures to accelerate the learning of signed distance functions and enhance the fidelity of local details in surface reconstruction. To further improve the training stability of grid-based tri-planes, we propose to exploit numerical gradients, replacing conventional analytical computations. Additionally, we present a progressive plane expansion strategy to facilitate faster signed distance function convergence and design a data sampling strategy to mitigate reconstruction artifacts. Our extensive experiments across a variety of benchmarks demonstrate the effectiveness and robustness of our approach. Code is available at https://github.com/CuiRuikai/NumGrad-Pull", "date": "2024-11-26", "categories": []}
{"id": "2412.04827", "title": "PanoDreamer: Optimization-Based Single Image to 360 3D Scene With Diffusion", "abstract": "In this paper, we present PanoDreamer, a novel method for producing a coherent 360{\\deg} 3D scene from a single input image. Unlike existing methods that generate the scene sequentially, we frame the problem as single-image panorama and depth estimation. Once the coherent panoramic image and its corresponding depth are obtained, the scene can be reconstructed by inpainting the small occluded regions and projecting them into 3D space. Our key contribution is formulating single-image panorama and depth estimation as two optimization tasks and introducing alternating minimization strategies to effectively solve their objectives. We demonstrate that our approach outperforms existing techniques in single-image 360{\\deg} 3D scene reconstruction in terms of consistency and overall quality.", "date": "2024-12-06", "categories": []}
{"id": "2511.06810", "title": "ConeGS: Error-Guided Densification Using Pixel Cones for Improved Reconstruction With Fewer Primitives", "abstract": "3D Gaussian Splatting (3DGS) achieves state-of-the-art image quality and real-time performance in novel view synthesis but often suffers from a suboptimal spatial distribution of primitives. This issue stems from cloning-based densification, which propagates Gaussians along existing geometry, limiting exploration and requiring many primitives to adequately cover the scene. We present ConeGS, an image-space-informed densification framework that is independent of existing scene geometry state. ConeGS first creates a fast Instant Neural Graphics Primitives (iNGP) reconstruction as a geometric proxy to estimate per-pixel depth. During the subsequent 3DGS optimization, it identifies high-error pixels and inserts new Gaussians along the corresponding viewing cones at the predicted depth values, initializing their size according to the cone diameter. A pre-activation opacity penalty rapidly removes redundant Gaussians, while a primitive budgeting strategy controls the total number of primitives, either by a fixed budget or by adapting to scene complexity, ensuring high reconstruction quality. Experiments show that ConeGS consistently enhances reconstruction quality and rendering performance across Gaussian budgets, with especially strong gains under tight primitive constraints where efficient placement is crucial.", "date": "2025-11-10", "categories": []}
{"id": "2512.23437", "title": "RealX3D: A Physically-Degraded 3D Benchmark for Multi-view Visual Restoration and Reconstruction", "abstract": "We introduce RealX3D, a real-capture benchmark for multi-view visual restoration and 3D reconstruction under diverse physical degradations. RealX3D groups corruptions into four families, including illumination, scattering, occlusion, and blurring, and captures each at multiple severity levels using a unified acquisition protocol that yields pixel-aligned LQ/GT views. Each scene includes high-resolution capture, RAW images, and dense laser scans, from which we derive world-scale meshes and metric depth. Benchmarking a broad range of optimization-based and feed-forward methods shows substantial degradation in reconstruction quality under physical corruptions, underscoring the fragility of current multi-view pipelines in real-world challenging environments.", "date": "2025-12-29", "categories": []}
{"id": "2601.15250", "title": "FlowSSC: Universal Generative Monocular Semantic Scene Completion via One-Step Latent Diffusion", "abstract": "Semantic Scene Completion (SSC) from monocular RGB images is a fundamental yet challenging task due to the inherent ambiguity of inferring occluded 3D geometry from a single view. While feed-forward methods have made progress, they often struggle to generate plausible details in occluded regions and preserve the fundamental spatial relationships of objects. Such accurate generative reasoning capability for the entire 3D space is critical in real-world applications. In this paper, we present FlowSSC, the first generative framework applied directly to monocular semantic scene completion. FlowSSC treats the SSC task as a conditional generation problem and can seamlessly integrate with existing feed-forward SSC methods to significantly boost their performance. To achieve real-time inference without compromising quality, we introduce Shortcut Flow-matching that operates in a compact triplane latent space. Unlike standard diffusion models that require hundreds of steps, our method utilizes a shortcut mechanism to achieve high-fidelity generation in a single step, enabling practical deployment in autonomous systems. Extensive experiments on SemanticKITTI demonstrate that FlowSSC achieves state-of-the-art performance, significantly outperforming existing baselines.", "date": "2026-01-21", "categories": []}
{"id": "2407.16396", "title": "VRP-UDF: Towards Unbiased Learning of Unsigned Distance Functions from Multi-view Images with Volume Rendering Priors", "abstract": "Unsigned distance functions (UDFs) have been a vital representation for open surfaces. With different differentiable renderers, current methods are able to train neural networks to infer a UDF by minimizing the rendering errors with the UDF to the multi-view ground truth. However, these differentiable renderers are mainly handcrafted, which makes them either biased on ray-surface intersections, or sensitive to unsigned distance outliers, or not scalable to large scenes. To resolve these issues, we present a novel differentiable renderer to infer UDFs more accurately. Instead of using handcrafted equations, our differentiable renderer is a neural network which is pre-trained in a data-driven manner. It learns how to render unsigned distances into depth images, leading to a prior knowledge, dubbed volume rendering priors. To infer a UDF for an unseen scene from multiple RGB images, we generalize the learned volume rendering priors to map inferred unsigned distances in alpha blending for RGB image rendering. To reduce the bias of sampling in UDF inference, we utilize an auxiliary point sampling prior as an indicator of ray-surface intersection, and propose novel schemes towards more accurate and uniform sampling near the zero-level sets. We also propose a new strategy that leverages our pretrained volume rendering prior to serve as a general surface refiner, which can be integrated with various Gaussian reconstruction methods to optimize the Gaussian distributions and refine geometric details. Our results show that the learned volume rendering prior is unbiased, robust, scalable, 3D aware, and more importantly, easy to learn. Further experiments show that the volume rendering prior is also a general strategy to enhance other neural implicit representations such as signed distance function and occupancy.", "date": "2024-07-23", "categories": []}
{"id": "2503.08930", "title": "Acoustic Neural 3D Reconstruction Under Pose Drift", "abstract": "We consider the problem of optimizing neural implicit surfaces for 3D reconstruction using acoustic images collected with drifting sensor poses. The accuracy of current state-of-the-art 3D acoustic modeling algorithms is highly dependent on accurate pose estimation; small errors in sensor pose can lead to severe reconstruction artifacts. In this paper, we propose an algorithm that jointly optimizes the neural scene representation and sonar poses. Our algorithm does so by parameterizing the 6DoF poses as learnable parameters and backpropagating gradients through the neural renderer and implicit representation. We validated our algorithm on both real and simulated datasets. It produces high-fidelity 3D reconstructions even under significant pose drift.", "date": "2025-03-11", "categories": []}
{"id": "2601.03024", "title": "SA-ResGS: Self-Augmented Residual 3D Gaussian Splatting for Next Best View Selection", "abstract": "We propose Self-Augmented Residual 3D Gaussian Splatting (SA-ResGS), a novel framework to stabilize uncertainty quantification and enhancing uncertainty-aware supervision in next-best-view (NBV) selection for active scene reconstruction. SA-ResGS improves both the reliability of uncertainty estimates and their effectiveness for supervision by generating Self-Augmented point clouds (SA-Points) via triangulation between a training view and a rasterized extrapolated view, enabling efficient scene coverage estimation. While improving scene coverage through physically guided view selection, SA-ResGS also addresses the challenge of under-supervised Gaussians, exacerbated by sparse and wide-baseline views, by introducing the first residual learning strategy tailored for 3D Gaussian Splatting. This targeted supervision enhances gradient flow in high-uncertainty Gaussians by combining uncertainty-driven filtering with dropout- and hard-negative-mining-inspired sampling. Our contributions are threefold: (1) a physically grounded view selection strategy that promotes efficient and uniform scene coverage; (2) an uncertainty-aware residual supervision scheme that amplifies learning signals for weakly contributing Gaussians, improving training stability and uncertainty estimation across scenes with diverse camera distributions; (3) an implicit unbiasing of uncertainty quantification as a consequence of constrained view selection and residual supervision, which together mitigate conflicting effects of wide-baseline exploration and sparse-view ambiguity in NBV planning. Experiments on active view selection demonstrate that SA-ResGS outperforms state-of-the-art baselines in both reconstruction quality and view selection robustness.", "date": "2026-01-06", "categories": []}
{"id": "2601.04754", "title": "ProFuse: Efficient Cross-View Context Fusion for Open-Vocabulary 3D Gaussian Splatting", "abstract": "We present ProFuse, an efficient context-aware framework for open-vocabulary 3D scene understanding with 3D Gaussian Splatting (3DGS). The pipeline enhances cross-view consistency and intra-mask cohesion within a direct registration setup, adding minimal overhead and requiring no render-supervised fine-tuning. Instead of relying on a pretrained 3DGS scene, we introduce a dense correspondence-guided pre-registration phase that initializes Gaussians with accurate geometry while jointly constructing 3D Context Proposals via cross-view clustering. Each proposal carries a global feature obtained through weighted aggregation of member embeddings, and this feature is fused onto Gaussians during direct registration to maintain per-primitive language coherence across views. With associations established in advance, semantic fusion requires no additional optimization beyond standard reconstruction, and the model retains geometric refinement without densification. ProFuse achieves strong open-vocabulary 3DGS understanding while completing semantic attachment in about five minutes per scene, which is two times faster than SOTA. Additional details are available at our project page https://chiou1203.github.io/ProFuse/.", "date": "2026-01-08", "categories": []}
{"id": "2601.11772", "title": "studentSplat: Your Student Model Learns Single-view 3D Gaussian Splatting", "abstract": "Recent advance in feed-forward 3D Gaussian splatting has enable remarkable multi-view 3D scene reconstruction or single-view 3D object reconstruction but single-view 3D scene reconstruction remain under-explored due to inherited ambiguity in single-view. We present \\textbf{studentSplat}, a single-view 3D Gaussian splatting method for scene reconstruction. To overcome the scale ambiguity and extrapolation problems inherent in novel-view supervision from a single input, we introduce two techniques: 1) a teacher-student architecture where a multi-view teacher model provides geometric supervision to the single-view student during training, addressing scale ambiguity and encourage geometric validity; and 2) an extrapolation network that completes missing scene context, enabling high-quality extrapolation. Extensive experiments show studentSplat achieves state-of-the-art single-view novel-view reconstruction quality and comparable performance to multi-view methods at the scene level. Furthermore, studentSplat demonstrates competitive performance as a self-supervised single-view depth estimation method, highlighting its potential for general single-view 3D understanding tasks.", "date": "2026-01-16", "categories": []}
{"id": "2601.12155", "title": "Inverse Rendering for High-Genus 3D Surface Meshes from Multi-view Images with Persistent Homology Priors", "abstract": "Reconstructing 3D objects from images is inherently an ill-posed problem due to ambiguities in geometry, appearance, and topology. This paper introduces collaborative inverse rendering with persistent homology priors, a novel strategy that leverages topological constraints to resolve these ambiguities. By incorporating priors that capture critical features such as tunnel loops and handle loops, our approach directly addresses the difficulty of reconstructing high-genus surfaces. The collaboration between photometric consistency from multi-view images and homology-based guidance enables recovery of complex high-genus geometry while circumventing catastrophic failures such as collapsing tunnels or losing high-genus structure. Instead of neural networks, our method relies on gradient-based optimization within a mesh-based inverse rendering framework to highlight the role of topological priors. Experimental results show that incorporating persistent homology priors leads to lower Chamfer Distance (CD) and higher Volume IoU compared to state-of-the-art mesh-based methods, demonstrating improved geometric accuracy and robustness against topological failure.", "date": "2026-01-17", "categories": []}
{"id": "2601.13706", "title": "ParkingTwin: Training-Free Streaming 3D Reconstruction for Parking-Lot Digital Twins", "abstract": "High-fidelity parking-lot digital twins provide essential priors for path planning, collision checking, and perception validation in Automated Valet Parking (AVP). Yet robot-oriented reconstruction faces a trilemma: sparse forward-facing views cause weak parallax and ill-posed geometry; dynamic occlusions and extreme lighting hinder stable texture fusion; and neural rendering typically needs expensive offline optimization, violating edge-side streaming constraints. We propose ParkingTwin, a training-free, lightweight system for online streaming 3D reconstruction. First, OSM-prior-driven geometric construction uses OpenStreetMap semantic topology to directly generate a metric-consistent TSDF, replacing blind geometric search with deterministic mapping and avoiding costly optimization. Second, geometry-aware dynamic filtering employs a quad-modal constraint field (normal/height/depth consistency) to reject moving vehicles and transient occlusions in real time. Third, illumination-robust fusion in CIELAB decouples luminance and chromaticity via adaptive L-channel weighting and depth-gradient suppression, reducing seams under abrupt lighting changes. ParkingTwin runs at 30+ FPS on an entry-level GTX 1660. On a 68,000 m^2 real-world dataset, it achieves SSIM 0.87 (+16.0%), delivers about 15x end-to-end speedup, and reduces GPU memory by 83.3% compared with state-of-the-art 3D Gaussian Splatting (3DGS) that typically requires high-end GPUs (RTX 4090D). The system outputs explicit triangle meshes compatible with Unity/Unreal digital-twin pipelines. Project page: https://mihoutao-liu.github.io/ParkingTwin/", "date": "2026-01-20", "categories": []}
{"id": "2601.14208", "title": "Rig-Aware 3D Reconstruction of Vehicle Undercarriages using Gaussian Splatting", "abstract": "Inspecting the undercarriage of used vehicles is a labor-intensive task that requires inspectors to crouch or crawl underneath each vehicle to thoroughly examine it. Additionally, online buyers rarely see undercarriage photos. We present an end-to-end pipeline that utilizes a three-camera rig to capture videos of the undercarriage as the vehicle drives over it, and produces an interactive 3D model of the undercarriage. The 3D model enables inspectors and customers to rotate, zoom, and slice through the undercarriage, allowing them to detect rust, leaks, or impact damage in seconds, thereby improving both workplace safety and buyer confidence. Our primary contribution is a rig-aware Structure-from-Motion (SfM) pipeline specifically designed to overcome the challenges of wide-angle lens distortion and low-parallax scenes. Our method overcomes the challenges of wide-angle lens distortion and low-parallax scenes by integrating precise camera calibration, synchronized video streams, and strong geometric priors from the camera rig. We use a constrained matching strategy with learned components, the DISK feature extractor, and the attention-based LightGlue matcher to generate high-quality sparse point clouds that are often unattainable with standard SfM pipelines. These point clouds seed the Gaussian splatting process to generate photorealistic undercarriage models that render in real-time. Our experiments and ablation studies demonstrate that our design choices are essential to achieve state-of-the-art quality.", "date": "2026-01-20", "categories": []}
{"id": "2312.09076", "title": "ProSGNeRF: Progressive Dynamic Neural Scene Graph with Frequency Modulated Foundation Model in Urban Scenes", "abstract": "Implicit neural representation has demonstrated promising results in 3D reconstruction on various scenes. However, existing approaches either struggle to model fast-moving objects or are incapable of handling large-scale camera ego-motions in urban environments. This leads to low-quality synthesized views of the large-scale urban scenes. In this paper, we aim to jointly solve the problems caused by large-scale scenes and fast-moving vehicles, which are more practical and challenging. To this end, we propose a progressive scene graph network architecture to learn the local scene representations of dynamic objects and global urban scenes. The progressive learning architecture dynamically allocates a new local scene graph trained on frames within a temporal window, with the window size automatically determined, allowing us to scale up the representation to arbitrarily large scenes. Besides, according to our observations, the training views of dynamic objects are relatively sparse according to rapid movements, which leads to a significant decline in reconstruction accuracy for dynamic objects. Therefore, we utilize a foundation model network to encode the latent code. Specifically, we leverage the generalization capability of the visual foundation model DINOv2 to extract appearance and shape codes, and train the network on a large-scale urban scene object dataset to enhance its prior modeling ability for handling sparse-view dynamic inputs. In parallel, we introduce a frequency-modulated module that regularizes the frequency spectrum of objects, thereby addressing the challenge of modeling sparse image inputs from a frequency-domain perspective. Experimental results demonstrate that our method achieves state-of-the-art view synthesis accuracy, object manipulation, and scene roaming ability in various scenes.", "date": "2023-12-14", "categories": []}
{"id": "2601.11030", "title": "IDDR-NGP: Incorporating Detectors for Distractor Removal with Instant Neural Radiance Field", "abstract": "This paper presents the first unified distractor removal method, named IDDR-NGP, which directly operates on Instant-NPG. The method is able to remove a wide range of distractors in 3D scenes, such as snowflakes, confetti, defoliation and petals, whereas existing methods usually focus on a specific type of distractors. By incorporating implicit 3D representations with 2D detectors, we demonstrate that it is possible to efficiently restore 3D scenes from multiple corrupted images. We design the learned perceptual image patch similarity~( LPIPS) loss and the multi-view compensation loss (MVCL) to jointly optimize the rendering results of IDDR-NGP, which could aggregate information from multi-view corrupted images. All of them can be trained in an end-to-end manner to synthesize high-quality 3D scenes. To support the research on distractors removal in implicit 3D representations, we build a new benchmark dataset that consists of both synthetic and real-world distractors. To validate the effectiveness and robustness of IDDR-NGP, we provide a wide range of distractors with corresponding annotated labels added to both realistic and synthetic scenes. Extensive experimental results demonstrate the effectiveness and robustness of IDDR-NGP in removing multiple types of distractors. In addition, our approach achieves results comparable with the existing SOTA desnow methods and is capable of accurately removing both realistic and synthetic distractors.", "date": "2026-01-16", "categories": []}
{"id": "2510.26117", "title": "JOGS: Joint Optimization of Pose Estimation and 3D Gaussian Splatting", "abstract": "Traditional novel view synthesis methods heavily rely on external camera pose estimation tools such as COLMAP, which often introduce computational bottlenecks and propagate errors. To address these challenges, we propose a unified framework that jointly optimizes 3D Gaussian points and camera poses without requiring pre-calibrated inputs. Our approach iteratively refines 3D Gaussian parameters and updates camera poses through a novel co-optimization strategy, ensuring simultaneous improvements in scene reconstruction fidelity and pose estimation accuracy. The key innovation lies in decoupling the joint optimization into two interleaved phases: first, updating 3D Gaussian parameters via differentiable rendering with fixed poses, and second, refining camera poses using a customized 3D optical flow algorithm that incorporates geometric and photometric constraints. This formulation progressively reduces projection errors, particularly in challenging scenarios with large viewpoint variations and sparse feature distributions, where traditional methods struggle. Extensive evaluations on multiple datasets demonstrate that our approach significantly outperforms existing COLMAP-free techniques in reconstruction quality, and also surpasses the standard COLMAP-based baseline in general.", "date": "2025-10-30", "categories": []}
{"id": "2601.09499", "title": "V-DPM: 4D Video Reconstruction with Dynamic Point Maps", "abstract": "Powerful 3D representations such as DUSt3R invariant point maps, which encode 3D shape and camera parameters, have significantly advanced feed forward 3D reconstruction. While point maps assume static scenes, Dynamic Point Maps (DPMs) extend this concept to dynamic 3D content by additionally representing scene motion. However, existing DPMs are limited to image pairs and, like DUSt3R, require post processing via optimization when more than two views are involved. We argue that DPMs are more useful when applied to videos and introduce V-DPM to demonstrate this. First, we show how to formulate DPMs for video input in a way that maximizes representational power, facilitates neural prediction, and enables reuse of pretrained models. Second, we implement these ideas on top of VGGT, a recent and powerful 3D reconstructor. Although VGGT was trained on static scenes, we show that a modest amount of synthetic data is sufficient to adapt it into an effective V-DPM predictor. Our approach achieves state of the art performance in 3D and 4D reconstruction for dynamic scenes. In particular, unlike recent dynamic extensions of VGGT such as P3, DPMs recover not only dynamic depth but also the full 3D motion of every point in the scene.", "date": "2026-01-14", "categories": []}
{"id": "2506.05280", "title": "Unifying Appearance Codes and Bilateral Grids for Driving Scene Gaussian Splatting", "abstract": "Neural rendering techniques, including NeRF and Gaussian Splatting (GS), rely on photometric consistency to produce high-quality reconstructions. However, in real-world scenarios, it is challenging to guarantee perfect photometric consistency in acquired images. Appearance codes have been widely used to address this issue, but their modeling capability is limited, as a single code is applied to the entire image. Recently, the bilateral grid was introduced to perform pixel-wise color mapping, but it is difficult to optimize and constrain effectively. In this paper, we propose a novel multi-scale bilateral grid that unifies appearance codes and bilateral grids. We demonstrate that this approach significantly improves geometric accuracy in dynamic, decoupled autonomous driving scene reconstruction, outperforming both appearance codes and bilateral grids. This is crucial for autonomous driving, where accurate geometry is important for obstacle avoidance and control. Our method shows strong results across four datasets: Waymo, NuScenes, Argoverse, and PandaSet. We further demonstrate that the improvement in geometry is driven by the multi-scale bilateral grid, which effectively reduces floaters caused by photometric inconsistency.", "date": "2025-06-05", "categories": []}
{"id": "2601.08175", "title": "CogniMap3D: Cognitive 3D Mapping and Rapid Retrieval", "abstract": "We present CogniMap3D, a bioinspired framework for dynamic 3D scene understanding and reconstruction that emulates human cognitive processes. Our approach maintains a persistent memory bank of static scenes, enabling efficient spatial knowledge storage and rapid retrieval. CogniMap3D integrates three core capabilities: a multi-stage motion cue framework for identifying dynamic objects, a cognitive mapping system for storing, recalling, and updating static scenes across multiple visits, and a factor graph optimization strategy for refining camera poses. Given an image stream, our model identifies dynamic regions through motion cues with depth and camera pose priors, then matches static elements against its memory bank. When revisiting familiar locations, CogniMap3D retrieves stored scenes, relocates cameras, and updates memory with new observations. Evaluations on video depth estimation, camera pose reconstruction, and 3D mapping tasks demonstrate its state-of-the-art performance, while effectively supporting continuous scene understanding across extended sequences and multiple visits.", "date": "2026-01-13", "categories": []}
{"id": "2405.20031", "title": "MG-SLAM: Structure Gaussian Splatting SLAM with Manhattan World Hypothesis", "abstract": "Gaussian Splatting SLAMs have made significant advancements in improving the efficiency and fidelity of real-time reconstructions. However, these systems often encounter incomplete reconstructions in complex indoor environments, characterized by substantial holes due to unobserved geometry caused by obstacles or limited view angles. To address this challenge, we present Manhattan Gaussian SLAM, an RGB-D system that leverages the Manhattan World hypothesis to enhance geometric accuracy and completeness. By seamlessly integrating fused line segments derived from structured scenes, our method ensures robust tracking in textureless indoor areas. Moreover, The extracted lines and planar surface assumption allow strategic interpolation of new Gaussians in regions of missing geometry, enabling efficient scene completion. Extensive experiments conducted on both synthetic and real-world scenes demonstrate that these advancements enable our method to achieve state-of-the-art performance, marking a substantial improvement in the capabilities of Gaussian SLAM systems.", "date": "2024-05-30", "categories": []}
{"id": "2502.02283", "title": "GP-GS: Gaussian Processes Densification for 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) enables photorealistic rendering but suffers from artefacts due to sparse Structure-from-Motion (SfM) initialisation. To address this limitation, we propose GP-GS, a Gaussian Process (GP) based densification framework for 3DGS optimisation. GP-GS formulates point cloud densification as a continuous regression problem, where a GP learns a local mapping from 2D pixel coordinates to 3D position and colour attributes. An adaptive neighbourhood-based sampling strategy generates candidate pixels for inference, while GP-predicted uncertainty is used to filter unreliable predictions, reducing noise and preserving geometric structure. Extensive experiments on synthetic and real-world benchmarks demonstrate that GP-GS consistently improves reconstruction quality and rendering fidelity, achieving up to 1.12 dB PSNR improvement over strong baselines.", "date": "2025-02-04", "categories": []}
{"id": "2505.01235", "title": "Compensating Spatiotemporally Inconsistent Observations for Online Dynamic 3D Gaussian Splatting", "abstract": "Online reconstruction of dynamic scenes is significant as it enables learning scenes from live-streaming video inputs, while existing offline dynamic reconstruction methods rely on recorded video inputs. However, previous online reconstruction approaches have primarily focused on efficiency and rendering quality, overlooking the temporal consistency of their results, which often contain noticeable artifacts in static regions. This paper identifies that errors such as noise in real-world recordings affect temporal inconsistency in online reconstruction. We propose a method that enhances temporal consistency in online reconstruction from observations with temporal inconsistency which is inevitable in cameras. We show that our method restores the ideal observation by subtracting the learned error. We demonstrate that applying our method to various baselines significantly enhances both temporal consistency and rendering quality across datasets. Code, video results, and checkpoints are available at https://bbangsik13.github.io/OR2.", "date": "2025-05-02", "categories": []}
{"id": "2506.04115", "title": "Multi-view Surface Reconstruction Using Normal and Reflectance Cues", "abstract": "Achieving high-fidelity 3D surface reconstruction while preserving fine details remains challenging, especially in the presence of materials with complex reflectance properties and without a dense-view setup. In this paper, we introduce a versatile framework that incorporates multi-view normal and optionally reflectance maps into radiance-based surface reconstruction. Our approach employs a pixel-wise joint re-parametrization of reflectance and surface normals, representing them as a vector of radiances under simulated, varying illumination. This formulation enables seamless incorporation into standard surface reconstruction pipelines, such as traditional multi-view stereo (MVS) frameworks or modern neural volume rendering (NVR) ones. Combined with the latter, our approach achieves state-of-the-art performance on multi-view photometric stereo (MVPS) benchmark datasets, including DiLiGenT-MV, LUCES-MV and Skoltech3D. In particular, our method excels in reconstructing fine-grained details and handling challenging visibility conditions. The present paper is an extended version of the earlier conference paper by Brument et al. (in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024), featuring an accelerated and more robust algorithm as well as a broader empirical evaluation. The code and data relative to this article is available at https://github.com/RobinBruneau/RNb-NeuS2.", "date": "2025-06-04", "categories": []}
{"id": "2508.13287", "title": "InnerGS: Internal Scenes Reconstruction and Segmentation via Factorized 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has recently gained popularity for efficient scene rendering by representing scenes as explicit sets of anisotropic 3D Gaussians. However, most existing work focuses primarily on modeling external surfaces. In this work, we target the reconstruction of internal scenes, which is crucial for applications that require a deep understanding of an object's interior. By directly modeling a continuous volumetric density through the inner 3D Gaussian distribution, our model effectively reconstructs smooth and detailed internal structures from sparse sliced data. Beyond high-fidelity reconstruction, we further demonstrate the framework's potential for downstream tasks such as segmentation. By integrating language features, we extend our approach to enable text-guided segmentation of medical scenes via natural language queries. Our approach eliminates the need for camera poses, is plug-and-play, and is inherently compatible with any data modalities. We provide cuda implementation at: https://github.com/Shuxin-Liang/InnerGS.", "date": "2025-08-18", "categories": []}
{"id": "2509.20607", "title": "Reflect3r: Single-View 3D Stereo Reconstruction Aided by Mirror Reflections", "abstract": "Mirror reflections are common in everyday environments and can provide stereo information within a single capture, as the real and reflected virtual views are visible simultaneously. We exploit this property by treating the reflection as an auxiliary view and designing a transformation that constructs a physically valid virtual camera, allowing direct pixel-domain generation of the virtual view while adhering to the real-world imaging process. This enables a multi-view stereo setup from a single image, simplifying the imaging process, making it compatible with powerful feed-forward reconstruction models for generalizable and robust 3D reconstruction. To further exploit the geometric symmetry introduced by mirrors, we propose a symmetric-aware loss to refine pose estimation. Our framework also naturally extends to dynamic scenes, where each frame contains a mirror reflection, enabling efficient per-frame geometry recovery. For quantitative evaluation, we provide a fully customizable synthetic dataset of 16 Blender scenes, each with ground-truth point clouds and camera poses. Extensive experiments on real-world data and synthetic data are conducted to illustrate the effectiveness of our method.", "date": "2025-09-24", "categories": []}
{"id": "2601.05368", "title": "MOSAIC-GS: Monocular Scene Reconstruction via Advanced Initialization for Complex Dynamic Environments", "abstract": "We present MOSAIC-GS, a novel, fully explicit, and computationally efficient approach for high-fidelity dynamic scene reconstruction from monocular videos using Gaussian Splatting. Monocular reconstruction is inherently ill-posed due to the lack of sufficient multiview constraints, making accurate recovery of object geometry and temporal coherence particularly challenging. To address this, we leverage multiple geometric cues, such as depth, optical flow, dynamic object segmentation, and point tracking. Combined with rigidity-based motion constraints, these cues allow us to estimate preliminary 3D scene dynamics during an initialization stage. Recovering scene dynamics prior to the photometric optimization reduces reliance on motion inference from visual appearance alone, which is often ambiguous in monocular settings. To enable compact representations, fast training, and real-time rendering while supporting non-rigid deformations, the scene is decomposed into static and dynamic components. Each Gaussian in the dynamic part of the scene is assigned a trajectory represented as time-dependent Poly-Fourier curve for parameter-efficient motion encoding. We demonstrate that MOSAIC-GS achieves substantially faster optimization and rendering compared to existing methods, while maintaining reconstruction quality on par with state-of-the-art approaches across standard monocular dynamic scene benchmarks.", "date": "2026-01-08", "categories": []}
{"id": "2601.00913", "title": "Clean-GS: Semantic Mask-Guided Pruning for 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting produces high-quality scene reconstructions but generates hundreds of thousands of spurious Gaussians (floaters) scattered throughout the environment. These artifacts obscure objects of interest and inflate model sizes, hindering deployment in bandwidth-constrained applications. We present Clean-GS, a method for removing background clutter and floaters from 3DGS reconstructions using sparse semantic masks. Our approach combines whitelist-based spatial filtering with color-guided validation and outlier removal to achieve 60-80\\% model compression while preserving object quality. Unlike existing 3DGS pruning methods that rely on global importance metrics, Clean-GS uses semantic information from as few as 3 segmentation masks (1\\% of views) to identify and remove Gaussians not belonging to the target object. Our multi-stage approach consisting of (1) whitelist filtering via projection to masked regions, (2) depth-buffered color validation, and (3) neighbor-based outlier removal isolates monuments and objects from complex outdoor scenes. Experiments on Tanks and Temples show that Clean-GS reduces file sizes from 125MB to 47MB while maintaining rendering quality, making 3DGS models practical for web deployment and AR/VR applications. Our code is available at https://github.com/smlab-niser/clean-gs", "date": "2026-01-01", "categories": []}
{"id": "2601.00939", "title": "ShadowGS: Shadow-Aware 3D Gaussian Splatting for Satellite Imagery", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a novel paradigm for 3D reconstruction from satellite imagery. However, in multi-temporal satellite images, prevalent shadows exhibit significant inconsistencies due to varying illumination conditions. To address this, we propose ShadowGS, a novel framework based on 3DGS. It leverages a physics-based rendering equation from remote sensing, combined with an efficient ray marching technique, to precisely model geometrically consistent shadows while maintaining efficient rendering. Additionally, it effectively disentangles different illumination components and apparent attributes in the scene. Furthermore, we introduce a shadow consistency constraint that significantly enhances the geometric accuracy of 3D reconstruction. We also incorporate a novel shadow map prior to improve performance with sparse-view inputs. Extensive experiments demonstrate that ShadowGS outperforms current state-of-the-art methods in shadow decoupling accuracy, 3D reconstruction precision, and novel view synthesis quality, with only a few minutes of training. ShadowGS exhibits robust performance across various settings, including RGB, pansharpened, and sparse-view satellite inputs.", "date": "2026-01-04", "categories": []}
{"id": "2601.01204", "title": "XStreamVGGT: Extremely Memory-Efficient Streaming Vision Geometry Grounded Transformer with KV Cache Compression", "abstract": "Learning-based 3D visual geometry models have benefited substantially from large-scale transformers. Among these, StreamVGGT leverages frame-wise causal attention for strong streaming reconstruction, but suffers from unbounded KV cache growth, leading to escalating memory consumption and inference latency as input frames accumulate. We propose XStreamVGGT, a tuning-free approach that systematically compresses the KV cache through joint pruning and quantization, enabling extremely memory-efficient streaming inference. Specifically, redundant KVs originating from multi-view inputs are pruned through efficient token importance identification, enabling a fixed memory budget. Leveraging the unique distribution of KV tensors, we incorporate KV quantization to further reduce memory consumption. Extensive evaluations show that XStreamVGGT achieves mostly negligible performance degradation while substantially reducing memory usage by 4.42$\\times$ and accelerating inference by 5.48$\\times$, enabling scalable and practical streaming 3D applications. The code is available at https://github.com/ywh187/XStreamVGGT/.", "date": "2026-01-03", "categories": []}
{"id": "2601.01222", "title": "UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass", "abstract": "We present UniSH, a unified, feed-forward framework for joint metric-scale 3D scene and human reconstruction. A key challenge in this domain is the scarcity of large-scale, annotated real-world data, forcing a reliance on synthetic datasets. This reliance introduces a significant sim-to-real domain gap, leading to poor generalization, low-fidelity human geometry, and poor alignment on in-the-wild videos. To address this, we propose an innovative training paradigm that effectively leverages unlabeled in-the-wild data. Our framework bridges strong, disparate priors from scene reconstruction and HMR, and is trained with two core components: (1) a robust distillation strategy to refine human surface details by distilling high-frequency details from an expert depth model, and (2) a two-stage supervision scheme, which first learns coarse localization on synthetic data, then fine-tunes on real data by directly optimizing the geometric correspondence between the SMPL mesh and the human point cloud. This approach enables our feed-forward model to jointly recover high-fidelity scene geometry, human point clouds, camera parameters, and coherent, metric-scale SMPL bodies, all in a single forward pass. Extensive experiments demonstrate that our model achieves state-of-the-art performance on human-centric scene reconstruction and delivers highly competitive results on global human motion estimation, comparing favorably against both optimization-based frameworks and HMR-only methods. Project page: https://murphylmf.github.io/UniSH/", "date": "2026-01-03", "categories": []}
{"id": "2601.01386", "title": "ParkGaussian: Surround-view 3D Gaussian Splatting for Autonomous Parking", "abstract": "Parking is a critical task for autonomous driving systems (ADS), with unique challenges in crowded parking slots and GPS-denied environments. However, existing works focus on 2D parking slot perception, mapping, and localization, 3D reconstruction remains underexplored, which is crucial for capturing complex spatial geometry in parking scenarios. Naively improving the visual quality of reconstructed parking scenes does not directly benefit autonomous parking, as the key entry point for parking is the slots perception module. To address these limitations, we curate the first benchmark named ParkRecon3D, specifically designed for parking scene reconstruction. It includes sensor data from four surround-view fisheye cameras with calibrated extrinsics and dense parking slot annotations. We then propose ParkGaussian, the first framework that integrates 3D Gaussian Splatting (3DGS) for parking scene reconstruction. To further improve the alignment between reconstruction and downstream parking slot detection, we introduce a slot-aware reconstruction strategy that leverages existing parking perception methods to enhance the synthesis quality of slot regions. Experiments on ParkRecon3D demonstrate that ParkGaussian achieves state-of-the-art reconstruction quality and better preserves perception consistency for downstream tasks. The code and dataset will be released at: https://github.com/wm-research/ParkGaussian", "date": "2026-01-04", "categories": []}
{"id": "2601.02072", "title": "SketchRodGS: Sketch-based Extraction of Slender Geometries for Animating Gaussian Splatting Scenes", "abstract": "Physics simulation of slender elastic objects often requires discretization as a polyline. However, constructing a polyline from Gaussian splatting is challenging as Gaussian splatting lacks connectivity information and the configuration of Gaussian primitives contains much noise. This paper presents a method to extract a polyline representation of the slender part of the objects in a Gaussian splatting scene from the user's sketching input. Our method robustly constructs a polyline mesh that represents the slender parts using the screen-space shortest path analysis that can be efficiently solved using dynamic programming. We demonstrate the effectiveness of our approach in several in-the-wild examples.", "date": "2026-01-05", "categories": []}
{"id": "2601.02102", "title": "360-GeoGS: Geometrically Consistent Feed-Forward 3D Gaussian Splatting Reconstruction for 360 Images", "abstract": "3D scene reconstruction is fundamental for spatial intelligence applications such as AR, robotics, and digital twins. Traditional multi-view stereo struggles with sparse viewpoints or low-texture regions, while neural rendering approaches, though capable of producing high-quality results, require per-scene optimization and lack real-time efficiency. Explicit 3D Gaussian Splatting (3DGS) enables efficient rendering, but most feed-forward variants focus on visual quality rather than geometric consistency, limiting accurate surface reconstruction and overall reliability in spatial perception tasks. This paper presents a novel feed-forward 3DGS framework for 360 images, capable of generating geometrically consistent Gaussian primitives while maintaining high rendering quality. A Depth-Normal geometric regularization is introduced to couple rendered depth gradients with normal information, supervising Gaussian rotation, scale, and position to improve point cloud and surface accuracy. Experimental results show that the proposed method maintains high rendering quality while significantly improving geometric consistency, providing an effective solution for 3D reconstruction in spatial perception tasks.", "date": "2026-01-05", "categories": []}
{"id": "2601.02281", "title": "InfiniteVGGT: Visual Geometry Grounded Transformer for Endless Streams", "abstract": "The grand vision of enabling persistent, large-scale 3D visual geometry understanding is shackled by the irreconcilable demands of scalability and long-term stability. While offline models like VGGT achieve inspiring geometry capability, their batch-based nature renders them irrelevant for live systems. Streaming architectures, though the intended solution for live operation, have proven inadequate. Existing methods either fail to support truly infinite-horizon inputs or suffer from catastrophic drift over long sequences. We shatter this long-standing dilemma with InfiniteVGGT, a causal visual geometry transformer that operationalizes the concept of a rolling memory through a bounded yet adaptive and perpetually expressive KV cache. Capitalizing on this, we devise a training-free, attention-agnostic pruning strategy that intelligently discards obsolete information, effectively ``rolling'' the memory forward with each new frame. Fully compatible with FlashAttention, InfiniteVGGT finally alleviates the compromise, enabling infinite-horizon streaming while outperforming existing streaming methods in long-term stability. The ultimate test for such a system is its performance over a truly infinite horizon, a capability that has been impossible to rigorously validate due to the lack of extremely long-term, continuous benchmarks. To address this critical gap, we introduce the Long3D benchmark, which, for the first time, enables a rigorous evaluation of continuous 3D geometry estimation on sequences about 10,000 frames. This provides the definitive evaluation platform for future research in long-term 3D geometry understanding. Code is available at: https://github.com/AutoLab-SAI-SJTU/InfiniteVGGT", "date": "2026-01-05", "categories": []}
{"id": "2601.00796", "title": "AdaGaR: Adaptive Gabor Representation for Dynamic Scene Reconstruction", "abstract": "Reconstructing dynamic 3D scenes from monocular videos requires simultaneously capturing high-frequency appearance details and temporally continuous motion. Existing methods using single Gaussian primitives are limited by their low-pass filtering nature, while standard Gabor functions introduce energy instability. Moreover, lack of temporal continuity constraints often leads to motion artifacts during interpolation. We propose AdaGaR, a unified framework addressing both frequency adaptivity and temporal continuity in explicit dynamic scene modeling. We introduce Adaptive Gabor Representation, extending Gaussians through learnable frequency weights and adaptive energy compensation to balance detail capture and stability. For temporal continuity, we employ Cubic Hermite Splines with Temporal Curvature Regularization to ensure smooth motion evolution. An Adaptive Initialization mechanism combining depth estimation, point tracking, and foreground masks establishes stable point cloud distributions in early training. Experiments on Tap-Vid DAVIS demonstrate state-of-the-art performance (PSNR 35.49, SSIM 0.9433, LPIPS 0.0723) and strong generalization across frame interpolation, depth consistency, video editing, and stereo view synthesis. Project page: https://jiewenchan.github.io/AdaGaR/", "date": "2026-01-02", "categories": []}
{"id": "2508.02261", "title": "SplatSSC: Decoupled Depth-Guided Gaussian Splatting for Semantic Scene Completion", "abstract": "Monocular 3D Semantic Scene Completion (SSC) is a challenging yet promising task that aims to infer dense geometric and semantic descriptions of a scene from a single image. While recent object-centric paradigms significantly improve efficiency by leveraging flexible 3D Gaussian primitives, they still rely heavily on a large number of randomly initialized primitives, which inevitably leads to 1) inefficient primitive initialization and 2) outlier primitives that introduce erroneous artifacts. In this paper, we propose SplatSSC, a novel framework that resolves these limitations with a depth-guided initialization strategy and a principled Gaussian aggregator. Instead of random initialization, SplatSSC utilizes a dedicated depth branch composed of a Group-wise Multi-scale Fusion (GMF) module, which integrates multi-scale image and depth features to generate a sparse yet representative set of initial Gaussian primitives. To mitigate noise from outlier primitives, we develop the Decoupled Gaussian Aggregator (DGA), which enhances robustness by decomposing geometric and semantic predictions during the Gaussian-to-voxel splatting process. Complemented with a specialized Probability Scale Loss, our method achieves state-of-the-art performance on the Occ-ScanNet dataset, outperforming prior approaches by over 6.3% in IoU and 4.1% in mIoU, while reducing both latency and memory cost by more than 9.3%.", "date": "2025-08-04", "categories": []}
{"id": "2512.23998", "title": "Improved 3D Gaussian Splatting of Unknown Spacecraft Structure Using Space Environment Illumination Knowledge", "abstract": "This work presents a novel pipeline to recover the 3D structure of an unknown target spacecraft from a sequence of images captured during Rendezvous and Proximity Operations (RPO) in space. The target's geometry and appearance are represented as a 3D Gaussian Splatting (3DGS) model. However, learning 3DGS requires static scenes, an assumption in contrast to dynamic lighting conditions encountered in spaceborne imagery. The trained 3DGS model can also be used for camera pose estimation through photometric optimization. Therefore, in addition to recovering a geometrically accurate 3DGS model, the photometric accuracy of the rendered images is imperative to downstream pose estimation tasks during the RPO process. This work proposes to incorporate the prior knowledge of the Sun's position, estimated and maintained by the servicer spacecraft, into the training pipeline for improved photometric quality of 3DGS rasterization. Experimental studies demonstrate the effectiveness of the proposed solution, as 3DGS models trained on a sequence of images learn to adapt to rapidly changing illumination conditions in space and reflect global shadowing and self-occlusion.", "date": "2025-12-30", "categories": []}
{"id": "2507.04749", "title": "MatDecompSDF: High-Fidelity 3D Shape and PBR Material Decomposition from Multi-View Images", "abstract": "We present MatDecompSDF, a novel framework for recovering high-fidelity 3D shapes and decomposing their physically-based material properties from multi-view images. The core challenge of inverse rendering lies in the ill-posed disentanglement of geometry, materials, and illumination from 2D observations. Our method addresses this by jointly optimizing three neural components: a neural Signed Distance Function (SDF) to represent complex geometry, a spatially-varying neural field for predicting PBR material parameters (albedo, roughness, metallic), and an MLP-based model for capturing unknown environmental lighting. The key to our approach is a physically-based differentiable rendering layer that connects these 3D properties to the input images, allowing for end-to-end optimization. We introduce a set of carefully designed physical priors and geometric regularizations, including a material smoothness loss and an Eikonal loss, to effectively constrain the problem and achieve robust decomposition. Extensive experiments on both synthetic and real-world datasets (e.g., DTU) demonstrate that MatDecompSDF surpasses state-of-the-art methods in geometric accuracy, material fidelity, and novel view synthesis. Crucially, our method produces editable and relightable assets that can be seamlessly integrated into standard graphics pipelines, validating its practical utility for digital content creation.", "date": "2025-07-07", "categories": []}
{"id": "2510.17719", "title": "RaindropGS: A Benchmark for 3D Gaussian Splatting under Raindrop Conditions", "abstract": "3D Gaussian Splatting (3DGS) under raindrop conditions suffers from severe occlusions and optical distortions caused by raindrop contamination on the camera lens, substantially degrading reconstruction quality. Existing benchmarks typically evaluate 3DGS using synthetic raindrop images with known camera poses (constrained images), assuming ideal conditions. However, in real-world scenarios, raindrops often interfere with accurate camera pose estimation and point cloud initialization. Moreover, a significant domain gap between synthetic and real raindrops further impairs generalization. To tackle these issues, we introduce RaindropGS, a comprehensive benchmark designed to evaluate the full 3DGS pipeline-from unconstrained, raindrop-corrupted images to clear 3DGS reconstructions. Specifically, the whole benchmark pipeline consists of three parts: data preparation, data processing, and raindrop-aware 3DGS evaluation, including types of raindrop interference, camera pose estimation and point cloud initialization, single image rain removal comparison, and 3D Gaussian training comparison. First, we collect a real-world raindrop reconstruction dataset, in which each scene contains three aligned image sets: raindrop-focused, background-focused, and rain-free ground truth, enabling a comprehensive evaluation of reconstruction quality under different focus conditions. Through comprehensive experiments and analyses, we reveal critical insights into the performance limitations of existing 3DGS methods on unconstrained raindrop images and the varying impact of different pipeline components: the impact of camera focus position on 3DGS reconstruction performance, and the interference caused by inaccurate pose and point cloud initialization on reconstruction. These insights establish clear directions for developing more robust 3DGS methods under raindrop conditions.", "date": "2025-10-20", "categories": []}
{"id": "2510.23928", "title": "Adaptive Keyframe Selection for Scalable 3D Scene Reconstruction in Dynamic Environments", "abstract": "In this paper, we propose an adaptive keyframe selection method for improved 3D scene reconstruction in dynamic environments. The proposed method integrates two complementary modules: an error-based selection module utilizing photometric and structural similarity (SSIM) errors, and a momentum-based update module that dynamically adjusts keyframe selection thresholds according to scene motion dynamics. By dynamically curating the most informative frames, our approach addresses a key data bottleneck in real-time perception. This allows for the creation of high-quality 3D world representations from a compressed data stream, a critical step towards scalable robot learning and deployment in complex, dynamic environments. Experimental results demonstrate significant improvements over traditional static keyframe selection strategies, such as fixed temporal intervals or uniform frame skipping. These findings highlight a meaningful advancement toward adaptive perception systems that can dynamically respond to complex and evolving visual scenes. We evaluate our proposed adaptive keyframe selection module on two recent state-of-the-art 3D reconstruction networks, Spann3r and CUT3R, and observe consistent improvements in reconstruction quality across both frameworks. Furthermore, an extensive ablation study confirms the effectiveness of each individual component in our method, underlining their contribution to the overall performance gains.", "date": "2025-10-27", "categories": []}
{"id": "2512.10369", "title": "Breaking the Vicious Cycle: Coherent 3D Gaussian Splatting from Sparse and Motion-Blurred Views", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a state-of-the-art method for novel view synthesis. However, its performance heavily relies on dense, high-quality input imagery, an assumption that is often violated in real-world applications, where data is typically sparse and motion-blurred. These two issues create a vicious cycle: sparse views ignore the multi-view constraints necessary to resolve motion blur, while motion blur erases high-frequency details crucial for aligning the limited views. Thus, reconstruction often fails catastrophically, with fragmented views and a low-frequency bias. To break this cycle, we introduce CoherentGS, a novel framework for high-fidelity 3D reconstruction from sparse and blurry images. Our key insight is to address these compound degradations using a dual-prior strategy. Specifically, we combine two pre-trained generative models: a specialized deblurring network for restoring sharp details and providing photometric guidance, and a diffusion model that offers geometric priors to fill in unobserved regions of the scene. This dual-prior strategy is supported by several key techniques, including a consistency-guided camera exploration module that adaptively guides the generative process, and a depth regularization loss that ensures geometric plausibility. We evaluate CoherentGS through both quantitative and qualitative experiments on synthetic and real-world scenes, using as few as 3, 6, and 9 input views. Our results demonstrate that CoherentGS significantly outperforms existing methods, setting a new state-of-the-art for this challenging task. The code and video demos are available at https://potatobigroom.github.io/CoherentGS/.", "date": "2025-12-11", "categories": []}
{"id": "2512.22295", "title": "Learning Dynamic Scene Reconstruction with Sinusoidal Geometric Priors", "abstract": "We propose SirenPose, a novel loss function that combines the periodic activation properties of sinusoidal representation networks with geometric priors derived from keypoint structures to improve the accuracy of dynamic 3D scene reconstruction. Existing approaches often struggle to maintain motion modeling accuracy and spatiotemporal consistency in fast moving and multi target scenes. By introducing physics inspired constraint mechanisms, SirenPose enforces coherent keypoint predictions across both spatial and temporal dimensions. We further expand the training dataset to 600,000 annotated instances to support robust learning. Experimental results demonstrate that models trained with SirenPose achieve significant improvements in spatiotemporal consistency metrics compared to prior methods, showing superior performance in handling rapid motion and complex scene changes.", "date": "2025-12-25", "categories": []}
{"id": "2512.22745", "title": "Split4D: Decomposed 4D Scene Reconstruction Without Video Segmentation", "abstract": "This paper addresses the problem of decomposed 4D scene reconstruction from multi-view videos. Recent methods achieve this by lifting video segmentation results to a 4D representation through differentiable rendering techniques. Therefore, they heavily rely on the quality of video segmentation maps, which are often unstable, leading to unreliable reconstruction results. To overcome this challenge, our key idea is to represent the decomposed 4D scene with the Freetime FeatureGS and design a streaming feature learning strategy to accurately recover it from per-image segmentation maps, eliminating the need for video segmentation. Freetime FeatureGS models the dynamic scene as a set of Gaussian primitives with learnable features and linear motion ability, allowing them to move to neighboring regions over time. We apply a contrastive loss to Freetime FeatureGS, forcing primitive features to be close or far apart based on whether their projections belong to the same instance in the 2D segmentation map. As our Gaussian primitives can move across time, it naturally extends the feature learning to the temporal dimension, achieving 4D segmentation. Furthermore, we sample observations for training in a temporally ordered manner, enabling the streaming propagation of features over time and effectively avoiding local minima during the optimization process. Experimental results on several datasets show that the reconstruction quality of our method outperforms recent methods by a large margin.", "date": "2025-12-28", "categories": []}
{"id": "2509.17329", "title": "SmokeSeer: 3D Gaussian Splatting for Smoke Removal and Scene Reconstruction", "abstract": "Smoke in real-world scenes can severely degrade image quality and hamper visibility. Recent image restoration methods either rely on data-driven priors that are susceptible to hallucinations, or are limited to static low-density smoke. We introduce SmokeSeer, a method for simultaneous 3D scene reconstruction and smoke removal from multi-view video sequences. Our method uses thermal and RGB images, leveraging the reduced scattering in thermal images to see through smoke. We build upon 3D Gaussian splatting to fuse information from the two image modalities, and decompose the scene into smoke and non-smoke components. Unlike prior work, SmokeSeer handles a broad range of smoke densities and adapts to temporally varying smoke. We validate our method on synthetic data and a new real-world smoke dataset with RGB and thermal images. We provide an open-source implementation and data on the project website.", "date": "2025-09-22", "categories": []}
{"id": "2404.06050", "title": "Incremental Joint Learning of Depth, Pose and Implicit Scene Representation on Monocular Camera in Large-scale Scenes", "abstract": "Dense scene reconstruction for photo-realistic view synthesis has various applications, such as VR/AR, autonomous vehicles. However, most existing methods have difficulties in large-scale scenes due to three core challenges: \\textit{(a) inaccurate depth input.} Accurate depth input is impossible to get in real-world large-scale scenes. \\textit{(b) inaccurate pose estimation.} Most existing approaches rely on accurate pre-estimated camera poses. \\textit{(c) insufficient scene representation capability.} A single global radiance field lacks the capacity to effectively scale to large-scale scenes. To this end, we propose an incremental joint learning framework, which can achieve accurate depth, pose estimation, and large-scale scene reconstruction. A vision transformer-based network is adopted as the backbone to enhance performance in scale information estimation. For pose estimation, a feature-metric bundle adjustment (FBA) method is designed for accurate and robust camera tracking in large-scale scenes. In terms of implicit scene representation, we propose an incremental scene representation method to construct the entire large-scale scene as multiple local radiance fields to enhance the scalability of 3D scene representation. Extended experiments have been conducted to demonstrate the effectiveness and accuracy of our method in depth estimation, pose estimation, and large-scale scene reconstruction.", "date": "2024-04-09", "categories": []}
{"id": "2512.20531", "title": "SirenPose: Dynamic Scene Reconstruction via Geometric Supervision", "abstract": "We introduce SirenPose, a geometry-aware loss formulation that integrates the periodic activation properties of sinusoidal representation networks with keypoint-based geometric supervision, enabling accurate and temporally consistent reconstruction of dynamic 3D scenes from monocular videos. Existing approaches often struggle with motion fidelity and spatiotemporal coherence in challenging settings involving fast motion, multi-object interaction, occlusion, and rapid scene changes. SirenPose incorporates physics inspired constraints to enforce coherent keypoint predictions across both spatial and temporal dimensions, while leveraging high frequency signal modeling to capture fine grained geometric details. We further expand the UniKPT dataset to 600,000 annotated instances and integrate graph neural networks to model keypoint relationships and structural correlations. Extensive experiments on benchmarks including Sintel, Bonn, and DAVIS demonstrate that SirenPose consistently outperforms state-of-the-art methods. On DAVIS, SirenPose achieves a 17.8 percent reduction in FVD, a 28.7 percent reduction in FID, and a 6.0 percent improvement in LPIPS compared to MoSCA. It also improves temporal consistency, geometric accuracy, user score, and motion smoothness. In pose estimation, SirenPose outperforms Monst3R with lower absolute trajectory error as well as reduced translational and rotational relative pose error, highlighting its effectiveness in handling rapid motion, complex dynamics, and physically plausible reconstruction.", "date": "2025-12-23", "categories": []}
{"id": "2501.00625", "title": "Gaussian Building Mesh (GBM): Extract a Building's 3D Mesh with Google Earth and Gaussian Splatting", "abstract": "Recently released open-source pre-trained foundational image segmentation and object detection models (SAM2+GroundingDINO) allow for geometrically consistent segmentation of objects of interest in multi-view 2D images. Users can use text-based or click-based prompts to segment objects of interest without requiring labeled training datasets. Gaussian Splatting allows for the learning of the 3D representation of a scene's geometry and radiance based on 2D images. Combining Google Earth Studio, SAM2+GroundingDINO, 2D Gaussian Splatting, and our improvements in mask refinement based on morphological operations and contour simplification, we created a pipeline to extract the 3D mesh of any building based on its name, address, or geographic coordinates.", "date": "2024-12-31", "categories": []}
{"id": "2507.14501", "title": "Advances in Feed-Forward 3D Reconstruction and View Synthesis: A Survey", "abstract": "3D reconstruction and view synthesis are foundational problems in computer vision, graphics, and immersive technologies such as augmented reality (AR), virtual reality (VR), and digital twins. Traditional methods rely on computationally intensive iterative optimization in a complex chain, limiting their applicability in real-world scenarios. Recent advances in feed-forward approaches, driven by deep learning, have revolutionized this field by enabling fast and generalizable 3D reconstruction and view synthesis. This survey offers a comprehensive review of feed-forward techniques for 3D reconstruction and view synthesis, with a taxonomy according to the underlying representation architectures including point cloud, 3D Gaussian Splatting (3DGS), Neural Radiance Fields (NeRF), etc. We examine key tasks such as pose-free reconstruction, dynamic 3D reconstruction, and 3D-aware image and video synthesis, highlighting their applications in digital humans, SLAM, robotics, and beyond. In addition, we review commonly used datasets with detailed statistics, along with evaluation protocols for various downstream tasks. We conclude by discussing open research challenges and promising directions for future work, emphasizing the potential of feed-forward approaches to advance the state of the art in 3D vision.", "date": "2025-07-19", "categories": []}
{"id": "2508.01171", "title": "No Pose at All: Self-Supervised Pose-Free 3D Gaussian Splatting from Sparse Views", "abstract": "We introduce SPFSplat, an efficient framework for 3D Gaussian splatting from sparse multi-view images, requiring no ground-truth poses during training or inference. It employs a shared feature extraction backbone, enabling simultaneous prediction of 3D Gaussian primitives and camera poses in a canonical space from unposed inputs within a single feed-forward step. Alongside the rendering loss based on estimated novel-view poses, a reprojection loss is integrated to enforce the learning of pixel-aligned Gaussian primitives for enhanced geometric constraints. This pose-free training paradigm and efficient one-step feed-forward design make SPFSplat well-suited for practical applications. Remarkably, despite the absence of pose supervision, SPFSplat achieves state-of-the-art performance in novel view synthesis even under significant viewpoint changes and limited image overlap. It also surpasses recent methods trained with geometry priors in relative pose estimation. Code and trained models are available on our project page: https://ranrhuang.github.io/spfsplat/.", "date": "2025-08-02", "categories": []}
{"id": "2512.08930", "title": "Selfi: Self Improving Reconstruction Engine via 3D Geometric Feature Alignment", "abstract": "Novel View Synthesis (NVS) has traditionally relied on models with explicit 3D inductive biases combined with known camera parameters from Structure-from-Motion (SfM) beforehand. Recent vision foundation models like VGGT take an orthogonal approach -- 3D knowledge is gained implicitly through training data and loss objectives, enabling feed-forward prediction of both camera parameters and 3D representations directly from a set of uncalibrated images. While flexible, VGGT features lack explicit multi-view geometric consistency, and we find that improving such 3D feature consistency benefits both NVS and pose estimation tasks. We introduce Selfi, a self-improving 3D reconstruction pipeline via feature alignment, transforming a VGGT backbone into a high-fidelity 3D reconstruction engine by leveraging its own outputs as pseudo-ground-truth. Specifically, we train a lightweight feature adapter using a reprojection-based consistency loss, which distills VGGT outputs into a new geometrically-aligned feature space that captures spatial proximity in 3D. This enables state-of-the-art performance in both NVS and camera pose estimation, demonstrating that feature alignment is a highly beneficial step for downstream 3D reasoning.", "date": "2025-12-09", "categories": []}
{"id": "2512.17955", "title": "A Modular Framework for Single-View 3D Reconstruction of Indoor Environments", "abstract": "We propose a modular framework for single-view indoor scene 3D reconstruction, where several core modules are powered by diffusion techniques. Traditional approaches for this task often struggle with the complex instance shapes and occlusions inherent in indoor environments. They frequently overshoot by attempting to predict 3D shapes directly from incomplete 2D images, which results in limited reconstruction quality. We aim to overcome this limitation by splitting the process into two steps: first, we employ diffusion-based techniques to predict the complete views of the room background and occluded indoor instances, then transform them into 3D. Our modular framework makes contributions to this field through the following components: an amodal completion module for restoring the full view of occluded instances, an inpainting model specifically trained to predict room layouts, a hybrid depth estimation technique that balances overall geometric accuracy with fine detail expressiveness, and a view-space alignment method that exploits both 2D and 3D cues to ensure precise placement of instances within the scene. This approach effectively reconstructs both foreground instances and the room background from a single image. Extensive experiments on the 3D-Front dataset demonstrate that our method outperforms current state-of-the-art (SOTA) approaches in terms of both visual quality and reconstruction accuracy. The framework holds promising potential for applications in interior design, real estate, and augmented reality.", "date": "2025-12-17", "categories": []}
{"id": "2512.18386", "title": "RecurGS: Interactive Scene Modeling via Discrete-State Recurrent Gaussian Fusion", "abstract": "Recent advances in 3D scene representations have enabled high-fidelity novel view synthesis, yet adapting to discrete scene changes and constructing interactive 3D environments remain open challenges in vision and robotics. Existing approaches focus solely on updating a single scene without supporting novel-state synthesis. Others rely on diffusion-based object-background decoupling that works on one state at a time and cannot fuse information across multiple observations. To address these limitations, we introduce RecurGS, a recurrent fusion framework that incrementally integrates discrete Gaussian scene states into a single evolving representation capable of interaction. RecurGS detects object-level changes across consecutive states, aligns their geometric motion using semantic correspondence and Lie-algebra based SE(3) refinement, and performs recurrent updates that preserve historical structures through replay supervision. A voxelized, visibility-aware fusion module selectively incorporates newly observed regions while keeping stable areas fixed, mitigating catastrophic forgetting and enabling efficient long-horizon updates. RecurGS supports object-level manipulation, synthesizes novel scene states without requiring additional scans, and maintains photorealistic fidelity across evolving environments. Extensive experiments across synthetic and real-world datasets demonstrate that our framework delivers high-quality reconstructions with substantially improved update efficiency, providing a scalable step toward continuously interactive Gaussian worlds.", "date": "2025-12-20", "categories": []}
{"id": "2512.18692", "title": "EcoSplat: Efficiency-controllable Feed-forward 3D Gaussian Splatting from Multi-view Images", "abstract": "Feed-forward 3D Gaussian Splatting (3DGS) enables efficient one-pass scene reconstruction, providing 3D representations for novel view synthesis without per-scene optimization. However, existing methods typically predict pixel-aligned primitives per-view, producing an excessive number of primitives in dense-view settings and offering no explicit control over the number of predicted Gaussians. To address this, we propose EcoSplat, the first efficiency-controllable feed-forward 3DGS framework that adaptively predicts the 3D representation for any given target primitive count at inference time. EcoSplat adopts a two-stage optimization process. The first stage is Pixel-aligned Gaussian Training (PGT) where our model learns initial primitive prediction. The second stage is Importance-aware Gaussian Finetuning (IGF) stage where our model learns rank primitives and adaptively adjust their parameters based on the target primitive count. Extensive experiments across multiple dense-view settings show that EcoSplat is robust and outperforms state-of-the-art methods under strict primitive-count constraints, making it well-suited for flexible downstream rendering tasks.", "date": "2025-12-21", "categories": []}
{"id": "2411.15355", "title": "UniGaussian: Driving Scene Reconstruction from Multiple Camera Models via Unified Gaussian Representations", "abstract": "Urban scene reconstruction is crucial for real-world autonomous driving simulators. Although existing methods have achieved photorealistic reconstruction, they mostly focus on pinhole cameras and neglect fisheye cameras. In fact, how to effectively simulate fisheye cameras in driving scene remains an unsolved problem. In this work, we propose UniGaussian, a novel approach that learns a unified 3D Gaussian representation from multiple camera models for urban scene reconstruction in autonomous driving. Our contributions are two-fold. First, we propose a new differentiable rendering method that distorts 3D Gaussians using a series of affine transformations tailored to fisheye camera models. This addresses the compatibility issue of 3D Gaussian splatting with fisheye cameras, which is hindered by light ray distortion caused by lenses or mirrors. Besides, our method maintains real-time rendering while ensuring differentiability. Second, built on the differentiable rendering method, we design a new framework that learns a unified Gaussian representation from multiple camera models. By applying affine transformations to adapt different camera models and regularizing the shared Gaussians with supervision from different modalities, our framework learns a unified 3D Gaussian representation with input data from multiple sources and achieves holistic driving scene understanding. As a result, our approach models multiple sensors (pinhole and fisheye cameras) and modalities (depth, semantic, normal and LiDAR point clouds). Our experiments show that our method achieves superior rendering quality and fast rendering speed for driving scene simulation.", "date": "2024-11-22", "categories": []}
{"id": "2512.17459", "title": "3D-RE-GEN: 3D Reconstruction of Indoor Scenes with a Generative Framework", "abstract": "Recent advances in 3D scene generation produce visually appealing output, but current representations hinder artists' workflows that require modifiable 3D textured mesh scenes for visual effects and game development. Despite significant advances, current textured mesh scene reconstruction methods are far from artist ready, suffering from incorrect object decomposition, inaccurate spatial relationships, and missing backgrounds. We present 3D-RE-GEN, a compositional framework that reconstructs a single image into textured 3D objects and a background. We show that combining state of the art models from specific domains achieves state of the art scene reconstruction performance, addressing artists' requirements.\n  Our reconstruction pipeline integrates models for asset detection, reconstruction, and placement, pushing certain models beyond their originally intended domains. Obtaining occluded objects is treated as an image editing task with generative models to infer and reconstruct with scene level reasoning under consistent lighting and geometry. Unlike current methods, 3D-RE-GEN generates a comprehensive background that spatially constrains objects during optimization and provides a foundation for realistic lighting and simulation tasks in visual effects and games. To obtain physically realistic layouts, we employ a novel 4-DoF differentiable optimization that aligns reconstructed objects with the estimated ground plane. 3D-RE-GEN~achieves state of the art performance in single image 3D scene reconstruction, producing coherent, modifiable scenes through compositional generation guided by precise camera recovery and spatial optimization.", "date": "2025-12-19", "categories": []}
{"id": "2512.17547", "title": "G3Splat: Geometrically Consistent Generalizable Gaussian Splatting", "abstract": "3D Gaussians have recently emerged as an effective scene representation for real-time splatting and accurate novel-view synthesis, motivating several works to adapt multi-view structure prediction networks to regress per-pixel 3D Gaussians from images. However, most prior work extends these networks to predict additional Gaussian parameters -- orientation, scale, opacity, and appearance -- while relying almost exclusively on view-synthesis supervision. We show that a view-synthesis loss alone is insufficient to recover geometrically meaningful splats in this setting. We analyze and address the ambiguities of learning 3D Gaussian splats under self-supervision for pose-free generalizable splatting, and introduce G3Splat, which enforces geometric priors to obtain geometrically consistent 3D scene representations. Trained on RE10K, our approach achieves state-of-the-art performance in (i) geometrically consistent reconstruction, (ii) relative pose estimation, and (iii) novel-view synthesis. We further demonstrate strong zero-shot generalization on ScanNet, substantially outperforming prior work in both geometry recovery and relative pose estimation. Code and pretrained models are released on our project page (https://m80hz.github.io/g3splat/).", "date": "2025-12-19", "categories": []}
{"id": "2409.15466", "title": "Mat\\'ern Kernels for Tunable Implicit Surface Reconstruction", "abstract": "We propose to use the family of Mat\\'ern kernels for implicit surface reconstruction, building upon the recent success of kernel methods for 3D reconstruction of oriented point clouds. As we show from a theoretical and practical perspective, Mat\\'ern kernels have some appealing properties which make them particularly well suited for surface reconstruction -- outperforming state-of-the-art methods based on the arc-cosine kernel while being significantly easier to implement, faster to compute, and scalable. Being stationary, we demonstrate that Mat\\'ern kernels allow for tunable surface reconstruction in the same way as Fourier feature mappings help coordinate-based MLPs overcome spectral bias. Moreover, we theoretically analyze Mat\\'ern kernels' connection to SIREN networks as well as their relation to previously employed arc-cosine kernels. Finally, based on recently introduced Neural Kernel Fields, we present data-dependent Mat\\'ern kernels and conclude that especially the Laplace kernel (being part of the Mat\\'ern family) is extremely competitive, performing almost on par with state-of-the-art methods in the noise-free case while having a more than five times shorter training time.", "date": "2024-09-23", "categories": []}
{"id": "2512.16564", "title": "4D Primitive-M\\^ach\\'e: Glueing Primitives for Persistent 4D Scene Reconstruction", "abstract": "We present a dynamic reconstruction system that receives a casual monocular RGB video as input, and outputs a complete and persistent reconstruction of the scene. In other words, we reconstruct not only the the currently visible parts of the scene, but also all previously viewed parts, which enables replaying the complete reconstruction across all timesteps.\n  Our method decomposes the scene into a set of rigid 3D primitives, which are assumed to be moving throughout the scene. Using estimated dense 2D correspondences, we jointly infer the rigid motion of these primitives through an optimisation pipeline, yielding a 4D reconstruction of the scene, i.e. providing 3D geometry dynamically moving through time. To achieve this, we also introduce a mechanism to extrapolate motion for objects that become invisible, employing motion-grouping techniques to maintain continuity.\n  The resulting system enables 4D spatio-temporal awareness, offering capabilities such as replayable 3D reconstructions of articulated objects through time, multi-object scanning, and object permanence. On object scanning and multi-object datasets, our system significantly outperforms existing methods both quantitatively and qualitatively.", "date": "2025-12-18", "categories": []}
{"id": "2512.16706", "title": "SDFoam: Signed-Distance Foam for explicit surface reconstruction", "abstract": "Neural radiance fields (NeRF) have driven impressive progress in view synthesis by using ray-traced volumetric rendering. Splatting-based methods such as 3D Gaussian Splatting (3DGS) provide faster rendering by rasterizing 3D primitives. RadiantFoam (RF) brought ray tracing back, achieving throughput comparable to Gaussian Splatting by organizing radiance with an explicit Voronoi Diagram (VD). Yet, all the mentioned methods still struggle with precise mesh reconstruction. We address this gap by jointly learning an explicit VD with an implicit Signed Distance Field (SDF). The scene is optimized via ray tracing and regularized by an Eikonal objective. The SDF introduces metric-consistent isosurfaces, which, in turn, bias near-surface Voronoi cell faces to align with the zero level set. The resulting model produces crisper, view-consistent surfaces with fewer floaters and improved topology, while preserving photometric quality and maintaining training speed on par with RadiantFoam. Across diverse scenes, our hybrid implicit-explicit formulation, which we name SDFoam, substantially improves mesh reconstruction accuracy (Chamfer distance) with comparable appearance (PSNR, SSIM), without sacrificing efficiency.", "date": "2025-12-18", "categories": []}
{"id": "2406.20066", "title": "ASSR-NeRF: Arbitrary-Scale Super-Resolution on Voxel Grid for High-Quality Radiance Fields Reconstruction", "abstract": "NeRF-based methods reconstruct 3D scenes by building a radiance field with implicit or explicit representations. While NeRF-based methods can perform novel view synthesis (NVS) at arbitrary scale, the performance in high-resolution novel view synthesis (HRNVS) with low-resolution (LR) optimization often results in oversmoothing. On the other hand, single-image super-resolution (SR) aims to enhance LR images to HR counterparts but lacks multi-view consistency. To address these challenges, we propose Arbitrary-Scale Super-Resolution NeRF (ASSR-NeRF), a novel framework for super-resolution novel view synthesis (SRNVS). We propose an attention-based VoxelGridSR model to directly perform 3D super-resolution (SR) on the optimized volume. Our model is trained on diverse scenes to ensure generalizability. For unseen scenes trained with LR views, we then can directly apply our VoxelGridSR to further refine the volume and achieve multi-view consistent SR. We demonstrate quantitative and qualitatively that the proposed method achieves significant performance in SRNVS.", "date": "2024-06-28", "categories": []}
{"id": "2512.14087", "title": "GaussianPlant: Structure-aligned Gaussian Splatting for 3D Reconstruction of Plants", "abstract": "We present a method for jointly recovering the appearance and internal structure of botanical plants from multi-view images based on 3D Gaussian Splatting (3DGS). While 3DGS exhibits robust reconstruction of scene appearance for novel-view synthesis, it lacks structural representations underlying those appearances (e.g., branching patterns of plants), which limits its applicability to tasks such as plant phenotyping. To achieve both high-fidelity appearance and structural reconstruction, we introduce GaussianPlant, a hierarchical 3DGS representation, which disentangles structure and appearance. Specifically, we employ structure primitives (StPs) to explicitly represent branch and leaf geometry, and appearance primitives (ApPs) to the plants' appearance using 3D Gaussians. StPs represent a simplified structure of the plant, i.e., modeling branches as cylinders and leaves as disks. To accurately distinguish the branches and leaves, StP's attributes (i.e., branches or leaves) are optimized in a self-organized manner. ApPs are bound to each StP to represent the appearance of branches or leaves as in conventional 3DGS. StPs and ApPs are jointly optimized using a re-rendering loss on the input multi-view images, as well as the gradient flow from ApP to StP using the binding correspondence information. We conduct experiments to qualitatively evaluate the reconstruction accuracy of both appearance and structure, as well as real-world experiments to qualitatively validate the practical performance. Experiments show that the GaussianPlant achieves both high-fidelity appearance reconstruction via ApPs and accurate structural reconstruction via StPs, enabling the extraction of branch structure and leaf instances.", "date": "2025-12-16", "categories": []}
{"id": "2512.14200", "title": "Beyond a Single Light: A Large-Scale Aerial Dataset for Urban Scene Reconstruction Under Varying Illumination", "abstract": "Recent advances in Neural Radiance Fields and 3D Gaussian Splatting have demonstrated strong potential for large-scale UAV-based 3D reconstruction tasks by fitting the appearance of images. However, real-world large-scale captures are often based on multi-temporal data capture, where illumination inconsistencies across different times of day can significantly lead to color artifacts, geometric inaccuracies, and inconsistent appearance. Due to the lack of UAV datasets that systematically capture the same areas under varying illumination conditions, this challenge remains largely underexplored. To fill this gap, we introduceSkyLume, a large-scale, real-world UAV dataset specifically designed for studying illumination robust 3D reconstruction in urban scene modeling: (1) We collect data from 10 urban regions data comprising more than 100k high resolution UAV images (four oblique views and nadir), where each region is captured at three periods of the day to systematically isolate illumination changes. (2) To support precise evaluation of geometry and appearance, we provide per-scene LiDAR scans and accurate 3D ground-truth for assessing depth, surface normals, and reconstruction quality under varying illumination. (3) For the inverse rendering task, we introduce the Temporal Consistency Coefficient (TCC), a metric that measuress cross-time albedo stability and directly evaluates the robustness of the disentanglement of light and material. We aim for this resource to serve as a foundation that advances research and real-world evaluation in large-scale inverse rendering, geometry reconstruction, and novel view synthesis.", "date": "2025-12-16", "categories": []}
{"id": "2412.16809", "title": "GeoTexDensifier: Geometry-Texture-Aware Densification for High-Quality Photorealistic 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has recently attracted wide attentions in various areas such as 3D navigation, Virtual Reality (VR) and 3D simulation, due to its photorealistic and efficient rendering performance. High-quality reconstrution of 3DGS relies on sufficient splats and a reasonable distribution of these splats to fit real geometric surface and texture details, which turns out to be a challenging problem. We present GeoTexDensifier, a novel geometry-texture-aware densification strategy to reconstruct high-quality Gaussian splats which better comply with the geometric structure and texture richness of the scene. Specifically, our GeoTexDensifier framework carries out an auxiliary texture-aware densification method to produce a denser distribution of splats in fully textured areas, while keeping sparsity in low-texture regions to maintain the quality of Gaussian point cloud. Meanwhile, a geometry-aware splitting strategy takes depth and normal priors to guide the splitting sampling and filter out the noisy splats whose initial positions are far from the actual geometric surfaces they aim to fit, under a Validation of Depth Ratio Change checking. With the help of relative monocular depth prior, such geometry-aware validation can effectively reduce the influence of scattered Gaussians to the final rendering quality, especially in regions with weak textures or without sufficient training views. The texture-aware densification and geometry-aware splitting strategies are fully combined to obtain a set of high-quality Gaussian splats. We experiment our GeoTexDensifier framework on various datasets and compare our Novel View Synthesis results to other state-of-the-art 3DGS approaches, with detailed quantitative and qualitative evaluations to demonstrate the effectiveness of our method in producing more photorealistic 3DGS models.", "date": "2024-12-22", "categories": []}
{"id": "2509.11574", "title": "Gaussian-Plus-SDF SLAM: High-fidelity 3D Reconstruction at 150+ fps", "abstract": "While recent Gaussian-based SLAM methods achieve photorealistic reconstruction from RGB-D data, their computational performance remains a critical bottleneck. State-of-the-art techniques operate at less than 20 fps, significantly lagging behind geometry-based approaches like KinectFusion (hundreds of fps). This limitation stems from the heavy computational burden: modeling scenes requires numerous Gaussians and complex iterative optimization to fit RGB-D data; insufficient Gaussian counts or optimization iterations cause severe quality degradation. To address this, we propose a Gaussian-SDF hybrid representation, combining a colorized signed distance field (SDF) for smooth geometry and appearance with 3D Gaussians to capture underrepresented details. The SDF is efficiently constructed via RGB-D fusion (as in geometry-based methods), while Gaussians undergo iterative optimization. Our representation enables significant Gaussian reduction (50% fewer) by avoiding full-scene Gaussian modeling, and efficient Gaussian optimization (75% fewer iterations) through targeted appearance refinement. Building upon this representation, we develop GPS-SLAM (Gaussian-plus-SDF SLAM), a real-time 3D reconstruction system achieving over 150 fps on real-world Azure Kinect sequences, faster by an order-of-magnitude than state-of-the-art techniques while maintaining comparable reconstruction quality. The source code and data are available at https://gapszju.github.io/GPS-SLAM.", "date": "2025-09-15", "categories": []}
{"id": "2512.12683", "title": "Quantum Implicit Neural Representations for 3D Scene Reconstruction and Novel View Synthesis", "abstract": "Implicit neural representations (INRs) have become a powerful paradigm for continuous signal modeling and 3D scene reconstruction, yet classical networks suffer from a well-known spectral bias that limits their ability to capture high-frequency details. Quantum Implicit Representation Networks (QIREN) mitigate this limitation by employing parameterized quantum circuits with inherent Fourier structures, enabling compact and expressive frequency modeling beyond classical MLPs. In this paper, we present Quantum Neural Radiance Fields (Q-NeRF), the first hybrid quantum-classical framework for neural radiance field rendering. Q-NeRF integrates QIREN modules into the Nerfacto backbone, preserving its efficient sampling, pose refinement, and volumetric rendering strategies while replacing selected density and radiance prediction components with quantum-enhanced counterparts. We systematically evaluate three hybrid configurations on standard multi-view indoor datasets, comparing them to classical baselines using PSNR, SSIM, and LPIPS metrics. Results show that hybrid quantum-classical models achieve competitive reconstruction quality under limited computational resources, with quantum modules particularly effective in representing fine-scale, view-dependent appearance. Although current implementations rely on quantum circuit simulators constrained to few-qubit regimes, the results highlight the potential of quantum encodings to alleviate spectral bias in implicit representations. Q-NeRF provides a foundational step toward scalable quantum-enabled 3D scene reconstruction and a baseline for future quantum neural rendering research.", "date": "2025-12-14", "categories": []}
{"id": "2506.19139", "title": "SOF: Sorted Opacity Fields for Fast Unbounded Surface Reconstruction", "abstract": "Recent advances in 3D Gaussian representations have significantly improved the quality and efficiency of image-based scene reconstruction. Their explicit nature facilitates real-time rendering and fast optimization, yet extracting accurate surfaces - particularly in large-scale, unbounded environments - remains a difficult task. Many existing methods rely on approximate depth estimates and global sorting heuristics, which can introduce artifacts and limit the fidelity of the reconstructed mesh. In this paper, we present Sorted Opacity Fields (SOF), a method designed to recover detailed surfaces from 3D Gaussians with both speed and precision. Our approach improves upon prior work by introducing hierarchical resorting and a robust formulation of Gaussian depth, which better aligns with the level-set. To enhance mesh quality, we incorporate a level-set regularizer operating on the opacity field and introduce losses that encourage geometrically-consistent primitive shapes. In addition, we develop a parallelized Marching Tetrahedra algorithm tailored to our opacity formulation, reducing meshing time by up to an order of magnitude. As demonstrated by our quantitative evaluation, SOF achieves higher reconstruction accuracy while cutting total processing time by more than a factor of three. These results mark a step forward in turning efficient Gaussian-based rendering into equally efficient geometry extraction.", "date": "2025-06-23", "categories": []}
{"id": "2506.21520", "title": "MADrive: Memory-Augmented Driving Scene Modeling", "abstract": "Recent advances in scene reconstruction have pushed toward highly realistic modeling of autonomous driving (AD) environments using 3D Gaussian splatting. However, the resulting reconstructions remain closely tied to the original observations and struggle to support photorealistic synthesis of significantly altered or novel driving scenarios. This work introduces MADrive, a memory-augmented reconstruction framework designed to extend the capabilities of existing scene reconstruction methods by replacing observed vehicles with visually similar 3D assets retrieved from a large-scale external memory bank. Specifically, we release MAD-Cars, a curated dataset of ${\\sim}70$K 360{\\deg} car videos captured in the wild and present a retrieval module that finds the most similar car instances in the memory bank, reconstructs the corresponding 3D assets from video, and integrates them into the target scene through orientation alignment and relighting. The resulting replacements provide complete multi-view representations of vehicles in the scene, enabling photorealistic synthesis of substantially altered configurations, as demonstrated in our experiments. Project page: https://yandex-research.github.io/madrive/", "date": "2025-06-26", "categories": []}
{"id": "2512.11301", "title": "MultiEgo: A Multi-View Egocentric Video Dataset for 4D Scene Reconstruction", "abstract": "Multi-view egocentric dynamic scene reconstruction holds significant research value for applications in holographic documentation of social interactions. However, existing reconstruction datasets focus on static multi-view or single-egocentric view setups, lacking multi-view egocentric datasets for dynamic scene reconstruction. Therefore, we present MultiEgo, the first multi-view egocentric dataset for 4D dynamic scene reconstruction. The dataset comprises five canonical social interaction scenes: meetings, performances, and a presentation. Each scene provides five authentic egocentric videos captured by participants wearing AR glasses. We design a hardware-based data acquisition system and processing pipeline, achieving sub-millisecond temporal synchronization across views, coupled with accurate pose annotations. Experiment validation demonstrates the practical utility and effectiveness of our dataset for free-viewpoint video (FVV) applications, establishing MultiEgo as a foundational resource for advancing multi-view egocentric dynamic scene reconstruction research.", "date": "2025-12-12", "categories": []}
{"id": "2512.11356", "title": "Prior-Enhanced Gaussian Splatting for Dynamic Scene Reconstruction from Casual Video", "abstract": "We introduce a fully automatic pipeline for dynamic scene reconstruction from casually captured monocular RGB videos. Rather than designing a new scene representation, we enhance the priors that drive Dynamic Gaussian Splatting. Video segmentation combined with epipolar-error maps yields object-level masks that closely follow thin structures; these masks (i) guide an object-depth loss that sharpens the consistent video depth, and (ii) support skeleton-based sampling plus mask-guided re-identification to produce reliable, comprehensive 2-D tracks. Two additional objectives embed the refined priors in the reconstruction stage: a virtual-view depth loss removes floaters, and a scaffold-projection loss ties motion nodes to the tracks, preserving fine geometry and coherent motion. The resulting system surpasses previous monocular dynamic scene reconstruction methods and delivers visibly superior renderings", "date": "2025-12-12", "categories": []}
{"id": "2505.17951", "title": "SplatCo: Structure-View Collaborative Gaussian Splatting for Detail-Preserving Rendering of Large-Scale Unbounded Scenes", "abstract": "We present SplatCo, a structure-view collaborative Gaussian splatting framework for high-fidelity rendering of complex outdoor scenes. SplatCo builds upon three novel components: 1) a cross-structure collaboration module that combines global tri-plane representations, which capture coarse scene layouts, with local context grid features representing fine details. This fusion is achieved through a hierarchical compensation mechanism, ensuring both global spatial awareness and local detail preservation; 2) a cross-view pruning mechanism that removes overfitted or inaccurate Gaussians based on structural consistency, thereby improving storage efficiency and preventing rendering artifacts; 3) a structure view co-learning module that aggregates structural gradients with view gradients,thereby steering the optimization of Gaussian geometric and appearance attributes more robustly. By combining these key components, SplatCo effectively achieves high-fidelity rendering for large-scale scenes. Code and project page are available at https://splatco-tech.github.io.", "date": "2025-05-23", "categories": []}
{"id": "2510.07752", "title": "DEGS: Deformable Event-based 3D Gaussian Splatting from RGB and Event Stream", "abstract": "Reconstructing Dynamic 3D Gaussian Splatting (3DGS) from low-framerate RGB videos is challenging. This is because large inter-frame motions will increase the uncertainty of the solution space. For example, one pixel in the first frame might have more choices to reach the corresponding pixel in the second frame. Event cameras can asynchronously capture rapid visual changes and are robust to motion blur, but they do not provide color information. Intuitively, the event stream can provide deterministic constraints for the inter-frame large motion by the event trajectories. Hence, combining low-temporal-resolution images with high-framerate event streams can address this challenge. However, it is challenging to jointly optimize Dynamic 3DGS using both RGB and event modalities due to the significant discrepancy between these two data modalities. This paper introduces a novel framework that jointly optimizes dynamic 3DGS from the two modalities. The key idea is to adopt event motion priors to guide the optimization of the deformation fields. First, we extract the motion priors encoded in event streams by using the proposed LoCM unsupervised fine-tuning framework to adapt an event flow estimator to a certain unseen scene. Then, we present the geometry-aware data association method to build the event-Gaussian motion correspondence, which is the primary foundation of the pipeline, accompanied by two useful strategies, namely motion decomposition and inter-frame pseudo-label. Extensive experiments show that our method outperforms existing image and event-based approaches across synthetic and real scenes and prove that our method can effectively optimize dynamic 3DGS with the help of event data.", "date": "2025-10-09", "categories": []}
{"id": "2512.10095", "title": "TraceFlow: Dynamic 3D Reconstruction of Specular Scenes Driven by Ray Tracing", "abstract": "We present TraceFlow, a novel framework for high-fidelity rendering of dynamic specular scenes by addressing two key challenges: precise reflection direction estimation and physically accurate reflection modeling. To achieve this, we propose a Residual Material-Augmented 2D Gaussian Splatting representation that models dynamic geometry and material properties, allowing accurate reflection ray computation. Furthermore, we introduce a Dynamic Environment Gaussian and a hybrid rendering pipeline that decomposes rendering into diffuse and specular components, enabling physically grounded specular synthesis via rasterization and ray tracing. Finally, we devise a coarse-to-fine training strategy to improve optimization stability and promote physically meaningful decomposition. Extensive experiments on dynamic scene benchmarks demonstrate that TraceFlow outperforms prior methods both quantitatively and qualitatively, producing sharper and more realistic specular reflections in complex dynamic environments.", "date": "2025-12-10", "categories": []}
{"id": "2512.10935", "title": "Any4D: Unified Feed-Forward Metric 4D Reconstruction", "abstract": "We present Any4D, a scalable multi-view transformer for metric-scale, dense feed-forward 4D reconstruction. Any4D directly generates per-pixel motion and geometry predictions for N frames, in contrast to prior work that typically focuses on either 2-view dense scene flow or sparse 3D point tracking. Moreover, unlike other recent methods for 4D reconstruction from monocular RGB videos, Any4D can process additional modalities and sensors such as RGB-D frames, IMU-based egomotion, and Radar Doppler measurements, when available. One of the key innovations that allows for such a flexible framework is a modular representation of a 4D scene; specifically, per-view 4D predictions are encoded using a variety of egocentric factors (depthmaps and camera intrinsics) represented in local camera coordinates, and allocentric factors (camera extrinsics and scene flow) represented in global world coordinates. We achieve superior performance across diverse setups - both in terms of accuracy (2-3X lower error) and compute efficiency (15X faster), opening avenues for multiple downstream applications.", "date": "2025-12-11", "categories": []}
{"id": "2512.08924", "title": "Efficiently Reconstructing Dynamic Scenes One D4RT at a Time", "abstract": "Understanding and reconstructing the complex geometry and motion of dynamic scenes from video remains a formidable challenge in computer vision. This paper introduces D4RT, a simple yet powerful feedforward model designed to efficiently solve this task. D4RT utilizes a unified transformer architecture to jointly infer depth, spatio-temporal correspondence, and full camera parameters from a single video. Its core innovation is a novel querying mechanism that sidesteps the heavy computation of dense, per-frame decoding and the complexity of managing multiple, task-specific decoders. Our decoding interface allows the model to independently and flexibly probe the 3D position of any point in space and time. The result is a lightweight and highly scalable method that enables remarkably efficient training and inference. We demonstrate that our approach sets a new state of the art, outperforming previous methods across a wide spectrum of 4D reconstruction tasks. We refer to the project webpage for animated results: https://d4rt-paper.github.io/.", "date": "2025-12-09", "categories": []}
{"id": "2512.09201", "title": "Residual Primitive Fitting of 3D Shapes with SuperFrusta", "abstract": "We introduce a framework for converting 3D shapes into compact and editable assemblies of analytic primitives, directly addressing the persistent trade-off between reconstruction fidelity and parsimony. Our approach combines two key contributions: a novel primitive, termed SuperFrustum, and an iterative fiting algorithm, Residual Primitive Fitting (ResFit). SuperFrustum is an analytical primitive that is simultaneously (1) expressive, being able to model various common solids such as cylinders, spheres, cones & their tapered and bent forms, (2) editable, being compactly parameterized with 8 parameters, and (3) optimizable, with a sign distance field differentiable w.r.t. its parameters almost everywhere. ResFit is an unsupervised procedure that interleaves global shape analysis with local optimization, iteratively fitting primitives to the unexplained residual of a shape to discover a parsimonious yet accurate decompositions for each input shape. On diverse 3D benchmarks, our method achieves state-of-the-art results, improving IoU by over 9 points while using nearly half as many primitives as prior work. The resulting assemblies bridge the gap between dense 3D data and human-controllable design, producing high-fidelity and editable shape programs.", "date": "2025-12-09", "categories": []}
{"id": "2501.13104", "title": "Neural Radiance Fields for the Real World: A Survey", "abstract": "Neural Radiance Fields (NeRFs) have remodeled 3D scene representation since release. NeRFs can effectively reconstruct complex 3D scenes from 2D images, advancing different fields and applications such as scene understanding, 3D content generation, and robotics. Despite significant research progress, a thorough review of recent innovations, applications, and challenges is lacking. This survey compiles key theoretical advancements and alternative representations and investigates emerging challenges. It further explores applications on reconstruction, highlights NeRFs' impact on computer vision and robotics, and reviews essential datasets and toolkits. By identifying gaps in the literature, this survey discusses open challenges and offers directions for future research.", "date": "2025-01-22", "categories": []}
{"id": "2512.06368", "title": "HuPrior3R: Incorporating Human Priors for Better 3D Dynamic Reconstruction from Monocular Videos", "abstract": "Monocular dynamic video reconstruction faces significant challenges in dynamic human scenes due to geometric inconsistencies and resolution degradation issues. Existing methods lack 3D human structural understanding, producing geometrically inconsistent results with distorted limb proportions and unnatural human-object fusion, while memory-constrained downsampling causes human boundary drift toward background geometry. To address these limitations, we propose to incorporate hybrid geometric priors that combine SMPL human body models with monocular depth estimation. Our approach leverages structured human priors to maintain surface consistency while capturing fine-grained geometric details in human regions. We introduce HuPrior3R, featuring a hierarchical pipeline with refinement components that processes full-resolution images for overall scene geometry, then applies strategic cropping and cross-attention fusion for human-specific detail enhancement. The method integrates SMPL priors through a Feature Fusion Module to ensure geometrically plausible reconstruction while preserving fine-grained human boundaries. Extensive experiments on TUM Dynamics and GTA-IM datasets demonstrate superior performance in dynamic human reconstruction.", "date": "2025-12-06", "categories": []}
{"id": "2512.08498", "title": "On-the-fly Large-scale 3D Reconstruction from Multi-Camera Rigs", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have enabled efficient free-viewpoint rendering and photorealistic scene reconstruction. While on-the-fly extensions of 3DGS have shown promise for real-time reconstruction from monocular RGB streams, they often fail to achieve complete 3D coverage due to the limited field of view (FOV). Employing a multi-camera rig fundamentally addresses this limitation. In this paper, we present the first on-the-fly 3D reconstruction framework for multi-camera rigs. Our method incrementally fuses dense RGB streams from multiple overlapping cameras into a unified Gaussian representation, achieving drift-free trajectory estimation and efficient online reconstruction. We propose a hierarchical camera initialization scheme that enables coarse inter-camera alignment without calibration, followed by a lightweight multi-camera bundle adjustment that stabilizes trajectories while maintaining real-time performance. Furthermore, we introduce a redundancy-free Gaussian sampling strategy and a frequency-aware optimization scheduler to reduce the number of Gaussian primitives and the required optimization iterations, thereby maintaining both efficiency and reconstruction fidelity. Our method reconstructs hundreds of meters of 3D scenes within just 2 minutes using only raw multi-camera video streams, demonstrating unprecedented speed, robustness, and Fidelity for on-the-fly 3D scene reconstruction.", "date": "2025-12-09", "categories": []}
{"id": "2512.08506", "title": "OCCDiff: Occupancy Diffusion Model for High-Fidelity 3D Building Reconstruction from Noisy Point Clouds", "abstract": "A major challenge in reconstructing buildings from LiDAR point clouds lies in accurately capturing building surfaces under varying point densities and noise interference. To flexibly gather high-quality 3D profiles of the building in diverse resolution, we propose OCCDiff applying latent diffusion in the occupancy function space. Our OCCDiff combines a latent diffusion process with a function autoencoder architecture to generate continuous occupancy functions evaluable at arbitrary locations. Moreover, a point encoder is proposed to provide condition features to diffusion learning, constraint the final occupancy prediction for occupancy decoder, and insert multi-modal features for latent generation to latent encoder. To further enhance the model performance, a multi-task training strategy is employed, ensuring that the point encoder learns diverse and robust feature representations. Empirical results show that our method generates physically consistent samples with high fidelity to the target distribution and exhibits robustness to noisy data.", "date": "2025-12-09", "categories": []}
{"id": "2512.08905", "title": "Self-Evolving 3D Scene Generation from a Single Image", "abstract": "Generating high-quality, textured 3D scenes from a single image remains a fundamental challenge in vision and graphics. Recent image-to-3D generators recover reasonable geometry from single views, but their object-centric training limits generalization to complex, large-scale scenes with faithful structure and texture. We present EvoScene, a self-evolving, training-free framework that progressively reconstructs complete 3D scenes from single images. The key idea is combining the complementary strengths of existing models: geometric reasoning from 3D generation models and visual knowledge from video generation models. Through three iterative stages--Spatial Prior Initialization, Visual-guided 3D Scene Mesh Generation, and Spatial-guided Novel View Generation--EvoScene alternates between 2D and 3D domains, gradually improving both structure and appearance. Experiments on diverse scenes demonstrate that EvoScene achieves superior geometric stability, view-consistent textures, and unseen-region completion compared to strong baselines, producing ready-to-use 3D meshes for practical applications.", "date": "2025-12-09", "categories": []}
{"id": "2511.04283", "title": "FastGS: Training 3D Gaussian Splatting in 100 Seconds", "abstract": "The dominant 3D Gaussian splatting (3DGS) acceleration methods fail to properly regulate the number of Gaussians during training, causing redundant computational time overhead. In this paper, we propose FastGS, a novel, simple, and general acceleration framework that fully considers the importance of each Gaussian based on multi-view consistency, efficiently solving the trade-off between training time and rendering quality. We innovatively design a densification and pruning strategy based on multi-view consistency, dispensing with the budgeting mechanism. Extensive experiments on Mip-NeRF 360, Tanks & Temples, and Deep Blending datasets demonstrate that our method significantly outperforms the state-of-the-art methods in training speed, achieving a 3.32$\\times$ training acceleration and comparable rendering quality compared with DashGaussian on the Mip-NeRF 360 dataset and a 15.45$\\times$ acceleration compared with vanilla 3DGS on the Deep Blending dataset. We demonstrate that FastGS exhibits strong generality, delivering 2-7$\\times$ training acceleration across various tasks, including dynamic scene reconstruction, surface reconstruction, sparse-view reconstruction, large-scale reconstruction, and simultaneous localization and mapping. The project page is available at https://fastgs.github.io/", "date": "2025-11-06", "categories": []}
{"id": "2512.05113", "title": "Splannequin: Freezing Monocular Mannequin-Challenge Footage with Dual-Detection Splatting", "abstract": "Synthesizing high-fidelity frozen 3D scenes from monocular Mannequin-Challenge (MC) videos is a unique problem distinct from standard dynamic scene reconstruction. Instead of focusing on modeling motion, our goal is to create a frozen scene while strategically preserving subtle dynamics to enable user-controlled instant selection. To achieve this, we introduce a novel application of dynamic Gaussian splatting: the scene is modeled dynamically, which retains nearby temporal variation, and a static scene is rendered by fixing the model's time parameter. However, under this usage, monocular capture with sparse temporal supervision introduces artifacts like ghosting and blur for Gaussians that become unobserved or occluded at weakly supervised timestamps. We propose Splannequin, an architecture-agnostic regularization that detects two states of Gaussian primitives, hidden and defective, and applies temporal anchoring. Under predominantly forward camera motion, hidden states are anchored to their recent well-observed past states, while defective states are anchored to future states with stronger supervision. Our method integrates into existing dynamic Gaussian pipelines via simple loss terms, requires no architectural changes, and adds zero inference overhead. This results in markedly improved visual quality, enabling high-fidelity, user-selectable frozen-time renderings, validated by a 96% user preference. Project page: https://chien90190.github.io/splannequin/", "date": "2025-12-04", "categories": []}
{"id": "2512.05988", "title": "VG3T: Visual Geometry Grounded Gaussian Transformer", "abstract": "Generating a coherent 3D scene representation from multi-view images is a fundamental yet challenging task. Existing methods often struggle with multi-view fusion, leading to fragmented 3D representations and sub-optimal performance. To address this, we introduce VG3T, a novel multi-view feed-forward network that predicts a 3D semantic occupancy via a 3D Gaussian representation. Unlike prior methods that infer Gaussians from single-view images, our model directly predicts a set of semantically attributed Gaussians in a joint, multi-view fashion. This novel approach overcomes the fragmentation and inconsistency inherent in view-by-view processing, offering a unified paradigm to represent both geometry and semantics. We also introduce two key components, Grid-Based Sampling and Positional Refinement, to mitigate the distance-dependent density bias common in pixel-aligned Gaussian initialization methods. Our VG3T shows a notable 1.7%p improvement in mIoU while using 46% fewer primitives than the previous state-of-the-art on the nuScenes benchmark, highlighting its superior efficiency and performance.", "date": "2025-11-28", "categories": []}
{"id": "2512.06080", "title": "Shoot-Bounce-3D: Single-Shot Occlusion-Aware 3D from Lidar by Decomposing Two-Bounce Light", "abstract": "3D scene reconstruction from a single measurement is challenging, especially in the presence of occluded regions and specular materials, such as mirrors. We address these challenges by leveraging single-photon lidars. These lidars estimate depth from light that is emitted into the scene and reflected directly back to the sensor. However, they can also measure light that bounces multiple times in the scene before reaching the sensor. This multi-bounce light contains additional information that can be used to recover dense depth, occluded geometry, and material properties. Prior work with single-photon lidar, however, has only demonstrated these use cases when a laser sequentially illuminates one scene point at a time. We instead focus on the more practical - and challenging - scenario of illuminating multiple scene points simultaneously. The complexity of light transport due to the combined effects of multiplexed illumination, two-bounce light, shadows, and specular reflections is challenging to invert analytically. Instead, we propose a data-driven method to invert light transport in single-photon lidar. To enable this approach, we create the first large-scale simulated dataset of ~100k lidar transients for indoor scenes. We use this dataset to learn a prior on complex light transport, enabling measured two-bounce light to be decomposed into the constituent contributions from each laser spot. Finally, we experimentally demonstrate how this decomposed light can be used to infer 3D geometry in scenes with occlusions and mirrors from a single measurement. Our code and dataset are released at https://shoot-bounce-3d.github.io.", "date": "2025-12-05", "categories": []}
{"id": "2512.06818", "title": "MeshSplatting: Differentiable Rendering with Opaque Meshes", "abstract": "Primitive-based splatting methods like 3D Gaussian Splatting have revolutionized novel view synthesis with real-time rendering. However, their point-based representations remain incompatible with mesh-based pipelines that power AR/VR and game engines. We present MeshSplatting, a mesh-based reconstruction approach that jointly optimizes geometry and appearance through differentiable rendering. By enforcing connectivity via restricted Delaunay triangulation and refining surface consistency, MeshSplatting creates end-to-end smooth, visually high-quality meshes that render efficiently in real-time 3D engines. On Mip-NeRF360, it boosts PSNR by +0.69 dB over the current state-of-the-art MiLo for mesh-based novel view synthesis, while training 2x faster and using 2x less memory, bridging neural rendering and interactive 3D graphics for seamless real-time scene interaction. The project page is available at https://meshsplatting.github.io/.", "date": "2025-12-07", "categories": []}
{"id": "2512.07165", "title": "MuSASplat: Efficient Sparse-View 3D Gaussian Splats via Lightweight Multi-Scale Adaptation", "abstract": "Sparse-view 3D Gaussian splatting seeks to render high-quality novel views of 3D scenes from a limited set of input images. While recent pose-free feed-forward methods leveraging pre-trained 3D priors have achieved impressive results, most of them rely on full fine-tuning of large Vision Transformer (ViT) backbones and incur substantial GPU costs. In this work, we introduce MuSASplat, a novel framework that dramatically reduces the computational burden of training pose-free feed-forward 3D Gaussian splats models with little compromise of rendering quality. Central to our approach is a lightweight Multi-Scale Adapter that enables efficient fine-tuning of ViT-based architectures with only a small fraction of training parameters. This design avoids the prohibitive GPU overhead associated with previous full-model adaptation techniques while maintaining high fidelity in novel view synthesis, even with very sparse input views. In addition, we introduce a Feature Fusion Aggregator that integrates features across input views effectively and efficiently. Unlike widely adopted memory banks, the Feature Fusion Aggregator ensures consistent geometric integration across input views and meanwhile mitigates the memory usage, training complexity, and computational costs significantly. Extensive experiments across diverse datasets show that MuSASplat achieves state-of-the-art rendering quality but has significantly reduced parameters and training resource requirements as compared with existing methods.", "date": "2025-12-08", "categories": []}
{"id": "2512.07230", "title": "STRinGS: Selective Text Refinement in Gaussian Splatting", "abstract": "Text as signs, labels, or instructions is a critical element of real-world scenes as they can convey important contextual information. 3D representations such as 3D Gaussian Splatting (3DGS) struggle to preserve fine-grained text details, while achieving high visual fidelity. Small errors in textual element reconstruction can lead to significant semantic loss. We propose STRinGS, a text-aware, selective refinement framework to address this issue for 3DGS reconstruction. Our method treats text and non-text regions separately, refining text regions first and merging them with non-text regions later for full-scene optimization. STRinGS produces sharp, readable text even in challenging configurations. We introduce a text readability measure OCR Character Error Rate (CER) to evaluate the efficacy on text regions. STRinGS results in a 63.6% relative improvement over 3DGS at just 7K iterations. We also introduce a curated dataset STRinGS-360 with diverse text scenarios to evaluate text readability in 3D reconstruction. Our method and dataset together push the boundaries of 3D scene understanding in text-rich environments, paving the way for more robust text-aware reconstruction methods.", "date": "2025-12-08", "categories": []}
{"id": "2512.07381", "title": "Tessellation GS: Neural Mesh Gaussians for Robust Monocular Reconstruction of Dynamic Objects", "abstract": "3D Gaussian Splatting (GS) enables highly photorealistic scene reconstruction from posed image sequences but struggles with viewpoint extrapolation due to its anisotropic nature, leading to overfitting and poor generalization, particularly in sparse-view and dynamic scene reconstruction. We propose Tessellation GS, a structured 2D GS approach anchored on mesh faces, to reconstruct dynamic scenes from a single continuously moving or static camera. Our method constrains 2D Gaussians to localized regions and infers their attributes via hierarchical neural features on mesh faces. Gaussian subdivision is guided by an adaptive face subdivision strategy driven by a detail-aware loss function. Additionally, we leverage priors from a reconstruction foundation model to initialize Gaussian deformations, enabling robust reconstruction of general dynamic objects from a single static camera, previously extremely challenging for optimization-based methods. Our method outperforms previous SOTA method, reducing LPIPS by 29.1% and Chamfer distance by 49.2% on appearance and mesh reconstruction tasks.", "date": "2025-12-08", "categories": []}
{"id": "2506.13348", "title": "TextureSplat: Per-Primitive Texture Mapping for Reflective Gaussian Splatting", "abstract": "Gaussian Splatting have demonstrated remarkable novel view synthesis performance at high rendering frame rates. Optimization-based inverse rendering within complex capture scenarios remains however a challenging problem. A particular case is modelling complex surface light interactions for highly reflective scenes, which results in intricate high frequency specular radiance components. We hypothesize that such challenging settings can benefit from increased representation power. We hence propose a method that tackles this issue through a geometrically and physically grounded Gaussian Splatting borne radiance field, where normals and material properties are spatially variable in the primitive's local space. Using per-primitive texture maps for this purpose, we also propose to harness the GPU hardware to accelerate rendering at test time via unified material texture atlas. Code will be available at https://github.com/maeyounes/TextureSplat", "date": "2025-06-16", "categories": []}
{"id": "2512.05131", "title": "AREA3D: Active Reconstruction Agent with Unified Feed-Forward 3D Perception and Vision-Language Guidance", "abstract": "Active 3D reconstruction enables an agent to autonomously select viewpoints to efficiently obtain accurate and complete scene geometry, rather than passively reconstructing scenes from pre-collected images. However, existing active reconstruction methods often rely on hand-crafted geometric heuristics, which can lead to redundant observations without substantially improving reconstruction quality. To address this limitation, we propose AREA3D, an active reconstruction agent that leverages feed-forward 3D reconstruction models and vision-language guidance. Our framework decouples view-uncertainty modeling from the underlying feed-forward reconstructor, enabling precise uncertainty estimation without expensive online optimization. In addition, an integrated vision-language model provides high-level semantic guidance, encouraging informative and diverse viewpoints beyond purely geometric cues. Extensive experiments on both scene-level and object-level benchmarks demonstrate that AREA3D achieves state-of-the-art reconstruction accuracy, particularly in the sparse-view regime. Code will be made available at: https://github.com/TianlingXu/AREA3D .", "date": "2025-11-28", "categories": []}
{"id": "2512.05783", "title": "Curvature-Regularized Variational Autoencoder for 3D Scene Reconstruction from Sparse Depth", "abstract": "When depth sensors provide only 5% of needed measurements, reconstructing complete 3D scenes becomes difficult. Autonomous vehicles and robots cannot tolerate the geometric errors that sparse reconstruction introduces. We propose curvature regularization through a discrete Laplacian operator, achieving 18.1% better reconstruction accuracy than standard variational autoencoders. Our contribution challenges an implicit assumption in geometric deep learning: that combining multiple geometric constraints improves performance. A single well-designed regularization term not only matches but exceeds the effectiveness of complex multi-term formulations. The discrete Laplacian offers stable gradients and noise suppression with just 15% training overhead and zero inference cost. Code and models are available at https://github.com/Maryousefi/GeoVAE-3D.", "date": "2025-12-05", "categories": []}
{"id": "2405.02568", "title": "Surface-Based Visibility-Guided Uncertainty for Continuous Active 3D Neural Reconstruction", "abstract": "View selection is critical in active 3D neural reconstruction as it impacts the contents of training set and resulting final output quality. Recent view selection strategies emphasize the visibility when evaluating model uncertainty in active 3D reconstruction. However, existing approaches estimate visibility only after the model fully converges, which has confined their application primarily to non-continuous active learning settings. This paper proposes Surface-Based Visibility field (SBV) that successfully estimates the visibility-guided uncertainty in continuous active 3D neural reconstruction. During learning neural implicit surfaces, our model learns rendering uncertainties and infers surface confidence values derived from signed distance functions. It then updates surface confidences using a voxel grid, robustly deducing the surface-based visibility for uncertainties. This approach captures uncertainties across all regions, whether well-defined surfaces or ambiguous areas, ensuring accurate visibility measurement in continuous active learning. Experiments on benchmark datasets-Tanks and Temples, BlendedMVS, Blender, DTU-and the newly proposed imbalanced viewpoint dataset (ImBView) show that view selection based on SBV-guided uncertainty improves performance by up to 11.6% over existing methods, highlighting its effectiveness in challenging reconstruction scenarios.", "date": "2024-05-04", "categories": []}
{"id": "2512.04528", "title": "Auto3R: Automated 3D Reconstruction and Scanning via Data-driven Uncertainty Quantification", "abstract": "Traditional high-quality 3D scanning and reconstruction typically relies on human labor to plan the scanning procedure. With the rapid development of embodied systems such as drones and robots, there is a growing demand of performing accurate 3D scanning and reconstruction in an fully automated manner. We introduce Auto3R, a data-driven uncertainty quantification model that is designed to automate the 3D scanning and reconstruction of scenes and objects, including objects with non-lambertian and specular materials. Specifically, in a process of iterative 3D reconstruction and scanning, Auto3R can make efficient and accurate prediction of uncertainty distribution over potential scanning viewpoints, without knowing the ground truth geometry and appearance. Through extensive experiments, Auto3R achieves superior performance that outperforms the state-of-the-art methods by a large margin. We also deploy Auto3R on a robot arm equipped with a camera and demonstrate that Auto3R can be used to effectively digitize real-world 3D objects and delivers ready-to-use and photorealistic digital assets. Our homepage: https://tomatoma00.github.io/auto3r.github.io .", "date": "2025-12-04", "categories": []}
{"id": "2512.04542", "title": "Gaussian Entropy Fields: Driving Adaptive Sparsity in 3D Gaussian Optimization", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a leading technique for novel view synthesis, demonstrating exceptional rendering efficiency. \\replaced[]{Well-reconstructed surfaces can be characterized by low configurational entropy, where dominant primitives clearly define surface geometry while redundant components are suppressed.}{The key insight is that well-reconstructed surfaces naturally exhibit low configurational entropy, where dominant primitives clearly define surface geometry while suppressing redundant components.} Three complementary technical contributions are introduced: (1) entropy-driven surface modeling via entropy minimization for low configurational entropy in primitive distributions; (2) adaptive spatial regularization using the Surface Neighborhood Redundancy Index (SNRI) and image entropy-guided weighting; (3) multi-scale geometric preservation through competitive cross-scale entropy alignment. Extensive experiments demonstrate that GEF achieves competitive geometric precision on DTU and T\\&T benchmarks, while delivering superior rendering quality compared to existing methods on Mip-NeRF 360. Notably, superior Chamfer Distance (0.64) on DTU and F1 score (0.44) on T\\&T are obtained, alongside the best SSIM (0.855) and LPIPS (0.136) among baselines on Mip-NeRF 360, validating the framework's ability to enhance surface reconstruction accuracy without compromising photometric fidelity.", "date": "2025-12-04", "categories": []}
{"id": "2503.06859", "title": "ActiveInitSplat: How Active Image Selection Helps Gaussian Splatting", "abstract": "Gaussian splatting (GS) along with its extensions and variants provides outstanding performance in real-time scene rendering while meeting reduced storage demands and computational efficiency. While the selection of 2D images capturing the scene of interest is crucial for the proper initialization and training of GS, hence markedly affecting the rendering performance, prior works rely on passively and typically densely selected 2D images. In contrast, this paper proposes `ActiveInitSplat', a novel framework for active selection of training images for proper initialization and training of GS. ActiveInitSplat relies on density and occupancy criteria of the resultant 3D scene representation from the selected 2D images, to ensure that the latter are captured from diverse viewpoints leading to better scene coverage and that the initialized Gaussian functions are well aligned with the actual 3D structure. Numerical tests on well-known simulated and real environments demonstrate the merits of ActiveInitSplat resulting in significant GS rendering performance improvement over passive GS baselines in both dense- and sparse-view settings, in the widely adopted LPIPS, SSIM, and PSNR metrics.", "date": "2025-03-10", "categories": []}
{"id": "2512.03939", "title": "MUT3R: Motion-aware Updating Transformer for Dynamic 3D Reconstruction", "abstract": "Recent stateful recurrent neural networks have achieved remarkable progress on static 3D reconstruction but remain vulnerable to motion-induced artifacts, where non-rigid regions corrupt attention propagation between the spatial memory and image feature. By analyzing the internal behaviors of the state and image token updating mechanism, we find that aggregating self-attention maps across layers reveals a consistent pattern: dynamic regions are naturally down-weighted, exposing an implicit motion cue that the pretrained transformer already encodes but never explicitly uses. Motivated by this observation, we introduce MUT3R, a training-free framework that applies the attention-derived motion cue to suppress dynamic content in the early layers of the transformer during inference. Our attention-level gating module suppresses the influence of dynamic regions before their artifacts propagate through the feature hierarchy. Notably, we do not retrain or fine-tune the model; we let the pretrained transformer diagnose its own motion cues and correct itself. This early regulation stabilizes geometric reasoning in streaming scenarios and leads to improvements in temporal consistency and camera pose robustness across multiple dynamic benchmarks, offering a simple and training-free pathway toward motion-aware streaming reconstruction.", "date": "2025-12-03", "categories": []}
{"id": "2512.04012", "title": "Emergent Outlier View Rejection in Visual Geometry Grounded Transformers", "abstract": "Reliable 3D reconstruction from in-the-wild image collections is often hindered by \"noisy\" images-irrelevant inputs with little or no view overlap with others. While traditional Structure-from-Motion pipelines handle such cases through geometric verification and outlier rejection, feed-forward 3D reconstruction models lack these explicit mechanisms, leading to degraded performance under in-the-wild conditions. In this paper, we discover that the existing feed-forward reconstruction model, e.g., VGGT, despite lacking explicit outlier-rejection mechanisms or noise-aware training, can inherently distinguish distractor images. Through an in-depth analysis under varying proportions of synthetic distractors, we identify a specific layer that naturally exhibits outlier-suppressing behavior. Further probing reveals that this layer encodes discriminative internal representations that enable an effective noise-filtering capability, which we simply leverage to perform outlier-view rejection in feed-forward 3D reconstruction without any additional fine-tuning or supervision. Extensive experiments on both controlled and in-the-wild datasets demonstrate that this implicit filtering mechanism is consistent and generalizes well across diverse scenarios.", "date": "2025-12-03", "categories": []}
{"id": "2512.04076", "title": "Radiance Meshes for Volumetric Reconstruction", "abstract": "We introduce radiance meshes, a technique for representing radiance fields with constant density tetrahedral cells produced with a Delaunay tetrahedralization. Unlike a Voronoi diagram, a Delaunay tetrahedralization yields simple triangles that are natively supported by existing hardware. As such, our model is able to perform exact and fast volume rendering using both rasterization and ray-tracing. We introduce a new rasterization method that achieves faster rendering speeds than all prior radiance field representations (assuming an equivalent number of primitives and resolution) across a variety of platforms. Optimizing the positions of Delaunay vertices introduces topological discontinuities (edge flips). To solve this, we use a Zip-NeRF-style backbone which allows us to express a smoothly varying field even when the topology changes. Our rendering method exactly evaluates the volume rendering equation and enables high quality, real-time view synthesis on standard consumer hardware. Our tetrahedral meshes also lend themselves to a variety of exciting applications including fisheye lens distortion, physics-based simulation, editing, and mesh extraction.", "date": "2025-12-03", "categories": []}
{"id": "2503.14537", "title": "Learning-based 3D Reconstruction in Autonomous Driving: A Comprehensive Survey", "abstract": "Learning-based 3D reconstruction has emerged as a transformative technique in autonomous driving, enabling precise modeling of environments through advanced neural representations. It has inspired pioneering solutions for vital tasks in autonomous driving, such as dense mapping and closed-loop simulation, as well as comprehensive scene feature for driving scene understanding and reasoning. Given the rapid growth in related research, this survey provides a comprehensive review of both technical evolutions and practical applications in autonomous driving. We begin with an introduction to the preliminaries of learning-based 3D reconstruction to provide a solid technical background foundation, then progress to a rigorous, multi-dimensional examination of cutting-edge methodologies, systematically organized according to the distinctive technical requirements and fundamental challenges of autonomous driving. Through analyzing and summarizing development trends and cutting-edge research, we identify existing technical challenges, along with insufficient disclosure of on-board validation and safety verification details in the current literature, and ultimately suggest potential directions to guide future studies.", "date": "2025-03-17", "categories": []}
{"id": "2504.03059", "title": "Compressing 3D Gaussian Splatting by Noise-Substituted Vector Quantization", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated remarkable effectiveness in 3D reconstruction, achieving high-quality results with real-time radiance field rendering. However, a key challenge is the substantial storage cost: reconstructing a single scene typically requires millions of Gaussian splats, each represented by 59 floating-point parameters, resulting in approximately 1 GB of memory. To address this challenge, we propose a compression method by building separate attribute codebooks and storing only discrete code indices. Specifically, we employ noise-substituted vector quantization technique to jointly train the codebooks and model features, ensuring consistency between gradient descent optimization and parameter discretization. Our method reduces the memory consumption efficiently (around $45\\times$) while maintaining competitive reconstruction quality on standard 3D benchmark scenes. Experiments on different codebook sizes show the trade-off between compression ratio and image quality. Furthermore, the trained compressed model remains fully compatible with popular 3DGS viewers and enables faster rendering speed, making it well-suited for practical applications.", "date": "2025-04-03", "categories": []}
{"id": "2507.22052", "title": "Ov3R: Open-Vocabulary Semantic 3D Reconstruction from RGB Videos", "abstract": "We present Ov3R, a novel framework for open-vocabulary semantic 3D reconstruction from RGB video streams, designed to advance Spatial AI. The system features two key components: CLIP3R, a CLIP-informed 3D reconstruction module that predicts dense point maps from overlapping clips while embedding object-level semantics; and 2D-3D OVS, a 2D-3D open-vocabulary semantic module that lifts 2D features into 3D by learning fused descriptors integrating spatial, geometric, and semantic cues. Unlike prior methods, Ov3R incorporates CLIP semantics directly into the reconstruction process, enabling globally consistent geometry and fine-grained semantic alignment. Our framework achieves state-of-the-art performance in both dense 3D reconstruction and open-vocabulary 3D segmentation, marking a step forward toward real-time, semantics-aware Spatial AI.", "date": "2025-07-29", "categories": []}
{"id": "2512.02621", "title": "Content-Aware Texturing for Gaussian Splatting", "abstract": "Gaussian Splatting has become the method of choice for 3D reconstruction and real-time rendering of captured real scenes. However, fine appearance details need to be represented as a large number of small Gaussian primitives, which can be wasteful when geometry and appearance exhibit different frequency characteristics.\n  Inspired by the long tradition of texture mapping, we propose to use texture to represent detailed appearance where possible. Our main focus is to incorporate per-primitive texture maps that adapt to the scene in a principled manner during Gaussian Splatting optimization. We do this by proposing a new appearance representation for 2D Gaussian primitives with textures where the size of a texel is bounded by the image sampling frequency and adapted to the content of the input images. We achieve this by adaptively upscaling or downscaling the texture resolution during optimization. In addition, our approach enables control of the number of primitives during optimization based on texture resolution. We show that our approach performs favorably in image quality and total number of parameters used compared to alternative solutions for textured Gaussian primitives. Project page: https://repo-sam.inria.fr/nerphys/gs-texturing/", "date": "2025-12-02", "categories": []}
{"id": "2512.03004", "title": "DGGT: Feedforward 4D Reconstruction of Dynamic Driving Scenes using Unposed Images", "abstract": "Autonomous driving needs fast, scalable 4D reconstruction and re-simulation for training and evaluation, yet most methods for dynamic driving scenes still rely on per-scene optimization, known camera calibration, or short frame windows, making them slow and impractical. We revisit this problem from a feedforward perspective and introduce \\textbf{Driving Gaussian Grounded Transformer (DGGT)}, a unified framework for pose-free dynamic scene reconstruction. We note that the existing formulations, treating camera pose as a required input, limit flexibility and scalability. Instead, we reformulate pose as an output of the model, enabling reconstruction directly from sparse, unposed images and supporting an arbitrary number of views for long sequences. Our approach jointly predicts per-frame 3D Gaussian maps and camera parameters, disentangles dynamics with a lightweight dynamic head, and preserves temporal consistency with a lifespan head that modulates visibility over time. A diffusion-based rendering refinement further reduces motion/interpolation artifacts and improves novel-view quality under sparse inputs. The result is a single-pass, pose-free algorithm that achieves state-of-the-art performance and speed. Trained and evaluated on large-scale driving benchmarks (Waymo, nuScenes, Argoverse2), our method outperforms prior work both when trained on each dataset and in zero-shot transfer across datasets, and it scales well as the number of input frames increases.", "date": "2025-12-02", "categories": []}
{"id": "2506.08015", "title": "4DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videos", "abstract": "We propose 4DGT, a 4D Gaussian-based Transformer model for dynamic scene reconstruction, trained entirely on real-world monocular posed videos. Using 4D Gaussian as an inductive bias, 4DGT unifies static and dynamic components, enabling the modeling of complex, time-varying environments with varying object lifespans. We proposed a novel density control strategy in training, which enables our 4DGT to handle longer space-time input and remain efficient rendering at runtime. Our model processes 64 consecutive posed frames in a rolling-window fashion, predicting consistent 4D Gaussians in the scene. Unlike optimization-based methods, 4DGT performs purely feed-forward inference, reducing reconstruction time from hours to seconds and scaling effectively to long video sequences. Trained only on large-scale monocular posed video datasets, 4DGT can outperform prior Gaussian-based networks significantly in real-world videos and achieve on-par accuracy with optimization-based methods on cross-domain videos. Project page: https://4dgt.github.io", "date": "2025-06-09", "categories": []}
{"id": "2512.00300", "title": "TGSFormer: Scalable Temporal Gaussian Splatting for Embodied Semantic Scene Completion", "abstract": "Embodied 3D Semantic Scene Completion (SSC) infers dense geometry and semantics from continuous egocentric observations. Most existing Gaussian-based methods rely on random initialization of many primitives within predefined spatial bounds, resulting in redundancy and poor scalability to unbounded scenes. Recent depth-guided approach alleviates this issue but remains local, suffering from latency and memory overhead as scale increases. To overcome these challenges, we propose TGSFormer, a scalable Temporal Gaussian Splatting framework for embodied SSC. It maintains a persistent Gaussian memory for temporal prediction, without relying on image coherence or frame caches. For temporal fusion, a Dual Temporal Encoder jointly processes current and historical Gaussian features through confidence-aware cross-attention. Subsequently, a Confidence-aware Voxel Fusion module merges overlapping primitives into voxel-aligned representations, regulating density and maintaining compactness. Extensive experiments demonstrate that TGSFormer achieves state-of-the-art results on both local and embodied SSC benchmarks, offering superior accuracy and scalability with significantly fewer primitives while maintaining consistent long-term scene integrity. The code will be released upon acceptance.", "date": "2025-11-29", "categories": []}
{"id": "2512.00534", "title": "Cross-Temporal 3D Gaussian Splatting for Sparse-View Guided Scene Update", "abstract": "Maintaining consistent 3D scene representations over time is a significant challenge in computer vision. Updating 3D scenes from sparse-view observations is crucial for various real-world applications, including urban planning, disaster assessment, and historical site preservation, where dense scans are often unavailable or impractical. In this paper, we propose Cross-Temporal 3D Gaussian Splatting (Cross-Temporal 3DGS), a novel framework for efficiently reconstructing and updating 3D scenes across different time periods, using sparse images and previously captured scene priors. Our approach comprises three stages: 1) Cross-temporal camera alignment for estimating and aligning camera poses across different timestamps; 2) Interference-based confidence initialization to identify unchanged regions between timestamps, thereby guiding updates; and 3) Progressive cross-temporal optimization, which iteratively integrates historical prior information into the 3D scene to enhance reconstruction quality. Our method supports non-continuous capture, enabling not only updates using new sparse views to refine existing scenes, but also recovering past scenes from limited data with the help of current captures. Furthermore, we demonstrate the potential of this approach to achieve temporal changes using only sparse images, which can later be reconstructed into detailed 3D representations as needed. Experimental results show significant improvements over baseline methods in reconstruction quality and data efficiency, making this approach a promising solution for scene versioning, cross-temporal digital twins, and long-term spatial documentation.", "date": "2025-11-29", "categories": []}
{"id": "2512.00547", "title": "Asset-Driven Sematic Reconstruction of Dynamic Scene with Multi-Human-Object Interactions", "abstract": "Real-world human-built environments are highly dynamic, involving multiple humans and their complex interactions with surrounding objects. While 3D geometry modeling of such scenes is crucial for applications like AR/VR, gaming, and embodied AI, it remains underexplored due to challenges like diverse motion patterns and frequent occlusions. Beyond novel view rendering, 3D Gaussian Splatting (GS) has demonstrated remarkable progress in producing detailed, high-quality surface geometry with fast optimization of the underlying structure. However, very few GS-based methods address multihuman, multiobject scenarios, primarily due to the above-mentioned inherent challenges. In a monocular setup, these challenges are further amplified, as maintaining structural consistency under severe occlusion becomes difficult when the scene is optimized solely based on GS-based rendering loss. To tackle the challenges of such a multihuman, multiobject dynamic scene, we propose a hybrid approach that effectively combines the advantages of 1) 3D generative models for generating high-fidelity meshes of the scene elements, 2) Semantic-aware deformation, \\ie rigid transformation of the rigid objects and LBS-based deformation of the humans, and mapping of the deformed high-fidelity meshes in the dynamic scene, and 3) GS-based optimization of the individual elements for further refining their alignments in the scene. Such a hybrid approach helps maintain the object structures even under severe occlusion and can produce multiview and temporally consistent geometry. We choose HOI-M3 for evaluation, as, to the best of our knowledge, this is the only dataset featuring multihuman, multiobject interactions in a dynamic scene. Our method outperforms the state-of-the-art method in producing better surface reconstruction of such scenes.", "date": "2025-11-29", "categories": []}
{"id": "2512.00794", "title": "PolarGS: Polarimetric Cues for Ambiguity-Free Gaussian Splatting with Accurate Geometry Recovery", "abstract": "Recent advances in surface reconstruction for 3D Gaussian Splatting (3DGS) have enabled remarkable geometric accuracy. However, their performance degrades in photometrically ambiguous regions such as reflective and textureless surfaces, where unreliable cues disrupt photometric consistency and hinder accurate geometry estimation. Reflected light is often partially polarized in a manner that reveals surface orientation, making polarization an optic complement to photometric cues in resolving such ambiguities. Therefore, we propose PolarGS, an optics-aware extension of RGB-based 3DGS that leverages polarization as an optical prior to resolve photometric ambiguities and enhance reconstruction accuracy. Specifically, we introduce two complementary modules: a polarization-guided photometric correction strategy, which ensures photometric consistency by identifying reflective regions via the Degree of Linear Polarization (DoLP) and refining reflective Gaussians with Color Refinement Maps; and a polarization-enhanced Gaussian densification mechanism for textureless area geometry recovery, which integrates both Angle and Degree of Linear Polarization (A/DoLP) into a PatchMatch-based depth completion process. This enables the back-projection and fusion of new Gaussians, leading to more complete reconstruction. PolarGS is framework-agnostic and achieves superior geometric accuracy compared to state-of-the-art methods.", "date": "2025-11-30", "categories": []}
{"id": "2512.01296", "title": "EGG-Fusion: Efficient 3D Reconstruction with Geometry-aware Gaussian Surfel on the Fly", "abstract": "Real-time 3D reconstruction is a fundamental task in computer graphics. Recently, differentiable-rendering-based SLAM system has demonstrated significant potential, enabling photorealistic scene rendering through learnable scene representations such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Current differentiable rendering methods face dual challenges in real-time computation and sensor noise sensitivity, leading to degraded geometric fidelity in scene reconstruction and limited practicality. To address these challenges, we propose a novel real-time system EGG-Fusion, featuring robust sparse-to-dense camera tracking and a geometry-aware Gaussian surfel mapping module, introducing an information filter-based fusion method that explicitly accounts for sensor noise to achieve high-precision surface reconstruction. The proposed differentiable Gaussian surfel mapping effectively models multi-view consistent surfaces while enabling efficient parameter optimization. Extensive experimental results demonstrate that the proposed system achieves a surface reconstruction error of 0.6\\textit{cm} on standardized benchmark datasets including Replica and ScanNet++, representing over 20\\% improvement in accuracy compared to state-of-the-art (SOTA) GS-based methods. Notably, the system maintains real-time processing capabilities at 24 FPS, establishing it as one of the most accurate differentiable-rendering-based real-time reconstruction systems. Project Page: https://zju3dv.github.io/eggfusion/", "date": "2025-12-01", "categories": []}
{"id": "2512.01329", "title": "TagSplat: Topology-Aware Gaussian Splatting for Dynamic Mesh Modeling and Tracking", "abstract": "Topology-consistent dynamic model sequences are essential for applications such as animation and model editing. However, existing 4D reconstruction methods face challenges in generating high-quality topology-consistent meshes. To address this, we propose a topology-aware dynamic reconstruction framework based on Gaussian Splatting. We introduce a Gaussian topological structure that explicitly encodes spatial connectivity. This structure enables topology-aware densification and pruning, preserving the manifold consistency of the Gaussian representation. Temporal regularization terms further ensure topological coherence over time, while differentiable mesh rasterization improves mesh quality. Experimental results demonstrate that our method reconstructs topology-consistent mesh sequences with significantly higher accuracy than existing approaches. Moreover, the resulting meshes enable precise 3D keypoint tracking. Project page: https://haza628.github.io/tagSplat/", "date": "2025-12-01", "categories": []}
{"id": "2507.02863", "title": "Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory", "abstract": "Dense 3D scene reconstruction from an ordered sequence or unordered image collections is a critical step when bringing research in computer vision into practical scenarios. Following the paradigm introduced by DUSt3R, which unifies an image pair densely into a shared coordinate system, subsequent methods maintain an implicit memory to achieve dense 3D reconstruction from more images. However, such implicit memory is limited in capacity and may suffer from information loss of earlier frames. We propose Point3R, an online framework targeting dense streaming 3D reconstruction. To be specific, we maintain an explicit spatial pointer memory directly associated with the 3D structure of the current scene. Each pointer in this memory is assigned a specific 3D position and aggregates scene information nearby in the global coordinate system into a changing spatial feature. Information extracted from the latest frame interacts explicitly with this pointer memory, enabling dense integration of the current observation into the global coordinate system. We design a 3D hierarchical position embedding to promote this interaction and design a simple yet effective fusion mechanism to ensure that our pointer memory is uniform and efficient. Our method achieves competitive or state-of-the-art performance on various tasks with low training costs. Code: https://github.com/YkiWu/Point3R.", "date": "2025-07-03", "categories": []}
{"id": "2510.07119", "title": "MoRe: Monocular Geometry Refinement via Graph Optimization for Cross-View Consistency", "abstract": "Monocular 3D foundation models offer an extensible solution for perception tasks, making them attractive for broader 3D vision applications. In this paper, we propose MoRe, a training-free Monocular Geometry Refinement method designed to improve cross-view consistency and achieve scale alignment. To induce inter-frame relationships, our method employs feature matching between frames to establish correspondences. Rather than applying simple least squares optimization on these matched points, we formulate a graph-based optimization framework that performs local planar approximation using the estimated 3D points and surface normals estimated by monocular foundation models. This formulation addresses the scale ambiguity inherent in monocular geometric priors while preserving the underlying 3D structure. We further demonstrate that MoRe not only enhances 3D reconstruction but also improves novel view synthesis, particularly in sparse view rendering scenarios.", "date": "2025-10-08", "categories": []}
{"id": "2511.22233", "title": "IE-SRGS: An Internal-External Knowledge Fusion Framework for High-Fidelity 3D Gaussian Splatting Super-Resolution", "abstract": "Reconstructing high-resolution (HR) 3D Gaussian Splatting (3DGS) models from low-resolution (LR) inputs remains challenging due to the lack of fine-grained textures and geometry. Existing methods typically rely on pre-trained 2D super-resolution (2DSR) models to enhance textures, but suffer from 3D Gaussian ambiguity arising from cross-view inconsistencies and domain gaps inherent in 2DSR models. We propose IE-SRGS, a novel 3DGS SR paradigm that addresses this issue by jointly leveraging the complementary strengths of external 2DSR priors and internal 3DGS features. Specifically, we use 2DSR and depth estimation models to generate HR images and depth maps as external knowledge, and employ multi-scale 3DGS models to produce cross-view consistent, domain-adaptive counterparts as internal knowledge. A mask-guided fusion strategy is introduced to integrate these two sources and synergistically exploit their complementary strengths, effectively guiding the 3D Gaussian optimization toward high-fidelity reconstruction. Extensive experiments on both synthetic and real-world benchmarks show that IE-SRGS consistently outperforms state-of-the-art methods in both quantitative accuracy and visual fidelity.", "date": "2025-11-27", "categories": []}
{"id": "2511.22429", "title": "Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation", "abstract": "We present Fin3R, a simple, effective, and general fine-tuning method for feed-forward 3D reconstruction models. The family of feed-forward reconstruction model regresses pointmap of all input images to a reference frame coordinate system, along with other auxiliary outputs, in a single forward pass. However, we find that current models struggle with fine geometry and robustness due to (\\textit{i}) the scarcity of high-fidelity depth and pose supervision and (\\textit{ii}) the inherent geometric misalignment from multi-view pointmap regression. Fin3R jointly tackles two issues with an extra lightweight fine-tuning step. We freeze the decoder, which handles view matching, and fine-tune only the image encoder-the component dedicated to feature extraction. The encoder is enriched with fine geometric details distilled from a strong monocular teacher model on large, unlabeled datasets, using a custom, lightweight LoRA adapter. We validate our method on a wide range of models, including DUSt3R, MASt3R, CUT3R, and VGGT. The fine-tuned models consistently deliver sharper boundaries, recover complex structures, and achieve higher geometric accuracy in both single- and multi-view settings, while adding only the tiny LoRA weights, which leave test-time memory and latency virtually unchanged. Project page: \\href{http://visual-ai.github.io/fin3r}{https://visual-ai.github.io/fin3r}", "date": "2025-11-27", "categories": []}
{"id": "2511.22459", "title": "Gaussians on Fire: High-Frequency Reconstruction of Flames", "abstract": "We propose a method to reconstruct dynamic fire in 3D from a limited set of camera views with a Gaussian-based spatiotemporal representation. Capturing and reconstructing fire and its dynamics is highly challenging due to its volatile nature, transparent quality, and multitude of high-frequency features. Despite these challenges, we aim to reconstruct fire from only three views, which consequently requires solving for under-constrained geometry. We solve this by separating the static background from the dynamic fire region by combining dense multi-view stereo images with monocular depth priors. The fire is initialized as a 3D flow field, obtained by fusing per-view dense optical flow projections. To capture the high frequency features of fire, each 3D Gaussian encodes a lifetime and linear velocity to match the dense optical flow. To ensure sub-frame temporal alignment across cameras we employ a custom hardware synchronization pattern -- allowing us to reconstruct fire with affordable commodity hardware. Our quantitative and qualitative validations across numerous reconstruction experiments demonstrate robust performance for diverse and challenging real fire scenarios.", "date": "2025-11-27", "categories": []}
{"id": "2511.22997", "title": "MrGS: Multi-modal Radiance Fields with 3D Gaussian Splatting for RGB-Thermal Novel View Synthesis", "abstract": "Recent advances in Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) have achieved considerable performance in RGB scene reconstruction. However, multi-modal rendering that incorporates thermal infrared imagery remains largely underexplored. Existing approaches tend to neglect distinctive thermal characteristics, such as heat conduction and the Lambertian property. In this study, we introduce MrGS, a multi-modal radiance field based on 3DGS that simultaneously reconstructs both RGB and thermal 3D scenes. Specifically, MrGS derives RGB- and thermal-related information from a single appearance feature through orthogonal feature extraction and employs view-dependent or view-independent embedding strategies depending on the degree of Lambertian reflectance exhibited by each modality. Furthermore, we leverage two physics-based principles to effectively model thermal-domain phenomena. First, we integrate Fourier's law of heat conduction prior to alpha blending to model intensity interpolation caused by thermal conduction between neighboring Gaussians. Second, we apply the Stefan-Boltzmann law and the inverse-square law to formulate a depth-aware thermal radiation map that imposes additional geometric constraints on thermal rendering. Experimental results demonstrate that the proposed MrGS achieves high-fidelity RGB-T scene reconstruction while reducing the number of Gaussians.", "date": "2025-11-28", "categories": []}
{"id": "2511.23052", "title": "Image Valuation in NeRF-based 3D reconstruction", "abstract": "Data valuation and monetization are becoming increasingly important across domains such as eXtended Reality (XR) and digital media. In the context of 3D scene reconstruction from a set of images -- whether casually or professionally captured -- not all inputs contribute equally to the final output. Neural Radiance Fields (NeRFs) enable photorealistic 3D reconstruction of scenes by optimizing a volumetric radiance field given a set of images. However, in-the-wild scenes often include image captures of varying quality, occlusions, and transient objects, resulting in uneven utility across inputs. In this paper we propose a method to quantify the individual contribution of each image to NeRF-based reconstructions of in-the-wild image sets. Contribution is assessed through reconstruction quality metrics based on PSNR and MSE. We validate our approach by removing low-contributing images during training and measuring the resulting impact on reconstruction fidelity.", "date": "2025-11-28", "categories": []}
{"id": "2511.21098", "title": "Pygmalion Effect in Vision: Image-to-Clay Translation for Reflective Geometry Reconstruction", "abstract": "Understanding reflection remains a long-standing challenge in 3D reconstruction due to the entanglement of appearance and geometry under view-dependent reflections. In this work, we present the Pygmalion Effect in Vision, a novel framework that metaphorically \"sculpts\" reflective objects into clay-like forms through image-to-clay translation. Inspired by the myth of Pygmalion, our method learns to suppress specular cues while preserving intrinsic geometric consistency, enabling robust reconstruction from multi-view images containing complex reflections. Specifically, we introduce a dual-branch network in which a BRDF-based reflective branch is complemented by a clay-guided branch that stabilizes geometry and refines surface normals. The two branches are trained jointly using the synthesized clay-like images, which provide a neutral, reflection-free supervision signal that complements the reflective views. Experiments on both synthetic and real datasets demonstrate substantial improvement in normal accuracy and mesh completeness over existing reflection-handling methods. Beyond technical gains, our framework reveals that seeing by unshining, translating radiance into neutrality, can serve as a powerful inductive bias for reflective object geometry learning.", "date": "2025-11-26", "categories": []}
{"id": "2511.21113", "title": "FaithFusion: Harmonizing Reconstruction and Generation via Pixel-wise Information Gain", "abstract": "In controllable driving-scene reconstruction and 3D scene generation, maintaining geometric fidelity while synthesizing visually plausible appearance under large viewpoint shifts is crucial. However, effective fusion of geometry-based 3DGS and appearance-driven diffusion models faces inherent challenges, as the absence of pixel-wise, 3D-consistent editing criteria often leads to over-restoration and geometric drift. To address these issues, we introduce \\textbf{FaithFusion}, a 3DGS-diffusion fusion framework driven by pixel-wise Expected Information Gain (EIG). EIG acts as a unified policy for coherent spatio-temporal synthesis: it guides diffusion as a spatial prior to refine high-uncertainty regions, while its pixel-level weighting distills the edits back into 3DGS. The resulting plug-and-play system is free from extra prior conditions and structural modifications.Extensive experiments on the Waymo dataset demonstrate that our approach attains SOTA performance across NTA-IoU, NTL-IoU, and FID, maintaining an FID of 107.47 even at 6 meters lane shift. Our code is available at https://github.com/wangyuanbiubiubiu/FaithFusion.", "date": "2025-11-26", "categories": []}
{"id": "2511.21317", "title": "HTTM: Head-wise Temporal Token Merging for Faster VGGT", "abstract": "The Visual Geometry Grounded Transformer (VGGT) marks a significant leap forward in 3D scene reconstruction, as it is the first model that directly infers all key 3D attributes (camera poses, depths, and dense geometry) jointly in one pass. However, this joint inference mechanism requires global attention layers that perform all-to-all attention computation on tokens from all views. For reconstruction of large scenes with long-sequence inputs, this causes a significant latency bottleneck. In this paper, we propose head-wise temporal merging (HTTM), a training-free 3D token merging method for accelerating VGGT. Existing merging techniques merge tokens uniformly across different attention heads, resulting in identical tokens in the layers' output, which hinders the model's representational ability. HTTM tackles this problem by merging tokens in multi-head granularity, which preserves the uniqueness of feature tokens after head concatenation. Additionally, this enables HTTM to leverage the spatial locality and temporal correspondence observed at the head level to achieve higher merging ratios with lower merging costs compared to existing methods. Thus, HTTM achieves up to 7x acceleration with negligible performance drops in a GPU-based inference.", "date": "2025-11-26", "categories": []}
{"id": "2511.21367", "title": "Endo-G$^{2}$T: Geometry-Guided & Temporally Aware Time-Embedded 4DGS For Endoscopic Scenes", "abstract": "Endoscopic (endo) video exhibits strong view-dependent effects such as specularities, wet reflections, and occlusions. Pure photometric supervision misaligns with geometry and triggers early geometric drift, where erroneous shapes are reinforced during densification and become hard to correct. We ask how to anchor geometry early for 4D Gaussian splatting (4DGS) while maintaining temporal consistency and efficiency in dynamic endoscopic scenes. Thus, we present Endo-G$^{2}$T, a geometry-guided and temporally aware training scheme for time-embedded 4DGS. First, geo-guided prior distillation converts confidence-gated monocular depth into supervision with scale-invariant depth and depth-gradient losses, using a warm-up-to-cap schedule to inject priors softly and avoid early overfitting. Second, a time-embedded Gaussian field represents dynamics in XYZT with a rotor-like rotation parameterization, yielding temporally coherent geometry with lightweight regularization that favors smooth motion and crisp opacity boundaries. Third, keyframe-constrained streaming improves efficiency and long-horizon stability through keyframe-focused optimization under a max-points budget, while non-keyframes advance with lightweight updates. Across EndoNeRF and StereoMIS-P1 datasets, Endo-G$^{2}$T achieves state-of-the-art results among monocular reconstruction baselines.", "date": "2025-11-26", "categories": []}
{"id": "2412.15447", "title": "LiHi-GS: LiDAR-Supervised Gaussian Splatting for Highway Driving Scene Reconstruction", "abstract": "Photorealistic 3D scene reconstruction plays an important role in autonomous driving, enabling the generation of novel data from existing datasets to simulate safety-critical scenarios and expand training data without additional acquisition costs. Gaussian Splatting (GS) facilitates real-time, photorealistic rendering with an explicit 3D Gaussian representation of the scene, providing faster processing and more intuitive scene editing than the implicit Neural Radiance Fields (NeRFs). While extensive GS research has yielded promising advancements in autonomous driving applications, they overlook two critical aspects: First, existing methods mainly focus on low-speed and feature-rich urban scenes and ignore the fact that highway scenarios play a significant role in autonomous driving. Second, while LiDARs are commonplace in autonomous driving platforms, existing methods learn primarily from images and use LiDAR only for initial estimates or without precise sensor modeling, thus missing out on leveraging the rich depth information LiDAR offers and limiting the ability to synthesize LiDAR data. In this paper, we propose a novel GS method for dynamic scene synthesis and editing with improved scene reconstruction through LiDAR supervision and support for LiDAR rendering. Unlike prior works that are tested mostly on urban datasets, to the best of our knowledge, we are the first to focus on the more challenging and highly relevant highway scenes for autonomous driving, with sparse sensor views and monotone backgrounds. Visit our project page at: https://umautobots.github.io/lihi_gs", "date": "2024-12-19", "categories": []}
{"id": "2508.19786", "title": "MAPo : Motion-Aware Partitioning of Deformable 3D Gaussian Splatting for High-Fidelity Dynamic Scene Reconstruction", "abstract": "3D Gaussian Splatting, known for enabling high-quality static scene reconstruction with fast rendering, is increasingly being applied to multi-view dynamic scene reconstruction. A common strategy involves learning a deformation field to model the temporal changes of a canonical set of 3D Gaussians. However, these deformation-based methods often produce blurred renderings and lose fine motion details in highly dynamic regions due to the inherent limitations of a single, unified model in representing diverse motion patterns. To address these challenges, we introduce Motion-Aware Partitioning of Deformable 3D Gaussian Splatting (MAPo), a novel framework for high-fidelity dynamic scene reconstruction. Its core is a dynamic score-based partitioning strategy that distinguishes between high- and low-dynamic 3D Gaussians. For high-dynamic 3D Gaussians, we recursively partition them temporally and duplicate their deformation networks for each new temporal segment, enabling specialized modeling to capture intricate motion details. Concurrently, low-dynamic 3DGs are treated as static to reduce computational costs. However, this temporal partitioning strategy for high-dynamic 3DGs can introduce visual discontinuities across frames at the partition boundaries. To address this, we introduce a cross-frame consistency loss, which not only ensures visual continuity but also further enhances rendering quality. Extensive experiments demonstrate that MAPo achieves superior rendering quality compared to baselines while maintaining comparable computational costs, particularly in regions with complex or rapid motions.", "date": "2025-08-27", "categories": []}
{"id": "2511.19971", "title": "VGGT4D: Mining Motion Cues in Visual Geometry Transformers for 4D Scene Reconstruction", "abstract": "Reconstructing dynamic 4D scenes is challenging, as it requires robust disentanglement of dynamic objects from the static background. While 3D foundation models like VGGT provide accurate 3D geometry, their performance drops markedly when moving objects dominate. Existing 4D approaches often rely on external priors, heavy post-optimization, or require fine-tuning on 4D datasets. In this paper, we propose VGGT4D, a training-free framework that extends the 3D foundation model VGGT for robust 4D scene reconstruction. Our approach is motivated by the key finding that VGGT's global attention layers already implicitly encode rich, layer-wise dynamic cues. To obtain masks that decouple static and dynamic elements, we mine and amplify global dynamic cues via gram similarity and aggregate them across a temporal window. To further sharpen mask boundaries, we introduce a refinement strategy driven by projection gradient. We then integrate these precise masks into VGGT's early-stage inference, effectively mitigating motion interference in both pose estimation and geometric reconstruction. Across six datasets, our method achieves superior performance in dynamic object segmentation, camera pose estimation, and dense reconstruction. It also supports single-pass inference on sequences longer than 500 frames.", "date": "2025-11-25", "categories": []}
{"id": "2511.20343", "title": "AMB3R: Accurate Feed-forward Metric-scale 3D Reconstruction with Backend", "abstract": "We present AMB3R, a multi-view feed-forward model for dense 3D reconstruction on a metric-scale that addresses diverse 3D vision tasks. The key idea is to leverage a sparse, yet compact, volumetric scene representation as our backend, enabling geometric reasoning with spatial compactness. Although trained solely for multi-view reconstruction, we demonstrate that AMB3R can be seamlessly extended to uncalibrated visual odometry (online) or large-scale structure from motion without the need for task-specific fine-tuning or test-time optimization. Compared to prior pointmap-based models, our approach achieves state-of-the-art performance in camera pose, depth, and metric-scale estimation, 3D reconstruction, and even surpasses optimization-based SLAM and SfM methods with dense reconstruction priors on common benchmarks.", "date": "2025-11-25", "categories": []}
{"id": "2412.07608", "title": "Faster and Better 3D Splatting via Group Training", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a powerful technique for novel view synthesis, demonstrating remarkable capability in high-fidelity scene reconstruction through its Gaussian primitive representations. However, the computational overhead induced by the massive number of primitives poses a significant bottleneck to training efficiency. To overcome this challenge, we propose Group Training, a simple yet effective strategy that organizes Gaussian primitives into manageable groups, optimizing training efficiency and improving rendering quality. This approach shows universal compatibility with existing 3DGS frameworks, including vanilla 3DGS and Mip-Splatting, consistently achieving accelerated training while maintaining superior synthesis quality. Extensive experiments reveal that our straightforward Group Training strategy achieves up to 30\\% faster convergence and improved rendering quality across diverse scenarios. Project Website: https://chengbo-wang.github.io/3DGS-with-Group-Training/", "date": "2024-12-10", "categories": []}
{"id": "2501.01119", "title": "Leverage Cross-Attention for End-to-End Open-Vocabulary Panoptic Reconstruction", "abstract": "Open-vocabulary panoptic reconstruction offers comprehensive scene understanding, enabling advances in embodied robotics and photorealistic simulation. In this paper, we propose PanopticRecon++, an end-to-end method that formulates panoptic reconstruction through a novel cross-attention perspective. This perspective models the relationship between 3D instances (as queries) and the scene's 3D embedding field (as keys) through their attention map. Unlike existing methods that separate the optimization of queries and keys or overlook spatial proximity, PanopticRecon++ introduces learnable 3D Gaussians as instance queries. This formulation injects 3D spatial priors to preserve proximity while maintaining end-to-end optimizability. Moreover, this query formulation facilitates the alignment of 2D open-vocabulary instance IDs across frames by leveraging optimal linear assignment with instance masks rendered from the queries. Additionally, we ensure semantic-instance segmentation consistency by fusing query-based instance segmentation probabilities with semantic probabilities in a novel panoptic head supervised by a panoptic loss. During training, the number of instance query tokens dynamically adapts to match the number of objects. PanopticRecon++ shows competitive performance in terms of 3D and 2D segmentation and reconstruction performance on both simulation and real-world datasets, and demonstrates a user case as a robot simulator. Our project website is at: https://yuxuan1206.github.io/panopticrecon_pp/", "date": "2025-01-02", "categories": []}
{"id": "2510.20776", "title": "CUPID: Generative 3D Reconstruction via Joint Object and Pose Modeling", "abstract": "We introduce Cupid, a generative 3D reconstruction framework that jointly models the full distribution over both canonical objects and camera poses. Our two-stage flow-based model first generates a coarse 3D structure and 2D-3D correspondences to estimate the camera pose robustly. Conditioned on this pose, a refinement stage injects pixel-aligned image features directly into the generative process, marrying the rich prior of a generative model with the geometric fidelity of reconstruction. This strategy achieves exceptional faithfulness, outperforming state-of-the-art reconstruction methods by over 3 dB PSNR and 10% in Chamfer Distance. As a unified generative model that decouples the object and camera pose, Cupid naturally extends to multi-view and scene-level reconstruction tasks without requiring post-hoc optimization or fine-tuning.", "date": "2025-10-23", "categories": []}
{"id": "2511.17824", "title": "QAL: A Loss for Recall Precision Balance in 3D Reconstruction", "abstract": "Volumetric learning underpins many 3D vision tasks such as completion, reconstruction, and mesh generation, yet training objectives still rely on Chamfer Distance (CD) or Earth Mover's Distance (EMD), which fail to balance recall and precision. We propose Quality-Aware Loss (QAL), a drop-in replacement for CD/EMD that combines a coverage-weighted nearest-neighbor term with an uncovered-ground-truth attraction term, explicitly decoupling recall and precision into tunable components.\n  Across diverse pipelines, QAL achieves consistent coverage gains, improving by an average of +4.3 pts over CD and +2.8 pts over the best alternatives. Though modest in percentage, these improvements reliably recover thin structures and under-represented regions that CD/EMD overlook. Extensive ablations confirm stable performance across hyperparameters and across output resolutions, while full retraining on PCN and ShapeNet demonstrates generalization across datasets and backbones. Moreover, QAL-trained completions yield higher grasp scores under GraspNet evaluation, showing that improved coverage translates directly into more reliable robotic manipulation.\n  QAL thus offers a principled, interpretable, and practical objective for robust 3D vision and safety-critical robotics pipelines", "date": "2025-11-21", "categories": []}
{"id": "2511.17904", "title": "CUS-GS: A Compact Unified Structured Gaussian Splatting Framework for Multimodal Scene Representation", "abstract": "Recent advances in Gaussian Splatting based 3D scene representation have shown two major trends: semantics-oriented approaches that focus on high-level understanding but lack explicit 3D geometry modeling, and structure-oriented approaches that capture spatial structures yet provide limited semantic abstraction. To bridge this gap, we present CUS-GS, a compact unified structured Gaussian Splatting representation, which connects multimodal semantic features with structured 3D geometry. Specifically, we design a voxelized anchor structure that constructs a spatial scaffold, while extracting multimodal semantic features from a set of foundation models (e.g., CLIP, DINOv2, SEEM). Moreover, we introduce a multimodal latent feature allocation mechanism to unify appearance, geometry, and semantics across heterogeneous feature spaces, ensuring a consistent representation across multiple foundation models. Finally, we propose a feature-aware significance evaluation strategy to dynamically guide anchor growing and pruning, effectively removing redundant or invalid anchors while maintaining semantic integrity. Extensive experiments show that CUS-GS achieves competitive performance compared to state-of-the-art methods using as few as 6M parameters - an order of magnitude smaller than the closest rival at 35M - highlighting the excellent trade off between performance and model efficiency of the proposed framework.", "date": "2025-11-22", "categories": []}
{"id": "2511.17918", "title": "Frequency-Adaptive Sharpness Regularization for Improving 3D Gaussian Splatting Generalization", "abstract": "Despite 3D Gaussian Splatting (3DGS) excelling in most configurations, it lacks generalization across novel viewpoints in a few-shot scenario because it overfits to the sparse observations. We revisit 3DGS optimization from a machine learning perspective, framing novel view synthesis as a generalization problem to unseen viewpoints-an underexplored direction. We propose Frequency-Adaptive Sharpness Regularization (FASR), which reformulates the 3DGS training objective, thereby guiding 3DGS to converge toward a better generalization solution. Although Sharpness-Aware Minimization (SAM) similarly reduces the sharpness of the loss landscape to improve generalization of classification models, directly employing it to 3DGS is suboptimal due to the discrepancy between the tasks. Specifically, it hinders reconstructing high-frequency details due to excessive regularization, while reducing its strength leads to under-penalizing sharpness. To address this, we reflect the local frequency of images to set the regularization weight and the neighborhood radius when estimating the local sharpness. It prevents floater artifacts in novel viewpoints and reconstructs fine details that SAM tends to oversmooth. Across datasets with various configurations, our method consistently improves a wide range of baselines. Code will be available at https://bbangsik13.github.io/FASR.", "date": "2025-11-22", "categories": []}
{"id": "2511.18290", "title": "SwiftVGGT: A Scalable Visual Geometry Grounded Transformer for Large-Scale Scenes", "abstract": "3D reconstruction in large-scale scenes is a fundamental task in 3D perception, but the inherent trade-off between accuracy and computational efficiency remains a significant challenge. Existing methods either prioritize speed and produce low-quality results, or achieve high-quality reconstruction at the cost of slow inference times. In this paper, we propose SwiftVGGT, a training-free method that significantly reduce inference time while preserving high-quality dense 3D reconstruction. To maintain global consistency in large-scale scenes, SwiftVGGT performs loop closure without relying on the external Visual Place Recognition (VPR) model. This removes redundant computation and enables accurate reconstruction over kilometer-scale environments. Furthermore, we propose a simple yet effective point sampling method to align neighboring chunks using a single Sim(3)-based Singular Value Decomposition (SVD) step. This eliminates the need for the Iteratively Reweighted Least Squares (IRLS) optimization commonly used in prior work, leading to substantial speed-ups. We evaluate SwiftVGGT on multiple datasets and show that it achieves state-of-the-art reconstruction quality while requiring only 33% of the inference time of recent VGGT-based large-scale reconstruction approaches.", "date": "2025-11-23", "categories": []}
{"id": "2511.18386", "title": "SegSplat: Feed-forward Gaussian Splatting and Open-Set Semantic Segmentation", "abstract": "We have introduced SegSplat, a novel framework designed to bridge the gap between rapid, feed-forward 3D reconstruction and rich, open-vocabulary semantic understanding. By constructing a compact semantic memory bank from multi-view 2D foundation model features and predicting discrete semantic indices alongside geometric and appearance attributes for each 3D Gaussian in a single pass, SegSplat efficiently imbues scenes with queryable semantics. Our experiments demonstrate that SegSplat achieves geometric fidelity comparable to state-of-the-art feed-forward 3D Gaussian Splatting methods while simultaneously enabling robust open-set semantic segmentation, crucially \\textit{without} requiring any per-scene optimization for semantic feature integration. This work represents a significant step towards practical, on-the-fly generation of semantically aware 3D environments, vital for advancing robotic interaction, augmented reality, and other intelligent systems.", "date": "2025-11-23", "categories": []}
{"id": "2511.18416", "title": "4D-VGGT: A General Foundation Model with SpatioTemporal Awareness for Dynamic Scene Geometry Estimation", "abstract": "We investigate a challenging task of dynamic scene geometry estimation, which requires representing both spatial and temporal features. Typically, existing methods align the two features into a unified latent space to model scene geometry. However, this unified paradigm suffers from potential mismatched representation due to the heterogeneous nature between spatial and temporal features. In this work, we propose 4D-VGGT, a general foundation model with divide-and-conquer spatiotemporal representation for dynamic scene geometry. Our model is divided into three aspects: 1) Multi-setting input. We design an adaptive visual grid that supports input sequences with arbitrary numbers of views and time steps. 2) Multi-level representation. We propose a cross-view global fusion for spatial representation and a cross-time local fusion for temporal representation. 3) Multi-task prediction. We append multiple task-specific heads to spatiotemporal representations, enabling a comprehensive visual geometry estimation for dynamic scenes. Under this unified framework, these components enhance the feature discriminability and application universality of our model for dynamic scenes. In addition, we integrate multiple geometry datasets to train our model and conduct extensive experiments to verify the effectiveness of our method across various tasks on multiple dynamic scene geometry benchmarks.", "date": "2025-11-23", "categories": []}
{"id": "2511.18873", "title": "Neural Texture Splatting: Expressive 3D Gaussian Splatting for View Synthesis, Geometry, and Dynamic Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a leading approach for high-quality novel view synthesis, with numerous variants extending its applicability to a broad spectrum of 3D and 4D scene reconstruction tasks. Despite its success, the representational capacity of 3DGS remains limited by the use of 3D Gaussian kernels to model local variations. Recent works have proposed to augment 3DGS with additional per-primitive capacity, such as per-splat textures, to enhance its expressiveness. However, these per-splat texture approaches primarily target dense novel view synthesis with a reduced number of Gaussian primitives, and their effectiveness tends to diminish when applied to more general reconstruction scenarios. In this paper, we aim to achieve concrete performance improvement over state-of-the-art 3DGS variants across a wide range of reconstruction tasks, including novel view synthesis, geometry and dynamic reconstruction, under both sparse and dense input settings. To this end, we introduce Neural Texture Splatting (NTS). At the core of our approach is a global neural field (represented as a hybrid of a tri-plane and a neural decoder) that predicts local appearance and geometric fields for each primitive. By leveraging this shared global representation that models local texture fields across primitives, we significantly reduce model size and facilitate efficient global information exchange, demonstrating strong generalization across tasks. Furthermore, our neural modeling of local texture fields introduces expressive view- and time-dependent effects, a critical aspect that existing methods fail to account for. Extensive experiments show that Neural Texture Splatting consistently improves models and achieves state-of-the-art results across multiple benchmarks.", "date": "2025-11-24", "categories": []}
{"id": "2411.16898", "title": "MonoGSDF: Exploring Monocular Geometric Cues for Gaussian Splatting-Guided Implicit Surface Reconstruction", "abstract": "Accurate meshing from monocular images remains a key challenge in 3D vision. While state-of-the-art 3D Gaussian Splatting (3DGS) methods excel at synthesizing photorealistic novel views through rasterization-based rendering, their reliance on sparse, explicit primitives severely limits their ability to recover watertight and topologically consistent 3D surfaces.We introduce MonoGSDF, a novel method that couples Gaussian-based primitives with a neural Signed Distance Field (SDF) for high-quality reconstruction. During training, the SDF guides Gaussians' spatial distribution, while at inference, Gaussians serve as priors to reconstruct surfaces, eliminating the need for memory-intensive Marching Cubes. To handle arbitrary-scale scenes, we propose a scaling strategy for robust generalization. A multi-resolution training scheme further refines details and monocular geometric cues from off-the-shelf estimators enhance reconstruction quality. Experiments on real-world datasets show MonoGSDF outperforms prior methods while maintaining efficiency.", "date": "2024-11-25", "categories": []}
{"id": "2511.13278", "title": "SF-Recon: Simplification-Free Lightweight Building Reconstruction via 3D Gaussian Splatting", "abstract": "Lightweight building surface models are crucial for digital city, navigation, and fast geospatial analytics, yet conventional multi-view geometry pipelines remain cumbersome and quality-sensitive due to their reliance on dense reconstruction, meshing, and subsequent simplification. This work presents SF-Recon, a method that directly reconstructs lightweight building surfaces from multi-view images without post-hoc mesh simplification. We first train an initial 3D Gaussian Splatting (3DGS) field to obtain a view-consistent representation. Building structure is then distilled by a normal-gradient-guided Gaussian optimization that selects primitives aligned with roof and wall boundaries, followed by multi-view edge-consistency pruning to enhance structural sharpness and suppress non-structural artifacts without external supervision. Finally, a multi-view depth-constrained Delaunay triangulation converts the structured Gaussian field into a lightweight, structurally faithful building mesh. Based on a proposed SF dataset, the experimental results demonstrate that our SF-Recon can directly reconstruct lightweight building models from multi-view imagery, achieving substantially fewer faces and vertices while maintaining computational efficiency. Website:https://lzh282140127-cell.github.io/SF-Recon-project/", "date": "2025-11-17", "categories": []}
{"id": "2511.16980", "title": "Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting", "abstract": "3DGS employs a large number of Gaussian primitives to fit scenes, resulting in substantial storage and computational overhead. Existing pruning methods rely on manually designed criteria or introduce additional learnable parameters, yielding suboptimal results. To address this, we propose an natural selection inspired pruning framework that models survival pressure as a regularization gradient field applied to opacity, allowing the optimization gradients--driven by the goal of maximizing rendering quality--to autonomously determine which Gaussians to retain or prune. This process is fully learnable and requires no human intervention. We further introduce an opacity decay technique with a finite opacity prior, which accelerates the selection process without compromising pruning effectiveness. Compared to 3DGS, our method achieves over 0.6 dB PSNR gain under 15\\% budgets, establishing state-of-the-art performance for compact 3DGS. Project page https://xiaobin2001.github.io/GNS-web.", "date": "2025-11-21", "categories": []}
{"id": "2511.17364", "title": "SVRecon: Sparse Voxel Rasterization for Surface Reconstruction", "abstract": "We extend the recently proposed sparse voxel rasterization paradigm to the task of high-fidelity surface reconstruction by integrating Signed Distance Function (SDF), named SVRecon. Unlike 3D Gaussians, sparse voxels are spatially disentangled from their neighbors and have sharp boundaries, which makes them prone to local minima during optimization. Although SDF values provide a naturally smooth and continuous geometric field, preserving this smoothness across independently parameterized sparse voxels is nontrivial. To address this challenge, we promote coherent and smooth voxel-wise structure through (1) robust geometric initialization using a visual geometry model and (2) a spatial smoothness loss that enforces coherent relationships across parent-child and sibling voxel groups. Extensive experiments across various benchmarks show that our method achieves strong reconstruction accuracy while having consistently speedy convergence. The code will be made public.", "date": "2025-11-21", "categories": []}
{"id": "2511.16112", "title": "Clustered Error Correction with Grouped 4D Gaussian Splatting", "abstract": "Existing 4D Gaussian Splatting (4DGS) methods struggle to accurately reconstruct dynamic scenes, often failing to resolve ambiguous pixel correspondences and inadequate densification in dynamic regions. We address these issues by introducing a novel method composed of two key components: (1) Elliptical Error Clustering and Error Correcting Splat Addition that pinpoints dynamic areas to improve and initialize fitting splats, and (2) Grouped 4D Gaussian Splatting that improves consistency of mapping between splats and represented dynamic objects. Specifically, we classify rendering errors into missing-color and occlusion types, then apply targeted corrections via backprojection or foreground splitting guided by cross-view color consistency. Evaluations on Neural 3D Video and Technicolor datasets demonstrate that our approach significantly improves temporal consistency and achieves state-of-the-art perceptual rendering quality, improving 0.39dB of PSNR on the Technicolor Light Field dataset. Our visualization shows improved alignment between splats and dynamic objects, and the error correction method's capability to identify errors and properly initialize new splats. Our implementation details and source code are available at https://github.com/tho-kn/cem-4dgs.", "date": "2025-11-20", "categories": []}
{"id": "2511.16298", "title": "Optimizing 3D Gaussian Splattering for Mobile GPUs", "abstract": "Image-based 3D scene reconstruction, which transforms multi-view images into a structured 3D representation of the surrounding environment, is a common task across many modern applications. 3D Gaussian Splatting (3DGS) is a new paradigm to address this problem and offers considerable efficiency as compared to the previous methods. Motivated by this, and considering various benefits of mobile device deployment (data privacy, operating without internet connectivity, and potentially faster responses), this paper develops Texture3dgs, an optimized mapping of 3DGS for a mobile GPU. A critical challenge in this area turns out to be optimizing for the two-dimensional (2D) texture cache, which needs to be exploited for faster executions on mobile GPUs. As a sorting method dominates the computations in 3DGS on mobile platforms, the core of Texture3dgs is a novel sorting algorithm where the processing, data movement, and placement are highly optimized for 2D memory. The properties of this algorithm are analyzed in view of a cost model for the texture cache. In addition, we accelerate other steps of the 3DGS algorithm through improved variable layout design and other optimizations. End-to-end evaluation shows that Texture3dgs delivers up to 4.1$\\times$ and 1.7$\\times$ speedup for the sorting and overall 3D scene reconstruction, respectively -- while also reducing memory usage by up to 1.6$\\times$ -- demonstrating the effectiveness of our design for efficient mobile 3D scene reconstruction.", "date": "2025-11-20", "categories": []}
{"id": "2509.11853", "title": "Segmentation-Driven Initialization for Sparse-view 3D Gaussian Splatting", "abstract": "Sparse-view synthesis remains a challenging problem due to the difficulty of recovering accurate geometry and appearance from limited observations. While recent advances in 3D Gaussian Splatting (3DGS) have enabled real-time rendering with competitive quality, existing pipelines often rely on Structure-from-Motion (SfM) for camera pose estimation, an approach that struggles in genuinely sparse-view settings. Moreover, several SfM-free methods replace SfM with multi-view stereo (MVS) models, but generate massive numbers of 3D Gaussians by back-projecting every pixel into 3D space, leading to high memory costs. We propose Segmentation-Driven Initialization for Gaussian Splatting (SDI-GS), a method that mitigates inefficiency by leveraging region-based segmentation to identify and retain only structurally significant regions. This enables selective downsampling of the dense point cloud, preserving scene fidelity while substantially reducing Gaussian count. Experiments across diverse benchmarks show that SDI-GS reduces Gaussian count by up to 50% and achieves comparable or superior rendering quality in PSNR and SSIM, with only marginal degradation in LPIPS. It further enables faster training and lower memory footprint, advancing the practicality of 3DGS for constrained-view scenarios.", "date": "2025-09-15", "categories": []}
{"id": "2511.14283", "title": "NeuralSSD: A Neural Solver for Signed Distance Surface Reconstruction", "abstract": "We proposed a generalized method, NeuralSSD, for reconstructing a 3D implicit surface from the widely-available point cloud data. NeuralSSD is a solver-based on the neural Galerkin method, aimed at reconstructing higher-quality and accurate surfaces from input point clouds. Implicit method is preferred due to its ability to accurately represent shapes and its robustness in handling topological changes. However, existing parameterizations of implicit fields lack explicit mechanisms to ensure a tight fit between the surface and input data. To address this, we propose a novel energy equation that balances the reliability of point cloud information. Additionally, we introduce a new convolutional network that learns three-dimensional information to achieve superior optimization results. This approach ensures that the reconstructed surface closely adheres to the raw input points and infers valuable inductive biases from point clouds, resulting in a highly accurate and stable surface reconstruction. NeuralSSD is evaluated on a variety of challenging datasets, including the ShapeNet and Matterport datasets, and achieves state-of-the-art results in terms of both surface reconstruction accuracy and generalizability.", "date": "2025-11-18", "categories": []}
{"id": "2511.14291", "title": "GEN3D: Generating Domain-Free 3D Scenes from a Single Image", "abstract": "Despite recent advancements in neural 3D reconstruction, the dependence on dense multi-view captures restricts their broader applicability. Additionally, 3D scene generation is vital for advancing embodied AI and world models, which depend on diverse, high-quality scenes for learning and evaluation. In this work, we propose Gen3d, a novel method for generation of high-quality, wide-scope, and generic 3D scenes from a single image. After the initial point cloud is created by lifting the RGBD image, Gen3d maintains and expands its world model. The 3D scene is finalized through optimizing a Gaussian splatting representation. Extensive experiments on diverse datasets demonstrate the strong generalization capability and superior performance of our method in generating a world model and Synthesizing high-fidelity and consistent novel views.", "date": "2025-11-18", "categories": []}
{"id": "2511.14315", "title": "Dental3R: Geometry-Aware Pairing for Intraoral 3D Reconstruction from Sparse-View Photographs", "abstract": "Intraoral 3D reconstruction is fundamental to digital orthodontics, yet conventional methods like intraoral scanning are inaccessible for remote tele-orthodontics, which typically relies on sparse smartphone imagery. While 3D Gaussian Splatting (3DGS) shows promise for novel view synthesis, its application to the standard clinical triad of unposed anterior and bilateral buccal photographs is challenging. The large view baselines, inconsistent illumination, and specular surfaces common in intraoral settings can destabilize simultaneous pose and geometry estimation. Furthermore, sparse-view photometric supervision often induces a frequency bias, leading to over-smoothed reconstructions that lose critical diagnostic details. To address these limitations, we propose \\textbf{Dental3R}, a pose-free, graph-guided pipeline for robust, high-fidelity reconstruction from sparse intraoral photographs. Our method first constructs a Geometry-Aware Pairing Strategy (GAPS) to intelligently select a compact subgraph of high-value image pairs. The GAPS focuses on correspondence matching, thereby improving the stability of the geometry initialization and reducing memory usage. Building on the recovered poses and point cloud, we train the 3DGS model with a wavelet-regularized objective. By enforcing band-limited fidelity using a discrete wavelet transform, our approach preserves fine enamel boundaries and interproximal edges while suppressing high-frequency artifacts. We validate our approach on a large-scale dataset of 950 clinical cases and an additional video-based test set of 195 cases. Experimental results demonstrate that Dental3R effectively handles sparse, unposed inputs and achieves superior novel view synthesis quality for dental occlusion visualization, outperforming state-of-the-art methods.", "date": "2025-11-18", "categories": []}
{"id": "2511.14539", "title": "Learning Compact Latent Space for Representing Neural Signed Distance Functions with High-fidelity Geometry Details", "abstract": "Neural signed distance functions (SDFs) have been a vital representation to represent 3D shapes or scenes with neural networks. An SDF is an implicit function that can query signed distances at specific coordinates for recovering a 3D surface. Although implicit functions work well on a single shape or scene, they pose obstacles when analyzing multiple SDFs with high-fidelity geometry details, due to the limited information encoded in the latent space for SDFs and the loss of geometry details. To overcome these obstacles, we introduce a method to represent multiple SDFs in a common space, aiming to recover more high-fidelity geometry details with more compact latent representations. Our key idea is to take full advantage of the benefits of generalization-based and overfitting-based learning strategies, which manage to preserve high-fidelity geometry details with compact latent codes. Based on this framework, we also introduce a novel sampling strategy to sample training queries. The sampling can improve the training efficiency and eliminate artifacts caused by the influence of other SDFs. We report numerical and visual evaluations on widely used benchmarks to validate our designs and show advantages over the latest methods in terms of the representative ability and compactness.", "date": "2025-11-18", "categories": []}
{"id": "2511.14540", "title": "Interaction-Aware 4D Gaussian Splatting for Dynamic Hand-Object Interaction Reconstruction", "abstract": "This paper focuses on a challenging setting of simultaneously modeling geometry and appearance of hand-object interaction scenes without any object priors. We follow the trend of dynamic 3D Gaussian Splatting based methods, and address several significant challenges. To model complex hand-object interaction with mutual occlusion and edge blur, we present interaction-aware hand-object Gaussians with newly introduced optimizable parameters aiming to adopt piecewise linear hypothesis for clearer structural representation. Moreover, considering the complementarity and tightness of hand shape and object shape during interaction dynamics, we incorporate hand information into object deformation field, constructing interaction-aware dynamic fields to model flexible motions. To further address difficulties in the optimization process, we propose a progressive strategy that handles dynamic regions and static background step by step. Correspondingly, explicit regularizations are designed to stabilize the hand-object representations for smooth motion transition, physical interaction reality, and coherent lighting. Experiments show that our approach surpasses existing dynamic 3D-GS-based methods and achieves state-of-the-art performance in reconstructing dynamic hand-object interaction.", "date": "2025-11-18", "categories": []}
{"id": "2511.14633", "title": "SparseSurf: Sparse-View 3D Gaussian Splatting for Surface Reconstruction", "abstract": "Recent advances in optimizing Gaussian Splatting for scene geometry have enabled efficient reconstruction of detailed surfaces from images. However, when input views are sparse, such optimization is prone to overfitting, leading to suboptimal reconstruction quality. Existing approaches address this challenge by employing flattened Gaussian primitives to better fit surface geometry, combined with depth regularization to alleviate geometric ambiguities under limited viewpoints. Nevertheless, the increased anisotropy inherent in flattened Gaussians exacerbates overfitting in sparse-view scenarios, hindering accurate surface fitting and degrading novel view synthesis performance. In this paper, we propose \\net{}, a method that reconstructs more accurate and detailed surfaces while preserving high-quality novel view rendering. Our key insight is to introduce Stereo Geometry-Texture Alignment, which bridges rendering quality and geometry estimation, thereby jointly enhancing both surface reconstruction and view synthesis. In addition, we present a Pseudo-Feature Enhanced Geometry Consistency that enforces multi-view geometric consistency by incorporating both training and unseen views, effectively mitigating overfitting caused by sparse supervision. Extensive experiments on the DTU, BlendedMVS, and Mip-NeRF360 datasets demonstrate that our method achieves the state-of-the-art performance.", "date": "2025-11-18", "categories": []}
{"id": "2501.14659", "title": "GUSLO: General and Unified Structured Light Optimization", "abstract": "Structured light (SL) 3D reconstruction captures the precise surface shape of objects, providing high-accuracy 3D data essential for industrial inspection and cultural heritage digitization. However, existing methods suffer from two key limitations: reliance on scene-specific calibration with manual parameter tuning, and optimization frameworks tailored to specific SL patterns, limiting their generalizability across varied scenarios. We propose General and Unified Structured Light Optimization (GUSLO), a novel framework addressing these issues through two coordinated innovations: (1) single-shot calibration via 2D triangulation-based interpolation that converts sparse matches into dense correspondence fields, and (2) artifact-aware photometric adaptation via explicit transfer functions, balancing generalization and color fidelity. We conduct diverse experiments covering binary, speckle, and color-coded settings. Results show that GUSLO consistently improves accuracy and cross-encoding robustness over conventional methods in challenging industrial and cultural scenarios.", "date": "2025-01-24", "categories": []}
{"id": "2511.12040", "title": "SRSplat: Feed-Forward Super-Resolution Gaussian Splatting from Sparse Multi-View Images", "abstract": "Feed-forward 3D reconstruction from sparse, low-resolution (LR) images is a crucial capability for real-world applications, such as autonomous driving and embodied AI. However, existing methods often fail to recover fine texture details. This limitation stems from the inherent lack of high-frequency information in LR inputs. To address this, we propose \\textbf{SRSplat}, a feed-forward framework that reconstructs high-resolution 3D scenes from only a few LR views. Our main insight is to compensate for the deficiency of texture information by jointly leveraging external high-quality reference images and internal texture cues. We first construct a scene-specific reference gallery, generated for each scene using Multimodal Large Language Models (MLLMs) and diffusion models. To integrate this external information, we introduce the \\textit{Reference-Guided Feature Enhancement (RGFE)} module, which aligns and fuses features from the LR input images and their reference twin image. Subsequently, we train a decoder to predict the Gaussian primitives using the multi-view fused feature obtained from \\textit{RGFE}. To further refine predicted Gaussian primitives, we introduce \\textit{Texture-Aware Density Control (TADC)}, which adaptively adjusts Gaussian density based on the internal texture richness of the LR inputs. Extensive experiments demonstrate that our SRSplat outperforms existing methods on various datasets, including RealEstate10K, ACID, and DTU, and exhibits strong cross-dataset and cross-resolution generalization capabilities.", "date": "2025-11-15", "categories": []}
{"id": "2511.12304", "title": "LiDAR-GS++:Improving LiDAR Gaussian Reconstruction via Diffusion Priors", "abstract": "Recent GS-based rendering has made significant progress for LiDAR, surpassing Neural Radiance Fields (NeRF) in both quality and speed. However, these methods exhibit artifacts in extrapolated novel view synthesis due to the incomplete reconstruction from single traversal scans. To address this limitation, we present LiDAR-GS++, a LiDAR Gaussian Splatting reconstruction method enhanced by diffusion priors for real-time and high-fidelity re-simulation on public urban roads. Specifically, we introduce a controllable LiDAR generation model conditioned on coarsely extrapolated rendering to produce extra geometry-consistent scans and employ an effective distillation mechanism for expansive reconstruction. By extending reconstruction to under-fitted regions, our approach ensures global geometric consistency for extrapolative novel views while preserving detailed scene surfaces captured by sensors. Experiments on multiple public datasets demonstrate that LiDAR-GS++ achieves state-of-the-art performance for both interpolated and extrapolated viewpoints, surpassing existing GS and NeRF-based methods.", "date": "2025-11-15", "categories": []}
{"id": "2511.13011", "title": "Beyond Darkness: Thermal-Supervised 3D Gaussian Splatting for Low-Light Novel View Synthesis", "abstract": "Under extremely low-light conditions, novel view synthesis (NVS) faces severe degradation in terms of geometry, color consistency, and radiometric stability. Standard 3D Gaussian Splatting (3DGS) pipelines fail when applied directly to underexposed inputs, as independent enhancement across views causes illumination inconsistencies and geometric distortion. To address this, we present DTGS, a unified framework that tightly couples Retinex-inspired illumination decomposition with thermal-guided 3D Gaussian Splatting for illumination-invariant reconstruction. Unlike prior approaches that treat enhancement as a pre-processing step, DTGS performs joint optimization across enhancement, geometry, and thermal supervision through a cyclic enhancement-reconstruction mechanism. A thermal supervisory branch stabilizes both color restoration and geometry learning by dynamically balancing enhancement, structural, and thermal losses. Moreover, a Retinex-based decomposition module embedded within the 3DGS loop provides physically interpretable reflectance-illumination separation, ensuring consistent color and texture across viewpoints. To evaluate our method, we construct RGBT-LOW, a new multi-view low-light thermal dataset capturing severe illumination degradation. Extensive experiments show that DTGS significantly outperforms existing low-light enhancement and 3D reconstruction baselines, achieving superior radiometric consistency, geometric fidelity, and color stability under extreme illumination.", "date": "2025-11-17", "categories": []}
{"id": "2505.16533", "title": "Motion Matters: Compact Gaussian Streaming for Free-Viewpoint Video Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a high-fidelity and efficient paradigm for online free-viewpoint video (FVV) reconstruction, offering viewers rapid responsiveness and immersive experiences. However, existing online methods face challenge in prohibitive storage requirements primarily due to point-wise modeling that fails to exploit the motion properties. To address this limitation, we propose a novel Compact Gaussian Streaming (ComGS) framework, leveraging the locality and consistency of motion in dynamic scene, that models object-consistent Gaussian point motion through keypoint-driven motion representation. By transmitting only the keypoint attributes, this framework provides a more storage-efficient solution. Specifically, we first identify a sparse set of motion-sensitive keypoints localized within motion regions using a viewspace gradient difference strategy. Equipped with these keypoints, we propose an adaptive motion-driven mechanism that predicts a spatial influence field for propagating keypoint motion to neighboring Gaussian points with similar motion. Moreover, ComGS adopts an error-aware correction strategy for key frame reconstruction that selectively refines erroneous regions and mitigates error accumulation without unnecessary overhead. Overall, ComGS achieves a remarkable storage reduction of over 159 X compared to 3DGStream and 14 X compared to the SOTA method QUEEN, while maintaining competitive visual fidelity and rendering speed.", "date": "2025-05-22", "categories": []}
{"id": "2511.11175", "title": "Dynamic Gaussian Scene Reconstruction from Unsynchronized Videos", "abstract": "Multi-view video reconstruction plays a vital role in computer vision, enabling applications in film production, virtual reality, and motion analysis. While recent advances such as 4D Gaussian Splatting (4DGS) have demonstrated impressive capabilities in dynamic scene reconstruction, they typically rely on the assumption that input video streams are temporally synchronized. However, in real-world scenarios, this assumption often fails due to factors like camera trigger delays or independent recording setups, leading to temporal misalignment across views and reduced reconstruction quality. To address this challenge, a novel temporal alignment strategy is proposed for high-quality 4DGS reconstruction from unsynchronized multi-view videos. Our method features a coarse-to-fine alignment module that estimates and compensates for each camera's time shift. The method first determines a coarse, frame-level offset and then refines it to achieve sub-frame accuracy. This strategy can be integrated as a readily integrable module into existing 4DGS frameworks, enhancing their robustness when handling asynchronous data. Experiments show that our approach effectively processes temporally misaligned videos and significantly enhances baseline methods.", "date": "2025-11-14", "categories": []}
{"id": "2511.11213", "title": "RealisticDreamer: Guidance Score Distillation for Few-shot Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has recently gained great attention in the 3D scene representation for its high-quality real-time rendering capabilities. However, when the input comprises sparse training views, 3DGS is prone to overfitting, primarily due to the lack of intermediate-view supervision. Inspired by the recent success of Video Diffusion Models (VDM), we propose a framework called Guidance Score Distillation (GSD) to extract the rich multi-view consistency priors from pretrained VDMs. Building on the insights from Score Distillation Sampling (SDS), GSD supervises rendered images from multiple neighboring views, guiding the Gaussian splatting representation towards the generative direction of VDM. However, the generative direction often involves object motion and random camera trajectories, making it challenging for direct supervision in the optimization process. To address this problem, we introduce an unified guidance form to correct the noise prediction result of VDM. Specifically, we incorporate both a depth warp guidance based on real depth maps and a guidance based on semantic image features, ensuring that the score update direction from VDM aligns with the correct camera pose and accurate geometry. Experimental results show that our method outperforms existing approaches across multiple datasets.", "date": "2025-11-14", "categories": []}
{"id": "2503.06161", "title": "Feature-EndoGaussian: Feature Distilled Gaussian Splatting in Surgical Deformable Scene Reconstruction", "abstract": "Minimally invasive surgery (MIS) requires high-fidelity, real-time visual feedback of dynamic and low-texture surgical scenes. To address these requirements, we introduce FeatureEndo-4DGS (FE-4DGS), the first real time pipeline leveraging feature-distilled 4D Gaussian Splatting for simultaneous reconstruction and semantic segmentation of deformable surgical environments. Unlike prior feature-distilled methods restricted to static scenes, and existing 4D approaches that lack semantic integration, FE-4DGS seamlessly leverages pre-trained 2D semantic embeddings to produce a unified 4D representation-where semantics also deform with tissue motion. This unified approach enables the generation of real-time RGB and semantic outputs through a single, parallelized rasterization process. Despite the additional complexity from feature distillation, FE-4DGS sustains real-time rendering (61 FPS) with a compact footprint, achieves state-of-the-art rendering fidelity on EndoNeRF (39.1 PSNR) and SCARED (27.3 PSNR), and delivers competitive EndoVis18 segmentation, matching or exceeding strong 2D baselines for binary segmentation tasks (0.93 DSC) and remaining competitive for multi-label segmentation (0.77 DSC).", "date": "2025-03-08", "categories": []}
{"id": "2511.09944", "title": "TSPE-GS: Probabilistic Depth Extraction for Semi-Transparent Surface Reconstruction via 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting offers a strong speed-quality trade-off but struggles to reconstruct semi-transparent surfaces because most methods assume a single depth per pixel, which fails when multiple surfaces are visible. We propose TSPE-GS (Transparent Surface Probabilistic Extraction for Gaussian Splatting), which uniformly samples transmittance to model a pixel-wise multi-modal distribution of opacity and depth, replacing the prior single-peak assumption and resolving cross-surface depth ambiguity. By progressively fusing truncated signed distance functions, TSPE-GS reconstructs external and internal surfaces separately within a unified framework. The method generalizes to other Gaussian-based reconstruction pipelines without extra training overhead. Extensive experiments on public and self-collected semi-transparent and opaque datasets show TSPE-GS significantly improves semi-transparent geometry reconstruction while maintaining performance on opaque scenes.", "date": "2025-11-13", "categories": []}
{"id": "2411.19461", "title": "Robust Bayesian Scene Reconstruction with Retrieval-Augmented Priors for Precise Grasping and Planning", "abstract": "Constructing 3D representations of object geometry is critical for many robotics tasks, particularly manipulation problems. These representations must be built from potentially noisy partial observations. In this work, we focus on the problem of reconstructing a multi-object scene from a single RGBD image using a fixed camera. Traditional scene representation methods generally cannot infer the geometry of unobserved regions of the objects in the image. Attempts have been made to leverage deep learning to train on a dataset of known objects and representations, and then generalize to new observations. However, this can be brittle to noisy real-world observations and objects not contained in the dataset, and do not provide well-calibrated reconstruction confidences. We propose BRRP, a reconstruction method that leverages preexisting mesh datasets to build an informative prior during robust probabilistic reconstruction. We introduce the concept of a retrieval-augmented prior, where we retrieve relevant components of our prior distribution from a database of objects during inference. The resulting prior enables estimation of the geometry of occluded portions of the in-scene objects. Our method produces a distribution over object shape that can be used for reconstruction and measuring uncertainty. We evaluate our method in both simulated scenes and in the real world. We demonstrate the robustness of our method against deep learning-only approaches while being more accurate than a method without an informative prior. Through real-world experiments, we particularly highlight the capability of BRRP to enable successful dexterous manipulation in clutter.", "date": "2024-11-29", "categories": []}
{"id": "2507.12114", "title": "LidarPainter: One-Step Away From Any Lidar View To Novel Guidance", "abstract": "Dynamic driving scene reconstruction is of great importance in fields like digital twin system and autonomous driving simulation. However, unacceptable degradation occurs when the view deviates from the input trajectory, leading to corrupted background and vehicle models. To improve reconstruction quality on novel trajectory, existing methods are subject to various limitations including inconsistency, deformation, and time consumption. This paper proposes LidarPainter, a one-step diffusion model that recovers consistent driving views from sparse LiDAR condition and artifact-corrupted renderings in real-time, enabling high-fidelity lane shifts in driving scene reconstruction. Extensive experiments show that LidarPainter outperforms state-of-the-art methods in speed, quality and resource efficiency, specifically 7 x faster than StreetCrafter with only one fifth of GPU memory required. LidarPainter also supports stylized generation using text prompts such as \"foggy\" and \"night\", allowing for a diverse expansion of the existing asset library.", "date": "2025-07-16", "categories": []}
{"id": "2511.08545", "title": "RePose-NeRF: Robust Radiance Fields for Mesh Reconstruction under Noisy Camera Poses", "abstract": "Accurate 3D reconstruction from multi-view images is essential for downstream robotic tasks such as navigation, manipulation, and environment understanding. However, obtaining precise camera poses in real-world settings remains challenging, even when calibration parameters are known. This limits the practicality of existing NeRF-based methods that rely heavily on accurate extrinsic estimates. Furthermore, their implicit volumetric representations differ significantly from the widely adopted polygonal meshes, making rendering and manipulation inefficient in standard 3D software. In this work, we propose a robust framework that reconstructs high-quality, editable 3D meshes directly from multi-view images with noisy extrinsic parameters. Our approach jointly refines camera poses while learning an implicit scene representation that captures fine geometric detail and photorealistic appearance. The resulting meshes are compatible with common 3D graphics and robotics tools, enabling efficient downstream use. Experiments on standard benchmarks demonstrate that our method achieves accurate and robust 3D reconstruction under pose uncertainty, bridging the gap between neural implicit representations and practical robotic applications.", "date": "2025-11-11", "categories": []}
{"id": "2502.14520", "title": "Learning Temporal 3D Semantic Scene Completion via Optical Flow Guidance", "abstract": "3D Semantic Scene Completion (SSC) provides comprehensive scene geometry and semantics for autonomous driving perception, which is crucial for enabling accurate and reliable decision-making. However, existing SSC methods are limited to capturing sparse information from the current frame or naively stacking multi-frame temporal features, thereby failing to acquire effective scene context. These approaches ignore critical motion dynamics and struggle to achieve temporal consistency. To address the above challenges, we propose a novel temporal SSC method FlowScene: Learning Temporal 3D Semantic Scene Completion via Optical Flow Guidance. By leveraging optical flow, FlowScene can integrate motion, different viewpoints, occlusions, and other contextual cues, thereby significantly improving the accuracy of 3D scene completion. Specifically, our framework introduces two key components: (1) a Flow-Guided Temporal Aggregation module that aligns and aggregates temporal features using optical flow, capturing motion-aware context and deformable structures; and (2) an Occlusion-Guided Voxel Refinement module that injects occlusion masks and temporally aggregated features into 3D voxel space, adaptively refining voxel representations for explicit geometric modeling. Experimental results demonstrate that FlowScene achieves state-of-the-art performance on the SemanticKITTI and SSCBench-KITTI-360 benchmarks.", "date": "2025-02-20", "categories": []}
{"id": "2505.08239", "title": "ACT-R: Adaptive Camera Trajectories for Single View 3D Reconstruction", "abstract": "We introduce the simple idea of adaptive view planning to multi-view synthesis, aiming to improve both occlusion revelation and 3D consistency for single-view 3D reconstruction. Instead of producing an unordered set of views independently or simultaneously, we generate a sequence of views, leveraging temporal consistency to enhance 3D coherence. More importantly, our view sequence is not determined by a pre-determined and fixed camera setup. Instead, we compute an adaptive camera trajectory (ACT), forming an orbit, which seeks to maximize the visibility of occluded regions of the 3D object to be reconstructed. Once the best orbit is found, we feed it to a video diffusion model to generate novel views around the orbit, which can then be passed to any multi-view 3D reconstruction model to obtain the final result. Our multi-view synthesis pipeline is quite efficient since it involves no run-time training/optimization, only forward inferences by applying pre-trained models for occlusion analysis and multi-view synthesis. Our method predicts camera trajectories that reveal occlusions effectively and produce consistent novel views, significantly improving 3D reconstruction over SOTA alternatives on the unseen GSO dataset. Project Page: https://mingrui-zhao.github.io/ACT-R/", "date": "2025-05-13", "categories": []}
{"id": "2509.11169", "title": "Multispectral-NeRF:a multispectral modeling approach based on neural radiance fields", "abstract": "3D reconstruction technology generates three-dimensional representations of real-world objects, scenes, or environments using sensor data such as 2D images, with extensive applications in robotics, autonomous vehicles, and virtual reality systems. Traditional 3D reconstruction techniques based on 2D images typically relies on RGB spectral information. With advances in sensor technology, additional spectral bands beyond RGB have been increasingly incorporated into 3D reconstruction workflows. Existing methods that integrate these expanded spectral data often suffer from expensive scheme prices, low accuracy and poor geometric features. Three - dimensional reconstruction based on NeRF can effectively address the various issues in current multispectral 3D reconstruction methods, producing high - precision and high - quality reconstruction results. However, currently, NeRF and some improved models such as NeRFacto are trained on three - band data and cannot take into account the multi - band information. To address this problem, we propose Multispectral-NeRF, an enhanced neural architecture derived from NeRF that can effectively integrates multispectral information. Our technical contributions comprise threefold modifications: Expanding hidden layer dimensionality to accommodate 6-band spectral inputs; Redesigning residual functions to optimize spectral discrepancy calculations between reconstructed and reference images; Adapting data compression modules to address the increased bit-depth requirements of multispectral imagery. Experimental results confirm that Multispectral-NeRF successfully processes multi-band spectral features while accurately preserving the original scenes' spectral characteristics.", "date": "2025-09-14", "categories": []}
{"id": "2510.14270", "title": "GauSSmart: Enhanced 3D Reconstruction through 2D Foundation Models and Geometric Filtering", "abstract": "Scene reconstruction has emerged as a central challenge in computer vision, with approaches such as Neural Radiance Fields (NeRF) and Gaussian Splatting achieving remarkable progress. While Gaussian Splatting demonstrates strong performance on large-scale datasets, it often struggles to capture fine details or maintain realism in regions with sparse coverage, largely due to the inherent limitations of sparse 3D training data.\n  In this work, we propose GauSSmart, a hybrid method that effectively bridges 2D foundational models and 3D Gaussian Splatting reconstruction. Our approach integrates established 2D computer vision techniques, including convex filtering and semantic feature supervision from foundational models such as DINO, to enhance Gaussian-based scene reconstruction. By leveraging 2D segmentation priors and high-dimensional feature embeddings, our method guides the densification and refinement of Gaussian splats, improving coverage in underrepresented areas and preserving intricate structural details.\n  We validate our approach across three datasets, where GauSSmart consistently outperforms existing Gaussian Splatting in the majority of evaluated scenes. Our results demonstrate the significant potential of hybrid 2D-3D approaches, highlighting how the thoughtful combination of 2D foundational models with 3D reconstruction pipelines can overcome the limitations inherent in either approach alone.", "date": "2025-10-16", "categories": []}
{"id": "2511.06310", "title": "Adaptive 3D Reconstruction via Diffusion Priors and Forward Curvature-Matching Likelihood Updates", "abstract": "Reconstructing high-quality point clouds from images remains challenging in computer vision. Existing generative-model-based approaches, particularly diffusion-model approaches that directly learn the posterior, may suffer from inflexibility -- they require conditioning signals during training, support only a fixed number of input views, and need complete retraining for different measurements. Recent diffusion-based methods have attempted to address this by combining prior models with likelihood updates, but they rely on heuristic fixed step sizes for the likelihood update that lead to slow convergence and suboptimal reconstruction quality. We advance this line of approach by integrating our novel Forward Curvature-Matching (FCM) update method with diffusion sampling. Our method dynamically determines optimal step sizes using only forward automatic differentiation and finite-difference curvature estimates, enabling precise optimization of the likelihood update. This formulation enables high-fidelity reconstruction from both single-view and multi-view inputs, and supports various input modalities through simple operator substitution -- all without retraining. Experiments on ShapeNet and CO3D datasets demonstrate that our method achieves superior reconstruction quality at matched or lower NFEs, yielding higher F-score and lower CD and EMD, validating its efficiency and adaptability for practical applications. Code is available at https://github.com/Seunghyeok0715/FCM", "date": "2025-11-09", "categories": []}
{"id": "2511.06632", "title": "DIAL-GS: Dynamic Instance Aware Reconstruction for Label-free Street Scenes with 4D Gaussian Splatting", "abstract": "Urban scene reconstruction is critical for autonomous driving, enabling structured 3D representations for data synthesis and closed-loop testing. Supervised approaches rely on costly human annotations and lack scalability, while current self-supervised methods often confuse static and dynamic elements and fail to distinguish individual dynamic objects, limiting fine-grained editing. We propose DIAL-GS, a novel dynamic instance-aware reconstruction method for label-free street scenes with 4D Gaussian Splatting. We first accurately identify dynamic instances by exploiting appearance-position inconsistency between warped rendering and actual observation. Guided by instance-level dynamic perception, we employ instance-aware 4D Gaussians as the unified volumetric representation, realizing dynamic-adaptive and instance-aware reconstruction. Furthermore, we introduce a reciprocal mechanism through which identity and dynamics reinforce each other, enhancing both integrity and consistency. Experiments on urban driving scenarios show that DIAL-GS surpasses existing self-supervised baselines in reconstruction quality and instance-level editing, offering a concise yet powerful solution for urban scene modeling.", "date": "2025-11-10", "categories": []}
{"id": "2511.06734", "title": "Rethinking Rainy 3D Scene Reconstruction via Perspective Transforming and Brightness Tuning", "abstract": "Rain degrades the visual quality of multi-view images, which are essential for 3D scene reconstruction, resulting in inaccurate and incomplete reconstruction results. Existing datasets often overlook two critical characteristics of real rainy 3D scenes: the viewpoint-dependent variation in the appearance of rain streaks caused by their projection onto 2D images, and the reduction in ambient brightness resulting from cloud coverage during rainfall. To improve data realism, we construct a new dataset named OmniRain3D that incorporates perspective heterogeneity and brightness dynamicity, enabling more faithful simulation of rain degradation in 3D scenes. Based on this dataset, we propose an end-to-end reconstruction framework named REVR-GSNet (Rain Elimination and Visibility Recovery for 3D Gaussian Splatting). Specifically, REVR-GSNet integrates recursive brightness enhancement, Gaussian primitive optimization, and GS-guided rain elimination into a unified architecture through joint alternating optimization, achieving high-fidelity reconstruction of clean 3D scenes from rain-degraded inputs. Extensive experiments show the effectiveness of our dataset and method. Our dataset and method provide a foundation for future research on multi-view image deraining and rainy 3D scene reconstruction.", "date": "2025-11-10", "categories": []}
{"id": "2511.07122", "title": "Sparse4DGS: 4D Gaussian Splatting for Sparse-Frame Dynamic Scene Reconstruction", "abstract": "Dynamic Gaussian Splatting approaches have achieved remarkable performance for 4D scene reconstruction. However, these approaches rely on dense-frame video sequences for photorealistic reconstruction. In real-world scenarios, due to equipment constraints, sometimes only sparse frames are accessible. In this paper, we propose Sparse4DGS, the first method for sparse-frame dynamic scene reconstruction. We observe that dynamic reconstruction methods fail in both canonical and deformed spaces under sparse-frame settings, especially in areas with high texture richness. Sparse4DGS tackles this challenge by focusing on texture-rich areas. For the deformation network, we propose Texture-Aware Deformation Regularization, which introduces a texture-based depth alignment loss to regulate Gaussian deformation. For the canonical Gaussian field, we introduce Texture-Aware Canonical Optimization, which incorporates texture-based noise into the gradient descent process of canonical Gaussians. Extensive experiments show that when taking sparse frames as inputs, our method outperforms existing dynamic or few-shot techniques on NeRF-Synthetic, HyperNeRF, NeRF-DS, and our iPhone-4D datasets.", "date": "2025-11-10", "categories": []}
{"id": "2511.07206", "title": "Geometric implicit neural representations for signed distance functions", "abstract": "\\textit{Implicit neural representations} (INRs) have emerged as a promising framework for representing signals in low-dimensional spaces. This survey reviews the existing literature on the specialized INR problem of approximating \\textit{signed distance functions} (SDFs) for surface scenes, using either oriented point clouds or a set of posed images. We refer to neural SDFs that incorporate differential geometry tools, such as normals and curvatures, in their loss functions as \\textit{geometric} INRs. The key idea behind this 3D reconstruction approach is to include additional \\textit{regularization} terms in the loss function, ensuring that the INR satisfies certain global properties that the function should hold -- such as having unit gradient in the case of SDFs. We explore key methodological components, including the definition of INR, the construction of geometric loss functions, and sampling schemes from a differential geometry perspective. Our review highlights the significant advancements enabled by geometric INRs in surface reconstruction from oriented point clouds and posed images.", "date": "2025-11-10", "categories": []}
{"id": "2509.18090", "title": "GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction", "abstract": "Reconstructing accurate surfaces with radiance fields has achieved remarkable progress in recent years. However, prevailing approaches, primarily based on Gaussian Splatting, are increasingly constrained by representational bottlenecks. In this paper, we introduce GeoSVR, an explicit voxel-based framework that explores and extends the under-investigated potential of sparse voxels for achieving accurate, detailed, and complete surface reconstruction. As strengths, sparse voxels support preserving the coverage completeness and geometric clarity, while corresponding challenges also arise from absent scene constraints and locality in surface refinement. To ensure correct scene convergence, we first propose a Voxel-Uncertainty Depth Constraint that maximizes the effect of monocular depth cues while presenting a voxel-oriented uncertainty to avoid quality degradation, enabling effective and robust scene constraints yet preserving highly accurate geometries. Subsequently, Sparse Voxel Surface Regularization is designed to enhance geometric consistency for tiny voxels and facilitate the voxel-based formation of sharp and accurate surfaces. Extensive experiments demonstrate our superior performance compared to existing methods across diverse challenging scenarios, excelling in geometric accuracy, detail preservation, and reconstruction completeness while maintaining high efficiency. Code is available at https://github.com/Fictionarry/GeoSVR.", "date": "2025-09-22", "categories": []}
{"id": "2511.05229", "title": "4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos", "abstract": "Novel view synthesis from monocular videos of dynamic scenes with unknown camera poses remains a fundamental challenge in computer vision and graphics. While recent advances in 3D representations such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have shown promising results for static scenes, they struggle with dynamic content and typically rely on pre-computed camera poses. We present 4D3R, a pose-free dynamic neural rendering framework that decouples static and dynamic components through a two-stage approach. Our method first leverages 3D foundational models for initial pose and geometry estimation, followed by motion-aware refinement. 4D3R introduces two key technical innovations: (1) a motion-aware bundle adjustment (MA-BA) module that combines transformer-based learned priors with SAM2 for robust dynamic object segmentation, enabling more accurate camera pose refinement; and (2) an efficient Motion-Aware Gaussian Splatting (MA-GS) representation that uses control points with a deformation field MLP and linear blend skinning to model dynamic motion, significantly reducing computational cost while maintaining high-quality reconstruction. Extensive experiments on real-world dynamic datasets demonstrate that our approach achieves up to 1.8dB PSNR improvement over state-of-the-art methods, particularly in challenging scenarios with large dynamic objects, while reducing computational requirements by 5x compared to previous dynamic scene representations.", "date": "2025-11-07", "categories": []}
{"id": "2504.14516", "title": "Back on Track: Bundle Adjustment for Dynamic Scene Reconstruction", "abstract": "Traditional SLAM systems, which rely on bundle adjustment, struggle with highly dynamic scenes commonly found in casual videos. Such videos entangle the motion of dynamic elements, undermining the assumption of static environments required by traditional systems. Existing techniques either filter out dynamic elements or model their motion independently. However, the former often results in incomplete reconstructions, whereas the latter can lead to inconsistent motion estimates. Taking a novel approach, this work leverages a 3D point tracker to separate the camera-induced motion from the observed motion of dynamic objects. By considering only the camera-induced component, bundle adjustment can operate reliably on all scene elements as a result. We further ensure depth consistency across video frames with lightweight post-processing based on scale maps. Our framework combines the core of traditional SLAM -- bundle adjustment -- with a robust learning-based 3D tracker front-end. Integrating motion decomposition, bundle adjustment and depth refinement, our unified framework, BA-Track, accurately tracks the camera motion and produces temporally coherent and scale-consistent dense reconstructions, accommodating both static and dynamic elements. Our experiments on challenging datasets reveal significant improvements in camera pose estimation and 3D reconstruction accuracy.", "date": "2025-04-20", "categories": []}
{"id": "2511.03970", "title": "Room Envelopes: A Synthetic Dataset for Indoor Layout Reconstruction from Images", "abstract": "Modern scene reconstruction methods are able to accurately recover 3D surfaces that are visible in one or more images. However, this leads to incomplete reconstructions, missing all occluded surfaces. While much progress has been made on reconstructing entire objects given partial observations using generative models, the structural elements of a scene, like the walls, floors and ceilings, have received less attention. We argue that these scene elements should be relatively easy to predict, since they are typically planar, repetitive and simple, and so less costly approaches may be suitable. In this work, we present a synthetic dataset -- Room Envelopes -- that facilitates progress on this task by providing a set of RGB images and two associated pointmaps for each image: one capturing the visible surface and one capturing the first surface once fittings and fixtures are removed, that is, the structural layout. As we show, this enables direct supervision for feed-forward monocular geometry estimators that predict both the first visible surface and the first layout surface. This confers an understanding of the scene's extent, as well as the shape and location of its objects.", "date": "2025-11-06", "categories": []}
{"id": "2511.04595", "title": "UniSplat: Unified Spatio-Temporal Fusion via 3D Latent Scaffolds for Dynamic Driving Scene Reconstruction", "abstract": "Feed-forward 3D reconstruction for autonomous driving has advanced rapidly, yet existing methods struggle with the joint challenges of sparse, non-overlapping camera views and complex scene dynamics. We present UniSplat, a general feed-forward framework that learns robust dynamic scene reconstruction through unified latent spatio-temporal fusion. UniSplat constructs a 3D latent scaffold, a structured representation that captures geometric and semantic scene context by leveraging pretrained foundation models. To effectively integrate information across spatial views and temporal frames, we introduce an efficient fusion mechanism that operates directly within the 3D scaffold, enabling consistent spatio-temporal alignment. To ensure complete and detailed reconstructions, we design a dual-branch decoder that generates dynamic-aware Gaussians from the fused scaffold by combining point-anchored refinement with voxel-based generation, and maintain a persistent memory of static Gaussians to enable streaming scene completion beyond current camera coverage. Extensive experiments on real-world datasets demonstrate that UniSplat achieves state-of-the-art performance in novel view synthesis, while providing robust and high-quality renderings even for viewpoints outside the original camera coverage.", "date": "2025-11-06", "categories": []}
{"id": "2411.19588", "title": "Gaussian Splashing: Direct Volumetric Rendering Underwater", "abstract": "In underwater images, most useful features are occluded by water. The extent of the occlusion depends on imaging geometry and can vary even across a sequence of burst images. As a result, 3D reconstruction methods robust on in-air scenes, like Neural Radiance Field methods (NeRFs) or 3D Gaussian Splatting (3DGS), fail on underwater scenes. While a recent underwater adaptation of NeRFs achieved state-of-the-art results, it is impractically slow: reconstruction takes hours and its rendering rate, in frames per second (FPS), is less than 1. Here, we present a new method that takes only a few minutes for reconstruction and renders novel underwater scenes at 140 FPS. Named Gaussian Splashing, our method unifies the strengths and speed of 3DGS with an image formation model for capturing scattering, introducing innovations in the rendering and depth estimation procedures and in the 3DGS loss function. Despite the complexities of underwater adaptation, our method produces images at unparalleled speeds with superior details. Moreover, it reveals distant scene details with far greater clarity than other methods, dramatically improving reconstructed and rendered images. We demonstrate results on existing datasets and a new dataset we have collected.\n  Additional visual results are available at: https://bgu-cs-vil.github.io/gaussiansplashingUW.github.io/ .", "date": "2024-11-29", "categories": []}
{"id": "2510.25173", "title": "D$^2$GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction", "abstract": "Recently, Gaussian Splatting (GS) has shown great potential for urban scene reconstruction in the field of autonomous driving. However, current urban scene reconstruction methods often depend on multimodal sensors as inputs, \\textit{i.e.} LiDAR and images. Though the geometry prior provided by LiDAR point clouds can largely mitigate ill-posedness in reconstruction, acquiring such accurate LiDAR data is still challenging in practice: i) precise spatiotemporal calibration between LiDAR and other sensors is required, as they may not capture data simultaneously; ii) reprojection errors arise from spatial misalignment when LiDAR and cameras are mounted at different locations. To avoid the difficulty of acquiring accurate LiDAR depth, we propose D$^2$GS, a LiDAR-free urban scene reconstruction framework. In this work, we obtain geometry priors that are as effective as LiDAR while being denser and more accurate. $\\textbf{First}$, we initialize a dense point cloud by back-projecting multi-view metric depth predictions. This point cloud is then optimized by a Progressive Pruning strategy to improve the global consistency. $\\textbf{Second}$, we jointly refine Gaussian geometry and predicted dense metric depth via a Depth Enhancer. Specifically, we leverage diffusion priors from a depth foundation model to enhance the depth maps rendered by Gaussians. In turn, the enhanced depths provide stronger geometric constraints during Gaussian training. $\\textbf{Finally}$, we improve the accuracy of ground geometry by constraining the shape and normal attributes of Gaussians within road regions. Extensive experiments on the Waymo dataset demonstrate that our method consistently outperforms state-of-the-art methods, producing more accurate geometry even when compared with those using ground-truth LiDAR data.", "date": "2025-10-29", "categories": []}
{"id": "2511.00362", "title": "Oitijjo-3D: Generative AI Framework for Rapid 3D Heritage Reconstruction from Street View Imagery", "abstract": "Cultural heritage restoration in Bangladesh faces a dual challenge of limited resources and scarce technical expertise. Traditional 3D digitization methods, such as photogrammetry or LiDAR scanning, require expensive hardware, expert operators, and extensive on-site access, which are often infeasible in developing contexts. As a result, many of Bangladesh's architectural treasures, from the Paharpur Buddhist Monastery to Ahsan Manzil, remain vulnerable to decay and inaccessible in digital form. This paper introduces Oitijjo-3D, a cost-free generative AI framework that democratizes 3D cultural preservation. By using publicly available Google Street View imagery, Oitijjo-3D reconstructs faithful 3D models of heritage structures through a two-stage pipeline - multimodal visual reasoning with Gemini 2.5 Flash Image for structure-texture synthesis, and neural image-to-3D generation through Hexagen for geometry recovery. The system produces photorealistic, metrically coherent reconstructions in seconds, achieving significant speedups compared to conventional Structure-from-Motion pipelines, without requiring any specialized hardware or expert supervision. Experiments on landmarks such as Ahsan Manzil, Choto Sona Mosque, and Paharpur demonstrate that Oitijjo-3D preserves both visual and structural fidelity while drastically lowering economic and technical barriers. By turning open imagery into digital heritage, this work reframes preservation as a community-driven, AI-assisted act of cultural continuity for resource-limited nations.", "date": "2025-11-01", "categories": []}
{"id": "2511.01186", "title": "LiDAR-VGGT: Cross-Modal Coarse-to-Fine Fusion for Globally Consistent and Metric-Scale Dense Mapping", "abstract": "Reconstructing large-scale colored point clouds is an important task in robotics, supporting perception, navigation, and scene understanding. Despite advances in LiDAR inertial visual odometry (LIVO), its performance remains highly sensitive to extrinsic calibration. Meanwhile, 3D vision foundation models, such as VGGT, suffer from limited scalability in large environments and inherently lack metric scale. To overcome these limitations, we propose LiDAR-VGGT, a novel framework that tightly couples LiDAR inertial odometry with the state-of-the-art VGGT model through a two-stage coarse- to-fine fusion pipeline: First, a pre-fusion module with robust initialization refinement efficiently estimates VGGT poses and point clouds with coarse metric scale within each session. Then, a post-fusion module enhances cross-modal 3D similarity transformation, using bounding-box-based regularization to reduce scale distortions caused by inconsistent FOVs between LiDAR and camera sensors. Extensive experiments across multiple datasets demonstrate that LiDAR-VGGT achieves dense, globally consistent colored point clouds and outperforms both VGGT-based methods and LIVO baselines. The implementation of our proposed novel color point cloud evaluation toolkit will be released as open source.", "date": "2025-11-03", "categories": []}
{"id": "2410.15392", "title": "EF-3DGS: Event-Aided Free-Trajectory 3D Gaussian Splatting", "abstract": "Scene reconstruction from casually captured videos has wide applications in real-world scenarios. With recent advancements in differentiable rendering techniques, several methods have attempted to simultaneously optimize scene representations (NeRF or 3DGS) and camera poses. Despite recent progress, existing methods relying on traditional camera input tend to fail in high-speed (or equivalently low-frame-rate) scenarios. Event cameras, inspired by biological vision, record pixel-wise intensity changes asynchronously with high temporal resolution, providing valuable scene and motion information in blind inter-frame intervals. In this paper, we introduce the event camera to aid scene construction from a casually captured video for the first time, and propose Event-Aided Free-Trajectory 3DGS, called EF-3DGS, which seamlessly integrates the advantages of event cameras into 3DGS through three key components. First, we leverage the Event Generation Model (EGM) to fuse events and frames, supervising the rendered views observed by the event stream. Second, we adopt the Contrast Maximization (CMax) framework in a piece-wise manner to extract motion information by maximizing the contrast of the Image of Warped Events (IWE), thereby calibrating the estimated poses. Besides, based on the Linear Event Generation Model (LEGM), the brightness information encoded in the IWE is also utilized to constrain the 3DGS in the gradient domain. Third, to mitigate the absence of color information of events, we introduce photometric bundle adjustment (PBA) to ensure view consistency across events and frames. We evaluate our method on the public Tanks and Temples benchmark and a newly collected real-world dataset, RealEv-DAVIS. Our project page is https://lbh666.github.io/ef-3dgs/.", "date": "2024-10-20", "categories": []}
{"id": "2504.06264", "title": "D$^2$USt3R: Enhancing 3D Reconstruction for Dynamic Scenes", "abstract": "In this work, we address the task of 3D reconstruction in dynamic scenes, where object motions frequently degrade the quality of previous 3D pointmap regression methods, such as DUSt3R, that are originally designed for static 3D scene reconstruction. Although these methods provide an elegant and powerful solution in static settings, they struggle in the presence of dynamic motions that disrupt alignment based solely on camera poses. To overcome this, we propose $D^2USt3R$ that directly regresses Static-Dynamic Aligned Pointmaps (SDAP) that simultaneiously capture both static and dynamic 3D scene geometry. By explicitly incorporating both spatial and temporal aspects, our approach successfully encapsulates 3D dense correspondence to the proposed pointmaps, enhancing downstream tasks. Extensive experimental evaluations demonstrate that our proposed approach consistently achieves superior 3D reconstruction performance across various datasets featuring complex motions.", "date": "2025-04-08", "categories": []}
{"id": "2510.22706", "title": "IGGT: Instance-Grounded Geometry Transformer for Semantic 3D Reconstruction", "abstract": "Humans naturally perceive the geometric structure and semantic content of a 3D world as intertwined dimensions, enabling coherent and accurate understanding of complex scenes. However, most prior approaches prioritize training large geometry models for low-level 3D reconstruction and treat high-level spatial understanding in isolation, overlooking the crucial interplay between these two fundamental aspects of 3D-scene analysis, thereby limiting generalization and leading to poor performance in downstream 3D understanding tasks. Recent attempts have mitigated this issue by simply aligning 3D models with specific language models, thus restricting perception to the aligned model's capacity and limiting adaptability to downstream tasks. In this paper, we propose InstanceGrounded Geometry Transformer (IGGT), an end-to-end large unified transformer to unify the knowledge for both spatial reconstruction and instance-level contextual understanding. Specifically, we design a 3D-Consistent Contrastive Learning strategy that guides IGGT to encode a unified representation with geometric structures and instance-grounded clustering through only 2D visual inputs. This representation supports consistent lifting of 2D visual inputs into a coherent 3D scene with explicitly distinct object instances. To facilitate this task, we further construct InsScene-15K, a large-scale dataset with high-quality RGB images, poses, depth maps, and 3D-consistent instance-level mask annotations with a novel data curation pipeline.", "date": "2025-10-26", "categories": []}
{"id": "2510.26921", "title": "DC4GS: Directional Consistency-Driven Adaptive Density Control for 3D Gaussian Splatting", "abstract": "We present a Directional Consistency (DC)-driven Adaptive Density Control (ADC) for 3D Gaussian Splatting (DC4GS). Whereas the conventional ADC bases its primitive splitting on the magnitudes of positional gradients, we further incorporate the DC of the gradients into ADC, and realize it through the angular coherence of the gradients. Our DC better captures local structural complexities in ADC, avoiding redundant splitting. When splitting is required, we again utilize the DC to define optimal split positions so that sub-primitives best align with the local structures than the conventional random placement. As a consequence, our DC4GS greatly reduces the number of primitives (up to 30% in our experiments) than the existing ADC, and also enhances reconstruction fidelity greatly.", "date": "2025-10-30", "categories": []}
{"id": "2510.27234", "title": "MoRE: 3D Visual Geometry Reconstruction Meets Mixture-of-Experts", "abstract": "Recent advances in language and vision have demonstrated that scaling up model capacity consistently improves performance across diverse tasks. In 3D visual geometry reconstruction, large-scale training has likewise proven effective for learning versatile representations. However, further scaling of 3D models is challenging due to the complexity of geometric supervision and the diversity of 3D data. To overcome these limitations, we propose MoRE, a dense 3D visual foundation model based on a Mixture-of-Experts (MoE) architecture that dynamically routes features to task-specific experts, allowing them to specialize in complementary data aspects and enhance both scalability and adaptability. Aiming to improve robustness under real-world conditions, MoRE incorporates a confidence-based depth refinement module that stabilizes and refines geometric estimation. In addition, it integrates dense semantic features with globally aligned 3D backbone representations for high-fidelity surface normal prediction. MoRE is further optimized with tailored loss functions to ensure robust learning across diverse inputs and multiple geometric tasks. Extensive experiments demonstrate that MoRE achieves state-of-the-art performance across multiple benchmarks and supports effective downstream applications without extra computation.", "date": "2025-10-31", "categories": []}
{"id": "2509.24325", "title": "ReCon-GS: Continuum-Preserved Gaussian Streaming for Fast and Compact Reconstruction of Dynamic Scenes", "abstract": "Online free-viewpoint video (FVV) reconstruction is challenged by slow per-frame optimization, inconsistent motion estimation, and unsustainable storage demands. To address these challenges, we propose the Reconfigurable Continuum Gaussian Stream, dubbed ReCon-GS, a novel storage-aware framework that enables high fidelity online dynamic scene reconstruction and real-time rendering. Specifically, we dynamically allocate multi-level Anchor Gaussians in a density-adaptive fashion to capture inter-frame geometric deformations, thereby decomposing scene motion into compact coarse-to-fine representations. Then, we design a dynamic hierarchy reconfiguration strategy that preserves localized motion expressiveness through on-demand anchor re-hierarchization, while ensuring temporal consistency through intra-hierarchical deformation inheritance that confines transformation priors to their respective hierarchy levels. Furthermore, we introduce a storage-aware optimization mechanism that flexibly adjusts the density of Anchor Gaussians at different hierarchy levels, enabling a controllable trade-off between reconstruction fidelity and memory usage. Extensive experiments on three widely used datasets demonstrate that, compared to state-of-the-art methods, ReCon-GS improves training efficiency by approximately 15% and achieves superior FVV synthesis quality with enhanced robustness and stability. Moreover, at equivalent rendering quality, ReCon-GS slashes memory requirements by over 50% compared to leading state-of-the-art methods.", "date": "2025-09-29", "categories": []}
{"id": "2510.26358", "title": "AgriGS-SLAM: Orchard Mapping Across Seasons via Multi-View Gaussian Splatting SLAM", "abstract": "Autonomous robots in orchards require real-time 3D scene understanding despite repetitive row geometry, seasonal appearance changes, and wind-driven foliage motion. We present AgriGS-SLAM, a Visual--LiDAR SLAM framework that couples direct LiDAR odometry and loop closures with multi-camera 3D Gaussian Splatting (3DGS) rendering. Batch rasterization across complementary viewpoints recovers orchard structure under occlusions, while a unified gradient-driven map lifecycle executed between keyframes preserves fine details and bounds memory. Pose refinement is guided by a probabilistic LiDAR-based depth consistency term, back-propagated through the camera projection to tighten geometry-appearance coupling. We deploy the system on a field platform in apple and pear orchards across dormancy, flowering, and harvesting, using a standardized trajectory protocol that evaluates both training-view and novel-view synthesis to reduce 3DGS overfitting in evaluation. Across seasons and sites, AgriGS-SLAM delivers sharper, more stable reconstructions and steadier trajectories than recent state-of-the-art 3DGS-SLAM baselines while maintaining real-time performance on-tractor. While demonstrated in orchard monitoring, the approach can be applied to other outdoor domains requiring robust multimodal perception.", "date": "2025-10-30", "categories": []}
{"id": "2510.26694", "title": "The Impact and Outlook of 3D Gaussian Splatting", "abstract": "Since its introduction, 3D Gaussian Splatting (3DGS) has rapidly transformed the landscape of 3D scene representations, inspiring an extensive body of associated research. Follow-up work includes analyses and contributions that enhance the efficiency, scalability, and real-world applicability of 3DGS. In this summary, we present an overview of several key directions that have emerged in the wake of 3DGS. We highlight advances enabling resource-efficient training and rendering, the evolution toward dynamic (or four-dimensional, 4DGS) representations, and deeper exploration of the mathematical foundations underlying its appearance modeling and rendering process. Furthermore, we examine efforts to bring 3DGS to mobile and virtual reality platforms, its extension to massive-scale environments, and recent progress toward near-instant radiance field reconstruction via feed-forward or distributed computation. Collectively, these developments illustrate how 3DGS has evolved from a breakthrough representation into a versatile and foundational tool for 3D vision and graphics.", "date": "2025-10-30", "categories": []}
{"id": "2506.09518", "title": "HAIF-GS: Hierarchical and Induced Flow-Guided Gaussian Splatting for Dynamic Scene", "abstract": "Reconstructing dynamic 3D scenes from monocular videos remains a fundamental challenge in 3D vision. While 3D Gaussian Splatting (3DGS) achieves real-time rendering in static settings, extending it to dynamic scenes is challenging due to the difficulty of learning structured and temporally consistent motion representations. This challenge often manifests as three limitations in existing methods: redundant Gaussian updates, insufficient motion supervision, and weak modeling of complex non-rigid deformations. These issues collectively hinder coherent and efficient dynamic reconstruction. To address these limitations, we propose HAIF-GS, a unified framework that enables structured and consistent dynamic modeling through sparse anchor-driven deformation. It first identifies motion-relevant regions via an Anchor Filter to suppress redundant updates in static areas. A self-supervised Induced Flow-Guided Deformation module induces anchor motion using multi-frame feature aggregation, eliminating the need for explicit flow labels. To further handle fine-grained deformations, a Hierarchical Anchor Propagation mechanism increases anchor resolution based on motion complexity and propagates multi-level transformations. Extensive experiments on synthetic and real-world benchmarks validate that HAIF-GS significantly outperforms prior dynamic 3DGS methods in rendering quality, temporal coherence, and reconstruction efficiency.", "date": "2025-06-11", "categories": []}
{"id": "2506.24096", "title": "MILo: Mesh-In-the-Loop Gaussian Splatting for Detailed and Efficient Surface Reconstruction", "abstract": "While recent advances in Gaussian Splatting have enabled fast reconstruction of high-quality 3D scenes from images, extracting accurate surface meshes remains a challenge. Current approaches extract the surface through costly post-processing steps, resulting in the loss of fine geometric details or requiring significant time and leading to very dense meshes with millions of vertices. More fundamentally, the a posteriori conversion from a volumetric to a surface representation limits the ability of the final mesh to preserve all geometric structures captured during training. We present MILo, a novel Gaussian Splatting framework that bridges the gap between volumetric and surface representations by differentiably extracting a mesh from the 3D Gaussians. We design a fully differentiable procedure that constructs the mesh-including both vertex locations and connectivity-at every iteration directly from the parameters of the Gaussians, which are the only quantities optimized during training. Our method introduces three key technical contributions: a bidirectional consistency framework ensuring both representations-Gaussians and the extracted mesh-capture the same underlying geometry during training; an adaptive mesh extraction process performed at each training iteration, which uses Gaussians as differentiable pivots for Delaunay triangulation; a novel method for computing signed distance values from the 3D Gaussians that enables precise surface extraction while avoiding geometric erosion. Our approach can reconstruct complete scenes, including backgrounds, with state-of-the-art quality while requiring an order of magnitude fewer mesh vertices than previous methods. Due to their light weight and empty interior, our meshes are well suited for downstream applications such as physics simulations or animation.", "date": "2025-06-30", "categories": []}
{"id": "2508.12015", "title": "InstDrive: Instance-Aware 3D Gaussian Splatting for Driving Scenes", "abstract": "Reconstructing dynamic driving scenes from dashcam videos has attracted increasing attention due to its significance in autonomous driving and scene understanding. While recent advances have made impressive progress, most methods still unify all background elements into a single representation, hindering both instance-level understanding and flexible scene editing. Some approaches attempt to lift 2D segmentation into 3D space, but often rely on pre-processed instance IDs or complex pipelines to map continuous features to discrete identities. Moreover, these methods are typically designed for indoor scenes with rich viewpoints, making them less applicable to outdoor driving scenarios. In this paper, we present InstDrive, an instance-aware 3D Gaussian Splatting framework tailored for the interactive reconstruction of dynamic driving scene. We use masks generated by SAM as pseudo ground-truth to guide 2D feature learning via contrastive loss and pseudo-supervised objectives. At the 3D level, we introduce regularization to implicitly encode instance identities and enforce consistency through a voxel-based loss. A lightweight static codebook further bridges continuous features and discrete identities without requiring data pre-processing or complex optimization. Quantitative and qualitative experiments demonstrate the effectiveness of InstDrive, and to the best of our knowledge, it is the first framework to achieve 3D instance segmentation in dynamic, open-world driving scenes.More visualizations are available at our project page.", "date": "2025-08-16", "categories": []}
{"id": "2510.24734", "title": "DrivingScene: A Multi-Task Online Feed-Forward 3D Gaussian Splatting Method for Dynamic Driving Scenes", "abstract": "Real-time, high-fidelity reconstruction of dynamic driving scenes is challenged by complex dynamics and sparse views, with prior methods struggling to balance quality and efficiency. We propose DrivingScene, an online, feed-forward framework that reconstructs 4D dynamic scenes from only two consecutive surround-view images. Our key innovation is a lightweight residual flow network that predicts the non-rigid motion of dynamic objects per camera on top of a learned static scene prior, explicitly modeling dynamics via scene flow. We also introduce a coarse-to-fine training paradigm that circumvents the instabilities common to end-to-end approaches. Experiments on nuScenes dataset show our image-only method simultaneously generates high-quality depth, scene flow, and 3D Gaussian point clouds online, significantly outperforming state-of-the-art methods in both dynamic reconstruction and novel view synthesis.", "date": "2025-10-14", "categories": []}
{"id": "2510.25129", "title": "AtlasGS: Atlanta-world Guided Surface Reconstruction with Implicit Structured Gaussians", "abstract": "3D reconstruction of indoor and urban environments is a prominent research topic with various downstream applications. However, existing geometric priors for addressing low-texture regions in indoor and urban settings often lack global consistency. Moreover, Gaussian Splatting and implicit SDF fields often suffer from discontinuities or exhibit computational inefficiencies, resulting in a loss of detail. To address these issues, we propose an Atlanta-world guided implicit-structured Gaussian Splatting that achieves smooth indoor and urban scene reconstruction while preserving high-frequency details and rendering efficiency. By leveraging the Atlanta-world model, we ensure the accurate surface reconstruction for low-texture regions, while the proposed novel implicit-structured GS representations provide smoothness without sacrificing efficiency and high-frequency details. Specifically, we propose a semantic GS representation to predict the probability of all semantic regions and deploy a structure plane regularization with learnable plane indicators for global accurate surface reconstruction. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in both indoor and urban scenes, delivering superior surface reconstruction quality.", "date": "2025-10-29", "categories": []}
{"id": "2510.25146", "title": "EA3D: Online Open-World 3D Object Extraction from Streaming Videos", "abstract": "Current 3D scene understanding methods are limited by offline-collected multi-view data or pre-constructed 3D geometry. In this paper, we present ExtractAnything3D (EA3D), a unified online framework for open-world 3D object extraction that enables simultaneous geometric reconstruction and holistic scene understanding. Given a streaming video, EA3D dynamically interprets each frame using vision-language and 2D vision foundation encoders to extract object-level knowledge. This knowledge is integrated and embedded into a Gaussian feature map via a feed-forward online update strategy. We then iteratively estimate visual odometry from historical frames and incrementally update online Gaussian features with new observations. A recurrent joint optimization module directs the model's attention to regions of interest, simultaneously enhancing both geometric reconstruction and semantic understanding. Extensive experiments across diverse benchmarks and tasks, including photo-realistic rendering, semantic and instance segmentation, 3D bounding box and semantic occupancy estimation, and 3D mesh generation, demonstrate the effectiveness of EA3D. Our method establishes a unified and efficient framework for joint online 3D reconstruction and holistic scene understanding, enabling a broad range of downstream tasks.", "date": "2025-10-29", "categories": []}
{"id": "2505.23642", "title": "Radiant Triangle Soup with Soft Connectivity Forces for 3D Reconstruction and Novel View Synthesis", "abstract": "We introduce an inference-time scene optimization algorithm utilizing triangle soup, a collection of disconnected translucent triangle primitives, as the representation for the geometry and appearance of a scene. Unlike full-rank Gaussian kernels, triangles are a natural, locally-flat proxy for surfaces that can be connected to achieve highly complex geometry. When coupled with per-vertex Spherical Harmonics (SH), triangles provide a rich visual representation without incurring an expensive increase in primitives. We leverage our new representation to incorporate optimization objectives and enforce spatial regularization directly on the underlying primitives. The main differentiator of our approach is the definition and enforcement of soft connectivity forces between triangles during optimization, encouraging explicit, but soft, surface continuity in 3D. Experiments on representative 3D reconstruction and novel view synthesis datasets show improvements in geometric accuracy compared to current state-of-the-art algorithms without sacrificing visual fidelity.", "date": "2025-05-29", "categories": []}
{"id": "2507.00371", "title": "PlantSegNeRF: A few-shot, cross-species method for plant 3D instance point cloud reconstruction via joint-channel NeRF with multi-view image instance matching", "abstract": "Organ segmentation of plant point clouds is a prerequisite for the high-resolution and accurate extraction of organ-level phenotypic traits. Although the fast development of deep learning has boosted much research on segmentation of plant point clouds, the existing techniques for organ segmentation still face limitations in resolution, segmentation accuracy, and generalizability across various plant species. In this study, we proposed a novel approach called plant segmentation neural radiance fields (PlantSegNeRF), aiming to directly generate high-precision instance point clouds from multi-view RGB image sequences for a wide range of plant species. PlantSegNeRF performed 2D instance segmentation on the multi-view images to generate instance masks for each organ with a corresponding ID. The multi-view instance IDs corresponding to the same plant organ were then matched and refined using a specially designed instance matching module. The instance NeRF was developed to render an implicit scene, containing color, density, semantic and instance information. The implicit scene was ultimately converted into high-precision plant instance point clouds based on the volume density. The results proved that in semantic segmentation of point clouds, PlantSegNeRF outperformed the commonly used methods, demonstrating an average improvement of 16.1%, 18.3%, 17.8%, and 24.2% in precision, recall, F1-score, and IoU compared to the second-best results on structurally complex species. More importantly, PlantSegNeRF exhibited significant advantages in plant point cloud instance segmentation tasks. Across all plant species, it achieved average improvements of 11.7%, 38.2%, 32.2% and 25.3% in mPrec, mRec, mCov, mWCov, respectively. This study extends the organ-level plant phenotyping and provides a high-throughput way to supply high-quality 3D data for the development of large-scale models in plant science.", "date": "2025-07-01", "categories": []}
{"id": "2509.25744", "title": "Image-Plane Geometric Decoding for View-Invariant Indoor Scene Reconstruction", "abstract": "Volume-based indoor scene reconstruction methods offer superior generalization capability and real-time deployment potential. However, existing methods rely on multi-view pixel back-projection ray intersections as weak geometric constraints to determine spatial positions. This dependence results in reconstruction quality being heavily influenced by input view density. Performance degrades in overlapping regions and unobserved areas.To address these limitations, we reduce dependency on inter-view geometric constraints by exploiting spatial information within individual views. We propose an image-plane decoding framework with three core components: Pixel-level Confidence Encoder, Affine Compensation Module, and Image-Plane Spatial Decoder. These modules decode three-dimensional structural information encoded in images through physical imaging processes. The framework effectively preserves spatial geometric features including edges, hollow structures, and complex textures. It significantly enhances view-invariant reconstruction.Experiments on indoor scene reconstruction datasets confirm superior reconstruction stability. Our method maintains nearly identical quality when view count reduces by 40%. It achieves a coefficient of variation of 0.24%, performance retention rate of 99.7%, and maximum performance drop of 0.42%. These results demonstrate that exploiting intra-view spatial information provides a robust solution for view-limited scenarios in practical applications.", "date": "2025-09-30", "categories": []}
{"id": "2510.10097", "title": "Gesplat: Robust Pose-Free 3D Reconstruction via Geometry-Guided Gaussian Splatting", "abstract": "Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have advanced 3D reconstruction and novel view synthesis, but remain heavily dependent on accurate camera poses and dense viewpoint coverage. These requirements limit their applicability in sparse-view settings, where pose estimation becomes unreliable and supervision is insufficient. To overcome these challenges, we introduce Gesplat, a 3DGS-based framework that enables robust novel view synthesis and geometrically consistent reconstruction from unposed sparse images. Unlike prior works that rely on COLMAP for sparse point cloud initialization, we leverage the VGGT foundation model to obtain more reliable initial poses and dense point clouds. Our approach integrates several key innovations: 1) a hybrid Gaussian representation with dual position-shape optimization enhanced by inter-view matching consistency; 2) a graph-guided attribute refinement module to enhance scene details; and 3) flow-based depth regularization that improves depth estimation accuracy for more effective supervision. Comprehensive quantitative and qualitative experiments demonstrate that our approach achieves more robust performance on both forward-facing and large-scale complex datasets compared to other pose-free methods.", "date": "2025-10-11", "categories": []}
{"id": "2510.23087", "title": "EndoWave: Rational-Wavelet 4D Gaussian Splatting for Endoscopic Reconstruction", "abstract": "In robot-assisted minimally invasive surgery, accurate 3D reconstruction from endoscopic video is vital for downstream tasks and improved outcomes. However, endoscopic scenarios present unique challenges, including photometric inconsistencies, non-rigid tissue motion, and view-dependent highlights. Most 3DGS-based methods that rely solely on appearance constraints for optimizing 3DGS are often insufficient in this context, as these dynamic visual artifacts can mislead the optimization process and lead to inaccurate reconstructions. To address these limitations, we present EndoWave, a unified spatio-temporal Gaussian Splatting framework by incorporating an optical flow-based geometric constraint and a multi-resolution rational wavelet supervision. First, we adopt a unified spatio-temporal Gaussian representation that directly optimizes primitives in a 4D domain. Second, we propose a geometric constraint derived from optical flow to enhance temporal coherence and effectively constrain the 3D structure of the scene. Third, we propose a multi-resolution rational orthogonal wavelet as a constraint, which can effectively separate the details of the endoscope and enhance the rendering performance. Extensive evaluations on two real surgical datasets, EndoNeRF and StereoMIS, demonstrate that our method EndoWave achieves state-of-the-art reconstruction quality and visual accuracy compared to the baseline method.", "date": "2025-10-27", "categories": []}
{"id": "2506.12945", "title": "Metropolis-Hastings Sampling for 3D Gaussian Reconstruction", "abstract": "We propose an adaptive sampling framework for 3D Gaussian Splatting (3DGS) that leverages comprehensive multi-view photometric error signals within a unified Metropolis-Hastings approach. Vanilla 3DGS heavily relies on heuristic-based density-control mechanisms (e.g., cloning, splitting, and pruning), which can lead to redundant computations or premature removal of beneficial Gaussians. Our framework overcomes these limitations by reformulating densification and pruning as a probabilistic sampling process, dynamically inserting and relocating Gaussians based on aggregated multi-view errors and opacity scores. Guided by Bayesian acceptance tests derived from these error-based importance scores, our method substantially reduces reliance on heuristics, offers greater flexibility, and adaptively infers Gaussian distributions without requiring predefined scene complexity. Experiments on benchmark datasets, including Mip-NeRF360, Tanks and Temples and Deep Blending, show that our approach reduces the number of Gaussians needed, achieving faster convergence while matching or modestly surpassing the view-synthesis quality of state-of-the-art models.", "date": "2025-06-15", "categories": []}
{"id": "2507.14798", "title": "An Evaluation of DUSt3R/MASt3R/VGGT 3D Reconstruction on Photogrammetric Aerial Blocks", "abstract": "State-of-the-art 3D computer vision algorithms continue to advance in handling sparse, unordered image sets. Recently developed foundational models for 3D reconstruction, such as Dense and Unconstrained Stereo 3D Reconstruction (DUSt3R), Matching and Stereo 3D Reconstruction (MASt3R), and Visual Geometry Grounded Transformer (VGGT), have attracted attention due to their ability to handle very sparse image overlaps. Evaluating DUSt3R/MASt3R/VGGT on typical aerial images matters, as these models may handle extremely low image overlaps, stereo occlusions, and textureless regions. For redundant collections, they can accelerate 3D reconstruction by using extremely sparsified image sets. Despite tests on various computer vision benchmarks, their potential on photogrammetric aerial blocks remains unexplored. This paper conducts a comprehensive evaluation of the pre-trained DUSt3R/MASt3R/VGGT models on the aerial blocks of the UseGeo dataset for pose estimation and dense 3D reconstruction. Results show these methods can accurately reconstruct dense point clouds from very sparse image sets (fewer than 10 images, up to 518 pixels resolution), with completeness gains up to +50% over COLMAP. VGGT also demonstrates higher computational efficiency, scalability, and more reliable camera pose estimation. However, all exhibit limitations with high-resolution images and large sets, as pose reliability declines with more images and geometric complexity. These findings suggest transformer-based methods cannot fully replace traditional SfM and MVS, but offer promise as complementary approaches, especially in challenging, low-resolution, and sparse scenarios.", "date": "2025-07-20", "categories": []}
{"id": "2510.20605", "title": "OnlineSplatter: Pose-Free Online 3D Reconstruction for Free-Moving Objects", "abstract": "Free-moving object reconstruction from monocular video remains challenging, particularly without reliable pose or depth cues and under arbitrary object motion. We introduce OnlineSplatter, a novel online feed-forward framework generating high-quality, object-centric 3D Gaussians directly from RGB frames without requiring camera pose, depth priors, or bundle optimization. Our approach anchors reconstruction using the first frame and progressively refines the object representation through a dense Gaussian primitive field, maintaining constant computational cost regardless of video sequence length. Our core contribution is a dual-key memory module combining latent appearance-geometry keys with explicit directional keys, robustly fusing current frame features with temporally aggregated object states. This design enables effective handling of free-moving objects via spatial-guided memory readout and an efficient sparsification mechanism, ensuring comprehensive yet compact object coverage. Evaluations on real-world datasets demonstrate that OnlineSplatter significantly outperforms state-of-the-art pose-free reconstruction baselines, consistently improving with more observations while maintaining constant memory and runtime.", "date": "2025-10-23", "categories": []}
{"id": "2510.19578", "title": "VGD: Visual Geometry Gaussian Splatting for Feed-Forward Surround-view Driving Reconstruction", "abstract": "Feed-forward surround-view autonomous driving scene reconstruction offers fast, generalizable inference ability, which faces the core challenge of ensuring generalization while elevating novel view quality. Due to the surround-view with minimal overlap regions, existing methods typically fail to ensure geometric consistency and reconstruction quality for novel views. To tackle this tension, we claim that geometric information must be learned explicitly, and the resulting features should be leveraged to guide the elevating of semantic quality in novel views. In this paper, we introduce \\textbf{Visual Gaussian Driving (VGD)}, a novel feed-forward end-to-end learning framework designed to address this challenge. To achieve generalizable geometric estimation, we design a lightweight variant of the VGGT architecture to efficiently distill its geometric priors from the pre-trained VGGT to the geometry branch. Furthermore, we design a Gaussian Head that fuses multi-scale geometry tokens to predict Gaussian parameters for novel view rendering, which shares the same patch backbone as the geometry branch. Finally, we integrate multi-scale features from both geometry and Gaussian head branches to jointly supervise a semantic refinement model, optimizing rendering quality through feature-consistent learning. Experiments on nuScenes demonstrate that our approach significantly outperforms state-of-the-art methods in both objective metrics and subjective quality under various settings, which validates VGD's scalability and high-fidelity surround-view reconstruction.", "date": "2025-10-22", "categories": []}
{"id": "2408.13036", "title": "H3D-DGS: Exploring Heterogeneous 3D Motion Representation for Deformable 3D Gaussian Splatting", "abstract": "Dynamic scene reconstruction poses a persistent challenge in 3D vision. Deformable 3D Gaussian Splatting has emerged as an effective method for this task, offering real-time rendering and high visual fidelity. This approach decomposes a dynamic scene into a static representation in a canonical space and time-varying scene motion. Scene motion is defined as the collective movement of all Gaussian points, and for compactness, existing approaches commonly adopt implicit neural fields or sparse control points. However, these methods predominantly rely on gradient-based optimization for all motion information. Due to the high degree of freedom, they struggle to converge on real-world datasets exhibiting complex motion. To preserve the compactness of motion representation and address convergence challenges, this paper proposes heterogeneous 3D control points, termed \\textbf{H3D control points}, whose attributes are obtained using a hybrid strategy combining optical flow back-projection and gradient-based methods. This design decouples directly observable motion components from those that are geometrically occluded. Specifically, components of 3D motion that project onto the image plane are directly acquired via optical flow back projection, while unobservable portions are refined through gradient-based optimization. Experiments on the Neu3DV and CMU-Panoptic datasets demonstrate that our method achieves superior performance over state-of-the-art deformable 3D Gaussian splatting techniques. Remarkably, our method converges within just 100 iterations and achieves a per-frame processing speed of 2 seconds on a single NVIDIA RTX 4070 GPU.", "date": "2024-08-23", "categories": []}
{"id": "2505.22978", "title": "Pose-free 3D Gaussian splatting via shape-ray estimation", "abstract": "While generalizable 3D Gaussian splatting enables efficient, high-quality rendering of unseen scenes, it heavily depends on precise camera poses for accurate geometry. In real-world scenarios, obtaining accurate poses is challenging, leading to noisy pose estimates and geometric misalignments. To address this, we introduce SHARE, a pose-free, feed-forward Gaussian splatting framework that overcomes these ambiguities by joint shape and camera rays estimation. Instead of relying on explicit 3D transformations, SHARE builds a pose-aware canonical volume representation that seamlessly integrates multi-view information, reducing misalignment caused by inaccurate pose estimates. Additionally, anchor-aligned Gaussian prediction enhances scene reconstruction by refining local geometry around coarse anchors, allowing for more precise Gaussian placement. Extensive experiments on diverse real-world datasets show that our method achieves robust performance in pose-free generalizable Gaussian splatting. Code is avilable at https://github.com/youngju-na/SHARE", "date": "2025-05-29", "categories": []}
{"id": "2509.12458", "title": "Neural 3D Object Reconstruction with Small-Scale Unmanned Aerial Vehicles", "abstract": "Small Unmanned Aerial Vehicles (UAVs) exhibit immense potential for navigating indoor and hard-to-reach areas, yet their significant constraints in payload and autonomy have largely prevented their use for complex tasks like high-quality 3-Dimensional (3D) reconstruction. To overcome this challenge, we introduce a novel system architecture that enables fully autonomous, high-fidelity 3D scanning of static objects using UAVs weighing under 100 grams. Our core innovation lies in a dual-reconstruction pipeline that creates a real-time feedback loop between data capture and flight control. A near-real-time (near-RT) process uses Structure from Motion (SfM) to generate an instantaneous pointcloud of the object. The system analyzes the model quality on the fly and dynamically adapts the UAV's trajectory to intelligently capture new images of poorly covered areas. This ensures comprehensive data acquisition. For the final, detailed output, a non-real-time (non-RT) pipeline employs a Neural Radiance Fields (NeRF)-based Neural 3D Reconstruction (N3DR) approach, fusing SfM-derived camera poses with precise Ultra Wide-Band (UWB) location data to achieve superior accuracy. We implemented and validated this architecture using Crazyflie 2.1 UAVs. Our experiments, conducted in both single- and multi-UAV configurations, conclusively show that dynamic trajectory adaptation consistently improves reconstruction quality over static flight paths. This work demonstrates a scalable and autonomous solution that unlocks the potential of miniaturized UAVs for fine-grained 3D reconstruction in constrained environments, a capability previously limited to much larger platforms.", "date": "2025-09-15", "categories": []}
{"id": "2510.18172", "title": "Adapting Stereo Vision From Objects To 3D Lunar Surface Reconstruction with the StereoLunar Dataset", "abstract": "Accurate 3D reconstruction of lunar surfaces is essential for space exploration. However, existing stereo vision reconstruction methods struggle in this context due to the Moon's lack of texture, difficult lighting variations, and atypical orbital trajectories. State-of-the-art deep learning models, trained on human-scale datasets, have rarely been tested on planetary imagery and cannot be transferred directly to lunar conditions. To address this issue, we introduce LunarStereo, the first open dataset of photorealistic stereo image pairs of the Moon, simulated using ray tracing based on high-resolution topography and reflectance models. It covers diverse altitudes, lighting conditions, and viewing angles around the lunar South Pole, offering physically grounded supervision for 3D reconstruction tasks. Based on this dataset, we adapt the MASt3R model to the lunar domain through fine-tuning on LunarStereo. We validate our approach through extensive qualitative and quantitative experiments on both synthetic and real lunar data, evaluating 3D surface reconstruction and relative pose estimation. Extensive experiments on synthetic and real lunar data validate the approach, demonstrating significant improvements over zero-shot baselines and paving the way for robust cross-scale generalization in extraterrestrial environments.", "date": "2025-10-20", "categories": []}
{"id": "2502.20154", "title": "Cutting-edge 3D reconstruction solutions for underwater coral reef images: A review and comparison", "abstract": "Corals serve as the foundational habitat-building organisms within reef ecosystems, constructing extensive structures that extend over vast distances. However, their inherent fragility and vulnerability to various threats render them susceptible to significant damage and destruction. The application of advanced 3D reconstruction technologies for high-quality modeling is crucial for preserving them. These technologies help scientists to accurately document and monitor the state of coral reefs, including their structure, species distribution and changes over time. Photogrammetry-based approaches stand out among existing solutions, especially with recent advancements in underwater videography, photogrammetric computer vision, and machine learning. Despite continuous progress in image-based 3D reconstruction techniques, there remains a lack of systematic reviews and comprehensive evaluations of cutting-edge solutions specifically applied to underwater coral reef images. The emerging advanced methods may have difficulty coping with underwater imaging environments, complex coral structures, and computational resource constraints. They need to be reviewed and evaluated to bridge the gap between many cutting-edge technical studies and practical applications. This paper focuses on the two critical stages of these approaches: camera pose estimation and dense surface reconstruction. We systematically review and summarize classical and emerging methods, conducting comprehensive evaluations through real-world and simulated datasets. Based on our findings, we offer reference recommendations and discuss the development potential and challenges of existing approaches in depth. This work equips scientists and managers with a technical foundation and practical guidance for processing underwater coral reef images for 3D reconstruction....", "date": "2025-02-27", "categories": []}
{"id": "2504.10809", "title": "GaSLight: Gaussian Splats for Spatially-Varying Lighting in HDR", "abstract": "We present GaSLight, a method that generates spatially-varying lighting from regular images. Our method proposes using HDR Gaussian Splats as light source representation, marking the first time regular images can serve as light sources in a 3D renderer. Our two-stage process first enhances the dynamic range of images plausibly and accurately by leveraging the priors embedded in diffusion models. Next, we employ Gaussian Splats to model 3D lighting, achieving spatially variant lighting. Our approach yields state-of-the-art results on HDR estimations and their applications in illuminating virtual objects and scenes. To facilitate the benchmarking of images as light sources, we introduce a novel dataset of calibrated and unsaturated HDR to evaluate images as light sources. We assess our method using a combination of this novel dataset and an existing dataset from the literature. Project page: https://lvsn.github.io/gaslight/", "date": "2025-04-15", "categories": []}
{"id": "2505.21060", "title": "Styl3R: Instant 3D Stylized Reconstruction for Arbitrary Scenes and Styles", "abstract": "Stylizing 3D scenes instantly while maintaining multi-view consistency and faithfully resembling a style image remains a significant challenge. Current state-of-the-art 3D stylization methods typically involve computationally intensive test-time optimization to transfer artistic features into a pretrained 3D representation, often requiring dense posed input images. In contrast, leveraging recent advances in feed-forward reconstruction models, we demonstrate a novel approach to achieve direct 3D stylization in less than a second using unposed sparse-view scene images and an arbitrary style image. To address the inherent decoupling between reconstruction and stylization, we introduce a branched architecture that separates structure modeling and appearance shading, effectively preventing stylistic transfer from distorting the underlying 3D scene structure. Furthermore, we adapt an identity loss to facilitate pre-training our stylization model through the novel view synthesis task. This strategy also allows our model to retain its original reconstruction capabilities while being fine-tuned for stylization. Comprehensive evaluations, using both in-domain and out-of-domain datasets, demonstrate that our approach produces high-quality stylized 3D content that achieve a superior blend of style and scene appearance, while also outperforming existing methods in terms of multi-view consistency and efficiency.", "date": "2025-05-27", "categories": []}
{"id": "2510.16272", "title": "Proactive Scene Decomposition and Reconstruction", "abstract": "Human behaviors are the major causes of scene dynamics and inherently contain rich cues regarding the dynamics. This paper formalizes a new task of proactive scene decomposition and reconstruction, an online approach that leverages human-object interactions to iteratively disassemble and reconstruct the environment. By observing these intentional interactions, we can dynamically refine the decomposition and reconstruction process, addressing inherent ambiguities in static object-level reconstruction. The proposed system effectively integrates multiple tasks in dynamic environments such as accurate camera and object pose estimation, instance decomposition, and online map updating, capitalizing on cues from human-object interactions in egocentric live streams for a flexible, progressive alternative to conventional object-level reconstruction methods. Aided by the Gaussian splatting technique, accurate and consistent dynamic scene modeling is achieved with photorealistic and efficient rendering. The efficacy is validated in multiple real-world scenarios with promising advantages.", "date": "2025-10-17", "categories": []}
{"id": "2510.17095", "title": "GSPlane: Concise and Accurate Planar Reconstruction via Structured Representation", "abstract": "Planes are fundamental primitives of 3D sences, especially in man-made environments such as indoor spaces and urban streets. Representing these planes in a structured and parameterized format facilitates scene editing and physical simulations in downstream applications. Recently, Gaussian Splatting (GS) has demonstrated remarkable effectiveness in the Novel View Synthesis task, with extensions showing great potential in accurate surface reconstruction. However, even state-of-the-art GS representations often struggle to reconstruct planar regions with sufficient smoothness and precision. To address this issue, we propose GSPlane, which recovers accurate geometry and produces clean and well-structured mesh connectivity for plane regions in the reconstructed scene. By leveraging off-the-shelf segmentation and normal prediction models, GSPlane extracts robust planar priors to establish structured representations for planar Gaussian coordinates, which help guide the training process by enforcing geometric consistency. To further enhance training robustness, a Dynamic Gaussian Re-classifier is introduced to adaptively reclassify planar Gaussians with persistently high gradients as non-planar, ensuring more reliable optimization. Furthermore, we utilize the optimized planar priors to refine the mesh layouts, significantly improving topological structure while reducing the number of vertices and faces. We also explore applications of the structured planar representation, which enable decoupling and flexible manipulation of objects on supportive planes. Extensive experiments demonstrate that, with no sacrifice in rendering quality, the introduction of planar priors significantly improves the geometric accuracy of the extracted meshes across various baselines.", "date": "2025-10-20", "categories": []}
{"id": "2508.02493", "title": "Low-Frequency First: Eliminating Floating Artifacts in 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) is a powerful and computationally efficient representation for 3D reconstruction. Despite its strengths, 3DGS often produces floating artifacts, which are erroneous structures detached from the actual geometry and significantly degrade visual fidelity. The underlying mechanisms causing these artifacts, particularly in low-quality initialization scenarios, have not been fully explored. In this paper, we investigate the origins of floating artifacts from a frequency-domain perspective and identify under-optimized Gaussians as the primary source. Based on our analysis, we propose \\textit{Eliminating-Floating-Artifacts} Gaussian Splatting (EFA-GS), which selectively expands under-optimized Gaussians to prioritize accurate low-frequency learning. Additionally, we introduce complementary depth-based and scale-based strategies to dynamically refine Gaussian expansion, effectively mitigating detail erosion. Extensive experiments on both synthetic and real-world datasets demonstrate that EFA-GS substantially reduces floating artifacts while preserving high-frequency details, achieving an improvement of 1.68 dB in PSNR over baseline method on our RWLQ dataset. Furthermore, we validate the effectiveness of our approach in downstream 3D editing tasks. Project Website: https://jcwang-gh.github.io/EFA-GS", "date": "2025-08-04", "categories": []}
{"id": "2510.01978", "title": "ROI-GS: Interest-based Local Quality 3D Gaussian Splatting", "abstract": "We tackle the challenge of efficiently reconstructing 3D scenes with high detail on objects of interest. Existing 3D Gaussian Splatting (3DGS) methods allocate resources uniformly across the scene, limiting fine detail to Regions Of Interest (ROIs) and leading to inflated model size. We propose ROI-GS, an object-aware framework that enhances local details through object-guided camera selection, targeted Object training, and seamless integration of high-fidelity object of interest reconstructions into the global scene. Our method prioritizes higher resolution details on chosen objects while maintaining real-time performance. Experiments show that ROI-GS significantly improves local quality (up to 2.96 dB PSNR), while reducing overall model size by $\\approx 17\\%$ of baseline and achieving faster training for a scene with a single object of interest, outperforming existing methods.", "date": "2025-10-02", "categories": []}
{"id": "2510.15072", "title": "SaLon3R: Structure-aware Long-term Generalizable 3D Reconstruction from Unposed Images", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have enabled generalizable, on-the-fly reconstruction of sequential input views. However, existing methods often predict per-pixel Gaussians and combine Gaussians from all views as the scene representation, leading to substantial redundancies and geometric inconsistencies in long-duration video sequences. To address this, we propose SaLon3R, a novel framework for Structure-aware, Long-term 3DGS Reconstruction. To our best knowledge, SaLon3R is the first online generalizable GS method capable of reconstructing over 50 views in over 10 FPS, with 50% to 90% redundancy removal. Our method introduces compact anchor primitives to eliminate redundancy through differentiable saliency-aware Gaussian quantization, coupled with a 3D Point Transformer that refines anchor attributes and saliency to resolve cross-frame geometric and photometric inconsistencies. Specifically, we first leverage a 3D reconstruction backbone to predict dense per-pixel Gaussians and a saliency map encoding regional geometric complexity. Redundant Gaussians are compressed into compact anchors by prioritizing high-complexity regions. The 3D Point Transformer then learns spatial structural priors in 3D space from training data to refine anchor attributes and saliency, enabling regionally adaptive Gaussian decoding for geometric fidelity. Without known camera parameters or test-time optimization, our approach effectively resolves artifacts and prunes the redundant 3DGS in a single feed-forward pass. Experiments on multiple datasets demonstrate our state-of-the-art performance on both novel view synthesis and depth estimation, demonstrating superior efficiency, robustness, and generalization ability for long-term generalizable 3D reconstruction. Project Page: https://wrld.github.io/SaLon3R/.", "date": "2025-10-16", "categories": []}
{"id": "2510.15467", "title": "MRASfM: Multi-Camera Reconstruction and Aggregation through Structure-from-Motion in Driving Scenes", "abstract": "Structure from Motion (SfM) estimates camera poses and reconstructs point clouds, forming a foundation for various tasks. However, applying SfM to driving scenes captured by multi-camera systems presents significant difficulties, including unreliable pose estimation, excessive outliers in road surface reconstruction, and low reconstruction efficiency. To address these limitations, we propose a Multi-camera Reconstruction and Aggregation Structure-from-Motion (MRASfM) framework specifically designed for driving scenes. MRASfM enhances the reliability of camera pose estimation by leveraging the fixed spatial relationships within the multi-camera system during the registration process. To improve the quality of road surface reconstruction, our framework employs a plane model to effectively remove erroneous points from the triangulated road surface. Moreover, treating the multi-camera set as a single unit in Bundle Adjustment (BA) helps reduce optimization variables to boost efficiency. In addition, MRASfM achieves multi-scene aggregation through scene association and assembly modules in a coarse-to-fine fashion. We deployed multi-camera systems on actual vehicles to validate the generalizability of MRASfM across various scenes and its robustness in challenging conditions through real-world applications. Furthermore, large-scale validation results on public datasets show the state-of-the-art performance of MRASfM, achieving 0.124 absolute pose error on the nuScenes dataset.", "date": "2025-10-17", "categories": []}
{"id": "2510.15491", "title": "Iterative Motion Compensation for Canonical 3D Reconstruction from UAV Plant Images Captured in Windy Conditions", "abstract": "3D phenotyping of plants plays a crucial role for understanding plant growth, yield prediction, and disease control. We present a pipeline capable of generating high-quality 3D reconstructions of individual agricultural plants. To acquire data, a small commercially available UAV captures images of a selected plant. Apart from placing ArUco markers, the entire image acquisition process is fully autonomous, controlled by a self-developed Android application running on the drone's controller. The reconstruction task is particularly challenging due to environmental wind and downwash of the UAV. Our proposed pipeline supports the integration of arbitrary state-of-the-art 3D reconstruction methods. To mitigate errors caused by leaf motion during image capture, we use an iterative method that gradually adjusts the input images through deformation. Motion is estimated using optical flow between the original input images and intermediate 3D reconstructions rendered from the corresponding viewpoints. This alignment gradually reduces scene motion, resulting in a canonical representation. After a few iterations, our pipeline improves the reconstruction of state-of-the-art methods and enables the extraction of high-resolution 3D meshes. We will publicly release the source code of our reconstruction pipeline. Additionally, we provide a dataset consisting of multiple plants from various crops, captured across different points in time.", "date": "2025-10-17", "categories": []}
{"id": "2407.13764", "title": "Shape of Motion: 4D Reconstruction from a Single Video", "abstract": "Monocular dynamic reconstruction is a challenging and long-standing vision problem due to the highly ill-posed nature of the task. Existing approaches depend on templates, are effective only in quasi-static scenes, or fail to model 3D motion explicitly. We introduce a method for reconstructing generic dynamic scenes, featuring explicit, persistent 3D motion trajectories in the world coordinate frame, from casually captured monocular videos. We tackle the problem with two key insights: First, we exploit the low-dimensional structure of 3D motion by representing scene motion with a compact set of SE(3) motion bases. Each point's motion is expressed as a linear combination of these bases, facilitating soft decomposition of the scene into multiple rigidly-moving groups. Second, we take advantage of off-the-shelf data-driven priors such as monocular depth maps and long-range 2D tracks, and devise a method to effectively consolidate these noisy supervisory signals, resulting in a globally consistent representation of the dynamic scene. Experiments show that our method achieves state-of-the-art performance for both long-range 3D/2D motion estimation and novel view synthesis on dynamic scenes. Project Page: https://shape-of-motion.github.io/", "date": "2024-07-18", "categories": []}
{"id": "2506.21117", "title": "CL-Splats: Continual Learning of Gaussian Splatting with Local Optimization", "abstract": "In dynamic 3D environments, accurately updating scene representations over time is crucial for applications in robotics, mixed reality, and embodied AI. As scenes evolve, efficient methods to incorporate changes are needed to maintain up-to-date, high-quality reconstructions without the computational overhead of re-optimizing the entire scene. This paper introduces CL-Splats, which incrementally updates Gaussian splatting-based 3D representations from sparse scene captures. CL-Splats integrates a robust change-detection module that segments updated and static components within the scene, enabling focused, local optimization that avoids unnecessary re-computation. Moreover, CL-Splats supports storing and recovering previous scene states, facilitating temporal segmentation and new scene-analysis applications. Our extensive experiments demonstrate that CL-Splats achieves efficient updates with improved reconstruction quality over the state-of-the-art. This establishes a robust foundation for future real-time adaptation in 3D scene reconstruction tasks.", "date": "2025-06-26", "categories": []}
{"id": "2510.14251", "title": "MACE: Mixture-of-Experts Accelerated Coordinate Encoding for Large-Scale Scene Localization and Rendering", "abstract": "Efficient localization and high-quality rendering in large-scale scenes remain a significant challenge due to the computational cost involved. While Scene Coordinate Regression (SCR) methods perform well in small-scale localization, they are limited by the capacity of a single network when extended to large-scale scenes. To address these challenges, we propose the Mixed Expert-based Accelerated Coordinate Encoding method (MACE), which enables efficient localization and high-quality rendering in large-scale scenes. Inspired by the remarkable capabilities of MOE in large model domains, we introduce a gating network to implicitly classify and select sub-networks, ensuring that only a single sub-network is activated during each inference. Furtheremore, we present Auxiliary-Loss-Free Load Balancing(ALF-LB) strategy to enhance the localization accuracy on large-scale scene. Our framework provides a significant reduction in costs while maintaining higher precision, offering an efficient solution for large-scale scene applications. Additional experiments on the Cambridge test set demonstrate that our method achieves high-quality rendering results with merely 10 minutes of training.", "date": "2025-10-16", "categories": []}
{"id": "2510.14564", "title": "BalanceGS: Algorithm-System Co-design for Efficient 3D Gaussian Splatting Training on GPU", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a promising 3D reconstruction technique. The traditional 3DGS training pipeline follows three sequential steps: Gaussian densification, Gaussian projection, and color splatting. Despite its promising reconstruction quality, this conventional approach suffers from three critical inefficiencies: (1) Skewed density allocation during Gaussian densification, (2) Imbalanced computation workload during Gaussian projection and (3) Fragmented memory access during color splatting.\n  To tackle the above challenges, we introduce BalanceGS, the algorithm-system co-design for efficient training in 3DGS. (1) At the algorithm level, we propose heuristic workload-sensitive Gaussian density control to automatically balance point distributions - removing 80% redundant Gaussians in dense regions while filling gaps in sparse areas. (2) At the system level, we propose Similarity-based Gaussian sampling and merging, which replaces the static one-to-one thread-pixel mapping with adaptive workload distribution - threads now dynamically process variable numbers of Gaussians based on local cluster density. (3) At the mapping level, we propose reordering-based memory access mapping strategy that restructures RGB storage and enables batch loading in shared memory.\n  Extensive experiments demonstrate that compared with 3DGS, our approach achieves a 1.44$\\times$ training speedup on a NVIDIA A100 GPU with negligible quality degradation.", "date": "2025-10-16", "categories": []}
{"id": "2412.11579", "title": "SweepEvGS: Event-Based 3D Gaussian Splatting for Macro and Micro Radiance Field Rendering from a Single Sweep", "abstract": "Recent advancements in 3D Gaussian Splatting (3D-GS) have demonstrated the potential of using 3D Gaussian primitives for high-speed, high-fidelity, and cost-efficient novel view synthesis from continuously calibrated input views. However, conventional methods require high-frame-rate dense and high-quality sharp images, which are time-consuming and inefficient to capture, especially in dynamic environments. Event cameras, with their high temporal resolution and ability to capture asynchronous brightness changes, offer a promising alternative for more reliable scene reconstruction without motion blur. In this paper, we propose SweepEvGS, a novel hardware-integrated method that leverages event cameras for robust and accurate novel view synthesis across various imaging settings from a single sweep. SweepEvGS utilizes the initial static frame with dense event streams captured during a single camera sweep to effectively reconstruct detailed scene views. We also introduce different real-world hardware imaging systems for real-world data collection and evaluation for future research. We validate the robustness and efficiency of SweepEvGS through experiments in three different imaging settings: synthetic objects, real-world macro-level, and real-world micro-level view synthesis. Our results demonstrate that SweepEvGS surpasses existing methods in visual rendering quality, rendering speed, and computational efficiency, highlighting its potential for dynamic practical applications.", "date": "2024-12-16", "categories": []}
{"id": "2509.15459", "title": "CAGE: Continuity-Aware edGE Network Unlocks Robust Floorplan Reconstruction", "abstract": "We present CAGE (Continuity-Aware edGE) network, a robust framework for reconstructing vector floorplans directly from point-cloud density maps. Traditional corner-based polygon representations are highly sensitive to noise and incomplete observations, often resulting in fragmented or implausible layouts.Recent line grouping methods leverage structural cues to improve robustness but still struggle to recover fine geometric details. To address these limitations,we propose a native edge-centric formulation, modeling each wall segment as a directed, geometrically continuous edge. This representation enables inference of coherent floorplan structures, ensuring watertight, topologically valid room boundaries while improving robustness and reducing artifacts. Towards this design, we develop a dual-query transformer decoder that integrates perturbed and latent queries within a denoising framework, which not only stabilizes optimization but also accelerates convergence. Extensive experiments on Structured3D and SceneCAD show that CAGE achieves state-of-the-art performance, with F1 scores of 99.1% (rooms), 91.7% (corners), and 89.3% (angles). The method also demonstrates strong cross-dataset generalization, underscoring the efficacy of our architectural innovations. Code and pretrained models are available on our project page: https://github.com/ee-Liu/CAGE.git.", "date": "2025-09-18", "categories": []}
{"id": "2510.12282", "title": "PAGS: Priority-Adaptive Gaussian Splatting for Dynamic Driving Scenes", "abstract": "Reconstructing dynamic 3D urban scenes is crucial for autonomous driving, yet current methods face a stark trade-off between fidelity and computational cost. This inefficiency stems from their semantically agnostic design, which allocates resources uniformly, treating static backgrounds and safety-critical objects with equal importance. To address this, we introduce Priority-Adaptive Gaussian Splatting (PAGS), a framework that injects task-aware semantic priorities directly into the 3D reconstruction and rendering pipeline. PAGS introduces two core contributions: (1) Semantically-Guided Pruning and Regularization strategy, which employs a hybrid importance metric to aggressively simplify non-critical scene elements while preserving fine-grained details on objects vital for navigation. (2) Priority-Driven Rendering pipeline, which employs a priority-based depth pre-pass to aggressively cull occluded primitives and accelerate the final shading computations. Extensive experiments on the Waymo and KITTI datasets demonstrate that PAGS achieves exceptional reconstruction quality, particularly on safety-critical objects, while significantly reducing training time and boosting rendering speeds to over 350 FPS.", "date": "2025-10-14", "categories": []}
{"id": "2510.12387", "title": "Scene Coordinate Reconstruction Priors", "abstract": "Scene coordinate regression (SCR) models have proven to be powerful implicit scene representations for 3D vision, enabling visual relocalization and structure-from-motion. SCR models are trained specifically for one scene. If training images imply insufficient multi-view constraints SCR models degenerate. We present a probabilistic reinterpretation of training SCR models, which allows us to infuse high-level reconstruction priors. We investigate multiple such priors, ranging from simple priors over the distribution of reconstructed depth values to learned priors over plausible scene coordinate configurations. For the latter, we train a 3D point cloud diffusion model on a large corpus of indoor scans. Our priors push predicted 3D scene points towards plausible geometry at each training step to increase their likelihood. On three indoor datasets our priors help learning better scene representations, resulting in more coherent scene point clouds, higher registration rates and better camera poses, with a positive effect on down-stream tasks such as novel view synthesis and camera relocalization.", "date": "2025-10-14", "categories": []}
{"id": "2410.05111", "title": "LiDAR-GS:Real-time LiDAR Re-Simulation using Gaussian Splatting", "abstract": "We present LiDAR-GS, a Gaussian Splatting (GS) method for real-time, high-fidelity re-simulation of LiDAR scans in public urban road scenes. Recent GS methods proposed for cameras have achieved significant advancements in real-time rendering beyond Neural Radiance Fields (NeRF). However, applying GS representation to LiDAR, an active 3D sensor type, poses several challenges that must be addressed to preserve high accuracy and unique characteristics. Specifically, LiDAR-GS designs a differentiable laser beam splatting, using range-view representation for precise surface splatting by projecting lasers onto micro cross-sections, effectively eliminating artifacts associated with local affine approximations. Furthermore, LiDAR-GS leverages Neural Gaussian Representation, which further integrate view-dependent clues, to represent key LiDAR properties that are influenced by the incident direction and external factors. Combining these practices with some essential adaptations, e.g., dynamic instances decomposition, LiDAR-GS succeeds in simultaneously re-simulating depth, intensity, and ray-drop channels, achieving state-of-the-art results in both rendering frame rate and quality on publically available large scene datasets when compared with the methods using explicit mesh or implicit NeRF. Our source code is publicly available at https://www.github.com/cqf7419/LiDAR-GS.", "date": "2024-10-07", "categories": []}
{"id": "2501.06927", "title": "CULTURE3D: A Large-Scale and Diverse Dataset of Cultural Landmarks and Terrains for Gaussian-Based Scene Rendering", "abstract": "Current state-of-the-art 3D reconstruction models face limitations in building extra-large scale outdoor scenes, primarily due to the lack of sufficiently large-scale and detailed datasets. In this paper, we present a extra-large fine-grained dataset with 10 billion points composed of 41,006 drone-captured high-resolution aerial images, covering 20 diverse and culturally significant scenes from worldwide locations such as Cambridge Uni main buildings, the Pyramids, and the Forbidden City Palace. Compared to existing datasets, ours offers significantly larger scale and higher detail, uniquely suited for fine-grained 3D applications. Each scene contains an accurate spatial layout and comprehensive structural information, supporting detailed 3D reconstruction tasks. By reconstructing environments using these detailed images, our dataset supports multiple applications, including outputs in the widely adopted COLMAP format, establishing a novel benchmark for evaluating state-of-the-art large-scale Gaussian Splatting methods.The dataset's flexibility encourages innovations and supports model plug-ins, paving the way for future 3D breakthroughs. All datasets and code will be open-sourced for community use.", "date": "2025-01-12", "categories": []}
{"id": "2510.10257", "title": "Opacity-Gradient Driven Density Control for Compact and Efficient Few-Shot 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) struggles in few-shot scenarios, where its standard adaptive density control (ADC) can lead to overfitting and bloated reconstructions. While state-of-the-art methods like FSGS improve quality, they often do so by significantly increasing the primitive count. This paper presents a framework that revises the core 3DGS optimization to prioritize efficiency. We replace the standard positional gradient heuristic with a novel densification trigger that uses the opacity gradient as a lightweight proxy for rendering error. We find this aggressive densification is only effective when paired with a more conservative pruning schedule, which prevents destructive optimization cycles. Combined with a standard depth-correlation loss for geometric guidance, our framework demonstrates a fundamental improvement in efficiency. On the 3-view LLFF dataset, our model is over 40% more compact (32k vs. 57k primitives) than FSGS, and on the Mip-NeRF 360 dataset, it achieves a reduction of approximately 70%. This dramatic gain in compactness is achieved with a modest trade-off in reconstruction metrics, establishing a new state-of-the-art on the quality-vs-efficiency Pareto frontier for few-shot view synthesis.", "date": "2025-10-11", "categories": []}
{"id": "2510.10726", "title": "WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting", "abstract": "We present WorldMirror, an all-in-one, feed-forward model for versatile 3D geometric prediction tasks. Unlike existing methods constrained to image-only inputs or customized for a specific task, our framework flexibly integrates diverse geometric priors, including camera poses, intrinsics, and depth maps, while simultaneously generating multiple 3D representations: dense point clouds, multi-view depth maps, camera parameters, surface normals, and 3D Gaussians. This elegant and unified architecture leverages available prior information to resolve structural ambiguities and delivers geometrically consistent 3D outputs in a single forward pass. WorldMirror achieves state-of-the-art performance across diverse benchmarks from camera, point map, depth, and surface normal estimation to novel view synthesis, while maintaining the efficiency of feed-forward inference. Code and models will be publicly available soon.", "date": "2025-10-12", "categories": []}
{"id": "2510.11649", "title": "PhySIC: Physically Plausible 3D Human-Scene Interaction and Contact from a Single Image", "abstract": "Reconstructing metrically accurate humans and their surrounding scenes from a single image is crucial for virtual reality, robotics, and comprehensive 3D scene understanding. However, existing methods struggle with depth ambiguity, occlusions, and physically inconsistent contacts. To address these challenges, we introduce PhySIC, a framework for physically plausible Human-Scene Interaction and Contact reconstruction. PhySIC recovers metrically consistent SMPL-X human meshes, dense scene surfaces, and vertex-level contact maps within a shared coordinate frame from a single RGB image. Starting from coarse monocular depth and body estimates, PhySIC performs occlusion-aware inpainting, fuses visible depth with unscaled geometry for a robust metric scaffold, and synthesizes missing support surfaces like floors. A confidence-weighted optimization refines body pose, camera parameters, and global scale by jointly enforcing depth alignment, contact priors, interpenetration avoidance, and 2D reprojection consistency. Explicit occlusion masking safeguards invisible regions against implausible configurations. PhySIC is efficient, requiring only 9 seconds for joint human-scene optimization and under 27 seconds end-to-end. It naturally handles multiple humans, enabling reconstruction of diverse interactions. Empirically, PhySIC outperforms single-image baselines, reducing mean per-vertex scene error from 641 mm to 227 mm, halving PA-MPJPE to 42 mm, and improving contact F1 from 0.09 to 0.51. Qualitative results show realistic foot-floor interactions, natural seating, and plausible reconstructions of heavily occluded furniture. By converting a single image into a physically plausible 3D human-scene pair, PhySIC advances scalable 3D scene understanding. Our implementation is publicly available at https://yuxuan-xue.com/physic.", "date": "2025-10-13", "categories": []}
{"id": "2411.15966", "title": "Gaussian Scenes: Pose-Free Sparse-View Scene Reconstruction using Depth-Enhanced Diffusion Priors", "abstract": "In this work, we introduce a generative approach for pose-free (without camera parameters) reconstruction of 360 scenes from a sparse set of 2D images. Pose-free scene reconstruction from incomplete, pose-free observations is usually regularized with depth estimation or 3D foundational priors. While recent advances have enabled sparse-view reconstruction of large complex scenes (with high degree of foreground and background detail) with known camera poses using view-conditioned generative priors, these methods cannot be directly adapted for the pose-free setting when ground-truth poses are not available during evaluation. To address this, we propose an image-to-image generative model designed to inpaint missing details and remove artifacts in novel view renders and depth maps of a 3D scene. We introduce context and geometry conditioning using Feature-wise Linear Modulation (FiLM) modulation layers as a lightweight alternative to cross-attention and also propose a novel confidence measure for 3D Gaussian splat representations to allow for better detection of these artifacts. By progressively integrating these novel views in a Gaussian-SLAM-inspired process, we achieve a multi-view-consistent 3D representation. Evaluations on the MipNeRF360 and DL3DV-10K benchmark dataset demonstrate that our method surpasses existing pose-free techniques and performs competitively with state-of-the-art posed (precomputed camera parameters are given) reconstruction methods in complex 360 scenes. Our project page provides additional results, videos, and code.", "date": "2024-11-24", "categories": []}
{"id": "2505.23044", "title": "SpatialSplat: Efficient Semantic 3D from Sparse Unposed Images", "abstract": "A major breakthrough in 3D reconstruction is the feedforward paradigm to generate pixel-wise 3D points or Gaussian primitives from sparse, unposed images. To further incorporate semantics while avoiding the significant memory and storage costs of high-dimensional semantic features, existing methods extend this paradigm by associating each primitive with a compressed semantic feature vector. However, these methods have two major limitations: (a) the naively compressed feature compromises expressiveness, affecting the model's ability to capture fine-grained semantics, and (b) the pixel-wise primitive prediction introduces redundancy in overlapping areas, causing unnecessary memory overhead. To this end, we introduce \\textbf{SpatialSplat}, a feedforward framework that produces redundancy-aware Gaussians and capitalizes on a dual-field semantic representation. Particularly, with the insight that primitives within the same instance exhibit high semantic consistency, we decompose the semantic representation into a coarse feature field that encodes uncompressed semantics with minimal primitives, and a fine-grained yet low-dimensional feature field that captures detailed inter-instance relationships. Moreover, we propose a selective Gaussian mechanism, which retains only essential Gaussians in the scene, effectively eliminating redundant primitives. Our proposed Spatialsplat learns accurate semantic information and detailed instances prior with more compact 3D Gaussians, making semantic 3D reconstruction more applicable. We conduct extensive experiments to evaluate our method, demonstrating a remarkable 60\\% reduction in scene representation parameters while achieving superior performance over state-of-the-art methods. The code is available at https://github.com/shengyuuu/SpatialSplat.git", "date": "2025-05-29", "categories": []}
{"id": "2510.08279", "title": "Learning Neural Exposure Fields for View Synthesis", "abstract": "Recent advances in neural scene representations have led to unprecedented quality in 3D reconstruction and view synthesis. Despite achieving high-quality results for common benchmarks with curated data, outputs often degrade for data that contain per image variations such as strong exposure changes, present, e.g., in most scenes with indoor and outdoor areas or rooms with windows. In this paper, we introduce Neural Exposure Fields (NExF), a novel technique for robustly reconstructing 3D scenes with high quality and 3D-consistent appearance from challenging real-world captures. In the core, we propose to learn a neural field predicting an optimal exposure value per 3D point, enabling us to optimize exposure along with the neural scene representation. While capture devices such as cameras select optimal exposure per image/pixel, we generalize this concept and perform optimization in 3D instead. This enables accurate view synthesis in high dynamic range scenarios, bypassing the need of post-processing steps or multi-exposure captures. Our contributions include a novel neural representation for exposure prediction, a system for joint optimization of the scene representation and the exposure field via a novel neural conditioning mechanism, and demonstrated superior performance on challenging real-world data. We find that our approach trains faster than prior works and produces state-of-the-art results on several benchmarks improving by over 55% over best-performing baselines.", "date": "2025-10-09", "categories": []}
{"id": "2502.02657", "title": "SiLVR: Scalable Lidar-Visual Radiance Field Reconstruction with Uncertainty Quantification", "abstract": "We present a neural radiance field (NeRF) based large-scale reconstruction system that fuses lidar and vision data to generate high-quality reconstructions that are geometrically accurate and capture photorealistic texture. Our system adopts the state-of-the-art NeRF representation to incorporate lidar. Adding lidar data adds strong geometric constraints on the depth and surface normals, which is particularly useful when modelling uniform texture surfaces which contain ambiguous visual reconstruction cues. A key contribution of this work is a novel method to quantify the epistemic uncertainty of the lidar-visual NeRF reconstruction by estimating the spatial variance of each point location in the radiance field given the sensor observations from the cameras and lidar. This provides a principled approach to evaluate the contribution of each sensor modality to the final reconstruction. In this way, reconstructions that are uncertain (due to e.g. uniform visual texture, limited observation viewpoints, or little lidar coverage) can be identified and removed. Our system is integrated with a real-time lidar SLAM system which is used to bootstrap a Structure-from-Motion (SfM) reconstruction procedure. It also helps to properly constrain the overall metric scale which is essential for the lidar depth loss. The refined SLAM trajectory can then be divided into submaps using Spectral Clustering to group sets of co-visible images together. This submapping approach is more suitable for visual reconstruction than distance-based partitioning. Our uncertainty estimation is particularly effective when merging submaps as their boundaries often contain artefacts due to limited observations. We demonstrate the reconstruction system using a multi-camera, lidar sensor suite in experiments involving both robot-mounted and handheld scanning. Our test datasets cover a total area of more than 20,000 square metres.", "date": "2025-02-04", "categories": []}
{"id": "2503.15742", "title": "Uncertainty-Aware Diffusion Guided Refinement of 3D Scenes", "abstract": "Reconstructing 3D scenes from a single image is a fundamentally ill-posed task due to the severely under-constrained nature of the problem. Consequently, when the scene is rendered from novel camera views, existing single image to 3D reconstruction methods render incoherent and blurry views. This problem is exacerbated when the unseen regions are far away from the input camera. In this work, we address these inherent limitations in existing single image-to-3D scene feedforward networks. To alleviate the poor performance due to insufficient information beyond the input image's view, we leverage a strong generative prior in the form of a pre-trained latent video diffusion model, for iterative refinement of a coarse scene represented by optimizable Gaussian parameters. To ensure that the style and texture of the generated images align with that of the input image, we incorporate on-the-fly Fourier-style transfer between the generated images and the input image. Additionally, we design a semantic uncertainty quantification module that calculates the per-pixel entropy and yields uncertainty maps used to guide the refinement process from the most confident pixels while discarding the remaining highly uncertain ones. We conduct extensive experiments on real-world scene datasets, including in-domain RealEstate-10K and out-of-domain KITTI-v2, showing that our approach can provide more realistic and high-fidelity novel view synthesis results compared to existing state-of-the-art methods.", "date": "2025-03-19", "categories": []}
{"id": "2510.07839", "title": "AlignGS: Aligning Geometry and Semantics for Robust Indoor Reconstruction from Sparse Views", "abstract": "The demand for semantically rich 3D models of indoor scenes is rapidly growing, driven by applications in augmented reality, virtual reality, and robotics. However, creating them from sparse views remains a challenge due to geometric ambiguity. Existing methods often treat semantics as a passive feature painted on an already-formed, and potentially flawed, geometry. We posit that for robust sparse-view reconstruction, semantic understanding instead be an active, guiding force. This paper introduces AlignGS, a novel framework that actualizes this vision by pioneering a synergistic, end-to-end optimization of geometry and semantics. Our method distills rich priors from 2D foundation models and uses them to directly regularize the 3D representation through a set of novel semantic-to-geometry guidance mechanisms, including depth consistency and multi-faceted normal regularization. Extensive evaluations on standard benchmarks demonstrate that our approach achieves state-of-the-art results in novel view synthesis and produces reconstructions with superior geometric accuracy. The results validate that leveraging semantic priors as a geometric regularizer leads to more coherent and complete 3D models from limited input views. Our code is avaliable at https://github.com/MediaX-SJTU/AlignGS .", "date": "2025-10-09", "categories": []}
{"id": "2510.08551", "title": "ARTDECO: Towards Efficient and High-Fidelity On-the-Fly 3D Reconstruction with Structured Scene Representation", "abstract": "On-the-fly 3D reconstruction from monocular image sequences is a long-standing challenge in computer vision, critical for applications such as real-to-sim, AR/VR, and robotics. Existing methods face a major tradeoff: per-scene optimization yields high fidelity but is computationally expensive, whereas feed-forward foundation models enable real-time inference but struggle with accuracy and robustness. In this work, we propose ARTDECO, a unified framework that combines the efficiency of feed-forward models with the reliability of SLAM-based pipelines. ARTDECO uses 3D foundation models for pose estimation and point prediction, coupled with a Gaussian decoder that transforms multi-scale features into structured 3D Gaussians. To sustain both fidelity and efficiency at scale, we design a hierarchical Gaussian representation with a LoD-aware rendering strategy, which improves rendering fidelity while reducing redundancy. Experiments on eight diverse indoor and outdoor benchmarks show that ARTDECO delivers interactive performance comparable to SLAM, robustness similar to feed-forward systems, and reconstruction quality close to per-scene optimization, providing a practical path toward on-the-fly digitization of real-world environments with both accurate geometry and high visual fidelity. Explore more demos on our project page: https://city-super.github.io/artdeco/.", "date": "2025-10-09", "categories": []}
{"id": "2510.08566", "title": "D$^2$GS: Depth-and-Density Guided Gaussian Splatting for Stable and Accurate Sparse-View Reconstruction", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) enable real-time, high-fidelity novel view synthesis (NVS) with explicit 3D representations. However, performance degradation and instability remain significant under sparse-view conditions. In this work, we identify two key failure modes under sparse-view conditions: overfitting in regions with excessive Gaussian density near the camera, and underfitting in distant areas with insufficient Gaussian coverage. To address these challenges, we propose a unified framework D$^2$GS, comprising two key components: a Depth-and-Density Guided Dropout strategy that suppresses overfitting by adaptively masking redundant Gaussians based on density and depth, and a Distance-Aware Fidelity Enhancement module that improves reconstruction quality in under-fitted far-field areas through targeted supervision. Moreover, we introduce a new evaluation metric to quantify the stability of learned Gaussian distributions, providing insights into the robustness of the sparse-view 3DGS. Extensive experiments on multiple datasets demonstrate that our method significantly improves both visual quality and robustness under sparse view conditions. The project page can be found at: https://insta360-research-team.github.io/DDGS-website/.", "date": "2025-10-09", "categories": []}
{"id": "2506.03538", "title": "Robust Neural Rendering in the Wild with Asymmetric Dual 3D Gaussian Splatting", "abstract": "3D reconstruction from in-the-wild images remains a challenging task due to inconsistent lighting conditions and transient distractors. Existing methods typically rely on heuristic strategies to handle the low-quality training data, which often struggle to produce stable and consistent reconstructions, frequently resulting in visual artifacts. In this work, we propose \\modelname{}, a novel framework that leverages the stochastic nature of these artifacts: they tend to vary across different training runs due to minor randomness. Specifically, our method trains two 3D Gaussian Splatting (3DGS) models in parallel, enforcing a consistency constraint that encourages convergence on reliable scene geometry while suppressing inconsistent artifacts. To prevent the two models from collapsing into similar failure modes due to confirmation bias, we introduce a divergent masking strategy that applies two complementary masks: a multi-cue adaptive mask and a self-supervised soft mask, which leads to an asymmetric training process of the two models, reducing shared error modes. In addition, to improve the efficiency of model training, we introduce a lightweight variant called Dynamic EMA Proxy, which replaces one of the two models with a dynamically updated Exponential Moving Average (EMA) proxy, and employs an alternating masking strategy to preserve divergence. Extensive experiments on challenging real-world datasets demonstrate that our method consistently outperforms existing approaches while achieving high efficiency. See the project website at https://steveli88.github.io/AsymGS.", "date": "2025-06-04", "categories": []}
{"id": "2507.15690", "title": "DWTGS: Rethinking Frequency Regularization for Sparse-view 3D Gaussian Splatting", "abstract": "Sparse-view 3D Gaussian Splatting (3DGS) presents significant challenges in reconstructing high-quality novel views, as it often overfits to the widely-varying high-frequency (HF) details of the sparse training views. While frequency regularization can be a promising approach, its typical reliance on Fourier transforms causes difficult parameter tuning and biases towards detrimental HF learning. We propose DWTGS, a framework that rethinks frequency regularization by leveraging wavelet-space losses that provide additional spatial supervision. Specifically, we supervise only the low-frequency (LF) LL subbands at multiple DWT levels, while enforcing sparsity on the HF HH subband in a self-supervised manner. Experiments across benchmarks show that DWTGS consistently outperforms Fourier-based counterparts, as this LF-centric strategy improves generalization and reduces HF hallucinations.", "date": "2025-07-21", "categories": []}
{"id": "2509.24893", "title": "HBSplat: Robust Sparse-View Gaussian Reconstruction with Hybrid-Loss Guided Depth and Bidirectional Warping", "abstract": "Novel View Synthesis (NVS) from sparse views presents a formidable challenge in 3D reconstruction, where limited multi-view constraints lead to severe overfitting, geometric distortion, and fragmented scenes. While 3D Gaussian Splatting (3DGS) delivers real-time, high-fidelity rendering, its performance drastically deteriorates under sparse inputs, plagued by floating artifacts and structural failures. To address these challenges, we introduce HBSplat, a unified framework that elevates 3DGS by seamlessly integrating robust structural cues, virtual view constraints, and occluded region completion. Our core contributions are threefold: a Hybrid-Loss Depth Estimation module that ensures multi-view consistency by leveraging dense matching priors and integrating reprojection, point propagation, and smoothness constraints; a Bidirectional Warping Virtual View Synthesis method that enforces substantially stronger constraints by creating high-fidelity virtual views through bidirectional depth-image warping and multi-view fusion; and an Occlusion-Aware Reconstruction component that recovers occluded areas using a depth-difference mask and a learning-based inpainting model. Extensive evaluations on LLFF, Blender, and DTU benchmarks validate that HBSplat sets a new state-of-the-art, achieving up to 21.13 dB PSNR and 0.189 LPIPS, while maintaining real-time inference. Code is available at: https://github.com/eternalland/HBSplat.", "date": "2025-09-29", "categories": []}
{"id": "2506.22118", "title": "Pipe Reconstruction from Point Cloud Data", "abstract": "Accurate digital twins of industrial assets, such as ships and offshore platforms, rely on the precise reconstruction of complex pipe networks. However, manual modelling of pipes from laser scan data is a time-consuming and labor-intensive process. This paper presents a pipeline for automated pipe reconstruction from incomplete laser scan data. The approach estimates a skeleton curve using Laplacian-based contraction, followed by curve elongation. The skeleton axis is then recentred using a rolling sphere technique combined with 2D circle fitting, and refined with a 3D smoothing step. This enables the determination of pipe properties, including radius, length and orientation, and facilitates the creation of detailed 3D models of complex pipe networks. By automating pipe reconstruction, this approach supports the development of digital twins, allowing for rapid and accurate modeling while reducing costs.", "date": "2025-06-27", "categories": []}
{"id": "2510.05560", "title": "HoloScene: Simulation-Ready Interactive 3D Worlds from a Single Video", "abstract": "Digitizing the physical world into accurate simulation-ready virtual environments offers significant opportunities in a variety of fields such as augmented and virtual reality, gaming, and robotics. However, current 3D reconstruction and scene-understanding methods commonly fall short in one or more critical aspects, such as geometry completeness, object interactivity, physical plausibility, photorealistic rendering, or realistic physical properties for reliable dynamic simulation. To address these limitations, we introduce HoloScene, a novel interactive 3D reconstruction framework that simultaneously achieves these requirements. HoloScene leverages a comprehensive interactive scene-graph representation, encoding object geometry, appearance, and physical properties alongside hierarchical and inter-object relationships. Reconstruction is formulated as an energy-based optimization problem, integrating observational data, physical constraints, and generative priors into a unified, coherent objective. Optimization is efficiently performed via a hybrid approach combining sampling-based exploration with gradient-based refinement. The resulting digital twins exhibit complete and precise geometry, physical stability, and realistic rendering from novel viewpoints. Evaluations conducted on multiple benchmark datasets demonstrate superior performance, while practical use-cases in interactive gaming and real-time digital-twin manipulation illustrate HoloScene's broad applicability and effectiveness. Project page: https://xiahongchi.github.io/HoloScene.", "date": "2025-10-07", "categories": []}
{"id": "2404.05606", "title": "Reconstructing Topology-Consistent Face Mesh by Volume Rendering from Multi-View Images", "abstract": "Industrial 3D face assets creation typically reconstructs topology-consistent face meshes from multi-view images for downstream production. However, high-quality reconstruction usually requires manual processing or specific capture settings. Recently NeRF has shown great advantages in 3D reconstruction, by representing scenes as density and radiance fields and utilizing neural volume rendering for novel view synthesis. Inspired by this, we introduce a novel method which combines explicit mesh with neural volume rendering to optimize geometry of an artist-made template face mesh from multi-view images while keeping the topology unchanged. Our method derives density fields from meshes using distance fields as an intermediary and encodes radiance field in compact tri-planes. To improve convergence, several adaptions tailored for meshes are introduced to the volume rendering. Experiments demonstrate that our method achieves superior reconstruction quality compared to previous approaches, validating the feasibility of integrating mesh and neural volume rendering.", "date": "2024-04-08", "categories": []}
{"id": "2411.11941", "title": "TimeFormer: Capturing Temporal Relationships of Deformable 3D Gaussians for Robust Reconstruction", "abstract": "Dynamic scene reconstruction is a long-term challenge in 3D vision. Recent methods extend 3D Gaussian Splatting to dynamic scenes via additional deformation fields and apply explicit constraints like motion flow to guide the deformation. However, they learn motion changes from individual timestamps independently, making it challenging to reconstruct complex scenes, particularly when dealing with violent movement, extreme-shaped geometries, or reflective surfaces. To address the above issue, we design a plug-and-play module called TimeFormer to enable existing deformable 3D Gaussians reconstruction methods with the ability to implicitly model motion patterns from a learning perspective. Specifically, TimeFormer includes a Cross-Temporal Transformer Encoder, which adaptively learns the temporal relationships of deformable 3D Gaussians. Furthermore, we propose a two-stream optimization strategy that transfers the motion knowledge learned from TimeFormer to the base stream during the training phase. This allows us to remove TimeFormer during inference, thereby preserving the original rendering speed. Extensive experiments in the multi-view and monocular dynamic scenes validate qualitative and quantitative improvement brought by TimeFormer. Project Page: https://patrickddj.github.io/TimeFormer/", "date": "2024-11-18", "categories": []}
{"id": "2411.18625", "title": "Textured Gaussians for Enhanced 3D Scene Appearance Modeling", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as a state-of-the-art 3D reconstruction and rendering technique due to its high-quality results and fast training and rendering time. However, pixels covered by the same Gaussian are always shaded in the same color up to a Gaussian falloff scaling factor. Furthermore, the finest geometric detail any individual Gaussian can represent is a simple ellipsoid. These properties of 3DGS greatly limit the expressivity of individual Gaussian primitives. To address these issues, we draw inspiration from texture and alpha mapping in traditional graphics and integrate it with 3DGS. Specifically, we propose a new generalized Gaussian appearance representation that augments each Gaussian with alpha~(A), RGB, or RGBA texture maps to model spatially varying color and opacity across the extent of each Gaussian. As such, each Gaussian can represent a richer set of texture patterns and geometric structures, instead of just a single color and ellipsoid as in naive Gaussian Splatting. Surprisingly, we found that the expressivity of Gaussians can be greatly improved by using alpha-only texture maps, and further augmenting Gaussians with RGB texture maps achieves the highest expressivity. We validate our method on a wide variety of standard benchmark datasets and our own custom captures at both the object and scene levels. We demonstrate image quality improvements over existing methods while using a similar or lower number of Gaussians.", "date": "2024-11-27", "categories": []}
{"id": "2503.09040", "title": "Motion Blender Gaussian Splatting for Dynamic Scene Reconstruction", "abstract": "Gaussian splatting has emerged as a powerful tool for high-fidelity reconstruction of dynamic scenes. However, existing methods primarily rely on implicit motion representations, such as encoding motions into neural networks or per-Gaussian parameters, which makes it difficult to further manipulate the reconstructed motions. This lack of explicit controllability limits existing methods to replaying recorded motions only, which hinders a wider application in robotics. To address this, we propose Motion Blender Gaussian Splatting (MBGS), a novel framework that uses motion graphs as an explicit and sparse motion representation. The motion of a graph's links is propagated to individual Gaussians via dual quaternion skinning, with learnable weight painting functions that determine the influence of each link. The motion graphs and 3D Gaussians are jointly optimized from input videos via differentiable rendering. Experiments show that MBGS achieves state-of-the-art performance on the highly challenging iPhone dataset while being competitive on HyperNeRF. We demonstrate the application potential of our method in animating novel object poses, synthesizing real robot demonstrations, and predicting robot actions through visual planning. The source code, models, video demonstrations can be found at http://mlzxy.github.io/motion-blender-gs.", "date": "2025-03-12", "categories": []}
{"id": "2510.03857", "title": "Optimized Minimal 4D Gaussian Splatting", "abstract": "4D Gaussian Splatting has emerged as a new paradigm for dynamic scene representation, enabling real-time rendering of scenes with complex motions. However, it faces a major challenge of storage overhead, as millions of Gaussians are required for high-fidelity reconstruction. While several studies have attempted to alleviate this memory burden, they still face limitations in compression ratio or visual quality. In this work, we present OMG4 (Optimized Minimal 4D Gaussian Splatting), a framework that constructs a compact set of salient Gaussians capable of faithfully representing 4D Gaussian models. Our method progressively prunes Gaussians in three stages: (1) Gaussian Sampling to identify primitives critical to reconstruction fidelity, (2) Gaussian Pruning to remove redundancies, and (3) Gaussian Merging to fuse primitives with similar characteristics. In addition, we integrate implicit appearance compression and generalize Sub-Vector Quantization (SVQ) to 4D representations, further reducing storage while preserving quality. Extensive experiments on standard benchmark datasets demonstrate that OMG4 significantly outperforms recent state-of-the-art methods, reducing model sizes by over 60% while maintaining reconstruction quality. These results position OMG4 as a significant step forward in compact 4D scene representation, opening new possibilities for a wide range of applications. Our source code is available at https://minshirley.github.io/OMG4/.", "date": "2025-10-04", "categories": []}
{"id": "2510.01669", "title": "UniVerse: Unleashing the Scene Prior of Video Diffusion Models for Robust Radiance Field Reconstruction", "abstract": "This paper tackles the challenge of robust reconstruction, i.e., the task of reconstructing a 3D scene from a set of inconsistent multi-view images. Some recent works have attempted to simultaneously remove image inconsistencies and perform reconstruction by integrating image degradation modeling into neural 3D scene representations. However, these methods rely heavily on dense observations for robustly optimizing model parameters. To address this issue, we propose to decouple robust reconstruction into two subtasks: restoration and reconstruction, which naturally simplifies the optimization process. To this end, we introduce UniVerse, a unified framework for robust reconstruction based on a video diffusion model. Specifically, UniVerse first converts inconsistent images into initial videos, then uses a specially designed video diffusion model to restore them into consistent images, and finally reconstructs the 3D scenes from these restored images. Compared with case-by-case per-view degradation modeling, the diffusion model learns a general scene prior from large-scale data, making it applicable to diverse image inconsistencies. Extensive experiments on both synthetic and real-world datasets demonstrate the strong generalization capability and superior performance of our method in robust reconstruction. Moreover, UniVerse can control the style of the reconstructed 3D scene. Project page: https://jin-cao-tma.github.io/UniVerse.github.io/", "date": "2025-10-02", "categories": []}
{"id": "2510.02732", "title": "From Tokens to Nodes: Semantic-Guided Motion Control for Dynamic 3D Gaussian Splatting", "abstract": "Dynamic 3D reconstruction from monocular videos remains difficult due to the ambiguity inferring 3D motion from limited views and computational demands of modeling temporally varying scenes. While recent sparse control methods alleviate computation by reducing millions of Gaussians to thousands of control points, they suffer from a critical limitation: they allocate points purely by geometry, leading to static redundancy and dynamic insufficiency. We propose a motion-adaptive framework that aligns control density with motion complexity. Leveraging semantic and motion priors from vision foundation models, we establish patch-token-node correspondences and apply motion-adaptive compression to concentrate control points in dynamic regions while suppressing redundancy in static backgrounds. Our approach achieves flexible representational density adaptation through iterative voxelization and motion tendency scoring, directly addressing the fundamental mismatch between control point allocation and motion complexity. To capture temporal evolution, we introduce spline-based trajectory parameterization initialized by 2D tracklets, replacing MLP-based deformation fields to achieve smoother motion representation and more stable optimization. Extensive experiments demonstrate significant improvements in reconstruction quality and efficiency over existing state-of-the-art methods.", "date": "2025-10-03", "categories": []}
{"id": "2510.01119", "title": "Instant4D: 4D Gaussian Splatting in Minutes", "abstract": "Dynamic view synthesis has seen significant advances, yet reconstructing scenes from uncalibrated, casual video remains challenging due to slow optimization and complex parameter estimation. In this work, we present Instant4D, a monocular reconstruction system that leverages native 4D representation to efficiently process casual video sequences within minutes, without calibrated cameras or depth sensors. Our method begins with geometric recovery through deep visual SLAM, followed by grid pruning to optimize scene representation. Our design significantly reduces redundancy while maintaining geometric integrity, cutting model size to under 10% of its original footprint. To handle temporal dynamics efficiently, we introduce a streamlined 4D Gaussian representation, achieving a 30x speed-up and reducing training time to within two minutes, while maintaining competitive performance across several benchmarks. Our method reconstruct a single video within 10 minutes on the Dycheck dataset or for a typical 200-frame video. We further apply our model to in-the-wild videos, showcasing its generalizability. Our project website is published at https://instant4d.github.io/.", "date": "2025-10-01", "categories": []}
{"id": "2502.12545", "title": "IM360: Large-scale Indoor Mapping with 360 Cameras", "abstract": "We present a novel 3D mapping pipeline for large-scale indoor environments. To address the significant challenges in large-scale indoor scenes, such as prevalent occlusions and textureless regions, we propose IM360, a novel approach that leverages the wide field of view of omnidirectional images and integrates the spherical camera model into the Structure-from-Motion (SfM) pipeline. Our SfM utilizes dense matching features specifically designed for 360 images, demonstrating superior capability in image registration. Furthermore, with the aid of mesh-based neural rendering techniques, we introduce a texture optimization method that refines texture maps and accurately captures view-dependent properties by combining diffuse and specular components. We evaluate our pipeline on large-scale indoor scenes, demonstrating its effectiveness in real-world scenarios. In practice, IM360 demonstrates superior performance, achieving a 3.5 PSNR increase in textured mesh reconstruction. We attain state-of-the-art performance in terms of camera localization and registration on Matterport3D and Stanford2D3D.", "date": "2025-02-18", "categories": []}
{"id": "2508.11379", "title": "G-CUT3R: Guided 3D Reconstruction with Camera and Depth Prior Integration", "abstract": "We introduce G-CUT3R, a novel feed-forward approach for guided 3D scene reconstruction that enhances the CUT3R model by integrating prior information. Unlike existing feed-forward methods that rely solely on input images, our method leverages auxiliary data, such as depth, camera calibrations, or camera positions, commonly available in real-world scenarios. We propose a lightweight modification to CUT3R, incorporating a dedicated encoder for each modality to extract features, which are fused with RGB image tokens via zero convolution. This flexible design enables seamless integration of any combination of prior information during inference. Evaluated across multiple benchmarks, including 3D reconstruction and other multi-view tasks, our approach demonstrates significant performance improvements, showing its ability to effectively utilize available priors while maintaining compatibility with varying input modalities.", "date": "2025-08-15", "categories": []}
{"id": "2509.23492", "title": "Orientation-anchored Hyper-Gaussian for 4D Reconstruction from Casual Videos", "abstract": "We present Orientation-anchored Gaussian Splatting (OriGS), a novel framework for high-quality 4D reconstruction from casually captured monocular videos. While recent advances extend 3D Gaussian Splatting to dynamic scenes via various motion anchors, such as graph nodes or spline control points, they often rely on low-rank assumptions and fall short in modeling complex, region-specific deformations inherent to unconstrained dynamics. OriGS addresses this by introducing a hyperdimensional representation grounded in scene orientation. We first estimate a Global Orientation Field that propagates principal forward directions across space and time, serving as stable structural guidance for dynamic modeling. Built upon this, we propose Orientation-aware Hyper-Gaussian, a unified formulation that embeds time, space, geometry, and orientation into a coherent probabilistic state. This enables inferring region-specific deformation through principled conditioned slicing, adaptively capturing diverse local dynamics in alignment with global motion intent. Experiments demonstrate the superior reconstruction fidelity of OriGS over mainstream methods in challenging real-world dynamic scenes.", "date": "2025-09-27", "categories": []}
{"id": "2509.24308", "title": "OMeGa: Joint Optimization of Explicit Meshes and Gaussian Splats for Robust Scene-Level Surface Reconstruction", "abstract": "Neural rendering with Gaussian splatting has advanced novel view synthesis, and most methods reconstruct surfaces via post-hoc mesh extraction. However, existing methods suffer from two limitations: (i) inaccurate geometry in texture-less indoor regions, and (ii) the decoupling of mesh extraction from optimization, thereby missing the opportunity to leverage mesh geometry to guide splat optimization. In this paper, we present OMeGa, an end-to-end framework that jointly optimizes an explicit triangle mesh and 2D Gaussian splats via a flexible binding strategy, where spatial attributes of Gaussian Splats are expressed in the mesh frame and texture attributes are retained on splats. To further improve reconstruction accuracy, we integrate mesh constraints and monocular normal supervision into the optimization, thereby regularizing geometry learning. In addition, we propose a heuristic, iterative mesh-refinement strategy that splits high-error faces and prunes unreliable ones to further improve the detail and accuracy of the reconstructed mesh. OMeGa achieves state-of-the-art performance on challenging indoor reconstruction benchmarks, reducing Chamfer-$L_1$ by 47.3\\% over the 2DGS baseline while maintaining competitive novel-view rendering quality. The experimental results demonstrate that OMeGa effectively addresses prior limitations in indoor texture-less reconstruction.", "date": "2025-09-29", "categories": []}
{"id": "2509.25001", "title": "LVT: Large-Scale Scene Reconstruction via Local View Transformers", "abstract": "Large transformer models are proving to be a powerful tool for 3D vision and novel view synthesis. However, the standard Transformer's well-known quadratic complexity makes it difficult to scale these methods to large scenes. To address this challenge, we propose the Local View Transformer (LVT), a large-scale scene reconstruction and novel view synthesis architecture that circumvents the need for the quadratic attention operation. Motivated by the insight that spatially nearby views provide more useful signal about the local scene composition than distant views, our model processes all information in a local neighborhood around each view. To attend to tokens in nearby views, we leverage a novel positional encoding that conditions on the relative geometric transformation between the query and nearby views. We decode the output of our model into a 3D Gaussian Splat scene representation that includes both color and opacity view-dependence. Taken together, the Local View Transformer enables reconstruction of arbitrarily large, high-resolution scenes in a single forward pass. See our project page for results and interactive demos https://toobaimt.github.io/lvt/.", "date": "2025-09-29", "categories": []}
{"id": "2509.25122", "title": "Triangle Splatting+: Differentiable Rendering with Opaque Triangles", "abstract": "Reconstructing 3D scenes and synthesizing novel views has seen rapid progress in recent years. Neural Radiance Fields demonstrated that continuous volumetric radiance fields can achieve high-quality image synthesis, but their long training and rendering times limit practicality. 3D Gaussian Splatting (3DGS) addressed these issues by representing scenes with millions of Gaussians, enabling real-time rendering and fast optimization. However, Gaussian primitives are not natively compatible with the mesh-based pipelines used in VR headsets, and real-time graphics applications. Existing solutions attempt to convert Gaussians into meshes through post-processing or two-stage pipelines, which increases complexity and degrades visual quality. In this work, we introduce Triangle Splatting+, which directly optimizes triangles, the fundamental primitive of computer graphics, within a differentiable splatting framework. We formulate triangle parametrization to enable connectivity through shared vertices, and we design a training strategy that enforces opaque triangles. The final output is immediately usable in standard graphics engines without post-processing. Experiments on the Mip-NeRF360 and Tanks & Temples datasets show that Triangle Splatting+achieves state-of-the-art performance in mesh-based novel view synthesis. Our method surpasses prior splatting approaches in visual fidelity while remaining efficient and fast to training. Moreover, the resulting semi-connected meshes support downstream applications such as physics-based simulation or interactive walkthroughs. The project page is https://trianglesplatting2.github.io/trianglesplatting2/.", "date": "2025-09-29", "categories": []}
{"id": "2509.25183", "title": "PAD3R: Pose-Aware Dynamic 3D Reconstruction from Casual Videos", "abstract": "We present PAD3R, a method for reconstructing deformable 3D objects from casually captured, unposed monocular videos. Unlike existing approaches, PAD3R handles long video sequences featuring substantial object deformation, large-scale camera movement, and limited view coverage that typically challenge conventional systems. At its core, our approach trains a personalized, object-centric pose estimator, supervised by a pre-trained image-to-3D model. This guides the optimization of deformable 3D Gaussian representation. The optimization is further regularized by long-term 2D point tracking over the entire input video. By combining generative priors and differentiable rendering, PAD3R reconstructs high-fidelity, articulated 3D representations of objects in a category-agnostic way. Extensive qualitative and quantitative results show that PAD3R is robust and generalizes well across challenging scenarios, highlighting its potential for dynamic scene understanding and 3D content creation.", "date": "2025-09-29", "categories": []}
{"id": "2503.01199", "title": "LiteGS: A High-performance Framework to Train 3DGS in Subminutes via System and Algorithm Codesign", "abstract": "3D Gaussian Splatting (3DGS) has emerged as promising alternative in 3D representation. However, it still suffers from high training cost. This paper introduces LiteGS, a high performance framework that systematically optimizes the 3DGS training pipeline from multiple aspects. At the low-level computation layer, we design a ``warp-based raster'' associated with two hardware-aware optimizations to significantly reduce gradient reduction overhead. At the mid-level data management layer, we introduce dynamic spatial sorting based on Morton coding to enable a performant ``Cluster-Cull-Compact'' pipeline and improve data locality, therefore reducing cache misses. At the top-level algorithm layer, we establish a new robust densification criterion based on the variance of the opacity gradient, paired with a more stable opacity control mechanism, to achieve more precise parameter growth. Experimental results demonstrate that LiteGS accelerates the original 3DGS training by up to 13.4x with comparable or superior quality and surpasses the current SOTA in lightweight models by up to 1.4x speedup. For high-quality reconstruction tasks, LiteGS sets a new accuracy record and decreases the training time by an order of magnitude.", "date": "2025-03-03", "categories": []}
{"id": "2507.02705", "title": "SIU3R: Simultaneous Scene Understanding and 3D Reconstruction Beyond Feature Alignment", "abstract": "Simultaneous understanding and 3D reconstruction plays an important role in developing end-to-end embodied intelligent systems. To achieve this, recent approaches resort to 2D-to-3D feature alignment paradigm, which leads to limited 3D understanding capability and potential semantic information loss. In light of this, we propose SIU3R, the first alignment-free framework for generalizable simultaneous understanding and 3D reconstruction from unposed images. Specifically, SIU3R bridges reconstruction and understanding tasks via pixel-aligned 3D representation, and unifies multiple understanding (segmentation) tasks into a set of unified learnable queries, enabling native 3D understanding without the need of alignment with 2D models. To encourage collaboration between the two tasks with shared representation, we further conduct in-depth analyses of their mutual benefits, and propose two lightweight modules to facilitate their interaction. Extensive experiments demonstrate that our method achieves state-of-the-art performance not only on the individual tasks of 3D reconstruction and understanding, but also on the task of simultaneous understanding and 3D reconstruction, highlighting the advantages of our alignment-free framework and the effectiveness of the mutual benefit designs. Project page: https://insomniaaac.github.io/siu3r/", "date": "2025-07-03", "categories": []}
{"id": "2509.19726", "title": "PolGS: Polarimetric Gaussian Splatting for Fast Reflective Surface Reconstruction", "abstract": "Efficient shape reconstruction for surfaces with complex reflectance properties is crucial for real-time virtual reality. While 3D Gaussian Splatting (3DGS)-based methods offer fast novel view rendering by leveraging their explicit surface representation, their reconstruction quality lags behind that of implicit neural representations, particularly in the case of recovering surfaces with complex reflective reflectance. To address these problems, we propose PolGS, a Polarimetric Gaussian Splatting model allowing fast reflective surface reconstruction in 10 minutes. By integrating polarimetric constraints into the 3DGS framework, PolGS effectively separates specular and diffuse components, enhancing reconstruction quality for challenging reflective materials. Experimental results on the synthetic and real-world dataset validate the effectiveness of our method.", "date": "2025-09-24", "categories": []}
{"id": "2410.03592", "title": "Variational Bayes Gaussian Splatting", "abstract": "Recently, 3D Gaussian Splatting has emerged as a promising approach for modeling 3D scenes using mixtures of Gaussians. The predominant optimization method for these models relies on backpropagating gradients through a differentiable rendering pipeline, which struggles with catastrophic forgetting when dealing with continuous streams of data. To address this limitation, we propose Variational Bayes Gaussian Splatting (VBGS), a novel approach that frames training a Gaussian splat as variational inference over model parameters. By leveraging the conjugacy properties of multivariate Gaussians, we derive a closed-form variational update rule, allowing efficient updates from partial, sequential observations without the need for replay buffers. Our experiments show that VBGS not only matches state-of-the-art performance on static datasets, but also enables continual learning from sequentially streamed 2D and 3D data, drastically improving performance in this setting.", "date": "2024-10-04", "categories": []}
{"id": "2509.16423", "title": "3D Gaussian Flats: Hybrid 2D/3D Photometric Scene Reconstruction", "abstract": "Recent advances in radiance fields and novel view synthesis enable creation of realistic digital twins from photographs. However, current methods struggle with flat, texture-less surfaces, creating uneven and semi-transparent reconstructions, due to an ill-conditioned photometric reconstruction objective. Surface reconstruction methods solve this issue but sacrifice visual quality. We propose a novel hybrid 2D/3D representation that jointly optimizes constrained planar (2D) Gaussians for modeling flat surfaces and freeform (3D) Gaussians for the rest of the scene. Our end-to-end approach dynamically detects and refines planar regions, improving both visual fidelity and geometric accuracy. It achieves state-of-the-art depth estimation on ScanNet++ and ScanNetv2, and excels at mesh extraction without overfitting to a specific camera model, showing its effectiveness in producing high-quality reconstruction of indoor scenes.", "date": "2025-09-19", "categories": []}
{"id": "2509.18566", "title": "Event-guided 3D Gaussian Splatting for Dynamic Human and Scene Reconstruction", "abstract": "Reconstructing dynamic humans together with static scenes from monocular videos remains difficult, especially under fast motion, where RGB frames suffer from motion blur. Event cameras exhibit distinct advantages, e.g., microsecond temporal resolution, making them a superior sensing choice for dynamic human reconstruction. Accordingly, we present a novel event-guided human-scene reconstruction framework that jointly models human and scene from a single monocular event camera via 3D Gaussian Splatting. Specifically, a unified set of 3D Gaussians carries a learnable semantic attribute; only Gaussians classified as human undergo deformation for animation, while scene Gaussians stay static. To combat blur, we propose an event-guided loss that matches simulated brightness changes between consecutive renderings with the event stream, improving local fidelity in fast-moving regions. Our approach removes the need for external human masks and simplifies managing separate Gaussian sets. On two benchmark datasets, ZJU-MoCap-Blur and MMHPSD-Blur, it delivers state-of-the-art human-scene reconstruction, with notable gains over strong baselines in PSNR/SSIM and reduced LPIPS, especially for high-speed subjects.", "date": "2025-09-23", "categories": []}
{"id": "2509.18759", "title": "FixingGS: Enhancing 3D Gaussian Splatting via Training-Free Score Distillation", "abstract": "Recently, 3D Gaussian Splatting (3DGS) has demonstrated remarkable success in 3D reconstruction and novel view synthesis. However, reconstructing 3D scenes from sparse viewpoints remains highly challenging due to insufficient visual information, which results in noticeable artifacts persisting across the 3D representation. To address this limitation, recent methods have resorted to generative priors to remove artifacts and complete missing content in under-constrained areas. Despite their effectiveness, these approaches struggle to ensure multi-view consistency, resulting in blurred structures and implausible details. In this work, we propose FixingGS, a training-free method that fully exploits the capabilities of the existing diffusion model for sparse-view 3DGS reconstruction enhancement. At the core of FixingGS is our distillation approach, which delivers more accurate and cross-view coherent diffusion priors, thereby enabling effective artifact removal and inpainting. In addition, we propose an adaptive progressive enhancement scheme that further refines reconstructions in under-constrained regions. Extensive experiments demonstrate that FixingGS surpasses existing state-of-the-art methods with superior visual quality and reconstruction performance. Our code will be released publicly.", "date": "2025-09-23", "categories": []}
{"id": "2509.18956", "title": "Seeing Through Reflections: Advancing 3D Scene Reconstruction in Mirror-Containing Environments with Gaussian Splatting", "abstract": "Mirror-containing environments pose unique challenges for 3D reconstruction and novel view synthesis (NVS), as reflective surfaces introduce view-dependent distortions and inconsistencies. While cutting-edge methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) excel in typical scenes, their performance deteriorates in the presence of mirrors. Existing solutions mainly focus on handling mirror surfaces through symmetry mapping but often overlook the rich information carried by mirror reflections. These reflections offer complementary perspectives that can fill in absent details and significantly enhance reconstruction quality. To advance 3D reconstruction in mirror-rich environments, we present MirrorScene3D, a comprehensive dataset featuring diverse indoor scenes, 1256 high-quality images, and annotated mirror masks, providing a benchmark for evaluating reconstruction methods in reflective settings. Building on this, we propose ReflectiveGS, an extension of 3D Gaussian Splatting that utilizes mirror reflections as complementary viewpoints rather than simple symmetry artifacts, enhancing scene geometry and recovering absent details. Experiments on MirrorScene3D show that ReflectiveGaussian outperforms existing methods in SSIM, PSNR, LPIPS, and training speed, setting a new benchmark for 3D reconstruction in mirror-rich environments.", "date": "2025-09-23", "categories": []}
{"id": "2509.19296", "title": "Lyra: Generative 3D Scene Reconstruction via Video Diffusion Model Self-Distillation", "abstract": "The ability to generate virtual environments is crucial for applications ranging from gaming to physical AI domains such as robotics, autonomous driving, and industrial AI. Current learning-based 3D reconstruction methods rely on the availability of captured real-world multi-view data, which is not always readily available. Recent advancements in video diffusion models have shown remarkable imagination capabilities, yet their 2D nature limits the applications to simulation where a robot needs to navigate and interact with the environment. In this paper, we propose a self-distillation framework that aims to distill the implicit 3D knowledge in the video diffusion models into an explicit 3D Gaussian Splatting (3DGS) representation, eliminating the need for multi-view training data. Specifically, we augment the typical RGB decoder with a 3DGS decoder, which is supervised by the output of the RGB decoder. In this approach, the 3DGS decoder can be purely trained with synthetic data generated by video diffusion models. At inference time, our model can synthesize 3D scenes from either a text prompt or a single image for real-time rendering. Our framework further extends to dynamic 3D scene generation from a monocular input video. Experimental results show that our framework achieves state-of-the-art performance in static and dynamic 3D scene generation.", "date": "2025-09-23", "categories": []}
{"id": "2412.03526", "title": "Feed-Forward Bullet-Time Reconstruction of Dynamic Scenes from Monocular Videos", "abstract": "Recent advancements in static feed-forward scene reconstruction have demonstrated significant progress in high-quality novel view synthesis. However, these models often struggle with generalizability across diverse environments and fail to effectively handle dynamic content. We present BTimer (short for BulletTimer), the first motion-aware feed-forward model for real-time reconstruction and novel view synthesis of dynamic scenes. Our approach reconstructs the full scene in a 3D Gaussian Splatting representation at a given target ('bullet') timestamp by aggregating information from all the context frames. Such a formulation allows BTimer to gain scalability and generalization by leveraging both static and dynamic scene datasets. Given a casual monocular dynamic video, BTimer reconstructs a bullet-time scene within 150ms while reaching state-of-the-art performance on both static and dynamic scene datasets, even compared with optimization-based approaches.", "date": "2024-12-04", "categories": []}
{"id": "2506.23835", "title": "SCORP: Scene-Consistent Object Refinement via Proxy Generation and Tuning", "abstract": "Viewpoint missing of objects is common in scene reconstruction, as camera paths typically prioritize capturing the overall scene structure rather than individual objects. This makes it highly challenging to achieve high-fidelity object-level modeling while maintaining accurate scene-level representation. Addressing this issue is critical for advancing downstream tasks requiring high-fidelity object reconstruction. In this paper, we introduce Scene-Consistent Object Refinement via Proxy Generation and Tuning (SCORP), a novel 3D enhancement framework that leverages 3D generative priors to recover fine-grained object geometry and appearance under missing views. Starting with proxy generation by substituting degraded objects using a 3D generation model, SCORP then progressively refines geometry and texture by aligning each proxy to its degraded counterpart in 7-DoF pose, followed by correcting spatial and appearance inconsistencies through registration-constrained enhancement. This two-stage proxy tuning ensures the high-fidelity geometry and appearance of the original object in unseen views while maintaining consistency in spatial positioning, observed geometry, and appearance. Across challenging benchmarks, SCORP achieves consistent gains over recent state-of-the-art baselines on both novel view synthesis and geometry completion tasks. SCORP is available at https://github.com/PolySummit/SCORP.", "date": "2025-06-30", "categories": []}
{"id": "2509.07493", "title": "Accurate and Complete Surface Reconstruction from 3D Gaussians via Direct SDF Learning", "abstract": "3D Gaussian Splatting (3DGS) has recently emerged as a powerful paradigm for photorealistic view synthesis, representing scenes with spatially distributed Gaussian primitives. While highly effective for rendering, achieving accurate and complete surface reconstruction remains challenging due to the unstructured nature of the representation and the absence of explicit geometric supervision. In this work, we propose DiGS, a unified framework that embeds Signed Distance Field (SDF) learning directly into the 3DGS pipeline, thereby enforcing strong and interpretable surface priors. By associating each Gaussian with a learnable SDF value, DiGS explicitly aligns primitives with underlying geometry and improves cross-view consistency. To further ensure dense and coherent coverage, we design a geometry-guided grid growth strategy that adaptively distributes Gaussians along geometry-consistent regions under a multi-scale hierarchy. Extensive experiments on standard benchmarks, including DTU, Mip-NeRF 360, and Tanks& Temples, demonstrate that DiGS consistently improves reconstruction accuracy and completeness while retaining high rendering fidelity.", "date": "2025-09-09", "categories": []}
{"id": "2509.11003", "title": "AD-GS: Alternating Densification for Sparse-Input 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has shown impressive results in real-time novel view synthesis. However, it often struggles under sparse-view settings, producing undesirable artifacts such as floaters, inaccurate geometry, and overfitting due to limited observations. We find that a key contributing factor is uncontrolled densification, where adding Gaussian primitives rapidly without guidance can harm geometry and cause artifacts. We propose AD-GS, a novel alternating densification framework that interleaves high and low densification phases. During high densification, the model densifies aggressively, followed by photometric loss based training to capture fine-grained scene details. Low densification then primarily involves aggressive opacity pruning of Gaussians followed by regularizing their geometry through pseudo-view consistency and edge-aware depth smoothness. This alternating approach helps reduce overfitting by carefully controlling model capacity growth while progressively refining the scene representation. Extensive experiments on challenging datasets demonstrate that AD-GS significantly improves rendering quality and geometric consistency compared to existing methods. The source code for our model can be found on our project page: https://gurutvapatle.github.io/publications/2025/ADGS.html .", "date": "2025-09-13", "categories": []}
{"id": "2509.16863", "title": "ConfidentSplat: Confidence-Weighted Depth Fusion for Accurate 3D Gaussian Splatting SLAM", "abstract": "We introduce ConfidentSplat, a novel 3D Gaussian Splatting (3DGS)-based SLAM system for robust, highfidelity RGB-only reconstruction. Addressing geometric inaccuracies in existing RGB-only 3DGS SLAM methods that stem from unreliable depth estimation, ConfidentSplat incorporates a core innovation: a confidence-weighted fusion mechanism. This mechanism adaptively integrates depth cues from multiview geometry with learned monocular priors (Omnidata ViT), dynamically weighting their contributions based on explicit reliability estimates-derived predominantly from multi-view geometric consistency-to generate high-fidelity proxy depth for map supervision. The resulting proxy depth guides the optimization of a deformable 3DGS map, which efficiently adapts online to maintain global consistency following pose updates from a DROID-SLAM-inspired frontend and backend optimizations (loop closure, global bundle adjustment). Extensive validation on standard benchmarks (TUM-RGBD, ScanNet) and diverse custom mobile datasets demonstrates significant improvements in reconstruction accuracy (L1 depth error) and novel view synthesis fidelity (PSNR, SSIM, LPIPS) over baselines, particularly in challenging conditions. ConfidentSplat underscores the efficacy of principled, confidence-aware sensor fusion for advancing state-of-the-art dense visual SLAM.", "date": "2025-09-21", "categories": []}
{"id": "2509.17027", "title": "Efficient 3D Scene Reconstruction and Simulation from Sparse Endoscopic Views", "abstract": "Surgical simulation is essential for medical training, enabling practitioners to develop crucial skills in a risk-free environment while improving patient safety and surgical outcomes. However, conventional methods for building simulation environments are cumbersome, time-consuming, and difficult to scale, often resulting in poor details and unrealistic simulations. In this paper, we propose a Gaussian Splatting-based framework to directly reconstruct interactive surgical scenes from endoscopic data while ensuring efficiency, rendering quality, and realism. A key challenge in this data-driven simulation paradigm is the restricted movement of endoscopic cameras, which limits viewpoint diversity. As a result, the Gaussian Splatting representation overfits specific perspectives, leading to reduced geometric accuracy. To address this issue, we introduce a novel virtual camera-based regularization method that adaptively samples virtual viewpoints around the scene and incorporates them into the optimization process to mitigate overfitting. An effective depth-based regularization is applied to both real and virtual views to further refine the scene geometry. To enable fast deformation simulation, we propose a sparse control node-based Material Point Method, which integrates physical properties into the reconstructed scene while significantly reducing computational costs. Experimental results on representative surgical data demonstrate that our method can efficiently reconstruct and simulate surgical scenes from sparse endoscopic views. Notably, our method takes only a few minutes to reconstruct the surgical scene and is able to produce physically plausible deformations in real-time with user-defined interactions.", "date": "2025-09-21", "categories": []}
{"id": "2509.17246", "title": "SPFSplatV2: Efficient Self-Supervised Pose-Free 3D Gaussian Splatting from Sparse Views", "abstract": "We introduce SPFSplatV2, an efficient feed-forward framework for 3D Gaussian splatting from sparse multi-view images, requiring no ground-truth poses during training and inference. It employs a shared feature extraction backbone, enabling simultaneous prediction of 3D Gaussian primitives and camera poses in a canonical space from unposed inputs. A masked attention mechanism is introduced to efficiently estimate target poses during training, while a reprojection loss enforces pixel-aligned Gaussian primitives, providing stronger geometric constraints. We further demonstrate the compatibility of our training framework with different reconstruction architectures, resulting in two model variants. Remarkably, despite the absence of pose supervision, our method achieves state-of-the-art performance in both in-domain and out-of-domain novel view synthesis, even under extreme viewpoint changes and limited image overlap, and surpasses recent methods that rely on geometric supervision for relative pose estimation. By eliminating dependence on ground-truth poses, our method offers the scalability to leverage larger and more diverse datasets. Code and pretrained models will be available on our project page: https://ranrhuang.github.io/spfsplatv2/.", "date": "2025-09-21", "categories": []}
{"id": "2509.17762", "title": "Neural-MMGS: Multi-modal Neural Gaussian Splats for Large-Scale Scene Reconstruction", "abstract": "This paper proposes Neural-MMGS, a novel neural 3DGS framework for multimodal large-scale scene reconstruction that fuses multiple sensing modalities in a per-gaussian compact, learnable embedding. While recent works focusing on large-scale scene reconstruction have incorporated LiDAR data to provide more accurate geometric constraints, we argue that LiDAR's rich physical properties remain underexplored. Similarly, semantic information has been used for object retrieval, but could provide valuable high-level context for scene reconstruction. Traditional approaches append these properties to Gaussians as separate parameters, increasing memory usage and limiting information exchange across modalities. Instead, our approach fuses all modalities -- image, LiDAR, and semantics -- into a compact, learnable embedding that implicitly encodes optical, physical, and semantic features in each Gaussian. We then train lightweight neural decoders to map these embeddings to Gaussian parameters, enabling the reconstruction of each sensing modality with lower memory overhead and improved scalability. We evaluate Neural-MMGS on the Oxford Spires and KITTI-360 datasets. On Oxford Spires, we achieve higher-quality reconstructions, while on KITTI-360, our method reaches competitive results with less storage consumption compared with current approaches in LiDAR-based novel-view synthesis.", "date": "2025-09-22", "categories": []}
{"id": "2509.17789", "title": "From Restoration to Reconstruction: Rethinking 3D Gaussian Splatting for Underwater Scenes", "abstract": "Underwater image degradation poses significant challenges for 3D reconstruction, where simplified physical models often fail in complex scenes. We propose \\textbf{R-Splatting}, a unified framework that bridges underwater image restoration (UIR) with 3D Gaussian Splatting (3DGS) to improve both rendering quality and geometric fidelity. Our method integrates multiple enhanced views produced by diverse UIR models into a single reconstruction pipeline. During inference, a lightweight illumination generator samples latent codes to support diverse yet coherent renderings, while a contrastive loss ensures disentangled and stable illumination representations. Furthermore, we propose \\textit{Uncertainty-Aware Opacity Optimization (UAOO)}, which models opacity as a stochastic function to regularize training. This suppresses abrupt gradient responses triggered by illumination variation and mitigates overfitting to noisy or view-specific artifacts. Experiments on Seathru-NeRF and our new BlueCoral3D dataset demonstrate that R-Splatting outperforms strong baselines in both rendering quality and geometric accuracy.", "date": "2025-09-22", "categories": []}
{"id": "2509.17864", "title": "ProDyG: Progressive Dynamic Scene Reconstruction via Gaussian Splatting from Monocular Videos", "abstract": "Achieving truly practical dynamic 3D reconstruction requires online operation, global pose and map consistency, detailed appearance modeling, and the flexibility to handle both RGB and RGB-D inputs. However, existing SLAM methods typically merely remove the dynamic parts or require RGB-D input, while offline methods are not scalable to long video sequences, and current transformer-based feedforward methods lack global consistency and appearance details. To this end, we achieve online dynamic scene reconstruction by disentangling the static and dynamic parts within a SLAM system. The poses are tracked robustly with a novel motion masking strategy, and dynamic parts are reconstructed leveraging a progressive adaptation of a Motion Scaffolds graph. Our method yields novel view renderings competitive to offline methods and achieves on-par tracking with state-of-the-art dynamic SLAM methods.", "date": "2025-09-22", "categories": []}
{"id": "2509.15750", "title": "FloorSAM: SAM-Guided Floorplan Reconstruction with Semantic-Geometric Fusion", "abstract": "Reconstructing building floor plans from point cloud data is key for indoor navigation, BIM, and precise measurements. Traditional methods like geometric algorithms and Mask R-CNN-based deep learning often face issues with noise, limited generalization, and loss of geometric details. We propose FloorSAM, a framework that integrates point cloud density maps with the Segment Anything Model (SAM) for accurate floor plan reconstruction from LiDAR data. Using grid-based filtering, adaptive resolution projection, and image enhancement, we create robust top-down density maps. FloorSAM uses SAM's zero-shot learning for precise room segmentation, improving reconstruction across diverse layouts. Room masks are generated via adaptive prompt points and multistage filtering, followed by joint mask and point cloud analysis for contour extraction and regularization. This produces accurate floor plans and recovers room topological relationships. Tests on Giblayout and ISPRS datasets show better accuracy, recall, and robustness than traditional methods, especially in noisy and complex settings. Code and materials: github.com/Silentbarber/FloorSAM.", "date": "2025-09-19", "categories": []}
{"id": "2509.16050", "title": "Graph-based Point Cloud Surface Reconstruction using B-Splines", "abstract": "Generating continuous surfaces from discrete point cloud data is a fundamental task in several 3D vision applications. Real-world point clouds are inherently noisy due to various technical and environmental factors. Existing data-driven surface reconstruction algorithms rely heavily on ground truth normals or compute approximate normals as an intermediate step. This dependency makes them extremely unreliable for noisy point cloud datasets, even if the availability of ground truth training data is ensured, which is not always the case. B-spline reconstruction techniques provide compact surface representations of point clouds and are especially known for their smoothening properties. However, the complexity of the surfaces approximated using B-splines is directly influenced by the number and location of the spline control points. Existing spline-based modeling methods predict the locations of a fixed number of control points for a given point cloud, which makes it very difficult to match the complexity of its underlying surface. In this work, we develop a Dictionary-Guided Graph Convolutional Network-based surface reconstruction strategy where we simultaneously predict both the location and the number of control points for noisy point cloud data to generate smooth surfaces without the use of any point normals. We compare our reconstruction method with several well-known as well as recent baselines by employing widely-used evaluation metrics, and demonstrate that our method outperforms all of them both qualitatively and quantitatively.", "date": "2025-09-19", "categories": []}
{"id": "2509.16132", "title": "Recovering Parametric Scenes from Very Few Time-of-Flight Pixels", "abstract": "We aim to recover the geometry of 3D parametric scenes using very few depth measurements from low-cost, commercially available time-of-flight sensors. These sensors offer very low spatial resolution (i.e., a single pixel), but image a wide field-of-view per pixel and capture detailed time-of-flight data in the form of time-resolved photon counts. This time-of-flight data encodes rich scene information and thus enables recovery of simple scenes from sparse measurements. We investigate the feasibility of using a distributed set of few measurements (e.g., as few as 15 pixels) to recover the geometry of simple parametric scenes with a strong prior, such as estimating the 6D pose of a known object. To achieve this, we design a method that utilizes both feed-forward prediction to infer scene parameters, and differentiable rendering within an analysis-by-synthesis framework to refine the scene parameter estimate. We develop hardware prototypes and demonstrate that our method effectively recovers object pose given an untextured 3D model in both simulations and controlled real-world captures, and show promising initial results for other parametric scenes. We additionally conduct experiments to explore the limits and capabilities of our imaging solution.", "date": "2025-09-19", "categories": []}
{"id": "2307.13756", "title": "PlaneRecTR++: Unified Query Learning for Joint 3D Planar Reconstruction and Pose Estimation", "abstract": "The challenging task of 3D planar reconstruction from images involves several sub-tasks including frame-wise plane detection, segmentation, parameter regression and possibly depth prediction, along with cross-frame plane correspondence and relative camera pose estimation. Previous works adopt a divide and conquer strategy, addressing above sub-tasks with distinct network modules in a two-stage paradigm. Specifically, given an initial camera pose and per-frame plane predictions from the first stage, further exclusively designed modules relying on external plane correspondence labeling are applied to merge multi-view plane entities and produce refined camera pose. Notably, existing work fails to integrate these closely related sub-tasks into a unified framework, and instead addresses them separately and sequentially, which we identify as a primary source of performance limitations. Motivated by this finding and the success of query-based learning in enriching reasoning among semantic entities, in this paper, we propose PlaneRecTR++, a Transformer-based architecture, which for the first time unifies all tasks of multi-view planar reconstruction and pose estimation within a compact single-stage framework, eliminating the need for the initial pose estimation and supervision of plane correspondence. Extensive quantitative and qualitative experiments demonstrate that our proposed unified learning achieves mutual benefits across sub-tasks, achieving a new state-of-the-art performance on the public ScanNetv1, ScanNetv2, NYUv2-Plane, and MatterPort3D datasets. Codes are available at https://github.com/SJingjia/PlaneRecTR-PP.", "date": "2023-07-25", "categories": []}
{"id": "2509.13536", "title": "MemGS: Memory-Efficient Gaussian Splatting for Real-Time SLAM", "abstract": "Recent advancements in 3D Gaussian Splatting (3DGS) have made a significant impact on rendering and reconstruction techniques. Current research predominantly focuses on improving rendering performance and reconstruction quality using high-performance desktop GPUs, largely overlooking applications for embedded platforms like micro air vehicles (MAVs). These devices, with their limited computational resources and memory, often face a trade-off between system performance and reconstruction quality. In this paper, we improve existing methods in terms of GPU memory usage while enhancing rendering quality. Specifically, to address redundant 3D Gaussian primitives in SLAM, we propose merging them in voxel space based on geometric similarity. This reduces GPU memory usage without impacting system runtime performance. Furthermore, rendering quality is improved by initializing 3D Gaussian primitives via Patch-Grid (PG) point sampling, enabling more accurate modeling of the entire scene. Quantitative and qualitative evaluations on publicly available datasets demonstrate the effectiveness of our improvements.", "date": "2025-09-16", "categories": []}
{"id": "2509.12931", "title": "4DRadar-GS: Self-Supervised Dynamic Driving Scene Reconstruction with 4D Radar", "abstract": "3D reconstruction and novel view synthesis are critical for validating autonomous driving systems and training advanced perception models. Recent self-supervised methods have gained significant attention due to their cost-effectiveness and enhanced generalization in scenarios where annotated bounding boxes are unavailable. However, existing approaches, which often rely on frequency-domain decoupling or optical flow, struggle to accurately reconstruct dynamic objects due to imprecise motion estimation and weak temporal consistency, resulting in incomplete or distorted representations of dynamic scene elements. To address these challenges, we propose 4DRadar-GS, a 4D Radar-augmented self-supervised 3D reconstruction framework tailored for dynamic driving scenes. Specifically, we first present a 4D Radar-assisted Gaussian initialization scheme that leverages 4D Radar's velocity and spatial information to segment dynamic objects and recover monocular depth scale, generating accurate Gaussian point representations. In addition, we propose a Velocity-guided PointTrack (VGPT) model, which is jointly trained with the reconstruction pipeline under scene flow supervision, to track fine-grained dynamic trajectories and construct temporally consistent representations. Evaluated on the OmniHD-Scenes dataset, 4DRadar-GS achieves state-of-the-art performance in dynamic driving scene 3D reconstruction.", "date": "2025-09-16", "categories": []}
{"id": "2504.00763", "title": "UnIRe: Unsupervised Instance Decomposition for Dynamic Urban Scene Reconstruction", "abstract": "Reconstructing and decomposing dynamic urban scenes is crucial for autonomous driving, urban planning, and scene editing. However, existing methods fail to perform instance-aware decomposition without manual annotations, which is crucial for instance-level scene editing.We propose UnIRe, a 3D Gaussian Splatting (3DGS) based approach that decomposes a scene into a static background and individual dynamic instances using only RGB images and LiDAR point clouds. At its core, we introduce 4D superpoints, a novel representation that clusters multi-frame LiDAR points in 4D space, enabling unsupervised instance separation based on spatiotemporal correlations. These 4D superpoints serve as the foundation for our decomposed 4D initialization, i.e., providing spatial and temporal initialization to train a dynamic 3DGS for arbitrary dynamic classes without requiring bounding boxes or object templates.Furthermore, we introduce a smoothness regularization strategy in both 2D and 3D space, further improving the temporal stability.Experiments on benchmark datasets show that our method outperforms existing methods in decomposed dynamic scene reconstruction while enabling accurate and flexible instance-level editing, making it a practical solution for real-world applications.", "date": "2025-04-01", "categories": []}
{"id": "2504.12356", "title": "Regist3R: Incremental Registration with Stereo Foundation Model", "abstract": "Multi-view 3D reconstruction has remained an essential yet challenging problem in the field of computer vision. While DUSt3R and its successors have achieved breakthroughs in 3D reconstruction from unposed images, these methods exhibit significant limitations when scaling to multi-view scenarios, including high computational cost and cumulative error induced by global alignment. To address these challenges, we propose Regist3R, a novel stereo foundation model tailored for efficient and scalable incremental reconstruction. Regist3R leverages an incremental reconstruction paradigm, enabling large-scale 3D reconstructions from unordered and many-view image collections. We evaluate Regist3R on public datasets for camera pose estimation and 3D reconstruction. Our experiments demonstrate that Regist3R achieves comparable performance with optimization-based methods while significantly improving computational efficiency, and outperforms existing multi-view reconstruction models. Furthermore, to assess its performance in real-world applications, we introduce a challenging oblique aerial dataset which has long spatial spans and hundreds of views. The results highlight the effectiveness of Regist3R. We also demonstrate the first attempt to reconstruct large-scale scenes encompassing over thousands of views through pointmap-based foundation models, showcasing its potential for practical applications in large-scale 3D reconstruction tasks, including urban modeling, aerial mapping, and beyond.", "date": "2025-04-16", "categories": []}
{"id": "2507.17594", "title": "RemixFusion: Residual-based Mixed Representation for Large-scale Online RGB-D Reconstruction", "abstract": "The introduction of the neural implicit representation has notably propelled the advancement of online dense reconstruction techniques. Compared to traditional explicit representations, such as TSDF, it improves the mapping completeness and memory efficiency. However, the lack of reconstruction details and the time-consuming learning of neural representations hinder the widespread application of neural-based methods to large-scale online reconstruction. We introduce RemixFusion, a novel residual-based mixed representation for scene reconstruction and camera pose estimation dedicated to high-quality and large-scale online RGB-D reconstruction. In particular, we propose a residual-based map representation comprised of an explicit coarse TSDF grid and an implicit neural module that produces residuals representing fine-grained details to be added to the coarse grid. Such mixed representation allows for detail-rich reconstruction with bounded time and memory budget, contrasting with the overly-smoothed results by the purely implicit representations, thus paving the way for high-quality camera tracking. Furthermore, we extend the residual-based representation to handle multi-frame joint pose optimization via bundle adjustment (BA). In contrast to the existing methods, which optimize poses directly, we opt to optimize pose changes. Combined with a novel technique for adaptive gradient amplification, our method attains better optimization convergence and global optimality. Furthermore, we adopt a local moving volume to factorize the mixed scene representation with a divide-and-conquer design to facilitate efficient online learning in our residual-based framework. Extensive experiments demonstrate that our method surpasses all state-of-the-art ones, including those based either on explicit or implicit representations, in terms of the accuracy of both mapping and tracking on large-scale scenes.", "date": "2025-07-23", "categories": []}
{"id": "2509.11275", "title": "ROSGS: Relightable Outdoor Scenes With Gaussian Splatting", "abstract": "Image data captured outdoors often exhibit unbounded scenes and unconstrained, varying lighting conditions, making it challenging to decompose them into geometry, reflectance, and illumination. Recent works have focused on achieving this decomposition using Neural Radiance Fields (NeRF) or the 3D Gaussian Splatting (3DGS) representation but remain hindered by two key limitations: the high computational overhead associated with neural networks of NeRF and the use of low-frequency lighting representations, which often result in inefficient rendering and suboptimal relighting accuracy. We propose ROSGS, a two-stage pipeline designed to efficiently reconstruct relightable outdoor scenes using the Gaussian Splatting representation. By leveraging monocular normal priors, ROSGS first reconstructs the scene's geometry with the compact 2D Gaussian Splatting (2DGS) representation, providing an efficient and accurate geometric foundation. Building upon this reconstructed geometry, ROSGS then decomposes the scene's texture and lighting through a hybrid lighting model. This model effectively represents typical outdoor lighting by employing a spherical Gaussian function to capture the directional, high-frequency components of sunlight, while learning a radiance transfer function via Spherical Harmonic coefficients to model the remaining low-frequency skylight comprehensively. Both quantitative metrics and qualitative comparisons demonstrate that ROSGS achieves state-of-the-art performance in relighting outdoor scenes and highlight its ability to deliver superior relighting accuracy and rendering efficiency.", "date": "2025-09-14", "categories": []}
{"id": "2503.06179", "title": "ForestSplats: Deformable transient field for Gaussian Splatting in the Wild", "abstract": "Recently, 3D Gaussian Splatting (3D-GS) has emerged, showing real-time rendering speeds and high-quality results in static scenes. Although 3D-GS shows effectiveness in static scenes, their performance significantly degrades in real-world environments due to transient objects, lighting variations, and diverse levels of occlusion. To tackle this, existing methods estimate occluders or transient elements by leveraging pre-trained models or integrating additional transient field pipelines. However, these methods still suffer from two defects: 1) Using semantic features from the Vision Foundation model (VFM) causes additional computational costs. 2) The transient field requires significant memory to handle transient elements with per-view Gaussians and struggles to define clear boundaries for occluders, solely relying on photometric errors. To address these problems, we propose ForestSplats, a novel approach that leverages the deformable transient field and a superpixel-aware mask to efficiently represent transient elements in the 2D scene across unconstrained image collections and effectively decompose static scenes from transient distractors without VFM. We designed the transient field to be deformable, capturing per-view transient elements. Furthermore, we introduce a superpixel-aware mask that clearly defines the boundaries of occluders by considering photometric errors and superpixels. Additionally, we propose uncertainty-aware densification to avoid generating Gaussians within the boundaries of occluders during densification. Through extensive experiments across several benchmark datasets, we demonstrate that ForestSplats outperforms existing methods without VFM and shows significant memory efficiency in representing transient elements.", "date": "2025-03-08", "categories": []}
{"id": "2509.08828", "title": "SAFT: Shape and Appearance of Fabrics from Template via Differentiable Physical Simulations from Monocular Video", "abstract": "The reconstruction of three-dimensional dynamic scenes is a well-established yet challenging task within the domain of computer vision. In this paper, we propose a novel approach that combines the domains of 3D geometry reconstruction and appearance estimation for physically based rendering and present a system that is able to perform both tasks for fabrics, utilizing only a single monocular RGB video sequence as input. In order to obtain realistic and high-quality deformations and renderings, a physical simulation of the cloth geometry and differentiable rendering are employed. In this paper, we introduce two novel regularization terms for the 3D reconstruction task that improve the plausibility of the reconstruction by addressing the depth ambiguity problem in monocular video. In comparison with the most recent methods in the field, we have reduced the error in the 3D reconstruction by a factor of 2.64 while requiring a medium runtime of 30 min per scene. Furthermore, the optimized motion achieves sufficient quality to perform an appearance estimation of the deforming object, recovering sharp details from this single monocular RGB video.", "date": "2025-09-10", "categories": []}
{"id": "2509.07932", "title": "Dynamic Scene 3D Reconstruction of an Uncooperative Resident Space Object", "abstract": "Characterization of uncooperative Resident Space Objects (RSO) play a crucial role in On-Orbit Servicing (OOS) and Active Debris Removal (ADR) missions to assess the geometry and motion properties. To address the challenges of reconstructing tumbling uncooperative targets, this study evaluates the performance of existing state-of-the-art 3D reconstruction algorithms for dynamic scenes, focusing on their ability to generate geometrically accurate models with high-fidelity. To support our evaluation, we developed a simulation environment using Isaac Sim to generate physics-accurate 2D image sequences of tumbling satellite under realistic orbital lighting conditions. Our preliminary results on static scenes using Neuralangelo demonstrate promising reconstruction quality. The generated 3D meshes closely match the original CAD models with minimal errors and artifacts when compared using Cloud Compare (CC). The reconstructed models were able to capture critical fine details for mission planning. This provides a baseline for our ongoing evaluation of dynamic scene reconstruction.", "date": "2025-09-09", "categories": []}
{"id": "2412.08511", "title": "NF3DM: Combining Neural Fields and Deformation Models for 3D Non-Rigid Motion Reconstruction", "abstract": "We introduce a novel, data-driven approach for reconstructing temporally coherent 3D motion from unstructured and potentially partial observations of non-rigidly deforming shapes. Our goal is to achieve high-fidelity motion reconstructions for shapes that undergo near-isometric deformations, such as humans wearing loose clothing. The key novelty of our work lies in its ability to combine implicit shape representations with explicit mesh-based deformation models, enabling detailed and temporally coherent motion reconstructions without relying on parametric shape models or decoupling shape and motion. Each frame is represented as a neural field decoded from a feature space where observations over time are fused, hence preserving geometric details present in the input data. Temporal coherence is enforced with a near-isometric deformation constraint between adjacent frames that applies to the underlying surface in the neural field. Our method outperforms state-of-the-art approaches, as demonstrated by its application to human and animal motion sequences reconstructed from monocular depth videos.", "date": "2024-12-11", "categories": []}
{"id": "2501.06897", "title": "ActiveGAMER: Active GAussian Mapping through Efficient Rendering", "abstract": "We introduce ActiveGAMER, an active mapping system that utilizes 3D Gaussian Splatting (3DGS) to achieve high-quality, real-time scene mapping and exploration. Unlike traditional NeRF-based methods, which are computationally demanding and restrict active mapping performance, our approach leverages the efficient rendering capabilities of 3DGS, allowing effective and efficient exploration in complex environments. The core of our system is a rendering-based information gain module that dynamically identifies the most informative viewpoints for next-best-view planning, enhancing both geometric and photometric reconstruction accuracy. ActiveGAMER also integrates a carefully balanced framework, combining coarse-to-fine exploration, post-refinement, and a global-local keyframe selection strategy to maximize reconstruction completeness and fidelity. Our system autonomously explores and reconstructs environments with state-of-the-art geometric and photometric accuracy and completeness, significantly surpassing existing approaches in both aspects. Extensive evaluations on benchmark datasets such as Replica and MP3D highlight ActiveGAMER's effectiveness in active mapping tasks.", "date": "2025-01-12", "categories": []}
{"id": "2507.06269", "title": "BayesSDF: Surface-Based Laplacian Uncertainty Estimation for 3D Geometry with Neural Signed Distance Fields", "abstract": "Accurate surface estimation is critical for downstream tasks in scientific simulation, and quantifying uncertainty in implicit neural 3D representations still remains a substantial challenge due to computational inefficiencies, scalability issues, and geometric inconsistencies. However, current neural implicit surface models do not offer a principled way to quantify uncertainty, limiting their reliability in real-world applications. Inspired by recent probabilistic rendering approaches, we introduce BayesSDF, a novel probabilistic framework for uncertainty estimation in neural implicit 3D representations. Unlike radiance-based models such as Neural Radiance Fields (NeRF) or 3D Gaussian Splatting, Signed Distance Functions (SDFs) provide continuous, differentiable surface representations, making them especially well-suited for uncertainty-aware modeling. BayesSDF applies a Laplace approximation over SDF weights and derives Hessian-based metrics to estimate local geometric instability. We empirically demonstrate that these uncertainty estimates correlate strongly with surface reconstruction error across both synthetic and real-world benchmarks. By enabling surface-aware uncertainty quantification, BayesSDF lays the groundwork for more robust, interpretable, and actionable 3D perception systems.", "date": "2025-07-08", "categories": []}
{"id": "2508.14014", "title": "Online 3D Gaussian Splatting Modeling with Novel View Selection", "abstract": "This study addresses the challenge of generating online 3D Gaussian Splatting (3DGS) models from RGB-only frames. Previous studies have employed dense SLAM techniques to estimate 3D scenes from keyframes for 3DGS model construction. However, these methods are limited by their reliance solely on keyframes, which are insufficient to capture an entire scene, resulting in incomplete reconstructions. Moreover, building a generalizable model requires incorporating frames from diverse viewpoints to achieve broader scene coverage. However, online processing restricts the use of many frames or extensive training iterations. Therefore, we propose a novel method for high-quality 3DGS modeling that improves model completeness through adaptive view selection. By analyzing reconstruction quality online, our approach selects optimal non-keyframes for additional training. By integrating both keyframes and selected non-keyframes, the method refines incomplete regions from diverse viewpoints, significantly enhancing completeness. We also present a framework that incorporates an online multi-view stereo approach, ensuring consistency in 3D information throughout the 3DGS modeling process. Experimental results demonstrate that our method outperforms state-of-the-art methods, delivering exceptional performance in complex outdoor scenes.", "date": "2025-08-19", "categories": []}
{"id": "2509.05296", "title": "WinT3R: Window-Based Streaming Reconstruction with Camera Token Pool", "abstract": "We present WinT3R, a feed-forward reconstruction model capable of online prediction of precise camera poses and high-quality point maps. Previous methods suffer from a trade-off between reconstruction quality and real-time performance. To address this, we first introduce a sliding window mechanism that ensures sufficient information exchange among frames within the window, thereby improving the quality of geometric predictions without large computation. In addition, we leverage a compact representation of cameras and maintain a global camera token pool, which enhances the reliability of camera pose estimation without sacrificing efficiency. These designs enable WinT3R to achieve state-of-the-art performance in terms of online reconstruction quality, camera pose estimation, and reconstruction speed, as validated by extensive experiments on diverse datasets. Code and model are publicly available at https://github.com/LiZizun/WinT3R.", "date": "2025-09-05", "categories": []}
{"id": "2502.08352", "title": "Sat-DN: Implicit Surface Reconstruction from Multi-View Satellite Images with Depth and Normal Supervision", "abstract": "With advancements in satellite imaging technology, acquiring high-resolution multi-view satellite imagery has become increasingly accessible, enabling rapid and location-independent ground model reconstruction. However, traditional stereo matching methods struggle to capture fine details, and while neural radiance fields (NeRFs) achieve high-quality reconstructions, their training time is prohibitively long. Moreover, challenges such as low visibility of building facades, illumination and style differences between pixels, and weakly textured regions in satellite imagery further make it hard to reconstruct reasonable terrain geometry and detailed building facades. To address these issues, we propose Sat-DN, a novel framework leveraging a progressively trained multi-resolution hash grid reconstruction architecture with explicit depth guidance and surface normal consistency constraints to enhance reconstruction quality. The multi-resolution hash grid accelerates training, while the progressive strategy incrementally increases the learning frequency, using coarse low-frequency geometry to guide the reconstruction of fine high-frequency details. The depth and normal constraints ensure a clear building outline and correct planar distribution. Extensive experiments on the DFC2019 dataset demonstrate that Sat-DN outperforms existing methods, achieving state-of-the-art results in both qualitative and quantitative evaluations. The code is available at https://github.com/costune/SatDN.", "date": "2025-02-12", "categories": []}
{"id": "2509.03775", "title": "ContraGS: Codebook-Condensed and Trainable Gaussian Splatting for Fast, Memory-Efficient Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) is a state-of-art technique to model real-world scenes with high quality and real-time rendering. Typically, a higher quality representation can be achieved by using a large number of 3D Gaussians. However, using large 3D Gaussian counts significantly increases the GPU device memory for storing model parameters. A large model thus requires powerful GPUs with high memory capacities for training and has slower training/rendering latencies due to the inefficiencies of memory access and data movement. In this work, we introduce ContraGS, a method to enable training directly on compressed 3DGS representations without reducing the Gaussian Counts, and thus with a little loss in model quality. ContraGS leverages codebooks to compactly store a set of Gaussian parameter vectors throughout the training process, thereby significantly reducing memory consumption. While codebooks have been demonstrated to be highly effective at compressing fully trained 3DGS models, directly training using codebook representations is an unsolved challenge. ContraGS solves the problem of learning non-differentiable parameters in codebook-compressed representations by posing parameter estimation as a Bayesian inference problem. To this end, ContraGS provides a framework that effectively uses MCMC sampling to sample over a posterior distribution of these compressed representations. With ContraGS, we demonstrate that ContraGS significantly reduces the peak memory during training (on average 3.49X) and accelerated training and rendering (1.36X and 1.88X on average, respectively), while retraining close to state-of-art quality.", "date": "2025-09-03", "categories": []}
{"id": "2412.05548", "title": "Street Gaussians without 3D Object Tracker", "abstract": "Realistic scene reconstruction in driving scenarios poses significant challenges due to fast-moving objects. Most existing methods rely on labor-intensive manual labeling of object poses to reconstruct dynamic objects in canonical space and move them based on these poses during rendering. While some approaches attempt to use 3D object trackers to replace manual annotations, the limited generalization of 3D trackers -- caused by the scarcity of large-scale 3D datasets -- results in inferior reconstructions in real-world settings. In contrast, 2D foundation models demonstrate strong generalization capabilities. To eliminate the reliance on 3D trackers and enhance robustness across diverse environments, we propose a stable object tracking module by leveraging associations from 2D deep trackers within a 3D object fusion strategy. We address inevitable tracking errors by further introducing a motion learning strategy in an implicit feature space that autonomously corrects trajectory errors and recovers missed detections. Experimental results on Waymo-NOTR and KITTI show that our method outperforms existing approaches. Our code will be released on https://lolrudy.github.io/No3DTrackSG/.", "date": "2024-12-07", "categories": []}
{"id": "2412.09573", "title": "FreeSplatter: Pose-free Gaussian Splatting for Sparse-view 3D Reconstruction", "abstract": "Sparse-view reconstruction models typically require precise camera poses, yet obtaining these parameters from sparse-view images remains challenging. We introduce FreeSplatter, a scalable feed-forward framework that generates high-quality 3D Gaussians from uncalibrated sparse-view images while estimating camera parameters within seconds. Our approach employs a streamlined transformer architecture where self-attention blocks facilitate information exchange among multi-view image tokens, decoding them into pixel-aligned 3D Gaussian primitives within a unified reference frame. This representation enables both high-fidelity 3D modeling and efficient camera parameter estimation using off-the-shelf solvers. We develop two specialized variants--for object-centric and scene-level reconstruction--trained on comprehensive datasets. Remarkably, FreeSplatter outperforms several pose-dependent Large Reconstruction Models (LRMs) by a notable margin while achieving comparable or even better pose estimation accuracy compared to state-of-the-art pose-free reconstruction approach MASt3R in challenging benchmarks. Beyond technical benchmarks, FreeSplatter streamlines text/image-to-3D content creation pipelines, eliminating the complexity of camera pose management while delivering exceptional visual fidelity.", "date": "2024-12-12", "categories": []}
{"id": "2504.05740", "title": "Micro-splatting: Multistage Isotropy-informed Covariance Regularization Optimization for High-Fidelity 3D Gaussian Splatting", "abstract": "High-fidelity 3D Gaussian Splatting methods excel at capturing fine textures but often overlook model compactness, resulting in massive splat counts, bloated memory, long training, and complex post-processing. We present Micro-Splatting: Two-Stage Adaptive Growth and Refinement, a unified, in-training pipeline that preserves visual detail while drastically reducing model complexity without any post-processing or auxiliary neural modules. In Stage I (Growth), we introduce a trace-based covariance regularization to maintain near-isotropic Gaussians, mitigating low-pass filtering in high-frequency regions and improving spherical-harmonic color fitting. We then apply gradient-guided adaptive densification that subdivides splats only in visually complex regions, leaving smooth areas sparse. In Stage II (Refinement), we prune low-impact splats using a simple opacity-scale importance score and merge redundant neighbors via lightweight spatial and feature thresholds, producing a lean yet detail-rich model. On four object-centric benchmarks, Micro-Splatting reduces splat count and model size by up to 60% and shortens training by 20%, while matching or surpassing state-of-the-art PSNR, SSIM, and LPIPS in real-time rendering. These results demonstrate that Micro-Splatting delivers both compactness and high fidelity in a single, efficient, end-to-end framework.", "date": "2025-04-08", "categories": []}
{"id": "2505.15737", "title": "RUSplatting: Robust 3D Gaussian Splatting for Sparse-View Underwater Scene Reconstruction", "abstract": "Reconstructing high-fidelity underwater scenes remains a challenging task due to light absorption, scattering, and limited visibility inherent in aquatic environments. This paper presents an enhanced Gaussian Splatting-based framework that improves both the visual quality and geometric accuracy of deep underwater rendering. We propose decoupled learning for RGB channels, guided by the physics of underwater attenuation, to enable more accurate colour restoration. To address sparse-view limitations and improve view consistency, we introduce a frame interpolation strategy with a novel adaptive weighting scheme. Additionally, we introduce a new loss function aimed at reducing noise while preserving edges, which is essential for deep-sea content. We also release a newly collected dataset, Submerged3D, captured specifically in deep-sea environments. Experimental results demonstrate that our framework consistently outperforms state-of-the-art methods with PSNR gains up to 1.90dB, delivering superior perceptual quality and robustness, and offering promising directions for marine robotics and underwater visual analytics. The code of RUSplatting is available at https://github.com/theflash987/RUSplatting and the dataset Submerged3D can be downloaded at https://zenodo.org/records/15482420.", "date": "2025-05-21", "categories": []}
{"id": "2506.18677", "title": "Reconstructing Tornadoes in 3D with Gaussian Splatting", "abstract": "Accurately reconstructing the 3D structure of tornadoes is critically important for understanding and preparing for this highly destructive weather phenomenon. While modern 3D scene reconstruction techniques, such as 3D Gaussian splatting (3DGS), could provide a valuable tool for reconstructing the 3D structure of tornados, at present we are critically lacking a controlled tornado dataset with which to develop and validate these tools. In this work we capture and release a novel multiview dataset of a small lab-based tornado. We demonstrate one can effectively reconstruct and visualize the 3D structure of this tornado using 3DGS.", "date": "2025-06-23", "categories": []}
{"id": "2509.00989", "title": "Towards Integrating Multi-Spectral Imaging with Gaussian Splatting", "abstract": "We present a study of how to integrate color (RGB) and multi-spectral imagery (red, green, red-edge, and near-infrared) into the 3D Gaussian Splatting (3DGS) framework, a state-of-the-art explicit radiance-field-based method for fast and high-fidelity 3D reconstruction from multi-view images. While 3DGS excels on RGB data, naive per-band optimization of additional spectra yields poor reconstructions due to inconsistently appearing geometry in the spectral domain. This problem is prominent, even though the actual geometry is the same, regardless of spectral modality. To investigate this, we evaluate three strategies: 1) Separate per-band reconstruction with no shared structure. 2) Splitting optimization, in which we first optimize RGB geometry, copy it, and then fit each new band to the model by optimizing both geometry and band representation. 3) Joint, in which the modalities are jointly optimized, optionally with an initial RGB-only phase. We showcase through quantitative metrics and qualitative novel-view renderings on multi-spectral datasets the effectiveness of our dedicated optimized Joint strategy, increasing overall spectral reconstruction as well as enhancing RGB results through spectral cross-talk. We therefore suggest integrating multi-spectral data directly into the spherical harmonics color components to compactly model each Gaussian's multi-spectral reflectance. Moreover, our analysis reveals several key trade-offs in when and how to introduce spectral bands during optimization, offering practical insights for robust multi-modal 3DGS reconstruction.", "date": "2025-08-31", "categories": []}
{"id": "2509.01873", "title": "Doctoral Thesis: Geometric Deep Learning For Camera Pose Prediction, Registration, Depth Estimation, and 3D Reconstruction", "abstract": "Modern deep learning developments create new opportunities for 3D mapping technology, scene reconstruction pipelines, and virtual reality development. Despite advances in 3D deep learning technology, direct training of deep learning models on 3D data faces challenges due to the high dimensionality inherent in 3D data and the scarcity of labeled datasets. Structure-from-motion (SfM) and Simultaneous Localization and Mapping (SLAM) exhibit robust performance when applied to structured indoor environments but often struggle with ambiguous features in unstructured environments. These techniques often struggle to generate detailed geometric representations effective for downstream tasks such as rendering and semantic analysis. Current limitations require the development of 3D representation methods that combine traditional geometric techniques with deep learning capabilities to generate robust geometry-aware deep learning models.\n  The dissertation provides solutions to the fundamental challenges in 3D vision by developing geometric deep learning methods tailored for essential tasks such as camera pose estimation, point cloud registration, depth prediction, and 3D reconstruction. The integration of geometric priors or constraints, such as including depth information, surface normals, and equivariance into deep learning models, enhances both the accuracy and robustness of geometric representations. This study systematically investigates key components of 3D vision, including camera pose estimation, point cloud registration, depth estimation, and high-fidelity 3D reconstruction, demonstrating their effectiveness across real-world applications such as digital cultural heritage preservation and immersive VR/AR environments.", "date": "2025-09-02", "categories": []}
{"id": "2508.21542", "title": "Complete Gaussian Splats from a Single Image with Denoising Diffusion Models", "abstract": "Gaussian splatting typically requires dense observations of the scene and can fail to reconstruct occluded and unobserved areas. We propose a latent diffusion model to reconstruct a complete 3D scene with Gaussian splats, including the occluded parts, from only a single image during inference. Completing the unobserved surfaces of a scene is challenging due to the ambiguity of the plausible surfaces. Conventional methods use a regression-based formulation to predict a single \"mode\" for occluded and out-of-frustum surfaces, leading to blurriness, implausibility, and failure to capture multiple possible explanations. Thus, they often address this problem partially, focusing either on objects isolated from the background, reconstructing only visible surfaces, or failing to extrapolate far from the input views. In contrast, we propose a generative formulation to learn a distribution of 3D representations of Gaussian splats conditioned on a single input image. To address the lack of ground-truth training data, we propose a Variational AutoReconstructor to learn a latent space only from 2D images in a self-supervised manner, over which a diffusion model is trained. Our method generates faithful reconstructions and diverse samples with the ability to complete the occluded surfaces for high-quality 360-degree renderings.", "date": "2025-08-29", "categories": []}
{"id": "2508.20965", "title": "DrivingGaussian++: Towards Realistic Reconstruction and Editable Simulation for Surrounding Dynamic Driving Scenes", "abstract": "We present DrivingGaussian++, an efficient and effective framework for realistic reconstructing and controllable editing of surrounding dynamic autonomous driving scenes. DrivingGaussian++ models the static background using incremental 3D Gaussians and reconstructs moving objects with a composite dynamic Gaussian graph, ensuring accurate positions and occlusions. By integrating a LiDAR prior, it achieves detailed and consistent scene reconstruction, outperforming existing methods in dynamic scene reconstruction and photorealistic surround-view synthesis. DrivingGaussian++ supports training-free controllable editing for dynamic driving scenes, including texture modification, weather simulation, and object manipulation, leveraging multi-view images and depth priors. By integrating large language models (LLMs) and controllable editing, our method can automatically generate dynamic object motion trajectories and enhance their realism during the optimization process. DrivingGaussian++ demonstrates consistent and realistic editing results and generates dynamic multi-view driving scenarios, while significantly enhancing scene diversity. More results and code can be found at the project site: https://xiong-creator.github.io/DrivingGaussian_plus.github.io", "date": "2025-08-28", "categories": []}
{"id": "2508.19508", "title": "DATR: Diffusion-based 3D Apple Tree Reconstruction Framework with Sparse-View", "abstract": "Digital twin applications offered transformative potential by enabling real-time monitoring and robotic simulation through accurate virtual replicas of physical assets. The key to these systems is 3D reconstruction with high geometrical fidelity. However, existing methods struggled under field conditions, especially with sparse and occluded views. This study developed a two-stage framework (DATR) for the reconstruction of apple trees from sparse views. The first stage leverages onboard sensors and foundation models to semi-automatically generate tree masks from complex field images. Tree masks are used to filter out background information in multi-modal data for the single-image-to-3D reconstruction at the second stage. This stage consists of a diffusion model and a large reconstruction model for respective multi view and implicit neural field generation. The training of the diffusion model and LRM was achieved by using realistic synthetic apple trees generated by a Real2Sim data generator. The framework was evaluated on both field and synthetic datasets. The field dataset includes six apple trees with field-measured ground truth, while the synthetic dataset featured structurally diverse trees. Evaluation results showed that our DATR framework outperformed existing 3D reconstruction methods across both datasets and achieved domain-trait estimation comparable to industrial-grade stationary laser scanners while improving the throughput by $\\sim$360 times, demonstrating strong potential for scalable agricultural digital twin systems.", "date": "2025-08-27", "categories": []}
{"id": "2508.18314", "title": "SERES: Semantic-aware neural reconstruction from sparse views", "abstract": "We propose a semantic-aware neural reconstruction method to generate 3D high-fidelity models from sparse images. To tackle the challenge of severe radiance ambiguity caused by mismatched features in sparse input, we enrich neural implicit representations by adding patch-based semantic logits that are optimized together with the signed distance field and the radiance field. A novel regularization based on the geometric primitive masks is introduced to mitigate shape ambiguity. The performance of our approach has been verified in experimental evaluation. The average chamfer distances of our reconstruction on the DTU dataset can be reduced by 44% for SparseNeuS and 20% for VolRecon. When working as a plugin for those dense reconstruction baselines such as NeuS and Neuralangelo, the average error on the DTU dataset can be reduced by 69% and 68% respectively.", "date": "2025-08-23", "categories": []}
{"id": "2508.18696", "title": "ColorGS: High-fidelity Surgical Scene Reconstruction with Colored Gaussian Splatting", "abstract": "High-fidelity reconstruction of deformable tissues from endoscopic videos remains challenging due to the limitations of existing methods in capturing subtle color variations and modeling global deformations. While 3D Gaussian Splatting (3DGS) enables efficient dynamic reconstruction, its fixed per-Gaussian color assignment struggles with intricate textures, and linear deformation modeling fails to model consistent global deformation. To address these issues, we propose ColorGS, a novel framework that integrates spatially adaptive color encoding and enhanced deformation modeling for surgical scene reconstruction. First, we introduce Colored Gaussian Primitives, which employ dynamic anchors with learnable color parameters to adaptively encode spatially varying textures, significantly improving color expressiveness under complex lighting and tissue similarity. Second, we design an Enhanced Deformation Model (EDM) that combines time-aware Gaussian basis functions with learnable time-independent deformations, enabling precise capture of both localized tissue deformations and global motion consistency caused by surgical interactions. Extensive experiments on DaVinci robotic surgery videos and benchmark datasets (EndoNeRF, StereoMIS) demonstrate that ColorGS achieves state-of-the-art performance, attaining a PSNR of 39.85 (1.5 higher than prior 3DGS-based methods) and superior SSIM (97.25\\%) while maintaining real-time rendering efficiency. Our work advances surgical scene reconstruction by balancing high fidelity with computational practicality, critical for intraoperative guidance and AR/VR applications.", "date": "2025-08-26", "categories": []}
{"id": "2508.19204", "title": "LSD-3D: Large-Scale 3D Driving Scene Generation with Geometry Grounding", "abstract": "Large-scale scene data is essential for training and testing in robot learning. Neural reconstruction methods have promised the capability of reconstructing large physically-grounded outdoor scenes from captured sensor data. However, these methods have baked-in static environments and only allow for limited scene control -- they are functionally constrained in scene and trajectory diversity by the captures from which they are reconstructed. In contrast, generating driving data with recent image or video diffusion models offers control, however, at the cost of geometry grounding and causality. In this work, we aim to bridge this gap and present a method that directly generates large-scale 3D driving scenes with accurate geometry, allowing for causal novel view synthesis with object permanence and explicit 3D geometry estimation. The proposed method combines the generation of a proxy geometry and environment representation with score distillation from learned 2D image priors. We find that this approach allows for high controllability, enabling the prompt-guided geometry and high-fidelity texture and structure that can be conditioned on map layouts -- producing realistic and geometrically consistent 3D generations of complex driving scenes.", "date": "2025-08-26", "categories": []}
{"id": "2409.16296", "title": "LiDAR-3DGS: LiDAR Reinforced 3D Gaussian Splatting for Multimodal Radiance Field Rendering", "abstract": "In this paper, we explore the capabilities of multimodal inputs to 3D Gaussian Splatting (3DGS) based Radiance Field Rendering. We present LiDAR-3DGS, a novel method of reinforcing 3DGS inputs with LiDAR generated point clouds to significantly improve the accuracy and detail of 3D models. We demonstrate a systematic approach of LiDAR reinforcement to 3DGS to enable capturing of important features such as bolts, apertures, and other details that are often missed by image-based features alone. These details are crucial for engineering applications such as remote monitoring and maintenance. Without modifying the underlying 3DGS algorithm, we demonstrate that even a modest addition of LiDAR generated point cloud significantly enhances the perceptual quality of the models. At 30k iterations, the model generated by our method resulted in an increase of 7.064% in PSNR and 0.565% in SSIM, respectively. Since the LiDAR used in this research was a commonly used commercial-grade device, the improvements observed were modest and can be further enhanced with higher-grade LiDAR systems. Additionally, these improvements can be supplementary to other derivative works of Radiance Field Rendering and also provide a new insight for future LiDAR and computer vision integrated modeling.", "date": "2024-09-09", "categories": []}
{"id": "2504.19938", "title": "Mesh-Learner: Texturing Mesh with Spherical Harmonics", "abstract": "In this paper, we present a 3D reconstruction and rendering framework termed Mesh-Learner that is natively compatible with traditional rasterization pipelines. It integrates mesh and spherical harmonic (SH) texture (i.e., texture filled with SH coefficients) into the learning process to learn each mesh s view-dependent radiance end-to-end. Images are rendered by interpolating surrounding SH Texels at each pixel s sampling point using a novel interpolation method. Conversely, gradients from each pixel are back-propagated to the related SH Texels in SH textures. Mesh-Learner exploits graphic features of rasterization pipeline (texture sampling, deferred rendering) to render, which makes Mesh-Learner naturally compatible with tools (e.g., Blender) and tasks (e.g., 3D reconstruction, scene rendering, reinforcement learning for robotics) that are based on rasterization pipelines. Our system can train vast, unlimited scenes because we transfer only the SH textures within the frustum to the GPU for training. At other times, the SH textures are stored in CPU RAM, which results in moderate GPU memory usage. The rendering results on interpolation and extrapolation sequences in the Replica and FAST-LIVO2 datasets achieve state-of-the-art performance compared to existing state-of-the-art methods (e.g., 3D Gaussian Splatting and M2-Mapping). To benefit the society, the code will be available at https://github.com/hku-mars/Mesh-Learner.", "date": "2025-04-28", "categories": []}
{"id": "2505.06219", "title": "VIN-NBV: A View Introspection Network for Next-Best-View Selection", "abstract": "Next Best View (NBV) algorithms aim to maximize 3D scene acquisition quality using minimal resources, e.g. number of acquisitions, time taken, or distance traversed. Prior methods often rely on coverage maximization as a proxy for reconstruction quality, but for complex scenes with occlusions and finer details, this is not always sufficient and leads to poor reconstructions. Our key insight is to train an acquisition policy that directly optimizes for reconstruction quality rather than just coverage. To achieve this, we introduce the View Introspection Network (VIN): a lightweight neural network that predicts the Relative Reconstruction Improvement (RRI) of a potential next viewpoint without making any new acquisitions. We use this network to power a simple, yet effective, sequential samplingbased greedy NBV policy. Our approach, VIN-NBV, generalizes to unseen object categories, operates without prior scene knowledge, is adaptable to resource constraints, and can handle occlusions. We show that our RRI fitness criterion leads to a ~30% gain in reconstruction quality over a coverage-based criterion using the same greedy strategy. Furthermore, VIN-NBV also outperforms deep reinforcement learning methods, Scan-RL and GenNBV, by ~40%.", "date": "2025-05-09", "categories": []}
{"id": "2508.09912", "title": "E-4DGS: High-Fidelity Dynamic Reconstruction from the Multi-view Event Cameras", "abstract": "Novel view synthesis and 4D reconstruction techniques predominantly rely on RGB cameras, thereby inheriting inherent limitations such as the dependence on adequate lighting, susceptibility to motion blur, and a limited dynamic range. Event cameras, offering advantages of low power, high temporal resolution and high dynamic range, have brought a new perspective to addressing the scene reconstruction challenges in high-speed motion and", "date": "2025-08-13", "categories": []}
{"id": "2508.17044", "title": "M3DMap: Object-aware Multimodal 3D Mapping for Dynamic Environments", "abstract": "3D mapping in dynamic environments poses a challenge for modern researchers in robotics and autonomous transportation. There are no universal representations for dynamic 3D scenes that incorporate multimodal data such as images, point clouds, and text. This article takes a step toward solving this problem. It proposes a taxonomy of methods for constructing multimodal 3D maps, classifying contemporary approaches based on scene types and representations, learning methods, and practical applications. Using this taxonomy, a brief structured analysis of recent methods is provided. The article also describes an original modular method called M3DMap, designed for object-aware construction of multimodal 3D maps for both static and dynamic scenes. It consists of several interconnected components: a neural multimodal object segmentation and tracking module; an odometry estimation module, including trainable algorithms; a module for 3D map construction and updating with various implementations depending on the desired scene representation; and a multimodal data retrieval module. The article highlights original implementations of these modules and their advantages in solving various practical tasks, from 3D object grounding to mobile manipulation. Additionally, it presents theoretical propositions demonstrating the positive effect of using multimodal data and modern foundational models in 3D mapping methods. Details of the taxonomy and method implementation are available at https://yuddim.github.io/M3DMap.", "date": "2025-08-23", "categories": []}
{"id": "2508.17172", "title": "VROOM - Visual Reconstruction over Onboard Multiview", "abstract": "We introduce VROOM, a system for reconstructing 3D models of Formula 1 circuits using only onboard camera footage from racecars. Leveraging video data from the 2023 Monaco Grand Prix, we address video challenges such as high-speed motion and sharp cuts in camera frames. Our pipeline analyzes different methods such as DROID-SLAM, AnyCam, and Monst3r and combines preprocessing techniques such as different methods of masking, temporal chunking, and resolution scaling to account for dynamic motion and computational constraints. We show that Vroom is able to partially recover track and vehicle trajectories in complex environments. These findings indicate the feasibility of using onboard video for scalable 4D reconstruction in real-world settings. The project page can be found at https://varun-bharadwaj.github.io/vroom, and our code is available at https://github.com/yajatyadav/vroom.", "date": "2025-08-24", "categories": []}
{"id": "2508.17436", "title": "Disentangled Geometry and Appearance for Efficient Multi-View Surface Reconstruction and Rendering", "abstract": "This paper addresses the limitations of neural rendering-based multi-view surface reconstruction methods, which require an additional mesh extraction step that is inconvenient and would produce poor-quality surfaces with mesh aliasing, restricting downstream applications. Building on the explicit mesh representation and differentiable rasterization framework, this work proposes an efficient solution that preserves the high efficiency of this framework while significantly improving reconstruction quality and versatility. Specifically, we introduce a disentangled geometry and appearance model that does not rely on deep networks, enhancing learning and broadening applicability. A neural deformation field is constructed to incorporate global geometric context, enhancing geometry learning, while a novel regularization constrains geometric features passed to a neural shader to ensure its accuracy and boost shading. For appearance, a view-invariant diffuse term is separated and baked into mesh vertices, further improving rendering efficiency. Experimental results demonstrate that the proposed method achieves state-of-the-art training (4.84 minutes) and rendering (0.023 seconds) speeds, with reconstruction quality that is competitive with top-performing methods. Moreover, the method enables practical applications such as mesh and texture editing, showcasing its versatility and application potential. This combination of efficiency, competitive quality, and broad applicability makes our approach a valuable contribution to multi-view surface reconstruction and rendering.", "date": "2025-08-24", "categories": []}
{"id": "2508.17579", "title": "IDU: Incremental Dynamic Update of Existing 3D Virtual Environments with New Imagery Data", "abstract": "For simulation and training purposes, military organizations have made substantial investments in developing high-resolution 3D virtual environments through extensive imaging and 3D scanning. However, the dynamic nature of battlefield conditions-where objects may appear or vanish over time-makes frequent full-scale updates both time-consuming and costly. In response, we introduce the Incremental Dynamic Update (IDU) pipeline, which efficiently updates existing 3D reconstructions, such as 3D Gaussian Splatting (3DGS), with only a small set of newly acquired images. Our approach starts with camera pose estimation to align new images with the existing 3D model, followed by change detection to pinpoint modifications in the scene. A 3D generative AI model is then used to create high-quality 3D assets of the new elements, which are seamlessly integrated into the existing 3D model. The IDU pipeline incorporates human guidance to ensure high accuracy in object identification and placement, with each update focusing on a single new object at a time. Experimental results confirm that our proposed IDU pipeline significantly reduces update time and labor, offering a cost-effective and targeted solution for maintaining up-to-date 3D models in rapidly evolving military scenarios.", "date": "2025-08-25", "categories": []}
{"id": "2409.12892", "title": "3DGS-LM: Faster Gaussian-Splatting Optimization with Levenberg-Marquardt", "abstract": "We present 3DGS-LM, a new method that accelerates the reconstruction of 3D Gaussian Splatting (3DGS) by replacing its ADAM optimizer with a tailored Levenberg-Marquardt (LM). Existing methods reduce the optimization time by decreasing the number of Gaussians or by improving the implementation of the differentiable rasterizer. However, they still rely on the ADAM optimizer to fit Gaussian parameters of a scene in thousands of iterations, which can take up to an hour. To this end, we change the optimizer to LM that runs in conjunction with the 3DGS differentiable rasterizer. For efficient GPU parallization, we propose a caching data structure for intermediate gradients that allows us to efficiently calculate Jacobian-vector products in custom CUDA kernels. In every LM iteration, we calculate update directions from multiple image subsets using these kernels and combine them in a weighted mean. Overall, our method is 20% faster than the original 3DGS while obtaining the same reconstruction quality. Our optimization is also agnostic to other methods that acclerate 3DGS, thus enabling even faster speedups compared to vanilla 3DGS.", "date": "2024-09-19", "categories": []}
{"id": "2504.09588", "title": "TextSplat: Text-Guided Semantic Fusion for Generalizable Gaussian Splatting", "abstract": "Recent advancements in Generalizable Gaussian Splatting have enabled robust 3D reconstruction from sparse input views by utilizing feed-forward Gaussian Splatting models, achieving superior cross-scene generalization. However, while many methods focus on geometric consistency, they often neglect the potential of text-driven guidance to enhance semantic understanding, which is crucial for accurately reconstructing fine-grained details in complex scenes. To address this limitation, we propose TextSplat--the first text-driven Generalizable Gaussian Splatting framework. By employing a text-guided fusion of diverse semantic cues, our framework learns robust cross-modal feature representations that improve the alignment of geometric and semantic information, producing high-fidelity 3D reconstructions. Specifically, our framework employs three parallel modules to obtain complementary representations: the Diffusion Prior Depth Estimator for accurate depth information, the Semantic Aware Segmentation Network for detailed semantic information, and the Multi-View Interaction Network for refined cross-view features. Then, in the Text-Guided Semantic Fusion Module, these representations are integrated via the text-guided and attention-based feature aggregation mechanism, resulting in enhanced 3D Gaussian parameters enriched with detailed semantic cues. Experimental results on various benchmark datasets demonstrate improved performance compared to existing methods across multiple evaluation metrics, validating the effectiveness of our framework. The code will be publicly available.", "date": "2025-04-13", "categories": []}
{"id": "2508.14278", "title": "GALA: Guided Attention with Language Alignment for Open Vocabulary Gaussian Splatting", "abstract": "3D scene reconstruction and understanding have gained increasing popularity, yet existing methods still struggle to capture fine-grained, language-aware 3D representations from 2D images. In this paper, we present GALA, a novel framework for open-vocabulary 3D scene understanding with 3D Gaussian Splatting (3DGS). GALA distills a scene-specific 3D instance feature field via self-supervised contrastive learning. To extend to generalized language feature fields, we introduce the core contribution of GALA, a cross-attention module with two learnable codebooks that encode view-independent semantic embeddings. This design not only ensures intra-instance feature similarity but also supports seamless 2D and 3D open-vocabulary queries. It reduces memory consumption by avoiding per-Gaussian high-dimensional feature learning. Extensive experiments on real-world datasets demonstrate GALA's remarkable open-vocabulary performance on both 2D and 3D.", "date": "2025-08-19", "categories": []}
{"id": "2508.15158", "title": "Reliable Multi-view 3D Reconstruction for `Just-in-time' Edge Environments", "abstract": "Multi-view 3D reconstruction applications are revolutionizing critical use cases that require rapid situational-awareness, such as emergency response, tactical scenarios, and public safety. In many cases, their near-real-time latency requirements and ad-hoc needs for compute resources necessitate adoption of `Just-in-time' edge environments where the system is set up on the fly to support the applications during the mission lifetime. However, reliability issues can arise from the inherent dynamism and operational adversities of such edge environments, resulting in spatiotemporally correlated disruptions that impact the camera operations, which can lead to sustained degradation of reconstruction quality. In this paper, we propose a novel portfolio theory inspired edge resource management strategy for reliable multi-view 3D reconstruction against possible system disruptions. Our proposed methodology can guarantee reconstruction quality satisfaction even when the cameras are prone to spatiotemporally correlated disruptions. The portfolio theoretic optimization problem is solved using a genetic algorithm that converges quickly for realistic system settings. Using publicly available and customized 3D datasets, we demonstrate the proposed camera selection strategy's benefits in guaranteeing reliable 3D reconstruction against traditional baseline strategies, under spatiotemporal disruptions.", "date": "2025-08-21", "categories": []}
{"id": "2508.15672", "title": "CM2LoD3: Reconstructing LoD3 Building Models Using Semantic Conflict Maps", "abstract": "Detailed 3D building models are crucial for urban planning, digital twins, and disaster management applications. While Level of Detail 1 (LoD)1 and LoD2 building models are widely available, they lack detailed facade elements essential for advanced urban analysis. In contrast, LoD3 models address this limitation by incorporating facade elements such as windows, doors, and underpasses. However, their generation has traditionally required manual modeling, making large-scale adoption challenging. In this contribution, CM2LoD3, we present a novel method for reconstructing LoD3 building models leveraging Conflict Maps (CMs) obtained from ray-to-model-prior analysis. Unlike previous works, we concentrate on semantically segmenting real-world CMs with synthetically generated CMs from our developed Semantic Conflict Map Generator (SCMG). We also observe that additional segmentation of textured models can be fused with CMs using confidence scores to further increase segmentation performance and thus increase 3D reconstruction accuracy. Experimental results demonstrate the effectiveness of our CM2LoD3 method in segmenting and reconstructing building openings, with the 61% performance with uncertainty-aware fusion of segmented building textures. This research contributes to the advancement of automated LoD3 model reconstruction, paving the way for scalable and efficient 3D city modeling. Our project is available: https://github.com/InFraHank/CM2LoD3", "date": "2025-08-21", "categories": []}
{"id": "2406.20077", "title": "HouseCrafter: Lifting Floorplans to 3D Scenes with 2D Diffusion Model", "abstract": "We introduce HouseCrafter, a novel approach that can lift a floorplan into a complete large 3D indoor scene (e.g., a house). Our key insight is to adapt a 2D diffusion model, which is trained on web-scale images, to generate consistent multi-view color (RGB) and depth (D) images across different locations of the scene. Specifically, the RGB-D images are generated autoregressively in a batch-wise manner along sampled locations based on the floorplan, where previously generated images are used as condition to the diffusion model to produce images at nearby locations. The global floorplan and attention design in the diffusion model ensures the consistency of the generated images, from which a 3D scene can be reconstructed. Through extensive evaluation on the 3D-Front dataset, we demonstrate that HouseCraft can generate high-quality house-scale 3D scenes. Ablation studies also validate the effectiveness of different design choices. We will release our code and model weights. Project page: https://neu-vi.github.io/houseCrafter/", "date": "2024-06-28", "categories": []}
{"id": "2504.07961", "title": "Geo4D: Leveraging Video Generators for Geometric 4D Scene Reconstruction", "abstract": "We introduce Geo4D, a method to repurpose video diffusion models for monocular 3D reconstruction of dynamic scenes. By leveraging the strong dynamic priors captured by large-scale pre-trained video models, Geo4D can be trained using only synthetic data while generalizing well to real data in a zero-shot manner. Geo4D predicts several complementary geometric modalities, namely point, disparity, and ray maps. We propose a new multi-modal alignment algorithm to align and fuse these modalities, as well as a sliding window approach at inference time, thus enabling robust and accurate 4D reconstruction of long videos. Extensive experiments across multiple benchmarks show that Geo4D significantly surpasses state-of-the-art video depth estimation methods.", "date": "2025-04-10", "categories": []}
{"id": "2508.13228", "title": "PreSem-Surf: RGB-D Surface Reconstruction with Progressive Semantic Modeling and SG-MLP Pre-Rendering Mechanism", "abstract": "This paper proposes PreSem-Surf, an optimized method based on the Neural Radiance Field (NeRF) framework, capable of reconstructing high-quality scene surfaces from RGB-D sequences in a short time. The method integrates RGB, depth, and semantic information to improve reconstruction performance. Specifically, a novel SG-MLP sampling structure combined with PR-MLP (Preconditioning Multilayer Perceptron) is introduced for voxel pre-rendering, allowing the model to capture scene-related information earlier and better distinguish noise from local details. Furthermore, progressive semantic modeling is adopted to extract semantic information at increasing levels of precision, reducing training time while enhancing scene understanding. Experiments on seven synthetic scenes with six evaluation metrics show that PreSem-Surf achieves the best performance in C-L1, F-score, and IoU, while maintaining competitive results in NC, Accuracy, and Completeness, demonstrating its effectiveness and practical applicability.", "date": "2025-08-17", "categories": []}
{"id": "2411.16392", "title": "Quadratic Gaussian Splatting: High Quality Surface Reconstruction with Second-order Geometric Primitives", "abstract": "We propose Quadratic Gaussian Splatting (QGS), a novel representation that replaces static primitives with deformable quadric surfaces (e.g., ellipse, paraboloids) to capture intricate geometry. Unlike prior works that rely on Euclidean distance for primitive density modeling--a metric misaligned with surface geometry under deformation--QGS introduces geodesic distance-based density distributions. This innovation ensures that density weights adapt intrinsically to the primitive curvature, preserving consistency during shape changes (e.g., from planar disks to curved paraboloids). By solving geodesic distances in closed form on quadric surfaces, QGS enables surface-aware splatting, where a single primitive can represent complex curvature that previously required dozens of planar surfels, potentially reducing memory usage while maintaining efficient rendering via fast ray-quadric intersection. Experiments on DTU, Tanks and Temples, and MipNeRF360 datasets demonstrate state-of-the-art surface reconstruction, with QGS reducing geometric error (chamfer distance) by 33% over 2DGS and 27% over GOF on the DTU dataset. Crucially, QGS retains competitive appearance quality, bridging the gap between geometric precision and visual fidelity for applications like robotics and immersive reality.", "date": "2024-11-25", "categories": []}
{"id": "2508.12313", "title": "Improving Densification in 3D Gaussian Splatting for High-Fidelity Rendering", "abstract": "Although 3D Gaussian Splatting (3DGS) has achieved impressive performance in real-time rendering, its densification strategy often results in suboptimal reconstruction quality. In this work, we present a comprehensive improvement to the densification pipeline of 3DGS from three perspectives: when to densify, how to densify, and how to mitigate overfitting. Specifically, we propose an Edge-Aware Score to effectively select candidate Gaussians for splitting. We further introduce a Long-Axis Split strategy that reduces geometric distortions introduced by clone and split operations. To address overfitting, we design a set of techniques, including Recovery-Aware Pruning, Multi-step Update, and Growth Control. Our method enhances rendering fidelity without introducing additional training or inference overhead, achieving state-of-the-art performance with fewer Gaussians.", "date": "2025-08-17", "categories": []}
{"id": "2508.13153", "title": "IGFuse: Interactive 3D Gaussian Scene Reconstruction via Multi-Scans Fusion", "abstract": "Reconstructing complete and interactive 3D scenes remains a fundamental challenge in computer vision and robotics, particularly due to persistent object occlusions and limited sensor coverage. Multiview observations from a single scene scan often fail to capture the full structural details. Existing approaches typically rely on multi stage pipelines, such as segmentation, background completion, and inpainting or require per-object dense scanning, both of which are error-prone, and not easily scalable. We propose IGFuse, a novel framework that reconstructs interactive Gaussian scene by fusing observations from multiple scans, where natural object rearrangement between captures reveal previously occluded regions. Our method constructs segmentation aware Gaussian fields and enforces bi-directional photometric and semantic consistency across scans. To handle spatial misalignments, we introduce a pseudo-intermediate scene state for unified alignment, alongside collaborative co-pruning strategies to refine geometry. IGFuse enables high fidelity rendering and object level scene manipulation without dense observations or complex pipelines. Extensive experiments validate the framework's strong generalization to novel scene configurations, demonstrating its effectiveness for real world 3D reconstruction and real-to-simulation transfer. Our project page is available online.", "date": "2025-08-18", "categories": []}
{"id": "2412.05908", "title": "GBR: Generative Bundle Refinement for High-fidelity Gaussian Splatting with Enhanced Mesh Reconstruction", "abstract": "Gaussian splatting has gained attention for its efficient representation and rendering of 3D scenes using continuous Gaussian primitives. However, it struggles with sparse-view inputs due to limited geometric and photometric information, causing ambiguities in depth, shape, and texture.\n  we propose GBR: Generative Bundle Refinement, a method for high-fidelity Gaussian splatting and meshing using only 4-6 input views. GBR integrates a neural bundle adjustment module to enhance geometry accuracy and a generative depth refinement module to improve geometry fidelity. More specifically, the neural bundle adjustment module integrates a foundation network to produce initial 3D point maps and point matches from unposed images, followed by bundle adjustment optimization to improve multiview consistency and point cloud accuracy. The generative depth refinement module employs a diffusion-based strategy to enhance geometric details and fidelity while preserving the scale. Finally, for Gaussian splatting optimization, we propose a multimodal loss function incorporating depth and normal consistency, geometric regularization, and pseudo-view supervision, providing robust guidance under sparse-view conditions. Experiments on widely used datasets show that GBR significantly outperforms existing methods under sparse-view inputs. Additionally, GBR demonstrates the ability to reconstruct and render large-scale real-world scenes, such as the Pavilion of Prince Teng and the Great Wall, with remarkable details using only 6 views.", "date": "2024-12-08", "categories": []}
{"id": "2503.06587", "title": "Introducing Unbiased Depth into 2D Gaussian Splatting for High-accuracy Surface Reconstruction", "abstract": "Recently, 2D Gaussian Splatting (2DGS) has demonstrated superior geometry reconstruction quality than the popular 3DGS by using 2D surfels to approximate thin surfaces. However, it falls short when dealing with glossy surfaces, resulting in visible holes in these areas. We find that the reflection discontinuity causes the issue. To fit the jump from diffuse to specular reflection at different viewing angles, depth bias is introduced in the optimized Gaussian primitives. To address that, we first replace the depth distortion loss in 2DGS with a novel depth convergence loss, which imposes a strong constraint on depth continuity. Then, we rectify the depth criterion in determining the actual surface, which fully accounts for all the intersecting Gaussians along the ray. Qualitative and quantitative evaluations across various datasets reveal that our method significantly improves reconstruction quality, with more complete and accurate surfaces than 2DGS. Code is available at https://github.com/XiaoXinyyx/Unbiased_Surfel.", "date": "2025-03-09", "categories": []}
{"id": "2412.03910", "title": "DGNS: Deformable Gaussian Splatting and Dynamic Neural Surface for Monocular Dynamic 3D Reconstruction", "abstract": "Dynamic scene reconstruction from monocular video is essential for real-world applications. We introduce DGNS, a hybrid framework integrating \\underline{D}eformable \\underline{G}aussian Splatting and Dynamic \\underline{N}eural \\underline{S}urfaces, effectively addressing dynamic novel-view synthesis and 3D geometry reconstruction simultaneously. During training, depth maps generated by the deformable Gaussian splatting module guide the ray sampling for faster processing and provide depth supervision within the dynamic neural surface module to improve geometry reconstruction. Conversely, the dynamic neural surface directs the distribution of Gaussian primitives around the surface, enhancing rendering quality. In addition, we propose a depth-filtering approach to further refine depth supervision. Extensive experiments conducted on public datasets demonstrate that DGNS achieves state-of-the-art performance in 3D reconstruction, along with competitive results in novel-view synthesis.", "date": "2024-12-05", "categories": []}
{"id": "2508.10507", "title": "Multi-Sample Anti-Aliasing and Constrained Optimization for 3D Gaussian Splatting", "abstract": "Recent advances in 3D Gaussian splatting have significantly improved real-time novel view synthesis, yet insufficient geometric constraints during scene optimization often result in blurred reconstructions of fine-grained details, particularly in regions with high-frequency textures and sharp discontinuities. To address this, we propose a comprehensive optimization framework integrating multisample anti-aliasing (MSAA) with dual geometric constraints. Our system computes pixel colors through adaptive blending of quadruple subsamples, effectively reducing aliasing artifacts in high-frequency components. The framework introduces two constraints: (a) an adaptive weighting strategy that prioritizes under-reconstructed regions through dynamic gradient analysis, and (b) gradient differential constraints enforcing geometric regularization at object boundaries. This targeted optimization enables the model to allocate computational resources preferentially to critical regions requiring refinement while maintaining global consistency. Extensive experimental evaluations across multiple benchmarks demonstrate that our method achieves state-of-the-art performance in detail preservation, particularly in preserving high-frequency textures and sharp discontinuities, while maintaining real-time rendering efficiency. Quantitative metrics and perceptual studies confirm statistically significant improvements over baseline approaches in both structural similarity (SSIM) and perceptual quality (LPIPS).", "date": "2025-08-14", "categories": []}
{"id": "2508.10893", "title": "STream3R: Scalable Sequential 3D Reconstruction with Causal Transformer", "abstract": "We present STream3R, a novel approach to 3D reconstruction that reformulates pointmap prediction as a decoder-only Transformer problem. Existing state-of-the-art methods for multi-view reconstruction either depend on expensive global optimization or rely on simplistic memory mechanisms that scale poorly with sequence length. In contrast, STream3R introduces an streaming framework that processes image sequences efficiently using causal attention, inspired by advances in modern language modeling. By learning geometric priors from large-scale 3D datasets, STream3R generalizes well to diverse and challenging scenarios, including dynamic scenes where traditional methods often fail. Extensive experiments show that our method consistently outperforms prior work across both static and dynamic scene benchmarks. Moreover, STream3R is inherently compatible with LLM-style training infrastructure, enabling efficient large-scale pretraining and fine-tuning for various downstream 3D tasks. Our results underscore the potential of causal Transformer models for online 3D perception, paving the way for real-time 3D understanding in streaming environments. More details can be found in our project page: https://nirvanalan.github.io/projects/stream3r.", "date": "2025-08-14", "categories": []}
{"id": "2508.09667", "title": "GSFixer: Improving 3D Gaussian Splatting with Reference-Guided Video Diffusion Priors", "abstract": "Reconstructing 3D scenes using 3D Gaussian Splatting (3DGS) from sparse views is an ill-posed problem due to insufficient information, often resulting in noticeable artifacts. While recent approaches have sought to leverage generative priors to complete information for under-constrained regions, they struggle to generate content that remains consistent with input observations. To address this challenge, we propose GSFixer, a novel framework designed to improve the quality of 3DGS representations reconstructed from sparse inputs. The core of our approach is the reference-guided video restoration model, built upon a DiT-based video diffusion model trained on paired artifact 3DGS renders and clean frames with additional reference-based conditions. Considering the input sparse views as references, our model integrates both 2D semantic features and 3D geometric features of reference views extracted from the visual geometry foundation model, enhancing the semantic coherence and 3D consistency when fixing artifact novel views. Furthermore, considering the lack of suitable benchmarks for 3DGS artifact restoration evaluation, we present DL3DV-Res which contains artifact frames rendered using low-quality 3DGS. Extensive experiments demonstrate our GSFixer outperforms current state-of-the-art methods in 3DGS artifact restoration and sparse-view 3D reconstruction. Project page: https://github.com/GVCLab/GSFixer.", "date": "2025-08-13", "categories": []}
{"id": "2508.09811", "title": "TRACE: Learning 3D Gaussian Physical Dynamics from Multi-view Videos", "abstract": "In this paper, we aim to model 3D scene geometry, appearance, and physical information just from dynamic multi-view videos in the absence of any human labels. By leveraging physics-informed losses as soft constraints or integrating simple physics models into neural nets, existing works often fail to learn complex motion physics, or doing so requires additional labels such as object types or masks. We propose a new framework named TRACE to model the motion physics of complex dynamic 3D scenes. The key novelty of our method is that, by formulating each 3D point as a rigid particle with size and orientation in space, we directly learn a translation rotation dynamics system for each particle, explicitly estimating a complete set of physical parameters to govern the particle's motion over time. Extensive experiments on three existing dynamic datasets and one newly created challenging synthetic datasets demonstrate the extraordinary performance of our method over baselines in the task of future frame extrapolation. A nice property of our framework is that multiple objects or parts can be easily segmented just by clustering the learned physical parameters.", "date": "2025-08-13", "categories": []}
{"id": "2508.09830", "title": "RayletDF: Raylet Distance Fields for Generalizable 3D Surface Reconstruction from Point Clouds or Gaussians", "abstract": "In this paper, we present a generalizable method for 3D surface reconstruction from raw point clouds or pre-estimated 3D Gaussians by 3DGS from RGB images. Unlike existing coordinate-based methods which are often computationally intensive when rendering explicit surfaces, our proposed method, named RayletDF, introduces a new technique called raylet distance field, which aims to directly predict surface points from query rays. Our pipeline consists of three key modules: a raylet feature extractor, a raylet distance field predictor, and a multi-raylet blender. These components work together to extract fine-grained local geometric features, predict raylet distances, and aggregate multiple predictions to reconstruct precise surface points. We extensively evaluate our method on multiple public real-world datasets, demonstrating superior performance in surface reconstruction from point clouds or 3D Gaussians. Most notably, our method achieves exceptional generalization ability, successfully recovering 3D surfaces in a single-forward pass across unseen datasets in testing.", "date": "2025-08-13", "categories": []}
{"id": "2503.14786", "title": "SketchSplat: 3D Edge Reconstruction via Differentiable Multi-view Sketch Splatting", "abstract": "Edges are one of the most basic parametric primitives to describe structural information in 3D. In this paper, we study parametric 3D edge reconstruction from calibrated multi-view images. Previous methods usually reconstruct a 3D edge point set from multi-view 2D edge images, and then fit 3D edges to the point set. However, noise in the point set may cause gaps among fitted edges, and the recovered edges may not align with input multi-view images since the edge fitting depends only on the reconstructed 3D point set. To mitigate these problems, we propose SketchSplat, a method to reconstruct accurate, complete, and compact 3D edges via differentiable multi-view sketch splatting. We represent 3D edges as sketches, which are parametric lines and curves defined by attributes including control points, scales, and opacity. During reconstruction, we iteratively sample Gaussian points from a set of sketches and rasterize the Gaussians onto 2D edge images. Then the gradient of the image loss can be back-propagated to optimize the sketch attributes. Our method bridges 2D edge images and 3D edges in a differentiable manner, which ensures that 3D edges align well with 2D images and leads to accurate and complete results. We also propose a series of adaptive topological operations to reduce redundant edges and apply them along with the sketch optimization, yielding a more compact reconstruction. Finally, we contribute an accurate 2D edge detector that improves the performance of both ours and existing methods. Experiments show that our method achieves state-of-the-art accuracy, completeness, and compactness on a benchmark CAD dataset.", "date": "2025-03-18", "categories": []}
{"id": "2508.07908", "title": "Mem4D: Decoupling Static and Dynamic Memory for Dynamic Scene Reconstruction", "abstract": "Reconstructing dense geometry for dynamic scenes from a monocular video is a critical yet challenging task. Recent memory-based methods enable efficient online reconstruction, but they fundamentally suffer from a Memory Demand Dilemma: The memory representation faces an inherent conflict between the long-term stability required for static structures and the rapid, high-fidelity detail retention needed for dynamic motion. This conflict forces existing methods into a compromise, leading to either geometric drift in static structures or blurred, inaccurate reconstructions of dynamic objects. To address this dilemma, we propose Mem4D, a novel framework that decouples the modeling of static geometry and dynamic motion. Guided by this insight, we design a dual-memory architecture: 1) The Transient Dynamics Memory (TDM) focuses on capturing high-frequency motion details from recent frames, enabling accurate and fine-grained modeling of dynamic content; 2) The Persistent Structure Memory (PSM) compresses and preserves long-term spatial information, ensuring global consistency and drift-free reconstruction for static elements. By alternating queries to these specialized memories, Mem4D simultaneously maintains static geometry with global consistency and reconstructs dynamic elements with high fidelity. Experiments on challenging benchmarks demonstrate that our method achieves state-of-the-art or competitive performance while maintaining high efficiency. Codes will be publicly available.", "date": "2025-08-11", "categories": []}
{"id": "2505.05591", "title": "QuickSplat: Fast 3D Surface Reconstruction via Learned Gaussian Initialization", "abstract": "Surface reconstruction is fundamental to computer vision and graphics, enabling applications in 3D modeling, mixed reality, robotics, and more. Existing approaches based on volumetric rendering obtain promising results, but optimize on a per-scene basis, resulting in a slow optimization that can struggle to model under-observed or textureless regions. We introduce QuickSplat, which learns data-driven priors to generate dense initializations for 2D gaussian splatting optimization of large-scale indoor scenes. This provides a strong starting point for the reconstruction, which accelerates the convergence of the optimization and improves the geometry of flat wall structures. We further learn to jointly estimate the densification and update of the scene parameters during each iteration; our proposed densifier network predicts new Gaussians based on the rendering gradients of existing ones, removing the needs of heuristics for densification. Extensive experiments on large-scale indoor scene reconstruction demonstrate the superiority of our data-driven optimization. Concretely, we accelerate runtime by 8x, while decreasing depth errors by up to 48% in comparison to state of the art methods.", "date": "2025-05-08", "categories": []}
{"id": "2505.17966", "title": "Is Single-View Mesh Reconstruction Ready for Robotics?", "abstract": "This paper evaluates single-view mesh reconstruction models for their potential in enabling instant digital twin creation for real-time planning and dynamics prediction using physics simulators for robotic manipulation. Recent single-view 3D reconstruction advances offer a promising avenue toward an automated real-to-sim pipeline: directly mapping a single observation of a scene into a simulation instance by reconstructing scene objects as individual, complete, and physically plausible 3D meshes. However, their suitability for physics simulations and robotics applications under immediacy, physical fidelity, and simulation readiness remains underexplored. We establish robotics-specific benchmarking criteria for 3D reconstruction, including handling typical inputs, collision-free and stable geometry, occlusions robustness, and meeting computational constraints. Our empirical evaluation using realistic robotics datasets shows that despite success on computer vision benchmarks, existing approaches fail to meet robotics-specific requirements. We quantitively examine limitations of single-view reconstruction for practical robotics implementation, in contrast to prior work that focuses on multi-view approaches. Our findings highlight critical gaps between computer vision advances and robotics needs, guiding future research at this intersection.", "date": "2025-05-23", "categories": []}
{"id": "2508.06715", "title": "Restage4D: Reanimating Deformable 3D Reconstruction from a Single Video", "abstract": "Creating deformable 3D content has gained increasing attention with the rise of text-to-image and image-to-video generative models. While these models provide rich semantic priors for appearance, they struggle to capture the physical realism and motion dynamics needed for authentic 4D scene synthesis. In contrast, real-world videos can provide physically grounded geometry and articulation cues that are difficult to hallucinate. One question is raised: \\textit{Can we generate physically consistent 4D content by leveraging the motion priors of the real-world video}? In this work, we explore the task of reanimating deformable 3D scenes from a single video, using the original sequence as a supervisory signal to correct artifacts from synthetic motion. We introduce \\textbf{Restage4D}, a geometry-preserving pipeline for video-conditioned 4D restaging. Our approach uses a video-rewinding training strategy to temporally bridge a real base video and a synthetic driving video via a shared motion representation. We further incorporate an occlusion-aware rigidity loss and a disocclusion backtracing mechanism to improve structural and geometry consistency under challenging motion. We validate Restage4D on DAVIS and PointOdyssey, demonstrating improved geometry consistency, motion quality, and 3D tracking performance. Our method not only preserves deformable structure under novel motion, but also automatically corrects errors introduced by generative models, revealing the potential of video prior in 4D restaging task. Source code and trained models will be released.", "date": "2025-08-08", "categories": []}
{"id": "2508.07355", "title": "GS4Buildings: Prior-Guided Gaussian Splatting for 3D Building Reconstruction", "abstract": "Recent advances in Gaussian Splatting (GS) have demonstrated its effectiveness in photo-realistic rendering and 3D reconstruction. Among these, 2D Gaussian Splatting (2DGS) is particularly suitable for surface reconstruction due to its flattened Gaussian representation and integrated normal regularization. However, its performance often degrades in large-scale and complex urban scenes with frequent occlusions, leading to incomplete building reconstructions. We propose GS4Buildings, a novel prior-guided Gaussian Splatting method leveraging the ubiquity of semantic 3D building models for robust and scalable building surface reconstruction. Instead of relying on traditional Structure-from-Motion (SfM) pipelines, GS4Buildings initializes Gaussians directly from low-level Level of Detail (LoD)2 semantic 3D building models. Moreover, we generate prior depth and normal maps from the planar building geometry and incorporate them into the optimization process, providing strong geometric guidance for surface consistency and structural accuracy. We also introduce an optional building-focused mode that limits reconstruction to building regions, achieving a 71.8% reduction in Gaussian primitives and enabling a more efficient and compact representation. Experiments on urban datasets demonstrate that GS4Buildings improves reconstruction completeness by 20.5% and geometric accuracy by 32.8%. These results highlight the potential of semantic building model integration to advance GS-based reconstruction toward real-world urban applications such as smart cities and digital twins. Our project is available: https://github.com/zqlin0521/GS4Buildings.", "date": "2025-08-10", "categories": []}
{"id": "2508.07372", "title": "DIP-GS: Deep Image Prior For Gaussian Splatting Sparse View Recovery", "abstract": "3D Gaussian Splatting (3DGS) is a leading 3D scene reconstruction method, obtaining high-quality reconstruction with real-time rendering runtime performance. The main idea behind 3DGS is to represent the scene as a collection of 3D gaussians, while learning their parameters to fit the given views of the scene. While achieving superior performance in the presence of many views, 3DGS struggles with sparse view reconstruction, where the input views are sparse and do not fully cover the scene and have low overlaps. In this paper, we propose DIP-GS, a Deep Image Prior (DIP) 3DGS representation. By using the DIP prior, which utilizes internal structure and patterns, with coarse-to-fine manner, DIP-based 3DGS can operate in scenarios where vanilla 3DGS fails, such as sparse view recovery. Note that our approach does not use any pre-trained models such as generative models and depth estimation, but rather relies only on the input frames. Among such methods, DIP-GS obtains state-of-the-art (SOTA) competitive results on various sparse-view reconstruction tasks, demonstrating its capabilities.", "date": "2025-08-10", "categories": []}
{"id": "2508.07483", "title": "Novel View Synthesis with Gaussian Splatting: Impact on Photogrammetry Model Accuracy and Resolution", "abstract": "In this paper, I present a comprehensive study comparing Photogrammetry and Gaussian Splatting techniques for 3D model reconstruction and view synthesis. I created a dataset of images from a real-world scene and constructed 3D models using both methods. To evaluate the performance, I compared the models using structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), learned perceptual image patch similarity (LPIPS), and lp/mm resolution based on the USAF resolution chart. A significant contribution of this work is the development of a modified Gaussian Splatting repository, which I forked and enhanced to enable rendering images from novel camera poses generated in the Blender environment. This innovation allows for the synthesis of high-quality novel views, showcasing the flexibility and potential of Gaussian Splatting. My investigation extends to an augmented dataset that includes both original ground images and novel views synthesized via Gaussian Splatting. This augmented dataset was employed to generate a new photogrammetry model, which was then compared against the original photogrammetry model created using only the original images. The results demonstrate the efficacy of using Gaussian Splatting to generate novel high-quality views and its potential to improve photogrammetry-based 3D reconstructions. The comparative analysis highlights the strengths and limitations of both approaches, providing valuable information for applications in extended reality (XR), photogrammetry, and autonomous vehicle simulations. Code is available at https://github.com/pranavc2255/gaussian-splatting-novel-view-render.git.", "date": "2025-08-10", "categories": []}
{"id": "2508.08086", "title": "Matrix-3D: Omnidirectional Explorable 3D World Generation", "abstract": "Explorable 3D world generation from a single image or text prompt forms a cornerstone of spatial intelligence. Recent works utilize video model to achieve wide-scope and generalizable 3D world generation. However, existing approaches often suffer from a limited scope in the generated scenes. In this work, we propose Matrix-3D, a framework that utilize panoramic representation for wide-coverage omnidirectional explorable 3D world generation that combines conditional video generation and panoramic 3D reconstruction. We first train a trajectory-guided panoramic video diffusion model that employs scene mesh renders as condition, to enable high-quality and geometrically consistent scene video generation. To lift the panorama scene video to 3D world, we propose two separate methods: (1) a feed-forward large panorama reconstruction model for rapid 3D scene reconstruction and (2) an optimization-based pipeline for accurate and detailed 3D scene reconstruction. To facilitate effective training, we also introduce the Matrix-Pano dataset, the first large-scale synthetic collection comprising 116K high-quality static panoramic video sequences with depth and trajectory annotations. Extensive experiments demonstrate that our proposed framework achieves state-of-the-art performance in panoramic video generation and 3D world generation. See more in https://matrix-3d.github.io.", "date": "2025-08-11", "categories": []}
{"id": "2504.05692", "title": "POMATO: Marrying Pointmap Matching with Temporal Motion for Dynamic 3D Reconstruction", "abstract": "3D reconstruction in dynamic scenes primarily relies on the combination of geometry estimation and matching modules where the latter task is pivotal for distinguishing dynamic regions which can help to mitigate the interference introduced by camera and object motion. Furthermore, the matching module explicitly models object motion, enabling the tracking of specific targets and advancing motion understanding in complex scenarios. Recently, the proposed representation of pointmap in DUSt3R suggests a potential solution to unify both geometry estimation and matching in 3D space, but it still struggles with ambiguous matching in dynamic regions, which may hamper further improvement. In this work, we present POMATO, a unified framework for dynamic 3D reconstruction by marrying pointmap matching with temporal motion. Specifically, our method first learns an explicit matching relationship by mapping RGB pixels from both dynamic and static regions across different views to 3D pointmaps within a unified coordinate system. Furthermore, we introduce a temporal motion module for dynamic motions that ensures scale consistency across different frames and enhances performance in tasks requiring both precise geometry and reliable matching, most notably 3D point tracking. We show the effectiveness of the proposed pointmap matching and temporal fusion paradigm by demonstrating the remarkable performance across multiple downstream tasks, including video depth estimation, 3D point tracking, and pose estimation. Code and models are publicly available at https://github.com/wyddmw/POMATO.", "date": "2025-04-08", "categories": []}
{"id": "2508.05907", "title": "Neural Field Representations of Mobile Computational Photography", "abstract": "Over the past two decades, mobile imaging has experienced a profound transformation, with cell phones rapidly eclipsing all other forms of digital photography in popularity. Today's cell phones are equipped with a diverse range of imaging technologies - laser depth ranging, multi-focal camera arrays, and split-pixel sensors - alongside non-visual sensors such as gyroscopes, accelerometers, and magnetometers. This, combined with on-board integrated chips for image and signal processing, makes the cell phone a versatile pocket-sized computational imaging platform. Parallel to this, we have seen in recent years how neural fields - small neural networks trained to map continuous spatial input coordinates to output signals - enable the reconstruction of complex scenes without explicit data representations such as pixel arrays or point clouds. In this thesis, I demonstrate how carefully designed neural field models can compactly represent complex geometry and lighting effects. Enabling applications such as depth estimation, layer separation, and image stitching directly from collected in-the-wild mobile photography data. These methods outperform state-of-the-art approaches without relying on complex pre-processing steps, labeled ground truth data, or machine learning priors. Instead, they leverage well-constructed, self-regularized models that tackle challenging inverse problems through stochastic gradient descent, fitting directly to raw measurements from a smartphone.", "date": "2025-08-08", "categories": []}
{"id": "2508.06014", "title": "ExploreGS: Explorable 3D Scene Reconstruction with Virtual Camera Samplings and Diffusion Priors", "abstract": "Recent advances in novel view synthesis (NVS) have enabled real-time rendering with 3D Gaussian Splatting (3DGS). However, existing methods struggle with artifacts and missing regions when rendering from viewpoints that deviate from the training trajectory, limiting seamless scene exploration. To address this, we propose a 3DGS-based pipeline that generates additional training views to enhance reconstruction. We introduce an information-gain-driven virtual camera placement strategy to maximize scene coverage, followed by video diffusion priors to refine rendered results. Fine-tuning 3D Gaussians with these enhanced views significantly improves reconstruction quality. To evaluate our method, we present Wild-Explore, a benchmark designed for challenging scene exploration. Experiments demonstrate that our approach outperforms existing 3DGS-based methods, enabling high-quality, artifact-free rendering from arbitrary viewpoints.\n  https://exploregs.github.io", "date": "2025-08-08", "categories": []}
{"id": "2508.06169", "title": "UW-3DGS: Underwater 3D Reconstruction with Physics-Aware Gaussian Splatting", "abstract": "Underwater 3D scene reconstruction faces severe challenges from light absorption, scattering, and turbidity, which degrade geometry and color fidelity in traditional methods like Neural Radiance Fields (NeRF). While NeRF extensions such as SeaThru-NeRF incorporate physics-based models, their MLP reliance limits efficiency and spatial resolution in hazy environments. We introduce UW-3DGS, a novel framework adapting 3D Gaussian Splatting (3DGS) for robust underwater reconstruction. Key innovations include: (1) a plug-and-play learnable underwater image formation module using voxel-based regression for spatially varying attenuation and backscatter; and (2) a Physics-Aware Uncertainty Pruning (PAUP) branch that adaptively removes noisy floating Gaussians via uncertainty scoring, ensuring artifact-free geometry. The pipeline operates in training and rendering stages. During training, noisy Gaussians are optimized end-to-end with underwater parameters, guided by PAUP pruning and scattering modeling. In rendering, refined Gaussians produce clean Unattenuated Radiance Images (URIs) free from media effects, while learned physics enable realistic Underwater Images (UWIs) with accurate light transport. Experiments on SeaThru-NeRF and UWBundle datasets show superior performance, achieving PSNR of 27.604, SSIM of 0.868, and LPIPS of 0.104 on SeaThru-NeRF, with ~65% reduction in floating artifacts.", "date": "2025-08-08", "categories": []}
{"id": "2406.05649", "title": "GTR: Improving Large 3D Reconstruction Models through Geometry and Texture Refinement", "abstract": "We propose a novel approach for 3D mesh reconstruction from multi-view images. Our method takes inspiration from large reconstruction models like LRM that use a transformer-based triplane generator and a Neural Radiance Field (NeRF) model trained on multi-view images. However, in our method, we introduce several important modifications that allow us to significantly enhance 3D reconstruction quality. First of all, we examine the original LRM architecture and find several shortcomings. Subsequently, we introduce respective modifications to the LRM architecture, which lead to improved multi-view image representation and more computationally efficient training. Second, in order to improve geometry reconstruction and enable supervision at full image resolution, we extract meshes from the NeRF field in a differentiable manner and fine-tune the NeRF model through mesh rendering. These modifications allow us to achieve state-of-the-art performance on both 2D and 3D evaluation metrics, such as a PSNR of 28.67 on Google Scanned Objects (GSO) dataset. Despite these superior results, our feed-forward model still struggles to reconstruct complex textures, such as text and portraits on assets. To address this, we introduce a lightweight per-instance texture refinement procedure. This procedure fine-tunes the triplane representation and the NeRF color estimation model on the mesh surface using the input multi-view images in just 4 seconds. This refinement improves the PSNR to 29.79 and achieves faithful reconstruction of complex textures, such as text. Additionally, our approach enables various downstream applications, including text- or image-to-3D generation.", "date": "2024-06-09", "categories": []}
{"id": "2508.04966", "title": "Laplacian Analysis Meets Dynamics Modelling: Gaussian Splatting for 4D Reconstruction", "abstract": "While 3D Gaussian Splatting (3DGS) excels in static scene modeling, its extension to dynamic scenes introduces significant challenges. Existing dynamic 3DGS methods suffer from either over-smoothing due to low-rank decomposition or feature collision from high-dimensional grid sampling. This is because of the inherent spectral conflicts between preserving motion details and maintaining deformation consistency at different frequency. To address these challenges, we propose a novel dynamic 3DGS framework with hybrid explicit-implicit functions. Our approach contains three key innovations: a spectral-aware Laplacian encoding architecture which merges Hash encoding and Laplacian-based module for flexible frequency motion control, an enhanced Gaussian dynamics attribute that compensates for photometric distortions caused by geometric deformation, and an adaptive Gaussian split strategy guided by KDTree-based primitive control to efficiently query and optimize dynamic areas. Through extensive experiments, our method demonstrates state-of-the-art performance in reconstructing complex dynamic scenes, achieving better reconstruction fidelity.", "date": "2025-08-07", "categories": []}
{"id": "2508.04968", "title": "UGOD: Uncertainty-Guided Differentiable Opacity and Soft Dropout for Enhanced Sparse-View 3DGS", "abstract": "3D Gaussian Splatting (3DGS) has become a competitive approach for novel view synthesis (NVS) due to its advanced rendering efficiency through 3D Gaussian projection and blending. However, Gaussians are treated equally weighted for rendering in most 3DGS methods, making them prone to overfitting, which is particularly the case in sparse-view scenarios. To address this, we investigate how adaptive weighting of Gaussians affects rendering quality, which is characterised by learned uncertainties proposed. This learned uncertainty serves two key purposes: first, it guides the differentiable update of Gaussian opacity while preserving the 3DGS pipeline integrity; second, the uncertainty undergoes soft differentiable dropout regularisation, which strategically transforms the original uncertainty into continuous drop probabilities that govern the final Gaussian projection and blending process for rendering. Extensive experimental results over widely adopted datasets demonstrate that our method outperforms rivals in sparse-view 3D synthesis, achieving higher quality reconstruction with fewer Gaussians in most datasets compared to existing sparse-view approaches, e.g., compared to DropGaussian, our method achieves 3.27\\% PSNR improvements on the MipNeRF 360 dataset.", "date": "2025-08-07", "categories": []}
{"id": "2508.05187", "title": "Refining Gaussian Splatting: A Volumetric Densification Approach", "abstract": "Achieving high-quality novel view synthesis in 3D Gaussian Splatting (3DGS) often depends on effective point primitive management. The underlying Adaptive Density Control (ADC) process addresses this issue by automating densification and pruning. Yet, the vanilla 3DGS densification strategy shows key shortcomings. To address this issue, in this paper we introduce a novel density control method, which exploits the volumes of inertia associated to each Gaussian function to guide the refinement process. Furthermore, we study the effect of both traditional Structure from Motion (SfM) and Deep Image Matching (DIM) methods for point cloud initialization. Extensive experimental evaluations on the Mip-NeRF 360 dataset demonstrate that our approach surpasses 3DGS in reconstruction quality, delivering encouraging performance across diverse scenes.", "date": "2025-08-07", "categories": []}
{"id": "2412.20720", "title": "4D Gaussian Splatting: Modeling Dynamic Scenes with Native 4D Primitives", "abstract": "Dynamic 3D scene representation and novel view synthesis are crucial for enabling immersive experiences required by AR/VR and metaverse applications. It is a challenging task due to the complexity of unconstrained real-world scenes and their temporal dynamics. In this paper, we reformulate the reconstruction of a time-varying 3D scene as approximating its underlying spatiotemporal 4D volume by optimizing a collection of native 4D primitives, i.e., 4D Gaussians, with explicit geometry and appearance modeling. Equipped with a tailored rendering pipeline, our representation can be end-to-end optimized using only photometric supervision while free viewpoint viewing at interactive frame rate, making it suitable for representing real world scene with complex dynamic. This approach has been the first solution to achieve real-time rendering of high-resolution, photorealistic novel views for complex dynamic scenes. To facilitate real-world applications, we derive several compact variants that effectively reduce the memory footprint to address its storage bottleneck. Extensive experiments validate the superiority of 4DGS in terms of visual quality and efficiency across a range of dynamic scene-related tasks (e.g., novel view synthesis, 4D generation, scene understanding) and scenarios (e.g., single object, indoor scenes, driving environments, synthetic and real data).", "date": "2024-12-30", "categories": []}
{"id": "2508.04078", "title": "RLGS: Reinforcement Learning-Based Adaptive Hyperparameter Tuning for Gaussian Splatting", "abstract": "Hyperparameter tuning in 3D Gaussian Splatting (3DGS) is a labor-intensive and expert-driven process, often resulting in inconsistent reconstructions and suboptimal results. We propose RLGS, a plug-and-play reinforcement learning framework for adaptive hyperparameter tuning in 3DGS through lightweight policy modules, dynamically adjusting critical hyperparameters such as learning rates and densification thresholds. The framework is model-agnostic and seamlessly integrates into existing 3DGS pipelines without architectural modifications. We demonstrate its generalization ability across multiple state-of-the-art 3DGS variants, including Taming-3DGS and 3DGS-MCMC, and validate its robustness across diverse datasets. RLGS consistently enhances rendering quality. For example, it improves Taming-3DGS by 0.7dB PSNR on the Tanks and Temple (TNT) dataset, under a fixed Gaussian budget, and continues to yield gains even when baseline performance saturates. Our results suggest that RLGS provides an effective and general solution for automating hyperparameter tuning in 3DGS training, bridging a gap in applying reinforcement learning to 3DGS.", "date": "2025-08-06", "categories": []}
{"id": "2508.04099", "title": "DET-GS: Depth- and Edge-Aware Regularization for High-Fidelity 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) represents a significant advancement in the field of efficient and high-fidelity novel view synthesis. Despite recent progress, achieving accurate geometric reconstruction under sparse-view conditions remains a fundamental challenge. Existing methods often rely on non-local depth regularization, which fails to capture fine-grained structures and is highly sensitive to depth estimation noise. Furthermore, traditional smoothing methods neglect semantic boundaries and indiscriminately degrade essential edges and textures, consequently limiting the overall quality of reconstruction. In this work, we propose DET-GS, a unified depth and edge-aware regularization framework for 3D Gaussian Splatting. DET-GS introduces a hierarchical geometric depth supervision framework that adaptively enforces multi-level geometric consistency, significantly enhancing structural fidelity and robustness against depth estimation noise. To preserve scene boundaries, we design an edge-aware depth regularization guided by semantic masks derived from Canny edge detection. Furthermore, we introduce an RGB-guided edge-preserving Total Variation loss that selectively smooths homogeneous regions while rigorously retaining high-frequency details and textures. Extensive experiments demonstrate that DET-GS achieves substantial improvements in both geometric accuracy and visual fidelity, outperforming state-of-the-art (SOTA) methods on sparse-view novel view synthesis benchmarks.", "date": "2025-08-06", "categories": []}
{"id": "2508.04224", "title": "SplitGaussian: Reconstructing Dynamic Scenes via Visual Geometry Decomposition", "abstract": "Reconstructing dynamic 3D scenes from monocular video remains fundamentally challenging due to the need to jointly infer motion, structure, and appearance from limited observations. Existing dynamic scene reconstruction methods based on Gaussian Splatting often entangle static and dynamic elements in a shared representation, leading to motion leakage, geometric distortions, and temporal flickering. We identify that the root cause lies in the coupled modeling of geometry and appearance across time, which hampers both stability and interpretability. To address this, we propose \\textbf{SplitGaussian}, a novel framework that explicitly decomposes scene representations into static and dynamic components. By decoupling motion modeling from background geometry and allowing only the dynamic branch to deform over time, our method prevents motion artifacts in static regions while supporting view- and time-dependent appearance refinement. This disentangled design not only enhances temporal consistency and reconstruction fidelity but also accelerates convergence. Extensive experiments demonstrate that SplitGaussian outperforms prior state-of-the-art methods in rendering quality, geometric stability, and motion separation.", "date": "2025-08-06", "categories": []}
{"id": "2508.04508", "title": "Surf3R: Rapid Surface Reconstruction from Sparse RGB Views in Seconds", "abstract": "Current multi-view 3D reconstruction methods rely on accurate camera calibration and pose estimation, requiring complex and time-intensive pre-processing that hinders their practical deployment. To address this challenge, we introduce Surf3R, an end-to-end feedforward approach that reconstructs 3D surfaces from sparse views without estimating camera poses and completes an entire scene in under 10 seconds. Our method employs a multi-branch and multi-view decoding architecture in which multiple reference views jointly guide the reconstruction process. Through the proposed branch-wise processing, cross-view attention, and inter-branch fusion, the model effectively captures complementary geometric cues without requiring camera calibration. Moreover, we introduce a D-Normal regularizer based on an explicit 3D Gaussian representation for surface reconstruction. It couples surface normals with other geometric parameters to jointly optimize the 3D geometry, significantly improving 3D consistency and surface detail accuracy. Experimental results demonstrate that Surf3R achieves state-of-the-art performance on multiple surface reconstruction metrics on ScanNet++ and Replica datasets, exhibiting excellent generalization and efficiency.", "date": "2025-08-06", "categories": []}
{"id": "2404.16429", "title": "Depth Supervised Neural Surface Reconstruction from Airborne Imagery", "abstract": "While originally developed for novel view synthesis, Neural Radiance Fields (NeRFs) have recently emerged as an alternative to multi-view stereo (MVS). Triggered by a manifold of research activities, promising results have been gained especially for texture-less, transparent, and reflecting surfaces, while such scenarios remain challenging for traditional MVS-based approaches. However, most of these investigations focus on close-range scenarios, with studies for airborne scenarios still missing. For this task, NeRFs face potential difficulties at areas of low image redundancy and weak data evidence, as often found in street canyons, facades or building shadows. Furthermore, training such networks is computationally expensive. Thus, the aim of our work is twofold: First, we investigate the applicability of NeRFs for aerial image blocks representing different characteristics like nadir-only, oblique and high-resolution imagery. Second, during these investigations we demonstrate the benefit of integrating depth priors from tie-point measures, which are provided during presupposed Bundle Block Adjustment. Our work is based on the state-of-the-art framework VolSDF, which models 3D scenes by signed distance functions (SDFs), since this is more applicable for surface reconstruction compared to the standard volumetric representation in vanilla NeRFs. For evaluation, the NeRF-based reconstructions are compared to results of a publicly available benchmark dataset for airborne images.", "date": "2024-04-25", "categories": []}
{"id": "2411.17044", "title": "4D Scaffold Gaussian Splatting with Dynamic-Aware Anchor Growing for Efficient and High-Fidelity Dynamic Scene Reconstruction", "abstract": "Modeling dynamic scenes through 4D Gaussians offers high visual fidelity and fast rendering speeds, but comes with significant storage overhead. Recent approaches mitigate this cost by aggressively reducing the number of Gaussians. However, this inevitably removes Gaussians essential for high-quality rendering, leading to severe degradation in dynamic regions. In this paper, we introduce a novel 4D anchor-based framework that tackles the storage cost in different perspective. Rather than reducing the number of Gaussians, our method retains a sufficient quantity to accurately model dynamic contents, while compressing them into compact, grid-aligned 4D anchor features. Each anchor is processed by an MLP to spawn a set of neural 4D Gaussians, which represent a local spatiotemporal region. We design these neural 4D Gaussians to capture temporal changes with minimal parameters, making them well-suited for the MLP-based spawning. Moreover, we introduce a dynamic-aware anchor growing strategy to effectively assign additional anchors to under-reconstructed dynamic regions. Our method adjusts the accumulated gradients with Gaussians' temporal coverage, significantly improving reconstruction quality in dynamic regions. Experimental results highlight that our method achieves state-of-the-art visual quality in dynamic regions, outperforming all baselines by a large margin with practical storage costs.", "date": "2024-11-26", "categories": []}
{"id": "2504.01647", "title": "FlowR: Flowing from Sparse to Dense 3D Reconstructions", "abstract": "3D Gaussian splatting enables high-quality novel view synthesis (NVS) at real-time frame rates. However, its quality drops sharply as we depart from the training views. Thus, dense captures are needed to match the high-quality expectations of applications like Virtual Reality (VR). However, such dense captures are very laborious and expensive to obtain. Existing works have explored using 2D generative models to alleviate this requirement by distillation or generating additional training views. These models typically rely on a noise-to-data generative process conditioned only on a handful of reference input views, leading to hallucinations, inconsistent generation results, and subsequent reconstruction artifacts. Instead, we propose a multi-view, flow matching model that learns a flow to directly connect novel view renderings from possibly sparse reconstructions to renderings that we expect from dense reconstructions. This enables augmenting scene captures with consistent, generated views to improve reconstruction quality. Our model is trained on a novel dataset of 3.6M image pairs and can process up to 45 views at 540x960 resolution (91K tokens) on one H100 GPU in a single forward pass. Our pipeline consistently improves NVS in sparse- and dense-view scenarios, leading to higher-quality reconstructions than prior works across multiple, widely-used NVS benchmarks.", "date": "2025-04-02", "categories": []}
{"id": "2507.12137", "title": "AD-GS: Object-Aware B-Spline Gaussian Splatting for Self-Supervised Autonomous Driving", "abstract": "Modeling and rendering dynamic urban driving scenes is crucial for self-driving simulation. Current high-quality methods typically rely on costly manual object tracklet annotations, while self-supervised approaches fail to capture dynamic object motions accurately and decompose scenes properly, resulting in rendering artifacts. We introduce AD-GS, a novel self-supervised framework for high-quality free-viewpoint rendering of driving scenes from a single log. At its core is a novel learnable motion model that integrates locality-aware B-spline curves with global-aware trigonometric functions, enabling flexible yet precise dynamic object modeling. Rather than requiring comprehensive semantic labeling, AD-GS automatically segments scenes into objects and background with the simplified pseudo 2D segmentation, representing objects using dynamic Gaussians and bidirectional temporal visibility masks. Further, our model incorporates visibility reasoning and physically rigid regularization to enhance robustness. Extensive evaluations demonstrate that our annotation-free model significantly outperforms current state-of-the-art annotation-free methods and is competitive with annotation-dependent approaches.", "date": "2025-07-16", "categories": []}
{"id": "2508.03118", "title": "H3R: Hybrid Multi-view Correspondence for Generalizable 3D Reconstruction", "abstract": "Despite recent advances in feed-forward 3D Gaussian Splatting, generalizable 3D reconstruction remains challenging, particularly in multi-view correspondence modeling. Existing approaches face a fundamental trade-off: explicit methods achieve geometric precision but struggle with ambiguous regions, while implicit methods provide robustness but suffer from slow convergence. We present H3R, a hybrid framework that addresses this limitation by integrating volumetric latent fusion with attention-based feature aggregation. Our framework consists of two complementary components: an efficient latent volume that enforces geometric consistency through epipolar constraints, and a camera-aware Transformer that leverages Pl\\\"ucker coordinates for adaptive correspondence refinement. By integrating both paradigms, our approach enhances generalization while converging 2$\\times$ faster than existing methods. Furthermore, we show that spatial-aligned foundation models (e.g., SD-VAE) substantially outperform semantic-aligned models (e.g., DINOv2), resolving the mismatch between semantic representations and spatial reconstruction requirements. Our method supports variable-number and high-resolution input views while demonstrating robust cross-dataset generalization. Extensive experiments show that our method achieves state-of-the-art performance across multiple benchmarks, with significant PSNR improvements of 0.59 dB, 1.06 dB, and 0.22 dB on the RealEstate10K, ACID, and DTU datasets, respectively. Code is available at https://github.com/JiaHeng-DLUT/H3R.", "date": "2025-08-05", "categories": []}
{"id": "2412.04887", "title": "Momentum-GS: Momentum Gaussian Self-Distillation for High-Quality Large Scene Reconstruction", "abstract": "3D Gaussian Splatting has demonstrated notable success in large-scale scene reconstruction, but challenges persist due to high training memory consumption and storage overhead. Hybrid representations that integrate implicit and explicit features offer a way to mitigate these limitations. However, when applied in parallelized block-wise training, two critical issues arise since reconstruction accuracy deteriorates due to reduced data diversity when training each block independently, and parallel training restricts the number of divided blocks to the available number of GPUs. To address these issues, we propose Momentum-GS, a novel approach that leverages momentum-based self-distillation to promote consistency and accuracy across the blocks while decoupling the number of blocks from the physical GPU count. Our method maintains a teacher Gaussian decoder updated with momentum, ensuring a stable reference during training. This teacher provides each block with global guidance in a self-distillation manner, promoting spatial consistency in reconstruction. To further ensure consistency across the blocks, we incorporate block weighting, dynamically adjusting each block's weight according to its reconstruction accuracy. Extensive experiments on large-scale scenes show that our method consistently outperforms existing techniques, achieving a 12.8% improvement in LPIPS over CityGaussian with much fewer divided blocks and establishing a new state of the art. Project page: https://jixuan-fan.github.io/Momentum-GS_Page/", "date": "2024-12-06", "categories": []}
{"id": "2503.18458", "title": "StableGS: A Floater-Free Framework for 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) reconstructions are plagued by stubborn ``floater\" artifacts that degrade their geometric and visual fidelity. We are the first to reveal the root cause: a fundamental conflict in the 3DGS optimization process where the opacity gradients of floaters vanish when their blended color reaches a pseudo-equilibrium of canceling errors against the background, trapping them in a spurious local minimum. To resolve this, we propose StableGS, a novel framework that decouples geometric regularization from final appearance rendering. Its core is a Dual Opacity architecture that creates two separate rendering paths: a ``Geometric Regularization Path\" to bear strong depth-based constraints for structural correctness, and an ``Appearance Refinement Path\" to generate high-fidelity details upon this stable foundation. We complement this with a synergistic set of geometric constraints: a self-supervised depth consistency loss and an external geometric prior enabled by our efficient global scale optimization algorithm. Experiments on multiple benchmarks show StableGS not only eliminates floaters but also resolves the common blur-artifact trade-off, achieving state-of-the-art geometric accuracy and visual quality.", "date": "2025-03-24", "categories": []}
{"id": "2507.19451", "title": "GS-Occ3D: Scaling Vision-only Occupancy Reconstruction with Gaussian Splatting", "abstract": "Occupancy is crucial for autonomous driving, providing essential geometric priors for perception and planning. However, existing methods predominantly rely on LiDAR-based occupancy annotations, which limits scalability and prevents leveraging vast amounts of potential crowdsourced data for auto-labeling. To address this, we propose GS-Occ3D, a scalable vision-only framework that directly reconstructs occupancy. Vision-only occupancy reconstruction poses significant challenges due to sparse viewpoints, dynamic scene elements, severe occlusions, and long-horizon motion. Existing vision-based methods primarily rely on mesh representation, which suffer from incomplete geometry and additional post-processing, limiting scalability. To overcome these issues, GS-Occ3D optimizes an explicit occupancy representation using an Octree-based Gaussian Surfel formulation, ensuring efficiency and scalability. Additionally, we decompose scenes into static background, ground, and dynamic objects, enabling tailored modeling strategies: (1) Ground is explicitly reconstructed as a dominant structural element, significantly improving large-area consistency; (2) Dynamic vehicles are separately modeled to better capture motion-related occupancy patterns. Extensive experiments on the Waymo dataset demonstrate that GS-Occ3D achieves state-of-the-art geometry reconstruction results. By curating vision-only binary occupancy labels from diverse urban scenes, we show their effectiveness for downstream occupancy models on Occ3D-Waymo and superior zero-shot generalization on Occ3D-nuScenes. It highlights the potential of large-scale vision-based occupancy reconstruction as a new paradigm for scalable auto-labeling. Project Page: https://gs-occ3d.github.io/", "date": "2025-07-25", "categories": []}
{"id": "2507.21045", "title": "Reconstructing 4D Spatial Intelligence: A Survey", "abstract": "Reconstructing 4D spatial intelligence from visual observations has long been a central yet challenging task in computer vision, with broad real-world applications. These range from entertainment domains like movies, where the focus is often on reconstructing fundamental visual elements, to embodied AI, which emphasizes interaction modeling and physical realism. Fueled by rapid advances in 3D representations and deep learning architectures, the field has evolved quickly, outpacing the scope of previous surveys. Additionally, existing surveys rarely offer a comprehensive analysis of the hierarchical structure of 4D scene reconstruction. To address this gap, we present a new perspective that organizes existing methods into five progressive levels of 4D spatial intelligence: (1) Level 1 -- reconstruction of low-level 3D attributes (e.g., depth, pose, and point maps); (2) Level 2 -- reconstruction of 3D scene components (e.g., objects, humans, structures); (3) Level 3 -- reconstruction of 4D dynamic scenes; (4) Level 4 -- modeling of interactions among scene components; and (5) Level 5 -- incorporation of physical laws and constraints. We conclude the survey by discussing the key challenges at each level and highlighting promising directions for advancing toward even richer levels of 4D spatial intelligence. To track ongoing developments, we maintain an up-to-date project page: https://github.com/yukangcao/Awesome-4D-Spatial-Intelligence.", "date": "2025-07-28", "categories": []}
{"id": "2508.00967", "title": "Cooperative Perception: A Resource-Efficient Framework for Multi-Drone 3D Scene Reconstruction Using Federated Diffusion and NeRF", "abstract": "The proposal introduces an innovative drone swarm perception system that aims to solve problems related to computational limitations and low-bandwidth communication, and real-time scene reconstruction. The framework enables efficient multi-agent 3D/4D scene synthesis through federated learning of shared diffusion model and YOLOv12 lightweight semantic extraction and local NeRF updates while maintaining privacy and scalability. The framework redesigns generative diffusion models for joint scene reconstruction, and improves cooperative scene understanding, while adding semantic-aware compression protocols. The approach can be validated through simulations and potential real-world deployment on drone testbeds, positioning it as a disruptive advancement in multi-agent AI for autonomous systems.", "date": "2025-08-01", "categories": []}
{"id": "2508.01150", "title": "OpenGS-Fusion: Open-Vocabulary Dense Mapping with Hybrid 3D Gaussian Splatting for Refined Object-Level Understanding", "abstract": "Recent advancements in 3D scene understanding have made significant strides in enabling interaction with scenes using open-vocabulary queries, particularly for VR/AR and robotic applications. Nevertheless, existing methods are hindered by rigid offline pipelines and the inability to provide precise 3D object-level understanding given open-ended queries. In this paper, we present OpenGS-Fusion, an innovative open-vocabulary dense mapping framework that improves semantic modeling and refines object-level understanding. OpenGS-Fusion combines 3D Gaussian representation with a Truncated Signed Distance Field to facilitate lossless fusion of semantic features on-the-fly. Furthermore, we introduce a novel multimodal language-guided approach named MLLM-Assisted Adaptive Thresholding, which refines the segmentation of 3D objects by adaptively adjusting similarity thresholds, achieving an improvement 17\\% in 3D mIoU compared to the fixed threshold strategy. Extensive experiments demonstrate that our method outperforms existing methods in 3D object understanding and scene reconstruction quality, as well as showcasing its effectiveness in language-guided scene interaction. The code is available at https://young-bit.github.io/opengs-fusion.github.io/ .", "date": "2025-08-02", "categories": []}
{"id": "2508.01239", "title": "OCSplats: Observation Completeness Quantification and Label Noise Separation in 3DGS", "abstract": "3D Gaussian Splatting (3DGS) has become one of the most promising 3D reconstruction technologies. However, label noise in real-world scenarios-such as moving objects, non-Lambertian surfaces, and shadows-often leads to reconstruction errors. Existing 3DGS-Bsed anti-noise reconstruction methods either fail to separate noise effectively or require scene-specific fine-tuning of hyperparameters, making them difficult to apply in practice. This paper re-examines the problem of anti-noise reconstruction from the perspective of epistemic uncertainty, proposing a novel framework, OCSplats. By combining key technologies such as hybrid noise assessment and observation-based cognitive correction, the accuracy of noise classification in areas with cognitive differences has been significantly improved. Moreover, to address the issue of varying noise proportions in different scenarios, we have designed a label noise classification pipeline based on dynamic anchor points. This pipeline enables OCSplats to be applied simultaneously to scenarios with vastly different noise proportions without adjusting parameters. Extensive experiments demonstrate that OCSplats always achieve leading reconstruction performance and precise label noise classification in scenes of different complexity levels.", "date": "2025-08-02", "categories": []}
{"id": "2508.02323", "title": "Dream-to-Recon: Monocular 3D Reconstruction with Diffusion-Depth Distillation from Single Images", "abstract": "Volumetric scene reconstruction from a single image is crucial for a broad range of applications like autonomous driving and robotics. Recent volumetric reconstruction methods achieve impressive results, but generally require expensive 3D ground truth or multi-view supervision. We propose to leverage pre-trained 2D diffusion models and depth prediction models to generate synthetic scene geometry from a single image. This can then be used to distill a feed-forward scene reconstruction model. Our experiments on the challenging KITTI-360 and Waymo datasets demonstrate that our method matches or outperforms state-of-the-art baselines that use multi-view supervision, and offers unique advantages, for example regarding dynamic scenes.", "date": "2025-08-04", "categories": []}
{"id": "2410.12781", "title": "Long-LRM: Long-sequence Large Reconstruction Model for Wide-coverage Gaussian Splats", "abstract": "We propose Long-LRM, a feed-forward 3D Gaussian reconstruction model for instant, high-resolution, 360{\\deg} wide-coverage, scene-level reconstruction. Specifically, it takes in 32 input images at a resolution of 960x540 and produces the Gaussian reconstruction in just 1 second on a single A100 GPU. To handle the long sequence of 250K tokens brought by the large input size, Long-LRM features a mixture of the recent Mamba2 blocks and the classical transformer blocks, enhanced by a light-weight token merging module and Gaussian pruning steps that balance between quality and efficiency. We evaluate Long-LRM on the large-scale DL3DV benchmark and Tanks&Temples, demonstrating reconstruction quality comparable to the optimization-based methods while achieving an 800x speedup w.r.t. the optimization-based approaches and an input size at least 60x larger than the previous feed-forward approaches. We conduct extensive ablation studies on our model design choices for both rendering quality and computation efficiency. We also explore Long-LRM's compatibility with other Gaussian variants such as 2D GS, which enhances Long-LRM's ability in geometry reconstruction. Project page: https://arthurhero.github.io/projects/llrm", "date": "2024-10-16", "categories": []}
{"id": "2504.12811", "title": "AAA-Gaussians: Anti-Aliased and Artifact-Free 3D Gaussian Rendering", "abstract": "Although 3D Gaussian Splatting (3DGS) has revolutionized 3D reconstruction, it still faces challenges such as aliasing, projection artifacts, and view inconsistencies, primarily due to the simplification of treating splats as 2D entities. We argue that incorporating full 3D evaluation of Gaussians throughout the 3DGS pipeline can effectively address these issues while preserving rasterization efficiency. Specifically, we introduce an adaptive 3D smoothing filter to mitigate aliasing and present a stable view-space bounding method that eliminates popping artifacts when Gaussians extend beyond the view frustum. Furthermore, we promote tile-based culling to 3D with screen-space planes, accelerating rendering and reducing sorting costs for hierarchical rasterization. Our method achieves state-of-the-art quality on in-distribution evaluation sets and significantly outperforms other approaches for out-of-distribution views. Our qualitative evaluations further demonstrate the effective removal of aliasing, distortions, and popping artifacts, ensuring real-time, artifact-free rendering.", "date": "2025-04-17", "categories": []}
{"id": "2508.00366", "title": "SparseRecon: Neural Implicit Surface Reconstruction from Sparse Views with Feature and Depth Consistencies", "abstract": "Surface reconstruction from sparse views aims to reconstruct a 3D shape or scene from few RGB images. The latest methods are either generalization-based or overfitting-based. However, the generalization-based methods do not generalize well on views that were unseen during training, while the reconstruction quality of overfitting-based methods is still limited by the limited geometry clues. To address this issue, we propose SparseRecon, a novel neural implicit reconstruction method for sparse views with volume rendering-based feature consistency and uncertainty-guided depth constraint. Firstly, we introduce a feature consistency loss across views to constrain the neural implicit field. This design alleviates the ambiguity caused by insufficient consistency information of views and ensures completeness and smoothness in the reconstruction results. Secondly, we employ an uncertainty-guided depth constraint to back up the feature consistency loss in areas with occlusion and insignificant features, which recovers geometry details for better reconstruction quality. Experimental results demonstrate that our method outperforms the state-of-the-art methods, which can produce high-quality geometry with sparse-view input, especially in the scenarios with small overlapping views. Project page: https://hanl2010.github.io/SparseRecon/.", "date": "2025-08-01", "categories": []}
{"id": "2503.17856", "title": "ClaraVid: A Holistic Scene Reconstruction Benchmark From Aerial Perspective With Delentropy-Based Complexity Profiling", "abstract": "The development of aerial holistic scene understanding algorithms is hindered by the scarcity of comprehensive datasets that enable both semantic and geometric reconstruction. While synthetic datasets offer an alternative, existing options exhibit task-specific limitations, unrealistic scene compositions, and rendering artifacts that compromise real-world applicability. We introduce ClaraVid, a synthetic aerial dataset specifically designed to overcome these limitations. Comprising 16,917 high-resolution images captured at 4032x3024 from multiple viewpoints across diverse landscapes, ClaraVid provides dense depth maps, panoptic segmentation, sparse point clouds, and dynamic object masks, while mitigating common rendering artifacts. To further advance neural reconstruction, we introduce the Delentropic Scene Profile (DSP), a novel complexity metric derived from differential entropy analysis, designed to quantitatively assess scene difficulty and inform reconstruction tasks. Utilizing DSP, we systematically benchmark neural reconstruction methods, uncovering a consistent, measurable correlation between scene complexity and reconstruction accuracy. Empirical results indicate that higher delentropy strongly correlates with increased reconstruction errors, validating DSP as a reliable complexity prior. The data and code are available on the project page at https://rdbch.github.io/claravid/", "date": "2025-03-22", "categories": []}
{"id": "2505.02178", "title": "Sparfels: Fast Reconstruction from Sparse Unposed Imagery", "abstract": "We present a method for Sparse view reconstruction with surface element splatting that runs within 3 minutes on a consumer grade GPU. While few methods address sparse radiance field learning from noisy or unposed sparse cameras, shape recovery remains relatively underexplored in this setting. Several radiance and shape learning test-time optimization methods address the sparse posed setting by learning data priors or using combinations of external monocular geometry priors. Differently, we propose an efficient and simple pipeline harnessing a single recent 3D foundation model. We leverage its various task heads, notably point maps and camera initializations to instantiate a bundle adjusting 2D Gaussian Splatting (2DGS) model, and image correspondences to guide camera optimization midst 2DGS training. Key to our contribution is a novel formulation of splatted color variance along rays, which can be computed efficiently. Reducing this moment in training leads to more accurate shape reconstructions. We demonstrate state-of-the-art performances in the sparse uncalibrated setting in reconstruction and novel view benchmarks based on established multi-view datasets.", "date": "2025-05-04", "categories": []}
{"id": "2507.01631", "title": "Tile and Slide : A New Framework for Scaling NeRF from Local to Global 3D Earth Observation", "abstract": "Neural Radiance Fields (NeRF) have recently emerged as a paradigm for 3D reconstruction from multiview satellite imagery. However, state-of-the-art NeRF methods are typically constrained to small scenes due to the memory footprint during training, which we study in this paper. Previous work on large-scale NeRFs palliate this by dividing the scene into NeRFs. This paper introduces Snake-NeRF, a framework that scales to large scenes. Our out-of-core method eliminates the need to load all images and networks simultaneously, and operates on a single device. We achieve this by dividing the region of interest into NeRFs that 3D tile without overlap. Importantly, we crop the images with overlap to ensure each NeRFs is trained with all the necessary pixels. We introduce a novel $2\\times 2$ 3D tile progression strategy and segmented sampler, which together prevent 3D reconstruction errors along the tile edges. Our experiments conclude that large satellite images can effectively be processed with linear time complexity, on a single GPU, and without compromise in quality.", "date": "2025-07-02", "categories": []}
{"id": "2507.23006", "title": "Robust and Efficient 3D Gaussian Splatting for Urban Scene Reconstruction", "abstract": "We present a framework that enables fast reconstruction and real-time rendering of urban-scale scenes while maintaining robustness against appearance variations across multi-view captures. Our approach begins with scene partitioning for parallel training, employing a visibility-based image selection strategy to optimize training efficiency. A controllable level-of-detail (LOD) strategy explicitly regulates Gaussian density under a user-defined budget, enabling efficient training and rendering while maintaining high visual fidelity. The appearance transformation module mitigates the negative effects of appearance inconsistencies across images while enabling flexible adjustments. Additionally, we utilize enhancement modules, such as depth regularization, scale regularization, and antialiasing, to improve reconstruction fidelity. Experimental results demonstrate that our method effectively reconstructs urban-scale scenes and outperforms previous approaches in both efficiency and quality. The source code is available at: https://yzslab.github.io/REUrbanGS.", "date": "2025-07-30", "categories": []}
{"id": "2507.23340", "title": "MagicRoad: Semantic-Aware 3D Road Surface Reconstruction via Obstacle Inpainting", "abstract": "Road surface reconstruction is essential for autonomous driving, supporting centimeter-accurate lane perception and high-definition mapping in complex urban environments.While recent methods based on mesh rendering or 3D Gaussian splatting (3DGS) achieve promising results under clean and static conditions, they remain vulnerable to occlusions from dynamic agents, visual clutter from static obstacles, and appearance degradation caused by lighting and weather changes. We present a robust reconstruction framework that integrates occlusion-aware 2D Gaussian surfels with semantic-guided color enhancement to recover clean, consistent road surfaces. Our method leverages a planar-adapted Gaussian representation for efficient large-scale modeling, employs segmentation-guided video inpainting to remove both dynamic and static foreground objects, and enhances color coherence via semantic-aware correction in HSV space. Extensive experiments on urban-scale datasets demonstrate that our framework produces visually coherent and geometrically faithful reconstructions, significantly outperforming prior methods under real-world conditions.", "date": "2025-07-31", "categories": []}
{"id": "2507.23374", "title": "NeRF Is a Valuable Assistant for 3D Gaussian Splatting", "abstract": "We introduce NeRF-GS, a novel framework that jointly optimizes Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). This framework leverages the inherent continuous spatial representation of NeRF to mitigate several limitations of 3DGS, including sensitivity to Gaussian initialization, limited spatial awareness, and weak inter-Gaussian correlations, thereby enhancing its performance. In NeRF-GS, we revisit the design of 3DGS and progressively align its spatial features with NeRF, enabling both representations to be optimized within the same scene through shared 3D spatial information. We further address the formal distinctions between the two approaches by optimizing residual vectors for both implicit features and Gaussian positions to enhance the personalized capabilities of 3DGS. Experimental results on benchmark datasets show that NeRF-GS surpasses existing methods and achieves state-of-the-art performance. This outcome confirms that NeRF and 3DGS are complementary rather than competing, offering new insights into hybrid approaches that combine 3DGS and NeRF for efficient 3D scene representation.", "date": "2025-07-31", "categories": []}
{"id": "2507.23704", "title": "Enhanced Velocity Field Modeling for Gaussian Video Reconstruction", "abstract": "High-fidelity 3D video reconstruction is essential for enabling real-time rendering of dynamic scenes with realistic motion in virtual and augmented reality (VR/AR). The deformation field paradigm of 3D Gaussian splatting has achieved near-photorealistic results in video reconstruction due to the great representation capability of deep deformation networks. However, in videos with complex motion and significant scale variations, deformation networks often overfit to irregular Gaussian trajectories, leading to suboptimal visual quality. Moreover, the gradient-based densification strategy designed for static scene reconstruction proves inadequate to address the absence of dynamic content. In light of these challenges, we propose a flow-empowered velocity field modeling scheme tailored for Gaussian video reconstruction, dubbed FlowGaussian-VR. It consists of two core components: a velocity field rendering (VFR) pipeline which enables optical flow-based optimization, and a flow-assisted adaptive densification (FAD) strategy that adjusts the number and size of Gaussians in dynamic regions. We validate our model's effectiveness on multi-view dynamic reconstruction and novel view synthesis with multiple real-world datasets containing challenging motion scenarios, demonstrating not only notable visual improvements (over 2.5 dB gain in PSNR) and less blurry artifacts in dynamic textures, but also regularized and trackable per-Gaussian trajectories.", "date": "2025-07-31", "categories": []}
{"id": "2507.22825", "title": "DepR: Depth Guided Single-view Scene Reconstruction with Instance-level Diffusion", "abstract": "We propose DepR, a depth-guided single-view scene reconstruction framework that integrates instance-level diffusion within a compositional paradigm. Instead of reconstructing the entire scene holistically, DepR generates individual objects and subsequently composes them into a coherent 3D layout. Unlike previous methods that use depth solely for object layout estimation during inference and therefore fail to fully exploit its rich geometric information, DepR leverages depth throughout both training and inference. Specifically, we introduce depth-guided conditioning to effectively encode shape priors into diffusion models. During inference, depth further guides DDIM sampling and layout optimization, enhancing alignment between the reconstruction and the input image. Despite being trained on limited synthetic data, DepR achieves state-of-the-art performance and demonstrates strong generalization in single-view scene reconstruction, as shown through evaluations on both synthetic and real-world datasets.", "date": "2025-07-30", "categories": []}
{"id": "2503.10170", "title": "GS-SDF: LiDAR-Augmented Gaussian Splatting and Neural SDF for Geometrically Consistent Rendering and Reconstruction", "abstract": "Digital twins are fundamental to the development of autonomous driving and embodied artificial intelligence. However, achieving high-granularity surface reconstruction and high-fidelity rendering remains a challenge. Gaussian splatting offers efficient photorealistic rendering but struggles with geometric inconsistencies due to fragmented primitives and sparse observational data in robotics applications. Existing regularization methods, which rely on render-derived constraints, often fail in complex environments. Moreover, effectively integrating sparse LiDAR data with Gaussian splatting remains challenging. We propose a unified LiDAR-visual system that synergizes Gaussian splatting with a neural signed distance field. The accurate LiDAR point clouds enable a trained neural signed distance field to offer a manifold geometry field. This motivates us to offer an SDF-based Gaussian initialization for physically grounded primitive placement and a comprehensive geometric regularization for geometrically consistent rendering and reconstruction. Experiments demonstrate superior reconstruction accuracy and rendering quality across diverse trajectories. To benefit the community, the codes are released at https://github.com/hku-mars/GS-SDF.", "date": "2025-03-13", "categories": []}
{"id": "2507.19141", "title": "DASH: 4D Hash Encoding with Self-Supervised Decomposition for Real-Time Dynamic Scene Rendering", "abstract": "Dynamic scene reconstruction is a long-term challenge in 3D vision. Existing plane-based methods in dynamic Gaussian splatting suffer from an unsuitable low-rank assumption, causing feature overlap and poor rendering quality. Although 4D hash encoding provides an explicit representation without low-rank constraints, directly applying it to the entire dynamic scene leads to substantial hash collisions and redundancy. To address these challenges, we present DASH, a real-time dynamic scene rendering framework that employs 4D hash encoding coupled with self-supervised decomposition. Our approach begins with a self-supervised decomposition mechanism that separates dynamic and static components without manual annotations or precomputed masks. Next, we introduce a multiresolution 4D hash encoder for dynamic elements, providing an explicit representation that avoids the low-rank assumption. Finally, we present a spatio-temporal smoothness regularization strategy to mitigate unstable deformation artifacts. Experiments on real-world datasets demonstrate that DASH achieves state-of-the-art dynamic rendering performance, exhibiting enhanced visual quality at real-time speeds of 264 FPS on a single 4090 GPU. Code: https://github.com/chenj02/DASH.", "date": "2025-07-25", "categories": []}
{"id": "2507.21960", "title": "PanoSplatt3R: Leveraging Perspective Pretraining for Generalized Unposed Wide-Baseline Panorama Reconstruction", "abstract": "Wide-baseline panorama reconstruction has emerged as a highly effective and pivotal approach for not only achieving geometric reconstruction of the surrounding 3D environment, but also generating highly realistic and immersive novel views. Although existing methods have shown remarkable performance across various benchmarks, they are predominantly reliant on accurate pose information. In real-world scenarios, the acquisition of precise pose often requires additional computational resources and is highly susceptible to noise. These limitations hinder the broad applicability and practicality of such methods. In this paper, we present PanoSplatt3R, an unposed wide-baseline panorama reconstruction method. We extend and adapt the foundational reconstruction pretrainings from the perspective domain to the panoramic domain, thus enabling powerful generalization capabilities. To ensure a seamless and efficient domain-transfer process, we introduce RoPE rolling that spans rolled coordinates in rotary positional embeddings across different attention heads, maintaining a minimal modification to RoPE's mechanism, while modeling the horizontal periodicity of panorama images. Comprehensive experiments demonstrate that PanoSplatt3R, even in the absence of pose information, significantly outperforms current state-of-the-art methods. This superiority is evident in both the generation of high-quality novel views and the accuracy of depth estimation, thereby showcasing its great potential for practical applications. Project page: https://npucvr.github.io/PanoSplatt3R", "date": "2025-07-29", "categories": []}
{"id": "2409.03456", "title": "LM-Gaussian: Boost Sparse-view 3D Gaussian Splatting with Large Model Priors", "abstract": "We aim to address sparse-view reconstruction of a 3D scene by leveraging priors from large-scale vision models. While recent advancements such as 3D Gaussian Splatting (3DGS) have demonstrated remarkable successes in 3D reconstruction, these methods typically necessitate hundreds of input images that densely capture the underlying scene, making them time-consuming and impractical for real-world applications. However, sparse-view reconstruction is inherently ill-posed and under-constrained, often resulting in inferior and incomplete outcomes. This is due to issues such as failed initialization, overfitting on input images, and a lack of details. To mitigate these challenges, we introduce LM-Gaussian, a method capable of generating high-quality reconstructions from a limited number of images. Specifically, we propose a robust initialization module that leverages stereo priors to aid in the recovery of camera poses and the reliable point clouds. Additionally, a diffusion-based refinement is iteratively applied to incorporate image diffusion priors into the Gaussian optimization process to preserve intricate scene details. Finally, we utilize video diffusion priors to further enhance the rendered images for realistic visual effects. Overall, our approach significantly reduces the data acquisition requirements compared to previous 3DGS methods. We validate the effectiveness of our framework through experiments on various public datasets, demonstrating its potential for high-quality 360-degree scene reconstruction. Visual results are on our website.", "date": "2024-09-05", "categories": []}
{"id": "2411.11921", "title": "DeSiRe-GS: 4D Street Gaussians for Static-Dynamic Decomposition and Surface Reconstruction for Urban Driving Scenes", "abstract": "We present DeSiRe-GS, a self-supervised gaussian splatting representation, enabling effective static-dynamic decomposition and high-fidelity surface reconstruction in complex driving scenarios. Our approach employs a two-stage optimization pipeline of dynamic street Gaussians. In the first stage, we extract 2D motion masks based on the observation that 3D Gaussian Splatting inherently can reconstruct only the static regions in dynamic environments. These extracted 2D motion priors are then mapped into the Gaussian space in a differentiable manner, leveraging an efficient formulation of dynamic Gaussians in the second stage. Combined with the introduced geometric regularizations, our method are able to address the over-fitting issues caused by data sparsity in autonomous driving, reconstructing physically plausible Gaussians that align with object surfaces rather than floating in air. Furthermore, we introduce temporal cross-view consistency to ensure coherence across time and viewpoints, resulting in high-quality surface reconstruction. Comprehensive experiments demonstrate the efficiency and effectiveness of DeSiRe-GS, surpassing prior self-supervised arts and achieving accuracy comparable to methods relying on external 3D bounding box annotations. Code is available at https://github.com/chengweialan/DeSiRe-GS", "date": "2024-11-18", "categories": []}
{"id": "2412.03515", "title": "Distilling Diffusion Models to Efficient 3D LiDAR Scene Completion", "abstract": "Diffusion models have been applied to 3D LiDAR scene completion due to their strong training stability and high completion quality. However, the slow sampling speed limits the practical application of diffusion-based scene completion models since autonomous vehicles require an efficient perception of surrounding environments. This paper proposes a novel distillation method tailored for 3D Li- DAR scene completion models, dubbed ScoreLiDAR, which achieves efficient yet high-quality scene completion. Score- LiDAR enables the distilled model to sample in significantly fewer steps after distillation. To improve completion quality, we also introduce a novel Structural Loss, which encourages the distilled model to capture the geometric structure of the 3D LiDAR scene. The loss contains a scene-wise term constraining the holistic structure and a point-wise term constraining the key landmark points and their relative configuration. Extensive experiments demonstrate that ScoreLiDAR significantly accelerates the completion time from 30.55 to 5.37 seconds per frame (>5x) on SemanticKITTI and achieves superior performance compared to state-of-the-art 3D LiDAR scene completion models. Our model and code are publicly available on https://github.com/happyw1nd/ScoreLiDAR.", "date": "2024-12-04", "categories": []}
{"id": "2507.15602", "title": "SurfaceSplat: Connecting Surface Reconstruction and Gaussian Splatting", "abstract": "Surface reconstruction and novel view rendering from sparse-view images are challenging. Signed Distance Function (SDF)-based methods struggle with fine details, while 3D Gaussian Splatting (3DGS)-based approaches lack global geometry coherence. We propose a novel hybrid method that combines the strengths of both approaches: SDF captures coarse geometry to enhance 3DGS-based rendering, while newly rendered images from 3DGS refine the details of SDF for accurate surface reconstruction. As a result, our method surpasses state-of-the-art approaches in surface reconstruction and novel view synthesis on the DTU and MobileBrick datasets. Code will be released at https://github.com/aim-uofa/SurfaceSplat.", "date": "2025-07-21", "categories": []}
{"id": "2507.20239", "title": "Decomposing Densification in Gaussian Splatting for Faster 3D Scene Reconstruction", "abstract": "3D Gaussian Splatting (GS) has emerged as a powerful representation for high-quality scene reconstruction, offering compelling rendering quality. However, the training process of GS often suffers from slow convergence due to inefficient densification and suboptimal spatial distribution of Gaussian primitives. In this work, we present a comprehensive analysis of the split and clone operations during the densification phase, revealing their distinct roles in balancing detail preservation and computational efficiency. Building upon this analysis, we propose a global-to-local densification strategy, which facilitates more efficient growth of Gaussians across the scene space, promoting both global coverage and local refinement. To cooperate with the proposed densification strategy and promote sufficient diffusion of Gaussian primitives in space, we introduce an energy-guided coarse-to-fine multi-resolution training framework, which gradually increases resolution based on energy density in 2D images. Additionally, we dynamically prune unnecessary Gaussian primitives to speed up the training. Extensive experiments on MipNeRF-360, Deep Blending, and Tanks & Temples datasets demonstrate that our approach significantly accelerates training,achieving over 2x speedup with fewer Gaussian primitives and superior reconstruction performance.", "date": "2025-07-27", "categories": []}
{"id": "2506.22800", "title": "RGE-GS: Reward-Guided Expansive Driving Scene Reconstruction via Diffusion Priors", "abstract": "A single-pass driving clip frequently results in incomplete scanning of the road structure, making reconstructed scene expanding a critical requirement for sensor simulators to effectively regress driving actions. Although contemporary 3D Gaussian Splatting (3DGS) techniques achieve remarkable reconstruction quality, their direct extension through the integration of diffusion priors often introduces cumulative physical inconsistencies and compromises training efficiency. To address these limitations, we present RGE-GS, a novel expansive reconstruction framework that synergizes diffusion-based generation with reward-guided Gaussian integration. The RGE-GS framework incorporates two key innovations: First, we propose a reward network that learns to identify and prioritize consistently generated patterns prior to reconstruction phases, thereby enabling selective retention of diffusion outputs for spatial stability. Second, during the reconstruction process, we devise a differentiated training strategy that automatically adjust Gaussian optimization progress according to scene converge metrics, which achieving better convergence than baseline methods. Extensive evaluations of publicly available datasets demonstrate that RGE-GS achieves state-of-the-art performance in reconstruction quality. Our source-code will be made publicly available at https://github.com/CN-ADLab/RGE-GS.", "date": "2025-06-28", "categories": []}
{"id": "2507.06230", "title": "Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion", "abstract": "Semantic scene completion (SSC) aims to infer both the 3D geometry and semantics of a scene from single images. In contrast to prior work on SSC that heavily relies on expensive ground-truth annotations, we approach SSC in an unsupervised setting. Our novel method, SceneDINO, adapts techniques from self-supervised representation learning and 2D unsupervised scene understanding to SSC. Our training exclusively utilizes multi-view consistency self-supervision without any form of semantic or geometric ground truth. Given a single input image, SceneDINO infers the 3D geometry and expressive 3D DINO features in a feed-forward manner. Through a novel 3D feature distillation approach, we obtain unsupervised 3D semantics. In both 3D and 2D unsupervised scene understanding, SceneDINO reaches state-of-the-art segmentation accuracy. Linear probing our 3D features matches the segmentation accuracy of a current supervised SSC approach. Additionally, we showcase the domain generalization and multi-view consistency of SceneDINO, taking the first steps towards a strong foundation for single image 3D scene understanding.", "date": "2025-07-08", "categories": []}
{"id": "2507.19188", "title": "VisHall3D: Monocular Semantic Scene Completion from Reconstructing the Visible Regions to Hallucinating the Invisible Regions", "abstract": "This paper introduces VisHall3D, a novel two-stage framework for monocular semantic scene completion that aims to address the issues of feature entanglement and geometric inconsistency prevalent in existing methods. VisHall3D decomposes the scene completion task into two stages: reconstructing the visible regions (vision) and inferring the invisible regions (hallucination). In the first stage, VisFrontierNet, a visibility-aware projection module, is introduced to accurately trace the visual frontier while preserving fine-grained details. In the second stage, OcclusionMAE, a hallucination network, is employed to generate plausible geometries for the invisible regions using a noise injection mechanism. By decoupling scene completion into these two distinct stages, VisHall3D effectively mitigates feature entanglement and geometric inconsistency, leading to significantly improved reconstruction quality.\n  The effectiveness of VisHall3D is validated through extensive experiments on two challenging benchmarks: SemanticKITTI and SSCBench-KITTI-360. VisHall3D achieves state-of-the-art performance, outperforming previous methods by a significant margin and paves the way for more accurate and reliable scene understanding in autonomous driving and other applications.", "date": "2025-07-25", "categories": []}
{"id": "2507.19474", "title": "DINO-SLAM: DINO-informed RGB-D SLAM for Neural Implicit and Explicit Representations", "abstract": "This paper presents DINO-SLAM, a DINO-informed design strategy to enhance neural implicit (Neural Radiance Field -- NeRF) and explicit representations (3D Gaussian Splatting -- 3DGS) in SLAM systems through more comprehensive scene representations. Purposely, we rely on a Scene Structure Encoder (SSE) that enriches DINO features into Enhanced DINO ones (EDINO) to capture hierarchical scene elements and their structural relationships. Building upon it, we propose two foundational paradigms for NeRF and 3DGS SLAM systems integrating EDINO features. Our DINO-informed pipelines achieve superior performance on the Replica, ScanNet, and TUM compared to state-of-the-art methods.", "date": "2025-07-25", "categories": []}
{"id": "2410.22128", "title": "PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting", "abstract": "We consider the problem of novel view synthesis from unposed images in a single feed-forward. Our framework capitalizes on fast speed, scalability, and high-quality 3D reconstruction and view synthesis capabilities of 3DGS, where we further extend it to offer a practical solution that relaxes common assumptions such as dense image views, accurate camera poses, and substantial image overlaps. We achieve this through identifying and addressing unique challenges arising from the use of pixel-aligned 3DGS: misaligned 3D Gaussians across different views induce noisy or sparse gradients that destabilize training and hinder convergence, especially when above assumptions are not met. To mitigate this, we employ pre-trained monocular depth estimation and visual correspondence models to achieve coarse alignments of 3D Gaussians. We then introduce lightweight, learnable modules to refine depth and pose estimates from the coarse alignments, improving the quality of 3D reconstruction and novel view synthesis. Furthermore, the refined estimates are leveraged to estimate geometry confidence scores, which assess the reliability of 3D Gaussian centers and condition the prediction of Gaussian parameters accordingly. Extensive evaluations on large-scale real-world datasets demonstrate that PF3plat sets a new state-of-the-art across all benchmarks, supported by comprehensive ablation studies validating our design choices. project page: https://cvlab-kaist.github.io/PF3plat/", "date": "2024-10-29", "categories": []}
{"id": "2503.13176", "title": "DeGauss: Dynamic-Static Decomposition with Gaussian Splatting for Distractor-free 3D Reconstruction", "abstract": "Reconstructing clean, distractor-free 3D scenes from real-world captures remains a significant challenge, particularly in highly dynamic and cluttered settings such as egocentric videos. To tackle this problem, we introduce DeGauss, a simple and robust self-supervised framework for dynamic scene reconstruction based on a decoupled dynamic-static Gaussian Splatting design. DeGauss models dynamic elements with foreground Gaussians and static content with background Gaussians, using a probabilistic mask to coordinate their composition and enable independent yet complementary optimization. DeGauss generalizes robustly across a wide range of real-world scenarios, from casual image collections to long, dynamic egocentric videos, without relying on complex heuristics or extensive supervision. Experiments on benchmarks including NeRF-on-the-go, ADT, AEA, Hot3D, and EPIC-Fields demonstrate that DeGauss consistently outperforms existing methods, establishing a strong baseline for generalizable, distractor-free 3D reconstructionin highly dynamic, interaction-rich environments. Project page: https://batfacewayne.github.io/DeGauss.io/", "date": "2025-03-17", "categories": []}
{"id": "2507.18023", "title": "High-fidelity 3D Gaussian Inpainting: preserving multi-view consistency and photorealistic details", "abstract": "Recent advancements in multi-view 3D reconstruction and novel-view synthesis, particularly through Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have greatly enhanced the fidelity and efficiency of 3D content creation. However, inpainting 3D scenes remains a challenging task due to the inherent irregularity of 3D structures and the critical need for maintaining multi-view consistency. In this work, we propose a novel 3D Gaussian inpainting framework that reconstructs complete 3D scenes by leveraging sparse inpainted views. Our framework incorporates an automatic Mask Refinement Process and region-wise Uncertainty-guided Optimization. Specifically, we refine the inpainting mask using a series of operations, including Gaussian scene filtering and back-projection, enabling more accurate localization of occluded regions and realistic boundary restoration. Furthermore, our Uncertainty-guided Fine-grained Optimization strategy, which estimates the importance of each region across multi-view images during training, alleviates multi-view inconsistencies and enhances the fidelity of fine details in the inpainted results. Comprehensive experiments conducted on diverse datasets demonstrate that our approach outperforms existing state-of-the-art methods in both visual quality and view consistency.", "date": "2025-07-24", "categories": []}
{"id": "2507.18231", "title": "PS-GS: Gaussian Splatting for Multi-View Photometric Stereo", "abstract": "Integrating inverse rendering with multi-view photometric stereo (MVPS) yields more accurate 3D reconstructions than the inverse rendering approaches that rely on fixed environment illumination. However, efficient inverse rendering with MVPS remains challenging. To fill this gap, we introduce the Gaussian Splatting for Multi-view Photometric Stereo (PS-GS), which efficiently and jointly estimates the geometry, materials, and lighting of the object that is illuminated by diverse directional lights (multi-light). Our method first reconstructs a standard 2D Gaussian splatting model as the initial geometry. Based on the initialization model, it then proceeds with the deferred inverse rendering by the full rendering equation containing a lighting-computing multi-layer perceptron. During the whole optimization, we regularize the rendered normal maps by the uncalibrated photometric stereo estimated normals. We also propose the 2D Gaussian ray-tracing for single directional light to refine the incident lighting. The regularizations and the use of multi-view and multi-light images mitigate the ill-posed problem of inverse rendering. After optimization, the reconstructed object can be used for novel-view synthesis, relighting, and material and shape editing. Experiments on both synthetic and real datasets demonstrate that our method outperforms prior works in terms of reconstruction accuracy and computational efficiency.", "date": "2025-07-24", "categories": []}
{"id": "2507.18255", "title": "LONG3R: Long Sequence Streaming 3D Reconstruction", "abstract": "Recent advancements in multi-view scene reconstruction have been significant, yet existing methods face limitations when processing streams of input images. These methods either rely on time-consuming offline optimization or are restricted to shorter sequences, hindering their applicability in real-time scenarios. In this work, we propose LONG3R (LOng sequence streaming 3D Reconstruction), a novel model designed for streaming multi-view 3D scene reconstruction over longer sequences. Our model achieves real-time processing by operating recurrently, maintaining and updating memory with each new observation. We first employ a memory gating mechanism to filter relevant memory, which, together with a new observation, is fed into a dual-source refined decoder for coarse-to-fine interaction. To effectively capture long-sequence memory, we propose a 3D spatio-temporal memory that dynamically prunes redundant spatial information while adaptively adjusting resolution along the scene. To enhance our model's performance on long sequences while maintaining training efficiency, we employ a two-stage curriculum training strategy, each stage targeting specific capabilities. Experiments demonstrate that LONG3R outperforms state-of-the-art streaming methods, particularly for longer sequences, while maintaining real-time inference speed. Project page: https://zgchen33.github.io/LONG3R/.", "date": "2025-07-24", "categories": []}
{"id": "2507.18473", "title": "CRUISE: Cooperative Reconstruction and Editing in V2X Scenarios using Gaussian Splatting", "abstract": "Vehicle-to-everything (V2X) communication plays a crucial role in autonomous driving, enabling cooperation between vehicles and infrastructure. While simulation has significantly contributed to various autonomous driving tasks, its potential for data generation and augmentation in V2X scenarios remains underexplored. In this paper, we introduce CRUISE, a comprehensive reconstruction-and-synthesis framework designed for V2X driving environments. CRUISE employs decomposed Gaussian Splatting to accurately reconstruct real-world scenes while supporting flexible editing. By decomposing dynamic traffic participants into editable Gaussian representations, CRUISE allows for seamless modification and augmentation of driving scenes. Furthermore, the framework renders images from both ego-vehicle and infrastructure views, enabling large-scale V2X dataset augmentation for training and evaluation. Our experimental results demonstrate that: 1) CRUISE reconstructs real-world V2X driving scenes with high fidelity; 2) using CRUISE improves 3D detection across ego-vehicle, infrastructure, and cooperative views, as well as cooperative 3D tracking on the V2X-Seq benchmark; and 3) CRUISE effectively generates challenging corner cases.", "date": "2025-07-24", "categories": []}
{"id": "2507.17661", "title": "Monocular Semantic Scene Completion via Masked Recurrent Networks", "abstract": "Monocular Semantic Scene Completion (MSSC) aims to predict the voxel-wise occupancy and semantic category from a single-view RGB image. Existing methods adopt a single-stage framework that aims to simultaneously achieve visible region segmentation and occluded region hallucination, while also being affected by inaccurate depth estimation. Such methods often achieve suboptimal performance, especially in complex scenes. We propose a novel two-stage framework that decomposes MSSC into coarse MSSC followed by the Masked Recurrent Network. Specifically, we propose the Masked Sparse Gated Recurrent Unit (MS-GRU) which concentrates on the occupied regions by the proposed mask updating mechanism, and a sparse GRU design is proposed to reduce the computation cost. Additionally, we propose the distance attention projection to reduce projection errors by assigning different attention scores according to the distance to the observed surface. Experimental results demonstrate that our proposed unified framework, MonoMRN, effectively supports both indoor and outdoor scenes and achieves state-of-the-art performance on the NYUv2 and SemanticKITTI datasets. Furthermore, we conduct robustness analysis under various disturbances, highlighting the role of the Masked Recurrent Network in enhancing the model's resilience to such challenges. The source code is publicly available.", "date": "2025-07-23", "categories": []}
{"id": "2410.13613", "title": "MEGA: Memory-Efficient 4D Gaussian Splatting for Dynamic Scenes", "abstract": "4D Gaussian Splatting (4DGS) has recently emerged as a promising technique for capturing complex dynamic 3D scenes with high fidelity. It utilizes a 4D Gaussian representation and a GPU-friendly rasterizer, enabling rapid rendering speeds. Despite its advantages, 4DGS faces significant challenges, notably the requirement of millions of 4D Gaussians, each with extensive associated attributes, leading to substantial memory and storage cost. This paper introduces a memory-efficient framework for 4DGS. We streamline the color attribute by decomposing it into a per-Gaussian direct color component with only 3 parameters and a shared lightweight alternating current color predictor. This approach eliminates the need for spherical harmonics coefficients, which typically involve up to 144 parameters in classic 4DGS, thereby creating a memory-efficient 4D Gaussian representation. Furthermore, we introduce an entropy-constrained Gaussian deformation technique that uses a deformation field to expand the action range of each Gaussian and integrates an opacity-based entropy loss to limit the number of Gaussians, thus forcing our model to use as few Gaussians as possible to fit a dynamic scene well. With simple half-precision storage and zip compression, our framework achieves a storage reduction by approximately 190$\\times$ and 125$\\times$ on the Technicolor and Neural 3D Video datasets, respectively, compared to the original 4DGS. Meanwhile, it maintains comparable rendering speeds and scene representation quality, setting a new standard in the field. Code is available at https://github.com/Xinjie-Q/MEGA.", "date": "2024-10-17", "categories": []}
{"id": "2507.16144", "title": "LongSplat: Online Generalizable 3D Gaussian Splatting from Long Sequence Images", "abstract": "3D Gaussian Splatting achieves high-fidelity novel view synthesis, but its application to online long-sequence scenarios is still limited. Existing methods either rely on slow per-scene optimization or fail to provide efficient incremental updates, hindering continuous performance. In this paper, we propose LongSplat, an online real-time 3D Gaussian reconstruction framework designed for long-sequence image input. The core idea is a streaming update mechanism that incrementally integrates current-view observations while selectively compressing redundant historical Gaussians. Crucial to this mechanism is our Gaussian-Image Representation (GIR), a representation that encodes 3D Gaussian parameters into a structured, image-like 2D format. GIR simultaneously enables efficient fusion of current-view and historical Gaussians and identity-aware redundancy compression. These functions enable online reconstruction and adapt the model to long sequences without overwhelming memory or computational costs. Furthermore, we leverage an existing image compression method to guide the generation of more compact and higher-quality 3D Gaussians. Extensive evaluations demonstrate that LongSplat achieves state-of-the-art efficiency-quality trade-offs in real-time novel view synthesis, delivering real-time reconstruction while reducing Gaussian counts by 44\\% compared to existing per-pixel Gaussian prediction methods.", "date": "2025-07-22", "categories": []}
{"id": "2507.16290", "title": "Dens3R: A Foundation Model for 3D Geometry Prediction", "abstract": "Recent advances in dense 3D reconstruction have led to significant progress, yet achieving accurate unified geometric prediction remains a major challenge. Most existing methods are limited to predicting a single geometry quantity from input images. However, geometric quantities such as depth, surface normals, and point maps are inherently correlated, and estimating them in isolation often fails to ensure consistency, thereby limiting both accuracy and practical applicability. This motivates us to explore a unified framework that explicitly models the structural coupling among different geometric properties to enable joint regression. In this paper, we present Dens3R, a 3D foundation model designed for joint geometric dense prediction and adaptable to a wide range of downstream tasks. Dens3R adopts a two-stage training framework to progressively build a pointmap representation that is both generalizable and intrinsically invariant. Specifically, we design a lightweight shared encoder-decoder backbone and introduce position-interpolated rotary positional encoding to maintain expressive power while enhancing robustness to high-resolution inputs. By integrating image-pair matching features with intrinsic invariance modeling, Dens3R accurately regresses multiple geometric quantities such as surface normals and depth, achieving consistent geometry perception from single-view to multi-view inputs. Additionally, we propose a post-processing pipeline that supports geometrically consistent multi-view inference. Extensive experiments demonstrate the superior performance of Dens3R across various dense 3D prediction tasks and highlight its potential for broader applications.", "date": "2025-07-22", "categories": []}
{"id": "2507.16406", "title": "Sparse-View 3D Reconstruction: Recent Advances and Open Challenges", "abstract": "Sparse-view 3D reconstruction is essential for applications in which dense image acquisition is impractical, such as robotics, augmented/virtual reality (AR/VR), and autonomous systems. In these settings, minimal image overlap prevents reliable correspondence matching, causing traditional methods, such as structure-from-motion (SfM) and multiview stereo (MVS), to fail. This survey reviews the latest advances in neural implicit models (e.g., NeRF and its regularized versions), explicit point-cloud-based approaches (e.g., 3D Gaussian Splatting), and hybrid frameworks that leverage priors from diffusion and vision foundation models (VFMs).We analyze how geometric regularization, explicit shape modeling, and generative inference are used to mitigate artifacts such as floaters and pose ambiguities in sparse-view settings. Comparative results on standard benchmarks reveal key trade-offs between the reconstruction accuracy, efficiency, and generalization. Unlike previous reviews, our survey provides a unified perspective on geometry-based, neural implicit, and generative (diffusion-based) methods. We highlight the persistent challenges in domain generalization and pose-free reconstruction and outline future directions for developing 3D-native generative priors and achieving real-time, unconstrained sparse-view reconstruction.", "date": "2025-07-22", "categories": []}
{"id": "2408.10789", "title": "Self-supervised Learning of Hybrid Part-aware 3D Representations of 2D Gaussians and Superquadrics", "abstract": "Low-level 3D representations, such as point clouds, meshes, NeRFs and 3D Gaussians, are commonly used for modeling 3D objects and scenes. However, cognitive studies indicate that human perception operates at higher levels and interprets 3D environments by decomposing them into meaningful structural parts, rather than low-level elements like points or voxels. Structured geometric decomposition enhances scene interpretability and facilitates downstream tasks requiring component-level manipulation. In this work, we introduce PartGS, a self-supervised part-aware reconstruction framework that integrates 2D Gaussians and superquadrics to parse objects and scenes into an interpretable decomposition, leveraging multi-view image inputs to uncover 3D structural information. Our method jointly optimizes superquadric meshes and Gaussians by coupling their parameters within a hybrid representation. On one hand, superquadrics enable the representation of a wide range of shape primitives, facilitating flexible and meaningful decompositions. On the other hand, 2D Gaussians capture detailed texture and geometric details, ensuring high-fidelity appearance and geometry reconstruction. Operating in a self-supervised manner, our approach demonstrates superior performance compared to state-of-the-art methods across extensive experiments on the DTU, ShapeNet, and real-world datasets.", "date": "2024-08-20", "categories": []}
{"id": "2412.14939", "title": "GURecon: Learning Detailed 3D Geometric Uncertainties for Neural Surface Reconstruction", "abstract": "Neural surface representation has demonstrated remarkable success in the areas of novel view synthesis and 3D reconstruction. However, assessing the geometric quality of 3D reconstructions in the absence of ground truth mesh remains a significant challenge, due to its rendering-based optimization process and entangled learning of appearance and geometry with photometric losses. In this paper, we present a novel framework, i.e, GURecon, which establishes a geometric uncertainty field for the neural surface based on geometric consistency. Different from existing methods that rely on rendering-based measurement, GURecon models a continuous 3D uncertainty field for the reconstructed surface, and is learned by an online distillation approach without introducing real geometric information for supervision. Moreover, in order to mitigate the interference of illumination on geometric consistency, a decoupled field is learned and exploited to finetune the uncertainty field. Experiments on various datasets demonstrate the superiority of GURecon in modeling 3D geometric uncertainty, as well as its plug-and-play extension to various neural surface representations and improvement on downstream tasks such as incremental reconstruction. The code and supplementary material are available on the project website: https://zju3dv.github.io/GURecon/.", "date": "2024-12-19", "categories": []}
{"id": "2507.08416", "title": "InstaScene: Towards Complete 3D Instance Decomposition and Reconstruction from Cluttered Scenes", "abstract": "Humans can naturally identify and mentally complete occluded objects in cluttered environments. However, imparting similar cognitive ability to robotics remains challenging even with advanced reconstruction techniques, which models scenes as undifferentiated wholes and fails to recognize complete object from partial observations. In this paper, we propose InstaScene, a new paradigm towards holistic 3D perception of complex scenes with a primary goal: decomposing arbitrary instances while ensuring complete reconstruction. To achieve precise decomposition, we develop a novel spatial contrastive learning by tracing rasterization of each instance across views, significantly enhancing semantic supervision in cluttered scenes. To overcome incompleteness from limited observations, we introduce in-situ generation that harnesses valuable observations and geometric cues, effectively guiding 3D generative models to reconstruct complete instances that seamlessly align with the real world. Experiments on scene decomposition and object completion across complex real-world and synthetic scenes demonstrate that our method achieves superior decomposition accuracy while producing geometrically faithful and visually intact objects.", "date": "2025-07-11", "categories": []}
{"id": "2507.14624", "title": "Real-Time Scene Reconstruction using Light Field Probes", "abstract": "Reconstructing photo-realistic large-scale scenes from images, for example at city scale, is a long-standing problem in computer graphics. Neural rendering is an emerging technique that enables photo-realistic image synthesis from previously unobserved viewpoints; however, state-of-the-art neural rendering methods have difficulty efficiently rendering a high complex large-scale scene because these methods typically trade scene size, fidelity, and rendering speed for quality. The other stream of techniques utilizes scene geometries for reconstruction. But the cost of building and maintaining a large set of geometry data increases as scene size grows. Our work explores novel view synthesis methods that efficiently reconstruct complex scenes without explicit use of scene geometries. Specifically, given sparse images of the scene (captured from the real world), we reconstruct intermediate, multi-scale, implicit representations of scene geometries. In this way, our method avoids explicitly relying on scene geometry, significantly reducing the computational cost of maintaining large 3D data. Unlike current methods, we reconstruct the scene using a probe data structure. Probe data hold highly accurate depth information of dense data points, enabling the reconstruction of highly complex scenes. By reconstructing the scene using probe data, the rendering cost is independent of the complexity of the scene. As such, our approach combines geometry reconstruction and novel view synthesis. Moreover, when rendering large-scale scenes, compressing and streaming probe data is more efficient than using explicit scene geometry. Therefore, our neural representation approach can potentially be applied to virtual reality (VR) and augmented reality (AR) applications.", "date": "2025-07-19", "categories": []}
{"id": "2507.15454", "title": "ObjectGS: Object-aware Scene Reconstruction and Scene Understanding via Gaussian Splatting", "abstract": "3D Gaussian Splatting is renowned for its high-fidelity reconstructions and real-time novel view synthesis, yet its lack of semantic understanding limits object-level perception. In this work, we propose ObjectGS, an object-aware framework that unifies 3D scene reconstruction with semantic understanding. Instead of treating the scene as a unified whole, ObjectGS models individual objects as local anchors that generate neural Gaussians and share object IDs, enabling precise object-level reconstruction. During training, we dynamically grow or prune these anchors and optimize their features, while a one-hot ID encoding with a classification loss enforces clear semantic constraints. We show through extensive experiments that ObjectGS not only outperforms state-of-the-art methods on open-vocabulary and panoptic segmentation tasks, but also integrates seamlessly with applications like mesh extraction and scene editing. Project page: https://ruijiezhu94.github.io/ObjectGS_page", "date": "2025-07-21", "categories": []}
{"id": "2507.13719", "title": "Augmented Reality in Cultural Heritage: A Dual-Model Pipeline for 3D Artwork Reconstruction", "abstract": "This paper presents an innovative augmented reality pipeline tailored for museum environments, aimed at recognizing artworks and generating accurate 3D models from single images. By integrating two complementary pre-trained depth estimation models, i.e., GLPN for capturing global scene structure and Depth-Anything for detailed local reconstruction, the proposed approach produces optimized depth maps that effectively represent complex artistic features. These maps are then converted into high-quality point clouds and meshes, enabling the creation of immersive AR experiences. The methodology leverages state-of-the-art neural network architectures and advanced computer vision techniques to overcome challenges posed by irregular contours and variable textures in artworks. Experimental results demonstrate significant improvements in reconstruction accuracy and visual realism, making the system a highly robust tool for museums seeking to enhance visitor engagement through interactive digital content.", "date": "2025-07-18", "categories": []}
{"id": "2412.06770", "title": "Dynamic EventNeRF: Reconstructing General Dynamic Scenes from Multi-view RGB and Event Streams", "abstract": "Volumetric reconstruction of dynamic scenes is an important problem in computer vision. It is especially challenging in poor lighting and with fast motion. This is partly due to limitations of RGB cameras: To capture frames under low lighting, the exposure time needs to be increased, which leads to more motion blur. In contrast, event cameras, which record changes in pixel brightness asynchronously, are much less dependent on lighting, making them more suitable for recording fast motion. We hence propose the first method to spatiotemporally reconstruct a scene from sparse multi-view event streams and sparse RGB frames. We train a sequence of cross-faded time-conditioned NeRF models, one per short recording segment. The individual segments are supervised with a set of event- and RGB-based losses and sparse-view regularisation. We assemble a real-world multi-view camera rig with six static event cameras around the object and record a benchmark multi-view event stream dataset of challenging motions. Our work outperforms RGB-based baselines, producing state-of-the-art results, and opens up the topic of multi-view event-based reconstruction as a new path for fast scene capture beyond RGB cameras. The code and the data are released at https://4dqv.mpi-inf.mpg.de/DynEventNeRF/", "date": "2024-12-09", "categories": []}
{"id": "2507.11129", "title": "MMOne: Representing Multiple Modalities in One Scene", "abstract": "Humans perceive the world through multimodal cues to understand and interact with the environment. Learning a scene representation for multiple modalities enhances comprehension of the physical world. However, modality conflicts, arising from inherent distinctions among different modalities, present two critical challenges: property disparity and granularity disparity. To address these challenges, we propose a general framework, MMOne, to represent multiple modalities in one scene, which can be readily extended to additional modalities. Specifically, a modality modeling module with a novel modality indicator is proposed to capture the unique properties of each modality. Additionally, we design a multimodal decomposition mechanism to separate multi-modal Gaussians into single-modal Gaussians based on modality differences. We address the essential distinctions among modalities by disentangling multimodal information into shared and modality-specific components, resulting in a more compact and efficient multimodal scene representation. Extensive experiments demonstrate that our method consistently enhances the representation capability for each modality and is scalable to additional modalities. The code is available at https://github.com/Neal2020GitHub/MMOne.", "date": "2025-07-15", "categories": []}
{"id": "2503.12335", "title": "GS-I$^{3}$: Gaussian Splatting for Surface Reconstruction from Illumination-Inconsistent Images", "abstract": "Accurate geometric surface reconstruction, providing essential environmental information for navigation and manipulation tasks, is critical for enabling robotic self-exploration and interaction. Recently, 3D Gaussian Splatting (3DGS) has gained significant attention in the field of surface reconstruction due to its impressive geometric quality and computational efficiency. While recent relevant advancements in novel view synthesis under inconsistent illumination using 3DGS have shown promise, the challenge of robust surface reconstruction under such conditions is still being explored. To address this challenge, we propose a method called GS-3I. Specifically, to mitigate 3D Gaussian optimization bias caused by underexposed regions in single-view images, based on Convolutional Neural Network (CNN), a tone mapping correction framework is introduced. Furthermore, inconsistent lighting across multi-view images, resulting from variations in camera settings and complex scene illumination, often leads to geometric constraint mismatches and deviations in the reconstructed surface. To overcome this, we propose a normal compensation mechanism that integrates reference normals extracted from single-view image with normals computed from multi-view observations to effectively constrain geometric inconsistencies. Extensive experimental evaluations demonstrate that GS-3I can achieve robust and accurate surface reconstruction across complex illumination scenarios, highlighting its effectiveness and versatility in this critical challenge. https://github.com/TFwang-9527/GS-3I", "date": "2025-03-16", "categories": []}
{"id": "2506.05935", "title": "SurGSplat: Progressive Geometry-Constrained Gaussian Splatting for Surgical Scene Reconstruction", "abstract": "Intraoperative navigation relies heavily on precise 3D reconstruction to ensure accuracy and safety during surgical procedures. However, endoscopic scenarios present unique challenges, including sparse features and inconsistent lighting, which render many existing Structure-from-Motion (SfM)-based methods inadequate and prone to reconstruction failure. To mitigate these constraints, we propose SurGSplat, a novel paradigm designed to progressively refine 3D Gaussian Splatting (3DGS) through the integration of geometric constraints. By enabling the detailed reconstruction of vascular structures and other critical features, SurGSplat provides surgeons with enhanced visual clarity, facilitating precise intraoperative decision-making. Experimental evaluations demonstrate that SurGSplat achieves superior performance in both novel view synthesis (NVS) and pose estimation accuracy, establishing it as a high-fidelity and efficient solution for surgical scene reconstruction. More information and results can be found on the page https://surgsplat.github.io/.", "date": "2025-06-06", "categories": []}
{"id": "2507.12095", "title": "BRUM: Robust 3D Vehicle Reconstruction from 360 Sparse Images", "abstract": "Accurate 3D reconstruction of vehicles is vital for applications such as vehicle inspection, predictive maintenance, and urban planning. Existing methods like Neural Radiance Fields and Gaussian Splatting have shown impressive results but remain limited by their reliance on dense input views, which hinders real-world applicability. This paper addresses the challenge of reconstructing vehicles from sparse-view inputs, leveraging depth maps and a robust pose estimation architecture to synthesize novel views and augment training data. Specifically, we enhance Gaussian Splatting by integrating a selective photometric loss, applied only to high-confidence pixels, and replacing standard Structure-from-Motion pipelines with the DUSt3R architecture to improve camera pose estimation. Furthermore, we present a novel dataset featuring both synthetic and real-world public transportation vehicles, enabling extensive evaluation of our approach. Experimental results demonstrate state-of-the-art performance across multiple benchmarks, showcasing the method's ability to achieve high-quality reconstructions even under constrained input conditions.", "date": "2025-07-16", "categories": []}
{"id": "2507.11321", "title": "A Mixed-Primitive-based Gaussian Splatting Method for Surface Reconstruction", "abstract": "Recently, Gaussian Splatting (GS) has received a lot of attention in surface reconstruction. However, while 3D objects can be of complex and diverse shapes in the real world, existing GS-based methods only limitedly use a single type of splatting primitive (Gaussian ellipse or Gaussian ellipsoid) to represent object surfaces during their reconstruction. In this paper, we highlight that this can be insufficient for object surfaces to be represented in high quality. Thus, we propose a novel framework that, for the first time, enables Gaussian Splatting to incorporate multiple types of (geometrical) primitives during its surface reconstruction process. Specifically, in our framework, we first propose a compositional splatting strategy, enabling the splatting and rendering of different types of primitives in the Gaussian Splatting pipeline. In addition, we also design our framework with a mixed-primitive-based initialization strategy and a vertex pruning mechanism to further promote its surface representation learning process to be well executed leveraging different types of primitives. Extensive experiments show the efficacy of our framework and its accurate surface reconstruction performance.", "date": "2025-07-15", "categories": []}
{"id": "2411.09145", "title": "Self-Supervised Monocular 4D Scene Reconstruction for Egocentric Videos", "abstract": "Egocentric videos provide valuable insights into human interactions with the physical world, which has sparked growing interest in the computer vision and robotics communities. A critical challenge in fully understanding the geometry and dynamics of egocentric videos is dense scene reconstruction. However, the lack of high-quality labeled datasets in this field has hindered the effectiveness of current supervised learning methods. In this work, we aim to address this issue by exploring an self-supervised dynamic scene reconstruction approach. We introduce EgoMono4D, a novel model that unifies the estimation of multiple variables necessary for Egocentric Monocular 4D reconstruction, including camera intrinsic, camera poses, and video depth, all within a fast feed-forward framework. Starting from pretrained single-frame depth and intrinsic estimation model, we extend it with camera poses estimation and align multi-frame results on large-scale unlabeled egocentric videos. We evaluate EgoMono4D in both in-domain and zero-shot generalization settings, achieving superior performance in dense pointclouds sequence reconstruction compared to all baselines. EgoMono4D represents the first attempt to apply self-supervised learning for pointclouds sequence reconstruction to the label-scarce egocentric field, enabling fast, dense, and generalizable reconstruction. The interactable visualization, code and trained models are released https://egomono4d.github.io/", "date": "2024-11-14", "categories": []}
{"id": "2505.11992", "title": "SpatialCrafter: Unleashing the Imagination of Video Diffusion Models for Scene Reconstruction from Limited Observations", "abstract": "Novel view synthesis (NVS) boosts immersive experiences in computer vision and graphics. Existing techniques, though progressed, rely on dense multi-view observations, restricting their application. This work takes on the challenge of reconstructing photorealistic 3D scenes from sparse or single-view inputs. We introduce SpatialCrafter, a framework that leverages the rich knowledge in video diffusion models to generate plausible additional observations, thereby alleviating reconstruction ambiguity. Through a trainable camera encoder and an epipolar attention mechanism for explicit geometric constraints, we achieve precise camera control and 3D consistency, further reinforced by a unified scale estimation strategy to handle scale discrepancies across datasets. Furthermore, by integrating monocular depth priors with semantic features in the video latent space, our framework directly regresses 3D Gaussian primitives and efficiently processes long-sequence features using a hybrid network structure. Extensive experiments show our method enhances sparse view reconstruction and restores the realistic appearance of 3D scenes.", "date": "2025-05-17", "categories": []}
{"id": "2507.07521", "title": "Spline Deformation Field", "abstract": "Trajectory modeling of dense points usually employs implicit deformation fields, represented as neural networks that map coordinates to relate canonical spatial positions to temporal offsets. However, the inductive biases inherent in neural networks can hinder spatial coherence in ill-posed scenarios. Current methods focus either on enhancing encoding strategies for deformation fields, often resulting in opaque and less intuitive models, or adopt explicit techniques like linear blend skinning, which rely on heuristic-based node initialization. Additionally, the potential of implicit representations for interpolating sparse temporal signals remains under-explored. To address these challenges, we propose a spline-based trajectory representation, where the number of knots explicitly determines the degrees of freedom. This approach enables efficient analytical derivation of velocities, preserving spatial coherence and accelerations, while mitigating temporal fluctuations. To model knot characteristics in both spatial and temporal domains, we introduce a novel low-rank time-variant spatial encoding, replacing conventional coupled spatiotemporal techniques. Our method demonstrates superior performance in temporal interpolation for fitting continuous fields with sparse inputs. Furthermore, it achieves competitive dynamic scene reconstruction quality compared to state-of-the-art methods while enhancing motion coherence without relying on linear blend skinning or as-rigid-as-possible constraints.", "date": "2025-07-10", "categories": []}
{"id": "2507.08448", "title": "Review of Feed-forward 3D Reconstruction: From DUSt3R to VGGT", "abstract": "3D reconstruction, which aims to recover the dense three-dimensional structure of a scene, is a cornerstone technology for numerous applications, including augmented/virtual reality, autonomous driving, and robotics. While traditional pipelines like Structure from Motion (SfM) and Multi-View Stereo (MVS) achieve high precision through iterative optimization, they are limited by complex workflows, high computational cost, and poor robustness in challenging scenarios like texture-less regions. Recently, deep learning has catalyzed a paradigm shift in 3D reconstruction. A new family of models, exemplified by DUSt3R, has pioneered a feed-forward approach. These models employ a unified deep network to jointly infer camera poses and dense geometry directly from an Unconstrained set of images in a single forward pass. This survey provides a systematic review of this emerging domain. We begin by dissecting the technical framework of these feed-forward models, including their Transformer-based correspondence modeling, joint pose and geometry regression mechanisms, and strategies for scaling from two-view to multi-view scenarios. To highlight the disruptive nature of this new paradigm, we contrast it with both traditional pipelines and earlier learning-based methods like MVSNet. Furthermore, we provide an overview of relevant datasets and evaluation metrics. Finally, we discuss the technology's broad application prospects and identify key future challenges and opportunities, such as model accuracy and scalability, and handling dynamic scenes.", "date": "2025-07-11", "categories": []}
{"id": "2503.04720", "title": "FluidNexus: 3D Fluid Reconstruction and Prediction from a Single Video", "abstract": "We study reconstructing and predicting 3D fluid appearance and velocity from a single video. Current methods require multi-view videos for fluid reconstruction. We present FluidNexus, a novel framework that bridges video generation and physics simulation to tackle this task. Our key insight is to synthesize multiple novel-view videos as references for reconstruction. FluidNexus consists of two key components: (1) a novel-view video synthesizer that combines frame-wise view synthesis with video diffusion refinement for generating realistic videos, and (2) a physics-integrated particle representation coupling differentiable simulation and rendering to simultaneously facilitate 3D fluid reconstruction and prediction. To evaluate our approach, we collect two new real-world fluid datasets featuring textured backgrounds and object interactions. Our method enables dynamic novel view synthesis, future prediction, and interaction simulation from a single fluid video. Project website: https://yuegao.me/FluidNexus.", "date": "2025-03-06", "categories": []}
{"id": "2506.01933", "title": "E3D-Bench: A Benchmark for End-to-End 3D Geometric Foundation Models", "abstract": "Spatial intelligence, encompassing 3D reconstruction, perception, and reasoning, is fundamental to applications such as robotics, aerial imaging, and extended reality. A key enabler is the real-time, accurate estimation of core 3D attributes (camera parameters, point clouds, depth maps, and 3D point tracks) from unstructured or streaming imagery. Inspired by the success of large foundation models in language and 2D vision, a new class of end-to-end 3D geometric foundation models (GFMs) has emerged, directly predicting dense 3D representations in a single feed-forward pass, eliminating the need for slow or unavailable precomputed camera parameters. Since late 2023, the field has exploded with diverse variants, but systematic evaluation is lacking. In this work, we present the first comprehensive benchmark for 3D GFMs, covering five core tasks: sparse-view depth estimation, video depth estimation, 3D reconstruction, multi-view pose estimation, novel view synthesis, and spanning both standard and challenging out-of-distribution datasets. Our standardized toolkit automates dataset handling, evaluation protocols, and metric computation to ensure fair, reproducible comparisons. We evaluate 16 state-of-the-art GFMs, revealing their strengths and limitations across tasks and domains, and derive key insights to guide future model scaling and optimization. All code, evaluation scripts, and processed data will be publicly released to accelerate research in 3D spatial intelligence.", "date": "2025-06-02", "categories": []}
{"id": "2507.07465", "title": "SD-GS: Structured Deformable 3D Gaussians for Efficient Dynamic Scene Reconstruction", "abstract": "Current 4D Gaussian frameworks for dynamic scene reconstruction deliver impressive visual fidelity and rendering speed, however, the inherent trade-off between storage costs and the ability to characterize complex physical motions significantly limits the practical application of these methods. To tackle these problems, we propose SD-GS, a compact and efficient dynamic Gaussian splatting framework for complex dynamic scene reconstruction, featuring two key contributions. First, we introduce a deformable anchor grid, a hierarchical and memory-efficient scene representation where each anchor point derives multiple 3D Gaussians in its local spatiotemporal region and serves as the geometric backbone of the 3D scene. Second, to enhance modeling capability for complex motions, we present a deformation-aware densification strategy that adaptively grows anchors in under-reconstructed high-dynamic regions while reducing redundancy in static areas, achieving superior visual quality with fewer anchors. Experimental results demonstrate that, compared to state-of-the-art methods, SD-GS achieves an average of 60\\% reduction in model size and an average of 100\\% improvement in FPS, significantly enhancing computational efficiency while maintaining or even surpassing visual quality.", "date": "2025-07-10", "categories": []}
{"id": "2411.12510", "title": "PR-ENDO: Physically Based Relightable Gaussian Splatting for Endoscopy", "abstract": "Endoluminal endoscopic procedures are essential for diagnosing colorectal cancer and other severe conditions in the digestive tract, urogenital system, and airways. 3D reconstruction and novel-view synthesis from endoscopic images are promising tools for enhancing diagnosis. Moreover, integrating physiological deformations and interaction with the endoscope enables the development of simulation tools from real video data. However, constrained camera trajectories and view-dependent lighting create artifacts, leading to inaccurate or overfitted reconstructions. We present PR-ENDO, a novel 3D reconstruction framework leveraging the unique property of endoscopic imaging, where a single light source is closely aligned with the camera. Our method separates light effects from tissue properties. PR-ENDO enhances 3D Gaussian Splatting with a physically based relightable model. We boost the traditional light transport formulation with a specialized MLP capturing complex light-related effects while ensuring reduced artifacts and better generalization across novel views. PR-ENDO achieves superior reconstruction quality compared to baseline methods on both public and in-house datasets. Unlike existing approaches, PR-ENDO enables tissue modifications while preserving a physically accurate response to light, making it closer to real-world clinical use.", "date": "2024-11-19", "categories": []}
{"id": "2502.16351", "title": "AquaNeRF: Neural Radiance Fields in Underwater Media with Distractor Removal", "abstract": "Neural radiance field (NeRF) research has made significant progress in modeling static video content captured in the wild. However, current models and rendering processes rarely consider scenes captured underwater, which are useful for studying and filming ocean life. They fail to address visual artifacts unique to underwater scenes, such as moving fish and suspended particles. This paper introduces a novel NeRF renderer and optimization scheme for an implicit MLP-based NeRF model. Our renderer reduces the influence of floaters and moving objects that interfere with static objects of interest by estimating a single surface per ray. We use a Gaussian weight function with a small offset to ensure that the transmittance of the surrounding media remains constant. Additionally, we enhance our model with a depth-based scaling function to upscale gradients for near-camera volumes. Overall, our method outperforms the baseline Nerfacto by approximately 7.5\\% and SeaThru-NeRF by 6.2% in terms of PSNR. Subjective evaluation also shows a significant reduction of artifacts while preserving details of static targets and background compared to the state of the arts.", "date": "2025-02-22", "categories": []}
{"id": "2507.06684", "title": "Photometric Stereo using Gaussian Splatting and inverse rendering", "abstract": "Recent state-of-the-art algorithms in photometric stereo rely on neural networks and operate either through prior learning or inverse rendering optimization. Here, we revisit the problem of calibrated photometric stereo by leveraging recent advances in 3D inverse rendering using the Gaussian Splatting formalism. This allows us to parameterize the 3D scene to be reconstructed and optimize it in a more interpretable manner. Our approach incorporates a simplified model for light representation and demonstrates the potential of the Gaussian Splatting rendering engine for the photometric stereo problem.", "date": "2025-07-09", "categories": []}
{"id": "2412.01717", "title": "Driving View Synthesis on Free-form Trajectories with Generative Prior", "abstract": "Driving view synthesis along free-form trajectories is essential for realistic driving simulations, enabling closed-loop evaluation of end-to-end driving policies. Existing methods excel at view interpolation along recorded paths but struggle to generalize to novel trajectories due to limited viewpoints in driving videos. To tackle this challenge, we propose DriveX, a novel free-form driving view synthesis framework, that progressively distills generative prior into the 3D Gaussian model during its optimization. Within this framework, we utilize a video diffusion model to refine the degraded novel trajectory renderings from the in-training Gaussian model, while the restored videos in turn serve as additional supervision for optimizing the 3D Gaussian. Concretely, we craft an inpainting-based video restoration task, which can disentangle the identification of degraded regions from the generative capability of the diffusion model and remove the need of simulating specific degraded pattern in the training of the diffusion model. To further enhance the consistency and fidelity of generated contents, the pseudo ground truth is progressively updated with gradually improved novel trajectory rendering, allowing both components to co-adapt and reinforce each other while minimizing the disruption on the optimization. By tightly integrating 3D scene representation with generative prior, DriveX achieves high-quality view synthesis beyond recorded trajectories in real time--unlocking new possibilities for flexible and realistic driving simulations on free-form trajectories.", "date": "2024-12-02", "categories": []}
{"id": "2506.22099", "title": "B\\'ezierGS: Dynamic Urban Scene Reconstruction with B\\'ezier Curve Gaussian Splatting", "abstract": "The realistic reconstruction of street scenes is critical for developing real-world simulators in autonomous driving. Most existing methods rely on object pose annotations, using these poses to reconstruct dynamic objects and move them during the rendering process. This dependence on high-precision object annotations limits large-scale and extensive scene reconstruction. To address this challenge, we propose B\\'ezier curve Gaussian splatting (B\\'ezierGS), which represents the motion trajectories of dynamic objects using learnable B\\'ezier curves. This approach fully leverages the temporal information of dynamic objects and, through learnable curve modeling, automatically corrects pose errors. By introducing additional supervision on dynamic object rendering and inter-curve consistency constraints, we achieve reasonable and accurate separation and reconstruction of scene elements. Extensive experiments on the Waymo Open Dataset and the nuPlan benchmark demonstrate that B\\'ezierGS outperforms state-of-the-art alternatives in both dynamic and static scene components reconstruction and novel view synthesis.", "date": "2025-06-27", "categories": []}
{"id": "2507.05627", "title": "DreamGrasp: Zero-Shot 3D Multi-Object Reconstruction from Partial-View Images for Robotic Manipulation", "abstract": "Partial-view 3D recognition -- reconstructing 3D geometry and identifying object instances from a few sparse RGB images -- is an exceptionally challenging yet practically essential task, particularly in cluttered, occluded real-world settings where full-view or reliable depth data are often unavailable. Existing methods, whether based on strong symmetry priors or supervised learning on curated datasets, fail to generalize to such scenarios. In this work, we introduce DreamGrasp, a framework that leverages the imagination capability of large-scale pre-trained image generative models to infer the unobserved parts of a scene. By combining coarse 3D reconstruction, instance segmentation via contrastive learning, and text-guided instance-wise refinement, DreamGrasp circumvents limitations of prior methods and enables robust 3D reconstruction in complex, multi-object environments. Our experiments show that DreamGrasp not only recovers accurate object geometry but also supports downstream tasks like sequential decluttering and target retrieval with high success rates.", "date": "2025-07-08", "categories": []}
{"id": "2507.05952", "title": "High-Fidelity and Generalizable Neural Surface Reconstruction with Sparse Feature Volumes", "abstract": "Generalizable neural surface reconstruction has become a compelling technique to reconstruct from few images without per-scene optimization, where dense 3D feature volume has proven effective as a global representation of scenes. However, the dense representation does not scale well to increasing voxel resolutions, severely limiting the reconstruction quality. We thus present a sparse representation method, that maximizes memory efficiency and enables significantly higher resolution reconstructions on standard hardware. We implement this through a two-stage approach: First training a network to predict voxel occupancies from posed images and associated depth maps, then computing features and performing volume rendering only in voxels with sufficiently high occupancy estimates. To support this sparse representation, we developed custom algorithms for efficient sampling, feature aggregation, and querying from sparse volumes-overcoming the dense-volume assumptions inherent in existing works. Experiments on public datasets demonstrate that our approach reduces storage requirements by more than 50 times without performance degradation, enabling reconstructions at $512^3$ resolution compared to the typical $128^3$ on similar hardware, and achieving superior reconstruction accuracy over current state-of-the-art methods.", "date": "2025-07-08", "categories": []}
{"id": "2403.20309", "title": "InstantSplat: Sparse-view Gaussian Splatting in Seconds", "abstract": "While neural 3D reconstruction has advanced substantially, its performance significantly degrades with sparse-view data, which limits its broader applicability, since SfM is often unreliable in sparse-view scenarios where feature matches are scarce. In this paper, we introduce InstantSplat, a novel approach for addressing sparse-view 3D scene reconstruction at lightning-fast speed. InstantSplat employs a self-supervised framework that optimizes 3D scene representation and camera poses by unprojecting 2D pixels into 3D space and aligning them using differentiable neural rendering. The optimization process is initialized with a large-scale trained geometric foundation model, which provides dense priors that yield initial points through model inference, after which we further optimize all scene parameters using photometric errors. To mitigate redundancy introduced by the prior model, we propose a co-visibility-based geometry initialization, and a Gaussian-based bundle adjustment is employed to rapidly adapt both the scene representation and camera parameters without relying on a complex adaptive density control process. Overall, InstantSplat is compatible with multiple point-based representations for view synthesis and surface reconstruction. It achieves an acceleration of over 30x in reconstruction and improves visual quality (SSIM) from 0.3755 to 0.7624 compared to traditional SfM with 3D-GS.", "date": "2024-03-29", "categories": []}
{"id": "2501.01949", "title": "VideoLifter: Lifting Videos to 3D with Fast Hierarchical Stereo Alignment", "abstract": "Efficiently reconstructing 3D scenes from monocular video remains a core challenge in computer vision, vital for applications in virtual reality, robotics, and scene understanding. Recently, frame-by-frame progressive reconstruction without camera poses is commonly adopted, incurring high computational overhead and compounding errors when scaling to longer videos. To overcome these issues, we introduce VideoLifter, a novel video-to-3D pipeline that leverages a local-to-global strategy on a fragment basis, achieving both extreme efficiency and SOTA quality. Locally, VideoLifter leverages learnable 3D priors to register fragments, extracting essential information for subsequent 3D Gaussian initialization with enforced inter-fragment consistency and optimized efficiency. Globally, it employs a tree-based hierarchical merging method with key frame guidance for inter-fragment alignment, pairwise merging with Gaussian point pruning, and subsequent joint optimization to ensure global consistency while efficiently mitigating cumulative errors. This approach significantly accelerates the reconstruction process, reducing training time by over 82% while holding better visual quality than current SOTA methods.", "date": "2025-01-03", "categories": []}
{"id": "2503.08135", "title": "ArticulatedGS: Self-supervised Digital Twin Modeling of Articulated Objects using 3D Gaussian Splatting", "abstract": "We tackle the challenge of concurrent reconstruction at the part level with the RGB appearance and estimation of motion parameters for building digital twins of articulated objects using the 3D Gaussian Splatting (3D-GS) method. With two distinct sets of multi-view imagery, each depicting an object in separate static articulation configurations, we reconstruct the articulated object in 3D Gaussian representations with both appearance and geometry information at the same time. Our approach decoupled multiple highly interdependent parameters through a multi-step optimization process, thereby achieving a stable optimization procedure and high-quality outcomes. We introduce ArticulatedGS, a self-supervised, comprehensive framework that autonomously learns to model shapes and appearances at the part level and synchronizes the optimization of motion parameters, all without reliance on 3D supervision, motion cues, or semantic labels. Our experimental results demonstrate that, among comparable methodologies, our approach has achieved optimal outcomes in terms of part segmentation accuracy, motion estimation accuracy, and visual quality.", "date": "2025-03-11", "categories": []}
{"id": "2507.04408", "title": "A View-consistent Sampling Method for Regularized Training of Neural Radiance Fields", "abstract": "Neural Radiance Fields (NeRF) has emerged as a compelling framework for scene representation and 3D recovery. To improve its performance on real-world data, depth regularizations have proven to be the most effective ones. However, depth estimation models not only require expensive 3D supervision in training, but also suffer from generalization issues. As a result, the depth estimations can be erroneous in practice, especially for outdoor unbounded scenes. In this paper, we propose to employ view-consistent distributions instead of fixed depth value estimations to regularize NeRF training. Specifically, the distribution is computed by utilizing both low-level color features and high-level distilled features from foundation models at the projected 2D pixel-locations from per-ray sampled 3D points. By sampling from the view-consistency distributions, an implicit regularization is imposed on the training of NeRF. We also utilize a depth-pushing loss that works in conjunction with the sampling technique to jointly provide effective regularizations for eliminating the failure modes. Extensive experiments conducted on various scenes from public datasets demonstrate that our proposed method can generate significantly better novel view synthesis results than state-of-the-art NeRF variants as well as different depth regularization methods.", "date": "2025-07-06", "categories": []}
{"id": "2507.02363", "title": "LocalDyGS: Multi-view Global Dynamic Scene Modeling via Adaptive Local Implicit Feature Decoupling", "abstract": "Due to the complex and highly dynamic motions in the real world, synthesizing dynamic videos from multi-view inputs for arbitrary viewpoints is challenging. Previous works based on neural radiance field or 3D Gaussian splatting are limited to modeling fine-scale motion, greatly restricting their application. In this paper, we introduce LocalDyGS, which consists of two parts to adapt our method to both large-scale and fine-scale motion scenes: 1) We decompose a complex dynamic scene into streamlined local spaces defined by seeds, enabling global modeling by capturing motion within each local space. 2) We decouple static and dynamic features for local space motion modeling. A static feature shared across time steps captures static information, while a dynamic residual field provides time-specific features. These are combined and decoded to generate Temporal Gaussians, modeling motion within each local space. As a result, we propose a novel dynamic scene reconstruction framework to model highly dynamic real-world scenes more realistically. Our method not only demonstrates competitive performance on various fine-scale datasets compared to state-of-the-art (SOTA) methods, but also represents the first attempt to model larger and more complex highly dynamic scenes. Project page: https://wujh2001.github.io/LocalDyGS/.", "date": "2025-07-03", "categories": []}
{"id": "2507.02546", "title": "MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details", "abstract": "We propose MoGe-2, an advanced open-domain geometry estimation model that recovers a metric scale 3D point map of a scene from a single image. Our method builds upon the recent monocular geometry estimation approach, MoGe, which predicts affine-invariant point maps with unknown scales. We explore effective strategies to extend MoGe for metric geometry prediction without compromising the relative geometry accuracy provided by the affine-invariant point representation. Additionally, we discover that noise and errors in real data diminish fine-grained detail in the predicted geometry. We address this by developing a unified data refinement approach that filters and completes real data from different sources using sharp synthetic labels, significantly enhancing the granularity of the reconstructed geometry while maintaining the overall accuracy. We train our model on a large corpus of mixed datasets and conducted comprehensive evaluations, demonstrating its superior performance in achieving accurate relative geometry, precise metric scale, and fine-grained detail recovery -- capabilities that no previous methods have simultaneously achieved.", "date": "2025-07-03", "categories": []}
{"id": "2507.02813", "title": "LangScene-X: Reconstruct Generalizable 3D Language-Embedded Scenes with TriMap Video Diffusion", "abstract": "Recovering 3D structures with open-vocabulary scene understanding from 2D images is a fundamental but daunting task. Recent developments have achieved this by performing per-scene optimization with embedded language information. However, they heavily rely on the calibrated dense-view reconstruction paradigm, thereby suffering from severe rendering artifacts and implausible semantic synthesis when limited views are available. In this paper, we introduce a novel generative framework, coined LangScene-X, to unify and generate 3D consistent multi-modality information for reconstruction and understanding. Powered by the generative capability of creating more consistent novel observations, we can build generalizable 3D language-embedded scenes from only sparse views. Specifically, we first train a TriMap video diffusion model that can generate appearance (RGBs), geometry (normals), and semantics (segmentation maps) from sparse inputs through progressive knowledge integration. Furthermore, we propose a Language Quantized Compressor (LQC), trained on large-scale image datasets, to efficiently encode language embeddings, enabling cross-scene generalization without per-scene retraining. Finally, we reconstruct the language surface fields by aligning language information onto the surface of 3D scenes, enabling open-ended language queries. Extensive experiments on real-world data demonstrate the superiority of our LangScene-X over state-of-the-art methods in terms of quality and generalizability. Project Page: https://liuff19.github.io/LangScene-X.", "date": "2025-07-03", "categories": []}
{"id": "2507.01912", "title": "3D Reconstruction and Information Fusion between Dormant and Canopy Seasons in Commercial Orchards Using Deep Learning and Fast GICP", "abstract": "In orchard automation, dense foliage during the canopy season severely occludes tree structures, minimizing visibility to various canopy parts such as trunks and branches, which limits the ability of a machine vision system. However, canopy structure is more open and visible during the dormant season when trees are defoliated. In this work, we present an information fusion framework that integrates multi-seasonal structural data to support robotic and automated crop load management during the entire growing season. The framework combines high-resolution RGB-D imagery from both dormant and canopy periods using YOLOv9-Seg for instance segmentation, Kinect Fusion for 3D reconstruction, and Fast Generalized Iterative Closest Point (Fast GICP) for model alignment. Segmentation outputs from YOLOv9-Seg were used to extract depth-informed masks, which enabled accurate 3D point cloud reconstruction via Kinect Fusion; these reconstructed models from each season were subsequently aligned using Fast GICP to achieve spatially coherent multi-season fusion. The YOLOv9-Seg model, trained on manually annotated images, achieved a mean squared error (MSE) of 0.0047 and segmentation mAP@50 scores up to 0.78 for trunks in dormant season dataset. Kinect Fusion enabled accurate reconstruction of tree geometry, validated with field measurements resulting in root mean square errors (RMSE) of 5.23 mm for trunk diameter, 4.50 mm for branch diameter, and 13.72 mm for branch spacing. Fast GICP achieved precise cross-seasonal registration with a minimum fitness score of 0.00197, allowing integrated, comprehensive tree structure modeling despite heavy occlusions during the growing season. This fused structural representation enables robotic systems to access otherwise obscured architectural information, improving the precision of pruning, thinning, and other automated orchard operations.", "date": "2025-07-02", "categories": []}
{"id": "2506.23309", "title": "SurgTPGS: Semantic 3D Surgical Scene Understanding with Text Promptable Gaussian Splatting", "abstract": "In contemporary surgical research and practice, accurately comprehending 3D surgical scenes with text-promptable capabilities is particularly crucial for surgical planning and real-time intra-operative guidance, where precisely identifying and interacting with surgical tools and anatomical structures is paramount. However, existing works focus on surgical vision-language model (VLM), 3D reconstruction, and segmentation separately, lacking support for real-time text-promptable 3D queries. In this paper, we present SurgTPGS, a novel text-promptable Gaussian Splatting method to fill this gap. We introduce a 3D semantics feature learning strategy incorporating the Segment Anything model and state-of-the-art vision-language models. We extract the segmented language features for 3D surgical scene reconstruction, enabling a more in-depth understanding of the complex surgical environment. We also propose semantic-aware deformation tracking to capture the seamless deformation of semantic features, providing a more precise reconstruction for both texture and semantic features. Furthermore, we present semantic region-aware optimization, which utilizes regional-based semantic information to supervise the training, particularly promoting the reconstruction quality and semantic smoothness. We conduct comprehensive experiments on two real-world surgical datasets to demonstrate the superiority of SurgTPGS over state-of-the-art methods, highlighting its potential to revolutionize surgical practices. SurgTPGS paves the way for developing next-generation intelligent surgical systems by enhancing surgical precision and safety. Our code is available at: https://github.com/lastbasket/SurgTPGS.", "date": "2025-06-29", "categories": []}
{"id": "2507.00916", "title": "Masks make discriminative models great again!", "abstract": "We present Image2GS, a novel approach that addresses the challenging problem of reconstructing photorealistic 3D scenes from a single image by focusing specifically on the image-to-3D lifting component of the reconstruction process. By decoupling the lifting problem (converting an image to a 3D model representing what is visible) from the completion problem (hallucinating content not present in the input), we create a more deterministic task suitable for discriminative models. Our method employs visibility masks derived from optimized 3D Gaussian splats to exclude areas not visible from the source view during training. This masked training strategy significantly improves reconstruction quality in visible regions compared to strong baselines. Notably, despite being trained only on masked regions, Image2GS remains competitive with state-of-the-art discriminative models trained on full target images when evaluated on complete scenes. Our findings highlight the fundamental struggle discriminative models face when fitting unseen regions and demonstrate the advantages of addressing image-to-3D lifting as a distinct problem with specialized techniques.", "date": "2025-07-01", "categories": []}
{"id": "2411.18066", "title": "GLS: Geometry-aware 3D Language Gaussian Splatting", "abstract": "Recently, 3D Gaussian Splatting (3DGS) has achieved impressive performance on indoor surface reconstruction and 3D open-vocabulary segmentation. This paper presents GLS, a unified framework of 3D surface reconstruction and open-vocabulary segmentation based on 3DGS. GLS extends two fields by improving their sharpness and smoothness. For indoor surface reconstruction, we introduce surface normal prior as a geometric cue to guide the rendered normal, and use the normal error to optimize the rendered depth. For 3D open-vocabulary segmentation, we employ 2D CLIP features to guide instance features and enhance the surface smoothness, then utilize DEVA masks to maintain their view consistency. Extensive experiments demonstrate the effectiveness of jointly optimizing surface reconstruction and 3D open-vocabulary segmentation, where GLS surpasses state-of-the-art approaches of each task on MuSHRoom, ScanNet++ and LERF-OVS datasets. Project webpage: https://jiaxiongq.github.io/GLS_ProjectPage.", "date": "2024-11-27", "categories": []}
{"id": "2506.23157", "title": "STD-GS: Exploring Frame-Event Interaction for SpatioTemporal-Disentangled Gaussian Splatting to Reconstruct High-Dynamic Scene", "abstract": "High-dynamic scene reconstruction aims to represent static background with rigid spatial features and dynamic objects with deformed continuous spatiotemporal features. Typically, existing methods adopt unified representation model (e.g., Gaussian) to directly match the spatiotemporal features of dynamic scene from frame camera. However, this unified paradigm fails in the potential discontinuous temporal features of objects due to frame imaging and the heterogeneous spatial features between background and objects. To address this issue, we disentangle the spatiotemporal features into various latent representations to alleviate the spatiotemporal mismatching between background and objects. In this work, we introduce event camera to compensate for frame camera, and propose a spatiotemporal-disentangled Gaussian splatting framework for high-dynamic scene reconstruction. As for dynamic scene, we figure out that background and objects have appearance discrepancy in frame-based spatial features and motion discrepancy in event-based temporal features, which motivates us to distinguish the spatiotemporal features between background and objects via clustering. As for dynamic object, we discover that Gaussian representations and event data share the consistent spatiotemporal characteristic, which could serve as a prior to guide the spatiotemporal disentanglement of object Gaussians. Within Gaussian splatting framework, the cumulative scene-object disentanglement can improve the spatiotemporal discrimination between background and objects to render the time-continuous dynamic scene. Extensive experiments have been performed to verify the superiority of the proposed method.", "date": "2025-06-29", "categories": []}
{"id": "2506.23308", "title": "Endo-4DGX: Robust Endoscopic Scene Reconstruction and Illumination Correction with Gaussian Splatting", "abstract": "Accurate reconstruction of soft tissue is crucial for advancing automation in image-guided robotic surgery. The recent 3D Gaussian Splatting (3DGS) techniques and their variants, 4DGS, achieve high-quality renderings of dynamic surgical scenes in real-time. However, 3D-GS-based methods still struggle in scenarios with varying illumination, such as low light and over-exposure. Training 3D-GS in such extreme light conditions leads to severe optimization problems and devastating rendering quality. To address these challenges, we present Endo-4DGX, a novel reconstruction method with illumination-adaptive Gaussian Splatting designed specifically for endoscopic scenes with uneven lighting. By incorporating illumination embeddings, our method effectively models view-dependent brightness variations. We introduce a region-aware enhancement module to model the sub-area lightness at the Gaussian level and a spatial-aware adjustment module to learn the view-consistent brightness adjustment. With the illumination adaptive design, Endo-4DGX achieves superior rendering performance under both low-light and over-exposure conditions while maintaining geometric accuracy. Additionally, we employ an exposure control loss to restore the appearance from adverse exposure to the normal level for illumination-adaptive optimization. Experimental results demonstrate that Endo-4DGX significantly outperforms combinations of state-of-the-art reconstruction and restoration methods in challenging lighting environments, underscoring its potential to advance robot-assisted surgical applications. Our code is available at https://github.com/lastbasket/Endo-4DGX.", "date": "2025-06-29", "categories": []}
{"id": "2506.23863", "title": "Puzzles: Unbounded Video-Depth Augmentation for Scalable End-to-End 3D Reconstruction", "abstract": "Multi-view 3D reconstruction remains a core challenge in computer vision. Recent methods, such as DUST3R and its successors, directly regress pointmaps from image pairs without relying on known scene geometry or camera parameters. However, the performance of these models is constrained by the diversity and scale of available training data. In this work, we introduce Puzzles, a data augmentation strategy that synthesizes an unbounded volume of high-quality posed video-depth data from a single image or video clip. By simulating diverse camera trajectories and realistic scene geometry through targeted image transformations, Puzzles significantly enhances data variety. Extensive experiments show that integrating Puzzles into existing video-based 3D reconstruction pipelines consistently boosts performance without modifying the underlying network architecture. Notably, models trained on only ten percent of the original data augmented with Puzzles still achieve accuracy comparable to those trained on the full dataset. Code is available at https://jiahao-ma.github.io/puzzles/.", "date": "2025-06-30", "categories": []}
{"id": "2506.21680", "title": "PhotonSplat: 3D Scene Reconstruction and Colorization from SPAD Sensors", "abstract": "Advances in 3D reconstruction using neural rendering have enabled high-quality 3D capture. However, they often fail when the input imagery is corrupted by motion blur, due to fast motion of the camera or the objects in the scene. This work advances neural rendering techniques in such scenarios by using single-photon avalanche diode (SPAD) arrays, an emerging sensing technology capable of sensing images at extremely high speeds. However, the use of SPADs presents its own set of unique challenges in the form of binary images, that are driven by stochastic photon arrivals. To address this, we introduce PhotonSplat, a framework designed to reconstruct 3D scenes directly from SPAD binary images, effectively navigating the noise vs. blur trade-off. Our approach incorporates a novel 3D spatial filtering technique to reduce noise in the renderings. The framework also supports both no-reference using generative priors and reference-based colorization from a single blurry image, enabling downstream applications such as segmentation, object detection and appearance editing tasks. Additionally, we extend our method to incorporate dynamic scene representations, making it suitable for scenes with moving objects. We further contribute PhotonScenes, a real-world multi-view dataset captured with the SPAD sensors.", "date": "2025-06-26", "categories": []}
{"id": "2506.20979", "title": "3D Scene-Camera Representation with Joint Camera Photometric Optimization", "abstract": "Representing scenes from multi-view images is a crucial task in computer vision with extensive applications. However, inherent photometric distortions in the camera imaging can significantly degrade image quality. Without accounting for these distortions, the 3D scene representation may inadvertently incorporate erroneous information unrelated to the scene, diminishing the quality of the representation. In this paper, we propose a novel 3D scene-camera representation with joint camera photometric optimization. By introducing internal and external photometric model, we propose a full photometric model and corresponding camera representation. Based on simultaneously optimizing the parameters of the camera representation, the proposed method effectively separates scene-unrelated information from the 3D scene representation. Additionally, during the optimization of the photometric parameters, we introduce a depth regularization to prevent the 3D scene representation from fitting scene-unrelated information. By incorporating the camera model as part of the mapping process, the proposed method constructs a complete map that includes both the scene radiance field and the camera photometric model. Experimental results demonstrate that the proposed method can achieve high-quality 3D scene representations, even under conditions of imaging degradation, such as vignetting and dirt.", "date": "2025-06-26", "categories": []}
{"id": "2405.03633", "title": "Neural Graph Map: Dense Mapping with Efficient Loop Closure Integration", "abstract": "Neural field-based SLAM methods typically employ a single, monolithic field as their scene representation. This prevents efficient incorporation of loop closure constraints and limits scalability. To address these shortcomings, we propose a novel RGB-D neural mapping framework in which the scene is represented by a collection of lightweight neural fields which are dynamically anchored to the pose graph of a sparse visual SLAM system. Our approach shows the ability to integrate large-scale loop closures, while requiring only minimal reintegration. Furthermore, we verify the scalability of our approach by demonstrating successful building-scale mapping taking multiple loop closures into account during the optimization, and show that our method outperforms existing state-of-the-art approaches on large scenes in terms of quality and runtime. Our code is available open-source at https://github.com/KTH-RPL/neural_graph_mapping.", "date": "2024-05-06", "categories": []}
{"id": "2408.16767", "title": "ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model", "abstract": "Advancements in 3D scene reconstruction have transformed 2D images from the real world into 3D models, producing realistic 3D results from hundreds of input photos. Despite great success in dense-view reconstruction scenarios, rendering a detailed scene from insufficient captured views is still an ill-posed optimization problem, often resulting in artifacts and distortions in unseen areas. In this paper, we propose ReconX, a novel 3D scene reconstruction paradigm that reframes the ambiguous reconstruction challenge as a temporal generation task. The key insight is to unleash the strong generative prior of large pre-trained video diffusion models for sparse-view reconstruction. However, 3D view consistency struggles to be accurately preserved in directly generated video frames from pre-trained models. To address this, given limited input views, the proposed ReconX first constructs a global point cloud and encodes it into a contextual space as the 3D structure condition. Guided by the condition, the video diffusion model then synthesizes video frames that are both detail-preserved and exhibit a high degree of 3D consistency, ensuring the coherence of the scene from various perspectives. Finally, we recover the 3D scene from the generated video through a confidence-aware 3D Gaussian Splatting optimization scheme. Extensive experiments on various real-world datasets show the superiority of our ReconX over state-of-the-art methods in terms of quality and generalizability.", "date": "2024-08-29", "categories": []}
{"id": "2411.10504", "title": "USP-Gaussian: Unifying Spike-based Image Reconstruction, Pose Correction and Gaussian Splatting", "abstract": "Spike cameras, as an innovative neuromorphic camera that captures scenes with the 0-1 bit stream at 40 kHz, are increasingly employed for the 3D reconstruction task via Neural Radiance Fields (NeRF) or 3D Gaussian Splatting (3DGS). Previous spike-based 3D reconstruction approaches often employ a casecased pipeline: starting with high-quality image reconstruction from spike streams based on established spike-to-image reconstruction algorithms, then progressing to camera pose estimation and 3D reconstruction. However, this cascaded approach suffers from substantial cumulative errors, where quality limitations of initial image reconstructions negatively impact pose estimation, ultimately degrading the fidelity of the 3D reconstruction. To address these issues, we propose a synergistic optimization framework, \\textbf{USP-Gaussian}, that unifies spike-based image reconstruction, pose correction, and Gaussian splatting into an end-to-end framework. Leveraging the multi-view consistency afforded by 3DGS and the motion capture capability of the spike camera, our framework enables a joint iterative optimization that seamlessly integrates information between the spike-to-image network and 3DGS. Experiments on synthetic datasets with accurate poses demonstrate that our method surpasses previous approaches by effectively eliminating cascading errors. Moreover, we integrate pose optimization to achieve robust 3D reconstruction in real-world scenarios with inaccurate initial poses, outperforming alternative methods by effectively reducing noise and preserving fine texture details. Our code, data and trained models will be available at https://github.com/chenkang455/USP-Gaussian.", "date": "2024-11-15", "categories": []}
{"id": "2412.01402", "title": "ULSR-GS: Ultra Large-scale Surface Reconstruction Gaussian Splatting with Multi-View Geometric Consistency", "abstract": "While Gaussian Splatting (GS) demonstrates efficient and high-quality scene rendering and small area surface extraction ability, it falls short in handling large-scale aerial image surface extraction tasks. To overcome this, we present ULSR-GS, a framework dedicated to high-fidelity surface extraction in ultra-large-scale scenes, addressing the limitations of existing GS-based mesh extraction methods. Specifically, we propose a point-to-photo partitioning approach combined with a multi-view optimal view matching principle to select the best training images for each sub-region. Additionally, during training, ULSR-GS employs a densification strategy based on multi-view geometric consistency to enhance surface extraction details. Experimental results demonstrate that ULSR-GS outperforms other state-of-the-art GS-based works on large-scale aerial photogrammetry benchmark datasets, significantly improving surface extraction accuracy in complex urban environments. Project page: https://ulsrgs.github.io.", "date": "2024-12-02", "categories": []}
{"id": "2501.07113", "title": "Matching-Free Depth Recovery from Structured Light", "abstract": "We introduce a novel approach for depth estimation using images obtained from monocular structured light systems. In contrast to many existing methods that depend on image matching, our technique employs a density voxel grid to represent scene geometry. This grid is trained through self-supervised differentiable volume rendering. Our method leverages color fields derived from the projected patterns in structured light systems during the rendering process, facilitating the isolated optimization of the geometry field. This innovative approach leads to faster convergence and high-quality results. Additionally, we integrate normalized device coordinates (NDC), a distortion loss, and a distinctive surface-based color loss to enhance geometric fidelity. Experimental results demonstrate that our method outperforms current matching-based techniques in terms of geometric performance in few-shot scenarios, achieving an approximately 30% reduction in average estimated depth errors for both synthetic scenes and real-world captured scenes. Moreover, our approach allows for rapid training, being approximately three times faster than previous matching-free methods that utilize implicit representations.", "date": "2025-01-13", "categories": []}
{"id": "2506.20638", "title": "Joint attitude estimation and 3D neural reconstruction of non-cooperative space objects", "abstract": "Obtaining a better knowledge of the current state and behavior of objects orbiting Earth has proven to be essential for a range of applications such as active debris removal, in-orbit maintenance, or anomaly detection. 3D models represent a valuable source of information in the field of Space Situational Awareness (SSA). In this work, we leveraged Neural Radiance Fields (NeRF) to perform 3D reconstruction of non-cooperative space objects from simulated images. This scenario is challenging for NeRF models due to unusual camera characteristics and environmental conditions : mono-chromatic images, unknown object orientation, limited viewing angles, absence of diffuse lighting etc. In this work we focus primarly on the joint optimization of camera poses alongside the NeRF. Our experimental results show that the most accurate 3D reconstruction is achieved when training with successive images one-by-one. We estimate camera poses by optimizing an uniform rotation and use regularization to prevent successive poses from being too far apart.", "date": "2025-06-25", "categories": []}
{"id": "2506.15242", "title": "RA-NeRF: Robust Neural Radiance Field Reconstruction with Accurate Camera Pose Estimation under Complex Trajectories", "abstract": "Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have emerged as powerful tools for 3D reconstruction and SLAM tasks. However, their performance depends heavily on accurate camera pose priors. Existing approaches attempt to address this issue by introducing external constraints but fall short of achieving satisfactory accuracy, particularly when camera trajectories are complex. In this paper, we propose a novel method, RA-NeRF, capable of predicting highly accurate camera poses even with complex camera trajectories. Following the incremental pipeline, RA-NeRF reconstructs the scene using NeRF with photometric consistency and incorporates flow-driven pose regulation to enhance robustness during initialization and localization. Additionally, RA-NeRF employs an implicit pose filter to capture the camera movement pattern and eliminate the noise for pose estimation. To validate our method, we conduct extensive experiments on the Tanks\\&Temple dataset for standard evaluation, as well as the NeRFBuster dataset, which presents challenging camera pose trajectories. On both datasets, RA-NeRF achieves state-of-the-art results in both camera pose estimation and visual quality, demonstrating its effectiveness and robustness in scene reconstruction under complex pose trajectories.", "date": "2025-06-18", "categories": []}
{"id": "2506.19291", "title": "HoliGS: Holistic Gaussian Splatting for Embodied View Synthesis", "abstract": "We propose HoliGS, a novel deformable Gaussian splatting framework that addresses embodied view synthesis from long monocular RGB videos. Unlike prior 4D Gaussian splatting and dynamic NeRF pipelines, which struggle with training overhead in minute-long captures, our method leverages invertible Gaussian Splatting deformation networks to reconstruct large-scale, dynamic environments accurately. Specifically, we decompose each scene into a static background plus time-varying objects, each represented by learned Gaussian primitives undergoing global rigid transformations, skeleton-driven articulation, and subtle non-rigid deformations via an invertible neural flow. This hierarchical warping strategy enables robust free-viewpoint novel-view rendering from various embodied camera trajectories by attaching Gaussians to a complete canonical foreground shape (\\eg, egocentric or third-person follow), which may involve substantial viewpoint changes and interactions between multiple actors. Our experiments demonstrate that \\ourmethod~ achieves superior reconstruction quality on challenging datasets while significantly reducing both training and rendering time compared to state-of-the-art monocular deformable NeRFs. These results highlight a practical and scalable solution for EVS in real-world scenarios. The source code will be released.", "date": "2025-06-24", "categories": []}
{"id": "2506.19491", "title": "Experimental Assessment of Neural 3D Reconstruction for Small UAV-based Applications", "abstract": "The increasing miniaturization of Unmanned Aerial Vehicles (UAVs) has expanded their deployment potential to indoor and hard-to-reach areas. However, this trend introduces distinct challenges, particularly in terms of flight dynamics and power consumption, which limit the UAVs' autonomy and mission capabilities. This paper presents a novel approach to overcoming these limitations by integrating Neural 3D Reconstruction (N3DR) with small UAV systems for fine-grained 3-Dimensional (3D) digital reconstruction of small static objects. Specifically, we design, implement, and evaluate an N3DR-based pipeline that leverages advanced models, i.e., Instant-ngp, Nerfacto, and Splatfacto, to improve the quality of 3D reconstructions using images of the object captured by a fleet of small UAVs. We assess the performance of the considered models using various imagery and pointcloud metrics, comparing them against the baseline Structure from Motion (SfM) algorithm. The experimental results demonstrate that the N3DR-enhanced pipeline significantly improves reconstruction quality, making it feasible for small UAVs to support high-precision 3D mapping and anomaly detection in constrained environments. In more general terms, our results highlight the potential of N3DR in advancing the capabilities of miniaturized UAV systems.", "date": "2025-06-24", "categories": []}
{"id": "2505.23481", "title": "PhysicsNeRF: Physics-Guided 3D Reconstruction from Sparse Views", "abstract": "PhysicsNeRF is a physically grounded framework for 3D reconstruction from sparse views, extending Neural Radiance Fields with four complementary constraints: depth ranking, RegNeRF-style consistency, sparsity priors, and cross-view alignment. While standard NeRFs fail under sparse supervision, PhysicsNeRF employs a compact 0.67M-parameter architecture and achieves 21.4 dB average PSNR using only 8 views, outperforming prior methods. A generalization gap of 5.7-6.2 dB is consistently observed and analyzed, revealing fundamental limitations of sparse-view reconstruction. PhysicsNeRF enables physically consistent, generalizable 3D representations for agent interaction and simulation, and clarifies the expressiveness-generalization trade-off in constrained NeRF models.", "date": "2025-05-29", "categories": []}
{"id": "2506.17636", "title": "3D Gaussian Splatting for Fine-Detailed Surface Reconstruction in Large-Scale Scene", "abstract": "Recent developments in 3D Gaussian Splatting have made significant advances in surface reconstruction. However, scaling these methods to large-scale scenes remains challenging due to high computational demands and the complex dynamic appearances typical of outdoor environments. These challenges hinder the application in aerial surveying and autonomous driving. This paper proposes a novel solution to reconstruct large-scale surfaces with fine details, supervised by full-sized images. Firstly, we introduce a coarse-to-fine strategy to reconstruct a coarse model efficiently, followed by adaptive scene partitioning and sub-scene refining from image segments. Additionally, we integrate a decoupling appearance model to capture global appearance variations and a transient mask model to mitigate interference from moving objects. Finally, we expand the multi-view constraint and introduce a single-view regularization for texture-less areas. Our experiments were conducted on the publicly available dataset GauU-Scene V2, which was captured using unmanned aerial vehicles. To the best of our knowledge, our method outperforms existing NeRF-based and Gaussian-based methods, achieving high-fidelity visual results and accurate surface from full-size image optimization. Open-source code will be available on GitHub.", "date": "2025-06-21", "categories": []}
{"id": "2506.18208", "title": "Limitations of NERF with pre-trained Vision Features for Few-Shot 3D Reconstruction", "abstract": "Neural Radiance Fields (NeRF) have revolutionized 3D scene reconstruction from sparse image collections. Recent work has explored integrating pre-trained vision features, particularly from DINO, to enhance few-shot reconstruction capabilities. However, the effectiveness of such approaches remains unclear, especially in extreme few-shot scenarios. In this paper, we present a systematic evaluation of DINO-enhanced NeRF models, comparing baseline NeRF, frozen DINO features, LoRA fine-tuned features, and multi-scale feature fusion. Surprisingly, our experiments reveal that all DINO variants perform worse than the baseline NeRF, achieving PSNR values around 12.9 to 13.0 compared to the baseline's 14.71. This counterintuitive result suggests that pre-trained vision features may not be beneficial for few-shot 3D reconstruction and may even introduce harmful biases. We analyze potential causes including feature-task mismatch, overfitting to limited data, and integration challenges. Our findings challenge common assumptions in the field and suggest that simpler architectures focusing on geometric consistency may be more effective for few-shot scenarios.", "date": "2025-06-22", "categories": []}
{"id": "2506.18890", "title": "4D-LRM: Large Space-Time Reconstruction Model From and To Any View at Any Time", "abstract": "Can we scale 4D pretraining to learn general space-time representations that reconstruct an object from a few views at some times to any view at any time? We provide an affirmative answer with 4D-LRM, the first large-scale 4D reconstruction model that takes input from unconstrained views and timestamps and renders arbitrary novel view-time combinations. Unlike prior 4D approaches, e.g., optimization-based, geometry-based, or generative, that struggle with efficiency, generalization, or faithfulness, 4D-LRM learns a unified space-time representation and directly predicts per-pixel 4D Gaussian primitives from posed image tokens across time, enabling fast, high-quality rendering at, in principle, infinite frame rate. Our results demonstrate that scaling spatiotemporal pretraining enables accurate and efficient 4D reconstruction. We show that 4D-LRM generalizes to novel objects, interpolates across time, and handles diverse camera setups. It reconstructs 24-frame sequences in one forward pass with less than 1.5 seconds on a single A100 GPU.", "date": "2025-06-23", "categories": []}
{"id": "2506.15806", "title": "Implicit 3D scene reconstruction using deep learning towards efficient collision understanding in autonomous driving", "abstract": "In crowded urban environments where traffic is dense, current technologies struggle to oversee tight navigation, but surface-level understanding allows autonomous vehicles to safely assess proximity to surrounding obstacles. 3D or 2D scene mapping of the surrounding objects is an essential task in addressing the above problem. Despite its importance in dense vehicle traffic conditions, 3D scene reconstruction of object shapes with higher boundary level accuracy is not yet entirely considered in current literature. The sign distance function represents any shape through parameters that calculate the distance from any point in space to the closest obstacle surface, making it more efficient in terms of storage. In recent studies, researchers have started to formulate problems with Implicit 3D reconstruction methods in the autonomous driving domain, highlighting the possibility of using sign distance function to map obstacles effectively. This research addresses this gap by developing a learning-based 3D scene reconstruction methodology that leverages LiDAR data and a deep neural network to build a the static Signed Distance Function (SDF) maps. Unlike traditional polygonal representations, this approach has the potential to map 3D obstacle shapes with more boundary-level details. Our preliminary results demonstrate that this method would significantly enhance collision detection performance, particularly in congested and dynamic environments.", "date": "2025-06-18", "categories": []}
{"id": "2506.16421", "title": "Structured Semantic 3D Reconstruction (S23DR) Challenge 2025 -- Winning solution", "abstract": "This paper presents the winning solution for the S23DR Challenge 2025, which involves predicting a house's 3D roof wireframe from a sparse point cloud and semantic segmentations. Our method operates directly in 3D, first identifying vertex candidates from the COLMAP point cloud using Gestalt segmentations. We then employ two PointNet-like models: one to refine and classify these candidates by analyzing local cubic patches, and a second to predict edges by processing the cylindrical regions connecting vertex pairs. This two-stage, 3D deep learning approach achieved a winning Hybrid Structure Score (HSS) of 0.43 on the private leaderboard.", "date": "2025-06-19", "categories": []}
{"id": "2309.10815", "title": "PanopticNeRF-360: Panoramic 3D-to-2D Label Transfer in Urban Scenes", "abstract": "Training perception systems for self-driving cars requires substantial 2D annotations that are labor-intensive to manual label. While existing datasets provide rich annotations on pre-recorded sequences, they fall short in labeling rarely encountered viewpoints, potentially hampering the generalization ability for perception models. In this paper, we present PanopticNeRF-360, a novel approach that combines coarse 3D annotations with noisy 2D semantic cues to generate high-quality panoptic labels and images from any viewpoint. Our key insight lies in exploiting the complementarity of 3D and 2D priors to mutually enhance geometry and semantics. Specifically, we propose to leverage coarse 3D bounding primitives and noisy 2D semantic and instance predictions to guide geometry optimization, by encouraging predicted labels to match panoptic pseudo ground truth. Simultaneously, the improved geometry assists in filtering 3D&2D annotation noise by fusing semantics in 3D space via a learned semantic field. To further enhance appearance, we combine MLP and hash grids to yield hybrid scene features, striking a balance between high-frequency appearance and contiguous semantics. Our experiments demonstrate PanopticNeRF-360's state-of-the-art performance over label transfer methods on the challenging urban scenes of the KITTI-360 dataset. Moreover, PanopticNeRF-360 enables omnidirectional rendering of high-fidelity, multi-view and spatiotemporally consistent appearance, semantic and instance labels. We make our code and data available at https://github.com/fuxiao0719/PanopticNeRF", "date": "2023-09-19", "categories": []}
{"id": "2506.14771", "title": "Empirical Studies of Large Scale Environment Scanning by Consumer Electronics", "abstract": "This paper presents an empirical evaluation of the Matterport Pro3, a consumer-grade 3D scanning device, for large-scale environment reconstruction. We conduct detailed scanning (1,099 scanning points) of a six-floor building (17,567 square meters) and assess the device's effectiveness, limitations, and performance enhancements in diverse scenarios. Challenges encountered during the scanning are addressed through proposed solutions, while we also explore advanced methods to overcome them more effectively. Comparative analysis with another consumer-grade device (iPhone) highlights the Pro3's balance between cost-effectiveness and performance. The Matterport Pro3 achieves a denser point cloud with 1,877,324 points compared to the iPhone's 506,961 points and higher alignment accuracy with an RMSE of 0.0118 meters. The cloud-to-cloud (C2C) average distance error between the two point cloud models is 0.0408 meters, with a standard deviation of 0.0715 meters. The study demonstrates the Pro3's ability to generate high-quality 3D models suitable for large-scale applications, leveraging features such as LiDAR and advanced alignment techniques.", "date": "2025-03-27", "categories": []}
{"id": "2503.12553", "title": "Niagara: Normal-Integrated Geometric Affine Field for Scene Reconstruction from a Single View", "abstract": "Recent advances in single-view 3D scene reconstruction have highlighted the challenges in capturing fine geometric details and ensuring structural consistency, particularly in high-fidelity outdoor scene modeling. This paper presents Niagara, a new single-view 3D scene reconstruction framework that can faithfully reconstruct challenging outdoor scenes from a single input image for the first time.\n  Our approach integrates monocular depth and normal estimation as input, which substantially improves its ability to capture fine details, mitigating common issues like geometric detail loss and deformation.\n  Additionally, we introduce a geometric affine field (GAF) and 3D self-attention as geometry-constraint, which combines the structural properties of explicit geometry with the adaptability of implicit feature fields, striking a balance between efficient rendering and high-fidelity reconstruction.\n  Our framework finally proposes a specialized encoder-decoder architecture, where a depth-based 3D Gaussian decoder is proposed to predict 3D Gaussian parameters, which can be used for novel view synthesis. Extensive results and analyses suggest that our Niagara surpasses prior SoTA approaches such as Flash3D in both single-view and dual-view settings, significantly enhancing the geometric accuracy and visual fidelity, especially in outdoor scenes.", "date": "2025-03-16", "categories": []}
{"id": "2506.12727", "title": "Efficient multi-view training for 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting (3DGS) has emerged as a preferred choice alongside Neural Radiance Fields (NeRF) in inverse rendering due to its superior rendering speed. Currently, the common approach in 3DGS is to utilize \"single-view\" mini-batch training, where only one image is processed per iteration, in contrast to NeRF's \"multi-view\" mini-batch training, which leverages multiple images. We observe that such single-view training can lead to suboptimal optimization due to increased variance in mini-batch stochastic gradients, highlighting the necessity for multi-view training. However, implementing multi-view training in 3DGS poses challenges. Simply rendering multiple images per iteration incurs considerable overhead and may result in suboptimal Gaussian densification due to its reliance on single-view assumptions. To address these issues, we modify the rasterization process to minimize the overhead associated with multi-view training and propose a 3D distance-aware D-SSIM loss and multi-view adaptive density control that better suits multi-view scenarios. Our experiments demonstrate that the proposed methods significantly enhance the performance of 3DGS and its variants, freeing 3DGS from the constraints of single-view training.", "date": "2025-06-15", "categories": []}
{"id": "2410.21955", "title": "ActiveSplat: High-Fidelity Scene Reconstruction through Active Gaussian Splatting", "abstract": "We propose ActiveSplat, an autonomous high-fidelity reconstruction system leveraging Gaussian splatting. Taking advantage of efficient and realistic rendering, the system establishes a unified framework for online mapping, viewpoint selection, and path planning. The key to ActiveSplat is a hybrid map representation that integrates both dense information about the environment and a sparse abstraction of the workspace. Therefore, the system leverages sparse topology for efficient viewpoint sampling and path planning, while exploiting view-dependent dense prediction for viewpoint selection, facilitating efficient decision-making with promising accuracy and completeness. A hierarchical planning strategy based on the topological map is adopted to mitigate repetitive trajectories and improve local granularity given limited time budgets, ensuring high-fidelity reconstruction with photorealistic view synthesis. Extensive experiments and ablation studies validate the efficacy of the proposed method in terms of reconstruction accuracy, data coverage, and exploration efficiency. The released code will be available on our project page: https://li-yuetao.github.io/ActiveSplat/.", "date": "2024-10-29", "categories": []}
{"id": "2412.20522", "title": "MaskGaussian: Adaptive 3D Gaussian Representation from Probabilistic Masks", "abstract": "While 3D Gaussian Splatting (3DGS) has demonstrated remarkable performance in novel view synthesis and real-time rendering, the high memory consumption due to the use of millions of Gaussians limits its practicality. To mitigate this issue, improvements have been made by pruning unnecessary Gaussians, either through a hand-crafted criterion or by using learned masks. However, these methods deterministically remove Gaussians based on a snapshot of the pruning moment, leading to sub-optimized reconstruction performance from a long-term perspective. To address this issue, we introduce MaskGaussian, which models Gaussians as probabilistic entities rather than permanently removing them, and utilize them according to their probability of existence. To achieve this, we propose a masked-rasterization technique that enables unused yet probabilistically existing Gaussians to receive gradients, allowing for dynamic assessment of their contribution to the evolving scene and adjustment of their probability of existence. Hence, the importance of Gaussians iteratively changes and the pruned Gaussians are selected diversely. Extensive experiments demonstrate the superiority of the proposed method in achieving better rendering quality with fewer Gaussians than previous pruning methods, pruning over 60% of Gaussians on average with only a 0.02 PSNR decline. Our code can be found at: https://github.com/kaikai23/MaskGaussian", "date": "2024-12-29", "categories": []}
{"id": "2506.12716", "title": "Generative 4D Scene Gaussian Splatting with Object View-Synthesis Priors", "abstract": "We tackle the challenge of generating dynamic 4D scenes from monocular, multi-object videos with heavy occlusions, and introduce GenMOJO, a novel approach that integrates rendering-based deformable 3D Gaussian optimization with generative priors for view synthesis. While existing models perform well on novel view synthesis for isolated objects, they struggle to generalize to complex, cluttered scenes. To address this, GenMOJO decomposes the scene into individual objects, optimizing a differentiable set of deformable Gaussians per object. This object-wise decomposition allows leveraging object-centric diffusion models to infer unobserved regions in novel viewpoints. It performs joint Gaussian splatting to render the full scene, capturing cross-object occlusions, and enabling occlusion-aware supervision. To bridge the gap between object-centric priors and the global frame-centric coordinate system of videos, GenMOJO uses differentiable transformations that align generative and rendering constraints within a unified framework. The resulting model generates 4D object reconstructions over space and time, and produces accurate 2D and 3D point tracks from monocular input. Quantitative evaluations and perceptual human studies confirm that GenMOJO generates more realistic novel views of scenes and produces more accurate point tracks compared to existing approaches.", "date": "2025-06-15", "categories": []}
{"id": "2506.13508", "title": "Multiview Geometric Regularization of Gaussian Splatting for Accurate Radiance Fields", "abstract": "Recent methods, such as 2D Gaussian Splatting and Gaussian Opacity Fields, have aimed to address the geometric inaccuracies of 3D Gaussian Splatting while retaining its superior rendering quality. However, these approaches still struggle to reconstruct smooth and reliable geometry, particularly in scenes with significant color variation across viewpoints, due to their per-point appearance modeling and single-view optimization constraints. In this paper, we propose an effective multiview geometric regularization strategy that integrates multiview stereo (MVS) depth, RGB, and normal constraints into Gaussian Splatting initialization and optimization. Our key insight is the complementary relationship between MVS-derived depth points and Gaussian Splatting-optimized positions: MVS robustly estimates geometry in regions of high color variation through local patch-based matching and epipolar constraints, whereas Gaussian Splatting provides more reliable and less noisy depth estimates near object boundaries and regions with lower color variation. To leverage this insight, we introduce a median depth-based multiview relative depth loss with uncertainty estimation, effectively integrating MVS depth information into Gaussian Splatting optimization. We also propose an MVS-guided Gaussian Splatting initialization to avoid Gaussians falling into suboptimal positions. Extensive experiments validate that our approach successfully combines these strengths, enhancing both geometric accuracy and rendering quality across diverse indoor and outdoor scenes.", "date": "2025-06-16", "categories": []}
{"id": "2506.13516", "title": "Micro-macro Gaussian Splatting with Enhanced Scalability for Unconstrained Scene Reconstruction", "abstract": "Reconstructing 3D scenes from unconstrained image collections poses significant challenges due to variations in appearance. In this paper, we propose Scalable Micro-macro Wavelet-based Gaussian Splatting (SMW-GS), a novel method that enhances 3D reconstruction across diverse scales by decomposing scene representations into global, refined, and intrinsic components. SMW-GS incorporates the following innovations: Micro-macro Projection, which enables Gaussian points to sample multi-scale details with improved diversity; and Wavelet-based Sampling, which refines feature representations using frequency-domain information to better capture complex scene appearances. To achieve scalability, we further propose a large-scale scene promotion strategy, which optimally assigns camera views to scene partitions by maximizing their contributions to Gaussian points, achieving consistent and high-quality reconstructions even in expansive environments. Extensive experiments demonstrate that SMW-GS significantly outperforms existing methods in both reconstruction quality and scalability, particularly excelling in large-scale urban environments with challenging illumination variations. Project is available at https://github.com/Kidleyh/SMW-GS.", "date": "2025-06-16", "categories": []}
{"id": "2506.13750", "title": "Test3R: Learning to Reconstruct 3D at Test Time", "abstract": "Dense matching methods like DUSt3R regress pairwise pointmaps for 3D reconstruction. However, the reliance on pairwise prediction and the limited generalization capability inherently restrict the global geometric consistency. In this work, we introduce Test3R, a surprisingly simple test-time learning technique that significantly boosts geometric accuracy. Using image triplets ($I_1,I_2,I_3$), Test3R generates reconstructions from pairs ($I_1,I_2$) and ($I_1,I_3$). The core idea is to optimize the network at test time via a self-supervised objective: maximizing the geometric consistency between these two reconstructions relative to the common image $I_1$. This ensures the model produces cross-pair consistent outputs, regardless of the inputs. Extensive experiments demonstrate that our technique significantly outperforms previous state-of-the-art methods on the 3D reconstruction and multi-view depth estimation tasks. Moreover, it is universally applicable and nearly cost-free, making it easily applied to other models and implemented with minimal test-time training overhead and parameter footprint. Code is available at https://github.com/nopQAQ/Test3R.", "date": "2025-06-16", "categories": []}
{"id": "2506.09565", "title": "SemanticSplat: Feed-Forward 3D Scene Understanding with Language-Aware Gaussian Fields", "abstract": "Holistic 3D scene understanding, which jointly models geometry, appearance, and semantics, is crucial for applications like augmented reality and robotic interaction. Existing feed-forward 3D scene understanding methods (e.g., LSM) are limited to extracting language-based semantics from scenes, failing to achieve holistic scene comprehension. Additionally, they suffer from low-quality geometry reconstruction and noisy artifacts. In contrast, per-scene optimization methods rely on dense input views, which reduces practicality and increases complexity during deployment. In this paper, we propose SemanticSplat, a feed-forward semantic-aware 3D reconstruction method, which unifies 3D Gaussians with latent semantic attributes for joint geometry-appearance-semantics modeling. To predict the semantic anisotropic Gaussians, SemanticSplat fuses diverse feature fields (e.g., LSeg, SAM) with a cost volume representation that stores cross-view feature similarities, enhancing coherent and accurate scene comprehension. Leveraging a two-stage distillation framework, SemanticSplat reconstructs a holistic multi-modal semantic feature field from sparse-view images. Experiments demonstrate the effectiveness of our method for 3D scene understanding tasks like promptable and open-vocabulary segmentation. Video results are available at https://semanticsplat.github.io.", "date": "2025-06-11", "categories": []}
{"id": "2506.10980", "title": "InstaInpaint: Instant 3D-Scene Inpainting with Masked Large Reconstruction Model", "abstract": "Recent advances in 3D scene reconstruction enable real-time viewing in virtual and augmented reality. To support interactive operations for better immersiveness, such as moving or editing objects, 3D scene inpainting methods are proposed to repair or complete the altered geometry. However, current approaches rely on lengthy and computationally intensive optimization, making them impractical for real-time or online applications. We propose InstaInpaint, a reference-based feed-forward framework that produces 3D-scene inpainting from a 2D inpainting proposal within 0.4 seconds. We develop a self-supervised masked-finetuning strategy to enable training of our custom large reconstruction model (LRM) on the large-scale dataset. Through extensive experiments, we analyze and identify several key designs that improve generalization, textural consistency, and geometric correctness. InstaInpaint achieves a 1000x speed-up from prior methods while maintaining a state-of-the-art performance across two standard benchmarks. Moreover, we show that InstaInpaint generalizes well to flexible downstream applications such as object insertion and multi-region inpainting. More video results are available at our project page: https://dhmbb2.github.io/InstaInpaint_page/.", "date": "2025-06-12", "categories": []}
{"id": "2311.09652", "title": "Event-based Motion-Robust Accurate Shape Estimation for Mixed Reflectance Scenes", "abstract": "Event-based structured light systems have recently been introduced as an exciting alternative to conventional frame-based triangulation systems for the 3D measurements of diffuse surfaces. Important benefits include the fast capture speed and the high dynamic range provided by the event camera - albeit at the cost of lower data quality. So far, both low-accuracy event-based and high-accuracy frame-based 3D imaging systems are tailored to a specific surface type, such as diffuse or specular, and can not be used for a broader class of object surfaces (\"mixed reflectance scenes\"). In this work, we present a novel event-based structured light system that enables fast 3D imaging of mixed reflectance scenes with high accuracy. On the captured events, we use epipolar constraints that intrinsically enable decomposing the measured reflections into diffuse, two-bounce specular, and other multi-bounce reflections. The diffuse surfaces in the scene are reconstructed using triangulation. Then, the reconstructed diffuse scene parts are leveraged as a \"display\" to evaluate the specular scene parts via deflectometry. This novel procedure allows us to use the entire scene as a virtual screen, using only a scanning laser and an event camera. The resulting system achieves fast and motion-robust (14Hz) reconstructions of mixed reflectance scenes with < 600 ${\\mu}m$ depth error. Moreover, we introduce an \"ultrafast\" capture mode (250Hz) for the 3D measurement of diffuse scenes.", "date": "2023-11-16", "categories": []}
{"id": "2506.09378", "title": "UniForward: Unified 3D Scene and Semantic Field Reconstruction via Feed-Forward Gaussian Splatting from Only Sparse-View Images", "abstract": "We propose a feed-forward Gaussian Splatting model that unifies 3D scene and semantic field reconstruction. Combining 3D scenes with semantic fields facilitates the perception and understanding of the surrounding environment. However, key challenges include embedding semantics into 3D representations, achieving generalizable real-time reconstruction, and ensuring practical applicability by using only images as input without camera parameters or ground truth depth. To this end, we propose UniForward, a feed-forward model to predict 3D Gaussians with anisotropic semantic features from only uncalibrated and unposed sparse-view images. To enable the unified representation of the 3D scene and semantic field, we embed semantic features into 3D Gaussians and predict them through a dual-branch decoupled decoder. During training, we propose a loss-guided view sampler to sample views from easy to hard, eliminating the need for ground truth depth or masks required by previous methods and stabilizing the training process. The whole model can be trained end-to-end using a photometric loss and a distillation loss that leverages semantic features from a pre-trained 2D semantic model. At the inference stage, our UniForward can reconstruct 3D scenes and the corresponding semantic fields in real time from only sparse-view images. The reconstructed 3D scenes achieve high-quality rendering, and the reconstructed 3D semantic field enables the rendering of view-consistent semantic features from arbitrary views, which can be further decoded into dense segmentation masks in an open-vocabulary manner. Experiments on novel view synthesis and novel view segmentation demonstrate that our method achieves state-of-the-art performances for unifying 3D scene and semantic field reconstruction.", "date": "2025-06-11", "categories": []}
{"id": "2506.09836", "title": "DynaSplat: Dynamic-Static Gaussian Splatting with Hierarchical Motion Decomposition for Scene Reconstruction", "abstract": "Reconstructing intricate, ever-changing environments remains a central ambition in computer vision, yet existing solutions often crumble before the complexity of real-world dynamics. We present DynaSplat, an approach that extends Gaussian Splatting to dynamic scenes by integrating dynamic-static separation and hierarchical motion modeling. First, we classify scene elements as static or dynamic through a novel fusion of deformation offset statistics and 2D motion flow consistency, refining our spatial representation to focus precisely where motion matters. We then introduce a hierarchical motion modeling strategy that captures both coarse global transformations and fine-grained local movements, enabling accurate handling of intricate, non-rigid motions. Finally, we integrate physically-based opacity estimation to ensure visually coherent reconstructions, even under challenging occlusions and perspective shifts. Extensive experiments on challenging datasets reveal that DynaSplat not only surpasses state-of-the-art alternatives in accuracy and realism but also provides a more intuitive, compact, and efficient route to dynamic scene reconstruction.", "date": "2025-06-11", "categories": []}
{"id": "2506.09997", "title": "DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos", "abstract": "We introduce the Deformable Gaussian Splats Large Reconstruction Model (DGS-LRM), the first feed-forward method predicting deformable 3D Gaussian splats from a monocular posed video of any dynamic scene. Feed-forward scene reconstruction has gained significant attention for its ability to rapidly create digital replicas of real-world environments. However, most existing models are limited to static scenes and fail to reconstruct the motion of moving objects. Developing a feed-forward model for dynamic scene reconstruction poses significant challenges, including the scarcity of training data and the need for appropriate 3D representations and training paradigms. To address these challenges, we introduce several key technical contributions: an enhanced large-scale synthetic dataset with ground-truth multi-view videos and dense 3D scene flow supervision; a per-pixel deformable 3D Gaussian representation that is easy to learn, supports high-quality dynamic view synthesis, and enables long-range 3D tracking; and a large transformer network that achieves real-time, generalizable dynamic scene reconstruction. Extensive qualitative and quantitative experiments demonstrate that DGS-LRM achieves dynamic scene reconstruction quality comparable to optimization-based methods, while significantly outperforming the state-of-the-art predictive dynamic reconstruction method on real-world examples. Its predicted physically grounded 3D deformation is accurate and can readily adapt for long-range 3D tracking tasks, achieving performance on par with state-of-the-art monocular video 3D tracking methods.", "date": "2025-06-11", "categories": []}
{"id": "2506.08619", "title": "A Probability-guided Sampler for Neural Implicit Surface Rendering", "abstract": "Several variants of Neural Radiance Fields (NeRFs) have significantly improved the accuracy of synthesized images and surface reconstruction of 3D scenes/objects. In all of these methods, a key characteristic is that none can train the neural network with every possible input data, specifically, every pixel and potential 3D point along the projection rays due to scalability issues. While vanilla NeRFs uniformly sample both the image pixels and 3D points along the projection rays, some variants focus only on guiding the sampling of the 3D points along the projection rays. In this paper, we leverage the implicit surface representation of the foreground scene and model a probability density function in a 3D image projection space to achieve a more targeted sampling of the rays toward regions of interest, resulting in improved rendering. Additionally, a new surface reconstruction loss is proposed for improved performance. This new loss fully explores the proposed 3D image projection space model and incorporates near-to-surface and empty space components. By integrating our novel sampling strategy and novel loss into current state-of-the-art neural implicit surface renderers, we achieve more accurate and detailed 3D reconstructions and improved image rendering, especially for the regions of interest in any given scene.", "date": "2025-06-10", "categories": []}
{"id": "2506.08635", "title": "SurfR: Surface Reconstruction with Multi-scale Attention", "abstract": "We propose a fast and accurate surface reconstruction algorithm for unorganized point clouds using an implicit representation. Recent learning methods are either single-object representations with small neural models that allow for high surface details but require per-object training or generalized representations that require larger models and generalize to newer shapes but lack details, and inference is slow. We propose a new implicit representation for general 3D shapes that is faster than all the baselines at their optimum resolution, with only a marginal loss in performance compared to the state-of-the-art. We achieve the best accuracy-speed trade-off using three key contributions. Many implicit methods extract features from the point cloud to classify whether a query point is inside or outside the object. First, to speed up the reconstruction, we show that this feature extraction does not need to use the query point at an early stage (lazy query). Second, we use a parallel multi-scale grid representation to develop robust features for different noise levels and input resolutions. Finally, we show that attention across scales can provide improved reconstruction results.", "date": "2025-06-10", "categories": []}
{"id": "2506.08704", "title": "TraGraph-GS: Trajectory Graph-based Gaussian Splatting for Arbitrary Large-Scale Scene Rendering", "abstract": "High-quality novel view synthesis for large-scale scenes presents a challenging dilemma in 3D computer vision. Existing methods typically partition large scenes into multiple regions, reconstruct a 3D representation using Gaussian splatting for each region, and eventually merge them for novel view rendering. They can accurately render specific scenes, yet they do not generalize effectively for two reasons: (1) rigid spatial partition techniques struggle with arbitrary camera trajectories, and (2) the merging of regions results in Gaussian overlap to distort texture details. To address these challenges, we propose TraGraph-GS, leveraging a trajectory graph to enable high-precision rendering for arbitrarily large-scale scenes. We present a spatial partitioning method for large-scale scenes based on graphs, which incorporates a regularization constraint to enhance the rendering of textures and distant objects, as well as a progressive rendering strategy to mitigate artifacts caused by Gaussian overlap. Experimental results demonstrate its superior performance both on four aerial and four ground datasets and highlight its remarkable efficiency: our method achieves an average improvement of 1.86 dB in PSNR on aerial datasets and 1.62 dB on ground datasets compared to state-of-the-art approaches.", "date": "2025-06-10", "categories": []}
{"id": "2412.04457", "title": "Monocular Dynamic Gaussian Splatting: Fast, Brittle, and Scene Complexity Rules", "abstract": "Gaussian splatting methods are emerging as a popular approach for converting multi-view image data into scene representations that allow view synthesis. In particular, there is interest in enabling view synthesis for dynamic scenes using only monocular input data -- an ill-posed and challenging problem. The fast pace of work in this area has produced multiple simultaneous papers that claim to work best, which cannot all be true. In this work, we organize, benchmark, and analyze many Gaussian-splatting-based methods, providing apples-to-apples comparisons that prior works have lacked. We use multiple existing datasets and a new instructive synthetic dataset designed to isolate factors that affect reconstruction quality. We systematically categorize Gaussian splatting methods into specific motion representation types and quantify how their differences impact performance. Empirically, we find that their rank order is well-defined in synthetic data, but the complexity of real-world data currently overwhelms the differences. Furthermore, the fast rendering speed of all Gaussian-based methods comes at the cost of brittleness in optimization. We summarize our experiments into a list of findings that can help to further progress in this lively problem setting.", "date": "2024-12-05", "categories": []}
{"id": "2505.21041", "title": "CityGo: Lightweight Urban Modeling and Rendering with Proxy Buildings and Residual Gaussians", "abstract": "Accurate and efficient modeling of large-scale urban scenes is critical for applications such as AR navigation, UAV based inspection, and smart city digital twins. While aerial imagery offers broad coverage and complements limitations of ground-based data, reconstructing city-scale environments from such views remains challenging due to occlusions, incomplete geometry, and high memory demands. Recent advances like 3D Gaussian Splatting (3DGS) improve scalability and visual quality but remain limited by dense primitive usage, long training times, and poor suit ability for edge devices. We propose CityGo, a hybrid framework that combines textured proxy geometry with residual and surrounding 3D Gaussians for lightweight, photorealistic rendering of urban scenes from aerial perspectives. Our approach first extracts compact building proxy meshes from MVS point clouds, then uses zero order SH Gaussians to generate occlusion-free textures via image-based rendering and back-projection. To capture high-frequency details, we introduce residual Gaussians placed based on proxy-photo discrepancies and guided by depth priors. Broader urban context is represented by surrounding Gaussians, with importance-aware downsampling applied to non-critical regions to reduce redundancy. A tailored optimization strategy jointly refines proxy textures and Gaussian parameters, enabling real-time rendering of complex urban scenes on mobile GPUs with significantly reduced training and memory requirements. Extensive experiments on real-world aerial datasets demonstrate that our hybrid representation significantly reduces training time, achieving on average 1.4x speedup, while delivering comparable visual fidelity to pure 3D Gaussian Splatting approaches. Furthermore, CityGo enables real-time rendering of large-scale urban scenes on mobile consumer GPUs, with substantially reduced memory usage and energy consumption.", "date": "2025-05-27", "categories": []}
{"id": "2506.06462", "title": "Splat and Replace: 3D Reconstruction with Repetitive Elements", "abstract": "We leverage repetitive elements in 3D scenes to improve novel view synthesis. Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have greatly improved novel view synthesis but renderings of unseen and occluded parts remain low-quality if the training views are not exhaustive enough. Our key observation is that our environment is often full of repetitive elements. We propose to leverage those repetitions to improve the reconstruction of low-quality parts of the scene due to poor coverage and occlusions. We propose a method that segments each repeated instance in a 3DGS reconstruction, registers them together, and allows information to be shared among instances. Our method improves the geometry while also accounting for appearance variations across instances. We demonstrate our method on a variety of synthetic and real scenes with typical repetitive elements, leading to a substantial improvement in the quality of novel view synthesis.", "date": "2025-06-06", "categories": []}
{"id": "2506.06988", "title": "Hybrid Mesh-Gaussian Representation for Efficient Indoor Scene Reconstruction", "abstract": "3D Gaussian splatting (3DGS) has demonstrated exceptional performance in image-based 3D reconstruction and real-time rendering. However, regions with complex textures require numerous Gaussians to capture significant color variations accurately, leading to inefficiencies in rendering speed. To address this challenge, we introduce a hybrid representation for indoor scenes that combines 3DGS with textured meshes. Our approach uses textured meshes to handle texture-rich flat areas, while retaining Gaussians to model intricate geometries. The proposed method begins by pruning and refining the extracted mesh to eliminate geometrically complex regions. We then employ a joint optimization for 3DGS and mesh, incorporating a warm-up strategy and transmittance-aware supervision to balance their contributions seamlessly.Extensive experiments demonstrate that the hybrid representation maintains comparable rendering quality and achieves superior frames per second FPS with fewer Gaussian primitives.", "date": "2025-06-08", "categories": []}
{"id": "2506.07628", "title": "HuSc3D: Human Sculpture dataset for 3D object reconstruction", "abstract": "3D scene reconstruction from 2D images is one of the most important tasks in computer graphics. Unfortunately, existing datasets and benchmarks concentrate on idealized synthetic or meticulously captured realistic data. Such benchmarks fail to convey the inherent complexities encountered in newly acquired real-world scenes. In such scenes especially those acquired outside, the background is often dynamic, and by popular usage of cell phone cameras, there might be discrepancies in, e.g., white balance. To address this gap, we present HuSc3D, a novel dataset specifically designed for rigorous benchmarking of 3D reconstruction models under realistic acquisition challenges. Our dataset uniquely features six highly detailed, fully white sculptures characterized by intricate perforations and minimal textural and color variation. Furthermore, the number of images per scene varies significantly, introducing the additional challenge of limited training data for some instances alongside scenes with a standard number of views. By evaluating popular 3D reconstruction methods on this diverse dataset, we demonstrate the distinctiveness of HuSc3D in effectively differentiating model performance, particularly highlighting the sensitivity of methods to fine geometric details, color ambiguity, and varying data availability--limitations often masked by more conventional datasets.", "date": "2025-06-09", "categories": []}
{"id": "2506.07865", "title": "FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity", "abstract": "In this paper, we aim to model 3D scene geometry, appearance, and the underlying physics purely from multi-view videos. By applying various governing PDEs as PINN losses or incorporating physics simulation into neural networks, existing works often fail to learn complex physical motions at boundaries or require object priors such as masks or types. In this paper, we propose FreeGave to learn the physics of complex dynamic 3D scenes without needing any object priors. The key to our approach is to introduce a physics code followed by a carefully designed divergence-free module for estimating a per-Gaussian velocity field, without relying on the inefficient PINN losses. Extensive experiments on three public datasets and a newly collected challenging real-world dataset demonstrate the superior performance of our method for future frame extrapolation and motion segmentation. Most notably, our investigation into the learned physics codes reveals that they truly learn meaningful 3D physical motion patterns in the absence of any human labels in training.", "date": "2025-06-09", "categories": []}
{"id": "2506.05348", "title": "FreeTimeGS: Free Gaussian Primitives at Anytime and Anywhere for Dynamic Scene Reconstruction", "abstract": "This paper addresses the challenge of reconstructing dynamic 3D scenes with complex motions. Some recent works define 3D Gaussian primitives in the canonical space and use deformation fields to map canonical primitives to observation spaces, achieving real-time dynamic view synthesis. However, these methods often struggle to handle scenes with complex motions due to the difficulty of optimizing deformation fields. To overcome this problem, we propose FreeTimeGS, a novel 4D representation that allows Gaussian primitives to appear at arbitrary time and locations. In contrast to canonical Gaussian primitives, our representation possesses the strong flexibility, thus improving the ability to model dynamic 3D scenes. In addition, we endow each Gaussian primitive with an motion function, allowing it to move to neighboring regions over time, which reduces the temporal redundancy. Experiments results on several datasets show that the rendering quality of our method outperforms recent methods by a large margin. Project page: https://zju3dv.github.io/freetimegs/ .", "date": "2025-06-05", "categories": []}
{"id": "2503.19136", "title": "Stochastic Poisson Surface Reconstruction with One Solve using Geometric Gaussian Processes", "abstract": "Poisson Surface Reconstruction is a widely-used algorithm for reconstructing a surface from an oriented point cloud. To facilitate applications where only partial surface information is available, or scanning is performed sequentially, a recent line of work proposes to incorporate uncertainty into the reconstructed surface via Gaussian process models. The resulting algorithms first perform Gaussian process interpolation, then solve a set of volumetric partial differential equations globally in space, resulting in a computationally expensive two-stage procedure. In this work, we apply recently-developed techniques from geometric Gaussian processes to combine interpolation and surface reconstruction into a single stage, requiring only one linear solve per sample. The resulting reconstructed surface samples can be queried locally in space, without the use of problem-dependent volumetric meshes or grids. These capabilities enable one to (a) perform probabilistic collision detection locally around the region of interest, (b) perform ray casting without evaluating points not on the ray's trajectory, and (c) perform next-view planning on a per-ray basis. They also do not requiring one to approximate kernel matrix inverses with diagonal matrices as part of intermediate computations, unlike prior methods. Results show that our approach provides a cleaner, more-principled, and more-flexible stochastic surface reconstruction pipeline.", "date": "2025-03-24", "categories": []}
{"id": "2506.04444", "title": "Photoreal Scene Reconstruction from an Egocentric Device", "abstract": "In this paper, we investigate the challenges associated with using egocentric devices to photorealistic reconstruct the scene in high dynamic range. Existing methodologies typically assume using frame-rate 6DoF pose estimated from the device's visual-inertial odometry system, which may neglect crucial details necessary for pixel-accurate reconstruction. This study presents two significant findings. Firstly, in contrast to mainstream work treating RGB camera as global shutter frame-rate camera, we emphasize the importance of employing visual-inertial bundle adjustment (VIBA) to calibrate the precise timestamps and movement of the rolling shutter RGB sensing camera in a high frequency trajectory format, which ensures an accurate calibration of the physical properties of the rolling-shutter camera. Secondly, we incorporate a physical image formation model based into Gaussian Splatting, which effectively addresses the sensor characteristics, including the rolling-shutter effect of RGB cameras and the dynamic ranges measured by sensors. Our proposed formulation is applicable to the widely-used variants of Gaussian Splats representation. We conduct a comprehensive evaluation of our pipeline using the open-source Project Aria device under diverse indoor and outdoor lighting conditions, and further validate it on a Meta Quest3 device. Across all experiments, we observe a consistent visual enhancement of +1 dB in PSNR by incorporating VIBA, with an additional +1 dB achieved through our proposed image formation model. Our complete implementation, evaluation datasets, and recording profile are available at http://www.projectaria.com/photoreal-reconstruction/", "date": "2025-06-04", "categories": []}
{"id": "2506.05204", "title": "OGGSplat: Open Gaussian Growing for Generalizable Reconstruction with Expanded Field-of-View", "abstract": "Reconstructing semantic-aware 3D scenes from sparse views is a challenging yet essential research direction, driven by the demands of emerging applications such as virtual reality and embodied AI. Existing per-scene optimization methods require dense input views and incur high computational costs, while generalizable approaches often struggle to reconstruct regions outside the input view cone. In this paper, we propose OGGSplat, an open Gaussian growing method that expands the field-of-view in generalizable 3D reconstruction. Our key insight is that the semantic attributes of open Gaussians provide strong priors for image extrapolation, enabling both semantic consistency and visual plausibility. Specifically, once open Gaussians are initialized from sparse views, we introduce an RGB-semantic consistent inpainting module applied to selected rendered views. This module enforces bidirectional control between an image diffusion model and a semantic diffusion model. The inpainted regions are then lifted back into 3D space for efficient and progressive Gaussian parameter optimization. To evaluate our method, we establish a Gaussian Outpainting (GO) benchmark that assesses both semantic and generative quality of reconstructed open-vocabulary scenes. OGGSplat also demonstrates promising semantic-aware scene reconstruction capabilities when provided with two view images captured directly from a smartphone camera.", "date": "2025-06-05", "categories": []}
{"id": "2506.02112", "title": "SAB3R: Semantic-Augmented Backbone in 3D Reconstruction", "abstract": "We introduce a new task, Map and Locate, which unifies the traditionally distinct objectives of open-vocabulary segmentation - detecting and segmenting object instances based on natural language queries - and 3D reconstruction, the process of estimating a scene's 3D structure from visual inputs. Specifically, Map and Locate involves generating a point cloud from an unposed video and segmenting object instances based on open-vocabulary queries. This task serves as a critical step toward real-world embodied AI applications and introduces a practical task that bridges reconstruction, recognition and reorganization. To tackle this task, we introduce a simple yet effective baseline, which we denote as SAB3R. Our approach builds upon MASt3R, a recent breakthrough in 3D computer vision, and incorporates a lightweight distillation strategy. This method transfers dense, per-pixel semantic features from 2D vision backbones (eg, CLIP and DINOv2) to enhance MASt3R's capabilities. Without introducing any auxiliary frozen networks, our model generates per-pixel semantic features and constructs cohesive point maps in a single forward pass. Compared to separately deploying MASt3R and CLIP, our unified model, SAB3R, achieves superior performance on the Map and Locate benchmark. Furthermore, we evaluate SAB3R on both 2D semantic segmentation and 3D tasks to comprehensively validate its effectiveness.", "date": "2025-06-02", "categories": []}
{"id": "2506.03713", "title": "Pl\\\"uckeRF: A Line-based 3D Representation for Few-view Reconstruction", "abstract": "Feed-forward 3D reconstruction methods aim to predict the 3D structure of a scene directly from input images, providing a faster alternative to per-scene optimization approaches. Significant progress has been made in single-view and few-view reconstruction using learned priors that infer object shape and appearance, even for unobserved regions. However, there is substantial potential to enhance these methods by better leveraging information from multiple views when available. To address this, we propose a few-view reconstruction model that more effectively harnesses multi-view information. Our approach introduces a simple mechanism that connects the 3D representation with pixel rays from the input views, allowing for preferential sharing of information between nearby 3D locations and between 3D locations and nearby pixel rays. We achieve this by defining the 3D representation as a set of structured, feature-augmented lines; the Pl\\\"uckeRF representation. Using this representation, we demonstrate improvements in reconstruction quality over the equivalent triplane representation and state-of-the-art feedforward reconstruction methods.", "date": "2025-06-04", "categories": []}
{"id": "2506.03872", "title": "JointSplat: Probabilistic Joint Flow-Depth Optimization for Sparse-View Gaussian Splatting", "abstract": "Reconstructing 3D scenes from sparse viewpoints is a long-standing challenge with wide applications. Recent advances in feed-forward 3D Gaussian sparse-view reconstruction methods provide an efficient solution for real-time novel view synthesis by leveraging geometric priors learned from large-scale multi-view datasets and computing 3D Gaussian centers via back-projection. Despite offering strong geometric cues, both feed-forward multi-view depth estimation and flow-depth joint estimation face key limitations: the former suffers from mislocation and artifact issues in low-texture or repetitive regions, while the latter is prone to local noise and global inconsistency due to unreliable matches when ground-truth flow supervision is unavailable. To overcome this, we propose JointSplat, a unified framework that leverages the complementarity between optical flow and depth via a novel probabilistic optimization mechanism. Specifically, this pixel-level mechanism scales the information fusion between depth and flow based on the matching probability of optical flow during training. Building upon the above mechanism, we further propose a novel multi-view depth-consistency loss to leverage the reliability of supervision while suppressing misleading gradients in uncertain areas. Evaluated on RealEstate10K and ACID, JointSplat consistently outperforms state-of-the-art (SOTA) methods, demonstrating the effectiveness and robustness of our proposed probabilistic joint flow-depth optimization approach for high-fidelity sparse-view 3D reconstruction.", "date": "2025-06-04", "categories": []}
{"id": "2506.04174", "title": "FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting", "abstract": "3D Gaussian splatting (3DGS) has enabled various applications in 3D scene representation and novel view synthesis due to its efficient rendering capabilities. However, 3DGS demands relatively significant GPU memory, limiting its use on devices with restricted computational resources. Previous approaches have focused on pruning less important Gaussians, effectively compressing 3DGS but often requiring a fine-tuning stage and lacking adaptability for the specific memory needs of different devices. In this work, we present an elastic inference method for 3DGS. Given an input for the desired model size, our method selects and transforms a subset of Gaussians, achieving substantial rendering performance without additional fine-tuning. We introduce a tiny learnable module that controls Gaussian selection based on the input percentage, along with a transformation module that adjusts the selected Gaussians to complement the performance of the reduced model. Comprehensive experiments on ZipNeRF, MipNeRF and Tanks\\&Temples scenes demonstrate the effectiveness of our approach. Code is available at https://flexgs.github.io.", "date": "2025-06-04", "categories": []}
{"id": "2505.19919", "title": "Weather-Magician: Reconstruction and Rendering Framework for 4D Weather Synthesis In Real Time", "abstract": "For tasks such as urban digital twins, VR/AR/game scene design, or creating synthetic films, the traditional industrial approach often involves manually modeling scenes and using various rendering engines to complete the rendering process. This approach typically requires high labor costs and hardware demands, and can result in poor quality when replicating complex real-world scenes. A more efficient approach is to use data from captured real-world scenes, then apply reconstruction and rendering algorithms to quickly recreate the authentic scene. However, current algorithms are unable to effectively reconstruct and render real-world weather effects. To address this, we propose a framework based on gaussian splatting, that can reconstruct real scenes and render them under synthesized 4D weather effects. Our work can simulate various common weather effects by applying Gaussians modeling and rendering techniques. It supports continuous dynamic weather changes and can easily control the details of the effects. Additionally, our work has low hardware requirements and achieves real-time rendering performance. The result demos can be accessed on our project homepage: weathermagician.github.io", "date": "2025-05-26", "categories": []}
{"id": "2506.02265", "title": "Rig3R: Rig-Aware Conditioning for Learned 3D Reconstruction", "abstract": "Estimating agent pose and 3D scene structure from multi-camera rigs is a central task in embodied AI applications such as autonomous driving. Recent learned approaches such as DUSt3R have shown impressive performance in multiview settings. However, these models treat images as unstructured collections, limiting effectiveness in scenarios where frames are captured from synchronized rigs with known or inferable structure.\n  To this end, we introduce Rig3R, a generalization of prior multiview reconstruction models that incorporates rig structure when available, and learns to infer it when not. Rig3R conditions on optional rig metadata including camera ID, time, and rig poses to develop a rig-aware latent space that remains robust to missing information. It jointly predicts pointmaps and two types of raymaps: a pose raymap relative to a global frame, and a rig raymap relative to a rig-centric frame consistent across time. Rig raymaps allow the model to infer rig structure directly from input images when metadata is missing.\n  Rig3R achieves state-of-the-art performance in 3D reconstruction, camera pose estimation, and rig discovery, outperforming both traditional and learned methods by 17-45% mAA across diverse real-world rig datasets, all in a single forward pass without post-processing or iterative refinement.", "date": "2025-06-02", "categories": []}
{"id": "2506.02493", "title": "Towards In-the-wild 3D Plane Reconstruction from a Single Image", "abstract": "3D plane reconstruction from a single image is a crucial yet challenging topic in 3D computer vision. Previous state-of-the-art (SOTA) methods have focused on training their system on a single dataset from either indoor or outdoor domain, limiting their generalizability across diverse testing data. In this work, we introduce a novel framework dubbed ZeroPlane, a Transformer-based model targeting zero-shot 3D plane detection and reconstruction from a single image, over diverse domains and environments. To enable data-driven models across multiple domains, we have curated a large-scale planar benchmark, comprising over 14 datasets and 560,000 high-resolution, dense planar annotations for diverse indoor and outdoor scenes. To address the challenge of achieving desirable planar geometry on multi-dataset training, we propose to disentangle the representation of plane normal and offset, and employ an exemplar-guided, classification-then-regression paradigm to learn plane and offset respectively. Additionally, we employ advanced backbones as image encoder, and present an effective pixel-geometry-enhanced plane embedding module to further facilitate planar reconstruction. Extensive experiments across multiple zero-shot evaluation datasets have demonstrated that our approach significantly outperforms previous methods on both reconstruction accuracy and generalizability, especially over in-the-wild data. Our code and data are available at: https://github.com/jcliu0428/ZeroPlane.", "date": "2025-06-03", "categories": []}
{"id": "2404.09748", "title": "LetsGo: Large-Scale Garage Modeling and Rendering via LiDAR-Assisted Gaussian Primitives", "abstract": "Large garages are ubiquitous yet intricate scenes that present unique challenges due to their monotonous colors, repetitive patterns, reflective surfaces, and transparent vehicle glass. Conventional Structure from Motion (SfM) methods for camera pose estimation and 3D reconstruction often fail in these environments due to poor correspondence construction. To address these challenges, we introduce LetsGo, a LiDAR-assisted Gaussian splatting framework for large-scale garage modeling and rendering. We develop a handheld scanner, Polar, equipped with IMU, LiDAR, and a fisheye camera, to facilitate accurate data acquisition. Using this Polar device, we present the GarageWorld dataset, consisting of eight expansive garage scenes with diverse geometric structures, which will be made publicly available for further research. Our approach demonstrates that LiDAR point clouds collected by the Polar device significantly enhance a suite of 3D Gaussian splatting algorithms for garage scene modeling and rendering. We introduce a novel depth regularizer that effectively eliminates floating artifacts in rendered images. Additionally, we propose a multi-resolution 3D Gaussian representation designed for Level-of-Detail (LOD) rendering. This includes adapted scaling factors for individual levels and a random-resolution-level training scheme to optimize the Gaussians across different resolutions. This representation enables efficient rendering of large-scale garage scenes on lightweight devices via a web-based renderer. Experimental results on our GarageWorld dataset, as well as on ScanNet++ and KITTI-360, demonstrate the superiority of our method in terms of rendering quality and resource efficiency.", "date": "2024-04-15", "categories": []}
{"id": "2406.04343", "title": "Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image", "abstract": "We propose Flash3D, a method for scene reconstruction and novel view synthesis from a single image which is both very generalisable and efficient. For generalisability, we start from a \"foundation\" model for monocular depth estimation and extend it to a full 3D shape and appearance reconstructor. For efficiency, we base this extension on feed-forward Gaussian Splatting. Specifically, we predict a first layer of 3D Gaussians at the predicted depth, and then add additional layers of Gaussians that are offset in space, allowing the model to complete the reconstruction behind occlusions and truncations. Flash3D is very efficient, trainable on a single GPU in a day, and thus accessible to most researchers. It achieves state-of-the-art results when trained and tested on RealEstate10k. When transferred to unseen datasets like NYU it outperforms competitors by a large margin. More impressively, when transferred to KITTI, Flash3D achieves better PSNR than methods trained specifically on that dataset. In some instances, it even outperforms recent methods that use multiple views as input. Code, models, demo, and more results are available at https://www.robots.ox.ac.uk/~vgg/research/flash3d/.", "date": "2024-06-06", "categories": []}
{"id": "2408.08206", "title": "WaterSplatting: Fast Underwater 3D Scene Reconstruction Using Gaussian Splatting", "abstract": "The underwater 3D scene reconstruction is a challenging, yet interesting problem with applications ranging from naval robots to VR experiences. The problem was successfully tackled by fully volumetric NeRF-based methods which can model both the geometry and the medium (water). Unfortunately, these methods are slow to train and do not offer real-time rendering. More recently, 3D Gaussian Splatting (3DGS) method offered a fast alternative to NeRFs. However, because it is an explicit method that renders only the geometry, it cannot render the medium and is therefore unsuited for underwater reconstruction. Therefore, we propose a novel approach that fuses volumetric rendering with 3DGS to handle underwater data effectively. Our method employs 3DGS for explicit geometry representation and a separate volumetric field (queried once per pixel) for capturing the scattering medium. This dual representation further allows the restoration of the scenes by removing the scattering medium. Our method outperforms state-of-the-art NeRF-based methods in rendering quality on the underwater SeaThru-NeRF dataset. Furthermore, it does so while offering real-time rendering performance, addressing the efficiency limitations of existing methods. Web: https://water-splatting.github.io", "date": "2024-08-15", "categories": []}
{"id": "2409.17345", "title": "SeaSplat: Representing Underwater Scenes with 3D Gaussian Splatting and a Physically Grounded Image Formation Model", "abstract": "We introduce SeaSplat, a method to enable real-time rendering of underwater scenes leveraging recent advances in 3D radiance fields. Underwater scenes are challenging visual environments, as rendering through a medium such as water introduces both range and color dependent effects on image capture. We constrain 3D Gaussian Splatting (3DGS), a recent advance in radiance fields enabling rapid training and real-time rendering of full 3D scenes, with a physically grounded underwater image formation model. Applying SeaSplat to the real-world scenes from SeaThru-NeRF dataset, a scene collected by an underwater vehicle in the US Virgin Islands, and simulation-degraded real-world scenes, not only do we see increased quantitative performance on rendering novel viewpoints from the scene with the medium present, but are also able to recover the underlying true color of the scene and restore renders to be without the presence of the intervening medium. We show that the underwater image formation helps learn scene structure, with better depth maps, as well as show that our improvements maintain the significant computational improvements afforded by leveraging a 3D Gaussian representation.", "date": "2024-09-25", "categories": []}
{"id": "2503.19753", "title": "A Survey on Event-driven 3D Reconstruction: Development under Different Categories", "abstract": "Event cameras have gained increasing attention for 3D reconstruction due to their high temporal resolution, low latency, and high dynamic range. They capture per-pixel brightness changes asynchronously, allowing accurate reconstruction under fast motion and challenging lighting conditions. In this survey, we provide a comprehensive review of event-driven 3D reconstruction methods, including stereo, monocular, and multimodal systems. We further categorize recent developments based on geometric, learning-based, and hybrid approaches. Emerging trends, such as neural radiance fields and 3D Gaussian splatting with event data, are also covered. The related works are structured chronologically to illustrate the innovations and progression within the field. To support future research, we also highlight key research gaps and future research directions in dataset, experiment, evaluation, event representation, etc.", "date": "2025-03-25", "categories": []}
{"id": "2506.01379", "title": "RadarSplat: Radar Gaussian Splatting for High-Fidelity Data Synthesis and 3D Reconstruction of Autonomous Driving Scenes", "abstract": "High-Fidelity 3D scene reconstruction plays a crucial role in autonomous driving by enabling novel data generation from existing datasets. This allows simulating safety-critical scenarios and augmenting training datasets without incurring further data collection costs. While recent advances in radiance fields have demonstrated promising results in 3D reconstruction and sensor data synthesis using cameras and LiDAR, their potential for radar remains largely unexplored. Radar is crucial for autonomous driving due to its robustness in adverse weather conditions like rain, fog, and snow, where optical sensors often struggle. Although the state-of-the-art radar-based neural representation shows promise for 3D driving scene reconstruction, it performs poorly in scenarios with significant radar noise, including receiver saturation and multipath reflection. Moreover, it is limited to synthesizing preprocessed, noise-excluded radar images, failing to address realistic radar data synthesis. To address these limitations, this paper proposes RadarSplat, which integrates Gaussian Splatting with novel radar noise modeling to enable realistic radar data synthesis and enhanced 3D reconstruction. Compared to the state-of-the-art, RadarSplat achieves superior radar image synthesis (+3.4 PSNR / 2.6x SSIM) and improved geometric reconstruction (-40% RMSE / 1.5x Accuracy), demonstrating its effectiveness in generating high-fidelity radar data and scene reconstruction. A project page is available at https://umautobots.github.io/radarsplat.", "date": "2025-06-02", "categories": []}
{"id": "2506.01389", "title": "Neural shape reconstruction from multiple views with static pattern projection", "abstract": "Active-stereo-based 3D shape measurement is crucial for various purposes, such as industrial inspection, reverse engineering, and medical systems, due to its strong ability to accurately acquire the shape of textureless objects. Active stereo systems typically consist of a camera and a pattern projector, tightly fixed to each other, and precise calibration between a camera and a projector is required, which in turn decreases the usability of the system. If a camera and a projector can be freely moved during shape scanning process, it will drastically increase the convenience of the usability of the system. To realize it, we propose a technique to recover the shape of the target object by capturing multiple images while both the camera and the projector are in motion, and their relative poses are auto-calibrated by our neural signed-distance-field (NeuralSDF) using novel volumetric differential rendering technique. In the experiment, the proposed method is evaluated by performing 3D reconstruction using both synthetic and real images.", "date": "2025-06-02", "categories": []}
{"id": "2505.10923", "title": "GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats", "abstract": "Accurate temporal reconstructions of plant growth are essential for plant phenotyping and breeding, yet remain challenging due to complex geometries, occlusions, and non-rigid deformations of plants. We present a novel framework for building temporal digital twins of plants by combining 3D Gaussian Splatting with a robust sample alignment pipeline. Our method begins by reconstructing Gaussian Splats from multi-view camera data, then leverages a two-stage registration approach: coarse alignment through feature-based matching and Fast Global Registration, followed by fine alignment with Iterative Closest Point. This pipeline yields a consistent 4D model of plant development in discrete time steps. We evaluate the approach on data from the Netherlands Plant Eco-phenotyping Center, demonstrating detailed temporal reconstructions of Sequoia and Quinoa species. Videos and Images can be seen at https://berkeleyautomation.github.io/GrowSplat/", "date": "2025-05-16", "categories": []}
{"id": "2505.19854", "title": "Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud", "abstract": "Gaussian Splatting (GS) has gained attention as a fast and effective method for novel view synthesis. It has also been applied to 3D reconstruction using multi-view images and can achieve fast and accurate 3D reconstruction. However, GS assumes that the input contains a large number of multi-view images, and therefore, the reconstruction accuracy significantly decreases when only a limited number of input images are available. One of the main reasons is the insufficient number of 3D points in the sparse point cloud obtained through Structure from Motion (SfM), which results in a poor initialization for optimizing the Gaussian primitives. We propose a new 3D reconstruction method, called Sparse2DGS, to enhance 2DGS in reconstructing objects using only three images. Sparse2DGS employs DUSt3R, a fundamental model for stereo images, along with COLMAP MVS to generate highly accurate and dense 3D point clouds, which are then used to initialize 2D Gaussians. Through experiments on the DTU dataset, we show that Sparse2DGS can accurately reconstruct the 3D shapes of objects using just three images. The project page is available at https://gsisaoki.github.io/SPARSE2DGS/", "date": "2025-05-26", "categories": []}
{"id": "2505.23280", "title": "Holistic Large-Scale Scene Reconstruction via Mixed Gaussian Splatting", "abstract": "Recent advances in 3D Gaussian Splatting have shown remarkable potential for novel view synthesis. However, most existing large-scale scene reconstruction methods rely on the divide-and-conquer paradigm, which often leads to the loss of global scene information and requires complex parameter tuning due to scene partitioning and local optimization. To address these limitations, we propose MixGS, a novel holistic optimization framework for large-scale 3D scene reconstruction. MixGS models the entire scene holistically by integrating camera pose and Gaussian attributes into a view-aware representation, which is decoded into fine-detailed Gaussians. Furthermore, a novel mixing operation combines decoded and original Gaussians to jointly preserve global coherence and local fidelity. Extensive experiments on large-scale scenes demonstrate that MixGS achieves state-of-the-art rendering quality and competitive speed, while significantly reducing computational requirements, enabling large-scale scene reconstruction training on a single 24GB VRAM GPU. The code will be released at https://github.com/azhuantou/MixGS.", "date": "2025-05-29", "categories": []}
{"id": "2505.23434", "title": "UrbanCraft: Urban View Extrapolation via Hierarchical Sem-Geometric Priors", "abstract": "Existing neural rendering-based urban scene reconstruction methods mainly focus on the Interpolated View Synthesis (IVS) setting that synthesizes from views close to training camera trajectory. However, IVS can not guarantee the on-par performance of the novel view outside the training camera distribution (\\textit{e.g.}, looking left, right, or downwards), which limits the generalizability of the urban reconstruction application. Previous methods have optimized it via image diffusion, but they fail to handle text-ambiguous or large unseen view angles due to coarse-grained control of text-only diffusion. In this paper, we design UrbanCraft, which surmounts the Extrapolated View Synthesis (EVS) problem using hierarchical sem-geometric representations serving as additional priors. Specifically, we leverage the partially observable scene to reconstruct coarse semantic and geometric primitives, establishing a coarse scene-level prior through an occupancy grid as the base representation. Additionally, we incorporate fine instance-level priors from 3D bounding boxes to enhance object-level details and spatial relationships. Building on this, we propose the \\textbf{H}ierarchical \\textbf{S}emantic-Geometric-\\textbf{G}uided Variational Score Distillation (HSG-VSD), which integrates semantic and geometric constraints from pretrained UrbanCraft2D into the score distillation sampling process, forcing the distribution to be consistent with the observable scene. Qualitative and quantitative comparisons demonstrate the effectiveness of our methods on EVS problem.", "date": "2025-05-29", "categories": []}
{"id": "2505.22400", "title": "STDR: Spatio-Temporal Decoupling for Real-Time Dynamic Scene Rendering", "abstract": "Although dynamic scene reconstruction has long been a fundamental challenge in 3D vision, the recent emergence of 3D Gaussian Splatting (3DGS) offers a promising direction by enabling high-quality, real-time rendering through explicit Gaussian primitives. However, existing 3DGS-based methods for dynamic reconstruction often suffer from \\textit{spatio-temporal incoherence} during initialization, where canonical Gaussians are constructed by aggregating observations from multiple frames without temporal distinction. This results in spatio-temporally entangled representations, making it difficult to model dynamic motion accurately. To overcome this limitation, we propose \\textbf{STDR} (Spatio-Temporal Decoupling for Real-time rendering), a plug-and-play module that learns spatio-temporal probability distributions for each Gaussian. STDR introduces a spatio-temporal mask, a separated deformation field, and a consistency regularization to jointly disentangle spatial and temporal patterns. Extensive experiments demonstrate that incorporating our module into existing 3DGS-based dynamic scene reconstruction frameworks leads to notable improvements in both reconstruction quality and spatio-temporal consistency across synthetic and real-world benchmarks.", "date": "2025-05-28", "categories": []}
{"id": "2505.22616", "title": "PS4PRO: Pixel-to-pixel Supervision for Photorealistic Rendering and Optimization", "abstract": "Neural rendering methods have gained significant attention for their ability to reconstruct 3D scenes from 2D images. The core idea is to take multiple views as input and optimize the reconstructed scene by minimizing the uncertainty in geometry and appearance across the views. However, the reconstruction quality is limited by the number of input views. This limitation is further pronounced in complex and dynamic scenes, where certain angles of objects are never seen. In this paper, we propose to use video frame interpolation as the data augmentation method for neural rendering. Furthermore, we design a lightweight yet high-quality video frame interpolation model, PS4PRO (Pixel-to-pixel Supervision for Photorealistic Rendering and Optimization). PS4PRO is trained on diverse video datasets, implicitly modeling camera movement as well as real-world 3D geometry. Our model performs as an implicit world prior, enriching the photo supervision for 3D reconstruction. By leveraging the proposed method, we effectively augment existing datasets for neural rendering methods. Our experimental results indicate that our method improves the reconstruction performance on both static and dynamic scenes.", "date": "2025-05-28", "categories": []}
{"id": "2505.20610", "title": "OmniIndoor3D: Comprehensive Indoor 3D Reconstruction", "abstract": "We propose a novel framework for comprehensive indoor 3D reconstruction using Gaussian representations, called OmniIndoor3D. This framework enables accurate appearance, geometry, and panoptic reconstruction of diverse indoor scenes captured by a consumer-level RGB-D camera. Since 3DGS is primarily optimized for photorealistic rendering, it lacks the precise geometry critical for high-quality panoptic reconstruction. Therefore, OmniIndoor3D first combines multiple RGB-D images to create a coarse 3D reconstruction, which is then used to initialize the 3D Gaussians and guide the 3DGS training. To decouple the optimization conflict between appearance and geometry, we introduce a lightweight MLP that adjusts the geometric properties of 3D Gaussians. The introduced lightweight MLP serves as a low-pass filter for geometry reconstruction and significantly reduces noise in indoor scenes. To improve the distribution of Gaussian primitives, we propose a densification strategy guided by panoptic priors to encourage smoothness on planar surfaces. Through the joint optimization of appearance, geometry, and panoptic reconstruction, OmniIndoor3D provides comprehensive 3D indoor scene understanding, which facilitates accurate and robust robotic navigation. We perform thorough evaluations across multiple datasets, and OmniIndoor3D achieves state-of-the-art results in appearance, geometry, and panoptic reconstruction. We believe our work bridges a critical gap in indoor 3D reconstruction. The code will be released at: https://ucwxb.github.io/OmniIndoor3D/", "date": "2025-05-27", "categories": []}
{"id": "2505.20729", "title": "Intern-GS: Vision Model Guided Sparse-View 3D Gaussian Splatting", "abstract": "Sparse-view scene reconstruction often faces significant challenges due to the constraints imposed by limited observational data. These limitations result in incomplete information, leading to suboptimal reconstructions using existing methodologies. To address this, we present Intern-GS, a novel approach that effectively leverages rich prior knowledge from vision foundation models to enhance the process of sparse-view Gaussian Splatting, thereby enabling high-quality scene reconstruction. Specifically, Intern-GS utilizes vision foundation models to guide both the initialization and the optimization process of 3D Gaussian splatting, effectively addressing the limitations of sparse inputs. In the initialization process, our method employs DUSt3R to generate a dense and non-redundant gaussian point cloud. This approach significantly alleviates the limitations encountered by traditional structure-from-motion (SfM) methods, which often struggle under sparse-view constraints. During the optimization process, vision foundation models predict depth and appearance for unobserved views, refining the 3D Gaussians to compensate for missing information in unseen regions. Extensive experiments demonstrate that Intern-GS achieves state-of-the-art rendering quality across diverse datasets, including both forward-facing and large-scale scenes, such as LLFF, DTU, and Tanks and Temples.", "date": "2025-05-27", "categories": []}
{"id": "2505.21258", "title": "Plenodium: UnderWater 3D Scene Reconstruction with Plenoptic Medium Representation", "abstract": "We present Plenodium (plenoptic medium), an effective and efficient 3D representation framework capable of jointly modeling both objects and participating media. In contrast to existing medium representations that rely solely on view-dependent modeling, our novel plenoptic medium representation incorporates both directional and positional information through spherical harmonics encoding, enabling highly accurate underwater scene reconstruction. To address the initialization challenge in degraded underwater environments, we propose the pseudo-depth Gaussian complementation to augment COLMAP-derived point clouds with robust depth priors. In addition, a depth ranking regularized loss is developed to optimize the geometry of the scene and improve the ordinal consistency of the depth maps. Extensive experiments on real-world underwater datasets demonstrate that our method achieves significant improvements in 3D reconstruction. Furthermore, we conduct a simulated dataset with ground truth and the controllable scattering medium to demonstrate the restoration capability of our method in underwater scenarios. Our code and dataset are available at https://plenodium.github.io/.", "date": "2025-05-27", "categories": []}
{"id": "2312.00639", "title": "RefinedFields: Radiance Fields Refinement for Planar Scene Representations", "abstract": "Planar scene representations have recently witnessed increased interests for modeling scenes from images, as their lightweight planar structure enables compatibility with image-based models. Notably, K-Planes have gained particular attention as they extend planar scene representations to support in-the-wild scenes, in addition to object-level scenes. However, their visual quality has recently lagged behind that of state-of-the-art techniques. To reduce this gap, we propose RefinedFields, a method that leverages pre-trained networks to refine K-Planes scene representations via optimization guidance using an alternating training procedure. We carry out extensive experiments and verify the merit of our method on synthetic data and real tourism photo collections. RefinedFields enhances rendered scenes with richer details and improves upon its base representation on the task of novel view synthesis. Our project page can be found at https://refinedfields.github.io .", "date": "2023-12-01", "categories": []}
{"id": "2504.17670", "title": "DiMeR: Disentangled Mesh Reconstruction Model", "abstract": "We propose DiMeR, a novel geometry-texture disentangled feed-forward model with 3D supervision for sparse-view mesh reconstruction. Existing methods confront two persistent obstacles: (i) textures can conceal geometric errors, i.e., visually plausible images can be rendered even with wrong geometry, producing multiple ambiguous optimization objectives in geometry-texture mixed solution space for similar objects; and (ii) prevailing mesh extraction methods are redundant, unstable, and lack 3D supervision. To solve these challenges, we rethink the inductive bias for mesh reconstruction. First, we disentangle the unified geometry-texture solution space, where a single input admits multiple feasible solutions, into geometry and texture spaces individually. Specifically, given that normal maps are strictly consistent with geometry and accurately capture surface variations, the normal maps serve as the sole input for geometry prediction in DiMeR, while the texture is estimated from RGB images. Second, we streamline the algorithm of mesh extraction by eliminating modules with low performance/cost ratios and redesigning regularization losses with 3D supervision. Notably, DiMeR still accepts raw RGB images as input by leveraging foundation models for normal prediction. Extensive experiments demonstrate that DiMeR generalises across sparse-view-, single-image-, and text-to-3D tasks, consistently outperforming baselines. On the GSO and OmniObject3D datasets, DiMeR significantly reduces Chamfer Distance by more than 30%.", "date": "2025-04-24", "categories": []}
{"id": "2505.18649", "title": "SuperGS: Consistent and Detailed 3D Super-Resolution Scene Reconstruction via Gaussian Splatting", "abstract": "Recently, 3D Gaussian Splatting (3DGS) has excelled in novel view synthesis (NVS) with its real-time rendering capabilities and superior quality. However, it encounters challenges for high-resolution novel view synthesis (HRNVS) due to the coarse nature of primitives derived from low-resolution input views. To address this issue, we propose SuperGS, an expansion of Scaffold-GS designed with a two-stage coarse-to-fine training framework. In the low-resolution stage, we introduce a latent feature field to represent the low-resolution scene, which serves as both the initialization and foundational information for super-resolution optimization. In the high-resolution stage, we propose a multi-view consistent densification strategy that backprojects high-resolution depth maps based on error maps and employs a multi-view voting mechanism, mitigating ambiguities caused by multi-view inconsistencies in the pseudo labels provided by 2D prior models while avoiding Gaussian redundancy. Furthermore, we model uncertainty through variational feature learning and use it to guide further scene representation refinement and adjust the supervisory effect of pseudo-labels, ensuring consistent and detailed scene reconstruction. Extensive experiments demonstrate that SuperGS outperforms state-of-the-art HRNVS methods on both forward-facing and 360-degree datasets.", "date": "2025-05-24", "categories": []}
{"id": "2505.20126", "title": "OB3D: A New Dataset for Benchmarking Omnidirectional 3D Reconstruction Using Blender", "abstract": "Recent advancements in radiance field rendering, exemplified by Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have significantly progressed 3D modeling and reconstruction. The use of multiple 360-degree omnidirectional images for these tasks is increasingly favored due to advantages in data acquisition and comprehensive scene capture. However, the inherent geometric distortions in common omnidirectional representations, such as equirectangular projection (particularly severe in polar regions and varying with latitude), pose substantial challenges to achieving high-fidelity 3D reconstructions. Current datasets, while valuable, often lack the specific focus, scene composition, and ground truth granularity required to systematically benchmark and drive progress in overcoming these omnidirectional-specific challenges. To address this critical gap, we introduce Omnidirectional Blender 3D (OB3D), a new synthetic dataset curated for advancing 3D reconstruction from multiple omnidirectional images. OB3D features diverse and complex 3D scenes generated from Blender 3D projects, with a deliberate emphasis on challenging scenarios. The dataset provides comprehensive ground truth, including omnidirectional RGB images, precise omnidirectional camera parameters, and pixel-aligned equirectangular maps for depth and normals, alongside evaluation metrics. By offering a controlled yet challenging environment, OB3Daims to facilitate the rigorous evaluation of existing methods and prompt the development of new techniques to enhance the accuracy and reliability of 3D reconstruction from omnidirectional images.", "date": "2025-05-26", "categories": []}
{"id": "2505.20267", "title": "HaloGS: Loose Coupling of Compact Geometry and Gaussian Splats for 3D Scenes", "abstract": "High fidelity 3D reconstruction and rendering hinge on capturing precise geometry while preserving photo realistic detail. Most existing methods either fuse these goals into a single cumbersome model or adopt hybrid schemes whose uniform primitives lead to a trade off between efficiency and fidelity. In this paper, we introduce HaloGS, a dual representation that loosely couples coarse triangles for geometry with Gaussian primitives for appearance, motivated by the lightweight classic geometry representations and their proven efficiency in real world applications. Our design yields a compact yet expressive model capable of photo realistic rendering across both indoor and outdoor environments, seamlessly adapting to varying levels of scene complexity. Experiments on multiple benchmark datasets demonstrate that our method yields both compact, accurate geometry and high fidelity renderings, especially in challenging scenarios where robust geometric structure make a clear difference.", "date": "2025-05-26", "categories": []}
{"id": "2505.17402", "title": "From Flight to Insight: Semantic 3D Reconstruction for Aerial Inspection via Gaussian Splatting and Language-Guided Segmentation", "abstract": "High-fidelity 3D reconstruction is critical for aerial inspection tasks such as infrastructure monitoring, structural assessment, and environmental surveying. While traditional photogrammetry techniques enable geometric modeling, they lack semantic interpretability, limiting their effectiveness for automated inspection workflows. Recent advances in neural rendering and 3D Gaussian Splatting (3DGS) offer efficient, photorealistic reconstructions but similarly lack scene-level understanding.\n  In this work, we present a UAV-based pipeline that extends Feature-3DGS for language-guided 3D segmentation. We leverage LSeg-based feature fields with CLIP embeddings to generate heatmaps in response to language prompts. These are thresholded to produce rough segmentations, and the highest-scoring point is then used as a prompt to SAM or SAM2 for refined 2D segmentation on novel view renderings. Our results highlight the strengths and limitations of various feature field backbones (CLIP-LSeg, SAM, SAM2) in capturing meaningful structure in large-scale outdoor environments. We demonstrate that this hybrid approach enables flexible, language-driven interaction with photorealistic 3D reconstructions, opening new possibilities for semantic aerial inspection and scene understanding.", "date": "2025-05-23", "categories": []}
{"id": "2505.10049", "title": "Advances in Radiance Field for Dynamic Scene: From Neural Field to Gaussian Field", "abstract": "Dynamic scene representation and reconstruction have undergone transformative advances in recent years, catalyzed by breakthroughs in neural radiance fields and 3D Gaussian splatting techniques. While initially developed for static environments, these methodologies have rapidly evolved to address the complexities inherent in 4D dynamic scenes through an expansive body of research. Coupled with innovations in differentiable volumetric rendering, these approaches have significantly enhanced the quality of motion representation and dynamic scene reconstruction, thereby garnering substantial attention from the computer vision and graphics communities. This survey presents a systematic analysis of over 200 papers focused on dynamic scene representation using radiance field, spanning the spectrum from implicit neural representations to explicit Gaussian primitives. We categorize and evaluate these works through multiple critical lenses: motion representation paradigms, reconstruction techniques for varied scene dynamics, auxiliary information integration strategies, and regularization approaches that ensure temporal consistency and physical plausibility. We organize diverse methodological approaches under a unified representational framework, concluding with a critical examination of persistent challenges and promising research directions. By providing this comprehensive overview, we aim to establish a definitive reference for researchers entering this rapidly evolving field while offering experienced practitioners a systematic understanding of both conceptual principles and practical frontiers in dynamic scene reconstruction.", "date": "2025-05-15", "categories": []}
{"id": "2505.16535", "title": "SHaDe: Compact and Consistent Dynamic 3D Reconstruction via Tri-Plane Deformation and Latent Diffusion", "abstract": "We present a novel framework for dynamic 3D scene reconstruction that integrates three key components: an explicit tri-plane deformation field, a view-conditioned canonical radiance field with spherical harmonics (SH) attention, and a temporally-aware latent diffusion prior. Our method encodes 4D scenes using three orthogonal 2D feature planes that evolve over time, enabling efficient and compact spatiotemporal representation. These features are explicitly warped into a canonical space via a deformation offset field, eliminating the need for MLP-based motion modeling.\n  In canonical space, we replace traditional MLP decoders with a structured SH-based rendering head that synthesizes view-dependent color via attention over learned frequency bands improving both interpretability and rendering efficiency. To further enhance fidelity and temporal consistency, we introduce a transformer-guided latent diffusion module that refines the tri-plane and deformation features in a compressed latent space. This generative module denoises scene representations under ambiguous or out-of-distribution (OOD) motion, improving generalization.\n  Our model is trained in two stages: the diffusion module is first pre-trained independently, and then fine-tuned jointly with the full pipeline using a combination of image reconstruction, diffusion denoising, and temporal consistency losses. We demonstrate state-of-the-art results on synthetic benchmarks, surpassing recent methods such as HexPlane and 4D Gaussian Splatting in visual quality, temporal coherence, and robustness to sparse-view dynamic inputs.", "date": "2025-05-22", "categories": []}
{"id": "2412.17806", "title": "Reconstructing People, Places, and Cameras", "abstract": "We present \"Humans and Structure from Motion\" (HSfM), a method for jointly reconstructing multiple human meshes, scene point clouds, and camera parameters in a metric world coordinate system from a sparse set of uncalibrated multi-view images featuring people. Our approach combines data-driven scene reconstruction with the traditional Structure-from-Motion (SfM) framework to achieve more accurate scene reconstruction and camera estimation, while simultaneously recovering human meshes. In contrast to existing scene reconstruction and SfM methods that lack metric scale information, our method estimates approximate metric scale by leveraging a human statistical model. Furthermore, it reconstructs multiple human meshes within the same world coordinate system alongside the scene point cloud, effectively capturing spatial relationships among individuals and their positions in the environment. We initialize the reconstruction of humans, scenes, and cameras using robust foundational models and jointly optimize these elements. This joint optimization synergistically improves the accuracy of each component. We compare our method to existing approaches on two challenging benchmarks, EgoHumans and EgoExo4D, demonstrating significant improvements in human localization accuracy within the world coordinate frame (reducing error from 3.51m to 1.04m in EgoHumans and from 2.9m to 0.56m in EgoExo4D). Notably, our results show that incorporating human data into the SfM pipeline improves camera pose estimation (e.g., increasing RRA@15 by 20.3% on EgoHumans). Additionally, qualitative results show that our approach improves overall scene reconstruction quality. Our code is available at: https://github.com/hongsukchoi/HSfM_RELEASE", "date": "2024-12-23", "categories": []}
{"id": "2505.14747", "title": "LOD1 3D City Model from LiDAR: The Impact of Segmentation Accuracy on Quality of Urban 3D Modeling and Morphology Extraction", "abstract": "Three-dimensional reconstruction of buildings, particularly at Level of Detail 1 (LOD1), plays a crucial role in various applications such as urban planning, urban environmental studies, and designing optimized transportation networks. This study focuses on assessing the potential of LiDAR data for accurate 3D building reconstruction at LOD1 and extracting morphological features from these models. Four deep semantic segmentation models, U-Net, Attention U-Net, U-Net3+, and DeepLabV3+, were used, applying transfer learning to extract building footprints from LiDAR data. The results showed that U-Net3+ and Attention U-Net outperformed the others, achieving IoU scores of 0.833 and 0.814, respectively. Various statistical measures, including maximum, range, mode, median, and the 90th percentile, were used to estimate building heights, resulting in the generation of 3D models at LOD1. As the main contribution of the research, the impact of segmentation accuracy on the quality of 3D building modeling and the accuracy of morphological features like building area and external wall surface area was investigated. The results showed that the accuracy of building identification (segmentation performance) significantly affects the 3D model quality and the estimation of morphological features, depending on the height calculation method. Overall, the UNet3+ method, utilizing the 90th percentile and median measures, leads to accurate height estimation of buildings and the extraction of morphological features.", "date": "2025-05-20", "categories": []}
{"id": "2505.15185", "title": "MonoSplat: Generalizable 3D Gaussian Splatting from Monocular Depth Foundation Models", "abstract": "Recent advances in generalizable 3D Gaussian Splatting have demonstrated promising results in real-time high-fidelity rendering without per-scene optimization, yet existing approaches still struggle to handle unfamiliar visual content during inference on novel scenes due to limited generalizability. To address this challenge, we introduce MonoSplat, a novel framework that leverages rich visual priors from pre-trained monocular depth foundation models for robust Gaussian reconstruction. Our approach consists of two key components: a Mono-Multi Feature Adapter that transforms monocular features into multi-view representations, coupled with an Integrated Gaussian Prediction module that effectively fuses both feature types for precise Gaussian generation. Through the Adapter's lightweight attention mechanism, features are seamlessly aligned and aggregated across views while preserving valuable monocular priors, enabling the Prediction module to generate Gaussian primitives with accurate geometry and appearance. Through extensive experiments on diverse real-world datasets, we convincingly demonstrate that MonoSplat achieves superior reconstruction quality and generalization capability compared to existing methods while maintaining computational efficiency with minimal trainable parameters. Codes are available at https://github.com/CUHK-AIM-Group/MonoSplat.", "date": "2025-05-21", "categories": []}
{"id": "2505.15294", "title": "R3GS: Gaussian Splatting for Robust Reconstruction and Relocalization in Unconstrained Image Collections", "abstract": "We propose R3GS, a robust reconstruction and relocalization framework tailored for unconstrained datasets. Our method uses a hybrid representation during training. Each anchor combines a global feature from a convolutional neural network (CNN) with a local feature encoded by the multiresolution hash grids [2]. Subsequently, several shallow multi-layer perceptrons (MLPs) predict the attributes of each Gaussians, including color, opacity, and covariance. To mitigate the adverse effects of transient objects on the reconstruction process, we ffne-tune a lightweight human detection network. Once ffne-tuned, this network generates a visibility map that efffciently generalizes to other transient objects (such as posters, banners, and cars) with minimal need for further adaptation. Additionally, to address the challenges posed by sky regions in outdoor scenes, we propose an effective sky-handling technique that incorporates a depth prior as a constraint. This allows the inffnitely distant sky to be represented on the surface of a large-radius sky sphere, signiffcantly reducing ffoaters caused by errors in sky reconstruction. Furthermore, we introduce a novel relocalization method that remains robust to changes in lighting conditions while estimating the camera pose of a given image within the reconstructed 3DGS scene. As a result, R3GS significantly enhances rendering ffdelity, improves both training and rendering efffciency, and reduces storage requirements. Our method achieves state-of-the-art performance compared to baseline methods on in-the-wild datasets. The code will be made open-source following the acceptance of the paper.", "date": "2025-05-21", "categories": []}
{"id": "2505.15373", "title": "RAZER: Robust Accelerated Zero-Shot 3D Open-Vocabulary Panoptic Reconstruction with Spatio-Temporal Aggregation", "abstract": "Mapping and understanding complex 3D environments is fundamental to how autonomous systems perceive and interact with the physical world, requiring both precise geometric reconstruction and rich semantic comprehension. While existing 3D semantic mapping systems excel at reconstructing and identifying predefined object instances, they lack the flexibility to efficiently build semantic maps with open-vocabulary during online operation. Although recent vision-language models have enabled open-vocabulary object recognition in 2D images, they haven't yet bridged the gap to 3D spatial understanding. The critical challenge lies in developing a training-free unified system that can simultaneously construct accurate 3D maps while maintaining semantic consistency and supporting natural language interactions in real time. In this paper, we develop a zero-shot framework that seamlessly integrates GPU-accelerated geometric reconstruction with open-vocabulary vision-language models through online instance-level semantic embedding fusion, guided by hierarchical object association with spatial indexing. Our training-free system achieves superior performance through incremental processing and unified geometric-semantic updates, while robustly handling 2D segmentation inconsistencies. The proposed general-purpose 3D scene understanding framework can be used for various tasks including zero-shot 3D instance retrieval, segmentation, and object detection to reason about previously unseen objects and interpret natural language queries. The project page is available at https://razer-3d.github.io.", "date": "2025-05-21", "categories": []}
{"id": "2505.13839", "title": "MGStream: Motion-aware 3D Gaussian for Streamable Dynamic Scene Reconstruction", "abstract": "3D Gaussian Splatting (3DGS) has gained significant attention in streamable dynamic novel view synthesis (DNVS) for its photorealistic rendering capability and computational efficiency. Despite much progress in improving rendering quality and optimization strategies, 3DGS-based streamable dynamic scene reconstruction still suffers from flickering artifacts and storage inefficiency, and struggles to model the emerging objects. To tackle this, we introduce MGStream which employs the motion-related 3D Gaussians (3DGs) to reconstruct the dynamic and the vanilla 3DGs for the static. The motion-related 3DGs are implemented according to the motion mask and the clustering-based convex hull algorithm. The rigid deformation is applied to the motion-related 3DGs for modeling the dynamic, and the attention-based optimization on the motion-related 3DGs enables the reconstruction of the emerging objects. As the deformation and optimization are only conducted on the motion-related 3DGs, MGStream avoids flickering artifacts and improves the storage efficiency. Extensive experiments on real-world datasets N3DV and MeetRoom demonstrate that MGStream surpasses existing streaming 3DGS-based approaches in terms of rendering quality, training/storage efficiency and temporal consistency. Our code is available at: https://github.com/pcl3dv/MGStream.", "date": "2025-05-20", "categories": []}
{"id": "2505.14621", "title": "3D Reconstruction from Sketches", "abstract": "We consider the problem of reconstructing a 3D scene from multiple sketches. We propose a pipeline which involves (1) stitching together multiple sketches through use of correspondence points, (2) converting the stitched sketch into a realistic image using a CycleGAN, and (3) estimating that image's depth-map using a pre-trained convolutional neural network based architecture called MegaDepth. Our contribution includes constructing a dataset of image-sketch pairs, the images for which are from the Zurich Building Database, and sketches have been generated by us. We use this dataset to train a CycleGAN for our pipeline's second step. We end up with a stitching process that does not generalize well to real drawings, but the rest of the pipeline that creates a 3D reconstruction from a single sketch performs quite well on a wide variety of drawings.", "date": "2025-05-20", "categories": []}
{"id": "2505.13215", "title": "Hybrid 3D-4D Gaussian Splatting for Fast Dynamic Scene Representation", "abstract": "Recent advancements in dynamic 3D scene reconstruction have shown promising results, enabling high-fidelity 3D novel view synthesis with improved temporal consistency. Among these, 4D Gaussian Splatting (4DGS) has emerged as an appealing approach due to its ability to model high-fidelity spatial and temporal variations. However, existing methods suffer from substantial computational and memory overhead due to the redundant allocation of 4D Gaussians to static regions, which can also degrade image quality. In this work, we introduce hybrid 3D-4D Gaussian Splatting (3D-4DGS), a novel framework that adaptively represents static regions with 3D Gaussians while reserving 4D Gaussians for dynamic elements. Our method begins with a fully 4D Gaussian representation and iteratively converts temporally invariant Gaussians into 3D, significantly reducing the number of parameters and improving computational efficiency. Meanwhile, dynamic Gaussians retain their full 4D representation, capturing complex motions with high fidelity. Our approach achieves significantly faster training times compared to baseline 4D Gaussian Splatting methods while maintaining or improving the visual quality.", "date": "2025-05-19", "categories": []}
{"id": "2505.05848", "title": "RefRef: A Synthetic Dataset and Benchmark for Reconstructing Refractive and Reflective Objects", "abstract": "Modern 3D reconstruction and novel view synthesis approaches have demonstrated strong performance on scenes with opaque Lambertian objects. However, most assume straight light paths and therefore cannot properly handle refractive and reflective materials. Moreover, datasets specialized for these effects are limited, stymieing efforts to evaluate performance and develop suitable techniques. In this work, we introduce a synthetic RefRef dataset and benchmark for reconstructing scenes with refractive and reflective objects from posed images. Our dataset has 50 such objects of varying complexity, from single-material convex shapes to multi-material non-convex shapes, each placed in three different background types, resulting in 150 scenes. We also propose an oracle method that, given the object geometry and refractive indices, calculates accurate light paths for neural rendering, and an approach based on this that avoids these assumptions. We benchmark these against several state-of-the-art methods and show that all methods lag significantly behind the oracle, highlighting the challenges of the task and dataset.", "date": "2025-05-09", "categories": []}
{"id": "2505.10578", "title": "ExploreGS: a vision-based low overhead framework for 3D scene reconstruction", "abstract": "This paper proposes a low-overhead, vision-based 3D scene reconstruction framework for drones, named ExploreGS. By using RGB images, ExploreGS replaces traditional lidar-based point cloud acquisition process with a vision model, achieving a high-quality reconstruction at a lower cost. The framework integrates scene exploration and model reconstruction, and leverags a Bag-of-Words(BoW) model to enable real-time processing capabilities, therefore, the 3D Gaussian Splatting (3DGS) training can be executed on-board. Comprehensive experiments in both simulation and real-world environments demonstrate the efficiency and applicability of the ExploreGS framework on resource-constrained devices, while maintaining reconstruction quality comparable to state-of-the-art methods.", "date": "2025-05-14", "categories": []}
{"id": "2505.10787", "title": "EA-3DGS: Efficient and Adaptive 3D Gaussians with Highly Enhanced Quality for outdoor scenes", "abstract": "Efficient scene representations are essential for many real-world applications, especially those involving spatial measurement. Although current NeRF-based methods have achieved impressive results in reconstructing building-scale scenes, they still suffer from slow training and inference speeds due to time-consuming stochastic sampling. Recently, 3D Gaussian Splatting (3DGS) has demonstrated excellent performance with its high-quality rendering and real-time speed, especially for objects and small-scale scenes. However, in outdoor scenes, its point-based explicit representation lacks an effective adjustment mechanism, and the millions of Gaussian points required often lead to memory constraints during training. To address these challenges, we propose EA-3DGS, a high-quality real-time rendering method designed for outdoor scenes. First, we introduce a mesh structure to regulate the initialization of Gaussian components by leveraging an adaptive tetrahedral mesh that partitions the grid and initializes Gaussian components on each face, effectively capturing geometric structures in low-texture regions. Second, we propose an efficient Gaussian pruning strategy that evaluates each 3D Gaussian's contribution to the view and prunes accordingly. To retain geometry-critical Gaussian points, we also present a structure-aware densification strategy that densifies Gaussian points in low-curvature regions. Additionally, we employ vector quantization for parameter quantization of Gaussian components, significantly reducing disk space requirements with only a minimal impact on rendering quality. Extensive experiments on 13 scenes, including eight from four public datasets (MatrixCity-Aerial, Mill-19, Tanks \\& Temples, WHU) and five self-collected scenes acquired through UAV photogrammetry measurement from SCUT-CA and plateau regions, further demonstrate the superiority of our method.", "date": "2025-05-16", "categories": []}
{"id": "2403.07547", "title": "SMURF: Continuous Dynamics for Motion-Deblurring Radiance Fields", "abstract": "Neural radiance fields (NeRF) has attracted considerable attention for their exceptional ability in synthesizing novel views with high fidelity. However, the presence of motion blur, resulting from slight camera movements during extended shutter exposures, poses a significant challenge, potentially compromising the quality of the reconstructed 3D scenes. To effectively handle this issue, we propose sequential motion understanding radiance fields (SMURF), a novel approach that models continuous camera motion and leverages the explicit volumetric representation method for robustness to motion-blurred input images. The core idea of the SMURF is continuous motion blurring kernel (CMBK), a module designed to model a continuous camera movements for processing blurry inputs. Our model is evaluated against benchmark datasets and demonstrates state-of-the-art performance both quantitatively and qualitatively.", "date": "2024-03-12", "categories": []}
{"id": "2412.16028", "title": "CoCoGaussian: Leveraging Circle of Confusion for Gaussian Splatting from Defocused Images", "abstract": "3D Gaussian Splatting (3DGS) has attracted significant attention for its high-quality novel view rendering, inspiring research to address real-world challenges. While conventional methods depend on sharp images for accurate scene reconstruction, real-world scenarios are often affected by defocus blur due to finite depth of field, making it essential to account for realistic 3D scene representation. In this study, we propose CoCoGaussian, a Circle of Confusion-aware Gaussian Splatting that enables precise 3D scene representation using only defocused images. CoCoGaussian addresses the challenge of defocus blur by modeling the Circle of Confusion (CoC) through a physically grounded approach based on the principles of photographic defocus. Exploiting 3D Gaussians, we compute the CoC diameter from depth and learnable aperture information, generating multiple Gaussians to precisely capture the CoC shape. Furthermore, we introduce a learnable scaling factor to enhance robustness and provide more flexibility in handling unreliable depth in scenes with reflective or refractive surfaces. Experiments on both synthetic and real-world datasets demonstrate that CoCoGaussian achieves state-of-the-art performance across multiple benchmarks.", "date": "2024-12-20", "categories": []}
{"id": "2505.07887", "title": "Monocular Online Reconstruction with Enhanced Detail Preservation", "abstract": "We propose an online 3D Gaussian-based dense mapping framework for photorealistic details reconstruction from a monocular image stream. Our approach addresses two key challenges in monocular online reconstruction: distributing Gaussians without relying on depth maps and ensuring both local and global consistency in the reconstructed maps. To achieve this, we introduce two key modules: the Hierarchical Gaussian Management Module for effective Gaussian distribution and the Global Consistency Optimization Module for maintaining alignment and coherence at all scales. In addition, we present the Multi-level Occupancy Hash Voxels (MOHV), a structure that regularizes Gaussians for capturing details across multiple levels of granularity. MOHV ensures accurate reconstruction of both fine and coarse geometries and textures, preserving intricate details while maintaining overall structural integrity. Compared to state-of-the-art RGB-only and even RGB-D methods, our framework achieves superior reconstruction quality with high computational efficiency. Moreover, it integrates seamlessly with various tracking systems, ensuring generality and scalability.", "date": "2025-05-11", "categories": []}
{"id": "2505.09562", "title": "Camera-Only 3D Panoptic Scene Completion for Autonomous Driving through Differentiable Object Shapes", "abstract": "Autonomous vehicles need a complete map of their surroundings to plan and act. This has sparked research into the tasks of 3D occupancy prediction, 3D scene completion, and 3D panoptic scene completion, which predict a dense map of the ego vehicle's surroundings as a voxel grid. Scene completion extends occupancy prediction by predicting occluded regions of the voxel grid, and panoptic scene completion further extends this task by also distinguishing object instances within the same class; both aspects are crucial for path planning and decision-making. However, 3D panoptic scene completion is currently underexplored. This work introduces a novel framework for 3D panoptic scene completion that extends existing 3D semantic scene completion models. We propose an Object Module and Panoptic Module that can easily be integrated with 3D occupancy and scene completion methods presented in the literature. Our approach leverages the available annotations in occupancy benchmarks, allowing individual object shapes to be learned as a differentiable problem. The code is available at https://github.com/nicolamarinello/OffsetOcc .", "date": "2025-05-14", "categories": []}
{"id": "2502.12894", "title": "CAST: Component-Aligned 3D Scene Reconstruction from an RGB Image", "abstract": "Recovering high-quality 3D scenes from a single RGB image is a challenging task in computer graphics. Current methods often struggle with domain-specific limitations or low-quality object generation. To address these, we propose CAST (Component-Aligned 3D Scene Reconstruction from a Single RGB Image), a novel method for 3D scene reconstruction and recovery. CAST starts by extracting object-level 2D segmentation and relative depth information from the input image, followed by using a GPT-based model to analyze inter-object spatial relationships. This enables the understanding of how objects relate to each other within the scene, ensuring more coherent reconstruction. CAST then employs an occlusion-aware large-scale 3D generation model to independently generate each object's full geometry, using MAE and point cloud conditioning to mitigate the effects of occlusions and partial object information, ensuring accurate alignment with the source image's geometry and texture. To align each object with the scene, the alignment generation model computes the necessary transformations, allowing the generated meshes to be accurately placed and integrated into the scene's point cloud. Finally, CAST incorporates a physics-aware correction step that leverages a fine-grained relation graph to generate a constraint graph. This graph guides the optimization of object poses, ensuring physical consistency and spatial coherence. By utilizing Signed Distance Fields (SDF), the model effectively addresses issues such as occlusions, object penetration, and floating objects, ensuring that the generated scene accurately reflects real-world physical interactions. CAST can be leveraged in robotics, enabling efficient real-to-simulation workflows and providing realistic, scalable simulation environments for robotic systems.", "date": "2025-02-18", "categories": []}
{"id": "2504.21650", "title": "HoloTime: Taming Video Diffusion Models for Panoramic 4D Scene Generation", "abstract": "The rapid advancement of diffusion models holds the promise of revolutionizing the application of VR and AR technologies, which typically require scene-level 4D assets for user experience. Nonetheless, existing diffusion models predominantly concentrate on modeling static 3D scenes or object-level dynamics, constraining their capacity to provide truly immersive experiences. To address this issue, we propose HoloTime, a framework that integrates video diffusion models to generate panoramic videos from a single prompt or reference image, along with a 360-degree 4D scene reconstruction method that seamlessly transforms the generated panoramic video into 4D assets, enabling a fully immersive 4D experience for users. Specifically, to tame video diffusion models for generating high-fidelity panoramic videos, we introduce the 360World dataset, the first comprehensive collection of panoramic videos suitable for downstream 4D scene reconstruction tasks. With this curated dataset, we propose Panoramic Animator, a two-stage image-to-video diffusion model that can convert panoramic images into high-quality panoramic videos. Following this, we present Panoramic Space-Time Reconstruction, which leverages a space-time depth estimation method to transform the generated panoramic videos into 4D point clouds, enabling the optimization of a holistic 4D Gaussian Splatting representation to reconstruct spatially and temporally consistent 4D scenes. To validate the efficacy of our method, we conducted a comparative analysis with existing approaches, revealing its superiority in both panoramic video generation and 4D scene reconstruction. This demonstrates our method's capability to create more engaging and realistic immersive environments, thereby enhancing user experiences in VR and AR applications.", "date": "2025-04-30", "categories": []}
{"id": "2505.08196", "title": "ADC-GS: Anchor-Driven Deformable and Compressed Gaussian Splatting for Dynamic Scene Reconstruction", "abstract": "Existing 4D Gaussian Splatting methods rely on per-Gaussian deformation from a canonical space to target frames, which overlooks redundancy among adjacent Gaussian primitives and results in suboptimal performance. To address this limitation, we propose Anchor-Driven Deformable and Compressed Gaussian Splatting (ADC-GS), a compact and efficient representation for dynamic scene reconstruction. Specifically, ADC-GS organizes Gaussian primitives into an anchor-based structure within the canonical space, enhanced by a temporal significance-based anchor refinement strategy. To reduce deformation redundancy, ADC-GS introduces a hierarchical coarse-to-fine pipeline that captures motions at varying granularities. Moreover, a rate-distortion optimization is adopted to achieve an optimal balance between bitrate consumption and representation fidelity. Experimental results demonstrate that ADC-GS outperforms the per-Gaussian deformation approaches in rendering speed by 300%-800% while achieving state-of-the-art storage efficiency without compromising rendering quality. The code is released at https://github.com/H-Huang774/ADC-GS.git.", "date": "2025-05-13", "categories": []}
{"id": "2407.01330", "title": "A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions", "abstract": "Unsigned distance fields (UDFs) provide a versatile framework for representing a diverse array of 3D shapes, encompassing both watertight and non-watertight geometries. Traditional UDF learning methods typically require extensive training on large 3D shape datasets, which is costly and necessitates re-training for new datasets. This paper presents a novel neural framework, LoSF-UDF, for reconstructing surfaces from 3D point clouds by leveraging local shape functions to learn UDFs. We observe that 3D shapes manifest simple patterns in localized regions, prompting us to develop a training dataset of point cloud patches characterized by mathematical functions that represent a continuum from smooth surfaces to sharp edges and corners. Our approach learns features within a specific radius around each query point and utilizes an attention mechanism to focus on the crucial features for UDF estimation. Despite being highly lightweight, with only 653 KB of trainable parameters and a modest-sized training dataset with 0.5 GB storage, our method enables efficient and robust surface reconstruction from point clouds without requiring for shape-specific training. Furthermore, our method exhibits enhanced resilience to noise and outliers in point clouds compared to existing methods. We conduct comprehensive experiments and comparisons across various datasets, including synthetic and real-scanned point clouds, to validate our method's efficacy. Notably, our lightweight framework offers rapid and reliable initialization for other unsupervised iterative approaches, improving both the efficiency and accuracy of their reconstructions. Our project and code are available at https://jbhu67.github.io/LoSF-UDF.github.io.", "date": "2024-07-01", "categories": []}
{"id": "2505.07373", "title": "Geometric Prior-Guided Neural Implicit Surface Reconstruction in the Wild", "abstract": "Neural implicit surface reconstruction using volume rendering techniques has recently achieved significant advancements in creating high-fidelity surfaces from multiple 2D images. However, current methods primarily target scenes with consistent illumination and struggle to accurately reconstruct 3D geometry in uncontrolled environments with transient occlusions or varying appearances. While some neural radiance field (NeRF)-based variants can better manage photometric variations and transient objects in complex scenes, they are designed for novel view synthesis rather than precise surface reconstruction due to limited surface constraints. To overcome this limitation, we introduce a novel approach that applies multiple geometric constraints to the implicit surface optimization process, enabling more accurate reconstructions from unconstrained image collections. First, we utilize sparse 3D points from structure-from-motion (SfM) to refine the signed distance function estimation for the reconstructed surface, with a displacement compensation to accommodate noise in the sparse points. Additionally, we employ robust normal priors derived from a normal predictor, enhanced by edge prior filtering and multi-view consistency constraints, to improve alignment with the actual surface geometry. Extensive testing on the Heritage-Recon benchmark and other datasets has shown that the proposed method can accurately reconstruct surfaces from in-the-wild images, yielding geometries with superior accuracy and granularity compared to existing techniques. Our approach enables high-quality 3D reconstruction of various landmarks, making it applicable to diverse scenarios such as digital preservation of cultural heritage sites.", "date": "2025-05-12", "categories": []}
{"id": "2410.03825", "title": "MonST3R: A Simple Approach for Estimating Geometry in the Presence of\n  Motion", "abstract": "  Estimating geometry from dynamic scenes, where objects move and deform over\ntime, remains a core challenge in computer vision. Current approaches often\nrely on multi-stage pipelines or global optimizations that decompose the\nproblem into subtasks, like depth and flow, leading to complex systems prone to\nerrors. In this paper, we present Motion DUSt3R (MonST3R), a novel\ngeometry-first approach that directly estimates per-timestep geometry from\ndynamic scenes. Our key insight is that by simply estimating a pointmap for\neach timestep, we can effectively adapt DUST3R's representation, previously\nonly used for static scenes, to dynamic scenes. However, this approach presents\na significant challenge: the scarcity of suitable training data, namely\ndynamic, posed videos with depth labels. Despite this, we show that by posing\nthe problem as a fine-tuning task, identifying several suitable datasets, and\nstrategically training the model on this limited data, we can surprisingly\nenable the model to handle dynamics, even without an explicit motion\nrepresentation. Based on this, we introduce new optimizations for several\ndownstream video-specific tasks and demonstrate strong performance on video\ndepth and camera pose estimation, outperforming prior work in terms of\nrobustness and efficiency. Moreover, MonST3R shows promising results for\nprimarily feed-forward 4D reconstruction.\n", "date": "2024-10-04", "categories": []}
{"id": "2505.05356", "title": "Time of the Flight of the Gaussians: Optimizing Depth Indirectly in\n  Dynamic Radiance Fields", "abstract": "  We present a method to reconstruct dynamic scenes from monocular\ncontinuous-wave time-of-flight (C-ToF) cameras using raw sensor samples that\nachieves similar or better accuracy than neural volumetric approaches and is\n100x faster. Quickly achieving high-fidelity dynamic 3D reconstruction from a\nsingle viewpoint is a significant challenge in computer vision. In C-ToF\nradiance field reconstruction, the property of interest-depth-is not directly\nmeasured, causing an additional challenge. This problem has a large and\nunderappreciated impact upon the optimization when using a fast primitive-based\nscene representation like 3D Gaussian splatting, which is commonly used with\nmulti-view data to produce satisfactory results and is brittle in its\noptimization otherwise. We incorporate two heuristics into the optimization to\nimprove the accuracy of scene geometry represented by Gaussians. Experimental\nresults show that our approach produces accurate reconstructions under\nconstrained C-ToF sensing conditions, including for fast motions like swinging\nbaseball bats. https://visual.cs.brown.edu/gftorf\n", "date": "2025-05-08", "categories": []}
{"id": "2406.02720", "title": "3D-HGS: 3D Half-Gaussian Splatting", "abstract": "  Photo-realistic image rendering from 3D scene reconstruction has advanced\nsignificantly with neural rendering techniques. Among these, 3D Gaussian\nSplatting (3D-GS) outperforms Neural Radiance Fields (NeRFs) in quality and\nspeed but struggles with shape and color discontinuities. We propose 3D\nHalf-Gaussian (3D-HGS) kernels as a plug-and-play solution to address these\nlimitations. Our experiments show that 3D-HGS enhances existing 3D-GS methods,\nachieving state-of-the-art rendering quality without compromising speed.\n", "date": "2024-06-04", "categories": []}
{"id": "2406.13896", "title": "SMORE: Simultaneous Map and Object REconstruction", "abstract": "  We present a method for dynamic surface reconstruction of large-scale urban\nscenes from LiDAR. Depth-based reconstructions tend to focus on small-scale\nobjects or large-scale SLAM reconstructions that treat moving objects as\noutliers. We take a holistic perspective and optimize a compositional model of\na dynamic scene that decomposes the world into rigidly-moving objects and the\nbackground. To achieve this, we take inspiration from recent novel view\nsynthesis methods and frame the reconstruction problem as a global optimization\nover neural surfaces, ego poses, and object poses, which minimizes the error\nbetween composed spacetime surfaces and input LiDAR scans. In contrast to view\nsynthesis methods, which typically minimize 2D errors with gradient descent, we\nminimize a 3D point-to-surface error by coordinate descent, which we decompose\ninto registration and surface reconstruction steps. Each step can be handled\nwell by off-the-shelf methods without any re-training. We analyze the surface\nreconstruction step for rolling-shutter LiDARs, and show that deskewing\noperations common in continuous time SLAM can be applied to dynamic objects as\nwell, improving results over prior art by an order of magnitude. Beyond\npursuing dynamic reconstruction as a goal in and of itself, we propose that\nsuch a system can be used to auto-label partially annotated sequences and\nproduce ground truth annotation for hard-to-label problems such as depth\ncompletion and scene flow. Please see https://anishmadan23.github.io/smore/ for\nmore visual results.\n", "date": "2024-06-19", "categories": []}
{"id": "2501.18630", "title": "Deformable Beta Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has advanced radiance field reconstruction by\nenabling real-time rendering. However, its reliance on Gaussian kernels for\ngeometry and low-order Spherical Harmonics (SH) for color encoding limits its\nability to capture complex geometries and diverse colors. We introduce\nDeformable Beta Splatting (DBS), a deformable and compact approach that\nenhances both geometry and color representation. DBS replaces Gaussian kernels\nwith deformable Beta Kernels, which offer bounded support and adaptive\nfrequency control to capture fine geometric details with higher fidelity while\nachieving better memory efficiency. In addition, we extended the Beta Kernel to\ncolor encoding, which facilitates improved representation of diffuse and\nspecular components, yielding superior results compared to SH-based methods.\nFurthermore, Unlike prior densification techniques that depend on Gaussian\nproperties, we mathematically prove that adjusting regularized opacity alone\nensures distribution-preserved Markov chain Monte Carlo (MCMC), independent of\nthe splatting kernel type. Experimental results demonstrate that DBS achieves\nstate-of-the-art visual quality while utilizing only 45% of the parameters and\nrendering 1.5x faster than 3DGS-MCMC, highlighting the superior performance of\nDBS for real-time radiance field rendering. Interactive demonstrations and\nsource code are available on our project website:\nhttps://rongliu-leo.github.io/beta-splatting/.\n", "date": "2025-01-27", "categories": []}
{"id": "2505.03362", "title": "3D Surface Reconstruction with Enhanced High-Frequency Details", "abstract": "  Neural implicit 3D reconstruction can reproduce shapes without 3D\nsupervision, and it learns the 3D scene through volume rendering methods and\nneural implicit representations. Current neural surface reconstruction methods\ntend to randomly sample the entire image, making it difficult to learn\nhigh-frequency details on the surface, and thus the reconstruction results tend\nto be too smooth. We designed a method (FreNeuS) based on high-frequency\ninformation to solve the problem of insufficient surface detail. Specifically,\nFreNeuS uses pixel gradient changes to easily acquire high-frequency regions in\nan image and uses the obtained high-frequency information to guide surface\ndetail reconstruction. High-frequency information is first used to guide the\ndynamic sampling of rays, applying different sampling strategies according to\nvariations in high-frequency regions. To further enhance the focus on surface\ndetails, we have designed a high-frequency weighting method that constrains the\nrepresentation of high-frequency details during the reconstruction process.\nQualitative and quantitative results show that our method can reconstruct fine\nsurface details and obtain better surface reconstruction quality compared to\nexisting methods. In addition, our method is more applicable and can be\ngeneralized to any NeuS-based work.\n", "date": "2025-05-06", "categories": []}
{"id": "2411.02979", "title": "CAD-NeRF: Learning NeRFs from Uncalibrated Few-view Images by CAD Model\n  Retrieval", "abstract": "  Reconstructing from multi-view images is a longstanding problem in 3D vision,\nwhere neural radiance fields (NeRFs) have shown great potential and get\nrealistic rendered images of novel views. Currently, most NeRF methods either\nrequire accurate camera poses or a large number of input images, or even both.\nReconstructing NeRF from few-view images without poses is challenging and\nhighly ill-posed. To address this problem, we propose CAD-NeRF, a method\nreconstructed from less than 10 images without any known poses. Specifically,\nwe build a mini library of several CAD models from ShapeNet and render them\nfrom many random views. Given sparse-view input images, we run a model and pose\nretrieval from the library, to get a model with similar shapes, serving as the\ndensity supervision and pose initializations. Here we propose a multi-view pose\nretrieval method to avoid pose conflicts among views, which is a new and unseen\nproblem in uncalibrated NeRF methods. Then, the geometry of the object is\ntrained by the CAD guidance. The deformation of the density field and camera\nposes are optimized jointly. Then texture and density are trained and\nfine-tuned as well. All training phases are in self-supervised manners.\nComprehensive evaluations of synthetic and real images show that CAD-NeRF\nsuccessfully learns accurate densities with a large deformation from retrieved\nCAD models, showing the generalization abilities.\n", "date": "2024-11-05", "categories": []}
{"id": "2503.06222", "title": "Vision-based 3D Semantic Scene Completion via Capture Dynamic\n  Representations", "abstract": "  The vision-based semantic scene completion task aims to predict dense\ngeometric and semantic 3D scene representations from 2D images. However, the\npresence of dynamic objects in the scene seriously affects the accuracy of the\nmodel inferring 3D structures from 2D images. Existing methods simply stack\nmultiple frames of image input to increase dense scene semantic information,\nbut ignore the fact that dynamic objects and non-texture areas violate\nmulti-view consistency and matching reliability. To address these issues, we\npropose a novel method, CDScene: Vision-based Robust Semantic Scene Completion\nvia Capturing Dynamic Representations. First, we leverage a multimodal\nlarge-scale model to extract 2D explicit semantics and align them into 3D\nspace. Second, we exploit the characteristics of monocular and stereo depth to\ndecouple scene information into dynamic and static features. The dynamic\nfeatures contain structural relationships around dynamic objects, and the\nstatic features contain dense contextual spatial information. Finally, we\ndesign a dynamic-static adaptive fusion module to effectively extract and\naggregate complementary features, achieving robust and accurate semantic scene\ncompletion in autonomous driving scenarios. Extensive experimental results on\nthe SemanticKITTI, SSCBench-KITTI360, and SemanticKITTI-C datasets demonstrate\nthe superiority and robustness of CDScene over existing state-of-the-art\nmethods.\n", "date": "2025-03-08", "categories": []}
{"id": "2505.01799", "title": "AquaGS: Fast Underwater Scene Reconstruction with SfM-Free Gaussian\n  Splatting", "abstract": "  Underwater scene reconstruction is a critical tech-nology for underwater\noperations, enabling the generation of 3D models from images captured by\nunderwater platforms. However, the quality of underwater images is often\ndegraded due to medium interference, which limits the effectiveness of\nStructure-from-Motion (SfM) pose estimation, leading to subsequent\nreconstruction failures. Additionally, SfM methods typically operate at slower\nspeeds, further hindering their applicability in real-time scenarios. In this\npaper, we introduce AquaGS, an SfM-free underwater scene reconstruction model\nbased on the SeaThru algorithm, which facilitates rapid and accurate separation\nof scene details and medium features. Our approach initializes Gaussians by\nintegrating state-of-the-art multi-view stereo (MVS) technology, employs\nimplicit Neural Radiance Fields (NeRF) for rendering translucent media and\nutilizes the latest explicit 3D Gaussian Splatting (3DGS) technique to render\nobject surfaces, which effectively addresses the limitations of traditional\nmethods and accurately simulates underwater optical phenomena. Experimental\nresults on the data set and the robot platform show that our model can complete\nhigh-precision reconstruction in 30 seconds with only 3 image inputs,\nsignificantly enhancing the practical application of the algorithm in robotic\nplatforms.\n", "date": "2025-05-03", "categories": []}
{"id": "2505.02175", "title": "SparSplat: Fast Multi-View Reconstruction with Generalizable 2D Gaussian\n  Splatting", "abstract": "  Recovering 3D information from scenes via multi-view stereo reconstruction\n(MVS) and novel view synthesis (NVS) is inherently challenging, particularly in\nscenarios involving sparse-view setups. The advent of 3D Gaussian Splatting\n(3DGS) enabled real-time, photorealistic NVS. Following this, 2D Gaussian\nSplatting (2DGS) leveraged perspective accurate 2D Gaussian primitive\nrasterization to achieve accurate geometry representation during rendering,\nimproving 3D scene reconstruction while maintaining real-time performance.\nRecent approaches have tackled the problem of sparse real-time NVS using 3DGS\nwithin a generalizable, MVS-based learning framework to regress 3D Gaussian\nparameters. Our work extends this line of research by addressing the challenge\nof generalizable sparse 3D reconstruction and NVS jointly, and manages to\nperform successfully at both tasks. We propose an MVS-based learning pipeline\nthat regresses 2DGS surface element parameters in a feed-forward fashion to\nperform 3D shape reconstruction and NVS from sparse-view images. We further\nshow that our generalizable pipeline can benefit from preexisting foundational\nmulti-view deep visual features. The resulting model attains the\nstate-of-the-art results on the DTU sparse 3D reconstruction benchmark in terms\nof Chamfer distance to ground-truth, as-well as state-of-the-art NVS. It also\ndemonstrates strong generalization on the BlendedMVS and Tanks and Temples\ndatasets. We note that our model outperforms the prior state-of-the-art in\nfeed-forward sparse view reconstruction based on volume rendering of implicit\nrepresentations, while offering an almost 2 orders of magnitude higher\ninference speed.\n", "date": "2025-05-04", "categories": []}
{"id": "2505.00734", "title": "Unconstrained Large-scale 3D Reconstruction and Rendering across\n  Altitudes", "abstract": "  Production of photorealistic, navigable 3D site models requires a large\nvolume of carefully collected images that are often unavailable to first\nresponders for disaster relief or law enforcement. Real-world challenges\ninclude limited numbers of images, heterogeneous unposed cameras, inconsistent\nlighting, and extreme viewpoint differences for images collected from varying\naltitudes. To promote research aimed at addressing these challenges, we have\ndeveloped the first public benchmark dataset for 3D reconstruction and novel\nview synthesis based on multiple calibrated ground-level, security-level, and\nairborne cameras. We present datasets that pose real-world challenges,\nindependently evaluate calibration of unposed cameras and quality of novel\nrendered views, demonstrate baseline performance using recent state-of-practice\nmethods, and identify challenges for further research.\n", "date": "2025-04-29", "categories": []}
{"id": "2405.18416", "title": "3D StreetUnveiler with Semantic-aware 2DGS -- a simple baseline", "abstract": "  Unveiling an empty street from crowded observations captured by in-car\ncameras is crucial for autonomous driving. However, removing all temporarily\nstatic objects, such as stopped vehicles and standing pedestrians, presents a\nsignificant challenge. Unlike object-centric 3D inpainting, which relies on\nthorough observation in a small scene, street scene cases involve long\ntrajectories that differ from previous 3D inpainting tasks. The camera-centric\nmoving environment of captured videos further complicates the task due to the\nlimited degree and time duration of object observation. To address these\nobstacles, we introduce StreetUnveiler to reconstruct an empty street.\nStreetUnveiler learns a 3D representation of the empty street from crowded\nobservations. Our representation is based on the hard-label semantic 2D\nGaussian Splatting (2DGS) for its scalability and ability to identify Gaussians\nto be removed. We inpaint rendered image after removing unwanted Gaussians to\nprovide pseudo-labels and subsequently re-optimize the 2DGS. Given its temporal\ncontinuous movement, we divide the empty street scene into observed,\npartial-observed, and unobserved regions, which we propose to locate through a\nrendered alpha map. This decomposition helps us to minimize the regions that\nneed to be inpainted. To enhance the temporal consistency of the inpainting, we\nintroduce a novel time-reversal framework to inpaint frames in reverse order\nand use later frames as references for earlier frames to fully utilize the\nlong-trajectory observations. Our experiments conducted on the street scene\ndataset successfully reconstructed a 3D representation of the empty street. The\nmesh representation of the empty street can be extracted for further\napplications. The project page and more visualizations can be found at:\nhttps://streetunveiler.github.io\n", "date": "2024-05-28", "categories": []}
{"id": "2412.09621", "title": "Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos", "abstract": "  Learning to understand dynamic 3D scenes from imagery is crucial for\napplications ranging from robotics to scene reconstruction. Yet, unlike other\nproblems where large-scale supervised training has enabled rapid progress,\ndirectly supervising methods for recovering 3D motion remains challenging due\nto the fundamental difficulty of obtaining ground truth annotations. We present\na system for mining high-quality 4D reconstructions from internet stereoscopic,\nwide-angle videos. Our system fuses and filters the outputs of camera pose\nestimation, stereo depth estimation, and temporal tracking methods into\nhigh-quality dynamic 3D reconstructions. We use this method to generate\nlarge-scale data in the form of world-consistent, pseudo-metric 3D point clouds\nwith long-term motion trajectories. We demonstrate the utility of this data by\ntraining a variant of DUSt3R to predict structure and 3D motion from real-world\nimage pairs, showing that training on our reconstructed data enables\ngeneralization to diverse real-world scenes. Project page and data at:\nhttps://stereo4d.github.io\n", "date": "2024-12-12", "categories": []}
{"id": "2503.13435", "title": "WideRange4D: Enabling High-Quality 4D Reconstruction with Wide-Range\n  Movements and Scenes", "abstract": "  With the rapid development of 3D reconstruction technology, research in 4D\nreconstruction is also advancing, existing 4D reconstruction methods can\ngenerate high-quality 4D scenes. However, due to the challenges in acquiring\nmulti-view video data, the current 4D reconstruction benchmarks mainly display\nactions performed in place, such as dancing, within limited scenarios. In\npractical scenarios, many scenes involve wide-range spatial movements,\nhighlighting the limitations of existing 4D reconstruction datasets.\nAdditionally, existing 4D reconstruction methods rely on deformation fields to\nestimate the dynamics of 3D objects, but deformation fields struggle with\nwide-range spatial movements, which limits the ability to achieve high-quality\n4D scene reconstruction with wide-range spatial movements. In this paper, we\nfocus on 4D scene reconstruction with significant object spatial movements and\npropose a novel 4D reconstruction benchmark, WideRange4D. This benchmark\nincludes rich 4D scene data with large spatial variations, allowing for a more\ncomprehensive evaluation of the generation capabilities of 4D generation\nmethods. Furthermore, we introduce a new 4D reconstruction method, Progress4D,\nwhich generates stable and high-quality 4D results across various complex 4D\nscene reconstruction tasks. We conduct both quantitative and qualitative\ncomparison experiments on WideRange4D, showing that our Progress4D outperforms\nexisting state-of-the-art 4D reconstruction methods. Project:\nhttps://github.com/Gen-Verse/WideRange4D\n", "date": "2025-03-17", "categories": []}
{"id": "2504.21067", "title": "GauSS-MI: Gaussian Splatting Shannon Mutual Information for Active 3D\n  Reconstruction", "abstract": "  This research tackles the challenge of real-time active view selection and\nuncertainty quantification on visual quality for active 3D reconstruction.\nVisual quality is a critical aspect of 3D reconstruction. Recent advancements\nsuch as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have\nnotably enhanced the image rendering quality of reconstruction models.\nNonetheless, the efficient and effective acquisition of input images for\nreconstruction-specifically, the selection of the most informative\nviewpoint-remains an open challenge, which is crucial for active\nreconstruction. Existing studies have primarily focused on evaluating geometric\ncompleteness and exploring unobserved or unknown regions, without direct\nevaluation of the visual uncertainty within the reconstruction model. To\naddress this gap, this paper introduces a probabilistic model that quantifies\nvisual uncertainty for each Gaussian. Leveraging Shannon Mutual Information, we\nformulate a criterion, Gaussian Splatting Shannon Mutual Information\n(GauSS-MI), for real-time assessment of visual mutual information from novel\nviewpoints, facilitating the selection of next best view. GauSS-MI is\nimplemented within an active reconstruction system integrated with a view and\nmotion planner. Extensive experiments across various simulated and real-world\nscenes showcase the superior visual quality and reconstruction efficiency\nperformance of the proposed system.\n", "date": "2025-04-29", "categories": []}
{"id": "2501.18627", "title": "Radiance Surfaces: Optimizing Surface Representations with a 5D Radiance\n  Field Loss", "abstract": "  We present a fast and simple technique to convert images into a radiance\nsurface-based scene representation. Building on existing radiance volume\nreconstruction algorithms, we introduce a subtle yet impactful modification of\nthe loss function requiring changes to only a few lines of code: instead of\nintegrating the radiance field along rays and supervising the resulting images,\nwe project the training images into the scene to directly supervise the\nspatio-directional radiance field.\n  The primary outcome of this change is the complete removal of alpha blending\nand ray marching from the image formation model, instead moving these steps\ninto the loss computation. In addition to promoting convergence to surfaces,\nthis formulation assigns explicit semantic meaning to 2D subsets of the\nradiance field, turning them into well-defined radiance surfaces. We finally\nextract a level set from this representation, which results in a high-quality\nradiance surface model.\n  Our method retains much of the speed and quality of the baseline algorithm.\nFor instance, a suitably modified variant of Instant NGP maintains comparable\ncomputational efficiency, while achieving an average PSNR that is only 0.1 dB\nlower. Most importantly, our method generates explicit surfaces in place of an\nexponential volume, doing so with a level of simplicity not seen in prior work.\n", "date": "2025-01-27", "categories": []}
{"id": "2504.20378", "title": "Sparse2DGS: Geometry-Prioritized Gaussian Splatting for Surface\n  Reconstruction from Sparse Views", "abstract": "  We present a Gaussian Splatting method for surface reconstruction using\nsparse input views. Previous methods relying on dense views struggle with\nextremely sparse Structure-from-Motion points for initialization. While\nlearning-based Multi-view Stereo (MVS) provides dense 3D points, directly\ncombining it with Gaussian Splatting leads to suboptimal results due to the\nill-posed nature of sparse-view geometric optimization. We propose Sparse2DGS,\nan MVS-initialized Gaussian Splatting pipeline for complete and accurate\nreconstruction. Our key insight is to incorporate the geometric-prioritized\nenhancement schemes, allowing for direct and robust geometric learning under\nill-posed conditions. Sparse2DGS outperforms existing methods by notable\nmargins while being ${2}\\times$ faster than the NeRF-based fine-tuning\napproach.\n", "date": "2025-04-29", "categories": []}
{"id": "2409.01003", "title": "Free-DyGS: Camera-Pose-Free Scene Reconstruction for Dynamic Surgical\n  Videos with Gaussian Splatting", "abstract": "  High-fidelity reconstruction of surgical scene is a fundamentally crucial\ntask to support many applications, such as intra-operative navigation and\nsurgical education. However, most existing methods assume the ideal surgical\nscenarios - either focus on dynamic reconstruction with deforming tissue yet\nassuming a given fixed camera pose, or allow endoscope movement yet\nreconstructing the static scenes. In this paper, we target at a more realistic\nyet challenging setup - free-pose reconstruction with a moving camera for\nhighly dynamic surgical scenes. Meanwhile, we take the first step to introduce\nGaussian Splitting (GS) technique to tackle this challenging setting and\npropose a novel GS-based framework for fast reconstruction, termed\n\\textit{Free-DyGS}. Concretely, our model embraces a novel scene initialization\nin which a pre-trained Sparse Gaussian Regressor (SGR) can efficiently\nparameterize the initial attributes. For each subsequent frame, we propose to\njointly optimize the deformation model and 6D camera poses in a frame-by-frame\nmanner, easing training given the limited deformation differences between\nconsecutive frames. A Scene Expansion scheme is followed to expand the GS model\nfor the unseen regions introduced by the moving camera. Moreover, the framework\nis equipped with a novel Retrospective Deformation Recapitulation (RDR)\nstrategy to preserve the entire-clip deformations throughout the frame-by-frame\ntraining scheme. The efficacy of the proposed Free-DyGS is substantiated\nthrough extensive experiments on two datasets: StereoMIS and Hamlyn datasets.\nThe experimental outcomes underscore that Free-DyGS surpasses other advanced\nmethods in both rendering accuracy and efficiency. Code will be available.\n", "date": "2024-09-02", "categories": []}
{"id": "2412.12091", "title": "Wonderland: Navigating 3D Scenes from a Single Image", "abstract": "  How can one efficiently generate high-quality, wide-scope 3D scenes from\narbitrary single images? Existing methods suffer several drawbacks, such as\nrequiring multi-view data, time-consuming per-scene optimization, distorted\ngeometry in occluded areas, and low visual quality in backgrounds. Our novel 3D\nscene reconstruction pipeline overcomes these limitations to tackle the\naforesaid challenge. Specifically, we introduce a large-scale reconstruction\nmodel that leverages latents from a video diffusion model to predict 3D\nGaussian Splattings of scenes in a feed-forward manner. The video diffusion\nmodel is designed to create videos precisely following specified camera\ntrajectories, allowing it to generate compressed video latents that encode\nmulti-view information while maintaining 3D consistency. We train the 3D\nreconstruction model to operate on the video latent space with a progressive\nlearning strategy, enabling the efficient generation of high-quality,\nwide-scope, and generic 3D scenes. Extensive evaluations across various\ndatasets affirm that our model significantly outperforms existing single-view\n3D scene generation methods, especially with out-of-domain images. Thus, we\ndemonstrate for the first time that a 3D reconstruction model can effectively\nbe built upon the latent space of a diffusion model in order to realize\nefficient 3D scene generation.\n", "date": "2024-12-16", "categories": []}
{"id": "2504.20026", "title": "LIRM: Large Inverse Rendering Model for Progressive Reconstruction of\n  Shape, Materials and View-dependent Radiance Fields", "abstract": "  We present Large Inverse Rendering Model (LIRM), a transformer architecture\nthat jointly reconstructs high-quality shape, materials, and radiance fields\nwith view-dependent effects in less than a second. Our model builds upon the\nrecent Large Reconstruction Models (LRMs) that achieve state-of-the-art\nsparse-view reconstruction quality. However, existing LRMs struggle to\nreconstruct unseen parts accurately and cannot recover glossy appearance or\ngenerate relightable 3D contents that can be consumed by standard Graphics\nengines. To address these limitations, we make three key technical\ncontributions to build a more practical multi-view 3D reconstruction framework.\nFirst, we introduce an update model that allows us to progressively add more\ninput views to improve our reconstruction. Second, we propose a hexa-plane\nneural SDF representation to better recover detailed textures, geometry and\nmaterial parameters. Third, we develop a novel neural directional-embedding\nmechanism to handle view-dependent effects. Trained on a large-scale shape and\nmaterial dataset with a tailored coarse-to-fine training scheme, our model\nachieves compelling results. It compares favorably to optimization-based\ndense-view inverse rendering methods in terms of geometry and relighting\naccuracy, while requiring only a fraction of the inference time.\n", "date": "2025-04-28", "categories": []}
{"id": "2504.20040", "title": "MP-SfM: Monocular Surface Priors for Robust Structure-from-Motion", "abstract": "  While Structure-from-Motion (SfM) has seen much progress over the years,\nstate-of-the-art systems are prone to failure when facing extreme viewpoint\nchanges in low-overlap, low-parallax or high-symmetry scenarios. Because\ncapturing images that avoid these pitfalls is challenging, this severely limits\nthe wider use of SfM, especially by non-expert users. We overcome these\nlimitations by augmenting the classical SfM paradigm with monocular depth and\nnormal priors inferred by deep neural networks. Thanks to a tight integration\nof monocular and multi-view constraints, our approach significantly outperforms\nexisting ones under extreme viewpoint changes, while maintaining strong\nperformance in standard conditions. We also show that monocular priors can help\nreject faulty associations due to symmetries, which is a long-standing problem\nfor SfM. This makes our approach the first capable of reliably reconstructing\nchallenging indoor environments from few images. Through principled uncertainty\npropagation, it is robust to errors in the priors, can handle priors inferred\nby different models with little tuning, and will thus easily benefit from\nfuture progress in monocular depth and normal estimation. Our code is publicly\navailable at https://github.com/cvg/mpsfm.\n", "date": "2025-04-28", "categories": []}
{"id": "2410.01535", "title": "GaussianBlock: Building Part-Aware Compositional and Editable 3D Scene\n  by Primitives and Gaussians", "abstract": "  Recently, with the development of Neural Radiance Fields and Gaussian\nSplatting, 3D reconstruction techniques have achieved remarkably high fidelity.\nHowever, the latent representations learnt by these methods are highly\nentangled and lack interpretability. In this paper, we propose a novel\npart-aware compositional reconstruction method, called GaussianBlock, that\nenables semantically coherent and disentangled representations, allowing for\nprecise and physical editing akin to building blocks, while simultaneously\nmaintaining high fidelity. Our GaussianBlock introduces a hybrid representation\nthat leverages the advantages of both primitives, known for their flexible\nactionability and editability, and 3D Gaussians, which excel in reconstruction\nquality. Specifically, we achieve semantically coherent primitives through a\nnovel attention-guided centering loss derived from 2D semantic priors,\ncomplemented by a dynamic splitting and fusion strategy. Furthermore, we\nutilize 3D Gaussians that hybridize with primitives to refine structural\ndetails and enhance fidelity. Additionally, a binding inheritance strategy is\nemployed to strengthen and maintain the connection between the two. Our\nreconstructed scenes are evidenced to be disentangled, compositional, and\ncompact across diverse benchmarks, enabling seamless, direct and precise\nediting while maintaining high quality.\n", "date": "2024-10-02", "categories": []}
{"id": "2406.04861", "title": "Normal-guided Detail-Preserving Neural Implicit Function for\n  High-Fidelity 3D Surface Reconstruction", "abstract": "  Neural implicit representations have emerged as a powerful paradigm for 3D\nreconstruction. However, despite their success, existing methods fail to\ncapture fine geometric details and thin structures, especially in scenarios\nwhere only sparse multi-view RGB images of the objects of interest are\navailable. This paper shows that training neural representations with\nfirst-order differential properties (surface normals) leads to highly accurate\n3D surface reconstruction, even with as few as two RGB images. Using input RGB\nimages, we compute approximate ground-truth surface normals from depth maps\nproduced by an off-the-shelf monocular depth estimator. During training, we\ndirectly locate the surface point of the SDF network and supervise its normal\nwith the one estimated from the depth map. Extensive experiments demonstrate\nthat our method achieves state-of-the-art reconstruction accuracy with a\nminimal number of views, capturing intricate geometric details and thin\nstructures that were previously challenging to capture.\n", "date": "2024-06-07", "categories": []}
{"id": "2409.07200", "title": "ThermalGaussian: Thermal 3D Gaussian Splatting", "abstract": "  Thermography is especially valuable for the military and other users of\nsurveillance cameras. Some recent methods based on Neural Radiance Fields\n(NeRF) are proposed to reconstruct the thermal scenes in 3D from a set of\nthermal and RGB images. However, unlike NeRF, 3D Gaussian splatting (3DGS)\nprevails due to its rapid training and real-time rendering. In this work, we\npropose ThermalGaussian, the first thermal 3DGS approach capable of rendering\nhigh-quality images in RGB and thermal modalities. We first calibrate the RGB\ncamera and the thermal camera to ensure that both modalities are accurately\naligned. Subsequently, we use the registered images to learn the multimodal 3D\nGaussians. To prevent the overfitting of any single modality, we introduce\nseveral multimodal regularization constraints. We also develop smoothing\nconstraints tailored to the physical characteristics of the thermal modality.\nBesides, we contribute a real-world dataset named RGBT-Scenes, captured by a\nhand-hold thermal-infrared camera, facilitating future research on thermal\nscene reconstruction. We conduct comprehensive experiments to show that\nThermalGaussian achieves photorealistic rendering of thermal images and\nimproves the rendering quality of RGB images. With the proposed multimodal\nregularization constraints, we also reduced the model's storage cost by 90%.\nOur project page is at https://thermalgaussian.github.io/.\n", "date": "2024-09-11", "categories": []}
{"id": "2504.15278", "title": "DRAWER: Digital Reconstruction and Articulation With Environment Realism", "abstract": "  Creating virtual digital replicas from real-world data unlocks significant\npotential across domains like gaming and robotics. In this paper, we present\nDRAWER, a novel framework that converts a video of a static indoor scene into a\nphotorealistic and interactive digital environment. Our approach centers on two\nmain contributions: (i) a reconstruction module based on a dual scene\nrepresentation that reconstructs the scene with fine-grained geometric details,\nand (ii) an articulation module that identifies articulation types and hinge\npositions, reconstructs simulatable shapes and appearances and integrates them\ninto the scene. The resulting virtual environment is photorealistic,\ninteractive, and runs in real time, with compatibility for game engines and\nrobotic simulation platforms. We demonstrate the potential of DRAWER by using\nit to automatically create an interactive game in Unreal Engine and to enable\nreal-to-sim-to-real transfer for robotics applications.\n", "date": "2025-04-21", "categories": []}
{"id": "2405.06945", "title": "Direct Learning of Mesh and Appearance via 3D Gaussian Splatting", "abstract": "  Accurately reconstructing a 3D scene including explicit geometry information\nis both attractive and challenging. Geometry reconstruction can benefit from\nincorporating differentiable appearance models, such as Neural Radiance Fields\nand 3D Gaussian Splatting (3DGS). However, existing methods encounter\nefficiency issues due to indirect geometry learning and the paradigm of\nseparately modeling geometry and surface appearance. In this work, we propose a\nlearnable scene model that incorporates 3DGS with an explicit geometry\nrepresentation, namely a mesh. Our model learns the mesh and appearance in an\nend-to-end manner, where we bind 3D Gaussians to the mesh faces and perform\ndifferentiable rendering of 3DGS to obtain photometric supervision. The model\ncreates an effective information pathway to supervise the learning of both 3DGS\nand mesh. Experimental results demonstrate that the learned scene model not\nonly improves efficiency and rendering quality but also enables manipulation\nvia the explicit mesh. In addition, our model has a unique advantage in\nadapting to scene updates, thanks to the end-to-end learning of both mesh and\nappearance.\n", "date": "2024-05-11", "categories": []}
{"id": "2406.04251", "title": "Improving Gaussian Splatting with Localized Points Management", "abstract": "  Point management is critical for optimizing 3D Gaussian Splatting models, as\npoint initiation (e.g., via structure from motion) is often distributionally\ninappropriate. Typically, Adaptive Density Control (ADC) algorithm is adopted,\nleveraging view-averaged gradient magnitude thresholding for point\ndensification, opacity thresholding for pruning, and regular all-points opacity\nreset. We reveal that this strategy is limited in tackling intricate/special\nimage regions (e.g., transparent) due to inability of identifying all 3D zones\nrequiring point densification, and lacking an appropriate mechanism to handle\nill-conditioned points with negative impacts (e.g., occlusion due to false high\nopacity). To address these limitations, we propose a Localized Point Management\n(LPM) strategy, capable of identifying those error-contributing zones in\ngreatest need for both point addition and geometry calibration. Zone\nidentification is achieved by leveraging the underlying multiview geometry\nconstraints, subject to image rendering errors. We apply point densification in\nthe identified zones and then reset the opacity of the points in front of these\nregions, creating a new opportunity to correct poorly conditioned points.\nServing as a versatile plugin, LPM can be seamlessly integrated into existing\nstatic 3D and dynamic 4D Gaussian Splatting models with minimal additional\ncost. Experimental evaluations validate the efficacy of our LPM in boosting a\nvariety of existing 3D/4D models both quantitatively and qualitatively.\nNotably, LPM improves both static 3DGS and dynamic SpaceTimeGS to achieve\nstate-of-the-art rendering quality while retaining real-time speeds, excelling\non challenging datasets such as Tanks & Temples and the Neural 3D Video\ndataset.\n", "date": "2024-06-06", "categories": []}
{"id": "2408.16760", "title": "OmniRe: Omni Urban Scene Reconstruction", "abstract": "  We introduce OmniRe, a comprehensive system for efficiently creating\nhigh-fidelity digital twins of dynamic real-world scenes from on-device logs.\nRecent methods using neural fields or Gaussian Splatting primarily focus on\nvehicles, hindering a holistic framework for all dynamic foregrounds demanded\nby downstream applications, e.g., the simulation of human behavior. OmniRe\nextends beyond vehicle modeling to enable accurate, full-length reconstruction\nof diverse dynamic objects in urban scenes. Our approach builds scene graphs on\n3DGS and constructs multiple Gaussian representations in canonical spaces that\nmodel various dynamic actors, including vehicles, pedestrians, cyclists, and\nothers. OmniRe allows holistically reconstructing any dynamic object in the\nscene, enabling advanced simulations (~60Hz) that include human-participated\nscenarios, such as pedestrian behavior simulation and human-vehicle\ninteraction. This comprehensive simulation capability is unmatched by existing\nmethods. Extensive evaluations on the Waymo dataset show that our approach\noutperforms prior state-of-the-art methods quantitatively and qualitatively by\na large margin. We further extend our results to 5 additional popular driving\ndatasets to demonstrate its generalizability on common urban scenes.\n", "date": "2024-08-29", "categories": []}
{"id": "2504.00387", "title": "Scene4U: Hierarchical Layered 3D Scene Reconstruction from Single\n  Panoramic Image for Your Immerse Exploration", "abstract": "  The reconstruction of immersive and realistic 3D scenes holds significant\npractical importance in various fields of computer vision and computer\ngraphics. Typically, immersive and realistic scenes should be free from\nobstructions by dynamic objects, maintain global texture consistency, and allow\nfor unrestricted exploration. The current mainstream methods for image-driven\nscene construction involves iteratively refining the initial image using a\nmoving virtual camera to generate the scene. However, previous methods struggle\nwith visual discontinuities due to global texture inconsistencies under varying\ncamera poses, and they frequently exhibit scene voids caused by\nforeground-background occlusions. To this end, we propose a novel layered 3D\nscene reconstruction framework from panoramic image, named Scene4U.\nSpecifically, Scene4U integrates an open-vocabulary segmentation model with a\nlarge language model to decompose a real panorama into multiple layers. Then,\nwe employs a layered repair module based on diffusion model to restore occluded\nregions using visual cues and depth information, generating a hierarchical\nrepresentation of the scene. The multi-layer panorama is then initialized as a\n3D Gaussian Splatting representation, followed by layered optimization, which\nultimately produces an immersive 3D scene with semantic and structural\nconsistency that supports free exploration. Scene4U outperforms\nstate-of-the-art method, improving by 24.24% in LPIPS and 24.40% in BRISQUE,\nwhile also achieving the fastest training speed. Additionally, to demonstrate\nthe robustness of Scene4U and allow users to experience immersive scenes from\nvarious landmarks, we build WorldVista3D dataset for 3D scene reconstruction,\nwhich contains panoramic images of globally renowned sites. The implementation\ncode and dataset will be released at https://github.com/LongHZ140516/Scene4U .\n", "date": "2025-04-01", "categories": []}
{"id": "2504.14548", "title": "VGNC: Reducing the Overfitting of Sparse-view 3DGS via Validation-guided\n  Gaussian Number Control", "abstract": "  Sparse-view 3D reconstruction is a fundamental yet challenging task in\npractical 3D reconstruction applications. Recently, many methods based on the\n3D Gaussian Splatting (3DGS) framework have been proposed to address\nsparse-view 3D reconstruction. Although these methods have made considerable\nadvancements, they still show significant issues with overfitting. To reduce\nthe overfitting, we introduce VGNC, a novel Validation-guided Gaussian Number\nControl (VGNC) approach based on generative novel view synthesis (NVS) models.\nTo the best of our knowledge, this is the first attempt to alleviate the\noverfitting issue of sparse-view 3DGS with generative validation images.\nSpecifically, we first introduce a validation image generation method based on\na generative NVS model. We then propose a Gaussian number control strategy that\nutilizes generated validation images to determine the optimal Gaussian numbers,\nthereby reducing the issue of overfitting. We conducted detailed experiments on\nvarious sparse-view 3DGS baselines and datasets to evaluate the effectiveness\nof VGNC. Extensive experiments show that our approach not only reduces\noverfitting but also improves rendering quality on the test set while\ndecreasing the number of Gaussian points. This reduction lowers storage demands\nand accelerates both training and rendering. The code will be released.\n", "date": "2025-04-20", "categories": []}
{"id": "2504.11092", "title": "Vivid4D: Improving 4D Reconstruction from Monocular Video by Video\n  Inpainting", "abstract": "  Reconstructing 4D dynamic scenes from casually captured monocular videos is\nvaluable but highly challenging, as each timestamp is observed from a single\nviewpoint. We introduce Vivid4D, a novel approach that enhances 4D monocular\nvideo synthesis by augmenting observation views - synthesizing multi-view\nvideos from a monocular input. Unlike existing methods that either solely\nleverage geometric priors for supervision or use generative priors while\noverlooking geometry, we integrate both. This reformulates view augmentation as\na video inpainting task, where observed views are warped into new viewpoints\nbased on monocular depth priors. To achieve this, we train a video inpainting\nmodel on unposed web videos with synthetically generated masks that mimic\nwarping occlusions, ensuring spatially and temporally consistent completion of\nmissing regions. To further mitigate inaccuracies in monocular depth priors, we\nintroduce an iterative view augmentation strategy and a robust reconstruction\nloss. Experiments demonstrate that our method effectively improves monocular 4D\nscene reconstruction and completion. See our project page:\nhttps://xdimlab.github.io/Vivid4D/.\n", "date": "2025-04-15", "categories": []}
{"id": "2504.13167", "title": "ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from\n  Monocular Videos", "abstract": "  Creating a photorealistic scene and human reconstruction from a single\nmonocular in-the-wild video figures prominently in the perception of a\nhuman-centric 3D world. Recent neural rendering advances have enabled holistic\nhuman-scene reconstruction but require pre-calibrated camera and human poses,\nand days of training time. In this work, we introduce a novel unified framework\nthat simultaneously performs camera tracking, human pose estimation and\nhuman-scene reconstruction in an online fashion. 3D Gaussian Splatting is\nutilized to learn Gaussian primitives for humans and scenes efficiently, and\nreconstruction-based camera tracking and human pose estimation modules are\ndesigned to enable holistic understanding and effective disentanglement of pose\nand appearance. Specifically, we design a human deformation module to\nreconstruct the details and enhance generalizability to out-of-distribution\nposes faithfully. Aiming to learn the spatial correlation between human and\nscene accurately, we introduce occlusion-aware human silhouette rendering and\nmonocular geometric priors, which further improve reconstruction quality.\nExperiments on the EMDB and NeuMan datasets demonstrate superior or on-par\nperformance with existing methods in camera tracking, human pose estimation,\nnovel view synthesis and runtime. Our project page is at\nhttps://eth-ait.github.io/ODHSR.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.13419", "title": "Mono3R: Exploiting Monocular Cues for Geometric 3D Reconstruction", "abstract": "  Recent advances in data-driven geometric multi-view 3D reconstruction\nfoundation models (e.g., DUSt3R) have shown remarkable performance across\nvarious 3D vision tasks, facilitated by the release of large-scale,\nhigh-quality 3D datasets. However, as we observed, constrained by their\nmatching-based principles, the reconstruction quality of existing models\nsuffers significant degradation in challenging regions with limited matching\ncues, particularly in weakly textured areas and low-light conditions. To\nmitigate these limitations, we propose to harness the inherent robustness of\nmonocular geometry estimation to compensate for the inherent shortcomings of\nmatching-based methods. Specifically, we introduce a monocular-guided\nrefinement module that integrates monocular geometric priors into multi-view\nreconstruction frameworks. This integration substantially enhances the\nrobustness of multi-view reconstruction systems, leading to high-quality\nfeed-forward reconstructions. Comprehensive experiments across multiple\nbenchmarks demonstrate that our method achieves substantial improvements in\nboth mutli-view camera pose estimation and point cloud accuracy.\n", "date": "2025-04-18", "categories": []}
{"id": "2504.13540", "title": "EG-Gaussian: Epipolar Geometry and Graph Network Enhanced 3D Gaussian\n  Splatting", "abstract": "  In this paper, we explore an open research problem concerning the\nreconstruction of 3D scenes from images. Recent methods have adopt 3D Gaussian\nSplatting (3DGS) to produce 3D scenes due to its efficient training process.\nHowever, these methodologies may generate incomplete 3D scenes or blurred\nmultiviews. This is because of (1) inaccurate 3DGS point initialization and (2)\nthe tendency of 3DGS to flatten 3D Gaussians with the sparse-view input. To\naddress these issues, we propose a novel framework EG-Gaussian, which utilizes\nepipolar geometry and graph networks for 3D scene reconstruction. Initially, we\nintegrate epipolar geometry into the 3DGS initialization phase to enhance\ninitial 3DGS point construction. Then, we specifically design a graph learning\nmodule to refine 3DGS spatial features, in which we incorporate both spatial\ncoordinates and angular relationships among neighboring points. Experiments on\nindoor and outdoor benchmark datasets demonstrate that our approach\nsignificantly improves reconstruction accuracy compared to 3DGS-based methods.\n", "date": "2025-04-18", "categories": []}
{"id": "2503.06125", "title": "RGB-Phase Speckle: Cross-Scene Stereo 3D Reconstruction via Wrapped\n  Pre-Normalization", "abstract": "  3D reconstruction garners increasing attention alongside the advancement of\nhigh-level image applications, where dense stereo matching (DSM) serves as a\npivotal technique. Previous studies often rely on publicly available datasets\nfor training, focusing on modifying network architectures or incorporating\nspecialized modules to extract domain-invariant features and thus improve model\nrobustness. In contrast, inspired by single-frame structured-light\nphase-shifting encoding, this study introduces RGB-Speckle, a cross-scene 3D\nreconstruction framework based on an active stereo camera system, designed to\nenhance robustness. Specifically, we propose a novel phase pre-normalization\nencoding-decoding method: first, we randomly perturb phase-shift maps and embed\nthem into the three RGB channels to generate color speckle patterns;\nsubsequently, the camera captures phase-encoded images modulated by objects as\ninput to a stereo matching network. This technique effectively mitigates\nexternal interference and ensures consistent input data for RGB-Speckle,\nthereby bolstering cross-domain 3D reconstruction stability. To validate the\nproposed method, we conduct complex experiments: (1) construct a color speckle\ndataset for complex scenarios based on the proposed encoding scheme; (2)\nevaluate the impact of the phase pre-normalization encoding-decoding technique\non 3D reconstruction accuracy; and (3) further investigate its robustness\nacross diverse conditions. Experimental results demonstrate that the proposed\nRGB-Speckle model offers significant advantages in cross-domain and cross-scene\n3D reconstruction tasks, enhancing model generalization and reinforcing\nrobustness in challenging environments, thus providing a novel solution for\nrobust 3D reconstruction research.\n", "date": "2025-03-08", "categories": []}
{"id": "2504.13152", "title": "St4RTrack: Simultaneous 4D Reconstruction and Tracking in the World", "abstract": "  Dynamic 3D reconstruction and point tracking in videos are typically treated\nas separate tasks, despite their deep connection. We propose St4RTrack, a\nfeed-forward framework that simultaneously reconstructs and tracks dynamic\nvideo content in a world coordinate frame from RGB inputs. This is achieved by\npredicting two appropriately defined pointmaps for a pair of frames captured at\ndifferent moments. Specifically, we predict both pointmaps at the same moment,\nin the same world, capturing both static and dynamic scene geometry while\nmaintaining 3D correspondences. Chaining these predictions through the video\nsequence with respect to a reference frame naturally computes long-range\ncorrespondences, effectively combining 3D reconstruction with 3D tracking.\nUnlike prior methods that rely heavily on 4D ground truth supervision, we\nemploy a novel adaptation scheme based on a reprojection loss. We establish a\nnew extensive benchmark for world-frame reconstruction and tracking,\ndemonstrating the effectiveness and efficiency of our unified, data-driven\nframework. Our code, model, and benchmark will be released.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.13157", "title": "AerialMegaDepth: Learning Aerial-Ground Reconstruction and View\n  Synthesis", "abstract": "  We explore the task of geometric reconstruction of images captured from a\nmixture of ground and aerial views. Current state-of-the-art learning-based\napproaches fail to handle the extreme viewpoint variation between aerial-ground\nimage pairs. Our hypothesis is that the lack of high-quality, co-registered\naerial-ground datasets for training is a key reason for this failure. Such data\nis difficult to assemble precisely because it is difficult to reconstruct in a\nscalable way. To overcome this challenge, we propose a scalable framework\ncombining pseudo-synthetic renderings from 3D city-wide meshes (e.g., Google\nEarth) with real, ground-level crowd-sourced images (e.g., MegaDepth). The\npseudo-synthetic data simulates a wide range of aerial viewpoints, while the\nreal, crowd-sourced images help improve visual fidelity for ground-level images\nwhere mesh-based renderings lack sufficient detail, effectively bridging the\ndomain gap between real images and pseudo-synthetic renderings. Using this\nhybrid dataset, we fine-tune several state-of-the-art algorithms and achieve\nsignificant improvements on real-world, zero-shot aerial-ground tasks. For\nexample, we observe that baseline DUSt3R localizes fewer than 5% of\naerial-ground pairs within 5 degrees of camera rotation error, while\nfine-tuning with our data raises accuracy to nearly 56%, addressing a major\nfailure point in handling large viewpoint changes. Beyond camera estimation and\nscene reconstruction, our dataset also improves performance on downstream tasks\nlike novel-view synthesis in challenging aerial-ground scenarios, demonstrating\nthe practical value of our approach in real-world applications.\n", "date": "2025-04-17", "categories": []}
{"id": "2405.07847", "title": "SceneFactory: A Workflow-centric and Unified Framework for Incremental\n  Scene Modeling", "abstract": "  We present SceneFactory, a workflow-centric and unified framework for\nincremental scene modeling, that conveniently supports a wide range of\napplications, such as (unposed and/or uncalibrated) multi-view depth\nestimation, LiDAR completion, (dense) RGB-D/RGB-L/Mono/Depth-only\nreconstruction and SLAM. The workflow-centric design uses multiple blocks as\nthe basis for constructing different production lines. The supported\napplications, i.e., productions avoid redundancy in their designs. Thus, the\nfocus is placed on each block itself for independent expansion. To support all\ninput combinations, our implementation consists of four building blocks that\nform SceneFactory: (1) tracking, (2) flexion, (3) depth estimation, and (4)\nscene reconstruction. The tracking block is based on Mono SLAM and is extended\nto support RGB-D and RGB-LiDAR (RGB-L) inputs. Flexion is used to convert the\ndepth image (untrackable) into a trackable image. For general-purpose depth\nestimation, we propose an unposed \\& uncalibrated multi-view depth estimation\nmodel (U$^2$-MVD) to estimate dense geometry. U$^2$-MVD exploits dense bundle\nadjustment to solve for poses, intrinsics, and inverse depth. A semantic-aware\nScaleCov step is then introduced to complete the multi-view depth. Relying on\nU$^2$-MVD, SceneFactory both supports user-friendly 3D creation (with just\nimages) and bridges the applications of Dense RGB-D and Dense Mono. For\nhigh-quality surface and color reconstruction, we propose Dual-purpose\nMulti-resolutional Neural Points (DM-NPs) for the first surface accessible\nSurface Color Field design, where we introduce Improved Point Rasterization\n(IPR) for point cloud based surface query. ...\n", "date": "2024-05-13", "categories": []}
{"id": "2504.11447", "title": "Diffusion Distillation With Direct Preference Optimization For Efficient\n  3D LiDAR Scene Completion", "abstract": "  The application of diffusion models in 3D LiDAR scene completion is limited\ndue to diffusion's slow sampling speed. Score distillation accelerates\ndiffusion sampling but with performance degradation, while post-training with\ndirect policy optimization (DPO) boosts performance using preference data. This\npaper proposes Distillation-DPO, a novel diffusion distillation framework for\nLiDAR scene completion with preference aligment. First, the student model\ngenerates paired completion scenes with different initial noises. Second, using\nLiDAR scene evaluation metrics as preference, we construct winning and losing\nsample pairs. Such construction is reasonable, since most LiDAR scene metrics\nare informative but non-differentiable to be optimized directly. Third,\nDistillation-DPO optimizes the student model by exploiting the difference in\nscore functions between the teacher and student models on the paired completion\nscenes. Such procedure is repeated until convergence. Extensive experiments\ndemonstrate that, compared to state-of-the-art LiDAR scene completion diffusion\nmodels, Distillation-DPO achieves higher-quality scene completion while\naccelerating the completion speed by more than 5-fold. Our method is the first\nto explore adopting preference learning in distillation to the best of our\nknowledge and provide insights into preference-aligned distillation. Our code\nis public available on https://github.com/happyw1nd/DistillationDPO.\n", "date": "2025-04-15", "categories": []}
{"id": "2503.21958", "title": "SC-NeRF: NeRF-based Point Cloud Reconstruction using a Stationary Camera\n  for Agricultural Applications", "abstract": "  This paper presents a NeRF-based framework for point cloud (PCD)\nreconstruction, specifically designed for indoor high-throughput plant\nphenotyping facilities. Traditional NeRF-based reconstruction methods require\ncameras to move around stationary objects, but this approach is impractical for\nhigh-throughput environments where objects are rapidly imaged while moving on\nconveyors or rotating pedestals. To address this limitation, we develop a\nvariant of NeRF-based PCD reconstruction that uses a single stationary camera\nto capture images as the object rotates on a pedestal. Our workflow comprises\nCOLMAP-based pose estimation, a straightforward pose transformation to simulate\ncamera movement, and subsequent standard NeRF training. A defined Region of\nInterest (ROI) excludes irrelevant scene data, enabling the generation of\nhigh-resolution point clouds (10M points). Experimental results demonstrate\nexcellent reconstruction fidelity, with precision-recall analyses yielding an\nF-score close to 100.00 across all evaluated plant objects. Although pose\nestimation remains computationally intensive with a stationary camera setup,\noverall training and reconstruction times are competitive, validating the\nmethod's feasibility for practical high-throughput indoor phenotyping\napplications. Our findings indicate that high-quality NeRF-based 3D\nreconstructions are achievable using a stationary camera, eliminating the need\nfor complex camera motion or costly imaging equipment. This approach is\nespecially beneficial when employing expensive and delicate instruments, such\nas hyperspectral cameras, for 3D plant phenotyping. Future work will focus on\noptimizing pose estimation techniques and further streamlining the methodology\nto facilitate seamless integration into automated, high-throughput 3D\nphenotyping pipelines.\n", "date": "2025-03-27", "categories": []}
{"id": "2504.08410", "title": "PMNI: Pose-free Multi-view Normal Integration for Reflective and\n  Textureless Surface Reconstruction", "abstract": "  Reflective and textureless surfaces remain a challenge in multi-view 3D\nreconstruction. Both camera pose calibration and shape reconstruction often\nfail due to insufficient or unreliable cross-view visual features. To address\nthese issues, we present PMNI (Pose-free Multi-view Normal Integration), a\nneural surface reconstruction method that incorporates rich geometric\ninformation by leveraging surface normal maps instead of RGB images. By\nenforcing geometric constraints from surface normals and multi-view shape\nconsistency within a neural signed distance function (SDF) optimization\nframework, PMNI simultaneously recovers accurate camera poses and high-fidelity\nsurface geometry. Experimental results on synthetic and real-world datasets\nshow that our method achieves state-of-the-art performance in the\nreconstruction of reflective surfaces, even without reliable initial camera\nposes.\n", "date": "2025-04-11", "categories": []}
{"id": "2504.09129", "title": "A Constrained Optimization Approach for Gaussian Splatting from\n  Coarsely-posed Images and Noisy Lidar Point Clouds", "abstract": "  3D Gaussian Splatting (3DGS) is a powerful reconstruction technique, but it\nneeds to be initialized from accurate camera poses and high-fidelity point\nclouds. Typically, the initialization is taken from Structure-from-Motion (SfM)\nalgorithms; however, SfM is time-consuming and restricts the application of\n3DGS in real-world scenarios and large-scale scene reconstruction. We introduce\na constrained optimization method for simultaneous camera pose estimation and\n3D reconstruction that does not require SfM support. Core to our approach is\ndecomposing a camera pose into a sequence of camera-to-(device-)center and\n(device-)center-to-world optimizations. To facilitate, we propose two\noptimization constraints conditioned to the sensitivity of each parameter group\nand restricts each parameter's search space. In addition, as we learn the scene\ngeometry directly from the noisy point clouds, we propose geometric constraints\nto improve the reconstruction quality. Experiments demonstrate that the\nproposed method significantly outperforms the existing (multi-modal) 3DGS\nbaseline and methods supplemented by COLMAP on both our collected dataset and\ntwo public benchmarks.\n", "date": "2025-04-12", "categories": []}
{"id": "2410.16995", "title": "E-3DGS: Gaussian Splatting with Exposure and Motion Events", "abstract": "  Achieving 3D reconstruction from images captured under optimal conditions has\nbeen extensively studied in the vision and imaging fields. However, in\nreal-world scenarios, challenges such as motion blur and insufficient\nillumination often limit the performance of standard frame-based cameras in\ndelivering high-quality images. To address these limitations, we incorporate a\ntransmittance adjustment device at the hardware level, enabling event cameras\nto capture both motion and exposure events for diverse 3D reconstruction\nscenarios. Motion events (triggered by camera or object movement) are collected\nin fast-motion scenarios when the device is inactive, while exposure events\n(generated through controlled camera exposure) are captured during slower\nmotion to reconstruct grayscale images for high-quality training and\noptimization of event-based 3D Gaussian Splatting (3DGS). Our framework\nsupports three modes: High-Quality Reconstruction using exposure events, Fast\nReconstruction relying on motion events, and Balanced Hybrid optimizing with\ninitial exposure events followed by high-speed motion events. On the EventNeRF\ndataset, we demonstrate that exposure events significantly improve fine detail\nreconstruction compared to motion events and outperform frame-based cameras\nunder challenging conditions such as low illumination and overexposure.\nFurthermore, we introduce EME-3D, a real-world 3D dataset with exposure events,\nmotion events, camera calibration parameters, and sparse point clouds. Our\nmethod achieves faster and higher-quality reconstruction than event-based NeRF\nand is more cost-effective than methods combining event and RGB data. E-3DGS\nsets a new benchmark for event-based 3D reconstruction with robust performance\nin challenging conditions and lower hardware demands. The source code and\ndataset will be available at https://github.com/MasterHow/E-3DGS.\n", "date": "2024-10-22", "categories": []}
{"id": "2503.16825", "title": "SGFormer: Satellite-Ground Fusion for 3D Semantic Scene Completion", "abstract": "  Recently, camera-based solutions have been extensively explored for scene\nsemantic completion (SSC). Despite their success in visible areas, existing\nmethods struggle to capture complete scene semantics due to frequent visual\nocclusions. To address this limitation, this paper presents the first\nsatellite-ground cooperative SSC framework, i.e., SGFormer, exploring the\npotential of satellite-ground image pairs in the SSC task. Specifically, we\npropose a dual-branch architecture that encodes orthogonal satellite and ground\nviews in parallel, unifying them into a common domain. Additionally, we design\na ground-view guidance strategy that corrects satellite image biases during\nfeature encoding, addressing misalignment between satellite and ground views.\nMoreover, we develop an adaptive weighting strategy that balances contributions\nfrom satellite and ground views. Experiments demonstrate that SGFormer\noutperforms the state of the art on SemanticKITTI and SSCBench-KITTI-360\ndatasets. Our code is available on https://github.com/gxytcrc/SGFormer.\n", "date": "2025-03-21", "categories": []}
{"id": "2504.07370", "title": "View-Dependent Uncertainty Estimation of 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has become increasingly popular in 3D scene\nreconstruction for its high visual accuracy. However, uncertainty estimation of\n3DGS scenes remains underexplored and is crucial to downstream tasks such as\nasset extraction and scene completion. Since the appearance of 3D gaussians is\nview-dependent, the color of a gaussian can thus be certain from an angle and\nuncertain from another. We thus propose to model uncertainty in 3DGS as an\nadditional view-dependent per-gaussian feature that can be modeled with\nspherical harmonics. This simple yet effective modeling is easily interpretable\nand can be integrated into the traditional 3DGS pipeline. It is also\nsignificantly faster than ensemble methods while maintaining high accuracy, as\ndemonstrated in our experiments.\n", "date": "2025-04-10", "categories": []}
{"id": "2404.16323", "title": "LeanGaussian: Breaking Pixel or Point Cloud Correspondence in Modeling\n  3D Gaussians", "abstract": "  Recently, Gaussian splatting has demonstrated significant success in novel\nview synthesis. Current methods often regress Gaussians with pixel or point\ncloud correspondence, linking each Gaussian with a pixel or a 3D point. This\nleads to the redundancy of Gaussians being used to overfit the correspondence\nrather than the objects represented by the 3D Gaussians themselves,\nconsequently wasting resources and lacking accurate geometries or textures. In\nthis paper, we introduce LeanGaussian, a novel approach that treats each query\nin deformable Transformer as one 3D Gaussian ellipsoid, breaking the pixel or\npoint cloud correspondence constraints. We leverage deformable decoder to\niteratively refine the Gaussians layer-by-layer with the image features as keys\nand values. Notably, the center of each 3D Gaussian is defined as 3D reference\npoints, which are then projected onto the image for deformable attention in 2D\nspace. On both the ShapeNet SRN dataset (category level) and the Google Scanned\nObjects dataset (open-category level, trained with the Objaverse dataset), our\napproach, outperforms prior methods by approximately 6.1%, achieving a PSNR of\n25.44 and 22.36, respectively. Additionally, our method achieves a 3D\nreconstruction speed of 7.2 FPS and rendering speed 500 FPS. Codes are\navailable at https://github.com/jwubz123/LeanGaussian.\n", "date": "2024-04-25", "categories": []}
{"id": "2410.18931", "title": "Sort-free Gaussian Splatting via Weighted Sum Rendering", "abstract": "  Recently, 3D Gaussian Splatting (3DGS) has emerged as a significant\nadvancement in 3D scene reconstruction, attracting considerable attention due\nto its ability to recover high-fidelity details while maintaining low\ncomplexity. Despite the promising results achieved by 3DGS, its rendering\nperformance is constrained by its dependence on costly non-commutative\nalpha-blending operations. These operations mandate complex view dependent\nsorting operations that introduce computational overhead, especially on the\nresource-constrained platforms such as mobile phones. In this paper, we propose\nWeighted Sum Rendering, which approximates alpha blending with weighted sums,\nthereby removing the need for sorting. This simplifies implementation, delivers\nsuperior performance, and eliminates the \"popping\" artifacts caused by sorting.\nExperimental results show that optimizing a generalized Gaussian splatting\nformulation to the new differentiable rendering yields competitive image\nquality. The method was implemented and tested in a mobile device GPU,\nachieving on average $1.23\\times$ faster rendering.\n", "date": "2024-10-24", "categories": []}
{"id": "2504.01732", "title": "FIORD: A Fisheye Indoor-Outdoor Dataset with LIDAR Ground Truth for 3D\n  Scene Reconstruction and Benchmarking", "abstract": "  The development of large-scale 3D scene reconstruction and novel view\nsynthesis methods mostly rely on datasets comprising perspective images with\nnarrow fields of view (FoV). While effective for small-scale scenes, these\ndatasets require large image sets and extensive structure-from-motion (SfM)\nprocessing, limiting scalability. To address this, we introduce a fisheye image\ndataset tailored for scene reconstruction tasks. Using dual 200-degree fisheye\nlenses, our dataset provides full 360-degree coverage of 5 indoor and 5 outdoor\nscenes. Each scene has sparse SfM point clouds and precise LIDAR-derived dense\npoint clouds that can be used as geometric ground-truth, enabling robust\nbenchmarking under challenging conditions such as occlusions and reflections.\nWhile the baseline experiments focus on vanilla Gaussian Splatting and NeRF\nbased Nerfacto methods, the dataset supports diverse approaches for scene\nreconstruction, novel view synthesis, and image-based rendering.\n", "date": "2025-04-02", "categories": []}
{"id": "2504.06716", "title": "GSta: Efficient Training Scheme with Siestaed Gaussians for Monocular 3D\n  Scene Reconstruction", "abstract": "  Gaussian Splatting (GS) is a popular approach for 3D reconstruction, mostly\ndue to its ability to converge reasonably fast, faithfully represent the scene\nand render (novel) views in a fast fashion. However, it suffers from large\nstorage and memory requirements, and its training speed still lags behind the\nhash-grid based radiance field approaches (e.g. Instant-NGP), which makes it\nespecially difficult to deploy them in robotics scenarios, where 3D\nreconstruction is crucial for accurate operation. In this paper, we propose\nGSta that dynamically identifies Gaussians that have converged well during\ntraining, based on their positional and color gradient norms. By forcing such\nGaussians into a siesta and stopping their updates (freezing) during training,\nwe improve training speed with competitive accuracy compared to state of the\nart. We also propose an early stopping mechanism based on the PSNR values\ncomputed on a subset of training images. Combined with other improvements, such\nas integrating a learning rate scheduler, GSta achieves an improved Pareto\nfront in convergence speed, memory and storage requirements, while preserving\nquality. We also show that GSta can improve other methods and complement\northogonal approaches in efficiency improvement; once combined with Trick-GS,\nGSta achieves up to 5x faster training, 16x smaller disk size compared to\nvanilla GS, while having comparable accuracy and consuming only half the peak\nmemory. More visualisations are available at\nhttps://anilarmagan.github.io/SRUK-GSta.\n", "date": "2025-04-09", "categories": []}
{"id": "2408.12598", "title": "ND-SDF: Learning Normal Deflection Fields for High-Fidelity Indoor\n  Reconstruction", "abstract": "  Neural implicit reconstruction via volume rendering has demonstrated its\neffectiveness in recovering dense 3D surfaces. However, it is non-trivial to\nsimultaneously recover meticulous geometry and preserve smoothness across\nregions with differing characteristics. To address this issue, previous methods\ntypically employ geometric priors, which are often constrained by the\nperformance of the prior models. In this paper, we propose ND-SDF, which learns\na Normal Deflection field to represent the angular deviation between the scene\nnormal and the prior normal. Unlike previous methods that uniformly apply\ngeometric priors on all samples, introducing significant bias in accuracy, our\nproposed normal deflection field dynamically learns and adapts the utilization\nof samples based on their specific characteristics, thereby improving both the\naccuracy and effectiveness of the model. Our method not only obtains smooth\nweakly textured regions such as walls and floors but also preserves the\ngeometric details of complex structures. In addition, we introduce a novel ray\nsampling strategy based on the deflection angle to facilitate the unbiased\nrendering process, which significantly improves the quality and accuracy of\nintricate surfaces, especially on thin structures. Consistent improvements on\nvarious challenging datasets demonstrate the superiority of our method.\n", "date": "2024-08-22", "categories": []}
{"id": "2412.17769", "title": "ActiveGS: Active Scene Reconstruction Using Gaussian Splatting", "abstract": "  Robotics applications often rely on scene reconstructions to enable\ndownstream tasks. In this work, we tackle the challenge of actively building an\naccurate map of an unknown scene using an RGB-D camera on a mobile platform. We\npropose a hybrid map representation that combines a Gaussian splatting map with\na coarse voxel map, leveraging the strengths of both representations: the\nhigh-fidelity scene reconstruction capabilities of Gaussian splatting and the\nspatial modelling strengths of the voxel map. At the core of our framework is\nan effective confidence modelling technique for the Gaussian splatting map to\nidentify under-reconstructed areas, while utilising spatial information from\nthe voxel map to target unexplored areas and assist in collision-free path\nplanning. By actively collecting scene information in under-reconstructed and\nunexplored areas for map updates, our approach achieves superior Gaussian\nsplatting reconstruction results compared to state-of-the-art approaches.\nAdditionally, we demonstrate the real-world applicability of our framework\nusing an unmanned aerial vehicle.\n", "date": "2024-12-23", "categories": []}
{"id": "2504.00950", "title": "Neural Pruning for 3D Scene Reconstruction: Efficient NeRF Acceleration", "abstract": "  Neural Radiance Fields (NeRF) have become a popular 3D reconstruction\napproach in recent years. While they produce high-quality results, they also\ndemand lengthy training times, often spanning days. This paper studies neural\npruning as a strategy to address these concerns. We compare pruning approaches,\nincluding uniform sampling, importance-based methods, and coreset-based\ntechniques, to reduce the model size and speed up training. Our findings show\nthat coreset-driven pruning can achieve a 50% reduction in model size and a 35%\nspeedup in training, with only a slight decrease in accuracy. These results\nsuggest that pruning can be an effective method for improving the efficiency of\nNeRF models in resource-limited settings.\n", "date": "2025-04-01", "categories": []}
{"id": "2504.05458", "title": "Optimizing 4D Gaussians for Dynamic Scene Video from Single Landscape\n  Images", "abstract": "  To achieve realistic immersion in landscape images, fluids such as water and\nclouds need to move within the image while revealing new scenes from various\ncamera perspectives. Recently, a field called dynamic scene video has emerged,\nwhich combines single image animation with 3D photography. These methods use\npseudo 3D space, implicitly represented with Layered Depth Images (LDIs). LDIs\nseparate a single image into depth-based layers, which enables elements like\nwater and clouds to move within the image while revealing new scenes from\ndifferent camera perspectives. However, as landscapes typically consist of\ncontinuous elements, including fluids, the representation of a 3D space\nseparates a landscape image into discrete layers, and it can lead to diminished\ndepth perception and potential distortions depending on camera movement.\nFurthermore, due to its implicit modeling of 3D space, the output may be\nlimited to videos in the 2D domain, potentially reducing their versatility. In\nthis paper, we propose representing a complete 3D space for dynamic scene video\nby modeling explicit representations, specifically 4D Gaussians, from a single\nimage. The framework is focused on optimizing 3D Gaussians by generating\nmulti-view images from a single image and creating 3D motion to optimize 4D\nGaussians. The most important part of proposed framework is consistent 3D\nmotion estimation, which estimates common motion among multi-view images to\nbring the motion in 3D space closer to actual motions. As far as we know, this\nis the first attempt that considers animation while representing a complete 3D\nspace from a single landscape image. Our model demonstrates the ability to\nprovide realistic immersion in various landscape images through diverse\nexperiments and metrics. Extensive experimental results are\nhttps://cvsp-lab.github.io/ICLR2025_3D-MOM/.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.06178", "title": "Flash Sculptor: Modular 3D Worlds from Objects", "abstract": "  Existing text-to-3D and image-to-3D models often struggle with complex scenes\ninvolving multiple objects and intricate interactions. Although some recent\nattempts have explored such compositional scenarios, they still require an\nextensive process of optimizing the entire layout, which is highly cumbersome\nif not infeasible at all. To overcome these challenges, we propose Flash\nSculptor in this paper, a simple yet effective framework for compositional 3D\nscene/object reconstruction from a single image. At the heart of Flash Sculptor\nlies a divide-and-conquer strategy, which decouples compositional scene\nreconstruction into a sequence of sub-tasks, including handling the appearance,\nrotation, scale, and translation of each individual instance. Specifically, for\nrotation, we introduce a coarse-to-fine scheme that brings the best of both\nworlds--efficiency and accuracy--while for translation, we develop an\noutlier-removal-based algorithm that ensures robust and precise parameters in a\nsingle step, without any iterative optimization. Extensive experiments\ndemonstrate that Flash Sculptor achieves at least a 3 times speedup over\nexisting compositional 3D methods, while setting new benchmarks in\ncompositional 3D reconstruction performance. Codes are available at\nhttps://github.com/YujiaHu1109/Flash-Sculptor.\n", "date": "2025-04-08", "categories": []}
{"id": "2504.06210", "title": "HiMoR: Monocular Deformable Gaussian Reconstruction with Hierarchical\n  Motion Representation", "abstract": "  We present Hierarchical Motion Representation (HiMoR), a novel deformation\nrepresentation for 3D Gaussian primitives capable of achieving high-quality\nmonocular dynamic 3D reconstruction. The insight behind HiMoR is that motions\nin everyday scenes can be decomposed into coarser motions that serve as the\nfoundation for finer details. Using a tree structure, HiMoR's nodes represent\ndifferent levels of motion detail, with shallower nodes modeling coarse motion\nfor temporal smoothness and deeper nodes capturing finer motion. Additionally,\nour model uses a few shared motion bases to represent motions of different sets\nof nodes, aligning with the assumption that motion tends to be smooth and\nsimple. This motion representation design provides Gaussians with a more\nstructured deformation, maximizing the use of temporal relationships to tackle\nthe challenging task of monocular dynamic 3D reconstruction. We also propose\nusing a more reliable perceptual metric as an alternative, given that\npixel-level metrics for evaluating monocular dynamic 3D reconstruction can\nsometimes fail to accurately reflect the true quality of reconstruction.\nExtensive experiments demonstrate our method's efficacy in achieving superior\nnovel view synthesis from challenging monocular videos with complex motions.\n", "date": "2025-04-08", "categories": []}
{"id": "2404.12378", "title": "6Img-to-3D: Few-Image Large-Scale Outdoor Driving Scene Reconstruction", "abstract": "  Current 3D reconstruction techniques struggle to infer unbounded scenes from\na few images faithfully. Specifically, existing methods have high computational\ndemands, require detailed pose information, and cannot reconstruct occluded\nregions reliably. We introduce 6Img-to-3D, an efficient, scalable\ntransformer-based encoder-renderer method for single-shot image to 3D\nreconstruction. Our method outputs a 3D-consistent parameterized triplane from\nonly six outward-facing input images for large-scale, unbounded outdoor driving\nscenarios. We take a step towards resolving existing shortcomings by combining\ncontracted custom cross- and self-attention mechanisms for triplane\nparameterization, differentiable volume rendering, scene contraction, and image\nfeature projection. We showcase that six surround-view vehicle images from a\nsingle timestamp without global pose information are enough to reconstruct\n360$^{\\circ}$ scenes during inference time, taking 395 ms. Our method allows,\nfor example, rendering third-person images and birds-eye views. Our code is\navailable at https://github.com/continental/6Img-to-3D, and more examples can\nbe found at our website here https://6Img-to-3D.GitHub.io/.\n", "date": "2024-04-18", "categories": []}
{"id": "2406.00346", "title": "Details Enhancement in Unsigned Distance Field Learning for\n  High-fidelity 3D Surface Reconstruction", "abstract": "  While Signed Distance Fields (SDF) are well-established for modeling\nwatertight surfaces, Unsigned Distance Fields (UDF) broaden the scope to\ninclude open surfaces and models with complex inner structures. Despite their\nflexibility, UDFs encounter significant challenges in high-fidelity 3D\nreconstruction, such as non-differentiability at the zero level set, difficulty\nin achieving the exact zero value, numerous local minima, vanishing gradients,\nand oscillating gradient directions near the zero level set. To address these\nchallenges, we propose Details Enhanced UDF (DEUDF) learning that integrates\nnormal alignment and the SIREN network for capturing fine geometric details,\nadaptively weighted Eikonal constraints to address vanishing gradients near the\ntarget surface, unconditioned MLP-based UDF representation to relax\nnon-negativity constraints, and DCUDF for extracting the local minimal average\ndistance surface. These strategies collectively stabilize the learning process\nfrom unoriented point clouds and enhance the accuracy of UDFs. Our\ncomputational results demonstrate that DEUDF outperforms existing UDF learning\nmethods in both accuracy and the quality of reconstructed surfaces. Our source\ncode is at https://github.com/GiliAI/DEUDF.\n", "date": "2024-06-01", "categories": []}
{"id": "2411.17190", "title": "SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian\n  Splatting", "abstract": "  We propose SelfSplat, a novel 3D Gaussian Splatting model designed to perform\npose-free and 3D prior-free generalizable 3D reconstruction from unposed\nmulti-view images. These settings are inherently ill-posed due to the lack of\nground-truth data, learned geometric information, and the need to achieve\naccurate 3D reconstruction without finetuning, making it difficult for\nconventional methods to achieve high-quality results. Our model addresses these\nchallenges by effectively integrating explicit 3D representations with\nself-supervised depth and pose estimation techniques, resulting in reciprocal\nimprovements in both pose accuracy and 3D reconstruction quality. Furthermore,\nwe incorporate a matching-aware pose estimation network and a depth refinement\nmodule to enhance geometry consistency across views, ensuring more accurate and\nstable 3D reconstructions. To present the performance of our method, we\nevaluated it on large-scale real-world datasets, including RealEstate10K, ACID,\nand DL3DV. SelfSplat achieves superior results over previous state-of-the-art\nmethods in both appearance and geometry quality, also demonstrates strong\ncross-dataset generalization capabilities. Extensive ablation studies and\nanalysis also validate the effectiveness of our proposed methods. Code and\npretrained models are available at https://gynjn.github.io/selfsplat/\n", "date": "2024-11-26", "categories": []}
{"id": "2412.05826", "title": "Doppelgangers++: Improved Visual Disambiguation with Geometric 3D\n  Features", "abstract": "  Accurate 3D reconstruction is frequently hindered by visual aliasing, where\nvisually similar but distinct surfaces (aka, doppelgangers), are incorrectly\nmatched. These spurious matches distort the structure-from-motion (SfM)\nprocess, leading to misplaced model elements and reduced accuracy. Prior\nefforts addressed this with CNN classifiers trained on curated datasets, but\nthese approaches struggle to generalize across diverse real-world scenes and\ncan require extensive parameter tuning. In this work, we present\nDoppelgangers++, a method to enhance doppelganger detection and improve 3D\nreconstruction accuracy. Our contributions include a diversified training\ndataset that incorporates geo-tagged images from everyday scenes to expand\nrobustness beyond landmark-based datasets. We further propose a\nTransformer-based classifier that leverages 3D-aware features from the MASt3R\nmodel, achieving superior precision and recall across both in-domain and\nout-of-domain tests. Doppelgangers++ integrates seamlessly into standard SfM\nand MASt3R-SfM pipelines, offering efficiency and adaptability across varied\nscenes. To evaluate SfM accuracy, we introduce an automated, geotag-based\nmethod for validating reconstructed models, eliminating the need for manual\ninspection. Through extensive experiments, we demonstrate that Doppelgangers++\nsignificantly enhances pairwise visual disambiguation and improves 3D\nreconstruction quality in complex and diverse scenarios.\n", "date": "2024-12-08", "categories": []}
{"id": "2504.04000", "title": "View2CAD: Reconstructing View-Centric CAD Models from Single RGB-D Scans", "abstract": "  Parametric CAD models, represented as Boundary Representations (B-reps), are\nfoundational to modern design and manufacturing workflows, offering the\nprecision and topological breakdown required for downstream tasks such as\nanalysis, editing, and fabrication. However, B-Reps are often inaccessible due\nto conversion to more standardized, less expressive geometry formats. Existing\nmethods to recover B-Reps from measured data require complete, noise-free 3D\ndata, which are laborious to obtain. We alleviate this difficulty by enabling\nthe precise reconstruction of CAD shapes from a single RGB-D image. We propose\na method that addresses the challenge of reconstructing only the observed\ngeometry from a single view. To allow for these partial observations, and to\navoid hallucinating incorrect geometry, we introduce a novel view-centric B-rep\n(VB-Rep) representation, which incorporates structures to handle visibility\nlimits and encode geometric uncertainty. We combine panoptic image segmentation\nwith iterative geometric optimization to refine and improve the reconstruction\nprocess. Our results demonstrate high-quality reconstruction on synthetic and\nreal RGB-D data, showing that our method can bridge the reality gap.\n", "date": "2025-04-05", "categories": []}
{"id": "2504.04190", "title": "Interpretable Single-View 3D Gaussian Splatting using Unsupervised\n  Hierarchical Disentangled Representation Learning", "abstract": "  Gaussian Splatting (GS) has recently marked a significant advancement in 3D\nreconstruction, delivering both rapid rendering and high-quality results.\nHowever, existing 3DGS methods pose challenges in understanding underlying 3D\nsemantics, which hinders model controllability and interpretability. To address\nit, we propose an interpretable single-view 3DGS framework, termed 3DisGS, to\ndiscover both coarse- and fine-grained 3D semantics via hierarchical\ndisentangled representation learning (DRL). Specifically, the model employs a\ndual-branch architecture, consisting of a point cloud initialization branch and\na triplane-Gaussian generation branch, to achieve coarse-grained\ndisentanglement by separating 3D geometry and visual appearance features.\nSubsequently, fine-grained semantic representations within each modality are\nfurther discovered through DRL-based encoder-adapters. To our knowledge, this\nis the first work to achieve unsupervised interpretable 3DGS. Evaluations\nindicate that our model achieves 3D disentanglement while preserving\nhigh-quality and rapid reconstruction.\n", "date": "2025-04-05", "categories": []}
{"id": "2504.04679", "title": "DeclutterNeRF: Generative-Free 3D Scene Recovery for Occlusion Removal", "abstract": "  Recent novel view synthesis (NVS) techniques, including Neural Radiance\nFields (NeRF) and 3D Gaussian Splatting (3DGS) have greatly advanced 3D scene\nreconstruction with high-quality rendering and realistic detail recovery.\nEffectively removing occlusions while preserving scene details can further\nenhance the robustness and applicability of these techniques. However, existing\napproaches for object and occlusion removal predominantly rely on generative\npriors, which, despite filling the resulting holes, introduce new artifacts and\nblurriness. Moreover, existing benchmark datasets for evaluating occlusion\nremoval methods lack realistic complexity and viewpoint variations. To address\nthese issues, we introduce DeclutterSet, a novel dataset featuring diverse\nscenes with pronounced occlusions distributed across foreground, midground, and\nbackground, exhibiting substantial relative motion across viewpoints. We\nfurther introduce DeclutterNeRF, an occlusion removal method free from\ngenerative priors. DeclutterNeRF introduces joint multi-view optimization of\nlearnable camera parameters, occlusion annealing regularization, and employs an\nexplainable stochastic structural similarity loss, ensuring high-quality,\nartifact-free reconstructions from incomplete images. Experiments demonstrate\nthat DeclutterNeRF significantly outperforms state-of-the-art methods on our\nproposed DeclutterSet, establishing a strong baseline for future research.\n", "date": "2025-04-07", "categories": []}
{"id": "2504.05249", "title": "Texture2LoD3: Enabling LoD3 Building Reconstruction With Panoramic\n  Images", "abstract": "  Despite recent advancements in surface reconstruction, Level of Detail (LoD)\n3 building reconstruction remains an unresolved challenge. The main issue\npertains to the object-oriented modelling paradigm, which requires\ngeoreferencing, watertight geometry, facade semantics, and low-poly\nrepresentation -- Contrasting unstructured mesh-oriented models. In\nTexture2LoD3, we introduce a novel method leveraging the ubiquity of 3D\nbuilding model priors and panoramic street-level images, enabling the\nreconstruction of LoD3 building models. We observe that prior low-detail\nbuilding models can serve as valid planar targets for ortho-rectifying\nstreet-level panoramic images. Moreover, deploying segmentation on accurately\ntextured low-level building surfaces supports maintaining essential\ngeoreferencing, watertight geometry, and low-poly representation for LoD3\nreconstruction. In the absence of LoD3 validation data, we additionally\nintroduce the ReLoD3 dataset, on which we experimentally demonstrate that our\nmethod leads to improved facade segmentation accuracy by 11% and can replace\ncostly manual projections. We believe that Texture2LoD3 can scale the adoption\nof LoD3 models, opening applications in estimating building solar potential or\nenhancing autonomous driving simulations. The project website, code, and data\nare available here: https://wenzhaotang.github.io/Texture2LoD3/.\n", "date": "2025-04-07", "categories": []}
{"id": "2404.10620", "title": "PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape\n  Reconstruction", "abstract": "  We propose PyTorchGeoNodes, a differentiable module for reconstructing 3D\nobjects and their parameters from images using interpretable shape programs.\nUnlike traditional CAD model retrieval, shape programs allow reasoning about\nsemantic parameters, editing, and a low memory footprint. Despite their\npotential, shape programs for 3D scene understanding have been largely\noverlooked. Our key contribution is enabling gradient-based optimization by\nparsing shape programs, or more precisely procedural models designed in\nBlender, into efficient PyTorch code. While there are many possible\napplications of our PyTochGeoNodes, we show that a combination of\nPyTorchGeoNodes with genetic algorithm is a method of choice to optimize both\ndiscrete and continuous shape program parameters for 3D reconstruction and\nunderstanding of 3D object parameters. Our modular framework can be further\nintegrated with other reconstruction algorithms, and we demonstrate one such\nintegration to enable procedural Gaussian splatting. Our experiments on the\nScanNet dataset show that our method achieves accurate reconstructions while\nenabling, until now, unseen level of 3D scene understanding.\n", "date": "2024-04-16", "categories": []}
{"id": "2502.09563", "title": "Self-Calibrating Gaussian Splatting for Large Field of View\n  Reconstruction", "abstract": "  In this paper, we present a self-calibrating framework that jointly optimizes\ncamera parameters, lens distortion and 3D Gaussian representations, enabling\naccurate and efficient scene reconstruction. In particular, our technique\nenables high-quality scene reconstruction from Large field-of-view (FOV)\nimagery taken with wide-angle lenses, allowing the scene to be modeled from a\nsmaller number of images. Our approach introduces a novel method for modeling\ncomplex lens distortions using a hybrid network that combines invertible\nresidual networks with explicit grids. This design effectively regularizes the\noptimization process, achieving greater accuracy than conventional camera\nmodels. Additionally, we propose a cubemap-based resampling strategy to support\nlarge FOV images without sacrificing resolution or introducing distortion\nartifacts. Our method is compatible with the fast rasterization of Gaussian\nSplatting, adaptable to a wide variety of camera lens distortion, and\ndemonstrates state-of-the-art performance on both synthetic and real-world\ndatasets.\n", "date": "2025-02-13", "categories": []}
{"id": "2504.03198", "title": "Endo3R: Unified Online Reconstruction from Dynamic Monocular Endoscopic\n  Video", "abstract": "  Reconstructing 3D scenes from monocular surgical videos can enhance surgeon's\nperception and therefore plays a vital role in various computer-assisted\nsurgery tasks. However, achieving scale-consistent reconstruction remains an\nopen challenge due to inherent issues in endoscopic videos, such as dynamic\ndeformations and textureless surfaces. Despite recent advances, current methods\neither rely on calibration or instrument priors to estimate scale, or employ\nSfM-like multi-stage pipelines, leading to error accumulation and requiring\noffline optimization. In this paper, we present Endo3R, a unified 3D foundation\nmodel for online scale-consistent reconstruction from monocular surgical video,\nwithout any priors or extra optimization. Our model unifies the tasks by\npredicting globally aligned pointmaps, scale-consistent video depths, and\ncamera parameters without any offline optimization. The core contribution of\nour method is expanding the capability of the recent pairwise reconstruction\nmodel to long-term incremental dynamic reconstruction by an uncertainty-aware\ndual memory mechanism. The mechanism maintains history tokens of both\nshort-term dynamics and long-term spatial consistency. Notably, to tackle the\nhighly dynamic nature of surgical scenes, we measure the uncertainty of tokens\nvia Sampson distance and filter out tokens with high uncertainty. Regarding the\nscarcity of endoscopic datasets with ground-truth depth and camera poses, we\nfurther devise a self-supervised mechanism with a novel dynamics-aware flow\nloss. Abundant experiments on SCARED and Hamlyn datasets demonstrate our\nsuperior performance in zero-shot surgical video depth prediction and camera\npose estimation with online efficiency. Project page:\nhttps://wrld.github.io/Endo3R/.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.01956", "title": "VideoScene: Distilling Video Diffusion Model to Generate 3D Scenes in\n  One Step", "abstract": "  Recovering 3D scenes from sparse views is a challenging task due to its\ninherent ill-posed problem. Conventional methods have developed specialized\nsolutions (e.g., geometry regularization or feed-forward deterministic model)\nto mitigate the issue. However, they still suffer from performance degradation\nby minimal overlap across input views with insufficient visual information.\nFortunately, recent video generative models show promise in addressing this\nchallenge as they are capable of generating video clips with plausible 3D\nstructures. Powered by large pretrained video diffusion models, some pioneering\nresearch start to explore the potential of video generative prior and create 3D\nscenes from sparse views. Despite impressive improvements, they are limited by\nslow inference time and the lack of 3D constraint, leading to inefficiencies\nand reconstruction artifacts that do not align with real-world geometry\nstructure. In this paper, we propose VideoScene to distill the video diffusion\nmodel to generate 3D scenes in one step, aiming to build an efficient and\neffective tool to bridge the gap from video to 3D. Specifically, we design a\n3D-aware leap flow distillation strategy to leap over time-consuming redundant\ninformation and train a dynamic denoising policy network to adaptively\ndetermine the optimal leap timestep during inference. Extensive experiments\ndemonstrate that our VideoScene achieves faster and superior 3D scene\ngeneration results than previous video diffusion models, highlighting its\npotential as an efficient tool for future video to 3D applications. Project\nPage: https://hanyang-21.github.io/VideoScene\n", "date": "2025-04-02", "categories": []}
{"id": "2504.02764", "title": "Scene Splatter: Momentum 3D Scene Generation from Single Image with\n  Video Diffusion Model", "abstract": "  In this paper, we propose Scene Splatter, a momentum-based paradigm for video\ndiffusion to generate generic scenes from single image. Existing methods, which\nemploy video generation models to synthesize novel views, suffer from limited\nvideo length and scene inconsistency, leading to artifacts and distortions\nduring further reconstruction. To address this issue, we construct noisy\nsamples from original features as momentum to enhance video details and\nmaintain scene consistency. However, for latent features with the perception\nfield that spans both known and unknown regions, such latent-level momentum\nrestricts the generative ability of video diffusion in unknown regions.\nTherefore, we further introduce the aforementioned consistent video as a\npixel-level momentum to a directly generated video without momentum for better\nrecovery of unseen regions. Our cascaded momentum enables video diffusion\nmodels to generate both high-fidelity and consistent novel views. We further\nfinetune the global Gaussian representations with enhanced frames and render\nnew frames for momentum update in the next step. In this manner, we can\niteratively recover a 3D scene, avoiding the limitation of video length.\nExtensive experiments demonstrate the generalization capability and superior\nperformance of our method in high-fidelity and consistent scene generation.\n", "date": "2025-04-03", "categories": []}
{"id": "2504.01960", "title": "Diffusion-Guided Gaussian Splatting for Large-Scale Unconstrained 3D\n  Reconstruction and Novel View Synthesis", "abstract": "  Recent advancements in 3D Gaussian Splatting (3DGS) and Neural Radiance\nFields (NeRF) have achieved impressive results in real-time 3D reconstruction\nand novel view synthesis. However, these methods struggle in large-scale,\nunconstrained environments where sparse and uneven input coverage, transient\nocclusions, appearance variability, and inconsistent camera settings lead to\ndegraded quality. We propose GS-Diff, a novel 3DGS framework guided by a\nmulti-view diffusion model to address these limitations. By generating\npseudo-observations conditioned on multi-view inputs, our method transforms\nunder-constrained 3D reconstruction problems into well-posed ones, enabling\nrobust optimization even with sparse data. GS-Diff further integrates several\nenhancements, including appearance embedding, monocular depth priors, dynamic\nobject modeling, anisotropy regularization, and advanced rasterization\ntechniques, to tackle geometric and photometric challenges in real-world\nsettings. Experiments on four benchmarks demonstrate that GS-Diff consistently\noutperforms state-of-the-art baselines by significant margins.\n", "date": "2025-04-02", "categories": []}
{"id": "2405.19671", "title": "GaussianRoom: Improving 3D Gaussian Splatting with SDF Guidance and\n  Monocular Cues for Indoor Scene Reconstruction", "abstract": "  Embodied intelligence requires precise reconstruction and rendering to\nsimulate large-scale real-world data. Although 3D Gaussian Splatting (3DGS) has\nrecently demonstrated high-quality results with real-time performance, it still\nfaces challenges in indoor scenes with large, textureless regions, resulting in\nincomplete and noisy reconstructions due to poor point cloud initialization and\nunderconstrained optimization. Inspired by the continuity of signed distance\nfield (SDF), which naturally has advantages in modeling surfaces, we propose a\nunified optimization framework that integrates neural signed distance fields\n(SDFs) with 3DGS for accurate geometry reconstruction and real-time rendering.\nThis framework incorporates a neural SDF field to guide the densification and\npruning of Gaussians, enabling Gaussians to model scenes accurately even with\npoor initialized point clouds. Simultaneously, the geometry represented by\nGaussians improves the efficiency of the SDF field by piloting its point\nsampling. Additionally, we introduce two regularization terms based on normal\nand edge priors to resolve geometric ambiguities in textureless areas and\nenhance detail accuracy. Extensive experiments in ScanNet and ScanNet++ show\nthat our method achieves state-of-the-art performance in both surface\nreconstruction and novel view synthesis.\n", "date": "2024-05-30", "categories": []}
{"id": "2504.00639", "title": "Coca-Splat: Collaborative Optimization for Camera Parameters and 3D\n  Gaussians", "abstract": "  In this work, we introduce Coca-Splat, a novel approach to addressing the\nchallenges of sparse view pose-free scene reconstruction and novel view\nsynthesis (NVS) by jointly optimizing camera parameters with 3D Gaussians.\nInspired by deformable DEtection TRansformer, we design separate queries for 3D\nGaussians and camera parameters and update them layer by layer through\ndeformable Transformer layers, enabling joint optimization in a single network.\nThis design demonstrates better performance because to accurately render views\nthat closely approximate ground-truth images relies on precise estimation of\nboth 3D Gaussians and camera parameters. In such a design, the centers of 3D\nGaussians are projected onto each view by camera parameters to get projected\npoints, which are regarded as 2D reference points in deformable\ncross-attention. With camera-aware multi-view deformable cross-attention\n(CaMDFA), 3D Gaussians and camera parameters are intrinsically connected by\nsharing the 2D reference points. Additionally, 2D reference point determined\nrays (RayRef) defined from camera centers to the reference points assist in\nmodeling relationship between 3D Gaussians and camera parameters through\nRQ-decomposition on an overdetermined system of equations derived from the\nrays, enhancing the relationship between 3D Gaussians and camera parameters.\nExtensive evaluation shows that our approach outperforms previous methods, both\npose-required and pose-free, on RealEstate10K and ACID within the same\npose-free setting.\n", "date": "2025-04-01", "categories": []}
{"id": "2504.01016", "title": "GeometryCrafter: Consistent Geometry Estimation for Open-world Videos\n  with Diffusion Priors", "abstract": "  Despite remarkable advancements in video depth estimation, existing methods\nexhibit inherent limitations in achieving geometric fidelity through the\naffine-invariant predictions, limiting their applicability in reconstruction\nand other metrically grounded downstream tasks. We propose GeometryCrafter, a\nnovel framework that recovers high-fidelity point map sequences with temporal\ncoherence from open-world videos, enabling accurate 3D/4D reconstruction,\ncamera parameter estimation, and other depth-based applications. At the core of\nour approach lies a point map Variational Autoencoder (VAE) that learns a\nlatent space agnostic to video latent distributions for effective point map\nencoding and decoding. Leveraging the VAE, we train a video diffusion model to\nmodel the distribution of point map sequences conditioned on the input videos.\nExtensive evaluations on diverse datasets demonstrate that GeometryCrafter\nachieves state-of-the-art 3D accuracy, temporal consistency, and generalization\ncapability.\n", "date": "2025-04-01", "categories": []}
{"id": "2404.03421", "title": "Gen3DSR: Generalizable 3D Scene Reconstruction via Divide and Conquer\n  from a Single View", "abstract": "  Single-view 3D reconstruction is currently approached from two dominant\nperspectives: reconstruction of scenes with limited diversity using 3D data\nsupervision or reconstruction of diverse singular objects using large image\npriors. However, real-world scenarios are far more complex and exceed the\ncapabilities of these methods. We therefore propose a hybrid method following a\ndivide-and-conquer strategy. We first process the scene holistically,\nextracting depth and semantic information, and then leverage an object-level\nmethod for the detailed reconstruction of individual components. By splitting\nthe problem into simpler tasks, our system is able to generalize to various\ntypes of scenes without retraining or fine-tuning. We purposely design our\npipeline to be highly modular with independent, self-contained modules, to\navoid the need for end-to-end training of the whole system. This enables the\npipeline to naturally improve as future methods can replace the individual\nmodules. We demonstrate the reconstruction performance of our approach on both\nsynthetic and real-world scenes, comparing favorable against prior works.\nProject page: https://andreeadogaru.github.io/Gen3DSR\n", "date": "2024-04-04", "categories": []}
{"id": "2503.18361", "title": "NeRFPrior: Learning Neural Radiance Field as a Prior for Indoor Scene\n  Reconstruction", "abstract": "  Recently, it has shown that priors are vital for neural implicit functions to\nreconstruct high-quality surfaces from multi-view RGB images. However, current\npriors require large-scale pre-training, and merely provide geometric clues\nwithout considering the importance of color. In this paper, we present\nNeRFPrior, which adopts a neural radiance field as a prior to learn signed\ndistance fields using volume rendering for surface reconstruction. Our NeRF\nprior can provide both geometric and color clues, and also get trained fast\nunder the same scene without additional data. Based on the NeRF prior, we are\nenabled to learn a signed distance function (SDF) by explicitly imposing a\nmulti-view consistency constraint on each ray intersection for surface\ninference. Specifically, at each ray intersection, we use the density in the\nprior as a coarse geometry estimation, while using the color near the surface\nas a clue to check its visibility from another view angle. For the textureless\nareas where the multi-view consistency constraint does not work well, we\nfurther introduce a depth consistency loss with confidence weights to infer the\nSDF. Our experimental results outperform the state-of-the-art methods under the\nwidely used benchmarks.\n", "date": "2025-03-24", "categories": []}
{"id": "2503.21219", "title": "GenFusion: Closing the Loop between Reconstruction and Generation via\n  Videos", "abstract": "  Recently, 3D reconstruction and generation have demonstrated impressive novel\nview synthesis results, achieving high fidelity and efficiency. However, a\nnotable conditioning gap can be observed between these two fields, e.g.,\nscalable 3D scene reconstruction often requires densely captured views, whereas\n3D generation typically relies on a single or no input view, which\nsignificantly limits their applications. We found that the source of this\nphenomenon lies in the misalignment between 3D constraints and generative\npriors. To address this problem, we propose a reconstruction-driven video\ndiffusion model that learns to condition video frames on artifact-prone RGB-D\nrenderings. Moreover, we propose a cyclical fusion pipeline that iteratively\nadds restoration frames from the generative model to the training set, enabling\nprogressive expansion and addressing the viewpoint saturation limitations seen\nin previous reconstruction and generation pipelines. Our evaluation, including\nview synthesis from sparse view and masked input, validates the effectiveness\nof our approach. More details at https://genfusion.sibowu.com.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.23044", "title": "CityGS-X: A Scalable Architecture for Efficient and Geometrically\n  Accurate Large-Scale Scene Reconstruction", "abstract": "  Despite its significant achievements in large-scale scene reconstruction, 3D\nGaussian Splatting still faces substantial challenges, including slow\nprocessing, high computational costs, and limited geometric accuracy. These\ncore issues arise from its inherently unstructured design and the absence of\nefficient parallelization. To overcome these challenges simultaneously, we\nintroduce CityGS-X, a scalable architecture built on a novel parallelized\nhybrid hierarchical 3D representation (PH^2-3D). As an early attempt, CityGS-X\nabandons the cumbersome merge-and-partition process and instead adopts a\nnewly-designed batch-level multi-task rendering process. This architecture\nenables efficient multi-GPU rendering through dynamic Level-of-Detail voxel\nallocations, significantly improving scalability and performance. Through\nextensive experiments, CityGS-X consistently outperforms existing methods in\nterms of faster training times, larger rendering capacities, and more accurate\ngeometric details in large-scale scenes. Notably, CityGS-X can train and render\na scene with 5,000+ images in just 5 hours using only 4 * 4090 GPUs, a task\nthat would make other alternative methods encounter Out-Of-Memory (OOM) issues\nand fail completely. This implies that CityGS-X is far beyond the capacity of\nother existing methods.\n", "date": "2025-03-29", "categories": []}
{"id": "2503.23881", "title": "ExScene: Free-View 3D Scene Reconstruction with Gaussian Splatting from\n  a Single Image", "abstract": "  The increasing demand for augmented and virtual reality applications has\nhighlighted the importance of crafting immersive 3D scenes from a simple\nsingle-view image. However, due to the partial priors provided by single-view\ninput, existing methods are often limited to reconstruct low-consistency 3D\nscenes with narrow fields of view from single-view input. These limitations\nmake them less capable of generalizing to reconstruct immersive scenes. To\naddress this problem, we propose ExScene, a two-stage pipeline to reconstruct\nan immersive 3D scene from any given single-view image. ExScene designs a novel\nmultimodal diffusion model to generate a high-fidelity and globally consistent\npanoramic image. We then develop a panoramic depth estimation approach to\ncalculate geometric information from panorama, and we combine geometric\ninformation with high-fidelity panoramic image to train an initial 3D Gaussian\nSplatting (3DGS) model. Following this, we introduce a GS refinement technique\nwith 2D stable video diffusion priors. We add camera trajectory consistency and\ncolor-geometric priors into the denoising process of diffusion to improve color\nand spatial consistency across image sequences. These refined sequences are\nthen used to fine-tune the initial 3DGS model, leading to better reconstruction\nquality. Experimental results demonstrate that our ExScene achieves consistent\nand immersive scene reconstruction using only single-view input, significantly\nsurpassing state-of-the-art baselines.\n", "date": "2025-03-31", "categories": []}
{"id": "2503.24210", "title": "DiET-GS: Diffusion Prior and Event Stream-Assisted Motion Deblurring 3D\n  Gaussian Splatting", "abstract": "  Reconstructing sharp 3D representations from blurry multi-view images are\nlong-standing problem in computer vision. Recent works attempt to enhance\nhigh-quality novel view synthesis from the motion blur by leveraging\nevent-based cameras, benefiting from high dynamic range and microsecond\ntemporal resolution. However, they often reach sub-optimal visual quality in\neither restoring inaccurate color or losing fine-grained details. In this\npaper, we present DiET-GS, a diffusion prior and event stream-assisted motion\ndeblurring 3DGS. Our framework effectively leverages both blur-free event\nstreams and diffusion prior in a two-stage training strategy. Specifically, we\nintroduce the novel framework to constraint 3DGS with event double integral,\nachieving both accurate color and well-defined details. Additionally, we\npropose a simple technique to leverage diffusion prior to further enhance the\nedge details. Qualitative and quantitative results on both synthetic and\nreal-world data demonstrate that our DiET-GS is capable of producing\nsignificantly better quality of novel views compared to the existing baselines.\nOur project page is https://diet-gs.github.io\n", "date": "2025-03-31", "categories": []}
{"id": "2503.24382", "title": "Free360: Layered Gaussian Splatting for Unbounded 360-Degree View\n  Synthesis from Extremely Sparse and Unposed Views", "abstract": "  Neural rendering has demonstrated remarkable success in high-quality 3D\nneural reconstruction and novel view synthesis with dense input views and\naccurate poses. However, applying it to extremely sparse, unposed views in\nunbounded 360{\\deg} scenes remains a challenging problem. In this paper, we\npropose a novel neural rendering framework to accomplish the unposed and\nextremely sparse-view 3D reconstruction in unbounded 360{\\deg} scenes. To\nresolve the spatial ambiguity inherent in unbounded scenes with sparse input\nviews, we propose a layered Gaussian-based representation to effectively model\nthe scene with distinct spatial layers. By employing a dense stereo\nreconstruction model to recover coarse geometry, we introduce a layer-specific\nbootstrap optimization to refine the noise and fill occluded regions in the\nreconstruction. Furthermore, we propose an iterative fusion of reconstruction\nand generation alongside an uncertainty-aware training approach to facilitate\nmutual conditioning and enhancement between these two processes. Comprehensive\nexperiments show that our approach outperforms existing state-of-the-art\nmethods in terms of rendering quality and surface reconstruction accuracy.\nProject page: https://zju3dv.github.io/free360/\n", "date": "2025-03-31", "categories": []}
{"id": "2212.01959", "title": "INGeo: Accelerating Instant Neural Scene Reconstruction with Noisy\n  Geometry Priors", "abstract": "  We present a method that accelerates reconstruction of 3D scenes and objects,\naiming to enable instant reconstruction on edge devices such as mobile phones\nand AR/VR headsets. While recent works have accelerated scene reconstruction\ntraining to minute/second-level on high-end GPUs, there is still a large gap to\nthe goal of instant training on edge devices which is yet highly desired in\nmany emerging applications such as immersive AR/VR. To this end, this work aims\nto further accelerate training by leveraging geometry priors of the target\nscene. Our method proposes strategies to alleviate the noise of the imperfect\ngeometry priors to accelerate the training speed on top of the highly optimized\nInstant-NGP. On the NeRF Synthetic dataset, our work uses half of the training\niterations to reach an average test PSNR of >30.\n", "date": "2022-12-05", "categories": []}
{"id": "2411.17067", "title": "Geometry Field Splatting with Gaussian Surfels", "abstract": "  Geometric reconstruction of opaque surfaces from images is a longstanding\nchallenge in computer vision, with renewed interest from volumetric view\nsynthesis algorithms using radiance fields. We leverage the geometry field\nproposed in recent work for stochastic opaque surfaces, which can then be\nconverted to volume densities. We adapt Gaussian kernels or surfels to splat\nthe geometry field rather than the volume, enabling precise reconstruction of\nopaque solids. Our first contribution is to derive an efficient and almost\nexact differentiable rendering algorithm for geometry fields parameterized by\nGaussian surfels, while removing current approximations involving Taylor series\nand no self-attenuation. Next, we address the discontinuous loss landscape when\nsurfels cluster near geometry, showing how to guarantee that the rendered color\nis a continuous function of the colors of the kernels, irrespective of\nordering. Finally, we use latent representations with spherical harmonics\nencoded reflection vectors rather than spherical harmonics encoded colors to\nbetter address specular surfaces. We demonstrate significant improvement in the\nquality of reconstructed 3D surfaces on widely-used datasets.\n", "date": "2024-11-26", "categories": []}
{"id": "2501.05226", "title": "Light Transport-aware Diffusion Posterior Sampling for Single-View\n  Reconstruction of 3D Volumes", "abstract": "  We introduce a single-view reconstruction technique of volumetric fields in\nwhich multiple light scattering effects are omnipresent, such as in clouds. We\nmodel the unknown distribution of volumetric fields using an unconditional\ndiffusion model trained on a novel benchmark dataset comprising 1,000\nsynthetically simulated volumetric density fields. The neural diffusion model\nis trained on the latent codes of a novel, diffusion-friendly, monoplanar\nrepresentation. The generative model is used to incorporate a tailored\nparametric diffusion posterior sampling technique into different reconstruction\ntasks. A physically-based differentiable volume renderer is employed to provide\ngradients with respect to light transport in the latent space. This stands in\ncontrast to classic NeRF approaches and makes the reconstructions better\naligned with observed data. Through various experiments, we demonstrate\nsingle-view reconstruction of volumetric clouds at a previously unattainable\nquality.\n", "date": "2025-01-09", "categories": []}
{"id": "2503.19458", "title": "GaussianUDF: Inferring Unsigned Distance Functions through 3D Gaussian\n  Splatting", "abstract": "  Reconstructing open surfaces from multi-view images is vital in digitalizing\ncomplex objects in daily life. A widely used strategy is to learn unsigned\ndistance functions (UDFs) by checking if their appearance conforms to the image\nobservations through neural rendering. However, it is still hard to learn\ncontinuous and implicit UDF representations through 3D Gaussians splatting\n(3DGS) due to the discrete and explicit scene representation, i.e., 3D\nGaussians. To resolve this issue, we propose a novel approach to bridge the gap\nbetween 3D Gaussians and UDFs. Our key idea is to overfit thin and flat 2D\nGaussian planes on surfaces, and then, leverage the self-supervision and\ngradient-based inference to supervise unsigned distances in both near and far\narea to surfaces. To this end, we introduce novel constraints and strategies to\nconstrain the learning of 2D Gaussians to pursue more stable optimization and\nmore reliable self-supervision, addressing the challenges brought by\ncomplicated gradient field on or near the zero level set of UDFs. We report\nnumerical and visual comparisons with the state-of-the-art on widely used\nbenchmarks and real data to show our advantages in terms of accuracy,\nefficiency, completeness, and sharpness of reconstructed open surfaces with\nboundaries.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.21364", "title": "LandMarkSystem Technical Report", "abstract": "  3D reconstruction is vital for applications in autonomous driving, virtual\nreality, augmented reality, and the metaverse. Recent advancements such as\nNeural Radiance Fields(NeRF) and 3D Gaussian Splatting (3DGS) have transformed\nthe field, yet traditional deep learning frameworks struggle to meet the\nincreasing demands for scene quality and scale. This paper introduces\nLandMarkSystem, a novel computing framework designed to enhance multi-scale\nscene reconstruction and rendering. By leveraging a componentized model\nadaptation layer, LandMarkSystem supports various NeRF and 3DGS structures\nwhile optimizing computational efficiency through distributed parallel\ncomputing and model parameter offloading. Our system addresses the limitations\nof existing frameworks, providing dedicated operators for complex 3D sparse\ncomputations, thus facilitating efficient training and rapid inference over\nextensive scenes. Key contributions include a modular architecture, a dynamic\nloading strategy for limited resources, and proven capabilities across multiple\nrepresentative algorithms.This comprehensive solution aims to advance the\nefficiency and effectiveness of 3D reconstruction tasks.To facilitate further\nresearch and collaboration, the source code and documentation for the\nLandMarkSystem project are publicly available in an open-source repository,\naccessing the repository at: https://github.com/InternLandMark/LandMarkSystem.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.22437", "title": "EndoLRMGS: Complete Endoscopic Scene Reconstruction combining Large\n  Reconstruction Modelling and Gaussian Splatting", "abstract": "  Complete reconstruction of surgical scenes is crucial for robot-assisted\nsurgery (RAS). Deep depth estimation is promising but existing works struggle\nwith depth discontinuities, resulting in noisy predictions at object boundaries\nand do not achieve complete reconstruction omitting occluded surfaces. To\naddress these issues we propose EndoLRMGS, that combines Large Reconstruction\nModelling (LRM) and Gaussian Splatting (GS), for complete surgical scene\nreconstruction. GS reconstructs deformable tissues and LRM generates 3D models\nfor surgical tools while position and scale are subsequently optimized by\nintroducing orthogonal perspective joint projection optimization (OPjPO) to\nenhance accuracy. In experiments on four surgical videos from three public\ndatasets, our method improves the Intersection-over-union (IoU) of tool 3D\nmodels in 2D projections by>40%. Additionally, EndoLRMGS improves the PSNR of\nthe tools projection from 3.82% to 11.07%. Tissue rendering quality also\nimproves, with PSNR increasing from 0.46% to 49.87%, and SSIM from 1.53% to\n29.21% across all test videos.\n", "date": "2025-03-28", "categories": []}
{"id": "2503.22537", "title": "LIM: Large Interpolator Model for Dynamic Reconstruction", "abstract": "  Reconstructing dynamic assets from video data is central to many in computer\nvision and graphics tasks. Existing 4D reconstruction approaches are limited by\ncategory-specific models or slow optimization-based methods. Inspired by the\nrecent Large Reconstruction Model (LRM), we present the Large Interpolation\nModel (LIM), a transformer-based feed-forward solution, guided by a novel\ncausal consistency loss, for interpolating implicit 3D representations across\ntime. Given implicit 3D representations at times $t_0$ and $t_1$, LIM produces\na deformed shape at any continuous time $t\\in[t_0,t_1]$, delivering\nhigh-quality interpolated frames in seconds. Furthermore, LIM allows explicit\nmesh tracking across time, producing a consistently uv-textured mesh sequence\nready for integration into existing production pipelines. We also use LIM, in\nconjunction with a diffusion-based multiview generator, to produce dynamic 4D\nreconstructions from monocular videos. We evaluate LIM on various dynamic\ndatasets, benchmarking against image-space interpolation methods (e.g., FiLM)\nand direct triplane linear interpolation, and demonstrate clear advantages. In\nsummary, LIM is the first feed-forward model capable of high-speed tracked 4D\nasset reconstruction across diverse categories.\n", "date": "2025-03-28", "categories": []}
{"id": "2407.11309", "title": "Gaussian Splatting Lucas-Kanade", "abstract": "  Gaussian Splatting and its dynamic extensions are effective for\nreconstructing 3D scenes from 2D images when there is significant camera\nmovement to facilitate motion parallax and when scene objects remain relatively\nstatic. However, in many real-world scenarios, these conditions are not met. As\na consequence, data-driven semantic and geometric priors have been favored as\nregularizers, despite their bias toward training data and their neglect of\nbroader movement dynamics.\n  Departing from this practice, we propose a novel analytical approach that\nadapts the classical Lucas-Kanade method to dynamic Gaussian splatting. By\nleveraging the intrinsic properties of the forward warp field network, we\nderive an analytical velocity field that, through time integration, facilitates\naccurate scene flow computation. This enables the precise enforcement of motion\nconstraints on warp fields, thus constraining both 2D motion and 3D positions\nof the Gaussians. Our method excels in reconstructing highly dynamic scenes\nwith minimal camera movement, as demonstrated through experiments on both\nsynthetic and real-world scenes.\n", "date": "2024-07-16", "categories": []}
{"id": "2411.14847", "title": "Dynamics-Aware Gaussian Splatting Streaming Towards Fast On-the-Fly 4D\n  Reconstruction", "abstract": "  The recent development of 3D Gaussian Splatting (3DGS) has led to great\ninterest in 4D dynamic spatial reconstruction. Existing approaches mainly rely\non full-length multi-view videos, while there has been limited exploration of\nonline reconstruction methods that enable on-the-fly training and per-timestep\nstreaming. Current 3DGS-based streaming methods treat the Gaussian primitives\nuniformly and constantly renew the densified Gaussians, thereby overlooking the\ndifference between dynamic and static features as well as neglecting the\ntemporal continuity in the scene. To address these limitations, we propose a\nnovel three-stage pipeline for iterative streamable 4D dynamic spatial\nreconstruction. Our pipeline comprises a selective inheritance stage to\npreserve temporal continuity, a dynamics-aware shift stage to distinguish\ndynamic and static primitives and optimize their movements, and an error-guided\ndensification stage to accommodate emerging objects. Our method achieves\nstate-of-the-art performance in online 4D reconstruction, demonstrating the\nfastest on-the-fly training, superior representation quality, and real-time\nrendering capability. Project page: https://www.liuzhening.top/DASS\n", "date": "2024-11-22", "categories": []}
{"id": "2411.15482", "title": "SplatFlow: Self-Supervised Dynamic Gaussian Splatting in Neural Motion\n  Flow Field for Autonomous Driving", "abstract": "  Most existing Dynamic Gaussian Splatting methods for complex dynamic urban\nscenarios rely on accurate object-level supervision from expensive manual\nlabeling, limiting their scalability in real-world applications. In this paper,\nwe introduce SplatFlow, a Self-Supervised Dynamic Gaussian Splatting within\nNeural Motion Flow Fields (NMFF) to learn 4D space-time representations without\nrequiring tracked 3D bounding boxes, enabling accurate dynamic scene\nreconstruction and novel view RGB/depth/flow synthesis. SplatFlow designs a\nunified framework to seamlessly integrate time-dependent 4D Gaussian\nrepresentation within NMFF, where NMFF is a set of implicit functions to model\ntemporal motions of both LiDAR points and Gaussians as continuous motion flow\nfields. Leveraging NMFF, SplatFlow effectively decomposes static background and\ndynamic objects, representing them with 3D and 4D Gaussian primitives,\nrespectively. NMFF also models the correspondences of each 4D Gaussian across\ntime, which aggregates temporal features to enhance cross-view consistency of\ndynamic components. SplatFlow further improves dynamic object identification by\ndistilling features from 2D foundation models into 4D space-time\nrepresentation. Comprehensive evaluations conducted on the Waymo and KITTI\nDatasets validate SplatFlow's state-of-the-art (SOTA) performance for both\nimage reconstruction and novel view synthesis in dynamic urban scenarios.\n", "date": "2024-11-23", "categories": []}
{"id": "2411.16180", "title": "Event-boosted Deformable 3D Gaussians for Dynamic Scene Reconstruction", "abstract": "  Deformable 3D Gaussian Splatting (3D-GS) is limited by missing intermediate\nmotion information due to the low temporal resolution of RGB cameras. To\naddress this, we introduce the first approach combining event cameras, which\ncapture high-temporal-resolution, continuous motion data, with deformable 3D-GS\nfor dynamic scene reconstruction. We observe that threshold modeling for events\nplays a crucial role in achieving high-quality reconstruction. Therefore, we\npropose a GS-Threshold Joint Modeling strategy, creating a mutually reinforcing\nprocess that greatly improves both 3D reconstruction and threshold modeling.\nMoreover, we introduce a Dynamic-Static Decomposition strategy that first\nidentifies dynamic areas by exploiting the inability of static Gaussians to\nrepresent motions, then applies a buffer-based soft decomposition to separate\ndynamic and static areas. This strategy accelerates rendering by avoiding\nunnecessary deformation in static areas, and focuses on dynamic areas to\nenhance fidelity. Additionally, we contribute the first event-inclusive 4D\nbenchmark with synthetic and real-world dynamic scenes, on which our method\nachieves state-of-the-art performance.\n", "date": "2024-11-25", "categories": []}
{"id": "2412.15215", "title": "EnvGS: Modeling View-Dependent Appearance with Environment Gaussian", "abstract": "  Reconstructing complex reflections in real-world scenes from 2D images is\nessential for achieving photorealistic novel view synthesis. Existing methods\nthat utilize environment maps to model reflections from distant lighting often\nstruggle with high-frequency reflection details and fail to account for\nnear-field reflections. In this work, we introduce EnvGS, a novel approach that\nemploys a set of Gaussian primitives as an explicit 3D representation for\ncapturing reflections of environments. These environment Gaussian primitives\nare incorporated with base Gaussian primitives to model the appearance of the\nwhole scene. To efficiently render these environment Gaussian primitives, we\ndeveloped a ray-tracing-based renderer that leverages the GPU's RT core for\nfast rendering. This allows us to jointly optimize our model for high-quality\nreconstruction while maintaining real-time rendering speeds. Results from\nmultiple real-world and synthetic datasets demonstrate that our method produces\nsignificantly more detailed reflections, achieving the best rendering quality\nin real-time novel view synthesis. The code is available at\nhttps://zju3dv.github.io/envgs.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.16604", "title": "OmniSplat: Taming Feed-Forward 3D Gaussian Splatting for Omnidirectional\n  Images with Editable Capabilities", "abstract": "  Feed-forward 3D Gaussian splatting (3DGS) models have gained significant\npopularity due to their ability to generate scenes immediately without needing\nper-scene optimization. Although omnidirectional images are becoming more\npopular since they reduce the computation required for image stitching to\ncomposite a holistic scene, existing feed-forward models are only designed for\nperspective images. The unique optical properties of omnidirectional images\nmake it difficult for feature encoders to correctly understand the context of\nthe image and make the Gaussian non-uniform in space, which hinders the image\nquality synthesized from novel views. We propose OmniSplat, a training-free\nfast feed-forward 3DGS generation framework for omnidirectional images. We\nadopt a Yin-Yang grid and decompose images based on it to reduce the domain gap\nbetween omnidirectional and perspective images. The Yin-Yang grid can use the\nexisting CNN structure as it is, but its quasi-uniform characteristic allows\nthe decomposed image to be similar to a perspective image, so it can exploit\nthe strong prior knowledge of the learned feed-forward network. OmniSplat\ndemonstrates higher reconstruction accuracy than existing feed-forward networks\ntrained on perspective images. Our project page is available on:\nhttps://robot0321.github.io/omnisplat/index.html.\n", "date": "2024-12-21", "categories": []}
{"id": "2503.21082", "title": "Can Video Diffusion Model Reconstruct 4D Geometry?", "abstract": "  Reconstructing dynamic 3D scenes (i.e., 4D geometry) from monocular video is\nan important yet challenging problem. Conventional multiview geometry-based\napproaches often struggle with dynamic motion, whereas recent learning-based\nmethods either require specialized 4D representation or sophisticated\noptimization. In this paper, we present Sora3R, a novel framework that taps\ninto the rich spatiotemporal priors of large-scale video diffusion models to\ndirectly infer 4D pointmaps from casual videos. Sora3R follows a two-stage\npipeline: (1) we adapt a pointmap VAE from a pretrained video VAE, ensuring\ncompatibility between the geometry and video latent spaces; (2) we finetune a\ndiffusion backbone in combined video and pointmap latent space to generate\ncoherent 4D pointmaps for every frame. Sora3R operates in a fully feedforward\nmanner, requiring no external modules (e.g., depth, optical flow, or\nsegmentation) or iterative global alignment. Extensive experiments demonstrate\nthat Sora3R reliably recovers both camera poses and detailed scene geometry,\nachieving performance on par with state-of-the-art methods for dynamic 4D\nreconstruction across diverse scenarios.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.21761", "title": "Uni4D: Unifying Visual Foundation Models for 4D Modeling from a Single\n  Video", "abstract": "  This paper presents a unified approach to understanding dynamic scenes from\ncasual videos. Large pretrained vision foundation models, such as\nvision-language, video depth prediction, motion tracking, and segmentation\nmodels, offer promising capabilities. However, training a single model for\ncomprehensive 4D understanding remains challenging. We introduce Uni4D, a\nmulti-stage optimization framework that harnesses multiple pretrained models to\nadvance dynamic 3D modeling, including static/dynamic reconstruction, camera\npose estimation, and dense 3D motion tracking. Our results show\nstate-of-the-art performance in dynamic 4D modeling with superior visual\nquality. Notably, Uni4D requires no retraining or fine-tuning, highlighting the\neffectiveness of repurposing visual foundation models for 4D understanding.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.18402", "title": "DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds", "abstract": "  3D Gaussian Splatting (3DGS) renders pixels by rasterizing Gaussian\nprimitives, where the rendering resolution and the primitive number, concluded\nas the optimization complexity, dominate the time cost in primitive\noptimization. In this paper, we propose DashGaussian, a scheduling scheme over\nthe optimization complexity of 3DGS that strips redundant complexity to\naccelerate 3DGS optimization. Specifically, we formulate 3DGS optimization as\nprogressively fitting 3DGS to higher levels of frequency components in the\ntraining views, and propose a dynamic rendering resolution scheme that largely\nreduces the optimization complexity based on this formulation. Besides, we\nargue that a specific rendering resolution should cooperate with a proper\nprimitive number for a better balance between computing redundancy and fitting\nquality, where we schedule the growth of the primitives to synchronize with the\nrendering resolution. Extensive experiments show that our method accelerates\nthe optimization of various 3DGS backbones by 45.7% on average while preserving\nthe rendering quality.\n", "date": "2025-03-24", "categories": []}
{"id": "2503.19976", "title": "Thin-Shell-SfT: Fine-Grained Monocular Non-rigid 3D Surface Tracking\n  with Neural Deformation Fields", "abstract": "  3D reconstruction of highly deformable surfaces (e.g. cloths) from monocular\nRGB videos is a challenging problem, and no solution provides a consistent and\naccurate recovery of fine-grained surface details. To account for the ill-posed\nnature of the setting, existing methods use deformation models with\nstatistical, neural, or physical priors. They also predominantly rely on\nnonadaptive discrete surface representations (e.g. polygonal meshes), perform\nframe-by-frame optimisation leading to error propagation, and suffer from poor\ngradients of the mesh-based differentiable renderers. Consequently, fine\nsurface details such as cloth wrinkles are often not recovered with the desired\naccuracy. In response to these limitations, we propose ThinShell-SfT, a new\nmethod for non-rigid 3D tracking that represents a surface as an implicit and\ncontinuous spatiotemporal neural field. We incorporate continuous thin shell\nphysics prior based on the Kirchhoff-Love model for spatial regularisation,\nwhich starkly contrasts the discretised alternatives of earlier works. Lastly,\nwe leverage 3D Gaussian splatting to differentiably render the surface into\nimage space and optimise the deformations based on analysis-bysynthesis\nprinciples. Our Thin-Shell-SfT outperforms prior works qualitatively and\nquantitatively thanks to our continuous surface formulation in conjunction with\na specially tailored simulation prior and surface-induced 3D Gaussians. See our\nproject page at https://4dqv.mpiinf.mpg.de/ThinShellSfT.\n", "date": "2025-03-25", "categories": []}
{"id": "2410.13862", "title": "DepthSplat: Connecting Gaussian Splatting and Depth", "abstract": "  Gaussian splatting and single-view depth estimation are typically studied in\nisolation. In this paper, we present DepthSplat to connect Gaussian splatting\nand depth estimation and study their interactions. More specifically, we first\ncontribute a robust multi-view depth model by leveraging pre-trained monocular\ndepth features, leading to high-quality feed-forward 3D Gaussian splatting\nreconstructions. We also show that Gaussian splatting can serve as an\nunsupervised pre-training objective for learning powerful depth models from\nlarge-scale multi-view posed datasets. We validate the synergy between Gaussian\nsplatting and depth estimation through extensive ablation and cross-task\ntransfer experiments. Our DepthSplat achieves state-of-the-art performance on\nScanNet, RealEstate10K and DL3DV datasets in terms of both depth estimation and\nnovel view synthesis, demonstrating the mutual benefits of connecting both\ntasks. In addition, DepthSplat enables feed-forward reconstruction from 12\ninput views (512x960 resolutions) in 0.6 seconds.\n", "date": "2024-10-17", "categories": []}
{"id": "2410.17249", "title": "SpectroMotion: Dynamic 3D Reconstruction of Specular Scenes", "abstract": "  We present SpectroMotion, a novel approach that combines 3D Gaussian\nSplatting (3DGS) with physically-based rendering (PBR) and deformation fields\nto reconstruct dynamic specular scenes. Previous methods extending 3DGS to\nmodel dynamic scenes have struggled to represent specular surfaces accurately.\nOur method addresses this limitation by introducing a residual correction\ntechnique for accurate surface normal computation during deformation,\ncomplemented by a deformable environment map that adapts to time-varying\nlighting conditions. We implement a coarse-to-fine training strategy\nsignificantly enhancing scene geometry and specular color prediction. It is the\nonly existing 3DGS method capable of synthesizing photorealistic real-world\ndynamic specular scenes, outperforming state-of-the-art methods in rendering\ncomplex, dynamic, and specular scenes.\n", "date": "2024-10-22", "categories": []}
{"id": "2412.04120", "title": "CrossSDF: 3D Reconstruction of Thin Structures From Cross-Sections", "abstract": "  Reconstructing complex structures from planar cross-sections is a challenging\nproblem, with wide-reaching applications in medical imaging, manufacturing, and\ntopography. Out-of-the-box point cloud reconstruction methods can often fail\ndue to the data sparsity between slicing planes, while current bespoke methods\nstruggle to reconstruct thin geometric structures and preserve topological\ncontinuity. This is important for medical applications where thin vessel\nstructures are present in CT and MRI scans. This paper introduces CrossSDF, a\nnovel approach for extracting a 3D signed distance field from 2D signed\ndistances generated from planar contours. Our approach makes the training of\nneural SDFs contour-aware by using losses designed for the case where geometry\nis known within 2D slices. Our results demonstrate a significant improvement\nover existing methods, effectively reconstructing thin structures and producing\naccurate 3D models without the interpolation artifacts or over-smoothing of\nprior approaches.\n", "date": "2024-12-05", "categories": []}
{"id": "2503.00260", "title": "Seeing A 3D World in A Grain of Sand", "abstract": "  We present a snapshot imaging technique for recovering 3D surrounding views\nof miniature scenes. Due to their intricacy, miniature scenes with objects\nsized in millimeters are difficult to reconstruct, yet miniatures are common in\nlife and their 3D digitalization is desirable. We design a catadioptric imaging\nsystem with a single camera and eight pairs of planar mirrors for snapshot 3D\nreconstruction from a dollhouse perspective. We place paired mirrors on nested\npyramid surfaces for capturing surrounding multi-view images in a single shot.\nOur mirror design is customizable based on the size of the scene for optimized\nview coverage. We use the 3D Gaussian Splatting (3DGS) representation for scene\nreconstruction and novel view synthesis. We overcome the challenge posed by our\nsparse view input by integrating visual hull-derived depth constraint. Our\nmethod demonstrates state-of-the-art performance on a variety of synthetic and\nreal miniature scenes.\n", "date": "2025-03-01", "categories": []}
{"id": "2503.18527", "title": "AIM2PC: Aerial Image to 3D Building Point Cloud Reconstruction", "abstract": "  Three-dimensional urban reconstruction of buildings from single-view images\nhas attracted significant attention over the past two decades. However, recent\nmethods primarily focus on rooftops from aerial images, often overlooking\nessential geometrical details. Additionally, there is a notable lack of\ndatasets containing complete 3D point clouds for entire buildings, along with\nchallenges in obtaining reliable camera pose information for aerial images.\nThis paper addresses these challenges by presenting a novel methodology, AIM2PC\n, which utilizes our generated dataset that includes complete 3D point clouds\nand determined camera poses. Our approach takes features from a single aerial\nimage as input and concatenates them with essential additional conditions, such\nas binary masks and Sobel edge maps, to enable more edge-aware reconstruction.\nBy incorporating a point cloud diffusion model based on Centered denoising\nDiffusion Probabilistic Models (CDPM), we project these concatenated features\nonto the partially denoised point cloud using our camera poses at each\ndiffusion step. The proposed method is able to reconstruct the complete 3D\nbuilding point cloud, including wall information and demonstrates superior\nperformance compared to existing baseline techniques. To allow further\ncomparisons with our methodology the dataset has been made available at\nhttps://github.com/Soulaimene/AIM2PCDataset\n", "date": "2025-03-24", "categories": []}
{"id": "2503.19332", "title": "Divide-and-Conquer: Dual-Hierarchical Optimization for Semantic 4D\n  Gaussian Spatting", "abstract": "  Semantic 4D Gaussians can be used for reconstructing and understanding\ndynamic scenes, with temporal variations than static scenes. Directly applying\nstatic methods to understand dynamic scenes will fail to capture the temporal\nfeatures. Few works focus on dynamic scene understanding based on Gaussian\nSplatting, since once the same update strategy is employed for both dynamic and\nstatic parts, regardless of the distinction and interaction between Gaussians,\nsignificant artifacts and noise appear. We propose Dual-Hierarchical\nOptimization (DHO), which consists of Hierarchical Gaussian Flow and\nHierarchical Gaussian Guidance in a divide-and-conquer manner. The former\nimplements effective division of static and dynamic rendering and features. The\nlatter helps to mitigate the issue of dynamic foreground rendering distortion\nin textured complex scenes. Extensive experiments show that our method\nconsistently outperforms the baselines on both synthetic and real-world\ndatasets, and supports various downstream tasks. Project Page:\nhttps://sweety-yan.github.io/DHO.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.19340", "title": "BADGR: Bundle Adjustment Diffusion Conditioned by GRadients for\n  Wide-Baseline Floor Plan Reconstruction", "abstract": "  Reconstructing precise camera poses and floor plan layouts from wide-baseline\nRGB panoramas is a difficult and unsolved problem. We introduce BADGR, a novel\ndiffusion model that jointly performs reconstruction and bundle adjustment (BA)\nto refine poses and layouts from a coarse state, using 1D floor boundary\npredictions from dozens of images of varying input densities. Unlike a guided\ndiffusion model, BADGR is conditioned on dense per-entity outputs from a\nsingle-step Levenberg Marquardt (LM) optimizer and is trained to predict camera\nand wall positions while minimizing reprojection errors for view-consistency.\nThe objective of layout generation from denoising diffusion process complements\nBA optimization by providing additional learned layout-structural constraints\non top of the co-visible features across images. These constraints help BADGR\nto make plausible guesses on spatial relations which help constrain pose graph,\nsuch as wall adjacency, collinearity, and learn to mitigate errors from dense\nboundary observations with global contexts. BADGR trains exclusively on 2D\nfloor plans, simplifying data acquisition, enabling robust augmentation, and\nsupporting variety of input densities. Our experiments and analysis validate\nour method, which significantly outperforms the state-of-the-art pose and floor\nplan layout reconstruction with different input densities.\n", "date": "2025-03-25", "categories": []}
{"id": "2412.09401", "title": "SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos", "abstract": "  In this paper, we introduce SLAM3R, a novel and effective system for\nreal-time, high-quality, dense 3D reconstruction using RGB videos. SLAM3R\nprovides an end-to-end solution by seamlessly integrating local 3D\nreconstruction and global coordinate registration through feed-forward neural\nnetworks. Given an input video, the system first converts it into overlapping\nclips using a sliding window mechanism. Unlike traditional pose\noptimization-based methods, SLAM3R directly regresses 3D pointmaps from RGB\nimages in each window and progressively aligns and deforms these local\npointmaps to create a globally consistent scene reconstruction - all without\nexplicitly solving any camera parameters. Experiments across datasets\nconsistently show that SLAM3R achieves state-of-the-art reconstruction accuracy\nand completeness while maintaining real-time performance at 20+ FPS. Code\navailable at: https://github.com/PKU-VCL-3DV/SLAM3R.\n", "date": "2024-12-12", "categories": []}
{"id": "2501.03714", "title": "MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval\n  Adjustment for Compact Dynamic 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has made significant strides in scene\nrepresentation and neural rendering, with intense efforts focused on adapting\nit for dynamic scenes. Despite delivering remarkable rendering quality and\nspeed, existing methods struggle with storage demands and representing complex\nreal-world motions. To tackle these issues, we propose MoDecGS, a\nmemory-efficient Gaussian splatting framework designed for reconstructing novel\nviews in challenging scenarios with complex motions. We introduce GlobaltoLocal\nMotion Decomposition (GLMD) to effectively capture dynamic motions in a\ncoarsetofine manner. This approach leverages Global Canonical Scaffolds (Global\nCS) and Local Canonical Scaffolds (Local CS), extending static Scaffold\nrepresentation to dynamic video reconstruction. For Global CS, we propose\nGlobal Anchor Deformation (GAD) to efficiently represent global dynamics along\ncomplex motions, by directly deforming the implicit Scaffold attributes which\nare anchor position, offset, and local context features. Next, we finely adjust\nlocal motions via the Local Gaussian Deformation (LGD) of Local CS explicitly.\nAdditionally, we introduce Temporal Interval Adjustment (TIA) to automatically\ncontrol the temporal coverage of each Local CS during training, allowing\nMoDecGS to find optimal interval assignments based on the specified number of\ntemporal segments. Extensive evaluations demonstrate that MoDecGS achieves an\naverage 70% reduction in model size over stateoftheart methods for dynamic 3D\nGaussians from realworld dynamic videos while maintaining or even improving\nrendering quality.\n", "date": "2025-01-07", "categories": []}
{"id": "2503.12552", "title": "MTGS: Multi-Traversal Gaussian Splatting", "abstract": "  Multi-traversal data, commonly collected through daily commutes or by\nself-driving fleets, provides multiple viewpoints for scene reconstruction\nwithin a road block. This data offers significant potential for high-quality\nnovel view synthesis, which is crucial for applications such as autonomous\nvehicle simulators. However, inherent challenges in multi-traversal data often\nresult in suboptimal reconstruction quality, including variations in appearance\nand the presence of dynamic objects. To address these issues, we propose\nMulti-Traversal Gaussian Splatting (MTGS), a novel approach that reconstructs\nhigh-quality driving scenes from arbitrarily collected multi-traversal data by\nmodeling a shared static geometry while separately handling dynamic elements\nand appearance variations. Our method employs a multi-traversal dynamic scene\ngraph with a shared static node and traversal-specific dynamic nodes,\ncomplemented by color correction nodes with learnable spherical harmonics\ncoefficient residuals. This approach enables high-fidelity novel view synthesis\nand provides flexibility to navigate any viewpoint. We conduct extensive\nexperiments on a large-scale driving dataset, nuPlan, with multi-traversal\ndata. Our results demonstrate that MTGS improves LPIPS by 23.5% and geometry\naccuracy by 46.3% compared to single-traversal baselines. The code and data\nwould be available to the public.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.14912", "title": "Deep Polycuboid Fitting for Compact 3D Representation of Indoor Scenes", "abstract": "  This paper presents a novel framework for compactly representing a 3D indoor\nscene using a set of polycuboids through a deep learning-based fitting method.\nIndoor scenes mainly consist of man-made objects, such as furniture, which\noften exhibit rectilinear geometry. This property allows indoor scenes to be\nrepresented using combinations of polycuboids, providing a compact\nrepresentation that benefits downstream applications like furniture\nrearrangement. Our framework takes a noisy point cloud as input and first\ndetects six types of cuboid faces using a transformer network. Then, a graph\nneural network is used to validate the spatial relationships of the detected\nfaces to form potential polycuboids. Finally, each polycuboid instance is\nreconstructed by forming a set of boxes based on the aggregated face labels. To\ntrain our networks, we introduce a synthetic dataset encompassing a diverse\nrange of cuboid and polycuboid shapes that reflect the characteristics of\nindoor scenes. Our framework generalizes well to real-world indoor scene\ndatasets, including Replica, ScanNet, and scenes captured with an iPhone. The\nversatility of our method is demonstrated through practical applications, such\nas virtual room tours and scene editing.\n", "date": "2025-03-19", "categories": []}
{"id": "2503.17733", "title": "GS-LTS: 3D Gaussian Splatting-Based Adaptive Modeling for Long-Term\n  Service Robots", "abstract": "  3D Gaussian Splatting (3DGS) has garnered significant attention in robotics\nfor its explicit, high fidelity dense scene representation, demonstrating\nstrong potential for robotic applications. However, 3DGS-based methods in\nrobotics primarily focus on static scenes, with limited attention to the\ndynamic scene changes essential for long-term service robots. These robots\ndemand sustained task execution and efficient scene updates-challenges current\napproaches fail to meet. To address these limitations, we propose GS-LTS\n(Gaussian Splatting for Long-Term Service), a 3DGS-based system enabling indoor\nrobots to manage diverse tasks in dynamic environments over time. GS-LTS\ndetects scene changes (e.g., object addition or removal) via single-image\nchange detection, employs a rule-based policy to autonomously collect\nmulti-view observations, and efficiently updates the scene representation\nthrough Gaussian editing. Additionally, we propose a simulation-based benchmark\nthat automatically generates scene change data as compact configuration\nscripts, providing a standardized, user-friendly evaluation benchmark.\nExperimental results demonstrate GS-LTS's advantages in reconstruction,\nnavigation, and superior scene updates-faster and higher quality than the image\ntraining baseline-advancing 3DGS for long-term robotic operations. Code and\nbenchmark are available at: https://vipl-vsu.github.io/3DGS-LTS.\n", "date": "2025-03-22", "categories": []}
{"id": "2503.18073", "title": "PanopticSplatting: End-to-End Panoptic Gaussian Splatting", "abstract": "  Open-vocabulary panoptic reconstruction is a challenging task for\nsimultaneous scene reconstruction and understanding. Recently, methods have\nbeen proposed for 3D scene understanding based on Gaussian splatting. However,\nthese methods are multi-staged, suffering from the accumulated errors and the\ndependence of hand-designed components. To streamline the pipeline and achieve\nglobal optimization, we propose PanopticSplatting, an end-to-end system for\nopen-vocabulary panoptic reconstruction. Our method introduces query-guided\nGaussian segmentation with local cross attention, lifting 2D instance masks\nwithout cross-frame association in an end-to-end way. The local cross attention\nwithin view frustum effectively reduces the training memory, making our model\nmore accessible to large scenes with more Gaussians and objects. In addition,\nto address the challenge of noisy labels in 2D pseudo masks, we propose label\nblending to promote consistent 3D segmentation with less noisy floaters, as\nwell as label warping on 2D predictions which enhances multi-view coherence and\nsegmentation accuracy. Our method demonstrates strong performances in 3D scene\npanoptic reconstruction on the ScanNet-V2 and ScanNet++ datasets, compared with\nboth NeRF-based and Gaussian-based panoptic reconstruction methods. Moreover,\nPanopticSplatting can be easily generalized to numerous variants of Gaussian\nsplatting, and we demonstrate its robustness on different Gaussian base models.\n", "date": "2025-03-23", "categories": []}
{"id": "2503.18897", "title": "Online 3D Scene Reconstruction Using Neural Object Priors", "abstract": "  This paper addresses the problem of reconstructing a scene online at the\nlevel of objects given an RGB-D video sequence. While current object-aware\nneural implicit representations hold promise, they are limited in online\nreconstruction efficiency and shape completion. Our main contributions to\nalleviate the above limitations are twofold. First, we propose a feature grid\ninterpolation mechanism to continuously update grid-based object-centric neural\nimplicit representations as new object parts are revealed. Second, we construct\nan object library with previously mapped objects in advance and leverage the\ncorresponding shape priors to initialize geometric object models in new videos,\nsubsequently completing them with novel views as well as synthesized past views\nto avoid losing original object details. Extensive experiments on synthetic\nenvironments from the Replica dataset, real-world ScanNet sequences and videos\ncaptured in our laboratory demonstrate that our approach outperforms\nstate-of-the-art neural implicit models for this task in terms of\nreconstruction accuracy and completeness.\n", "date": "2025-03-24", "categories": []}
{"id": "2411.00239", "title": "Aquatic-GS: A Hybrid 3D Representation for Underwater Scenes", "abstract": "  Representing underwater 3D scenes is a valuable yet complex task, as\nattenuation and scattering effects during underwater imaging significantly\ncouple the information of the objects and the water. This coupling presents a\nsignificant challenge for existing methods in effectively representing both the\nobjects and the water medium simultaneously. To address this challenge, we\npropose Aquatic-GS, a hybrid 3D representation approach for underwater scenes\nthat effectively represents both the objects and the water medium.\nSpecifically, we construct a Neural Water Field (NWF) to implicitly model the\nwater parameters, while extending the latest 3D Gaussian Splatting (3DGS) to\nmodel the objects explicitly. Both components are integrated through a\nphysics-based underwater image formation model to represent complex underwater\nscenes. Moreover, to construct more precise scene geometry and details, we\ndesign a Depth-Guided Optimization (DGO) mechanism that uses a pseudo-depth map\nas auxiliary guidance. After optimization, Aquatic-GS enables the rendering of\nnovel underwater viewpoints and supports restoring the true appearance of\nunderwater scenes, as if the water medium were absent. Extensive experiments on\nboth simulated and real-world datasets demonstrate that Aquatic-GS surpasses\nstate-of-the-art underwater 3D representation methods, achieving better\nrendering quality and real-time rendering performance with a 410x increase in\nspeed. Furthermore, regarding underwater image restoration, Aquatic-GS\noutperforms representative dewatering methods in color correction, detail\nrecovery, and stability. Our models, code, and datasets can be accessed at\nhttps://aquaticgs.github.io.\n", "date": "2024-10-31", "categories": []}
{"id": "2412.03473", "title": "UrbanGS: Semantic-Guided Gaussian Splatting for Urban Scene\n  Reconstruction", "abstract": "  Reconstructing urban scenes is challenging due to their complex geometries\nand the presence of potentially dynamic objects. 3D Gaussian Splatting\n(3DGS)-based methods have shown strong performance, but existing approaches\noften incorporate manual 3D annotations to improve dynamic object modeling,\nwhich is impractical due to high labeling costs. Some methods leverage 4D\nGaussian Splatting (4DGS) to represent the entire scene, but they treat static\nand dynamic objects uniformly, leading to unnecessary updates for static\nelements and ultimately degrading reconstruction quality. To address these\nissues, we propose UrbanGS, which leverages 2D semantic maps and an existing\ndynamic Gaussian approach to distinguish static objects from the scene,\nenabling separate processing of definite static and potentially dynamic\nelements. Specifically, for definite static regions, we enforce global\nconsistency to prevent unintended changes in dynamic Gaussian and introduce a\nK-nearest neighbor (KNN)-based regularization to improve local coherence on\nlow-textured ground surfaces. Notably, for potentially dynamic objects, we\naggregate temporal information using learnable time embeddings, allowing each\nGaussian to model deformations over time. Extensive experiments on real-world\ndatasets demonstrate that our approach outperforms state-of-the-art methods in\nreconstruction quality and efficiency, accurately preserving static content\nwhile capturing dynamic elements.\n", "date": "2024-12-04", "categories": []}
{"id": "2503.16710", "title": "4D Gaussian Splatting SLAM", "abstract": "  Simultaneously localizing camera poses and constructing Gaussian radiance\nfields in dynamic scenes establish a crucial bridge between 2D images and the\n4D real world. Instead of removing dynamic objects as distractors and\nreconstructing only static environments, this paper proposes an efficient\narchitecture that incrementally tracks camera poses and establishes the 4D\nGaussian radiance fields in unknown scenarios by using a sequence of RGB-D\nimages. First, by generating motion masks, we obtain static and dynamic priors\nfor each pixel. To eliminate the influence of static scenes and improve the\nefficiency on learning the motion of dynamic objects, we classify the Gaussian\nprimitives into static and dynamic Gaussian sets, while the sparse control\npoints along with an MLP is utilized to model the transformation fields of the\ndynamic Gaussians. To more accurately learn the motion of dynamic Gaussians, a\nnovel 2D optical flow map reconstruction algorithm is designed to render\noptical flows of dynamic objects between neighbor images, which are further\nused to supervise the 4D Gaussian radiance fields along with traditional\nphotometric and geometric constraints. In experiments, qualitative and\nquantitative evaluation results show that the proposed method achieves robust\ntracking and high-quality view synthesis performance in real-world\nenvironments.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.16964", "title": "DroneSplat: 3D Gaussian Splatting for Robust 3D Reconstruction from\n  In-the-Wild Drone Imagery", "abstract": "  Drones have become essential tools for reconstructing wild scenes due to\ntheir outstanding maneuverability. Recent advances in radiance field methods\nhave achieved remarkable rendering quality, providing a new avenue for 3D\nreconstruction from drone imagery. However, dynamic distractors in wild\nenvironments challenge the static scene assumption in radiance fields, while\nlimited view constraints hinder the accurate capture of underlying scene\ngeometry. To address these challenges, we introduce DroneSplat, a novel\nframework designed for robust 3D reconstruction from in-the-wild drone imagery.\nOur method adaptively adjusts masking thresholds by integrating local-global\nsegmentation heuristics with statistical approaches, enabling precise\nidentification and elimination of dynamic distractors in static scenes. We\nenhance 3D Gaussian Splatting with multi-view stereo predictions and a\nvoxel-guided optimization strategy, supporting high-quality rendering under\nlimited view constraints. For comprehensive evaluation, we provide a\ndrone-captured 3D reconstruction dataset encompassing both dynamic and static\nscenes. Extensive experiments demonstrate that DroneSplat outperforms both 3DGS\nand NeRF baselines in handling in-the-wild drone imagery.\n", "date": "2025-03-21", "categories": []}
{"id": "2503.16979", "title": "Instant Gaussian Stream: Fast and Generalizable Streaming of Dynamic\n  Scene Reconstruction via Gaussian Splatting", "abstract": "  Building Free-Viewpoint Videos in a streaming manner offers the advantage of\nrapid responsiveness compared to offline training methods, greatly enhancing\nuser experience. However, current streaming approaches face challenges of high\nper-frame reconstruction time (10s+) and error accumulation, limiting their\nbroader application. In this paper, we propose Instant Gaussian Stream (IGS), a\nfast and generalizable streaming framework, to address these issues. First, we\nintroduce a generalized Anchor-driven Gaussian Motion Network, which projects\nmulti-view 2D motion features into 3D space, using anchor points to drive the\nmotion of all Gaussians. This generalized Network generates the motion of\nGaussians for each target frame in the time required for a single inference.\nSecond, we propose a Key-frame-guided Streaming Strategy that refines each key\nframe, enabling accurate reconstruction of temporally complex scenes while\nmitigating error accumulation. We conducted extensive in-domain and\ncross-domain evaluations, demonstrating that our approach can achieve streaming\nwith a average per-frame reconstruction time of 2s+, alongside a enhancement in\nview synthesis quality.\n", "date": "2025-03-21", "categories": []}
{"id": "2503.17316", "title": "Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene\n  Priors", "abstract": "  We present Pow3r, a novel large 3D vision regression model that is highly\nversatile in the input modalities it accepts. Unlike previous feed-forward\nmodels that lack any mechanism to exploit known camera or scene priors at test\ntime, Pow3r incorporates any combination of auxiliary information such as\nintrinsics, relative pose, dense or sparse depth, alongside input images,\nwithin a single network. Building upon the recent DUSt3R paradigm, a\ntransformer-based architecture that leverages powerful pre-training, our\nlightweight and versatile conditioning acts as additional guidance for the\nnetwork to predict more accurate estimates when auxiliary information is\navailable. During training we feed the model with random subsets of modalities\nat each iteration, which enables the model to operate under different levels of\nknown priors at test time. This in turn opens up new capabilities, such as\nperforming inference in native image resolution, or point-cloud completion. Our\nexperiments on 3D reconstruction, depth completion, multi-view depth\nprediction, multi-view stereo, and multi-view pose estimation tasks yield\nstate-of-the-art results and confirm the effectiveness of Pow3r at exploiting\nall available information. The project webpage is\nhttps://europe.naverlabs.com/pow3r.\n", "date": "2025-03-21", "categories": []}
{"id": "2501.13928", "title": "Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass", "abstract": "  Multi-view 3D reconstruction remains a core challenge in computer vision,\nparticularly in applications requiring accurate and scalable representations\nacross diverse perspectives. Current leading methods such as DUSt3R employ a\nfundamentally pairwise approach, processing images in pairs and necessitating\ncostly global alignment procedures to reconstruct from multiple views. In this\nwork, we propose Fast 3D Reconstruction (Fast3R), a novel multi-view\ngeneralization to DUSt3R that achieves efficient and scalable 3D reconstruction\nby processing many views in parallel. Fast3R's Transformer-based architecture\nforwards N images in a single forward pass, bypassing the need for iterative\nalignment. Through extensive experiments on camera pose estimation and 3D\nreconstruction, Fast3R demonstrates state-of-the-art performance, with\nsignificant improvements in inference speed and reduced error accumulation.\nThese results establish Fast3R as a robust alternative for multi-view\napplications, offering enhanced scalability without compromising reconstruction\naccuracy.\n", "date": "2025-01-23", "categories": []}
{"id": "2503.15835", "title": "BARD-GS: Blur-Aware Reconstruction of Dynamic Scenes via Gaussian\n  Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has shown remarkable potential for static scene\nreconstruction, and recent advancements have extended its application to\ndynamic scenes. However, the quality of reconstructions depends heavily on\nhigh-quality input images and precise camera poses, which are not that trivial\nto fulfill in real-world scenarios. Capturing dynamic scenes with handheld\nmonocular cameras, for instance, typically involves simultaneous movement of\nboth the camera and objects within a single exposure. This combined motion\nfrequently results in image blur that existing methods cannot adequately\nhandle. To address these challenges, we introduce BARD-GS, a novel approach for\nrobust dynamic scene reconstruction that effectively handles blurry inputs and\nimprecise camera poses. Our method comprises two main components: 1) camera\nmotion deblurring and 2) object motion deblurring. By explicitly decomposing\nmotion blur into camera motion blur and object motion blur and modeling them\nseparately, we achieve significantly improved rendering results in dynamic\nregions. In addition, we collect a real-world motion blur dataset of dynamic\nscenes to evaluate our approach. Extensive experiments demonstrate that BARD-GS\neffectively reconstructs high-quality dynamic scenes under realistic\nconditions, significantly outperforming existing methods.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.16318", "title": "Dynamic Point Maps: A Versatile Representation for Dynamic 3D\n  Reconstruction", "abstract": "  DUSt3R has recently shown that one can reduce many tasks in multi-view\ngeometry, including estimating camera intrinsics and extrinsics, reconstructing\nthe scene in 3D, and establishing image correspondences, to the prediction of a\npair of viewpoint-invariant point maps, i.e., pixel-aligned point clouds\ndefined in a common reference frame. This formulation is elegant and powerful,\nbut unable to tackle dynamic scenes. To address this challenge, we introduce\nthe concept of Dynamic Point Maps (DPM), extending standard point maps to\nsupport 4D tasks such as motion segmentation, scene flow estimation, 3D object\ntracking, and 2D correspondence. Our key intuition is that, when time is\nintroduced, there are several possible spatial and time references that can be\nused to define the point maps. We identify a minimal subset of such\ncombinations that can be regressed by a network to solve the sub tasks\nmentioned above. We train a DPM predictor on a mixture of synthetic and real\ndata and evaluate it across diverse benchmarks for video depth prediction,\ndynamic point cloud reconstruction, 3D scene flow and object pose tracking,\nachieving state-of-the-art performance. Code, models and additional results are\navailable at https://www.robots.ox.ac.uk/~vgg/research/dynamic-point-maps/.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.16422", "title": "1000+ FPS 4D Gaussian Splatting for Dynamic Scene Rendering", "abstract": "  4D Gaussian Splatting (4DGS) has recently gained considerable attention as a\nmethod for reconstructing dynamic scenes. Despite achieving superior quality,\n4DGS typically requires substantial storage and suffers from slow rendering\nspeed. In this work, we delve into these issues and identify two key sources of\ntemporal redundancy. (Q1) \\textbf{Short-Lifespan Gaussians}: 4DGS uses a large\nportion of Gaussians with short temporal span to represent scene dynamics,\nleading to an excessive number of Gaussians. (Q2) \\textbf{Inactive Gaussians}:\nWhen rendering, only a small subset of Gaussians contributes to each frame.\nDespite this, all Gaussians are processed during rasterization, resulting in\nredundant computation overhead. To address these redundancies, we present\n\\textbf{4DGS-1K}, which runs at over 1000 FPS on modern GPUs. For Q1, we\nintroduce the Spatial-Temporal Variation Score, a new pruning criterion that\neffectively removes short-lifespan Gaussians while encouraging 4DGS to capture\nscene dynamics using Gaussians with longer temporal spans. For Q2, we store a\nmask for active Gaussians across consecutive frames, significantly reducing\nredundant computations in rendering. Compared to vanilla 4DGS, our method\nachieves a $41\\times$ reduction in storage and $9\\times$ faster rasterization\nspeed on complex dynamic scenes, while maintaining comparable visual quality.\nPlease see our project page at https://4DGS-1K.github.io.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.13265", "title": "FlexWorld: Progressively Expanding 3D Scenes for Flexiable-View\n  Synthesis", "abstract": "  Generating flexible-view 3D scenes, including 360{\\deg} rotation and zooming,\nfrom single images is challenging due to a lack of 3D data. To this end, we\nintroduce FlexWorld, a novel framework consisting of two key components: (1) a\nstrong video-to-video (V2V) diffusion model to generate high-quality novel view\nimages from incomplete input rendered from a coarse scene, and (2) a\nprogressive expansion process to construct a complete 3D scene. In particular,\nleveraging an advanced pre-trained video model and accurate depth-estimated\ntraining pairs, our V2V model can generate novel views under large camera pose\nvariations. Building upon it, FlexWorld progressively generates new 3D content\nand integrates it into the global scene through geometry-aware scene fusion.\nExtensive experiments demonstrate the effectiveness of FlexWorld in generating\nhigh-quality novel view videos and flexible-view 3D scenes from single images,\nachieving superior visual quality under multiple popular metrics and datasets\ncompared to existing state-of-the-art methods. Qualitatively, we highlight that\nFlexWorld can generate high-fidelity scenes with flexible views like 360{\\deg}\nrotations and zooming. Project page: https://ml-gsai.github.io/FlexWorld.\n", "date": "2025-03-17", "categories": []}
{"id": "2503.14830", "title": "Decompositional Neural Scene Reconstruction with Generative Diffusion\n  Prior", "abstract": "  Decompositional reconstruction of 3D scenes, with complete shapes and\ndetailed texture of all objects within, is intriguing for downstream\napplications but remains challenging, particularly with sparse views as input.\nRecent approaches incorporate semantic or geometric regularization to address\nthis issue, but they suffer significant degradation in underconstrained areas\nand fail to recover occluded regions. We argue that the key to solving this\nproblem lies in supplementing missing information for these areas. To this end,\nwe propose DP-Recon, which employs diffusion priors in the form of Score\nDistillation Sampling (SDS) to optimize the neural representation of each\nindividual object under novel views. This provides additional information for\nthe underconstrained areas, but directly incorporating diffusion prior raises\npotential conflicts between the reconstruction and generative guidance.\nTherefore, we further introduce a visibility-guided approach to dynamically\nadjust the per-pixel SDS loss weights. Together these components enhance both\ngeometry and appearance recovery while remaining faithful to input images.\nExtensive experiments across Replica and ScanNet++ demonstrate that our method\nsignificantly outperforms SOTA methods. Notably, it achieves better object\nreconstruction under 10 views than the baselines under 100 views. Our method\nenables seamless text-based editing for geometry and appearance through SDS\noptimization and produces decomposed object meshes with detailed UV maps that\nsupport photorealistic Visual effects (VFX) editing. The project page is\navailable at https://dp-recon.github.io/.\n", "date": "2025-03-19", "categories": []}
{"id": "2309.15329", "title": "BASED: Bundle-Adjusting Surgical Endoscopic Dynamic Video Reconstruction\n  using Neural Radiance Fields", "abstract": "  Reconstruction of deformable scenes from endoscopic videos is important for\nmany applications such as intraoperative navigation, surgical visual\nperception, and robotic surgery. It is a foundational requirement for realizing\nautonomous robotic interventions for minimally invasive surgery. However,\nprevious approaches in this domain have been limited by their modular nature\nand are confined to specific camera and scene settings. Our work adopts the\nNeural Radiance Fields (NeRF) approach to learning 3D implicit representations\nof scenes that are both dynamic and deformable over time, and furthermore with\nunknown camera poses. We demonstrate this approach on endoscopic surgical\nscenes from robotic surgery. This work removes the constraints of known camera\nposes and overcomes the drawbacks of the state-of-the-art unstructured dynamic\nscene reconstruction technique, which relies on the static part of the scene\nfor accurate reconstruction. Through several experimental datasets, we\ndemonstrate the versatility of our proposed model to adapt to diverse camera\nand scene settings, and show its promise for both current and future robotic\nsurgical systems.\n", "date": "2023-09-27", "categories": []}
{"id": "2403.10344", "title": "ViiNeuS: Volumetric Initialization for Implicit Neural Surface\n  reconstruction of urban scenes with limited image overlap", "abstract": "  Neural implicit surface representation methods have recently shown impressive\n3D reconstruction results. However, existing solutions struggle to reconstruct\ndriving scenes due to their large size, highly complex nature and their limited\nvisual observation overlap. Hence, to achieve accurate reconstructions,\nadditional supervision data such as LiDAR, strong geometric priors, and long\ntraining times are required. To tackle such limitations, we present ViiNeuS, a\nnew hybrid implicit surface learning method that efficiently initializes the\nsigned distance field to reconstruct large driving scenes from 2D street view\nimages. ViiNeuS's hybrid architecture models two separate implicit fields: one\nrepresenting the volumetric density of the scene, and another one representing\nthe signed distance to the surface. To accurately reconstruct urban outdoor\ndriving scenarios, we introduce a novel volume-rendering strategy that relies\non self-supervised probabilistic density estimation to sample points near the\nsurface and transition progressively from volumetric to surface representation.\nOur solution permits a proper and fast initialization of the signed distance\nfield without relying on any geometric prior on the scene, compared to\nconcurrent methods. By conducting extensive experiments on four outdoor driving\ndatasets, we show that ViiNeuS can learn an accurate and detailed 3D surface\nrepresentation of various urban scene while being two times faster to train\ncompared to previous state-of-the-art solutions.\n", "date": "2024-03-15", "categories": []}
{"id": "2503.14219", "title": "Segmentation-Guided Neural Radiance Fields for Novel Street View\n  Synthesis", "abstract": "  Recent advances in Neural Radiance Fields (NeRF) have shown great potential\nin 3D reconstruction and novel view synthesis, particularly for indoor and\nsmall-scale scenes. However, extending NeRF to large-scale outdoor environments\npresents challenges such as transient objects, sparse cameras and textures, and\nvarying lighting conditions. In this paper, we propose a segmentation-guided\nenhancement to NeRF for outdoor street scenes, focusing on complex urban\nenvironments. Our approach extends ZipNeRF and utilizes Grounded SAM for\nsegmentation mask generation, enabling effective handling of transient objects,\nmodeling of the sky, and regularization of the ground. We also introduce\nappearance embeddings to adapt to inconsistent lighting across view sequences.\nExperimental results demonstrate that our method outperforms the baseline\nZipNeRF, improving novel view synthesis quality with fewer artifacts and\nsharper details.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14274", "title": "Improving Adaptive Density Control for 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has become one of the most influential works in\nthe past year. Due to its efficient and high-quality novel view synthesis\ncapabilities, it has been widely adopted in many research fields and\napplications. Nevertheless, 3DGS still faces challenges to properly manage the\nnumber of Gaussian primitives that are used during scene reconstruction.\nFollowing the adaptive density control (ADC) mechanism of 3D Gaussian\nSplatting, new Gaussians in under-reconstructed regions are created, while\nGaussians that do not contribute to the rendering quality are pruned. We\nobserve that those criteria for densifying and pruning Gaussians can sometimes\nlead to worse rendering by introducing artifacts. We especially observe\nunder-reconstructed background or overfitted foreground regions. To encounter\nboth problems, we propose three new improvements to the adaptive density\ncontrol mechanism. Those include a correction for the scene extent calculation\nthat does not only rely on camera positions, an exponentially ascending\ngradient threshold to improve training convergence, and significance-aware\npruning strategy to avoid background artifacts. With these adaptions, we show\nthat the rendering quality improves while using the same number of Gaussians\nprimitives. Furthermore, with our improvements, the training converges\nconsiderably faster, allowing for more than twice as fast training times while\nyielding better quality than 3DGS. Finally, our contributions are easily\ncompatible with most existing derivative works of 3DGS making them relevant for\nfuture works.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14483", "title": "Multi-view Reconstruction via SfM-guided Monocular Depth Estimation", "abstract": "  In this paper, we present a new method for multi-view geometric\nreconstruction. In recent years, large vision models have rapidly developed,\nperforming excellently across various tasks and demonstrating remarkable\ngeneralization capabilities. Some works use large vision models for monocular\ndepth estimation, which have been applied to facilitate multi-view\nreconstruction tasks in an indirect manner. Due to the ambiguity of the\nmonocular depth estimation task, the estimated depth values are usually not\naccurate enough, limiting their utility in aiding multi-view reconstruction. We\npropose to incorporate SfM information, a strong multi-view prior, into the\ndepth estimation process, thus enhancing the quality of depth prediction and\nenabling their direct application in multi-view geometric reconstruction.\nExperimental results on public real-world datasets show that our method\nsignificantly improves the quality of depth estimation compared to previous\nmonocular depth estimation works. Additionally, we evaluate the reconstruction\nquality of our approach in various types of scenes including indoor,\nstreetscape, and aerial views, surpassing state-of-the-art MVS methods. The\ncode and supplementary materials are available at\nhttps://zju3dv.github.io/murre/ .\n", "date": "2025-03-18", "categories": []}
{"id": "2312.00846", "title": "NeuSG: Neural Implicit Surface Reconstruction with 3D Gaussian Splatting\n  Guidance", "abstract": "  Existing neural implicit surface reconstruction methods have achieved\nimpressive performance in multi-view 3D reconstruction by leveraging explicit\ngeometry priors such as depth maps or point clouds as regularization. However,\nthe reconstruction results still lack fine details because of the over-smoothed\ndepth map or sparse point cloud. In this work, we propose a neural implicit\nsurface reconstruction pipeline with guidance from 3D Gaussian Splatting to\nrecover highly detailed surfaces. The advantage of 3D Gaussian Splatting is\nthat it can generate dense point clouds with detailed structure. Nonetheless, a\nnaive adoption of 3D Gaussian Splatting can fail since the generated points are\nthe centers of 3D Gaussians that do not necessarily lie on the surface. We thus\nintroduce a scale regularizer to pull the centers close to the surface by\nenforcing the 3D Gaussians to be extremely thin. Moreover, we propose to refine\nthe point cloud from 3D Gaussians Splatting with the normal priors from the\nsurface predicted by neural implicit models instead of using a fixed set of\npoints as guidance. Consequently, the quality of surface reconstruction\nimproves from the guidance of the more accurate 3D Gaussian splatting. By\njointly optimizing the 3D Gaussian Splatting and the neural implicit model, our\napproach benefits from both representations and generates complete surfaces\nwith intricate details. Experiments on Tanks and Temples verify the\neffectiveness of our proposed method.\n", "date": "2023-12-01", "categories": []}
{"id": "2411.19492", "title": "Diorama: Unleashing Zero-shot Single-view 3D Indoor Scene Modeling", "abstract": "  Reconstructing structured 3D scenes from RGB images using CAD objects unlocks\nefficient and compact scene representations that maintain compositionality and\ninteractability. Existing works propose training-heavy methods relying on\neither expensive yet inaccurate real-world annotations or controllable yet\nmonotonous synthetic data that do not generalize well to unseen objects or\ndomains. We present Diorama, the first zero-shot open-world system that\nholistically models 3D scenes from single-view RGB observations without\nrequiring end-to-end training or human annotations. We show the feasibility of\nour approach by decomposing the problem into subtasks and introduce robust,\ngeneralizable solutions to each: architecture reconstruction, 3D shape\nretrieval, object pose estimation, and scene layout optimization. We evaluate\nour system on both synthetic and real-world data to show we significantly\noutperform baselines from prior work. We also demonstrate generalization to\ninternet images and the text-to-scene task.\n", "date": "2024-11-29", "categories": []}
{"id": "2502.03465", "title": "Seeing World Dynamics in a Nutshell", "abstract": "  We consider the problem of efficiently representing casually captured\nmonocular videos in a spatially- and temporally-coherent manner. While existing\napproaches predominantly rely on 2D/2.5D techniques treating videos as\ncollections of spatiotemporal pixels, they struggle with complex motions,\nocclusions, and geometric consistency due to absence of temporal coherence and\nexplicit 3D structure. Drawing inspiration from monocular video as a projection\nof the dynamic 3D world, we explore representing videos in their intrinsic 3D\nform through continuous flows of Gaussian primitives in space-time. In this\npaper, we propose NutWorld, a novel framework that efficiently transforms\nmonocular videos into dynamic 3D Gaussian representations in a single forward\npass. At its core, NutWorld introduces a structured spatial-temporal aligned\nGaussian (STAG) representation, enabling optimization-free scene modeling with\neffective depth and flow regularization. Through comprehensive experiments, we\ndemonstrate that NutWorld achieves high-fidelity video reconstruction quality\nwhile enabling various downstream applications in real-time. Demos and code\nwill be available at https://github.com/Nut-World/NutWorld.\n", "date": "2025-02-05", "categories": []}
{"id": "2503.12307", "title": "Swift4D:Adaptive divide-and-conquer Gaussian Splatting for compact and\n  efficient reconstruction of dynamic scene", "abstract": "  Novel view synthesis has long been a practical but challenging task, although\nthe introduction of numerous methods to solve this problem, even combining\nadvanced representations like 3D Gaussian Splatting, they still struggle to\nrecover high-quality results and often consume too much storage memory and\ntraining time. In this paper we propose Swift4D, a divide-and-conquer 3D\nGaussian Splatting method that can handle static and dynamic primitives\nseparately, achieving a good trade-off between rendering quality and\nefficiency, motivated by the fact that most of the scene is the static\nprimitive and does not require additional dynamic properties. Concretely, we\nfocus on modeling dynamic transformations only for the dynamic primitives which\nbenefits both efficiency and quality. We first employ a learnable decomposition\nstrategy to separate the primitives, which relies on an additional parameter to\nclassify primitives as static or dynamic. For the dynamic primitives, we employ\na compact multi-resolution 4D Hash mapper to transform these primitives from\ncanonical space into deformation space at each timestamp, and then mix the\nstatic and dynamic primitives to produce the final output. This\ndivide-and-conquer method facilitates efficient training and reduces storage\nredundancy. Our method not only achieves state-of-the-art rendering quality\nwhile being 20X faster in training than previous SOTA methods with a minimum\nstorage requirement of only 30MB on real-world datasets. Code is available at\nhttps://github.com/WuJH2001/swift4d.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12492", "title": "Geometry-Aware Face Reconstruction Under Occluded Scenes", "abstract": "  Recently, deep learning-based 3D face reconstruction methods have\ndemonstrated promising advancements in terms of quality and efficiency.\nNevertheless, these techniques face challenges in effectively handling occluded\nscenes and fail to capture intricate geometric facial details. Inspired by the\nprinciples of GANs and bump mapping, we have successfully addressed these\nissues. Our approach aims to deliver comprehensive 3D facial reconstructions,\neven in the presence of occlusions.While maintaining the overall shape's\nrobustness, we introduce a mid-level shape refinement to the fundamental\nstructure. Furthermore, we illustrate how our method adeptly extends to\ngenerate plausible details for obscured facial regions. We offer numerous\nexamples that showcase the effectiveness of our framework in producing\nrealistic results, where traditional methods often struggle. To substantiate\nthe superior adaptability of our approach, we have conducted extensive\nexperiments in the context of general 3D face reconstruction tasks, serving as\nconcrete evidence of its regulatory prowess compared to manual occlusion\nremoval methods.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12535", "title": "SPC-GS: Gaussian Splatting with Semantic-Prompt Consistency for Indoor\n  Open-World Free-view Synthesis from Sparse Inputs", "abstract": "  3D Gaussian Splatting-based indoor open-world free-view synthesis approaches\nhave shown significant performance with dense input images. However, they\nexhibit poor performance when confronted with sparse inputs, primarily due to\nthe sparse distribution of Gaussian points and insufficient view supervision.\nTo relieve these challenges, we propose SPC-GS, leveraging Scene-layout-based\nGaussian Initialization (SGI) and Semantic-Prompt Consistency (SPC)\nRegularization for open-world free view synthesis with sparse inputs.\nSpecifically, SGI provides a dense, scene-layout-based Gaussian distribution by\nutilizing view-changed images generated from the video generation model and\nview-constraint Gaussian points densification. Additionally, SPC mitigates\nlimited view supervision by employing semantic-prompt-based consistency\nconstraints developed by SAM2. This approach leverages available semantics from\ntraining views, serving as instructive prompts, to optimize visually\noverlapping regions in novel views with 2D and 3D consistency constraints.\nExtensive experiments demonstrate the superior performance of SPC-GS across\nReplica and ScanNet benchmarks. Notably, our SPC-GS achieves a 3.06 dB gain in\nPSNR for reconstruction quality and a 7.3% improvement in mIoU for open-world\nsemantic segmentation.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.11172", "title": "Uncertainty-Aware Normal-Guided Gaussian Splatting for Surface\n  Reconstruction from Sparse Image Sequences", "abstract": "  3D Gaussian Splatting (3DGS) has achieved impressive rendering performance in\nnovel view synthesis. However, its efficacy diminishes considerably in sparse\nimage sequences, where inherent data sparsity amplifies geometric uncertainty\nduring optimization. This often leads to convergence at suboptimal local\nminima, resulting in noticeable structural artifacts in the reconstructed\nscenes.To mitigate these issues, we propose Uncertainty-aware Normal-Guided\nGaussian Splatting (UNG-GS), a novel framework featuring an explicit Spatial\nUncertainty Field (SUF) to quantify geometric uncertainty within the 3DGS\npipeline. UNG-GS enables high-fidelity rendering and achieves high-precision\nreconstruction without relying on priors. Specifically, we first integrate\nGaussian-based probabilistic modeling into the training of 3DGS to optimize the\nSUF, providing the model with adaptive error tolerance. An uncertainty-aware\ndepth rendering strategy is then employed to weight depth contributions based\non the SUF, effectively reducing noise while preserving fine details.\nFurthermore, an uncertainty-guided normal refinement method adjusts the\ninfluence of neighboring depth values in normal estimation, promoting robust\nresults. Extensive experiments demonstrate that UNG-GS significantly\noutperforms state-of-the-art methods in both sparse and dense sequences. The\ncode will be open-source.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.11651", "title": "VGGT: Visual Geometry Grounded Transformer", "abstract": "  We present VGGT, a feed-forward neural network that directly infers all key\n3D attributes of a scene, including camera parameters, point maps, depth maps,\nand 3D point tracks, from one, a few, or hundreds of its views. This approach\nis a step forward in 3D computer vision, where models have typically been\nconstrained to and specialized for single tasks. It is also simple and\nefficient, reconstructing images in under one second, and still outperforming\nalternatives that require post-processing with visual geometry optimization\ntechniques. The network achieves state-of-the-art results in multiple 3D tasks,\nincluding camera parameter estimation, multi-view depth estimation, dense point\ncloud reconstruction, and 3D point tracking. We also show that using pretrained\nVGGT as a feature backbone significantly enhances downstream tasks, such as\nnon-rigid point tracking and feed-forward novel view synthesis. Code and models\nare publicly available at https://github.com/facebookresearch/vggt.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.08093", "title": "MVGSR: Multi-View Consistency Gaussian Splatting for Robust Surface\n  Reconstruction", "abstract": "  3D Gaussian Splatting (3DGS) has gained significant attention for its\nhigh-quality rendering capabilities, ultra-fast training, and inference speeds.\nHowever, when we apply 3DGS to surface reconstruction tasks, especially in\nenvironments with dynamic objects and distractors, the method suffers from\nfloating artifacts and color errors due to inconsistency from different\nviewpoints. To address this challenge, we propose Multi-View Consistency\nGaussian Splatting for the domain of Robust Surface Reconstruction\n(\\textbf{MVGSR}), which takes advantage of lightweight Gaussian models and a\n{heuristics-guided distractor masking} strategy for robust surface\nreconstruction in non-static environments. Compared to existing methods that\nrely on MLPs for distractor segmentation strategies, our approach separates\ndistractors from static scene elements by comparing multi-view feature\nconsistency, allowing us to obtain precise distractor masks early in training.\nFurthermore, we introduce a pruning measure based on multi-view contributions\nto reset transmittance, effectively reducing floating artifacts. Finally, a\nmulti-view consistency loss is applied to achieve high-quality performance in\nsurface reconstruction tasks. Experimental results demonstrate that MVGSR\nachieves competitive geometric accuracy and rendering fidelity compared to the\nstate-of-the-art surface reconstruction algorithms. More information is\navailable on our project page (https://mvgsr.github.io).\n", "date": "2025-03-11", "categories": []}
{"id": "2503.10286", "title": "VicaSplat: A Single Run is All You Need for 3D Gaussian Splatting and\n  Camera Estimation from Unposed Video Frames", "abstract": "  We present VicaSplat, a novel framework for joint 3D Gaussians reconstruction\nand camera pose estimation from a sequence of unposed video frames, which is a\ncritical yet underexplored task in real-world 3D applications. The core of our\nmethod lies in a novel transformer-based network architecture. In particular,\nour model starts with an image encoder that maps each image to a list of visual\ntokens. All visual tokens are concatenated with additional inserted learnable\ncamera tokens. The obtained tokens then fully communicate with each other\nwithin a tailored transformer decoder. The camera tokens causally aggregate\nfeatures from visual tokens of different views, and further modulate them\nframe-wisely to inject view-dependent features. 3D Gaussian splats and camera\npose parameters can then be estimated via different prediction heads.\nExperiments show that VicaSplat surpasses baseline methods for multi-view\ninputs, and achieves comparable performance to prior two-view approaches.\nRemarkably, VicaSplat also demonstrates exceptional cross-dataset\ngeneralization capability on the ScanNet benchmark, achieving superior\nperformance without any fine-tuning. Project page:\nhttps://lizhiqi49.github.io/VicaSplat.\n", "date": "2025-03-13", "categories": []}
{"id": "2503.10464", "title": "Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within\n  Unified Neural Representations", "abstract": "  Learning accurate scene reconstruction without pose priors in neural radiance\nfields is challenging due to inherent geometric ambiguity. Recent development\neither relies on correspondence priors for regularization or uses off-the-shelf\nflow estimators to derive analytical poses. However, the potential for jointly\nlearning scene geometry, camera poses, and dense flow within a unified neural\nrepresentation remains largely unexplored. In this paper, we present Flow-NeRF,\na unified framework that simultaneously optimizes scene geometry, camera poses,\nand dense optical flow all on-the-fly. To enable the learning of dense flow\nwithin the neural radiance field, we design and build a bijective mapping for\nflow estimation, conditioned on pose. To make the scene reconstruction benefit\nfrom the flow estimation, we develop an effective feature enhancement mechanism\nto pass canonical space features to world space representations, significantly\nenhancing scene geometry. We validate our model across four important tasks,\ni.e., novel view synthesis, depth estimation, camera pose prediction, and dense\noptical flow estimation, using several datasets. Our approach surpasses\nprevious methods in almost all metrics for novel-view view synthesis and depth\nestimation and yields both qualitatively sound and quantitatively accurate\nnovel-view flow. Our project page is https://zhengxunzhi.github.io/flownerf/.\n", "date": "2025-03-13", "categories": []}
{"id": "2503.10604", "title": "MuDG: Taming Multi-modal Diffusion with Gaussian Splatting for Urban\n  Scene Reconstruction", "abstract": "  Recent breakthroughs in radiance fields have significantly advanced 3D scene\nreconstruction and novel view synthesis (NVS) in autonomous driving.\nNevertheless, critical limitations persist: reconstruction-based methods\nexhibit substantial performance deterioration under significant viewpoint\ndeviations from training trajectories, while generation-based techniques\nstruggle with temporal coherence and precise scene controllability. To overcome\nthese challenges, we present MuDG, an innovative framework that integrates\nMulti-modal Diffusion model with Gaussian Splatting (GS) for Urban Scene\nReconstruction. MuDG leverages aggregated LiDAR point clouds with RGB and\ngeometric priors to condition a multi-modal video diffusion model, synthesizing\nphotorealistic RGB, depth, and semantic outputs for novel viewpoints. This\nsynthesis pipeline enables feed-forward NVS without computationally intensive\nper-scene optimization, providing comprehensive supervision signals to refine\n3DGS representations for rendering robustness enhancement under extreme\nviewpoint changes. Experiments on the Open Waymo Dataset demonstrate that MuDG\noutperforms existing methods in both reconstruction and synthesis quality.\n", "date": "2025-03-13", "categories": []}
{"id": "2406.16038", "title": "LiveScene: Language Embedding Interactive Radiance Fields for Physical\n  Scene Rendering and Control", "abstract": "  This paper scales object-level reconstruction to complex scenes, advancing\ninteractive scene reconstruction. We introduce two datasets, OmniSim and\nInterReal, featuring 28 scenes with multiple interactive objects. To tackle the\nchallenge of inaccurate interactive motion recovery in complex scenes, we\npropose LiveScene, a scene-level language-embedded interactive radiance field\nthat efficiently reconstructs and controls multiple objects. By decomposing the\ninteractive scene into local deformable fields, LiveScene enables separate\nreconstruction of individual object motions, reducing memory consumption.\nAdditionally, our interaction-aware language embedding localizes individual\ninteractive objects, allowing for arbitrary control using natural language. Our\napproach demonstrates significant superiority in novel view synthesis,\ninteractive scene control, and language grounding performance through extensive\nexperiments. Project page: https://livescenes.github.io.\n", "date": "2024-06-23", "categories": []}
{"id": "2503.08929", "title": "HessianForge: Scalable LiDAR reconstruction with Physics-Informed Neural\n  Representation and Smoothness Energy Constraints", "abstract": "  Accurate and efficient 3D mapping of large-scale outdoor environments from\nLiDAR measurements is a fundamental challenge in robotics, particularly towards\nensuring smooth and artifact-free surface reconstructions. Although the\nstate-of-the-art methods focus on memory-efficient neural representations for\nhigh-fidelity surface generation, they often fail to produce artifact-free\nmanifolds, with artifacts arising due to noisy and sparse inputs. To address\nthis issue, we frame surface mapping as a physics-informed energy optimization\nproblem, enforcing surface smoothness by optimizing an energy functional that\npenalizes sharp surface ridges. Specifically, we propose a deep learning based\napproach that learns the signed distance field (SDF) of the surface manifold\nfrom raw LiDAR point clouds using a physics-informed loss function that\noptimizes the $L_2$-Hessian energy of the surface. Our learning framework\nincludes a hierarchical octree based input feature encoding and a multi-scale\nneural network to iteratively refine the signed distance field at different\nscales of resolution. Lastly, we introduce a test-time refinement strategy to\ncorrect topological inconsistencies and edge distortions that can arise in the\ngenerated mesh. We propose a \\texttt{CUDA}-accelerated least-squares\noptimization that locally adjusts vertex positions to enforce\nfeature-preserving smoothing. We evaluate our approach on large-scale outdoor\ndatasets and demonstrate that our approach outperforms current state-of-the-art\nmethods in terms of improved accuracy and smoothness. Our code is available at\n\\href{https://github.com/HrishikeshVish/HessianForge/}{https://github.com/HrishikeshVish/HessianForge/}\n", "date": "2025-03-11", "categories": []}
{"id": "2503.09332", "title": "SDD-4DGS: Static-Dynamic Aware Decoupling in Gaussian Splatting for 4D\n  Scene Reconstruction", "abstract": "  Dynamic and static components in scenes often exhibit distinct properties,\nyet most 4D reconstruction methods treat them indiscriminately, leading to\nsuboptimal performance in both cases. This work introduces SDD-4DGS, the first\nframework for static-dynamic decoupled 4D scene reconstruction based on\nGaussian Splatting. Our approach is built upon a novel probabilistic dynamic\nperception coefficient that is naturally integrated into the Gaussian\nreconstruction pipeline, enabling adaptive separation of static and dynamic\ncomponents. With carefully designed implementation strategies to realize this\ntheoretical framework, our method effectively facilitates explicit learning of\nmotion patterns for dynamic elements while maintaining geometric stability for\nstatic structures. Extensive experiments on five benchmark datasets demonstrate\nthat SDD-4DGS consistently outperforms state-of-the-art methods in\nreconstruction fidelity, with enhanced detail restoration for static structures\nand precise modeling of dynamic motions. The code will be released.\n", "date": "2025-03-12", "categories": []}
{"id": "2403.16862", "title": "INPC: Implicit Neural Point Clouds for Radiance Field Rendering", "abstract": "  We introduce a new approach for reconstruction and novel view synthesis of\nunbounded real-world scenes. In contrast to previous methods using either\nvolumetric fields, grid-based models, or discrete point cloud proxies, we\npropose a hybrid scene representation, which implicitly encodes the geometry in\na continuous octree-based probability field and view-dependent appearance in a\nmulti-resolution hash grid. This allows for extraction of arbitrary explicit\npoint clouds, which can be rendered using rasterization. In doing so, we\ncombine the benefits of both worlds and retain favorable behavior during\noptimization: Our novel implicit point cloud representation and differentiable\nbilinear rasterizer enable fast rendering while preserving the fine geometric\ndetail captured by volumetric neural fields. Furthermore, this representation\ndoes not depend on priors like structure-from-motion point clouds. Our method\nachieves state-of-the-art image quality on common benchmarks. Furthermore, we\nachieve fast inference at interactive frame rates, and can convert our trained\nmodel into a large, explicit point cloud to further enhance performance.\n", "date": "2024-03-25", "categories": []}
{"id": "2503.07739", "title": "SIRE: SE(3) Intrinsic Rigidity Embeddings", "abstract": "  Motion serves as a powerful cue for scene perception and understanding by\nseparating independently moving surfaces and organizing the physical world into\ndistinct entities. We introduce SIRE, a self-supervised method for motion\ndiscovery of objects and dynamic scene reconstruction from casual scenes by\nlearning intrinsic rigidity embeddings from videos. Our method trains an image\nencoder to estimate scene rigidity and geometry, supervised by a simple 4D\nreconstruction loss: a least-squares solver uses the estimated geometry and\nrigidity to lift 2D point track trajectories into SE(3) tracks, which are\nsimply re-projected back to 2D and compared against the original 2D\ntrajectories for supervision. Crucially, our framework is fully end-to-end\ndifferentiable and can be optimized either on video datasets to learn\ngeneralizable image priors, or even on a single video to capture scene-specific\nstructure - highlighting strong data efficiency. We demonstrate the\neffectiveness of our rigidity embeddings and geometry across multiple settings,\nincluding downstream object segmentation, SE(3) rigid motion estimation, and\nself-supervised depth estimation. Our findings suggest that SIRE can learn\nstrong geometry and motion rigidity priors from video data, with minimal\nsupervision.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.08166", "title": "Dynamic Scene Reconstruction: Recent Advance in Real-time Rendering and\n  Streaming", "abstract": "  Representing and rendering dynamic scenes from 2D images is a fundamental yet\nchallenging problem in computer vision and graphics. This survey provides a\ncomprehensive review of the evolution and advancements in dynamic scene\nrepresentation and rendering, with a particular emphasis on recent progress in\nNeural Radiance Fields based and 3D Gaussian Splatting based reconstruction\nmethods. We systematically summarize existing approaches, categorize them\naccording to their core principles, compile relevant datasets, compare the\nperformance of various methods on these benchmarks, and explore the challenges\nand future research directions in this rapidly evolving field. In total, we\nreview over 170 relevant papers, offering a broad perspective on the state of\nthe art in this domain.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.08382", "title": "Twinner: Shining Light on Digital Twins in a Few Snaps", "abstract": "  We present the first large reconstruction model, Twinner, capable of\nrecovering a scene's illumination as well as an object's geometry and material\nproperties from only a few posed images. Twinner is based on the Large\nReconstruction Model and innovates in three key ways: 1) We introduce a\nmemory-efficient voxel-grid transformer whose memory scales only quadratically\nwith the size of the voxel grid. 2) To deal with scarcity of high-quality\nground-truth PBR-shaded models, we introduce a large fully-synthetic dataset of\nprocedurally-generated PBR-textured objects lit with varied illumination. 3) To\nnarrow the synthetic-to-real gap, we finetune the model on real life datasets\nby means of a differentiable physically-based shading model, eschewing the need\nfor ground-truth illumination or material properties which are challenging to\nobtain in real life. We demonstrate the efficacy of our model on the real life\nStanfordORB benchmark where, given few input views, we achieve reconstruction\nquality significantly superior to existing feedforward reconstruction networks,\nand comparable to significantly slower per-scene optimization methods.\n", "date": "2025-03-11", "categories": []}
{"id": "2410.15636", "title": "LucidFusion: Reconstructing 3D Gaussians with Arbitrary Unposed Images", "abstract": "  Recent large reconstruction models have made notable progress in generating\nhigh-quality 3D objects from single images. However, current reconstruction\nmethods often rely on explicit camera pose estimation or fixed viewpoints,\nrestricting their flexibility and practical applicability. We reformulate 3D\nreconstruction as image-to-image translation and introduce the Relative\nCoordinate Map (RCM), which aligns multiple unposed images to a main view\nwithout pose estimation. While RCM simplifies the process, its lack of global\n3D supervision can yield noisy outputs. To address this, we propose Relative\nCoordinate Gaussians (RCG) as an extension to RCM, which treats each pixel's\ncoordinates as a Gaussian center and employs differentiable rasterization for\nconsistent geometry and pose recovery. Our LucidFusion framework handles an\narbitrary number of unposed inputs, producing robust 3D reconstructions within\nseconds and paving the way for more flexible, pose-free 3D pipelines.\n", "date": "2024-10-21", "categories": []}
{"id": "2411.14514", "title": "NexusSplats: Efficient 3D Gaussian Splatting in the Wild", "abstract": "  Photorealistic 3D reconstruction of unstructured real-world scenes remains\nchallenging due to complex illumination variations and transient occlusions.\nExisting methods based on Neural Radiance Fields (NeRF) and 3D Gaussian\nSplatting (3DGS) struggle with inefficient light decoupling and\nstructure-agnostic occlusion handling. To address these limitations, we propose\nNexusSplats, an approach tailored for efficient and high-fidelity 3D scene\nreconstruction under complex lighting and occlusion conditions. In particular,\nNexusSplats leverages a hierarchical light decoupling strategy that performs\ncentralized appearance learning, efficiently and effectively decoupling varying\nlighting conditions. Furthermore, a structure-aware occlusion handling\nmechanism is developed, establishing a nexus between 3D and 2D structures for\nfine-grained occlusion handling. Experimental results demonstrate that\nNexusSplats achieves state-of-the-art rendering quality and reduces the number\nof total parameters by 65.4\\%, leading to 2.7$\\times$ faster reconstruction.\n", "date": "2024-11-21", "categories": []}
{"id": "2412.00155", "title": "T-3DGS: Removing Transient Objects for 3D Scene Reconstruction", "abstract": "  Transient objects in video sequences can significantly degrade the quality of\n3D scene reconstructions. To address this challenge, we propose T-3DGS, a novel\nframework that robustly filters out transient distractors during 3D\nreconstruction using Gaussian Splatting. Our framework consists of two steps.\nFirst, we employ an unsupervised classification network that distinguishes\ntransient objects from static scene elements by leveraging their distinct\ntraining dynamics within the reconstruction process. Second, we refine these\ninitial detections by integrating an off-the-shelf segmentation method with a\nbidirectional tracking module, which together enhance boundary accuracy and\ntemporal coherence. Evaluations on both sparsely and densely captured video\ndatasets demonstrate that T-3DGS significantly outperforms state-of-the-art\napproaches, enabling high-fidelity 3D reconstructions in challenging,\nreal-world scenarios.\n", "date": "2024-11-29", "categories": []}
{"id": "2412.01807", "title": "Occam's LGS: An Efficient Approach for Language Gaussian Splatting", "abstract": "  TL;DR: Gaussian Splatting is a widely adopted approach for 3D scene\nrepresentation, offering efficient, high-quality reconstruction and rendering.\nA key reason for its success is the simplicity of representing scenes with sets\nof Gaussians, making it interpretable and adaptable. To enhance understanding\nbeyond visual representation, recent approaches extend Gaussian Splatting with\nsemantic vision-language features, enabling open-set tasks. Typically, these\nlanguage features are aggregated from multiple 2D views, however, existing\nmethods rely on cumbersome techniques, resulting in high computational costs\nand longer training times.\n  In this work, we show that the complicated pipelines for language 3D Gaussian\nSplatting are simply unnecessary. Instead, we follow a probabilistic\nformulation of Language Gaussian Splatting and apply Occam's razor to the task\nat hand, leading to a highly efficient weighted multi-view feature aggregation\ntechnique. Doing so offers us state-of-the-art results with a speed-up of two\norders of magnitude without any compression, allowing for easy scene\nmanipulation. Project Page: https://insait-institute.github.io/OccamLGS/\n", "date": "2024-12-02", "categories": []}
{"id": "2412.19089", "title": "Humans as a Calibration Pattern: Dynamic 3D Scene Reconstruction from\n  Unsynchronized and Uncalibrated Videos", "abstract": "  Recent works on dynamic 3D neural field reconstruction assume the input from\nsynchronized multi-view videos whose poses are known. The input constraints are\noften not satisfied in real-world setups, making the approach impractical. We\nshow that unsynchronized videos from unknown poses can generate dynamic neural\nfields as long as the videos capture human motion. Humans are one of the most\ncommon dynamic subjects captured in videos, and their shapes and poses can be\nestimated using state-of-the-art libraries. While noisy, the estimated human\nshape and pose parameters provide a decent initialization point to start the\nhighly non-convex and under-constrained problem of training a consistent\ndynamic neural representation. Given the shape and pose parameters of humans in\nindividual frames, we formulate methods to calculate the time offsets between\nvideos, followed by camera pose estimations that analyze the 3D joint\npositions. Then, we train the dynamic neural fields employing multiresolution\ngrids while we concurrently refine both time offsets and camera poses. The\nsetup still involves optimizing many parameters; therefore, we introduce a\nrobust progressive learning strategy to stabilize the process. Experiments show\nthat our approach achieves accurate spatio-temporal calibration and\nhigh-quality scene reconstruction in challenging conditions.\n", "date": "2024-12-26", "categories": []}
{"id": "2501.09347", "title": "UVRM: A Scalable 3D Reconstruction Model from Unposed Videos", "abstract": "  Large Reconstruction Models (LRMs) have recently become a popular method for\ncreating 3D foundational models. Training 3D reconstruction models with 2D\nvisual data traditionally requires prior knowledge of camera poses for the\ntraining samples, a process that is both time-consuming and prone to errors.\nConsequently, 3D reconstruction training has been confined to either synthetic\n3D datasets or small-scale datasets with annotated poses. In this study, we\ninvestigate the feasibility of 3D reconstruction using unposed video data of\nvarious objects. We introduce UVRM, a novel 3D reconstruction model capable of\nbeing trained and evaluated on monocular videos without requiring any\ninformation about the pose. UVRM uses a transformer network to implicitly\naggregate video frames into a pose-invariant latent feature space, which is\nthen decoded into a tri-plane 3D representation. To obviate the need for\nground-truth pose annotations during training, UVRM employs a combination of\nthe score distillation sampling (SDS) method and an analysis-by-synthesis\napproach, progressively synthesizing pseudo novel-views using a pre-trained\ndiffusion model. We qualitatively and quantitatively evaluate UVRM's\nperformance on the G-Objaverse and CO3D datasets without relying on pose\ninformation. Extensive experiments show that UVRM is capable of effectively and\nefficiently reconstructing a wide range of 3D objects from unposed videos.\n", "date": "2025-01-16", "categories": []}
{"id": "2503.06569", "title": "Global-Aware Monocular Semantic Scene Completion with State Space Models", "abstract": "  Monocular Semantic Scene Completion (MonoSSC) reconstructs and interprets 3D\nenvironments from a single image, enabling diverse real-world applications.\nHowever, existing methods are often constrained by the local receptive field of\nConvolutional Neural Networks (CNNs), making it challenging to handle the\nnon-uniform distribution of projected points (Fig. \\ref{fig:perspective}) and\neffectively reconstruct missing information caused by the 3D-to-2D projection.\nIn this work, we introduce GA-MonoSSC, a hybrid architecture for MonoSSC that\neffectively captures global context in both the 2D image domain and 3D space.\nSpecifically, we propose a Dual-Head Multi-Modality Encoder, which leverages a\nTransformer architecture to capture spatial relationships across all features\nin the 2D image domain, enabling more comprehensive 2D feature extraction.\nAdditionally, we introduce the Frustum Mamba Decoder, built on the State Space\nModel (SSM), to efficiently capture long-range dependencies in 3D space.\nFurthermore, we propose a frustum reordering strategy within the Frustum Mamba\nDecoder to mitigate feature discontinuities in the reordered voxel sequence,\nensuring better alignment with the scan mechanism of the State Space Model\n(SSM) for improved 3D representation learning. We conduct extensive experiments\non the widely used Occ-ScanNet and NYUv2 datasets, demonstrating that our\nproposed method achieves state-of-the-art performance, validating its\neffectiveness. The code will be released upon acceptance.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.06818", "title": "Sub-Image Recapture for Multi-View 3D Reconstruction", "abstract": "  3D reconstruction of high-resolution target remains a challenge task due to\nthe large memory required from the large input image size. Recently developed\nlearning based algorithms provide promising reconstruction performance than\ntraditional ones, however, they generally require more memory than the\ntraditional algorithms and facing scalability issue. In this paper, we\ndeveloped a generic approach, sub-image recapture (SIR), to split large image\ninto smaller sub-images and process them individually. As a result of this\nframework, the existing 3D reconstruction algorithms can be implemented based\non sub-image recapture with significantly reduced memory and substantially\nimproved scalability\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07476", "title": "SOGS: Second-Order Anchor for Advanced 3D Gaussian Splatting", "abstract": "  Anchor-based 3D Gaussian splatting (3D-GS) exploits anchor features in 3D\nGaussian prediction, which has achieved impressive 3D rendering quality with\nreduced Gaussian redundancy. On the other hand, it often encounters the dilemma\namong anchor features, model size, and rendering quality - large anchor\nfeatures lead to large 3D models and high-quality rendering whereas reducing\nanchor features degrades Gaussian attribute prediction which leads to clear\nartifacts in the rendered textures and geometries. We design SOGS, an\nanchor-based 3D-GS technique that introduces second-order anchors to achieve\nsuperior rendering quality and reduced anchor features and model size\nsimultaneously. Specifically, SOGS incorporates covariance-based second-order\nstatistics and correlation across feature dimensions to augment features within\neach anchor, compensating for the reduced feature size and improving rendering\nquality effectively. In addition, it introduces a selective gradient loss to\nenhance the optimization of scene textures and scene geometries, leading to\nhigh-quality rendering with small anchor features. Extensive experiments over\nmultiple widely adopted benchmarks show that SOGS achieves superior rendering\nquality in novel view synthesis with clearly reduced model size.\n", "date": "2025-03-10", "categories": []}
{"id": "2412.06234", "title": "Generative Densification: Learning to Densify Gaussians for\n  High-Fidelity Generalizable 3D Reconstruction", "abstract": "  Generalized feed-forward Gaussian models have achieved significant progress\nin sparse-view 3D reconstruction by leveraging prior knowledge from large\nmulti-view datasets. However, these models often struggle to represent\nhigh-frequency details due to the limited number of Gaussians. While the\ndensification strategy used in per-scene 3D Gaussian splatting (3D-GS)\noptimization can be adapted to the feed-forward models, it may not be ideally\nsuited for generalized scenarios. In this paper, we propose Generative\nDensification, an efficient and generalizable method to densify Gaussians\ngenerated by feed-forward models. Unlike the 3D-GS densification strategy,\nwhich iteratively splits and clones raw Gaussian parameters, our method\nup-samples feature representations from the feed-forward models and generates\ntheir corresponding fine Gaussians in a single forward pass, leveraging the\nembedded prior knowledge for enhanced generalization. Experimental results on\nboth object-level and scene-level reconstruction tasks demonstrate that our\nmethod outperforms state-of-the-art approaches with comparable or smaller model\nsizes, achieving notable improvements in representing fine details.\n", "date": "2024-12-09", "categories": []}
{"id": "2503.05086", "title": "Fake It To Make It: Virtual Multiviews to Enhance Monocular Indoor\n  Semantic Scene Completion", "abstract": "  Monocular Indoor Semantic Scene Completion (SSC) aims to reconstruct a 3D\nsemantic occupancy map from a single RGB image of an indoor scene, inferring\nspatial layout and object categories from 2D image cues. The challenge of this\ntask arises from the depth, scale, and shape ambiguities that emerge when\ntransforming a 2D image into 3D space, particularly within the complex and\noften heavily occluded environments of indoor scenes. Current SSC methods often\nstruggle with these ambiguities, resulting in distorted or missing object\nrepresentations. To overcome these limitations, we introduce an innovative\napproach that leverages novel view synthesis and multiview fusion.\nSpecifically, we demonstrate how virtual cameras can be placed around the scene\nto emulate multiview inputs that enhance contextual scene information. We also\nintroduce a Multiview Fusion Adaptor (MVFA) to effectively combine the\nmultiview 3D scene predictions into a unified 3D semantic occupancy map.\nFinally, we identify and study the inherent limitation of generative techniques\nwhen applied to SSC, specifically the Novelty-Consistency tradeoff. Our system,\nGenFuSE, demonstrates IoU score improvements of up to 2.8% for Scene Completion\nand 4.9% for Semantic Scene Completion when integrated with existing SSC\nnetworks on the NYUv2 dataset. This work introduces GenFuSE as a standard\nframework for advancing monocular SSC with synthesized inputs.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05162", "title": "EvolvingGS: High-Fidelity Streamable Volumetric Video via Evolving 3D\n  Gaussian Representation", "abstract": "  We have recently seen great progress in 3D scene reconstruction through\nexplicit point-based 3D Gaussian Splatting (3DGS), notable for its high quality\nand fast rendering speed. However, reconstructing dynamic scenes such as\ncomplex human performances with long durations remains challenging. Prior\nefforts fall short of modeling a long-term sequence with drastic motions,\nfrequent topology changes or interactions with props, and resort to segmenting\nthe whole sequence into groups of frames that are processed independently,\nwhich undermines temporal stability and thereby leads to an unpleasant viewing\nexperience and inefficient storage footprint. In view of this, we introduce\nEvolvingGS, a two-stage strategy that first deforms the Gaussian model to\ncoarsely align with the target frame, and then refines it with minimal point\naddition/subtraction, particularly in fast-changing areas. Owing to the\nflexibility of the incrementally evolving representation, our method\noutperforms existing approaches in terms of both per-frame and temporal quality\nmetrics while maintaining fast rendering through its purely explicit\nrepresentation. Moreover, by exploiting temporal coherence between successive\nframes, we propose a simple yet effective compression algorithm that achieves\nover 50x compression rate. Extensive experiments on both public benchmarks and\nchallenging custom datasets demonstrate that our method significantly advances\nthe state-of-the-art in dynamic scene reconstruction, particularly for extended\nsequences with complex human performances.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05484", "title": "DecoupledGaussian: Object-Scene Decoupling for Physics-Based Interaction", "abstract": "  We present DecoupledGaussian, a novel system that decouples static objects\nfrom their contacted surfaces captured in-the-wild videos, a key prerequisite\nfor realistic Newtonian-based physical simulations. Unlike prior methods\nfocused on synthetic data or elastic jittering along the contact surface, which\nprevent objects from fully detaching or moving independently, DecoupledGaussian\nallows for significant positional changes without being constrained by the\ninitial contacted surface. Recognizing the limitations of current 2D inpainting\ntools for restoring 3D locations, our approach proposes joint Poisson fields to\nrepair and expand the Gaussians of both objects and contacted scenes after\nseparation. This is complemented by a multi-carve strategy to refine the\nobject's geometry. Our system enables realistic simulations of decoupling\nmotions, collisions, and fractures driven by user-specified impulses,\nsupporting complex interactions within and across multiple scenes. We validate\nDecoupledGaussian through a comprehensive user study and quantitative\nbenchmarks. This system enhances digital interaction with objects and scenes in\nreal-world environments, benefiting industries such as VR, robotics, and\nautonomous driving. Our project page is at:\nhttps://wangmiaowei.github.io/DecoupledGaussian.github.io/.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.03115", "title": "NTR-Gaussian: Nighttime Dynamic Thermal Reconstruction with 4D Gaussian\n  Splatting Based on Thermodynamics", "abstract": "  Thermal infrared imaging offers the advantage of all-weather capability,\nenabling non-intrusive measurement of an object's surface temperature.\nConsequently, thermal infrared images are employed to reconstruct 3D models\nthat accurately reflect the temperature distribution of a scene, aiding in\napplications such as building monitoring and energy management. However,\nexisting approaches predominantly focus on static 3D reconstruction for a\nsingle time period, overlooking the impact of environmental factors on thermal\nradiation and failing to predict or analyze temperature variations over time.\nTo address these challenges, we propose the NTR-Gaussian method, which treats\ntemperature as a form of thermal radiation, incorporating elements like\nconvective heat transfer and radiative heat dissipation. Our approach utilizes\nneural networks to predict thermodynamic parameters such as emissivity,\nconvective heat transfer coefficient, and heat capacity. By integrating these\npredictions, we can accurately forecast thermal temperatures at various times\nthroughout a nighttime scene. Furthermore, we introduce a dynamic dataset\nspecifically for nighttime thermal imagery. Extensive experiments and\nevaluations demonstrate that NTR-Gaussian significantly outperforms comparison\nmethods in thermal reconstruction, achieving a predicted temperature error\nwithin 1 degree Celsius.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03664", "title": "A Generative Approach to High Fidelity 3D Reconstruction from Text Data", "abstract": "  The convergence of generative artificial intelligence and advanced computer\nvision technologies introduces a groundbreaking approach to transforming\ntextual descriptions into three-dimensional representations. This research\nproposes a fully automated pipeline that seamlessly integrates text-to-image\ngeneration, various image processing techniques, and deep learning methods for\nreflection removal and 3D reconstruction. By leveraging state-of-the-art\ngenerative models like Stable Diffusion, the methodology translates natural\nlanguage inputs into detailed 3D models through a multi-stage workflow.\n  The reconstruction process begins with the generation of high-quality images\nfrom textual prompts, followed by enhancement by a reinforcement learning agent\nand reflection removal using the Stable Delight model. Advanced image upscaling\nand background removal techniques are then applied to further enhance visual\nfidelity. These refined two-dimensional representations are subsequently\ntransformed into volumetric 3D models using sophisticated machine learning\nalgorithms, capturing intricate spatial relationships and geometric\ncharacteristics. This process achieves a highly structured and detailed output,\nensuring that the final 3D models reflect both semantic accuracy and geometric\nprecision.\n  This approach addresses key challenges in generative reconstruction, such as\nmaintaining semantic coherence, managing geometric complexity, and preserving\ndetailed visual information. Comprehensive experimental evaluations will assess\nreconstruction quality, semantic accuracy, and geometric fidelity across\ndiverse domains and varying levels of complexity. By demonstrating the\npotential of AI-driven 3D reconstruction techniques, this research offers\nsignificant implications for fields such as augmented reality (AR), virtual\nreality (VR), and digital content creation.\n", "date": "2025-03-05", "categories": []}
{"id": "2404.18669", "title": "Bootstrap-GS: Self-Supervised Augmentation for High-Fidelity Gaussian\n  Splatting", "abstract": "  Recent advancements in 3D Gaussian Splatting (3D-GS) have established new\nbenchmarks for rendering quality and efficiency in 3D reconstruction. However,\n3D-GS faces critical limitations when generating novel views that significantly\ndeviate from those encountered during training. Moreover, issues such as\ndilation and aliasing arise during zoom operations. These challenges stem from\na fundamental issue: training sampling deficiency. In this paper, we introduce\na bootstrapping framework to address this problem. Our approach synthesizes\npseudo-ground truth from novel views that align with the limited training set\nand reintegrates these synthesized views into the training pipeline.\nExperimental results demonstrate that our bootstrapping technique not only\nreduces artifacts but also improves quantitative metrics. Furthermore, our\ntechnique is highly adaptable, allowing various Gaussian-based method to\nbenefit from its integration.\n", "date": "2024-04-29", "categories": []}
{"id": "2502.16779", "title": "Unposed Sparse Views Room Layout Reconstruction in the Age of Pretrain\n  Model", "abstract": "  Room layout estimation from multiple-perspective images is poorly\ninvestigated due to the complexities that emerge from multi-view geometry,\nwhich requires muti-step solutions such as camera intrinsic and extrinsic\nestimation, image matching, and triangulation. However, in 3D reconstruction,\nthe advancement of recent 3D foundation models such as DUSt3R has shifted the\nparadigm from the traditional multi-step structure-from-motion process to an\nend-to-end single-step approach. To this end, we introduce Plane-DUSt3R, a\nnovel method for multi-view room layout estimation leveraging the 3D foundation\nmodel DUSt3R. Plane-DUSt3R incorporates the DUSt3R framework and fine-tunes on\na room layout dataset (Structure3D) with a modified objective to estimate\nstructural planes. By generating uniform and parsimonious results, Plane-DUSt3R\nenables room layout estimation with only a single post-processing step and 2D\ndetection results. Unlike previous methods that rely on single-perspective or\npanorama image, Plane-DUSt3R extends the setting to handle multiple-perspective\nimages. Moreover, it offers a streamlined, end-to-end solution that simplifies\nthe process and reduces error accumulation. Experimental results demonstrate\nthat Plane-DUSt3R not only outperforms state-of-the-art methods on the\nsynthetic dataset but also proves robust and effective on in the wild data with\ndifferent image styles such as cartoon. Our code is available at:\nhttps://github.com/justacar/Plane-DUSt3R\n", "date": "2025-02-24", "categories": []}
{"id": "2404.12379", "title": "Dynamic Gaussians Mesh: Consistent Mesh Reconstruction from Dynamic\n  Scenes", "abstract": "  Modern 3D engines and graphics pipelines require mesh as a memory-efficient\nrepresentation, which allows efficient rendering, geometry processing, texture\nediting, and many other downstream operations. However, it is still highly\ndifficult to obtain high-quality mesh in terms of detailed structure and time\nconsistency from dynamic observations. To this end, we introduce Dynamic\nGaussians Mesh (DG-Mesh), a framework to reconstruct a high-fidelity and\ntime-consistent mesh from dynamic input. Our work leverages the recent\nadvancement in 3D Gaussian Splatting to construct the mesh sequence with\ntemporal consistency from dynamic observations. Building on top of this\nrepresentation, DG-Mesh recovers high-quality meshes from the Gaussian points\nand can track the mesh vertices over time, which enables applications such as\ntexture editing on dynamic objects. We introduce the Gaussian-Mesh Anchoring,\nwhich encourages evenly distributed Gaussians, resulting better mesh\nreconstruction through mesh-guided densification and pruning on the deformed\nGaussians. By applying cycle-consistent deformation between the canonical and\nthe deformed space, we can project the anchored Gaussian back to the canonical\nspace and optimize Gaussians across all time frames. During the evaluation on\ndifferent datasets, DG-Mesh provides significantly better mesh reconstruction\nand rendering than baselines. Project page: https://www.liuisabella.com/DG-Mesh\n", "date": "2024-04-18", "categories": []}
{"id": "2410.11019", "title": "ET-Former: Efficient Triplane Deformable Attention for 3D Semantic Scene\n  Completion From Monocular Camera", "abstract": "  We introduce ET-Former, a novel end-to-end algorithm for semantic scene\ncompletion using a single monocular camera. Our approach generates a semantic\noccupancy map from single RGB observation while simultaneously providing\nuncertainty estimates for semantic predictions. By designing a triplane-based\ndeformable attention mechanism, our approach improves geometric understanding\nof the scene than other SOTA approaches and reduces noise in semantic\npredictions. Additionally, through the use of a Conditional Variational\nAutoEncoder (CVAE), we estimate the uncertainties of these predictions. The\ngenerated semantic and uncertainty maps will help formulate navigation\nstrategies that facilitate safe and permissible decision making in the future.\nEvaluated on the Semantic-KITTI dataset, ET-Former achieves the highest\nIntersection over Union (IoU) and mean IoU (mIoU) scores while maintaining the\nlowest GPU memory usage, surpassing state-of-the-art (SOTA) methods. It\nimproves the SOTA scores of IoU from 44.71 to 51.49 and mIoU from 15.04 to\n16.30 on SeamnticKITTI test, with a notably low training memory consumption of\n10.9 GB. Project page: https://github.com/jingGM/ET-Former.git.\n", "date": "2024-10-14", "categories": []}
{"id": "2502.21093", "title": "FlexDrive: Toward Trajectory Flexibility in Driving Scene Reconstruction\n  and Rendering", "abstract": "  Driving scene reconstruction and rendering have advanced significantly using\nthe 3D Gaussian Splatting. However, most prior research has focused on the\nrendering quality along a pre-recorded vehicle path and struggles to generalize\nto out-of-path viewpoints, which is caused by the lack of high-quality\nsupervision in those out-of-path views. To address this issue, we introduce an\nInverse View Warping technique to create compact and high-quality images as\nsupervision for the reconstruction of the out-of-path views, enabling\nhigh-quality rendering results for those views. For accurate and robust inverse\nview warping, a depth bootstrap strategy is proposed to obtain on-the-fly dense\ndepth maps during the optimization process, overcoming the sparsity and\nincompleteness of LiDAR depth data. Our method achieves superior in-path and\nout-of-path reconstruction and rendering performance on the widely used Waymo\nOpen dataset. In addition, a simulator-based benchmark is proposed to obtain\nthe out-of-path ground truth and quantitatively evaluate the performance of\nout-of-path rendering, where our method outperforms previous methods by a\nsignificant margin.\n", "date": "2025-02-28", "categories": []}
{"id": "2503.00726", "title": "Enhancing Monocular 3D Scene Completion with Diffusion Model", "abstract": "  3D scene reconstruction is essential for applications in virtual reality,\nrobotics, and autonomous driving, enabling machines to understand and interact\nwith complex environments. Traditional 3D Gaussian Splatting techniques rely on\nimages captured from multiple viewpoints to achieve optimal performance, but\nthis dependence limits their use in scenarios where only a single image is\navailable. In this work, we introduce FlashDreamer, a novel approach for\nreconstructing a complete 3D scene from a single image, significantly reducing\nthe need for multi-view inputs. Our approach leverages a pre-trained\nvision-language model to generate descriptive prompts for the scene, guiding a\ndiffusion model to produce images from various perspectives, which are then\nfused to form a cohesive 3D reconstruction. Extensive experiments show that our\nmethod effectively and robustly expands single-image inputs into a\ncomprehensive 3D scene, extending monocular 3D reconstruction capabilities\nwithout further training. Our code is available\nhttps://github.com/CharlieSong1999/FlashDreamer/tree/main.\n", "date": "2025-03-02", "categories": []}
{"id": "2503.00853", "title": "MTReD: 3D Reconstruction Dataset for Fly-over Videos of Maritime Domain", "abstract": "  This work tackles 3D scene reconstruction for a video fly-over perspective\nproblem in the maritime domain, with a specific emphasis on geometrically and\nvisually sound reconstructions. This will allow for downstream tasks such as\nsegmentation, navigation, and localization. To our knowledge, there is no\ndataset available in this domain. As such, we propose a novel maritime 3D scene\nreconstruction benchmarking dataset, named as MTReD (Maritime Three-Dimensional\nReconstruction Dataset). The MTReD comprises 19 fly-over videos curated from\nthe Internet containing ships, islands, and coastlines. As the task is aimed\ntowards geometrical consistency and visual completeness, the dataset uses two\nmetrics: (1) Reprojection error; and (2) Perception based metrics. We find that\nexisting perception-based metrics, such as Learned Perceptual Image Patch\nSimilarity (LPIPS), do not appropriately measure the completeness of a\nreconstructed image. Thus, we propose a novel semantic similarity metric\nutilizing DINOv2 features coined DiFPS (DinoV2 Features Perception Similarity).\nWe perform initial evaluation on two baselines: (1) Structured from Motion\n(SfM) through Colmap; and (2) the recent state-of-the-art MASt3R model. We find\nthat the reconstructed scenes by MASt3R have higher reprojection errors, but\nsuperior perception based metric scores. To this end, some pre-processing\nmethods are explored, and we find a pre-processing method which improves both\nthe reprojection error and perception-based score. We envisage our proposed\nMTReD to stimulate further research in these directions. The dataset and all\nthe code will be made available in https://github.com/RuiYiYong/MTReD.\n", "date": "2025-03-02", "categories": []}
{"id": "2503.01661", "title": "MUSt3R: Multi-view Network for Stereo 3D Reconstruction", "abstract": "  DUSt3R introduced a novel paradigm in geometric computer vision by proposing\na model that can provide dense and unconstrained Stereo 3D Reconstruction of\narbitrary image collections with no prior information about camera calibration\nnor viewpoint poses. Under the hood, however, DUSt3R processes image pairs,\nregressing local 3D reconstructions that need to be aligned in a global\ncoordinate system. The number of pairs, growing quadratically, is an inherent\nlimitation that becomes especially concerning for robust and fast optimization\nin the case of large image collections. In this paper, we propose an extension\nof DUSt3R from pairs to multiple views, that addresses all aforementioned\nconcerns. Indeed, we propose a Multi-view Network for Stereo 3D Reconstruction,\nor MUSt3R, that modifies the DUSt3R architecture by making it symmetric and\nextending it to directly predict 3D structure for all views in a common\ncoordinate frame. Second, we entail the model with a multi-layer memory\nmechanism which allows to reduce the computational complexity and to scale the\nreconstruction to large collections, inferring thousands of 3D pointmaps at\nhigh frame-rates with limited added complexity. The framework is designed to\nperform 3D reconstruction both offline and online, and hence can be seamlessly\napplied to SfM and visual SLAM scenarios showing state-of-the-art performance\non various 3D downstream tasks, including uncalibrated Visual Odometry,\nrelative camera pose, scale and focal estimation, 3D reconstruction and\nmulti-view depth estimation.\n", "date": "2025-03-03", "categories": []}
{"id": "2503.01774", "title": "Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models", "abstract": "  Neural Radiance Fields and 3D Gaussian Splatting have revolutionized 3D\nreconstruction and novel-view synthesis task. However, achieving photorealistic\nrendering from extreme novel viewpoints remains challenging, as artifacts\npersist across representations. In this work, we introduce Difix3D+, a novel\npipeline designed to enhance 3D reconstruction and novel-view synthesis through\nsingle-step diffusion models. At the core of our approach is Difix, a\nsingle-step image diffusion model trained to enhance and remove artifacts in\nrendered novel views caused by underconstrained regions of the 3D\nrepresentation. Difix serves two critical roles in our pipeline. First, it is\nused during the reconstruction phase to clean up pseudo-training views that are\nrendered from the reconstruction and then distilled back into 3D. This greatly\nenhances underconstrained regions and improves the overall 3D representation\nquality. More importantly, Difix also acts as a neural enhancer during\ninference, effectively removing residual artifacts arising from imperfect 3D\nsupervision and the limited capacity of current reconstruction models. Difix3D+\nis a general solution, a single model compatible with both NeRF and 3DGS\nrepresentations, and it achieves an average 2$\\times$ improvement in FID score\nover baselines while maintaining 3D consistency.\n", "date": "2025-03-03", "categories": []}
{"id": "2411.00705", "title": "ReMatching Dynamic Reconstruction Flow", "abstract": "  Reconstructing a dynamic scene from image inputs is a fundamental computer\nvision task with many downstream applications. Despite recent advancements,\nexisting approaches still struggle to achieve high-quality reconstructions from\nunseen viewpoints and timestamps. This work introduces the ReMatching\nframework, designed to improve reconstruction quality by incorporating\ndeformation priors into dynamic reconstruction models. Our approach advocates\nfor velocity-field based priors, for which we suggest a matching procedure that\ncan seamlessly supplement existing dynamic reconstruction pipelines. The\nframework is highly adaptable and can be applied to various dynamic\nrepresentations. Moreover, it supports integrating multiple types of model\npriors and enables combining simpler ones to create more complex classes. Our\nevaluations on popular benchmarks involving both synthetic and real-world\ndynamic scenes demonstrate that augmenting current state-of-the-art methods\nwith our approach leads to a clear improvement in reconstruction accuracy.\n", "date": "2024-11-01", "categories": []}
{"id": "2502.20861", "title": "MESC-3D:Mining Effective Semantic Cues for 3D Reconstruction from a\n  Single Image", "abstract": "  Reconstructing 3D shapes from a single image plays an important role in\ncomputer vision. Many methods have been proposed and achieve impressive\nperformance. However, existing methods mainly focus on extracting semantic\ninformation from images and then simply concatenating it with 3D point clouds\nwithout further exploring the concatenated semantics. As a result, these\nentangled semantic features significantly hinder the reconstruction\nperformance. In this paper, we propose a novel single-image 3D reconstruction\nmethod called Mining Effective Semantic Cues for 3D Reconstruction from a\nSingle Image (MESC-3D), which can actively mine effective semantic cues from\nentangled features. Specifically, we design an Effective Semantic Mining Module\nto establish connections between point clouds and image semantic attributes,\nenabling the point clouds to autonomously select the necessary information.\nFurthermore, to address the potential insufficiencies in semantic information\nfrom a single image, such as occlusions, inspired by the human ability to\nrepresent 3D objects using prior knowledge drawn from daily experiences, we\nintroduce a 3D Semantic Prior Learning Module. This module incorporates\nsemantic understanding of spatial structures, enabling the model to interpret\nand reconstruct 3D objects with greater accuracy and realism, closely mirroring\nhuman perception of complex 3D environments. Extensive evaluations show that\nour method achieves significant improvements in reconstruction quality and\nrobustness compared to prior works. Additionally, further experiments validate\nthe strong generalization capabilities and excels in zero-shot preformance on\nunseen classes. Code is available at https://github.com/QINGQINGLE/MESC-3D.\n", "date": "2025-02-28", "categories": []}
{"id": "2412.06273", "title": "Omni-Scene: Omni-Gaussian Representation for Ego-Centric Sparse-View\n  Scene Reconstruction", "abstract": "  Prior works employing pixel-based Gaussian representation have demonstrated\nefficacy in feed-forward sparse-view reconstruction. However, such\nrepresentation necessitates cross-view overlap for accurate depth estimation,\nand is challenged by object occlusions and frustum truncations. As a result,\nthese methods require scene-centric data acquisition to maintain cross-view\noverlap and complete scene visibility to circumvent occlusions and truncations,\nwhich limits their applicability to scene-centric reconstruction. In contrast,\nin autonomous driving scenarios, a more practical paradigm is ego-centric\nreconstruction, which is characterized by minimal cross-view overlap and\nfrequent occlusions and truncations. The limitations of pixel-based\nrepresentation thus hinder the utility of prior works in this task. In light of\nthis, this paper conducts an in-depth analysis of different representations,\nand introduces Omni-Gaussian representation with tailored network design to\ncomplement their strengths and mitigate their drawbacks. Experiments show that\nour method significantly surpasses state-of-the-art methods, pixelSplat and\nMVSplat, in ego-centric reconstruction, and achieves comparable performance to\nprior works in scene-centric reconstruction.\n", "date": "2024-12-09", "categories": []}
{"id": "2502.20378", "title": "Efficient Gaussian Splatting for Monocular Dynamic Scene Rendering via\n  Sparse Time-Variant Attribute Modeling", "abstract": "  Rendering dynamic scenes from monocular videos is a crucial yet challenging\ntask. The recent deformable Gaussian Splatting has emerged as a robust solution\nto represent real-world dynamic scenes. However, it often leads to heavily\nredundant Gaussians, attempting to fit every training view at various time\nsteps, leading to slower rendering speeds. Additionally, the attributes of\nGaussians in static areas are time-invariant, making it unnecessary to model\nevery Gaussian, which can cause jittering in static regions. In practice, the\nprimary bottleneck in rendering speed for dynamic scenes is the number of\nGaussians. In response, we introduce Efficient Dynamic Gaussian Splatting\n(EDGS), which represents dynamic scenes via sparse time-variant attribute\nmodeling. Our approach formulates dynamic scenes using a sparse anchor-grid\nrepresentation, with the motion flow of dense Gaussians calculated via a\nclassical kernel representation. Furthermore, we propose an unsupervised\nstrategy to efficiently filter out anchors corresponding to static areas. Only\nanchors associated with deformable objects are input into MLPs to query\ntime-variant attributes. Experiments on two real-world datasets demonstrate\nthat our EDGS significantly improves the rendering speed with superior\nrendering quality compared to previous state-of-the-art methods.\n", "date": "2025-02-27", "categories": []}
{"id": "2502.19318", "title": "Does 3D Gaussian Splatting Need Accurate Volumetric Rendering?", "abstract": "  Since its introduction, 3D Gaussian Splatting (3DGS) has become an important\nreference method for learning 3D representations of a captured scene, allowing\nreal-time novel-view synthesis with high visual quality and fast training\ntimes. Neural Radiance Fields (NeRFs), which preceded 3DGS, are based on a\nprincipled ray-marching approach for volumetric rendering. In contrast, while\nsharing a similar image formation model with NeRF, 3DGS uses a hybrid rendering\nsolution that builds on the strengths of volume rendering and primitive\nrasterization. A crucial benefit of 3DGS is its performance, achieved through a\nset of approximations, in many cases with respect to volumetric rendering\ntheory. A naturally arising question is whether replacing these approximations\nwith more principled volumetric rendering solutions can improve the quality of\n3DGS. In this paper, we present an in-depth analysis of the various\napproximations and assumptions used by the original 3DGS solution. We\ndemonstrate that, while more accurate volumetric rendering can help for low\nnumbers of primitives, the power of efficient optimization and the large number\nof Gaussians allows 3DGS to outperform volumetric rendering despite its\napproximations.\n", "date": "2025-02-26", "categories": []}
{"id": "2403.17888", "title": "2D Gaussian Splatting for Geometrically Accurate Radiance Fields", "abstract": "  3D Gaussian Splatting (3DGS) has recently revolutionized radiance field\nreconstruction, achieving high quality novel view synthesis and fast rendering\nspeed without baking. However, 3DGS fails to accurately represent surfaces due\nto the multi-view inconsistent nature of 3D Gaussians. We present 2D Gaussian\nSplatting (2DGS), a novel approach to model and reconstruct geometrically\naccurate radiance fields from multi-view images. Our key idea is to collapse\nthe 3D volume into a set of 2D oriented planar Gaussian disks. Unlike 3D\nGaussians, 2D Gaussians provide view-consistent geometry while modeling\nsurfaces intrinsically. To accurately recover thin surfaces and achieve stable\noptimization, we introduce a perspective-correct 2D splatting process utilizing\nray-splat intersection and rasterization. Additionally, we incorporate depth\ndistortion and normal consistency terms to further enhance the quality of the\nreconstructions. We demonstrate that our differentiable renderer allows for\nnoise-free and detailed geometry reconstruction while maintaining competitive\nappearance quality, fast training speed, and real-time rendering.\n", "date": "2024-03-26", "categories": []}
{"id": "2412.14166", "title": "MegaSynth: Scaling Up 3D Scene Reconstruction with Synthesized Data", "abstract": "  We propose scaling up 3D scene reconstruction by training with synthesized\ndata. At the core of our work is MegaSynth, a procedurally generated 3D dataset\ncomprising 700K scenes - over 50 times larger than the prior real dataset DL3DV\n- dramatically scaling the training data. To enable scalable data generation,\nour key idea is eliminating semantic information, removing the need to model\ncomplex semantic priors such as object affordances and scene composition.\nInstead, we model scenes with basic spatial structures and geometry primitives,\noffering scalability. Besides, we control data complexity to facilitate\ntraining while loosely aligning it with real-world data distribution to benefit\nreal-world generalization. We explore training LRMs with both MegaSynth and\navailable real data. Experiment results show that joint training or\npre-training with MegaSynth improves reconstruction quality by 1.2 to 1.8 dB\nPSNR across diverse image domains. Moreover, models trained solely on MegaSynth\nperform comparably to those trained on real data, underscoring the low-level\nnature of 3D reconstruction. Additionally, we provide an in-depth analysis of\nMegaSynth's properties for enhancing model capability, training stability, and\ngeneralization, as well as application to other tasks.\n", "date": "2024-12-18", "categories": []}
{"id": "2502.17377", "title": "Graph-Guided Scene Reconstruction from Images with 3D Gaussian Splatting", "abstract": "  This paper investigates an open research challenge of reconstructing\nhigh-quality, large 3D open scenes from images. It is observed existing methods\nhave various limitations, such as requiring precise camera poses for input and\ndense viewpoints for supervision. To perform effective and efficient 3D scene\nreconstruction, we propose a novel graph-guided 3D scene reconstruction\nframework, GraphGS. Specifically, given a set of images captured by RGB cameras\non a scene, we first design a spatial prior-based scene structure estimation\nmethod. This is then used to create a camera graph that includes information\nabout the camera topology. Further, we propose to apply the graph-guided\nmulti-view consistency constraint and adaptive sampling strategy to the 3D\nGaussian Splatting optimization process. This greatly alleviates the issue of\nGaussian points overfitting to specific sparse viewpoints and expedites the 3D\nreconstruction process. We demonstrate GraphGS achieves high-fidelity 3D\nreconstruction from images, which presents state-of-the-art performance through\nquantitative and qualitative evaluation across multiple datasets. Project Page:\nhttps://3dagentworld.github.io/graphgs.\n", "date": "2025-02-24", "categories": []}
{"id": "2309.16850", "title": "Sketch2CAD: 3D CAD Model Reconstruction from 2D Sketch using Visual\n  Transformer", "abstract": "  Current 3D reconstruction methods typically generate outputs in the form of\nvoxels, point clouds, or meshes. However, each of these formats has inherent\nlimitations, such as rough surfaces and distorted structures. Additionally,\nthese data types are not ideal for further manual editing and post-processing.\nIn this paper, we present a novel 3D reconstruction method designed to overcome\nthese disadvantages by reconstructing CAD-compatible models. We trained a\nvisual transformer to predict a \"scene descriptor\" from a single 2D wire-frame\nimage. This descriptor includes essential information, such as object types and\nparameters like position, rotation, and size. Using the predicted parameters, a\n3D scene can be reconstructed with 3D modeling software that has programmable\ninterfaces, such as Rhino Grasshopper, to build highly editable 3D models in\nthe form of B-rep. To evaluate our proposed model, we created two datasets: one\nconsisting of simple scenes and another with more complex scenes. The test\nresults indicate the model's capability to accurately reconstruct simple scenes\nwhile highlighting its difficulties with more complex ones.\n", "date": "2023-09-28", "categories": []}
{"id": "2502.14235", "title": "OG-Gaussian: Occupancy Based Street Gaussians for Autonomous Driving", "abstract": "  Accurate and realistic 3D scene reconstruction enables the lifelike creation\nof autonomous driving simulation environments. With advancements in 3D Gaussian\nSplatting (3DGS), previous studies have applied it to reconstruct complex\ndynamic driving scenes. These methods typically require expensive LiDAR sensors\nand pre-annotated datasets of dynamic objects. To address these challenges, we\npropose OG-Gaussian, a novel approach that replaces LiDAR point clouds with\nOccupancy Grids (OGs) generated from surround-view camera images using\nOccupancy Prediction Network (ONet). Our method leverages the semantic\ninformation in OGs to separate dynamic vehicles from static street background,\nconverting these grids into two distinct sets of initial point clouds for\nreconstructing both static and dynamic objects. Additionally, we estimate the\ntrajectories and poses of dynamic objects through a learning-based approach,\neliminating the need for complex manual annotations. Experiments on Waymo Open\ndataset demonstrate that OG-Gaussian is on par with the current\nstate-of-the-art in terms of reconstruction quality and rendering speed,\nachieving an average PSNR of 35.13 and a rendering speed of 143 FPS, while\nsignificantly reducing computational costs and economic overhead.\n", "date": "2025-02-20", "categories": []}
{"id": "2502.14684", "title": "CDGS: Confidence-Aware Depth Regularization for 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has shown significant advantages in novel view\nsynthesis (NVS), particularly in achieving high rendering speeds and\nhigh-quality results. However, its geometric accuracy in 3D reconstruction\nremains limited due to the lack of explicit geometric constraints during\noptimization. This paper introduces CDGS, a confidence-aware depth\nregularization approach developed to enhance 3DGS. We leverage multi-cue\nconfidence maps of monocular depth estimation and sparse Structure-from-Motion\ndepth to adaptively adjust depth supervision during the optimization process.\nOur method demonstrates improved geometric detail preservation in early\ntraining stages and achieves competitive performance in both NVS quality and\ngeometric accuracy. Experiments on the publicly available Tanks and Temples\nbenchmark dataset show that our method achieves more stable convergence\nbehavior and more accurate geometric reconstruction results, with improvements\nof up to 2.31 dB in PSNR for NVS and consistently lower geometric errors in\nM3C2 distance metrics. Notably, our method reaches comparable F-scores to the\noriginal 3DGS with only 50% of the training iterations. We expect this work\nwill facilitate the development of efficient and accurate 3D reconstruction\nsystems for real-world applications such as digital twin creation, heritage\npreservation, or forestry applications.\n", "date": "2025-02-20", "categories": []}
{"id": "2502.12534", "title": "NoKSR: Kernel-Free Neural Surface Reconstruction via Point Cloud\n  Serialization", "abstract": "  We present a novel approach to large-scale point cloud surface reconstruction\nby developing an efficient framework that converts an irregular point cloud\ninto a signed distance field (SDF). Our backbone builds upon recent\ntransformer-based architectures (i.e., PointTransformerV3), that serializes the\npoint cloud into a locality-preserving sequence of tokens. We efficiently\npredict the SDF value at a point by aggregating nearby tokens, where fast\napproximate neighbors can be retrieved thanks to the serialization. We\nserialize the point cloud at different levels/scales, and non-linearly\naggregate a feature to predict the SDF value. We show that aggregating across\nmultiple scales is critical to overcome the approximations introduced by the\nserialization (i.e. false negatives in the neighborhood). Our frameworks sets\nthe new state-of-the-art in terms of accuracy and efficiency (better or similar\nperformance with half the latency of the best prior method, coupled with a\nsimpler implementation), particularly on outdoor datasets where sparse-grid\nmethods have shown limited performance.\n", "date": "2025-02-18", "categories": []}
{"id": "2502.13968", "title": "Betsu-Betsu: Multi-View Separable 3D Reconstruction of Two Interacting\n  Objects", "abstract": "  Separable 3D reconstruction of multiple objects from multi-view RGB images --\nresulting in two different 3D shapes for the two objects with a clear\nseparation between them -- remains a sparsely researched problem. It is\nchallenging due to severe mutual occlusions and ambiguities along the objects'\ninteraction boundaries. This paper investigates the setting and introduces a\nnew neuro-implicit method that can reconstruct the geometry and appearance of\ntwo objects undergoing close interactions while disjoining both in 3D, avoiding\nsurface inter-penetrations and enabling novel-view synthesis of the observed\nscene. The framework is end-to-end trainable and supervised using a novel\nalpha-blending regularisation that ensures that the two geometries are well\nseparated even under extreme occlusions. Our reconstruction method is\nmarkerless and can be applied to rigid as well as articulated objects. We\nintroduce a new dataset consisting of close interactions between a human and an\nobject and also evaluate on two scenes of humans performing martial arts. The\nexperiments confirm the effectiveness of our framework and substantial\nimprovements using 3D and novel view synthesis metrics compared to several\nexisting approaches applicable in our setting.\n", "date": "2025-02-19", "categories": []}
{"id": "2402.07376", "title": "Unsupervised Discovery of Object-Centric Neural Fields", "abstract": "  We study inferring 3D object-centric scene representations from a single\nimage. While recent methods have shown potential in unsupervised 3D object\ndiscovery from simple synthetic images, they fail to generalize to real-world\nscenes with visually rich and diverse objects. This limitation stems from their\nobject representations, which entangle objects' intrinsic attributes like shape\nand appearance with extrinsic, viewer-centric properties such as their 3D\nlocation. To address this bottleneck, we propose Unsupervised discovery of\nObject-Centric neural Fields (uOCF). uOCF focuses on learning the intrinsics of\nobjects and models the extrinsics separately. Our approach significantly\nimproves systematic generalization, thus enabling unsupervised learning of\nhigh-fidelity object-centric scene representations from sparse real-world\nimages. To evaluate our approach, we collect three new datasets, including two\nreal kitchen environments. Extensive experiments show that uOCF enables\nunsupervised discovery of visually rich objects from a single real image,\nallowing applications such as 3D object segmentation and scene manipulation.\nNotably, uOCF demonstrates zero-shot generalization to unseen objects from a\nsingle real image. Project page: https://red-fairy.github.io/uOCF/\n", "date": "2024-02-12", "categories": []}
{"id": "2403.06877", "title": "SiLVR: Scalable Lidar-Visual Reconstruction with Neural Radiance Fields\n  for Robotic Inspection", "abstract": "  We present a neural-field-based large-scale reconstruction system that fuses\nlidar and vision data to generate high-quality reconstructions that are\ngeometrically accurate and capture photo-realistic textures. This system adapts\nthe state-of-the-art neural radiance field (NeRF) representation to also\nincorporate lidar data which adds strong geometric constraints on the depth and\nsurface normals. We exploit the trajectory from a real-time lidar SLAM system\nto bootstrap a Structure-from-Motion (SfM) procedure to both significantly\nreduce the computation time and to provide metric scale which is crucial for\nlidar depth loss. We use submapping to scale the system to large-scale\nenvironments captured over long trajectories. We demonstrate the reconstruction\nsystem with data from a multi-camera, lidar sensor suite onboard a legged\nrobot, hand-held while scanning building scenes for 600 metres, and onboard an\naerial robot surveying a multi-storey mock disaster site-building. Website:\nhttps://ori-drs.github.io/projects/silvr/\n", "date": "2024-03-11", "categories": []}
{"id": "2501.18643", "title": "3D Reconstruction of Shoes for Augmented Reality", "abstract": "  This paper introduces a mobile-based solution that enhances online shoe\nshopping through 3D modeling and Augmented Reality (AR), leveraging the\nefficiency of 3D Gaussian Splatting. Addressing the limitations of static 2D\nimages, the framework generates realistic 3D shoe models from 2D images,\nachieving an average Peak Signal-to-Noise Ratio (PSNR) of 32, and enables\nimmersive AR interactions via smartphones. A custom shoe segmentation dataset\nof 3120 images was created, with the best-performing segmentation model\nachieving an Intersection over Union (IoU) score of 0.95. This paper\ndemonstrates the potential of 3D modeling and AR to revolutionize online\nshopping by offering realistic virtual interactions, with applicability across\nbroader fashion categories.\n", "date": "2025-01-29", "categories": []}
{"id": "2502.10492", "title": "Multi-view 3D surface reconstruction from SAR images by inverse\n  rendering", "abstract": "  3D reconstruction of a scene from Synthetic Aperture Radar (SAR) images\nmainly relies on interferometric measurements, which involve strict constraints\non the acquisition process. These last years, progress in deep learning has\nsignificantly advanced 3D reconstruction from multiple views in optical\nimaging, mainly through reconstruction-by-synthesis approaches pioneered by\nNeural Radiance Fields. In this paper, we propose a new inverse rendering\nmethod for 3D reconstruction from unconstrained SAR images, drawing inspiration\nfrom optical approaches. First, we introduce a new simplified differentiable\nSAR rendering model, able to synthesize images from a digital elevation model\nand a radar backscattering coefficients map. Then, we introduce a\ncoarse-to-fine strategy to train a Multi-Layer Perceptron (MLP) to fit the\nheight and appearance of a given radar scene from a few SAR views. Finally, we\ndemonstrate the surface reconstruction capabilities of our method on synthetic\nSAR images produced by ONERA's physically-based EMPRISE simulator. Our method\nshowcases the potential of exploiting geometric disparities in SAR images and\npaves the way for multi-sensor data fusion.\n", "date": "2025-02-14", "categories": []}
{"id": "2407.20908", "title": "Dynamic Scene Understanding through Object-Centric Voxelization and\n  Neural Rendering", "abstract": "  Learning object-centric representations from unsupervised videos is\nchallenging. Unlike most previous approaches that focus on decomposing 2D\nimages, we present a 3D generative model named DynaVol-S for dynamic scenes\nthat enables object-centric learning within a differentiable volume rendering\nframework. The key idea is to perform object-centric voxelization to capture\nthe 3D nature of the scene, which infers per-object occupancy probabilities at\nindividual spatial locations. These voxel features evolve through a\ncanonical-space deformation function and are optimized in an inverse rendering\npipeline with a compositional NeRF. Additionally, our approach integrates 2D\nsemantic features to create 3D semantic grids, representing the scene through\nmultiple disentangled voxel grids. DynaVol-S significantly outperforms existing\nmodels in both novel view synthesis and unsupervised decomposition tasks for\ndynamic scenes. By jointly considering geometric structures and semantic\nfeatures, it effectively addresses challenging real-world scenarios involving\ncomplex object interactions. Furthermore, once trained, the explicitly\nmeaningful voxel features enable additional capabilities that 2D scene\ndecomposition methods cannot achieve, such as novel scene generation through\nediting geometric shapes or manipulating the motion trajectories of objects.\n", "date": "2024-07-30", "categories": []}
{"id": "2502.09613", "title": "Latent Radiance Fields with 3D-aware 2D Representations", "abstract": "  Latent 3D reconstruction has shown great promise in empowering 3D semantic\nunderstanding and 3D generation by distilling 2D features into the 3D space.\nHowever, existing approaches struggle with the domain gap between 2D feature\nspace and 3D representations, resulting in degraded rendering performance. To\naddress this challenge, we propose a novel framework that integrates 3D\nawareness into the 2D latent space. The framework consists of three stages: (1)\na correspondence-aware autoencoding method that enhances the 3D consistency of\n2D latent representations, (2) a latent radiance field (LRF) that lifts these\n3D-aware 2D representations into 3D space, and (3) a VAE-Radiance Field\n(VAE-RF) alignment strategy that improves image decoding from the rendered 2D\nrepresentations. Extensive experiments demonstrate that our method outperforms\nthe state-of-the-art latent 3D reconstruction approaches in terms of synthesis\nperformance and cross-dataset generalizability across diverse indoor and\noutdoor scenes. To our knowledge, this is the first work showing the radiance\nfield representations constructed from 2D latent representations can yield\nphotorealistic 3D reconstruction performance.\n", "date": "2025-02-13", "categories": []}
{"id": "2412.18862", "title": "WeatherGS: 3D Scene Reconstruction in Adverse Weather Conditions via\n  Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has gained significant attention for 3D scene\nreconstruction, but still suffers from complex outdoor environments, especially\nunder adverse weather. This is because 3DGS treats the artifacts caused by\nadverse weather as part of the scene and will directly reconstruct them,\nlargely reducing the clarity of the reconstructed scene. To address this\nchallenge, we propose WeatherGS, a 3DGS-based framework for reconstructing\nclear scenes from multi-view images under different weather conditions.\nSpecifically, we explicitly categorize the multi-weather artifacts into the\ndense particles and lens occlusions that have very different characters, in\nwhich the former are caused by snowflakes and raindrops in the air, and the\nlatter are raised by the precipitation on the camera lens. In light of this, we\npropose a dense-to-sparse preprocess strategy, which sequentially removes the\ndense particles by an Atmospheric Effect Filter (AEF) and then extracts the\nrelatively sparse occlusion masks with a Lens Effect Detector (LED). Finally,\nwe train a set of 3D Gaussians by the processed images and generated masks for\nexcluding occluded areas, and accurately recover the underlying clear scene by\nGaussian splatting. We conduct a diverse and challenging benchmark to\nfacilitate the evaluation of 3D reconstruction under complex weather scenarios.\nExtensive experiments on this benchmark demonstrate that our WeatherGS\nconsistently produces high-quality, clean scenes across various weather\nscenarios, outperforming existing state-of-the-art methods. See project\npage:https://jumponthemoon.github.io/weather-gs.\n", "date": "2024-12-25", "categories": []}
{"id": "2409.12886", "title": "EdgeGaussians -- 3D Edge Mapping via Gaussian Splatting", "abstract": "  With their meaningful geometry and their omnipresence in the 3D world, edges\nare extremely useful primitives in computer vision. 3D edges comprise of lines\nand curves, and methods to reconstruct them use either multi-view images or\npoint clouds as input. State-of-the-art image-based methods first learn a 3D\nedge point cloud then fit 3D edges to it. The edge point cloud is obtained by\nlearning a 3D neural implicit edge field from which the 3D edge points are\nsampled on a specific level set (0 or 1). However, such methods present two\nimportant drawbacks: i) it is not realistic to sample points on exact level\nsets due to float imprecision and training inaccuracies. Instead, they are\nsampled within a range of levels so the points do not lie accurately on the 3D\nedges and require further processing. ii) Such implicit representations are\ncomputationally expensive and require long training times. In this paper, we\naddress these two limitations and propose a 3D edge mapping that is simpler,\nmore efficient, and preserves accuracy. Our method learns explicitly the 3D\nedge points and their edge direction hence bypassing the need for point\nsampling. It casts a 3D edge point as the center of a 3D Gaussian and the edge\ndirection as the principal axis of the Gaussian. Such a representation has the\nadvantage of being not only geometrically meaningful but also compatible with\nthe efficient training optimization defined in Gaussian Splatting. Results show\nthat the proposed method produces edges as accurate and complete as the\nstate-of-the-art while being an order of magnitude faster. Code is released at\nhttps://github.com/kunalchelani/EdgeGaussians.\n", "date": "2024-09-19", "categories": []}
{"id": "2502.07140", "title": "Few-Shot Multi-Human Neural Rendering Using Geometry Constraints", "abstract": "  We present a method for recovering the shape and radiance of a scene\nconsisting of multiple people given solely a few images. Multi-human scenes are\ncomplex due to additional occlusion and clutter. For single-human settings,\nexisting approaches using implicit neural representations have achieved\nimpressive results that deliver accurate geometry and appearance. However, it\nremains challenging to extend these methods for estimating multiple humans from\nsparse views. We propose a neural implicit reconstruction method that addresses\nthe inherent challenges of this task through the following contributions:\nFirst, we propose to use geometry constraints by exploiting pre-computed meshes\nusing a human body model (SMPL). Specifically, we regularize the signed\ndistances using the SMPL mesh and leverage bounding boxes for improved\nrendering. Second, we propose a ray regularization scheme to minimize rendering\ninconsistencies, and a saturation regularization for robust optimization in\nvariable illumination. Extensive experiments on both real and synthetic\ndatasets demonstrate the benefits of our approach and show state-of-the-art\nperformance against existing neural reconstruction methods.\n", "date": "2025-02-11", "categories": []}
{"id": "2502.07615", "title": "Flow Distillation Sampling: Regularizing 3D Gaussians with Pre-trained\n  Matching Priors", "abstract": "  3D Gaussian Splatting (3DGS) has achieved excellent rendering quality with\nfast training and rendering speed. However, its optimization process lacks\nexplicit geometric constraints, leading to suboptimal geometric reconstruction\nin regions with sparse or no observational input views. In this work, we try to\nmitigate the issue by incorporating a pre-trained matching prior to the 3DGS\noptimization process. We introduce Flow Distillation Sampling (FDS), a\ntechnique that leverages pre-trained geometric knowledge to bolster the\naccuracy of the Gaussian radiance field. Our method employs a strategic\nsampling technique to target unobserved views adjacent to the input views,\nutilizing the optical flow calculated from the matching model (Prior Flow) to\nguide the flow analytically calculated from the 3DGS geometry (Radiance Flow).\nComprehensive experiments in depth rendering, mesh reconstruction, and novel\nview synthesis showcase the significant advantages of FDS over state-of-the-art\nmethods. Additionally, our interpretive experiments and analysis aim to shed\nlight on the effects of FDS on geometric accuracy and rendering quality,\npotentially providing readers with insights into its performance. Project page:\nhttps://nju-3dv.github.io/projects/fds\n", "date": "2025-02-11", "categories": []}
{"id": "2411.18311", "title": "Neural Surface Priors for Editable Gaussian Splatting", "abstract": "  In computer graphics and vision, recovering easily modifiable scene\nappearance from image data is crucial for applications such as content\ncreation. We introduce a novel method that integrates 3D Gaussian Splatting\nwith an implicit surface representation, enabling intuitive editing of\nrecovered scenes through mesh manipulation. Starting with a set of input images\nand camera poses, our approach reconstructs the scene surface using a neural\nsigned distance field. This neural surface acts as a geometric prior guiding\nthe training of Gaussian Splatting components, ensuring their alignment with\nthe scene geometry. To facilitate editing, we encode the visual and geometric\ninformation into a lightweight triangle soup proxy. Edits applied to the mesh\nextracted from the neural surface propagate seamlessly through this\nintermediate structure to update the recovered appearance. Unlike previous\nmethods relying on the triangle soup proxy representation, our approach\nsupports a wider range of modifications and fully leverages the mesh topology,\nenabling a more flexible and intuitive editing process. The complete source\ncode for this project can be accessed at:\nhttps://github.com/WJakubowska/NeuralSurfacePriors.\n", "date": "2024-11-27", "categories": []}
{"id": "2502.05222", "title": "VistaFlow: Photorealistic Volumetric Reconstruction with Dynamic\n  Resolution Management via Q-Learning", "abstract": "  We introduce VistaFlow, a scalable three-dimensional imaging technique\ncapable of reconstructing fully interactive 3D volumetric images from a set of\n2D photographs. Our model synthesizes novel viewpoints through a differentiable\nrendering system capable of dynamic resolution management on photorealistic 3D\nscenes. We achieve this through the introduction of QuiQ, a novel intermediate\nvideo controller trained through Q-learning to maintain a consistently high\nframerate by adjusting render resolution with millisecond precision. Notably,\nVistaFlow runs natively on integrated CPU graphics, making it viable for mobile\nand entry-level devices while still delivering high-performance rendering.\nVistaFlow bypasses Neural Radiance Fields (NeRFs), using the PlenOctree data\nstructure to render complex light interactions such as reflection and\nsubsurface scattering with minimal hardware requirements. Our model is capable\nof outperforming state-of-the-art methods with novel view synthesis at a\nresolution of 1080p at over 100 frames per second on consumer hardware. By\ntailoring render quality to the capabilities of each device, VistaFlow has the\npotential to improve the efficiency and accessibility of photorealistic 3D\nscene rendering across a wide spectrum of hardware, from high-end workstations\nto inexpensive microcontrollers.\n", "date": "2025-02-05", "categories": []}
{"id": "2502.06047", "title": "Neural Shortest Path for Surface Reconstruction from Point Clouds", "abstract": "  In this paper, we propose the neural shortest path (NSP), a vector-valued\nimplicit neural representation (INR) that approximates a distance function and\nits gradient. The key feature of NSP is to learn the exact shortest path (ESP),\nwhich directs an arbitrary point to its nearest point on the target surface.\nThe NSP is decomposed into its magnitude and direction, and a variable\nsplitting method is used that each decomposed component approximates a distance\nfunction and its gradient, respectively. Unlike to existing methods of learning\nthe distance function itself, the NSP ensures the simultaneous recovery of the\ndistance function and its gradient. We mathematically prove that the decomposed\nrepresentation of NSP guarantees the convergence of the magnitude of NSP in the\n$H^1$ norm. Furthermore, we devise a novel loss function that enforces the\nproperty of ESP, demonstrating that its global minimum is the ESP. We evaluate\nthe performance of the NSP through comprehensive experiments on diverse\ndatasets, validating its capacity to reconstruct high-quality surfaces with the\nrobustness to noise and data sparsity. The numerical results show substantial\nimprovements over state-of-the-art methods, highlighting the importance of\nlearning the ESP, the product of distance function and its gradient, for\nrepresenting a wide variety of complex surfaces.\n", "date": "2025-02-09", "categories": []}
{"id": "2502.04630", "title": "High-Speed Dynamic 3D Imaging with Sensor Fusion Splatting", "abstract": "  Capturing and reconstructing high-speed dynamic 3D scenes has numerous\napplications in computer graphics, vision, and interdisciplinary fields such as\nrobotics, aerodynamics, and evolutionary biology. However, achieving this using\na single imaging modality remains challenging. For instance, traditional RGB\ncameras suffer from low frame rates, limited exposure times, and narrow\nbaselines. To address this, we propose a novel sensor fusion approach using\nGaussian splatting, which combines RGB, depth, and event cameras to capture and\nreconstruct deforming scenes at high speeds. The key insight of our method lies\nin leveraging the complementary strengths of these imaging modalities: RGB\ncameras capture detailed color information, event cameras record rapid scene\nchanges with microsecond resolution, and depth cameras provide 3D scene\ngeometry. To unify the underlying scene representation across these modalities,\nwe represent the scene using deformable 3D Gaussians. To handle rapid scene\nmovements, we jointly optimize the 3D Gaussian parameters and their temporal\ndeformation fields by integrating data from all three sensor modalities. This\nfusion enables efficient, high-quality imaging of fast and complex scenes, even\nunder challenging conditions such as low light, narrow baselines, or rapid\nmotion. Experiments on synthetic and real datasets captured with our prototype\nsensor fusion setup demonstrate that our method significantly outperforms\nstate-of-the-art techniques, achieving noticeable improvements in both\nrendering fidelity and structural accuracy.\n", "date": "2025-02-07", "categories": []}
{"id": "2502.04734", "title": "SC-OmniGS: Self-Calibrating Omnidirectional Gaussian Splatting", "abstract": "  360-degree cameras streamline data collection for radiance field 3D\nreconstruction by capturing comprehensive scene data. However, traditional\nradiance field methods do not address the specific challenges inherent to\n360-degree images. We present SC-OmniGS, a novel self-calibrating\nomnidirectional Gaussian splatting system for fast and accurate omnidirectional\nradiance field reconstruction using 360-degree images. Rather than converting\n360-degree images to cube maps and performing perspective image calibration, we\ntreat 360-degree images as a whole sphere and derive a mathematical framework\nthat enables direct omnidirectional camera pose calibration accompanied by 3D\nGaussians optimization. Furthermore, we introduce a differentiable\nomnidirectional camera model in order to rectify the distortion of real-world\ndata for performance enhancement. Overall, the omnidirectional camera intrinsic\nmodel, extrinsic poses, and 3D Gaussians are jointly optimized by minimizing\nweighted spherical photometric loss. Extensive experiments have demonstrated\nthat our proposed SC-OmniGS is able to recover a high-quality radiance field\nfrom noisy camera poses or even no pose prior in challenging scenarios\ncharacterized by wide baselines and non-object-centric configurations. The\nnoticeable performance gain in the real-world dataset captured by\nconsumer-grade omnidirectional cameras verifies the effectiveness of our\ngeneral omnidirectional camera model in reducing the distortion of 360-degree\nimages.\n", "date": "2025-02-07", "categories": []}
{"id": "2502.04318", "title": "sshELF: Single-Shot Hierarchical Extrapolation of Latent Features for 3D\n  Reconstruction from Sparse-Views", "abstract": "  Reconstructing unbounded outdoor scenes from sparse outward-facing views\nposes significant challenges due to minimal view overlap. Previous methods\noften lack cross-scene understanding and their primitive-centric formulations\noverload local features to compensate for missing global context, resulting in\nblurriness in unseen parts of the scene. We propose sshELF, a fast, single-shot\npipeline for sparse-view 3D scene reconstruction via hierarchal extrapolation\nof latent features. Our key insights is that disentangling information\nextrapolation from primitive decoding allows efficient transfer of structural\npatterns across training scenes. Our method: (1) learns cross-scene priors to\ngenerate intermediate virtual views to extrapolate to unobserved regions, (2)\noffers a two-stage network design separating virtual view generation from 3D\nprimitive decoding for efficient training and modular model design, and (3)\nintegrates a pre-trained foundation model for joint inference of latent\nfeatures and texture, improving scene understanding and generalization. sshELF\ncan reconstruct 360 degree scenes from six sparse input views and achieves\ncompetitive results on synthetic and real-world datasets. We find that sshELF\nfaithfully reconstructs occluded regions, supports real-time rendering, and\nprovides rich latent features for downstream applications. The code will be\nreleased.\n", "date": "2025-02-06", "categories": []}
{"id": "2502.02144", "title": "DOC-Depth: A novel approach for dense depth ground truth generation", "abstract": "  Accurate depth information is essential for many computer vision\napplications. Yet, no available dataset recording method allows for fully dense\naccurate depth estimation in a large scale dynamic environment. In this paper,\nwe introduce DOC-Depth, a novel, efficient and easy-to-deploy approach for\ndense depth generation from any LiDAR sensor. After reconstructing consistent\ndense 3D environment using LiDAR odometry, we address dynamic objects\nocclusions automatically thanks to DOC, our state-of-the art dynamic object\nclassification method. Additionally, DOC-Depth is fast and scalable, allowing\nfor the creation of unbounded datasets in terms of size and time. We\ndemonstrate the effectiveness of our approach on the KITTI dataset, improving\nits density from 16.1% to 71.2% and release this new fully dense depth\nannotation, to facilitate future research in the domain. We also showcase\nresults using various LiDAR sensors and in multiple environments. All software\ncomponents are publicly available for the research community.\n", "date": "2025-02-04", "categories": []}
{"id": "2412.12906", "title": "CATSplat: Context-Aware Transformer with Spatial Guidance for\n  Generalizable 3D Gaussian Splatting from A Single-View Image", "abstract": "  Recently, generalizable feed-forward methods based on 3D Gaussian Splatting\nhave gained significant attention for their potential to reconstruct 3D scenes\nusing finite resources. These approaches create a 3D radiance field,\nparameterized by per-pixel 3D Gaussian primitives, from just a few images in a\nsingle forward pass. However, unlike multi-view methods that benefit from\ncross-view correspondences, 3D scene reconstruction with a single-view image\nremains an underexplored area. In this work, we introduce CATSplat, a novel\ngeneralizable transformer-based framework designed to break through the\ninherent constraints in monocular settings. First, we propose leveraging\ntextual guidance from a visual-language model to complement insufficient\ninformation from a single image. By incorporating scene-specific contextual\ndetails from text embeddings through cross-attention, we pave the way for\ncontext-aware 3D scene reconstruction beyond relying solely on visual cues.\nMoreover, we advocate utilizing spatial guidance from 3D point features toward\ncomprehensive geometric understanding under single-view settings. With 3D\npriors, image features can capture rich structural insights for predicting 3D\nGaussians without multi-view techniques. Extensive experiments on large-scale\ndatasets demonstrate the state-of-the-art performance of CATSplat in\nsingle-view 3D scene reconstruction with high-quality novel view synthesis.\n", "date": "2024-12-17", "categories": []}
{"id": "2501.16737", "title": "Consistency Diffusion Models for Single-Image 3D Reconstruction with\n  Priors", "abstract": "  This paper delves into the study of 3D point cloud reconstruction from a\nsingle image. Our objective is to develop the Consistency Diffusion Model,\nexploring synergistic 2D and 3D priors in the Bayesian framework to ensure\nsuperior consistency in the reconstruction process, a challenging yet critical\nrequirement in this field. Specifically, we introduce a pioneering training\nframework under diffusion models that brings two key innovations. First, we\nconvert 3D structural priors derived from the initial 3D point cloud as a bound\nterm to increase evidence in the variational Bayesian framework, leveraging\nthese robust intrinsic priors to tightly govern the diffusion training process\nand bolster consistency in reconstruction. Second, we extract and incorporate\n2D priors from the single input image, projecting them onto the 3D point cloud\nto enrich the guidance for diffusion training. Our framework not only sidesteps\npotential model learning shifts that may arise from directly imposing\nadditional constraints during training but also precisely transposes the 2D\npriors into the 3D domain. Extensive experimental evaluations reveal that our\napproach sets new benchmarks in both synthetic and real-world datasets. The\ncode is included with the submission.\n", "date": "2025-01-28", "categories": []}
{"id": "2501.18804", "title": "Zero-Shot Novel View and Depth Synthesis with Multi-View Geometric\n  Diffusion", "abstract": "  Current methods for 3D scene reconstruction from sparse posed images employ\nintermediate 3D representations such as neural fields, voxel grids, or 3D\nGaussians, to achieve multi-view consistent scene appearance and geometry. In\nthis paper we introduce MVGD, a diffusion-based architecture capable of direct\npixel-level generation of images and depth maps from novel viewpoints, given an\narbitrary number of input views. Our method uses raymap conditioning to both\naugment visual features with spatial information from different viewpoints, as\nwell as to guide the generation of images and depth maps from novel views. A\nkey aspect of our approach is the multi-task generation of images and depth\nmaps, using learnable task embeddings to guide the diffusion process towards\nspecific modalities. We train this model on a collection of more than 60\nmillion multi-view samples from publicly available datasets, and propose\ntechniques to enable efficient and consistent learning in such diverse\nconditions. We also propose a novel strategy that enables the efficient\ntraining of larger models by incrementally fine-tuning smaller ones, with\npromising scaling behavior. Through extensive experiments, we report\nstate-of-the-art results in multiple novel view synthesis benchmarks, as well\nas multi-view stereo and video depth estimation.\n", "date": "2025-01-30", "categories": []}
{"id": "2501.19319", "title": "Advancing Dense Endoscopic Reconstruction with Gaussian Splatting-driven\n  Surface Normal-aware Tracking and Mapping", "abstract": "  Simultaneous Localization and Mapping (SLAM) is essential for precise\nsurgical interventions and robotic tasks in minimally invasive procedures.\nWhile recent advancements in 3D Gaussian Splatting (3DGS) have improved SLAM\nwith high-quality novel view synthesis and fast rendering, these systems\nstruggle with accurate depth and surface reconstruction due to multi-view\ninconsistencies. Simply incorporating SLAM and 3DGS leads to mismatches between\nthe reconstructed frames. In this work, we present Endo-2DTAM, a real-time\nendoscopic SLAM system with 2D Gaussian Splatting (2DGS) to address these\nchallenges. Endo-2DTAM incorporates a surface normal-aware pipeline, which\nconsists of tracking, mapping, and bundle adjustment modules for geometrically\naccurate reconstruction. Our robust tracking module combines point-to-point and\npoint-to-plane distance metrics, while the mapping module utilizes normal\nconsistency and depth distortion to enhance surface reconstruction quality. We\nalso introduce a pose-consistent strategy for efficient and geometrically\ncoherent keyframe sampling. Extensive experiments on public endoscopic datasets\ndemonstrate that Endo-2DTAM achieves an RMSE of $1.87\\pm 0.63$ mm for depth\nreconstruction of surgical scenes while maintaining computationally efficient\ntracking, high-quality visual appearance, and real-time rendering. Our code\nwill be released at github.com/lastbasket/Endo-2DTAM.\n", "date": "2025-01-31", "categories": []}
{"id": "2501.16617", "title": "Predicting 3D representations for Dynamic Scenes", "abstract": "  We present a novel framework for dynamic radiance field prediction given\nmonocular video streams. Unlike previous methods that primarily focus on\npredicting future frames, our method goes a step further by generating explicit\n3D representations of the dynamic scene. The framework builds on two core\ndesigns. First, we adopt an ego-centric unbounded triplane to explicitly\nrepresent the dynamic physical world. Second, we develop a 4D-aware transformer\nto aggregate features from monocular videos to update the triplane. Coupling\nthese two designs enables us to train the proposed model with large-scale\nmonocular videos in a self-supervised manner. Our model achieves top results in\ndynamic radiance field prediction on NVIDIA dynamic scenes, demonstrating its\nstrong performance on 4D physical world modeling. Besides, our model shows a\nsuperior generalizability to unseen scenarios. Notably, we find that our\napproach emerges capabilities for geometry and semantic learning.\n", "date": "2025-01-28", "categories": []}
{"id": "2501.13975", "title": "3DGS$^2$: Near Second-order Converging 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has emerged as a mainstream solution for novel\nview synthesis and 3D reconstruction. By explicitly encoding a 3D scene using a\ncollection of Gaussian kernels, 3DGS achieves high-quality rendering with\nsuperior efficiency. As a learning-based approach, 3DGS training has been dealt\nwith the standard stochastic gradient descent (SGD) method, which offers at\nmost linear convergence. Consequently, training often requires tens of minutes,\neven with GPU acceleration. This paper introduces a (near) second-order\nconvergent training algorithm for 3DGS, leveraging its unique properties. Our\napproach is inspired by two key observations. First, the attributes of a\nGaussian kernel contribute independently to the image-space loss, which\nendorses isolated and local optimization algorithms. We exploit this by\nsplitting the optimization at the level of individual kernel attributes,\nanalytically constructing small-size Newton systems for each parameter group,\nand efficiently solving these systems on GPU threads. This achieves Newton-like\nconvergence per training image without relying on the global Hessian. Second,\nkernels exhibit sparse and structured coupling across input images. This\nproperty allows us to effectively utilize spatial information to mitigate\novershoot during stochastic training. Our method converges an order faster than\nstandard GPU-based 3DGS training, requiring over $10\\times$ fewer iterations\nwhile maintaining or surpassing the quality of the compared with the SGD-based\n3DGS reconstructions.\n", "date": "2025-01-22", "categories": []}
{"id": "2501.16101", "title": "3D Reconstruction of non-visible surfaces of objects from a Single Depth\n  View -- Comparative Study", "abstract": "  Scene and object reconstruction is an important problem in robotics, in\nparticular in planning collision-free trajectories or in object manipulation.\nThis paper compares two strategies for the reconstruction of nonvisible parts\nof the object surface from a single RGB-D camera view. The first method, named\nDeepSDF predicts the Signed Distance Transform to the object surface for a\ngiven point in 3D space. The second method, named MirrorNet reconstructs the\noccluded objects' parts by generating images from the other side of the\nobserved object. Experiments performed with objects from the ShapeNet dataset,\nshow that the view-dependent MirrorNet is faster and has smaller reconstruction\nerrors in most categories.\n", "date": "2025-01-27", "categories": []}
{"id": "2501.14231", "title": "Micro-macro Wavelet-based Gaussian Splatting for 3D Reconstruction from\n  Unconstrained Images", "abstract": "  3D reconstruction from unconstrained image collections presents substantial\nchallenges due to varying appearances and transient occlusions. In this paper,\nwe introduce Micro-macro Wavelet-based Gaussian Splatting (MW-GS), a novel\napproach designed to enhance 3D reconstruction by disentangling scene\nrepresentations into global, refined, and intrinsic components. The proposed\nmethod features two key innovations: Micro-macro Projection, which allows\nGaussian points to capture details from feature maps across multiple scales\nwith enhanced diversity; and Wavelet-based Sampling, which leverages frequency\ndomain information to refine feature representations and significantly improve\nthe modeling of scene appearances. Additionally, we incorporate a Hierarchical\nResidual Fusion Network to seamlessly integrate these features. Extensive\nexperiments demonstrate that MW-GS delivers state-of-the-art rendering\nperformance, surpassing existing methods.\n", "date": "2025-01-24", "categories": []}
{"id": "2501.14319", "title": "Scalable Benchmarking and Robust Learning for Noise-Free Ego-Motion and\n  3D Reconstruction from Noisy Video", "abstract": "  We aim to redefine robust ego-motion estimation and photorealistic 3D\nreconstruction by addressing a critical limitation: the reliance on noise-free\ndata in existing models. While such sanitized conditions simplify evaluation,\nthey fail to capture the unpredictable, noisy complexities of real-world\nenvironments. Dynamic motion, sensor imperfections, and synchronization\nperturbations lead to sharp performance declines when these models are deployed\nin practice, revealing an urgent need for frameworks that embrace and excel\nunder real-world noise. To bridge this gap, we tackle three core challenges:\nscalable data generation, comprehensive benchmarking, and model robustness\nenhancement. First, we introduce a scalable noisy data synthesis pipeline that\ngenerates diverse datasets simulating complex motion, sensor imperfections, and\nsynchronization errors. Second, we leverage this pipeline to create\nRobust-Ego3D, a benchmark rigorously designed to expose noise-induced\nperformance degradation, highlighting the limitations of current learning-based\nmethods in ego-motion accuracy and 3D reconstruction quality. Third, we propose\nCorrespondence-guided Gaussian Splatting (CorrGS), a novel test-time adaptation\nmethod that progressively refines an internal clean 3D representation by\naligning noisy observations with rendered RGB-D frames from clean 3D map,\nenhancing geometric alignment and appearance restoration through visual\ncorrespondence. Extensive experiments on synthetic and real-world data\ndemonstrate that CorrGS consistently outperforms prior state-of-the-art\nmethods, particularly in scenarios involving rapid motion and dynamic\nillumination.\n", "date": "2025-01-24", "categories": []}
{"id": "2501.14534", "title": "Trick-GS: A Balanced Bag of Tricks for Efficient Gaussian Splatting", "abstract": "  Gaussian splatting (GS) for 3D reconstruction has become quite popular due to\ntheir fast training, inference speeds and high quality reconstruction. However,\nGS-based reconstructions generally consist of millions of Gaussians, which\nmakes them hard to use on computationally constrained devices such as\nsmartphones. In this paper, we first propose a principled analysis of advances\nin efficient GS methods. Then, we propose Trick-GS, which is a careful\ncombination of several strategies including (1) progressive training with\nresolution, noise and Gaussian scales, (2) learning to prune and mask\nprimitives and SH bands by their significance, and (3) accelerated GS training\nframework. Trick-GS takes a large step towards resource-constrained GS, where\nfaster run-time, smaller and faster-convergence of models is of paramount\nconcern. Our results on three datasets show that Trick-GS achieves up to 2x\nfaster training, 40x smaller disk size and 2x faster rendering speed compared\nto vanilla GS, while having comparable accuracy.\n", "date": "2025-01-24", "categories": []}
{"id": "2404.00409", "title": "3DGSR: Implicit Surface Reconstruction with 3D Gaussian Splatting", "abstract": "  In this paper, we present an implicit surface reconstruction method with 3D\nGaussian Splatting (3DGS), namely 3DGSR, that allows for accurate 3D\nreconstruction with intricate details while inheriting the high efficiency and\nrendering quality of 3DGS. The key insight is incorporating an implicit signed\ndistance field (SDF) within 3D Gaussians to enable them to be aligned and\njointly optimized. First, we introduce a differentiable SDF-to-opacity\ntransformation function that converts SDF values into corresponding Gaussians'\nopacities. This function connects the SDF and 3D Gaussians, allowing for\nunified optimization and enforcing surface constraints on the 3D Gaussians.\nDuring learning, optimizing the 3D Gaussians provides supervisory signals for\nSDF learning, enabling the reconstruction of intricate details. However, this\nonly provides sparse supervisory signals to the SDF at locations occupied by\nGaussians, which is insufficient for learning a continuous SDF. Then, to\naddress this limitation, we incorporate volumetric rendering and align the\nrendered geometric attributes (depth, normal) with those derived from 3D\nGaussians. This consistency regularization introduces supervisory signals to\nlocations not covered by discrete 3D Gaussians, effectively eliminating\nredundant surfaces outside the Gaussian sampling range. Our extensive\nexperimental results demonstrate that our 3DGSR method enables high-quality 3D\nsurface reconstruction while preserving the efficiency and rendering quality of\n3DGS. Besides, our method competes favorably with leading surface\nreconstruction techniques while offering a more efficient learning process and\nmuch better rendering qualities. The code will be available at\nhttps://github.com/CVMI-Lab/3DGSR.\n", "date": "2024-03-30", "categories": []}
{"id": "2404.12385", "title": "MeshLRM: Large Reconstruction Model for High-Quality Meshes", "abstract": "  We propose MeshLRM, a novel LRM-based approach that can reconstruct a\nhigh-quality mesh from merely four input images in less than one second.\nDifferent from previous large reconstruction models (LRMs) that focus on\nNeRF-based reconstruction, MeshLRM incorporates differentiable mesh extraction\nand rendering within the LRM framework. This allows for end-to-end mesh\nreconstruction by fine-tuning a pre-trained NeRF LRM with mesh rendering.\nMoreover, we improve the LRM architecture by simplifying several complex\ndesigns in previous LRMs. MeshLRM's NeRF initialization is sequentially trained\nwith low- and high-resolution images; this new LRM training strategy enables\nsignificantly faster convergence and thereby leads to better quality with less\ncompute. Our approach achieves state-of-the-art mesh reconstruction from\nsparse-view inputs and also allows for many downstream applications, including\ntext-to-3D and single-image-to-3D generation. Project page:\nhttps://sarahweiii.github.io/meshlrm/\n", "date": "2024-04-18", "categories": []}
{"id": "2408.03503", "title": "Opening the Black Box of 3D Reconstruction Error Analysis with VECTOR", "abstract": "  Reconstruction of 3D scenes from 2D images is a technical challenge that\nimpacts domains from Earth and planetary sciences and space exploration to\naugmented and virtual reality. Typically, reconstruction algorithms first\nidentify common features across images and then minimize reconstruction errors\nafter estimating the shape of the terrain. This bundle adjustment (BA) step\noptimizes around a single, simplifying scalar value that obfuscates many\npossible causes of reconstruction errors (e.g., initial estimate of the\nposition and orientation of the camera, lighting conditions, ease of feature\ndetection in the terrain). Reconstruction errors can lead to inaccurate\nscientific inferences or endanger a spacecraft exploring a remote environment.\nTo address this challenge, we present VECTOR, a visual analysis tool that\nimproves error inspection for stereo reconstruction BA. VECTOR provides\nanalysts with previously unavailable visibility into feature locations, camera\npose, and computed 3D points. VECTOR was developed in partnership with the\nPerseverance Mars Rover and Ingenuity Mars Helicopter terrain reconstruction\nteam at the NASA Jet Propulsion Laboratory. We report on how this tool was used\nto debug and improve terrain reconstruction for the Mars 2020 mission.\n", "date": "2024-08-07", "categories": []}
{"id": "2411.05362", "title": "From Transparent to Opaque: Rethinking Neural Implicit Surfaces with\n  $\\alpha$-NeuS", "abstract": "  Traditional 3D shape reconstruction techniques from multi-view images, such\nas structure from motion and multi-view stereo, face challenges in\nreconstructing transparent objects. Recent advances in neural radiance fields\nand its variants primarily address opaque or transparent objects, encountering\ndifficulties to reconstruct both transparent and opaque objects simultaneously.\nThis paper introduces $\\alpha$-Neus -- an extension of NeuS -- that proves NeuS\nis unbiased for materials from fully transparent to fully opaque. We find that\ntransparent and opaque surfaces align with the non-negative local minima and\nthe zero iso-surface, respectively, in the learned distance field of NeuS.\nTraditional iso-surfacing extraction algorithms, such as marching cubes, which\nrely on fixed iso-values, are ill-suited for such data. We develop a method to\nextract the transparent and opaque surface simultaneously based on DCUDF. To\nvalidate our approach, we construct a benchmark that includes both real-world\nand synthetic scenes, demonstrating its practical utility and effectiveness.\nOur data and code are publicly available at\nhttps://github.com/728388808/alpha-NeuS.\n", "date": "2024-11-08", "categories": []}
{"id": "2501.10474", "title": "Poxel: Voxel Reconstruction for 3D Printing", "abstract": "  Recent advancements in 3D reconstruction, especially through neural rendering\napproaches like Neural Radiance Fields (NeRF) and Plenoxel, have led to\nhigh-quality 3D visualizations. However, these methods are optimized for\ndigital environments and employ view-dependent color models (RGB) and 2D\nsplatting techniques, which do not translate well to physical 3D printing. This\npaper introduces \"Poxel\", which stands for Printable-Voxel, a voxel-based 3D\nreconstruction framework optimized for photopolymer jetting 3D printing, which\nallows for high-resolution, full-color 3D models using a CMYKWCl color model.\nOur framework directly outputs printable voxel grids by removing\nview-dependency and converting the digital RGB color space to a physical\nCMYKWCl color space suitable for multi-material jetting. The proposed system\nachieves better fidelity and quality in printed models, aligning with the\nrequirements of physical 3D objects.\n", "date": "2025-01-16", "categories": []}
{"id": "2501.11884", "title": "Fast Underwater Scene Reconstruction using Multi-View Stereo and\n  Physical Imaging", "abstract": "  Underwater scene reconstruction poses a substantial challenge because of the\nintricate interplay between light and the medium, resulting in scattering and\nabsorption effects that make both depth estimation and rendering more complex.\nWhile recent Neural Radiance Fields (NeRF) based methods for underwater scenes\nachieve high-quality results by modeling and separating the scattering medium,\nthey still suffer from slow training and rendering speeds. To address these\nlimitations, we propose a novel method that integrates Multi-View Stereo (MVS)\nwith a physics-based underwater image formation model. Our approach consists of\ntwo branches: one for depth estimation using the traditional cost volume\npipeline of MVS, and the other for rendering based on the physics-based image\nformation model. The depth branch improves scene geometry, while the medium\nbranch determines the scattering parameters to achieve precise scene rendering.\nUnlike traditional MVSNet methods that rely on ground-truth depth, our method\ndoes not necessitate the use of depth truth, thus allowing for expedited\ntraining and rendering processes. By leveraging the medium subnet to estimate\nthe medium parameters and combining this with a color MLP for rendering, we\nrestore the true colors of underwater scenes and achieve higher-fidelity\ngeometric representations. Experimental results show that our method enables\nhigh-quality synthesis of novel views in scattering media, clear views\nrestoration by removing the medium, and outperforms existing methods in\nrendering quality and training efficiency.\n", "date": "2025-01-21", "categories": []}
{"id": "2501.12387", "title": "Continuous 3D Perception Model with Persistent State", "abstract": "  We present a unified framework capable of solving a broad range of 3D tasks.\nOur approach features a stateful recurrent model that continuously updates its\nstate representation with each new observation. Given a stream of images, this\nevolving state can be used to generate metric-scale pointmaps (per-pixel 3D\npoints) for each new input in an online fashion. These pointmaps reside within\na common coordinate system, and can be accumulated into a coherent, dense scene\nreconstruction that updates as new images arrive. Our model, called CUT3R\n(Continuous Updating Transformer for 3D Reconstruction), captures rich priors\nof real-world scenes: not only can it predict accurate pointmaps from image\nobservations, but it can also infer unseen regions of the scene by probing at\nvirtual, unobserved views. Our method is simple yet highly flexible, naturally\naccepting varying lengths of images that may be either video streams or\nunordered photo collections, containing both static and dynamic content. We\nevaluate our method on various 3D/4D tasks and demonstrate competitive or\nstate-of-the-art performance in each. Project Page: https://cut3r.github.io/\n", "date": "2025-01-21", "categories": []}
{"id": "2305.17134", "title": "NeuManifold: Neural Watertight Manifold Reconstruction with Efficient\n  and High-Quality Rendering Support", "abstract": "  We present a method for generating high-quality watertight manifold meshes\nfrom multi-view input images. Existing volumetric rendering methods are robust\nin optimization but tend to generate noisy meshes with poor topology.\nDifferentiable rasterization-based methods can generate high-quality meshes but\nare sensitive to initialization. Our method combines the benefits of both\nworlds; we take the geometry initialization obtained from neural volumetric\nfields, and further optimize the geometry as well as a compact neural texture\nrepresentation with differentiable rasterizers. Through extensive experiments,\nwe demonstrate that our method can generate accurate mesh reconstructions with\nfaithful appearance that are comparable to previous volume rendering methods\nwhile being an order of magnitude faster in rendering. We also show that our\ngenerated mesh and neural texture reconstruction is compatible with existing\ngraphics pipelines and enables downstream 3D applications such as simulation.\nProject page: https://sarahweiii.github.io/neumanifold/\n", "date": "2023-05-26", "categories": []}
{"id": "2501.08577", "title": "Scalable and High-Quality Neural Implicit Representation for 3D\n  Reconstruction", "abstract": "  Various SDF-based neural implicit surface reconstruction methods have been\nproposed recently, and have demonstrated remarkable modeling capabilities.\nHowever, due to the global nature and limited representation ability of a\nsingle network, existing methods still suffer from many drawbacks, such as\nlimited accuracy and scale of the reconstruction. In this paper, we propose a\nversatile, scalable and high-quality neural implicit representation to address\nthese issues. We integrate a divide-and-conquer approach into the neural\nSDF-based reconstruction. Specifically, we model the object or scene as a\nfusion of multiple independent local neural SDFs with overlapping regions. The\nconstruction of our representation involves three key steps: (1) constructing\nthe distribution and overlap relationship of the local radiance fields based on\nobject structure or data distribution, (2) relative pose registration for\nadjacent local SDFs, and (3) SDF blending. Thanks to the independent\nrepresentation of each local region, our approach can not only achieve\nhigh-fidelity surface reconstruction, but also enable scalable scene\nreconstruction. Extensive experimental results demonstrate the effectiveness\nand practicality of our proposed method.\n", "date": "2025-01-15", "categories": []}
{"id": "2501.09259", "title": "OpticFusion: Multi-Modal Neural Implicit 3D Reconstruction of\n  Microstructures by Fusing White Light Interferometry and Optical Microscopy", "abstract": "  White Light Interferometry (WLI) is a precise optical tool for measuring the\n3D topography of microstructures. However, conventional WLI cannot capture the\nnatural color of a sample's surface, which is essential for many microscale\nresearch applications that require both 3D geometry and color information.\nPrevious methods have attempted to overcome this limitation by modifying WLI\nhardware and analysis software, but these solutions are often costly. In this\nwork, we address this challenge from a computer vision multi-modal\nreconstruction perspective for the first time. We introduce OpticFusion, a\nnovel approach that uses an additional digital optical microscope (OM) to\nachieve 3D reconstruction with natural color textures using multi-view WLI and\nOM images. Our method employs a two-step data association process to obtain the\nposes of WLI and OM data. By leveraging the neural implicit representation, we\nfuse multi-modal data and apply color decomposition technology to extract the\nsample's natural color. Tested on our multi-modal dataset of various microscale\nsamples, OpticFusion achieves detailed 3D reconstructions with color textures.\nOur method provides an effective tool for practical applications across\nnumerous microscale research fields. The source code and our real-world dataset\nare available at https://github.com/zju3dv/OpticFusion.\n", "date": "2025-01-16", "categories": []}
{"id": "2403.08640", "title": "Refractive COLMAP: Refractive Structure-from-Motion Revisited", "abstract": "  In this paper, we present a complete refractive Structure-from-Motion (RSfM)\nframework for underwater 3D reconstruction using refractive camera setups (for\nboth, flat- and dome-port underwater housings). Despite notable achievements in\nrefractive multi-view geometry over the past decade, a robust, complete and\npublicly available solution for such tasks is not available at present, and\noften practical applications have to resort to approximating refraction effects\nby the intrinsic (distortion) parameters of a pinhole camera model. To fill\nthis gap, we have integrated refraction considerations throughout the entire\nSfM process within the state-of-the-art, open-source SfM framework COLMAP.\nNumerical simulations and reconstruction results on synthetically generated but\nphoto-realistic images with ground truth validate that enabling refraction does\nnot compromise accuracy or robustness as compared to in-air reconstructions.\nFinally, we demonstrate the capability of our approach for large-scale\nrefractive scenarios using a dataset consisting of nearly 6000 images. The\nimplementation is released as open-source at:\nhttps://cau-git.rz.uni-kiel.de/inf-ag-koeser/colmap_underwater.\n", "date": "2024-03-13", "categories": []}
{"id": "2501.08370", "title": "3D Gaussian Splatting with Normal Information for Mesh Extraction and\n  Improved Rendering", "abstract": "  Differentiable 3D Gaussian splatting has emerged as an efficient and flexible\nrendering technique for representing complex scenes from a collection of 2D\nviews and enabling high-quality real-time novel-view synthesis. However, its\nreliance on photometric losses can lead to imprecisely reconstructed geometry\nand extracted meshes, especially in regions with high curvature or fine detail.\nWe propose a novel regularization method using the gradients of a signed\ndistance function estimated from the Gaussians, to improve the quality of\nrendering while also extracting a surface mesh. The regularizing normal\nsupervision facilitates better rendering and mesh reconstruction, which is\ncrucial for downstream applications in video generation, animation, AR-VR and\ngaming. We demonstrate the effectiveness of our approach on datasets such as\nMip-NeRF360, Tanks and Temples, and Deep-Blending. Our method scores higher on\nphotorealism metrics compared to other mesh extracting rendering methods\nwithout compromising mesh quality.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.07260", "title": "Skip Mamba Diffusion for Monocular 3D Semantic Scene Completion", "abstract": "  3D semantic scene completion is critical for multiple downstream tasks in\nautonomous systems. It estimates missing geometric and semantic information in\nthe acquired scene data. Due to the challenging real-world conditions, this\ntask usually demands complex models that process multi-modal data to achieve\nacceptable performance. We propose a unique neural model, leveraging advances\nfrom the state space and diffusion generative modeling to achieve remarkable 3D\nsemantic scene completion performance with monocular image input. Our technique\nprocesses the data in the conditioned latent space of a variational autoencoder\nwhere diffusion modeling is carried out with an innovative state space\ntechnique. A key component of our neural network is the proposed Skimba (Skip\nMamba) denoiser, which is adept at efficiently processing long-sequence data.\nThe Skimba diffusion model is integral to our 3D scene completion network,\nincorporating a triple Mamba structure, dimensional decomposition residuals and\nvarying dilations along three directions. We also adopt a variant of this\nnetwork for the subsequent semantic segmentation stage of our method. Extensive\nevaluation on the standard SemanticKITTI and SSCBench-KITTI360 datasets show\nthat our approach not only outperforms other monocular techniques by a large\nmargin, it also achieves competitive performance against stereo methods. The\ncode is available at https://github.com/xrkong/skimba\n", "date": "2025-01-13", "categories": []}
{"id": "2501.07478", "title": "3DGS-to-PC: Convert a 3D Gaussian Splatting Scene into a Dense Point\n  Cloud or Mesh", "abstract": "  3D Gaussian Splatting (3DGS) excels at producing highly detailed 3D\nreconstructions, but these scenes often require specialised renderers for\neffective visualisation. In contrast, point clouds are a widely used 3D\nrepresentation and are compatible with most popular 3D processing software, yet\nconverting 3DGS scenes into point clouds is a complex challenge. In this work\nwe introduce 3DGS-to-PC, a flexible and highly customisable framework that is\ncapable of transforming 3DGS scenes into dense, high-accuracy point clouds. We\nsample points probabilistically from each Gaussian as a 3D density function. We\nadditionally threshold new points using the Mahalanobis distance to the\nGaussian centre, preventing extreme outliers. The result is a point cloud that\nclosely represents the shape encoded into the 3D Gaussian scene. Individual\nGaussians use spherical harmonics to adapt colours depending on view, and each\npoint may contribute only subtle colour hints to the resulting rendered scene.\nTo avoid spurious or incorrect colours that do not fit with the final point\ncloud, we recalculate Gaussian colours via a customised image rendering\napproach, assigning each Gaussian the colour of the pixel to which it\ncontributes most across all views. 3DGS-to-PC also supports mesh generation\nthrough Poisson Surface Reconstruction, applied to points sampled from\npredicted surface Gaussians. This allows coloured meshes to be generated from\n3DGS scenes without the need for re-training. This package is highly\ncustomisable and capability of simple integration into existing 3DGS pipelines.\n3DGS-to-PC provides a powerful tool for converting 3DGS data into point cloud\nand surface-based formats.\n", "date": "2025-01-13", "categories": []}
{"id": "2412.18696", "title": "STITCH: Surface reconstrucTion using Implicit neural representations\n  with Topology Constraints and persistent Homology", "abstract": "  We present STITCH, a novel approach for neural implicit surface\nreconstruction of a sparse and irregularly spaced point cloud while enforcing\ntopological constraints (such as having a single connected component). We\ndevelop a new differentiable framework based on persistent homology to\nformulate topological loss terms that enforce the prior of a single 2-manifold\nobject. Our method demonstrates excellent performance in preserving the\ntopology of complex 3D geometries, evident through both visual and empirical\ncomparisons. We supplement this with a theoretical analysis, and provably show\nthat optimizing the loss with stochastic (sub)gradient descent leads to\nconvergence and enables reconstructing shapes with a single connected\ncomponent. Our approach showcases the integration of differentiable topological\ndata analysis tools for implicit surface reconstruction.\n", "date": "2024-12-24", "categories": []}
{"id": "2501.04074", "title": "NeRFs are Mirror Detectors: Using Structural Similarity for Multi-View\n  Mirror Scene Reconstruction with 3D Surface Primitives", "abstract": "  While neural radiance fields (NeRF) led to a breakthrough in photorealistic\nnovel view synthesis, handling mirroring surfaces still denotes a particular\nchallenge as they introduce severe inconsistencies in the scene representation.\nPrevious attempts either focus on reconstructing single reflective objects or\nrely on strong supervision guidance in terms of additional user-provided\nannotations of visible image regions of the mirrors, thereby limiting the\npractical usability. In contrast, in this paper, we present NeRF-MD, a method\nwhich shows that NeRFs can be considered as mirror detectors and which is\ncapable of reconstructing neural radiance fields of scenes containing mirroring\nsurfaces without the need for prior annotations. To this end, we first compute\nan initial estimate of the scene geometry by training a standard NeRF using a\ndepth reprojection loss. Our key insight lies in the fact that parts of the\nscene corresponding to a mirroring surface will still exhibit a significant\nphotometric inconsistency, whereas the remaining parts are already\nreconstructed in a plausible manner. This allows us to detect mirror surfaces\nby fitting geometric primitives to such inconsistent regions in this initial\nstage of the training. Using this information, we then jointly optimize the\nradiance field and mirror geometry in a second training stage to refine their\nquality. We demonstrate the capability of our method to allow the faithful\ndetection of mirrors in the scene as well as the reconstruction of a single\nconsistent scene representation, and demonstrate its potential in comparison to\nbaseline and mirror-aware approaches.\n", "date": "2025-01-07", "categories": []}
{"id": "2501.04689", "title": "SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single\n  Images", "abstract": "  We study the problem of single-image 3D object reconstruction. Recent works\nhave diverged into two directions: regression-based modeling and generative\nmodeling. Regression methods efficiently infer visible surfaces, but struggle\nwith occluded regions. Generative methods handle uncertain regions better by\nmodeling distributions, but are computationally expensive and the generation is\noften misaligned with visible surfaces. In this paper, we present SPAR3D, a\nnovel two-stage approach aiming to take the best of both directions. The first\nstage of SPAR3D generates sparse 3D point clouds using a lightweight point\ndiffusion model, which has a fast sampling speed. The second stage uses both\nthe sampled point cloud and the input image to create highly detailed meshes.\nOur two-stage design enables probabilistic modeling of the ill-posed\nsingle-image 3D task while maintaining high computational efficiency and great\noutput fidelity. Using point clouds as an intermediate representation further\nallows for interactive user edits. Evaluated on diverse datasets, SPAR3D\ndemonstrates superior performance over previous state-of-the-art methods, at an\ninference speed of 0.7 seconds. Project page with code and model:\nhttps://spar3d.github.io\n", "date": "2025-01-08", "categories": []}
{"id": "2501.02807", "title": "AE-NeRF: Augmenting Event-Based Neural Radiance Fields for Non-ideal\n  Conditions and Larger Scene", "abstract": "  Compared to frame-based methods, computational neuromorphic imaging using\nevent cameras offers significant advantages, such as minimal motion blur,\nenhanced temporal resolution, and high dynamic range. The multi-view\nconsistency of Neural Radiance Fields combined with the unique benefits of\nevent cameras, has spurred recent research into reconstructing NeRF from data\ncaptured by moving event cameras. While showing impressive performance,\nexisting methods rely on ideal conditions with the availability of uniform and\nhigh-quality event sequences and accurate camera poses, and mainly focus on the\nobject level reconstruction, thus limiting their practical applications. In\nthis work, we propose AE-NeRF to address the challenges of learning event-based\nNeRF from non-ideal conditions, including non-uniform event sequences, noisy\nposes, and various scales of scenes. Our method exploits the density of event\nstreams and jointly learn a pose correction module with an event-based NeRF\n(e-NeRF) framework for robust 3D reconstruction from inaccurate camera poses.\nTo generalize to larger scenes, we propose hierarchical event distillation with\na proposal e-NeRF network and a vanilla e-NeRF network to resample and refine\nthe reconstruction process. We further propose an event reconstruction loss and\na temporal loss to improve the view consistency of the reconstructed scene. We\nestablished a comprehensive benchmark that includes large-scale scenes to\nsimulate practical non-ideal conditions, incorporating both synthetic and\nchallenging real-world event datasets. The experimental results show that our\nmethod achieves a new state-of-the-art in event-based 3D reconstruction.\n", "date": "2025-01-06", "categories": []}
{"id": "2410.22715", "title": "SCRREAM : SCan, Register, REnder And Map:A Framework for Annotating\n  Accurate and Dense 3D Indoor Scenes with a Benchmark", "abstract": "  Traditionally, 3d indoor datasets have generally prioritized scale over\nground-truth accuracy in order to obtain improved generalization. However,\nusing these datasets to evaluate dense geometry tasks, such as depth rendering,\ncan be problematic as the meshes of the dataset are often incomplete and may\nproduce wrong ground truth to evaluate the details. In this paper, we propose\nSCRREAM, a dataset annotation framework that allows annotation of fully dense\nmeshes of objects in the scene and registers camera poses on the real image\nsequence, which can produce accurate ground truth for both sparse 3D as well as\ndense 3D tasks. We show the details of the dataset annotation pipeline and\nshowcase four possible variants of datasets that can be obtained from our\nframework with example scenes, such as indoor reconstruction and SLAM, scene\nediting & object removal, human reconstruction and 6d pose estimation. Recent\npipelines for indoor reconstruction and SLAM serve as new benchmarks. In\ncontrast to previous indoor dataset, our design allows to evaluate dense\ngeometry tasks on eleven sample scenes against accurately rendered ground truth\ndepth maps.\n", "date": "2024-10-30", "categories": []}
{"id": "2405.18426", "title": "GFlow: Recovering 4D World from Monocular Video", "abstract": "  Recovering 4D world from monocular video is a crucial yet challenging task.\nConventional methods usually rely on the assumptions of multi-view videos,\nknown camera parameters, or static scenes. In this paper, we relax all these\nconstraints and tackle a highly ambitious but practical task: With only one\nmonocular video without camera parameters, we aim to recover the dynamic 3D\nworld alongside the camera poses. To solve this, we introduce GFlow, a new\nframework that utilizes only 2D priors (depth and optical flow) to lift a video\nto a 4D scene, as a flow of 3D Gaussians through space and time. GFlow starts\nby segmenting the video into still and moving parts, then alternates between\noptimizing camera poses and the dynamics of the 3D Gaussian points. This method\nensures consistency among adjacent points and smooth transitions between\nframes. Since dynamic scenes always continually introduce new visual content,\nwe present prior-driven initialization and pixel-wise densification strategy\nfor Gaussian points to integrate new content. By combining all those\ntechniques, GFlow transcends the boundaries of 4D recovery from causal videos;\nit naturally enables tracking of points and segmentation of moving objects\nacross frames. Additionally, GFlow estimates the camera poses for each frame,\nenabling novel view synthesis by changing camera pose. This capability\nfacilitates extensive scene-level or object-level editing, highlighting GFlow's\nversatility and effectiveness. Visit our project page at:\nhttps://littlepure2333.github.io/GFlow\n", "date": "2024-05-28", "categories": []}
{"id": "2408.04831", "title": "AugGS: Self-augmented Gaussians with Structural Masks for Sparse-view 3D\n  Reconstruction", "abstract": "  Sparse-view 3D reconstruction is a major challenge in computer vision, aiming\nto create complete three-dimensional models from limited viewing angles. Key\nobstacles include: 1) a small number of input images with inconsistent\ninformation; 2) dependence on input image quality; and 3) large model parameter\nsizes. To tackle these issues, we propose a self-augmented two-stage Gaussian\nsplatting framework enhanced with structural masks for sparse-view 3D\nreconstruction. Initially, our method generates a basic 3D Gaussian\nrepresentation from sparse inputs and renders multi-view images. We then\nfine-tune a pre-trained 2D diffusion model to enhance these images, using them\nas augmented data to further optimize the 3D Gaussians. Additionally, a\nstructural masking strategy during training enhances the model's robustness to\nsparse inputs and noise. Experiments on benchmarks like MipNeRF360,\nOmniObject3D, and OpenIllumination demonstrate that our approach achieves\nstate-of-the-art performance in perceptual quality and multi-view consistency\nwith sparse inputs.\n", "date": "2024-08-09", "categories": []}
{"id": "2501.00352", "title": "PanoSLAM: Panoptic 3D Scene Reconstruction via Gaussian SLAM", "abstract": "  Understanding geometric, semantic, and instance information in 3D scenes from\nsequential video data is essential for applications in robotics and augmented\nreality. However, existing Simultaneous Localization and Mapping (SLAM) methods\ngenerally focus on either geometric or semantic reconstruction. In this paper,\nwe introduce PanoSLAM, the first SLAM system to integrate geometric\nreconstruction, 3D semantic segmentation, and 3D instance segmentation within a\nunified framework. Our approach builds upon 3D Gaussian Splatting, modified\nwith several critical components to enable efficient rendering of depth, color,\nsemantic, and instance information from arbitrary viewpoints. To achieve\npanoptic 3D scene reconstruction from sequential RGB-D videos, we propose an\nonline Spatial-Temporal Lifting (STL) module that transfers 2D panoptic\npredictions from vision models into 3D Gaussian representations. This STL\nmodule addresses the challenges of label noise and inconsistencies in 2D\npredictions by refining the pseudo labels across multi-view inputs, creating a\ncoherent 3D representation that enhances segmentation accuracy. Our experiments\nshow that PanoSLAM outperforms recent semantic SLAM methods in both mapping and\ntracking accuracy. For the first time, it achieves panoptic 3D reconstruction\nof open-world environments directly from the RGB-D video.\n(https://github.com/runnanchen/PanoSLAM)\n", "date": "2024-12-31", "categories": []}
{"id": "2501.00602", "title": "STORM: Spatio-Temporal Reconstruction Model for Large-Scale Outdoor\n  Scenes", "abstract": "  We present STORM, a spatio-temporal reconstruction model designed for\nreconstructing dynamic outdoor scenes from sparse observations. Existing\ndynamic reconstruction methods often rely on per-scene optimization, dense\nobservations across space and time, and strong motion supervision, resulting in\nlengthy optimization times, limited generalization to novel views or scenes,\nand degenerated quality caused by noisy pseudo-labels for dynamics. To address\nthese challenges, STORM leverages a data-driven Transformer architecture that\ndirectly infers dynamic 3D scene representations--parameterized by 3D Gaussians\nand their velocities--in a single forward pass. Our key design is to aggregate\n3D Gaussians from all frames using self-supervised scene flows, transforming\nthem to the target timestep to enable complete (i.e., \"amodal\") reconstructions\nfrom arbitrary viewpoints at any moment in time. As an emergent property, STORM\nautomatically captures dynamic instances and generates high-quality masks using\nonly reconstruction losses. Extensive experiments on public datasets show that\nSTORM achieves precise dynamic scene reconstruction, surpassing\nstate-of-the-art per-scene optimization methods (+4.3 to 6.6 PSNR) and existing\nfeed-forward approaches (+2.1 to 4.7 PSNR) in dynamic regions. STORM\nreconstructs large-scale outdoor scenes in 200ms, supports real-time rendering,\nand outperforms competitors in scene flow estimation, improving 3D EPE by\n0.422m and Acc5 by 28.02%. Beyond reconstruction, we showcase four additional\napplications of our model, illustrating the potential of self-supervised\nlearning for broader dynamic scene understanding.\n", "date": "2024-12-31", "categories": []}
{"id": "2501.01049", "title": "TS-SatMVSNet: Slope Aware Height Estimation for Large-Scale Earth\n  Terrain Multi-view Stereo", "abstract": "  3D terrain reconstruction with remote sensing imagery achieves cost-effective\nand large-scale earth observation and is crucial for safeguarding natural\ndisasters, monitoring ecological changes, and preserving the\nenvironment.Recently, learning-based multi-view stereo~(MVS) methods have shown\npromise in this task. However, these methods simply modify the general\nlearning-based MVS framework for height estimation, which overlooks the terrain\ncharacteristics and results in insufficient accuracy. Considering that the\nEarth's surface generally undulates with no drastic changes and can be measured\nby slope, integrating slope considerations into MVS frameworks could enhance\nthe accuracy of terrain reconstructions. To this end, we propose an end-to-end\nslope-aware height estimation network named TS-SatMVSNet for large-scale remote\nsensing terrain reconstruction.To effectively obtain the slope representation,\ndrawing from mathematical gradient concepts, we innovatively proposed a\nheight-based slope calculation strategy to first calculate a slope map from a\nheight map to measure the terrain undulation. To fully integrate slope\ninformation into the MVS pipeline, we separately design two slope-guided\nmodules to enhance reconstruction outcomes at both micro and macro levels.\nSpecifically, at the micro level, we designed a slope-guided interval partition\nmodule for refined height estimation using slope values. At the macro level, a\nheight correction module is proposed, using a learnable Gaussian smoothing\noperator to amend the inaccurate height values. Additionally, to enhance the\nefficacy of height estimation, we proposed a slope direction loss for\nimplicitly optimizing height estimation results. Extensive experiments on the\nWHU-TLC dataset and MVS3D dataset show that our proposed method achieves\nstate-of-the-art performance and demonstrates competitive generalization\nability.\n", "date": "2025-01-02", "categories": []}
{"id": "2501.01196", "title": "Sparis: Neural Implicit Surface Reconstruction of Indoor Scenes from\n  Sparse Views", "abstract": "  In recent years, reconstructing indoor scene geometry from multi-view images\nhas achieved encouraging accomplishments. Current methods incorporate monocular\npriors into neural implicit surface models to achieve high-quality\nreconstructions. However, these methods require hundreds of images for scene\nreconstruction. When only a limited number of views are available as input, the\nperformance of monocular priors deteriorates due to scale ambiguity, leading to\nthe collapse of the reconstructed scene geometry. In this paper, we propose a\nnew method, named Sparis, for indoor surface reconstruction from sparse views.\nSpecifically, we investigate the impact of monocular priors on sparse scene\nreconstruction, introducing a novel prior based on inter-image matching\ninformation. Our prior offers more accurate depth information while ensuring\ncross-view matching consistency. Additionally, we employ an angular filter\nstrategy and an epipolar matching weight function, aiming to reduce errors due\nto view matching inaccuracies, thereby refining the inter-image prior for\nimproved reconstruction accuracy. The experiments conducted on widely used\nbenchmarks demonstrate superior performance in sparse-view scene\nreconstruction.\n", "date": "2025-01-02", "categories": []}
{"id": "2412.19849", "title": "3D Face Reconstruction With Geometry Details From a Single Color Image\n  Under Occluded Scenes", "abstract": "  3D face reconstruction technology aims to generate a face stereo model\nnaturally and realistically. Previous deep face reconstruction approaches are\ntypically designed to generate convincing textures and cannot generalize well\nto multiple occluded scenarios simultaneously. By introducing bump mapping, we\nsuccessfully added mid-level details to coarse 3D faces. More innovatively, our\nmethod takes into account occlusion scenarios. Thus on top of common 3D face\nreconstruction approaches, we in this paper propose a unified framework to\nhandle multiple types of obstruction simultaneously (e.g., hair, palms and\nglasses et al.).Extensive experiments and comparisons demonstrate that our\nmethod can generate high-quality reconstruction results with geometry details\nfrom captured facial images under occluded scenes.\n", "date": "2024-12-25", "categories": []}
{"id": "2412.20328", "title": "Dual-Level Precision Edges Guided Multi-View Stereo with Accurate\n  Planarization", "abstract": "  The reconstruction of low-textured areas is a prominent research focus in\nmulti-view stereo (MVS). In recent years, traditional MVS methods have\nperformed exceptionally well in reconstructing low-textured areas by\nconstructing plane models. However, these methods often encounter issues such\nas crossing object boundaries and limited perception ranges, which undermine\nthe robustness of plane model construction. Building on previous work\n(APD-MVS), we propose the DPE-MVS method. By introducing dual-level precision\nedge information, including fine and coarse edges, we enhance the robustness of\nplane model construction, thereby improving reconstruction accuracy in\nlow-textured areas. Furthermore, by leveraging edge information, we refine the\nsampling strategy in conventional PatchMatch MVS and propose an adaptive patch\nsize adjustment approach to optimize matching cost calculation in both\nstochastic and low-textured areas. This additional use of edge information\nallows for more precise and robust matching. Our method achieves\nstate-of-the-art performance on the ETH3D and Tanks & Temples benchmarks.\nNotably, our method outperforms all published methods on the ETH3D benchmark.\n", "date": "2024-12-29", "categories": []}
{"id": "2412.19130", "title": "MVS-GS: High-Quality 3D Gaussian Splatting Mapping via Online Multi-View\n  Stereo", "abstract": "  This study addresses the challenge of online 3D model generation for neural\nrendering using an RGB image stream. Previous research has tackled this issue\nby incorporating Neural Radiance Fields (NeRF) or 3D Gaussian Splatting (3DGS)\nas scene representations within dense SLAM methods. However, most studies focus\nprimarily on estimating coarse 3D scenes rather than achieving detailed\nreconstructions. Moreover, depth estimation based solely on images is often\nambiguous, resulting in low-quality 3D models that lead to inaccurate\nrenderings. To overcome these limitations, we propose a novel framework for\nhigh-quality 3DGS modeling that leverages an online multi-view stereo (MVS)\napproach. Our method estimates MVS depth using sequential frames from a local\ntime window and applies comprehensive depth refinement techniques to filter out\noutliers, enabling accurate initialization of Gaussians in 3DGS. Furthermore,\nwe introduce a parallelized backend module that optimizes the 3DGS model\nefficiently, ensuring timely updates with each new keyframe. Experimental\nresults demonstrate that our method outperforms state-of-the-art dense SLAM\nmethods, particularly excelling in challenging outdoor environments.\n", "date": "2024-12-26", "categories": []}
{"id": "2412.19518", "title": "Dust to Tower: Coarse-to-Fine Photo-Realistic Scene Reconstruction from\n  Sparse Uncalibrated Images", "abstract": "  Photo-realistic scene reconstruction from sparse-view, uncalibrated images is\nhighly required in practice. Although some successes have been made, existing\nmethods are either Sparse-View but require accurate camera parameters (i.e.,\nintrinsic and extrinsic), or SfM-free but need densely captured images. To\ncombine the advantages of both methods while addressing their respective\nweaknesses, we propose Dust to Tower (D2T), an accurate and efficient\ncoarse-to-fine framework to optimize 3DGS and image poses simultaneously from\nsparse and uncalibrated images. Our key idea is to first construct a coarse\nmodel efficiently and subsequently refine it using warped and inpainted images\nat novel viewpoints. To do this, we first introduce a Coarse Construction\nModule (CCM) which exploits a fast Multi-View Stereo model to initialize a 3D\nGaussian Splatting (3DGS) and recover initial camera poses. To refine the 3D\nmodel at novel viewpoints, we propose a Confidence Aware Depth Alignment (CADA)\nmodule to refine the coarse depth maps by aligning their confident parts with\nestimated depths by a Mono-depth model. Then, a Warped Image-Guided Inpainting\n(WIGI) module is proposed to warp the training images to novel viewpoints by\nthe refined depth maps, and inpainting is applied to fulfill the ``holes\" in\nthe warped images caused by view-direction changes, providing high-quality\nsupervision to further optimize the 3D model and the camera poses. Extensive\nexperiments and ablation studies demonstrate the validity of D2T and its design\nchoices, achieving state-of-the-art performance in both tasks of novel view\nsynthesis and pose estimation while keeping high efficiency. Codes will be\npublicly available.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.19584", "title": "DAS3R: Dynamics-Aware Gaussian Splatting for Static Scene Reconstruction", "abstract": "  We propose a novel framework for scene decomposition and static background\nreconstruction from everyday videos. By integrating the trained motion masks\nand modeling the static scene as Gaussian splats with dynamics-aware\noptimization, our method achieves more accurate background reconstruction\nresults than previous works. Our proposed method is termed DAS3R, an\nabbreviation for Dynamics-Aware Gaussian Splatting for Static Scene\nReconstruction. Compared to existing methods, DAS3R is more robust in complex\nmotion scenarios, capable of handling videos where dynamic objects occupy a\nsignificant portion of the scene, and does not require camera pose inputs or\npoint cloud data from SLAM-based methods. We compared DAS3R against recent\ndistractor-free approaches on the DAVIS and Sintel datasets; DAS3R demonstrates\nenhanced performance and robustness with a margin of more than 2 dB in PSNR.\nThe project's webpage can be accessed via \\url{https://kai422.github.io/DAS3R/}\n", "date": "2024-12-27", "categories": []}
{"id": "2412.19720", "title": "Sharpening Neural Implicit Functions with Frequency Consolidation Priors", "abstract": "  Signed Distance Functions (SDFs) are vital implicit representations to\nrepresent high fidelity 3D surfaces. Current methods mainly leverage a neural\nnetwork to learn an SDF from various supervisions including signed distances,\n3D point clouds, or multi-view images. However, due to various reasons\nincluding the bias of neural network on low frequency content, 3D unaware\nsampling, sparsity in point clouds, or low resolutions of images, neural\nimplicit representations still struggle to represent geometries with high\nfrequency components like sharp structures, especially for the ones learned\nfrom images or point clouds. To overcome this challenge, we introduce a method\nto sharpen a low frequency SDF observation by recovering its high frequency\ncomponents, pursuing a sharper and more complete surface. Our key idea is to\nlearn a mapping from a low frequency observation to a full frequency coverage\nin a data-driven manner, leading to a prior knowledge of shape consolidation in\nthe frequency domain, dubbed frequency consolidation priors. To better\ngeneralize a learned prior to unseen shapes, we introduce to represent\nfrequency components as embeddings and disentangle the embedding of the low\nfrequency component from the embedding of the full frequency component. This\ndisentanglement allows the prior to generalize on an unseen low frequency\nobservation by simply recovering its full frequency embedding through a\ntest-time self-reconstruction. Our evaluations under widely used benchmarks or\nreal scenes show that our method can recover high frequency component and\nproduce more accurate surfaces than the latest methods. The code, data, and\npre-trained models are available at \\url{https://github.com/chenchao15/FCP}.\n", "date": "2024-12-27", "categories": []}
{"id": "2407.03771", "title": "SpikeGS: Reconstruct 3D scene via fast-moving bio-inspired sensors", "abstract": "  3D Gaussian Splatting (3DGS) demonstrates unparalleled superior performance\nin 3D scene reconstruction. However, 3DGS heavily relies on the sharp images.\nFulfilling this requirement can be challenging in real-world scenarios\nespecially when the camera moves fast, which severely limits the application of\n3DGS. To address these challenges, we proposed Spike Gausian Splatting\n(SpikeGS), the first framework that integrates the spike streams into 3DGS\npipeline to reconstruct 3D scenes via a fast-moving bio-inspired camera. With\naccumulation rasterization, interval supervision, and a specially designed\npipeline, SpikeGS extracts detailed geometry and texture from high temporal\nresolution but texture lacking spike stream, reconstructs 3D scenes captured in\n1 second. Extensive experiments on multiple synthetic and real-world datasets\ndemonstrate the superiority of SpikeGS compared with existing spike-based and\ndeblur 3D scene reconstruction methods. Codes and data will be released soon.\n", "date": "2024-07-04", "categories": []}
{"id": "2409.12753", "title": "DrivingForward: Feed-forward 3D Gaussian Splatting for Driving Scene\n  Reconstruction from Flexible Surround-view Input", "abstract": "  We propose DrivingForward, a feed-forward Gaussian Splatting model that\nreconstructs driving scenes from flexible surround-view input. Driving scene\nimages from vehicle-mounted cameras are typically sparse, with limited overlap,\nand the movement of the vehicle further complicates the acquisition of camera\nextrinsics. To tackle these challenges and achieve real-time reconstruction, we\njointly train a pose network, a depth network, and a Gaussian network to\npredict the Gaussian primitives that represent the driving scenes. The pose\nnetwork and depth network determine the position of the Gaussian primitives in\na self-supervised manner, without using depth ground truth and camera\nextrinsics during training. The Gaussian network independently predicts\nprimitive parameters from each input image, including covariance, opacity, and\nspherical harmonics coefficients. At the inference stage, our model can achieve\nfeed-forward reconstruction from flexible multi-frame surround-view input.\nExperiments on the nuScenes dataset show that our model outperforms existing\nstate-of-the-art feed-forward and scene-optimized reconstruction methods in\nterms of reconstruction.\n", "date": "2024-09-19", "categories": []}
{"id": "2412.16467", "title": "Sensing Surface Patches in Volume Rendering for Inferring Signed\n  Distance Functions", "abstract": "  It is vital to recover 3D geometry from multi-view RGB images in many 3D\ncomputer vision tasks. The latest methods infer the geometry represented as a\nsigned distance field by minimizing the rendering error on the field through\nvolume rendering. However, it is still challenging to explicitly impose\nconstraints on surfaces for inferring more geometry details due to the limited\nability of sensing surfaces in volume rendering. To resolve this problem, we\nintroduce a method to infer signed distance functions (SDFs) with a better\nsense of surfaces through volume rendering. Using the gradients and signed\ndistances, we establish a small surface patch centered at the estimated\nintersection along a ray by pulling points randomly sampled nearby. Hence, we\nare able to explicitly impose surface constraints on the sensed surface patch,\nsuch as multi-view photo consistency and supervision from depth or normal\npriors, through volume rendering. We evaluate our method by numerical and\nvisual comparisons on scene benchmarks. Our superiority over the latest methods\njustifies our effectiveness.\n", "date": "2024-12-21", "categories": []}
{"id": "2411.15723", "title": "GSurf: 3D Reconstruction via Signed Distance Fields with Direct Gaussian\n  Supervision", "abstract": "  Surface reconstruction from multi-view images is a core challenge in 3D\nvision. Recent studies have explored signed distance fields (SDF) within Neural\nRadiance Fields (NeRF) to achieve high-fidelity surface reconstructions.\nHowever, these approaches often suffer from slow training and rendering speeds\ncompared to 3D Gaussian splatting (3DGS). Current state-of-the-art techniques\nattempt to fuse depth information to extract geometry from 3DGS, but frequently\nresult in incomplete reconstructions and fragmented surfaces. In this paper, we\nintroduce GSurf, a novel end-to-end method for learning a signed distance field\ndirectly from Gaussian primitives. The continuous and smooth nature of SDF\naddresses common issues in the 3DGS family, such as holes resulting from noisy\nor missing depth data. By using Gaussian splatting for rendering, GSurf avoids\nthe redundant volume rendering typically required in other GS and SDF\nintegrations. Consequently, GSurf achieves faster training and rendering speeds\nwhile delivering 3D reconstruction quality comparable to neural implicit\nsurface methods, such as VolSDF and NeuS. Experimental results across various\nbenchmark datasets demonstrate the effectiveness of our method in producing\nhigh-fidelity 3D reconstructions.\n", "date": "2024-11-24", "categories": []}
{"id": "2412.15400", "title": "SolidGS: Consolidating Gaussian Surfel Splatting for Sparse-View Surface\n  Reconstruction", "abstract": "  Gaussian splatting has achieved impressive improvements for both novel-view\nsynthesis and surface reconstruction from multi-view images. However, current\nmethods still struggle to reconstruct high-quality surfaces from only sparse\nview input images using Gaussian splatting. In this paper, we propose a novel\nmethod called SolidGS to address this problem. We observed that the\nreconstructed geometry can be severely inconsistent across multi-views, due to\nthe property of Gaussian function in geometry rendering. This motivates us to\nconsolidate all Gaussians by adopting a more solid kernel function, which\neffectively improves the surface reconstruction quality. With the additional\nhelp of geometrical regularization and monocular normal estimation, our method\nachieves superior performance on the sparse view surface reconstruction than\nall the Gaussian splatting methods and neural field methods on the widely used\nDTU, Tanks-and-Temples, and LLFF datasets.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.15550", "title": "EGSRAL: An Enhanced 3D Gaussian Splatting based Renderer with Automated\n  Labeling for Large-Scale Driving Scene", "abstract": "  3D Gaussian Splatting (3D GS) has gained popularity due to its faster\nrendering speed and high-quality novel view synthesis. Some researchers have\nexplored using 3D GS for reconstructing driving scenes. However, these methods\noften rely on various data types, such as depth maps, 3D boxes, and\ntrajectories of moving objects. Additionally, the lack of annotations for\nsynthesized images limits their direct application in downstream tasks. To\naddress these issues, we propose EGSRAL, a 3D GS-based method that relies\nsolely on training images without extra annotations. EGSRAL enhances 3D GS's\ncapability to model both dynamic objects and static backgrounds and introduces\na novel adaptor for auto labeling, generating corresponding annotations based\non existing annotations. We also propose a grouping strategy for vanilla 3D GS\nto address perspective issues in rendering large-scale, complex scenes. Our\nmethod achieves state-of-the-art performance on multiple datasets without any\nextra annotation. For example, the PSNR metric reaches 29.04 on the nuScenes\ndataset. Moreover, our automated labeling can significantly improve the\nperformance of 2D/3D detection tasks. Code is available at\nhttps://github.com/jiangxb98/EGSRAL.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.15909", "title": "CCNDF: Curvature Constrained Neural Distance Fields from 3D LiDAR\n  Sequences", "abstract": "  Neural distance fields (NDF) have emerged as a powerful tool for addressing\nchallenges in 3D computer vision and graphics downstream problems. While\nsignificant progress has been made to learn NDF from various kind of sensor\ndata, a crucial aspect that demands attention is the supervision of neural\nfields during training as the ground-truth NDFs are not available for\nlarge-scale outdoor scenes. Previous works have utilized various forms of\nexpected signed distance to guide model learning. Yet, these approaches often\nneed to pay more attention to critical considerations of surface geometry and\nare limited to small-scale implementations. To this end, we propose a novel\nmethodology leveraging second-order derivatives of the signed distance field\nfor improved neural field learning. Our approach addresses limitations by\naccurately estimating signed distance, offering a more comprehensive\nunderstanding of underlying geometry. To assess the efficacy of our\nmethodology, we conducted comparative evaluations against prevalent methods for\nmapping and localization tasks, which are primary application areas of NDF. Our\nresults demonstrate the superiority of the proposed approach, highlighting its\npotential for advancing the capabilities of neural distance fields in computer\nvision and graphics applications.\n", "date": "2024-12-20", "categories": []}
{"id": "2411.18613", "title": "CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models", "abstract": "  We present CAT4D, a method for creating 4D (dynamic 3D) scenes from monocular\nvideo. CAT4D leverages a multi-view video diffusion model trained on a diverse\ncombination of datasets to enable novel view synthesis at any specified camera\nposes and timestamps. Combined with a novel sampling approach, this model can\ntransform a single monocular video into a multi-view video, enabling robust 4D\nreconstruction via optimization of a deformable 3D Gaussian representation. We\ndemonstrate competitive performance on novel view synthesis and dynamic scene\nreconstruction benchmarks, and highlight the creative capabilities for 4D scene\ngeneration from real or generated videos. See our project page for results and\ninteractive demos: https://cat-4d.github.io/.\n", "date": "2024-11-27", "categories": []}
{"id": "2412.14418", "title": "An Immersive Multi-Elevation Multi-Seasonal Dataset for 3D\n  Reconstruction and Visualization", "abstract": "  Significant progress has been made in photo-realistic scene reconstruction\nover recent years. Various disparate efforts have enabled capabilities such as\nmulti-appearance or large-scale modeling; however, there lacks a welldesigned\ndataset that can evaluate the holistic progress of scene reconstruction. We\nintroduce a collection of imagery of the Johns Hopkins Homewood Campus,\nacquired at different seasons, times of day, in multiple elevations, and across\na large scale. We perform a multi-stage calibration process, which efficiently\nrecover camera parameters from phone and drone cameras. This dataset can enable\nresearchers to rigorously explore challenges in unconstrained settings,\nincluding effects of inconsistent illumination, reconstruction from large scale\nand from significantly different perspectives, etc.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.14568", "title": "Improving Geometry in Sparse-View 3DGS via Reprojection-based DoF\n  Separation", "abstract": "  Recent learning-based Multi-View Stereo models have demonstrated\nstate-of-the-art performance in sparse-view 3D reconstruction. However,\ndirectly applying 3D Gaussian Splatting (3DGS) as a refinement step following\nthese models presents challenges. We hypothesize that the excessive positional\ndegrees of freedom (DoFs) in Gaussians induce geometry distortion, fitting\ncolor patterns at the cost of structural fidelity. To address this, we propose\nreprojection-based DoF separation, a method distinguishing positional DoFs in\nterms of uncertainty: image-plane-parallel DoFs and ray-aligned DoF. To\nindependently manage each DoF, we introduce a reprojection process along with\ntailored constraints for each DoF. Through experiments across various datasets,\nwe confirm that separating the positional DoFs of Gaussians and applying\ntargeted constraints effectively suppresses geometric artifacts, producing\nreconstruction results that are both visually and geometrically plausible.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.15211", "title": "Generative Multiview Relighting for 3D Reconstruction under Extreme\n  Illumination Variation", "abstract": "  Reconstructing the geometry and appearance of objects from photographs taken\nin different environments is difficult as the illumination and therefore the\nobject appearance vary across captured images. This is particularly challenging\nfor more specular objects whose appearance strongly depends on the viewing\ndirection. Some prior approaches model appearance variation across images using\na per-image embedding vector, while others use physically-based rendering to\nrecover the materials and per-image illumination. Such approaches fail at\nfaithfully recovering view-dependent appearance given significant variation in\ninput illumination and tend to produce mostly diffuse results. We present an\napproach that reconstructs objects from images taken under different\nilluminations by first relighting the images under a single reference\nillumination with a multiview relighting diffusion model and then\nreconstructing the object's geometry and appearance with a radiance field\narchitecture that is robust to the small remaining inconsistencies among the\nrelit images. We validate our proposed approach on both synthetic and real\ndatasets and demonstrate that it greatly outperforms existing techniques at\nreconstructing high-fidelity appearance from images taken under extreme\nillumination variation. Moreover, our approach is particularly effective at\nrecovering view-dependent \"shiny\" appearance which cannot be reconstructed by\nprior methods.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.09982", "title": "SplineGS: Robust Motion-Adaptive Spline for Real-Time Dynamic 3D\n  Gaussians from Monocular Video", "abstract": "  Synthesizing novel views from in-the-wild monocular videos is challenging due\nto scene dynamics and the lack of multi-view cues. To address this, we propose\nSplineGS, a COLMAP-free dynamic 3D Gaussian Splatting (3DGS) framework for\nhigh-quality reconstruction and fast rendering from monocular videos. At its\ncore is a novel Motion-Adaptive Spline (MAS) method, which represents\ncontinuous dynamic 3D Gaussian trajectories using cubic Hermite splines with a\nsmall number of control points. For MAS, we introduce a Motion-Adaptive Control\npoints Pruning (MACP) method to model the deformation of each dynamic 3D\nGaussian across varying motions, progressively pruning control points while\nmaintaining dynamic modeling integrity. Additionally, we present a joint\noptimization strategy for camera parameter estimation and 3D Gaussian\nattributes, leveraging photometric and geometric consistency. This eliminates\nthe need for Structure-from-Motion preprocessing and enhances SplineGS's\nrobustness in real-world conditions. Experiments show that SplineGS\nsignificantly outperforms state-of-the-art methods in novel view synthesis\nquality for dynamic scenes from monocular videos, achieving thousands times\nfaster rendering speed.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.12734", "title": "Gaussian Billboards: Expressive 2D Gaussian Splatting with Textures", "abstract": "  Gaussian Splatting has recently emerged as the go-to representation for\nreconstructing and rendering 3D scenes. The transition from 3D to 2D Gaussian\nprimitives has further improved multi-view consistency and surface\nreconstruction accuracy. In this work we highlight the similarity between 2D\nGaussian Splatting (2DGS) and billboards from traditional computer graphics.\nBoth use flat semi-transparent 2D geometry that is positioned, oriented and\nscaled in 3D space. However 2DGS uses a solid color per splat and an opacity\nmodulated by a Gaussian distribution, where billboards are more expressive,\nmodulating the color with a uv-parameterized texture. We propose to unify these\nconcepts by presenting Gaussian Billboards, a modification of 2DGS to add\nspatially-varying color achieved using per-splat texture interpolation. The\nresult is a mixture of the two representations, which benefits from both the\nrobust scene optimization power of 2DGS and the expressiveness of texture\nmapping. We show that our method can improve the sharpness and quality of the\nscene representation in a wide range of qualitative and quantitative\nevaluations compared to the original 2DGS implementation.\n", "date": "2024-12-17", "categories": []}
{"id": "2311.02778", "title": "MuSHRoom: Multi-Sensor Hybrid Room Dataset for Joint 3D Reconstruction\n  and Novel View Synthesis", "abstract": "  Metaverse technologies demand accurate, real-time, and immersive modeling on\nconsumer-grade hardware for both non-human perception (e.g.,\ndrone/robot/autonomous car navigation) and immersive technologies like AR/VR,\nrequiring both structural accuracy and photorealism. However, there exists a\nknowledge gap in how to apply geometric reconstruction and photorealism\nmodeling (novel view synthesis) in a unified framework. To address this gap and\npromote the development of robust and immersive modeling and rendering with\nconsumer-grade devices, we propose a real-world Multi-Sensor Hybrid Room\nDataset (MuSHRoom). Our dataset presents exciting challenges and requires\nstate-of-the-art methods to be cost-effective, robust to noisy data and\ndevices, and can jointly learn 3D reconstruction and novel view synthesis\ninstead of treating them as separate tasks, making them ideal for real-world\napplications. We benchmark several famous pipelines on our dataset for joint 3D\nmesh reconstruction and novel view synthesis. Our dataset and benchmark show\ngreat potential in promoting the improvements for fusing 3D reconstruction and\nhigh-quality rendering in a robust and computationally efficient end-to-end\nfashion. The dataset and code are available at the project website:\nhttps://xuqianren.github.io/publications/MuSHRoom/.\n", "date": "2023-11-05", "categories": []}
{"id": "2411.19271", "title": "AGS-Mesh: Adaptive Gaussian Splatting and Meshing with Geometric Priors\n  for Indoor Room Reconstruction Using Smartphones", "abstract": "  Geometric priors are often used to enhance 3D reconstruction. With many\nsmartphones featuring low-resolution depth sensors and the prevalence of\noff-the-shelf monocular geometry estimators, incorporating geometric priors as\nregularization signals has become common in 3D vision tasks. However, the\naccuracy of depth estimates from mobile devices is typically poor for highly\ndetailed geometry, and monocular estimators often suffer from poor multi-view\nconsistency and precision. In this work, we propose an approach for joint\nsurface depth and normal refinement of Gaussian Splatting methods for accurate\n3D reconstruction of indoor scenes. We develop supervision strategies that\nadaptively filters low-quality depth and normal estimates by comparing the\nconsistency of the priors during optimization. We mitigate regularization in\nregions where prior estimates have high uncertainty or ambiguities. Our\nfiltering strategy and optimization design demonstrate significant improvements\nin both mesh estimation and novel-view synthesis for both 3D and 2D Gaussian\nSplatting-based methods on challenging indoor room datasets. Furthermore, we\nexplore the use of alternative meshing strategies for finer geometry\nextraction. We develop a scale-aware meshing strategy inspired by TSDF and\noctree-based isosurface extraction, which recovers finer details from Gaussian\nmodels compared to other commonly used open-source meshing tools. Our code is\nreleased in https://xuqianren.github.io/ags_mesh_website/.\n", "date": "2024-11-28", "categories": []}
{"id": "2412.11241", "title": "Volumetric Mapping with Panoptic Refinement via Kernel Density\n  Estimation for Mobile Robots", "abstract": "  Reconstructing three-dimensional (3D) scenes with semantic understanding is\nvital in many robotic applications. Robots need to identify which objects,\nalong with their positions and shapes, to manipulate them precisely with given\ntasks. Mobile robots, especially, usually use lightweight networks to segment\nobjects on RGB images and then localize them via depth maps; however, they\noften encounter out-of-distribution scenarios where masks over-cover the\nobjects. In this paper, we address the problem of panoptic segmentation quality\nin 3D scene reconstruction by refining segmentation errors using non-parametric\nstatistical methods. To enhance mask precision, we map the predicted masks into\na depth frame to estimate their distribution via kernel densities. The outliers\nin depth perception are then rejected without the need for additional\nparameters in an adaptive manner to out-of-distribution scenarios, followed by\n3D reconstruction using projective signed distance functions (SDFs). We\nvalidate our method on a synthetic dataset, which shows improvements in both\nquantitative and qualitative results for panoptic mapping. Through real-world\ntesting, the results furthermore show our method's capability to be deployed on\na real-robot system. Our source code is available at:\nhttps://github.com/mkhangg/refined panoptic mapping.\n", "date": "2024-12-15", "categories": []}
{"id": "2302.03531", "title": "Structured Generative Models for Scene Understanding", "abstract": "  This position paper argues for the use of \\emph{structured generative models}\n(SGMs) for the understanding of static scenes. This requires the reconstruction\nof a 3D scene from an input image (or a set of multi-view images), whereby the\ncontents of the image(s) are causally explained in terms of models of\ninstantiated objects, each with their own type, shape, appearance and pose,\nalong with global variables like scene lighting and camera parameters. This\napproach also requires scene models which account for the co-occurrences and\ninter-relationships of objects in a scene. The SGM approach has the merits that\nit is compositional and generative, which lead to interpretability and\neditability. \\\\\\\\ To pursue the SGM agenda, we need models for objects and\nscenes, and approaches to carry out inference. We first review models for\nobjects, which include ``things'' (object categories that have a well defined\nshape), and ``stuff'' (categories which have amorphous spatial extent). We then\nmove on to review \\emph{scene models} which describe the inter-relationships of\nobjects. Perhaps the most challenging problem for SGMs is \\emph{inference} of\nthe objects, lighting and camera parameters, and scene inter-relationships from\ninput consisting of a single or multiple images. We conclude with a discussion\nof issues that need addressing to advance the SGM agenda.\n", "date": "2023-02-07", "categories": []}
{"id": "2412.00477", "title": "LineGS : 3D Line Segment Representation on 3D Gaussian Splatting", "abstract": "  Abstract representations of 3D scenes play a crucial role in computer vision,\nenabling a wide range of applications such as mapping, localization, surface\nreconstruction, and even advanced tasks like SLAM and rendering. Among these\nrepresentations, line segments are widely used because of their ability to\nsuccinctly capture the structural features of a scene. However, existing 3D\nreconstruction methods often face significant challenges. Methods relying on 2D\nprojections suffer from instability caused by errors in multi-view matching and\nocclusions, while direct 3D approaches are hampered by noise and sparsity in 3D\npoint cloud data. This paper introduces LineGS, a novel method that combines\ngeometry-guided 3D line reconstruction with a 3D Gaussian splatting model to\naddress these challenges and improve representation ability. The method\nleverages the high-density Gaussian point distributions along the edge of the\nscene to refine and optimize initial line segments generated from traditional\ngeometric approaches. By aligning these segments with the underlying geometric\nfeatures of the scene, LineGS achieves a more precise and reliable\nrepresentation of 3D structures. The results show significant improvements in\nboth geometric accuracy and model compactness compared to baseline methods.\n", "date": "2024-11-30", "categories": []}
{"id": "2412.09723", "title": "MAC-Ego3D: Multi-Agent Gaussian Consensus for Real-Time Collaborative\n  Ego-Motion and Photorealistic 3D Reconstruction", "abstract": "  Real-time multi-agent collaboration for ego-motion estimation and\nhigh-fidelity 3D reconstruction is vital for scalable spatial intelligence.\nHowever, traditional methods produce sparse, low-detail maps, while recent\ndense mapping approaches struggle with high latency. To overcome these\nchallenges, we present MAC-Ego3D, a novel framework for real-time collaborative\nphotorealistic 3D reconstruction via Multi-Agent Gaussian Consensus. MAC-Ego3D\nenables agents to independently construct, align, and iteratively refine local\nmaps using a unified Gaussian splat representation. Through Intra-Agent\nGaussian Consensus, it enforces spatial coherence among neighboring Gaussian\nsplats within an agent. For global alignment, parallelized Inter-Agent Gaussian\nConsensus, which asynchronously aligns and optimizes local maps by regularizing\nmulti-agent Gaussian splats, seamlessly integrates them into a high-fidelity 3D\nmodel. Leveraging Gaussian primitives, MAC-Ego3D supports efficient RGB-D\nrendering, enabling rapid inter-agent Gaussian association and alignment.\nMAC-Ego3D bridges local precision and global coherence, delivering higher\nefficiency, largely reducing localization error, and improving mapping\nfidelity. It establishes a new SOTA on synthetic and real-world benchmarks,\nachieving a 15x increase in inference speed, order-of-magnitude reductions in\nego-motion estimation error for partial cases, and RGB PSNR gains of 4 to 10\ndB. Our code will be made publicly available at\nhttps://github.com/Xiaohao-Xu/MAC-Ego3D .\n", "date": "2024-12-12", "categories": []}
{"id": "2412.10051", "title": "TSGaussian: Semantic and Depth-Guided Target-Specific Gaussian Splatting\n  from Sparse Views", "abstract": "  Recent advances in Gaussian Splatting have significantly advanced the field,\nachieving both panoptic and interactive segmentation of 3D scenes. However,\nexisting methodologies often overlook the critical need for reconstructing\nspecified targets with complex structures from sparse views. To address this\nissue, we introduce TSGaussian, a novel framework that combines semantic\nconstraints with depth priors to avoid geometry degradation in challenging\nnovel view synthesis tasks. Our approach prioritizes computational resources on\ndesignated targets while minimizing background allocation. Bounding boxes from\nYOLOv9 serve as prompts for Segment Anything Model to generate 2D mask\npredictions, ensuring semantic accuracy and cost efficiency. TSGaussian\neffectively clusters 3D gaussians by introducing a compact identity encoding\nfor each Gaussian ellipsoid and incorporating 3D spatial consistency\nregularization. Leveraging these modules, we propose a pruning strategy to\neffectively reduce redundancy in 3D gaussians. Extensive experiments\ndemonstrate that TSGaussian outperforms state-of-the-art methods on three\nstandard datasets and a new challenging dataset we collected, achieving\nsuperior results in novel view synthesis of specific objects. Code is available\nat: https://github.com/leon2000-ai/TSGaussian.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.10294", "title": "Coherent 3D Scene Diffusion From a Single RGB Image", "abstract": "  We present a novel diffusion-based approach for coherent 3D scene\nreconstruction from a single RGB image. Our method utilizes an\nimage-conditioned 3D scene diffusion model to simultaneously denoise the 3D\nposes and geometries of all objects within the scene. Motivated by the\nill-posed nature of the task and to obtain consistent scene reconstruction\nresults, we learn a generative scene prior by conditioning on all scene objects\nsimultaneously to capture the scene context and by allowing the model to learn\ninter-object relationships throughout the diffusion process. We further propose\nan efficient surface alignment loss to facilitate training even in the absence\nof full ground-truth annotation, which is common in publicly available\ndatasets. This loss leverages an expressive shape representation, which enables\ndirect point sampling from intermediate shape predictions. By framing the task\nof single RGB image 3D scene reconstruction as a conditional diffusion process,\nour approach surpasses current state-of-the-art methods, achieving a 12.04%\nimprovement in AP3D on SUN RGB-D and a 13.43% increase in F-Score on Pix3D.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.08412", "title": "Pragmatist: Multiview Conditional Diffusion Models for High-Fidelity 3D\n  Reconstruction from Unposed Sparse Views", "abstract": "  Inferring 3D structures from sparse, unposed observations is challenging due\nto its unconstrained nature. Recent methods propose to predict implicit\nrepresentations directly from unposed inputs in a data-driven manner, achieving\npromising results. However, these methods do not utilize geometric priors and\ncannot hallucinate the appearance of unseen regions, thus making it challenging\nto reconstruct fine geometric and textural details. To tackle this challenge,\nour key idea is to reformulate this ill-posed problem as conditional novel view\nsynthesis, aiming to generate complete observations from limited input views to\nfacilitate reconstruction. With complete observations, the poses of the input\nviews can be easily recovered and further used to optimize the reconstructed\nobject. To this end, we propose a novel pipeline Pragmatist. First, we generate\na complete observation of the object via a multiview conditional diffusion\nmodel. Then, we use a feed-forward large reconstruction model to obtain the\nreconstructed mesh. To further improve the reconstruction quality, we recover\nthe poses of input views by inverting the obtained 3D representations and\nfurther optimize the texture using detailed input views. Unlike previous\napproaches, our pipeline improves reconstruction by efficiently leveraging\nunposed inputs and generative priors, circumventing the direct resolution of\nhighly ill-posed problems. Extensive experiments show that our approach\nachieves promising performance in several benchmarks.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.09043", "title": "DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous\n  Driving", "abstract": "  Photorealistic 4D reconstruction of street scenes is essential for developing\nreal-world simulators in autonomous driving. However, most existing methods\nperform this task offline and rely on time-consuming iterative processes,\nlimiting their practical applications. To this end, we introduce the Large 4D\nGaussian Reconstruction Model (DrivingRecon), a generalizable driving scene\nreconstruction model, which directly predicts 4D Gaussian from surround view\nvideos. To better integrate the surround-view images, the Prune and Dilate\nBlock (PD-Block) is proposed to eliminate overlapping Gaussian points between\nadjacent views and remove redundant background points. To enhance\ncross-temporal information, dynamic and static decoupling is tailored to better\nlearn geometry and motion features. Experimental results demonstrate that\nDrivingRecon significantly improves scene reconstruction quality and novel view\nsynthesis compared to existing methods. Furthermore, we explore applications of\nDrivingRecon in model pre-training, vehicle adaptation, and scene editing. Our\ncode is available at https://github.com/EnVision-Research/DriveRecon.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09055", "title": "Hyperbolic-constraint Point Cloud Reconstruction from Single RGB-D\n  Images", "abstract": "  Reconstructing desired objects and scenes has long been a primary goal in 3D\ncomputer vision. Single-view point cloud reconstruction has become a popular\ntechnique due to its low cost and accurate results. However, single-view\nreconstruction methods often rely on expensive CAD models and complex geometric\npriors. Effectively utilizing prior knowledge about the data remains a\nchallenge. In this paper, we introduce hyperbolic space to 3D point cloud\nreconstruction, enabling the model to represent and understand complex\nhierarchical structures in point clouds with low distortion. We build upon\nprevious methods by proposing a hyperbolic Chamfer distance and a regularized\ntriplet loss to enhance the relationship between partial and complete point\nclouds. Additionally, we design adaptive boundary conditions to improve the\nmodel's understanding and reconstruction of 3D structures. Our model\noutperforms most existing models, and ablation studies demonstrate the\nsignificance of our model and its components. Experimental results show that\nour method significantly improves feature extraction capabilities. Our model\nachieves outstanding performance in 3D reconstruction tasks.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09597", "title": "LiftImage3D: Lifting Any Single Image to 3D Gaussians with Video\n  Generation Priors", "abstract": "  Single-image 3D reconstruction remains a fundamental challenge in computer\nvision due to inherent geometric ambiguities and limited viewpoint information.\nRecent advances in Latent Video Diffusion Models (LVDMs) offer promising 3D\npriors learned from large-scale video data. However, leveraging these priors\neffectively faces three key challenges: (1) degradation in quality across large\ncamera motions, (2) difficulties in achieving precise camera control, and (3)\ngeometric distortions inherent to the diffusion process that damage 3D\nconsistency. We address these challenges by proposing LiftImage3D, a framework\nthat effectively releases LVDMs' generative priors while ensuring 3D\nconsistency. Specifically, we design an articulated trajectory strategy to\ngenerate video frames, which decomposes video sequences with large camera\nmotions into ones with controllable small motions. Then we use robust neural\nmatching models, i.e. MASt3R, to calibrate the camera poses of generated frames\nand produce corresponding point clouds. Finally, we propose a distortion-aware\n3D Gaussian splatting representation, which can learn independent distortions\nbetween frames and output undistorted canonical Gaussians. Extensive\nexperiments demonstrate that LiftImage3D achieves state-of-the-art performance\non two challenging datasets, i.e. LLFF, DL3DV, and Tanks and Temples, and\ngeneralizes well to diverse in-the-wild images, from cartoon illustrations to\ncomplex real-world scenes.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09608", "title": "Representing Long Volumetric Video with Temporal Gaussian Hierarchy", "abstract": "  This paper aims to address the challenge of reconstructing long volumetric\nvideos from multi-view RGB videos. Recent dynamic view synthesis methods\nleverage powerful 4D representations, like feature grids or point cloud\nsequences, to achieve high-quality rendering results. However, they are\ntypically limited to short (1~2s) video clips and often suffer from large\nmemory footprints when dealing with longer videos. To solve this issue, we\npropose a novel 4D representation, named Temporal Gaussian Hierarchy, to\ncompactly model long volumetric videos. Our key observation is that there are\ngenerally various degrees of temporal redundancy in dynamic scenes, which\nconsist of areas changing at different speeds. Motivated by this, our approach\nbuilds a multi-level hierarchy of 4D Gaussian primitives, where each level\nseparately describes scene regions with different degrees of content change,\nand adaptively shares Gaussian primitives to represent unchanged scene content\nover different temporal segments, thus effectively reducing the number of\nGaussian primitives. In addition, the tree-like structure of the Gaussian\nhierarchy allows us to efficiently represent the scene at a particular moment\nwith a subset of Gaussian primitives, leading to nearly constant GPU memory\nusage during the training or rendering regardless of the video length.\nExtensive experimental results demonstrate the superiority of our method over\nalternative methods in terms of training cost, rendering speed, and storage\nusage. To our knowledge, this work is the first approach capable of efficiently\nhandling minutes of volumetric video data while maintaining state-of-the-art\nrendering quality. Our project page is available at:\nhttps://zju3dv.github.io/longvolcap.\n", "date": "2024-12-12", "categories": []}
{"id": "2312.13285", "title": "UniSDF: Unifying Neural Representations for High-Fidelity 3D\n  Reconstruction of Complex Scenes with Reflections", "abstract": "  Neural 3D scene representations have shown great potential for 3D\nreconstruction from 2D images. However, reconstructing real-world captures of\ncomplex scenes still remains a challenge. Existing generic 3D reconstruction\nmethods often struggle to represent fine geometric details and do not\nadequately model reflective surfaces of large-scale scenes. Techniques that\nexplicitly focus on reflective surfaces can model complex and detailed\nreflections by exploiting better reflection parameterizations. However, we\nobserve that these methods are often not robust in real scenarios where\nnon-reflective as well as reflective components are present. In this work, we\npropose UniSDF, a general purpose 3D reconstruction method that can reconstruct\nlarge complex scenes with reflections. We investigate both camera view as well\nas reflected view-based color parameterization techniques and find that\nexplicitly blending these representations in 3D space enables reconstruction of\nsurfaces that are more geometrically accurate, especially for reflective\nsurfaces. We further combine this representation with a multi-resolution grid\nbackbone that is trained in a coarse-to-fine manner, enabling faster\nreconstructions than prior methods. Extensive experiments on object-level\ndatasets DTU, Shiny Blender as well as unbounded datasets Mip-NeRF 360 and\nRef-NeRF real demonstrate that our method is able to robustly reconstruct\ncomplex large-scale scenes with fine details and reflective surfaces, leading\nto the best overall performance. Project page:\n\\url{https://fangjinhuawang.github.io/UniSDF}.\n", "date": "2023-12-20", "categories": []}
{"id": "2405.10320", "title": "Toon3D: Seeing Cartoons from New Perspectives", "abstract": "  We recover the underlying 3D structure from images of cartoons and anime\ndepicting the same scene. This is an interesting problem domain because images\nin creative media are often depicted without explicit geometric consistency for\nstorytelling and creative expression-they are only 3D in a qualitative sense.\nWhile humans can easily perceive the underlying 3D scene from these images,\nexisting Structure-from-Motion (SfM) methods that assume 3D consistency fail\ncatastrophically. We present Toon3D for reconstructing geometrically\ninconsistent images. Our key insight is to deform the input images while\nrecovering camera poses and scene geometry, effectively explaining away\ngeometrical inconsistencies to achieve consistency. This process is guided by\nthe structure inferred from monocular depth predictions. We curate a dataset\nwith multi-view imagery from cartoons and anime that we annotate with reliable\nsparse correspondences using our user-friendly annotation tool. Our recovered\npoint clouds can be plugged into novel-view synthesis methods to experience\ncartoons from viewpoints never drawn before. We evaluate against classical and\nrecent learning-based SfM methods, where Toon3D is able to obtain more reliable\ncamera poses and scene geometry.\n", "date": "2024-05-16", "categories": []}
{"id": "2412.06974", "title": "MV-DUSt3R+: Single-Stage Scene Reconstruction from Sparse Views In 2\n  Seconds", "abstract": "  Recent sparse multi-view scene reconstruction advances like DUSt3R and MASt3R\nno longer require camera calibration and camera pose estimation. However, they\nonly process a pair of views at a time to infer pixel-aligned pointmaps. When\ndealing with more than two views, a combinatorial number of error prone\npairwise reconstructions are usually followed by an expensive global\noptimization, which often fails to rectify the pairwise reconstruction errors.\nTo handle more views, reduce errors, and improve inference time, we propose the\nfast single-stage feed-forward network MV-DUSt3R. At its core are multi-view\ndecoder blocks which exchange information across any number of views while\nconsidering one reference view. To make our method robust to reference view\nselection, we further propose MV-DUSt3R+, which employs cross-reference-view\nblocks to fuse information across different reference view choices. To further\nenable novel view synthesis, we extend both by adding and jointly training\nGaussian splatting heads. Experiments on multi-view stereo reconstruction,\nmulti-view pose estimation, and novel view synthesis confirm that our methods\nimprove significantly upon prior art. Code will be released.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.07371", "title": "PRM: Photometric Stereo based Large Reconstruction Model", "abstract": "  We propose PRM, a novel photometric stereo based large reconstruction model\nto reconstruct high-quality meshes with fine-grained local details. Unlike\nprevious large reconstruction models that prepare images under fixed and simple\nlighting as both input and supervision, PRM renders photometric stereo images\nby varying materials and lighting for the purposes, which not only improves the\nprecise local details by providing rich photometric cues but also increases the\nmodel robustness to variations in the appearance of input images. To offer\nenhanced flexibility of images rendering, we incorporate a real-time\nphysically-based rendering (PBR) method and mesh rasterization for online\nimages rendering. Moreover, in employing an explicit mesh as our 3D\nrepresentation, PRM ensures the application of differentiable PBR, which\nsupports the utilization of multiple photometric supervisions and better models\nthe specular color for high-quality geometry optimization. Our PRM leverages\nphotometric stereo images to achieve high-quality reconstructions with\nfine-grained local details, even amidst sophisticated image appearances.\nExtensive experiments demonstrate that PRM significantly outperforms other\nmodels.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.07746", "title": "LoRA3D: Low-Rank Self-Calibration of 3D Geometric Foundation Models", "abstract": "  Emerging 3D geometric foundation models, such as DUSt3R, offer a promising\napproach for in-the-wild 3D vision tasks. However, due to the high-dimensional\nnature of the problem space and scarcity of high-quality 3D data, these\npre-trained models still struggle to generalize to many challenging\ncircumstances, such as limited view overlap or low lighting. To address this,\nwe propose LoRA3D, an efficient self-calibration pipeline to\n$\\textit{specialize}$ the pre-trained models to target scenes using their own\nmulti-view predictions. Taking sparse RGB images as input, we leverage robust\noptimization techniques to refine multi-view predictions and align them into a\nglobal coordinate frame. In particular, we incorporate prediction confidence\ninto the geometric optimization process, automatically re-weighting the\nconfidence to better reflect point estimation accuracy. We use the calibrated\nconfidence to generate high-quality pseudo labels for the calibrating views and\nuse low-rank adaptation (LoRA) to fine-tune the models on the pseudo-labeled\ndata. Our method does not require any external priors or manual labels. It\ncompletes the self-calibration process on a $\\textbf{single standard GPU within\njust 5 minutes}$. Each low-rank adapter requires only $\\textbf{18MB}$ of\nstorage. We evaluated our method on $\\textbf{more than 160 scenes}$ from the\nReplica, TUM and Waymo Open datasets, achieving up to $\\textbf{88% performance\nimprovement}$ on 3D reconstruction, multi-view pose estimation and novel-view\nrendering.\n", "date": "2024-12-10", "categories": []}
{"id": "2403.19495", "title": "CoherentGS: Sparse Novel View Synthesis with Coherent 3D Gaussians", "abstract": "  The field of 3D reconstruction from images has rapidly evolved in the past\nfew years, first with the introduction of Neural Radiance Field (NeRF) and more\nrecently with 3D Gaussian Splatting (3DGS). The latter provides a significant\nedge over NeRF in terms of the training and inference speed, as well as the\nreconstruction quality. Although 3DGS works well for dense input images, the\nunstructured point-cloud like representation quickly overfits to the more\nchallenging setup of extremely sparse input images (e.g., 3 images), creating a\nrepresentation that appears as a jumble of needles from novel views. To address\nthis issue, we propose regularized optimization and depth-based initialization.\nOur key idea is to introduce a structured Gaussian representation that can be\ncontrolled in 2D image space. We then constraint the Gaussians, in particular\ntheir position, and prevent them from moving independently during optimization.\nSpecifically, we introduce single and multiview constraints through an implicit\nconvolutional decoder and a total variation loss, respectively. With the\ncoherency introduced to the Gaussians, we further constrain the optimization\nthrough a flow-based loss function. To support our regularized optimization, we\npropose an approach to initialize the Gaussians using monocular depth estimates\nat each input view. We demonstrate significant improvements compared to the\nstate-of-the-art sparse-view NeRF-based approaches on a variety of scenes.\n", "date": "2024-03-28", "categories": []}
{"id": "2406.11672", "title": "Effective Rank Analysis and Regularization for Enhanced 3D Gaussian\n  Splatting", "abstract": "  3D reconstruction from multi-view images is one of the fundamental challenges\nin computer vision and graphics. Recently, 3D Gaussian Splatting (3DGS) has\nemerged as a promising technique capable of real-time rendering with\nhigh-quality 3D reconstruction. This method utilizes 3D Gaussian representation\nand tile-based splatting techniques, bypassing the expensive neural field\nquerying. Despite its potential, 3DGS encounters challenges such as needle-like\nartifacts, suboptimal geometries, and inaccurate normals caused by the\nGaussians converging into anisotropic shapes with one dominant variance. We\npropose using the effective rank analysis to examine the shape statistics of 3D\nGaussian primitives, and identify the Gaussians indeed converge into\nneedle-like shapes with the effective rank 1. To address this, we introduce the\neffective rank as a regularization, which constrains the structure of the\nGaussians. Our new regularization method enhances normal and geometry\nreconstruction while reducing needle-like artifacts. The approach can be\nintegrated as an add-on module to other 3DGS variants, improving their quality\nwithout compromising visual fidelity. The project page is available at\nhttps://junhahyung.github.io/erankgs.github.io.\n", "date": "2024-06-17", "categories": []}
{"id": "2408.04474", "title": "LumiGauss: Relightable Gaussian Splatting in the Wild", "abstract": "  Decoupling lighting from geometry using unconstrained photo collections is\nnotoriously challenging. Solving it would benefit many users as creating\ncomplex 3D assets takes days of manual labor. Many previous works have\nattempted to address this issue, often at the expense of output fidelity, which\nquestions the practicality of such methods. We introduce LumiGauss - a\ntechnique that tackles 3D reconstruction of scenes and environmental lighting\nthrough 2D Gaussian Splatting. Our approach yields high-quality scene\nreconstructions and enables realistic lighting synthesis under novel\nenvironment maps. We also propose a method for enhancing the quality of\nshadows, common in outdoor scenes, by exploiting spherical harmonics\nproperties. Our approach facilitates seamless integration with game engines and\nenables the use of fast precomputed radiance transfer. We validate our method\non the NeRF-OSR dataset, demonstrating superior performance over baseline\nmethods. Moreover, LumiGauss can synthesize realistic images for unseen\nenvironment maps. Our code: https://github.com/joaxkal/lumigauss.\n", "date": "2024-08-06", "categories": []}
{"id": "2412.05546", "title": "Radiant: Large-scale 3D Gaussian Rendering based on Hierarchical\n  Framework", "abstract": "  With the advancement of computer vision, the recently emerged 3D Gaussian\nSplatting (3DGS) has increasingly become a popular scene reconstruction\nalgorithm due to its outstanding performance. Distributed 3DGS can efficiently\nutilize edge devices to directly train on the collected images, thereby\noffloading computational demands and enhancing efficiency. However, traditional\ndistributed frameworks often overlook computational and communication\nchallenges in real-world environments, hindering large-scale deployment and\npotentially posing privacy risks. In this paper, we propose Radiant, a\nhierarchical 3DGS algorithm designed for large-scale scene reconstruction that\nconsiders system heterogeneity, enhancing the model performance and training\nefficiency. Via extensive empirical study, we find that it is crucial to\npartition the regions for each edge appropriately and allocate varying camera\npositions to each device for image collection and training. The core of Radiant\nis partitioning regions based on heterogeneous environment information and\nallocating workloads to each device accordingly. Furthermore, we provide a 3DGS\nmodel aggregation algorithm that enhances the quality and ensures the\ncontinuity of models' boundaries. Finally, we develop a testbed, and\nexperiments demonstrate that Radiant improved reconstruction quality by up to\n25.7\\% and reduced up to 79.6\\% end-to-end latency.\n", "date": "2024-12-07", "categories": []}
{"id": "2412.06299", "title": "4D Gaussian Splatting with Scale-aware Residual Field and Adaptive\n  Optimization for Real-time Rendering of Temporally Complex Dynamic Scenes", "abstract": "  Reconstructing dynamic scenes from video sequences is a highly promising task\nin the multimedia domain. While previous methods have made progress, they often\nstruggle with slow rendering and managing temporal complexities such as\nsignificant motion and object appearance/disappearance. In this paper, we\npropose SaRO-GS as a novel dynamic scene representation capable of achieving\nreal-time rendering while effectively handling temporal complexities in dynamic\nscenes. To address the issue of slow rendering speed, we adopt a Gaussian\nprimitive-based representation and optimize the Gaussians in 4D space, which\nfacilitates real-time rendering with the assistance of 3D Gaussian Splatting.\nAdditionally, to handle temporally complex dynamic scenes, we introduce a\nScale-aware Residual Field. This field considers the size information of each\nGaussian primitive while encoding its residual feature and aligns with the\nself-splitting behavior of Gaussian primitives. Furthermore, we propose an\nAdaptive Optimization Schedule, which assigns different optimization strategies\nto Gaussian primitives based on their distinct temporal properties, thereby\nexpediting the reconstruction of dynamic regions. Through evaluations on\nmonocular and multi-view datasets, our method has demonstrated state-of-the-art\nperformance. Please see our project page at\nhttps://yjb6.github.io/SaRO-GS.github.io.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06767", "title": "MAtCha Gaussians: Atlas of Charts for High-Quality Geometry and\n  Photorealism From Sparse Views", "abstract": "  We present a novel appearance model that simultaneously realizes explicit\nhigh-quality 3D surface mesh recovery and photorealistic novel view synthesis\nfrom sparse view samples. Our key idea is to model the underlying scene\ngeometry Mesh as an Atlas of Charts which we render with 2D Gaussian surfels\n(MAtCha Gaussians). MAtCha distills high-frequency scene surface details from\nan off-the-shelf monocular depth estimator and refines it through Gaussian\nsurfel rendering. The Gaussian surfels are attached to the charts on the fly,\nsatisfying photorealism of neural volumetric rendering and crisp geometry of a\nmesh model, i.e., two seemingly contradicting goals in a single model. At the\ncore of MAtCha lies a novel neural deformation model and a structure loss that\npreserve the fine surface details distilled from learned monocular depths while\naddressing their fundamental scale ambiguities. Results of extensive\nexperimental validation demonstrate MAtCha's state-of-the-art quality of\nsurface reconstruction and photorealism on-par with top contenders but with\ndramatic reduction in the number of input views and computational time. We\nbelieve MAtCha will serve as a foundational tool for any visual application in\nvision, graphics, and robotics that require explicit geometry in addition to\nphotorealism. Our project page is the following:\nhttps://anttwo.github.io/matcha/\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06777", "title": "Driv3R: Learning Dense 4D Reconstruction for Autonomous Driving", "abstract": "  Realtime 4D reconstruction for dynamic scenes remains a crucial challenge for\nautonomous driving perception. Most existing methods rely on depth estimation\nthrough self-supervision or multi-modality sensor fusion. In this paper, we\npropose Driv3R, a DUSt3R-based framework that directly regresses per-frame\npoint maps from multi-view image sequences. To achieve streaming dense\nreconstruction, we maintain a memory pool to reason both spatial relationships\nacross sensors and dynamic temporal contexts to enhance multi-view 3D\nconsistency and temporal integration. Furthermore, we employ a 4D flow\npredictor to identify moving objects within the scene to direct our network\nfocus more on reconstructing these dynamic regions. Finally, we align all\nper-frame pointmaps consistently to the world coordinate system in an\noptimization-free manner. We conduct extensive experiments on the large-scale\nnuScenes dataset to evaluate the effectiveness of our method. Driv3R\noutperforms previous frameworks in 4D dynamic scene reconstruction, achieving\n15x faster inference speed compared to methods requiring global alignment.\nCode: https://github.com/Barrybarry-Smith/Driv3R.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.04469", "title": "QUEEN: QUantized Efficient ENcoding of Dynamic Gaussians for Streaming\n  Free-viewpoint Videos", "abstract": "  Online free-viewpoint video (FVV) streaming is a challenging problem, which\nis relatively under-explored. It requires incremental on-the-fly updates to a\nvolumetric representation, fast training and rendering to satisfy real-time\nconstraints and a small memory footprint for efficient transmission. If\nachieved, it can enhance user experience by enabling novel applications, e.g.,\n3D video conferencing and live volumetric video broadcast, among others. In\nthis work, we propose a novel framework for QUantized and Efficient ENcoding\n(QUEEN) for streaming FVV using 3D Gaussian Splatting (3D-GS). QUEEN directly\nlearns Gaussian attribute residuals between consecutive frames at each\ntime-step without imposing any structural constraints on them, allowing for\nhigh quality reconstruction and generalizability. To efficiently store the\nresiduals, we further propose a quantization-sparsity framework, which contains\na learned latent-decoder for effectively quantizing attribute residuals other\nthan Gaussian positions and a learned gating module to sparsify position\nresiduals. We propose to use the Gaussian viewspace gradient difference vector\nas a signal to separate the static and dynamic content of the scene. It acts as\na guide for effective sparsity learning and speeds up training. On diverse FVV\nbenchmarks, QUEEN outperforms the state-of-the-art online FVV methods on all\nmetrics. Notably, for several highly dynamic scenes, it reduces the model size\nto just 0.7 MB per frame while training in under 5 sec and rendering at 350\nFPS. Project website is at https://research.nvidia.com/labs/amri/projects/queen\n", "date": "2024-12-05", "categories": []}
{"id": "2311.05289", "title": "VoxNeRF: Bridging Voxel Representation and Neural Radiance Fields for\n  Enhanced Indoor View Synthesis", "abstract": "  The generation of high-fidelity view synthesis is essential for robotic\nnavigation and interaction but remains challenging, particularly in indoor\nenvironments and real-time scenarios. Existing techniques often require\nsignificant computational resources for both training and rendering, and they\nfrequently result in suboptimal 3D representations due to insufficient\ngeometric structuring. To address these limitations, we introduce VoxNeRF, a\nnovel approach that utilizes easy-to-obtain geometry priors to enhance both the\nquality and efficiency of neural indoor reconstruction and novel view\nsynthesis. We propose an efficient voxel-guided sampling technique that\nallocates computational resources selectively to the most relevant segments of\nrays based on a voxel-encoded geometry prior, significantly reducing training\nand rendering time. Additionally, we incorporate a robust depth loss to improve\nreconstruction and rendering quality in sparse view settings. Our approach is\nvalidated with extensive experiments on ScanNet and ScanNet++ where VoxNeRF\noutperforms existing state-of-the-art methods and establishes a new benchmark\nfor indoor immersive interpolation and extrapolation settings.\n", "date": "2023-11-09", "categories": []}
{"id": "2409.17459", "title": "TFS-NeRF: Template-Free NeRF for Semantic 3D Reconstruction of Dynamic\n  Scene", "abstract": "  Despite advancements in Neural Implicit models for 3D surface reconstruction,\nhandling dynamic environments with interactions between arbitrary rigid,\nnon-rigid, or deformable entities remains challenging. The generic\nreconstruction methods adaptable to such dynamic scenes often require\nadditional inputs like depth or optical flow or rely on pre-trained image\nfeatures for reasonable outcomes. These methods typically use latent codes to\ncapture frame-by-frame deformations. Another set of dynamic scene\nreconstruction methods, are entity-specific, mostly focusing on humans, and\nrelies on template models. In contrast, some template-free methods bypass these\nrequirements and adopt traditional LBS (Linear Blend Skinning) weights for a\ndetailed representation of deformable object motions, although they involve\ncomplex optimizations leading to lengthy training times. To this end, as a\nremedy, this paper introduces TFS-NeRF, a template-free 3D semantic NeRF for\ndynamic scenes captured from sparse or single-view RGB videos, featuring\ninteractions among two entities and more time-efficient than other LBS-based\napproaches. Our framework uses an Invertible Neural Network (INN) for LBS\nprediction, simplifying the training process. By disentangling the motions of\ninteracting entities and optimizing per-entity skinning weights, our method\nefficiently generates accurate, semantically separable geometries. Extensive\nexperiments demonstrate that our approach produces high-quality reconstructions\nof both deformable and non-deformable objects in complex interactions, with\nimproved training efficiency compared to existing methods.\n", "date": "2024-09-26", "categories": []}
{"id": "2411.01853", "title": "GVKF: Gaussian Voxel Kernel Functions for Highly Efficient Surface\n  Reconstruction in Open Scenes", "abstract": "  In this paper we present a novel method for efficient and effective 3D\nsurface reconstruction in open scenes. Existing Neural Radiance Fields (NeRF)\nbased works typically require extensive training and rendering time due to the\nadopted implicit representations. In contrast, 3D Gaussian splatting (3DGS)\nuses an explicit and discrete representation, hence the reconstructed surface\nis built by the huge number of Gaussian primitives, which leads to excessive\nmemory consumption and rough surface details in sparse Gaussian areas. To\naddress these issues, we propose Gaussian Voxel Kernel Functions (GVKF), which\nestablish a continuous scene representation based on discrete 3DGS through\nkernel regression. The GVKF integrates fast 3DGS rasterization and highly\neffective scene implicit representations, achieving high-fidelity open scene\nsurface reconstruction. Experiments on challenging scene datasets demonstrate\nthe efficiency and effectiveness of our proposed GVKF, featuring with high\nreconstruction quality, real-time rendering speed, significant savings in\nstorage and training memory consumption.\n", "date": "2024-11-04", "categories": []}
{"id": "2412.03428", "title": "2DGS-Room: Seed-Guided 2D Gaussian Splatting with Geometric Constrains\n  for High-Fidelity Indoor Scene Reconstruction", "abstract": "  The reconstruction of indoor scenes remains challenging due to the inherent\ncomplexity of spatial structures and the prevalence of textureless regions.\nRecent advancements in 3D Gaussian Splatting have improved novel view synthesis\nwith accelerated processing but have yet to deliver comparable performance in\nsurface reconstruction. In this paper, we introduce 2DGS-Room, a novel method\nleveraging 2D Gaussian Splatting for high-fidelity indoor scene reconstruction.\nSpecifically, we employ a seed-guided mechanism to control the distribution of\n2D Gaussians, with the density of seed points dynamically optimized through\nadaptive growth and pruning mechanisms. To further improve geometric accuracy,\nwe incorporate monocular depth and normal priors to provide constraints for\ndetails and textureless regions respectively. Additionally, multi-view\nconsistency constraints are employed to mitigate artifacts and further enhance\nreconstruction quality. Extensive experiments on ScanNet and ScanNet++ datasets\ndemonstrate that our method achieves state-of-the-art performance in indoor\nscene reconstruction.\n", "date": "2024-12-04", "categories": []}
{"id": "2412.03451", "title": "PlanarSplatting: Accurate Planar Surface Reconstruction in 3 Minutes", "abstract": "  This paper presents PlanarSplatting, an ultra-fast and accurate surface\nreconstruction approach for multiview indoor images. We take the 3D planes as\nthe main objective due to their compactness and structural expressiveness in\nindoor scenes, and develop an explicit optimization framework that learns to\nfit the expected surface of indoor scenes by splatting the 3D planes into 2.5D\ndepth and normal maps. As our PlanarSplatting operates directly on the 3D plane\nprimitives, it eliminates the dependencies on 2D/3D plane detection and plane\nmatching and tracking for planar surface reconstruction. Furthermore, the\nessential merits of plane-based representation plus CUDA-based implementation\nof planar splatting functions, PlanarSplatting reconstructs an indoor scene in\n3 minutes while having significantly better geometric accuracy. Thanks to our\nultra-fast reconstruction speed, the largest quantitative evaluation on the\nScanNet and ScanNet++ datasets over hundreds of scenes clearly demonstrated the\nadvantages of our method. We believe that our accurate and ultrafast planar\nsurface reconstruction method will be applied in the structured data curation\nfor surface reconstruction in the future. The code of our CUDA implementation\nwill be publicly available. Project page:\nhttps://icetttb.github.io/PlanarSplatting/\n", "date": "2024-12-04", "categories": []}
{"id": "2412.03570", "title": "Sparse-view Pose Estimation and Reconstruction via Analysis by\n  Generative Synthesis", "abstract": "  Inferring the 3D structure underlying a set of multi-view images typically\nrequires solving two co-dependent tasks -- accurate 3D reconstruction requires\nprecise camera poses, and predicting camera poses relies on (implicitly or\nexplicitly) modeling the underlying 3D. The classical framework of analysis by\nsynthesis casts this inference as a joint optimization seeking to explain the\nobserved pixels, and recent instantiations learn expressive 3D representations\n(e.g., Neural Fields) with gradient-descent-based pose refinement of initial\npose estimates. However, given a sparse set of observed views, the observations\nmay not provide sufficient direct evidence to obtain complete and accurate 3D.\nMoreover, large errors in pose estimation may not be easily corrected and can\nfurther degrade the inferred 3D. To allow robust 3D reconstruction and pose\nestimation in this challenging setup, we propose SparseAGS, a method that\nadapts this analysis-by-synthesis approach by: a) including\nnovel-view-synthesis-based generative priors in conjunction with photometric\nobjectives to improve the quality of the inferred 3D, and b) explicitly\nreasoning about outliers and using a discrete search with a continuous\noptimization-based strategy to correct them. We validate our framework across\nreal-world and synthetic datasets in combination with several off-the-shelf\npose estimation systems as initialization. We find that it significantly\nimproves the base systems' pose accuracy while yielding high-quality 3D\nreconstructions that outperform the results from current multi-view\nreconstruction baselines.\n", "date": "2024-12-04", "categories": []}
{"id": "2410.07266", "title": "Spiking GS: Towards High-Accuracy and Low-Cost Surface Reconstruction\n  via Spiking Neuron-based Gaussian Splatting", "abstract": "  3D Gaussian Splatting is capable of reconstructing 3D scenes in minutes.\nDespite recent advances in improving surface reconstruction accuracy, the\nreconstructed results still exhibit bias and suffer from inefficiency in\nstorage and training. This paper provides a different observation on the cause\nof the inefficiency and the reconstruction bias, which is attributed to the\nintegration of the low-opacity parts (LOPs) of the generated Gaussians. We show\nthat LOPs consist of Gaussians with overall low-opacity (LOGs) and the\nlow-opacity tails (LOTs) of Gaussians. We propose Spiking GS to reduce such two\ntypes of LOPs by integrating spiking neurons into the Gaussian Splatting\npipeline. Specifically, we introduce global and local full-precision\nintegrate-and-fire spiking neurons to the opacity and representation function\nof flattened 3D Gaussians, respectively. Furthermore, we enhance the density\ncontrol strategy with spiking neurons' thresholds and a new criterion on the\nscale of Gaussians. Our method can represent more accurate reconstructed\nsurfaces at a lower cost. The supplementary material and code are available at\nhttps://github.com/zju-bmi-lab/SpikingGS.\n", "date": "2024-10-09", "categories": []}
{"id": "2412.01931", "title": "Planar Gaussian Splatting", "abstract": "  This paper presents Planar Gaussian Splatting (PGS), a novel neural rendering\napproach to learn the 3D geometry and parse the 3D planes of a scene, directly\nfrom multiple RGB images. The PGS leverages Gaussian primitives to model the\nscene and employ a hierarchical Gaussian mixture approach to group them.\nSimilar Gaussians are progressively merged probabilistically in the\ntree-structured Gaussian mixtures to identify distinct 3D plane instances and\nform the overall 3D scene geometry. In order to enable the grouping, the\nGaussian primitives contain additional parameters, such as plane descriptors\nderived by lifting 2D masks from a general 2D segmentation model and surface\nnormals. Experiments show that the proposed PGS achieves state-of-the-art\nperformance in 3D planar reconstruction without requiring either 3D plane\nlabels or depth supervision. In contrast to existing supervised methods that\nhave limited generalizability and struggle under domain shift, PGS maintains\nits performance across datasets thanks to its neural rendering and\nscene-specific optimization mechanism, while also being significantly faster\nthan existing optimization-based approaches.\n", "date": "2024-12-02", "categories": []}
{"id": "2412.02075", "title": "Gaussian Object Carver: Object-Compositional Gaussian Splatting with\n  surfaces completion", "abstract": "  3D scene reconstruction is a foundational problem in computer vision. Despite\nrecent advancements in Neural Implicit Representations (NIR), existing methods\noften lack editability and compositional flexibility, limiting their use in\nscenarios requiring high interactivity and object-level manipulation. In this\npaper, we introduce the Gaussian Object Carver (GOC), a novel, efficient, and\nscalable framework for object-compositional 3D scene reconstruction. GOC\nleverages 3D Gaussian Splatting (GS), enriched with monocular geometry priors\nand multi-view geometry regularization, to achieve high-quality and flexible\nreconstruction. Furthermore, we propose a zero-shot Object Surface Completion\n(OSC) model, which uses 3D priors from 3d object data to reconstruct unobserved\nsurfaces, ensuring object completeness even in occluded areas. Experimental\nresults demonstrate that GOC improves reconstruction efficiency and geometric\nfidelity. It holds promise for advancing the practical application of digital\ntwins in embodied AI, AR/VR, and interactive simulation environments.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.02225", "title": "How to Use Diffusion Priors under Sparse Views?", "abstract": "  Novel view synthesis under sparse views has been a long-term important\nchallenge in 3D reconstruction. Existing works mainly rely on introducing\nexternal semantic or depth priors to supervise the optimization of 3D\nrepresentations. However, the diffusion model, as an external prior that can\ndirectly provide visual supervision, has always underperformed in sparse-view\n3D reconstruction using Score Distillation Sampling (SDS) due to the low\ninformation entropy of sparse views compared to text, leading to optimization\nchallenges caused by mode deviation. To this end, we present a thorough\nanalysis of SDS from the mode-seeking perspective and propose Inline Prior\nGuided Score Matching (IPSM), which leverages visual inline priors provided by\npose relationships between viewpoints to rectify the rendered image\ndistribution and decomposes the original optimization objective of SDS, thereby\noffering effective diffusion visual guidance without any fine-tuning or\npre-training. Furthermore, we propose the IPSM-Gaussian pipeline, which adopts\n3D Gaussian Splatting as the backbone and supplements depth and geometry\nconsistency regularization based on IPSM to further improve inline priors and\nrectified distribution. Experimental results on different public datasets show\nthat our method achieves state-of-the-art reconstruction quality. The code is\nreleased at https://github.com/iCVTEAM/IPSM.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.02249", "title": "Multi-robot autonomous 3D reconstruction using Gaussian splatting with\n  Semantic guidance", "abstract": "  Implicit neural representations and 3D Gaussian splatting (3DGS) have shown\ngreat potential for scene reconstruction. Recent studies have expanded their\napplications in autonomous reconstruction through task assignment methods.\nHowever, these methods are mainly limited to single robot, and rapid\nreconstruction of large-scale scenes remains challenging. Additionally,\ntask-driven planning based on surface uncertainty is prone to being trapped in\nlocal optima. To this end, we propose the first 3DGS-based centralized\nmulti-robot autonomous 3D reconstruction framework. To further reduce time cost\nof task generation and improve reconstruction quality, we integrate online\nopen-vocabulary semantic segmentation with surface uncertainty of 3DGS,\nfocusing view sampling on regions with high instance uncertainty. Finally, we\ndevelop a multi-robot collaboration strategy with mode and task assignments\nimproving reconstruction quality while ensuring planning efficiency. Our method\ndemonstrates the highest reconstruction quality among all planning methods and\nsuperior planning efficiency compared to existing multi-robot methods. We\ndeploy our method on multiple robots, and results show that it can effectively\nplan view paths and reconstruct scenes with high quality.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.02493", "title": "RelayGS: Reconstructing Dynamic Scenes with Large-Scale and Complex\n  Motions via Relay Gaussians", "abstract": "  Reconstructing dynamic scenes with large-scale and complex motions remains a\nsignificant challenge. Recent techniques like Neural Radiance Fields and 3D\nGaussian Splatting (3DGS) have shown promise but still struggle with scenes\ninvolving substantial movement. This paper proposes RelayGS, a novel method\nbased on 3DGS, specifically designed to represent and reconstruct highly\ndynamic scenes. Our RelayGS learns a complete 4D representation with canonical\n3D Gaussians and a compact motion field, consisting of three stages. First, we\nlearn a fundamental 3DGS from all frames, ignoring temporal scene variations,\nand use a learnable mask to separate the highly dynamic foreground from the\nminimally moving background. Second, we replicate multiple copies of the\ndecoupled foreground Gaussians from the first stage, each corresponding to a\ntemporal segment, and optimize them using pseudo-views constructed from\nmultiple frames within each segment. These Gaussians, termed Relay Gaussians,\nact as explicit relay nodes, simplifying and breaking down large-scale motion\ntrajectories into smaller, manageable segments. Finally, we jointly learn the\nscene's temporal motion and refine the canonical Gaussians learned from the\nfirst two stages. We conduct thorough experiments on two dynamic scene datasets\nfeaturing large and complex motions, where our RelayGS outperforms\nstate-of-the-arts by more than 1 dB in PSNR, and successfully reconstructs\nreal-world basketball game scenes in a much more complete and coherent manner,\nwhereas previous methods usually struggle to capture the complex motion of\nplayers. Code will be publicly available at https://github.com/gqk/RelayGS\n", "date": "2024-12-03", "categories": []}
{"id": "2301.13656", "title": "A Survey and Benchmark of Automatic Surface Reconstruction from Point\n  Clouds", "abstract": "  We present a comprehensive survey and benchmark of both traditional and\nlearning-based methods for surface reconstruction from point clouds. This task\nis particularly challenging for real-world acquisitions due to factors such as\nnoise, outliers, non-uniform sampling, and missing data. Traditional approaches\noften simplify the problem by imposing handcrafted priors on either the input\npoint clouds or the resulting surface, a process that can require tedious\nhyperparameter tuning. In contrast, deep learning models have the capability to\ndirectly learn the properties of input point clouds and desired surfaces from\ndata. We study the influence of handcrafted and learned priors on the precision\nand robustness of surface reconstruction techniques. We evaluate various\ntime-tested and contemporary methods in a standardized manner. When both\ntrained and evaluated on point clouds with identical characteristics, the\nlearning-based models consistently produce higher-quality surfaces compared to\ntheir traditional counterparts -- even in scenarios involving novel shape\ncategories. However, traditional methods demonstrate greater resilience to the\ndiverse anomalies commonly found in real-world 3D acquisitions. For the benefit\nof the research community, we make our code and datasets available, inviting\nfurther enhancements to learning-based surface reconstruction. This can be\naccessed at https://github.com/raphaelsulzer/dsr-benchmark .\n", "date": "2023-01-31", "categories": []}
{"id": "2312.14132", "title": "DUSt3R: Geometric 3D Vision Made Easy", "abstract": "  Multi-view stereo reconstruction (MVS) in the wild requires to first estimate\nthe camera parameters e.g. intrinsic and extrinsic parameters. These are\nusually tedious and cumbersome to obtain, yet they are mandatory to triangulate\ncorresponding pixels in 3D space, which is the core of all best performing MVS\nalgorithms. In this work, we take an opposite stance and introduce DUSt3R, a\nradically novel paradigm for Dense and Unconstrained Stereo 3D Reconstruction\nof arbitrary image collections, i.e. operating without prior information about\ncamera calibration nor viewpoint poses. We cast the pairwise reconstruction\nproblem as a regression of pointmaps, relaxing the hard constraints of usual\nprojective camera models. We show that this formulation smoothly unifies the\nmonocular and binocular reconstruction cases. In the case where more than two\nimages are provided, we further propose a simple yet effective global alignment\nstrategy that expresses all pairwise pointmaps in a common reference frame. We\nbase our network architecture on standard Transformer encoders and decoders,\nallowing us to leverage powerful pretrained models. Our formulation directly\nprovides a 3D model of the scene as well as depth information, but\ninterestingly, we can seamlessly recover from it, pixel matches, relative and\nabsolute camera. Exhaustive experiments on all these tasks showcase that the\nproposed DUSt3R can unify various 3D vision tasks and set new SoTAs on\nmonocular/multi-view depth estimation as well as relative pose estimation. In\nsummary, DUSt3R makes many geometric 3D vision tasks easy.\n", "date": "2023-12-21", "categories": []}
{"id": "2406.07648", "title": "Multi-View Large Reconstruction Model via Geometry-Aware Positional\n  Encoding and Attention", "abstract": "  Despite recent advancements in the Large Reconstruction Model (LRM)\ndemonstrating impressive results, when extending its input from single image to\nmultiple images, it exhibits inefficiencies, subpar geometric and texture\nquality, as well as slower convergence speed than expected. It is attributed to\nthat, LRM formulates 3D reconstruction as a naive images-to-3D translation\nproblem, ignoring the strong 3D coherence among the input images. In this\npaper, we propose a Multi-view Large Reconstruction Model (M-LRM) designed to\nreconstruct high-quality 3D shapes from multi-views in a 3D-aware manner.\nSpecifically, we introduce a multi-view consistent cross-attention scheme to\nenable M-LRM to accurately query information from the input images. Moreover,\nwe employ the 3D priors of the input multi-view images to initialize the\ntriplane tokens. Compared to previous methods, the proposed M-LRM can generate\n3D shapes of high fidelity. Experimental studies demonstrate that our model\nachieves a significant performance gain and faster training convergence.\nProject page: \\url{https://murphylmf.github.io/M-LRM/}.\n", "date": "2024-06-11", "categories": []}
{"id": "2411.12440", "title": "Beyond Gaussians: Fast and High-Fidelity 3D Splatting with Linear\n  Kernels", "abstract": "  Recent advancements in 3D Gaussian Splatting (3DGS) have substantially\nimproved novel view synthesis, enabling high-quality reconstruction and\nreal-time rendering. However, blurring artifacts, such as floating primitives\nand over-reconstruction, remain challenging. Current methods address these\nissues by refining scene structure, enhancing geometric representations,\naddressing blur in training images, improving rendering consistency, and\noptimizing density control, yet the role of kernel design remains\nunderexplored. We identify the soft boundaries of Gaussian ellipsoids as one of\nthe causes of these artifacts, limiting detail capture in high-frequency\nregions. To bridge this gap, we introduce 3D Linear Splatting (3DLS), which\nreplaces Gaussian kernels with linear kernels to achieve sharper and more\nprecise results, particularly in high-frequency regions. Through evaluations on\nthree datasets, 3DLS demonstrates state-of-the-art fidelity and accuracy, along\nwith a 30% FPS improvement over baseline 3DGS. The implementation will be made\npublicly available upon acceptance.\n", "date": "2024-11-19", "categories": []}
{"id": "2411.19454", "title": "GausSurf: Geometry-Guided 3D Gaussian Splatting for Surface\n  Reconstruction", "abstract": "  3D Gaussian Splatting has achieved impressive performance in novel view\nsynthesis with real-time rendering capabilities. However, reconstructing\nhigh-quality surfaces with fine details using 3D Gaussians remains a\nchallenging task. In this work, we introduce GausSurf, a novel approach to\nhigh-quality surface reconstruction by employing geometry guidance from\nmulti-view consistency in texture-rich areas and normal priors in texture-less\nareas of a scene. We observe that a scene can be mainly divided into two\nprimary regions: 1) texture-rich and 2) texture-less areas. To enforce\nmulti-view consistency at texture-rich areas, we enhance the reconstruction\nquality by incorporating a traditional patch-match based Multi-View Stereo\n(MVS) approach to guide the geometry optimization in an iterative scheme. This\nscheme allows for mutual reinforcement between the optimization of Gaussians\nand patch-match refinement, which significantly improves the reconstruction\nresults and accelerates the training process. Meanwhile, for the texture-less\nareas, we leverage normal priors from a pre-trained normal estimation model to\nguide optimization. Extensive experiments on the DTU and Tanks and Temples\ndatasets demonstrate that our method surpasses state-of-the-art methods in\nterms of reconstruction quality and computation time.\n", "date": "2024-11-29", "categories": []}
{"id": "2412.00242", "title": "Uni-SLAM: Uncertainty-Aware Neural Implicit SLAM for Real-Time Dense\n  Indoor Scene Reconstruction", "abstract": "  Neural implicit fields have recently emerged as a powerful representation\nmethod for multi-view surface reconstruction due to their simplicity and\nstate-of-the-art performance. However, reconstructing thin structures of indoor\nscenes while ensuring real-time performance remains a challenge for dense\nvisual SLAM systems. Previous methods do not consider varying quality of input\nRGB-D data and employ fixed-frequency mapping process to reconstruct the scene,\nwhich could result in the loss of valuable information in some frames. In this\npaper, we propose Uni-SLAM, a decoupled 3D spatial representation based on hash\ngrids for indoor reconstruction. We introduce a novel defined predictive\nuncertainty to reweight the loss function, along with strategic local-to-global\nbundle adjustment. Experiments on synthetic and real-world datasets demonstrate\nthat our system achieves state-of-the-art tracking and mapping accuracy while\nmaintaining real-time performance. It significantly improves over current\nmethods with a 25% reduction in depth L1 error and a 66.86% completion rate\nwithin 1 cm on the Replica dataset, reflecting a more accurate reconstruction\nof thin structures. Project page:\nhttps://shaoxiang777.github.io/project/uni-slam/\n", "date": "2024-11-29", "categories": []}
{"id": "2412.01431", "title": "Semantic Scene Completion with Multi-Feature Data Balancing Network", "abstract": "  Semantic Scene Completion (SSC) is a critical task in computer vision, that\nutilized in applications such as virtual reality (VR). SSC aims to construct\ndetailed 3D models from partial views by transforming a single 2D image into a\n3D representation, assigning each voxel a semantic label. The main challenge\nlies in completing 3D volumes with limited information, compounded by data\nimbalance, inter-class ambiguity, and intra-class diversity in indoor scenes.\nTo address this, we propose the Multi-Feature Data Balancing Network (MDBNet),\na dual-head model for RGB and depth data (F-TSDF) inputs. Our hybrid\nencoder-decoder architecture with identity transformation in a pre-activation\nresidual module (ITRM) effectively manages diverse signals within F-TSDF. We\nevaluate RGB feature fusion strategies and use a combined loss function cross\nentropy for 2D RGB features and weighted cross-entropy for 3D SSC predictions.\nMDBNet results surpass comparable state-of-the-art (SOTA) methods on NYU\ndatasets, demonstrating the effectiveness of our approach.\n", "date": "2024-12-02", "categories": []}
{"id": "2311.09806", "title": "EvaSurf: Efficient View-Aware Implicit Textured Surface Reconstruction", "abstract": "  Reconstructing real-world 3D objects has numerous applications in computer\nvision, such as virtual reality, video games, and animations. Ideally, 3D\nreconstruction methods should generate high-fidelity results with 3D\nconsistency in real-time. Traditional methods match pixels between images using\nphoto-consistency constraints or learned features, while differentiable\nrendering methods like Neural Radiance Fields (NeRF) use differentiable volume\nrendering or surface-based representation to generate high-fidelity scenes.\nHowever, these methods require excessive runtime for rendering, making them\nimpractical for daily applications. To address these challenges, we present\n$\\textbf{EvaSurf}$, an $\\textbf{E}$fficient $\\textbf{V}$iew-$\\textbf{A}$ware\nimplicit textured $\\textbf{Surf}$ace reconstruction method. In our method, we\nfirst employ an efficient surface-based model with a multi-view supervision\nmodule to ensure accurate mesh reconstruction. To enable high-fidelity\nrendering, we learn an implicit texture embedded with view-aware encoding to\ncapture view-dependent information. Furthermore, with the explicit geometry and\nthe implicit texture, we can employ a lightweight neural shader to reduce the\nexpense of computation and further support real-time rendering on common mobile\ndevices. Extensive experiments demonstrate that our method can reconstruct\nhigh-quality appearance and accurate mesh on both synthetic and real-world\ndatasets. Moreover, our method can be trained in just 1-2 hours using a single\nGPU and run on mobile devices at over 40 FPS (Frames Per Second), with a final\npackage required for rendering taking up only 40-50 MB.\n", "date": "2023-11-16", "categories": []}
{"id": "2312.13299", "title": "Compact 3D Scene Representation via Self-Organizing Gaussian Grids", "abstract": "  3D Gaussian Splatting has recently emerged as a highly promising technique\nfor modeling of static 3D scenes. In contrast to Neural Radiance Fields, it\nutilizes efficient rasterization allowing for very fast rendering at\nhigh-quality. However, the storage size is significantly higher, which hinders\npractical deployment, e.g. on resource constrained devices. In this paper, we\nintroduce a compact scene representation organizing the parameters of 3D\nGaussian Splatting (3DGS) into a 2D grid with local homogeneity, ensuring a\ndrastic reduction in storage requirements without compromising visual quality\nduring rendering. Central to our idea is the explicit exploitation of\nperceptual redundancies present in natural scenes. In essence, the inherent\nnature of a scene allows for numerous permutations of Gaussian parameters to\nequivalently represent it. To this end, we propose a novel highly parallel\nalgorithm that regularly arranges the high-dimensional Gaussian parameters into\na 2D grid while preserving their neighborhood structure. During training, we\nfurther enforce local smoothness between the sorted parameters in the grid. The\nuncompressed Gaussians use the same structure as 3DGS, ensuring a seamless\nintegration with established renderers. Our method achieves a reduction factor\nof 17x to 42x in size for complex scenes with no increase in training time,\nmarking a substantial leap forward in the domain of 3D scene distribution and\nconsumption. Additional information can be found on our project page:\nhttps://fraunhoferhhi.github.io/Self-Organizing-Gaussians/\n", "date": "2023-12-19", "categories": []}
{"id": "2405.17421", "title": "MoSca: Dynamic Gaussian Fusion from Casual Videos via 4D Motion\n  Scaffolds", "abstract": "  We introduce 4D Motion Scaffolds (MoSca), a modern 4D reconstruction system\ndesigned to reconstruct and synthesize novel views of dynamic scenes from\nmonocular videos captured casually in the wild. To address such a challenging\nand ill-posed inverse problem, we leverage prior knowledge from foundational\nvision models and lift the video data to a novel Motion Scaffold (MoSca)\nrepresentation, which compactly and smoothly encodes the underlying\nmotions/deformations. The scene geometry and appearance are then disentangled\nfrom the deformation field and are encoded by globally fusing the Gaussians\nanchored onto the MoSca and optimized via Gaussian Splatting. Additionally,\ncamera focal length and poses can be solved using bundle adjustment without the\nneed of any other pose estimation tools. Experiments demonstrate\nstate-of-the-art performance on dynamic rendering benchmarks and its\neffectiveness on real videos.\n", "date": "2024-05-27", "categories": []}
{"id": "2408.12191", "title": "Transientangelo: Few-Viewpoint Surface Reconstruction Using\n  Single-Photon Lidar", "abstract": "  We consider the problem of few-viewpoint 3D surface reconstruction using raw\nmeasurements from a lidar system. Lidar captures 3D scene geometry by emitting\npulses of light to a target and recording the speed-of-light time delay of the\nreflected light. However, conventional lidar systems do not output the raw,\ncaptured waveforms of backscattered light; instead, they pre-process these data\ninto a 3D point cloud. Since this procedure typically does not accurately model\nthe noise statistics of the system, exploit spatial priors, or incorporate\ninformation about downstream tasks, it ultimately discards useful information\nthat is encoded in raw measurements of backscattered light. Here, we propose to\nleverage raw measurements captured with a single-photon lidar system from\nmultiple viewpoints to optimize a neural surface representation of a scene. The\nmeasurements consist of time-resolved photon count histograms, or transients,\nwhich capture information about backscattered light at picosecond time scales.\nAdditionally, we develop new regularization strategies that improve robustness\nto photon noise, enabling accurate surface reconstruction with as few as 10\nphotons per pixel. Our method outperforms other techniques for few-viewpoint 3D\nreconstruction based on depth maps, point clouds, or conventional lidar as\ndemonstrated in simulation and with captured data.\n", "date": "2024-08-22", "categories": []}
{"id": "2411.18807", "title": "Reconstructing Animals and the Wild", "abstract": "  The idea of 3D reconstruction as scene understanding is foundational in\ncomputer vision. Reconstructing 3D scenes from 2D visual observations requires\nstrong priors to disambiguate structure. Much work has been focused on the\nanthropocentric, which, characterized by smooth surfaces, coherent normals, and\nregular edges, allows for the integration of strong geometric inductive biases.\nHere, we consider a more challenging problem where such assumptions do not\nhold: the reconstruction of natural scenes containing trees, bushes, boulders,\nand animals. While numerous works have attempted to tackle the problem of\nreconstructing animals in the wild, they have focused solely on the animal,\nneglecting environmental context. This limits their usefulness for analysis\ntasks, as animals exist inherently within the 3D world, and information is lost\nwhen environmental factors are disregarded. We propose a method to reconstruct\nnatural scenes from single images. We base our approach on recent advances\nleveraging the strong world priors ingrained in Large Language Models and train\nan autoregressive model to decode a CLIP embedding into a structured\ncompositional scene representation, encompassing both animals and the wild\n(RAW). To enable this, we propose a synthetic dataset comprising one million\nimages and thousands of assets. Our approach, having been trained solely on\nsynthetic data, generalizes to the task of reconstructing animals and their\nenvironments in real-world images. We will release our dataset and code to\nencourage future research at https://raw.is.tue.mpg.de/\n", "date": "2024-11-27", "categories": []}
{"id": "2411.18866", "title": "RIGI: Rectifying Image-to-3D Generation Inconsistency via\n  Uncertainty-aware Learning", "abstract": "  Given a single image of a target object, image-to-3D generation aims to\nreconstruct its texture and geometric shape. Recent methods often utilize\nintermediate media, such as multi-view images or videos, to bridge the gap\nbetween input image and the 3D target, thereby guiding the generation of both\nshape and texture. However, inconsistencies in the generated multi-view\nsnapshots frequently introduce noise and artifacts along object boundaries,\nundermining the 3D reconstruction process. To address this challenge, we\nleverage 3D Gaussian Splatting (3DGS) for 3D reconstruction, and explicitly\nintegrate uncertainty-aware learning into the reconstruction process. By\ncapturing the stochasticity between two Gaussian models, we estimate an\nuncertainty map, which is subsequently used for uncertainty-aware\nregularization to rectify the impact of inconsistencies. Specifically, we\noptimize both Gaussian models simultaneously, calculating the uncertainty map\nby evaluating the discrepancies between rendered images from identical\nviewpoints. Based on the uncertainty map, we apply adaptive pixel-wise loss\nweighting to regularize the models, reducing reconstruction intensity in\nhigh-uncertainty regions. This approach dynamically detects and mitigates\nconflicts in multi-view labels, leading to smoother results and effectively\nreducing artifacts. Extensive experiments show the effectiveness of our method\nin improving 3D generation quality by reducing inconsistencies and artifacts.\n", "date": "2024-11-28", "categories": []}
{"id": "2411.19102", "title": "360Recon: An Accurate Reconstruction Method Based on Depth Fusion from\n  360 Images", "abstract": "  360-degree images offer a significantly wider field of view compared to\ntraditional pinhole cameras, enabling sparse sampling and dense 3D\nreconstruction in low-texture environments. This makes them crucial for\napplications in VR, AR, and related fields. However, the inherent distortion\ncaused by the wide field of view affects feature extraction and matching,\nleading to geometric consistency issues in subsequent multi-view\nreconstruction. In this work, we propose 360Recon, an innovative MVS algorithm\nfor ERP images. The proposed spherical feature extraction module effectively\nmitigates distortion effects, and by combining the constructed 3D cost volume\nwith multi-scale enhanced features from ERP images, our approach achieves\nhigh-precision scene reconstruction while preserving local geometric\nconsistency. Experimental results demonstrate that 360Recon achieves\nstate-of-the-art performance and high efficiency in depth estimation and 3D\nreconstruction on existing public panoramic reconstruction datasets.\n", "date": "2024-11-28", "categories": []}
{"id": "2411.19474", "title": "Blurred LiDAR for Sharper 3D: Robust Handheld 3D Scanning with Diffuse\n  LiDAR and RGB", "abstract": "  3D surface reconstruction is essential across applications of virtual\nreality, robotics, and mobile scanning. However, RGB-based reconstruction often\nfails in low-texture, low-light, and low-albedo scenes. Handheld LiDARs, now\ncommon on mobile devices, aim to address these challenges by capturing depth\ninformation from time-of-flight measurements of a coarse grid of projected\ndots. Yet, these sparse LiDARs struggle with scene coverage on limited input\nviews, leaving large gaps in depth information. In this work, we propose using\nan alternative class of \"blurred\" LiDAR that emits a diffuse flash, greatly\nimproving scene coverage but introducing spatial ambiguity from mixed\ntime-of-flight measurements across a wide field of view. To handle these\nambiguities, we propose leveraging the complementary strengths of diffuse LiDAR\nwith RGB. We introduce a Gaussian surfel-based rendering framework with a\nscene-adaptive loss function that dynamically balances RGB and diffuse LiDAR\nsignals. We demonstrate that, surprisingly, diffuse LiDAR can outperform\ntraditional sparse LiDAR, enabling robust 3D scanning with accurate color and\ngeometry estimation in challenging environments.\n", "date": "2024-11-29", "categories": []}
{"id": "2411.19950", "title": "AlphaTablets: A Generic Plane Representation for 3D Planar\n  Reconstruction from Monocular Videos", "abstract": "  We introduce AlphaTablets, a novel and generic representation of 3D planes\nthat features continuous 3D surface and precise boundary delineation. By\nrepresenting 3D planes as rectangles with alpha channels, AlphaTablets combine\nthe advantages of current 2D and 3D plane representations, enabling accurate,\nconsistent and flexible modeling of 3D planes. We derive differentiable\nrasterization on top of AlphaTablets to efficiently render 3D planes into\nimages, and propose a novel bottom-up pipeline for 3D planar reconstruction\nfrom monocular videos. Starting with 2D superpixels and geometric cues from\npre-trained models, we initialize 3D planes as AlphaTablets and optimize them\nvia differentiable rendering. An effective merging scheme is introduced to\nfacilitate the growth and refinement of AlphaTablets. Through iterative\noptimization and merging, we reconstruct complete and accurate 3D planes with\nsolid surfaces and clear boundaries. Extensive experiments on the ScanNet\ndataset demonstrate state-of-the-art performance in 3D planar reconstruction,\nunderscoring the great potential of AlphaTablets as a generic 3D plane\nrepresentation for various applications. Project page is available at:\nhttps://hyzcluster.github.io/alphatablets\n", "date": "2024-11-29", "categories": []}
{"id": "2411.18072", "title": "SmileSplat: Generalizable Gaussian Splats for Unconstrained Sparse\n  Images", "abstract": "  Sparse Multi-view Images can be Learned to predict explicit radiance fields\nvia Generalizable Gaussian Splatting approaches, which can achieve wider\napplication prospects in real-life when ground-truth camera parameters are not\nrequired as inputs. In this paper, a novel generalizable Gaussian Splatting\nmethod, SmileSplat, is proposed to reconstruct pixel-aligned Gaussian surfels\nfor diverse scenarios only requiring unconstrained sparse multi-view images.\nFirst, Gaussian surfels are predicted based on the multi-head Gaussian\nregression decoder, which can are represented with less degree-of-freedom but\nhave better multi-view consistency. Furthermore, the normal vectors of Gaussian\nsurfel are enhanced based on high-quality of normal priors. Second, the\nGaussians and camera parameters (both extrinsic and intrinsic) are optimized to\nobtain high-quality Gaussian radiance fields for novel view synthesis tasks\nbased on the proposed Bundle-Adjusting Gaussian Splatting module. Extensive\nexperiments on novel view rendering and depth map prediction tasks are\nconducted on public datasets, demonstrating that the proposed method achieves\nstate-of-the-art performance in various 3D vision tasks. More information can\nbe found on our project page\n(https://yanyan-li.github.io/project/gs/smilesplat)\n", "date": "2024-11-27", "categories": []}
{"id": "2411.16779", "title": "NovelGS: Consistent Novel-view Denoising via Large Gaussian\n  Reconstruction Model", "abstract": "  We introduce NovelGS, a diffusion model for Gaussian Splatting (GS) given\nsparse-view images. Recent works leverage feed-forward networks to generate\npixel-aligned Gaussians, which could be fast rendered. Unfortunately, the\nmethod was unable to produce satisfactory results for areas not covered by the\ninput images due to the formulation of these methods. In contrast, we leverage\nthe novel view denoising through a transformer-based network to generate 3D\nGaussians. Specifically, by incorporating both conditional views and noisy\ntarget views, the network predicts pixel-aligned Gaussians for each view.\nDuring training, the rendered target and some additional views of the Gaussians\nare supervised. During inference, the target views are iteratively rendered and\ndenoised from pure noise. Our approach demonstrates state-of-the-art\nperformance in addressing the multi-view image reconstruction challenge. Due to\ngenerative modeling of unseen regions, NovelGS effectively reconstructs 3D\nobjects with consistent and sharp textures. Experimental results on publicly\navailable datasets indicate that NovelGS substantially surpasses existing\nimage-to-3D frameworks, both qualitatively and quantitatively. We also\ndemonstrate the potential of NovelGS in generative tasks, such as text-to-3D\nand image-to-3D, by integrating it with existing multiview diffusion models. We\nwill make the code publicly accessible.\n", "date": "2024-11-25", "categories": []}
{"id": "2411.16877", "title": "PreF3R: Pose-Free Feed-Forward 3D Gaussian Splatting from\n  Variable-length Image Sequence", "abstract": "  We present PreF3R, Pose-Free Feed-forward 3D Reconstruction from an image\nsequence of variable length. Unlike previous approaches, PreF3R removes the\nneed for camera calibration and reconstructs the 3D Gaussian field within a\ncanonical coordinate frame directly from a sequence of unposed images, enabling\nefficient novel-view rendering. We leverage DUSt3R's ability for pair-wise 3D\nstructure reconstruction, and extend it to sequential multi-view input via a\nspatial memory network, eliminating the need for optimization-based global\nalignment. Additionally, PreF3R incorporates a dense Gaussian parameter\nprediction head, which enables subsequent novel-view synthesis with\ndifferentiable rasterization. This allows supervising our model with the\ncombination of photometric loss and pointmap regression loss, enhancing both\nphotorealism and structural accuracy. Given a sequence of ordered images,\nPreF3R incrementally reconstructs the 3D Gaussian field at 20 FPS, therefore\nenabling real-time novel-view rendering. Empirical experiments demonstrate that\nPreF3R is an effective solution for the challenging task of pose-free\nfeed-forward novel-view synthesis, while also exhibiting robust generalization\nto unseen scenes.\n", "date": "2024-11-25", "categories": []}
{"id": "2406.03175", "title": "Dynamic 3D Gaussian Fields for Urban Areas", "abstract": "  We present an efficient neural 3D scene representation for novel-view\nsynthesis (NVS) in large-scale, dynamic urban areas. Existing works are not\nwell suited for applications like mixed-reality or closed-loop simulation due\nto their limited visual quality and non-interactive rendering speeds. Recently,\nrasterization-based approaches have achieved high-quality NVS at impressive\nspeeds. However, these methods are limited to small-scale, homogeneous data,\ni.e. they cannot handle severe appearance and geometry variations due to\nweather, season, and lighting and do not scale to larger, dynamic areas with\nthousands of images. We propose 4DGF, a neural scene representation that scales\nto large-scale dynamic urban areas, handles heterogeneous input data, and\nsubstantially improves rendering speeds. We use 3D Gaussians as an efficient\ngeometry scaffold while relying on neural fields as a compact and flexible\nappearance model. We integrate scene dynamics via a scene graph at global scale\nwhile modeling articulated motions on a local level via deformations. This\ndecomposed approach enables flexible scene composition suitable for real-world\napplications. In experiments, we surpass the state-of-the-art by over 3 dB in\nPSNR and more than 200 times in rendering speed.\n", "date": "2024-06-05", "categories": []}
{"id": "2411.09156", "title": "DyGASR: Dynamic Generalized Exponential Splatting with Surface Alignment\n  for Accelerated 3D Mesh Reconstruction", "abstract": "  Recent advancements in 3D Gaussian Splatting (3DGS), which lead to\nhigh-quality novel view synthesis and accelerated rendering, have remarkably\nimproved the quality of radiance field reconstruction. However, the extraction\nof mesh from a massive number of minute 3D Gaussian points remains great\nchallenge due to the large volume of Gaussians and difficulty of representation\nof sharp signals caused by their inherent low-pass characteristics. To address\nthis issue, we propose DyGASR, which utilizes generalized exponential function\ninstead of traditional 3D Gaussian to decrease the number of particles and\ndynamically optimize the representation of the captured signal. In addition, it\nis observed that reconstructing mesh with Generalized Exponential\nSplatting(GES) without modifications frequently leads to failures since the\ngeneralized exponential distribution centroids may not precisely align with the\nscene surface. To overcome this, we adopt Sugar's approach and introduce\nGeneralized Surface Regularization (GSR), which reduces the smallest scaling\nvector of each point cloud to zero and ensures normal alignment perpendicular\nto the surface, facilitating subsequent Poisson surface mesh reconstruction.\nAdditionally, we propose a dynamic resolution adjustment strategy that utilizes\na cosine schedule to gradually increase image resolution from low to high\nduring the training stage, thus avoiding constant full resolution, which\nsignificantly boosts the reconstruction speed. Our approach surpasses existing\n3DGS-based mesh reconstruction methods, as evidenced by extensive evaluations\non various scene datasets, demonstrating a 25\\% increase in speed, and a 30\\%\nreduction in memory usage.\n", "date": "2024-11-14", "categories": []}
{"id": "2411.15420", "title": "Semi-supervised Single-view 3D Reconstruction via Multi Shape Prior\n  Fusion Strategy and Self-Attention", "abstract": "  In the domain of single-view 3D reconstruction, traditional techniques have\nfrequently relied on expensive and time-intensive 3D annotation data. Facing\nthe challenge of annotation acquisition, semi-supervised learning strategies\noffer an innovative approach to reduce the dependence on labeled data. Despite\nthese developments, the utilization of this learning paradigm in 3D\nreconstruction tasks remains relatively constrained. In this research, we\ncreated an innovative semi-supervised framework for 3D reconstruction that\ndistinctively uniquely introduces a multi shape prior fusion strategy,\nintending to guide the creation of more realistic object structures.\nAdditionally, to improve the quality of shape generation, we integrated a\nself-attention module into the traditional decoder. In benchmark tests on the\nShapeNet dataset, our method substantially outperformed existing supervised\nlearning methods at diverse labeled ratios of 1\\%, 10\\%, and 20\\%. Moreover, it\nshowcased excellent performance on the real-world Pix3D dataset. Through\ncomprehensive experiments on ShapeNet, our framework demonstrated a 3.3\\%\nperformance improvement over the baseline. Moreover, stringent ablation studies\nfurther confirmed the notable effectiveness of our approach. Our code has been\nreleased on https://github.com/NWUzhouwei/SSMP\n", "date": "2024-11-23", "categories": []}
{"id": "2411.15779", "title": "ZeroGS: Training 3D Gaussian Splatting from Unposed Images", "abstract": "  Neural radiance fields (NeRF) and 3D Gaussian Splatting (3DGS) are popular\ntechniques to reconstruct and render photo-realistic images. However, the\npre-requisite of running Structure-from-Motion (SfM) to get camera poses limits\ntheir completeness. While previous methods can reconstruct from a few unposed\nimages, they are not applicable when images are unordered or densely captured.\nIn this work, we propose ZeroGS to train 3DGS from hundreds of unposed and\nunordered images. Our method leverages a pretrained foundation model as the\nneural scene representation. Since the accuracy of the predicted pointmaps does\nnot suffice for accurate image registration and high-fidelity image rendering,\nwe propose to mitigate the issue by initializing and finetuning the pretrained\nmodel from a seed image. Images are then progressively registered and added to\nthe training buffer, which is further used to train the model. We also propose\nto refine the camera poses and pointmaps by minimizing a point-to-camera ray\nconsistency loss across multiple views. Experiments on the LLFF dataset, the\nMipNeRF360 dataset, and the Tanks-and-Temples dataset show that our method\nrecovers more accurate camera poses than state-of-the-art pose-free NeRF/3DGS\nmethods, and even renders higher quality images than 3DGS with COLMAP poses.\nOur project page is available at https://aibluefisher.github.io/ZeroGS.\n", "date": "2024-11-24", "categories": []}
{"id": "2406.01593", "title": "MaGS: Reconstructing and Simulating Dynamic 3D Objects with\n  Mesh-adsorbed Gaussian Splatting", "abstract": "  3D reconstruction and simulation, although interrelated, have distinct\nobjectives: reconstruction requires a flexible 3D representation that can adapt\nto diverse scenes, while simulation needs a structured representation to model\nmotion principles effectively. This paper introduces the Mesh-adsorbed Gaussian\nSplatting (MaGS) method to address this challenge. MaGS constrains 3D Gaussians\nto roam near the mesh, creating a mutually adsorbed mesh-Gaussian 3D\nrepresentation. Such representation harnesses both the rendering flexibility of\n3D Gaussians and the structured property of meshes. To achieve this, we\nintroduce RMD-Net, a network that learns motion priors from video data to\nrefine mesh deformations, alongside RGD-Net, which models the relative\ndisplacement between the mesh and Gaussians to enhance rendering fidelity under\nmesh constraints. To generalize to novel, user-defined deformations beyond\ninput video without reliance on temporal data, we propose MPE-Net, which\nleverages inherent mesh information to bootstrap RMD-Net and RGD-Net. Due to\nthe universality of meshes, MaGS is compatible with various deformation priors\nsuch as ARAP, SMPL, and soft physics simulation. Extensive experiments on the\nD-NeRF, DG-Mesh, and PeopleSnapshot datasets demonstrate that MaGS achieves\nstate-of-the-art performance in both reconstruction and simulation.\n", "date": "2024-06-03", "categories": []}
{"id": "2410.23245", "title": "PointRecon: Online Point-based 3D Reconstruction via Ray-based 2D-3D\n  Matching", "abstract": "  We propose a novel online, point-based 3D reconstruction method from posed\nmonocular RGB videos. Our model maintains a global point cloud representation\nof the scene, continuously updating the features and 3D locations of points as\nnew images are observed. It expands the point cloud with newly detected points\nwhile carefully removing redundancies. The point cloud updates and the depth\npredictions for new points are achieved through a novel ray-based 2D-3D feature\nmatching technique, which is robust against errors in previous point position\npredictions. In contrast to offline methods, our approach processes\ninfinite-length sequences and provides real-time updates. Additionally, the\npoint cloud imposes no pre-defined resolution or scene size constraints, and\nits unified global representation ensures view consistency across perspectives.\nExperiments on the ScanNet dataset show that our method achieves comparable\nquality among online MVS approaches. Project page:\nhttps://arthurhero.github.io/projects/pointrecon\n", "date": "2024-10-30", "categories": []}
{"id": "2411.15018", "title": "Neural 4D Evolution under Large Topological Changes from 2D Images", "abstract": "  In the literature, it has been shown that the evolution of the known explicit\n3D surface to the target one can be learned from 2D images using the\ninstantaneous flow field, where the known and target 3D surfaces may largely\ndiffer in topology. We are interested in capturing 4D shapes whose topology\nchanges largely over time. We encounter that the straightforward extension of\nthe existing 3D-based method to the desired 4D case performs poorly.\n  In this work, we address the challenges in extending 3D neural evolution to\n4D under large topological changes by proposing two novel modifications. More\nprecisely, we introduce (i) a new architecture to discretize and encode the\ndeformation and learn the SDF and (ii) a technique to impose the temporal\nconsistency. (iii) Also, we propose a rendering scheme for color prediction\nbased on Gaussian splatting. Furthermore, to facilitate learning directly from\n2D images, we propose a learning framework that can disentangle the geometry\nand appearance from RGB images. This method of disentanglement, while also\nuseful for the 4D evolution problem that we are concentrating on, is also novel\nand valid for static scenes. Our extensive experiments on various data provide\nawesome results and, most importantly, open a new approach toward\nreconstructing challenging scenes with significant topological changes and\ndeformations. Our source code and the dataset are publicly available at\nhttps://github.com/insait-institute/N4DE.\n", "date": "2024-11-22", "categories": []}
{"id": "2308.08812", "title": "A Fusion of Variational Distribution Priors and Saliency Map Replay for\n  Continual 3D Reconstruction", "abstract": "  Single-image 3D reconstruction is a research challenge focused on predicting\n3D object shapes from single-view images. This task requires significant data\nacquisition to predict both visible and occluded portions of the shape.\nFurthermore, learning-based methods face the difficulty of creating a\ncomprehensive training dataset for all possible classes. To this end, we\npropose a continual learning-based 3D reconstruction method where our goal is\nto design a model using Variational Priors that can still reconstruct the\npreviously seen classes reasonably even after training on new classes.\nVariational Priors represent abstract shapes and combat forgetting, whereas\nsaliency maps preserve object attributes with less memory usage. This is vital\ndue to resource constraints in storing extensive training data. Additionally,\nwe introduce saliency map-based experience replay to capture global and\ndistinct object features. Thorough experiments show competitive results\ncompared to established methods, both quantitatively and qualitatively.\n", "date": "2023-08-17", "categories": []}
{"id": "2405.14342", "title": "RoGs: Large Scale Road Surface Reconstruction with Meshgrid Gaussian", "abstract": "  Road surface reconstruction plays a crucial role in autonomous driving, which\ncan be used for road lane perception and autolabeling. Recently, mesh-based\nroad surface reconstruction algorithms have shown promising reconstruction\nresults. However, these mesh-based methods suffer from slow speed and poor\nreconstruction quality. To address these limitations, we propose a novel\nlarge-scale road surface reconstruction approach with meshgrid Gaussian, named\nRoGs. Specifically, we model the road surface by placing Gaussian surfels in\nthe vertices of a uniformly distributed square mesh, where each surfel stores\ncolor, semantic, and geometric information. This square mesh-based layout\ncovers the entire road with fewer Gaussian surfels and reduces the overlap\nbetween Gaussian surfels during training. In addition, because the road surface\nhas no thickness, 2D Gaussian surfel is more consistent with the physical\nreality of the road surface than 3D Gaussian sphere. Then, unlike previous\ninitialization methods that rely on point clouds, we introduce a vehicle\npose-based initialization method to initialize the height and rotation of the\nGaussian surfel. Thanks to this meshgrid Gaussian modeling and pose-based\ninitialization, our method achieves significant speedups while improving\nreconstruction quality. We obtain excellent results in reconstruction of road\nsurfaces in a variety of challenging real-world scenes.\n", "date": "2024-05-23", "categories": []}
{"id": "2411.12309", "title": "DGTR: Distributed Gaussian Turbo-Reconstruction for Sparse-View Vast\n  Scenes", "abstract": "  Novel-view synthesis (NVS) approaches play a critical role in vast scene\nreconstruction. However, these methods rely heavily on dense image inputs and\nprolonged training times, making them unsuitable where computational resources\nare limited. Additionally, few-shot methods often struggle with poor\nreconstruction quality in vast environments. This paper presents DGTR, a novel\ndistributed framework for efficient Gaussian reconstruction for sparse-view\nvast scenes. Our approach divides the scene into regions, processed\nindependently by drones with sparse image inputs. Using a feed-forward Gaussian\nmodel, we predict high-quality Gaussian primitives, followed by a global\nalignment algorithm to ensure geometric consistency. Synthetic views and depth\npriors are incorporated to further enhance training, while a distillation-based\nmodel aggregation mechanism enables efficient reconstruction. Our method\nachieves high-quality large-scale scene reconstruction and novel-view synthesis\nin significantly reduced training times, outperforming existing approaches in\nboth speed and scalability. We demonstrate the effectiveness of our framework\non vast aerial scenes, achieving high-quality results within minutes. Code will\nreleased on our [https://3d-aigc.github.io/DGTR].\n", "date": "2024-11-19", "categories": []}
{"id": "2411.12635", "title": "M3D: Dual-Stream Selective State Spaces and Depth-Driven Framework for\n  High-Fidelity Single-View 3D Reconstruction", "abstract": "  The precise reconstruction of 3D objects from a single RGB image in complex\nscenes presents a critical challenge in virtual reality, autonomous driving,\nand robotics. Existing neural implicit 3D representation methods face\nsignificant difficulties in balancing the extraction of global and local\nfeatures, particularly in diverse and complex environments, leading to\ninsufficient reconstruction precision and quality. We propose M3D, a novel\nsingle-view 3D reconstruction framework, to tackle these challenges. This\nframework adopts a dual-stream feature extraction strategy based on Selective\nState Spaces to effectively balance the extraction of global and local\nfeatures, thereby improving scene comprehension and representation precision.\nAdditionally, a parallel branch extracts depth information, effectively\nintegrating visual and geometric features to enhance reconstruction quality and\npreserve intricate details. Experimental results indicate that the fusion of\nmulti-scale features with depth information via the dual-branch feature\nextraction significantly boosts geometric consistency and fidelity, achieving\nstate-of-the-art reconstruction performance.\n", "date": "2024-11-19", "categories": []}
{"id": "2411.12592", "title": "SPARS3R: Semantic Prior Alignment and Regularization for Sparse 3D\n  Reconstruction", "abstract": "  Recent efforts in Gaussian-Splat-based Novel View Synthesis can achieve\nphotorealistic rendering; however, such capability is limited in sparse-view\nscenarios due to sparse initialization and over-fitting floaters. Recent\nprogress in depth estimation and alignment can provide dense point cloud with\nfew views; however, the resulting pose accuracy is suboptimal. In this work, we\npresent SPARS3R, which combines the advantages of accurate pose estimation from\nStructure-from-Motion and dense point cloud from depth estimation. To this end,\nSPARS3R first performs a Global Fusion Alignment process that maps a prior\ndense point cloud to a sparse point cloud from Structure-from-Motion based on\ntriangulated correspondences. RANSAC is applied during this process to\ndistinguish inliers and outliers. SPARS3R then performs a second, Semantic\nOutlier Alignment step, which extracts semantically coherent regions around the\noutliers and performs local alignment in these regions. Along with several\nimprovements in the evaluation process, we demonstrate that SPARS3R can achieve\nphotorealistic rendering with sparse images and significantly outperforms\nexisting approaches.\n", "date": "2024-11-15", "categories": []}
{"id": "2406.10853", "title": "MV2Cyl: Reconstructing 3D Extrusion Cylinders from Multi-View Images", "abstract": "  We present MV2Cyl, a novel method for reconstructing 3D from 2D multi-view\nimages, not merely as a field or raw geometry but as a sketch-extrude CAD\nmodel. Extracting extrusion cylinders from raw 3D geometry has been extensively\nresearched in computer vision, while the processing of 3D data through neural\nnetworks has remained a bottleneck. Since 3D scans are generally accompanied by\nmulti-view images, leveraging 2D convolutional neural networks allows these\nimages to be exploited as a rich source for extracting extrusion cylinder\ninformation. However, we observe that extracting only the surface information\nof the extrudes and utilizing it results in suboptimal outcomes due to the\nchallenges in the occlusion and surface segmentation. By synergizing with the\nextracted base curve information, we achieve the optimal reconstruction result\nwith the best accuracy in 2D sketch and extrude parameter estimation. Our\nexperiments, comparing our method with previous work that takes a raw 3D point\ncloud as input, demonstrate the effectiveness of our approach by taking\nadvantage of multi-view images. Our project page can be found at\nhttp://mv2cyl.github.io .\n", "date": "2024-06-16", "categories": []}
{"id": "2411.10825", "title": "ARM: Appearance Reconstruction Model for Relightable 3D Generation", "abstract": "  Recent image-to-3D reconstruction models have greatly advanced geometry\ngeneration, but they still struggle to faithfully generate realistic\nappearance. To address this, we introduce ARM, a novel method that reconstructs\nhigh-quality 3D meshes and realistic appearance from sparse-view images. The\ncore of ARM lies in decoupling geometry from appearance, processing appearance\nwithin the UV texture space. Unlike previous methods, ARM improves texture\nquality by explicitly back-projecting measurements onto the texture map and\nprocessing them in a UV space module with a global receptive field. To resolve\nambiguities between material and illumination in input images, ARM introduces a\nmaterial prior that encodes semantic appearance information, enhancing the\nrobustness of appearance decomposition. Trained on just 8 H100 GPUs, ARM\noutperforms existing methods both quantitatively and qualitatively.\n", "date": "2024-11-16", "categories": []}
{"id": "2411.11196", "title": "PickScan: Object discovery and reconstruction from handheld interactions", "abstract": "  Reconstructing compositional 3D representations of scenes, where each object\nis represented with its own 3D model, is a highly desirable capability in\nrobotics and augmented reality. However, most existing methods rely heavily on\nstrong appearance priors for object discovery, therefore only working on those\nclasses of objects on which the method has been trained, or do not allow for\nobject manipulation, which is necessary to scan objects fully and to guide\nobject discovery in challenging scenarios. We address these limitations with a\nnovel interaction-guided and class-agnostic method based on object\ndisplacements that allows a user to move around a scene with an RGB-D camera,\nhold up objects, and finally outputs one 3D model per held-up object. Our main\ncontribution to this end is a novel approach to detecting user-object\ninteractions and extracting the masks of manipulated objects. On a\ncustom-captured dataset, our pipeline discovers manipulated objects with 78.3%\nprecision at 100% recall and reconstructs them with a mean chamfer distance of\n0.90cm. Compared to Co-Fusion, the only comparable interaction-based and\nclass-agnostic baseline, this corresponds to a reduction in chamfer distance of\n73% while detecting 99% fewer false positives.\n", "date": "2024-11-17", "categories": []}
{"id": "2408.04268", "title": "Evaluating Modern Approaches in 3D Scene Reconstruction: NeRF vs\n  Gaussian-Based Methods", "abstract": "  Exploring the capabilities of Neural Radiance Fields (NeRF) and\nGaussian-based methods in the context of 3D scene reconstruction, this study\ncontrasts these modern approaches with traditional Simultaneous Localization\nand Mapping (SLAM) systems. Utilizing datasets such as Replica and ScanNet, we\nassess performance based on tracking accuracy, mapping fidelity, and view\nsynthesis. Findings reveal that NeRF excels in view synthesis, offering unique\ncapabilities in generating new perspectives from existing data, albeit at\nslower processing speeds. Conversely, Gaussian-based methods provide rapid\nprocessing and significant expressiveness but lack comprehensive scene\ncompletion. Enhanced by global optimization and loop closure techniques, newer\nmethods like NICE-SLAM and SplaTAM not only surpass older frameworks such as\nORB-SLAM2 in terms of robustness but also demonstrate superior performance in\ndynamic and complex environments. This comparative analysis bridges theoretical\nresearch with practical implications, shedding light on future developments in\nrobust 3D scene reconstruction across various real-world applications.\n", "date": "2024-08-08", "categories": []}
{"id": "2408.12677", "title": "GSFusion: Online RGB-D Mapping Where Gaussian Splatting Meets TSDF\n  Fusion", "abstract": "  Traditional volumetric fusion algorithms preserve the spatial structure of 3D\nscenes, which is beneficial for many tasks in computer vision and robotics.\nHowever, they often lack realism in terms of visualization. Emerging 3D\nGaussian splatting bridges this gap, but existing Gaussian-based reconstruction\nmethods often suffer from artifacts and inconsistencies with the underlying 3D\nstructure, and struggle with real-time optimization, unable to provide users\nwith immediate feedback in high quality. One of the bottlenecks arises from the\nmassive amount of Gaussian parameters that need to be updated during\noptimization. Instead of using 3D Gaussian as a standalone map representation,\nwe incorporate it into a volumetric mapping system to take advantage of\ngeometric information and propose to use a quadtree data structure on images to\ndrastically reduce the number of splats initialized. In this way, we\nsimultaneously generate a compact 3D Gaussian map with fewer artifacts and a\nvolumetric map on the fly. Our method, GSFusion, significantly enhances\ncomputational efficiency without sacrificing rendering quality, as demonstrated\non both synthetic and real datasets. Code will be available at\nhttps://github.com/goldoak/GSFusion.\n", "date": "2024-08-22", "categories": []}
{"id": "2407.09733", "title": "Textured-GS: Gaussian Splatting with Spatially Defined Color and Opacity", "abstract": "  In this paper, we introduce Textured-GS, an innovative method for rendering\nGaussian splatting that incorporates spatially defined color and opacity\nvariations using Spherical Harmonics (SH). This approach enables each Gaussian\nto exhibit a richer representation by accommodating varying colors and\nopacities across its surface, significantly enhancing rendering quality\ncompared to traditional methods. To demonstrate the merits of our approach, we\nhave adapted the Mini-Splatting architecture to integrate textured Gaussians\nwithout increasing the number of Gaussians. Our experiments across multiple\nreal-world datasets show that Textured-GS consistently outperforms both the\nbaseline Mini-Splatting and standard 3DGS in terms of visual fidelity. The\nresults highlight the potential of Textured-GS to advance Gaussian-based\nrendering technologies, promising more efficient and high-quality scene\nreconstructions. Our implementation is available at\nhttps://github.com/ZhentaoHuang/Textured-GS.\n", "date": "2024-07-13", "categories": []}
{"id": "2405.12369", "title": "AtomGS: Atomizing Gaussian Splatting for High-Fidelity Radiance Field", "abstract": "  3D Gaussian Splatting (3DGS) has recently advanced radiance field\nreconstruction by offering superior capabilities for novel view synthesis and\nreal-time rendering speed. However, its strategy of blending optimization and\nadaptive density control might lead to sub-optimal results; it can sometimes\nyield noisy geometry and blurry artifacts due to prioritizing optimizing large\nGaussians at the cost of adequately densifying smaller ones. To address this,\nwe introduce AtomGS, consisting of Atomized Proliferation and Geometry-Guided\nOptimization. The Atomized Proliferation constrains ellipsoid Gaussians of\nvarious sizes into more uniform-sized Atom Gaussians. The strategy enhances the\nrepresentation of areas with fine features by placing greater emphasis on\ndensification in accordance with scene details. In addition, we proposed a\nGeometry-Guided Optimization approach that incorporates an Edge-Aware Normal\nLoss. This optimization method effectively smooths flat surfaces while\npreserving intricate details. Our evaluation shows that AtomGS outperforms\nexisting state-of-the-art methods in rendering quality. Additionally, it\nachieves competitive accuracy in geometry reconstruction and offers a\nsignificant improvement in training speed over other SDF-based methods. More\ninteractive demos can be found in our website\n(https://rongliu-leo.github.io/AtomGS/).\n", "date": "2024-05-20", "categories": []}
{"id": "2303.10083", "title": "$\\alpha$Surf: Implicit Surface Reconstruction for Semi-Transparent and\n  Thin Objects with Decoupled Geometry and Opacity", "abstract": "  Implicit surface representations such as the signed distance function (SDF)\nhave emerged as a promising approach for image-based surface reconstruction.\nHowever, existing optimization methods assume solid surfaces and are therefore\nunable to properly reconstruct semi-transparent surfaces and thin structures,\nwhich also exhibit low opacity due to the blending effect with the background.\nWhile neural radiance field (NeRF) based methods can model semi-transparency\nand achieve photo-realistic quality in synthesized novel views, their\nvolumetric geometry representation tightly couples geometry and opacity, and\ntherefore cannot be easily converted into surfaces without introducing\nartifacts. We present $\\alpha$Surf, a novel surface representation with\ndecoupled geometry and opacity for the reconstruction of semi-transparent and\nthin surfaces where the colors mix. Ray-surface intersections on our\nrepresentation can be found in closed-form via analytical solutions of cubic\npolynomials, avoiding Monte-Carlo sampling and is fully differentiable by\nconstruction. Our qualitative and quantitative evaluations show that our\napproach can accurately reconstruct surfaces with semi-transparent and thin\nparts with fewer artifacts, achieving better reconstruction quality than\nstate-of-the-art SDF and NeRF methods. Website: https://alphasurf.netlify.app/\n", "date": "2023-03-17", "categories": []}
{"id": "2403.17822", "title": "DN-Splatter: Depth and Normal Priors for Gaussian Splatting and Meshing", "abstract": "  High-fidelity 3D reconstruction of common indoor scenes is crucial for VR and\nAR applications. 3D Gaussian splatting, a novel differentiable rendering\ntechnique, has achieved state-of-the-art novel view synthesis results with high\nrendering speeds and relatively low training times. However, its performance on\nscenes commonly seen in indoor datasets is poor due to the lack of geometric\nconstraints during optimization. In this work, we explore the use of readily\naccessible geometric cues to enhance Gaussian splatting optimization in\nchallenging, ill-posed, and textureless scenes. We extend 3D Gaussian splatting\nwith depth and normal cues to tackle challenging indoor datasets and showcase\ntechniques for efficient mesh extraction. Specifically, we regularize the\noptimization procedure with depth information, enforce local smoothness of\nnearby Gaussians, and use off-the-shelf monocular networks to achieve better\nalignment with the true scene geometry. We propose an adaptive depth loss based\non the gradient of color images, improving depth estimation and novel view\nsynthesis results over various baselines. Our simple yet effective\nregularization technique enables direct mesh extraction from the Gaussian\nrepresentation, yielding more physically accurate reconstructions of indoor\nscenes.\n", "date": "2024-03-26", "categories": []}
{"id": "2410.21739", "title": "SS3DM: Benchmarking Street-View Surface Reconstruction with a Synthetic\n  3D Mesh Dataset", "abstract": "  Reconstructing accurate 3D surfaces for street-view scenarios is crucial for\napplications such as digital entertainment and autonomous driving simulation.\nHowever, existing street-view datasets, including KITTI, Waymo, and nuScenes,\nonly offer noisy LiDAR points as ground-truth data for geometric evaluation of\nreconstructed surfaces. These geometric ground-truths often lack the necessary\nprecision to evaluate surface positions and do not provide data for assessing\nsurface normals. To overcome these challenges, we introduce the SS3DM dataset,\ncomprising precise \\textbf{S}ynthetic \\textbf{S}treet-view \\textbf{3D}\n\\textbf{M}esh models exported from the CARLA simulator. These mesh models\nfacilitate accurate position evaluation and include normal vectors for\nevaluating surface normal. To simulate the input data in realistic driving\nscenarios for 3D reconstruction, we virtually drive a vehicle equipped with six\nRGB cameras and five LiDAR sensors in diverse outdoor scenes. Leveraging this\ndataset, we establish a benchmark for state-of-the-art surface reconstruction\nmethods, providing a comprehensive evaluation of the associated challenges.\n  For more information, visit our homepage at https://ss3dm.top.\n", "date": "2024-10-29", "categories": []}
{"id": "2411.04963", "title": "VAIR: Visuo-Acoustic Implicit Representations for Low-Cost, Multi-Modal\n  Transparent Surface Reconstruction in Indoor Scenes", "abstract": "  Mobile robots operating indoors must be prepared to navigate challenging\nscenes that contain transparent surfaces. This paper proposes a novel method\nfor the fusion of acoustic and visual sensing modalities through implicit\nneural representations to enable dense reconstruction of transparent surfaces\nin indoor scenes. We propose a novel model that leverages generative latent\noptimization to learn an implicit representation of indoor scenes consisting of\ntransparent surfaces. We demonstrate that we can query the implicit\nrepresentation to enable volumetric rendering in image space or 3D geometry\nreconstruction (point clouds or mesh) with transparent surface prediction. We\nevaluate our method's effectiveness qualitatively and quantitatively on a new\ndataset collected using a custom, low-cost sensing platform featuring RGB-D\ncameras and ultrasonic sensors. Our method exhibits significant improvement\nover state-of-the-art for transparent surface reconstruction.\n", "date": "2024-11-07", "categories": []}
{"id": "2404.03202", "title": "OmniGS: Fast Radiance Field Reconstruction using Omnidirectional\n  Gaussian Splatting", "abstract": "  Photorealistic reconstruction relying on 3D Gaussian Splatting has shown\npromising potential in various domains. However, the current 3D Gaussian\nSplatting system only supports radiance field reconstruction using undistorted\nperspective images. In this paper, we present OmniGS, a novel omnidirectional\nGaussian splatting system, to take advantage of omnidirectional images for fast\nradiance field reconstruction. Specifically, we conduct a theoretical analysis\nof spherical camera model derivatives in 3D Gaussian Splatting. According to\nthe derivatives, we then implement a new GPU-accelerated omnidirectional\nrasterizer that directly splats 3D Gaussians onto the equirectangular screen\nspace for omnidirectional image rendering. We realize differentiable\noptimization of the omnidirectional radiance field without the requirement of\ncube-map rectification or tangent-plane approximation. Extensive experiments\nconducted in egocentric and roaming scenarios demonstrate that our method\nachieves state-of-the-art reconstruction quality and high rendering speed using\nomnidirectional images. The code will be publicly available.\n", "date": "2024-04-04", "categories": []}
{"id": "2405.17705", "title": "DC-Gaussian: Improving 3D Gaussian Splatting for Reflective Dash Cam\n  Videos", "abstract": "  We present DC-Gaussian, a new method for generating novel views from\nin-vehicle dash cam videos. While neural rendering techniques have made\nsignificant strides in driving scenarios, existing methods are primarily\ndesigned for videos collected by autonomous vehicles. However, these videos are\nlimited in both quantity and diversity compared to dash cam videos, which are\nmore widely used across various types of vehicles and capture a broader range\nof scenarios. Dash cam videos often suffer from severe obstructions such as\nreflections and occlusions on the windshields, which significantly impede the\napplication of neural rendering techniques. To address this challenge, we\ndevelop DC-Gaussian based on the recent real-time neural rendering technique 3D\nGaussian Splatting (3DGS). Our approach includes an adaptive image\ndecomposition module to model reflections and occlusions in a unified manner.\nAdditionally, we introduce illumination-aware obstruction modeling to manage\nreflections and occlusions under varying lighting conditions. Lastly, we employ\na geometry-guided Gaussian enhancement strategy to improve rendering details by\nincorporating additional geometry priors. Experiments on self-captured and\npublic dash cam videos show that our method not only achieves state-of-the-art\nperformance in novel view synthesis, but also accurately reconstructing\ncaptured scenes getting rid of obstructions. See the project page for code,\ndata: https://linhanwang.github.io/dcgaussian/.\n", "date": "2024-05-27", "categories": []}
{"id": "2407.02077", "title": "Hierarchical Temporal Context Learning for Camera-based Semantic Scene\n  Completion", "abstract": "  Camera-based 3D semantic scene completion (SSC) is pivotal for predicting\ncomplicated 3D layouts with limited 2D image observations. The existing\nmainstream solutions generally leverage temporal information by roughly\nstacking history frames to supplement the current frame, such straightforward\ntemporal modeling inevitably diminishes valid clues and increases learning\ndifficulty. To address this problem, we present HTCL, a novel Hierarchical\nTemporal Context Learning paradigm for improving camera-based semantic scene\ncompletion. The primary innovation of this work involves decomposing temporal\ncontext learning into two hierarchical steps: (a) cross-frame affinity\nmeasurement and (b) affinity-based dynamic refinement. Firstly, to separate\ncritical relevant context from redundant information, we introduce the pattern\naffinity with scale-aware isolation and multiple independent learners for\nfine-grained contextual correspondence modeling. Subsequently, to dynamically\ncompensate for incomplete observations, we adaptively refine the feature\nsampling locations based on initially identified locations with high affinity\nand their neighboring relevant regions. Our method ranks $1^{st}$ on the\nSemanticKITTI benchmark and even surpasses LiDAR-based methods in terms of mIoU\non the OpenOccupancy benchmark. Our code is available on\nhttps://github.com/Arlo0o/HTCL.\n", "date": "2024-07-02", "categories": []}
{"id": "2411.02972", "title": "Exploring Seasonal Variability in the Context of Neural Radiance Fields\n  for 3D Reconstruction on Satellite Imagery", "abstract": "  In this work, the seasonal predictive capabilities of Neural Radiance Fields\n(NeRF) applied to satellite images are investigated. Focusing on the\nutilization of satellite data, the study explores how Sat-NeRF, a novel\napproach in computer vision, performs in predicting seasonal variations across\ndifferent months. Through comprehensive analysis and visualization, the study\nexamines the model's ability to capture and predict seasonal changes,\nhighlighting specific challenges and strengths. Results showcase the impact of\nthe sun direction on predictions, revealing nuanced details in seasonal\ntransitions, such as snow cover, color accuracy, and texture representation in\ndifferent landscapes. Given these results, we propose Planet-NeRF, an extension\nto Sat-NeRF capable of incorporating seasonal variability through a set of\nmonth embedding vectors. Comparative evaluations reveal that Planet-NeRF\noutperforms prior models in the case where seasonal changes are present. The\nextensive evaluation combined with the proposed method offers promising avenues\nfor future research in this domain.\n", "date": "2024-11-05", "categories": []}
{"id": "2408.06543", "title": "HDRGS: High Dynamic Range Gaussian Splatting", "abstract": "  Recent years have witnessed substantial advancements in the field of 3D\nreconstruction from 2D images, particularly following the introduction of the\nneural radiance field (NeRF) technique. However, reconstructing a 3D high\ndynamic range (HDR) radiance field, which aligns more closely with real-world\nconditions, from 2D multi-exposure low dynamic range (LDR) images continues to\npose significant challenges. Approaches to this issue fall into two categories:\ngrid-based and implicit-based. Implicit methods, using multi-layer perceptrons\n(MLP), face inefficiencies, limited solvability, and overfitting risks.\nConversely, grid-based methods require significant memory and struggle with\nimage quality and long training times. In this paper, we introduce Gaussian\nSplatting-a recent, high-quality, real-time 3D reconstruction technique-into\nthis domain. We further develop the High Dynamic Range Gaussian Splatting\n(HDR-GS) method, designed to address the aforementioned challenges. This method\nenhances color dimensionality by including luminance and uses an asymmetric\ngrid for tone-mapping, swiftly and precisely converting pixel irradiance to\ncolor. Our approach improves HDR scene recovery accuracy and integrates a novel\ncoarse-to-fine strategy to speed up model convergence, enhancing robustness\nagainst sparse viewpoints and exposure extremes, and preventing local optima.\nExtensive testing confirms that our method surpasses current state-of-the-art\ntechniques in both synthetic and real-world scenarios.\n", "date": "2024-08-13", "categories": []}
{"id": "2411.01208", "title": "MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level\n  Queries at Multi-Step", "abstract": "  Reconstructing a continuous surface from a raw 3D point cloud is a\nchallenging task. Recent methods usually train neural networks to overfit on\nsingle point clouds to infer signed distance functions (SDFs). However, neural\nnetworks tend to smooth local details due to the lack of ground truth signed\ndistances or normals, which limits the performance of overfitting-based methods\nin reconstruction tasks. To resolve this issue, we propose a novel method,\nnamed MultiPull, to learn multi-scale implicit fields from raw point clouds by\noptimizing accurate SDFs from coarse to fine. We achieve this by mapping 3D\nquery points into a set of frequency features, which makes it possible to\nleverage multi-level features during optimization. Meanwhile, we introduce\noptimization constraints from the perspective of spatial distance and normal\nconsistency, which play a key role in point cloud reconstruction based on\nmulti-scale optimization strategies. Our experiments on widely used object and\nscene benchmarks demonstrate that our method outperforms the state-of-the-art\nmethods in surface reconstruction.\n", "date": "2024-11-02", "categories": []}
{"id": "2411.01226", "title": "MonoPlane: Exploiting Monocular Geometric Cues for Generalizable 3D\n  Plane Reconstruction", "abstract": "  This paper presents a generalizable 3D plane detection and reconstruction\nframework named MonoPlane. Unlike previous robust estimator-based works (which\nrequire multiple images or RGB-D input) and learning-based works (which suffer\nfrom domain shift), MonoPlane combines the best of two worlds and establishes a\nplane reconstruction pipeline based on monocular geometric cues, resulting in\naccurate, robust and scalable 3D plane detection and reconstruction in the\nwild. Specifically, we first leverage large-scale pre-trained neural networks\nto obtain the depth and surface normals from a single image. These monocular\ngeometric cues are then incorporated into a proximity-guided RANSAC framework\nto sequentially fit each plane instance. We exploit effective 3D point\nproximity and model such proximity via a graph within RANSAC to guide the plane\nfitting from noisy monocular depths, followed by image-level multi-plane joint\noptimization to improve the consistency among all plane instances. We further\ndesign a simple but effective pipeline to extend this single-view solution to\nsparse-view 3D plane reconstruction. Extensive experiments on a list of\ndatasets demonstrate our superior zero-shot generalizability over baselines,\nachieving state-of-the-art plane reconstruction performance in a transferring\nsetting. Our code is available at https://github.com/thuzhaowang/MonoPlane .\n", "date": "2024-11-02", "categories": []}
{"id": "2411.01725", "title": "A Probabilistic Formulation of LiDAR Mapping with Neural Radiance Fields", "abstract": "  In this paper we reexamine the process through which a Neural Radiance Field\n(NeRF) can be trained to produce novel LiDAR views of a scene. Unlike image\napplications where camera pixels integrate light over time, LiDAR pulses arrive\nat specific times. As such, multiple LiDAR returns are possible for any given\ndetector and the classification of these returns is inherently probabilistic.\nApplying a traditional NeRF training routine can result in the network learning\nphantom surfaces in free space between conflicting range measurements, similar\nto how floater aberrations may be produced by an image model. We show that by\nformulating loss as an integral of probability (rather than as an integral of\noptical density) the network can learn multiple peaks for a given ray, allowing\nthe sampling of first, nth, or strongest returns from a single output channel.\nCode is available at https://github.com/mcdermatt/PLINK\n", "date": "2024-11-04", "categories": []}
{"id": "2406.09371", "title": "LRM-Zero: Training Large Reconstruction Models with Synthesized Data", "abstract": "  We present LRM-Zero, a Large Reconstruction Model (LRM) trained entirely on\nsynthesized 3D data, achieving high-quality sparse-view 3D reconstruction. The\ncore of LRM-Zero is our procedural 3D dataset, Zeroverse, which is\nautomatically synthesized from simple primitive shapes with random texturing\nand augmentations (e.g., height fields, boolean differences, and wireframes).\nUnlike previous 3D datasets (e.g., Objaverse) which are often captured or\ncrafted by humans to approximate real 3D data, Zeroverse completely ignores\nrealistic global semantics but is rich in complex geometric and texture details\nthat are locally similar to or even more intricate than real objects. We\ndemonstrate that our LRM-Zero, trained with our fully synthesized Zeroverse,\ncan achieve high visual quality in the reconstruction of real-world objects,\ncompetitive with models trained on Objaverse. We also analyze several critical\ndesign choices of Zeroverse that contribute to LRM-Zero's capability and\ntraining stability. Our work demonstrates that 3D reconstruction, one of the\ncore tasks in 3D vision, can potentially be addressed without the semantics of\nreal-world objects. The Zeroverse's procedural synthesis code and interactive\nvisualization are available at: https://desaixie.github.io/lrm-zero/.\n", "date": "2024-06-13", "categories": []}
{"id": "2403.10242", "title": "GeoGS3D: Single-view 3D Reconstruction via Geometric-aware Diffusion\n  Model and Gaussian Splatting", "abstract": "  We introduce GeoGS3D, a novel two-stage framework for reconstructing detailed\n3D objects from single-view images. Inspired by the success of pre-trained 2D\ndiffusion models, our method incorporates an orthogonal plane decomposition\nmechanism to extract 3D geometric features from the 2D input, facilitating the\ngeneration of multi-view consistent images. During the following Gaussian\nSplatting, these images are fused with epipolar attention, fully utilizing the\ngeometric correlations across views. Moreover, we propose a novel metric,\nGaussian Divergence Significance (GDS), to prune unnecessary operations during\noptimization, significantly accelerating the reconstruction process. Extensive\nexperiments demonstrate that GeoGS3D generates images with high consistency\nacross views and reconstructs high-quality 3D objects, both qualitatively and\nquantitatively.\n", "date": "2024-03-15", "categories": []}
{"id": "2404.16666", "title": "PhyRecon: Physically Plausible Neural Scene Reconstruction", "abstract": "  We address the issue of physical implausibility in multi-view neural\nreconstruction. While implicit representations have gained popularity in\nmulti-view 3D reconstruction, previous work struggles to yield physically\nplausible results, limiting their utility in domains requiring rigorous\nphysical accuracy. This lack of plausibility stems from the absence of physics\nmodeling in existing methods and their inability to recover intricate\ngeometrical structures. In this paper, we introduce PHYRECON, the first\napproach to leverage both differentiable rendering and differentiable physics\nsimulation to learn implicit surface representations. PHYRECON features a novel\ndifferentiable particle-based physical simulator built on neural implicit\nrepresentations. Central to this design is an efficient transformation between\nSDF-based implicit representations and explicit surface points via our proposed\nSurface Points Marching Cubes (SP-MC), enabling differentiable learning with\nboth rendering and physical losses. Additionally, PHYRECON models both\nrendering and physical uncertainty to identify and compensate for inconsistent\nand inaccurate monocular geometric priors. The physical uncertainty further\nfacilitates physics-guided pixel sampling to enhance the learning of slender\nstructures. By integrating these techniques, our model supports differentiable\njoint modeling of appearance, geometry, and physics. Extensive experiments\ndemonstrate that PHYRECON significantly improves the reconstruction quality.\nOur results also exhibit superior physical stability in physical simulators,\nwith at least a 40% improvement across all datasets, paving the way for future\nphysics-based applications.\n", "date": "2024-04-25", "categories": []}
{"id": "2406.12095", "title": "DistillNeRF: Perceiving 3D Scenes from Single-Glance Images by\n  Distilling Neural Fields and Foundation Model Features", "abstract": "  We propose DistillNeRF, a self-supervised learning framework addressing the\nchallenge of understanding 3D environments from limited 2D observations in\noutdoor autonomous driving scenes. Our method is a generalizable feedforward\nmodel that predicts a rich neural scene representation from sparse,\nsingle-frame multi-view camera inputs with limited view overlap, and is trained\nself-supervised with differentiable rendering to reconstruct RGB, depth, or\nfeature images. Our first insight is to exploit per-scene optimized Neural\nRadiance Fields (NeRFs) by generating dense depth and virtual camera targets\nfrom them, which helps our model to learn enhanced 3D geometry from sparse\nnon-overlapping image inputs. Second, to learn a semantically rich 3D\nrepresentation, we propose distilling features from pre-trained 2D foundation\nmodels, such as CLIP or DINOv2, thereby enabling various downstream tasks\nwithout the need for costly 3D human annotations. To leverage these two\ninsights, we introduce a novel model architecture with a two-stage\nlift-splat-shoot encoder and a parameterized sparse hierarchical voxel\nrepresentation. Experimental results on the NuScenes and Waymo NOTR datasets\ndemonstrate that DistillNeRF significantly outperforms existing comparable\nstate-of-the-art self-supervised methods for scene reconstruction, novel view\nsynthesis, and depth estimation; and it allows for competitive zero-shot 3D\nsemantic occupancy prediction, as well as open-world scene understanding\nthrough distilled foundation model features. Demos and code will be available\nat https://distillnerf.github.io/.\n", "date": "2024-06-17", "categories": []}
{"id": "2407.08447", "title": "WildGaussians: 3D Gaussian Splatting in the Wild", "abstract": "  While the field of 3D scene reconstruction is dominated by NeRFs due to their\nphotorealistic quality, 3D Gaussian Splatting (3DGS) has recently emerged,\noffering similar quality with real-time rendering speeds. However, both methods\nprimarily excel with well-controlled 3D scenes, while in-the-wild data -\ncharacterized by occlusions, dynamic objects, and varying illumination -\nremains challenging. NeRFs can adapt to such conditions easily through\nper-image embedding vectors, but 3DGS struggles due to its explicit\nrepresentation and lack of shared parameters. To address this, we introduce\nWildGaussians, a novel approach to handle occlusions and appearance changes\nwith 3DGS. By leveraging robust DINO features and integrating an appearance\nmodeling module within 3DGS, our method achieves state-of-the-art results. We\ndemonstrate that WildGaussians matches the real-time rendering speed of 3DGS\nwhile surpassing both 3DGS and NeRF baselines in handling in-the-wild data, all\nwithin a simple architectural framework.\n", "date": "2024-07-11", "categories": []}
{"id": "2410.14790", "title": "SSL-NBV: A Self-Supervised-Learning-Based Next-Best-View algorithm for\n  Efficient 3D Plant Reconstruction by a Robot", "abstract": "  The 3D reconstruction of plants is challenging due to their complex shape\ncausing many occlusions. Next-Best-View (NBV) methods address this by\niteratively selecting new viewpoints to maximize information gain (IG).\nDeep-learning-based NBV (DL-NBV) methods demonstrate higher computational\nefficiency over classic voxel-based NBV approaches but current methods require\nextensive training using ground-truth plant models, making them impractical for\nreal-world plants. These methods, moreover, rely on offline training with\npre-collected data, limiting adaptability in changing agricultural\nenvironments. This paper proposes a self-supervised learning-based NBV method\n(SSL-NBV) that uses a deep neural network to predict the IG for candidate\nviewpoints. The method allows the robot to gather its own training data during\ntask execution by comparing new 3D sensor data to the earlier gathered data and\nby employing weakly-supervised learning and experience replay for efficient\nonline learning. Comprehensive evaluations were conducted in simulation and\nreal-world environments using cross-validation. The results showed that SSL-NBV\nrequired fewer views for plant reconstruction than non-NBV methods and was over\n800 times faster than a voxel-based method. SSL-NBV reduced training\nannotations by over 90% compared to a baseline DL-NBV. Furthermore, SSL-NBV\ncould adapt to novel scenarios through online fine-tuning. Also using real\nplants, the results showed that the proposed method can learn to effectively\nplan new viewpoints for 3D plant reconstruction. Most importantly, SSL-NBV\nautomated the entire network training and uses continuous online learning,\nallowing it to operate in changing agricultural environments.\n", "date": "2024-10-02", "categories": []}
{"id": "2410.18956", "title": "Large Spatial Model: End-to-end Unposed Images to Semantic 3D", "abstract": "  Reconstructing and understanding 3D structures from a limited number of\nimages is a well-established problem in computer vision. Traditional methods\nusually break this task into multiple subtasks, each requiring complex\ntransformations between different data representations. For instance, dense\nreconstruction through Structure-from-Motion (SfM) involves converting images\ninto key points, optimizing camera parameters, and estimating structures.\nAfterward, accurate sparse reconstructions are required for further dense\nmodeling, which is subsequently fed into task-specific neural networks. This\nmulti-step process results in considerable processing time and increased\nengineering complexity.\n  In this work, we present the Large Spatial Model (LSM), which processes\nunposed RGB images directly into semantic radiance fields. LSM simultaneously\nestimates geometry, appearance, and semantics in a single feed-forward\noperation, and it can generate versatile label maps by interacting with\nlanguage at novel viewpoints. Leveraging a Transformer-based architecture, LSM\nintegrates global geometry through pixel-aligned point maps. To enhance spatial\nattribute regression, we incorporate local context aggregation with multi-scale\nfusion, improving the accuracy of fine local details. To tackle the scarcity of\nlabeled 3D semantic data and enable natural language-driven scene manipulation,\nwe incorporate a pre-trained 2D language-based segmentation model into a\n3D-consistent semantic feature field. An efficient decoder then parameterizes a\nset of semantic anisotropic Gaussians, facilitating supervised end-to-end\nlearning. Extensive experiments across various tasks show that LSM unifies\nmultiple 3D vision tasks directly from unposed images, achieving real-time\nsemantic 3D reconstruction for the first time.\n", "date": "2024-10-24", "categories": []}
{"id": "2410.23522", "title": "LBurst: Learning-Based Robotic Burst Feature Extraction for 3D\n  Reconstruction in Low Light", "abstract": "  Drones have revolutionized the fields of aerial imaging, mapping, and\ndisaster recovery. However, the deployment of drones in low-light conditions is\nconstrained by the image quality produced by their on-board cameras. In this\npaper, we present a learning architecture for improving 3D reconstructions in\nlow-light conditions by finding features in a burst. Our approach enhances\nvisual reconstruction by detecting and describing high quality true features\nand less spurious features in low signal-to-noise ratio images. We demonstrate\nthat our method is capable of handling challenging scenes in millilux\nillumination, making it a significant step towards drones operating at night\nand in extremely low-light applications such as underground mining and search\nand rescue operations.\n", "date": "2024-10-31", "categories": []}
{"id": "2410.24207", "title": "No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse\n  Unposed Images", "abstract": "  We introduce NoPoSplat, a feed-forward model capable of reconstructing 3D\nscenes parameterized by 3D Gaussians from \\textit{unposed} sparse multi-view\nimages. Our model, trained exclusively with photometric loss, achieves\nreal-time 3D Gaussian reconstruction during inference. To eliminate the need\nfor accurate pose input during reconstruction, we anchor one input view's local\ncamera coordinates as the canonical space and train the network to predict\nGaussian primitives for all views within this space. This approach obviates the\nneed to transform Gaussian primitives from local coordinates into a global\ncoordinate system, thus avoiding errors associated with per-frame Gaussians and\npose estimation. To resolve scale ambiguity, we design and compare various\nintrinsic embedding methods, ultimately opting to convert camera intrinsics\ninto a token embedding and concatenate it with image tokens as input to the\nmodel, enabling accurate scene scale prediction. We utilize the reconstructed\n3D Gaussians for novel view synthesis and pose estimation tasks and propose a\ntwo-stage coarse-to-fine pipeline for accurate pose estimation. Experimental\nresults demonstrate that our pose-free approach can achieve superior novel view\nsynthesis quality compared to pose-required methods, particularly in scenarios\nwith limited input image overlap. For pose estimation, our method, trained\nwithout ground truth depth or explicit matching loss, significantly outperforms\nthe state-of-the-art methods with substantial improvements. This work makes\nsignificant advances in pose-free generalizable 3D reconstruction and\ndemonstrates its applicability to real-world scenarios. Code and trained models\nare available at https://noposplat.github.io/.\n", "date": "2024-10-31", "categories": []}
{"id": "2410.06613", "title": "ES-Gaussian: Gaussian Splatting Mapping via Error Space-Based Gaussian\n  Completion", "abstract": "  Accurate and affordable indoor 3D reconstruction is critical for effective\nrobot navigation and interaction. Traditional LiDAR-based mapping provides high\nprecision but is costly, heavy, and power-intensive, with limited ability for\nnovel view rendering. Vision-based mapping, while cost-effective and capable of\ncapturing visual data, often struggles with high-quality 3D reconstruction due\nto sparse point clouds. We propose ES-Gaussian, an end-to-end system using a\nlow-altitude camera and single-line LiDAR for high-quality 3D indoor\nreconstruction. Our system features Visual Error Construction (VEC) to enhance\nsparse point clouds by identifying and correcting areas with insufficient\ngeometric detail from 2D error maps. Additionally, we introduce a novel 3DGS\ninitialization method guided by single-line LiDAR, overcoming the limitations\nof traditional multi-view setups and enabling effective reconstruction in\nresource-constrained environments. Extensive experimental results on our new\nDreame-SR dataset and a publicly available dataset demonstrate that ES-Gaussian\noutperforms existing methods, particularly in challenging scenarios. The\nproject page is available at https://chenlu-china.github.io/ES-Gaussian/.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.22383", "title": "Exploiting Semantic Scene Reconstruction for Estimating Building\n  Envelope Characteristics", "abstract": "  Achieving the EU's climate neutrality goal requires retrofitting existing\nbuildings to reduce energy use and emissions. A critical step in this process\nis the precise assessment of geometric building envelope characteristics to\ninform retrofitting decisions. Previous methods for estimating building\ncharacteristics, such as window-to-wall ratio, building footprint area, and the\nlocation of architectural elements, have primarily relied on applying\ndeep-learning-based detection or segmentation techniques on 2D images. However,\nthese approaches tend to focus on planar facade properties, limiting their\naccuracy and comprehensiveness when analyzing complete building envelopes in\n3D.\n  While neural scene representations have shown exceptional performance in\nindoor scene reconstruction, they remain under-explored for external building\nenvelope analysis. This work addresses this gap by leveraging cutting-edge\nneural surface reconstruction techniques based on signed distance function\n(SDF) representations for 3D building analysis. We propose BuildNet3D, a novel\nframework to estimate geometric building characteristics from 2D image inputs.\nBy integrating SDF-based representation with semantic modality, BuildNet3D\nrecovers fine-grained 3D geometry and semantics of building envelopes, which\nare then used to automatically extract building characteristics. Our framework\nis evaluated on a range of complex building structures, demonstrating high\naccuracy and generalizability in estimating window-to-wall ratio and building\nfootprint. The results underscore the effectiveness of BuildNet3D for practical\napplications in building analysis and retrofitting.\n", "date": "2024-10-29", "categories": []}
{"id": "2410.23213", "title": "ELMGS: Enhancing memory and computation scaLability through coMpression\n  for 3D Gaussian Splatting", "abstract": "  3D models have recently been popularized by the potentiality of end-to-end\ntraining offered first by Neural Radiance Fields and most recently by 3D\nGaussian Splatting models. The latter has the big advantage of naturally\nproviding fast training convergence and high editability. However, as the\nresearch around these is still in its infancy, there is still a gap in the\nliterature regarding the model's scalability. In this work, we propose an\napproach enabling both memory and computation scalability of such models. More\nspecifically, we propose an iterative pruning strategy that removes redundant\ninformation encoded in the model. We also enhance compressibility for the model\nby including in the optimization strategy a differentiable quantization and\nentropy coding estimator. Our results on popular benchmarks showcase the\neffectiveness of the proposed approach and open the road to the broad\ndeployability of such a solution even on resource-constrained devices.\n", "date": "2024-10-30", "categories": []}
{"id": "2407.04237", "title": "GSD: View-Guided Gaussian Splatting Diffusion for 3D Reconstruction", "abstract": "  We present GSD, a diffusion model approach based on Gaussian Splatting (GS)\nrepresentation for 3D object reconstruction from a single view. Prior works\nsuffer from inconsistent 3D geometry or mediocre rendering quality due to\nimproper representations. We take a step towards resolving these shortcomings\nby utilizing the recent state-of-the-art 3D explicit representation, Gaussian\nSplatting, and an unconditional diffusion model. This model learns to generate\n3D objects represented by sets of GS ellipsoids. With these strong generative\n3D priors, though learning unconditionally, the diffusion model is ready for\nview-guided reconstruction without further model fine-tuning. This is achieved\nby propagating fine-grained 2D features through the efficient yet flexible\nsplatting function and the guided denoising sampling process. In addition, a 2D\ndiffusion model is further employed to enhance rendering fidelity, and improve\nreconstructed GS quality by polishing and re-using the rendered images. The\nfinal reconstructed objects explicitly come with high-quality 3D structure and\ntexture, and can be efficiently rendered in arbitrary views. Experiments on the\nchallenging real-world CO3D dataset demonstrate the superiority of our\napproach. Project page: https://yxmu.foo/GSD/\n", "date": "2024-07-05", "categories": []}
{"id": "2403.14627", "title": "MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images", "abstract": "  We introduce MVSplat, an efficient model that, given sparse multi-view images\nas input, predicts clean feed-forward 3D Gaussians. To accurately localize the\nGaussian centers, we build a cost volume representation via plane sweeping,\nwhere the cross-view feature similarities stored in the cost volume can provide\nvaluable geometry cues to the estimation of depth. We also learn other Gaussian\nprimitives' parameters jointly with the Gaussian centers while only relying on\nphotometric supervision. We demonstrate the importance of the cost volume\nrepresentation in learning feed-forward Gaussians via extensive experimental\nevaluations. On the large-scale RealEstate10K and ACID benchmarks, MVSplat\nachieves state-of-the-art performance with the fastest feed-forward inference\nspeed (22~fps). More impressively, compared to the latest state-of-the-art\nmethod pixelSplat, MVSplat uses $10\\times$ fewer parameters and infers more\nthan $2\\times$ faster while providing higher appearance and geometry quality as\nwell as better cross-dataset generalization.\n", "date": "2024-03-21", "categories": []}
{"id": "2406.15333", "title": "GeoLRM: Geometry-Aware Large Reconstruction Model for High-Quality 3D\n  Gaussian Generation", "abstract": "  In this work, we introduce the Geometry-Aware Large Reconstruction Model\n(GeoLRM), an approach which can predict high-quality assets with 512k Gaussians\nand 21 input images in only 11 GB GPU memory. Previous works neglect the\ninherent sparsity of 3D structure and do not utilize explicit geometric\nrelationships between 3D and 2D images. This limits these methods to a\nlow-resolution representation and makes it difficult to scale up to the dense\nviews for better quality. GeoLRM tackles these issues by incorporating a novel\n3D-aware transformer structure that directly processes 3D points and uses\ndeformable cross-attention mechanisms to effectively integrate image features\ninto 3D representations. We implement this solution through a two-stage\npipeline: initially, a lightweight proposal network generates a sparse set of\n3D anchor points from the posed image inputs; subsequently, a specialized\nreconstruction transformer refines the geometry and retrieves textural details.\nExtensive experimental results demonstrate that GeoLRM significantly\noutperforms existing models, especially for dense view inputs. We also\ndemonstrate the practical applicability of our model with 3D generation tasks,\nshowcasing its versatility and potential for broader adoption in real-world\napplications. The project page: https://linshan-bin.github.io/GeoLRM/.\n", "date": "2024-06-21", "categories": []}
{"id": "2410.20030", "title": "SCube: Instant Large-Scale Scene Reconstruction using VoxSplats", "abstract": "  We present SCube, a novel method for reconstructing large-scale 3D scenes\n(geometry, appearance, and semantics) from a sparse set of posed images. Our\nmethod encodes reconstructed scenes using a novel representation VoxSplat,\nwhich is a set of 3D Gaussians supported on a high-resolution sparse-voxel\nscaffold. To reconstruct a VoxSplat from images, we employ a hierarchical voxel\nlatent diffusion model conditioned on the input images followed by a\nfeedforward appearance prediction model. The diffusion model generates\nhigh-resolution grids progressively in a coarse-to-fine manner, and the\nappearance network predicts a set of Gaussians within each voxel. From as few\nas 3 non-overlapping input images, SCube can generate millions of Gaussians\nwith a 1024^3 voxel grid spanning hundreds of meters in 20 seconds. Past works\ntackling scene reconstruction from images either rely on per-scene optimization\nand fail to reconstruct the scene away from input views (thus requiring dense\nview coverage as input) or leverage geometric priors based on low-resolution\nmodels, which produce blurry results. In contrast, SCube leverages\nhigh-resolution sparse networks and produces sharp outputs from few views. We\nshow the superiority of SCube compared to prior art using the Waymo\nself-driving dataset on 3D reconstruction and demonstrate its applications,\nsuch as LiDAR simulation and text-to-scene generation.\n", "date": "2024-10-26", "categories": []}
{"id": "2410.20686", "title": "ODGS: 3D Scene Reconstruction from Omnidirectional Images with 3D\n  Gaussian Splattings", "abstract": "  Omnidirectional (or 360-degree) images are increasingly being used for 3D\napplications since they allow the rendering of an entire scene with a single\nimage. Existing works based on neural radiance fields demonstrate successful 3D\nreconstruction quality on egocentric videos, yet they suffer from long training\nand rendering times. Recently, 3D Gaussian splatting has gained attention for\nits fast optimization and real-time rendering. However, directly using a\nperspective rasterizer to omnidirectional images results in severe distortion\ndue to the different optical properties between two image domains. In this\nwork, we present ODGS, a novel rasterization pipeline for omnidirectional\nimages, with geometric interpretation. For each Gaussian, we define a tangent\nplane that touches the unit sphere and is perpendicular to the ray headed\ntoward the Gaussian center. We then leverage a perspective camera rasterizer to\nproject the Gaussian onto the corresponding tangent plane. The projected\nGaussians are transformed and combined into the omnidirectional image,\nfinalizing the omnidirectional rasterization process. This interpretation\nreveals the implicit assumptions within the proposed pipeline, which we verify\nthrough mathematical proofs. The entire rasterization process is parallelized\nusing CUDA, achieving optimization and rendering speeds 100 times faster than\nNeRF-based methods. Our comprehensive experiments highlight the superiority of\nODGS by delivering the best reconstruction and perceptual quality across\nvarious datasets. Additionally, results on roaming datasets demonstrate that\nODGS restores fine details effectively, even when reconstructing large 3D\nscenes. The source code is available on our project page\n(https://github.com/esw0116/ODGS).\n", "date": "2024-10-28", "categories": []}
{"id": "2404.09426", "title": "ViFu: Multiple 360$^\\circ$ Objects Reconstruction with Clean Background\n  via Visible Part Fusion", "abstract": "  In this paper, we propose a method to segment and recover a static, clean\nbackground and multiple 360$^\\circ$ objects from observations of scenes at\ndifferent timestamps. Recent works have used neural radiance fields to model 3D\nscenes and improved the quality of novel view synthesis, while few studies have\nfocused on modeling the invisible or occluded parts of the training images.\nThese under-reconstruction parts constrain both scene editing and rendering\nview selection, thereby limiting their utility for synthetic data generation\nfor downstream tasks. Our basic idea is that, by observing the same set of\nobjects in various arrangement, so that parts that are invisible in one scene\nmay become visible in others. By fusing the visible parts from each scene,\nocclusion-free rendering of both background and foreground objects can be\nachieved.\n  We decompose the multi-scene fusion task into two main components: (1)\nobjects/background segmentation and alignment, where we leverage point\ncloud-based methods tailored to our novel problem formulation; (2) radiance\nfields fusion, where we introduce visibility field to quantify the visible\ninformation of radiance fields, and propose visibility-aware rendering for the\nfusion of series of scenes, ultimately obtaining clean background and\n360$^\\circ$ object rendering. Comprehensive experiments were conducted on\nsynthetic and real datasets, and the results demonstrate the effectiveness of\nour method.\n", "date": "2024-04-15", "categories": []}
{"id": "2410.18433", "title": "Segmentation-aware Prior Assisted Joint Global Information Aggregated 3D\n  Building Reconstruction", "abstract": "  Multi-View Stereo plays a pivotal role in civil engineering by facilitating\n3D modeling, precise engineering surveying, quantitative analysis, as well as\nmonitoring and maintenance. It serves as a valuable tool, offering\nhigh-precision and real-time spatial information crucial for various\nengineering projects. However, Multi-View Stereo algorithms encounter\nchallenges in reconstructing weakly-textured regions within large-scale\nbuilding scenes. In these areas, the stereo matching of pixels often fails,\nleading to inaccurate depth estimations. Based on the Segment Anything Model\nand RANSAC algorithm, we propose an algorithm that accurately segments\nweakly-textured regions and constructs their plane priors. These plane priors,\ncombined with triangulation priors, form a reliable prior candidate set.\nAdditionally, we introduce a novel global information aggregation cost\nfunction. This function selects optimal plane prior information based on global\ninformation in the prior candidate set, constrained by geometric consistency\nduring the depth estimation update process. Experimental results on both the\nETH3D benchmark dataset, aerial dataset, building dataset and real scenarios\nsubstantiate the superior performance of our method in producing 3D building\nmodels compared to other state-of-the-art methods. In summary, our work aims to\nenhance the completeness and density of 3D building reconstruction, carrying\nimplications for broader applications in urban planning and virtual reality.\n", "date": "2024-10-24", "categories": []}
{"id": "2410.18979", "title": "PixelGaussian: Generalizable 3D Gaussian Reconstruction from Arbitrary\n  Views", "abstract": "  We propose PixelGaussian, an efficient feed-forward framework for learning\ngeneralizable 3D Gaussian reconstruction from arbitrary views. Most existing\nmethods rely on uniform pixel-wise Gaussian representations, which learn a\nfixed number of 3D Gaussians for each view and cannot generalize well to more\ninput views. Differently, our PixelGaussian dynamically adapts both the\nGaussian distribution and quantity based on geometric complexity, leading to\nmore efficient representations and significant improvements in reconstruction\nquality. Specifically, we introduce a Cascade Gaussian Adapter to adjust\nGaussian distribution according to local geometry complexity identified by a\nkeypoint scorer. CGA leverages deformable attention in context-aware\nhypernetworks to guide Gaussian pruning and splitting, ensuring accurate\nrepresentation in complex regions while reducing redundancy. Furthermore, we\ndesign a transformer-based Iterative Gaussian Refiner module that refines\nGaussian representations through direct image-Gaussian interactions. Our\nPixelGaussian can effectively reduce Gaussian redundancy as input views\nincrease. We conduct extensive experiments on the large-scale ACID and\nRealEstate10K datasets, where our method achieves state-of-the-art performance\nwith good generalization to various numbers of views. Code:\nhttps://github.com/Barrybarry-Smith/PixelGaussian.\n", "date": "2024-10-24", "categories": []}
{"id": "2410.16892", "title": "VistaDream: Sampling multiview consistent images for single-view scene\n  reconstruction", "abstract": "  In this paper, we propose VistaDream a novel framework to reconstruct a 3D\nscene from a single-view image. Recent diffusion models enable generating\nhigh-quality novel-view images from a single-view input image. Most existing\nmethods only concentrate on building the consistency between the input image\nand the generated images while losing the consistency between the generated\nimages. VistaDream addresses this problem by a two-stage pipeline. In the first\nstage, VistaDream begins with building a global coarse 3D scaffold by zooming\nout a little step with inpainted boundaries and an estimated depth map. Then,\non this global scaffold, we use iterative diffusion-based RGB-D inpainting to\ngenerate novel-view images to inpaint the holes of the scaffold. In the second\nstage, we further enhance the consistency between the generated novel-view\nimages by a novel training-free Multiview Consistency Sampling (MCS) that\nintroduces multi-view consistency constraints in the reverse sampling process\nof diffusion models. Experimental results demonstrate that without training or\nfine-tuning existing diffusion models, VistaDream achieves consistent and\nhigh-quality novel view synthesis using just single-view images and outperforms\nbaseline methods by a large margin. The code, videos, and interactive demos are\navailable at https://vistadream-project-page.github.io/.\n", "date": "2024-10-22", "categories": []}
{"id": "2410.04354", "title": "StreetSurfGS: Scalable Urban Street Surface Reconstruction with\n  Planar-based Gaussian Splatting", "abstract": "  Reconstructing urban street scenes is crucial due to its vital role in\napplications such as autonomous driving and urban planning. These scenes are\ncharacterized by long and narrow camera trajectories, occlusion, complex object\nrelationships, and data sparsity across multiple scales. Despite recent\nadvancements, existing surface reconstruction methods, which are primarily\ndesigned for object-centric scenarios, struggle to adapt effectively to the\nunique characteristics of street scenes. To address this challenge, we\nintroduce StreetSurfGS, the first method to employ Gaussian Splatting\nspecifically tailored for scalable urban street scene surface reconstruction.\nStreetSurfGS utilizes a planar-based octree representation and segmented\ntraining to reduce memory costs, accommodate unique camera characteristics, and\nensure scalability. Additionally, to mitigate depth inaccuracies caused by\nobject overlap, we propose a guided smoothing strategy within regularization to\neliminate inaccurate boundary points and outliers. Furthermore, to address\nsparse views and multi-scale challenges, we use a dual-step matching strategy\nthat leverages adjacent and long-term information. Extensive experiments\nvalidate the efficacy of StreetSurfGS in both novel view synthesis and surface\nreconstruction.\n", "date": "2024-10-06", "categories": []}
{"id": "2410.11703", "title": "Robotic Arm Platform for Multi-View Image Acquisition and 3D\n  Reconstruction in Minimally Invasive Surgery", "abstract": "  Minimally invasive surgery (MIS) offers significant benefits such as reduced\nrecovery time and minimised patient trauma, but poses challenges in visibility\nand access, making accurate 3D reconstruction a significant tool in surgical\nplanning and navigation. This work introduces a robotic arm platform for\nefficient multi-view image acquisition and precise 3D reconstruction in MIS\nsettings. We adapted a laparoscope to a robotic arm and captured ex-vivo images\nof several ovine organs across varying lighting conditions (operating room and\nlaparoscopic) and trajectories (spherical and laparoscopic). We employed\nrecently released learning-based feature matchers combined with COLMAP to\nproduce our reconstructions. The reconstructions were evaluated against\nhigh-precision laser scans for quantitative evaluation. Our results show that\nwhilst reconstructions suffer most under realistic MIS lighting and trajectory,\nmany versions of our pipeline achieve close to sub-millimetre accuracy with an\naverage of 1.05 mm Root Mean Squared Error and 0.82 mm Chamfer distance. Our\nbest reconstruction results occur with operating room lighting and spherical\ntrajectories. Our robotic platform provides a tool for controlled, repeatable\nmulti-view data acquisition for 3D generation in MIS environments which we hope\nleads to new datasets for training learning-based models.\n", "date": "2024-10-15", "categories": []}
{"id": "2410.15376", "title": "ActiveNeuS: Neural Signed Distance Fields for Active Stereo", "abstract": "  3D-shape reconstruction in extreme environments, such as low illumination or\nscattering condition, has been an open problem and intensively researched.\nActive stereo is one of potential solution for such environments for its\nrobustness and high accuracy. However, active stereo systems usually consist of\nspecialized system configurations with complicated algorithms, which narrow\ntheir application. In this paper, we propose Neural Signed Distance Field for\nactive stereo systems to enable implicit correspondence search and\ntriangulation in generalized Structured Light. With our technique, textureless\nor equivalent surfaces by low light condition are successfully reconstructed\neven with a small number of captured images. Experiments were conducted to\nconfirm that the proposed method could achieve state-of-the-art reconstruction\nquality under such severe condition. We also demonstrated that the proposed\nmethod worked in an underwater scenario.\n", "date": "2024-10-20", "categories": []}
{"id": "2410.15971", "title": "Zero-Shot Scene Reconstruction from Single Images with Deep Prior\n  Assembly", "abstract": "  Large language and vision models have been leading a revolution in visual\ncomputing. By greatly scaling up sizes of data and model parameters, the large\nmodels learn deep priors which lead to remarkable performance in various tasks.\nIn this work, we present deep prior assembly, a novel framework that assembles\ndiverse deep priors from large models for scene reconstruction from single\nimages in a zero-shot manner. We show that this challenging task can be done\nwithout extra knowledge but just simply generalizing one deep prior in one\nsub-task. To this end, we introduce novel methods related to poses, scales, and\nocclusion parsing which are keys to enable deep priors to work together in a\nrobust way. Deep prior assembly does not require any 3D or 2D data-driven\ntraining in the task and demonstrates superior performance in generalizing\npriors to open-world scenes. We conduct evaluations on various datasets, and\nreport analysis, numerical and visual comparisons with the latest methods to\nshow our superiority. Project page:\nhttps://junshengzhou.github.io/DeepPriorAssembly.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.16266", "title": "3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with\n  View-consistent 2D Diffusion Priors", "abstract": "  Novel-view synthesis aims to generate novel views of a scene from multiple\ninput images or videos, and recent advancements like 3D Gaussian splatting\n(3DGS) have achieved notable success in producing photorealistic renderings\nwith efficient pipelines. However, generating high-quality novel views under\nchallenging settings, such as sparse input views, remains difficult due to\ninsufficient information in under-sampled areas, often resulting in noticeable\nartifacts. This paper presents 3DGS-Enhancer, a novel pipeline for enhancing\nthe representation quality of 3DGS representations. We leverage 2D video\ndiffusion priors to address the challenging 3D view consistency problem,\nreformulating it as achieving temporal consistency within a video generation\nprocess. 3DGS-Enhancer restores view-consistent latent features of rendered\nnovel views and integrates them with the input views through a spatial-temporal\ndecoder. The enhanced views are then used to fine-tune the initial 3DGS model,\nsignificantly improving its rendering performance. Extensive experiments on\nlarge-scale datasets of unbounded scenes demonstrate that 3DGS-Enhancer yields\nsuperior reconstruction performance and high-fidelity rendering results\ncompared to state-of-the-art methods. The project webpage is\nhttps://xiliu8006.github.io/3DGS-Enhancer-project .\n", "date": "2024-10-21", "categories": []}
{"id": "2410.14169", "title": "DaRePlane: Direction-aware Representations for Dynamic Scene\n  Reconstruction", "abstract": "  Numerous recent approaches to modeling and re-rendering dynamic scenes\nleverage plane-based explicit representations, addressing slow training times\nassociated with models like neural radiance fields (NeRF) and Gaussian\nsplatting (GS). However, merely decomposing 4D dynamic scenes into multiple 2D\nplane-based representations is insufficient for high-fidelity re-rendering of\nscenes with complex motions. In response, we present DaRePlane, a novel\ndirection-aware representation approach that captures scene dynamics from six\ndifferent directions. This learned representation undergoes an inverse\ndual-tree complex wavelet transformation (DTCWT) to recover plane-based\ninformation. Within NeRF pipelines, DaRePlane computes features for each\nspace-time point by fusing vectors from these recovered planes, then passed to\na tiny MLP for color regression. When applied to Gaussian splatting, DaRePlane\ncomputes the features of Gaussian points, followed by a tiny multi-head MLP for\nspatial-time deformation prediction. Notably, to address redundancy introduced\nby the six real and six imaginary direction-aware wavelet coefficients, we\nintroduce a trainable masking approach, mitigating storage issues without\nsignificant performance decline. To demonstrate the generality and efficiency\nof DaRePlane, we test it on both regular and surgical dynamic scenes, for both\nNeRF and GS systems. Extensive experiments show that DaRePlane yields\nstate-of-the-art performance in novel view synthesis for various complex\ndynamic scenes.\n", "date": "2024-10-18", "categories": []}
{"id": "2410.14189", "title": "Neural Signed Distance Function Inference through Splatting 3D Gaussians\n  Pulled on Zero-Level Set", "abstract": "  It is vital to infer a signed distance function (SDF) in multi-view based\nsurface reconstruction. 3D Gaussian splatting (3DGS) provides a novel\nperspective for volume rendering, and shows advantages in rendering efficiency\nand quality. Although 3DGS provides a promising neural rendering option, it is\nstill hard to infer SDFs for surface reconstruction with 3DGS due to the\ndiscreteness, the sparseness, and the off-surface drift of 3D Gaussians. To\nresolve these issues, we propose a method that seamlessly merge 3DGS with the\nlearning of neural SDFs. Our key idea is to more effectively constrain the SDF\ninference with the multi-view consistency. To this end, we dynamically align 3D\nGaussians on the zero-level set of the neural SDF using neural pulling, and\nthen render the aligned 3D Gaussians through the differentiable rasterization.\nMeanwhile, we update the neural SDF by pulling neighboring space to the pulled\n3D Gaussians, which progressively refine the signed distance field near the\nsurface. With both differentiable pulling and splatting, we jointly optimize 3D\nGaussians and the neural SDF with both RGB and geometry constraints, which\nrecovers more accurate, smooth, and complete surfaces with more geometry\ndetails. Our numerical and visual comparisons show our superiority over the\nstate-of-the-art results on the widely used benchmarks.\n", "date": "2024-10-18", "categories": []}
{"id": "2410.13349", "title": "GlossyGS: Inverse Rendering of Glossy Objects with 3D Gaussian Splatting", "abstract": "  Reconstructing objects from posed images is a crucial and complex task in\ncomputer graphics and computer vision. While NeRF-based neural reconstruction\nmethods have exhibited impressive reconstruction ability, they tend to be\ntime-comsuming. Recent strategies have adopted 3D Gaussian Splatting (3D-GS)\nfor inverse rendering, which have led to quick and effective outcomes. However,\nthese techniques generally have difficulty in producing believable geometries\nand materials for glossy objects, a challenge that stems from the inherent\nambiguities of inverse rendering. To address this, we introduce GlossyGS, an\ninnovative 3D-GS-based inverse rendering framework that aims to precisely\nreconstruct the geometry and materials of glossy objects by integrating\nmaterial priors. The key idea is the use of micro-facet geometry segmentation\nprior, which helps to reduce the intrinsic ambiguities and improve the\ndecomposition of geometries and materials. Additionally, we introduce a normal\nmap prefiltering strategy to more accurately simulate the normal distribution\nof reflective surfaces. These strategies are integrated into a hybrid geometry\nand material representation that employs both explicit and implicit methods to\ndepict glossy objects. We demonstrate through quantitative analysis and\nqualitative visualization that the proposed method is effective to reconstruct\nhigh-fidelity geometries and materials of glossy objects, and performs\nfavorably against state-of-the-arts.\n", "date": "2024-10-17", "categories": []}
{"id": "2410.12725", "title": "Optimizing 3D Geometry Reconstruction from Implicit Neural\n  Representations", "abstract": "  Implicit neural representations have emerged as a powerful tool in learning\n3D geometry, offering unparalleled advantages over conventional representations\nlike mesh-based methods. A common type of INR implicitly encodes a shape's\nboundary as the zero-level set of the learned continuous function and learns a\nmapping from a low-dimensional latent space to the space of all possible shapes\nrepresented by its signed distance function. However, most INRs struggle to\nretain high-frequency details, which are crucial for accurate geometric\ndepiction, and they are computationally expensive. To address these\nlimitations, we present a novel approach that both reduces computational\nexpenses and enhances the capture of fine details. Our method integrates\nperiodic activation functions, positional encodings, and normals into the\nneural network architecture. This integration significantly enhances the\nmodel's ability to learn the entire space of 3D shapes while preserving\nintricate details and sharp features, areas where conventional representations\noften fall short.\n", "date": "2024-10-16", "categories": []}
{"id": "2410.10927", "title": "Cultural Heritage 3D Reconstruction with Diffusion Networks", "abstract": "  This article explores the use of recent generative AI algorithms for\nrepairing cultural heritage objects, leveraging a conditional diffusion model\ndesigned to reconstruct 3D point clouds effectively. Our study evaluates the\nmodel's performance across general and cultural heritage-specific settings.\nResults indicate that, with considerations for object variability, the\ndiffusion model can accurately reproduce cultural heritage geometries. Despite\nencountering challenges like data diversity and outlier sensitivity, the model\ndemonstrates significant potential in artifact restoration research. This work\nlays groundwork for advancing restoration methodologies for ancient artifacts\nusing AI technologies.\n", "date": "2024-10-14", "categories": []}
{"id": "2410.11285", "title": "Scalable Indoor Novel-View Synthesis using Drone-Captured 360 Imagery\n  with 3D Gaussian Splatting", "abstract": "  Scene reconstruction and novel-view synthesis for large, complex,\nmulti-story, indoor scenes is a challenging and time-consuming task. Prior\nmethods have utilized drones for data capture and radiance fields for scene\nreconstruction, both of which present certain challenges. First, in order to\ncapture diverse viewpoints with the drone's front-facing camera, some\napproaches fly the drone in an unstable zig-zag fashion, which hinders\ndrone-piloting and generates motion blur in the captured data. Secondly, most\nradiance field methods do not easily scale to arbitrarily large number of\nimages. This paper proposes an efficient and scalable pipeline for indoor\nnovel-view synthesis from drone-captured 360 videos using 3D Gaussian\nSplatting. 360 cameras capture a wide set of viewpoints, allowing for\ncomprehensive scene capture under a simple straightforward drone trajectory. To\nscale our method to large scenes, we devise a divide-and-conquer strategy to\nautomatically split the scene into smaller blocks that can be reconstructed\nindividually and in parallel. We also propose a coarse-to-fine alignment\nstrategy to seamlessly match these blocks together to compose the entire scene.\nOur experiments demonstrate marked improvement in both reconstruction quality,\ni.e. PSNR and SSIM, and computation time compared to prior approaches.\n", "date": "2024-10-15", "categories": []}
{"id": "2307.08636", "title": "PolyGNN: Polyhedron-based Graph Neural Network for 3D Building\n  Reconstruction from Point Clouds", "abstract": "  We present PolyGNN, a polyhedron-based graph neural network for 3D building\nreconstruction from point clouds. PolyGNN learns to assemble primitives\nobtained by polyhedral decomposition via graph node classification, achieving a\nwatertight and compact reconstruction. To effectively represent\narbitrary-shaped polyhedra in the neural network, we propose a skeleton-based\nsampling strategy to generate polyhedron-wise queries. These queries are then\nincorporated with inter-polyhedron adjacency to enhance the classification.\nPolyGNN is end-to-end optimizable and is designed to accommodate variable-size\ninput points, polyhedra, and queries with an index-driven batching technique.\nTo address the abstraction gap between existing city-building models and the\nunderlying instances, and provide a fair evaluation of the proposed method, we\ndevelop our method on a large-scale synthetic dataset with well-defined ground\ntruths of polyhedral labels. We further conduct a transferability analysis\nacross cities and on real-world point clouds. Both qualitative and quantitative\nresults demonstrate the effectiveness of our method, particularly its\nefficiency for large-scale reconstructions. The source code and data are\navailable at https://github.com/chenzhaiyu/polygnn.\n", "date": "2023-07-17", "categories": []}
{"id": "2403.16964", "title": "GSDF: 3DGS Meets SDF for Improved Rendering and Reconstruction", "abstract": "  Presenting a 3D scene from multiview images remains a core and long-standing\nchallenge in computer vision and computer graphics. Two main requirements lie\nin rendering and reconstruction. Notably, SOTA rendering quality is usually\nachieved with neural volumetric rendering techniques, which rely on aggregated\npoint/primitive-wise color and neglect the underlying scene geometry. Learning\nof neural implicit surfaces is sparked from the success of neural rendering.\nCurrent works either constrain the distribution of density fields or the shape\nof primitives, resulting in degraded rendering quality and flaws on the learned\nscene surfaces. The efficacy of such methods is limited by the inherent\nconstraints of the chosen neural representation, which struggles to capture\nfine surface details, especially for larger, more intricate scenes. To address\nthese issues, we introduce GSDF, a novel dual-branch architecture that combines\nthe benefits of a flexible and efficient 3D Gaussian Splatting (3DGS)\nrepresentation with neural Signed Distance Fields (SDF). The core idea is to\nleverage and enhance the strengths of each branch while alleviating their\nlimitation through mutual guidance and joint supervision. We show on diverse\nscenes that our design unlocks the potential for more accurate and detailed\nsurface reconstructions, and at the meantime benefits 3DGS rendering with\nstructures that are more aligned with the underlying geometry.\n", "date": "2024-03-25", "categories": []}
{"id": "2406.02972", "title": "Event3DGS: Event-Based 3D Gaussian Splatting for High-Speed Robot\n  Egomotion", "abstract": "  By combining differentiable rendering with explicit point-based scene\nrepresentations, 3D Gaussian Splatting (3DGS) has demonstrated breakthrough 3D\nreconstruction capabilities. However, to date 3DGS has had limited impact on\nrobotics, where high-speed egomotion is pervasive: Egomotion introduces motion\nblur and leads to artifacts in existing frame-based 3DGS reconstruction\nmethods. To address this challenge, we introduce Event3DGS, an {\\em\nevent-based} 3DGS framework. By exploiting the exceptional temporal resolution\nof event cameras, Event3GDS can reconstruct high-fidelity 3D structure and\nappearance under high-speed egomotion. Extensive experiments on multiple\nsynthetic and real-world datasets demonstrate the superiority of Event3DGS\ncompared with existing event-based dense 3D scene reconstruction frameworks;\nEvent3DGS substantially improves reconstruction quality (+3dB) while reducing\ncomputational costs by 95\\%. Our framework also allows one to incorporate a few\nmotion-blurred frame-based measurements into the reconstruction process to\nfurther improve appearance fidelity without loss of structural accuracy.\n", "date": "2024-06-05", "categories": []}
{"id": "2410.09292", "title": "SurgicalGS: Dynamic 3D Gaussian Splatting for Accurate Robotic-Assisted\n  Surgical Scene Reconstruction", "abstract": "  Accurate 3D reconstruction of dynamic surgical scenes from endoscopic video\nis essential for robotic-assisted surgery. While recent 3D Gaussian Splatting\nmethods have shown promise in achieving high-quality reconstructions with fast\nrendering speeds, their use of inverse depth loss functions compresses depth\nvariations. This can lead to a loss of fine geometric details, limiting their\nability to capture precise 3D geometry and effectiveness in intraoperative\napplication. To address these challenges, we present SurgicalGS, a dynamic 3D\nGaussian Splatting framework specifically designed for surgical scene\nreconstruction with improved geometric accuracy. Our approach first initialises\na Gaussian point cloud using depth priors, employing binary motion masks to\nidentify pixels with significant depth variations and fusing point clouds from\ndepth maps across frames for initialisation. We use the Flexible Deformation\nModel to represent dynamic scene and introduce a normalised depth\nregularisation loss along with an unsupervised depth smoothness constraint to\nensure more accurate geometric reconstruction. Extensive experiments on two\nreal surgical datasets demonstrate that SurgicalGS achieves state-of-the-art\nreconstruction quality, especially in terms of accurate geometry, advancing the\nusability of 3D Gaussian Splatting in robotic-assisted surgery.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.07418", "title": "NeRF-Accelerated Ecological Monitoring in Mixed-Evergreen Redwood Forest", "abstract": "  Forest mapping provides critical observational data needed to understand the\ndynamics of forest environments. Notably, tree diameter at breast height (DBH)\nis a metric used to estimate forest biomass and carbon dioxide sequestration.\nManual methods of forest mapping are labor intensive and time consuming, a\nbottleneck for large-scale mapping efforts. Automated mapping relies on\nacquiring dense forest reconstructions, typically in the form of point clouds.\nTerrestrial laser scanning (TLS) and mobile laser scanning (MLS) generate point\nclouds using expensive LiDAR sensing, and have been used successfully to\nestimate tree diameter. Neural radiance fields (NeRFs) are an emergent\ntechnology enabling photorealistic, vision-based reconstruction by training a\nneural network on a sparse set of input views. In this paper, we present a\ncomparison of MLS and NeRF forest reconstructions for the purpose of trunk\ndiameter estimation in a mixed-evergreen Redwood forest. In addition, we\npropose an improved DBH-estimation method using convex-hull modeling. Using\nthis approach, we achieved 1.68 cm RMSE, which consistently outperformed\nstandard cylinder modeling approaches. Our code contributions and forest\ndatasets are freely available at https://github.com/harelab-ucsc/RedwoodNeRF.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.08282", "title": "FusionSense: Bridging Common Sense, Vision, and Touch for Robust\n  Sparse-View Reconstruction", "abstract": "  Humans effortlessly integrate common-sense knowledge with sensory input from\nvision and touch to understand their surroundings. Emulating this capability,\nwe introduce FusionSense, a novel 3D reconstruction framework that enables\nrobots to fuse priors from foundation models with highly sparse observations\nfrom vision and tactile sensors. FusionSense addresses three key challenges:\n(i) How can robots efficiently acquire robust global shape information about\nthe surrounding scene and objects? (ii) How can robots strategically select\ntouch points on the object using geometric and common-sense priors? (iii) How\ncan partial observations such as tactile signals improve the overall\nrepresentation of the object? Our framework employs 3D Gaussian Splatting as a\ncore representation and incorporates a hierarchical optimization strategy\ninvolving global structure construction, object visual hull pruning and local\ngeometric constraints. This advancement results in fast and robust perception\nin environments with traditionally challenging objects that are transparent,\nreflective, or dark, enabling more downstream manipulation or navigation tasks.\nExperiments on real-world data suggest that our framework outperforms\npreviously state-of-the-art sparse-view methods. All code and data are\nopen-sourced on the project website.\n", "date": "2024-10-10", "categories": []}
{"id": "2405.14959", "title": "EvGGS: A Collaborative Learning Framework for Event-based Generalizable\n  Gaussian Splatting", "abstract": "  Event cameras offer promising advantages such as high dynamic range and low\nlatency, making them well-suited for challenging lighting conditions and\nfast-moving scenarios. However, reconstructing 3D scenes from raw event streams\nis difficult because event data is sparse and does not carry absolute color\ninformation. To release its potential in 3D reconstruction, we propose the\nfirst event-based generalizable 3D reconstruction framework, called EvGGS,\nwhich reconstructs scenes as 3D Gaussians from only event input in a\nfeedforward manner and can generalize to unseen cases without any retraining.\nThis framework includes a depth estimation module, an intensity reconstruction\nmodule, and a Gaussian regression module. These submodules connect in a\ncascading manner, and we collaboratively train them with a designed joint loss\nto make them mutually promote. To facilitate related studies, we build a novel\nevent-based 3D dataset with various material objects and calibrated labels of\ngrayscale images, depth maps, camera poses, and silhouettes. Experiments show\nmodels that have jointly trained significantly outperform those trained\nindividually. Our approach performs better than all baselines in reconstruction\nquality, and depth/intensity predictions with satisfactory rendering speed.\n", "date": "2024-05-23", "categories": []}
{"id": "2410.06231", "title": "RelitLRM: Generative Relightable Radiance for Large Reconstruction\n  Models", "abstract": "  We propose RelitLRM, a Large Reconstruction Model (LRM) for generating\nhigh-quality Gaussian splatting representations of 3D objects under novel\nilluminations from sparse (4-8) posed images captured under unknown static\nlighting. Unlike prior inverse rendering methods requiring dense captures and\nslow optimization, often causing artifacts like incorrect highlights or shadow\nbaking, RelitLRM adopts a feed-forward transformer-based model with a novel\ncombination of a geometry reconstructor and a relightable appearance generator\nbased on diffusion. The model is trained end-to-end on synthetic multi-view\nrenderings of objects under varying known illuminations. This architecture\ndesign enables to effectively decompose geometry and appearance, resolve the\nambiguity between material and lighting, and capture the multi-modal\ndistribution of shadows and specularity in the relit appearance. We show our\nsparse-view feed-forward RelitLRM offers competitive relighting results to\nstate-of-the-art dense-view optimization-based baselines while being\nsignificantly faster. Our project page is available at:\nhttps://relit-lrm.github.io/.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.07597", "title": "Fine-detailed Neural Indoor Scene Reconstruction using multi-level\n  importance sampling and multi-view consistency", "abstract": "  Recently, neural implicit 3D reconstruction in indoor scenarios has become\npopular due to its simplicity and impressive performance. Previous works could\nproduce complete results leveraging monocular priors of normal or depth.\nHowever, they may suffer from over-smoothed reconstructions and long-time\noptimization due to unbiased sampling and inaccurate monocular priors. In this\npaper, we propose a novel neural implicit surface reconstruction method, named\nFD-NeuS, to learn fine-detailed 3D models using multi-level importance sampling\nstrategy and multi-view consistency methodology. Specifically, we leverage\nsegmentation priors to guide region-based ray sampling, and use piecewise\nexponential functions as weights to pilot 3D points sampling along the rays,\nensuring more attention on important regions. In addition, we introduce\nmulti-view feature consistency and multi-view normal consistency as supervision\nand uncertainty respectively, which further improve the reconstruction of\ndetails. Extensive quantitative and qualitative results show that FD-NeuS\noutperforms existing methods in various scenes.\n", "date": "2024-10-10", "categories": []}
{"id": "2410.07707", "title": "MotionGS: Exploring Explicit Motion Guidance for Deformable 3D Gaussian\n  Splatting", "abstract": "  Dynamic scene reconstruction is a long-term challenge in the field of 3D\nvision. Recently, the emergence of 3D Gaussian Splatting has provided new\ninsights into this problem. Although subsequent efforts rapidly extend static\n3D Gaussian to dynamic scenes, they often lack explicit constraints on object\nmotion, leading to optimization difficulties and performance degradation. To\naddress the above issues, we propose a novel deformable 3D Gaussian splatting\nframework called MotionGS, which explores explicit motion priors to guide the\ndeformation of 3D Gaussians. Specifically, we first introduce an optical flow\ndecoupling module that decouples optical flow into camera flow and motion flow,\ncorresponding to camera movement and object motion respectively. Then the\nmotion flow can effectively constrain the deformation of 3D Gaussians, thus\nsimulating the motion of dynamic objects. Additionally, a camera pose\nrefinement module is proposed to alternately optimize 3D Gaussians and camera\nposes, mitigating the impact of inaccurate camera poses. Extensive experiments\nin the monocular dynamic scenes validate that MotionGS surpasses\nstate-of-the-art methods and exhibits significant superiority in both\nqualitative and quantitative results. Project page:\nhttps://ruijiezhu94.github.io/MotionGS_page\n", "date": "2024-10-10", "categories": []}
{"id": "2410.08092", "title": "UW-SDF: Exploiting Hybrid Geometric Priors for Neural SDF Reconstruction\n  from Underwater Multi-view Monocular Images", "abstract": "  Due to the unique characteristics of underwater environments, accurate 3D\nreconstruction of underwater objects poses a challenging problem in tasks such\nas underwater exploration and mapping. Traditional methods that rely on\nmultiple sensor data for 3D reconstruction are time-consuming and face\nchallenges in data acquisition in underwater scenarios. We propose UW-SDF, a\nframework for reconstructing target objects from multi-view underwater images\nbased on neural SDF. We introduce hybrid geometric priors to optimize the\nreconstruction process, markedly enhancing the quality and efficiency of neural\nSDF reconstruction. Additionally, to address the challenge of segmentation\nconsistency in multi-view images, we propose a novel few-shot multi-view target\nsegmentation strategy using the general-purpose segmentation model (SAM),\nenabling rapid automatic segmentation of unseen objects. Through extensive\nqualitative and quantitative experiments on diverse datasets, we demonstrate\nthat our proposed method outperforms the traditional underwater 3D\nreconstruction method and other neural rendering approaches in the field of\nunderwater 3D reconstruction.\n", "date": "2024-10-10", "categories": []}
{"id": "2410.05772", "title": "Comparative Analysis of Novel View Synthesis and Photogrammetry for 3D\n  Forest Stand Reconstruction and extraction of individual tree parameters", "abstract": "  Accurate and efficient 3D reconstruction of trees is crucial for forest\nresource assessments and management. Close-Range Photogrammetry (CRP) is\ncommonly used for reconstructing forest scenes but faces challenges like low\nefficiency and poor quality. Recently, Novel View Synthesis (NVS) technologies,\nincluding Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have\nshown promise for 3D plant reconstruction with limited images. However,\nexisting research mainly focuses on small plants in orchards or individual\ntrees, leaving uncertainty regarding their application in larger, complex\nforest stands. In this study, we collected sequential images of forest plots\nwith varying complexity and performed dense reconstruction using NeRF and 3DGS.\nThe resulting point clouds were compared with those from photogrammetry and\nlaser scanning. Results indicate that NVS methods significantly enhance\nreconstruction efficiency. Photogrammetry struggles with complex stands,\nleading to point clouds with excessive canopy noise and incorrectly\nreconstructed trees, such as duplicated trunks. NeRF, while better for canopy\nregions, may produce errors in ground areas with limited views. The 3DGS method\ngenerates sparser point clouds, particularly in trunk areas, affecting diameter\nat breast height (DBH) accuracy. All three methods can extract tree height\ninformation, with NeRF yielding the highest accuracy; however, photogrammetry\nremains superior for DBH accuracy. These findings suggest that NVS methods have\nsignificant potential for 3D reconstruction of forest stands, offering valuable\nsupport for complex forest resource inventory and visualization tasks.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.06245", "title": "HiSplat: Hierarchical 3D Gaussian Splatting for Generalizable\n  Sparse-View Reconstruction", "abstract": "  Reconstructing 3D scenes from multiple viewpoints is a fundamental task in\nstereo vision. Recently, advances in generalizable 3D Gaussian Splatting have\nenabled high-quality novel view synthesis for unseen scenes from sparse input\nviews by feed-forward predicting per-pixel Gaussian parameters without extra\noptimization. However, existing methods typically generate single-scale 3D\nGaussians, which lack representation of both large-scale structure and texture\ndetails, resulting in mislocation and artefacts. In this paper, we propose a\nnovel framework, HiSplat, which introduces a hierarchical manner in\ngeneralizable 3D Gaussian Splatting to construct hierarchical 3D Gaussians via\na coarse-to-fine strategy. Specifically, HiSplat generates large coarse-grained\nGaussians to capture large-scale structures, followed by fine-grained Gaussians\nto enhance delicate texture details. To promote inter-scale interactions, we\npropose an Error Aware Module for Gaussian compensation and a Modulating Fusion\nModule for Gaussian repair. Our method achieves joint optimization of\nhierarchical representations, allowing for novel view synthesis using only\ntwo-view reference images. Comprehensive experiments on various datasets\ndemonstrate that HiSplat significantly enhances reconstruction quality and\ncross-dataset generalization compared to prior single-scale methods. The\ncorresponding ablation study and analysis of different-scale 3D Gaussians\nreveal the mechanism behind the effectiveness. Project website:\nhttps://open3dvlab.github.io/HiSplat/\n", "date": "2024-10-08", "categories": []}
{"id": "2303.04508", "title": "InFusionSurf: Refining Neural RGB-D Surface Reconstruction Using\n  Per-Frame Intrinsic Refinement and TSDF Fusion Prior Learning", "abstract": "  We introduce InFusionSurf, an innovative enhancement for neural radiance\nfield (NeRF) frameworks in 3D surface reconstruction using RGB-D video frames.\nBuilding upon previous methods that have employed feature encoding to improve\noptimization speed, we further improve the reconstruction quality with minimal\nimpact on optimization time by refining depth information. InFusionSurf\naddresses camera motion-induced blurs in each depth frame through a per-frame\nintrinsic refinement scheme. It incorporates the truncated signed distance\nfield (TSDF) Fusion, a classical real-time 3D surface reconstruction method, as\na pretraining tool for the feature grid, enhancing reconstruction details and\ntraining speed. Comparative quantitative and qualitative analyses show that\nInFusionSurf reconstructs scenes with high accuracy while maintaining\noptimization efficiency. The effectiveness of our intrinsic refinement and TSDF\nFusion-based pretraining is further validated through an ablation study.\n", "date": "2023-03-08", "categories": []}
{"id": "2410.04529", "title": "In-Place Panoptic Radiance Field Segmentation with Perceptual Prior for\n  3D Scene Understanding", "abstract": "  Accurate 3D scene representation and panoptic understanding are essential for\napplications such as virtual reality, robotics, and autonomous driving.\nHowever, challenges persist with existing methods, including precise 2D-to-3D\nmapping, handling complex scene characteristics like boundary ambiguity and\nvarying scales, and mitigating noise in panoptic pseudo-labels. This paper\nintroduces a novel perceptual-prior-guided 3D scene representation and panoptic\nunderstanding method, which reformulates panoptic understanding within neural\nradiance fields as a linear assignment problem involving 2D semantics and\ninstance recognition. Perceptual information from pre-trained 2D panoptic\nsegmentation models is incorporated as prior guidance, thereby synchronizing\nthe learning processes of appearance, geometry, and panoptic understanding\nwithin neural radiance fields. An implicit scene representation and\nunderstanding model is developed to enhance generalization across indoor and\noutdoor scenes by extending the scale-encoded cascaded grids within a\nreparameterized domain distillation framework. This model effectively manages\ncomplex scene attributes and generates 3D-consistent scene representations and\npanoptic understanding outcomes for various scenes. Experiments and ablation\nstudies under challenging conditions, including synthetic and real-world\nscenes, demonstrate the proposed method's effectiveness in enhancing 3D scene\nrepresentation and panoptic segmentation accuracy.\n", "date": "2024-10-06", "categories": []}
{"id": "2410.05044", "title": "PhotoReg: Photometrically Registering 3D Gaussian Splatting Models", "abstract": "  Building accurate representations of the environment is critical for\nintelligent robots to make decisions during deployment. Advances in\nphotorealistic environment models have enabled robots to develop\nhyper-realistic reconstructions, which can be used to generate images that are\nintuitive for human inspection. In particular, the recently introduced\n\\ac{3DGS}, which describes the scene with up to millions of primitive\nellipsoids, can be rendered in real time. \\ac{3DGS} has rapidly gained\nprominence. However, a critical unsolved problem persists: how can we fuse\nmultiple \\ac{3DGS} into a single coherent model? Solving this problem will\nenable robot teams to jointly build \\ac{3DGS} models of their surroundings. A\nkey insight of this work is to leverage the {duality} between photorealistic\nreconstructions, which render realistic 2D images from 3D structure, and\n\\emph{3D foundation models}, which predict 3D structure from image pairs. To\nthis end, we develop PhotoReg, a framework to register multiple photorealistic\n\\ac{3DGS} models with 3D foundation models. As \\ac{3DGS} models are generally\nbuilt from monocular camera images, they have \\emph{arbitrary scale}. To\nresolve this, PhotoReg actively enforces scale consistency among the different\n\\ac{3DGS} models by considering depth estimates within these models. Then, the\nalignment is iteratively refined with fine-grained photometric losses to\nproduce high-quality fused \\ac{3DGS} models. We rigorously evaluate PhotoReg on\nboth standard benchmark datasets and our custom-collected datasets, including\nwith two quadruped robots. The code is released at\n\\url{ziweny11.github.io/photoreg}.\n", "date": "2024-10-07", "categories": []}
{"id": "2410.05097", "title": "DreamSat: Towards a General 3D Model for Novel View Synthesis of Space\n  Objects", "abstract": "  Novel view synthesis (NVS) enables to generate new images of a scene or\nconvert a set of 2D images into a comprehensive 3D model. In the context of\nSpace Domain Awareness, since space is becoming increasingly congested, NVS can\naccurately map space objects and debris, improving the safety and efficiency of\nspace operations. Similarly, in Rendezvous and Proximity Operations missions,\n3D models can provide details about a target object's shape, size, and\norientation, allowing for better planning and prediction of the target's\nbehavior. In this work, we explore the generalization abilities of these\nreconstruction techniques, aiming to avoid the necessity of retraining for each\nnew scene, by presenting a novel approach to 3D spacecraft reconstruction from\nsingle-view images, DreamSat, by fine-tuning the Zero123 XL, a state-of-the-art\nsingle-view reconstruction model, on a high-quality dataset of 190 high-quality\nspacecraft models and integrating it into the DreamGaussian framework. We\ndemonstrate consistent improvements in reconstruction quality across multiple\nmetrics, including Contrastive Language-Image Pretraining (CLIP) score\n(+0.33%), Peak Signal-to-Noise Ratio (PSNR) (+2.53%), Structural Similarity\nIndex (SSIM) (+2.38%), and Learned Perceptual Image Patch Similarity (LPIPS)\n(+0.16%) on a test set of 30 previously unseen spacecraft images. Our method\naddresses the lack of domain-specific 3D reconstruction tools in the space\nindustry by leveraging state-of-the-art diffusion models and 3D Gaussian\nsplatting techniques. This approach maintains the efficiency of the\nDreamGaussian framework while enhancing the accuracy and detail of spacecraft\nreconstructions. The code for this work can be accessed on GitHub\n(https://github.com/ARCLab-MIT/space-nvs).\n", "date": "2024-10-07", "categories": []}
{"id": "2406.19811", "title": "EgoGaussian: Dynamic Scene Understanding from Egocentric Video with 3D\n  Gaussian Splatting", "abstract": "  Human activities are inherently complex, often involving numerous object\ninteractions. To better understand these activities, it is crucial to model\ntheir interactions with the environment captured through dynamic changes. The\nrecent availability of affordable head-mounted cameras and egocentric data\noffers a more accessible and efficient means to understand human-object\ninteractions in 3D environments. However, most existing methods for human\nactivity modeling neglect the dynamic interactions with objects, resulting in\nonly static representations. The few existing solutions often require inputs\nfrom multiple sources, including multi-camera setups, depth-sensing cameras, or\nkinesthetic sensors. To this end, we introduce EgoGaussian, the first method\ncapable of simultaneously reconstructing 3D scenes and dynamically tracking 3D\nobject motion from RGB egocentric input alone. We leverage the uniquely\ndiscrete nature of Gaussian Splatting and segment dynamic interactions from the\nbackground, with both having explicit representations. Our approach employs a\nclip-level online learning pipeline that leverages the dynamic nature of human\nactivities, allowing us to reconstruct the temporal evolution of the scene in\nchronological order and track rigid object motion. EgoGaussian shows\nsignificant improvements in terms of both dynamic object and background\nreconstruction quality compared to the state-of-the-art. We also qualitatively\ndemonstrate the high quality of the reconstructed models.\n", "date": "2024-06-28", "categories": []}
{"id": "2407.19837", "title": "VortSDF: 3D Modeling with Centroidal Voronoi Tesselation on Signed\n  Distance Field", "abstract": "  Volumetric shape representations have become ubiquitous in multi-view\nreconstruction tasks. They often build on regular voxel grids as discrete\nrepresentations of 3D shape functions, such as SDF or radiance fields, either\nas the full shape model or as sampled instantiations of continuous\nrepresentations, as with neural networks. Despite their proven efficiency,\nvoxel representations come with the precision versus complexity trade-off. This\ninherent limitation can significantly impact performance when moving away from\nsimple and uncluttered scenes. In this paper we investigate an alternative\ndiscretization strategy with the Centroidal Voronoi Tesselation (CVT). CVTs\nallow to better partition the observation space with respect to shape occupancy\nand to focus the discretization around shape surfaces. To leverage this\ndiscretization strategy for multi-view reconstruction, we introduce a\nvolumetric optimization framework that combines explicit SDF fields with a\nshallow color network, in order to estimate 3D shape properties over\ntetrahedral grids. Experimental results with Chamfer statistics validate this\napproach with unprecedented reconstruction quality on various scenarios such as\nobjects, open scenes or human.\n", "date": "2024-07-29", "categories": []}
{"id": "2408.11540", "title": "DeRainGS: Gaussian Splatting for Enhanced Scene Reconstruction in Rainy\n  Environments", "abstract": "  Reconstruction under adverse rainy conditions poses significant challenges\ndue to reduced visibility and the distortion of visual perception. These\nconditions can severely impair the quality of geometric maps, which is\nessential for applications ranging from autonomous planning to environmental\nmonitoring. In response to these challenges, this study introduces the novel\ntask of 3D Reconstruction in Rainy Environments (3DRRE), specifically designed\nto address the complexities of reconstructing 3D scenes under rainy conditions.\nTo benchmark this task, we construct the HydroViews dataset that comprises a\ndiverse collection of both synthesized and real-world scene images\ncharacterized by various intensities of rain streaks and raindrops.\nFurthermore, we propose DeRainGS, the first 3DGS method tailored for\nreconstruction in adverse rainy environments. Extensive experiments across a\nwide range of rain scenarios demonstrate that our method delivers\nstate-of-the-art performance, remarkably outperforming existing occlusion-free\nmethods.\n", "date": "2024-08-21", "categories": []}
{"id": "2408.14724", "title": "GeoTransfer : Generalizable Few-Shot Multi-View Reconstruction via\n  Transfer Learning", "abstract": "  This paper presents a novel approach for sparse 3D reconstruction by\nleveraging the expressive power of Neural Radiance Fields (NeRFs) and fast\ntransfer of their features to learn accurate occupancy fields. Existing 3D\nreconstruction methods from sparse inputs still struggle with capturing\nintricate geometric details and can suffer from limitations in handling\noccluded regions. On the other hand, NeRFs excel in modeling complex scenes but\ndo not offer means to extract meaningful geometry. Our proposed method offers\nthe best of both worlds by transferring the information encoded in NeRF\nfeatures to derive an accurate occupancy field representation. We utilize a\npre-trained, generalizable state-of-the-art NeRF network to capture detailed\nscene radiance information, and rapidly transfer this knowledge to train a\ngeneralizable implicit occupancy network. This process helps in leveraging the\nknowledge of the scene geometry encoded in the generalizable NeRF prior and\nrefining it to learn occupancy fields, facilitating a more precise\ngeneralizable representation of 3D space. The transfer learning approach leads\nto a dramatic reduction in training time, by orders of magnitude (i.e. from\nseveral days to 3.5 hrs), obviating the need to train generalizable sparse\nsurface reconstruction methods from scratch. Additionally, we introduce a novel\nloss on volumetric rendering weights that helps in the learning of accurate\noccupancy fields, along with a normal loss that helps in global smoothing of\nthe occupancy fields. We evaluate our approach on the DTU dataset and\ndemonstrate state-of-the-art performance in terms of reconstruction accuracy,\nespecially in challenging scenarios with sparse input data and occluded\nregions. We furthermore demonstrate the generalization capabilities of our\nmethod by showing qualitative results on the Blended MVS dataset without any\nretraining.\n", "date": "2024-08-27", "categories": []}
{"id": "2409.19039", "title": "Gaussian Heritage: 3D Digitization of Cultural Heritage with Integrated\n  Object Segmentation", "abstract": "  The creation of digital replicas of physical objects has valuable\napplications for the preservation and dissemination of tangible cultural\nheritage. However, existing methods are often slow, expensive, and require\nexpert knowledge. We propose a pipeline to generate a 3D replica of a scene\nusing only RGB images (e.g. photos of a museum) and then extract a model for\neach item of interest (e.g. pieces in the exhibit). We do this by leveraging\nthe advancements in novel view synthesis and Gaussian Splatting, modified to\nenable efficient 3D segmentation. This approach does not need manual\nannotation, and the visual inputs can be captured using a standard smartphone,\nmaking it both affordable and easy to deploy. We provide an overview of the\nmethod and baseline evaluation of the accuracy of object segmentation. The code\nis available at https://mahtaabdn.github.io/gaussian_heritage.github.io/.\n", "date": "2024-09-27", "categories": []}
{"id": "2409.19322", "title": "Scalable Cloud-Native Pipeline for Efficient 3D Model Reconstruction\n  from Monocular Smartphone Images", "abstract": "  In recent years, 3D models have gained popularity in various fields,\nincluding entertainment, manufacturing, and simulation. However, manually\ncreating these models can be a time-consuming and resource-intensive process,\nmaking it impractical for large-scale industrial applications. To address this\nissue, researchers are exploiting Artificial Intelligence and Machine Learning\nalgorithms to automatically generate 3D models effortlessly. In this paper, we\npresent a novel cloud-native pipeline that can automatically reconstruct 3D\nmodels from monocular 2D images captured using a smartphone camera. Our goal is\nto provide an efficient and easily-adoptable solution that meets the Industry\n4.0 standards for creating a Digital Twin model, which could enhance personnel\nexpertise through accelerated training. We leverage machine learning models\ndeveloped by NVIDIA Research Labs alongside a custom-designed pose recorder\nwith a unique pose compensation component based on the ARCore framework by\nGoogle. Our solution produces a reusable 3D model, with embedded materials and\ntextures, exportable and customizable in any external 3D modelling software or\n3D engine. Furthermore, the whole workflow is implemented by adopting the\nmicroservices architecture standard, enabling each component of the pipeline to\noperate as a standalone replaceable module.\n", "date": "2024-09-28", "categories": []}
{"id": "2409.19405", "title": "G3R: Gradient Guided Generalizable Reconstruction", "abstract": "  Large scale 3D scene reconstruction is important for applications such as\nvirtual reality and simulation. Existing neural rendering approaches (e.g.,\nNeRF, 3DGS) have achieved realistic reconstructions on large scenes, but\noptimize per scene, which is expensive and slow, and exhibit noticeable\nartifacts under large view changes due to overfitting. Generalizable approaches\nor large reconstruction models are fast, but primarily work for small\nscenes/objects and often produce lower quality rendering results. In this work,\nwe introduce G3R, a generalizable reconstruction approach that can efficiently\npredict high-quality 3D scene representations for large scenes. We propose to\nlearn a reconstruction network that takes the gradient feedback signals from\ndifferentiable rendering to iteratively update a 3D scene representation,\ncombining the benefits of high photorealism from per-scene optimization with\ndata-driven priors from fast feed-forward prediction methods. Experiments on\nurban-driving and drone datasets show that G3R generalizes across diverse large\nscenes and accelerates the reconstruction process by at least 10x while\nachieving comparable or better realism compared to 3DGS, and also being more\nrobust to large view changes.\n", "date": "2024-09-28", "categories": []}
{"id": "2409.20276", "title": "Active Neural Mapping at Scale", "abstract": "  We introduce a NeRF-based active mapping system that enables efficient and\nrobust exploration of large-scale indoor environments. The key to our approach\nis the extraction of a generalized Voronoi graph (GVG) from the continually\nupdated neural map, leading to the synergistic integration of scene geometry,\nappearance, topology, and uncertainty. Anchoring uncertain areas induced by the\nneural map to the vertices of GVG allows the exploration to undergo adaptive\ngranularity along a safe path that traverses unknown areas efficiently.\nHarnessing a modern hybrid NeRF representation, the proposed system achieves\ncompetitive results in terms of reconstruction accuracy, coverage completeness,\nand exploration efficiency even when scaling up to large indoor environments.\nExtensive results at different scales validate the efficacy of the proposed\nsystem.\n", "date": "2024-09-30", "categories": []}
{"id": "2409.20530", "title": "Dual Encoder GAN Inversion for High-Fidelity 3D Head Reconstruction from\n  Single Images", "abstract": "  3D GAN inversion aims to project a single image into the latent space of a 3D\nGenerative Adversarial Network (GAN), thereby achieving 3D geometry\nreconstruction. While there exist encoders that achieve good results in 3D GAN\ninversion, they are predominantly built on EG3D, which specializes in\nsynthesizing near-frontal views and is limiting in synthesizing comprehensive\n3D scenes from diverse viewpoints. In contrast to existing approaches, we\npropose a novel framework built on PanoHead, which excels in synthesizing\nimages from a 360-degree perspective. To achieve realistic 3D modeling of the\ninput image, we introduce a dual encoder system tailored for high-fidelity\nreconstruction and realistic generation from different viewpoints. Accompanying\nthis, we propose a stitching framework on the triplane domain to get the best\npredictions from both. To achieve seamless stitching, both encoders must output\nconsistent results despite being specialized for different tasks. For this\nreason, we carefully train these encoders using specialized losses, including\nan adversarial loss based on our novel occlusion-aware triplane discriminator.\nExperiments reveal that our approach surpasses the existing encoder training\nmethods qualitatively and quantitatively. Please visit the project page:\nhttps://berkegokmen1.github.io/dual-enc-3d-gan-inv.\n", "date": "2024-09-30", "categories": []}
{"id": "2311.18159", "title": "CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization", "abstract": "  3D Gaussian Splatting (3DGS) is a new method for modeling and rendering 3D\nradiance fields that achieves much faster learning and rendering time compared\nto SOTA NeRF methods. However, it comes with a drawback in the much larger\nstorage demand compared to NeRF methods since it needs to store the parameters\nfor several 3D Gaussians. We notice that many Gaussians may share similar\nparameters, so we introduce a simple vector quantization method based on\nK-means to quantize the Gaussian parameters while optimizing them. Then, we\nstore the small codebook along with the index of the code for each Gaussian. We\ncompress the indices further by sorting them and using a method similar to\nrun-length encoding. Moreover, we use a simple regularizer to encourage zero\nopacity (invisible Gaussians) to reduce the storage and rendering time by a\nlarge factor through reducing the number of Gaussians. We do extensive\nexperiments on standard benchmarks as well as an existing 3D dataset that is an\norder of magnitude larger than the standard benchmarks used in this field. We\nshow that our simple yet effective method can reduce the storage cost for 3DGS\nby 40 to 50x and rendering time by 2 to 3x with a very small drop in the\nquality of rendered images.\n", "date": "2023-11-30", "categories": []}
{"id": "2409.18852", "title": "Space-time 2D Gaussian Splatting for Accurate Surface Reconstruction\n  under Complex Dynamic Scenes", "abstract": "  Previous surface reconstruction methods either suffer from low geometric\naccuracy or lengthy training times when dealing with real-world complex dynamic\nscenes involving multi-person activities, and human-object interactions. To\ntackle the dynamic contents and the occlusions in complex scenes, we present a\nspace-time 2D Gaussian Splatting approach. Specifically, to improve geometric\nquality in dynamic scenes, we learn canonical 2D Gaussian splats and deform\nthese 2D Gaussian splats while enforcing the disks of the Gaussian located on\nthe surface of the objects by introducing depth and normal regularizers.\nFurther, to tackle the occlusion issues in complex scenes, we introduce a\ncompositional opacity deformation strategy, which further reduces the surface\nrecovery of those occluded areas. Experiments on real-world sparse-view video\ndatasets and monocular dynamic datasets demonstrate that our reconstructions\noutperform state-of-the-art methods, especially for the surface of the details.\nThe project page and more visualizations can be found at:\nhttps://tb2-sy.github.io/st-2dgs/.\n", "date": "2024-09-27", "categories": []}
{"id": "2407.15435", "title": "Enhancement of 3D Gaussian Splatting using Raw Mesh for Photorealistic\n  Recreation of Architectures", "abstract": "  The photorealistic reconstruction and rendering of architectural scenes have\nextensive applications in industries such as film, games, and transportation.\nIt also plays an important role in urban planning, architectural design, and\nthe city's promotion, especially in protecting historical and cultural relics.\nThe 3D Gaussian Splatting, due to better performance over NeRF, has become a\nmainstream technology in 3D reconstruction. Its only input is a set of images\nbut it relies heavily on geometric parameters computed by the SfM process. At\nthe same time, there is an existing abundance of raw 3D models, that could\ninform the structural perception of certain buildings but cannot be applied. In\nthis paper, we propose a straightforward method to harness these raw 3D models\nto guide 3D Gaussians in capturing the basic shape of the building and improve\nthe visual quality of textures and details when photos are captured\nnon-systematically. This exploration opens up new possibilities for improving\nthe effectiveness of 3D reconstruction techniques in the field of architectural\ndesign.\n", "date": "2024-07-22", "categories": []}
{"id": "2409.16470", "title": "Frequency-based View Selection in Gaussian Splatting Reconstruction", "abstract": "  Three-dimensional reconstruction is a fundamental problem in robotics\nperception. We examine the problem of active view selection to perform 3D\nGaussian Splatting reconstructions with as few input images as possible.\nAlthough 3D Gaussian Splatting has made significant progress in image rendering\nand 3D reconstruction, the quality of the reconstruction is strongly impacted\nby the selection of 2D images and the estimation of camera poses through\nStructure-from-Motion (SfM) algorithms. Current methods to select views that\nrely on uncertainties from occlusions, depth ambiguities, or neural network\npredictions directly are insufficient to handle the issue and struggle to\ngeneralize to new scenes. By ranking the potential views in the frequency\ndomain, we are able to effectively estimate the potential information gain of\nnew viewpoints without ground truth data. By overcoming current constraints on\nmodel architecture and efficacy, our method achieves state-of-the-art results\nin view selection, demonstrating its potential for efficient image-based 3D\nreconstruction.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.16944", "title": "Go-SLAM: Grounded Object Segmentation and Localization with Gaussian\n  Splatting SLAM", "abstract": "  We introduce Go-SLAM, a novel framework that utilizes 3D Gaussian Splatting\nSLAM to reconstruct dynamic environments while embedding object-level\ninformation within the scene representations. This framework employs advanced\nobject segmentation techniques, assigning a unique identifier to each Gaussian\nsplat that corresponds to the object it represents. Consequently, our system\nfacilitates open-vocabulary querying, allowing users to locate objects using\nnatural language descriptions. Furthermore, the framework features an optimal\npath generation module that calculates efficient navigation paths for robots\ntoward queried objects, considering obstacles and environmental uncertainties.\nComprehensive evaluations in various scene settings demonstrate the\neffectiveness of our approach in delivering high-fidelity scene\nreconstructions, precise object segmentation, flexible object querying, and\nefficient robot path planning. This work represents an additional step forward\nin bridging the gap between 3D scene reconstruction, semantic object\nunderstanding, and real-time environment interactions.\n", "date": "2024-09-25", "categories": []}
{"id": "2403.18711", "title": "SAT-NGP : Unleashing Neural Graphics Primitives for Fast Relightable\n  Transient-Free 3D reconstruction from Satellite Imagery", "abstract": "  Current stereo-vision pipelines produce high accuracy 3D reconstruction when\nusing multiple pairs or triplets of satellite images. However, these pipelines\nare sensitive to the changes between images that can occur as a result of\nmulti-date acquisitions. Such variations are mainly due to variable shadows,\nreflexions and transient objects (cars, vegetation). To take such changes into\naccount, Neural Radiance Fields (NeRF) have recently been applied to multi-date\nsatellite imagery. However, Neural methods are very compute-intensive, taking\ndozens of hours to learn, compared with minutes for standard stereo-vision\npipelines. Following the ideas of Instant Neural Graphics Primitives we propose\nto use an efficient sampling strategy and multi-resolution hash encoding to\naccelerate the learning. Our model, Satellite Neural Graphics Primitives\n(SAT-NGP) decreases the learning time to 15 minutes while maintaining the\nquality of the 3D reconstruction.\n", "date": "2024-03-27", "categories": []}
{"id": "2409.12774", "title": "GaRField++: Reinforced Gaussian Radiance Fields for Large-Scale 3D Scene\n  Reconstruction", "abstract": "  This paper proposes a novel framework for large-scale scene reconstruction\nbased on 3D Gaussian splatting (3DGS) and aims to address the scalability and\naccuracy challenges faced by existing methods. For tackling the scalability\nissue, we split the large scene into multiple cells, and the candidate\npoint-cloud and camera views of each cell are correlated through a\nvisibility-based camera selection and a progressive point-cloud extension. To\nreinforce the rendering quality, three highlighted improvements are made in\ncomparison with vanilla 3DGS, which are a strategy of the ray-Gaussian\nintersection and the novel Gaussians density control for learning efficiency,\nan appearance decoupling module based on ConvKAN network to solve uneven\nlighting conditions in large-scale scenes, and a refined final loss with the\ncolor loss, the depth distortion loss, and the normal consistency loss.\nFinally, the seamless stitching procedure is executed to merge the individual\nGaussian radiance field for novel view synthesis across different cells.\nEvaluation of Mill19, Urban3D, and MatrixCity datasets shows that our method\nconsistently generates more high-fidelity rendering results than\nstate-of-the-art methods of large-scale scene reconstruction. We further\nvalidate the generalizability of the proposed approach by rendering on\nself-collected video clips recorded by a commercial drone.\n", "date": "2024-09-19", "categories": []}
{"id": "2409.15715", "title": "Disentangled Generation and Aggregation for Robust Radiance Fields", "abstract": "  The utilization of the triplane-based radiance fields has gained attention in\nrecent years due to its ability to effectively disentangle 3D scenes with a\nhigh-quality representation and low computation cost. A key requirement of this\nmethod is the precise input of camera poses. However, due to the local update\nproperty of the triplane, a similar joint estimation as previous joint\npose-NeRF optimization works easily results in local minima. To this end, we\npropose the Disentangled Triplane Generation module to introduce global feature\ncontext and smoothness into triplane learning, which mitigates errors caused by\nlocal updating. Then, we propose the Disentangled Plane Aggregation to mitigate\nthe entanglement caused by the common triplane feature aggregation during\ncamera pose updating. In addition, we introduce a two-stage warm-start training\nstrategy to reduce the implicit constraints caused by the triplane generator.\nQuantitative and qualitative results demonstrate that our proposed method\nachieves state-of-the-art performance in novel view synthesis with noisy or\nunknown camera poses, as well as efficient convergence of optimization. Project\npage: https://gaohchen.github.io/DiGARR/.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.15959", "title": "Semantics-Controlled Gaussian Splatting for Outdoor Scene Reconstruction\n  and Rendering in Virtual Reality", "abstract": "  Advancements in 3D rendering like Gaussian Splatting (GS) allow novel view\nsynthesis and real-time rendering in virtual reality (VR). However, GS-created\n3D environments are often difficult to edit. For scene enhancement or to\nincorporate 3D assets, segmenting Gaussians by class is essential. Existing\nsegmentation approaches are typically limited to certain types of scenes, e.g.,\n''circular'' scenes, to determine clear object boundaries. However, this method\nis ineffective when removing large objects in non-''circling'' scenes such as\nlarge outdoor scenes. We propose Semantics-Controlled GS (SCGS), a\nsegmentation-driven GS approach, enabling the separation of large scene parts\nin uncontrolled, natural environments. SCGS allows scene editing and the\nextraction of scene parts for VR. Additionally, we introduce a challenging\noutdoor dataset, overcoming the ''circling'' setup. We outperform the\nstate-of-the-art in visual quality on our dataset and in segmentation quality\non the 3D-OVS dataset. We conducted an exploratory user study, comparing a\n360-video, plain GS, and SCGS in VR with a fixed viewpoint. In our subsequent\nmain study, users were allowed to move freely, evaluating plain GS and SCGS.\nOur main study results show that participants clearly prefer SCGS over plain\nGS. We overall present an innovative approach that surpasses the\nstate-of-the-art both technically and in user experience.\n", "date": "2024-09-24", "categories": []}
{"id": "2407.20219", "title": "Global Structure-from-Motion Revisited", "abstract": "  Recovering 3D structure and camera motion from images has been a\nlong-standing focus of computer vision research and is known as\nStructure-from-Motion (SfM). Solutions to this problem are categorized into\nincremental and global approaches. Until now, the most popular systems follow\nthe incremental paradigm due to its superior accuracy and robustness, while\nglobal approaches are drastically more scalable and efficient. With this work,\nwe revisit the problem of global SfM and propose GLOMAP as a new\ngeneral-purpose system that outperforms the state of the art in global SfM. In\nterms of accuracy and robustness, we achieve results on-par or superior to\nCOLMAP, the most widely used incremental SfM, while being orders of magnitude\nfaster. We share our system as an open-source implementation at\n{https://github.com/colmap/glomap}.\n", "date": "2024-07-29", "categories": []}
{"id": "2409.00381", "title": "3D Gaussian Splatting for Large-scale Surface Reconstruction from Aerial\n  Images", "abstract": "  Recently, 3D Gaussian Splatting (3DGS) has demonstrated excellent ability in\nsmall-scale 3D surface reconstruction. However, extending 3DGS to large-scale\nscenes remains a significant challenge. To address this gap, we propose a novel\n3DGS-based method for large-scale surface reconstruction using aerial\nmulti-view stereo (MVS) images, named Aerial Gaussian Splatting (AGS). First,\nwe introduce a data chunking method tailored for large-scale aerial images,\nmaking 3DGS feasible for surface reconstruction over extensive scenes. Second,\nwe integrate the Ray-Gaussian Intersection method into 3DGS to obtain depth and\nnormal information. Finally, we implement multi-view geometric consistency\nconstraints to enhance the geometric consistency across different views. Our\nexperiments on multiple datasets demonstrate, for the first time, the\n3DGS-based method can match conventional aerial MVS methods on geometric\naccuracy in aerial large-scale surface reconstruction, and our method also\nbeats state-of-the-art GS-based methods both on geometry and rendering quality.\n", "date": "2024-08-31", "categories": []}
{"id": "2409.14019", "title": "MOSE: Monocular Semantic Reconstruction Using NeRF-Lifted Noisy Priors", "abstract": "  Accurately reconstructing dense and semantically annotated 3D meshes from\nmonocular images remains a challenging task due to the lack of geometry\nguidance and imperfect view-dependent 2D priors. Though we have witnessed\nrecent advancements in implicit neural scene representations enabling precise\n2D rendering simply from multi-view images, there have been few works\naddressing 3D scene understanding with monocular priors alone. In this paper,\nwe propose MOSE, a neural field semantic reconstruction approach to lift\ninferred image-level noisy priors to 3D, producing accurate semantics and\ngeometry in both 3D and 2D space. The key motivation for our method is to\nleverage generic class-agnostic segment masks as guidance to promote local\nconsistency of rendered semantics during training. With the help of semantics,\nwe further apply a smoothness regularization to texture-less regions for better\ngeometric quality, thus achieving mutual benefits of geometry and semantics.\nExperiments on the ScanNet dataset show that our MOSE outperforms relevant\nbaselines across all metrics on tasks of 3D semantic segmentation, 2D semantic\nsegmentation and 3D surface reconstruction.\n", "date": "2024-09-21", "categories": []}
{"id": "2409.14331", "title": "PISR: Polarimetric Neural Implicit Surface Reconstruction for\n  Textureless and Specular Objects", "abstract": "  Neural implicit surface reconstruction has achieved remarkable progress\nrecently. Despite resorting to complex radiance modeling, state-of-the-art\nmethods still struggle with textureless and specular surfaces. Different from\nRGB images, polarization images can provide direct constraints on the azimuth\nangles of the surface normals. In this paper, we present PISR, a novel method\nthat utilizes a geometrically accurate polarimetric loss to refine shape\nindependently of appearance. In addition, PISR smooths surface normals in image\nspace to eliminate severe shape distortions and leverages the hash-grid-based\nneural signed distance function to accelerate the reconstruction. Experimental\nresults demonstrate that PISR achieves higher accuracy and robustness, with an\nL1 Chamfer distance of 0.5 mm and an F-score of 99.5% at 1 mm, while converging\n4~30 times faster than previous polarimetric surface reconstruction methods.\n", "date": "2024-09-22", "categories": []}
{"id": "2409.01807", "title": "EPRecon: An Efficient Framework for Real-Time Panoptic 3D Reconstruction\n  from Monocular Video", "abstract": "  Panoptic 3D reconstruction from a monocular video is a fundamental perceptual\ntask in robotic scene understanding. However, existing efforts suffer from\ninefficiency in terms of inference speed and accuracy, limiting their practical\napplicability. We present EPRecon, an efficient real-time panoptic 3D\nreconstruction framework. Current volumetric-based reconstruction methods\nusually utilize multi-view depth map fusion to obtain scene depth priors, which\nis time-consuming and poses challenges to real-time scene reconstruction. To\naddress this issue, we propose a lightweight module to directly estimate scene\ndepth priors in a 3D volume for reconstruction quality improvement by\ngenerating occupancy probabilities of all voxels. In addition, compared with\nexisting panoptic segmentation methods, EPRecon extracts panoptic features from\nboth voxel features and corresponding image features, obtaining more detailed\nand comprehensive instance-level semantic information and achieving more\naccurate segmentation results. Experimental results on the ScanNetV2 dataset\ndemonstrate the superiority of EPRecon over current state-of-the-art methods in\nterms of both panoptic 3D reconstruction quality and real-time inference. Code\nis available at https://github.com/zhen6618/EPRecon.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.13392", "title": "Elite-EvGS: Learning Event-based 3D Gaussian Splatting by Distilling\n  Event-to-Video Priors", "abstract": "  Event cameras are bio-inspired sensors that output asynchronous and sparse\nevent streams, instead of fixed frames. Benefiting from their distinct\nadvantages, such as high dynamic range and high temporal resolution, event\ncameras have been applied to address 3D reconstruction, important for robotic\nmapping. Recently, neural rendering techniques, such as 3D Gaussian splatting\n(3DGS), have been shown successful in 3D reconstruction. However, it still\nremains under-explored how to develop an effective event-based 3DGS pipeline.\nIn particular, as 3DGS typically depends on high-quality initialization and\ndense multiview constraints, a potential problem appears for the 3DGS\noptimization with events given its inherent sparse property. To this end, we\npropose a novel event-based 3DGS framework, named Elite-EvGS. Our key idea is\nto distill the prior knowledge from the off-the-shelf event-to-video (E2V)\nmodels to effectively reconstruct 3D scenes from events in a coarse-to-fine\noptimization manner. Specifically, to address the complexity of 3DGS\ninitialization from events, we introduce a novel warm-up initialization\nstrategy that optimizes a coarse 3DGS from the frames generated by E2V models\nand then incorporates events to refine the details. Then, we propose a\nprogressive event supervision strategy that employs the window-slicing\noperation to progressively reduce the number of events used for supervision.\nThis subtly relives the temporal randomness of the event frames, benefiting the\noptimization of local textural and global structural details. Experiments on\nthe benchmark datasets demonstrate that Elite-EvGS can reconstruct 3D scenes\nwith better textural and structural details. Meanwhile, our method yields\nplausible performance on the captured real-world data, including diverse\nchallenging conditions, such as fast motion and low light scenes.\n", "date": "2024-09-20", "categories": []}
{"id": "2405.16788", "title": "3D Reconstruction with Fast Dipole Sums", "abstract": "  We introduce a method for high-quality 3D reconstruction from multi-view\nimages. Our method uses a new point-based representation, the regularized\ndipole sum, which generalizes the winding number to allow for interpolation of\nper-point attributes in point clouds with noisy or outlier points. Using\nregularized dipole sums, we represent implicit geometry and radiance fields as\nper-point attributes of a dense point cloud, which we initialize from structure\nfrom motion. We additionally derive Barnes-Hut fast summation schemes for\naccelerated forward and adjoint dipole sum queries. These queries facilitate\nthe use of ray tracing to efficiently and differentiably render images with our\npoint-based representations, and thus update their point attributes to optimize\nscene geometry and appearance. We evaluate our method in inverse rendering\napplications against state-of-the-art alternatives, based on ray tracing of\nneural representations or rasterization of Gaussian point-based\nrepresentations. Our method significantly improves 3D reconstruction quality\nand robustness at equal runtimes, while also supporting more general rendering\nmethods such as shadow rays for direct illumination.\n", "date": "2024-05-27", "categories": []}
{"id": "2406.07037", "title": "PanoSSC: Exploring Monocular Panoptic 3D Scene Reconstruction for\n  Autonomous Driving", "abstract": "  Vision-centric occupancy networks, which represent the surrounding\nenvironment with uniform voxels with semantics, have become a new trend for\nsafe driving of camera-only autonomous driving perception systems, as they are\nable to detect obstacles regardless of their shape and occlusion. Modern\noccupancy networks mainly focus on reconstructing visible voxels from object\nsurfaces with voxel-wise semantic prediction. Usually, they suffer from\ninconsistent predictions of one object and mixed predictions for adjacent\nobjects. These confusions may harm the safety of downstream planning modules.\nTo this end, we investigate panoptic segmentation on 3D voxel scenarios and\npropose an instance-aware occupancy network, PanoSSC. We predict foreground\nobjects and backgrounds separately and merge both in post-processing. For\nforeground instance grouping, we propose a novel 3D instance mask decoder that\ncan efficiently extract individual objects. we unify geometric reconstruction,\n3D semantic segmentation, and 3D instance segmentation into PanoSSC framework\nand propose new metrics for evaluating panoptic voxels. Extensive experiments\nshow that our method achieves competitive results on SemanticKITTI semantic\nscene completion benchmark.\n", "date": "2024-06-11", "categories": []}
{"id": "2409.12193", "title": "Vista3D: Unravel the 3D Darkside of a Single Image", "abstract": "  We embark on the age-old quest: unveiling the hidden dimensions of objects\nfrom mere glimpses of their visible parts. To address this, we present Vista3D,\na framework that realizes swift and consistent 3D generation within a mere 5\nminutes. At the heart of Vista3D lies a two-phase approach: the coarse phase\nand the fine phase. In the coarse phase, we rapidly generate initial geometry\nwith Gaussian Splatting from a single image. In the fine phase, we extract a\nSigned Distance Function (SDF) directly from learned Gaussian Splatting,\noptimizing it with a differentiable isosurface representation. Furthermore, it\nelevates the quality of generation by using a disentangled representation with\ntwo independent implicit functions to capture both visible and obscured aspects\nof objects. Additionally, it harmonizes gradients from 2D diffusion prior with\n3D-aware diffusion priors by angular diffusion prior composition. Through\nextensive evaluation, we demonstrate that Vista3D effectively sustains a\nbalance between the consistency and diversity of the generated 3D objects.\nDemos and code will be available at https://github.com/florinshen/Vista3D.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11211", "title": "SplatFields: Neural Gaussian Splats for Sparse 3D and 4D Reconstruction", "abstract": "  Digitizing 3D static scenes and 4D dynamic events from multi-view images has\nlong been a challenge in computer vision and graphics. Recently, 3D Gaussian\nSplatting (3DGS) has emerged as a practical and scalable reconstruction method,\ngaining popularity due to its impressive reconstruction quality, real-time\nrendering capabilities, and compatibility with widely used visualization tools.\nHowever, the method requires a substantial number of input views to achieve\nhigh-quality scene reconstruction, introducing a significant practical\nbottleneck. This challenge is especially severe in capturing dynamic scenes,\nwhere deploying an extensive camera array can be prohibitively costly. In this\nwork, we identify the lack of spatial autocorrelation of splat features as one\nof the factors contributing to the suboptimal performance of the 3DGS technique\nin sparse reconstruction settings. To address the issue, we propose an\noptimization strategy that effectively regularizes splat features by modeling\nthem as the outputs of a corresponding implicit neural field. This results in a\nconsistent enhancement of reconstruction quality across various scenarios. Our\napproach effectively handles static and dynamic cases, as demonstrated by\nextensive testing across different setups and scene complexities.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.11307", "title": "GS-Net: Generalizable Plug-and-Play 3D Gaussian Splatting Module", "abstract": "  3D Gaussian Splatting (3DGS) integrates the strengths of primitive-based\nrepresentations and volumetric rendering techniques, enabling real-time,\nhigh-quality rendering. However, 3DGS models typically overfit to single-scene\ntraining and are highly sensitive to the initialization of Gaussian ellipsoids,\nheuristically derived from Structure from Motion (SfM) point clouds, which\nlimits both generalization and practicality. To address these limitations, we\npropose GS-Net, a generalizable, plug-and-play 3DGS module that densifies\nGaussian ellipsoids from sparse SfM point clouds, enhancing geometric structure\nrepresentation. To the best of our knowledge, GS-Net is the first plug-and-play\n3DGS module with cross-scene generalization capabilities. Additionally, we\nintroduce the CARLA-NVS dataset, which incorporates additional camera\nviewpoints to thoroughly evaluate reconstruction and rendering quality.\nExtensive experiments demonstrate that applying GS-Net to 3DGS yields a PSNR\nimprovement of 2.08 dB for conventional viewpoints and 1.86 dB for novel\nviewpoints, confirming the method's effectiveness and robustness.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.10041", "title": "DENSER: 3D Gaussians Splatting for Scene Reconstruction of Dynamic Urban\n  Environments", "abstract": "  This paper presents DENSER, an efficient and effective approach leveraging 3D\nGaussian splatting (3DGS) for the reconstruction of dynamic urban environments.\nWhile several methods for photorealistic scene representations, both implicitly\nusing neural radiance fields (NeRF) and explicitly using 3DGS have shown\npromising results in scene reconstruction of relatively complex dynamic scenes,\nmodeling the dynamic appearance of foreground objects tend to be challenging,\nlimiting the applicability of these methods to capture subtleties and details\nof the scenes, especially far dynamic objects. To this end, we propose DENSER,\na framework that significantly enhances the representation of dynamic objects\nand accurately models the appearance of dynamic objects in the driving scene.\nInstead of directly using Spherical Harmonics (SH) to model the appearance of\ndynamic objects, we introduce and integrate a new method aiming at dynamically\nestimating SH bases using wavelets, resulting in better representation of\ndynamic objects appearance in both space and time. Besides object appearance,\nDENSER enhances object shape representation through densification of its point\ncloud across multiple scene frames, resulting in faster convergence of model\ntraining. Extensive evaluations on KITTI dataset show that the proposed\napproach significantly outperforms state-of-the-art methods by a wide margin.\nSource codes and models will be uploaded to this repository\nhttps://github.com/sntubix/denser\n", "date": "2024-09-16", "categories": []}
{"id": "2409.08207", "title": "VI3DRM:Towards meticulous 3D Reconstruction from Sparse Views via\n  Photo-Realistic Novel View Synthesis", "abstract": "  Recently, methods like Zero-1-2-3 have focused on single-view based 3D\nreconstruction and have achieved remarkable success. However, their predictions\nfor unseen areas heavily rely on the inductive bias of large-scale pretrained\ndiffusion models. Although subsequent work, such as DreamComposer, attempts to\nmake predictions more controllable by incorporating additional views, the\nresults remain unrealistic due to feature entanglement in the vanilla latent\nspace, including factors such as lighting, material, and structure. To address\nthese issues, we introduce the Visual Isotropy 3D Reconstruction Model\n(VI3DRM), a diffusion-based sparse views 3D reconstruction model that operates\nwithin an ID consistent and perspective-disentangled 3D latent space. By\nfacilitating the disentanglement of semantic information, color, material\nproperties and lighting, VI3DRM is capable of generating highly realistic\nimages that are indistinguishable from real photographs. By leveraging both\nreal and synthesized images, our approach enables the accurate construction of\npointmaps, ultimately producing finely textured meshes or point clouds. On the\nNVS task, tested on the GSO dataset, VI3DRM significantly outperforms\nstate-of-the-art method DreamComposer, achieving a PSNR of 38.61, an SSIM of\n0.929, and an LPIPS of 0.027. Code will be made available upon publication.\n", "date": "2024-09-12", "categories": []}
{"id": "2309.07891", "title": "HandNeRF: Learning to Reconstruct Hand-Object Interaction Scene from a\n  Single RGB Image", "abstract": "  This paper presents a method to learn hand-object interaction prior for\nreconstructing a 3D hand-object scene from a single RGB image. The inference as\nwell as training-data generation for 3D hand-object scene reconstruction is\nchallenging due to the depth ambiguity of a single image and occlusions by the\nhand and object. We turn this challenge into an opportunity by utilizing the\nhand shape to constrain the possible relative configuration of the hand and\nobject geometry. We design a generalizable implicit function, HandNeRF, that\nexplicitly encodes the correlation of the 3D hand shape features and 2D object\nfeatures to predict the hand and object scene geometry. With experiments on\nreal-world datasets, we show that HandNeRF is able to reconstruct hand-object\nscenes of novel grasp configurations more accurately than comparable methods.\nMoreover, we demonstrate that object reconstruction from HandNeRF ensures more\naccurate execution of downstream tasks, such as grasping and motion planning\nfor robotic hand-over and manipulation. Homepage:\nhttps://samsunglabs.github.io/HandNeRF-project-page/\n", "date": "2023-09-14", "categories": []}
{"id": "2404.10772", "title": "Gaussian Opacity Fields: Efficient Adaptive Surface Reconstruction in\n  Unbounded Scenes", "abstract": "  Recently, 3D Gaussian Splatting (3DGS) has demonstrated impressive novel view\nsynthesis results, while allowing the rendering of high-resolution images in\nreal-time. However, leveraging 3D Gaussians for surface reconstruction poses\nsignificant challenges due to the explicit and disconnected nature of 3D\nGaussians. In this work, we present Gaussian Opacity Fields (GOF), a novel\napproach for efficient, high-quality, and adaptive surface reconstruction in\nunbounded scenes. Our GOF is derived from ray-tracing-based volume rendering of\n3D Gaussians, enabling direct geometry extraction from 3D Gaussians by\nidentifying its levelset, without resorting to Poisson reconstruction or TSDF\nfusion as in previous work. We approximate the surface normal of Gaussians as\nthe normal of the ray-Gaussian intersection plane, enabling the application of\nregularization that significantly enhances geometry. Furthermore, we develop an\nefficient geometry extraction method utilizing Marching Tetrahedra, where the\ntetrahedral grids are induced from 3D Gaussians and thus adapt to the scene's\ncomplexity. Our evaluations reveal that GOF surpasses existing 3DGS-based\nmethods in surface reconstruction and novel view synthesis. Further, it\ncompares favorably to or even outperforms, neural implicit methods in both\nquality and speed.\n", "date": "2024-04-16", "categories": []}
{"id": "2407.02174", "title": "BeNeRF: Neural Radiance Fields from a Single Blurry Image and Event\n  Stream", "abstract": "  Neural implicit representation of visual scenes has attracted a lot of\nattention in recent research of computer vision and graphics. Most prior\nmethods focus on how to reconstruct 3D scene representation from a set of\nimages. In this work, we demonstrate the possibility to recover the neural\nradiance fields (NeRF) from a single blurry image and its corresponding event\nstream. We model the camera motion with a cubic B-Spline in SE(3) space. Both\nthe blurry image and the brightness change within a time interval, can then be\nsynthesized from the 3D scene representation given the 6-DoF poses interpolated\nfrom the cubic B-Spline. Our method can jointly learn both the implicit neural\nscene representation and recover the camera motion by minimizing the\ndifferences between the synthesized data and the real measurements without\npre-computed camera poses from COLMAP. We evaluate the proposed method with\nboth synthetic and real datasets. The experimental results demonstrate that we\nare able to render view-consistent latent sharp images from the learned NeRF\nand bring a blurry image alive in high quality. Code and data are available at\nhttps://github.com/wu-cvgl/BeNeRF.\n", "date": "2024-07-02", "categories": []}
{"id": "2409.07245", "title": "Single-View 3D Reconstruction via SO(2)-Equivariant Gaussian Sculpting\n  Networks", "abstract": "  This paper introduces SO(2)-Equivariant Gaussian Sculpting Networks (GSNs) as\nan approach for SO(2)-Equivariant 3D object reconstruction from single-view\nimage observations.\n  GSNs take a single observation as input to generate a Gaussian splat\nrepresentation describing the observed object's geometry and texture. By using\na shared feature extractor before decoding Gaussian colors, covariances,\npositions, and opacities, GSNs achieve extremely high throughput (>150FPS).\nExperiments demonstrate that GSNs can be trained efficiently using a multi-view\nrendering loss and are competitive, in quality, with expensive diffusion-based\nreconstruction algorithms. The GSN model is validated on multiple benchmark\nexperiments. Moreover, we demonstrate the potential for GSNs to be used within\na robotic manipulation pipeline for object-centric grasping.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.07456", "title": "Self-Evolving Depth-Supervised 3D Gaussian Splatting from Rendered\n  Stereo Pairs", "abstract": "  3D Gaussian Splatting (GS) significantly struggles to accurately represent\nthe underlying 3D scene geometry, resulting in inaccuracies and floating\nartifacts when rendering depth maps. In this paper, we address this limitation,\nundertaking a comprehensive analysis of the integration of depth priors\nthroughout the optimization process of Gaussian primitives, and present a novel\nstrategy for this purpose. This latter dynamically exploits depth cues from a\nreadily available stereo network, processing virtual stereo pairs rendered by\nthe GS model itself during training and achieving consistent self-improvement\nof the scene representation. Experimental results on three popular datasets,\nbreaking ground as the first to assess depth accuracy for these models,\nvalidate our findings.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.06407", "title": "Sources of Uncertainty in 3D Scene Reconstruction", "abstract": "  The process of 3D scene reconstruction can be affected by numerous\nuncertainty sources in real-world scenes. While Neural Radiance Fields (NeRFs)\nand 3D Gaussian Splatting (GS) achieve high-fidelity rendering, they lack\nbuilt-in mechanisms to directly address or quantify uncertainties arising from\nthe presence of noise, occlusions, confounding outliers, and imprecise camera\npose inputs. In this paper, we introduce a taxonomy that categorizes different\nsources of uncertainty inherent in these methods. Moreover, we extend NeRF- and\nGS-based methods with uncertainty estimation techniques, including learning\nuncertainty outputs and ensembles, and perform an empirical study to assess\ntheir ability to capture the sensitivity of the reconstruction. Our study\nhighlights the need for addressing various uncertainty aspects when designing\nNeRF/GS-based methods for uncertainty-aware 3D reconstruction.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.06685", "title": "GigaGS: Scaling up Planar-Based 3D Gaussians for Large Scene Surface\n  Reconstruction", "abstract": "  3D Gaussian Splatting (3DGS) has shown promising performance in novel view\nsynthesis. Previous methods adapt it to obtaining surfaces of either individual\n3D objects or within limited scenes. In this paper, we make the first attempt\nto tackle the challenging task of large-scale scene surface reconstruction.\nThis task is particularly difficult due to the high GPU memory consumption,\ndifferent levels of details for geometric representation, and noticeable\ninconsistencies in appearance. To this end, we propose GigaGS, the first work\nfor high-quality surface reconstruction for large-scale scenes using 3DGS.\nGigaGS first applies a partitioning strategy based on the mutual visibility of\nspatial regions, which effectively grouping cameras for parallel processing. To\nenhance the quality of the surface, we also propose novel multi-view\nphotometric and geometric consistency constraints based on Level-of-Detail\nrepresentation. In doing so, our method can reconstruct detailed surface\nstructures. Comprehensive experiments are conducted on various datasets. The\nconsistent improvement demonstrates the superiority of GigaGS.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.05310", "title": "Neural Surface Reconstruction and Rendering for LiDAR-Visual Systems", "abstract": "  This paper presents a unified surface reconstruction and rendering framework\nfor LiDAR-visual systems, integrating Neural Radiance Fields (NeRF) and Neural\nDistance Fields (NDF) to recover both appearance and structural information\nfrom posed images and point clouds. We address the structural visible gap\nbetween NeRF and NDF by utilizing a visible-aware occupancy map to classify\nspace into the free, occupied, visible unknown, and background regions. This\nclassification facilitates the recovery of a complete appearance and structure\nof the scene. We unify the training of the NDF and NeRF using a spatial-varying\nscale SDF-to-density transformation for levels of detail for both structure and\nappearance. The proposed method leverages the learned NDF for structure-aware\nNeRF training by an adaptive sphere tracing sampling strategy for accurate\nstructure rendering. In return, NeRF further refines structural in recovering\nmissing or fuzzy structures in the NDF. Extensive experiments demonstrate the\nsuperior quality and versatility of the proposed method across various\nscenarios. To benefit the community, the codes will be released at\n\\url{https://github.com/hku-mars/M2Mapping}.\n", "date": "2024-09-09", "categories": []}
{"id": "2407.13520", "title": "EaDeblur-GS: Event assisted 3D Deblur Reconstruction with Gaussian\n  Splatting", "abstract": "  3D deblurring reconstruction techniques have recently seen significant\nadvancements with the development of Neural Radiance Fields (NeRF) and 3D\nGaussian Splatting (3DGS). Although these techniques can recover relatively\nclear 3D reconstructions from blurry image inputs, they still face limitations\nin handling severe blurring and complex camera motion. To address these issues,\nwe propose Event-assisted 3D Deblur Reconstruction with Gaussian Splatting\n(EaDeblur-GS), which integrates event camera data to enhance the robustness of\n3DGS against motion blur. By employing an Adaptive Deviation Estimator (ADE)\nnetwork to estimate Gaussian center deviations and using novel loss functions,\nEaDeblur-GS achieves sharp 3D reconstructions in real-time, demonstrating\nperformance comparable to state-of-the-art methods.\n", "date": "2024-07-18", "categories": []}
{"id": "2409.03061", "title": "Incorporating dense metric depth into neural 3D representations for view\n  synthesis and relighting", "abstract": "  Synthesizing accurate geometry and photo-realistic appearance of small scenes\nis an active area of research with compelling use cases in gaming, virtual\nreality, robotic-manipulation, autonomous driving, convenient product capture,\nand consumer-level photography. When applying scene geometry and appearance\nestimation techniques to robotics, we found that the narrow cone of possible\nviewpoints due to the limited range of robot motion and scene clutter caused\ncurrent estimation techniques to produce poor quality estimates or even fail.\nOn the other hand, in robotic applications, dense metric depth can often be\nmeasured directly using stereo and illumination can be controlled. Depth can\nprovide a good initial estimate of the object geometry to improve\nreconstruction, while multi-illumination images can facilitate relighting. In\nthis work we demonstrate a method to incorporate dense metric depth into the\ntraining of neural 3D representations and address an artifact observed while\njointly refining geometry and appearance by disambiguating between texture and\ngeometry edges. We also discuss a multi-flash stereo camera system developed to\ncapture the necessary data for our pipeline and show results on relighting and\nview synthesis with a few training views.\n", "date": "2024-09-04", "categories": []}
{"id": "2409.03213", "title": "Optimizing 3D Gaussian Splatting for Sparse Viewpoint Scene\n  Reconstruction", "abstract": "  3D Gaussian Splatting (3DGS) has emerged as a promising approach for 3D scene\nrepresentation, offering a reduction in computational overhead compared to\nNeural Radiance Fields (NeRF). However, 3DGS is susceptible to high-frequency\nartifacts and demonstrates suboptimal performance under sparse viewpoint\nconditions, thereby limiting its applicability in robotics and computer vision.\nTo address these limitations, we introduce SVS-GS, a novel framework for Sparse\nViewpoint Scene reconstruction that integrates a 3D Gaussian smoothing filter\nto suppress artifacts. Furthermore, our approach incorporates a Depth Gradient\nProfile Prior (DGPP) loss with a dynamic depth mask to sharpen edges and 2D\ndiffusion with Score Distillation Sampling (SDS) loss to enhance geometric\nconsistency in novel view synthesis. Experimental evaluations on the\nMipNeRF-360 and SeaThru-NeRF datasets demonstrate that SVS-GS markedly improves\n3D reconstruction from sparse viewpoints, offering a robust and efficient\nsolution for scene understanding in robotics and computer vision applications.\n", "date": "2024-09-05", "categories": []}
{"id": "2408.14279", "title": "Learning Local Pattern Modularization for Point Cloud Reconstruction\n  from Unseen Classes", "abstract": "  It is challenging to reconstruct 3D point clouds in unseen classes from\nsingle 2D images. Instead of object-centered coordinate system, current methods\ngeneralized global priors learned in seen classes to reconstruct 3D shapes from\nunseen classes in viewer-centered coordinate system. However, the\nreconstruction accuracy and interpretability are still eager to get improved.\nTo resolve this issue, we introduce to learn local pattern modularization for\nreconstructing 3D shapes in unseen classes, which achieves both good\ngeneralization ability and high reconstruction accuracy. Our insight is to\nlearn a local prior which is class-agnostic and easy to generalize in\nobject-centered coordinate system. Specifically, the local prior is learned via\na process of learning and customizing local pattern modularization in seen\nclasses. During this process, we first learn a set of patterns in local\nregions, which is the basis in the object-centered coordinate system to\nrepresent an arbitrary region on shapes across different classes. Then, we\nmodularize each region on an initially reconstructed shape using the learned\nlocal patterns. Based on that, we customize the local pattern modularization\nusing the input image by refining the reconstruction with more details. Our\nmethod enables to reconstruct high fidelity point clouds from unseen classes in\nobject-centered coordinate system without requiring a large number of patterns\nor any additional information, such as segmentation supervision or camera\nposes. Our experimental results under widely used benchmarks show that our\nmethod achieves the state-of-the-art reconstruction accuracy for shapes from\nunseen classes. The code is available at https://github.com/chenchao15/Unseen.\n", "date": "2024-08-26", "categories": []}
{"id": "2403.10814", "title": "DarkGS: Learning Neural Illumination and 3D Gaussians Relighting for\n  Robotic Exploration in the Dark", "abstract": "  Humans have the remarkable ability to construct consistent mental models of\nan environment, even under limited or varying levels of illumination. We wish\nto endow robots with this same capability. In this paper, we tackle the\nchallenge of constructing a photorealistic scene representation under poorly\nilluminated conditions and with a moving light source. We approach the task of\nmodeling illumination as a learning problem, and utilize the developed\nillumination model to aid in scene reconstruction. We introduce an innovative\nframework that uses a data-driven approach, Neural Light Simulators (NeLiS), to\nmodel and calibrate the camera-light system. Furthermore, we present DarkGS, a\nmethod that applies NeLiS to create a relightable 3D Gaussian scene model\ncapable of real-time, photorealistic rendering from novel viewpoints. We show\nthe applicability and robustness of our proposed simulator and system in a\nvariety of real-world environments.\n", "date": "2024-03-16", "categories": []}
{"id": "2312.00583", "title": "DeformGS: Scene Flow in Highly Deformable Scenes for Deformable Object\n  Manipulation", "abstract": "  Teaching robots to fold, drape, or reposition deformable objects such as\ncloth will unlock a variety of automation applications. While remarkable\nprogress has been made for rigid object manipulation, manipulating deformable\nobjects poses unique challenges, including frequent occlusions,\ninfinite-dimensional state spaces and complex dynamics. Just as object pose\nestimation and tracking have aided robots for rigid manipulation, dense 3D\ntracking (scene flow) of highly deformable objects will enable new applications\nin robotics while aiding existing approaches, such as imitation learning or\ncreating digital twins with real2sim transfer. We propose DeformGS, an approach\nto recover scene flow in highly deformable scenes, using simultaneous video\ncaptures of a dynamic scene from multiple cameras. DeformGS builds on recent\nadvances in Gaussian splatting, a method that learns the properties of a large\nnumber of Gaussians for state-of-the-art and fast novel-view synthesis.\nDeformGS learns a deformation function to project a set of Gaussians with\ncanonical properties into world space. The deformation function uses a\nneural-voxel encoding and a multilayer perceptron (MLP) to infer Gaussian\nposition, rotation, and a shadow scalar. We enforce physics-inspired\nregularization terms based on conservation of momentum and isometry, which\nleads to trajectories with smaller trajectory errors. We also leverage existing\nfoundation models SAM and XMEM to produce noisy masks, and learn a per-Gaussian\nmask for better physics-inspired regularization. DeformGS achieves high-quality\n3D tracking on highly deformable scenes with shadows and occlusions. In\nexperiments, DeformGS improves 3D tracking by an average of 55.8% compared to\nthe state-of-the-art. With sufficient texture, DeformGS achieves a median\ntracking error of 3.3 mm on a cloth of 1.5 x 1.5 m in area. Website:\nhttps://deformgs.github.io\n", "date": "2023-11-30", "categories": []}
{"id": "2403.14628", "title": "Zero-Shot Multi-Object Scene Completion", "abstract": "  We present a 3D scene completion method that recovers the complete geometry\nof multiple unseen objects in complex scenes from a single RGB-D image. Despite\nnotable advancements in single-object 3D shape completion, high-quality\nreconstructions in highly cluttered real-world multi-object scenes remains a\nchallenge. To address this issue, we propose OctMAE, an architecture that\nleverages an Octree U-Net and a latent 3D MAE to achieve high-quality and near\nreal-time multi-object scene completion through both local and global geometric\nreasoning. Because a naive 3D MAE can be computationally intractable and memory\nintensive even in the latent space, we introduce a novel occlusion masking\nstrategy and adopt 3D rotary embeddings, which significantly improves the\nruntime and scene completion quality. To generalize to a wide range of objects\nin diverse scenes, we create a large-scale photorealistic dataset, featuring a\ndiverse set of 12K 3D object models from the Objaverse dataset which are\nrendered in multi-object scenes with physics-based positioning. Our method\noutperforms the current state-of-the-art on both synthetic and real-world\ndatasets and demonstrates a strong zero-shot capability.\n", "date": "2024-03-21", "categories": []}
{"id": "2405.03417", "title": "Gaussian Splatting: 3D Reconstruction and Novel View Synthesis, a Review", "abstract": "  Image-based 3D reconstruction is a challenging task that involves inferring\nthe 3D shape of an object or scene from a set of input images. Learning-based\nmethods have gained attention for their ability to directly estimate 3D shapes.\nThis review paper focuses on state-of-the-art techniques for 3D reconstruction,\nincluding the generation of novel, unseen views. An overview of recent\ndevelopments in the Gaussian Splatting method is provided, covering input\ntypes, model structures, output representations, and training strategies.\nUnresolved challenges and future directions are also discussed. Given the rapid\nprogress in this domain and the numerous opportunities for enhancing 3D\nreconstruction methods, a comprehensive examination of algorithms appears\nessential. Consequently, this study offers a thorough overview of the latest\nadvancements in Gaussian Splatting.\n", "date": "2024-05-06", "categories": []}
{"id": "2408.16982", "title": "2DGH: 2D Gaussian-Hermite Splatting for High-quality Rendering and\n  Better Geometry Reconstruction", "abstract": "  2D Gaussian Splatting has recently emerged as a significant method in 3D\nreconstruction, enabling novel view synthesis and geometry reconstruction\nsimultaneously. While the well-known Gaussian kernel is broadly used, its lack\nof anisotropy and deformation ability leads to dim and vague edges at object\nsilhouettes, limiting the reconstruction quality of current Gaussian splatting\nmethods. To enhance the representation power, we draw inspiration from quantum\nphysics and propose to use the Gaussian-Hermite kernel as the new primitive in\nGaussian splatting. The new kernel takes a unified mathematical form and\nextends the Gaussian function, which serves as the zero-rank term in the\nupdated formulation. Our experiments demonstrate the extraordinary performance\nof Gaussian-Hermite kernel in both geometry reconstruction and novel-view\nsynthesis tasks. The proposed kernel outperforms traditional Gaussian Splatting\nkernels, showcasing its potential for high-quality 3D reconstruction and\nrendering.\n", "date": "2024-08-30", "categories": []}
{"id": "2408.17223", "title": "OG-Mapping: Octree-based Structured 3D Gaussians for Online Dense\n  Mapping", "abstract": "  3D Gaussian splatting (3DGS) has recently demonstrated promising advancements\nin RGB-D online dense mapping. Nevertheless, existing methods excessively rely\non per-pixel depth cues to perform map densification, which leads to\nsignificant redundancy and increased sensitivity to depth noise. Additionally,\nexplicitly storing 3D Gaussian parameters of room-scale scene poses a\nsignificant storage challenge. In this paper, we introduce OG-Mapping, which\nleverages the robust scene structural representation capability of sparse\noctrees, combined with structured 3D Gaussian representations, to achieve\nefficient and robust online dense mapping. Moreover, OG-Mapping employs an\nanchor-based progressive map refinement strategy to recover the scene\nstructures at multiple levels of detail. Instead of maintaining a small number\nof active keyframes with a fixed keyframe window as previous approaches do, a\ndynamic keyframe window is employed to allow OG-Mapping to better tackle false\nlocal minima and forgetting issues. Experimental results demonstrate that\nOG-Mapping delivers more robust and superior realism mapping results than\nexisting Gaussian-based RGB-D online mapping methods with a compact model, and\nno additional post-processing is required.\n", "date": "2024-08-30", "categories": []}
{"id": "2408.16061", "title": "3D Reconstruction with Spatial Memory", "abstract": "  We present Spann3R, a novel approach for dense 3D reconstruction from ordered\nor unordered image collections. Built on the DUSt3R paradigm, Spann3R uses a\ntransformer-based architecture to directly regress pointmaps from images\nwithout any prior knowledge of the scene or camera parameters. Unlike DUSt3R,\nwhich predicts per image-pair pointmaps each expressed in its local coordinate\nframe, Spann3R can predict per-image pointmaps expressed in a global coordinate\nsystem, thus eliminating the need for optimization-based global alignment. The\nkey idea of Spann3R is to manage an external spatial memory that learns to keep\ntrack of all previous relevant 3D information. Spann3R then queries this\nspatial memory to predict the 3D structure of the next frame in a global\ncoordinate system. Taking advantage of DUSt3R's pre-trained weights, and\nfurther fine-tuning on a subset of datasets, Spann3R shows competitive\nperformance and generalization ability on various unseen datasets and can\nprocess ordered image collections in real time. Project page:\n\\url{https://hengyiwang.github.io/projects/spanner}\n", "date": "2024-08-28", "categories": []}
{"id": "2403.17550", "title": "DeepMIF: Deep Monotonic Implicit Fields for Large-Scale LiDAR 3D Mapping", "abstract": "  Recently, significant progress has been achieved in sensing real large-scale\noutdoor 3D environments, particularly by using modern acquisition equipment\nsuch as LiDAR sensors. Unfortunately, they are fundamentally limited in their\nability to produce dense, complete 3D scenes. To address this issue, recent\nlearning-based methods integrate neural implicit representations and\noptimizable feature grids to approximate surfaces of 3D scenes. However,\nnaively fitting samples along raw LiDAR rays leads to noisy 3D mapping results\ndue to the nature of sparse, conflicting LiDAR measurements. Instead, in this\nwork we depart from fitting LiDAR data exactly, instead letting the network\noptimize a non-metric monotonic implicit field defined in 3D space. To fit our\nfield, we design a learning system integrating a monotonicity loss that enables\noptimizing neural monotonic fields and leverages recent progress in large-scale\n3D mapping. Our algorithm achieves high-quality dense 3D mapping performance as\ncaptured by multiple quantitative and perceptual measures and visual results\nobtained for Mai City, Newer College, and KITTI benchmarks. The code of our\napproach will be made publicly available.\n", "date": "2024-03-26", "categories": []}
{"id": "2404.15891", "title": "OMEGAS: Object Mesh Extraction from Large Scenes Guided by Gaussian\n  Segmentation", "abstract": "  Recent advancements in 3D reconstruction technologies have paved the way for\nhigh-quality and real-time rendering of complex 3D scenes. Despite these\nachievements, a notable challenge persists: it is difficult to precisely\nreconstruct specific objects from large scenes. Current scene reconstruction\ntechniques frequently result in the loss of object detail textures and are\nunable to reconstruct object portions that are occluded or unseen in views. To\naddress this challenge, we delve into the meticulous 3D reconstruction of\nspecific objects within large scenes and propose a framework termed OMEGAS:\nObject Mesh Extraction from Large Scenes Guided by Gaussian Segmentation.\nSpecifically, we proposed a novel 3D target segmentation technique based on 2D\nGaussian Splatting, which segments 3D consistent target masks in multi-view\nscene images and generates a preliminary target model. Moreover, to reconstruct\nthe unseen portions of the target, we propose a novel target replenishment\ntechnique driven by large-scale generative diffusion priors. We demonstrate\nthat our method can accurately reconstruct specific targets from large scenes,\nboth quantitatively and qualitatively. Our experiments show that OMEGAS\nsignificantly outperforms existing reconstruction methods across various\nscenarios. Our project page is at: https://github.com/CrystalWlz/OMEGAS\n", "date": "2024-04-24", "categories": []}
{"id": "2408.13912", "title": "Splatt3R: Zero-shot Gaussian Splatting from Uncalibrated Image Pairs", "abstract": "  In this paper, we introduce Splatt3R, a pose-free, feed-forward method for\nin-the-wild 3D reconstruction and novel view synthesis from stereo pairs. Given\nuncalibrated natural images, Splatt3R can predict 3D Gaussian Splats without\nrequiring any camera parameters or depth information. For generalizability, we\nbuild Splatt3R upon a ``foundation'' 3D geometry reconstruction method, MASt3R,\nby extending it to deal with both 3D structure and appearance. Specifically,\nunlike the original MASt3R which reconstructs only 3D point clouds, we predict\nthe additional Gaussian attributes required to construct a Gaussian primitive\nfor each point. Hence, unlike other novel view synthesis methods, Splatt3R is\nfirst trained by optimizing the 3D point cloud's geometry loss, and then a\nnovel view synthesis objective. By doing this, we avoid the local minima\npresent in training 3D Gaussian Splats from stereo views. We also propose a\nnovel loss masking strategy that we empirically find is critical for strong\nperformance on extrapolated viewpoints. We train Splatt3R on the ScanNet++\ndataset and demonstrate excellent generalisation to uncalibrated, in-the-wild\nimages. Splatt3R can reconstruct scenes at 4FPS at 512 x 512 resolution, and\nthe resultant splats can be rendered in real-time.\n", "date": "2024-08-25", "categories": []}
{"id": "2408.15524", "title": "Ray-Distance Volume Rendering for Neural Scene Reconstruction", "abstract": "  Existing methods in neural scene reconstruction utilize the Signed Distance\nFunction (SDF) to model the density function. However, in indoor scenes, the\ndensity computed from the SDF for a sampled point may not consistently reflect\nits real importance in volume rendering, often due to the influence of\nneighboring objects. To tackle this issue, our work proposes a novel approach\nfor indoor scene reconstruction, which instead parameterizes the density\nfunction with the Signed Ray Distance Function (SRDF). Firstly, the SRDF is\npredicted by the network and transformed to a ray-conditioned density function\nfor volume rendering. We argue that the ray-specific SRDF only considers the\nsurface along the camera ray, from which the derived density function is more\nconsistent to the real occupancy than that from the SDF. Secondly, although\nSRDF and SDF represent different aspects of scene geometries, their values\nshould share the same sign indicating the underlying spatial occupancy.\nTherefore, this work introduces a SRDF-SDF consistency loss to constrain the\nsigns of the SRDF and SDF outputs. Thirdly, this work proposes a\nself-supervised visibility task, introducing the physical visibility geometry\nto the reconstruction task. The visibility task combines prior from predicted\nSRDF and SDF as pseudo labels, and contributes to generating more accurate 3D\ngeometry. Our method implemented with different representations has been\nvalidated on indoor datasets, achieving improved performance in both\nreconstruction and view synthesis.\n", "date": "2024-08-28", "categories": []}
{"id": "2408.15608", "title": "Geometry-guided Feature Learning and Fusion for Indoor Scene\n  Reconstruction", "abstract": "  In addition to color and textural information, geometry provides important\ncues for 3D scene reconstruction. However, current reconstruction methods only\ninclude geometry at the feature level thus not fully exploiting the geometric\ninformation.\n  In contrast, this paper proposes a novel geometry integration mechanism for\n3D scene reconstruction. Our approach incorporates 3D geometry at three levels,\ni.e. feature learning, feature fusion, and network supervision. First,\ngeometry-guided feature learning encodes geometric priors to contain\nview-dependent information. Second, a geometry-guided adaptive feature fusion\nis introduced which utilizes the geometric priors as a guidance to adaptively\ngenerate weights for multiple views. Third, at the supervision level, taking\nthe consistency between 2D and 3D normals into account, a consistent 3D normal\nloss is designed to add local constraints.\n  Large-scale experiments are conducted on the ScanNet dataset, showing that\nvolumetric methods with our geometry integration mechanism outperform\nstate-of-the-art methods quantitatively as well as qualitatively. Volumetric\nmethods with ours also show good generalization on the 7-Scenes and TUM RGB-D\ndatasets.\n", "date": "2024-08-28", "categories": []}
{"id": "2408.11413", "title": "Pano2Room: Novel View Synthesis from a Single Indoor Panorama", "abstract": "  Recent single-view 3D generative methods have made significant advancements\nby leveraging knowledge distilled from extensive 3D object datasets. However,\nchallenges persist in the synthesis of 3D scenes from a single view, primarily\ndue to the complexity of real-world environments and the limited availability\nof high-quality prior resources. In this paper, we introduce a novel approach\ncalled Pano2Room, designed to automatically reconstruct high-quality 3D indoor\nscenes from a single panoramic image. These panoramic images can be easily\ngenerated using a panoramic RGBD inpainter from captures at a single location\nwith any camera. The key idea is to initially construct a preliminary mesh from\nthe input panorama, and iteratively refine this mesh using a panoramic RGBD\ninpainter while collecting photo-realistic 3D-consistent pseudo novel views.\nFinally, the refined mesh is converted into a 3D Gaussian Splatting field and\ntrained with the collected pseudo novel views. This pipeline enables the\nreconstruction of real-world 3D scenes, even in the presence of large\nocclusions, and facilitates the synthesis of photo-realistic novel views with\ndetailed geometry. Extensive qualitative and quantitative experiments have been\nconducted to validate the superiority of our method in single-panorama indoor\nnovel synthesis compared to the state-of-the-art. Our code and data are\navailable at \\url{https://github.com/TrickyGo/Pano2Room}.\n", "date": "2024-08-21", "categories": []}
{"id": "2408.13770", "title": "TranSplat: Generalizable 3D Gaussian Splatting from Sparse Multi-View\n  Images with Transformers", "abstract": "  Compared with previous 3D reconstruction methods like Nerf, recent\nGeneralizable 3D Gaussian Splatting (G-3DGS) methods demonstrate impressive\nefficiency even in the sparse-view setting. However, the promising\nreconstruction performance of existing G-3DGS methods relies heavily on\naccurate multi-view feature matching, which is quite challenging. Especially\nfor the scenes that have many non-overlapping areas between various views and\ncontain numerous similar regions, the matching performance of existing methods\nis poor and the reconstruction precision is limited. To address this problem,\nwe develop a strategy that utilizes a predicted depth confidence map to guide\naccurate local feature matching. In addition, we propose to utilize the\nknowledge of existing monocular depth estimation models as prior to boost the\ndepth estimation precision in non-overlapping areas between views. Combining\nthe proposed strategies, we present a novel G-3DGS method named TranSplat,\nwhich obtains the best performance on both the RealEstate10K and ACID\nbenchmarks while maintaining competitive speed and presenting strong\ncross-dataset generalization ability. Our code, and demos will be available at:\nhttps://xingyoujun.github.io/transplat.\n", "date": "2024-08-25", "categories": []}
{"id": "2408.13972", "title": "DynaSurfGS: Dynamic Surface Reconstruction with Planar-based Gaussian\n  Splatting", "abstract": "  Dynamic scene reconstruction has garnered significant attention in recent\nyears due to its capabilities in high-quality and real-time rendering. Among\nvarious methodologies, constructing a 4D spatial-temporal representation, such\nas 4D-GS, has gained popularity for its high-quality rendered images. However,\nthese methods often produce suboptimal surfaces, as the discrete 3D Gaussian\npoint clouds fail to align with the object's surface precisely. To address this\nproblem, we propose DynaSurfGS to achieve both photorealistic rendering and\nhigh-fidelity surface reconstruction of dynamic scenarios. Specifically, the\nDynaSurfGS framework first incorporates Gaussian features from 4D neural voxels\nwith the planar-based Gaussian Splatting to facilitate precise surface\nreconstruction. It leverages normal regularization to enforce the smoothness of\nthe surface of dynamic objects. It also incorporates the as-rigid-as-possible\n(ARAP) constraint to maintain the approximate rigidity of local neighborhoods\nof 3D Gaussians between timesteps and ensure that adjacent 3D Gaussians remain\nclosely aligned throughout. Extensive experiments demonstrate that DynaSurfGS\nsurpasses state-of-the-art methods in both high-fidelity surface reconstruction\nand photorealistic rendering.\n", "date": "2024-08-26", "categories": []}
{"id": "2311.09077", "title": "Spiking NeRF: Representing the Real-World Geometry by a Discontinuous\n  Representation", "abstract": "  A crucial reason for the success of existing NeRF-based methods is to build a\nneural density field for the geometry representation via multiple perceptron\nlayers (MLPs). MLPs are continuous functions, however, real geometry or density\nfield is frequently discontinuous at the interface between the air and the\nsurface. Such a contrary brings the problem of unfaithful geometry\nrepresentation. To this end, this paper proposes spiking NeRF, which leverages\nspiking neurons and a hybrid Artificial Neural Network (ANN)-Spiking Neural\nNetwork (SNN) framework to build a discontinuous density field for faithful\ngeometry representation. Specifically, we first demonstrate the reason why\ncontinuous density fields will bring inaccuracy. Then, we propose to use the\nspiking neurons to build a discontinuous density field. We conduct a\ncomprehensive analysis for the problem of existing spiking neuron models and\nthen provide the numerical relationship between the parameter of the spiking\nneuron and the theoretical accuracy of geometry. Based on this, we propose a\nbounded spiking neuron to build the discontinuous density field. Our method\nachieves SOTA performance. The source code and the supplementary material are\navailable at https://github.com/liaozhanfeng/Spiking-NeRF.\n", "date": "2023-11-15", "categories": []}
{"id": "2408.11465", "title": "MeTTA: Single-View to 3D Textured Mesh Reconstruction with Test-Time\n  Adaptation", "abstract": "  Reconstructing 3D from a single view image is a long-standing challenge. One\nof the popular approaches to tackle this problem is learning-based methods, but\ndealing with the test cases unfamiliar with training data (Out-of-distribution;\nOoD) introduces an additional challenge. To adapt for unseen samples in test\ntime, we propose MeTTA, a test-time adaptation (TTA) exploiting generative\nprior. We design joint optimization of 3D geometry, appearance, and pose to\nhandle OoD cases with only a single view image. However, the alignment between\nthe reference image and the 3D shape via the estimated viewpoint could be\nerroneous, which leads to ambiguity. To address this ambiguity, we carefully\ndesign learnable virtual cameras and their self-calibration. In our\nexperiments, we demonstrate that MeTTA effectively deals with OoD scenarios at\nfailure cases of existing learning-based 3D reconstruction models and enables\nobtaining a realistic appearance with physically based rendering (PBR)\ntextures.\n", "date": "2024-08-21", "categories": []}
{"id": "2408.11697", "title": "Robust 3D Gaussian Splatting for Novel View Synthesis in Presence of\n  Distractors", "abstract": "  3D Gaussian Splatting has shown impressive novel view synthesis results;\nnonetheless, it is vulnerable to dynamic objects polluting the input data of an\notherwise static scene, so called distractors. Distractors have severe impact\non the rendering quality as they get represented as view-dependent effects or\nresult in floating artifacts. Our goal is to identify and ignore such\ndistractors during the 3D Gaussian optimization to obtain a clean\nreconstruction. To this end, we take a self-supervised approach that looks at\nthe image residuals during the optimization to determine areas that have likely\nbeen falsified by a distractor. In addition, we leverage a pretrained\nsegmentation network to provide object awareness, enabling more accurate\nexclusion of distractors. This way, we obtain segmentation masks of distractors\nto effectively ignore them in the loss formulation. We demonstrate that our\napproach is robust to various distractors and strongly improves rendering\nquality on distractor-polluted scenes, improving PSNR by 1.86dB compared to 3D\nGaussian Splatting.\n", "date": "2024-08-21", "categories": []}
{"id": "2401.01339", "title": "Street Gaussians: Modeling Dynamic Urban Scenes with Gaussian Splatting", "abstract": "  This paper aims to tackle the problem of modeling dynamic urban streets for\nautonomous driving scenes. Recent methods extend NeRF by incorporating tracked\nvehicle poses to animate vehicles, enabling photo-realistic view synthesis of\ndynamic urban street scenes. However, significant limitations are their slow\ntraining and rendering speed. We introduce Street Gaussians, a new explicit\nscene representation that tackles these limitations. Specifically, the dynamic\nurban scene is represented as a set of point clouds equipped with semantic\nlogits and 3D Gaussians, each associated with either a foreground vehicle or\nthe background. To model the dynamics of foreground object vehicles, each\nobject point cloud is optimized with optimizable tracked poses, along with a 4D\nspherical harmonics model for the dynamic appearance. The explicit\nrepresentation allows easy composition of object vehicles and background, which\nin turn allows for scene editing operations and rendering at 135 FPS (1066\n$\\times$ 1600 resolution) within half an hour of training. The proposed method\nis evaluated on multiple challenging benchmarks, including KITTI and Waymo Open\ndatasets. Experiments show that the proposed method consistently outperforms\nstate-of-the-art methods across all datasets. The code will be released to\nensure reproducibility.\n", "date": "2024-01-02", "categories": []}
{"id": "2407.16600", "title": "DHGS: Decoupled Hybrid Gaussian Splatting for Driving Scene", "abstract": "  Existing Gaussian splatting methods often fall short in achieving\nsatisfactory novel view synthesis in driving scenes, primarily due to the\nabsence of crafty designs and geometric constraints for the involved elements.\nThis paper introduces a novel neural rendering method termed Decoupled Hybrid\nGaussian Splatting (DHGS), targeting at promoting the rendering quality of\nnovel view synthesis for static driving scenes. The novelty of this work lies\nin the decoupled and hybrid pixel-level blender for road and non-road layers,\nwithout the conventional unified differentiable rendering logic for the entire\nscene. Still, consistency and continuity in superimposition are preserved\nthrough the proposed depth-ordered hybrid rendering strategy. Additionally, an\nimplicit road representation comprised of a Signed Distance Function (SDF) is\ntrained to supervise the road surface with subtle geometric attributes.\nAccompanied by the use of auxiliary transmittance loss and consistency loss,\nnovel images with imperceptible boundary and elevated fidelity are ultimately\nobtained. Substantial experiments on the Waymo dataset prove that DHGS\noutperforms the state-of-the-art methods. The project page where more video\nevidences are given is: https://ironbrotherstyle.github.io/dhgs_web.\n", "date": "2024-07-23", "categories": []}
{"id": "2408.10198", "title": "MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction\n  Model", "abstract": "  Open-world 3D reconstruction models have recently garnered significant\nattention. However, without sufficient 3D inductive bias, existing methods\ntypically entail expensive training costs and struggle to extract high-quality\n3D meshes. In this work, we introduce MeshFormer, a sparse-view reconstruction\nmodel that explicitly leverages 3D native structure, input guidance, and\ntraining supervision. Specifically, instead of using a triplane representation,\nwe store features in 3D sparse voxels and combine transformers with 3D\nconvolutions to leverage an explicit 3D structure and projective bias. In\naddition to sparse-view RGB input, we require the network to take input and\ngenerate corresponding normal maps. The input normal maps can be predicted by\n2D diffusion models, significantly aiding in the guidance and refinement of the\ngeometry's learning. Moreover, by combining Signed Distance Function (SDF)\nsupervision with surface rendering, we directly learn to generate high-quality\nmeshes without the need for complex multi-stage training processes. By\nincorporating these explicit 3D biases, MeshFormer can be trained efficiently\nand deliver high-quality textured meshes with fine-grained geometric details.\nIt can also be integrated with 2D diffusion models to enable fast\nsingle-image-to-3D and text-to-3D tasks. Project page:\nhttps://meshformer3d.github.io\n", "date": "2024-08-19", "categories": []}
{"id": "2408.08086", "title": "Single-image coherent reconstruction of objects and humans", "abstract": "  Existing methods for reconstructing objects and humans from a monocular image\nsuffer from severe mesh collisions and performance limitations for interacting\noccluding objects. This paper introduces a method to obtain a globally\nconsistent 3D reconstruction of interacting objects and people from a single\nimage. Our contributions include: 1) an optimization framework, featuring a\ncollision loss, tailored to handle human-object and human-human interactions,\nensuring spatially coherent scene reconstruction; and 2) a novel technique to\nrobustly estimate 6 degrees of freedom (DOF) poses, specifically for heavily\noccluded objects, exploiting image inpainting. Notably, our proposed method\noperates effectively on images from real-world scenarios, without necessitating\nscene or object-level 3D supervision. Extensive qualitative and quantitative\nevaluation against existing methods demonstrates a significant reduction in\ncollisions in the final reconstructions of scenes with multiple interacting\nhumans and objects and a more coherent scene reconstruction.\n", "date": "2024-08-15", "categories": []}
{"id": "2408.07540", "title": "3D Gaussian Editing with A Single Image", "abstract": "  The modeling and manipulation of 3D scenes captured from the real world are\npivotal in various applications, attracting growing research interest. While\nprevious works on editing have achieved interesting results through\nmanipulating 3D meshes, they often require accurately reconstructed meshes to\nperform editing, which limits their application in 3D content generation. To\naddress this gap, we introduce a novel single-image-driven 3D scene editing\napproach based on 3D Gaussian Splatting, enabling intuitive manipulation via\ndirectly editing the content on a 2D image plane. Our method learns to optimize\nthe 3D Gaussians to align with an edited version of the image rendered from a\nuser-specified viewpoint of the original scene. To capture long-range object\ndeformation, we introduce positional loss into the optimization process of 3D\nGaussian Splatting and enable gradient propagation through reparameterization.\nTo handle occluded 3D Gaussians when rendering from the specified viewpoint, we\nbuild an anchor-based structure and employ a coarse-to-fine optimization\nstrategy capable of handling long-range deformation while maintaining\nstructural stability. Furthermore, we design a novel masking strategy to\nadaptively identify non-rigid deformation regions for fine-scale modeling.\nExtensive experiments show the effectiveness of our method in handling\ngeometric details, long-range, and non-rigid deformation, demonstrating\nsuperior editing flexibility and quality compared to previous approaches.\n", "date": "2024-08-14", "categories": []}
{"id": "2309.07640", "title": "Indoor Scene Reconstruction with Fine-Grained Details Using Hybrid\n  Representation and Normal Prior Enhancement", "abstract": "  The reconstruction of indoor scenes from multi-view RGB images is challenging\ndue to the coexistence of flat and texture-less regions alongside delicate and\nfine-grained regions. Recent methods leverage neural radiance fields aided by\npredicted surface normal priors to recover the scene geometry. These methods\nexcel in producing complete and smooth results for floor and wall areas.\nHowever, they struggle to capture complex surfaces with high-frequency\nstructures due to the inadequate neural representation and the inaccurately\npredicted normal priors. This work aims to reconstruct high-fidelity surfaces\nwith fine-grained details by addressing the above limitations. To improve the\ncapacity of the implicit representation, we propose a hybrid architecture to\nrepresent low-frequency and high-frequency regions separately. To enhance the\nnormal priors, we introduce a simple yet effective image sharpening and\ndenoising technique, coupled with a network that estimates the pixel-wise\nuncertainty of the predicted surface normal vectors. Identifying such\nuncertainty can prevent our model from being misled by unreliable surface\nnormal supervisions that hinder the accurate reconstruction of intricate\ngeometries. Experiments on the benchmark datasets show that our method\noutperforms existing methods in terms of reconstruction quality. Furthermore,\nthe proposed method also generalizes well to real-world indoor scenarios\ncaptured by our hand-held mobile phones. Our code is publicly available at:\nhttps://github.com/yec22/Fine-Grained-Indoor-Recon.\n", "date": "2023-09-14", "categories": []}
{"id": "2310.13356", "title": "Sync-NeRF: Generalizing Dynamic NeRFs to Unsynchronized Videos", "abstract": "  Recent advancements in 4D scene reconstruction using neural radiance fields\n(NeRF) have demonstrated the ability to represent dynamic scenes from\nmulti-view videos. However, they fail to reconstruct the dynamic scenes and\nstruggle to fit even the training views in unsynchronized settings. It happens\nbecause they employ a single latent embedding for a frame while the multi-view\nimages at the same frame were actually captured at different moments. To\naddress this limitation, we introduce time offsets for individual\nunsynchronized videos and jointly optimize the offsets with NeRF. By design,\nour method is applicable for various baselines and improves them with large\nmargins. Furthermore, finding the offsets naturally works as synchronizing the\nvideos without manual effort. Experiments are conducted on the common Plenoptic\nVideo Dataset and a newly built Unsynchronized Dynamic Blender Dataset to\nverify the performance of our method. Project page:\nhttps://seoha-kim.github.io/sync-nerf\n", "date": "2023-10-20", "categories": []}
{"id": "2408.06592", "title": "ActiveNeRF: Learning Accurate 3D Geometry by Active Pattern Projection", "abstract": "  NeRFs have achieved incredible success in novel view synthesis. However, the\naccuracy of the implicit geometry is unsatisfactory because the passive static\nenvironmental illumination has low spatial frequency and cannot provide enough\ninformation for accurate geometry reconstruction. In this work, we propose\nActiveNeRF, a 3D geometry reconstruction framework, which improves the geometry\nquality of NeRF by actively projecting patterns of high spatial frequency onto\nthe scene using a projector which has a constant relative pose to the camera.\nWe design a learnable active pattern rendering pipeline which jointly learns\nthe scene geometry and the active pattern. We find that, by adding the active\npattern and imposing its consistency across different views, our proposed\nmethod outperforms state of the art geometry reconstruction methods\nqualitatively and quantitatively in both simulation and real experiments. Code\nis avaliable at https://github.com/hcp16/active_nerf\n", "date": "2024-08-13", "categories": []}
{"id": "2408.05402", "title": "Mesh deformation-based single-view 3D reconstruction of thin eyeglasses\n  frames with differentiable rendering", "abstract": "  With the support of Virtual Reality (VR) and Augmented Reality (AR)\ntechnologies, the 3D virtual eyeglasses try-on application is well on its way\nto becoming a new trending solution that offers a \"try on\" option to select the\nperfect pair of eyeglasses at the comfort of your own home. Reconstructing\neyeglasses frames from a single image with traditional depth and image-based\nmethods is extremely difficult due to their unique characteristics such as lack\nof sufficient texture features, thin elements, and severe self-occlusions. In\nthis paper, we propose the first mesh deformation-based reconstruction\nframework for recovering high-precision 3D full-frame eyeglasses models from a\nsingle RGB image, leveraging prior and domain-specific knowledge. Specifically,\nbased on the construction of a synthetic eyeglasses frame dataset, we first\ndefine a class-specific eyeglasses frame template with pre-defined keypoints.\nThen, given an input eyeglasses frame image with thin structure and few texture\nfeatures, we design a keypoint detector and refiner to detect predefined\nkeypoints in a coarse-to-fine manner to estimate the camera pose accurately.\nAfter that, using differentiable rendering, we propose a novel optimization\napproach for producing correct geometry by progressively performing free-form\ndeformation (FFD) on the template mesh. We define a series of loss functions to\nenforce consistency between the rendered result and the corresponding RGB\ninput, utilizing constraints from inherent structure, silhouettes, keypoints,\nper-pixel shading information, and so on. Experimental results on both the\nsynthetic dataset and real images demonstrate the effectiveness of the proposed\nalgorithm.\n", "date": "2024-08-10", "categories": []}
{"id": "2306.02099", "title": "Enhancing Surface Neural Implicits with Curvature-Guided Sampling and\n  Uncertainty-Augmented Representations", "abstract": "  Neural implicit representations have become a popular choice for modeling\nsurfaces due to their adaptability in resolution and support for complex\ntopology. While previous works have achieved impressive reconstruction quality\nby training on ground truth point clouds or meshes, they often do not discuss\nthe data acquisition and ignore the effect of input quality and sampling\nmethods during reconstruction. In this paper, we introduce a method that\ndirectly digests depth images for the task of high-fidelity 3D reconstruction.\nTo this end, a simple sampling strategy is proposed to generate highly\neffective training data, by incorporating differentiable geometric features\ncomputed directly based on the input depth images with only marginal\ncomputational cost. Due to its simplicity, our sampling strategy can be easily\nincorporated into diverse popular methods, allowing their training process to\nbe more stable and efficient. Despite its simplicity, our method outperforms a\nrange of both classical and learning-based baselines and demonstrates\nstate-of-the-art results in both synthetic and real-world datasets.\n", "date": "2023-06-03", "categories": []}
{"id": "2403.11796", "title": "OpenOcc: Open Vocabulary 3D Scene Reconstruction via Occupancy\n  Representation", "abstract": "  3D reconstruction has been widely used in autonomous navigation fields of\nmobile robotics. However, the former research can only provide the basic\ngeometry structure without the capability of open-world scene understanding,\nlimiting advanced tasks like human interaction and visual navigation. Moreover,\ntraditional 3D scene understanding approaches rely on expensive labeled 3D\ndatasets to train a model for a single task with supervision. Thus, geometric\nreconstruction with zero-shot scene understanding i.e. Open vocabulary 3D\nUnderstanding and Reconstruction, is crucial for the future development of\nmobile robots. In this paper, we propose OpenOcc, a novel framework unifying\nthe 3D scene reconstruction and open vocabulary understanding with neural\nradiance fields. We model the geometric structure of the scene with occupancy\nrepresentation and distill the pre-trained open vocabulary model into a 3D\nlanguage field via volume rendering for zero-shot inference. Furthermore, a\nnovel semantic-aware confidence propagation (SCP) method has been proposed to\nrelieve the issue of language field representation degeneracy caused by\ninconsistent measurements in distilled features. Experimental results show that\nour approach achieves competitive performance in 3D scene understanding tasks,\nespecially for small and long-tail objects.\n", "date": "2024-03-18", "categories": []}
{"id": "2406.08759", "title": "GaussianForest: Hierarchical-Hybrid 3D Gaussian Splatting for Compressed\n  Scene Modeling", "abstract": "  The field of novel-view synthesis has recently witnessed the emergence of 3D\nGaussian Splatting, which represents scenes in a point-based manner and renders\nthrough rasterization. This methodology, in contrast to Radiance Fields that\nrely on ray tracing, demonstrates superior rendering quality and speed.\nHowever, the explicit and unstructured nature of 3D Gaussians poses a\nsignificant storage challenge, impeding its broader application. To address\nthis challenge, we introduce the Gaussian-Forest modeling framework, which\nhierarchically represents a scene as a forest of hybrid 3D Gaussians. Each\nhybrid Gaussian retains its unique explicit attributes while sharing implicit\nones with its sibling Gaussians, thus optimizing parameterization with\nsignificantly fewer variables. Moreover, adaptive growth and pruning strategies\nare designed, ensuring detailed representation in complex regions and a notable\nreduction in the number of required Gaussians. Extensive experiments\ndemonstrate that Gaussian-Forest not only maintains comparable speed and\nquality but also achieves a compression rate surpassing 10 times, marking a\nsignificant advancement in efficient scene modeling. Codes will be available at\nhttps://github.com/Xian-Bei/GaussianForest.\n", "date": "2024-06-13", "categories": []}
{"id": "2404.00875", "title": "DPA-Net: Structured 3D Abstraction from Sparse Views via Differentiable\n  Primitive Assembly", "abstract": "  We present a differentiable rendering framework to learn structured 3D\nabstractions in the form of primitive assemblies from sparse RGB images\ncapturing a 3D object. By leveraging differentiable volume rendering, our\nmethod does not require 3D supervision. Architecturally, our network follows\nthe general pipeline of an image-conditioned neural radiance field (NeRF)\nexemplified by pixelNeRF for color prediction. As our core contribution, we\nintroduce differential primitive assembly (DPA) into NeRF to output a 3D\noccupancy field in place of density prediction, where the predicted occupancies\nserve as opacity values for volume rendering. Our network, coined DPA-Net,\nproduces a union of convexes, each as an intersection of convex quadric\nprimitives, to approximate the target 3D object, subject to an abstraction loss\nand a masking loss, both defined in the image space upon volume rendering. With\ntest-time adaptation and additional sampling and loss designs aimed at\nimproving the accuracy and compactness of the obtained assemblies, our method\ndemonstrates superior performance over state-of-the-art alternatives for 3D\nprimitive abstraction from sparse views.\n", "date": "2024-04-01", "categories": []}
{"id": "2408.01126", "title": "IG-SLAM: Instant Gaussian SLAM", "abstract": "  3D Gaussian Splatting has recently shown promising results as an alternative\nscene representation in SLAM systems to neural implicit representations.\nHowever, current methods either lack dense depth maps to supervise the mapping\nprocess or detailed training designs that consider the scale of the\nenvironment. To address these drawbacks, we present IG-SLAM, a dense RGB-only\nSLAM system that employs robust Dense-SLAM methods for tracking and combines\nthem with Gaussian Splatting. A 3D map of the environment is constructed using\naccurate pose and dense depth provided by tracking. Additionally, we utilize\ndepth uncertainty in map optimization to improve 3D reconstruction. Our decay\nstrategy in map optimization enhances convergence and allows the system to run\nat 10 fps in a single process. We demonstrate competitive performance with\nstate-of-the-art RGB-only SLAM systems while achieving faster operation speeds.\nWe present our experiments on the Replica, TUM-RGBD, ScanNet, and EuRoC\ndatasets. The system achieves photo-realistic 3D reconstruction in large-scale\nsequences, particularly in the EuRoC dataset.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.03822", "title": "Compact 3D Gaussian Splatting for Static and Dynamic Radiance Fields", "abstract": "  3D Gaussian splatting (3DGS) has recently emerged as an alternative\nrepresentation that leverages a 3D Gaussian-based representation and introduces\nan approximated volumetric rendering, achieving very fast rendering speed and\npromising image quality. Furthermore, subsequent studies have successfully\nextended 3DGS to dynamic 3D scenes, demonstrating its wide range of\napplications. However, a significant drawback arises as 3DGS and its following\nmethods entail a substantial number of Gaussians to maintain the high fidelity\nof the rendered images, which requires a large amount of memory and storage. To\naddress this critical issue, we place a specific emphasis on two key\nobjectives: reducing the number of Gaussian points without sacrificing\nperformance and compressing the Gaussian attributes, such as view-dependent\ncolor and covariance. To this end, we propose a learnable mask strategy that\nsignificantly reduces the number of Gaussians while preserving high\nperformance. In addition, we propose a compact but effective representation of\nview-dependent color by employing a grid-based neural field rather than relying\non spherical harmonics. Finally, we learn codebooks to compactly represent the\ngeometric and temporal attributes by residual vector quantization. With model\ncompression techniques such as quantization and entropy coding, we consistently\nshow over 25x reduced storage and enhanced rendering speed compared to 3DGS for\nstatic scenes, while maintaining the quality of the scene representation. For\ndynamic scenes, our approach achieves more than 12x storage efficiency and\nretains a high-quality reconstruction compared to the existing state-of-the-art\nmethods. Our work provides a comprehensive framework for 3D scene\nrepresentation, achieving high performance, fast training, compactness, and\nreal-time rendering. Our project page is available at\nhttps://maincold2.github.io/c3dgs/.\n", "date": "2024-08-07", "categories": []}
{"id": "2402.10344", "title": "Evaluating Neural Radiance Fields (NeRFs) for 3D Plant Geometry\n  Reconstruction in Field Conditions", "abstract": "  We evaluate different Neural Radiance Fields (NeRFs) techniques for the 3D\nreconstruction of plants in varied environments, from indoor settings to\noutdoor fields. Traditional methods usually fail to capture the complex\ngeometric details of plants, which is crucial for phenotyping and breeding\nstudies. We evaluate the reconstruction fidelity of NeRFs in three scenarios\nwith increasing complexity and compare the results with the point cloud\nobtained using LiDAR as ground truth. In the most realistic field scenario, the\nNeRF models achieve a 74.6% F1 score after 30 minutes of training on the GPU,\nhighlighting the efficacy of NeRFs for 3D reconstruction in challenging\nenvironments. Additionally, we propose an early stopping technique for NeRF\ntraining that almost halves the training time while achieving only a reduction\nof 7.4% in the average F1 score. This optimization process significantly\nenhances the speed and efficiency of 3D reconstruction using NeRFs. Our\nfindings demonstrate the potential of NeRFs in detailed and realistic 3D plant\nreconstruction and suggest practical approaches for enhancing the speed and\nefficiency of NeRFs in the 3D reconstruction process.\n", "date": "2024-02-15", "categories": []}
{"id": "2408.03060", "title": "MGFs: Masked Gaussian Fields for Meshing Building based on Multi-View\n  Images", "abstract": "  Over the last few decades, image-based building surface reconstruction has\ngarnered substantial research interest and has been applied across various\nfields, such as heritage preservation, architectural planning, etc. Compared to\nthe traditional photogrammetric and NeRF-based solutions, recently, Gaussian\nfields-based methods have exhibited significant potential in generating surface\nmeshes due to their time-efficient training and detailed 3D information\npreservation. However, most gaussian fields-based methods are trained with all\nimage pixels, encompassing building and nonbuilding areas, which results in a\nsignificant noise for building meshes and degeneration in time efficiency. This\npaper proposes a novel framework, Masked Gaussian Fields (MGFs), designed to\ngenerate accurate surface reconstruction for building in a time-efficient way.\nThe framework first applies EfficientSAM and COLMAP to generate multi-level\nmasks of building and the corresponding masked point clouds. Subsequently, the\nmasked gaussian fields are trained by integrating two innovative losses: a\nmulti-level perceptual masked loss focused on constructing building regions and\na boundary loss aimed at enhancing the details of the boundaries between\ndifferent masks. Finally, we improve the tetrahedral surface mesh extraction\nmethod based on the masked gaussian spheres. Comprehensive experiments on UAV\nimages demonstrate that, compared to the traditional method and several\nNeRF-based and Gaussian-based SOTA solutions, our approach significantly\nimproves both the accuracy and efficiency of building surface reconstruction.\nNotably, as a byproduct, there is an additional gain in the novel view\nsynthesis of building.\n", "date": "2024-08-06", "categories": []}
{"id": "2402.03309", "title": "AONeuS: A Neural Rendering Framework for Acoustic-Optical Sensor Fusion", "abstract": "  Underwater perception and 3D surface reconstruction are challenging problems\nwith broad applications in construction, security, marine archaeology, and\nenvironmental monitoring. Treacherous operating conditions, fragile\nsurroundings, and limited navigation control often dictate that submersibles\nrestrict their range of motion and, thus, the baseline over which they can\ncapture measurements. In the context of 3D scene reconstruction, it is\nwell-known that smaller baselines make reconstruction more challenging. Our\nwork develops a physics-based multimodal acoustic-optical neural surface\nreconstruction framework (AONeuS) capable of effectively integrating\nhigh-resolution RGB measurements with low-resolution depth-resolved imaging\nsonar measurements. By fusing these complementary modalities, our framework can\nreconstruct accurate high-resolution 3D surfaces from measurements captured\nover heavily-restricted baselines. Through extensive simulations and in-lab\nexperiments, we demonstrate that AONeuS dramatically outperforms recent\nRGB-only and sonar-only inverse-differentiable-rendering--based surface\nreconstruction methods. A website visualizing the results of our paper is\nlocated at this address: https://aoneus.github.io/\n", "date": "2024-02-05", "categories": []}
{"id": "2408.01677", "title": "HIVE: HIerarchical Volume Encoding for Neural Implicit Surface\n  Reconstruction", "abstract": "  Neural implicit surface reconstruction has become a new trend in\nreconstructing a detailed 3D shape from images. In previous methods, however,\nthe 3D scene is only encoded by the MLPs which do not have an explicit 3D\nstructure. To better represent 3D shapes, we introduce a volume encoding to\nexplicitly encode the spatial information. We further design hierarchical\nvolumes to encode the scene structures in multiple scales. The high-resolution\nvolumes capture the high-frequency geometry details since spatially varying\nfeatures could be learned from different 3D points, while the low-resolution\nvolumes enforce the spatial consistency to keep the shape smooth since adjacent\nlocations possess the same low-resolution feature. In addition, we adopt a\nsparse structure to reduce the memory consumption at high-resolution volumes,\nand two regularization terms to enhance results smoothness. This hierarchical\nvolume encoding could be appended to any implicit surface reconstruction method\nas a plug-and-play module, and can generate a smooth and clean reconstruction\nwith more details. Superior performance is demonstrated in DTU, EPFL, and\nBlendedMVS datasets with significant improvement on the standard metrics.\n", "date": "2024-08-03", "categories": []}
{"id": "2403.00228", "title": "DISORF: A Distributed Online 3D Reconstruction Framework for Mobile\n  Robots", "abstract": "  We present a framework, DISORF, to enable online 3D reconstruction and\nvisualization of scenes captured by resource-constrained mobile robots and edge\ndevices. To address the limited computing capabilities of edge devices and\npotentially limited network availability, we design a framework that\nefficiently distributes computation between the edge device and the remote\nserver. We leverage on-device SLAM systems to generate posed keyframes and\ntransmit them to remote servers that can perform high-quality 3D reconstruction\nand visualization at runtime by leveraging recent advances in neural 3D\nmethods. We identify a key challenge with online training where naive image\nsampling strategies can lead to significant degradation in rendering quality.\nWe propose a novel shifted exponential frame sampling method that addresses\nthis challenge for online training. We demonstrate the effectiveness of our\nframework in enabling high-quality real-time reconstruction and visualization\nof unknown scenes as they are captured and streamed from cameras in mobile\nrobots and edge devices.\n", "date": "2024-03-01", "categories": []}
{"id": "2407.12939", "title": "GenRC: Generative 3D Room Completion from Sparse Image Collections", "abstract": "  Sparse RGBD scene completion is a challenging task especially when\nconsidering consistent textures and geometries throughout the entire scene.\nDifferent from existing solutions that rely on human-designed text prompts or\npredefined camera trajectories, we propose GenRC, an automated training-free\npipeline to complete a room-scale 3D mesh with high-fidelity textures. To\nachieve this, we first project the sparse RGBD images to a highly incomplete 3D\nmesh. Instead of iteratively generating novel views to fill in the void, we\nutilized our proposed E-Diffusion to generate a view-consistent panoramic RGBD\nimage which ensures global geometry and appearance consistency. Furthermore, we\nmaintain the input-output scene stylistic consistency through textual inversion\nto replace human-designed text prompts. To bridge the domain gap among\ndatasets, E-Diffusion leverages models trained on large-scale datasets to\ngenerate diverse appearances. GenRC outperforms state-of-the-art methods under\nmost appearance and geometric metrics on ScanNet and ARKitScenes datasets, even\nthough GenRC is not trained on these datasets nor using predefined camera\ntrajectories. Project page: https://minfenli.github.io/GenRC\n", "date": "2024-07-17", "categories": []}
{"id": "2402.16174", "title": "GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction", "abstract": "  While recent advances in neural radiance field enable realistic digitization\nfor large-scale scenes, the image-capturing process is still time-consuming and\nlabor-intensive. Previous works attempt to automate this process using the\nNext-Best-View (NBV) policy for active 3D reconstruction. However, the existing\nNBV policies heavily rely on hand-crafted criteria, limited action space, or\nper-scene optimized representations. These constraints limit their\ncross-dataset generalizability. To overcome them, we propose GenNBV, an\nend-to-end generalizable NBV policy. Our policy adopts a reinforcement learning\n(RL)-based framework and extends typical limited action space to 5D free space.\nIt empowers our agent drone to scan from any viewpoint, and even interact with\nunseen geometries during training. To boost the cross-dataset generalizability,\nwe also propose a novel multi-source state embedding, including geometric,\nsemantic, and action representations. We establish a benchmark using the Isaac\nGym simulator with the Houses3K and OmniObject3D datasets to evaluate this NBV\npolicy. Experiments demonstrate that our policy achieves a 98.26% and 97.12%\ncoverage ratio on unseen building-scale objects from these datasets,\nrespectively, outperforming prior solutions.\n", "date": "2024-02-25", "categories": []}
{"id": "2403.16292", "title": "latentSplat: Autoencoding Variational Gaussians for Fast Generalizable\n  3D Reconstruction", "abstract": "  We present latentSplat, a method to predict semantic Gaussians in a 3D latent\nspace that can be splatted and decoded by a light-weight generative 2D\narchitecture. Existing methods for generalizable 3D reconstruction either do\nnot scale to large scenes and resolutions, or are limited to interpolation of\nclose input views. latentSplat combines the strengths of regression-based and\ngenerative approaches while being trained purely on readily available real\nvideo data. The core of our method are variational 3D Gaussians, a\nrepresentation that efficiently encodes varying uncertainty within a latent\nspace consisting of 3D feature Gaussians. From these Gaussians, specific\ninstances can be sampled and rendered via efficient splatting and a fast,\ngenerative decoder. We show that latentSplat outperforms previous works in\nreconstruction quality and generalization, while being fast and scalable to\nhigh-resolution data.\n", "date": "2024-03-24", "categories": []}
{"id": "2406.20055", "title": "SpotlessSplats: Ignoring Distractors in 3D Gaussian Splatting", "abstract": "  3D Gaussian Splatting (3DGS) is a promising technique for 3D reconstruction,\noffering efficient training and rendering speeds, making it suitable for\nreal-time applications.However, current methods require highly controlled\nenvironments (no moving people or wind-blown elements, and consistent lighting)\nto meet the inter-view consistency assumption of 3DGS. This makes\nreconstruction of real-world captures problematic. We present SpotLessSplats,\nan approach that leverages pre-trained and general-purpose features coupled\nwith robust optimization to effectively ignore transient distractors. Our\nmethod achieves state-of-the-art reconstruction quality both visually and\nquantitatively, on casual captures. Additional results available at:\nhttps://spotlesssplats.github.io\n", "date": "2024-06-28", "categories": []}
{"id": "2407.16988", "title": "DreamCar: Leveraging Car-specific Prior for in-the-wild 3D Car\n  Reconstruction", "abstract": "  Self-driving industries usually employ professional artists to build\nexquisite 3D cars. However, it is expensive to craft large-scale digital\nassets. Since there are already numerous datasets available that contain a vast\nnumber of images of cars, we focus on reconstructing high-quality 3D car models\nfrom these datasets. However, these datasets only contain one side of cars in\nthe forward-moving scene. We try to use the existing generative models to\nprovide more supervision information, but they struggle to generalize well in\ncars since they are trained on synthetic datasets not car-specific. In\naddition, The reconstructed 3D car texture misaligns due to a large error in\ncamera pose estimation when dealing with in-the-wild images. These restrictions\nmake it challenging for previous methods to reconstruct complete 3D cars. To\naddress these problems, we propose a novel method, named DreamCar, which can\nreconstruct high-quality 3D cars given a few images even a single image. To\ngeneralize the generative model, we collect a car dataset, named Car360, with\nover 5,600 vehicles. With this dataset, we make the generative model more\nrobust to cars. We use this generative prior specific to the car to guide its\nreconstruction via Score Distillation Sampling. To further complement the\nsupervision information, we utilize the geometric and appearance symmetry of\ncars. Finally, we propose a pose optimization method that rectifies poses to\ntackle texture misalignment. Extensive experiments demonstrate that our method\nsignificantly outperforms existing methods in reconstructing high-quality 3D\ncars. \\href{https://xiaobiaodu.github.io/dreamcar-project/}{Our code is\navailable.}\n", "date": "2024-07-24", "categories": []}
{"id": "2403.02136", "title": "Point2Building: Reconstructing Buildings from Airborne LiDAR Point\n  Clouds", "abstract": "  We present a learning-based approach to reconstruct buildings as 3D polygonal\nmeshes from airborne LiDAR point clouds. What makes 3D building reconstruction\nfrom airborne LiDAR hard is the large diversity of building designs and\nespecially roof shapes, the low and varying point density across the scene, and\nthe often incomplete coverage of building facades due to occlusions by\nvegetation or to the viewing angle of the sensor. To cope with the diversity of\nshapes and inhomogeneous and incomplete object coverage, we introduce a\ngenerative model that directly predicts 3D polygonal meshes from input point\nclouds. Our autoregressive model, called Point2Building, iteratively builds up\nthe mesh by generating sequences of vertices and faces. This approach enables\nour model to adapt flexibly to diverse geometries and building structures.\nUnlike many existing methods that rely heavily on pre-processing steps like\nexhaustive plane detection, our model learns directly from the point cloud\ndata, thereby reducing error propagation and increasing the fidelity of the\nreconstruction. We experimentally validate our method on a collection of\nairborne LiDAR data of Zurich, Berlin and Tallinn. Our method shows good\ngeneralization to diverse urban styles.\n", "date": "2024-03-04", "categories": []}
{"id": "2407.19108", "title": "ObjectCarver: Semi-automatic segmentation, reconstruction and separation\n  of 3D objects", "abstract": "  Implicit neural fields have made remarkable progress in reconstructing 3D\nsurfaces from multiple images; however, they encounter challenges when it comes\nto separating individual objects within a scene. Previous work has attempted to\ntackle this problem by introducing a framework to train separate signed\ndistance fields (SDFs) simultaneously for each of N objects and using a\nregularization term to prevent objects from overlapping. However, all of these\nmethods require segmentation masks to be provided, which are not always readily\navailable. We introduce our method, ObjectCarver, to tackle the problem of\nobject separation from just click input in a single view. Given posed\nmulti-view images and a set of user-input clicks to prompt segmentation of the\nindividual objects, our method decomposes the scene into separate objects and\nreconstructs a high-quality 3D surface for each one. We introduce a loss\nfunction that prevents floaters and avoids inappropriate carving-out due to\nocclusion. In addition, we introduce a novel scene initialization method that\nsignificantly speeds up the process while preserving geometric details compared\nto previous approaches. Despite requiring neither ground truth masks nor\nmonocular cues, our method outperforms baselines both qualitatively and\nquantitatively. In addition, we introduce a new benchmark dataset for\nevaluation.\n", "date": "2024-07-26", "categories": []}
{"id": "2407.19763", "title": "TeleOR: Real-time Telemedicine System for Full-Scene Operating Room", "abstract": "  The advent of telemedicine represents a transformative development in\nleveraging technology to extend the reach of specialized medical expertise to\nremote surgeries, a field where the immediacy of expert guidance is paramount.\nHowever, the intricate dynamics of Operating Room (OR) scene pose unique\nchallenges for telemedicine, particularly in achieving high-fidelity, real-time\nscene reconstruction and transmission amidst obstructions and bandwidth\nlimitations. This paper introduces TeleOR, a pioneering system designed to\naddress these challenges through real-time OR scene reconstruction for\nTele-intervention. TeleOR distinguishes itself with three innovative\napproaches: dynamic self-calibration, which leverages inherent scene features\nfor calibration without the need for preset markers, allowing for obstacle\navoidance and real-time camera adjustment; selective OR reconstruction,\nfocusing on dynamically changing scene segments to reduce reconstruction\ncomplexity; and viewport-adaptive transmission, optimizing data transmission\nbased on real-time client feedback to efficiently deliver high-quality 3D\nreconstructions within bandwidth constraints. Comprehensive experiments on the\n4D-OR surgical scene dataset demostrate the superiority and applicability of\nTeleOR, illuminating the potential to revolutionize tele-interventions by\novercoming the spatial and technical barriers inherent in remote surgical\nguidance.\n", "date": "2024-07-29", "categories": []}
{"id": "2407.11238", "title": "Evaluating geometric accuracy of NeRF reconstructions compared to SLAM\n  method", "abstract": "  As Neural Radiance Field (NeRF) implementations become faster, more efficient\nand accurate, their applicability to real world mapping tasks becomes more\naccessible. Traditionally, 3D mapping, or scene reconstruction, has relied on\nexpensive LiDAR sensing. Photogrammetry can perform image-based 3D\nreconstruction but is computationally expensive and requires extremely dense\nimage representation to recover complex geometry and photorealism. NeRFs\nperform 3D scene reconstruction by training a neural network on sparse image\nand pose data, achieving superior results to photogrammetry with less input\ndata. This paper presents an evaluation of two NeRF scene reconstructions for\nthe purpose of estimating the diameter of a vertical PVC cylinder. One of these\nare trained on commodity iPhone data and the other is trained on robot-sourced\nimagery and poses. This neural-geometry is compared to state-of-the-art\nlidar-inertial SLAM in terms of scene noise and metric-accuracy.\n", "date": "2024-07-15", "categories": []}
{"id": "2310.05483", "title": "HarmonicNeRF: Geometry-Informed Synthetic View Augmentation for 3D Scene\n  Reconstruction in Driving Scenarios", "abstract": "  In the realm of autonomous driving, achieving precise 3D reconstruction of\nthe driving environment is critical for ensuring safety and effective\nnavigation. Neural Radiance Fields (NeRF) have shown promise in creating highly\ndetailed and accurate models of complex environments. However, the application\nof NeRF in autonomous driving scenarios encounters several challenges,\nprimarily due to the sparsity of viewpoints inherent in camera trajectories and\nthe constraints on data collection in unbounded outdoor scenes, which typically\noccur along predetermined paths. This limitation not only reduces the available\nscene information but also poses significant challenges for NeRF training, as\nthe sparse and path-distributed observational data leads to\nunder-representation of the scene's geometry. In this paper, we introduce\nHarmonicNeRF, a novel approach for outdoor self-supervised monocular scene\nreconstruction. HarmonicNeRF capitalizes on the strengths of NeRF and enhances\nsurface reconstruction accuracy by augmenting the input space with\ngeometry-informed synthetic views. This is achieved through the application of\nspherical harmonics to generate novel radiance values, taking into careful\nconsideration the color observations from the limited available real-world\nviews. Additionally, our method incorporates proxy geometry to effectively\nmanage occlusion, generating radiance pseudo-labels that circumvent the\nlimitations of traditional image-warping techniques, which often fail in sparse\ndata conditions typical of autonomous driving environments. Extensive\nexperiments conducted on the KITTI, Argoverse, and NuScenes datasets\ndemonstrate our approach establishes new benchmarks in synthesizing novel depth\nviews and reconstructing scenes, significantly outperforming existing methods.\nProject page: https://github.com/Jiawei-Yao0812/HarmonicNeRF\n", "date": "2023-10-09", "categories": []}
{"id": "2407.16173", "title": "Integrating Meshes and 3D Gaussians for Indoor Scene Reconstruction with\n  SAM Mask Guidance", "abstract": "  We present a novel approach for 3D indoor scene reconstruction that combines\n3D Gaussian Splatting (3DGS) with mesh representations. We use meshes for the\nroom layout of the indoor scene, such as walls, ceilings, and floors, while\nemploying 3D Gaussians for other objects. This hybrid approach leverages the\nstrengths of both representations, offering enhanced flexibility and ease of\nediting. However, joint training of meshes and 3D Gaussians is challenging\nbecause it is not clear which primitive should affect which part of the\nrendered image. Objects close to the room layout often struggle during\ntraining, particularly when the room layout is textureless, which can lead to\nincorrect optimizations and unnecessary 3D Gaussians. To overcome these\nchallenges, we employ Segment Anything Model (SAM) to guide the selection of\nprimitives. The SAM mask loss enforces each instance to be represented by\neither Gaussians or meshes, ensuring clear separation and stable training.\nFurthermore, we introduce an additional densification stage without resetting\nthe opacity after the standard densification. This stage mitigates the\ndegradation of image quality caused by a limited number of 3D Gaussians after\nthe standard densification.\n", "date": "2024-07-23", "categories": []}
{"id": "2407.16503", "title": "HDRSplat: Gaussian Splatting for High Dynamic Range 3D Scene\n  Reconstruction from Raw Images", "abstract": "  The recent advent of 3D Gaussian Splatting (3DGS) has revolutionized the 3D\nscene reconstruction space enabling high-fidelity novel view synthesis in\nreal-time. However, with the exception of RawNeRF, all prior 3DGS and\nNeRF-based methods rely on 8-bit tone-mapped Low Dynamic Range (LDR) images for\nscene reconstruction. Such methods struggle to achieve accurate reconstructions\nin scenes that require a higher dynamic range. Examples include scenes captured\nin nighttime or poorly lit indoor spaces having a low signal-to-noise ratio, as\nwell as daylight scenes with shadow regions exhibiting extreme contrast. Our\nproposed method HDRSplat tailors 3DGS to train directly on 14-bit linear raw\nimages in near darkness which preserves the scenes' full dynamic range and\ncontent. Our key contributions are two-fold: Firstly, we propose a linear HDR\nspace-suited loss that effectively extracts scene information from noisy dark\nregions and nearly saturated bright regions simultaneously, while also handling\nview-dependent colors without increasing the degree of spherical harmonics.\nSecondly, through careful rasterization tuning, we implicitly overcome the\nheavy reliance and sensitivity of 3DGS on point cloud initialization. This is\ncritical for accurate reconstruction in regions of low texture, high depth of\nfield, and low illumination. HDRSplat is the fastest method to date that does\n14-bit (HDR) 3D scene reconstruction in $\\le$15 minutes/scene ($\\sim$30x faster\nthan prior state-of-the-art RawNeRF). It also boasts the fastest inference\nspeed at $\\ge$120fps. We further demonstrate the applicability of our HDR scene\nreconstruction by showcasing various applications like synthetic defocus, dense\ndepth map extraction, and post-capture control of exposure, tone-mapping and\nview-point.\n", "date": "2024-07-23", "categories": []}
{"id": "2407.16575", "title": "Timeliness-Fidelity Tradeoff in 3D Scene Representations", "abstract": "  Real-time three-dimensional (3D) scene representations serve as one of the\nbuilding blocks that bolster various innovative applications, e.g., digital\nmanufacturing, Virtual/Augmented/Extended/Mixed Reality (VR/AR/XR/MR), and the\nmetaverse. Despite substantial efforts that have been made to real-time\ncommunications and computing, real-time 3D scene representations remain a\nchallenging task. This paper investigates the tradeoff between timeliness and\nfidelity in real-time 3D scene representations. Specifically, we establish a\nframework to evaluate the impact of communication delay on the tradeoff, where\nthe real-world scenario is monitored by multiple cameras that communicate with\nan edge server. To improve fidelity for 3D scene representations, we propose to\nuse a single-step Proximal Policy Optimization (PPO) method that leverages the\nAge of Information (AoI) to decide if the received image needs to be involved\nin 3D scene representations and rendering. We test our framework and the\nproposed approach with different well-known 3D scene representation methods.\nSimulation results reveal that real-time 3D scene representation can be\nsensitively affected by communication delay, and our proposed method can\nachieve optimal 3D scene representation results.\n", "date": "2024-07-23", "categories": []}
{"id": "2401.14726", "title": "3D Reconstruction and New View Synthesis of Indoor Environments based on\n  a Dual Neural Radiance Field", "abstract": "  Simultaneously achieving 3D reconstruction and new view synthesis for indoor\nenvironments has widespread applications but is technically very challenging.\nState-of-the-art methods based on implicit neural functions can achieve\nexcellent 3D reconstruction results, but their performances on new view\nsynthesis can be unsatisfactory. The exciting development of neural radiance\nfield (NeRF) has revolutionized new view synthesis, however, NeRF-based models\ncan fail to reconstruct clean geometric surfaces. We have developed a dual\nneural radiance field (Du-NeRF) to simultaneously achieve high-quality geometry\nreconstruction and view rendering. Du-NeRF contains two geometric fields, one\nderived from the SDF field to facilitate geometric reconstruction and the other\nderived from the density field to boost new view synthesis. One of the\ninnovative features of Du-NeRF is that it decouples a view-independent\ncomponent from the density field and uses it as a label to supervise the\nlearning process of the SDF field. This reduces shape-radiance ambiguity and\nenables geometry and color to benefit from each other during the learning\nprocess. Extensive experiments demonstrate that Du-NeRF can significantly\nimprove the performance of novel view synthesis and 3D reconstruction for\nindoor environments and it is particularly effective in constructing areas\ncontaining fine geometries that do not obey multi-view color consistency.\n", "date": "2024-01-26", "categories": []}
{"id": "2407.14059", "title": "Regularizing Dynamic Radiance Fields with Kinematic Fields", "abstract": "  This paper presents a novel approach for reconstructing dynamic radiance\nfields from monocular videos. We integrate kinematics with dynamic radiance\nfields, bridging the gap between the sparse nature of monocular videos and the\nreal-world physics. Our method introduces the kinematic field, capturing motion\nthrough kinematic quantities: velocity, acceleration, and jerk. The kinematic\nfield is jointly learned with the dynamic radiance field by minimizing the\nphotometric loss without motion ground truth. We further augment our method\nwith physics-driven regularizers grounded in kinematics. We propose\nphysics-driven regularizers that ensure the physical validity of predicted\nkinematic quantities, including advective acceleration and jerk. Additionally,\nwe control the motion trajectory based on rigidity equations formed with the\npredicted kinematic quantities. In experiments, our method outperforms the\nstate-of-the-arts by capturing physical motion patterns within challenging\nreal-world monocular videos.\n", "date": "2024-07-19", "categories": []}
{"id": "2407.14225", "title": "Fast Learning of Signed Distance Functions from Noisy Point Clouds via\n  Noise to Noise Mapping", "abstract": "  Learning signed distance functions (SDFs) from point clouds is an important\ntask in 3D computer vision. However, without ground truth signed distances,\npoint normals or clean point clouds, current methods still struggle from\nlearning SDFs from noisy point clouds. To overcome this challenge, we propose\nto learn SDFs via a noise to noise mapping, which does not require any clean\npoint cloud or ground truth supervision. Our novelty lies in the noise to noise\nmapping which can infer a highly accurate SDF of a single object or scene from\nits multiple or even single noisy observations. We achieve this by a novel loss\nwhich enables statistical reasoning on point clouds and maintains geometric\nconsistency although point clouds are irregular, unordered and have no point\ncorrespondence among noisy observations. To accelerate training, we use\nmulti-resolution hash encodings implemented in CUDA in our framework, which\nreduces our training time by a factor of ten, achieving convergence within one\nminute. We further introduce a novel schema to improve multi-view\nreconstruction by estimating SDFs as a prior. Our evaluations under widely-used\nbenchmarks demonstrate our superiority over the state-of-the-art methods in\nsurface reconstruction from point clouds or multi-view images, point cloud\ndenoising and upsampling.\n", "date": "2024-07-04", "categories": []}
{"id": "2407.14257", "title": "SparseCraft: Few-Shot Neural Reconstruction through Stereopsis Guided\n  Geometric Linearization", "abstract": "  We present a novel approach for recovering 3D shape and view dependent\nappearance from a few colored images, enabling efficient 3D reconstruction and\nnovel view synthesis. Our method learns an implicit neural representation in\nthe form of a Signed Distance Function (SDF) and a radiance field. The model is\ntrained progressively through ray marching enabled volumetric rendering, and\nregularized with learning-free multi-view stereo (MVS) cues. Key to our\ncontribution is a novel implicit neural shape function learning strategy that\nencourages our SDF field to be as linear as possible near the level-set, hence\nrobustifying the training against noise emanating from the supervision and\nregularization signals. Without using any pretrained priors, our method, called\nSparseCraft, achieves state-of-the-art performances both in novel-view\nsynthesis and reconstruction from sparse views in standard benchmarks, while\nrequiring less than 10 minutes for training.\n", "date": "2024-07-19", "categories": []}
{"id": "2406.13515", "title": "MVSBoost: An Efficient Point Cloud-based 3D Reconstruction", "abstract": "  Efficient and accurate 3D reconstruction is crucial for various applications,\nincluding augmented and virtual reality, medical imaging, and cinematic special\neffects. While traditional Multi-View Stereo (MVS) systems have been\nfundamental in these applications, using neural implicit fields in implicit 3D\nscene modeling has introduced new possibilities for handling complex topologies\nand continuous surfaces. However, neural implicit fields often suffer from\ncomputational inefficiencies, overfitting, and heavy reliance on data quality,\nlimiting their practical use. This paper presents an enhanced MVS framework\nthat integrates multi-view 360-degree imagery with robust camera pose\nestimation via Structure from Motion (SfM) and advanced image processing for\npoint cloud densification, mesh reconstruction, and texturing. Our approach\nsignificantly improves upon traditional MVS methods, offering superior accuracy\nand precision as validated using Chamfer distance metrics on the Realistic\nSynthetic 360 dataset. The developed MVS technique enhances the detail and\nclarity of 3D reconstructions and demonstrates superior computational\nefficiency and robustness in complex scene reconstruction, effectively handling\nocclusions and varying viewpoints. These improvements suggest that our MVS\nframework can compete with and potentially exceed current state-of-the-art\nneural implicit field methods, especially in scenarios requiring real-time\nprocessing and scalability.\n", "date": "2024-06-19", "categories": []}
{"id": "2403.13327", "title": "Gaussian Splatting on the Move: Blur and Rolling Shutter Compensation\n  for Natural Camera Motion", "abstract": "  High-quality scene reconstruction and novel view synthesis based on Gaussian\nSplatting (3DGS) typically require steady, high-quality photographs, often\nimpractical to capture with handheld cameras. We present a method that adapts\nto camera motion and allows high-quality scene reconstruction with handheld\nvideo data suffering from motion blur and rolling shutter distortion. Our\napproach is based on detailed modelling of the physical image formation process\nand utilizes velocities estimated using visual-inertial odometry (VIO). Camera\nposes are considered non-static during the exposure time of a single image\nframe and camera poses are further optimized in the reconstruction process. We\nformulate a differentiable rendering pipeline that leverages screen space\napproximation to efficiently incorporate rolling-shutter and motion blur\neffects into the 3DGS framework. Our results with both synthetic and real data\ndemonstrate superior performance in mitigating camera motion over existing\nmethods, thereby advancing 3DGS in naturalistic settings.\n", "date": "2024-03-20", "categories": []}
{"id": "2404.01810", "title": "GS2Mesh: Surface Reconstruction from Gaussian Splatting via Novel Stereo\n  Views", "abstract": "  Recently, 3D Gaussian Splatting (3DGS) has emerged as an efficient approach\nfor accurately representing scenes. However, despite its superior novel view\nsynthesis capabilities, extracting the geometry of the scene directly from the\nGaussian properties remains a challenge, as those are optimized based on a\nphotometric loss. While some concurrent models have tried adding geometric\nconstraints during the Gaussian optimization process, they still produce noisy,\nunrealistic surfaces.\n  We propose a novel approach for bridging the gap between the noisy 3DGS\nrepresentation and the smooth 3D mesh representation, by injecting real-world\nknowledge into the depth extraction process. Instead of extracting the geometry\nof the scene directly from the Gaussian properties, we instead extract the\ngeometry through a pre-trained stereo-matching model. We render stereo-aligned\npairs of images corresponding to the original training poses, feed the pairs\ninto a stereo model to get a depth profile, and finally fuse all of the\nprofiles together to get a single mesh.\n  The resulting reconstruction is smoother, more accurate and shows more\nintricate details compared to other methods for surface reconstruction from\nGaussian Splatting, while only requiring a small overhead on top of the fairly\nshort 3DGS optimization process.\n  We performed extensive testing of the proposed method on in-the-wild scenes,\nobtained using a smartphone, showcasing its superior reconstruction abilities.\nAdditionally, we tested the method on the Tanks and Temples and DTU benchmarks,\nachieving state-of-the-art results.\n", "date": "2024-04-02", "categories": []}
{"id": "2407.12661", "title": "InfoNorm: Mutual Information Shaping of Normals for Sparse-View\n  Reconstruction", "abstract": "  3D surface reconstruction from multi-view images is essential for scene\nunderstanding and interaction. However, complex indoor scenes pose challenges\nsuch as ambiguity due to limited observations. Recent implicit surface\nrepresentations, such as Neural Radiance Fields (NeRFs) and signed distance\nfunctions (SDFs), employ various geometric priors to resolve the lack of\nobserved information. Nevertheless, their performance heavily depends on the\nquality of the pre-trained geometry estimation models. To ease such dependence,\nwe propose regularizing the geometric modeling by explicitly encouraging the\nmutual information among surface normals of highly correlated scene points. In\nthis way, the geometry learning process is modulated by the second-order\ncorrelations from noisy (first-order) geometric priors, thus eliminating the\nbias due to poor generalization. Additionally, we introduce a simple yet\neffective scheme that utilizes semantic and geometric features to identify\ncorrelated points, enhancing their mutual information accordingly. The proposed\ntechnique can serve as a plugin for SDF-based neural surface representations.\nOur experiments demonstrate the effectiveness of the proposed in improving the\nsurface reconstruction quality of major states of the arts. Our code is\navailable at: \\url{https://github.com/Muliphein/InfoNorm}.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.12667", "title": "SG-NeRF: Neural Surface Reconstruction with Scene Graph Optimization", "abstract": "  3D surface reconstruction from images is essential for numerous applications.\nRecently, Neural Radiance Fields (NeRFs) have emerged as a promising framework\nfor 3D modeling. However, NeRFs require accurate camera poses as input, and\nexisting methods struggle to handle significantly noisy pose estimates (i.e.,\noutliers), which are commonly encountered in real-world scenarios. To tackle\nthis challenge, we present a novel approach that optimizes radiance fields with\nscene graphs to mitigate the influence of outlier poses. Our method\nincorporates an adaptive inlier-outlier confidence estimation scheme based on\nscene graphs, emphasizing images of high compatibility with the neighborhood\nand consistency in the rendering quality. We also introduce an effective\nintersection-over-union (IoU) loss to optimize the camera pose and surface\ngeometry, together with a coarse-to-fine strategy to facilitate the training.\nFurthermore, we propose a new dataset containing typical outlier poses for a\ndetailed evaluation. Experimental results on various datasets consistently\ndemonstrate the effectiveness and superiority of our method over existing\napproaches, showcasing its robustness in handling outliers and producing\nhigh-quality 3D reconstructions. Our code and data are available at:\n\\url{https://github.com/Iris-cyy/SG-NeRF}.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.04699", "title": "LaRa: Efficient Large-Baseline Radiance Fields", "abstract": "  Radiance field methods have achieved photorealistic novel view synthesis and\ngeometry reconstruction. But they are mostly applied in per-scene optimization\nor small-baseline settings. While several recent works investigate feed-forward\nreconstruction with large baselines by utilizing transformers, they all operate\nwith a standard global attention mechanism and hence ignore the local nature of\n3D reconstruction. We propose a method that unifies local and global reasoning\nin transformer layers, resulting in improved quality and faster convergence.\nOur model represents scenes as Gaussian Volumes and combines this with an image\nencoder and Group Attention Layers for efficient feed-forward reconstruction.\nExperimental results demonstrate that our model, trained for two days on four\nGPUs, demonstrates high fidelity in reconstructing 360 deg radiance fields, and\nrobustness to zero-shot and out-of-domain testing. Our project Page:\nhttps://apchenstu.github.io/LaRa/.\n", "date": "2024-07-05", "categories": []}
{"id": "2407.11343", "title": "Ev-GS: Event-based Gaussian splatting for Efficient and Accurate\n  Radiance Field Rendering", "abstract": "  Computational neuromorphic imaging (CNI) with event cameras offers advantages\nsuch as minimal motion blur and enhanced dynamic range, compared to\nconventional frame-based methods. Existing event-based radiance field rendering\nmethods are built on neural radiance field, which is computationally heavy and\nslow in reconstruction speed. Motivated by the two aspects, we introduce Ev-GS,\nthe first CNI-informed scheme to infer 3D Gaussian splatting from a monocular\nevent camera, enabling efficient novel view synthesis. Leveraging 3D Gaussians\nwith pure event-based supervision, Ev-GS overcomes challenges such as the\ndetection of fast-moving objects and insufficient lighting. Experimental\nresults show that Ev-GS outperforms the method that takes frame-based signals\nas input by rendering realistic views with reduced blurring and improved visual\nquality. Moreover, it demonstrates competitive reconstruction quality and\nreduced computing occupancy compared to existing methods, which paves the way\nto a highly efficient CNI approach for signal processing.\n", "date": "2024-07-16", "categories": []}
{"id": "2407.11431", "title": "MRIo3DS-Net: A Mutually Reinforcing Images to 3D Surface RNN-like\n  framework for model-adaptation indoor 3D reconstruction", "abstract": "  This paper is the first to propose an end-to-end framework of mutually\nreinforcing images to 3D surface recurrent neural network-like for\nmodel-adaptation indoor 3D reconstruction,where multi-view dense matching and\npoint cloud surface optimization are mutually reinforced by a RNN-like\nstructure rather than being treated as a separate issue.The characteristics are\nas follows:In the multi-view dense matching module, the model-adaptation\nstrategy is used to fine-tune and optimize a Transformer-based multi-view dense\nmatching DNN,so that it has the higher image feature for matching and detail\nexpression capabilities;In the point cloud surface optimization module,the 3D\nsurface reconstruction network based on 3D implicit field is optimized by using\nmodel-adaptation strategy,which solves the problem of point cloud surface\noptimization without knowing normal vector of 3D surface.To improve and finely\nreconstruct 3D surfaces from point cloud,smooth loss is proposed and added to\nthis module;The MRIo3DS-Net is a RNN-like framework,which utilizes the finely\noptimized 3D surface obtained by PCSOM to recursively reinforce the\ndifferentiable warping for optimizing MVDMM.This refinement leads to achieving\nbetter dense matching results, and better dense matching results leads to\nachieving better 3D surface results recursively and mutually.Hence,\nmodel-adaptation strategy can better collaborate the differences between the\ntwo network modules,so that they complement each other to achieve the better\neffect;To accelerate the transfer learning and training convergence from source\ndomain to target domain,a multi-task loss function based on Bayesian\nuncertainty is used to adaptively adjust the weights between the two networks\nloss functions of MVDMM and PCSOM;In this multi-task cascade network\nframework,any modules can be replaced by any state-of-the-art networks to\nachieve better 3D reconstruction results.\n", "date": "2024-07-16", "categories": []}
{"id": "2407.11610", "title": "MergeNet: Explicit Mesh Reconstruction from Sparse Point Clouds via Edge\n  Prediction", "abstract": "  This paper introduces a novel method for reconstructing meshes from sparse\npoint clouds by predicting edge connection. Existing implicit methods usually\nproduce superior smooth and watertight meshes due to the isosurface extraction\nalgorithms~(e.g., Marching Cubes). However, these methods become memory and\ncomputationally intensive with increasing resolution. Explicit methods are more\nefficient by directly forming the face from points. Nevertheless, the challenge\nof selecting appropriate faces from enormous candidates often leads to\nundesirable faces and holes. Moreover, the reconstruction performance of both\napproaches tends to degrade when the point cloud gets sparse. To this end, we\npropose MEsh Reconstruction via edGE~(MergeNet), which converts mesh\nreconstruction into local connectivity prediction problems. Specifically,\nMergeNet learns to extract the features of candidate edges and regress their\ndistances to the underlying surface. Consequently, the predicted distance is\nutilized to filter out edges that lay on surfaces. Finally, the meshes are\nreconstructed by refining the triangulations formed by these edges. Extensive\nexperiments on synthetic and real-scanned datasets demonstrate the superiority\nof MergeNet to SoTA explicit methods.\n", "date": "2024-07-16", "categories": []}
{"id": "2404.02410", "title": "TCLC-GS: Tightly Coupled LiDAR-Camera Gaussian Splatting for Autonomous\n  Driving", "abstract": "  Most 3D Gaussian Splatting (3D-GS) based methods for urban scenes initialize\n3D Gaussians directly with 3D LiDAR points, which not only underutilizes LiDAR\ndata capabilities but also overlooks the potential advantages of fusing LiDAR\nwith camera data. In this paper, we design a novel tightly coupled LiDAR-Camera\nGaussian Splatting (TCLC-GS) to fully leverage the combined strengths of both\nLiDAR and camera sensors, enabling rapid, high-quality 3D reconstruction and\nnovel view RGB/depth synthesis. TCLC-GS designs a hybrid explicit (colorized 3D\nmesh) and implicit (hierarchical octree feature) 3D representation derived from\nLiDAR-camera data, to enrich the properties of 3D Gaussians for splatting. 3D\nGaussian's properties are not only initialized in alignment with the 3D mesh\nwhich provides more completed 3D shape and color information, but are also\nendowed with broader contextual information through retrieved octree implicit\nfeatures. During the Gaussian Splatting optimization process, the 3D mesh\noffers dense depth information as supervision, which enhances the training\nprocess by learning of a robust geometry. Comprehensive evaluations conducted\non the Waymo Open Dataset and nuScenes Dataset validate our method's\nstate-of-the-art (SOTA) performance. Utilizing a single NVIDIA RTX 3090 Ti, our\nmethod demonstrates fast training and achieves real-time RGB and depth\nrendering at 90 FPS in resolution of 1920x1280 (Waymo), and 120 FPS in\nresolution of 1600x900 (nuScenes) in urban scenarios.\n", "date": "2024-04-03", "categories": []}
{"id": "2405.12218", "title": "MVSGaussian: Fast Generalizable Gaussian Splatting Reconstruction from\n  Multi-View Stereo", "abstract": "  We present MVSGaussian, a new generalizable 3D Gaussian representation\napproach derived from Multi-View Stereo (MVS) that can efficiently reconstruct\nunseen scenes. Specifically, 1) we leverage MVS to encode geometry-aware\nGaussian representations and decode them into Gaussian parameters. 2) To\nfurther enhance performance, we propose a hybrid Gaussian rendering that\nintegrates an efficient volume rendering design for novel view synthesis. 3) To\nsupport fast fine-tuning for specific scenes, we introduce a multi-view\ngeometric consistent aggregation strategy to effectively aggregate the point\nclouds generated by the generalizable model, serving as the initialization for\nper-scene optimization. Compared with previous generalizable NeRF-based\nmethods, which typically require minutes of fine-tuning and seconds of\nrendering per image, MVSGaussian achieves real-time rendering with better\nsynthesis quality for each scene. Compared with the vanilla 3D-GS, MVSGaussian\nachieves better view synthesis with less training computational cost. Extensive\nexperiments on DTU, Real Forward-facing, NeRF Synthetic, and Tanks and Temples\ndatasets validate that MVSGaussian attains state-of-the-art performance with\nconvincing generalizability, real-time rendering speed, and fast per-scene\noptimization.\n", "date": "2024-05-20", "categories": []}
{"id": "2407.03939", "title": "SfM on-the-fly: Get better 3D from What You Capture", "abstract": "  In the last twenty years, Structure from Motion (SfM) has been a constant\nresearch hotspot in the fields of photogrammetry, computer vision, robotics\netc., whereas real-time performance is just a recent topic of growing interest.\nThis work builds upon the original on-the-fly SfM (Zhan et al., 2024) and\npresents an updated version with three new advancements to get better 3D from\nwhat you capture: (i) real-time image matching is further boosted by employing\nthe Hierarchical Navigable Small World (HNSW) graphs, thus more true positive\noverlapping image candidates are faster identified; (ii) a self-adaptive\nweighting strategy is proposed for robust hierarchical local bundle adjustment\nto improve the SfM results; (iii) multiple agents are included for supporting\ncollaborative SfM and seamlessly merge multiple 3D reconstructions into a\ncomplete 3D scene when commonly registered images appear. Various comprehensive\nexperiments demonstrate that the proposed SfM method (named on-the-fly SfMv2)\ncan generate more complete and robust 3D reconstructions in a high\ntime-efficient way. Code is available at\nhttp://yifeiyu225.github.io/on-the-flySfMv2.github.io/.\n", "date": "2024-07-04", "categories": []}
{"id": "2407.10062", "title": "SpikeGS: 3D Gaussian Splatting from Spike Streams with High-Speed Camera\n  Motion", "abstract": "  Novel View Synthesis plays a crucial role by generating new 2D renderings\nfrom multi-view images of 3D scenes. However, capturing high-speed scenes with\nconventional cameras often leads to motion blur, hindering the effectiveness of\n3D reconstruction. To address this challenge, high-frame-rate dense 3D\nreconstruction emerges as a vital technique, enabling detailed and accurate\nmodeling of real-world objects or scenes in various fields, including Virtual\nReality or embodied AI. Spike cameras, a novel type of neuromorphic sensor,\ncontinuously record scenes with an ultra-high temporal resolution, showing\npotential for accurate 3D reconstruction. Despite their promise, existing\napproaches, such as applying Neural Radiance Fields (NeRF) to spike cameras,\nencounter challenges due to the time-consuming rendering process. To address\nthis issue, we make the first attempt to introduce the 3D Gaussian Splatting\n(3DGS) into spike cameras in high-speed capture, providing 3DGS as dense and\ncontinuous clues of views, then constructing SpikeGS. Specifically, to train\nSpikeGS, we establish computational equations between the rendering process of\n3DGS and the processes of instantaneous imaging and exposing-like imaging of\nthe continuous spike stream. Besides, we build a very lightweight but effective\nmapping process from spikes to instant images to support training. Furthermore,\nwe introduced a new spike-based 3D rendering dataset for validation. Extensive\nexperiments have demonstrated our method possesses the high quality of novel\nview rendering, proving the tremendous potential of spike cameras in modeling\n3D scenes.\n", "date": "2024-07-14", "categories": []}
{"id": "2407.10439", "title": "PolyRoom: Room-aware Transformer for Floorplan Reconstruction", "abstract": "  Reconstructing geometry and topology structures from raw unstructured data\nhas always been an important research topic in indoor mapping research. In this\npaper, we aim to reconstruct the floorplan with a vectorized representation\nfrom point clouds. Despite significant advancements achieved in recent years,\ncurrent methods still encounter several challenges, such as missing corners or\nedges, inaccuracies in corner positions or angles, self-intersecting or\noverlapping polygons, and potentially implausible topology. To tackle these\nchallenges, we present PolyRoom, a room-aware Transformer that leverages\nuniform sampling representation, room-aware query initialization, and\nroom-aware self-attention for floorplan reconstruction. Specifically, we adopt\na uniform sampling floorplan representation to enable dense supervision during\ntraining and effective utilization of angle information. Additionally, we\npropose a room-aware query initialization scheme to prevent non-polygonal\nsequences and introduce room-aware self-attention to enhance memory efficiency\nand model performance. Experimental results on two widely used datasets\ndemonstrate that PolyRoom surpasses current state-of-the-art methods both\nquantitatively and qualitatively. Our code is available at:\nhttps://github.com/3dv-casia/PolyRoom/.\n", "date": "2024-07-15", "categories": []}
{"id": "2407.08795", "title": "Feasibility of Neural Radiance Fields for Crime Scene Video\n  Reconstruction", "abstract": "  This paper aims to review and determine the feasibility of using variations\nof NeRF models in order to reconstruct crime scenes given input videos of the\nscene. We focus on three main innovations of NeRF when it comes to\nreconstructing crime scenes: Multi-object Synthesis, Deformable Synthesis, and\nLighting. From there, we analyse its innovation progress against the\nrequirements to be met in order to be able to reconstruct crime scenes with\ngiven videos of such scenes.\n", "date": "2024-07-11", "categories": []}
{"id": "2308.15536", "title": "DebSDF: Delving into the Details and Bias of Neural Indoor Scene\n  Reconstruction", "abstract": "  In recent years, the neural implicit surface has emerged as a powerful\nrepresentation for multi-view surface reconstruction due to its simplicity and\nstate-of-the-art performance. However, reconstructing smooth and detailed\nsurfaces in indoor scenes from multi-view images presents unique challenges.\nIndoor scenes typically contain large texture-less regions, making the\nphotometric loss unreliable for optimizing the implicit surface. Previous work\nutilizes monocular geometry priors to improve the reconstruction in indoor\nscenes. However, monocular priors often contain substantial errors in thin\nstructure regions due to domain gaps and the inherent inconsistencies when\nderived independently from different views. This paper presents \\textbf{DebSDF}\nto address these challenges, focusing on the utilization of uncertainty in\nmonocular priors and the bias in SDF-based volume rendering. We propose an\nuncertainty modeling technique that associates larger uncertainties with larger\nerrors in the monocular priors. High-uncertainty priors are then excluded from\noptimization to prevent bias. This uncertainty measure also informs an\nimportance-guided ray sampling and adaptive smoothness regularization,\nenhancing the learning of fine structures. We further introduce a bias-aware\nsigned distance function to density transformation that takes into account the\ncurvature and the angle between the view direction and the SDF normals to\nreconstruct fine details better. Our approach has been validated through\nextensive experiments on several challenging datasets, demonstrating improved\nqualitative and quantitative results in reconstructing thin structures in\nindoor scenes, thereby outperforming previous work.\n", "date": "2023-08-29", "categories": []}
{"id": "2407.08137", "title": "Survey on Fundamental Deep Learning 3D Reconstruction Techniques", "abstract": "  This survey aims to investigate fundamental deep learning (DL) based 3D\nreconstruction techniques that produce photo-realistic 3D models and scenes,\nhighlighting Neural Radiance Fields (NeRFs), Latent Diffusion Models (LDM), and\n3D Gaussian Splatting. We dissect the underlying algorithms, evaluate their\nstrengths and tradeoffs, and project future research trajectories in this\nrapidly evolving field. We provide a comprehensive overview of the fundamental\nin DL-driven 3D scene reconstruction, offering insights into their potential\napplications and limitations.\n", "date": "2024-07-11", "categories": []}
{"id": "2309.14737", "title": "Volumetric Semantically Consistent 3D Panoptic Mapping", "abstract": "  We introduce an online 2D-to-3D semantic instance mapping algorithm aimed at\ngenerating comprehensive, accurate, and efficient semantic 3D maps suitable for\nautonomous agents in unstructured environments. The proposed approach is based\non a Voxel-TSDF representation used in recent algorithms. It introduces novel\nways of integrating semantic prediction confidence during mapping, producing\nsemantic and instance-consistent 3D regions. Further improvements are achieved\nby graph optimization-based semantic labeling and instance refinement. The\nproposed method achieves accuracy superior to the state of the art on public\nlarge-scale datasets, improving on a number of widely used metrics. We also\nhighlight a downfall in the evaluation of recent studies: using the ground\ntruth trajectory as input instead of a SLAM-estimated one substantially affects\nthe accuracy, creating a large gap between the reported results and the actual\nperformance on real-world data.\n", "date": "2023-09-26", "categories": []}
{"id": "2407.05023", "title": "SurgicalGaussian: Deformable 3D Gaussians for High-Fidelity Surgical\n  Scene Reconstruction", "abstract": "  Dynamic reconstruction of deformable tissues in endoscopic video is a key\ntechnology for robot-assisted surgery. Recent reconstruction methods based on\nneural radiance fields (NeRFs) have achieved remarkable results in the\nreconstruction of surgical scenes. However, based on implicit representation,\nNeRFs struggle to capture the intricate details of objects in the scene and\ncannot achieve real-time rendering. In addition, restricted single view\nperception and occluded instruments also propose special challenges in surgical\nscene reconstruction. To address these issues, we develop SurgicalGaussian, a\ndeformable 3D Gaussian Splatting method to model dynamic surgical scenes. Our\napproach models the spatio-temporal features of soft tissues at each time stamp\nvia a forward-mapping deformation MLP and regularization to constrain local 3D\nGaussians to comply with consistent movement. With the depth initialization\nstrategy and tool mask-guided training, our method can remove surgical\ninstruments and reconstruct high-fidelity surgical scenes. Through experiments\non various surgical videos, our network outperforms existing method on many\naspects, including rendering quality, rendering speed and GPU usage. The\nproject page can be found at https://surgicalgaussian.github.io.\n", "date": "2024-07-06", "categories": []}
{"id": "2407.05615", "title": "OSN: Infinite Representations of Dynamic 3D Scenes from Monocular Videos", "abstract": "  It has long been challenging to recover the underlying dynamic 3D scene\nrepresentations from a monocular RGB video. Existing works formulate this\nproblem into finding a single most plausible solution by adding various\nconstraints such as depth priors and strong geometry constraints, ignoring the\nfact that there could be infinitely many 3D scene representations corresponding\nto a single dynamic video. In this paper, we aim to learn all plausible 3D\nscene configurations that match the input video, instead of just inferring a\nspecific one. To achieve this ambitious goal, we introduce a new framework,\ncalled OSN. The key to our approach is a simple yet innovative object scale\nnetwork together with a joint optimization module to learn an accurate scale\nrange for every dynamic 3D object. This allows us to sample as many faithful 3D\nscene configurations as possible. Extensive experiments show that our method\nsurpasses all baselines and achieves superior accuracy in dynamic novel view\nsynthesis on multiple synthetic and real-world datasets. Most notably, our\nmethod demonstrates a clear advantage in learning fine-grained 3D scene\ngeometry. Our code and data are available at https://github.com/vLAR-group/OSN\n", "date": "2024-07-08", "categories": []}
{"id": "2407.02598", "title": "AutoSplat: Constrained Gaussian Splatting for Autonomous Driving Scene\n  Reconstruction", "abstract": "  Realistic scene reconstruction and view synthesis are essential for advancing\nautonomous driving systems by simulating safety-critical scenarios. 3D Gaussian\nSplatting excels in real-time rendering and static scene reconstructions but\nstruggles with modeling driving scenarios due to complex backgrounds, dynamic\nobjects, and sparse views. We propose AutoSplat, a framework employing Gaussian\nsplatting to achieve highly realistic reconstructions of autonomous driving\nscenes. By imposing geometric constraints on Gaussians representing the road\nand sky regions, our method enables multi-view consistent simulation of\nchallenging scenarios including lane changes. Leveraging 3D templates, we\nintroduce a reflected Gaussian consistency constraint to supervise both the\nvisible and unseen side of foreground objects. Moreover, to model the dynamic\nappearance of foreground objects, we estimate residual spherical harmonics for\neach foreground Gaussian. Extensive experiments on Pandaset and KITTI\ndemonstrate that AutoSplat outperforms state-of-the-art methods in scene\nreconstruction and novel view synthesis across diverse driving scenarios. Visit\nour project page at https://autosplat.github.io/.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.03594", "title": "UniPlane: Unified Plane Detection and Reconstruction from Posed\n  Monocular Videos", "abstract": "  We present UniPlane, a novel method that unifies plane detection and\nreconstruction from posed monocular videos. Unlike existing methods that detect\nplanes from local observations and associate them across the video for the\nfinal reconstruction, UniPlane unifies both the detection and the\nreconstruction tasks in a single network, which allows us to directly optimize\nfinal reconstruction quality and fully leverage temporal information.\nSpecifically, we build a Transformers-based deep neural network that jointly\nconstructs a 3D feature volume for the environment and estimates a set of\nper-plane embeddings as queries. UniPlane directly reconstructs the 3D planes\nby taking dot products between voxel embeddings and the plane embeddings\nfollowed by binary thresholding. Extensive experiments on real-world datasets\ndemonstrate that UniPlane outperforms state-of-the-art methods in both plane\ndetection and reconstruction tasks, achieving +4.6 in F-score in geometry as\nwell as consistent improvements in other geometry and segmentation metrics.\n", "date": "2024-07-04", "categories": []}
{"id": "2407.02918", "title": "Free-SurGS: SfM-Free 3D Gaussian Splatting for Surgical Scene\n  Reconstruction", "abstract": "  Real-time 3D reconstruction of surgical scenes plays a vital role in\ncomputer-assisted surgery, holding a promise to enhance surgeons' visibility.\nRecent advancements in 3D Gaussian Splatting (3DGS) have shown great potential\nfor real-time novel view synthesis of general scenes, which relies on accurate\nposes and point clouds generated by Structure-from-Motion (SfM) for\ninitialization. However, 3DGS with SfM fails to recover accurate camera poses\nand geometry in surgical scenes due to the challenges of minimal textures and\nphotometric inconsistencies. To tackle this problem, in this paper, we propose\nthe first SfM-free 3DGS-based method for surgical scene reconstruction by\njointly optimizing the camera poses and scene representation. Based on the\nvideo continuity, the key of our method is to exploit the immediate optical\nflow priors to guide the projection flow derived from 3D Gaussians. Unlike most\nprevious methods relying on photometric loss only, we formulate the pose\nestimation problem as minimizing the flow loss between the projection flow and\noptical flow. A consistency check is further introduced to filter the flow\noutliers by detecting the rigid and reliable points that satisfy the epipolar\ngeometry. During 3D Gaussian optimization, we randomly sample frames to\noptimize the scene representations to grow the 3D Gaussian progressively.\nExperiments on the SCARED dataset demonstrate our superior performance over\nexisting methods in novel view synthesis and pose estimation with high\nefficiency. Code is available at https://github.com/wrld/Free-SurGS.\n", "date": "2024-07-03", "categories": []}
{"id": "2407.03172", "title": "IMC 2024 Methods & Solutions Review", "abstract": "  For the past three years, Kaggle has been hosting the Image Matching\nChallenge, which focuses on solving a 3D image reconstruction problem using a\ncollection of 2D images. Each year, this competition fosters the development of\ninnovative and effective methodologies by its participants. In this paper, we\nintroduce an advanced ensemble technique that we developed, achieving a score\nof 0.153449 on the private leaderboard and securing the 160th position out of\nover 1,000 participants. Additionally, we conduct a comprehensive review of\nexisting methods and techniques employed by top-performing teams in the\ncompetition. Our solution, alongside the insights gathered from other leading\napproaches, contributes to the ongoing advancement in the field of 3D image\nreconstruction. This research provides valuable knowledge for future\nparticipants and researchers aiming to excel in similar image matching and\nreconstruction challenges.\n", "date": "2024-07-03", "categories": []}
{"id": "2407.01761", "title": "DRAGON: Drone and Ground Gaussian Splatting for 3D Building\n  Reconstruction", "abstract": "  3D building reconstruction from imaging data is an important task for many\napplications ranging from urban planning to reconnaissance. Modern Novel View\nsynthesis (NVS) methods like NeRF and Gaussian Splatting offer powerful\ntechniques for developing 3D models from natural 2D imagery in an unsupervised\nfashion. These algorithms generally require input training views surrounding\nthe scene of interest, which, in the case of large buildings, is typically not\navailable across all camera elevations. In particular, the most readily\navailable camera viewpoints at scale across most buildings are at near-ground\n(e.g., with mobile phones) and aerial (drones) elevations. However, due to the\nsignificant difference in viewpoint between drone and ground image sets, camera\nregistration - a necessary step for NVS algorithms - fails. In this work we\npropose a method, DRAGON, that can take drone and ground building imagery as\ninput and produce a 3D NVS model. The key insight of DRAGON is that\nintermediate elevation imagery may be extrapolated by an NVS algorithm itself\nin an iterative procedure with perceptual regularization, thereby bridging the\nvisual feature gap between the two elevations and enabling registration. We\ncompiled a semi-synthetic dataset of 9 large building scenes using Google Earth\nStudio, and quantitatively and qualitatively demonstrate that DRAGON can\ngenerate compelling renderings on this dataset compared to baseline strategies.\n", "date": "2024-07-01", "categories": []}
{"id": "2407.01945", "title": "Indoor 3D Reconstruction with an Unknown Camera-Projector Pair", "abstract": "  Structured light-based method with a camera-projector pair (CPP) plays a\nvital role in indoor 3D reconstruction, especially for scenes with weak\ntextures. Previous methods usually assume known intrinsics, which are\npre-calibrated from known objects, or self-calibrated from multi-view\nobservations. It is still challenging to reliably recover CPP intrinsics from\nonly two views without any known objects. In this paper, we provide a simple\nyet reliable solution. We demonstrate that, for the first time, sufficient\nconstraints on CPP intrinsics can be derived from an unknown cuboid corner\n(C2), e.g. a room's corner, which is a common structure in indoor scenes. In\naddition, with only known camera principal point, the complex multi-variable\nestimation of all CPP intrinsics can be simplified to a simple univariable\noptimization problem, leading to reliable calibration and thus direct 3D\nreconstruction with unknown CPP. Extensive results have demonstrated the\nsuperiority of the proposed method over both traditional and learning-based\ncounterparts. Furthermore, the proposed method also demonstrates impressive\npotential to solve similar tasks without active lighting, such as sparse-view\nstructure from motion.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.01349", "title": "PanopticRecon: Leverage Open-vocabulary Instance Segmentation for\n  Zero-shot Panoptic Reconstruction", "abstract": "  Panoptic reconstruction is a challenging task in 3D scene understanding.\nHowever, most existing methods heavily rely on pre-trained semantic\nsegmentation models and known 3D object bounding boxes for 3D panoptic\nsegmentation, which is not available for in-the-wild scenes. In this paper, we\npropose a novel zero-shot panoptic reconstruction method from RGB-D images of\nscenes. For zero-shot segmentation, we leverage open-vocabulary instance\nsegmentation, but it has to face partial labeling and instance association\nchallenges. We tackle both challenges by propagating partial labels with the\naid of dense generalized features and building a 3D instance graph for\nassociating 2D instance IDs. Specifically, we exploit partial labels to learn a\nclassifier for generalized semantic features to provide complete labels for\nscenes with dense distilled features. Moreover, we formulate instance\nassociation as a 3D instance graph segmentation problem, allowing us to fully\nutilize the scene geometry prior and all 2D instance masks to infer global\nunique pseudo 3D instance ID. Our method outperforms state-of-the-art methods\non the indoor dataset ScanNet V2 and the outdoor dataset KITTI-360,\ndemonstrating the effectiveness of our graph segmentation method and\nreconstruction network.\n", "date": "2024-07-01", "categories": []}
{"id": "2404.07097", "title": "Fast Encoder-Based 3D from Casual Videos via Point Track Processing", "abstract": "  This paper addresses the long-standing challenge of reconstructing 3D\nstructures from videos with dynamic content. Current approaches to this problem\nwere not designed to operate on casual videos recorded by standard cameras or\nrequire a long optimization time.\n  Aiming to significantly improve the efficiency of previous approaches, we\npresent TracksTo4D, a learning-based approach that enables inferring 3D\nstructure and camera positions from dynamic content originating from casual\nvideos using a single efficient feed-forward pass. To achieve this, we propose\noperating directly over 2D point tracks as input and designing an architecture\ntailored for processing 2D point tracks. Our proposed architecture is designed\nwith two key principles in mind: (1) it takes into account the inherent\nsymmetries present in the input point tracks data, and (2) it assumes that the\nmovement patterns can be effectively represented using a low-rank\napproximation. TracksTo4D is trained in an unsupervised way on a dataset of\ncasual videos utilizing only the 2D point tracks extracted from the videos,\nwithout any 3D supervision. Our experiments show that TracksTo4D can\nreconstruct a temporal point cloud and camera positions of the underlying video\nwith accuracy comparable to state-of-the-art methods, while drastically\nreducing runtime by up to 95\\%. We further show that TracksTo4D generalizes\nwell to unseen videos of unseen semantic categories at inference time.\n", "date": "2024-04-10", "categories": []}
{"id": "2406.19390", "title": "SALVe: Semantic Alignment Verification for Floorplan Reconstruction from\n  Sparse Panoramas", "abstract": "  We propose a new system for automatic 2D floorplan reconstruction that is\nenabled by SALVe, our novel pairwise learned alignment verifier. The inputs to\nour system are sparsely located 360$^\\circ$ panoramas, whose semantic features\n(windows, doors, and openings) are inferred and used to hypothesize pairwise\nroom adjacency or overlap. SALVe initializes a pose graph, which is\nsubsequently optimized using GTSAM. Once the room poses are computed, room\nlayouts are inferred using HorizonNet, and the floorplan is constructed by\nstitching the most confident layout boundaries. We validate our system\nqualitatively and quantitatively as well as through ablation studies, showing\nthat it outperforms state-of-the-art SfM systems in completeness by over 200%,\nwithout sacrificing accuracy. Our results point to the significance of our\nwork: poses of 81% of panoramas are localized in the first 2 connected\ncomponents (CCs), and 89% in the first 3 CCs. Code and models are publicly\navailable at https://github.com/zillow/salve.\n", "date": "2024-06-27", "categories": []}
{"id": "2406.18198", "title": "VDG: Vision-Only Dynamic Gaussian for Driving Simulation", "abstract": "  Dynamic Gaussian splatting has led to impressive scene reconstruction and\nimage synthesis advances in novel views. Existing methods, however, heavily\nrely on pre-computed poses and Gaussian initialization by Structure from Motion\n(SfM) algorithms or expensive sensors. For the first time, this paper addresses\nthis issue by integrating self-supervised VO into our pose-free dynamic\nGaussian method (VDG) to boost pose and depth initialization and static-dynamic\ndecomposition. Moreover, VDG can work with only RGB image input and construct\ndynamic scenes at a faster speed and larger scenes compared with the pose-free\ndynamic view-synthesis method. We demonstrate the robustness of our approach\nvia extensive quantitative and qualitative experiments. Our results show\nfavorable performance over the state-of-the-art dynamic view synthesis methods.\nAdditional video and source code will be posted on our project page at\nhttps://3d-aigc.github.io/VDG.\n", "date": "2024-06-26", "categories": []}
{"id": "2406.18199", "title": "GS-Octree: Octree-based 3D Gaussian Splatting for Robust Object-level 3D\n  Reconstruction Under Strong Lighting", "abstract": "  The 3D Gaussian Splatting technique has significantly advanced the\nconstruction of radiance fields from multi-view images, enabling real-time\nrendering. While point-based rasterization effectively reduces computational\ndemands for rendering, it often struggles to accurately reconstruct the\ngeometry of the target object, especially under strong lighting. To address\nthis challenge, we introduce a novel approach that combines octree-based\nimplicit surface representations with Gaussian splatting. Our method consists\nof four stages. Initially, it reconstructs a signed distance field (SDF) and a\nradiance field through volume rendering, encoding them in a low-resolution\noctree. The initial SDF represents the coarse geometry of the target object.\nSubsequently, it introduces 3D Gaussians as additional degrees of freedom,\nwhich are guided by the SDF. In the third stage, the optimized Gaussians\nfurther improve the accuracy of the SDF, allowing it to recover finer geometric\ndetails compared to the initial SDF obtained in the first stage. Finally, it\nadopts the refined SDF to further optimize the 3D Gaussians via splatting,\neliminating those that contribute little to visual appearance. Experimental\nresults show that our method, which leverages the distribution of 3D Gaussians\nwith SDFs, reconstructs more accurate geometry, particularly in images with\nspecular highlights caused by strong lighting.\n", "date": "2024-06-26", "categories": []}
{"id": "2403.05005", "title": "DITTO: Dual and Integrated Latent Topologies for Implicit 3D\n  Reconstruction", "abstract": "  We propose a novel concept of dual and integrated latent topologies (DITTO in\nshort) for implicit 3D reconstruction from noisy and sparse point clouds. Most\nexisting methods predominantly focus on single latent type, such as point or\ngrid latents. In contrast, the proposed DITTO leverages both point and grid\nlatents (i.e., dual latent) to enhance their strengths, the stability of grid\nlatents and the detail-rich capability of point latents. Concretely, DITTO\nconsists of dual latent encoder and integrated implicit decoder. In the dual\nlatent encoder, a dual latent layer, which is the key module block composing\nthe encoder, refines both latents in parallel, maintaining their distinct\nshapes and enabling recursive interaction. Notably, a newly proposed dynamic\nsparse point transformer within the dual latent layer effectively refines point\nlatents. Then, the integrated implicit decoder systematically combines these\nrefined latents, achieving high-fidelity 3D reconstruction and surpassing\nprevious state-of-the-art methods on object- and scene-level datasets,\nespecially in thin and detailed structures.\n", "date": "2024-03-08", "categories": []}
{"id": "2406.01467", "title": "RaDe-GS: Rasterizing Depth in Gaussian Splatting", "abstract": "  Gaussian Splatting (GS) has proven to be highly effective in novel view\nsynthesis, achieving high-quality and real-time rendering. However, its\npotential for reconstructing detailed 3D shapes has not been fully explored.\nExisting methods often suffer from limited shape accuracy due to the discrete\nand unstructured nature of Gaussian splats, which complicates the shape\nextraction. While recent techniques like 2D GS have attempted to improve shape\nreconstruction, they often reformulate the Gaussian primitives in ways that\nreduce both rendering quality and computational efficiency. To address these\nproblems, our work introduces a rasterized approach to render the depth maps\nand surface normal maps of general 3D Gaussian splats. Our method not only\nsignificantly enhances shape reconstruction accuracy but also maintains the\ncomputational efficiency intrinsic to Gaussian Splatting. It achieves a Chamfer\ndistance error comparable to NeuraLangelo on the DTU dataset and maintains\nsimilar computational efficiency as the original 3D GS methods. Our method is a\nsignificant advancement in Gaussian Splatting and can be directly integrated\ninto existing Gaussian Splatting-based methods.\n", "date": "2024-06-03", "categories": []}
{"id": "2406.11836", "title": "RetinaGS: Scalable Training for Dense Scene Rendering with Billion-Scale\n  3D Gaussians", "abstract": "  In this work, we explore the possibility of training high-parameter 3D\nGaussian splatting (3DGS) models on large-scale, high-resolution datasets. We\ndesign a general model parallel training method for 3DGS, named RetinaGS, which\nuses a proper rendering equation and can be applied to any scene and arbitrary\ndistribution of Gaussian primitives. It enables us to explore the scaling\nbehavior of 3DGS in terms of primitive numbers and training resolutions that\nwere difficult to explore before and surpass previous state-of-the-art\nreconstruction quality. We observe a clear positive trend of increasing visual\nquality when increasing primitive numbers with our method. We also demonstrate\nthe first attempt at training a 3DGS model with more than one billion\nprimitives on the full MatrixCity dataset that attains a promising visual\nquality.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.15643", "title": "Taming 3DGS: High-Quality Radiance Fields with Limited Resources", "abstract": "  3D Gaussian Splatting (3DGS) has transformed novel-view synthesis with its\nfast, interpretable, and high-fidelity rendering. However, its resource\nrequirements limit its usability. Especially on constrained devices, training\nperformance degrades quickly and often cannot complete due to excessive memory\nconsumption of the model. The method converges with an indefinite number of\nGaussians -- many of them redundant -- making rendering unnecessarily slow and\npreventing its usage in downstream tasks that expect fixed-size inputs. To\naddress these issues, we tackle the challenges of training and rendering 3DGS\nmodels on a budget. We use a guided, purely constructive densification process\nthat steers densification toward Gaussians that raise the reconstruction\nquality. Model size continuously increases in a controlled manner towards an\nexact budget, using score-based densification of Gaussians with training-time\npriors that measure their contribution. We further address training speed\nobstacles: following a careful analysis of 3DGS' original pipeline, we derive\nfaster, numerically equivalent solutions for gradient computation and attribute\nupdates, including an alternative parallelization for efficient\nbackpropagation. We also propose quality-preserving approximations where\nsuitable to reduce training time even further. Taken together, these\nenhancements yield a robust, scalable solution with reduced training times,\nlower compute and memory requirements, and high quality. Our evaluation shows\nthat in a budgeted setting, we obtain competitive quality metrics with 3DGS\nwhile achieving a 4--5x reduction in both model size and training time. With\nmore generous budgets, our measured quality surpasses theirs. These advances\nopen the door for novel-view synthesis in constrained environments, e.g.,\nmobile devices.\n", "date": "2024-06-21", "categories": []}
{"id": "2306.11368", "title": "RoMe: Towards Large Scale Road Surface Reconstruction via Mesh\n  Representation", "abstract": "  In autonomous driving applications, accurate and efficient road surface\nreconstruction is paramount. This paper introduces RoMe, a novel framework\ndesigned for the robust reconstruction of large-scale road surfaces. Leveraging\na unique mesh representation, RoMe ensures that the reconstructed road surfaces\nare accurate and seamlessly aligned with semantics. To address challenges in\ncomputational efficiency, we propose a waypoint sampling strategy, enabling\nRoMe to reconstruct vast environments by focusing on sub-areas and subsequently\nmerging them. Furthermore, we incorporate an extrinsic optimization module to\nenhance the robustness against inaccuracies in extrinsic calibration. Our\nextensive evaluations of both public datasets and wild data underscore RoMe's\nsuperiority in terms of speed, accuracy, and robustness. For instance, it costs\nonly 2 GPU hours to recover a road surface of 600*600 square meters from\nthousands of images. Notably, RoMe's capability extends beyond mere\nreconstruction, offering significant value for autolabeling tasks in autonomous\ndriving applications. All related data and code are available at\nhttps://github.com/DRosemei/RoMe.\n", "date": "2023-06-20", "categories": []}
{"id": "2406.12054", "title": "FAWN: Floor-And-Walls Normal Regularization for Direct Neural TSDF\n  Reconstruction", "abstract": "  Leveraging 3D semantics for direct 3D reconstruction has a great potential\nyet unleashed. For instance, by assuming that walls are vertical, and a floor\nis planar and horizontal, we can correct distorted room shapes and eliminate\nlocal artifacts such as holes, pits, and hills. In this paper, we propose FAWN,\na modification of truncated signed distance function (TSDF) reconstruction\nmethods, which considers scene structure by detecting walls and floor in a\nscene, and penalizing the corresponding surface normals for deviating from the\nhorizontal and vertical directions. Implemented as a 3D sparse convolutional\nmodule, FAWN can be incorporated into any trainable pipeline that predicts\nTSDF. Since FAWN requires 3D semantics only for training, no additional\nlimitations on further use are imposed. We demonstrate, that FAWN-modified\nmethods use semantics more effectively, than existing semantic-based\napproaches. Besides, we apply our modification to state-of-the-art TSDF\nreconstruction methods, and demonstrate a quality gain in SCANNET, ICL-NUIM,\nTUM RGB-D, and 7SCENES benchmarks.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.10167", "title": "4DRecons: 4D Neural Implicit Deformable Objects Reconstruction from a\n  single RGB-D Camera with Geometrical and Topological Regularizations", "abstract": "  This paper presents a novel approach 4DRecons that takes a single camera\nRGB-D sequence of a dynamic subject as input and outputs a complete textured\ndeforming 3D model over time. 4DRecons encodes the output as a 4D neural\nimplicit surface and presents an optimization procedure that combines a data\nterm and two regularization terms. The data term fits the 4D implicit surface\nto the input partial observations. We address fundamental challenges in fitting\na complete implicit surface to partial observations. The first regularization\nterm enforces that the deformation among adjacent frames is as rigid as\npossible (ARAP). To this end, we introduce a novel approach to compute\ncorrespondences between adjacent textured implicit surfaces, which are used to\ndefine the ARAP regularization term. The second regularization term enforces\nthat the topology of the underlying object remains fixed over time. This\nregularization is critical for avoiding self-intersections that are typical in\nimplicit-based reconstructions. We have evaluated the performance of 4DRecons\non a variety of datasets. Experimental results show that 4DRecons can handle\nlarge deformations and complex inter-part interactions and outperform\nstate-of-the-art approaches considerably.\n", "date": "2024-06-14", "categories": []}
{"id": "2206.01290", "title": "Points2NeRF: Generating Neural Radiance Fields from 3D point cloud", "abstract": "  Contemporary registration devices for 3D visual information, such as LIDARs\nand various depth cameras, capture data as 3D point clouds. In turn, such\nclouds are challenging to be processed due to their size and complexity.\nExisting methods address this problem by fitting a mesh to the point cloud and\nrendering it instead. This approach, however, leads to the reduced fidelity of\nthe resulting visualization and misses color information of the objects crucial\nin computer graphics applications. In this work, we propose to mitigate this\nchallenge by representing 3D objects as Neural Radiance Fields (NeRFs). We\nleverage a hypernetwork paradigm and train the model to take a 3D point cloud\nwith the associated color values and return a NeRF network's weights that\nreconstruct 3D objects from input 2D images. Our method provides efficient 3D\nobject representation and offers several advantages over the existing\napproaches, including the ability to condition NeRFs and improved\ngeneralization beyond objects seen in training. The latter we also confirmed in\nthe results of our empirical evaluation.\n", "date": "2022-06-02", "categories": []}
{"id": "2312.13729", "title": "Gaussian Splatting with NeRF-based Color and Opacity", "abstract": "  Neural Radiance Fields (NeRFs) have demonstrated the remarkable potential of\nneural networks to capture the intricacies of 3D objects. By encoding the shape\nand color information within neural network weights, NeRFs excel at producing\nstrikingly sharp novel views of 3D objects. Recently, numerous generalizations\nof NeRFs utilizing generative models have emerged, expanding its versatility.\nIn contrast, Gaussian Splatting (GS) offers a similar render quality with\nfaster training and inference as it does not need neural networks to work. It\nencodes information about the 3D objects in the set of Gaussian distributions\nthat can be rendered in 3D similarly to classical meshes. Unfortunately, GS are\ndifficult to condition since they usually require circa hundred thousand\nGaussian components. To mitigate the caveats of both models, we propose a\nhybrid model Viewing Direction Gaussian Splatting (VDGS) that uses GS\nrepresentation of the 3D object's shape and NeRF-based encoding of color and\nopacity. Our model uses Gaussian distributions with trainable positions (i.e.\nmeans of Gaussian), shape (i.e. covariance of Gaussian), color and opacity, and\na neural network that takes Gaussian parameters and viewing direction to\nproduce changes in the said color and opacity. As a result, our model better\ndescribes shadows, light reflections, and the transparency of 3D objects\nwithout adding additional texture and light components.\n", "date": "2023-12-21", "categories": []}
{"id": "2406.08009", "title": "OpenObj: Open-Vocabulary Object-Level Neural Radiance Fields with\n  Fine-Grained Understanding", "abstract": "  In recent years, there has been a surge of interest in open-vocabulary 3D\nscene reconstruction facilitated by visual language models (VLMs), which\nshowcase remarkable capabilities in open-set retrieval. However, existing\nmethods face some limitations: they either focus on learning point-wise\nfeatures, resulting in blurry semantic understanding, or solely tackle\nobject-level reconstruction, thereby overlooking the intricate details of the\nobject's interior. To address these challenges, we introduce OpenObj, an\ninnovative approach to build open-vocabulary object-level Neural Radiance\nFields (NeRF) with fine-grained understanding. In essence, OpenObj establishes\na robust framework for efficient and watertight scene modeling and\ncomprehension at the object-level. Moreover, we incorporate part-level features\ninto the neural fields, enabling a nuanced representation of object interiors.\nThis approach captures object-level instances while maintaining a fine-grained\nunderstanding. The results on multiple datasets demonstrate that OpenObj\nachieves superior performance in zero-shot semantic segmentation and retrieval\ntasks. Additionally, OpenObj supports real-world robotics tasks at multiple\nscales, including global movement and local manipulation.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.08176", "title": "Category-level Neural Field for Reconstruction of Partially Observed\n  Objects in Indoor Environment", "abstract": "  Neural implicit representation has attracted attention in 3D reconstruction\nthrough various success cases. For further applications such as scene\nunderstanding or editing, several works have shown progress towards object\ncompositional reconstruction. Despite their superior performance in observed\nregions, their performance is still limited in reconstructing objects that are\npartially observed. To better treat this problem, we introduce category-level\nneural fields that learn meaningful common 3D information among objects\nbelonging to the same category present in the scene. Our key idea is to\nsubcategorize objects based on their observed shape for better training of the\ncategory-level model. Then we take advantage of the neural field to conduct the\nchallenging task of registering partially observed objects by selecting and\naligning against representative objects selected by ray-based uncertainty.\nExperiments on both simulation and real-world datasets demonstrate that our\nmethod improves the reconstruction of unobserved parts for several categories.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.08479", "title": "Real3D: Scaling Up Large Reconstruction Models with Real-World Images", "abstract": "  The default strategy for training single-view Large Reconstruction Models\n(LRMs) follows the fully supervised route using large-scale datasets of\nsynthetic 3D assets or multi-view captures. Although these resources simplify\nthe training procedure, they are hard to scale up beyond the existing datasets\nand they are not necessarily representative of the real distribution of object\nshapes. To address these limitations, in this paper, we introduce Real3D, the\nfirst LRM system that can be trained using single-view real-world images.\nReal3D introduces a novel self-training framework that can benefit from both\nthe existing synthetic data and diverse single-view real images. We propose two\nunsupervised losses that allow us to supervise LRMs at the pixel- and\nsemantic-level, even for training examples without ground-truth 3D or novel\nviews. To further improve performance and scale up the image data, we develop\nan automatic data curation approach to collect high-quality examples from\nin-the-wild images. Our experiments show that Real3D consistently outperforms\nprior work in four diverse evaluation settings that include real and synthetic\ndata, as well as both in-domain and out-of-domain shapes. Code and model can be\nfound here: https://hwjiang1510.github.io/Real3D/\n", "date": "2024-06-12", "categories": []}
{"id": "2402.01915", "title": "Robust Inverse Graphics via Probabilistic Inference", "abstract": "  How do we infer a 3D scene from a single image in the presence of corruptions\nlike rain, snow or fog? Straightforward domain randomization relies on knowing\nthe family of corruptions ahead of time. Here, we propose a Bayesian\napproach-dubbed robust inverse graphics (RIG)-that relies on a strong scene\nprior and an uninformative uniform corruption prior, making it applicable to a\nwide range of corruptions. Given a single image, RIG performs posterior\ninference jointly over the scene and the corruption. We demonstrate this idea\nby training a neural radiance field (NeRF) scene prior and using a secondary\nNeRF to represent the corruptions over which we place an uninformative prior.\nRIG, trained only on clean data, outperforms depth estimators and alternative\nNeRF approaches that perform point estimation instead of full inference. The\nresults hold for a number of scene prior architectures based on normalizing\nflows and diffusion models. For the latter, we develop reconstruction-guidance\nwith auxiliary latents (ReGAL)-a diffusion conditioning algorithm that is\napplicable in the presence of auxiliary latent variables such as the\ncorruption. RIG demonstrates how scene priors can be used beyond generation\ntasks.\n", "date": "2024-02-02", "categories": []}
{"id": "2406.03461", "title": "Polarization Wavefront Lidar: Learning Large Scene Reconstruction from\n  Polarized Wavefronts", "abstract": "  Lidar has become a cornerstone sensing modality for 3D vision, especially for\nlarge outdoor scenarios and autonomous driving. Conventional lidar sensors are\ncapable of providing centimeter-accurate distance information by emitting laser\npulses into a scene and measuring the time-of-flight (ToF) of the reflection.\nHowever, the polarization of the received light that depends on the surface\norientation and material properties is usually not considered. As such, the\npolarization modality has the potential to improve scene reconstruction beyond\ndistance measurements. In this work, we introduce a novel long-range\npolarization wavefront lidar sensor (PolLidar) that modulates the polarization\nof the emitted and received light. Departing from conventional lidar sensors,\nPolLidar allows access to the raw time-resolved polarimetric wavefronts. We\nleverage polarimetric wavefronts to estimate normals, distance, and material\nproperties in outdoor scenarios with a novel learned reconstruction method. To\ntrain and evaluate the method, we introduce a simulated and real-world\nlong-range dataset with paired raw lidar data, ground truth distance, and\nnormal maps. We find that the proposed method improves normal and distance\nreconstruction by 53\\% mean angular error and 41\\% mean absolute error compared\nto existing shape-from-polarization (SfP) and ToF methods. Code and data are\nopen-sourced at https://light.princeton.edu/pollidar.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.06972", "title": "Generative Lifting of Multiview to 3D from Unknown Pose: Wrapping NeRF\n  inside Diffusion", "abstract": "  We cast multiview reconstruction from unknown pose as a generative modeling\nproblem. From a collection of unannotated 2D images of a scene, our approach\nsimultaneously learns both a network to predict camera pose from 2D image\ninput, as well as the parameters of a Neural Radiance Field (NeRF) for the 3D\nscene. To drive learning, we wrap both the pose prediction network and NeRF\ninside a Denoising Diffusion Probabilistic Model (DDPM) and train the system\nvia the standard denoising objective. Our framework requires the system\naccomplish the task of denoising an input 2D image by predicting its pose and\nrendering the NeRF from that pose. Learning to denoise thus forces the system\nto concurrently learn the underlying 3D NeRF representation and a mapping from\nimages to camera extrinsic parameters. To facilitate the latter, we design a\ncustom network architecture to represent pose as a distribution, granting\nimplicit capacity for discovering view correspondences when trained end-to-end\nfor denoising alone. This technique allows our system to successfully build\nNeRFs, without pose knowledge, for challenging scenes where competing methods\nfail. At the conclusion of training, our learned NeRF can be extracted and used\nas a 3D scene model; our full system can be used to sample novel camera poses\nand generate novel-view images.\n", "date": "2024-06-11", "categories": []}
{"id": "2406.07111", "title": "NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse\n  Polarized Images", "abstract": "  We present NeRSP, a Neural 3D reconstruction technique for Reflective\nsurfaces with Sparse Polarized images. Reflective surface reconstruction is\nextremely challenging as specular reflections are view-dependent and thus\nviolate the multiview consistency for multiview stereo. On the other hand,\nsparse image inputs, as a practical capture setting, commonly cause incomplete\nor distorted results due to the lack of correspondence matching. This paper\njointly handles the challenges from sparse inputs and reflective surfaces by\nleveraging polarized images. We derive photometric and geometric cues from the\npolarimetric image formation model and multiview azimuth consistency, which\njointly optimize the surface geometry modeled via implicit neural\nrepresentation. Based on the experiments on our synthetic and real datasets, we\nachieve the state-of-the-art surface reconstruction results with only 6 views\nas input.\n", "date": "2024-06-11", "categories": []}
{"id": "2406.07499", "title": "Trim 3D Gaussian Splatting for Accurate Geometry Representation", "abstract": "  In this paper, we introduce Trim 3D Gaussian Splatting (TrimGS) to\nreconstruct accurate 3D geometry from images. Previous arts for geometry\nreconstruction from 3D Gaussians mainly focus on exploring strong geometry\nregularization. Instead, from a fresh perspective, we propose to obtain\naccurate 3D geometry of a scene by Gaussian trimming, which selectively removes\nthe inaccurate geometry while preserving accurate structures. To achieve this,\nwe analyze the contributions of individual 3D Gaussians and propose a\ncontribution-based trimming strategy to remove the redundant or inaccurate\nGaussians. Furthermore, our experimental and theoretical analyses reveal that a\nrelatively small Gaussian scale is a non-negligible factor in representing and\noptimizing the intricate details. Therefore the proposed TrimGS maintains\nrelatively small Gaussian scales. In addition, TrimGS is also compatible with\nthe effective geometry regularization strategies in previous arts. When\ncombined with the original 3DGS and the state-of-the-art 2DGS, TrimGS\nconsistently yields more accurate geometry and higher perceptual quality. Our\nproject page is https://trimgs.github.io\n", "date": "2024-06-11", "categories": []}
{"id": "2406.05533", "title": "PAPR in Motion: Seamless Point-level 3D Scene Interpolation", "abstract": "  We propose the problem of point-level 3D scene interpolation, which aims to\nsimultaneously reconstruct a 3D scene in two states from multiple views,\nsynthesize smooth point-level interpolations between them, and render the scene\nfrom novel viewpoints, all without any supervision between the states. The\nprimary challenge is on achieving a smooth transition between states that may\ninvolve significant and non-rigid changes. To address these challenges, we\nintroduce \"PAPR in Motion\", a novel approach that builds upon the recent\nProximity Attention Point Rendering (PAPR) technique, which can deform a point\ncloud to match a significantly different shape and render a visually coherent\nscene even after non-rigid deformations. Our approach is specifically designed\nto maintain the temporal consistency of the geometric structure by introducing\nvarious regularization techniques for PAPR. The result is a method that can\neffectively bridge large scene changes and produce visually coherent and\ntemporally smooth interpolations in both geometry and appearance. Evaluation\nacross diverse motion types demonstrates that \"PAPR in Motion\" outperforms the\nleading neural renderer for dynamic scenes. For more results and code, please\nvisit our project website at https://niopeng.github.io/PAPR-in-Motion/ .\n", "date": "2024-06-08", "categories": []}
{"id": "2402.08876", "title": "DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling", "abstract": "  In recent years, there has been a growing interest in training Neural\nNetworks to approximate Unsigned Distance Fields (UDFs) for representing open\nsurfaces in the context of 3D reconstruction. However, UDFs are\nnon-differentiable at the zero level set which leads to significant errors in\ndistances and gradients, generally resulting in fragmented and discontinuous\nsurfaces. In this paper, we propose to learn a hyperbolic scaling of the\nunsigned distance field, which defines a new Eikonal problem with distinct\nboundary conditions. This allows our formulation to integrate seamlessly with\nstate-of-the-art continuously differentiable implicit neural representation\nnetworks, largely applied in the literature to represent signed distance\nfields. Our approach not only addresses the challenge of open surface\nrepresentation but also demonstrates significant improvement in reconstruction\nquality and training performance. Moreover, the unlocked field's\ndifferentiability allows the accurate computation of essential topological\nproperties such as normal directions and curvatures, pervasive in downstream\ntasks such as rendering. Through extensive experiments, we validate our\napproach across various data sets and against competitive baselines. The\nresults demonstrate enhanced accuracy and up to an order of magnitude increase\nin speed compared to previous methods.\n", "date": "2024-02-14", "categories": []}
{"id": "2406.03697", "title": "Superpoint Gaussian Splatting for Real-Time High-Fidelity Dynamic Scene\n  Reconstruction", "abstract": "  Rendering novel view images in dynamic scenes is a crucial yet challenging\ntask. Current methods mainly utilize NeRF-based methods to represent the static\nscene and an additional time-variant MLP to model scene deformations, resulting\nin relatively low rendering quality as well as slow inference speed. To tackle\nthese challenges, we propose a novel framework named Superpoint Gaussian\nSplatting (SP-GS). Specifically, our framework first employs explicit 3D\nGaussians to reconstruct the scene and then clusters Gaussians with similar\nproperties (e.g., rotation, translation, and location) into superpoints.\nEmpowered by these superpoints, our method manages to extend 3D Gaussian\nsplatting to dynamic scenes with only a slight increase in computational\nexpense. Apart from achieving state-of-the-art visual quality and real-time\nrendering under high resolutions, the superpoint representation provides a\nstronger manipulation capability. Extensive experiments demonstrate the\npracticality and effectiveness of our approach on both synthetic and real-world\ndatasets. Please see our project page at\nhttps://dnvtmf.github.io/SP_GS.github.io.\n", "date": "2024-06-06", "categories": []}
{"id": "2302.03640", "title": "SSR-2D: Semantic 3D Scene Reconstruction from 2D Images", "abstract": "  Most deep learning approaches to comprehensive semantic modeling of 3D indoor\nspaces require costly dense annotations in the 3D domain. In this work, we\nexplore a central 3D scene modeling task, namely, semantic scene reconstruction\nwithout using any 3D annotations. The key idea of our approach is to design a\ntrainable model that employs both incomplete 3D reconstructions and their\ncorresponding source RGB-D images, fusing cross-domain features into volumetric\nembeddings to predict complete 3D geometry, color, and semantics with only 2D\nlabeling which can be either manual or machine-generated. Our key technical\ninnovation is to leverage differentiable rendering of color and semantics to\nbridge 2D observations and unknown 3D space, using the observed RGB images and\n2D semantics as supervision, respectively. We additionally develop a learning\npipeline and corresponding method to enable learning from imperfect predicted\n2D labels, which could be additionally acquired by synthesizing in an augmented\nset of virtual training views complementing the original real captures,\nenabling more efficient self-supervision loop for semantics. As a result, our\nend-to-end trainable solution jointly addresses geometry completion,\ncolorization, and semantic mapping from limited RGB-D images, without relying\non any 3D ground-truth information. Our method achieves the state-of-the-art\nperformance of semantic scene completion on two large-scale benchmark datasets\nMatterPort3D and ScanNet, surpasses baselines even with costly 3D annotations\nin predicting both geometry and semantics. To our knowledge, our method is also\nthe first 2D-driven method addressing completion and semantic segmentation of\nreal-world 3D scans simultaneously.\n", "date": "2023-02-07", "categories": []}
{"id": "2406.02495", "title": "GenS: Generalizable Neural Surface Reconstruction from Multi-View Images", "abstract": "  Combining the signed distance function (SDF) and differentiable volume\nrendering has emerged as a powerful paradigm for surface reconstruction from\nmulti-view images without 3D supervision. However, current methods are impeded\nby requiring long-time per-scene optimizations and cannot generalize to new\nscenes. In this paper, we present GenS, an end-to-end generalizable neural\nsurface reconstruction model. Unlike coordinate-based methods that train a\nseparate network for each scene, we construct a generalized multi-scale volume\nto directly encode all scenes. Compared with existing solutions, our\nrepresentation is more powerful, which can recover high-frequency details while\nmaintaining global smoothness. Meanwhile, we introduce a multi-scale\nfeature-metric consistency to impose the multi-view consistency in a more\ndiscriminative multi-scale feature space, which is robust to the failures of\nthe photometric consistency. And the learnable feature can be self-enhanced to\ncontinuously improve the matching accuracy and mitigate aggregation ambiguity.\nFurthermore, we design a view contrast loss to force the model to be robust to\nthose regions covered by few viewpoints through distilling the geometric prior\nfrom dense input to sparse input. Extensive experiments on popular benchmarks\nshow that our model can generalize well to new scenes and outperform existing\nstate-of-the-art methods even those employing ground-truth depth supervision.\nCode is available at https://github.com/prstrive/GenS.\n", "date": "2024-06-04", "categories": []}
{"id": "2405.11021", "title": "Enhanced 3D Urban Scene Reconstruction and Point Cloud Densification\n  using Gaussian Splatting and Google Earth Imagery", "abstract": "  3D urban scene reconstruction and modelling is a crucial research area in\nremote sensing with numerous applications in academia, commerce, industry, and\nadministration. Recent advancements in view synthesis models have facilitated\nphotorealistic 3D reconstruction solely from 2D images. Leveraging Google Earth\nimagery, we construct a 3D Gaussian Splatting model of the Waterloo region\ncentered on the University of Waterloo and are able to achieve view-synthesis\nresults far exceeding previous 3D view-synthesis results based on neural\nradiance fields which we demonstrate in our benchmark. Additionally, we\nretrieved the 3D geometry of the scene using the 3D point cloud extracted from\nthe 3D Gaussian Splatting model which we benchmarked against our Multi-\nView-Stereo dense reconstruction of the scene, thereby reconstructing both the\n3D geometry and photorealistic lighting of the large-scale urban scene through\n3D Gaussian Splatting\n", "date": "2024-05-17", "categories": []}
{"id": "2405.16517", "title": "Sp2360: Sparse-view 360 Scene Reconstruction using Cascaded 2D Diffusion\n  Priors", "abstract": "  We aim to tackle sparse-view reconstruction of a 360 3D scene using priors\nfrom latent diffusion models (LDM). The sparse-view setting is ill-posed and\nunderconstrained, especially for scenes where the camera rotates 360 degrees\naround a point, as no visual information is available beyond some frontal views\nfocused on the central object(s) of interest. In this work, we show that\npretrained 2D diffusion models can strongly improve the reconstruction of a\nscene with low-cost fine-tuning. Specifically, we present SparseSplat360\n(Sp2360), a method that employs a cascade of in-painting and artifact removal\nmodels to fill in missing details and clean novel views. Due to superior\ntraining and rendering speeds, we use an explicit scene representation in the\nform of 3D Gaussians over NeRF-based implicit representations. We propose an\niterative update strategy to fuse generated pseudo novel views with existing 3D\nGaussians fitted to the initial sparse inputs. As a result, we obtain a\nmulti-view consistent scene representation with details coherent with the\nobserved inputs. Our evaluation on the challenging Mip-NeRF360 dataset shows\nthat our proposed 2D to 3D distillation algorithm considerably improves the\nperformance of a regularized version of 3DGS adapted to a sparse-view setting\nand outperforms existing sparse-view reconstruction methods in 360 scene\nreconstruction. Qualitatively, our method generates entire 360 scenes from as\nfew as 9 input views, with a high degree of foreground and background detail.\n", "date": "2024-05-26", "categories": []}
{"id": "2405.20310", "title": "A Pixel Is Worth More Than One 3D Gaussians in Single-View 3D\n  Reconstruction", "abstract": "  Learning 3D scene representation from a single-view image is a long-standing\nfundamental problem in computer vision, with the inherent ambiguity in\npredicting contents unseen from the input view. Built on the recently proposed\n3D Gaussian Splatting (3DGS), the Splatter Image method has made promising\nprogress on fast single-image novel view synthesis via learning a single 3D\nGaussian for each pixel based on the U-Net feature map of an input image.\nHowever, it has limited expressive power to represent occluded components that\nare not observable in the input view. To address this problem, this paper\npresents a Hierarchical Splatter Image method in which a pixel is worth more\nthan one 3D Gaussians. Specifically, each pixel is represented by a parent 3D\nGaussian and a small number of child 3D Gaussians. Parent 3D Gaussians are\nlearned as done in the vanilla Splatter Image. Child 3D Gaussians are learned\nvia a lightweight Multi-Layer Perceptron (MLP) which takes as input the\nprojected image features of a parent 3D Gaussian and the embedding of a target\ncamera view. Both parent and child 3D Gaussians are learned end-to-end in a\nstage-wise way. The joint condition of input image features from eyes of the\nparent Gaussians and the target camera position facilitates learning to\nallocate child Gaussians to ``see the unseen'', recovering the occluded details\nthat are often missed by parent Gaussians.\n  In experiments, the proposed method is tested on the ShapeNet-SRN and CO3D\ndatasets with state-of-the-art performance obtained, especially showing\npromising capabilities of reconstructing occluded contents in the input view.\n", "date": "2024-05-30", "categories": []}
{"id": "2406.00750", "title": "Freeplane: Unlocking Free Lunch in Triplane-Based Sparse-View\n  Reconstruction Models", "abstract": "  Creating 3D assets from single-view images is a complex task that demands a\ndeep understanding of the world. Recently, feed-forward 3D generative models\nhave made significant progress by training large reconstruction models on\nextensive 3D datasets, with triplanes being the preferred 3D geometry\nrepresentation. However, effectively utilizing the geometric priors of\ntriplanes, while minimizing artifacts caused by generated inconsistent\nmulti-view images, remains a challenge. In this work, we present\n\\textbf{Fre}quency modulat\\textbf{e}d tri\\textbf{plane} (\\textbf{Freeplane}), a\nsimple yet effective method to improve the generation quality of feed-forward\nmodels without additional training. We first analyze the role of triplanes in\nfeed-forward methods and find that the inconsistent multi-view images introduce\nhigh-frequency artifacts on triplanes, leading to low-quality 3D meshes. Based\non this observation, we propose strategically filtering triplane features and\ncombining triplanes before and after filtering to produce high-quality textured\nmeshes. These techniques incur no additional cost and can be seamlessly\nintegrated into pre-trained feed-forward models to enhance their robustness\nagainst the inconsistency of generated multi-view images. Both qualitative and\nquantitative results demonstrate that our method improves the performance of\nfeed-forward models by simply modulating triplanes. All you need is to modulate\nthe triplanes during inference.\n", "date": "2024-06-02", "categories": []}
{"id": "2405.17187", "title": "Memorize What Matters: Emergent Scene Decomposition from Multitraverse", "abstract": "  Humans naturally retain memories of permanent elements, while ephemeral\nmoments often slip through the cracks of memory. This selective retention is\ncrucial for robotic perception, localization, and mapping. To endow robots with\nthis capability, we introduce 3D Gaussian Mapping (3DGM), a self-supervised,\ncamera-only offline mapping framework grounded in 3D Gaussian Splatting. 3DGM\nconverts multitraverse RGB videos from the same region into a Gaussian-based\nenvironmental map while concurrently performing 2D ephemeral object\nsegmentation. Our key observation is that the environment remains consistent\nacross traversals, while objects frequently change. This allows us to exploit\nself-supervision from repeated traversals to achieve environment-object\ndecomposition. More specifically, 3DGM formulates multitraverse environmental\nmapping as a robust differentiable rendering problem, treating pixels of the\nenvironment and objects as inliers and outliers, respectively. Using robust\nfeature distillation, feature residuals mining, and robust optimization, 3DGM\njointly performs 2D segmentation and 3D mapping without human intervention. We\nbuild the Mapverse benchmark, sourced from the Ithaca365 and nuPlan datasets,\nto evaluate our method in unsupervised 2D segmentation, 3D reconstruction, and\nneural rendering. Extensive results verify the effectiveness and potential of\nour method for self-driving and robotics.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.17835", "title": "Deform3DGS: Flexible Deformation for Fast Surgical Scene Reconstruction\n  with Gaussian Splatting", "abstract": "  Tissue deformation poses a key challenge for accurate surgical scene\nreconstruction. Despite yielding high reconstruction quality, existing methods\nsuffer from slow rendering speeds and long training times, limiting their\nintraoperative applicability. Motivated by recent progress in 3D Gaussian\nSplatting, an emerging technology in real-time 3D rendering, this work presents\na novel fast reconstruction framework, termed Deform3DGS, for deformable\ntissues during endoscopic surgery. Specifically, we introduce 3D GS into\nsurgical scenes by integrating a point cloud initialization to improve\nreconstruction. Furthermore, we propose a novel flexible deformation modeling\nscheme (FDM) to learn tissue deformation dynamics at the level of individual\nGaussians. Our FDM can model the surface deformation with efficient\nrepresentations, allowing for real-time rendering performance. More\nimportantly, FDM significantly accelerates surgical scene reconstruction,\ndemonstrating considerable clinical values, particularly in intraoperative\nsettings where time efficiency is crucial. Experiments on DaVinci robotic\nsurgery videos indicate the efficacy of our approach, showcasing superior\nreconstruction fidelity PSNR: (37.90) and rendering speed (338.8 FPS) while\nsubstantially reducing training time to only 1 minute/scene. Our code is\navailable at https://github.com/jinlab-imvr/Deform3DGS.\n", "date": "2024-05-28", "categories": []}
{"id": "2405.19657", "title": "Uncertainty-guided Optimal Transport in Depth Supervised Sparse-View 3D\n  Gaussian", "abstract": "  3D Gaussian splatting has demonstrated impressive performance in real-time\nnovel view synthesis. However, achieving successful reconstruction from RGB\nimages generally requires multiple input views captured under static\nconditions. To address the challenge of sparse input views, previous approaches\nhave incorporated depth supervision into the training of 3D Gaussians to\nmitigate overfitting, using dense predictions from pretrained depth networks as\npseudo-ground truth. Nevertheless, depth predictions from monocular depth\nestimation models inherently exhibit significant uncertainty in specific areas.\nRelying solely on pixel-wise L2 loss may inadvertently incorporate detrimental\nnoise from these uncertain areas. In this work, we introduce a novel method to\nsupervise the depth distribution of 3D Gaussians, utilizing depth priors with\nintegrated uncertainty estimates. To address these localized errors in depth\npredictions, we integrate a patch-wise optimal transport strategy to complement\ntraditional L2 loss in depth supervision. Extensive experiments conducted on\nthe LLFF, DTU, and Blender datasets demonstrate that our approach, UGOT,\nachieves superior novel view synthesis and consistently outperforms\nstate-of-the-art methods.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.19819", "title": "Gated Fields: Learning Scene Reconstruction from Gated Videos", "abstract": "  Reconstructing outdoor 3D scenes from temporal observations is a challenge\nthat recent work on neural fields has offered a new avenue for. However,\nexisting methods that recover scene properties, such as geometry, appearance,\nor radiance, solely from RGB captures often fail when handling poorly-lit or\ntexture-deficient regions. Similarly, recovering scenes with scanning LiDAR\nsensors is also difficult due to their low angular sampling rate which makes\nrecovering expansive real-world scenes difficult. Tackling these gaps, we\nintroduce Gated Fields - a neural scene reconstruction method that utilizes\nactive gated video sequences. To this end, we propose a neural rendering\napproach that seamlessly incorporates time-gated capture and illumination. Our\nmethod exploits the intrinsic depth cues in the gated videos, achieving precise\nand dense geometry reconstruction irrespective of ambient illumination\nconditions. We validate the method across day and night scenarios and find that\nGated Fields compares favorably to RGB and LiDAR reconstruction methods. Our\ncode and datasets are available at https://light.princeton.edu/gatedfields/.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.20323", "title": "$\\textit{S}^3$Gaussian: Self-Supervised Street Gaussians for Autonomous\n  Driving", "abstract": "  Photorealistic 3D reconstruction of street scenes is a critical technique for\ndeveloping real-world simulators for autonomous driving. Despite the efficacy\nof Neural Radiance Fields (NeRF) for driving scenes, 3D Gaussian Splatting\n(3DGS) emerges as a promising direction due to its faster speed and more\nexplicit representation. However, most existing street 3DGS methods require\ntracked 3D vehicle bounding boxes to decompose the static and dynamic elements\nfor effective reconstruction, limiting their applications for in-the-wild\nscenarios. To facilitate efficient 3D scene reconstruction without costly\nannotations, we propose a self-supervised street Gaussian\n($\\textit{S}^3$Gaussian) method to decompose dynamic and static elements from\n4D consistency. We represent each scene with 3D Gaussians to preserve the\nexplicitness and further accompany them with a spatial-temporal field network\nto compactly model the 4D dynamics. We conduct extensive experiments on the\nchallenging Waymo-Open dataset to evaluate the effectiveness of our method. Our\n$\\textit{S}^3$Gaussian demonstrates the ability to decompose static and dynamic\nscenes and achieves the best performance without using 3D annotations. Code is\navailable at: https://github.com/nnanhuang/S3Gaussian/.\n", "date": "2024-05-30", "categories": []}
{"id": "2311.17089", "title": "Multi-Scale 3D Gaussian Splatting for Anti-Aliased Rendering", "abstract": "  3D Gaussians have recently emerged as a highly efficient representation for\n3D reconstruction and rendering. Despite its high rendering quality and speed\nat high resolutions, they both deteriorate drastically when rendered at lower\nresolutions or from far away camera position. During low resolution or far away\nrendering, the pixel size of the image can fall below the Nyquist frequency\ncompared to the screen size of each splatted 3D Gaussian and leads to aliasing\neffect. The rendering is also drastically slowed down by the sequential alpha\nblending of more splatted Gaussians per pixel. To address these issues, we\npropose a multi-scale 3D Gaussian splatting algorithm, which maintains\nGaussians at different scales to represent the same scene. Higher-resolution\nimages are rendered with more small Gaussians, and lower-resolution images are\nrendered with fewer larger Gaussians. With similar training time, our algorithm\ncan achieve 13\\%-66\\% PSNR and 160\\%-2400\\% rendering speed improvement at\n4$\\times$-128$\\times$ scale rendering on Mip-NeRF360 dataset compared to the\nsingle scale 3D Gaussian splitting. Our code and more results are available on\nour project website https://jokeryan.github.io/projects/ms-gs/\n", "date": "2023-11-28", "categories": []}
{"id": "2405.16829", "title": "PyGS: Large-scale Scene Representation with Pyramidal 3D Gaussian\n  Splatting", "abstract": "  Neural Radiance Fields (NeRFs) have demonstrated remarkable proficiency in\nsynthesizing photorealistic images of large-scale scenes. However, they are\noften plagued by a loss of fine details and long rendering durations. 3D\nGaussian Splatting has recently been introduced as a potent alternative,\nachieving both high-fidelity visual results and accelerated rendering\nperformance. Nonetheless, scaling 3D Gaussian Splatting is fraught with\nchallenges. Specifically, large-scale scenes grapples with the integration of\nobjects across multiple scales and disparate viewpoints, which often leads to\ncompromised efficacy as the Gaussians need to balance between detail levels.\nFurthermore, the generation of initialization points via COLMAP from\nlarge-scale dataset is both computationally demanding and prone to incomplete\nreconstructions. To address these challenges, we present Pyramidal 3D Gaussian\nSplatting (PyGS) with NeRF Initialization. Our approach represent the scene\nwith a hierarchical assembly of Gaussians arranged in a pyramidal fashion. The\ntop level of the pyramid is composed of a few large Gaussians, while each\nsubsequent layer accommodates a denser collection of smaller Gaussians. We\neffectively initialize these pyramidal Gaussians through sampling a rapidly\ntrained grid-based NeRF at various frequencies. We group these pyramidal\nGaussians into clusters and use a compact weighting network to dynamically\ndetermine the influence of each pyramid level of each cluster considering\ncamera viewpoint during rendering. Our method achieves a significant\nperformance leap across multiple large-scale datasets and attains a rendering\ntime that is over 400 times faster than current state-of-the-art approaches.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.19295", "title": "3D Neural Edge Reconstruction", "abstract": "  Real-world objects and environments are predominantly composed of edge\nfeatures, including straight lines and curves. Such edges are crucial elements\nfor various applications, such as CAD modeling, surface meshing, lane mapping,\netc. However, existing traditional methods only prioritize lines over curves\nfor simplicity in geometric modeling. To this end, we introduce EMAP, a new\nmethod for learning 3D edge representations with a focus on both lines and\ncurves. Our method implicitly encodes 3D edge distance and direction in\nUnsigned Distance Functions (UDF) from multi-view edge maps. On top of this\nneural representation, we propose an edge extraction algorithm that robustly\nabstracts parametric 3D edges from the inferred edge points and their\ndirections. Comprehensive evaluations demonstrate that our method achieves\nbetter 3D edge reconstruction on multiple challenging datasets. We further show\nthat our learned UDF field enhances neural surface reconstruction by capturing\nmore details.\n", "date": "2024-05-29", "categories": []}
{"id": "2303.16739", "title": "Active Implicit Object Reconstruction using Uncertainty-guided\n  Next-Best-View Optimization", "abstract": "  Actively planning sensor views during object reconstruction is crucial for\nautonomous mobile robots. An effective method should be able to strike a\nbalance between accuracy and efficiency. In this paper, we propose a seamless\nintegration of the emerging implicit representation with the active\nreconstruction task. We build an implicit occupancy field as our geometry\nproxy. While training, the prior object bounding box is utilized as auxiliary\ninformation to generate clean and detailed reconstructions. To evaluate view\nuncertainty, we employ a sampling-based approach that directly extracts entropy\nfrom the reconstructed occupancy probability field as our measure of view\ninformation gain. This eliminates the need for additional uncertainty maps or\nlearning. Unlike previous methods that compare view uncertainty within a finite\nset of candidates, we aim to find the next-best-view (NBV) on a continuous\nmanifold. Leveraging the differentiability of the implicit representation, the\nNBV can be optimized directly by maximizing the view uncertainty using gradient\ndescent. It significantly enhances the method's adaptability to different\nscenarios. Simulation and real-world experiments demonstrate that our approach\neffectively improves reconstruction accuracy and efficiency of view planning in\nactive reconstruction tasks. The proposed system will open source at\nhttps://github.com/HITSZ-NRSL/ActiveImplicitRecon.git.\n", "date": "2023-03-29", "categories": []}
{"id": "2403.09413", "title": "Relaxing Accurate Initialization Constraint for 3D Gaussian Splatting", "abstract": "  3D Gaussian splatting (3DGS) has recently demonstrated impressive\ncapabilities in real-time novel view synthesis and 3D reconstruction. However,\n3DGS heavily depends on the accurate initialization derived from\nStructure-from-Motion (SfM) methods. When the quality of the initial point\ncloud deteriorates, such as in the presence of noise or when using randomly\ninitialized point cloud, 3DGS often undergoes large performance drops. To\naddress this limitation, we propose a novel optimization strategy dubbed\nRAIN-GS (Relaing Accurate Initialization Constraint for 3D Gaussian Splatting).\nOur approach is based on an in-depth analysis of the original 3DGS optimization\nscheme and the analysis of the SfM initialization in the frequency domain.\nLeveraging simple modifications based on our analyses, RAIN-GS successfully\ntrains 3D Gaussians from sub-optimal point cloud (e.g., randomly initialized\npoint cloud), effectively relaxing the need for accurate initialization. We\ndemonstrate the efficacy of our strategy through quantitative and qualitative\ncomparisons on multiple datasets, where RAIN-GS trained with random point cloud\nachieves performance on-par with or even better than 3DGS trained with accurate\nSfM point cloud. Our project page and code can be found at\nhttps://ku-cvlab.github.io/RAIN-GS.\n", "date": "2024-03-14", "categories": []}
{"id": "2405.10554", "title": "NeRO: Neural Road Surface Reconstruction", "abstract": "  Accurately reconstructing road surfaces is pivotal for various applications\nespecially in autonomous driving. This paper introduces a position encoding\nMulti-Layer Perceptrons (MLPs) framework to reconstruct road surfaces, with\ninput as world coordinates x and y, and output as height, color, and semantic\ninformation. The effectiveness of this method is demonstrated through its\ncompatibility with a variety of road height sources like vehicle camera poses,\nLiDAR point clouds, and SFM point clouds, robust to the semantic noise of\nimages like sparse labels and noise semantic prediction, and fast training\nspeed, which indicates a promising application for rendering road surfaces with\nsemantics, particularly in applications demanding visualization of road\nsurface, 4D labeling, and semantic groupings.\n", "date": "2024-05-17", "categories": []}
{"id": "2405.16923", "title": "SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction\n  with Geometry Constrain", "abstract": "  With the emergence of Gaussian Splats, recent efforts have focused on\nlarge-scale scene geometric reconstruction. However, most of these efforts\neither concentrate on memory reduction or spatial space division, neglecting\ninformation in the semantic space. In this paper, we propose a novel method,\nnamed SA-GS, for fine-grained 3D geometry reconstruction using semantic-aware\n3D Gaussian Splats. Specifically, we leverage prior information stored in large\nvision models such as SAM and DINO to generate semantic masks. We then\nintroduce a geometric complexity measurement function to serve as soft\nregularization, guiding the shape of each Gaussian Splat within specific\nsemantic areas. Additionally, we present a method that estimates the expected\nnumber of Gaussian Splats in different semantic areas, effectively providing a\nlower bound for Gaussian Splats in these areas. Subsequently, we extract the\npoint cloud using a novel probability density-based extraction method,\ntransforming Gaussian Splats into a point cloud crucial for downstream tasks.\nOur method also offers the potential for detailed semantic inquiries while\nmaintaining high image-based reconstruction results. We provide extensive\nexperiments on publicly available large-scale scene reconstruction datasets\nwith highly accurate point clouds as ground truth and our novel dataset. Our\nresults demonstrate the superiority of our method over current state-of-the-art\nGaussian Splats reconstruction methods by a significant margin in terms of\ngeometric-based measurement metrics. Code and additional results will soon be\navailable on our project page.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.17083", "title": "F-3DGS: Factorized Coordinates and Representations for 3D Gaussian\n  Splatting", "abstract": "  The neural radiance field (NeRF) has made significant strides in representing\n3D scenes and synthesizing novel views. Despite its advancements, the high\ncomputational costs of NeRF have posed challenges for its deployment in\nresource-constrained environments and real-time applications. As an alternative\nto NeRF-like neural rendering methods, 3D Gaussian Splatting (3DGS) offers\nrapid rendering speeds while maintaining excellent image quality. However, as\nit represents objects and scenes using a myriad of Gaussians, it requires\nsubstantial storage to achieve high-quality representation. To mitigate the\nstorage overhead, we propose Factorized 3D Gaussian Splatting (F-3DGS), a novel\napproach that drastically reduces storage requirements while preserving image\nquality. Inspired by classical matrix and tensor factorization techniques, our\nmethod represents and approximates dense clusters of Gaussians with\nsignificantly fewer Gaussians through efficient factorization. We aim to\nefficiently represent dense 3D Gaussians by approximating them with a limited\namount of information for each axis and their combinations. This method allows\nus to encode a substantially large number of Gaussians along with their\nessential attributes -- such as color, scale, and rotation -- necessary for\nrendering using a relatively small number of elements. Extensive experimental\nresults demonstrate that F-3DGS achieves a significant reduction in storage\ncosts while maintaining comparable quality in rendered images.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.17891", "title": "A Refined 3D Gaussian Representation for High-Quality Dynamic Scene\n  Reconstruction", "abstract": "  In recent years, Neural Radiance Fields (NeRF) has revolutionized\nthree-dimensional (3D) reconstruction with its implicit representation.\nBuilding upon NeRF, 3D Gaussian Splatting (3D-GS) has departed from the\nimplicit representation of neural networks and instead directly represents\nscenes as point clouds with Gaussian-shaped distributions. While this shift has\nnotably elevated the rendering quality and speed of radiance fields but\ninevitably led to a significant increase in memory usage. Additionally,\neffectively rendering dynamic scenes in 3D-GS has emerged as a pressing\nchallenge. To address these concerns, this paper purposes a refined 3D Gaussian\nrepresentation for high-quality dynamic scene reconstruction. Firstly, we use a\ndeformable multi-layer perceptron (MLP) network to capture the dynamic offset\nof Gaussian points and express the color features of points through hash\nencoding and a tiny MLP to reduce storage requirements. Subsequently, we\nintroduce a learnable denoising mask coupled with denoising loss to eliminate\nnoise points from the scene, thereby further compressing 3D Gaussian model.\nFinally, motion noise of points is mitigated through static constraints and\nmotion consistency constraints. Experimental results demonstrate that our\nmethod surpasses existing approaches in rendering quality and speed, while\nsignificantly reducing the memory usage associated with 3D-GS, making it highly\nsuitable for various tasks such as novel view synthesis, and dynamic mapping.\n", "date": "2024-05-28", "categories": []}
{"id": "2403.18795", "title": "Gamba: Marry Gaussian Splatting with Mamba for single view 3D\n  reconstruction", "abstract": "  We tackle the challenge of efficiently reconstructing a 3D asset from a\nsingle image at millisecond speed. Existing methods for single-image 3D\nreconstruction are primarily based on Score Distillation Sampling (SDS) with\nNeural 3D representations. Despite promising results, these approaches\nencounter practical limitations due to lengthy optimizations and significant\nmemory consumption. In this work, we introduce Gamba, an end-to-end 3D\nreconstruction model from a single-view image, emphasizing two main insights:\n(1) Efficient Backbone Design: introducing a Mamba-based GambaFormer network to\nmodel 3D Gaussian Splatting (3DGS) reconstruction as sequential prediction with\nlinear scalability of token length, thereby accommodating a substantial number\nof Gaussians; (2) Robust Gaussian Constraints: deriving radial mask constraints\nfrom multi-view masks to eliminate the need for warmup supervision of 3D point\nclouds in training. We trained Gamba on Objaverse and assessed it against\nexisting optimization-based and feed-forward 3D reconstruction approaches on\nthe GSO Dataset, among which Gamba is the only end-to-end trained single-view\nreconstruction model with 3DGS. Experimental results demonstrate its\ncompetitive generation capabilities both qualitatively and quantitatively and\nhighlight its remarkable speed: Gamba completes reconstruction within 0.05\nseconds on a single NVIDIA A100 GPU, which is about $1,000\\times$ faster than\noptimization-based methods. Please see our project page at\nhttps://florinshen.github.io/gamba-project.\n", "date": "2024-03-27", "categories": []}
{"id": "2405.16822", "title": "Vidu4D: Single Generated Video to High-Fidelity 4D Reconstruction with\n  Dynamic Gaussian Surfels", "abstract": "  Video generative models are receiving particular attention given their\nability to generate realistic and imaginative frames. Besides, these models are\nalso observed to exhibit strong 3D consistency, significantly enhancing their\npotential to act as world simulators. In this work, we present Vidu4D, a novel\nreconstruction model that excels in accurately reconstructing 4D (i.e.,\nsequential 3D) representations from single generated videos, addressing\nchallenges associated with non-rigidity and frame distortion. This capability\nis pivotal for creating high-fidelity virtual contents that maintain both\nspatial and temporal coherence. At the core of Vidu4D is our proposed Dynamic\nGaussian Surfels (DGS) technique. DGS optimizes time-varying warping functions\nto transform Gaussian surfels (surface elements) from a static state to a\ndynamically warped state. This transformation enables a precise depiction of\nmotion and deformation over time. To preserve the structural integrity of\nsurface-aligned Gaussian surfels, we design the warped-state geometric\nregularization based on continuous warping fields for estimating normals.\nAdditionally, we learn refinements on rotation and scaling parameters of\nGaussian surfels, which greatly alleviates texture flickering during the\nwarping process and enhances the capture of fine-grained appearance details.\nVidu4D also contains a novel initialization state that provides a proper start\nfor the warping fields in DGS. Equipping Vidu4D with an existing video\ngenerative model, the overall framework demonstrates high-fidelity text-to-4D\ngeneration in both appearance and geometry.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.16888", "title": "Part123: Part-aware 3D Reconstruction from a Single-view Image", "abstract": "  Recently, the emergence of diffusion models has opened up new opportunities\nfor single-view reconstruction. However, all the existing methods represent the\ntarget object as a closed mesh devoid of any structural information, thus\nneglecting the part-based structure, which is crucial for many downstream\napplications, of the reconstructed shape. Moreover, the generated meshes\nusually suffer from large noises, unsmooth surfaces, and blurry textures,\nmaking it challenging to obtain satisfactory part segments using 3D\nsegmentation techniques. In this paper, we present Part123, a novel framework\nfor part-aware 3D reconstruction from a single-view image. We first use\ndiffusion models to generate multiview-consistent images from a given image,\nand then leverage Segment Anything Model (SAM), which demonstrates powerful\ngeneralization ability on arbitrary objects, to generate multiview segmentation\nmasks. To effectively incorporate 2D part-based information into 3D\nreconstruction and handle inconsistency, we introduce contrastive learning into\na neural rendering framework to learn a part-aware feature space based on the\nmultiview segmentation masks. A clustering-based algorithm is also developed to\nautomatically derive 3D part segmentation results from the reconstructed\nmodels. Experiments show that our method can generate 3D models with\nhigh-quality segmented parts on various objects. Compared to existing\nunstructured reconstruction methods, the part-aware 3D models from our method\nbenefit some important applications, including feature-preserving\nreconstruction, primitive fitting, and 3D shape editing.\n", "date": "2024-05-27", "categories": []}
{"id": "2210.02757", "title": "CAP-UDF: Learning Unsigned Distance Functions Progressively from Raw\n  Point Clouds with Consistency-Aware Field Optimization", "abstract": "  Surface reconstruction for point clouds is an important task in 3D computer\nvision. Most of the latest methods resolve this problem by learning signed\ndistance functions from point clouds, which are limited to reconstructing\nclosed surfaces. Some other methods tried to represent open surfaces using\nunsigned distance functions (UDF) which are learned from ground truth\ndistances. However, the learned UDF is hard to provide smooth distance fields\ndue to the discontinuous character of point clouds. In this paper, we propose\nCAP-UDF, a novel method to learn consistency-aware UDF from raw point clouds.\nWe achieve this by learning to move queries onto the surface with a field\nconsistency constraint, where we also enable to progressively estimate a more\naccurate surface. Specifically, we train a neural network to gradually infer\nthe relationship between queries and the approximated surface by searching for\nthe moving target of queries in a dynamic way. Meanwhile, we introduce a\npolygonization algorithm to extract surfaces using the gradients of the learned\nUDF. We conduct comprehensive experiments in surface reconstruction for point\nclouds, real scans or depth maps, and further explore our performance in\nunsupervised point normal estimation, which demonstrate non-trivial\nimprovements of CAP-UDF over the state-of-the-art methods.\n", "date": "2022-10-06", "categories": []}
{"id": "2402.10128", "title": "GES: Generalized Exponential Splatting for Efficient Radiance Field\n  Rendering", "abstract": "  Advancements in 3D Gaussian Splatting have significantly accelerated 3D\nreconstruction and generation. However, it may require a large number of\nGaussians, which creates a substantial memory footprint. This paper introduces\nGES (Generalized Exponential Splatting), a novel representation that employs\nGeneralized Exponential Function (GEF) to model 3D scenes, requiring far fewer\nparticles to represent a scene and thus significantly outperforming Gaussian\nSplatting methods in efficiency with a plug-and-play replacement ability for\nGaussian-based utilities. GES is validated theoretically and empirically in\nboth principled 1D setup and realistic 3D scenes.\n  It is shown to represent signals with sharp edges more accurately, which are\ntypically challenging for Gaussians due to their inherent low-pass\ncharacteristics. Our empirical analysis demonstrates that GEF outperforms\nGaussians in fitting natural-occurring signals (e.g. squares, triangles, and\nparabolic signals), thereby reducing the need for extensive splitting\noperations that increase the memory footprint of Gaussian Splatting. With the\naid of a frequency-modulated loss, GES achieves competitive performance in\nnovel-view synthesis benchmarks while requiring less than half the memory\nstorage of Gaussian Splatting and increasing the rendering speed by up to 39%.\nThe code is available on the project website https://abdullahamdi.com/ges .\n", "date": "2024-02-15", "categories": []}
{"id": "2404.12547", "title": "Evaluating Alternatives to SFM Point Cloud Initialization for Gaussian\n  Splatting", "abstract": "  3D Gaussian Splatting has recently been embraced as a versatile and effective\nmethod for scene reconstruction and novel view synthesis, owing to its\nhigh-quality results and compatibility with hardware rasterization. Despite its\nadvantages, Gaussian Splatting's reliance on high-quality point cloud\ninitialization by Structure-from-Motion (SFM) algorithms is a significant\nlimitation to be overcome. To this end, we investigate various initialization\nstrategies for Gaussian Splatting and delve into how volumetric reconstructions\nfrom Neural Radiance Fields (NeRF) can be utilized to bypass the dependency on\nSFM data. Our findings demonstrate that random initialization can perform much\nbetter if carefully designed and that by employing a combination of improved\ninitialization strategies and structure distillation from low-cost NeRF models,\nit is possible to achieve equivalent results, or at times even superior, to\nthose obtained from SFM initialization. Source code is available at\nhttps://theialab.github.io/nerf-3dgs .\n", "date": "2024-04-18", "categories": []}
{"id": "2405.15622", "title": "LAM3D: Large Image-Point-Cloud Alignment Model for 3D Reconstruction\n  from Single Image", "abstract": "  Large Reconstruction Models have made significant strides in the realm of\nautomated 3D content generation from single or multiple input images. Despite\ntheir success, these models often produce 3D meshes with geometric\ninaccuracies, stemming from the inherent challenges of deducing 3D shapes\nsolely from image data. In this work, we introduce a novel framework, the Large\nImage and Point Cloud Alignment Model (LAM3D), which utilizes 3D point cloud\ndata to enhance the fidelity of generated 3D meshes. Our methodology begins\nwith the development of a point-cloud-based network that effectively generates\nprecise and meaningful latent tri-planes, laying the groundwork for accurate 3D\nmesh reconstruction. Building upon this, our Image-Point-Cloud Feature\nAlignment technique processes a single input image, aligning to the latent\ntri-planes to imbue image features with robust 3D information. This process not\nonly enriches the image features but also facilitates the production of\nhigh-fidelity 3D meshes without the need for multi-view input, significantly\nreducing geometric distortions. Our approach achieves state-of-the-art\nhigh-fidelity 3D mesh reconstruction from a single image in just 6 seconds, and\nexperiments on various datasets demonstrate its effectiveness.\n", "date": "2024-05-24", "categories": []}
{"id": "2404.09412", "title": "DeferredGS: Decoupled and Editable Gaussian Splatting with Deferred\n  Shading", "abstract": "  Reconstructing and editing 3D objects and scenes both play crucial roles in\ncomputer graphics and computer vision. Neural radiance fields (NeRFs) can\nachieve realistic reconstruction and editing results but suffer from\ninefficiency in rendering. Gaussian splatting significantly accelerates\nrendering by rasterizing Gaussian ellipsoids. However, Gaussian splatting\nutilizes a single Spherical Harmonic (SH) function to model both texture and\nlighting, limiting independent editing capabilities of these components.\nRecently, attempts have been made to decouple texture and lighting with the\nGaussian splatting representation but may fail to produce plausible geometry\nand decomposition results on reflective scenes. Additionally, the forward\nshading technique they employ introduces noticeable blending artifacts during\nrelighting, as the geometry attributes of Gaussians are optimized under the\noriginal illumination and may not be suitable for novel lighting conditions. To\naddress these issues, we introduce DeferredGS, a method for decoupling and\nediting the Gaussian splatting representation using deferred shading. To\nachieve successful decoupling, we model the illumination with a learnable\nenvironment map and define additional attributes such as texture parameters and\nnormal direction on Gaussians, where the normal is distilled from a jointly\ntrained signed distance function. More importantly, we apply deferred shading,\nresulting in more realistic relighting effects compared to previous methods.\nBoth qualitative and quantitative experiments demonstrate the superior\nperformance of DeferredGS in novel view synthesis and editing tasks.\n", "date": "2024-04-15", "categories": []}
{"id": "2403.14836", "title": "Evaluating Panoramic 3D Estimation in Indoor Lighting Analysis", "abstract": "  This paper presents the use of panoramic 3D estimation in lighting\nsimulation. Conventional lighting simulation necessitates detailed modeling as\ninput, resulting in significant labor effort and time cost. The 3D layout\nestimation method directly takes a single panorama as input and generates a\nlighting simulation model with room geometry and window aperture. We evaluate\nthe simulation results by comparing the luminance errors between on-site High\nDynamic Range (HDR) photographs, 3D estimation model, and detailed model in\npanoramic representation and fisheye perspective. Given the selected scene, the\nresults demonstrate the estimated room layout is reliable for lighting\nsimulation.\n", "date": "2024-03-21", "categories": []}
{"id": "2405.12006", "title": "Depth Reconstruction with Neural Signed Distance Fields in Structured\n  Light Systems", "abstract": "  We introduce a novel depth estimation technique for multi-frame structured\nlight setups using neural implicit representations of 3D space. Our approach\nemploys a neural signed distance field (SDF), trained through self-supervised\ndifferentiable rendering. Unlike passive vision, where joint estimation of\nradiance and geometry fields is necessary, we capitalize on known radiance\nfields from projected patterns in structured light systems. This enables\nisolated optimization of the geometry field, ensuring convergence and network\nefficacy with fixed device positioning. To enhance geometric fidelity, we\nincorporate an additional color loss based on object surfaces during training.\nReal-world experiments demonstrate our method's superiority in geometric\nperformance for few-shot scenarios, while achieving comparable results with\nincreased pattern availability.\n", "date": "2024-05-20", "categories": []}
{"id": "2405.10314", "title": "CAT3D: Create Anything in 3D with Multi-View Diffusion Models", "abstract": "  Advances in 3D reconstruction have enabled high-quality 3D capture, but\nrequire a user to collect hundreds to thousands of images to create a 3D scene.\nWe present CAT3D, a method for creating anything in 3D by simulating this\nreal-world capture process with a multi-view diffusion model. Given any number\nof input images and a set of target novel viewpoints, our model generates\nhighly consistent novel views of a scene. These generated views can be used as\ninput to robust 3D reconstruction techniques to produce 3D representations that\ncan be rendered from any viewpoint in real-time. CAT3D can create entire 3D\nscenes in as little as one minute, and outperforms existing methods for single\nimage and few-view 3D scene creation. See our project page for results and\ninteractive demos at https://cat3d.github.io .\n", "date": "2024-05-16", "categories": []}
{"id": "2405.07178", "title": "Hologram: Realtime Holographic Overlays via LiDAR Augmented\n  Reconstruction", "abstract": "  Guided by the hologram technology of the infamous Star Wars franchise, I\npresent an application that creates real-time holographic overlays using LiDAR\naugmented 3D reconstruction. Prior attempts involve SLAM or NeRFs which either\nrequire highly calibrated scenes, incur steep computation costs, or fail to\nrender dynamic scenes. I propose 3 high-fidelity reconstruction tools that can\nrun on a portable device, such as a iPhone 14 Pro, which can allow for metric\naccurate facial reconstructions. My systems enable interactive and immersive\nholographic experiences that can be used for a wide range of applications,\nincluding augmented reality, telepresence, and entertainment.\n", "date": "2024-05-12", "categories": []}
{"id": "2404.19706", "title": "RTG-SLAM: Real-time 3D Reconstruction at Scale using Gaussian Splatting", "abstract": "  We present Real-time Gaussian SLAM (RTG-SLAM), a real-time 3D reconstruction\nsystem with an RGBD camera for large-scale environments using Gaussian\nsplatting. The system features a compact Gaussian representation and a highly\nefficient on-the-fly Gaussian optimization scheme. We force each Gaussian to be\neither opaque or nearly transparent, with the opaque ones fitting the surface\nand dominant colors, and transparent ones fitting residual colors. By rendering\ndepth in a different way from color rendering, we let a single opaque Gaussian\nwell fit a local surface region without the need of multiple overlapping\nGaussians, hence largely reducing the memory and computation cost. For\non-the-fly Gaussian optimization, we explicitly add Gaussians for three types\nof pixels per frame: newly observed, with large color errors, and with large\ndepth errors. We also categorize all Gaussians into stable and unstable ones,\nwhere the stable Gaussians are expected to well fit previously observed RGBD\nimages and otherwise unstable. We only optimize the unstable Gaussians and only\nrender the pixels occupied by unstable Gaussians. In this way, both the number\nof Gaussians to be optimized and pixels to be rendered are largely reduced, and\nthe optimization can be done in real time. We show real-time reconstructions of\na variety of large scenes. Compared with the state-of-the-art NeRF-based RGBD\nSLAM, our system achieves comparable high-quality reconstruction but with\naround twice the speed and half the memory cost, and shows superior performance\nin the realism of novel view synthesis and camera tracking accuracy.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.05446", "title": "GDGS: Gradient Domain Gaussian Splatting for Sparse Representation of\n  Radiance Fields", "abstract": "  The 3D Gaussian splatting methods are getting popular. However, they work\ndirectly on the signal, leading to a dense representation of the signal. Even\nwith some techniques such as pruning or distillation, the results are still\ndense. In this paper, we propose to model the gradient of the original signal.\nThe gradients are much sparser than the original signal. Therefore, the\ngradients use much less Gaussian splats, leading to the more efficient storage\nand thus higher computational performance during both training and rendering.\nThanks to the sparsity, during the view synthesis, only a small mount of pixels\nare needed, leading to much higher computational performance ($100\\sim\n1000\\times$ faster). And the 2D image can be recovered from the gradients via\nsolving a Poisson equation with linear computation complexity. Several\nexperiments are performed to confirm the sparseness of the gradients and the\ncomputation performance of the proposed method. The method can be applied\nvarious applications, such as human body modeling and indoor environment\nmodeling.\n", "date": "2024-05-08", "categories": []}
{"id": "2401.01970", "title": "FMGS: Foundation Model Embedded 3D Gaussian Splatting for Holistic 3D\n  Scene Understanding", "abstract": "  Precisely perceiving the geometric and semantic properties of real-world 3D\nobjects is crucial for the continued evolution of augmented reality and robotic\napplications. To this end, we present Foundation Model Embedded Gaussian\nSplatting (FMGS), which incorporates vision-language embeddings of foundation\nmodels into 3D Gaussian Splatting (GS). The key contribution of this work is an\nefficient method to reconstruct and represent 3D vision-language models. This\nis achieved by distilling feature maps generated from image-based foundation\nmodels into those rendered from our 3D model. To ensure high-quality rendering\nand fast training, we introduce a novel scene representation by integrating\nstrengths from both GS and multi-resolution hash encodings (MHE). Our effective\ntraining procedure also introduces a pixel alignment loss that makes the\nrendered feature distance of the same semantic entities close, following the\npixel-level semantic boundaries. Our results demonstrate remarkable multi-view\nsemantic consistency, facilitating diverse downstream tasks, beating\nstate-of-the-art methods by 10.2 percent on open-vocabulary language-based\nobject detection, despite that we are 851X faster for inference. This research\nexplores the intersection of vision, language, and 3D scene representation,\npaving the way for enhanced scene understanding in uncontrolled real-world\nenvironments. We plan to release the code on the project page.\n", "date": "2024-01-03", "categories": []}
{"id": "2403.15064", "title": "Recent Trends in 3D Reconstruction of General Non-Rigid Scenes", "abstract": "  Reconstructing models of the real world, including 3D geometry, appearance,\nand motion of real scenes, is essential for computer graphics and computer\nvision. It enables the synthesizing of photorealistic novel views, useful for\nthe movie industry and AR/VR applications. It also facilitates the content\ncreation necessary in computer games and AR/VR by avoiding laborious manual\ndesign processes. Further, such models are fundamental for intelligent\ncomputing systems that need to interpret real-world scenes and actions to act\nand interact safely with the human world. Notably, the world surrounding us is\ndynamic, and reconstructing models of dynamic, non-rigidly moving scenes is a\nseverely underconstrained and challenging problem. This state-of-the-art report\n(STAR) offers the reader a comprehensive summary of state-of-the-art techniques\nwith monocular and multi-view inputs such as data from RGB and RGB-D sensors,\namong others, conveying an understanding of different approaches, their\npotential applications, and promising further research directions. The report\ncovers 3D reconstruction of general non-rigid scenes and further addresses the\ntechniques for scene decomposition, editing and controlling, and generalizable\nand generative modeling. More specifically, we first review the common and\nfundamental concepts necessary to understand and navigate the field and then\ndiscuss the state-of-the-art techniques by reviewing recent approaches that use\ntraditional and machine-learning-based neural representations, including a\ndiscussion on the newly enabled applications. The STAR is concluded with a\ndiscussion of the remaining limitations and open challenges.\n", "date": "2024-03-22", "categories": []}
{"id": "2405.00900", "title": "LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes", "abstract": "  Photorealistic simulation plays a crucial role in applications such as\nautonomous driving, where advances in neural radiance fields (NeRFs) may allow\nbetter scalability through the automatic creation of digital 3D assets.\nHowever, reconstruction quality suffers on street scenes due to largely\ncollinear camera motions and sparser samplings at higher speeds. On the other\nhand, the application often demands rendering from camera views that deviate\nfrom the inputs to accurately simulate behaviors like lane changes. In this\npaper, we propose several insights that allow a better utilization of Lidar\ndata to improve NeRF quality on street scenes. First, our framework learns a\ngeometric scene representation from Lidar, which is fused with the implicit\ngrid-based representation for radiance decoding, thereby supplying stronger\ngeometric information offered by explicit point cloud. Second, we put forth a\nrobust occlusion-aware depth supervision scheme, which allows utilizing\ndensified Lidar points by accumulation. Third, we generate augmented training\nviews from Lidar points for further improvement. Our insights translate to\nlargely improved novel view synthesis under real driving scenes.\n", "date": "2024-05-01", "categories": []}
{"id": "2405.03388", "title": "3D LiDAR Mapping in Dynamic Environments Using a 4D Implicit Neural\n  Representation", "abstract": "  Building accurate maps is a key building block to enable reliable\nlocalization, planning, and navigation of autonomous vehicles. We propose a\nnovel approach for building accurate maps of dynamic environments utilizing a\nsequence of LiDAR scans. To this end, we propose encoding the 4D scene into a\nnovel spatio-temporal implicit neural map representation by fitting a\ntime-dependent truncated signed distance function to each point. Using our\nrepresentation, we extract the static map by filtering the dynamic parts. Our\nneural representation is based on sparse feature grids, a globally shared\ndecoder, and time-dependent basis functions, which we jointly optimize in an\nunsupervised fashion. To learn this representation from a sequence of LiDAR\nscans, we design a simple yet efficient loss function to supervise the map\noptimization in a piecewise way. We evaluate our approach on various scenes\ncontaining moving objects in terms of the reconstruction quality of static maps\nand the segmentation of dynamic point clouds. The experimental results\ndemonstrate that our method is capable of removing the dynamic part of the\ninput point clouds while reconstructing accurate and complete 3D maps,\noutperforming several state-of-the-art methods. Codes are available at:\nhttps://github.com/PRBonn/4dNDF\n", "date": "2024-05-06", "categories": []}
{"id": "2405.02005", "title": "HoloGS: Instant Depth-based 3D Gaussian Splatting with Microsoft\n  HoloLens 2", "abstract": "  In the fields of photogrammetry, computer vision and computer graphics, the\ntask of neural 3D scene reconstruction has led to the exploration of various\ntechniques. Among these, 3D Gaussian Splatting stands out for its explicit\nrepresentation of scenes using 3D Gaussians, making it appealing for tasks like\n3D point cloud extraction and surface reconstruction. Motivated by its\npotential, we address the domain of 3D scene reconstruction, aiming to leverage\nthe capabilities of the Microsoft HoloLens 2 for instant 3D Gaussian Splatting.\nWe present HoloGS, a novel workflow utilizing HoloLens sensor data, which\nbypasses the need for pre-processing steps like Structure from Motion by\ninstantly accessing the required input data i.e. the images, camera poses and\nthe point cloud from depth sensing. We provide comprehensive investigations,\nincluding the training process and the rendering quality, assessed through the\nPeak Signal-to-Noise Ratio, and the geometric 3D accuracy of the densified\npoint cloud from Gaussian centers, measured by Chamfer Distance. We evaluate\nour approach on two self-captured scenes: An outdoor scene of a cultural\nheritage statue and an indoor scene of a fine-structured plant. Our results\nshow that the HoloLens data, including RGB images, corresponding camera poses,\nand depth sensing based point clouds to initialize the Gaussians, are suitable\nas input for 3D Gaussian Splatting.\n", "date": "2024-05-03", "categories": []}
{"id": "2405.00340", "title": "NC-SDF: Enhancing Indoor Scene Reconstruction Using Neural SDFs with\n  View-Dependent Normal Compensation", "abstract": "  State-of-the-art neural implicit surface representations have achieved\nimpressive results in indoor scene reconstruction by incorporating monocular\ngeometric priors as additional supervision. However, we have observed that\nmulti-view inconsistency between such priors poses a challenge for high-quality\nreconstructions. In response, we present NC-SDF, a neural signed distance field\n(SDF) 3D reconstruction framework with view-dependent normal compensation (NC).\nSpecifically, we integrate view-dependent biases in monocular normal priors\ninto the neural implicit representation of the scene. By adaptively learning\nand correcting the biases, our NC-SDF effectively mitigates the adverse impact\nof inconsistent supervision, enhancing both the global consistency and local\ndetails in the reconstructions. To further refine the details, we introduce an\ninformative pixel sampling strategy to pay more attention to intricate geometry\nwith higher information content. Additionally, we design a hybrid geometry\nmodeling approach to improve the neural implicit representation. Experiments on\nsynthetic and real-world datasets demonstrate that NC-SDF outperforms existing\napproaches in terms of reconstruction quality.\n", "date": "2024-05-01", "categories": []}
{"id": "2405.00676", "title": "Spectrally Pruned Gaussian Fields with Neural Compensation", "abstract": "  Recently, 3D Gaussian Splatting, as a novel 3D representation, has garnered\nattention for its fast rendering speed and high rendering quality. However,\nthis comes with high memory consumption, e.g., a well-trained Gaussian field\nmay utilize three million Gaussian primitives and over 700 MB of memory. We\ncredit this high memory footprint to the lack of consideration for the\nrelationship between primitives. In this paper, we propose a memory-efficient\nGaussian field named SUNDAE with spectral pruning and neural compensation. On\none hand, we construct a graph on the set of Gaussian primitives to model their\nrelationship and design a spectral down-sampling module to prune out primitives\nwhile preserving desired signals. On the other hand, to compensate for the\nquality loss of pruning Gaussians, we exploit a lightweight neural network head\nto mix splatted features, which effectively compensates for quality losses\nwhile capturing the relationship between primitives in its weights. We\ndemonstrate the performance of SUNDAE with extensive results. For example,\nSUNDAE can achieve 26.80 PSNR at 145 FPS using 104 MB memory while the vanilla\nGaussian splatting algorithm achieves 25.60 PSNR at 160 FPS using 523 MB\nmemory, on the Mip-NeRF360 dataset. Codes are publicly available at\nhttps://runyiyang.github.io/projects/SUNDAE/.\n", "date": "2024-05-01", "categories": []}
{"id": "2404.17774", "title": "High-quality Surface Reconstruction using Gaussian Surfels", "abstract": "  We propose a novel point-based representation, Gaussian surfels, to combine\nthe advantages of the flexible optimization procedure in 3D Gaussian points and\nthe surface alignment property of surfels. This is achieved by directly setting\nthe z-scale of 3D Gaussian points to 0, effectively flattening the original 3D\nellipsoid into a 2D ellipse. Such a design provides clear guidance to the\noptimizer. By treating the local z-axis as the normal direction, it greatly\nimproves optimization stability and surface alignment. While the derivatives to\nthe local z-axis computed from the covariance matrix are zero in this setting,\nwe design a self-supervised normal-depth consistency loss to remedy this issue.\nMonocular normal priors and foreground masks are incorporated to enhance the\nquality of the reconstruction, mitigating issues related to highlights and\nbackground. We propose a volumetric cutting method to aggregate the information\nof Gaussian surfels so as to remove erroneous points in depth maps generated by\nalpha blending. Finally, we apply screened Poisson reconstruction method to the\nfused depth maps to extract the surface mesh. Experimental results show that\nour method demonstrates superior performance in surface reconstruction compared\nto state-of-the-art neural volume rendering and point-based rendering methods.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.19702", "title": "GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting", "abstract": "  We propose GS-LRM, a scalable large reconstruction model that can predict\nhigh-quality 3D Gaussian primitives from 2-4 posed sparse images in 0.23\nseconds on single A100 GPU. Our model features a very simple transformer-based\narchitecture; we patchify input posed images, pass the concatenated multi-view\nimage tokens through a sequence of transformer blocks, and decode final\nper-pixel Gaussian parameters directly from these tokens for differentiable\nrendering. In contrast to previous LRMs that can only reconstruct objects, by\npredicting per-pixel Gaussians, GS-LRM naturally handles scenes with large\nvariations in scale and complexity. We show that our model can work on both\nobject and scene captures by training it on Objaverse and RealEstate10K\nrespectively. In both scenarios, the models outperform state-of-the-art\nbaselines by a wide margin. We also demonstrate applications of our model in\ndownstream 3D generation tasks. Our project webpage is available at:\nhttps://sai-bi.github.io/project/gs-lrm/ .\n", "date": "2024-04-30", "categories": []}
{"id": "2404.19760", "title": "Lightplane: Highly-Scalable Components for Neural 3D Fields", "abstract": "  Contemporary 3D research, particularly in reconstruction and generation,\nheavily relies on 2D images for inputs or supervision. However, current designs\nfor these 2D-3D mapping are memory-intensive, posing a significant bottleneck\nfor existing methods and hindering new applications. In response, we propose a\npair of highly scalable components for 3D neural fields: Lightplane Render and\nSplatter, which significantly reduce memory usage in 2D-3D mapping. These\ninnovations enable the processing of vastly more and higher resolution images\nwith small memory and computational costs. We demonstrate their utility in\nvarious applications, from benefiting single-scene optimization with\nimage-level losses to realizing a versatile pipeline for dramatically scaling\n3D reconstruction and generation. Code:\n\\url{https://github.com/facebookresearch/lightplane}.\n", "date": "2024-04-30", "categories": []}
{"id": "2404.17974", "title": "HVOFusion: Incremental Mesh Reconstruction Using Hybrid Voxel Octree", "abstract": "  Incremental scene reconstruction is essential to the navigation in robotics.\nMost of the conventional methods typically make use of either TSDF (truncated\nsigned distance functions) volume or neural networks to implicitly represent\nthe surface. Due to the voxel representation or involving with time-consuming\nsampling, they have difficulty in balancing speed, memory storage, and surface\nquality. In this paper, we propose a novel hybrid voxel-octree approach to\neffectively fuse octree with voxel structures so that we can take advantage of\nboth implicit surface and explicit triangular mesh representation. Such sparse\nstructure preserves triangular faces in the leaf nodes and produces partial\nmeshes sequentially for incremental reconstruction. This storage scheme allows\nus to naturally optimize the mesh in explicit 3D space to achieve higher\nsurface quality. We iteratively deform the mesh towards the target and recovers\nvertex colors by optimizing a shading model. Experimental results on several\ndatasets show that our proposed approach is capable of quickly and accurately\nreconstructing a scene with realistic colors.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.18439", "title": "$\\nu$-DBA: Neural Implicit Dense Bundle Adjustment Enables Image-Only\n  Driving Scene Reconstruction", "abstract": "  The joint optimization of the sensor trajectory and 3D map is a crucial\ncharacteristic of bundle adjustment (BA), essential for autonomous driving.\nThis paper presents $\\nu$-DBA, a novel framework implementing geometric dense\nbundle adjustment (DBA) using 3D neural implicit surfaces for map\nparametrization, which optimizes both the map surface and trajectory poses\nusing geometric error guided by dense optical flow prediction. Additionally, we\nfine-tune the optical flow model with per-scene self-supervision to further\nimprove the quality of the dense mapping. Our experimental results on multiple\ndriving scene datasets demonstrate that our method achieves superior trajectory\noptimization and dense reconstruction accuracy. We also investigate the\ninfluences of photometric error and different neural geometric priors on the\nperformance of surface reconstruction and novel view synthesis. Our method\nstands as a significant step towards leveraging neural implicit representations\nin dense bundle adjustment for more accurate trajectories and detailed\nenvironmental mapping.\n", "date": "2024-04-29", "categories": []}
{"id": "2404.09105", "title": "EGGS: Edge Guided Gaussian Splatting for Radiance Fields", "abstract": "  The Gaussian splatting methods are getting popular. However, their loss\nfunction only contains the $\\ell_1$ norm and the structural similarity between\nthe rendered and input images, without considering the edges in these images.\nIt is well-known that the edges in an image provide important information.\nTherefore, in this paper, we propose an Edge Guided Gaussian Splatting (EGGS)\nmethod that leverages the edges in the input images. More specifically, we give\nthe edge region a higher weight than the flat region. With such edge guidance,\nthe resulting Gaussian particles focus more on the edges instead of the flat\nregions. Moreover, such edge guidance does not crease the computation cost\nduring the training and rendering stage. The experiments confirm that such\nsimple edge-weighted loss function indeed improves about $1\\sim2$ dB on several\ndifference data sets. With simply plugging in the edge guidance, the proposed\nmethod can improve all Gaussian splatting methods in different scenarios, such\nas human head modeling, building 3D reconstruction, etc.\n", "date": "2024-04-14", "categories": []}
{"id": "2404.14410", "title": "Guess The Unseen: Dynamic 3D Scene Reconstruction from Partial 2D\n  Glimpses", "abstract": "  In this paper, we present a method to reconstruct the world and multiple\ndynamic humans in 3D from a monocular video input. As a key idea, we represent\nboth the world and multiple humans via the recently emerging 3D Gaussian\nSplatting (3D-GS) representation, enabling to conveniently and efficiently\ncompose and render them together. In particular, we address the scenarios with\nseverely limited and sparse observations in 3D human reconstruction, a common\nchallenge encountered in the real world. To tackle this challenge, we introduce\na novel approach to optimize the 3D-GS representation in a canonical space by\nfusing the sparse cues in the common space, where we leverage a pre-trained 2D\ndiffusion model to synthesize unseen views while keeping the consistency with\nthe observed 2D appearances. We demonstrate our method can reconstruct\nhigh-quality animatable 3D humans in various challenging examples, in the\npresence of occlusion, image crops, few-shot, and extremely sparse\nobservations. After reconstruction, our method is capable of not only rendering\nthe scene in any novel views at arbitrary time instances, but also editing the\n3D scene by removing individual humans or applying different motions for each\nhuman. Through various experiments, we demonstrate the quality and efficiency\nof our methods over alternative existing approaches.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.12777", "title": "EfficientGS: Streamlining Gaussian Splatting for Large-Scale\n  High-Resolution Scene Representation", "abstract": "  In the domain of 3D scene representation, 3D Gaussian Splatting (3DGS) has\nemerged as a pivotal technology. However, its application to large-scale,\nhigh-resolution scenes (exceeding 4k$\\times$4k pixels) is hindered by the\nexcessive computational requirements for managing a large number of Gaussians.\nAddressing this, we introduce 'EfficientGS', an advanced approach that\noptimizes 3DGS for high-resolution, large-scale scenes. We analyze the\ndensification process in 3DGS and identify areas of Gaussian\nover-proliferation. We propose a selective strategy, limiting Gaussian increase\nto key primitives, thereby enhancing the representational efficiency.\nAdditionally, we develop a pruning mechanism to remove redundant Gaussians,\nthose that are merely auxiliary to adjacent ones. For further enhancement, we\nintegrate a sparse order increment for Spherical Harmonics (SH), designed to\nalleviate storage constraints and reduce training overhead. Our empirical\nevaluations, conducted on a range of datasets including extensive 4K+ aerial\nimages, demonstrate that 'EfficientGS' not only expedites training and\nrendering times but also achieves this with a model size approximately tenfold\nsmaller than conventional 3DGS while maintaining high rendering fidelity.\n", "date": "2024-04-19", "categories": []}
{"id": "2311.17116", "title": "REF$^2$-NeRF: Reflection and Refraction aware Neural Radiance Field", "abstract": "  Recently, significant progress has been made in the study of methods for 3D\nreconstruction from multiple images using implicit neural representations,\nexemplified by the neural radiance field (NeRF) method. Such methods, which are\nbased on volume rendering, can model various light phenomena, and various\nextended methods have been proposed to accommodate different scenes and\nsituations. However, when handling scenes with multiple glass objects, e.g.,\nobjects in a glass showcase, modeling the target scene accurately has been\nchallenging due to the presence of multiple reflection and refraction effects.\nThus, this paper proposes a NeRF-based modeling method for scenes containing a\nglass case. In the proposed method, refraction and reflection are modeled using\nelements that are dependent and independent of the viewer's perspective. This\napproach allows us to estimate the surfaces where refraction occurs, i.e.,\nglass surfaces, and enables the separation and modeling of both direct and\nreflected light components. The proposed method requires predetermined camera\nposes, but accurately estimating these poses in scenes with glass objects is\ndifficult. Therefore, we used a robotic arm with an attached camera to acquire\nimages with known poses. Compared to existing methods, the proposed method\nenables more accurate modeling of both glass refraction and the overall scene.\n", "date": "2023-11-28", "categories": []}
{"id": "2312.05889", "title": "SuperPrimitive: Scene Reconstruction at a Primitive Level", "abstract": "  Joint camera pose and dense geometry estimation from a set of images or a\nmonocular video remains a challenging problem due to its computational\ncomplexity and inherent visual ambiguities. Most dense incremental\nreconstruction systems operate directly on image pixels and solve for their 3D\npositions using multi-view geometry cues. Such pixel-level approaches suffer\nfrom ambiguities or violations of multi-view consistency (e.g. caused by\ntextureless or specular surfaces).\n  We address this issue with a new image representation which we call a\nSuperPrimitive. SuperPrimitives are obtained by splitting images into\nsemantically correlated local regions and enhancing them with estimated surface\nnormal directions, both of which are predicted by state-of-the-art single image\nneural networks. This provides a local geometry estimate per SuperPrimitive,\nwhile their relative positions are adjusted based on multi-view observations.\n  We demonstrate the versatility of our new representation by addressing three\n3D reconstruction tasks: depth completion, few-view structure from motion, and\nmonocular dense visual odometry.\n", "date": "2023-12-10", "categories": []}
{"id": "2404.02155", "title": "Alpha Invariance: On Inverse Scaling Between Distance and Volume Density\n  in Neural Radiance Fields", "abstract": "  Scale-ambiguity in 3D scene dimensions leads to magnitude-ambiguity of\nvolumetric densities in neural radiance fields, i.e., the densities double when\nscene size is halved, and vice versa. We call this property alpha invariance.\nFor NeRFs to better maintain alpha invariance, we recommend 1) parameterizing\nboth distance and volume densities in log space, and 2) a\ndiscretization-agnostic initialization strategy to guarantee high ray\ntransmittance. We revisit a few popular radiance field models and find that\nthese systems use various heuristics to deal with issues arising from scene\nscaling. We test their behaviors and show our recipe to be more robust.\n", "date": "2024-04-02", "categories": []}
{"id": "2303.15368", "title": "2S-UDF: A Novel Two-stage UDF Learning Method for Robust Non-watertight\n  Model Reconstruction from Multi-view Images", "abstract": "  Recently, building on the foundation of neural radiance field, various\ntechniques have emerged to learn unsigned distance fields (UDF) to reconstruct\n3D non-watertight models from multi-view images. Yet, a central challenge in\nUDF-based volume rendering is formulating a proper way to convert unsigned\ndistance values into volume density, ensuring that the resulting weight\nfunction remains unbiased and sensitive to occlusions. Falling short on these\nrequirements often results in incorrect topology or large reconstruction errors\nin resulting models. This paper addresses this challenge by presenting a novel\ntwo-stage algorithm, 2S-UDF, for learning a high-quality UDF from multi-view\nimages. Initially, the method applies an easily trainable density function\nthat, while slightly biased and transparent, aids in coarse reconstruction. The\nsubsequent stage then refines the geometry and appearance of the object to\nachieve a high-quality reconstruction by directly adjusting the weight function\nused in volume rendering to ensure that it is unbiased and occlusion-aware.\nDecoupling density and weight in two stages makes our training stable and\nrobust, distinguishing our technique from existing UDF learning approaches.\nEvaluations on the DeepFashion3D, DTU, and BlendedMVS datasets validate the\nrobustness and effectiveness of our proposed approach. In both quantitative\nmetrics and visual quality, the results indicate our superior performance over\nother UDF learning techniques in reconstructing 3D non-watertight models from\nmulti-view images. Our code is available at\nhttps://bitbucket.org/jkdeng/2sudf/.\n", "date": "2023-03-27", "categories": []}
{"id": "2312.13150", "title": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "abstract": "  We introduce the \\method, an ultra-efficient approach for monocular 3D object\nreconstruction. Splatter Image is based on Gaussian Splatting, which allows\nfast and high-quality reconstruction of 3D scenes from multiple images. We\napply Gaussian Splatting to monocular reconstruction by learning a neural\nnetwork that, at test time, performs reconstruction in a feed-forward manner,\nat 38 FPS. Our main innovation is the surprisingly straightforward design of\nthis network, which, using 2D operators, maps the input image to one 3D\nGaussian per pixel. The resulting set of Gaussians thus has the form an image,\nthe Splatter Image. We further extend the method take several images as input\nvia cross-view attention. Owning to the speed of the renderer (588 FPS), we use\na single GPU for training while generating entire images at each iteration to\noptimize perceptual metrics like LPIPS. On several synthetic, real,\nmulti-category and large-scale benchmark datasets, we achieve better results in\nterms of PSNR, LPIPS, and other metrics while training and evaluating much\nfaster than prior works. Code, models, demo and more results are available at\nhttps://szymanowiczs.github.io/splatter-image.\n", "date": "2023-12-20", "categories": []}
{"id": "2404.10218", "title": "Autonomous Implicit Indoor Scene Reconstruction with Frontier\n  Exploration", "abstract": "  Implicit neural representations have demonstrated significant promise for 3D\nscene reconstruction. Recent works have extended their applications to\nautonomous implicit reconstruction through the Next Best View (NBV) based\nmethod. However, the NBV method cannot guarantee complete scene coverage and\noften necessitates extensive viewpoint sampling, particularly in complex\nscenes. In the paper, we propose to 1) incorporate frontier-based exploration\ntasks for global coverage with implicit surface uncertainty-based\nreconstruction tasks to achieve high-quality reconstruction. and 2) introduce a\nmethod to achieve implicit surface uncertainty using color uncertainty, which\nreduces the time needed for view selection. Further with these two tasks, we\npropose an adaptive strategy for switching modes in view path planning, to\nreduce time and maintain superior reconstruction quality. Our method exhibits\nthe highest reconstruction quality among all planning methods and superior\nplanning efficiency in methods involving reconstruction tasks. We deploy our\nmethod on a UAV and the results show that our method can plan multi-task views\nand reconstruct a scene with high quality.\n", "date": "2024-04-16", "categories": []}
{"id": "2306.04166", "title": "BAA-NGP: Bundle-Adjusting Accelerated Neural Graphics Primitives", "abstract": "  Implicit neural representations have become pivotal in robotic perception,\nenabling robots to comprehend 3D environments from 2D images. Given a set of\ncamera poses and associated images, the models can be trained to synthesize\nnovel, unseen views. To successfully navigate and interact in dynamic settings,\nrobots require the understanding of their spatial surroundings driven by\nunassisted reconstruction of 3D scenes and camera poses from real-time video\nfootage. Existing approaches like COLMAP and bundle-adjusting neural radiance\nfield methods take hours to days to process due to the high computational\ndemands of feature matching, dense point sampling, and training of a\nmulti-layer perceptron structure with a large number of parameters. To address\nthese challenges, we propose a framework called bundle-adjusting accelerated\nneural graphics primitives (BAA-NGP) which leverages accelerated sampling and\nhash encoding to expedite automatic pose refinement/estimation and 3D scene\nreconstruction. Experimental results demonstrate 10 to 20 x speed improvement\ncompared to other bundle-adjusting neural radiance field methods without\nsacrificing the quality of pose estimation. The github repository can be found\nhere https://github.com/IntelLabs/baa-ngp.\n", "date": "2023-06-07", "categories": []}
{"id": "2311.12796", "title": "Physics-guided Shape-from-Template: Monocular Video Perception through\n  Neural Surrogate Models", "abstract": "  3D reconstruction of dynamic scenes is a long-standing problem in computer\ngraphics and increasingly difficult the less information is available.\nShape-from-Template (SfT) methods aim to reconstruct a template-based geometry\nfrom RGB images or video sequences, often leveraging just a single monocular\ncamera without depth information, such as regular smartphone recordings.\nUnfortunately, existing reconstruction methods are either unphysical and noisy\nor slow in optimization. To solve this problem, we propose a novel SfT\nreconstruction algorithm for cloth using a pre-trained neural surrogate model\nthat is fast to evaluate, stable, and produces smooth reconstructions due to a\nregularizing physics simulation. Differentiable rendering of the simulated mesh\nenables pixel-wise comparisons between the reconstruction and a target video\nsequence that can be used for a gradient-based optimization procedure to\nextract not only shape information but also physical parameters such as\nstretching, shearing, or bending stiffness of the cloth. This allows to retain\na precise, stable, and smooth reconstructed geometry while reducing the runtime\nby a factor of 400-500 compared to $\\phi$-SfT, a state-of-the-art physics-based\nSfT approach.\n", "date": "2023-11-21", "categories": []}
{"id": "2404.04880", "title": "GauU-Scene V2: Assessing the Reliability of Image-Based Metrics with\n  Expansive Lidar Image Dataset Using 3DGS and NeRF", "abstract": "  We introduce a novel, multimodal large-scale scene reconstruction benchmark\nthat utilizes newly developed 3D representation approaches: Gaussian Splatting\nand Neural Radiance Fields (NeRF). Our expansive U-Scene dataset surpasses any\npreviously existing real large-scale outdoor LiDAR and image dataset in both\narea and point count. GauU-Scene encompasses over 6.5 square kilometers and\nfeatures a comprehensive RGB dataset coupled with LiDAR ground truth.\nAdditionally, we are the first to propose a LiDAR and image alignment method\nfor a drone-based dataset. Our assessment of GauU-Scene includes a detailed\nanalysis across various novel viewpoints, employing image-based metrics such as\nSSIM, LPIPS, and PSNR on NeRF and Gaussian Splatting based methods. This\nanalysis reveals contradictory results when applying geometric-based metrics\nlike Chamfer distance. The experimental results on our multimodal dataset\nhighlight the unreliability of current image-based metrics and reveal\nsignificant drawbacks in geometric reconstruction using the current Gaussian\nSplatting-based method, further illustrating the necessity of our dataset for\nassessing geometry reconstruction tasks. We also provide detailed supplementary\ninformation on data collection protocols and make the dataset available on the\nfollowing anonymous project page\n", "date": "2024-04-07", "categories": []}
{"id": "2404.09458", "title": "CompGS: Efficient 3D Scene Representation via Compressed Gaussian\n  Splatting", "abstract": "  Gaussian splatting, renowned for its exceptional rendering quality and\nefficiency, has emerged as a prominent technique in 3D scene representation.\nHowever, the substantial data volume of Gaussian splatting impedes its\npractical utility in real-world applications. Herein, we propose an efficient\n3D scene representation, named Compressed Gaussian Splatting (CompGS), which\nharnesses compact Gaussian primitives for faithful 3D scene modeling with a\nremarkably reduced data size. To ensure the compactness of Gaussian primitives,\nwe devise a hybrid primitive structure that captures predictive relationships\nbetween each other. Then, we exploit a small set of anchor primitives for\nprediction, allowing the majority of primitives to be encapsulated into highly\ncompact residual forms. Moreover, we develop a rate-constrained optimization\nscheme to eliminate redundancies within such hybrid primitives, steering our\nCompGS towards an optimal trade-off between bitrate consumption and\nrepresentation efficacy. Experimental results show that the proposed CompGS\nsignificantly outperforms existing methods, achieving superior compactness in\n3D scene representation without compromising model accuracy and rendering\nquality. Our code will be released on GitHub for further research.\n", "date": "2024-04-15", "categories": []}
{"id": "2312.14937", "title": "SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes", "abstract": "  Novel view synthesis for dynamic scenes is still a challenging problem in\ncomputer vision and graphics. Recently, Gaussian splatting has emerged as a\nrobust technique to represent static scenes and enable high-quality and\nreal-time novel view synthesis. Building upon this technique, we propose a new\nrepresentation that explicitly decomposes the motion and appearance of dynamic\nscenes into sparse control points and dense Gaussians, respectively. Our key\nidea is to use sparse control points, significantly fewer in number than the\nGaussians, to learn compact 6 DoF transformation bases, which can be locally\ninterpolated through learned interpolation weights to yield the motion field of\n3D Gaussians. We employ a deformation MLP to predict time-varying 6 DoF\ntransformations for each control point, which reduces learning complexities,\nenhances learning abilities, and facilitates obtaining temporal and spatial\ncoherent motion patterns. Then, we jointly learn the 3D Gaussians, the\ncanonical space locations of control points, and the deformation MLP to\nreconstruct the appearance, geometry, and dynamics of 3D scenes. During\nlearning, the location and number of control points are adaptively adjusted to\naccommodate varying motion complexities in different regions, and an ARAP loss\nfollowing the principle of as rigid as possible is developed to enforce spatial\ncontinuity and local rigidity of learned motions. Finally, thanks to the\nexplicit sparse motion representation and its decomposition from appearance,\nour method can enable user-controlled motion editing while retaining\nhigh-fidelity appearances. Extensive experiments demonstrate that our approach\noutperforms existing approaches on novel view synthesis with a high rendering\nspeed and enables novel appearance-preserved motion editing applications.\nProject page: https://yihua7.github.io/SC-GS-web/\n", "date": "2023-12-04", "categories": []}
{"id": "2404.07933", "title": "Boosting Self-Supervision for Single-View Scene Completion via Knowledge\n  Distillation", "abstract": "  Inferring scene geometry from images via Structure from Motion is a\nlong-standing and fundamental problem in computer vision. While classical\napproaches and, more recently, depth map predictions only focus on the visible\nparts of a scene, the task of scene completion aims to reason about geometry\neven in occluded regions. With the popularity of neural radiance fields\n(NeRFs), implicit representations also became popular for scene completion by\npredicting so-called density fields. Unlike explicit approaches. e.g.\nvoxel-based methods, density fields also allow for accurate depth prediction\nand novel-view synthesis via image-based rendering. In this work, we propose to\nfuse the scene reconstruction from multiple images and distill this knowledge\ninto a more accurate single-view scene reconstruction. To this end, we propose\nMulti-View Behind the Scenes (MVBTS) to fuse density fields from multiple posed\nimages, trained fully self-supervised only from image data. Using knowledge\ndistillation, we use MVBTS to train a single-view scene completion network via\ndirect supervision called KDBTS. It achieves state-of-the-art performance on\noccupancy prediction, especially in occluded regions.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.06753", "title": "MonoSelfRecon: Purely Self-Supervised Explicit Generalizable 3D\n  Reconstruction of Indoor Scenes from Monocular RGB Views", "abstract": "  Current monocular 3D scene reconstruction (3DR) works are either\nfully-supervised, or not generalizable, or implicit in 3D representation. We\npropose a novel framework - MonoSelfRecon that for the first time achieves\nexplicit 3D mesh reconstruction for generalizable indoor scenes with monocular\nRGB views by purely self-supervision on voxel-SDF (signed distance function).\nMonoSelfRecon follows an Autoencoder-based architecture, decodes voxel-SDF and\na generalizable Neural Radiance Field (NeRF), which is used to guide voxel-SDF\nin self-supervision. We propose novel self-supervised losses, which not only\nsupport pure self-supervision, but can be used together with supervised signals\nto further boost supervised training. Our experiments show that \"MonoSelfRecon\"\ntrained in pure self-supervision outperforms current best self-supervised\nindoor depth estimation models and is comparable to 3DR models trained in fully\nsupervision with depth annotations. MonoSelfRecon is not restricted by specific\nmodel design, which can be used to any models with voxel-SDF for purely\nself-supervised manner.\n", "date": "2024-04-10", "categories": []}
{"id": "2403.14085", "title": "Surface Reconstruction from Point Clouds via Grid-based Intersection\n  Prediction", "abstract": "  Surface reconstruction from point clouds is a crucial task in the fields of\ncomputer vision and computer graphics. SDF-based methods excel at\nreconstructing smooth meshes with minimal error and artefacts but struggle with\nrepresenting open surfaces. On the other hand, UDF-based methods can\neffectively represent open surfaces but often introduce noise, leading to\nartefacts in the mesh. In this work, we propose a novel approach that directly\npredicts the intersection points between line segment of point pairs and\nimplicit surfaces. To achieve it, we propose two modules named Relative\nIntersection Module and Sign Module respectively with the feature of point pair\nas input. To preserve the continuity of the surface, we also integrate symmetry\ninto the two modules, which means the position of predicted intersection will\nnot change even if the input order of the point pair changes. This method not\nonly preserves the ability to represent open surfaces but also eliminates most\nartefacts on the mesh. Our approach demonstrates state-of-the-art performance\non three datasets: ShapeNet, MGN, and ScanNet. The code will be made available\nupon acceptance.\n", "date": "2024-03-21", "categories": []}
{"id": "2307.09555", "title": "Transient Neural Radiance Fields for Lidar View Synthesis and 3D\n  Reconstruction", "abstract": "  Neural radiance fields (NeRFs) have become a ubiquitous tool for modeling\nscene appearance and geometry from multiview imagery. Recent work has also\nbegun to explore how to use additional supervision from lidar or depth sensor\nmeasurements in the NeRF framework. However, previous lidar-supervised NeRFs\nfocus on rendering conventional camera imagery and use lidar-derived point\ncloud data as auxiliary supervision; thus, they fail to incorporate the\nunderlying image formation model of the lidar. Here, we propose a novel method\nfor rendering transient NeRFs that take as input the raw, time-resolved photon\ncount histograms measured by a single-photon lidar system, and we seek to\nrender such histograms from novel views. Different from conventional NeRFs, the\napproach relies on a time-resolved version of the volume rendering equation to\nrender the lidar measurements and capture transient light transport phenomena\nat picosecond timescales. We evaluate our method on a first-of-its-kind dataset\nof simulated and captured transient multiview scans from a prototype\nsingle-photon lidar. Overall, our work brings NeRFs to a new dimension of\nimaging at transient timescales, newly enabling rendering of transient imagery\nfrom novel views. Additionally, we show that our approach recovers improved\ngeometry and conventional appearance compared to point cloud-based supervision\nwhen training on few input viewpoints. Transient NeRFs may be especially useful\nfor applications which seek to simulate raw lidar measurements for downstream\ntasks in autonomous driving, robotics, and remote sensing.\n", "date": "2023-07-14", "categories": []}
{"id": "2311.11825", "title": "Holistic Inverse Rendering of Complex Facade via Aerial 3D Scanning", "abstract": "  In this work, we use multi-view aerial images to reconstruct the geometry,\nlighting, and material of facades using neural signed distance fields (SDFs).\nWithout the requirement of complex equipment, our method only takes simple RGB\nimages captured by a drone as inputs to enable physically based and\nphotorealistic novel-view rendering, relighting, and editing. However, a\nreal-world facade usually has complex appearances ranging from diffuse rocks\nwith subtle details to large-area glass windows with specular reflections,\nmaking it hard to attend to everything. As a result, previous methods can\npreserve the geometry details but fail to reconstruct smooth glass windows or\nverse vise. In order to address this challenge, we introduce three spatial- and\nsemantic-adaptive optimization strategies, including a semantic regularization\napproach based on zero-shot segmentation techniques to improve material\nconsistency, a frequency-aware geometry regularization to balance surface\nsmoothness and details in different surfaces, and a visibility probe-based\nscheme to enable efficient modeling of the local lighting in large-scale\noutdoor environments. In addition, we capture a real-world facade aerial 3D\nscanning image set and corresponding point clouds for training and\nbenchmarking. The experiment demonstrates the superior quality of our method on\nfacade holistic inverse rendering, novel view synthesis, and scene editing\ncompared to state-of-the-art baselines.\n", "date": "2023-11-20", "categories": []}
{"id": "2312.03203", "title": "Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled\n  Feature Fields", "abstract": "  3D scene representations have gained immense popularity in recent years.\nMethods that use Neural Radiance fields are versatile for traditional tasks\nsuch as novel view synthesis. In recent times, some work has emerged that aims\nto extend the functionality of NeRF beyond view synthesis, for semantically\naware tasks such as editing and segmentation using 3D feature field\ndistillation from 2D foundation models. However, these methods have two major\nlimitations: (a) they are limited by the rendering speed of NeRF pipelines, and\n(b) implicitly represented feature fields suffer from continuity artifacts\nreducing feature quality. Recently, 3D Gaussian Splatting has shown\nstate-of-the-art performance on real-time radiance field rendering. In this\nwork, we go one step further: in addition to radiance field rendering, we\nenable 3D Gaussian splatting on arbitrary-dimension semantic features via 2D\nfoundation model distillation. This translation is not straightforward: naively\nincorporating feature fields in the 3DGS framework encounters significant\nchallenges, notably the disparities in spatial resolution and channel\nconsistency between RGB images and feature maps. We propose architectural and\ntraining changes to efficiently avert this problem. Our proposed method is\ngeneral, and our experiments showcase novel view semantic segmentation,\nlanguage-guided editing and segment anything through learning feature fields\nfrom state-of-the-art 2D foundation models such as SAM and CLIP-LSeg. Across\nexperiments, our distillation method is able to provide comparable or better\nresults, while being significantly faster to both train and render.\nAdditionally, to the best of our knowledge, we are the first method to enable\npoint and bounding-box prompting for radiance field manipulation, by leveraging\nthe SAM model. Project website at: https://feature-3dgs.github.io/\n", "date": "2023-12-06", "categories": []}
{"id": "2404.04823", "title": "3D Building Reconstruction from Monocular Remote Sensing Images with\n  Multi-level Supervisions", "abstract": "  3D building reconstruction from monocular remote sensing images is an\nimportant and challenging research problem that has received increasing\nattention in recent years, owing to its low cost of data acquisition and\navailability for large-scale applications. However, existing methods rely on\nexpensive 3D-annotated samples for fully-supervised training, restricting their\napplication to large-scale cross-city scenarios. In this work, we propose\nMLS-BRN, a multi-level supervised building reconstruction network that can\nflexibly utilize training samples with different annotation levels to achieve\nbetter reconstruction results in an end-to-end manner. To alleviate the demand\non full 3D supervision, we design two new modules, Pseudo Building Bbox\nCalculator and Roof-Offset guided Footprint Extractor, as well as new tasks and\ntraining strategies for different types of samples. Experimental results on\nseveral public and new datasets demonstrate that our proposed MLS-BRN achieves\ncompetitive performance using much fewer 3D-annotated samples, and\nsignificantly improves the footprint extraction and 3D reconstruction\nperformance compared with current state-of-the-art. The code and datasets of\nthis work will be released at https://github.com/opendatalab/MLS-BRN.git.\n", "date": "2024-04-07", "categories": []}
{"id": "2404.04875", "title": "NeRF2Points: Large-Scale Point Cloud Generation From Street Views'\n  Radiance Field Optimization", "abstract": "  Neural Radiance Fields (NeRF) have emerged as a paradigm-shifting methodology\nfor the photorealistic rendering of objects and environments, enabling the\nsynthesis of novel viewpoints with remarkable fidelity. This is accomplished\nthrough the strategic utilization of object-centric camera poses characterized\nby significant inter-frame overlap. This paper explores a compelling,\nalternative utility of NeRF: the derivation of point clouds from aggregated\nurban landscape imagery. The transmutation of street-view data into point\nclouds is fraught with complexities, attributable to a nexus of interdependent\nvariables. First, high-quality point cloud generation hinges on precise camera\nposes, yet many datasets suffer from inaccuracies in pose metadata. Also, the\nstandard approach of NeRF is ill-suited for the distinct characteristics of\nstreet-view data from autonomous vehicles in vast, open settings. Autonomous\nvehicle cameras often record with limited overlap, leading to blurring,\nartifacts, and compromised pavement representation in NeRF-based point clouds.\nIn this paper, we present NeRF2Points, a tailored NeRF variant for urban point\ncloud synthesis, notable for its high-quality output from RGB inputs alone. Our\npaper is supported by a bespoke, high-resolution 20-kilometer urban street\ndataset, designed for point cloud generation and evaluation. NeRF2Points\nadeptly navigates the inherent challenges of NeRF-based point cloud synthesis\nthrough the implementation of the following strategic innovations: (1)\nIntegration of Weighted Iterative Geometric Optimization (WIGO) and Structure\nfrom Motion (SfM) for enhanced camera pose accuracy, elevating street-view data\nprecision. (2) Layered Perception and Integrated Modeling (LPiM) is designed\nfor distinct radiance field modeling in urban environments, resulting in\ncoherent point cloud representations.\n", "date": "2024-04-07", "categories": []}
{"id": "2404.05163", "title": "Semantic Flow: Learning Semantic Field of Dynamic Scenes from Monocular\n  Videos", "abstract": "  In this work, we pioneer Semantic Flow, a neural semantic representation of\ndynamic scenes from monocular videos. In contrast to previous NeRF methods that\nreconstruct dynamic scenes from the colors and volume densities of individual\npoints, Semantic Flow learns semantics from continuous flows that contain rich\n3D motion information. As there is 2D-to-3D ambiguity problem in the viewing\ndirection when extracting 3D flow features from 2D video frames, we consider\nthe volume densities as opacity priors that describe the contributions of flow\nfeatures to the semantics on the frames. More specifically, we first learn a\nflow network to predict flows in the dynamic scene, and propose a flow feature\naggregation module to extract flow features from video frames. Then, we propose\na flow attention module to extract motion information from flow features, which\nis followed by a semantic network to output semantic logits of flows. We\nintegrate the logits with volume densities in the viewing direction to\nsupervise the flow features with semantic labels on video frames. Experimental\nresults show that our model is able to learn from multiple dynamic scenes and\nsupports a series of new tasks such as instance-level scene editing, semantic\ncompletions, dynamic scene tracking and semantic adaption on novel scenes.\nCodes are available at https://github.com/tianfr/Semantic-Flow/.\n", "date": "2024-04-08", "categories": []}
{"id": "2308.04868", "title": "InstantAvatar: Efficient 3D Head Reconstruction via Surface Rendering", "abstract": "  Recent advances in full-head reconstruction have been obtained by optimizing\na neural field through differentiable surface or volume rendering to represent\na single scene. While these techniques achieve an unprecedented accuracy, they\ntake several minutes, or even hours, due to the expensive optimization process\nrequired. In this work, we introduce InstantAvatar, a method that recovers\nfull-head avatars from few images (down to just one) in a few seconds on\ncommodity hardware. In order to speed up the reconstruction process, we propose\na system that combines, for the first time, a voxel-grid neural field\nrepresentation with a surface renderer. Notably, a naive combination of these\ntwo techniques leads to unstable optimizations that do not converge to valid\nsolutions. In order to overcome this limitation, we present a novel statistical\nmodel that learns a prior distribution over 3D head signed distance functions\nusing a voxel-grid based architecture. The use of this prior model, in\ncombination with other design choices, results into a system that achieves 3D\nhead reconstructions with comparable accuracy as the state-of-the-art with a\n100x speed-up.\n", "date": "2023-08-09", "categories": []}
{"id": "2312.12337", "title": "pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable\n  Generalizable 3D Reconstruction", "abstract": "  We introduce pixelSplat, a feed-forward model that learns to reconstruct 3D\nradiance fields parameterized by 3D Gaussian primitives from pairs of images.\nOur model features real-time and memory-efficient rendering for scalable\ntraining as well as fast 3D reconstruction at inference time. To overcome local\nminima inherent to sparse and locally supported representations, we predict a\ndense probability distribution over 3D and sample Gaussian means from that\nprobability distribution. We make this sampling operation differentiable via a\nreparameterization trick, allowing us to back-propagate gradients through the\nGaussian splatting representation. We benchmark our method on wide-baseline\nnovel view synthesis on the real-world RealEstate10k and ACID datasets, where\nwe outperform state-of-the-art light field transformers and accelerate\nrendering by 2.5 orders of magnitude while reconstructing an interpretable and\neditable 3D radiance field.\n", "date": "2023-12-19", "categories": []}
{"id": "2312.14239", "title": "PlatoNeRF: 3D Reconstruction in Plato's Cave via Single-View Two-Bounce\n  Lidar", "abstract": "  3D reconstruction from a single-view is challenging because of the ambiguity\nfrom monocular cues and lack of information about occluded regions. Neural\nradiance fields (NeRF), while popular for view synthesis and 3D reconstruction,\nare typically reliant on multi-view images. Existing methods for single-view 3D\nreconstruction with NeRF rely on either data priors to hallucinate views of\noccluded regions, which may not be physically accurate, or shadows observed by\nRGB cameras, which are difficult to detect in ambient light and low albedo\nbackgrounds. We propose using time-of-flight data captured by a single-photon\navalanche diode to overcome these limitations. Our method models two-bounce\noptical paths with NeRF, using lidar transient data for supervision. By\nleveraging the advantages of both NeRF and two-bounce light measured by lidar,\nwe demonstrate that we can reconstruct visible and occluded geometry without\ndata priors or reliance on controlled ambient lighting or scene albedo. In\naddition, we demonstrate improved generalization under practical constraints on\nsensor spatial- and temporal-resolution. We believe our method is a promising\ndirection as single-photon lidars become ubiquitous on consumer devices, such\nas phones, tablets, and headsets.\n", "date": "2023-12-21", "categories": []}
{"id": "2404.03925", "title": "LightOctree: Lightweight 3D Spatially-Coherent Indoor Lighting\n  Estimation", "abstract": "  We present a lightweight solution for estimating spatially-coherent indoor\nlighting from a single RGB image. Previous methods for estimating illumination\nusing volumetric representations have overlooked the sparse distribution of\nlight sources in space, necessitating substantial memory and computational\nresources for achieving high-quality results. We introduce a unified, voxel\noctree-based illumination estimation framework to produce 3D spatially-coherent\nlighting. Additionally, a differentiable voxel octree cone tracing rendering\nlayer is proposed to eliminate regular volumetric representation throughout the\nentire process and ensure the retention of features across different frequency\ndomains. This reduction significantly decreases spatial usage and required\nfloating-point operations without substantially compromising precision.\nExperimental results demonstrate that our approach achieves high-quality\ncoherent estimation with minimal cost compared to previous methods.\n", "date": "2024-04-05", "categories": []}
{"id": "2404.04026", "title": "MM-Gaussian: 3D Gaussian-based Multi-modal Fusion for Localization and\n  Reconstruction in Unbounded Scenes", "abstract": "  Localization and mapping are critical tasks for various applications such as\nautonomous vehicles and robotics. The challenges posed by outdoor environments\npresent particular complexities due to their unbounded characteristics. In this\nwork, we present MM-Gaussian, a LiDAR-camera multi-modal fusion system for\nlocalization and mapping in unbounded scenes. Our approach is inspired by the\nrecently developed 3D Gaussians, which demonstrate remarkable capabilities in\nachieving high rendering quality and fast rendering speed. Specifically, our\nsystem fully utilizes the geometric structure information provided by\nsolid-state LiDAR to address the problem of inaccurate depth encountered when\nrelying solely on visual solutions in unbounded, outdoor scenarios.\nAdditionally, we utilize 3D Gaussian point clouds, with the assistance of\npixel-level gradient descent, to fully exploit the color information in photos,\nthereby achieving realistic rendering effects. To further bolster the\nrobustness of our system, we designed a relocalization module, which assists in\nreturning to the correct trajectory in the event of a localization failure.\nExperiments conducted in multiple scenarios demonstrate the effectiveness of\nour method.\n", "date": "2024-04-05", "categories": []}
{"id": "2404.04211", "title": "Robust Gaussian Splatting", "abstract": "  In this paper, we address common error sources for 3D Gaussian Splatting\n(3DGS) including blur, imperfect camera poses, and color inconsistencies, with\nthe goal of improving its robustness for practical applications like\nreconstructions from handheld phone captures. Our main contribution involves\nmodeling motion blur as a Gaussian distribution over camera poses, allowing us\nto address both camera pose refinement and motion blur correction in a unified\nway. Additionally, we propose mechanisms for defocus blur compensation and for\naddressing color in-consistencies caused by ambient light, shadows, or due to\ncamera-related factors like varying white balancing settings. Our proposed\nsolutions integrate in a seamless way with the 3DGS formulation while\nmaintaining its benefits in terms of training efficiency and rendering speed.\nWe experimentally validate our contributions on relevant benchmark datasets\nincluding Scannet++ and Deblur-NeRF, obtaining state-of-the-art results and\nthus consistent improvements over relevant baselines.\n", "date": "2024-04-05", "categories": []}
{"id": "2211.07459", "title": "Self-Aligning Depth-regularized Radiance Fields for Asynchronous RGB-D\n  Sequences", "abstract": "  It has been shown that learning radiance fields with depth rendering and\ndepth supervision can effectively promote the quality and convergence of view\nsynthesis. However, this paradigm requires input RGB-D sequences to be\nsynchronized, hindering its usage in the UAV city modeling scenario. As there\nexists asynchrony between RGB images and depth images due to high-speed flight,\nwe propose a novel time-pose function, which is an implicit network that maps\ntimestamps to $\\rm SE(3)$ elements. To simplify the training process, we also\ndesign a joint optimization scheme to jointly learn the large-scale\ndepth-regularized radiance fields and the time-pose function. Our algorithm\nconsists of three steps: (1) time-pose function fitting, (2) radiance field\nbootstrapping, (3) joint pose error compensation and radiance field refinement.\nIn addition, we propose a large synthetic dataset with diverse controlled\nmismatches and ground truth to evaluate this new problem setting\nsystematically. Through extensive experiments, we demonstrate that our method\noutperforms baselines without regularization. We also show qualitatively\nimproved results on a real-world asynchronous RGB-D sequence captured by drone.\nCodes, data, and models will be made publicly available.\n", "date": "2022-11-14", "categories": []}
{"id": "2312.00778", "title": "MorpheuS: Neural Dynamic 360{\\deg} Surface Reconstruction from Monocular\n  RGB-D Video", "abstract": "  Neural rendering has demonstrated remarkable success in dynamic scene\nreconstruction. Thanks to the expressiveness of neural representations, prior\nworks can accurately capture the motion and achieve high-fidelity\nreconstruction of the target object. Despite this, real-world video scenarios\noften feature large unobserved regions where neural representations struggle to\nachieve realistic completion. To tackle this challenge, we introduce MorpheuS,\na framework for dynamic 360{\\deg} surface reconstruction from a casually\ncaptured RGB-D video. Our approach models the target scene as a canonical field\nthat encodes its geometry and appearance, in conjunction with a deformation\nfield that warps points from the current frame to the canonical space. We\nleverage a view-dependent diffusion prior and distill knowledge from it to\nachieve realistic completion of unobserved regions. Experimental results on\nvarious real-world and synthetic datasets show that our method can achieve\nhigh-fidelity 360{\\deg} surface reconstruction of a deformable object from a\nmonocular RGB-D video.\n", "date": "2023-12-01", "categories": []}
{"id": "2404.03070", "title": "Behind the Veil: Enhanced Indoor 3D Scene Reconstruction with Occluded\n  Surfaces Completion", "abstract": "  In this paper, we present a novel indoor 3D reconstruction method with\noccluded surface completion, given a sequence of depth readings. Prior\nstate-of-the-art (SOTA) methods only focus on the reconstruction of the visible\nareas in a scene, neglecting the invisible areas due to the occlusions, e.g.,\nthe contact surface between furniture, occluded wall and floor. Our method\ntackles the task of completing the occluded scene surfaces, resulting in a\ncomplete 3D scene mesh. The core idea of our method is learning 3D geometry\nprior from various complete scenes to infer the occluded geometry of an unseen\nscene from solely depth measurements. We design a coarse-fine hierarchical\noctree representation coupled with a dual-decoder architecture, i.e.,\nGeo-decoder and 3D Inpainter, which jointly reconstructs the complete 3D scene\ngeometry. The Geo-decoder with detailed representation at fine levels is\noptimized online for each scene to reconstruct visible surfaces. The 3D\nInpainter with abstract representation at coarse levels is trained offline\nusing various scenes to complete occluded surfaces. As a result, while the\nGeo-decoder is specialized for an individual scene, the 3D Inpainter can be\ngenerally applied across different scenes. We evaluate the proposed method on\nthe 3D Completed Room Scene (3D-CRS) and iTHOR datasets, significantly\noutperforming the SOTA methods by a gain of 16.8% and 24.2% in terms of the\ncompleteness of 3D reconstruction. 3D-CRS dataset including a complete 3D mesh\nof each scene is provided at project webpage.\n", "date": "2024-04-03", "categories": []}
{"id": "2404.03652", "title": "The More You See in 2D, the More You Perceive in 3D", "abstract": "  Humans can infer 3D structure from 2D images of an object based on past\nexperience and improve their 3D understanding as they see more images. Inspired\nby this behavior, we introduce SAP3D, a system for 3D reconstruction and novel\nview synthesis from an arbitrary number of unposed images. Given a few unposed\nimages of an object, we adapt a pre-trained view-conditioned diffusion model\ntogether with the camera poses of the images via test-time fine-tuning. The\nadapted diffusion model and the obtained camera poses are then utilized as\ninstance-specific priors for 3D reconstruction and novel view synthesis. We\nshow that as the number of input images increases, the performance of our\napproach improves, bridging the gap between optimization-based prior-less 3D\nreconstruction methods and single-image-to-3D diffusion-based methods. We\ndemonstrate our system on real images as well as standard synthetic benchmarks.\nOur ablation studies confirm that this adaption behavior is key for more\naccurate 3D understanding.\n", "date": "2024-04-04", "categories": []}
{"id": "2404.03658", "title": "Know Your Neighbors: Improving Single-View Reconstruction via Spatial\n  Vision-Language Reasoning", "abstract": "  Recovering the 3D scene geometry from a single view is a fundamental yet\nill-posed problem in computer vision. While classical depth estimation methods\ninfer only a 2.5D scene representation limited to the image plane, recent\napproaches based on radiance fields reconstruct a full 3D representation.\nHowever, these methods still struggle with occluded regions since inferring\ngeometry without visual observation requires (i) semantic knowledge of the\nsurroundings, and (ii) reasoning about spatial context. We propose KYN, a novel\nmethod for single-view scene reconstruction that reasons about semantic and\nspatial context to predict each point's density. We introduce a vision-language\nmodulation module to enrich point features with fine-grained semantic\ninformation. We aggregate point representations across the scene through a\nlanguage-guided spatial attention mechanism to yield per-point density\npredictions aware of the 3D semantic context. We show that KYN improves 3D\nshape recovery compared to predicting density for each 3D point in isolation.\nWe achieve state-of-the-art results in scene and object reconstruction on\nKITTI-360, and show improved zero-shot generalization compared to prior work.\nProject page: https://ruili3.github.io/kyn.\n", "date": "2024-04-04", "categories": []}
{"id": "2307.10206", "title": "NEAT: Distilling 3D Wireframes from Neural Attraction Fields", "abstract": "  This paper studies the problem of structured 3D reconstruction using\nwireframes that consist of line segments and junctions, focusing on the\ncomputation of structured boundary geometries of scenes. Instead of leveraging\nmatching-based solutions from 2D wireframes (or line segments) for 3D wireframe\nreconstruction as done in prior arts, we present NEAT, a rendering-distilling\nformulation using neural fields to represent 3D line segments with 2D\nobservations, and bipartite matching for perceiving and distilling of a sparse\nset of 3D global junctions. The proposed {NEAT} enjoys the joint optimization\nof the neural fields and the global junctions from scratch, using\nview-dependent 2D observations without precomputed cross-view feature matching.\nComprehensive experiments on the DTU and BlendedMVS datasets demonstrate our\nNEAT's superiority over state-of-the-art alternatives for 3D wireframe\nreconstruction. Moreover, the distilled 3D global junctions by NEAT, are a\nbetter initialization than SfM points, for the recently-emerged 3D Gaussian\nSplatting for high-fidelity novel view synthesis using about 20 times fewer\ninitial 3D points. Project page: \\url{https://xuenan.net/neat}.\n", "date": "2023-07-14", "categories": []}
{"id": "2404.02391", "title": "APC2Mesh: Bridging the gap from occluded building fa\\c{c}ades to full 3D\n  models", "abstract": "  The benefits of having digital twins of urban buildings are numerous.\nHowever, a major difficulty encountered in their creation from airborne LiDAR\npoint clouds is the effective means of accurately reconstructing significant\nocclusions amidst point density variations and noise. To bridge the\nnoise/sparsity/occlusion gap and generate high fidelity 3D building models, we\npropose APC2Mesh which integrates point completion into a 3D reconstruction\npipeline, enabling the learning of dense geometrically accurate representation\nof buildings. Specifically, we leveraged complete points generated from\noccluded ones as input to a linearized skip attention-based deformation network\nfor 3D mesh reconstruction. In our experiments, conducted on 3 different\nscenes, we demonstrate that: (1) APC2Mesh delivers comparatively superior\nresults, indicating its efficacy in handling the challenges of occluded\nairborne building points of diverse styles and complexities. (2) The\ncombination of point completion with typical deep learning-based 3D point cloud\nreconstruction methods offers a direct and effective solution for\nreconstructing significantly occluded airborne building points. As such, this\nneural integration holds promise for advancing the creation of digital twins\nfor urban buildings with greater accuracy and fidelity.\n", "date": "2024-04-03", "categories": []}
{"id": "2404.02742", "title": "LiDAR4D: Dynamic Neural Fields for Novel Space-time View LiDAR Synthesis", "abstract": "  Although neural radiance fields (NeRFs) have achieved triumphs in image novel\nview synthesis (NVS), LiDAR NVS remains largely unexplored. Previous LiDAR NVS\nmethods employ a simple shift from image NVS methods while ignoring the dynamic\nnature and the large-scale reconstruction problem of LiDAR point clouds. In\nlight of this, we propose LiDAR4D, a differentiable LiDAR-only framework for\nnovel space-time LiDAR view synthesis. In consideration of the sparsity and\nlarge-scale characteristics, we design a 4D hybrid representation combined with\nmulti-planar and grid features to achieve effective reconstruction in a\ncoarse-to-fine manner. Furthermore, we introduce geometric constraints derived\nfrom point clouds to improve temporal consistency. For the realistic synthesis\nof LiDAR point clouds, we incorporate the global optimization of ray-drop\nprobability to preserve cross-region patterns. Extensive experiments on\nKITTI-360 and NuScenes datasets demonstrate the superiority of our method in\naccomplishing geometry-aware and time-consistent dynamic reconstruction. Codes\nare available at https://github.com/ispc-lab/LiDAR4D.\n", "date": "2024-04-03", "categories": []}
{"id": "2401.16416", "title": "Endo-4DGS: Endoscopic Monocular Scene Reconstruction with 4D Gaussian\n  Splatting", "abstract": "  In the realm of robot-assisted minimally invasive surgery, dynamic scene\nreconstruction can significantly enhance downstream tasks and improve surgical\noutcomes. Neural Radiance Fields (NeRF)-based methods have recently risen to\nprominence for their exceptional ability to reconstruct scenes but are hampered\nby slow inference speed, prolonged training, and inconsistent depth estimation.\nSome previous work utilizes ground truth depth for optimization but is hard to\nacquire in the surgical domain. To overcome these obstacles, we present\nEndo-4DGS, a real-time endoscopic dynamic reconstruction approach that utilizes\n3D Gaussian Splatting (GS) for 3D representation. Specifically, we propose\nlightweight MLPs to capture temporal dynamics with Gaussian deformation fields.\nTo obtain a satisfactory Gaussian Initialization, we exploit a powerful depth\nestimation foundation model, Depth-Anything, to generate pseudo-depth maps as a\ngeometry prior. We additionally propose confidence-guided learning to tackle\nthe ill-pose problems in monocular depth estimation and enhance the\ndepth-guided reconstruction with surface normal constraints and depth\nregularization. Our approach has been validated on two surgical datasets, where\nit can effectively render in real-time, compute efficiently, and reconstruct\nwith remarkable accuracy.\n", "date": "2024-01-29", "categories": []}
{"id": "2309.12790", "title": "NTO3D: Neural Target Object 3D Reconstruction with Segment Anything", "abstract": "  Neural 3D reconstruction from multi-view images has recently attracted\nincreasing attention from the community. Existing methods normally learn a\nneural field for the whole scene, while it is still under-explored how to\nreconstruct a target object indicated by users. Considering the Segment\nAnything Model (SAM) has shown effectiveness in segmenting any 2D images, in\nthis paper, we propose NTO3D, a novel high-quality Neural Target Object 3D\n(NTO3D) reconstruction method, which leverages the benefits of both neural\nfield and SAM. We first propose a novel strategy to lift the multi-view 2D\nsegmentation masks of SAM into a unified 3D occupancy field. The 3D occupancy\nfield is then projected into 2D space and generates the new prompts for SAM.\nThis process is iterative until convergence to separate the target object from\nthe scene. After this, we then lift the 2D features of the SAM encoder into a\n3D feature field in order to improve the reconstruction quality of the target\nobject. NTO3D lifts the 2D masks and features of SAM into the 3D neural field\nfor high-quality neural target object 3D reconstruction. We conduct detailed\nexperiments on several benchmark datasets to demonstrate the advantages of our\nmethod. The code will be available at: https://github.com/ucwxb/NTO3D.\n", "date": "2023-09-22", "categories": []}
{"id": "2403.17801", "title": "Towards 3D Vision with Low-Cost Single-Photon Cameras", "abstract": "  We present a method for reconstructing 3D shape of arbitrary Lambertian\nobjects based on measurements by miniature, energy-efficient, low-cost\nsingle-photon cameras. These cameras, operating as time resolved image sensors,\nilluminate the scene with a very fast pulse of diffuse light and record the\nshape of that pulse as it returns back from the scene at a high temporal\nresolution. We propose to model this image formation process, account for its\nnon-idealities, and adapt neural rendering to reconstruct 3D geometry from a\nset of spatially distributed sensors with known poses. We show that our\napproach can successfully recover complex 3D shapes from simulated data. We\nfurther demonstrate 3D object reconstruction from real-world captures,\nutilizing measurements from a commodity proximity sensor. Our work draws a\nconnection between image-based modeling and active range scanning and is a step\ntowards 3D vision with single-photon cameras.\n", "date": "2024-03-26", "categories": []}
{"id": "2403.19314", "title": "Total-Decom: Decomposed 3D Scene Reconstruction with Minimal Interaction", "abstract": "  Scene reconstruction from multi-view images is a fundamental problem in\ncomputer vision and graphics. Recent neural implicit surface reconstruction\nmethods have achieved high-quality results; however, editing and manipulating\nthe 3D geometry of reconstructed scenes remains challenging due to the absence\nof naturally decomposed object entities and complex object/background\ncompositions. In this paper, we present Total-Decom, a novel method for\ndecomposed 3D reconstruction with minimal human interaction. Our approach\nseamlessly integrates the Segment Anything Model (SAM) with hybrid\nimplicit-explicit neural surface representations and a mesh-based\nregion-growing technique for accurate 3D object decomposition. Total-Decom\nrequires minimal human annotations while providing users with real-time control\nover the granularity and quality of decomposition. We extensively evaluate our\nmethod on benchmark datasets and demonstrate its potential for downstream\napplications, such as animation and scene editing. The code is available at\nhttps://github.com/CVMI-Lab/Total-Decom.git.\n", "date": "2024-03-28", "categories": []}
{"id": "2404.00676", "title": "OmniLocalRF: Omnidirectional Local Radiance Fields from Dynamic Videos", "abstract": "  Omnidirectional cameras are extensively used in various applications to\nprovide a wide field of vision. However, they face a challenge in synthesizing\nnovel views due to the inevitable presence of dynamic objects, including the\nphotographer, in their wide field of view. In this paper, we introduce a new\napproach called Omnidirectional Local Radiance Fields (OmniLocalRF) that can\nrender static-only scene views, removing and inpainting dynamic objects\nsimultaneously. Our approach combines the principles of local radiance fields\nwith the bidirectional optimization of omnidirectional rays. Our input is an\nomnidirectional video, and we evaluate the mutual observations of the entire\nangle between the previous and current frames. To reduce ghosting artifacts of\ndynamic objects and inpaint occlusions, we devise a multi-resolution motion\nmask prediction module. Unlike existing methods that primarily separate dynamic\ncomponents through the temporal domain, our method uses multi-resolution neural\nfeature planes for precise segmentation, which is more suitable for long\n360-degree videos. Our experiments validate that OmniLocalRF outperforms\nexisting methods in both qualitative and quantitative metrics, especially in\nscenarios with complex real-world scenes. In particular, our approach\neliminates the need for manual interaction, such as drawing motion masks by\nhand and additional pose estimation, making it a highly effective and efficient\nsolution.\n", "date": "2024-03-31", "categories": []}
{"id": "2404.00678", "title": "OmniSDF: Scene Reconstruction using Omnidirectional Signed Distance\n  Functions and Adaptive Binoctrees", "abstract": "  We present a method to reconstruct indoor and outdoor static scene geometry\nand appearance from an omnidirectional video moving in a small circular sweep.\nThis setting is challenging because of the small baseline and large depth\nranges, making it difficult to find ray crossings. To better constrain the\noptimization, we estimate geometry as a signed distance field within a\nspherical binoctree data structure and use a complementary efficient tree\ntraversal strategy based on a breadth-first search for sampling. Unlike regular\ngrids or trees, the shape of this structure well-matches the camera setting,\ncreating a better memory-quality trade-off. From an initial depth estimate, the\nbinoctree is adaptively subdivided throughout the optimization; previous\nmethods use a fixed depth that leaves the scene undersampled. In comparison\nwith three neural optimization methods and two non-neural methods, ours shows\ndecreased geometry error on average, especially in a detailed scene, while\nsignificantly reducing the required number of voxels to represent such details.\n", "date": "2024-03-31", "categories": []}
{"id": "2404.01248", "title": "Scalable Scene Modeling from Perspective Imaging: Physics-based\n  Appearance and Geometry Inference", "abstract": "  3D scene modeling techniques serve as the bedrocks in the geospatial\nengineering and computer science, which drives many applications ranging from\nautomated driving, terrain mapping, navigation, virtual, augmented, mixed, and\nextended reality (for gaming and movie industry etc.). This dissertation\npresents a fraction of contributions that advances 3D scene modeling to its\nstate of the art, in the aspects of both appearance and geometry modeling. In\ncontrast to the prevailing deep learning methods, as a core contribution, this\nthesis aims to develop algorithms that follow first principles, where\nsophisticated physic-based models are introduced alongside with simpler\nlearning and inference tasks. The outcomes of these algorithms yield processes\nthat can consume much larger volume of data for highly accurate reconstructing\n3D scenes at a scale without losing methodological generality, which are not\npossible by contemporary complex-model based deep learning methods.\nSpecifically, the dissertation introduces three novel methodologies that\naddress the challenges of inferring appearance and geometry through\nphysics-based modeling.\n  Overall, the research encapsulated in this dissertation marks a series of\nmethodological triumphs in the processing of complex datasets. By navigating\nthe confluence of deep learning, computational geometry, and photogrammetry,\nthis work lays down a robust framework for future exploration and practical\napplication in the rapidly evolving field of 3D scene reconstruction. The\noutcomes of these studies are evidenced through rigorous experiments and\ncomparisons with existing state-of-the-art methods, demonstrating the efficacy\nand scalability of the proposed approaches.\n", "date": "2024-04-01", "categories": []}
{"id": "2312.01215", "title": "RNb-NeuS: Reflectance and Normal-based Multi-View 3D Reconstruction", "abstract": "  This paper introduces a versatile paradigm for integrating multi-view\nreflectance (optional) and normal maps acquired through photometric stereo. Our\napproach employs a pixel-wise joint re-parameterization of reflectance and\nnormal, considering them as a vector of radiances rendered under simulated,\nvarying illumination. This re-parameterization enables the seamless integration\nof reflectance and normal maps as input data in neural volume rendering-based\n3D reconstruction while preserving a single optimization objective. In\ncontrast, recent multi-view photometric stereo (MVPS) methods depend on\nmultiple, potentially conflicting objectives. Despite its apparent simplicity,\nour proposed approach outperforms state-of-the-art approaches in MVPS\nbenchmarks across F-score, Chamfer distance, and mean angular error metrics.\nNotably, it significantly improves the detailed 3D reconstruction of areas with\nhigh curvature or low visibility.\n", "date": "2023-12-02", "categories": []}
{"id": "2403.19985", "title": "Stable Surface Regularization for Fast Few-Shot NeRF", "abstract": "  This paper proposes an algorithm for synthesizing novel views under few-shot\nsetup. The main concept is to develop a stable surface regularization technique\ncalled Annealing Signed Distance Function (ASDF), which anneals the surface in\na coarse-to-fine manner to accelerate convergence speed. We observe that the\nEikonal loss - which is a widely known geometric regularization - requires\ndense training signal to shape different level-sets of SDF, leading to\nlow-fidelity results under few-shot training. In contrast, the proposed surface\nregularization successfully reconstructs scenes and produce high-fidelity\ngeometry with stable training. Our method is further accelerated by utilizing\ngrid representation and monocular geometric priors. Finally, the proposed\napproach is up to 45 times faster than existing few-shot novel view synthesis\nmethods, and it produces comparable results in the ScanNet dataset and\nNeRF-Real dataset.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.20013", "title": "DerainNeRF: 3D Scene Estimation with Adhesive Waterdrop Removal", "abstract": "  When capturing images through the glass during rainy or snowy weather\nconditions, the resulting images often contain waterdrops adhered on the glass\nsurface, and these waterdrops significantly degrade the image quality and\nperformance of many computer vision algorithms. To tackle these limitations, we\npropose a method to reconstruct the clear 3D scene implicitly from multi-view\nimages degraded by waterdrops. Our method exploits an attention network to\npredict the location of waterdrops and then train a Neural Radiance Fields to\nrecover the 3D scene implicitly. By leveraging the strong scene representation\ncapabilities of NeRF, our method can render high-quality novel-view images with\nwaterdrops removed. Extensive experimental results on both synthetic and real\ndatasets show that our method is able to generate clear 3D scenes and\noutperforms existing state-of-the-art (SOTA) image adhesive waterdrop removal\nmethods.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.20159", "title": "HGS-Mapping: Online Dense Mapping Using Hybrid Gaussian Representation\n  in Urban Scenes", "abstract": "  Online dense mapping of urban scenes forms a fundamental cornerstone for\nscene understanding and navigation of autonomous vehicles. Recent advancements\nin mapping methods are mainly based on NeRF, whose rendering speed is too slow\nto meet online requirements. 3D Gaussian Splatting (3DGS), with its rendering\nspeed hundreds of times faster than NeRF, holds greater potential in online\ndense mapping. However, integrating 3DGS into a street-view dense mapping\nframework still faces two challenges, including incomplete reconstruction due\nto the absence of geometric information beyond the LiDAR coverage area and\nextensive computation for reconstruction in large urban scenes. To this end, we\npropose HGS-Mapping, an online dense mapping framework in unbounded large-scale\nscenes. To attain complete construction, our framework introduces Hybrid\nGaussian Representation, which models different parts of the entire scene using\nGaussians with distinct properties. Furthermore, we employ a hybrid Gaussian\ninitialization mechanism and an adaptive update method to achieve high-fidelity\nand rapid reconstruction. To the best of our knowledge, we are the first to\nintegrate Gaussian representation into online dense mapping of urban scenes.\nOur approach achieves SOTA reconstruction accuracy while only employing 66%\nnumber of Gaussians, leading to 20% faster reconstruction speed.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.20275", "title": "Snap-it, Tap-it, Splat-it: Tactile-Informed 3D Gaussian Splatting for\n  Reconstructing Challenging Surfaces", "abstract": "  Touch and vision go hand in hand, mutually enhancing our ability to\nunderstand the world. From a research perspective, the problem of mixing touch\nand vision is underexplored and presents interesting challenges. To this end,\nwe propose Tactile-Informed 3DGS, a novel approach that incorporates touch data\n(local depth maps) with multi-view vision data to achieve surface\nreconstruction and novel view synthesis. Our method optimises 3D Gaussian\nprimitives to accurately model the object's geometry at points of contact. By\ncreating a framework that decreases the transmittance at touch locations, we\nachieve a refined surface reconstruction, ensuring a uniformly smooth depth\nmap. Touch is particularly useful when considering non-Lambertian objects (e.g.\nshiny or reflective surfaces) since contemporary methods tend to fail to\nreconstruct with fidelity specular highlights. By combining vision and tactile\nsensing, we achieve more accurate geometry reconstructions with fewer images\nthan prior methods. We conduct evaluation on objects with glossy and reflective\nsurfaces and demonstrate the effectiveness of our approach, offering\nsignificant improvements in reconstruction quality.\n", "date": "2024-03-29", "categories": []}
{"id": "2309.03008", "title": "Sparse 3D Reconstruction via Object-Centric Ray Sampling", "abstract": "  We propose a novel method for 3D object reconstruction from a sparse set of\nviews captured from a 360-degree calibrated camera rig. We represent the object\nsurface through a hybrid model that uses both an MLP-based neural\nrepresentation and a triangle mesh. A key contribution in our work is a novel\nobject-centric sampling scheme of the neural representation, where rays are\nshared among all views. This efficiently concentrates and reduces the number of\nsamples used to update the neural model at each iteration. This sampling scheme\nrelies on the mesh representation to ensure also that samples are\nwell-distributed along its normals. The rendering is then performed efficiently\nby a differentiable renderer. We demonstrate that this sampling scheme results\nin a more effective training of the neural representation, does not require the\nadditional supervision of segmentation masks, yields state of the art 3D\nreconstructions, and works with sparse views on the Google's Scanned Objects,\nTank and Temples and MVMC Car datasets. Code available at:\nhttps://github.com/llukmancerkezi/ROSTER\n", "date": "2023-09-06", "categories": []}
{"id": "2311.04154", "title": "High-fidelity 3D Reconstruction of Plants using Neural Radiance Field", "abstract": "  Accurate reconstruction of plant phenotypes plays a key role in optimising\nsustainable farming practices in the field of Precision Agriculture (PA).\nCurrently, optical sensor-based approaches dominate the field, but the need for\nhigh-fidelity 3D reconstruction of crops and plants in unstructured\nagricultural environments remains challenging. Recently, a promising\ndevelopment has emerged in the form of Neural Radiance Field (NeRF), a novel\nmethod that utilises neural density fields. This technique has shown impressive\nperformance in various novel vision synthesis tasks, but has remained\nrelatively unexplored in the agricultural context. In our study, we focus on\ntwo fundamental tasks within plant phenotyping: (1) the synthesis of 2D\nnovel-view images and (2) the 3D reconstruction of crop and plant models. We\nexplore the world of neural radiance fields, in particular two SOTA methods:\nInstant-NGP, which excels in generating high-quality images with impressive\ntraining and inference speed, and Instant-NSR, which improves the reconstructed\ngeometry by incorporating the Signed Distance Function (SDF) during training.\nIn particular, we present a novel plant phenotype dataset comprising real plant\nimages from production environments. This dataset is a first-of-its-kind\ninitiative aimed at comprehensively exploring the advantages and limitations of\nNeRF in agricultural contexts. Our experimental results show that NeRF\ndemonstrates commendable performance in the synthesis of novel-view images and\nis able to achieve reconstruction results that are competitive with Reality\nCapture, a leading commercial software for 3D Multi-View Stereo (MVS)-based\nreconstruction. However, our study also highlights certain drawbacks of NeRF,\nincluding relatively slow training speeds, performance limitations in cases of\ninsufficient sampling, and challenges in obtaining geometry quality in complex\nsetups.\n", "date": "2023-11-07", "categories": []}
{"id": "2403.19632", "title": "GauStudio: A Modular Framework for 3D Gaussian Splatting and Beyond", "abstract": "  We present GauStudio, a novel modular framework for modeling 3D Gaussian\nSplatting (3DGS) to provide standardized, plug-and-play components for users to\neasily customize and implement a 3DGS pipeline. Supported by our framework, we\npropose a hybrid Gaussian representation with foreground and skyball background\nmodels. Experiments demonstrate this representation reduces artifacts in\nunbounded outdoor scenes and improves novel view synthesis. Finally, we propose\nGaussian Splatting Surface Reconstruction (GauS), a novel render-then-fuse\napproach for high-fidelity mesh reconstruction from 3DGS inputs without\nfine-tuning. Overall, our GauStudio framework, hybrid representation, and GauS\napproach enhance 3DGS modeling and rendering capabilities, enabling\nhigher-quality novel view synthesis and surface reconstruction.\n", "date": "2024-03-28", "categories": []}
{"id": "2312.09138", "title": "Living Scenes: Multi-object Relocalization and Reconstruction in\n  Changing 3D Environments", "abstract": "  Research into dynamic 3D scene understanding has primarily focused on\nshort-term change tracking from dense observations, while little attention has\nbeen paid to long-term changes with sparse observations. We address this gap\nwith MoRE, a novel approach for multi-object relocalization and reconstruction\nin evolving environments. We view these environments as \"living scenes\" and\nconsider the problem of transforming scans taken at different points in time\ninto a 3D reconstruction of the object instances, whose accuracy and\ncompleteness increase over time. At the core of our method lies an\nSE(3)-equivariant representation in a single encoder-decoder network, trained\non synthetic data. This representation enables us to seamlessly tackle instance\nmatching, registration, and reconstruction. We also introduce a joint\noptimization algorithm that facilitates the accumulation of point clouds\noriginating from the same instance across multiple scans taken at different\npoints in time. We validate our method on synthetic and real-world data and\ndemonstrate state-of-the-art performance in both end-to-end performance and\nindividual subtasks.\n", "date": "2023-12-14", "categories": []}
{"id": "2403.17638", "title": "Learning with Unreliability: Fast Few-shot Voxel Radiance Fields with\n  Relative Geometric Consistency", "abstract": "  We propose a voxel-based optimization framework, ReVoRF, for few-shot\nradiance fields that strategically address the unreliability in pseudo novel\nview synthesis. Our method pivots on the insight that relative depth\nrelationships within neighboring regions are more reliable than the absolute\ncolor values in disoccluded areas. Consequently, we devise a bilateral\ngeometric consistency loss that carefully navigates the trade-off between color\nfidelity and geometric accuracy in the context of depth consistency for\nuncertain regions. Moreover, we present a reliability-guided learning strategy\nto discern and utilize the variable quality across synthesized views,\ncomplemented by a reliability-aware voxel smoothing algorithm that smoothens\nthe transition between reliable and unreliable data patches. Our approach\nallows for a more nuanced use of all available data, promoting enhanced\nlearning from regions previously considered unsuitable for high-quality\nreconstruction. Extensive experiments across diverse datasets reveal that our\napproach attains significant gains in efficiency and accuracy, delivering\nrendering speeds of 3 FPS, 7 mins to train a $360^\\circ$ scene, and a 5\\%\nimprovement in PSNR over existing few-shot methods. Code is available at\nhttps://github.com/HKCLynn/ReVoRF.\n", "date": "2024-03-26", "categories": []}
{"id": "2403.01861", "title": "AiSDF: Structure-aware Neural Signed Distance Fields in Indoor Scenes", "abstract": "  Indoor scenes we are living in are visually homogenous or textureless, while\nthey inherently have structural forms and provide enough structural priors for\n3D scene reconstruction. Motivated by this fact, we propose a structure-aware\nonline signed distance fields (SDF) reconstruction framework in indoor scenes,\nespecially under the Atlanta world (AW) assumption. Thus, we dub this\nincremental SDF reconstruction for AW as AiSDF. Within the online framework, we\ninfer the underlying Atlanta structure of a given scene and then estimate\nplanar surfel regions supporting the Atlanta structure. This Atlanta-aware\nsurfel representation provides an explicit planar map for a given scene. In\naddition, based on these Atlanta planar surfel regions, we adaptively sample\nand constrain the structural regularity in the SDF reconstruction, which\nenables us to improve the reconstruction quality by maintaining a high-level\nstructure while enhancing the details of a given scene. We evaluate the\nproposed AiSDF on the ScanNet and ReplicaCAD datasets, where we demonstrate\nthat the proposed framework is capable of reconstructing fine details of\nobjects implicitly, as well as structures explicitly in room-scale scenes.\n", "date": "2024-03-04", "categories": []}
{"id": "2403.04926", "title": "BAGS: Blur Agnostic Gaussian Splatting through Multi-Scale Kernel\n  Modeling", "abstract": "  Recent efforts in using 3D Gaussians for scene reconstruction and novel view\nsynthesis can achieve impressive results on curated benchmarks; however, images\ncaptured in real life are often blurry. In this work, we analyze the robustness\nof Gaussian-Splatting-based methods against various image blur, such as motion\nblur, defocus blur, downscaling blur, \\etc. Under these degradations,\nGaussian-Splatting-based methods tend to overfit and produce worse results than\nNeural-Radiance-Field-based methods. To address this issue, we propose Blur\nAgnostic Gaussian Splatting (BAGS). BAGS introduces additional 2D modeling\ncapacities such that a 3D-consistent and high quality scene can be\nreconstructed despite image-wise blur. Specifically, we model blur by\nestimating per-pixel convolution kernels from a Blur Proposal Network (BPN).\nBPN is designed to consider spatial, color, and depth variations of the scene\nto maximize modeling capacity. Additionally, BPN also proposes a\nquality-assessing mask, which indicates regions where blur occur. Finally, we\nintroduce a coarse-to-fine kernel optimization scheme; this optimization scheme\nis fast and avoids sub-optimal solutions due to a sparse point cloud\ninitialization, which often occurs when we apply Structure-from-Motion on\nblurry images. We demonstrate that BAGS achieves photorealistic renderings\nunder various challenging blur conditions and imaging geometry, while\nsignificantly improving upon existing approaches.\n", "date": "2024-03-07", "categories": []}
{"id": "2403.15530", "title": "Pixel-GS: Density Control with Pixel-aware Gradient for 3D Gaussian\n  Splatting", "abstract": "  3D Gaussian Splatting (3DGS) has demonstrated impressive novel view synthesis\nresults while advancing real-time rendering performance. However, it relies\nheavily on the quality of the initial point cloud, resulting in blurring and\nneedle-like artifacts in areas with insufficient initializing points. This is\nmainly attributed to the point cloud growth condition in 3DGS that only\nconsiders the average gradient magnitude of points from observable views,\nthereby failing to grow for large Gaussians that are observable for many\nviewpoints while many of them are only covered in the boundaries. To this end,\nwe propose a novel method, named Pixel-GS, to take into account the number of\npixels covered by the Gaussian in each view during the computation of the\ngrowth condition. We regard the covered pixel numbers as the weights to\ndynamically average the gradients from different views, such that the growth of\nlarge Gaussians can be prompted. As a result, points within the areas with\ninsufficient initializing points can be grown more effectively, leading to a\nmore accurate and detailed reconstruction. In addition, we propose a simple yet\neffective strategy to scale the gradient field according to the distance to the\ncamera, to suppress the growth of floaters near the camera. Extensive\nexperiments both qualitatively and quantitatively demonstrate that our method\nachieves state-of-the-art rendering quality while maintaining real-time\nrendering speed, on the challenging Mip-NeRF 360 and Tanks & Temples datasets.\n", "date": "2024-03-22", "categories": []}
{"id": "2403.16141", "title": "Entity-NeRF: Detecting and Removing Moving Entities in Urban Scenes", "abstract": "  Recent advancements in the study of Neural Radiance Fields (NeRF) for dynamic\nscenes often involve explicit modeling of scene dynamics. However, this\napproach faces challenges in modeling scene dynamics in urban environments,\nwhere moving objects of various categories and scales are present. In such\nsettings, it becomes crucial to effectively eliminate moving objects to\naccurately reconstruct static backgrounds. Our research introduces an\ninnovative method, termed here as Entity-NeRF, which combines the strengths of\nknowledge-based and statistical strategies. This approach utilizes entity-wise\nstatistics, leveraging entity segmentation and stationary entity classification\nthrough thing/stuff segmentation. To assess our methodology, we created an\nurban scene dataset masked with moving objects. Our comprehensive experiments\ndemonstrate that Entity-NeRF notably outperforms existing techniques in\nremoving moving objects and reconstructing static urban backgrounds, both\nquantitatively and qualitatively.\n", "date": "2024-03-24", "categories": []}
{"id": "2309.06030", "title": "Federated Learning for Large-Scale Scene Modeling with Neural Radiance\n  Fields", "abstract": "  We envision a system to continuously build and maintain a map based on\nearth-scale neural radiance fields (NeRF) using data collected from vehicles\nand drones in a lifelong learning manner. However, existing large-scale\nmodeling by NeRF has problems in terms of scalability and maintainability when\nmodeling earth-scale environments. Therefore, to address these problems, we\npropose a federated learning pipeline for large-scale modeling with NeRF. We\ntailor the model aggregation pipeline in federated learning for NeRF, thereby\nallowing local updates of NeRF. In the aggregation step, the accuracy of the\nclients' global pose is critical. Thus, we also propose global pose alignment\nto align the noisy global pose of clients before the aggregation step. In\nexperiments, we show the effectiveness of the proposed pose alignment and the\nfederated learning pipeline on the large-scale scene dataset, Mill19.\n", "date": "2023-09-12", "categories": []}
{"id": "2403.14366", "title": "SurroundSDF: Implicit 3D Scene Understanding Based on Signed Distance\n  Field", "abstract": "  Vision-centric 3D environment understanding is both vital and challenging for\nautonomous driving systems. Recently, object-free methods have attracted\nconsiderable attention. Such methods perceive the world by predicting the\nsemantics of discrete voxel grids but fail to construct continuous and accurate\nobstacle surfaces. To this end, in this paper, we propose SurroundSDF to\nimplicitly predict the signed distance field (SDF) and semantic field for the\ncontinuous perception from surround images. Specifically, we introduce a\nquery-based approach and utilize SDF constrained by the Eikonal formulation to\naccurately describe the surfaces of obstacles. Furthermore, considering the\nabsence of precise SDF ground truth, we propose a novel weakly supervised\nparadigm for SDF, referred to as the Sandwich Eikonal formulation, which\nemphasizes applying correct and dense constraints on both sides of the surface,\nthereby enhancing the perceptual accuracy of the surface. Experiments suggest\nthat our method achieves SOTA for both occupancy prediction and 3D scene\nreconstruction tasks on the nuScenes dataset.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.14619", "title": "ClusteringSDF: Self-Organized Neural Implicit Surfaces for 3D\n  Decomposition", "abstract": "  3D decomposition/segmentation still remains a challenge as large-scale 3D\nannotated data is not readily available. Contemporary approaches typically\nleverage 2D machine-generated segments, integrating them for 3D consistency.\nWhile the majority of these methods are based on NeRFs, they face a potential\nweakness that the instance/semantic embedding features derive from independent\nMLPs, thus preventing the segmentation network from learning the geometric\ndetails of the objects directly through radiance and density. In this paper, we\npropose ClusteringSDF, a novel approach to achieve both segmentation and\nreconstruction in 3D via the neural implicit surface representation,\nspecifically Signal Distance Function (SDF), where the segmentation rendering\nis directly integrated with the volume rendering of neural implicit surfaces.\nAlthough based on ObjectSDF++, ClusteringSDF no longer requires the\nground-truth segments for supervision while maintaining the capability of\nreconstructing individual object surfaces, but purely with the noisy and\ninconsistent labels from pre-trained models.As the core of ClusteringSDF, we\nintroduce a high-efficient clustering mechanism for lifting the 2D labels to 3D\nand the experimental results on the challenging scenes from ScanNet and Replica\ndatasets show that ClusteringSDF can achieve competitive performance compared\nagainst the state-of-the-art with significantly reduced training time.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.14621", "title": "GRM: Large Gaussian Reconstruction Model for Efficient 3D Reconstruction\n  and Generation", "abstract": "  We introduce GRM, a large-scale reconstructor capable of recovering a 3D\nasset from sparse-view images in around 0.1s. GRM is a feed-forward\ntransformer-based model that efficiently incorporates multi-view information to\ntranslate the input pixels into pixel-aligned Gaussians, which are unprojected\nto create a set of densely distributed 3D Gaussians representing a scene.\nTogether, our transformer architecture and the use of 3D Gaussians unlock a\nscalable and efficient reconstruction framework. Extensive experimental results\ndemonstrate the superiority of our method over alternatives regarding both\nreconstruction quality and efficiency. We also showcase the potential of GRM in\ngenerative tasks, i.e., text-to-3D and image-to-3D, by integrating it with\nexisting multi-view diffusion models. Our project website is at:\nhttps://justimyhxu.github.io/projects/grm/.\n", "date": "2024-03-21", "categories": []}
{"id": "2311.02313", "title": "LISNeRF Mapping: LiDAR-based Implicit Mapping via Semantic Neural Fields\n  for Large-Scale 3D Scenes", "abstract": "  Large-scale semantic mapping is crucial for outdoor autonomous agents to\nfulfill high-level tasks such as planning and navigation. This paper proposes a\nnovel method for large-scale 3D semantic reconstruction through implicit\nrepresentations from posed LiDAR measurements alone. We first leverage an\noctree-based and hierarchical structure to store implicit features, then these\nimplicit features are decoded to semantic information and signed distance value\nthrough shallow Multilayer Perceptrons (MLPs). We adopt off-the-shelf\nalgorithms to predict the semantic labels and instance IDs of point clouds. We\nthen jointly optimize the feature embeddings and MLPs parameters with a\nself-supervision paradigm for point cloud geometry and a pseudo-supervision\nparadigm for semantic and panoptic labels. Subsequently, categories and\ngeometric structures for novel points are regressed, and marching cubes are\nexploited to subdivide and visualize the scenes in the inferring stage. For\nscenarios with memory constraints, a map stitching strategy is also developed\nto merge sub-maps into a complete map. Experiments on two real-world datasets,\nSemanticKITTI and SemanticPOSS, demonstrate the superior segmentation\nefficiency and mapping effectiveness of our framework compared to current\nstate-of-the-art 3D LiDAR mapping methods.\n", "date": "2023-11-04", "categories": []}
{"id": "2312.07920", "title": "DrivingGaussian: Composite Gaussian Splatting for Surrounding Dynamic\n  Autonomous Driving Scenes", "abstract": "  We present DrivingGaussian, an efficient and effective framework for\nsurrounding dynamic autonomous driving scenes. For complex scenes with moving\nobjects, we first sequentially and progressively model the static background of\nthe entire scene with incremental static 3D Gaussians. We then leverage a\ncomposite dynamic Gaussian graph to handle multiple moving objects,\nindividually reconstructing each object and restoring their accurate positions\nand occlusion relationships within the scene. We further use a LiDAR prior for\nGaussian Splatting to reconstruct scenes with greater details and maintain\npanoramic consistency. DrivingGaussian outperforms existing methods in dynamic\ndriving scene reconstruction and enables photorealistic surround-view synthesis\nwith high-fidelity and multi-camera consistency. Our project page is at:\nhttps://github.com/VDIGPKU/DrivingGaussian.\n", "date": "2023-12-13", "categories": []}
{"id": "2403.13064", "title": "SceneScript: Reconstructing Scenes With An Autoregressive Structured\n  Language Model", "abstract": "  We introduce SceneScript, a method that directly produces full scene models\nas a sequence of structured language commands using an autoregressive,\ntoken-based approach. Our proposed scene representation is inspired by recent\nsuccesses in transformers & LLMs, and departs from more traditional methods\nwhich commonly describe scenes as meshes, voxel grids, point clouds or radiance\nfields. Our method infers the set of structured language commands directly from\nencoded visual data using a scene language encoder-decoder architecture. To\ntrain SceneScript, we generate and release a large-scale synthetic dataset\ncalled Aria Synthetic Environments consisting of 100k high-quality in-door\nscenes, with photorealistic and ground-truth annotated renders of egocentric\nscene walkthroughs. Our method gives state-of-the art results in architectural\nlayout estimation, and competitive results in 3D object detection. Lastly, we\nexplore an advantage for SceneScript, which is the ability to readily adapt to\nnew commands via simple additions to the structured language, which we\nillustrate for tasks such as coarse 3D object part reconstruction.\n", "date": "2024-03-19", "categories": []}
{"id": "2308.09591", "title": "O$^2$-Recon: Completing 3D Reconstruction of Occluded Objects in the\n  Scene with a Pre-trained 2D Diffusion Model", "abstract": "  Occlusion is a common issue in 3D reconstruction from RGB-D videos, often\nblocking the complete reconstruction of objects and presenting an ongoing\nproblem. In this paper, we propose a novel framework, empowered by a 2D\ndiffusion-based in-painting model, to reconstruct complete surfaces for the\nhidden parts of objects. Specifically, we utilize a pre-trained diffusion model\nto fill in the hidden areas of 2D images. Then we use these in-painted images\nto optimize a neural implicit surface representation for each instance for 3D\nreconstruction. Since creating the in-painting masks needed for this process is\ntricky, we adopt a human-in-the-loop strategy that involves very little human\nengagement to generate high-quality masks. Moreover, some parts of objects can\nbe totally hidden because the videos are usually shot from limited\nperspectives. To ensure recovering these invisible areas, we develop a cascaded\nnetwork architecture for predicting signed distance field, making use of\ndifferent frequency bands of positional encoding and maintaining overall\nsmoothness. Besides the commonly used rendering loss, Eikonal loss, and\nsilhouette loss, we adopt a CLIP-based semantic consistency loss to guide the\nsurface from unseen camera angles. Experiments on ScanNet scenes show that our\nproposed framework achieves state-of-the-art accuracy and completeness in\nobject-level reconstruction from scene-level RGB-D videos. Code:\nhttps://github.com/THU-LYJ-Lab/O2-Recon.\n", "date": "2023-08-18", "categories": []}
{"id": "2310.00145", "title": "3D Reconstruction in Noisy Agricultural Environments: A Bayesian\n  Optimization Perspective for View Planning", "abstract": "  3D reconstruction is a fundamental task in robotics that gained attention due\nto its major impact in a wide variety of practical settings, including\nagriculture, underwater, and urban environments. This task can be carried out\nvia view planning (VP), which aims to optimally place a certain number of\ncameras in positions that maximize the visual information, improving the\nresulting 3D reconstruction. Nonetheless, in most real-world settings, existing\nenvironmental noise can significantly affect the performance of 3D\nreconstruction. To that end, this work advocates a novel geometric-based\nreconstruction quality function for VP, that accounts for the existing noise of\nthe environment, without requiring its closed-form expression. With no analytic\nexpression of the objective function, this work puts forth an adaptive Bayesian\noptimization algorithm for accurate 3D reconstruction in the presence of noise.\nNumerical tests on noisy agricultural environments showcase the merits of the\nproposed approach for 3D reconstruction with even a small number of available\ncameras.\n", "date": "2023-09-29", "categories": []}
{"id": "2403.10357", "title": "ANIM: Accurate Neural Implicit Model for Human Reconstruction from a\n  single RGB-D image", "abstract": "  Recent progress in human shape learning, shows that neural implicit models\nare effective in generating 3D human surfaces from limited number of views, and\neven from a single RGB image. However, existing monocular approaches still\nstruggle to recover fine geometric details such as face, hands or cloth\nwrinkles. They are also easily prone to depth ambiguities that result in\ndistorted geometries along the camera optical axis. In this paper, we explore\nthe benefits of incorporating depth observations in the reconstruction process\nby introducing ANIM, a novel method that reconstructs arbitrary 3D human shapes\nfrom single-view RGB-D images with an unprecedented level of accuracy. Our\nmodel learns geometric details from both multi-resolution pixel-aligned and\nvoxel-aligned features to leverage depth information and enable spatial\nrelationships, mitigating depth ambiguities. We further enhance the quality of\nthe reconstructed shape by introducing a depth-supervision strategy, which\nimproves the accuracy of the signed distance field estimation of points that\nlie on the reconstructed surface. Experiments demonstrate that ANIM outperforms\nstate-of-the-art works that use RGB, surface normals, point cloud or RGB-D data\nas input. In addition, we introduce ANIM-Real, a new multi-modal dataset\ncomprising high-quality scans paired with consumer-grade RGB-D camera, and our\nprotocol to fine-tune ANIM, enabling high-quality reconstruction from\nreal-world human capture.\n", "date": "2024-03-15", "categories": []}
{"id": "2403.11831", "title": "BAD-Gaussians: Bundle Adjusted Deblur Gaussian Splatting", "abstract": "  While neural rendering has demonstrated impressive capabilities in 3D scene\nreconstruction and novel view synthesis, it heavily relies on high-quality\nsharp images and accurate camera poses. Numerous approaches have been proposed\nto train Neural Radiance Fields (NeRF) with motion-blurred images, commonly\nencountered in real-world scenarios such as low-light or long-exposure\nconditions. However, the implicit representation of NeRF struggles to\naccurately recover intricate details from severely motion-blurred images and\ncannot achieve real-time rendering. In contrast, recent advancements in 3D\nGaussian Splatting achieve high-quality 3D scene reconstruction and real-time\nrendering by explicitly optimizing point clouds as Gaussian spheres.\n  In this paper, we introduce a novel approach, named BAD-Gaussians (Bundle\nAdjusted Deblur Gaussian Splatting), which leverages explicit Gaussian\nrepresentation and handles severe motion-blurred images with inaccurate camera\nposes to achieve high-quality scene reconstruction. Our method models the\nphysical image formation process of motion-blurred images and jointly learns\nthe parameters of Gaussians while recovering camera motion trajectories during\nexposure time.\n  In our experiments, we demonstrate that BAD-Gaussians not only achieves\nsuperior rendering quality compared to previous state-of-the-art deblur neural\nrendering methods on both synthetic and real datasets but also enables\nreal-time rendering capabilities.\n  Our project page and source code is available at\nhttps://lingzhezhao.github.io/BAD-Gaussians/\n", "date": "2024-03-18", "categories": []}
{"id": "2403.12722", "title": "HUGS: Holistic Urban 3D Scene Understanding via Gaussian Splatting", "abstract": "  Holistic understanding of urban scenes based on RGB images is a challenging\nyet important problem. It encompasses understanding both the geometry and\nappearance to enable novel view synthesis, parsing semantic labels, and\ntracking moving objects. Despite considerable progress, existing approaches\noften focus on specific aspects of this task and require additional inputs such\nas LiDAR scans or manually annotated 3D bounding boxes. In this paper, we\nintroduce a novel pipeline that utilizes 3D Gaussian Splatting for holistic\nurban scene understanding. Our main idea involves the joint optimization of\ngeometry, appearance, semantics, and motion using a combination of static and\ndynamic 3D Gaussians, where moving object poses are regularized via physical\nconstraints. Our approach offers the ability to render new viewpoints in\nreal-time, yielding 2D and 3D semantic information with high accuracy, and\nreconstruct dynamic scenes, even in scenarios where 3D bounding box detection\nare highly noisy. Experimental results on KITTI, KITTI-360, and Virtual KITTI 2\ndemonstrate the effectiveness of our approach.\n", "date": "2024-03-19", "categories": []}
{"id": "2403.11233", "title": "STAIR: Semantic-Targeted Active Implicit Reconstruction", "abstract": "  Many autonomous robotic applications require object-level understanding when\ndeployed. Actively reconstructing objects of interest, i.e. objects with\nspecific semantic meanings, is therefore relevant for a robot to perform\ndownstream tasks in an initially unknown environment. In this work, we propose\na novel framework for semantic-targeted active reconstruction using posed RGB-D\nmeasurements and 2D semantic labels as input. The key components of our\nframework are a semantic implicit neural representation and a compatible\nplanning utility function based on semantic rendering and uncertainty\nestimation, enabling adaptive view planning to target objects of interest. Our\nplanning approach achieves better reconstruction performance in terms of mesh\nand novel view rendering quality compared to implicit reconstruction baselines\nthat do not consider semantics for view planning. Our framework further\noutperforms a state-of-the-art semantic-targeted active reconstruction pipeline\nbased on explicit maps, justifying our choice of utilising implicit neural\nrepresentations to tackle semantic-targeted active reconstruction problems.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.11364", "title": "Creating Seamless 3D Maps Using Radiance Fields", "abstract": "  It is desirable to create 3D object models and 3D maps from 2D input images\nfor applications such as navigation, virtual tourism, and urban planning. The\ntraditional methods of creating 3D maps, (such as photogrammetry), require a\nlarge number of images and odometry. Additionally, traditional methods have\ndifficulty with reflective surfaces and specular reflections; windows and\nchrome in the scene can be problematic. Google Road View is a familiar\napplication, which uses traditional methods to fuse a collection of 2D input\nimages into the illusion of a 3D map. However, Google Road View does not create\nan actual 3D object model, only a collection of views. The objective of this\nwork is to create an actual 3D object model using updated techniques. Neural\nRadiance Fields (NeRF[1]) has emerged as a potential solution, offering the\ncapability to produce more precise and intricate 3D maps. Gaussian Splatting[4]\nis another contemporary technique. This investigation compares Neural Radiance\nFields to Gaussian Splatting, and describes some of their inner workings. Our\nprimary contribution is a method for improving the results of the 3D\nreconstructed models. Our results indicate that Gaussian Splatting was superior\nto the NeRF technique.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.11447", "title": "Motion-aware 3D Gaussian Splatting for Efficient Dynamic Scene\n  Reconstruction", "abstract": "  3D Gaussian Splatting (3DGS) has become an emerging tool for dynamic scene\nreconstruction. However, existing methods focus mainly on extending static 3DGS\ninto a time-variant representation, while overlooking the rich motion\ninformation carried by 2D observations, thus suffering from performance\ndegradation and model redundancy. To address the above problem, we propose a\nnovel motion-aware enhancement framework for dynamic scene reconstruction,\nwhich mines useful motion cues from optical flow to improve different paradigms\nof dynamic 3DGS. Specifically, we first establish a correspondence between 3D\nGaussian movements and pixel-level flow. Then a novel flow augmentation method\nis introduced with additional insights into uncertainty and loss collaboration.\nMoreover, for the prevalent deformation-based paradigm that presents a harder\noptimization problem, a transient-aware deformation auxiliary module is\nproposed. We conduct extensive experiments on both multi-view and monocular\nscenes to verify the merits of our work. Compared with the baselines, our\nmethod shows significant superiority in both rendering quality and efficiency.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.11460", "title": "Fed3DGS: Scalable 3D Gaussian Splatting with Federated Learning", "abstract": "  In this work, we present Fed3DGS, a scalable 3D reconstruction framework\nbased on 3D Gaussian splatting (3DGS) with federated learning. Existing\ncity-scale reconstruction methods typically adopt a centralized approach, which\ngathers all data in a central server and reconstructs scenes. The approach\nhampers scalability because it places a heavy load on the server and demands\nextensive data storage when reconstructing scenes on a scale beyond city-scale.\nIn pursuit of a more scalable 3D reconstruction, we propose a federated\nlearning framework with 3DGS, which is a decentralized framework and can\npotentially use distributed computational resources across millions of clients.\nWe tailor a distillation-based model update scheme for 3DGS and introduce\nappearance modeling for handling non-IID data in the scenario of 3D\nreconstruction with federated learning. We simulate our method on several\nlarge-scale benchmarks, and our method demonstrates rendered image quality\ncomparable to centralized approaches. In addition, we also simulate our method\nwith data collected in different seasons, demonstrating that our framework can\nreflect changes in the scenes and our appearance modeling captures changes due\nto seasonal variations.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.11789", "title": "EMIE-MAP: Large-Scale Road Surface Reconstruction Based on Explicit Mesh\n  and Implicit Encoding", "abstract": "  Road surface reconstruction plays a vital role in autonomous driving systems,\nenabling road lane perception and high-precision mapping. Recently, neural\nimplicit encoding has achieved remarkable results in scene representation,\nparticularly in the realistic rendering of scene textures. However, it faces\nchallenges in directly representing geometric information for large-scale\nscenes. To address this, we propose EMIE-MAP, a novel method for large-scale\nroad surface reconstruction based on explicit mesh and implicit encoding. The\nroad geometry is represented using explicit mesh, where each vertex stores\nimplicit encoding representing the color and semantic information. To overcome\nthe difficulty in optimizing road elevation, we introduce a trajectory-based\nelevation initialization and an elevation residual learning method based on\nMulti-Layer Perceptron (MLP). Additionally, by employing implicit encoding and\nmulti-camera color MLPs decoding, we achieve separate modeling of scene\nphysical properties and camera characteristics, allowing surround-view\nreconstruction compatible with different camera models. Our method achieves\nremarkable road surface reconstruction performance in a variety of real-world\nchallenging scenarios.\n", "date": "2024-03-18", "categories": []}
{"id": "2310.07449", "title": "PoRF: Pose Residual Field for Accurate Neural Surface Reconstruction", "abstract": "  Neural surface reconstruction is sensitive to the camera pose noise, even if\nstate-of-the-art pose estimators like COLMAP or ARKit are used. More\nimportantly, existing Pose-NeRF joint optimisation methods have struggled to\nimprove pose accuracy in challenging real-world scenarios. To overcome the\nchallenges, we introduce the pose residual field (PoRF), a novel implicit\nrepresentation that uses an MLP for regressing pose updates. This is more\nrobust than the conventional pose parameter optimisation due to parameter\nsharing that leverages global information over the entire sequence.\nFurthermore, we propose an epipolar geometry loss to enhance the supervision\nthat leverages the correspondences exported from COLMAP results without the\nextra computational overhead. Our method yields promising results. On the DTU\ndataset, we reduce the rotation error by 78\\% for COLMAP poses, leading to the\ndecreased reconstruction Chamfer distance from 3.48mm to 0.85mm. On the\nMobileBrick dataset that contains casually captured unbounded 360-degree\nvideos, our method refines ARKit poses and improves the reconstruction F1 score\nfrom 69.18 to 75.67, outperforming that with the dataset provided ground-truth\npose (75.14). These achievements demonstrate the efficacy of our approach in\nrefining camera poses and improving the accuracy of neural surface\nreconstruction in real-world scenarios.\n", "date": "2023-10-11", "categories": []}
{"id": "2403.08768", "title": "3DFIRES: Few Image 3D REconstruction for Scenes with Hidden Surface", "abstract": "  This paper introduces 3DFIRES, a novel system for scene-level 3D\nreconstruction from posed images. Designed to work with as few as one view,\n3DFIRES reconstructs the complete geometry of unseen scenes, including hidden\nsurfaces. With multiple view inputs, our method produces full reconstruction\nwithin all camera frustums. A key feature of our approach is the fusion of\nmulti-view information at the feature level, enabling the production of\ncoherent and comprehensive 3D reconstruction. We train our system on\nnon-watertight scans from large-scale real scene dataset. We show it matches\nthe efficacy of single-view reconstruction methods with only one input and\nsurpasses existing techniques in both quantitative and qualitative measures for\nsparse-view 3D reconstruction.\n", "date": "2024-03-13", "categories": []}
{"id": "2311.11845", "title": "Entangled View-Epipolar Information Aggregation for Generalizable Neural\n  Radiance Fields", "abstract": "  Generalizable NeRF can directly synthesize novel views across new scenes,\neliminating the need for scene-specific retraining in vanilla NeRF. A critical\nenabling factor in these approaches is the extraction of a generalizable 3D\nrepresentation by aggregating source-view features. In this paper, we propose\nan Entangled View-Epipolar Information Aggregation method dubbed EVE-NeRF.\nDifferent from existing methods that consider cross-view and along-epipolar\ninformation independently, EVE-NeRF conducts the view-epipolar feature\naggregation in an entangled manner by injecting the scene-invariant appearance\ncontinuity and geometry consistency priors to the aggregation process. Our\napproach effectively mitigates the potential lack of inherent geometric and\nappearance constraint resulting from one-dimensional interactions, thus further\nboosting the 3D representation generalizablity. EVE-NeRF attains\nstate-of-the-art performance across various evaluation scenarios. Extensive\nexperiments demonstate that, compared to prevailing single-dimensional\naggregation, the entangled network excels in the accuracy of 3D scene geometry\nand appearance reconstruction. Our code is publicly available at\nhttps://github.com/tatakai1/EVENeRF.\n", "date": "2023-11-20", "categories": []}
{"id": "2311.04400", "title": "LRM: Large Reconstruction Model for Single Image to 3D", "abstract": "  We propose the first Large Reconstruction Model (LRM) that predicts the 3D\nmodel of an object from a single input image within just 5 seconds. In contrast\nto many previous methods that are trained on small-scale datasets such as\nShapeNet in a category-specific fashion, LRM adopts a highly scalable\ntransformer-based architecture with 500 million learnable parameters to\ndirectly predict a neural radiance field (NeRF) from the input image. We train\nour model in an end-to-end manner on massive multi-view data containing around\n1 million objects, including both synthetic renderings from Objaverse and real\ncaptures from MVImgNet. This combination of a high-capacity model and\nlarge-scale training data empowers our model to be highly generalizable and\nproduce high-quality 3D reconstructions from various testing inputs, including\nreal-world in-the-wild captures and images created by generative models. Video\ndemos and interactable 3D meshes can be found on our LRM project webpage:\nhttps://yiconghong.me/LRM.\n", "date": "2023-11-08", "categories": []}
{"id": "2403.05907", "title": "Lightning NeRF: Efficient Hybrid Scene Representation for Autonomous\n  Driving", "abstract": "  Recent studies have highlighted the promising application of NeRF in\nautonomous driving contexts. However, the complexity of outdoor environments,\ncombined with the restricted viewpoints in driving scenarios, complicates the\ntask of precisely reconstructing scene geometry. Such challenges often lead to\ndiminished quality in reconstructions and extended durations for both training\nand rendering. To tackle these challenges, we present Lightning NeRF. It uses\nan efficient hybrid scene representation that effectively utilizes the geometry\nprior from LiDAR in autonomous driving scenarios. Lightning NeRF significantly\nimproves the novel view synthesis performance of NeRF and reduces computational\noverheads. Through evaluations on real-world datasets, such as KITTI-360,\nArgoverse2, and our private dataset, we demonstrate that our approach not only\nexceeds the current state-of-the-art in novel view synthesis quality but also\nachieves a five-fold increase in training speed and a ten-fold improvement in\nrendering speed. Codes are available at\nhttps://github.com/VISION-SJTU/Lightning-NeRF .\n", "date": "2024-03-09", "categories": []}
{"id": "2402.08138", "title": "H2O-SDF: Two-phase Learning for 3D Indoor Reconstruction using Object\n  Surface Fields", "abstract": "  Advanced techniques using Neural Radiance Fields (NeRF), Signed Distance\nFields (SDF), and Occupancy Fields have recently emerged as solutions for 3D\nindoor scene reconstruction. We introduce a novel two-phase learning approach,\nH2O-SDF, that discriminates between object and non-object regions within indoor\nenvironments. This method achieves a nuanced balance, carefully preserving the\ngeometric integrity of room layouts while also capturing intricate surface\ndetails of specific objects. A cornerstone of our two-phase learning framework\nis the introduction of the Object Surface Field (OSF), a novel concept designed\nto mitigate the persistent vanishing gradient problem that has previously\nhindered the capture of high-frequency details in other methods. Our proposed\napproach is validated through several experiments that include ablation\nstudies.\n", "date": "2024-02-13", "categories": []}
{"id": "2312.02725", "title": "R3D-SWIN:Use Shifted Window Attention for Single-View 3D Reconstruction", "abstract": "  Recently, vision transformers have performed well in various computer vision\ntasks, including voxel 3D reconstruction. However, the windows of the vision\ntransformer are not multi-scale, and there is no connection between the\nwindows, which limits the accuracy of voxel 3D reconstruction. Therefore, we\npropose a voxel 3D reconstruction network based on shifted window attention. To\nthe best of our knowledge, this is the first work to apply shifted window\nattention to voxel 3D reconstruction. Experimental results on ShapeNet verify\nour method achieves SOTA accuracy in single-view reconstruction.\n", "date": "2023-12-05", "categories": []}
{"id": "2306.09348", "title": "Seeing the World through Your Eyes", "abstract": "  The reflective nature of the human eye is an underappreciated source of\ninformation about what the world around us looks like. By imaging the eyes of a\nmoving person, we can collect multiple views of a scene outside the camera's\ndirect line of sight through the reflections in the eyes. In this paper, we\nreconstruct a 3D scene beyond the camera's line of sight using portrait images\ncontaining eye reflections. This task is challenging due to 1) the difficulty\nof accurately estimating eye poses and 2) the entangled appearance of the eye\niris and the scene reflections. Our method jointly refines the cornea poses,\nthe radiance field depicting the scene, and the observer's eye iris texture. We\nfurther propose a simple regularization prior on the iris texture pattern to\nimprove reconstruction quality. Through various experiments on synthetic and\nreal-world captures featuring people with varied eye colors, we demonstrate the\nfeasibility of our approach to recover 3D scenes using eye reflections.\n", "date": "2023-06-15", "categories": []}
{"id": "2310.09463", "title": "HIO-SDF: Hierarchical Incremental Online Signed Distance Fields", "abstract": "  A good representation of a large, complex mobile robot workspace must be\nspace-efficient yet capable of encoding relevant geometric details. When\nexploring unknown environments, it needs to be updatable incrementally in an\nonline fashion. We introduce HIO-SDF, a new method that represents the\nenvironment as a Signed Distance Field (SDF). State of the art representations\nof SDFs are based on either neural networks or voxel grids. Neural networks are\ncapable of representing the SDF continuously. However, they are hard to update\nincrementally as neural networks tend to forget previously observed parts of\nthe environment unless an extensive sensor history is stored for training.\nVoxel-based representations do not have this problem but they are not\nspace-efficient especially in large environments with fine details. HIO-SDF\ncombines the advantages of these representations using a hierarchical approach\nwhich employs a coarse voxel grid that captures the observed parts of the\nenvironment together with high-resolution local information to train a neural\nnetwork. HIO-SDF achieves a 46% lower mean global SDF error across all test\nscenes than a state of the art continuous representation, and a 30% lower error\nthan a discrete representation at the same resolution as our coarse global SDF\ngrid. Videos and code are available at:\nhttps://samsunglabs.github.io/HIO-SDF-project-page/\n", "date": "2023-10-14", "categories": []}
{"id": "2402.18287", "title": "Windowed-FourierMixer: Enhancing Clutter-Free Room Modeling with Fourier\n  Transform", "abstract": "  With the growing demand for immersive digital applications, the need to\nunderstand and reconstruct 3D scenes has significantly increased. In this\ncontext, inpainting indoor environments from a single image plays a crucial\nrole in modeling the internal structure of interior spaces as it enables the\ncreation of textured and clutter-free reconstructions. While recent methods\nhave shown significant progress in room modeling, they rely on constraining\nlayout estimators to guide the reconstruction process. These methods are highly\ndependent on the performance of the structure estimator and its generative\nability in heavily occluded environments. In response to these issues, we\npropose an innovative approach based on a U-Former architecture and a new\nWindowed-FourierMixer block, resulting in a unified, single-phase network\ncapable of effectively handle human-made periodic structures such as indoor\nspaces. This new architecture proves advantageous for tasks involving indoor\nscenes where symmetry is prevalent, allowing the model to effectively capture\nfeatures such as horizon/ceiling height lines and cuboid-shaped rooms.\nExperiments show the proposed approach outperforms current state-of-the-art\nmethods on the Structured3D dataset demonstrating superior performance in both\nquantitative metrics and qualitative results. Code and models will be made\npublicly available.\n", "date": "2024-02-28", "categories": []}
{"id": "2402.16936", "title": "Disentangled 3D Scene Generation with Layout Learning", "abstract": "  We introduce a method to generate 3D scenes that are disentangled into their\ncomponent objects. This disentanglement is unsupervised, relying only on the\nknowledge of a large pretrained text-to-image model. Our key insight is that\nobjects can be discovered by finding parts of a 3D scene that, when rearranged\nspatially, still produce valid configurations of the same scene. Concretely,\nour method jointly optimizes multiple NeRFs from scratch - each representing\nits own object - along with a set of layouts that composite these objects into\nscenes. We then encourage these composited scenes to be in-distribution\naccording to the image generator. We show that despite its simplicity, our\napproach successfully generates 3D scenes decomposed into individual objects,\nenabling new capabilities in text-to-3D content creation. For results and an\ninteractive demo, see our project page at https://dave.ml/layoutlearning/\n", "date": "2024-02-26", "categories": []}
{"id": "2402.17427", "title": "VastGaussian: Vast 3D Gaussians for Large Scene Reconstruction", "abstract": "  Existing NeRF-based methods for large scene reconstruction often have\nlimitations in visual quality and rendering speed. While the recent 3D Gaussian\nSplatting works well on small-scale and object-centric scenes, scaling it up to\nlarge scenes poses challenges due to limited video memory, long optimization\ntime, and noticeable appearance variations. To address these challenges, we\npresent VastGaussian, the first method for high-quality reconstruction and\nreal-time rendering on large scenes based on 3D Gaussian Splatting. We propose\na progressive partitioning strategy to divide a large scene into multiple\ncells, where the training cameras and point cloud are properly distributed with\nan airspace-aware visibility criterion. These cells are merged into a complete\nscene after parallel optimization. We also introduce decoupled appearance\nmodeling into the optimization process to reduce appearance variations in the\nrendered images. Our approach outperforms existing NeRF-based methods and\nachieves state-of-the-art results on multiple large scene datasets, enabling\nfast optimization and high-fidelity real-time rendering.\n", "date": "2024-02-27", "categories": []}
{"id": "2402.17483", "title": "AlignMiF: Geometry-Aligned Multimodal Implicit Field for LiDAR-Camera\n  Joint Synthesis", "abstract": "  Neural implicit fields have been a de facto standard in novel view synthesis.\nRecently, there exist some methods exploring fusing multiple modalities within\na single field, aiming to share implicit features from different modalities to\nenhance reconstruction performance. However, these modalities often exhibit\nmisaligned behaviors: optimizing for one modality, such as LiDAR, can adversely\naffect another, like camera performance, and vice versa. In this work, we\nconduct comprehensive analyses on the multimodal implicit field of LiDAR-camera\njoint synthesis, revealing the underlying issue lies in the misalignment of\ndifferent sensors. Furthermore, we introduce AlignMiF, a geometrically aligned\nmultimodal implicit field with two proposed modules: Geometry-Aware Alignment\n(GAA) and Shared Geometry Initialization (SGI). These modules effectively align\nthe coarse geometry across different modalities, significantly enhancing the\nfusion process between LiDAR and camera data. Through extensive experiments\nacross various datasets and scenes, we demonstrate the effectiveness of our\napproach in facilitating better interaction between LiDAR and camera modalities\nwithin a unified neural field. Specifically, our proposed AlignMiF, achieves\nremarkable improvement over recent implicit fusion methods (+2.01 and +3.11\nimage PSNR on the KITTI-360 and Waymo datasets) and consistently surpasses\nsingle modality performance (13.8% and 14.2% reduction in LiDAR Chamfer\nDistance on the respective datasets).\n", "date": "2024-02-27", "categories": []}
{"id": "2402.11431", "title": "A Robust Error-Resistant View Selection Method for 3D Reconstruction", "abstract": "  To address the issue of increased triangulation uncertainty caused by\nselecting views with small camera baselines in Structure from Motion (SFM) view\nselection, this paper proposes a robust error-resistant view selection method.\nThe method utilizes a triangulation-based computation to obtain an\nerror-resistant model, which is then used to construct an error-resistant\nmatrix. The sorting results of each row in the error-resistant matrix determine\nthe candidate view set for each view. By traversing the candidate view sets of\nall views and completing the missing views based on the error-resistant matrix,\nthe integrity of 3D reconstruction is ensured. Experimental comparisons between\nthis method and the exhaustive method with the highest accuracy in the COLMAP\nprogram are conducted in terms of average reprojection error and absolute\ntrajectory error in the reconstruction results. The proposed method\ndemonstrates an average reduction of 29.40% in reprojection error accuracy and\n5.07% in absolute trajectory error on the TUM dataset and DTU dataset.\n", "date": "2024-02-18", "categories": []}
{"id": "2402.16739", "title": "Neural Mesh Fusion: Unsupervised 3D Planar Surface Understanding", "abstract": "  This paper presents Neural Mesh Fusion (NMF), an efficient approach for joint\noptimization of polygon mesh from multi-view image observations and\nunsupervised 3D planar-surface parsing of the scene. In contrast to implicit\nneural representations, NMF directly learns to deform surface triangle mesh and\ngenerate an embedding for unsupervised 3D planar segmentation through\ngradient-based optimization directly on the surface mesh. The conducted\nexperiments show that NMF obtains competitive results compared to\nstate-of-the-art multi-view planar reconstruction, while not requiring any\nground-truth 3D or planar supervision. Moreover, NMF is significantly more\ncomputationally efficient compared to implicit neural rendering-based scene\nreconstruction approaches.\n", "date": "2024-02-26", "categories": []}
{"id": "2310.10642", "title": "Real-time Photorealistic Dynamic Scene Representation and Rendering with\n  4D Gaussian Splatting", "abstract": "  Reconstructing dynamic 3D scenes from 2D images and generating diverse views\nover time is challenging due to scene complexity and temporal dynamics. Despite\nadvancements in neural implicit models, limitations persist: (i) Inadequate\nScene Structure: Existing methods struggle to reveal the spatial and temporal\nstructure of dynamic scenes from directly learning the complex 6D plenoptic\nfunction. (ii) Scaling Deformation Modeling: Explicitly modeling scene element\ndeformation becomes impractical for complex dynamics. To address these issues,\nwe consider the spacetime as an entirety and propose to approximate the\nunderlying spatio-temporal 4D volume of a dynamic scene by optimizing a\ncollection of 4D primitives, with explicit geometry and appearance modeling.\nLearning to optimize the 4D primitives enables us to synthesize novel views at\nany desired time with our tailored rendering routine. Our model is conceptually\nsimple, consisting of a 4D Gaussian parameterized by anisotropic ellipses that\ncan rotate arbitrarily in space and time, as well as view-dependent and\ntime-evolved appearance represented by the coefficient of 4D spherindrical\nharmonics. This approach offers simplicity, flexibility for variable-length\nvideo and end-to-end training, and efficient real-time rendering, making it\nsuitable for capturing complex dynamic scene motions. Experiments across\nvarious benchmarks, including monocular and multi-view scenarios, demonstrate\nour 4DGS model's superior visual quality and efficiency.\n", "date": "2023-10-16", "categories": []}
{"id": "2402.14650", "title": "GaussianPro: 3D Gaussian Splatting with Progressive Propagation", "abstract": "  The advent of 3D Gaussian Splatting (3DGS) has recently brought about a\nrevolution in the field of neural rendering, facilitating high-quality\nrenderings at real-time speed. However, 3DGS heavily depends on the initialized\npoint cloud produced by Structure-from-Motion (SfM) techniques. When tackling\nwith large-scale scenes that unavoidably contain texture-less surfaces, the SfM\ntechniques always fail to produce enough points in these surfaces and cannot\nprovide good initialization for 3DGS. As a result, 3DGS suffers from difficult\noptimization and low-quality renderings. In this paper, inspired by classical\nmulti-view stereo (MVS) techniques, we propose GaussianPro, a novel method that\napplies a progressive propagation strategy to guide the densification of the 3D\nGaussians. Compared to the simple split and clone strategies used in 3DGS, our\nmethod leverages the priors of the existing reconstructed geometries of the\nscene and patch matching techniques to produce new Gaussians with accurate\npositions and orientations. Experiments on both large-scale and small-scale\nscenes validate the effectiveness of our method, where our method significantly\nsurpasses 3DGS on the Waymo dataset, exhibiting an improvement of 1.15dB in\nterms of PSNR.\n", "date": "2024-02-22", "categories": []}
{"id": "2402.03445", "title": "Denoising Diffusion via Image-Based Rendering", "abstract": "  Generating 3D scenes is a challenging open problem, which requires\nsynthesizing plausible content that is fully consistent in 3D space. While\nrecent methods such as neural radiance fields excel at view synthesis and 3D\nreconstruction, they cannot synthesize plausible details in unobserved regions\nsince they lack a generative capability. Conversely, existing generative\nmethods are typically not capable of reconstructing detailed, large-scale\nscenes in the wild, as they use limited-capacity 3D scene representations,\nrequire aligned camera poses, or rely on additional regularizers. In this work,\nwe introduce the first diffusion model able to perform fast, detailed\nreconstruction and generation of real-world 3D scenes. To achieve this, we make\nthree contributions. First, we introduce a new neural scene representation,\nIB-planes, that can efficiently and accurately represent large 3D scenes,\ndynamically allocating more capacity as needed to capture details visible in\neach image. Second, we propose a denoising-diffusion framework to learn a prior\nover this novel 3D scene representation, using only 2D images without the need\nfor any additional supervision signal such as masks or depths. This supports 3D\nreconstruction and generation in a unified architecture. Third, we develop a\nprincipled approach to avoid trivial 3D solutions when integrating the\nimage-based rendering with the diffusion model, by dropping out representations\nof some images. We evaluate the model on several challenging datasets of real\nand synthetic images, and demonstrate superior results on generation, novel\nview synthesis and 3D reconstruction.\n", "date": "2024-02-05", "categories": []}
{"id": "2311.13681", "title": "Compact 3D Gaussian Representation for Radiance Field", "abstract": "  Neural Radiance Fields (NeRFs) have demonstrated remarkable potential in\ncapturing complex 3D scenes with high fidelity. However, one persistent\nchallenge that hinders the widespread adoption of NeRFs is the computational\nbottleneck due to the volumetric rendering. On the other hand, 3D Gaussian\nsplatting (3DGS) has recently emerged as an alternative representation that\nleverages a 3D Gaussisan-based representation and adopts the rasterization\npipeline to render the images rather than volumetric rendering, achieving very\nfast rendering speed and promising image quality. However, a significant\ndrawback arises as 3DGS entails a substantial number of 3D Gaussians to\nmaintain the high fidelity of the rendered images, which requires a large\namount of memory and storage. To address this critical issue, we place a\nspecific emphasis on two key objectives: reducing the number of Gaussian points\nwithout sacrificing performance and compressing the Gaussian attributes, such\nas view-dependent color and covariance. To this end, we propose a learnable\nmask strategy that significantly reduces the number of Gaussians while\npreserving high performance. In addition, we propose a compact but effective\nrepresentation of view-dependent color by employing a grid-based neural field\nrather than relying on spherical harmonics. Finally, we learn codebooks to\ncompactly represent the geometric attributes of Gaussian by vector\nquantization. With model compression techniques such as quantization and\nentropy coding, we consistently show over 25$\\times$ reduced storage and\nenhanced rendering speed, while maintaining the quality of the scene\nrepresentation, compared to 3DGS. Our work provides a comprehensive framework\nfor 3D scene representation, achieving high performance, fast training,\ncompactness, and real-time rendering. Our project page is available at\nhttps://maincold2.github.io/c3dgs/.\n", "date": "2023-11-22", "categories": []}
{"id": "2402.09325", "title": "PC-NeRF: Parent-Child Neural Radiance Fields Using Sparse LiDAR Frames\n  in Autonomous Driving Environments", "abstract": "  Large-scale 3D scene reconstruction and novel view synthesis are vital for\nautonomous vehicles, especially utilizing temporally sparse LiDAR frames.\nHowever, conventional explicit representations remain a significant bottleneck\ntowards representing the reconstructed and synthetic scenes at unlimited\nresolution. Although the recently developed neural radiance fields (NeRF) have\nshown compelling results in implicit representations, the problem of\nlarge-scale 3D scene reconstruction and novel view synthesis using sparse LiDAR\nframes remains unexplored. To bridge this gap, we propose a 3D scene\nreconstruction and novel view synthesis framework called parent-child neural\nradiance field (PC-NeRF). Based on its two modules, parent NeRF and child NeRF,\nthe framework implements hierarchical spatial partitioning and multi-level\nscene representation, including scene, segment, and point levels. The\nmulti-level scene representation enhances the efficient utilization of sparse\nLiDAR point cloud data and enables the rapid acquisition of an approximate\nvolumetric scene representation. With extensive experiments, PC-NeRF is proven\nto achieve high-precision novel LiDAR view synthesis and 3D reconstruction in\nlarge-scale scenes. Moreover, PC-NeRF can effectively handle situations with\nsparse LiDAR frames and demonstrate high deployment efficiency with limited\ntraining epochs. Our approach implementation and the pre-trained models are\navailable at https://github.com/biter0088/pc-nerf.\n", "date": "2024-02-14", "categories": []}
{"id": "2211.14310", "title": "Efficient 3D Reconstruction, Streaming and Visualization of Static and\n  Dynamic Scene Parts for Multi-client Live-telepresence in Large-scale\n  Environments", "abstract": "  Despite the impressive progress of telepresence systems for room-scale scenes\nwith static and dynamic scene entities, expanding their capabilities to\nscenarios with larger dynamic environments beyond a fixed size of a few\nsquare-meters remains challenging.\n  In this paper, we aim at sharing 3D live-telepresence experiences in\nlarge-scale environments beyond room scale with both static and dynamic scene\nentities at practical bandwidth requirements only based on light-weight scene\ncapture with a single moving consumer-grade RGB-D camera. To this end, we\npresent a system which is built upon a novel hybrid volumetric scene\nrepresentation in terms of the combination of a voxel-based scene\nrepresentation for the static contents, that not only stores the reconstructed\nsurface geometry but also contains information about the object semantics as\nwell as their accumulated dynamic movement over time, and a point-cloud-based\nrepresentation for dynamic scene parts, where the respective separation from\nstatic parts is achieved based on semantic and instance information extracted\nfor the input frames. With an independent yet simultaneous streaming of both\nstatic and dynamic content, where we seamlessly integrate potentially moving\nbut currently static scene entities in the static model until they are becoming\ndynamic again, as well as the fusion of static and dynamic data at the remote\nclient, our system is able to achieve VR-based live-telepresence at close to\nreal-time rates. Our evaluation demonstrates the potential of our novel\napproach in terms of visual quality, performance, and ablation studies\nregarding involved design choices.\n", "date": "2022-11-25", "categories": []}
{"id": "2402.04554", "title": "BirdNeRF: Fast Neural Reconstruction of Large-Scale Scenes From Aerial\n  Imagery", "abstract": "  In this study, we introduce BirdNeRF, an adaptation of Neural Radiance Fields\n(NeRF) designed specifically for reconstructing large-scale scenes using aerial\nimagery. Unlike previous research focused on small-scale and object-centric\nNeRF reconstruction, our approach addresses multiple challenges, including (1)\nAddressing the issue of slow training and rendering associated with large\nmodels. (2) Meeting the computational demands necessitated by modeling a\nsubstantial number of images, requiring extensive resources such as\nhigh-performance GPUs. (3) Overcoming significant artifacts and low visual\nfidelity commonly observed in large-scale reconstruction tasks due to limited\nmodel capacity. Specifically, we present a novel bird-view pose-based spatial\ndecomposition algorithm that decomposes a large aerial image set into multiple\nsmall sets with appropriately sized overlaps, allowing us to train individual\nNeRFs of sub-scene. This decomposition approach not only decouples rendering\ntime from the scene size but also enables rendering to scale seamlessly to\narbitrarily large environments. Moreover, it allows for per-block updates of\nthe environment, enhancing the flexibility and adaptability of the\nreconstruction process. Additionally, we propose a projection-guided novel view\nre-rendering strategy, which aids in effectively utilizing the independently\ntrained sub-scenes to generate superior rendering results. We evaluate our\napproach on existing datasets as well as against our own drone footage,\nimproving reconstruction speed by 10x over classical photogrammetry software\nand 50x over state-of-the-art large-scale NeRF solution, on a single GPU with\nsimilar rendering quality.\n", "date": "2024-02-07", "categories": []}
{"id": "2402.06826", "title": "Neural Rendering based Urban Scene Reconstruction for Autonomous Driving", "abstract": "  Dense 3D reconstruction has many applications in automated driving including\nautomated annotation validation, multimodal data augmentation, providing ground\ntruth annotations for systems lacking LiDAR, as well as enhancing auto-labeling\naccuracy. LiDAR provides highly accurate but sparse depth, whereas camera\nimages enable estimation of dense depth but noisy particularly at long ranges.\nIn this paper, we harness the strengths of both sensors and propose a\nmultimodal 3D scene reconstruction using a framework combining neural implicit\nsurfaces and radiance fields. In particular, our method estimates dense and\naccurate 3D structures and creates an implicit map representation based on\nsigned distance fields, which can be further rendered into RGB images, and\ndepth maps. A mesh can be extracted from the learned signed distance field and\nculled based on occlusion. Dynamic objects are efficiently filtered on the fly\nduring sampling using 3D object detection models. We demonstrate qualitative\nand quantitative results on challenging automotive scenes.\n", "date": "2024-02-09", "categories": []}
{"id": "2402.07301", "title": "LISR: Learning Linear 3D Implicit Surface Representation Using Compactly\n  Supported Radial Basis Functions", "abstract": "  Implicit 3D surface reconstruction of an object from its partial and noisy 3D\npoint cloud scan is the classical geometry processing and 3D computer vision\nproblem. In the literature, various 3D shape representations have been\ndeveloped, differing in memory efficiency and shape retrieval effectiveness,\nsuch as volumetric, parametric, and implicit surfaces. Radial basis functions\nprovide memory-efficient parameterization of the implicit surface. However, we\nshow that training a neural network using the mean squared error between the\nground-truth implicit surface and the linear basis-based implicit surfaces does\nnot converge to the global solution. In this work, we propose locally supported\ncompact radial basis functions for a linear representation of the implicit\nsurface. This representation enables us to generate 3D shapes with arbitrary\ntopologies at any resolution due to their continuous nature. We then propose a\nneural network architecture for learning the linear implicit shape\nrepresentation of the 3D surface of an object. We learn linear implicit shapes\nwithin a supervised learning framework using ground truth Signed-Distance Field\n(SDF) data for guidance. The classical strategies face difficulties in finding\nlinear implicit shapes from a given 3D point cloud due to numerical issues\n(requires solving inverse of a large matrix) in basis and query point\nselection. The proposed approach achieves better Chamfer distance and\ncomparable F-score than the state-of-the-art approach on the benchmark dataset.\nWe also show the effectiveness of the proposed approach by using it for the 3D\nshape completion task.\n", "date": "2024-02-11", "categories": []}
{"id": "2402.06288", "title": "MLS2LoD3: Refining low LoDs building models with MLS point clouds to\n  reconstruct semantic LoD3 building models", "abstract": "  Although highly-detailed LoD3 building models reveal great potential in\nvarious applications, they have yet to be available. The primary challenges in\ncreating such models concern not only automatic detection and reconstruction\nbut also standard-consistent modeling. In this paper, we introduce a novel\nrefinement strategy enabling LoD3 reconstruction by leveraging the ubiquity of\nlower LoD building models and the accuracy of MLS point clouds. Such a strategy\npromises at-scale LoD3 reconstruction and unlocks LoD3 applications, which we\nalso describe and illustrate in this paper. Additionally, we present guidelines\nfor reconstructing LoD3 facade elements and their embedding into the CityGML\nstandard model, disseminating gained knowledge to academics and professionals.\nWe believe that our method can foster development of LoD3 reconstruction\nalgorithms and subsequently enable their wider adoption.\n", "date": "2024-02-09", "categories": []}
{"id": "2402.06521", "title": "Reconstructing facade details using MLS point clouds and Bag-of-Words\n  approach", "abstract": "  In the reconstruction of fa\\c{c}ade elements, the identification of specific\nobject types remains challenging and is often circumvented by rectangularity\nassumptions or the use of bounding boxes. We propose a new approach for the\nreconstruction of 3D fa\\c{c}ade details. We combine MLS point clouds and a\npre-defined 3D model library using a BoW concept, which we augment by\nincorporating semi-global features. We conduct experiments on the models\nsuperimposed with random noise and on the TUM-FA\\c{C}ADE dataset. Our method\ndemonstrates promising results, improving the conventional BoW approach. It\nholds the potential to be utilized for more realistic facade reconstruction\nwithout rectangularity assumptions, which can be used in applications such as\ntesting automated driving functions or estimating fa\\c{c}ade solar potential.\n", "date": "2024-02-09", "categories": []}
{"id": "2401.08518", "title": "PPSURF: Combining Patches and Point Convolutions for Detailed Surface\n  Reconstruction", "abstract": "  3D surface reconstruction from point clouds is a key step in areas such as\ncontent creation, archaeology, digital cultural heritage, and engineering.\nCurrent approaches either try to optimize a non-data-driven surface\nrepresentation to fit the points, or learn a data-driven prior over the\ndistribution of commonly occurring surfaces and how they correlate with\npotentially noisy point clouds. Data-driven methods enable robust handling of\nnoise and typically either focus on a global or a local prior, which trade-off\nbetween robustness to noise on the global end and surface detail preservation\non the local end. We propose PPSurf as a method that combines a global prior\nbased on point convolutions and a local prior based on processing local point\ncloud patches. We show that this approach is robust to noise while recovering\nsurface details more accurately than the current state-of-the-art.\n  Our source code, pre-trained model and dataset are available at:\nhttps://github.com/cg-tuwien/ppsurf\n", "date": "2024-01-16", "categories": []}
{"id": "2308.11774", "title": "SAMSNeRF: Segment Anything Model (SAM) Guides Dynamic Surgical Scene\n  Reconstruction by Neural Radiance Field (NeRF)", "abstract": "  The accurate reconstruction of surgical scenes from surgical videos is\ncritical for various applications, including intraoperative navigation and\nimage-guided robotic surgery automation. However, previous approaches, mainly\nrelying on depth estimation, have limited effectiveness in reconstructing\nsurgical scenes with moving surgical tools. To address this limitation and\nprovide accurate 3D position prediction for surgical tools in all frames, we\npropose a novel approach called SAMSNeRF that combines Segment Anything Model\n(SAM) and Neural Radiance Field (NeRF) techniques. Our approach generates\naccurate segmentation masks of surgical tools using SAM, which guides the\nrefinement of the dynamic surgical scene reconstruction by NeRF. Our\nexperimental results on public endoscopy surgical videos demonstrate that our\napproach successfully reconstructs high-fidelity dynamic surgical scenes and\naccurately reflects the spatial information of surgical tools. Our proposed\napproach can significantly enhance surgical navigation and automation by\nproviding surgeons with accurate 3D position information of surgical tools\nduring surgery.The source code will be released soon.\n", "date": "2023-08-22", "categories": []}
{"id": "2303.12678", "title": "Uni-Fusion: Universal Continuous Mapping", "abstract": "  We present Uni-Fusion, a universal continuous mapping framework for surfaces,\nsurface properties (color, infrared, etc.) and more (latent features in CLIP\nembedding space, etc.). We propose the first universal implicit encoding model\nthat supports encoding of both geometry and different types of properties (RGB,\ninfrared, features, etc.) without requiring any training. Based on this, our\nframework divides the point cloud into regular grid voxels and generates a\nlatent feature in each voxel to form a Latent Implicit Map (LIM) for geometries\nand arbitrary properties. Then, by fusing a local LIM frame-wisely into a\nglobal LIM, an incremental reconstruction is achieved. Encoded with\ncorresponding types of data, our Latent Implicit Map is capable of generating\ncontinuous surfaces, surface property fields, surface feature fields, and all\nother possible options. To demonstrate the capabilities of our model, we\nimplement three applications: (1) incremental reconstruction for surfaces and\ncolor (2) 2D-to-3D transfer of fabricated properties (3) open-vocabulary scene\nunderstanding by creating a text CLIP feature field on surfaces. We evaluate\nUni-Fusion by comparing it in corresponding applications, from which Uni-Fusion\nshows high-flexibility in various applications while performing best or being\ncompetitive. The project page of Uni-Fusion is available at\nhttps://jarrome.github.io/Uni-Fusion/ .\n", "date": "2023-03-22", "categories": []}
{"id": "2402.00740", "title": "DRSM: efficient neural 4d decomposition for dynamic reconstruction in\n  stationary monocular cameras", "abstract": "  With the popularity of monocular videos generated by video sharing and live\nbroadcasting applications, reconstructing and editing dynamic scenes in\nstationary monocular cameras has become a special but anticipated technology.\nIn contrast to scene reconstructions that exploit multi-view observations, the\nproblem of modeling a dynamic scene from a single view is significantly more\nunder-constrained and ill-posed. Inspired by recent progress in neural\nrendering, we present a novel framework to tackle 4D decomposition problem for\ndynamic scenes in monocular cameras. Our framework utilizes decomposed static\nand dynamic feature planes to represent 4D scenes and emphasizes the learning\nof dynamic regions through dense ray casting. Inadequate 3D clues from a\nsingle-view and occlusion are also particular challenges in scene\nreconstruction. To overcome these difficulties, we propose deep supervised\noptimization and ray casting strategies. With experiments on various videos,\nour method generates higher-fidelity results than existing methods for\nsingle-view dynamic scene representation.\n", "date": "2024-02-01", "categories": []}
{"id": "2402.00763", "title": "360-GS: Layout-guided Panoramic Gaussian Splatting For Indoor Roaming", "abstract": "  3D Gaussian Splatting (3D-GS) has recently attracted great attention with\nreal-time and photo-realistic renderings. This technique typically takes\nperspective images as input and optimizes a set of 3D elliptical Gaussians by\nsplatting them onto the image planes, resulting in 2D Gaussians. However,\napplying 3D-GS to panoramic inputs presents challenges in effectively modeling\nthe projection onto the spherical surface of ${360^\\circ}$ images using 2D\nGaussians. In practical applications, input panoramas are often sparse, leading\nto unreliable initialization of 3D Gaussians and subsequent degradation of\n3D-GS quality. In addition, due to the under-constrained geometry of\ntexture-less planes (e.g., walls and floors), 3D-GS struggles to model these\nflat regions with elliptical Gaussians, resulting in significant floaters in\nnovel views. To address these issues, we propose 360-GS, a novel $360^{\\circ}$\nGaussian splatting for a limited set of panoramic inputs. Instead of splatting\n3D Gaussians directly onto the spherical surface, 360-GS projects them onto the\ntangent plane of the unit sphere and then maps them to the spherical\nprojections. This adaptation enables the representation of the projection using\nGaussians. We guide the optimization of 360-GS by exploiting layout priors\nwithin panoramas, which are simple to obtain and contain strong structural\ninformation about the indoor scene. Our experimental results demonstrate that\n360-GS allows panoramic rendering and outperforms state-of-the-art methods with\nfewer artifacts in novel view synthesis, thus providing immersive roaming in\nindoor scenarios.\n", "date": "2024-02-01", "categories": []}
{"id": "2401.17058", "title": "Atlanta Scaled layouts from non-central panoramas", "abstract": "  In this work we present a novel approach for 3D layout recovery of indoor\nenvironments using a non-central acquisition system. From a non-central\npanorama, full and scaled 3D lines can be independently recovered by geometry\nreasoning without geometric nor scale assumptions. However, their sensitivity\nto noise and complex geometric modeling has led these panoramas being little\ninvestigated. Our new pipeline aims to extract the boundaries of the structural\nlines of an indoor environment with a neural network and exploit the properties\nof non-central projection systems in a new geometrical processing to recover an\nscaled 3D layout. The results of our experiments show that we improve\nstate-of-the-art methods for layout reconstruction and line extraction in\nnon-central projection systems. We completely solve the problem in Manhattan\nand Atlanta environments, handling occlusions and retrieving the metric scale\nof the room without extra measurements. As far as the authors knowledge goes,\nour approach is the first work using deep learning on non-central panoramas and\nrecovering scaled layouts from single panoramas.\n", "date": "2024-01-30", "categories": []}
{"id": "2401.15841", "title": "2L3: Lifting Imperfect Generated 2D Images into Accurate 3D", "abstract": "  Reconstructing 3D objects from a single image is an intriguing but\nchallenging problem. One promising solution is to utilize multi-view (MV) 3D\nreconstruction to fuse generated MV images into consistent 3D objects. However,\nthe generated images usually suffer from inconsistent lighting, misaligned\ngeometry, and sparse views, leading to poor reconstruction quality. To cope\nwith these problems, we present a novel 3D reconstruction framework that\nleverages intrinsic decomposition guidance, transient-mono prior guidance, and\nview augmentation to cope with the three issues, respectively. Specifically, we\nfirst leverage to decouple the shading information from the generated images to\nreduce the impact of inconsistent lighting; then, we introduce mono prior with\nview-dependent transient encoding to enhance the reconstructed normal; and\nfinally, we design a view augmentation fusion strategy that minimizes\npixel-level loss in generated sparse views and semantic loss in augmented\nrandom views, resulting in view-consistent geometry and detailed textures. Our\napproach, therefore, enables the integration of a pre-trained MV image\ngenerator and a neural network-based volumetric signed distance function (SDF)\nrepresentation for a single image to 3D object reconstruction. We evaluate our\nframework on various datasets and demonstrate its superior performance in both\nquantitative and qualitative assessments, signifying a significant advancement\nin 3D object reconstruction. Compared with the latest state-of-the-art method\nSyncdreamer~\\cite{liu2023syncdreamer}, we reduce the Chamfer Distance error by\nabout 36\\% and improve PSNR by about 30\\% .\n", "date": "2024-01-29", "categories": []}
{"id": "2212.04492", "title": "Few-View Object Reconstruction with Unknown Categories and Camera Poses", "abstract": "  While object reconstruction has made great strides in recent years, current\nmethods typically require densely captured images and/or known camera poses,\nand generalize poorly to novel object categories. To step toward object\nreconstruction in the wild, this work explores reconstructing general\nreal-world objects from a few images without known camera poses or object\ncategories. The crux of our work is solving two fundamental 3D vision problems\n-- shape reconstruction and pose estimation -- in a unified approach. Our\napproach captures the synergies of these two problems: reliable camera pose\nestimation gives rise to accurate shape reconstruction, and the accurate\nreconstruction, in turn, induces robust correspondence between different views\nand facilitates pose estimation. Our method FORGE predicts 3D features from\neach view and leverages them in conjunction with the input images to establish\ncross-view correspondence for estimating relative camera poses. The 3D features\nare then transformed by the estimated poses into a shared space and are fused\ninto a neural radiance field. The reconstruction results are rendered by volume\nrendering techniques, enabling us to train the model without 3D shape\nground-truth. Our experiments show that FORGE reliably reconstructs objects\nfrom five views. Our pose estimation method outperforms existing ones by a\nlarge margin. The reconstruction results under predicted poses are comparable\nto the ones using ground-truth poses. The performance on novel testing\ncategories matches the results on categories seen during training. Project\npage: https://ut-austin-rpl.github.io/FORGE/\n", "date": "2022-12-08", "categories": []}
{"id": "2305.00393", "title": "DynaVol: Unsupervised Learning for Dynamic Scenes through Object-Centric\n  Voxelization", "abstract": "  Unsupervised learning of object-centric representations in dynamic visual\nscenes is challenging. Unlike most previous approaches that learn to decompose\n2D images, we present DynaVol, a 3D scene generative model that unifies\ngeometric structures and object-centric learning in a differentiable volume\nrendering framework. The key idea is to perform object-centric voxelization to\ncapture the 3D nature of the scene, which infers the probability distribution\nover objects at individual spatial locations. These voxel features evolve over\ntime through a canonical-space deformation function, forming the basis for\nglobal representation learning via slot attention. The voxel features and\nglobal features are complementary and are both leveraged by a compositional\nNeRF decoder for volume rendering. DynaVol remarkably outperforms existing\napproaches for unsupervised dynamic scene decomposition. Once trained, the\nexplicitly meaningful voxel features enable additional capabilities that 2D\nscene decomposition methods cannot achieve: it is possible to freely edit the\ngeometric shapes or manipulate the motion trajectories of the objects.\n", "date": "2023-04-30", "categories": []}
{"id": "2401.14032", "title": "GauU-Scene: A Scene Reconstruction Benchmark on Large Scale 3D\n  Reconstruction Dataset Using Gaussian Splatting", "abstract": "  We introduce a novel large-scale scene reconstruction benchmark using the\nnewly developed 3D representation approach, Gaussian Splatting, on our\nexpansive U-Scene dataset. U-Scene encompasses over one and a half square\nkilometres, featuring a comprehensive RGB dataset coupled with LiDAR ground\ntruth. For data acquisition, we employed the Matrix 300 drone equipped with the\nhigh-accuracy Zenmuse L1 LiDAR, enabling precise rooftop data collection. This\ndataset, offers a unique blend of urban and academic environments for advanced\nspatial analysis convers more than 1.5 km$^2$. Our evaluation of U-Scene with\nGaussian Splatting includes a detailed analysis across various novel\nviewpoints. We also juxtapose these results with those derived from our\naccurate point cloud dataset, highlighting significant differences that\nunderscore the importance of combine multi-modal information\n", "date": "2024-01-25", "categories": []}
{"id": "2401.14354", "title": "Learning Robust Generalizable Radiance Field with Visibility and Feature\n  Augmented Point Representation", "abstract": "  This paper introduces a novel paradigm for the generalizable neural radiance\nfield (NeRF). Previous generic NeRF methods combine multiview stereo techniques\nwith image-based neural rendering for generalization, yielding impressive\nresults, while suffering from three issues. First, occlusions often result in\ninconsistent feature matching. Then, they deliver distortions and artifacts in\ngeometric discontinuities and locally sharp shapes due to their individual\nprocess of sampled points and rough feature aggregation. Third, their\nimage-based representations experience severe degradations when source views\nare not near enough to the target view. To address challenges, we propose the\nfirst paradigm that constructs the generalizable neural field based on\npoint-based rather than image-based rendering, which we call the Generalizable\nneural Point Field (GPF). Our approach explicitly models visibilities by\ngeometric priors and augments them with neural features. We propose a novel\nnonuniform log sampling strategy to improve both rendering speed and\nreconstruction quality. Moreover, we present a learnable kernel spatially\naugmented with features for feature aggregations, mitigating distortions at\nplaces with drastically varying geometries. Besides, our representation can be\neasily manipulated. Experiments show that our model can deliver better\ngeometries, view consistencies, and rendering quality than all counterparts and\nbenchmarks on three datasets in both generalization and finetuning settings,\npreliminarily proving the potential of the new paradigm for generalizable NeRF.\n", "date": "2024-01-25", "categories": []}
{"id": "2401.11711", "title": "HG3-NeRF: Hierarchical Geometric, Semantic, and Photometric Guided\n  Neural Radiance Fields for Sparse View Inputs", "abstract": "  Neural Radiance Fields (NeRF) have garnered considerable attention as a\nparadigm for novel view synthesis by learning scene representations from\ndiscrete observations. Nevertheless, NeRF exhibit pronounced performance\ndegradation when confronted with sparse view inputs, consequently curtailing\nits further applicability. In this work, we introduce Hierarchical Geometric,\nSemantic, and Photometric Guided NeRF (HG3-NeRF), a novel methodology that can\naddress the aforementioned limitation and enhance consistency of geometry,\nsemantic content, and appearance across different views. We propose\nHierarchical Geometric Guidance (HGG) to incorporate the attachment of\nStructure from Motion (SfM), namely sparse depth prior, into the scene\nrepresentations. Different from direct depth supervision, HGG samples volume\npoints from local-to-global geometric regions, mitigating the misalignment\ncaused by inherent bias in the depth prior. Furthermore, we draw inspiration\nfrom notable variations in semantic consistency observed across images of\ndifferent resolutions and propose Hierarchical Semantic Guidance (HSG) to learn\nthe coarse-to-fine semantic content, which corresponds to the coarse-to-fine\nscene representations. Experimental results demonstrate that HG3-NeRF can\noutperform other state-of-the-art methods on different standard benchmarks and\nachieve high-fidelity synthesis results for sparse view inputs.\n", "date": "2024-01-22", "categories": []}
{"id": "2312.06946", "title": "WaterHE-NeRF: Water-ray Tracing Neural Radiance Fields for Underwater\n  Scene Reconstruction", "abstract": "  Neural Radiance Field (NeRF) technology demonstrates immense potential in\nnovel viewpoint synthesis tasks, due to its physics-based volumetric rendering\nprocess, which is particularly promising in underwater scenes. Addressing the\nlimitations of existing underwater NeRF methods in handling light attenuation\ncaused by the water medium and the lack of real Ground Truth (GT) supervision,\nthis study proposes WaterHE-NeRF. We develop a new water-ray tracing field by\nRetinex theory that precisely encodes color, density, and illuminance\nattenuation in three-dimensional space. WaterHE-NeRF, through its illuminance\nattenuation mechanism, generates both degraded and clear multi-view images and\noptimizes image restoration by combining reconstruction loss with Wasserstein\ndistance. Additionally, the use of histogram equalization (HE) as pseudo-GT\nenhances the network's accuracy in preserving original details and color\ndistribution. Extensive experiments on real underwater datasets and synthetic\ndatasets validate the effectiveness of WaterHE-NeRF. Our code will be made\npublicly available.\n", "date": "2023-12-12", "categories": []}
{"id": "2304.04351", "title": "Evaluate Geometry of Radiance Fields with Low-frequency Color Prior", "abstract": "  A radiance field is an effective representation of 3D scenes, which has been\nwidely adopted in novel-view synthesis and 3D reconstruction. It is still an\nopen and challenging problem to evaluate the geometry, i.e., the density field,\nas the ground-truth is almost impossible to obtain. One alternative indirect\nsolution is to transform the density field into a point-cloud and compute its\nChamfer Distance with the scanned ground-truth. However, many widely-used\ndatasets have no point-cloud ground-truth since the scanning process along with\nthe equipment is expensive and complicated. To this end, we propose a novel\nmetric, named Inverse Mean Residual Color (IMRC), which can evaluate the\ngeometry only with the observation images. Our key insight is that the better\nthe geometry, the lower-frequency the computed color field. From this insight,\ngiven a reconstructed density field and observation images, we design a\nclosed-form method to approximate the color field with low-frequency spherical\nharmonics, and compute the inverse mean residual color. Then the higher the\nIMRC, the better the geometry. Qualitative and quantitative experimental\nresults verify the effectiveness of our proposed IMRC metric. We also benchmark\nseveral state-of-the-art methods using IMRC to promote future related research.\nOur code is available at https://github.com/qihangGH/IMRC.\n", "date": "2023-04-10", "categories": []}
{"id": "2401.09252", "title": "3D Scene Geometry Estimation from 360$^\\circ$ Imagery: A Survey", "abstract": "  This paper provides a comprehensive survey on pioneer and state-of-the-art 3D\nscene geometry estimation methodologies based on single, two, or multiple\nimages captured under the omnidirectional optics. We first revisit the basic\nconcepts of the spherical camera model, and review the most common acquisition\ntechnologies and representation formats suitable for omnidirectional (also\ncalled 360$^\\circ$, spherical or panoramic) images and videos. We then survey\nmonocular layout and depth inference approaches, highlighting the recent\nadvances in learning-based solutions suited for spherical data. The classical\nstereo matching is then revised on the spherical domain, where methodologies\nfor detecting and describing sparse and dense features become crucial. The\nstereo matching concepts are then extrapolated for multiple view camera setups,\ncategorizing them among light fields, multi-view stereo, and structure from\nmotion (or visual simultaneous localization and mapping). We also compile and\ndiscuss commonly adopted datasets and figures of merit indicated for each\npurpose and list recent results for completeness. We conclude this paper by\npointing out current and future trends.\n", "date": "2024-01-17", "categories": []}
{"id": "2306.00965", "title": "BUOL: A Bottom-Up Framework with Occupancy-aware Lifting for Panoptic 3D\n  Scene Reconstruction From A Single Image", "abstract": "  Understanding and modeling the 3D scene from a single image is a practical\nproblem. A recent advance proposes a panoptic 3D scene reconstruction task that\nperforms both 3D reconstruction and 3D panoptic segmentation from a single\nimage. Although having made substantial progress, recent works only focus on\ntop-down approaches that fill 2D instances into 3D voxels according to\nestimated depth, which hinders their performance by two ambiguities. (1)\ninstance-channel ambiguity: The variable ids of instances in each scene lead to\nambiguity during filling voxel channels with 2D information, confusing the\nfollowing 3D refinement. (2) voxel-reconstruction ambiguity: 2D-to-3D lifting\nwith estimated single view depth only propagates 2D information onto the\nsurface of 3D regions, leading to ambiguity during the reconstruction of\nregions behind the frontal view surface. In this paper, we propose BUOL, a\nBottom-Up framework with Occupancy-aware Lifting to address the two issues for\npanoptic 3D scene reconstruction from a single image. For instance-channel\nambiguity, a bottom-up framework lifts 2D information to 3D voxels based on\ndeterministic semantic assignments rather than arbitrary instance id\nassignments. The 3D voxels are then refined and grouped into 3D instances\naccording to the predicted 2D instance centers. For voxel-reconstruction\nambiguity, the estimated multi-plane occupancy is leveraged together with depth\nto fill the whole regions of things and stuff. Our method shows a tremendous\nperformance advantage over state-of-the-art methods on synthetic dataset\n3D-Front and real-world dataset Matterport3D. Code and models are available in\nhttps://github.com/chtsy/buol.\n", "date": "2023-06-01", "categories": []}
{"id": "2401.05335", "title": "InseRF: Text-Driven Generative Object Insertion in Neural 3D Scenes", "abstract": "  We introduce InseRF, a novel method for generative object insertion in the\nNeRF reconstructions of 3D scenes. Based on a user-provided textual description\nand a 2D bounding box in a reference viewpoint, InseRF generates new objects in\n3D scenes. Recently, methods for 3D scene editing have been profoundly\ntransformed, owing to the use of strong priors of text-to-image diffusion\nmodels in 3D generative modeling. Existing methods are mostly effective in\nediting 3D scenes via style and appearance changes or removing existing\nobjects. Generating new objects, however, remains a challenge for such methods,\nwhich we address in this study. Specifically, we propose grounding the 3D\nobject insertion to a 2D object insertion in a reference view of the scene. The\n2D edit is then lifted to 3D using a single-view object reconstruction method.\nThe reconstructed object is then inserted into the scene, guided by the priors\nof monocular depth estimation methods. We evaluate our method on various 3D\nscenes and provide an in-depth analysis of the proposed components. Our\nexperiments with generative insertion of objects in several 3D scenes indicate\nthe effectiveness of our method compared to the existing methods. InseRF is\ncapable of controllable and 3D-consistent object insertion without requiring\nexplicit 3D information as input. Please visit our project page at\nhttps://mohamad-shahbazi.github.io/inserf.\n", "date": "2024-01-10", "categories": []}
{"id": "2310.11598", "title": "Learning Neural Implicit through Volume Rendering with Attentive Depth\n  Fusion Priors", "abstract": "  Learning neural implicit representations has achieved remarkable performance\nin 3D reconstruction from multi-view images. Current methods use volume\nrendering to render implicit representations into either RGB or depth images\nthat are supervised by multi-view ground truth. However, rendering a view each\ntime suffers from incomplete depth at holes and unawareness of occluded\nstructures from the depth supervision, which severely affects the accuracy of\ngeometry inference via volume rendering. To resolve this issue, we propose to\nlearn neural implicit representations from multi-view RGBD images through\nvolume rendering with an attentive depth fusion prior. Our prior allows neural\nnetworks to perceive coarse 3D structures from the Truncated Signed Distance\nFunction (TSDF) fused from all depth images available for rendering. The TSDF\nenables accessing the missing depth at holes on one depth image and the\noccluded parts that are invisible from the current view. By introducing a novel\nattention mechanism, we allow neural networks to directly use the depth fusion\nprior with the inferred occupancy as the learned implicit function. Our\nattention mechanism works with either a one-time fused TSDF that represents a\nwhole scene or an incrementally fused TSDF that represents a partial scene in\nthe context of Simultaneous Localization and Mapping (SLAM). Our evaluations on\nwidely used benchmarks including synthetic and real-world scans show our\nsuperiority over the latest neural implicit methods. Project page:\nhttps://machineperceptionlab.github.io/Attentive_DF_Prior/\n", "date": "2023-10-17", "categories": []}
{"id": "2312.16218", "title": "Hyper-VolTran: Fast and Generalizable One-Shot Image to 3D Object\n  Structure via HyperNetworks", "abstract": "  Solving image-to-3D from a single view is an ill-posed problem, and current\nneural reconstruction methods addressing it through diffusion models still rely\non scene-specific optimization, constraining their generalization capability.\nTo overcome the limitations of existing approaches regarding generalization and\nconsistency, we introduce a novel neural rendering technique. Our approach\nemploys the signed distance function as the surface representation and\nincorporates generalizable priors through geometry-encoding volumes and\nHyperNetworks. Specifically, our method builds neural encoding volumes from\ngenerated multi-view inputs. We adjust the weights of the SDF network\nconditioned on an input image at test-time to allow model adaptation to novel\nscenes in a feed-forward manner via HyperNetworks. To mitigate artifacts\nderived from the synthesized views, we propose the use of a volume transformer\nmodule to improve the aggregation of image features instead of processing each\nviewpoint separately. Through our proposed method, dubbed as Hyper-VolTran, we\navoid the bottleneck of scene-specific optimization and maintain consistency\nacross the images generated from multiple viewpoints. Our experiments show the\nadvantages of our proposed approach with consistent results and rapid\ngeneration.\n", "date": "2023-12-24", "categories": []}
{"id": "2401.03203", "title": "Hi-Map: Hierarchical Factorized Radiance Field for High-Fidelity\n  Monocular Dense Mapping", "abstract": "  In this paper, we introduce Hi-Map, a novel monocular dense mapping approach\nbased on Neural Radiance Field (NeRF). Hi-Map is exceptional in its capacity to\nachieve efficient and high-fidelity mapping using only posed RGB inputs. Our\nmethod eliminates the need for external depth priors derived from e.g., a depth\nestimation model. Our key idea is to represent the scene as a hierarchical\nfeature grid that encodes the radiance and then factorizes it into feature\nplanes and vectors. As such, the scene representation becomes simpler and more\ngeneralizable for fast and smooth convergence on new observations. This allows\nfor efficient computation while alleviating noise patterns by reducing the\ncomplexity of the scene representation. Buttressed by the hierarchical\nfactorized representation, we leverage the Sign Distance Field (SDF) as a proxy\nof rendering for inferring the volume density, demonstrating high mapping\nfidelity. Moreover, we introduce a dual-path encoding strategy to strengthen\nthe photometric cues and further boost the mapping quality, especially for the\ndistant and textureless regions. Extensive experiments demonstrate our method's\nsuperiority in geometric and textural accuracy over the state-of-the-art\nNeRF-based monocular mapping methods.\n", "date": "2024-01-06", "categories": []}
{"id": "2401.03257", "title": "RustNeRF: Robust Neural Radiance Field with Low-Quality Images", "abstract": "  Recent work on Neural Radiance Fields (NeRF) exploits multi-view 3D\nconsistency, achieving impressive results in 3D scene modeling and\nhigh-fidelity novel-view synthesis. However, there are limitations. First,\nexisting methods assume enough high-quality images are available for training\nthe NeRF model, ignoring real-world image degradation. Second, previous methods\nstruggle with ambiguity in the training set due to unmodeled inconsistencies\namong different views. In this work, we present RustNeRF for real-world\nhigh-quality NeRF. To improve NeRF's robustness under real-world inputs, we\ntrain a 3D-aware preprocessing network that incorporates real-world degradation\nmodeling. We propose a novel implicit multi-view guidance to address\ninformation loss during image degradation and restoration. Extensive\nexperiments demonstrate RustNeRF's advantages over existing approaches under\nreal-world degradation. The code will be released.\n", "date": "2024-01-06", "categories": []}
{"id": "2401.03704", "title": "Sur2f: A Hybrid Representation for High-Quality and Efficient Surface\n  Reconstruction from Multi-view Images", "abstract": "  Multi-view surface reconstruction is an ill-posed, inverse problem in 3D\nvision research. It involves modeling the geometry and appearance with\nappropriate surface representations. Most of the existing methods rely either\non explicit meshes, using surface rendering of meshes for reconstruction, or on\nimplicit field functions, using volume rendering of the fields for\nreconstruction. The two types of representations in fact have their respective\nmerits. In this work, we propose a new hybrid representation, termed Sur2f,\naiming to better benefit from both representations in a complementary manner.\nTechnically, we learn two parallel streams of an implicit signed distance field\nand an explicit surrogate surface Sur2f mesh, and unify volume rendering of the\nimplicit signed distance function (SDF) and surface rendering of the surrogate\nmesh with a shared, neural shader; the unified shading promotes their\nconvergence to the same, underlying surface. We synchronize learning of the\nsurrogate mesh by driving its deformation with functions induced from the\nimplicit SDF. In addition, the synchronized surrogate mesh enables\nsurface-guided volume sampling, which greatly improves the sampling efficiency\nper ray in volume rendering. We conduct thorough experiments showing that\nSur$^2$f outperforms existing reconstruction methods and surface\nrepresentations, including hybrid ones, in terms of both recovery quality and\nrecovery efficiency.\n", "date": "2024-01-08", "categories": []}
{"id": "2311.13398", "title": "Depth-Regularized Optimization for 3D Gaussian Splatting in Few-Shot\n  Images", "abstract": "  In this paper, we present a method to optimize Gaussian splatting with a\nlimited number of images while avoiding overfitting. Representing a 3D scene by\ncombining numerous Gaussian splats has yielded outstanding visual quality.\nHowever, it tends to overfit the training views when only a small number of\nimages are available. To address this issue, we introduce a dense depth map as\na geometry guide to mitigate overfitting. We obtained the depth map using a\npre-trained monocular depth estimation model and aligning the scale and offset\nusing sparse COLMAP feature points. The adjusted depth aids in the color-based\noptimization of 3D Gaussian splatting, mitigating floating artifacts, and\nensuring adherence to geometric constraints. We verify the proposed method on\nthe NeRF-LLFF dataset with varying numbers of few images. Our approach\ndemonstrates robust geometry compared to the original method that relies solely\non images. Project page: robot0321.github.io/DepthRegGS\n", "date": "2023-11-22", "categories": []}
{"id": "2401.02292", "title": "GridFormer: Point-Grid Transformer for Surface Reconstruction", "abstract": "  Implicit neural networks have emerged as a crucial technology in 3D surface\nreconstruction. To reconstruct continuous surfaces from discrete point clouds,\nencoding the input points into regular grid features (plane or volume) has been\ncommonly employed in existing approaches. However, these methods typically use\nthe grid as an index for uniformly scattering point features. Compared with the\nirregular point features, the regular grid features may sacrifice some\nreconstruction details but improve efficiency. To take full advantage of these\ntwo types of features, we introduce a novel and high-efficiency attention\nmechanism between the grid and point features named Point-Grid Transformer\n(GridFormer). This mechanism treats the grid as a transfer point connecting the\nspace and point cloud. Our method maximizes the spatial expressiveness of grid\nfeatures and maintains computational efficiency. Furthermore, optimizing\npredictions over the entire space could potentially result in blurred\nboundaries. To address this issue, we further propose a boundary optimization\nstrategy incorporating margin binary cross-entropy loss and boundary sampling.\nThis approach enables us to achieve a more precise representation of the object\nstructure. Our experiments validate that our method is effective and\noutperforms the state-of-the-art approaches under widely used benchmarks by\nproducing more precise geometry reconstructions. The code is available at\nhttps://github.com/list17/GridFormer.\n", "date": "2024-01-04", "categories": []}
{"id": "2312.16215", "title": "SUNDIAL: 3D Satellite Understanding through Direct, Ambient, and Complex\n  Lighting Decomposition", "abstract": "  3D modeling from satellite imagery is essential in areas of environmental\nscience, urban planning, agriculture, and disaster response. However,\ntraditional 3D modeling techniques face unique challenges in the remote sensing\ncontext, including limited multi-view baselines over extensive regions, varying\ndirect, ambient, and complex illumination conditions, and time-varying scene\nchanges across captures. In this work, we introduce SUNDIAL, a comprehensive\napproach to 3D reconstruction of satellite imagery using neural radiance\nfields. We jointly learn satellite scene geometry, illumination components, and\nsun direction in this single-model approach, and propose a secondary shadow ray\ncasting technique to 1) improve scene geometry using oblique sun angles to\nrender shadows, 2) enable physically-based disentanglement of scene albedo and\nillumination, and 3) determine the components of illumination from direct,\nambient (sky), and complex sources. To achieve this, we incorporate lighting\ncues and geometric priors from remote sensing literature in a neural rendering\napproach, modeling physical properties of satellite scenes such as shadows,\nscattered sky illumination, and complex illumination and shading of vegetation\nand water. We evaluate the performance of SUNDIAL against existing NeRF-based\ntechniques for satellite scene modeling and demonstrate improved scene and\nlighting disentanglement, novel view and lighting rendering, and geometry and\nsun direction estimation on challenging scenes with small baselines, sparse\ninputs, and variable illumination.\n", "date": "2023-12-24", "categories": []}
{"id": "2312.17192", "title": "HISR: Hybrid Implicit Surface Representation for Photorealistic 3D Human\n  Reconstruction", "abstract": "  Neural reconstruction and rendering strategies have demonstrated\nstate-of-the-art performances due, in part, to their ability to preserve high\nlevel shape details. Existing approaches, however, either represent objects as\nimplicit surface functions or neural volumes and still struggle to recover\nshapes with heterogeneous materials, in particular human skin, hair or clothes.\nTo this aim, we present a new hybrid implicit surface representation to model\nhuman shapes. This representation is composed of two surface layers that\nrepresent opaque and translucent regions on the clothed human body. We segment\ndifferent regions automatically using visual cues and learn to reconstruct two\nsigned distance functions (SDFs). We perform surface-based rendering on opaque\nregions (e.g., body, face, clothes) to preserve high-fidelity surface normals\nand volume rendering on translucent regions (e.g., hair). Experiments\ndemonstrate that our approach obtains state-of-the-art results on 3D human\nreconstructions, and also shows competitive performances on other objects.\n", "date": "2023-12-28", "categories": []}
{"id": "2307.05473", "title": "Differentiable Blocks World: Qualitative 3D Decomposition by Rendering\n  Primitives", "abstract": "  Given a set of calibrated images of a scene, we present an approach that\nproduces a simple, compact, and actionable 3D world representation by means of\n3D primitives. While many approaches focus on recovering high-fidelity 3D\nscenes, we focus on parsing a scene into mid-level 3D representations made of a\nsmall set of textured primitives. Such representations are interpretable, easy\nto manipulate and suited for physics-based simulations. Moreover, unlike\nexisting primitive decomposition methods that rely on 3D input data, our\napproach operates directly on images through differentiable rendering.\nSpecifically, we model primitives as textured superquadric meshes and optimize\ntheir parameters from scratch with an image rendering loss. We highlight the\nimportance of modeling transparency for each primitive, which is critical for\noptimization and also enables handling varying numbers of primitives. We show\nthat the resulting textured primitives faithfully reconstruct the input images\nand accurately model the visible 3D points, while providing amodal shape\ncompletions of unseen object regions. We compare our approach to the state of\nthe art on diverse scenes from DTU, and demonstrate its robustness on real-life\ncaptures from BlendedMVS and Nerfstudio. We also showcase how our results can\nbe used to effortlessly edit a scene or perform physical simulations. Code and\nvideo results are available at https://www.tmonnier.com/DBW .\n", "date": "2023-07-11", "categories": []}
{"id": "2312.13977", "title": "NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse\n  Input Views", "abstract": "  Recently, neural implicit functions have demonstrated remarkable results in\nthe field of multi-view reconstruction. However, most existing methods are\ntailored for dense views and exhibit unsatisfactory performance when dealing\nwith sparse views. Several latest methods have been proposed for generalizing\nimplicit reconstruction to address the sparse view reconstruction task, but\nthey still suffer from high training costs and are merely valid under carefully\nselected perspectives. In this paper, we propose a novel sparse view\nreconstruction framework that leverages on-surface priors to achieve highly\nfaithful surface reconstruction. Specifically, we design several constraints on\nglobal geometry alignment and local geometry refinement for jointly optimizing\ncoarse shapes and fine details. To achieve this, we train a neural network to\nlearn a global implicit field from the on-surface points obtained from SfM and\nthen leverage it as a coarse geometric constraint. To exploit local geometric\nconsistency, we project on-surface points onto seen and unseen views, treating\nthe consistent loss of projected features as a fine geometric constraint. The\nexperimental results with DTU and BlendedMVS datasets in two prevalent sparse\nsettings demonstrate significant improvements over the state-of-the-art\nmethods.\n", "date": "2023-12-21", "categories": []}
{"id": "2306.09077", "title": "Estimating Generic 3D Room Structures from 2D Annotations", "abstract": "  Indoor rooms are among the most common use cases in 3D scene understanding.\nCurrent state-of-the-art methods for this task are driven by large annotated\ndatasets. Room layouts are especially important, consisting of structural\nelements in 3D, such as wall, floor, and ceiling. However, they are difficult\nto annotate, especially on pure RGB video. We propose a novel method to produce\ngeneric 3D room layouts just from 2D segmentation masks, which are easy to\nannotate for humans. Based on these 2D annotations, we automatically\nreconstruct 3D plane equations for the structural elements and their spatial\nextent in the scene, and connect adjacent elements at the appropriate contact\nedges. We annotate and publicly release 2246 3D room layouts on the\nRealEstate10k dataset, containing YouTube videos. We demonstrate the high\nquality of these 3D layouts annotations with extensive experiments.\n", "date": "2023-06-15", "categories": []}
{"id": "2311.14981", "title": "Multi-task Planar Reconstruction with Feature Warping Guidance", "abstract": "  Piece-wise planar 3D reconstruction simultaneously segments plane instances\nand recovers their 3D plane parameters from an image, which is particularly\nuseful for indoor or man-made environments. Efficient reconstruction of 3D\nplanes coupled with semantic predictions offers advantages for a wide range of\napplications requiring scene understanding and concurrent spatial mapping.\nHowever, most existing planar reconstruction models either neglect semantic\npredictions or do not run efficiently enough for real-time applications. We\nintroduce SOLOPlanes, a real-time planar reconstruction model based on a\nmodified instance segmentation architecture which simultaneously predicts\nsemantics for each plane instance, along with plane parameters and piece-wise\nplane instance masks. We achieve an improvement in instance mask segmentation\nby including multi-view guidance for plane predictions in the training process.\nThis cross-task improvement, training for plane prediction but improving the\nmask segmentation, is due to the nature of feature sharing in multi-task\nlearning. Our model simultaneously predicts semantics using single images at\ninference time, while achieving real-time predictions at 43 FPS.\n", "date": "2023-11-25", "categories": []}
{"id": "2312.12961", "title": "Radar Fields: An Extension of Radiance Fields to SAR", "abstract": "  Radiance fields have been a major breakthrough in the field of inverse\nrendering, novel view synthesis and 3D modeling of complex scenes from\nmulti-view image collections. Since their introduction, it was shown that they\ncould be extended to other modalities such as LiDAR, radio frequencies, X-ray\nor ultrasound. In this paper, we show that, despite the important difference\nbetween optical and synthetic aperture radar (SAR) image formation models, it\nis possible to extend radiance fields to radar images thus presenting the first\n\"radar fields\". This allows us to learn surface models using only collections\nof radar images, similar to how regular radiance fields are learned and with\nthe same computational complexity on average. Thanks to similarities in how\nboth fields are defined, this work also shows a potential for hybrid methods\ncombining both optical and SAR images.\n", "date": "2023-12-20", "categories": []}
{"id": "2308.06962", "title": "Color-NeuS: Reconstructing Neural Implicit Surfaces with Color", "abstract": "  The reconstruction of object surfaces from multi-view images or monocular\nvideo is a fundamental issue in computer vision. However, much of the recent\nresearch concentrates on reconstructing geometry through implicit or explicit\nmethods. In this paper, we shift our focus towards reconstructing mesh in\nconjunction with color. We remove the view-dependent color from neural volume\nrendering while retaining volume rendering performance through a relighting\nnetwork. Mesh is extracted from the signed distance function (SDF) network for\nthe surface, and color for each surface vertex is drawn from the global color\nnetwork. To evaluate our approach, we conceived a in hand object scanning task\nfeaturing numerous occlusions and dramatic shifts in lighting conditions. We've\ngathered several videos for this task, and the results surpass those of any\nexisting methods capable of reconstructing mesh alongside color. Additionally,\nour method's performance was assessed using public datasets, including DTU,\nBlendedMVS, and OmniObject3D. The results indicated that our method performs\nwell across all these datasets. Project page:\nhttps://colmar-zlicheng.github.io/color_neus.\n", "date": "2023-08-14", "categories": []}
{"id": "2312.09147", "title": "Triplane Meets Gaussian Splatting: Fast and Generalizable Single-View 3D\n  Reconstruction with Transformers", "abstract": "  Recent advancements in 3D reconstruction from single images have been driven\nby the evolution of generative models. Prominent among these are methods based\non Score Distillation Sampling (SDS) and the adaptation of diffusion models in\nthe 3D domain. Despite their progress, these techniques often face limitations\ndue to slow optimization or rendering processes, leading to extensive training\nand optimization times. In this paper, we introduce a novel approach for\nsingle-view reconstruction that efficiently generates a 3D model from a single\nimage via feed-forward inference. Our method utilizes two transformer-based\nnetworks, namely a point decoder and a triplane decoder, to reconstruct 3D\nobjects using a hybrid Triplane-Gaussian intermediate representation. This\nhybrid representation strikes a balance, achieving a faster rendering speed\ncompared to implicit representations while simultaneously delivering superior\nrendering quality than explicit representations. The point decoder is designed\nfor generating point clouds from single images, offering an explicit\nrepresentation which is then utilized by the triplane decoder to query Gaussian\nfeatures for each point. This design choice addresses the challenges associated\nwith directly regressing explicit 3D Gaussian attributes characterized by their\nnon-structural nature. Subsequently, the 3D Gaussians are decoded by an MLP to\nenable rapid rendering through splatting. Both decoders are built upon a\nscalable, transformer-based architecture and have been efficiently trained on\nlarge-scale 3D datasets. The evaluations conducted on both synthetic datasets\nand real-world images demonstrate that our method not only achieves higher\nquality but also ensures a faster runtime in comparison to previous\nstate-of-the-art techniques. Please see our project page at\nhttps://zouzx.github.io/TriplaneGaussian/.\n", "date": "2023-12-14", "categories": []}
{"id": "2312.12418", "title": "LASA: Instance Reconstruction from Real Scans using A Large-scale\n  Aligned Shape Annotation Dataset", "abstract": "  Instance shape reconstruction from a 3D scene involves recovering the full\ngeometries of multiple objects at the semantic instance level. Many methods\nleverage data-driven learning due to the intricacies of scene complexity and\nsignificant indoor occlusions. Training these methods often requires a\nlarge-scale, high-quality dataset with aligned and paired shape annotations\nwith real-world scans. Existing datasets are either synthetic or misaligned,\nrestricting the performance of data-driven methods on real data. To this end,\nwe introduce LASA, a Large-scale Aligned Shape Annotation Dataset comprising\n10,412 high-quality CAD annotations aligned with 920 real-world scene scans\nfrom ArkitScenes, created manually by professional artists. On this top, we\npropose a novel Diffusion-based Cross-Modal Shape Reconstruction (DisCo)\nmethod. It is empowered by a hybrid feature aggregation design to fuse\nmulti-modal inputs and recover high-fidelity object geometries. Besides, we\npresent an Occupancy-Guided 3D Object Detection (OccGOD) method and demonstrate\nthat our shape annotations provide scene occupancy clues that can further\nimprove 3D object detection. Supported by LASA, extensive experiments show that\nour methods achieve state-of-the-art performance in both instance-level scene\nreconstruction and 3D object detection tasks.\n", "date": "2023-12-19", "categories": []}
{"id": "2310.19629", "title": "RayDF: Neural Ray-surface Distance Fields with Multi-view Consistency", "abstract": "  In this paper, we study the problem of continuous 3D shape representations.\nThe majority of existing successful methods are coordinate-based implicit\nneural representations. However, they are inefficient to render novel views or\nrecover explicit surface points. A few works start to formulate 3D shapes as\nray-based neural functions, but the learned structures are inferior due to the\nlack of multi-view geometry consistency. To tackle these challenges, we propose\na new framework called RayDF. It consists of three major components: 1) the\nsimple ray-surface distance field, 2) the novel dual-ray visibility classifier,\nand 3) a multi-view consistency optimization module to drive the learned\nray-surface distances to be multi-view geometry consistent. We extensively\nevaluate our method on three public datasets, demonstrating remarkable\nperformance in 3D surface point reconstruction on both synthetic and\nchallenging real-world 3D scenes, clearly surpassing existing coordinate-based\nand ray-based baselines. Most notably, our method achieves a 1000x faster speed\nthan coordinate-based methods to render an 800x800 depth image, showing the\nsuperiority of our method for 3D shape representation. Our code and data are\navailable at https://github.com/vLAR-group/RayDF\n", "date": "2023-10-30", "categories": []}
{"id": "2312.09005", "title": "Scene 3-D Reconstruction System in Scattering Medium", "abstract": "  The research on neural radiance fields for new view synthesis has experienced\nexplosive growth with the development of new models and extensions. The NERF\nalgorithm, suitable for underwater scenes or scattering media, is also\nevolving. Existing underwater 3D reconstruction systems still face challenges\nsuch as extensive training time and low rendering efficiency. This paper\nproposes an improved underwater 3D reconstruction system to address these\nissues and achieve rapid, high-quality 3D reconstruction.To begin with, we\nenhance underwater videos captured by a monocular camera to correct the poor\nimage quality caused by the physical properties of the water medium while\nensuring consistency in enhancement across adjacent frames. Subsequently, we\nperform keyframe selection on the video frames to optimize resource utilization\nand eliminate the impact of dynamic objects on the reconstruction results. The\nselected keyframes, after pose estimation using COLMAP, undergo a\nthree-dimensional reconstruction improvement process using neural radiance\nfields based on multi-resolution hash coding for model construction and\nrendering.\n", "date": "2023-12-14", "categories": []}
{"id": "2306.01642", "title": "Automatic Reconstruction of Semantic 3D Models from 2D Floor Plans", "abstract": "  Digitalization of existing buildings and the creation of 3D BIM models for\nthem has become crucial for many tasks. Of particular importance are floor\nplans, which contain information about building layouts and are vital for\nprocesses such as construction, maintenance or refurbishing. However, this data\nis not always available in digital form, especially for older buildings\nconstructed before CAD tools were widely available, or lacks semantic\ninformation. The digitalization of such information usually requires manual\nwork of an expert that must reconstruct the layouts by hand, which is a\ncumbersome and error-prone process. In this paper, we present a pipeline for\nreconstruction of vectorized 3D models from scanned 2D plans, aiming at\nincreasing the efficiency of this process. The method presented achieves\nstate-of-the-art results in the public dataset CubiCasa5k, and shows good\ngeneralization to different types of plans. Our vectorization approach is\nparticularly effective, outperforming previous methods.\n", "date": "2023-06-02", "categories": []}
{"id": "2312.02221", "title": "Slice3D: Multi-Slice, Occlusion-Revealing, Single View 3D Reconstruction", "abstract": "  We introduce multi-slice reasoning, a new notion for single-view 3D\nreconstruction which challenges the current and prevailing belief that\nmulti-view synthesis is the most natural conduit between single-view and 3D.\nOur key observation is that object slicing is more advantageous than altering\nviews to reveal occluded structures. Specifically, slicing is more\nocclusion-revealing since it can peel through any occluders without\nobstruction. In the limit, i.e., with infinitely many slices, it is guaranteed\nto unveil all hidden object parts. We realize our idea by developing Slice3D, a\nnovel method for single-view 3D reconstruction which first predicts multi-slice\nimages from a single RGB image and then integrates the slices into a 3D model\nusing a coordinate-based transformer network for signed distance prediction.\nThe slice images can be regressed or generated, both through a U-Net based\nnetwork. For the former, we inject a learnable slice indicator code to\ndesignate each decoded image into a spatial slice location, while the slice\ngenerator is a denoising diffusion model operating on the entirety of slice\nimages stacked on the input channels. We conduct extensive evaluation against\nstate-of-the-art alternatives to demonstrate superiority of our method,\nespecially in recovering complex and severely occluded shape structures, amid\nambiguities. All Slice3D results were produced by networks trained on a single\nNvidia A40 GPU, with an inference time less than 20 seconds.\n", "date": "2023-12-03", "categories": []}
{"id": "2312.06706", "title": "UNeR3D: Versatile and Scalable 3D RGB Point Cloud Generation from 2D\n  Images in Unsupervised Reconstruction", "abstract": "  In the realm of 3D reconstruction from 2D images, a persisting challenge is\nto achieve high-precision reconstructions devoid of 3D Ground Truth data\nreliance. We present UNeR3D, a pioneering unsupervised methodology that sets a\nnew standard for generating detailed 3D reconstructions solely from 2D views.\nOur model significantly cuts down the training costs tied to supervised\napproaches and introduces RGB coloration to 3D point clouds, enriching the\nvisual experience. Employing an inverse distance weighting technique for color\nrendering, UNeR3D ensures seamless color transitions, enhancing visual\nfidelity. Our model's flexible architecture supports training with any number\nof views, and uniquely, it is not constrained by the number of views used\nduring training when performing reconstructions. It can infer with an arbitrary\ncount of views during inference, offering unparalleled versatility.\nAdditionally, the model's continuous spatial input domain allows the generation\nof point clouds at any desired resolution, empowering the creation of\nhigh-resolution 3D RGB point clouds. We solidify the reconstruction process\nwith a novel multi-view geometric loss and color loss, demonstrating that our\nmodel excels with single-view inputs and beyond, thus reshaping the paradigm of\nunsupervised learning in 3D vision. Our contributions signal a substantial leap\nforward in 3D vision, offering new horizons for content creation across diverse\napplications. Code is available at https://github.com/HongbinLin3589/UNeR3D.\n", "date": "2023-12-10", "categories": []}
{"id": "2306.02956", "title": "Explicit Neural Surfaces: Learning Continuous Geometry With Deformation\n  Fields", "abstract": "  We introduce Explicit Neural Surfaces (ENS), an efficient smooth surface\nrepresentation that directly encodes topology with a deformation field from a\nknown base domain. We apply this representation to reconstruct explicit\nsurfaces from multiple views, where we use a series of neural deformation\nfields to progressively transform the base domain into a target shape. By using\nmeshes as discrete surface proxies, we train the deformation fields through\nefficient differentiable rasterization. Using a fixed base domain allows us to\nhave Laplace-Beltrami eigenfunctions as an intrinsic positional encoding\nalongside standard extrinsic Fourier features, with which our approach can\ncapture fine surface details. Compared to implicit surfaces, ENS trains faster\nand has several orders of magnitude faster inference times. The explicit nature\nof our approach also allows higher-quality mesh extraction whilst maintaining\ncompetitive surface reconstruction performance and real-time capabilities.\n", "date": "2023-06-05", "categories": []}
{"id": "2312.02617", "title": "DreaMo: Articulated 3D Reconstruction From A Single Casual Video", "abstract": "  Articulated 3D reconstruction has valuable applications in various domains,\nyet it remains costly and demands intensive work from domain experts. Recent\nadvancements in template-free learning methods show promising results with\nmonocular videos. Nevertheless, these approaches necessitate a comprehensive\ncoverage of all viewpoints of the subject in the input video, thus limiting\ntheir applicability to casually captured videos from online sources. In this\nwork, we study articulated 3D shape reconstruction from a single and casually\ncaptured internet video, where the subject's view coverage is incomplete. We\npropose DreaMo that jointly performs shape reconstruction while solving the\nchallenging low-coverage regions with view-conditioned diffusion prior and\nseveral tailored regularizations. In addition, we introduce a skeleton\ngeneration strategy to create human-interpretable skeletons from the learned\nneural bones and skinning weights. We conduct our study on a self-collected\ninternet video collection characterized by incomplete view coverage. DreaMo\nshows promising quality in novel-view rendering, detailed articulated shape\nreconstruction, and skeleton generation. Extensive qualitative and quantitative\nstudies validate the efficacy of each proposed component, and show existing\nmethods are unable to solve correct geometry due to the incomplete view\ncoverage.\n", "date": "2023-12-05", "categories": []}
{"id": "2312.04654", "title": "NeuSD: Surface Completion with Multi-View Text-to-Image Diffusion", "abstract": "  We present a novel method for 3D surface reconstruction from multiple images\nwhere only a part of the object of interest is captured. Our approach builds on\ntwo recent developments: surface reconstruction using neural radiance fields\nfor the reconstruction of the visible parts of the surface, and guidance of\npre-trained 2D diffusion models in the form of Score Distillation Sampling\n(SDS) to complete the shape in unobserved regions in a plausible manner. We\nintroduce three components. First, we suggest employing normal maps as a pure\ngeometric representation for SDS instead of color renderings which are\nentangled with the appearance information. Second, we introduce the freezing of\nthe SDS noise during training which results in more coherent gradients and\nbetter convergence. Third, we propose Multi-View SDS as a way to condition the\ngeneration of the non-observable part of the surface without fine-tuning or\nmaking changes to the underlying 2D Stable Diffusion model. We evaluate our\napproach on the BlendedMVS dataset demonstrating significant qualitative and\nquantitative improvements over competing methods.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.04803", "title": "SuperNormal: Neural Surface Reconstruction via Multi-View Normal\n  Integration", "abstract": "  We present SuperNormal, a fast, high-fidelity approach to multi-view 3D\nreconstruction using surface normal maps. With a few minutes, SuperNormal\nproduces detailed surfaces on par with 3D scanners. We harness volume rendering\nto optimize a neural signed distance function (SDF) powered by multi-resolution\nhash encoding. To accelerate training, we propose directional finite difference\nand patch-based ray marching to approximate the SDF gradients numerically.\nWhile not compromising reconstruction quality, this strategy is nearly twice as\nefficient as analytical gradients and about three times faster than\naxis-aligned finite difference. Experiments on the benchmark dataset\ndemonstrate the superiority of SuperNormal in efficiency and accuracy compared\nto existing multi-view photometric stereo methods. On our captured objects,\nSuperNormal produces more fine-grained geometry than recent neural 3D\nreconstruction methods.\n", "date": "2023-12-08", "categories": []}
{"id": "2307.11086", "title": "PAPR: Proximity Attention Point Rendering", "abstract": "  Learning accurate and parsimonious point cloud representations of scene\nsurfaces from scratch remains a challenge in 3D representation learning.\nExisting point-based methods often suffer from the vanishing gradient problem\nor require a large number of points to accurately model scene geometry and\ntexture. To address these limitations, we propose Proximity Attention Point\nRendering (PAPR), a novel method that consists of a point-based scene\nrepresentation and a differentiable renderer. Our scene representation uses a\npoint cloud where each point is characterized by its spatial position,\ninfluence score, and view-independent feature vector. The renderer selects the\nrelevant points for each ray and produces accurate colours using their\nassociated features. PAPR effectively learns point cloud positions to represent\nthe correct scene geometry, even when the initialization drastically differs\nfrom the target geometry. Notably, our method captures fine texture details\nwhile using only a parsimonious set of points. We also demonstrate four\npractical applications of our method: zero-shot geometry editing, object\nmanipulation, texture transfer, and exposure control. More results and code are\navailable on our project website at https://zvict.github.io/papr/.\n", "date": "2023-07-20", "categories": []}
{"id": "2312.04560", "title": "NeRFiller: Completing Scenes via Generative 3D Inpainting", "abstract": "  We propose NeRFiller, an approach that completes missing portions of a 3D\ncapture via generative 3D inpainting using off-the-shelf 2D visual generative\nmodels. Often parts of a captured 3D scene or object are missing due to mesh\nreconstruction failures or a lack of observations (e.g., contact regions, such\nas the bottom of objects, or hard-to-reach areas). We approach this challenging\n3D inpainting problem by leveraging a 2D inpainting diffusion model. We\nidentify a surprising behavior of these models, where they generate more 3D\nconsistent inpaints when images form a 2$\\times$2 grid, and show how to\ngeneralize this behavior to more than four images. We then present an iterative\nframework to distill these inpainted regions into a single consistent 3D scene.\nIn contrast to related works, we focus on completing scenes rather than\ndeleting foreground objects, and our approach does not require tight 2D object\nmasks or text. We compare our approach to relevant baselines adapted to our\nsetting on a variety of scenes, where NeRFiller creates the most 3D consistent\nand plausible scene completions. Our project page is at\nhttps://ethanweber.me/nerfiller.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.03266", "title": "SO-NeRF: Active View Planning for NeRF using Surrogate Objectives", "abstract": "  Despite the great success of Neural Radiance Fields (NeRF), its\ndata-gathering process remains vague with only a general rule of thumb of\nsampling as densely as possible. The lack of understanding of what actually\nconstitutes good views for NeRF makes it difficult to actively plan a sequence\nof views that yield the maximal reconstruction quality. We propose Surrogate\nObjectives for Active Radiance Fields (SOAR), which is a set of interpretable\nfunctions that evaluates the goodness of views using geometric and photometric\nvisual cues - surface coverage, geometric complexity, textural complexity, and\nray diversity. Moreover, by learning to infer the SOAR scores from a deep\nnetwork, SOARNet, we are able to effectively select views in mere seconds\ninstead of hours, without the need for prior visits to all the candidate views\nor training any radiance field during such planning. Our experiments show\nSOARNet outperforms the baselines with $\\sim$80x speed-up while achieving\nbetter or comparable reconstruction qualities. We finally show that SOAR is\nmodel-agnostic, thus it generalizes across fully neural-implicit to fully\nexplicit approaches.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.03372", "title": "Evaluating the point cloud of individual trees generated from images\n  based on Neural Radiance fields (NeRF) method", "abstract": "  Three-dimensional (3D) reconstruction of trees has always been a key task in\nprecision forestry management and research. Due to the complex branch\nmorphological structure of trees themselves and the occlusions from tree stems,\nbranches and foliage, it is difficult to recreate a complete three-dimensional\ntree model from a two-dimensional image by conventional photogrammetric\nmethods. In this study, based on tree images collected by various cameras in\ndifferent ways, the Neural Radiance Fields (NeRF) method was used for\nindividual tree reconstruction and the exported point cloud models are compared\nwith point cloud derived from photogrammetric reconstruction and laser scanning\nmethods. The results show that the NeRF method performs well in individual tree\n3D reconstruction, as it has higher successful reconstruction rate, better\nreconstruction in the canopy area, it requires less amount of images as input.\nCompared with photogrammetric reconstruction method, NeRF has significant\nadvantages in reconstruction efficiency and is adaptable to complex scenes, but\nthe generated point cloud tends to be noisy and low resolution. The accuracy of\ntree structural parameters (tree height and diameter at breast height)\nextracted from the photogrammetric point cloud is still higher than those of\nderived from the NeRF point cloud. The results of this study illustrate the\ngreat potential of NeRF method for individual tree reconstruction, and it\nprovides new ideas and research directions for 3D reconstruction and\nvisualization of complex forest scenes.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.03431", "title": "Gaussian-Flow: 4D Reconstruction with Dynamic 3D Gaussian Particle", "abstract": "  We introduce Gaussian-Flow, a novel point-based approach for fast dynamic\nscene reconstruction and real-time rendering from both multi-view and monocular\nvideos. In contrast to the prevalent NeRF-based approaches hampered by slow\ntraining and rendering speeds, our approach harnesses recent advancements in\npoint-based 3D Gaussian Splatting (3DGS). Specifically, a novel Dual-Domain\nDeformation Model (DDDM) is proposed to explicitly model attribute deformations\nof each Gaussian point, where the time-dependent residual of each attribute is\ncaptured by a polynomial fitting in the time domain, and a Fourier series\nfitting in the frequency domain. The proposed DDDM is capable of modeling\ncomplex scene deformations across long video footage, eliminating the need for\ntraining separate 3DGS for each frame or introducing an additional implicit\nneural field to model 3D dynamics. Moreover, the explicit deformation modeling\nfor discretized Gaussian points ensures ultra-fast training and rendering of a\n4D scene, which is comparable to the original 3DGS designed for static 3D\nreconstruction. Our proposed approach showcases a substantial efficiency\nimprovement, achieving a $5\\times$ faster training speed compared to the\nper-frame 3DGS modeling. In addition, quantitative results demonstrate that the\nproposed Gaussian-Flow significantly outperforms previous leading methods in\nnovel view rendering quality. Project page:\nhttps://nju-3dv.github.io/projects/Gaussian-Flow\n", "date": "2023-12-06", "categories": []}
{"id": "2312.02981", "title": "ReconFusion: 3D Reconstruction with Diffusion Priors", "abstract": "  3D reconstruction methods such as Neural Radiance Fields (NeRFs) excel at\nrendering photorealistic novel views of complex scenes. However, recovering a\nhigh-quality NeRF typically requires tens to hundreds of input images,\nresulting in a time-consuming capture process. We present ReconFusion to\nreconstruct real-world scenes using only a few photos. Our approach leverages a\ndiffusion prior for novel view synthesis, trained on synthetic and multiview\ndatasets, which regularizes a NeRF-based 3D reconstruction pipeline at novel\ncamera poses beyond those captured by the set of input images. Our method\nsynthesizes realistic geometry and texture in underconstrained regions while\npreserving the appearance of observed regions. We perform an extensive\nevaluation across various real-world datasets, including forward-facing and\n360-degree scenes, demonstrating significant performance improvements over\nprevious few-view NeRF reconstruction approaches.\n", "date": "2023-12-05", "categories": []}
{"id": "2306.07581", "title": "Binary Radiance Fields", "abstract": "  In this paper, we propose \\textit{binary radiance fields} (BiRF), a\nstorage-efficient radiance field representation employing binary feature\nencoding that encodes local features using binary encoding parameters in a\nformat of either $+1$ or $-1$. This binarization strategy lets us represent the\nfeature grid with highly compact feature encoding and a dramatic reduction in\nstorage size. Furthermore, our 2D-3D hybrid feature grid design enhances the\ncompactness of feature encoding as the 3D grid includes main components while\n2D grids capture details. In our experiments, binary radiance field\nrepresentation successfully outperforms the reconstruction performance of\nstate-of-the-art (SOTA) efficient radiance field models with lower storage\nallocation. In particular, our model achieves impressive results in static\nscene reconstruction, with a PSNR of 32.03 dB for Synthetic-NeRF scenes, 34.48\ndB for Synthetic-NSVF scenes, 28.20 dB for Tanks and Temples scenes while only\nutilizing 0.5 MB of storage space, respectively. We hope the proposed binary\nradiance field representation will make radiance fields more accessible without\na storage bottleneck.\n", "date": "2023-06-13", "categories": []}
{"id": "2311.12775", "title": "SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh\n  Reconstruction and High-Quality Mesh Rendering", "abstract": "  We propose a method to allow precise and extremely fast mesh extraction from\n3D Gaussian Splatting. Gaussian Splatting has recently become very popular as\nit yields realistic rendering while being significantly faster to train than\nNeRFs. It is however challenging to extract a mesh from the millions of tiny 3D\ngaussians as these gaussians tend to be unorganized after optimization and no\nmethod has been proposed so far. Our first key contribution is a regularization\nterm that encourages the gaussians to align well with the surface of the scene.\nWe then introduce a method that exploits this alignment to extract a mesh from\nthe Gaussians using Poisson reconstruction, which is fast, scalable, and\npreserves details, in contrast to the Marching Cubes algorithm usually applied\nto extract meshes from Neural SDFs. Finally, we introduce an optional\nrefinement strategy that binds gaussians to the surface of the mesh, and\njointly optimizes these Gaussians and the mesh through Gaussian splatting\nrendering. This enables easy editing, sculpting, rigging, animating,\ncompositing and relighting of the Gaussians using traditional softwares by\nmanipulating the mesh instead of the gaussians themselves. Retrieving such an\neditable mesh for realistic rendering is done within minutes with our method,\ncompared to hours with the state-of-the-art methods on neural SDFs, while\nproviding a better rendering quality. Our project page is the following:\nhttps://anttwo.github.io/sugar/\n", "date": "2023-11-21", "categories": []}
{"id": "2311.18068", "title": "ALSTER: A Local Spatio-Temporal Expert for Online 3D Semantic\n  Reconstruction", "abstract": "  We propose an online 3D semantic segmentation method that incrementally\nreconstructs a 3D semantic map from a stream of RGB-D frames. Unlike offline\nmethods, ours is directly applicable to scenarios with real-time constraints,\nsuch as robotics or mixed reality. To overcome the inherent challenges of\nonline methods, we make two main contributions. First, to effectively extract\ninformation from the input RGB-D video stream, we jointly estimate geometry and\nsemantic labels per frame in 3D. A key focus of our approach is to reason about\nsemantic entities both in the 2D input and the local 3D domain to leverage\ndifferences in spatial context and network architectures. Our method predicts\n2D features using an off-the-shelf segmentation network. The extracted 2D\nfeatures are refined by a lightweight 3D network to enable reasoning about the\nlocal 3D structure. Second, to efficiently deal with an infinite stream of\ninput RGB-D frames, a subsequent network serves as a temporal expert predicting\nthe incremental scene updates by leveraging 2D, 3D, and past information in a\nlearned manner. These updates are then integrated into a global scene\nrepresentation. Using these main contributions, our method can enable scenarios\nwith real-time constraints and can scale to arbitrary scene sizes by processing\nand updating the scene only in a local region defined by the new measurement.\nOur experiments demonstrate improved results compared to existing online\nmethods that purely operate in local regions and show that complementary\nsources of information can boost the performance. We provide a thorough\nablation study on the benefits of different architectural as well as\nalgorithmic design decisions. Our method yields competitive results on the\npopular ScanNet benchmark and SceneNN dataset.\n", "date": "2023-11-29", "categories": []}
{"id": "2311.18695", "title": "Seg2Reg: Differentiable 2D Segmentation to 1D Regression Rendering for\n  360 Room Layout Reconstruction", "abstract": "  State-of-the-art single-view 360-degree room layout reconstruction methods\nformulate the problem as a high-level 1D (per-column) regression task. On the\nother hand, traditional low-level 2D layout segmentation is simpler to learn\nand can represent occluded regions, but it requires complex post-processing for\nthe targeting layout polygon and sacrifices accuracy. We present Seg2Reg to\nrender 1D layout depth regression from the 2D segmentation map in a\ndifferentiable and occlusion-aware way, marrying the merits of both sides.\nSpecifically, our model predicts floor-plan density for the input\nequirectangular 360-degree image. Formulating the 2D layout representation as a\ndensity field enables us to employ `flattened' volume rendering to form 1D\nlayout depth regression. In addition, we propose a novel 3D warping\naugmentation on layout to improve generalization. Finally, we re-implement\nrecent room layout reconstruction methods into our codebase for benchmarking\nand explore modern backbones and training techniques to serve as the strong\nbaseline. Our model significantly outperforms previous arts. The code will be\nmade available upon publication.\n", "date": "2023-11-30", "categories": []}
{"id": "2311.17878", "title": "TSDF-Sampling: Efficient Sampling for Neural Surface Field using\n  Truncated Signed Distance Field", "abstract": "  Multi-view neural surface reconstruction has exhibited impressive results.\nHowever, a notable limitation is the prohibitively slow inference time when\ncompared to traditional techniques, primarily attributed to the dense sampling,\nrequired to maintain the rendering quality. This paper introduces a novel\napproach that substantially reduces the number of samplings by incorporating\nthe Truncated Signed Distance Field (TSDF) of the scene. While prior works have\nproposed importance sampling, their dependence on initial uniform samples over\nthe entire space makes them unable to avoid performance degradation when trying\nto use less number of samples. In contrast, our method leverages the TSDF\nvolume generated only by the trained views, and it proves to provide a\nreasonable bound on the sampling from upcoming novel views. As a result, we\nachieve high rendering quality by fully exploiting the continuous neural SDF\nestimation within the bounds given by the TSDF volume. Notably, our method is\nthe first approach that can be robustly plug-and-play into a diverse array of\nneural surface field models, as long as they use the volume rendering\ntechnique. Our empirical results show an 11-fold increase in inference speed\nwithout compromising performance. The result videos are available at our\nproject page: https://tsdf-sampling.github.io/\n", "date": "2023-11-29", "categories": []}
{"id": "2306.15670", "title": "Symphonize 3D Semantic Scene Completion with Contextual Instance Queries", "abstract": "  `3D Semantic Scene Completion (SSC) has emerged as a nascent and pivotal\nundertaking in autonomous driving, aiming to predict voxel occupancy within\nvolumetric scenes. However, prevailing methodologies primarily focus on\nvoxel-wise feature aggregation, while neglecting instance semantics and scene\ncontext. In this paper, we present a novel paradigm termed Symphonies\n(Scene-from-Insts), that delves into the integration of instance queries to\norchestrate 2D-to-3D reconstruction and 3D scene modeling. Leveraging our\nproposed Serial Instance-Propagated Attentions, Symphonies dynamically encodes\ninstance-centric semantics, facilitating intricate interactions between\nimage-based and volumetric domains. Simultaneously, Symphonies enables holistic\nscene comprehension by capturing context through the efficient fusion of\ninstance queries, alleviating geometric ambiguity such as occlusion and\nperspective errors through contextual scene reasoning. Experimental results\ndemonstrate that Symphonies achieves state-of-the-art performance on\nchallenging benchmarks SemanticKITTI and SSCBench-KITTI-360, yielding\nremarkable mIoU scores of 15.04 and 18.58, respectively. These results showcase\nthe paradigm's promising advancements. The code is available at\nhttps://github.com/hustvl/Symphonies.\n", "date": "2023-06-27", "categories": []}
{"id": "2307.09112", "title": "NU-MCC: Multiview Compressive Coding with Neighborhood Decoder and\n  Repulsive UDF", "abstract": "  Remarkable progress has been made in 3D reconstruction from single-view RGB-D\ninputs. MCC is the current state-of-the-art method in this field, which\nachieves unprecedented success by combining vision Transformers with\nlarge-scale training. However, we identified two key limitations of MCC: 1) The\nTransformer decoder is inefficient in handling large number of query points; 2)\nThe 3D representation struggles to recover high-fidelity details. In this\npaper, we propose a new approach called NU-MCC that addresses these\nlimitations. NU-MCC includes two key innovations: a Neighborhood decoder and a\nRepulsive Unsigned Distance Function (Repulsive UDF). First, our Neighborhood\ndecoder introduces center points as an efficient proxy of input visual\nfeatures, allowing each query point to only attend to a small neighborhood.\nThis design not only results in much faster inference speed but also enables\nthe exploitation of finer-scale visual features for improved recovery of 3D\ntextures. Second, our Repulsive UDF is a novel alternative to the occupancy\nfield used in MCC, significantly improving the quality of 3D object\nreconstruction. Compared to standard UDFs that suffer from holes in results,\nour proposed Repulsive UDF can achieve more complete surface reconstruction.\nExperimental results demonstrate that NU-MCC is able to learn a strong 3D\nrepresentation, significantly advancing the state of the art in single-view 3D\nreconstruction. Particularly, it outperforms MCC by 9.7% in terms of the\nF1-score on the CO3D-v2 dataset with more than 5x faster running speed.\n", "date": "2023-07-18", "categories": []}
{"id": "2309.13101", "title": "Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene\n  Reconstruction", "abstract": "  Implicit neural representation has paved the way for new approaches to\ndynamic scene reconstruction and rendering. Nonetheless, cutting-edge dynamic\nneural rendering methods rely heavily on these implicit representations, which\nfrequently struggle to capture the intricate details of objects in the scene.\nFurthermore, implicit methods have difficulty achieving real-time rendering in\ngeneral dynamic scenes, limiting their use in a variety of tasks. To address\nthe issues, we propose a deformable 3D Gaussians Splatting method that\nreconstructs scenes using 3D Gaussians and learns them in canonical space with\na deformation field to model monocular dynamic scenes. We also introduce an\nannealing smoothing training mechanism with no extra overhead, which can\nmitigate the impact of inaccurate poses on the smoothness of time interpolation\ntasks in real-world datasets. Through a differential Gaussian rasterizer, the\ndeformable 3D Gaussians not only achieve higher rendering quality but also\nreal-time rendering speed. Experiments show that our method outperforms\nexisting methods significantly in terms of both rendering quality and speed,\nmaking it well-suited for tasks such as novel-view synthesis, time\ninterpolation, and real-time rendering.\n", "date": "2023-09-22", "categories": []}
{"id": "2311.12062", "title": "PBWR: Parametric Building Wireframe Reconstruction from Aerial LiDAR\n  Point Clouds", "abstract": "  In this paper, we present an end-to-end 3D building wireframe reconstruction\nmethod to regress edges directly from aerial LiDAR point clouds.Our method,\nnamed Parametric Building Wireframe Reconstruction (PBWR), takes aerial LiDAR\npoint clouds and initial edge entities as input, and fully uses self-attention\nmechanism of transformers to regress edge parameters without any intermediate\nsteps such as corner prediction. We propose an edge non-maximum suppression\n(E-NMS) module based on edge similarityto remove redundant edges. Additionally,\na dedicated edge loss function is utilized to guide the PBWR in regressing\nedges parameters, where simple use of edge distance loss isn't suitable. In our\nexperiments, we demonstrate state-of-the-art results on the Building3D dataset,\nachieving an improvement of approximately 36% in entry-level dataset edge\naccuracy and around 42% improvement in the Tallinn dataset.\n", "date": "2023-11-18", "categories": []}
{"id": "2307.05832", "title": "Bag of Views: An Appearance-based Approach to Next-Best-View Planning\n  for 3D Reconstruction", "abstract": "  UAV-based intelligent data acquisition for 3D reconstruction and monitoring\nof infrastructure has experienced an increasing surge of interest due to recent\nadvancements in image processing and deep learning-based techniques. View\nplanning is an essential part of this task that dictates the information\ncapture strategy and heavily impacts the quality of the 3D model generated from\nthe captured data. Recent methods have used prior knowledge or partial\nreconstruction of the target to accomplish view planning for active\nreconstruction; the former approach poses a challenge for complex or newly\nidentified targets while the latter is computationally expensive. In this work,\nwe present Bag-of-Views (BoV), a fully appearance-based model used to assign\nutility to the captured views for both offline dataset refinement and online\nnext-best-view (NBV) planning applications targeting the task of 3D\nreconstruction. With this contribution, we also developed the View Planning\nToolbox (VPT), a lightweight package for training and testing machine\nlearning-based view planning frameworks, custom view dataset generation of\narbitrary 3D scenes, and 3D reconstruction. Through experiments which pair a\nBoV-based reinforcement learning model with VPT, we demonstrate the efficacy of\nour model in reducing the number of required views for high-quality\nreconstructions in dataset refinement and NBV planning.\n", "date": "2023-07-11", "categories": []}
{"id": "2212.05231", "title": "NeuS2: Fast Learning of Neural Implicit Surfaces for Multi-view\n  Reconstruction", "abstract": "  Recent methods for neural surface representation and rendering, for example\nNeuS, have demonstrated the remarkably high-quality reconstruction of static\nscenes. However, the training of NeuS takes an extremely long time (8 hours),\nwhich makes it almost impossible to apply them to dynamic scenes with thousands\nof frames. We propose a fast neural surface reconstruction approach, called\nNeuS2, which achieves two orders of magnitude improvement in terms of\nacceleration without compromising reconstruction quality. To accelerate the\ntraining process, we parameterize a neural surface representation by\nmulti-resolution hash encodings and present a novel lightweight calculation of\nsecond-order derivatives tailored to our networks to leverage CUDA parallelism,\nachieving a factor two speed up. To further stabilize and expedite training, a\nprogressive learning strategy is proposed to optimize multi-resolution hash\nencodings from coarse to fine. We extend our method for fast training of\ndynamic scenes, with a proposed incremental training strategy and a novel\nglobal transformation prediction component, which allow our method to handle\nchallenging long sequences with large movements and deformations. Our\nexperiments on various datasets demonstrate that NeuS2 significantly\noutperforms the state-of-the-arts in both surface reconstruction accuracy and\ntraining speed for both static and dynamic scenes. The code is available at our\nwebsite: https://vcai.mpi-inf.mpg.de/projects/NeuS2/ .\n", "date": "2022-12-10", "categories": []}
{"id": "2301.05206", "title": "ImMesh: An Immediate LiDAR Localization and Meshing Framework", "abstract": "  In this paper, we propose a novel LiDAR(-inertial) odometry and mapping\nframework to achieve the goal of simultaneous localization and meshing in\nreal-time. This proposed framework termed ImMesh comprises four tightly-coupled\nmodules: receiver, localization, meshing, and broadcaster. The localization\nmodule utilizes the prepossessed sensor data from the receiver, estimates the\nsensor pose online by registering LiDAR scans to maps, and dynamically grows\nthe map. Then, our meshing module takes the registered LiDAR scan for\nincrementally reconstructing the triangle mesh on the fly. Finally, the\nreal-time odometry, map, and mesh are published via our broadcaster. The key\ncontribution of this work is the meshing module, which represents a scene by an\nefficient hierarchical voxels structure, performs fast finding of voxels\nobserved by new scans, and reconstructs triangle facets in each voxel in an\nincremental manner. This voxel-wise meshing operation is delicately designed\nfor the purpose of efficiency; it first performs a dimension reduction by\nprojecting 3D points to a 2D local plane contained in the voxel, and then\nexecutes the meshing operation with pull, commit and push steps for incremental\nreconstruction of triangle facets. To the best of our knowledge, this is the\nfirst work in literature that can reconstruct online the triangle mesh of\nlarge-scale scenes, just relying on a standard CPU without GPU acceleration. To\nshare our findings and make contributions to the community, we make our code\npublicly available on our GitHub: https://github.com/hku-mars/ImMesh.\n", "date": "2023-01-12", "categories": []}
{"id": "2311.06659", "title": "3DFusion, A real-time 3D object reconstruction pipeline based on\n  streamed instance segmented data", "abstract": "  This paper presents a real-time segmentation and reconstruction system that\nutilizes RGB-D images to generate accurate and detailed individual 3D models of\nobjects within a captured scene. Leveraging state-of-the-art instance\nsegmentation techniques, the system performs pixel-level segmentation on RGB-D\ndata, effectively separating foreground objects from the background. The\nsegmented objects are then reconstructed into distinct 3D models in a\nhigh-performance computation platform. The real-time 3D modelling can be\napplied across various domains, including augmented/virtual reality, interior\ndesign, urban planning, road assistance, security systems, and more. To achieve\nreal-time performance, the paper proposes a method that effectively samples\nconsecutive frames to reduce network load while ensuring reconstruction\nquality. Additionally, a multi-process SLAM pipeline is adopted for parallel 3D\nreconstruction, enabling efficient cutting of the clustering objects into\nindividuals. This system employs the industry-leading framework YOLO for\ninstance segmentation. To improve YOLO's performance and accuracy,\nmodifications were made to resolve duplicated or false detection of similar\nobjects, ensuring the reconstructed models align with the targets. Overall,\nthis work establishes a robust real-time system with a significant enhancement\nfor object segmentation and reconstruction in the indoor environment. It can\npotentially be extended to the outdoor scenario, opening up numerous\nopportunities for real-world applications.\n", "date": "2023-11-11", "categories": []}
{"id": "2307.03404", "title": "RGB-D Mapping and Tracking in a Plenoxel Radiance Field", "abstract": "  The widespread adoption of Neural Radiance Fields (NeRFs) have ensured\nsignificant advances in the domain of novel view synthesis in recent years.\nThese models capture a volumetric radiance field of a scene, creating highly\nconvincing, dense, photorealistic models through the use of simple,\ndifferentiable rendering equations. Despite their popularity, these algorithms\nsuffer from severe ambiguities in visual data inherent to the RGB sensor, which\nmeans that although images generated with view synthesis can visually appear\nvery believable, the underlying 3D model will often be wrong. This considerably\nlimits the usefulness of these models in practical applications like Robotics\nand Extended Reality (XR), where an accurate dense 3D reconstruction otherwise\nwould be of significant value. In this paper, we present the vital differences\nbetween view synthesis models and 3D reconstruction models. We also comment on\nwhy a depth sensor is essential for modeling accurate geometry in general\noutward-facing scenes using the current paradigm of novel view synthesis\nmethods. Focusing on the structure-from-motion task, we practically demonstrate\nthis need by extending the Plenoxel radiance field model: Presenting an\nanalytical differential approach for dense mapping and tracking with radiance\nfields based on RGB-D data without a neural network. Our method achieves\nstate-of-the-art results in both mapping and tracking tasks, while also being\nfaster than competing neural network-based approaches. The code is available\nat: https://github.com/ysus33/RGB-D_Plenoxel_Mapping_Tracking.git\n", "date": "2023-07-07", "categories": []}
{"id": "2305.16914", "title": "PlaNeRF: SVD Unsupervised 3D Plane Regularization for NeRF Large-Scale\n  Scene Reconstruction", "abstract": "  Neural Radiance Fields (NeRF) enable 3D scene reconstruction from 2D images\nand camera poses for Novel View Synthesis (NVS). Although NeRF can produce\nphotorealistic results, it often suffers from overfitting to training views,\nleading to poor geometry reconstruction, especially in low-texture areas. This\nlimitation restricts many important applications which require accurate\ngeometry, such as extrapolated NVS, HD mapping and scene editing. To address\nthis limitation, we propose a new method to improve NeRF's 3D structure using\nonly RGB images and semantic maps. Our approach introduces a novel plane\nregularization based on Singular Value Decomposition (SVD), that does not rely\non any geometric prior. In addition, we leverage the Structural Similarity\nIndex Measure (SSIM) in our loss design to properly initialize the volumetric\nrepresentation of NeRF. Quantitative and qualitative results show that our\nmethod outperforms popular regularization approaches in accurate geometry\nreconstruction for large-scale outdoor scenes and achieves SoTA rendering\nquality on the KITTI-360 NVS benchmark.\n", "date": "2023-05-26", "categories": []}
{"id": "2306.11739", "title": "Multi-view 3D Object Reconstruction and Uncertainty Modelling with\n  Neural Shape Prior", "abstract": "  3D object reconstruction is important for semantic scene understanding. It is\nchallenging to reconstruct detailed 3D shapes from monocular images directly\ndue to a lack of depth information, occlusion and noise. Most current methods\ngenerate deterministic object models without any awareness of the uncertainty\nof the reconstruction. We tackle this problem by leveraging a neural object\nrepresentation which learns an object shape distribution from large dataset of\n3d object models and maps it into a latent space. We propose a method to model\nuncertainty as part of the representation and define an uncertainty-aware\nencoder which generates latent codes with uncertainty directly from individual\ninput images. Further, we propose a method to propagate the uncertainty in the\nlatent code to SDF values and generate a 3d object mesh with local uncertainty\nfor each mesh component. Finally, we propose an incremental fusion method under\na Bayesian framework to fuse the latent codes from multi-view observations. We\nevaluate the system in both synthetic and real datasets to demonstrate the\neffectiveness of uncertainty-based fusion to improve 3D object reconstruction\naccuracy.\n", "date": "2023-06-17", "categories": []}
{"id": "2311.02121", "title": "Enhancing Monocular Height Estimation from Aerial Images with\n  Street-view Images", "abstract": "  Accurate height estimation from monocular aerial imagery presents a\nsignificant challenge due to its inherently ill-posed nature. This limitation\nis rooted in the absence of adequate geometric constraints available to the\nmodel when training with monocular imagery. Without additional geometric\ninformation to supplement the monocular image data, the model's ability to\nprovide reliable estimations is compromised.\n  In this paper, we propose a method that enhances monocular height estimation\nby incorporating street-view images. Our insight is that street-view images\nprovide a distinct viewing perspective and rich structural details of the\nscene, serving as geometric constraints to enhance the performance of monocular\nheight estimation. Specifically, we aim to optimize an implicit 3D scene\nrepresentation, density field, with geometry constraints from street-view\nimages, thereby improving the accuracy and robustness of height estimation. Our\nexperimental results demonstrate the effectiveness of our proposed method,\noutperforming the baseline and offering significant improvements in terms of\naccuracy and structural consistency.\n", "date": "2023-11-03", "categories": []}
{"id": "2311.02552", "title": "IPVNet: Learning Implicit Point-Voxel Features for Open-Surface 3D\n  Reconstruction", "abstract": "  Reconstruction of 3D open surfaces (e.g., non-watertight meshes) is an\nunderexplored area of computer vision. Recent learning-based implicit\ntechniques have removed previous barriers by enabling reconstruction in\narbitrary resolutions. Yet, such approaches often rely on distinguishing\nbetween the inside and outside of a surface in order to extract a zero level\nset when reconstructing the target. In the case of open surfaces, this\ndistinction often leads to artifacts such as the artificial closing of surface\ngaps. However, real-world data may contain intricate details defined by salient\nsurface gaps. Implicit functions that regress an unsigned distance field have\nshown promise in reconstructing such open surfaces. Nonetheless, current\nunsigned implicit methods rely on a discretized representation of the raw data.\nThis not only bounds the learning process to the representation's resolution,\nbut it also introduces outliers in the reconstruction. To enable accurate\nreconstruction of open surfaces without introducing outliers, we propose a\nlearning-based implicit point-voxel model (IPVNet). IPVNet predicts the\nunsigned distance between a surface and a query point in 3D space by leveraging\nboth raw point cloud data and its discretized voxel counterpart. Experiments on\nsynthetic and real-world public datasets demonstrates that IPVNet outperforms\nthe state of the art while producing far fewer outliers in the resulting\nreconstruction.\n", "date": "2023-11-05", "categories": []}
{"id": "2311.02601", "title": "Optimizing Implicit Neural Representations from Point Clouds via\n  Energy-Based Models", "abstract": "  Reconstructing a continuous surface from an unoritented 3D point cloud is a\nfundamental task in 3D shape processing. In recent years, several methods have\nbeen proposed to address this problem using implicit neural representations\n(INRs). In this study, we propose a method to optimize INRs using energy-based\nmodels (EBMs). By employing the absolute value of the coordinate-based neural\nnetworks as the energy function, the INR can be optimized through the\nestimation of the point cloud distribution by the EBM. In addition, appropriate\nparameter settings of the EBM enable the model to consider the magnitude of\npoint cloud noise. Our experiments confirmed that the proposed method is more\nrobust against point cloud noise than conventional surface reconstruction\nmethods.\n", "date": "2023-11-05", "categories": []}
{"id": "2306.04699", "title": "DiViNeT: 3D Reconstruction from Disparate Views via Neural Template\n  Regularization", "abstract": "  We present a volume rendering-based neural surface reconstruction method that\ntakes as few as three disparate RGB images as input. Our key idea is to\nregularize the reconstruction, which is severely ill-posed and leaving\nsignificant gaps between the sparse views, by learning a set of neural\ntemplates to act as surface priors. Our method, coined DiViNet, operates in two\nstages. It first learns the templates, in the form of 3D Gaussian functions,\nacross different scenes, without 3D supervision. In the reconstruction stage,\nour predicted templates serve as anchors to help \"stitch'' the surfaces over\nsparse regions. We demonstrate that our approach is not only able to complete\nthe surface geometry but also reconstructs surface details to a reasonable\nextent from a few disparate input views. On the DTU and BlendedMVS datasets,\nour approach achieves the best reconstruction quality among existing methods in\nthe presence of such sparse views and performs on par, if not better, with\ncompeting methods when dense views are employed as inputs.\n", "date": "2023-06-07", "categories": []}
{"id": "2311.01292", "title": "Joint 3D Shape and Motion Estimation from Rolling Shutter Light-Field\n  Images", "abstract": "  In this paper, we propose an approach to address the problem of 3D\nreconstruction of scenes from a single image captured by a light-field camera\nequipped with a rolling shutter sensor. Our method leverages the 3D information\ncues present in the light-field and the motion information provided by the\nrolling shutter effect. We present a generic model for the imaging process of\nthis sensor and a two-stage algorithm that minimizes the re-projection error\nwhile considering the position and motion of the camera in a motion-shape\nbundle adjustment estimation strategy. Thereby, we provide an instantaneous 3D\nshape-and-pose-and-velocity sensing paradigm. To the best of our knowledge,\nthis is the first study to leverage this type of sensor for this purpose. We\nalso present a new benchmark dataset composed of different light-fields showing\nrolling shutter effects, which can be used as a common base to improve the\nevaluation and tracking the progress in the field. We demonstrate the\neffectiveness and advantages of our approach through several experiments\nconducted for different scenes and types of motions. The source code and\ndataset are publicly available at: https://github.com/ICB-Vision-AI/RSLF\n", "date": "2023-11-02", "categories": []}
{"id": "2311.00457", "title": "Single-view 3D Scene Reconstruction with High-fidelity Shape and Texture", "abstract": "  Reconstructing detailed 3D scenes from single-view images remains a\nchallenging task due to limitations in existing approaches, which primarily\nfocus on geometric shape recovery, overlooking object appearances and fine\nshape details. To address these challenges, we propose a novel framework for\nsimultaneous high-fidelity recovery of object shapes and textures from\nsingle-view images. Our approach utilizes the proposed Single-view neural\nimplicit Shape and Radiance field (SSR) representations to leverage both\nexplicit 3D shape supervision and volume rendering of color, depth, and surface\nnormal images. To overcome shape-appearance ambiguity under partial\nobservations, we introduce a two-stage learning curriculum incorporating both\n3D and 2D supervisions. A distinctive feature of our framework is its ability\nto generate fine-grained textured meshes while seamlessly integrating rendering\ncapabilities into the single-view 3D reconstruction model. This integration\nenables not only improved textured 3D object reconstruction by 27.7% and 11.6%\non the 3D-FRONT and Pix3D datasets, respectively, but also supports the\nrendering of images from novel viewpoints. Beyond individual objects, our\napproach facilitates composing object-level representations into flexible scene\nrepresentations, thereby enabling applications such as holistic scene\nunderstanding and 3D scene editing. We conduct extensive experiments to\ndemonstrate the effectiveness of our method.\n", "date": "2023-11-01", "categories": []}
{"id": "2310.20095", "title": "$p$-Poisson surface reconstruction in curl-free flow from point clouds", "abstract": "  The aim of this paper is the reconstruction of a smooth surface from an\nunorganized point cloud sampled by a closed surface, with the preservation of\ngeometric shapes, without any further information other than the point cloud.\nImplicit neural representations (INRs) have recently emerged as a promising\napproach to surface reconstruction. However, the reconstruction quality of\nexisting methods relies on ground truth implicit function values or surface\nnormal vectors. In this paper, we show that proper supervision of partial\ndifferential equations and fundamental properties of differential vector fields\nare sufficient to robustly reconstruct high-quality surfaces. We cast the\n$p$-Poisson equation to learn a signed distance function (SDF) and the\nreconstructed surface is implicitly represented by the zero-level set of the\nSDF. For efficient training, we develop a variable splitting structure by\nintroducing a gradient of the SDF as an auxiliary variable and impose the\n$p$-Poisson equation directly on the auxiliary variable as a hard constraint.\nBased on the curl-free property of the gradient field, we impose a curl-free\nconstraint on the auxiliary variable, which leads to a more faithful\nreconstruction. Experiments on standard benchmark datasets show that the\nproposed INR provides a superior and robust reconstruction. The code is\navailable at \\url{https://github.com/Yebbi/PINC}.\n", "date": "2023-10-31", "categories": []}
{"id": "2310.16831", "title": "PERF: Panoramic Neural Radiance Field from a Single Panorama", "abstract": "  Neural Radiance Field (NeRF) has achieved substantial progress in novel view\nsynthesis given multi-view images. Recently, some works have attempted to train\na NeRF from a single image with 3D priors. They mainly focus on a limited field\nof view with a few occlusions, which greatly limits their scalability to\nreal-world 360-degree panoramic scenarios with large-size occlusions. In this\npaper, we present PERF, a 360-degree novel view synthesis framework that trains\na panoramic neural radiance field from a single panorama. Notably, PERF allows\n3D roaming in a complex scene without expensive and tedious image collection.\nTo achieve this goal, we propose a novel collaborative RGBD inpainting method\nand a progressive inpainting-and-erasing method to lift up a 360-degree 2D\nscene to a 3D scene. Specifically, we first predict a panoramic depth map as\ninitialization given a single panorama and reconstruct visible 3D regions with\nvolume rendering. Then we introduce a collaborative RGBD inpainting approach\ninto a NeRF for completing RGB images and depth maps from random views, which\nis derived from an RGB Stable Diffusion model and a monocular depth estimator.\nFinally, we introduce an inpainting-and-erasing strategy to avoid inconsistent\ngeometry between a newly-sampled view and reference views. The two components\nare integrated into the learning of NeRFs in a unified optimization framework\nand achieve promising results. Extensive experiments on Replica and a new\ndataset PERF-in-the-wild demonstrate the superiority of our PERF over\nstate-of-the-art methods. Our PERF can be widely used for real-world\napplications, such as panorama-to-3D, text-to-3D, and 3D scene stylization\napplications. Project page and code are available at\nhttps://perf-project.github.io/ and https://github.com/perf-project/PeRF.\n", "date": "2023-10-25", "categories": []}
{"id": "2310.19188", "title": "3DMiner: Discovering Shapes from Large-Scale Unannotated Image Datasets", "abstract": "  We present 3DMiner -- a pipeline for mining 3D shapes from challenging\nlarge-scale unannotated image datasets. Unlike other unsupervised 3D\nreconstruction methods, we assume that, within a large-enough dataset, there\nmust exist images of objects with similar shapes but varying backgrounds,\ntextures, and viewpoints. Our approach leverages the recent advances in\nlearning self-supervised image representations to cluster images with\ngeometrically similar shapes and find common image correspondences between\nthem. We then exploit these correspondences to obtain rough camera estimates as\ninitialization for bundle-adjustment. Finally, for every image cluster, we\napply a progressive bundle-adjusting reconstruction method to learn a neural\noccupancy field representing the underlying shape. We show that this procedure\nis robust to several types of errors introduced in previous steps (e.g., wrong\ncamera poses, images containing dissimilar shapes, etc.), allowing us to obtain\nshape and pose annotations for images in-the-wild. When using images from Pix3D\nchairs, our method is capable of producing significantly better results than\nstate-of-the-art unsupervised 3D reconstruction techniques, both quantitatively\nand qualitatively. Furthermore, we show how 3DMiner can be applied to\nin-the-wild data by reconstructing shapes present in images from the LAION-5B\ndataset. Project Page: https://ttchengab.github.io/3dminerOfficial\n", "date": "2023-10-29", "categories": []}
{"id": "2310.15712", "title": "GNeSF: Generalizable Neural Semantic Fields", "abstract": "  3D scene segmentation based on neural implicit representation has emerged\nrecently with the advantage of training only on 2D supervision. However,\nexisting approaches still requires expensive per-scene optimization that\nprohibits generalization to novel scenes during inference. To circumvent this\nproblem, we introduce a generalizable 3D segmentation framework based on\nimplicit representation. Specifically, our framework takes in multi-view image\nfeatures and semantic maps as the inputs instead of only spatial information to\navoid overfitting to scene-specific geometric and semantic information. We\npropose a novel soft voting mechanism to aggregate the 2D semantic information\nfrom different views for each 3D point. In addition to the image features, view\ndifference information is also encoded in our framework to predict the voting\nscores. Intuitively, this allows the semantic information from nearby views to\ncontribute more compared to distant ones. Furthermore, a visibility module is\nalso designed to detect and filter out detrimental information from occluded\nviews. Due to the generalizability of our proposed method, we can synthesize\nsemantic maps or conduct 3D semantic segmentation for novel scenes with solely\n2D semantic supervision. Experimental results show that our approach achieves\ncomparable performance with scene-specific approaches. More importantly, our\napproach can even outperform existing strong supervision-based approaches with\nonly 2D annotations. Our source code is available at:\nhttps://github.com/HLinChen/GNeSF.\n", "date": "2023-10-24", "categories": []}
{"id": "2310.17527", "title": "Masked Space-Time Hash Encoding for Efficient Dynamic Scene\n  Reconstruction", "abstract": "  In this paper, we propose the Masked Space-Time Hash encoding (MSTH), a novel\nmethod for efficiently reconstructing dynamic 3D scenes from multi-view or\nmonocular videos. Based on the observation that dynamic scenes often contain\nsubstantial static areas that result in redundancy in storage and computations,\nMSTH represents a dynamic scene as a weighted combination of a 3D hash encoding\nand a 4D hash encoding. The weights for the two components are represented by a\nlearnable mask which is guided by an uncertainty-based objective to reflect the\nspatial and temporal importance of each 3D position. With this design, our\nmethod can reduce the hash collision rate by avoiding redundant queries and\nmodifications on static areas, making it feasible to represent a large number\nof space-time voxels by hash tables with small size.Besides, without the\nrequirements to fit the large numbers of temporally redundant features\nindependently, our method is easier to optimize and converge rapidly with only\ntwenty minutes of training for a 300-frame dynamic scene.As a result, MSTH\nobtains consistently better results than previous methods with only 20 minutes\nof training time and 130 MB of memory storage. Code is available at\nhttps://github.com/masked-spacetime-hashing/msth\n", "date": "2023-10-26", "categories": []}
{"id": "2308.14371", "title": "SuperUDF: Self-supervised UDF Estimation for Surface Reconstruction", "abstract": "  Learning-based surface reconstruction based on unsigned distance functions\n(UDF) has many advantages such as handling open surfaces. We propose SuperUDF,\na self-supervised UDF learning which exploits a learned geometry prior for\nefficient training and a novel regularization for robustness to sparse\nsampling. The core idea of SuperUDF draws inspiration from the classical\nsurface approximation operator of locally optimal projection (LOP). The key\ninsight is that if the UDF is estimated correctly, the 3D points should be\nlocally projected onto the underlying surface following the gradient of the\nUDF. Based on that, a number of inductive biases on UDF geometry and a\npre-learned geometry prior are devised to learn UDF estimation efficiently. A\nnovel regularization loss is proposed to make SuperUDF robust to sparse\nsampling. Furthermore, we also contribute a learning-based mesh extraction from\nthe estimated UDFs. Extensive evaluations demonstrate that SuperUDF outperforms\nthe state of the arts on several public datasets in terms of both quality and\nefficiency. Code url is https://github.com/THHHomas/SuperUDF.\n", "date": "2023-08-28", "categories": []}
{"id": "2310.14700", "title": "Interaction-Driven Active 3D Reconstruction with Object Interiors", "abstract": "  We introduce an active 3D reconstruction method which integrates visual\nperception, robot-object interaction, and 3D scanning to recover both the\nexterior and interior, i.e., unexposed, geometries of a target 3D object.\nUnlike other works in active vision which focus on optimizing camera viewpoints\nto better investigate the environment, the primary feature of our\nreconstruction is an analysis of the interactability of various parts of the\ntarget object and the ensuing part manipulation by a robot to enable scanning\nof occluded regions. As a result, an understanding of part articulations of the\ntarget object is obtained on top of complete geometry acquisition. Our method\noperates fully automatically by a Fetch robot with built-in RGBD sensors. It\niterates between interaction analysis and interaction-driven reconstruction,\nscanning and reconstructing detected moveable parts one at a time, where both\nthe articulated part detection and mesh reconstruction are carried out by\nneural networks. In the final step, all the remaining, non-articulated parts,\nincluding all the interior structures that had been exposed by prior part\nmanipulations and subsequently scanned, are reconstructed to complete the\nacquisition. We demonstrate the performance of our method via qualitative and\nquantitative evaluation, ablation studies, comparisons to alternatives, as well\nas experiments in a real environment.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.13373", "title": "Single-view 3D reconstruction via inverse procedural modeling", "abstract": "  We propose an approach to 3D reconstruction via inverse procedural modeling\nand investigate two variants of this approach. The first option consists in the\nfitting set of input parameters using a genetic algorithm. We demonstrate the\nresults of our work on tree models, complex objects, with the reconstruction of\nwhich most existing methods cannot handle. The second option allows us to\nsignificantly improve the precision by using gradients within memetic\nalgorithm, differentiable rendering and also differentiable procedural\ngenerators. In our work we see 2 main contributions. First, we propose a method\nto join differentiable rendering and inverse procedural modeling. This gives us\nan opportunity to reconstruct 3D model more accurately than existing approaches\nwhen a small number of input images are available (even for single image).\nSecond, we join both differentiable and non-differentiable procedural\ngenerators in a single framework which allow us to apply inverse procedural\nmodeling to fairly complex generators: when gradient is available,\nreconstructions is precise, when gradient is not available, reconstruction is\napproximate, but always high quality without visual artifacts.\n", "date": "2023-10-20", "categories": []}
{"id": "2310.09956", "title": "Tabletop Transparent Scene Reconstruction via Epipolar-Guided Optical\n  Flow with Monocular Depth Completion Prior", "abstract": "  Reconstructing transparent objects using affordable RGB-D cameras is a\npersistent challenge in robotic perception due to inconsistent appearances\nacross views in the RGB domain and inaccurate depth readings in each\nsingle-view. We introduce a two-stage pipeline for reconstructing transparent\nobjects tailored for mobile platforms. In the first stage, off-the-shelf\nmonocular object segmentation and depth completion networks are leveraged to\npredict the depth of transparent objects, furnishing single-view shape prior.\nSubsequently, we propose Epipolar-guided Optical Flow (EOF) to fuse several\nsingle-view shape priors from the first stage to a cross-view consistent 3D\nreconstruction given camera poses estimated from opaque part of the scene. Our\nkey innovation lies in EOF which employs boundary-sensitive sampling and\nepipolar-line constraints into optical flow to accurately establish 2D\ncorrespondences across multiple views on transparent objects. Quantitative\nevaluations demonstrate that our pipeline significantly outperforms baseline\nmethods in 3D reconstruction quality, paving the way for more adept robotic\nperception and interaction with transparent objects.\n", "date": "2023-10-15", "categories": []}
{"id": "2309.14616", "title": "NDC-Scene: Boost Monocular 3D Semantic Scene Completion in Normalized\n  Device Coordinates Space", "abstract": "  Monocular 3D Semantic Scene Completion (SSC) has garnered significant\nattention in recent years due to its potential to predict complex semantics and\ngeometry shapes from a single image, requiring no 3D inputs. In this paper, we\nidentify several critical issues in current state-of-the-art methods, including\nthe Feature Ambiguity of projected 2D features in the ray to the 3D space, the\nPose Ambiguity of the 3D convolution, and the Computation Imbalance in the 3D\nconvolution across different depth levels. To address these problems, we devise\na novel Normalized Device Coordinates scene completion network (NDC-Scene) that\ndirectly extends the 2D feature map to a Normalized Device Coordinates (NDC)\nspace, rather than to the world space directly, through progressive restoration\nof the dimension of depth with deconvolution operations. Experiment results\ndemonstrate that transferring the majority of computation from the target 3D\nspace to the proposed normalized device coordinates space benefits monocular\nSSC tasks. Additionally, we design a Depth-Adaptive Dual Decoder to\nsimultaneously upsample and fuse the 2D and 3D feature maps, further improving\noverall performance. Our extensive experiments confirm that the proposed method\nconsistently outperforms state-of-the-art methods on both outdoor SemanticKITTI\nand indoor NYUv2 datasets. Our code are available at\nhttps://github.com/Jiawei-Yao0812/NDCScene.\n", "date": "2023-09-26", "categories": []}
{"id": "2310.07522", "title": "S4C: Self-Supervised Semantic Scene Completion with Neural Fields", "abstract": "  3D semantic scene understanding is a fundamental challenge in computer\nvision. It enables mobile agents to autonomously plan and navigate arbitrary\nenvironments. SSC formalizes this challenge as jointly estimating dense\ngeometry and semantic information from sparse observations of a scene. Current\nmethods for SSC are generally trained on 3D ground truth based on aggregated\nLiDAR scans. This process relies on special sensors and annotation by hand\nwhich are costly and do not scale well. To overcome this issue, our work\npresents the first self-supervised approach to SSC called S4C that does not\nrely on 3D ground truth data. Our proposed method can reconstruct a scene from\na single image and only relies on videos and pseudo segmentation ground truth\ngenerated from off-the-shelf image segmentation network during training. Unlike\nexisting methods, which use discrete voxel grids, we represent scenes as\nimplicit semantic fields. This formulation allows querying any point within the\ncamera frustum for occupancy and semantic class. Our architecture is trained\nthrough rendering-based self-supervised losses. Nonetheless, our method\nachieves performance close to fully supervised state-of-the-art methods.\nAdditionally, our method demonstrates strong generalization capabilities and\ncan synthesize accurate segmentation maps for far away viewpoints.\n", "date": "2023-10-11", "categories": []}
{"id": "2307.11932", "title": "RIC: Rotate-Inpaint-Complete for Generalizable Scene Reconstruction", "abstract": "  General scene reconstruction refers to the task of estimating the full 3D\ngeometry and texture of a scene containing previously unseen objects. In many\npractical applications such as AR/VR, autonomous navigation, and robotics, only\na single view of the scene may be available, making the scene reconstruction\ntask challenging. In this paper, we present a method for scene reconstruction\nby structurally breaking the problem into two steps: rendering novel views via\ninpainting and 2D to 3D scene lifting. Specifically, we leverage the\ngeneralization capability of large visual language models (Dalle-2) to inpaint\nthe missing areas of scene color images rendered from different views. Next, we\nlift these inpainted images to 3D by predicting normals of the inpainted image\nand solving for the missing depth values. By predicting for normals instead of\ndepth directly, our method allows for robustness to changes in depth\ndistributions and scale. With rigorous quantitative evaluation, we show that\nour method outperforms multiple baselines while providing generalization to\nnovel objects and scenes.\n", "date": "2023-07-21", "categories": []}
{"id": "2310.03125", "title": "Shielding the Unseen: Privacy Protection through Poisoning NeRF with\n  Spatial Deformation", "abstract": "  In this paper, we introduce an innovative method of safeguarding user privacy\nagainst the generative capabilities of Neural Radiance Fields (NeRF) models.\nOur novel poisoning attack method induces changes to observed views that are\nimperceptible to the human eye, yet potent enough to disrupt NeRF's ability to\naccurately reconstruct a 3D scene. To achieve this, we devise a bi-level\noptimization algorithm incorporating a Projected Gradient Descent (PGD)-based\nspatial deformation. We extensively test our approach on two common NeRF\nbenchmark datasets consisting of 29 real-world scenes with high-quality images.\nOur results compellingly demonstrate that our privacy-preserving method\nsignificantly impairs NeRF's performance across these benchmark datasets.\nAdditionally, we show that our method is adaptable and versatile, functioning\nacross various perturbation strengths and NeRF architectures. This work offers\nvaluable insights into NeRF's vulnerabilities and emphasizes the need to\naccount for such potential privacy risks when developing robust 3D scene\nreconstruction algorithms. Our study contributes to the larger conversation\nsurrounding responsible AI and generative machine learning, aiming to protect\nuser privacy and respect creative ownership in the digital age.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.03578", "title": "Targeted Adversarial Attacks on Generalizable Neural Radiance Fields", "abstract": "  Neural Radiance Fields (NeRFs) have recently emerged as a powerful tool for\n3D scene representation and rendering. These data-driven models can learn to\nsynthesize high-quality images from sparse 2D observations, enabling realistic\nand interactive scene reconstructions. However, the growing usage of NeRFs in\ncritical applications such as augmented reality, robotics, and virtual\nenvironments could be threatened by adversarial attacks.\n  In this paper we present how generalizable NeRFs can be attacked by both\nlow-intensity adversarial attacks and adversarial patches, where the later\ncould be robust enough to be used in real world applications. We also\ndemonstrate targeted attacks, where a specific, predefined output scene is\ngenerated by these attack with success.\n", "date": "2023-10-05", "categories": []}
{"id": "2303.08605", "title": "RICO: Regularizing the Unobservable for Indoor Compositional\n  Reconstruction", "abstract": "  Recently, neural implicit surfaces have become popular for multi-view\nreconstruction. To facilitate practical applications like scene editing and\nmanipulation, some works extend the framework with semantic masks input for the\nobject-compositional reconstruction rather than the holistic perspective.\nThough achieving plausible disentanglement, the performance drops significantly\nwhen processing the indoor scenes where objects are usually partially observed.\nWe propose RICO to address this by regularizing the unobservable regions for\nindoor compositional reconstruction. Our key idea is to first regularize the\nsmoothness of the occluded background, which then in turn guides the foreground\nobject reconstruction in unobservable regions based on the object-background\nrelationship. Particularly, we regularize the geometry smoothness of occluded\nbackground patches. With the improved background surface, the signed distance\nfunction and the reversedly rendered depth of objects can be optimized to bound\nthem within the background range. Extensive experiments show our method\noutperforms other methods on synthetic and real-world indoor scenes and prove\nthe effectiveness of proposed regularizations. The code is available at\nhttps://github.com/kyleleey/RICO.\n", "date": "2023-03-15", "categories": []}
{"id": "2304.12317", "title": "Total-Recon: Deformable Scene Reconstruction for Embodied View Synthesis", "abstract": "  We explore the task of embodied view synthesis from monocular videos of\ndeformable scenes. Given a minute-long RGBD video of people interacting with\ntheir pets, we render the scene from novel camera trajectories derived from the\nin-scene motion of actors: (1) egocentric cameras that simulate the point of\nview of a target actor and (2) 3rd-person cameras that follow the actor.\nBuilding such a system requires reconstructing the root-body and articulated\nmotion of every actor, as well as a scene representation that supports\nfree-viewpoint synthesis. Longer videos are more likely to capture the scene\nfrom diverse viewpoints (which helps reconstruction) but are also more likely\nto contain larger motions (which complicates reconstruction). To address these\nchallenges, we present Total-Recon, the first method to photorealistically\nreconstruct deformable scenes from long monocular RGBD videos. Crucially, to\nscale to long videos, our method hierarchically decomposes the scene into the\nbackground and objects, whose motion is decomposed into carefully initialized\nroot-body motion and local articulations. To quantify such \"in-the-wild\"\nreconstruction and view synthesis, we collect ground-truth data from a\nspecialized stereo RGBD capture rig for 11 challenging videos, significantly\noutperforming prior methods. Our code, model, and data can be found at\nhttps://andrewsonga.github.io/totalrecon .\n", "date": "2023-04-24", "categories": []}
{"id": "2310.00874", "title": "PC-NeRF: Parent-Child Neural Radiance Fields under Partial Sensor Data\n  Loss in Autonomous Driving Environments", "abstract": "  Reconstructing large-scale 3D scenes is essential for autonomous vehicles,\nespecially when partial sensor data is lost. Although the recently developed\nneural radiance fields (NeRF) have shown compelling results in implicit\nrepresentations, the large-scale 3D scene reconstruction using partially lost\nLiDAR point cloud data still needs to be explored. To bridge this gap, we\npropose a novel 3D scene reconstruction framework called parent-child neural\nradiance field (PC-NeRF). The framework comprises two modules, the parent NeRF\nand the child NeRF, to simultaneously optimize scene-level, segment-level, and\npoint-level scene representations. Sensor data can be utilized more efficiently\nby leveraging the segment-level representation capabilities of child NeRFs, and\nan approximate volumetric representation of the scene can be quickly obtained\neven with limited observations. With extensive experiments, our proposed\nPC-NeRF is proven to achieve high-precision 3D reconstruction in large-scale\nscenes. Moreover, PC-NeRF can effectively tackle situations where partial\nsensor data is lost and has high deployment efficiency with limited training\ntime. Our approach implementation and the pre-trained models will be available\nat https://github.com/biter0088/pc-nerf.\n", "date": "2023-10-02", "categories": []}
{"id": "2310.01410", "title": "LEAP: Liberate Sparse-view 3D Modeling from Camera Poses", "abstract": "  Are camera poses necessary for multi-view 3D modeling? Existing approaches\npredominantly assume access to accurate camera poses. While this assumption\nmight hold for dense views, accurately estimating camera poses for sparse views\nis often elusive. Our analysis reveals that noisy estimated poses lead to\ndegraded performance for existing sparse-view 3D modeling methods. To address\nthis issue, we present LEAP, a novel pose-free approach, therefore challenging\nthe prevailing notion that camera poses are indispensable. LEAP discards\npose-based operations and learns geometric knowledge from data. LEAP is\nequipped with a neural volume, which is shared across scenes and is\nparameterized to encode geometry and texture priors. For each incoming scene,\nwe update the neural volume by aggregating 2D image features in a\nfeature-similarity-driven manner. The updated neural volume is decoded into the\nradiance field, enabling novel view synthesis from any viewpoint. On both\nobject-centric and scene-level datasets, we show that LEAP significantly\noutperforms prior methods when they employ predicted poses from\nstate-of-the-art pose estimators. Notably, LEAP performs on par with prior\napproaches that use ground-truth poses while running $400\\times$ faster than\nPixelNeRF. We show LEAP generalizes to novel object categories and scenes, and\nlearns knowledge closely resembles epipolar geometry. Project page:\nhttps://hwjiang1510.github.io/LEAP/\n", "date": "2023-10-02", "categories": []}
{"id": "2309.15164", "title": "3D Reconstruction with Generalizable Neural Fields using Scene Priors", "abstract": "  High-fidelity 3D scene reconstruction has been substantially advanced by\nrecent progress in neural fields. However, most existing methods train a\nseparate network from scratch for each individual scene. This is not scalable,\ninefficient, and unable to yield good results given limited views. While\nlearning-based multi-view stereo methods alleviate this issue to some extent,\ntheir multi-view setting makes it less flexible to scale up and to broad\napplications. Instead, we introduce training generalizable Neural Fields\nincorporating scene Priors (NFPs). The NFP network maps any single-view RGB-D\nimage into signed distance and radiance values. A complete scene can be\nreconstructed by merging individual frames in the volumetric space WITHOUT a\nfusion module, which provides better flexibility. The scene priors can be\ntrained on large-scale datasets, allowing for fast adaptation to the\nreconstruction of a new scene with fewer views. NFP not only demonstrates SOTA\nscene reconstruction performance and efficiency, but it also supports\nsingle-image novel-view synthesis, which is underexplored in neural fields.\nMore qualitative results are available at:\nhttps://oasisyang.github.io/neural-prior\n", "date": "2023-09-26", "categories": []}
{"id": "2309.16715", "title": "MV-DeepSDF: Implicit Modeling with Multi-Sweep Point Clouds for 3D\n  Vehicle Reconstruction in Autonomous Driving", "abstract": "  Reconstructing 3D vehicles from noisy and sparse partial point clouds is of\ngreat significance to autonomous driving. Most existing 3D reconstruction\nmethods cannot be directly applied to this problem because they are elaborately\ndesigned to deal with dense inputs with trivial noise. In this work, we propose\na novel framework, dubbed MV-DeepSDF, which estimates the optimal Signed\nDistance Function (SDF) shape representation from multi-sweep point clouds to\nreconstruct vehicles in the wild. Although there have been some SDF-based\nimplicit modeling methods, they only focus on single-view-based reconstruction,\nresulting in low fidelity. In contrast, we first analyze multi-sweep\nconsistency and complementarity in the latent feature space and propose to\ntransform the implicit space shape estimation problem into an element-to-set\nfeature extraction problem. Then, we devise a new architecture to extract\nindividual element-level representations and aggregate them to generate a\nset-level predicted latent code. This set-level latent code is an expression of\nthe optimal 3D shape in the implicit space, and can be subsequently decoded to\na continuous SDF of the vehicle. In this way, our approach learns consistent\nand complementary information among multi-sweeps for 3D vehicle reconstruction.\nWe conduct thorough experiments on two real-world autonomous driving datasets\n(Waymo and KITTI) to demonstrate the superiority of our approach over\nstate-of-the-art alternative methods both qualitatively and quantitatively.\n", "date": "2023-08-21", "categories": []}
{"id": "2212.01331", "title": "Surface Normal Clustering for Implicit Representation of Manhattan\n  Scenes", "abstract": "  Novel view synthesis and 3D modeling using implicit neural field\nrepresentation are shown to be very effective for calibrated multi-view\ncameras. Such representations are known to benefit from additional geometric\nand semantic supervision. Most existing methods that exploit additional\nsupervision require dense pixel-wise labels or localized scene priors. These\nmethods cannot benefit from high-level vague scene priors provided in terms of\nscenes' descriptions. In this work, we aim to leverage the geometric prior of\nManhattan scenes to improve the implicit neural radiance field representations.\nMore precisely, we assume that only the knowledge of the indoor scene (under\ninvestigation) being Manhattan is known -- with no additional information\nwhatsoever -- with an unknown Manhattan coordinate frame. Such high-level prior\nis used to self-supervise the surface normals derived explicitly in the\nimplicit neural fields. Our modeling allows us to cluster the derived normals\nand exploit their orthogonality constraints for self-supervision. Our\nexhaustive experiments on datasets of diverse indoor scenes demonstrate the\nsignificant benefit of the proposed method over the established baselines. The\nsource code is available at\nhttps://github.com/nikola3794/normal-clustering-nerf.\n", "date": "2022-12-02", "categories": []}
{"id": "2309.14824", "title": "Generalization of pixel-wise phase estimation by CNN and improvement of\n  phase-unwrapping by MRF optimization for one-shot 3D scan", "abstract": "  Active stereo technique using single pattern projection, a.k.a. one-shot 3D\nscan, have drawn a wide attention from industry, medical purposes, etc. One\nsevere drawback of one-shot 3D scan is sparse reconstruction. In addition,\nsince spatial pattern becomes complicated for the purpose of efficient\nembedding, it is easily affected by noise, which results in unstable decoding.\nTo solve the problems, we propose a pixel-wise interpolation technique for\none-shot scan, which is applicable to any types of static pattern if the\npattern is regular and periodic. This is achieved by U-net which is pre-trained\nby CG with efficient data augmentation algorithm. In the paper, to further\novercome the decoding instability, we propose a robust correspondence finding\nalgorithm based on Markov random field (MRF) optimization. We also propose a\nshape refinement algorithm based on b-spline and Gaussian kernel interpolation\nusing explicitly detected laser curves. Experiments are conducted to show the\neffectiveness of the proposed method using real data with strong noises and\ntextures.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.13240", "title": "NeRF-Enhanced Outpainting for Faithful Field-of-View Extrapolation", "abstract": "  In various applications, such as robotic navigation and remote visual\nassistance, expanding the field of view (FOV) of the camera proves beneficial\nfor enhancing environmental perception. Unlike image outpainting techniques\naimed solely at generating aesthetically pleasing visuals, these applications\ndemand an extended view that faithfully represents the scene. To achieve this,\nwe formulate a new problem of faithful FOV extrapolation that utilizes a set of\npre-captured images as prior knowledge of the scene. To address this problem,\nwe present a simple yet effective solution called NeRF-Enhanced Outpainting\n(NEO) that uses extended-FOV images generated through NeRF to train a\nscene-specific image outpainting model. To assess the performance of NEO, we\nconduct comprehensive evaluations on three photorealistic datasets and one\nreal-world dataset. Extensive experiments on the benchmark datasets showcase\nthe robustness and potential of our method in addressing this challenge. We\nbelieve our work lays a strong foundation for future exploration within the\nresearch community.\n", "date": "2023-09-23", "categories": []}
{"id": "2309.13294", "title": "MP-MVS: Multi-Scale Windows PatchMatch and Planar Prior Multi-View\n  Stereo", "abstract": "  Significant strides have been made in enhancing the accuracy of Multi-View\nStereo (MVS)-based 3D reconstruction. However, untextured areas with unstable\nphotometric consistency often remain incompletely reconstructed. In this paper,\nwe propose a resilient and effective multi-view stereo approach (MP-MVS). We\ndesign a multi-scale windows PatchMatch (mPM) to obtain reliable depth of\nuntextured areas. In contrast with other multi-scale approaches, which is\nfaster and can be easily extended to PatchMatch-based MVS approaches.\nSubsequently, we improve the existing checkerboard sampling schemes by limiting\nour sampling to distant regions, which can effectively improve the efficiency\nof spatial propagation while mitigating outlier generation. Finally, we\nintroduce and improve planar prior assisted PatchMatch of ACMP. Instead of\nrelying on photometric consistency, we utilize geometric consistency\ninformation between multi-views to select reliable triangulated vertices. This\nstrategy can obtain a more accurate planar prior model to rectify photometric\nconsistency measurements. Our approach has been tested on the ETH3D High-res\nmulti-view benchmark with several state-of-the-art approaches. The results\ndemonstrate that our approach can reach the state-of-the-art. The associated\ncodes will be accessible at https://github.com/RongxuanTan/MP-MVS.\n", "date": "2023-09-23", "categories": []}
{"id": "2309.11993", "title": "Neural Stochastic Screened Poisson Reconstruction", "abstract": "  Reconstructing a surface from a point cloud is an underdetermined problem. We\nuse a neural network to study and quantify this reconstruction uncertainty\nunder a Poisson smoothness prior. Our algorithm addresses the main limitations\nof existing work and can be fully integrated into the 3D scanning pipeline,\nfrom obtaining an initial reconstruction to deciding on the next best sensor\nposition and updating the reconstruction upon capturing more data.\n", "date": "2023-09-21", "categories": []}
{"id": "2309.11133", "title": "Shape Anchor Guided Holistic Indoor Scene Understanding", "abstract": "  This paper proposes a shape anchor guided learning strategy (AncLearn) for\nrobust holistic indoor scene understanding. We observe that the search space\nconstructed by current methods for proposal feature grouping and instance point\nsampling often introduces massive noise to instance detection and mesh\nreconstruction. Accordingly, we develop AncLearn to generate anchors that\ndynamically fit instance surfaces to (i) unmix noise and target-related\nfeatures for offering reliable proposals at the detection stage, and (ii)\nreduce outliers in object point sampling for directly providing well-structured\ngeometry priors without segmentation during reconstruction. We embed AncLearn\ninto a reconstruction-from-detection learning system (AncRec) to generate\nhigh-quality semantic scene models in a purely instance-oriented manner.\nExperiments conducted on the challenging ScanNetv2 dataset demonstrate that our\nshape anchor-based method consistently achieves state-of-the-art performance in\nterms of 3D object detection, layout estimation, and shape reconstruction. The\ncode will be available at https://github.com/Geo-Tell/AncRec.\n", "date": "2023-09-20", "categories": []}
{"id": "2309.10279", "title": "360$^\\circ$ Reconstruction From a Single Image Using Space Carved\n  Outpainting", "abstract": "  We introduce POP3D, a novel framework that creates a full $360^\\circ$-view 3D\nmodel from a single image. POP3D resolves two prominent issues that limit the\nsingle-view reconstruction. Firstly, POP3D offers substantial generalizability\nto arbitrary categories, a trait that previous methods struggle to achieve.\nSecondly, POP3D further improves reconstruction fidelity and naturalness, a\ncrucial aspect that concurrent works fall short of. Our approach marries the\nstrengths of four primary components: (1) a monocular depth and normal\npredictor that serves to predict crucial geometric cues, (2) a space carving\nmethod capable of demarcating the potentially unseen portions of the target\nobject, (3) a generative model pre-trained on a large-scale image dataset that\ncan complete unseen regions of the target, and (4) a neural implicit surface\nreconstruction method tailored in reconstructing objects using RGB images along\nwith monocular geometric cues. The combination of these components enables\nPOP3D to readily generalize across various in-the-wild images and generate\nstate-of-the-art reconstructions, outperforming similar works by a significant\nmargin. Project page: \\url{http://cg.postech.ac.kr/research/POP3D}\n", "date": "2023-09-19", "categories": []}
{"id": "2309.09622", "title": "Holistic Geometric Feature Learning for Structured Reconstruction", "abstract": "  The inference of topological principles is a key problem in structured\nreconstruction. We observe that wrongly predicted topological relationships are\noften incurred by the lack of holistic geometry clues in low-level features.\nInspired by the fact that massive signals can be compactly described with\nfrequency analysis, we experimentally explore the efficiency and tendency of\nlearning structure geometry in the frequency domain. Accordingly, we propose a\nfrequency-domain feature learning strategy (F-Learn) to fuse scattered\ngeometric fragments holistically for topology-intact structure reasoning.\nBenefiting from the parsimonious design, the F-Learn strategy can be easily\ndeployed into a deep reconstructor with a lightweight model modification.\nExperiments demonstrate that the F-Learn strategy can effectively introduce\nstructure awareness into geometric primitive detection and topology inference,\nbringing significant performance improvement to final structured\nreconstruction. Code and pre-trained models are available at\nhttps://github.com/Geo-Tell/F-Learn.\n", "date": "2023-09-18", "categories": []}
{"id": "2309.09739", "title": "Improving Neural Indoor Surface Reconstruction with Mask-Guided Adaptive\n  Consistency Constraints", "abstract": "  3D scene reconstruction from 2D images has been a long-standing task. Instead\nof estimating per-frame depth maps and fusing them in 3D, recent research\nleverages the neural implicit surface as a unified representation for 3D\nreconstruction. Equipped with data-driven pre-trained geometric cues, these\nmethods have demonstrated promising performance. However, inaccurate prior\nestimation, which is usually inevitable, can lead to suboptimal reconstruction\nquality, particularly in some geometrically complex regions. In this paper, we\npropose a two-stage training process, decouple view-dependent and\nview-independent colors, and leverage two novel consistency constraints to\nenhance detail reconstruction performance without requiring extra priors.\nAdditionally, we introduce an essential mask scheme to adaptively influence the\nselection of supervision constraints, thereby improving performance in a\nself-supervised paradigm. Experiments on synthetic and real-world datasets show\nthe capability of reducing the interference from prior estimation errors and\nachieving high-quality scene reconstruction with rich geometric details.\n", "date": "2023-09-18", "categories": []}
{"id": "2309.08596", "title": "Robust e-NeRF: NeRF from Sparse & Noisy Events under Non-Uniform Motion", "abstract": "  Event cameras offer many advantages over standard cameras due to their\ndistinctive principle of operation: low power, low latency, high temporal\nresolution and high dynamic range. Nonetheless, the success of many downstream\nvisual applications also hinges on an efficient and effective scene\nrepresentation, where Neural Radiance Field (NeRF) is seen as the leading\ncandidate. Such promise and potential of event cameras and NeRF inspired recent\nworks to investigate on the reconstruction of NeRF from moving event cameras.\nHowever, these works are mainly limited in terms of the dependence on dense and\nlow-noise event streams, as well as generalization to arbitrary contrast\nthreshold values and camera speed profiles. In this work, we propose Robust\ne-NeRF, a novel method to directly and robustly reconstruct NeRFs from moving\nevent cameras under various real-world conditions, especially from sparse and\nnoisy events generated under non-uniform motion. It consists of two key\ncomponents: a realistic event generation model that accounts for various\nintrinsic parameters (e.g. time-independent, asymmetric threshold and\nrefractory period) and non-idealities (e.g. pixel-to-pixel threshold\nvariation), as well as a complementary pair of normalized reconstruction losses\nthat can effectively generalize to arbitrary speed profiles and intrinsic\nparameter values without such prior knowledge. Experiments on real and novel\nrealistically simulated sequences verify our effectiveness. Our code, synthetic\ndataset and improved event simulator are public.\n", "date": "2023-09-15", "categories": []}
{"id": "2211.16799", "title": "NOPE-SAC: Neural One-Plane RANSAC for Sparse-View Planar 3D\n  Reconstruction", "abstract": "  This paper studies the challenging two-view 3D reconstruction in a rigorous\nsparse-view configuration, which is suffering from insufficient correspondences\nin the input image pairs for camera pose estimation. We present a novel Neural\nOne-PlanE RANSAC framework (termed NOPE-SAC in short) that exerts excellent\ncapability to learn one-plane pose hypotheses from 3D plane correspondences.\nBuilding on the top of a siamese plane detection network, our NOPE-SAC first\ngenerates putative plane correspondences with a coarse initial pose. It then\nfeeds the learned 3D plane parameters of correspondences into shared MLPs to\nestimate the one-plane camera pose hypotheses, which are subsequently reweighed\nin a RANSAC manner to obtain the final camera pose. Because the neural\none-plane pose minimizes the number of plane correspondences for adaptive pose\nhypotheses generation, it enables stable pose voting and reliable pose\nrefinement in a few plane correspondences for the sparse-view inputs. In the\nexperiments, we demonstrate that our NOPE-SAC significantly improves the camera\npose estimation for the two-view inputs with severe viewpoint changes, setting\nseveral new state-of-the-art performances on two challenging benchmarks, i.e.,\nMatterPort3D and ScanNet, for sparse-view 3D reconstruction. The source code is\nreleased at https://github.com/IceTTTb/NopeSAC for reproducible research.\n", "date": "2022-11-30", "categories": []}
{"id": "2309.07021", "title": "Exploiting Multiple Priors for Neural 3D Indoor Reconstruction", "abstract": "  Neural implicit modeling permits to achieve impressive 3D reconstruction\nresults on small objects, while it exhibits significant limitations in large\nindoor scenes. In this work, we propose a novel neural implicit modeling method\nthat leverages multiple regularization strategies to achieve better\nreconstructions of large indoor environments, while relying only on images. A\nsparse but accurate depth prior is used to anchor the scene to the initial\nmodel. A dense but less accurate depth prior is also introduced, flexible\nenough to still let the model diverge from it to improve the estimated\ngeometry. Then, a novel self-supervised strategy to regularize the estimated\nsurface normals is presented. Finally, a learnable exposure compensation scheme\npermits to cope with challenging lighting conditions. Experimental results show\nthat our approach produces state-of-the-art 3D reconstructions in challenging\nindoor scenarios.\n", "date": "2023-09-13", "categories": []}
{"id": "2209.04436", "title": "Multi-NeuS: 3D Head Portraits from Single Image with Neural Implicit\n  Functions", "abstract": "  We present an approach for the reconstruction of textured 3D meshes of human\nheads from one or few views. Since such few-shot reconstruction is\nunderconstrained, it requires prior knowledge which is hard to impose on\ntraditional 3D reconstruction algorithms. In this work, we rely on the recently\nintroduced 3D representation $\\unicode{x2013}$ neural implicit functions\n$\\unicode{x2013}$ which, being based on neural networks, allows to naturally\nlearn priors about human heads from data, and is directly convertible to\ntextured mesh. Namely, we extend NeuS, a state-of-the-art neural implicit\nfunction formulation, to represent multiple objects of a class (human heads in\nour case) simultaneously. The underlying neural net architecture is designed to\nlearn the commonalities among these objects and to generalize to unseen ones.\nOur model is trained on just a hundred smartphone videos and does not require\nany scanned 3D data. Afterwards, the model can fit novel heads in the few-shot\nor one-shot modes with good results.\n", "date": "2022-09-07", "categories": []}
{"id": "2303.09248", "title": "Cross-Dimensional Refined Learning for Real-Time 3D Visual Perception\n  from Monocular Video", "abstract": "  We present a novel real-time capable learning method that jointly perceives a\n3D scene's geometry structure and semantic labels. Recent approaches to\nreal-time 3D scene reconstruction mostly adopt a volumetric scheme, where a\nTruncated Signed Distance Function (TSDF) is directly regressed. However, these\nvolumetric approaches tend to focus on the global coherence of their\nreconstructions, which leads to a lack of local geometric detail. To overcome\nthis issue, we propose to leverage the latent geometric prior knowledge in 2D\nimage features by explicit depth prediction and anchored feature generation, to\nrefine the occupancy learning in TSDF volume. Besides, we find that this\ncross-dimensional feature refinement methodology can also be adopted for the\nsemantic segmentation task by utilizing semantic priors. Hence, we proposed an\nend-to-end cross-dimensional refinement neural network (CDRNet) to extract both\n3D mesh and 3D semantic labeling in real time. The experiment results show that\nthis method achieves a state-of-the-art 3D perception efficiency on multiple\ndatasets, which indicates the great potential of our method for industrial\napplications.\n", "date": "2023-03-16", "categories": []}
{"id": "2303.11219", "title": "NeTO:Neural Reconstruction of Transparent Objects with Self-Occlusion\n  Aware Refraction-Tracing", "abstract": "  We present a novel method, called NeTO, for capturing 3D geometry of solid\ntransparent objects from 2D images via volume rendering. Reconstructing\ntransparent objects is a very challenging task, which is ill-suited for\ngeneral-purpose reconstruction techniques due to the specular light transport\nphenomena. Although existing refraction-tracing based methods, designed\nspecially for this task, achieve impressive results, they still suffer from\nunstable optimization and loss of fine details, since the explicit surface\nrepresentation they adopted is difficult to be optimized, and the\nself-occlusion problem is ignored for refraction-tracing. In this paper, we\npropose to leverage implicit Signed Distance Function (SDF) as surface\nrepresentation, and optimize the SDF field via volume rendering with a\nself-occlusion aware refractive ray tracing. The implicit representation\nenables our method to be capable of reconstructing high-quality reconstruction\neven with a limited set of images, and the self-occlusion aware strategy makes\nit possible for our method to accurately reconstruct the self-occluded regions.\nExperiments show that our method achieves faithful reconstruction results and\noutperforms prior works by a large margin. Visit our project page at\nhttps://www.xxlong.site/NeTO/\n", "date": "2023-03-20", "categories": []}
{"id": "2309.01811", "title": "Instant Continual Learning of Neural Radiance Fields", "abstract": "  Neural radiance fields (NeRFs) have emerged as an effective method for\nnovel-view synthesis and 3D scene reconstruction. However, conventional\ntraining methods require access to all training views during scene\noptimization. This assumption may be prohibitive in continual learning\nscenarios, where new data is acquired in a sequential manner and a continuous\nupdate of the NeRF is desired, as in automotive or remote sensing applications.\nWhen naively trained in such a continual setting, traditional scene\nrepresentation frameworks suffer from catastrophic forgetting, where previously\nlearned knowledge is corrupted after training on new data. Prior works in\nalleviating forgetting with NeRFs suffer from low reconstruction quality and\nhigh latency, making them impractical for real-world application. We propose a\ncontinual learning framework for training NeRFs that leverages replay-based\nmethods combined with a hybrid explicit--implicit scene representation. Our\nmethod outperforms previous methods in reconstruction quality when trained in a\ncontinual setting, while having the additional benefit of being an order of\nmagnitude faster.\n", "date": "2023-09-04", "categories": []}
{"id": "2309.00385", "title": "Dense Voxel 3D Reconstruction Using a Monocular Event Camera", "abstract": "  Event cameras are sensors inspired by biological systems that specialize in\ncapturing changes in brightness. These emerging cameras offer many advantages\nover conventional frame-based cameras, including high dynamic range, high frame\nrates, and extremely low power consumption. Due to these advantages, event\ncameras have increasingly been adapted in various fields, such as frame\ninterpolation, semantic segmentation, odometry, and SLAM. However, their\napplication in 3D reconstruction for VR applications is underexplored. Previous\nmethods in this field mainly focused on 3D reconstruction through depth map\nestimation. Methods that produce dense 3D reconstruction generally require\nmultiple cameras, while methods that utilize a single event camera can only\nproduce a semi-dense result. Other single-camera methods that can produce dense\n3D reconstruction rely on creating a pipeline that either incorporates the\naforementioned methods or other existing Structure from Motion (SfM) or\nMulti-view Stereo (MVS) methods. In this paper, we propose a novel approach for\nsolving dense 3D reconstruction using only a single event camera. To the best\nof our knowledge, our work is the first attempt in this regard. Our preliminary\nresults demonstrate that the proposed method can produce visually\ndistinguishable dense 3D reconstructions directly without requiring pipelines\nlike those used by existing methods. Additionally, we have created a synthetic\ndataset with $39,739$ object scans using an event camera simulator. This\ndataset will help accelerate other relevant research in this field.\n", "date": "2023-09-01", "categories": []}
{"id": "2308.15733", "title": "Drone-NeRF: Efficient NeRF Based 3D Scene Reconstruction for Large-Scale\n  Drone Survey", "abstract": "  Neural rendering has garnered substantial attention owing to its capacity for\ncreating realistic 3D scenes. However, its applicability to extensive scenes\nremains challenging, with limitations in effectiveness. In this work, we\npropose the Drone-NeRF framework to enhance the efficient reconstruction of\nunbounded large-scale scenes suited for drone oblique photography using Neural\nRadiance Fields (NeRF). Our approach involves dividing the scene into uniform\nsub-blocks based on camera position and depth visibility. Sub-scenes are\ntrained in parallel using NeRF, then merged for a complete scene. We refine the\nmodel by optimizing camera poses and guiding NeRF with a uniform sampler.\nIntegrating chosen samples enhances accuracy. A hash-coded fusion MLP\naccelerates density representation, yielding RGB and Depth outputs. Our\nframework accounts for sub-scene constraints, reduces parallel-training noise,\nhandles shadow occlusion, and merges sub-regions for a polished rendering\nresult. This Drone-NeRF framework demonstrates promising capabilities in\naddressing challenges related to scene complexity, rendering efficiency, and\naccuracy in drone-obtained imagery.\n", "date": "2023-08-30", "categories": []}
{"id": "2303.14672", "title": "Sat2Density: Faithful Density Learning from Satellite-Ground Image Pairs", "abstract": "  This paper aims to develop an accurate 3D geometry representation of\nsatellite images using satellite-ground image pairs. Our focus is on the\nchallenging problem of 3D-aware ground-views synthesis from a satellite image.\nWe draw inspiration from the density field representation used in volumetric\nneural rendering and propose a new approach, called Sat2Density. Our method\nutilizes the properties of ground-view panoramas for the sky and non-sky\nregions to learn faithful density fields of 3D scenes in a geometric\nperspective. Unlike other methods that require extra depth information during\ntraining, our Sat2Density can automatically learn accurate and faithful 3D\ngeometry via density representation without depth supervision. This advancement\nsignificantly improves the ground-view panorama synthesis task. Additionally,\nour study provides a new geometric perspective to understand the relationship\nbetween satellite and ground-view images in 3D space.\n", "date": "2023-03-26", "categories": []}
{"id": "2306.08648", "title": "SimpleMapping: Real-Time Visual-Inertial Dense Mapping with Deep\n  Multi-View Stereo", "abstract": "  We present a real-time visual-inertial dense mapping method capable of\nperforming incremental 3D mesh reconstruction with high quality using only\nsequential monocular images and inertial measurement unit (IMU) readings. 6-DoF\ncamera poses are estimated by a robust feature-based visual-inertial odometry\n(VIO), which also generates noisy sparse 3D map points as a by-product. We\npropose a sparse point aided multi-view stereo neural network (SPA-MVSNet) that\ncan effectively leverage the informative but noisy sparse points from the VIO\nsystem. The sparse depth from VIO is firstly completed by a single-view depth\ncompletion network. This dense depth map, although naturally limited in\naccuracy, is then used as a prior to guide our MVS network in the cost volume\ngeneration and regularization for accurate dense depth prediction. Predicted\ndepth maps of keyframe images by the MVS network are incrementally fused into a\nglobal map using TSDF-Fusion. We extensively evaluate both the proposed\nSPA-MVSNet and the entire visual-inertial dense mapping system on several\npublic datasets as well as our own dataset, demonstrating the system's\nimpressive generalization capabilities and its ability to deliver high-quality\n3D mesh reconstruction online. Our proposed dense mapping system achieves a\n39.7% improvement in F-score over existing systems when evaluated on the\nchallenging scenarios of the EuRoC dataset.\n", "date": "2023-06-14", "categories": []}
{"id": "2308.14383", "title": "Multi-Modal Neural Radiance Field for Monocular Dense SLAM with a\n  Light-Weight ToF Sensor", "abstract": "  Light-weight time-of-flight (ToF) depth sensors are compact and\ncost-efficient, and thus widely used on mobile devices for tasks such as\nautofocus and obstacle detection. However, due to the sparse and noisy depth\nmeasurements, these sensors have rarely been considered for dense geometry\nreconstruction. In this work, we present the first dense SLAM system with a\nmonocular camera and a light-weight ToF sensor. Specifically, we propose a\nmulti-modal implicit scene representation that supports rendering both the\nsignals from the RGB camera and light-weight ToF sensor which drives the\noptimization by comparing with the raw sensor inputs. Moreover, in order to\nguarantee successful pose tracking and reconstruction, we exploit a predicted\ndepth as an intermediate supervision and develop a coarse-to-fine optimization\nstrategy for efficient learning of the implicit representation. At last, the\ntemporal information is explicitly exploited to deal with the noisy signals\nfrom light-weight ToF sensors to improve the accuracy and robustness of the\nsystem. Experiments demonstrate that our system well exploits the signals of\nlight-weight ToF sensors and achieves competitive results both on camera\ntracking and dense scene reconstruction. Project page:\n\\url{https://zju3dv.github.io/tof_slam/}.\n", "date": "2023-08-28", "categories": []}
{"id": "2308.14713", "title": "R3D3: Dense 3D Reconstruction of Dynamic Scenes from Multiple Cameras", "abstract": "  Dense 3D reconstruction and ego-motion estimation are key challenges in\nautonomous driving and robotics. Compared to the complex, multi-modal systems\ndeployed today, multi-camera systems provide a simpler, low-cost alternative.\nHowever, camera-based 3D reconstruction of complex dynamic scenes has proven\nextremely difficult, as existing solutions often produce incomplete or\nincoherent results. We propose R3D3, a multi-camera system for dense 3D\nreconstruction and ego-motion estimation. Our approach iterates between\ngeometric estimation that exploits spatial-temporal information from multiple\ncameras, and monocular depth refinement. We integrate multi-camera feature\ncorrelation and dense bundle adjustment operators that yield robust geometric\ndepth and pose estimates. To improve reconstruction where geometric depth is\nunreliable, e.g. for moving objects or low-textured regions, we introduce\nlearnable scene priors via a depth refinement network. We show that this design\nenables a dense, consistent 3D reconstruction of challenging, dynamic outdoor\nenvironments. Consequently, we achieve state-of-the-art dense depth prediction\non the DDAD and NuScenes benchmarks.\n", "date": "2023-08-28", "categories": []}
{"id": "2308.14737", "title": "Flexible Techniques for Differentiable Rendering with 3D Gaussians", "abstract": "  Fast, reliable shape reconstruction is an essential ingredient in many\ncomputer vision applications. Neural Radiance Fields demonstrated that\nphotorealistic novel view synthesis is within reach, but was gated by\nperformance requirements for fast reconstruction of real scenes and objects.\nSeveral recent approaches have built on alternative shape representations, in\nparticular, 3D Gaussians. We develop extensions to these renderers, such as\nintegrating differentiable optical flow, exporting watertight meshes and\nrendering per-ray normals. Additionally, we show how two of the recent methods\nare interoperable with each other. These reconstructions are quick, robust, and\neasily performed on GPU or CPU. For code and visual examples, see\nhttps://leonidk.github.io/fmb-plus\n", "date": "2023-08-28", "categories": []}
{"id": "2212.02501", "title": "SceneRF: Self-Supervised Monocular 3D Scene Reconstruction with Radiance\n  Fields", "abstract": "  3D reconstruction from a single 2D image was extensively covered in the\nliterature but relies on depth supervision at training time, which limits its\napplicability. To relax the dependence to depth we propose SceneRF, a\nself-supervised monocular scene reconstruction method using only posed image\nsequences for training. Fueled by the recent progress in neural radiance fields\n(NeRF) we optimize a radiance field though with explicit depth optimization and\na novel probabilistic sampling strategy to efficiently handle large scenes. At\ninference, a single input image suffices to hallucinate novel depth views which\nare fused together to obtain 3D scene reconstruction. Thorough experiments\ndemonstrate that we outperform all baselines for novel depth views synthesis\nand scene reconstruction, on indoor BundleFusion and outdoor SemanticKITTI.\nCode is available at https://astra-vision.github.io/SceneRF .\n", "date": "2022-12-05", "categories": []}
{"id": "2308.13175", "title": "GridPull: Towards Scalability in Learning Implicit Representations from\n  3D Point Clouds", "abstract": "  Learning implicit representations has been a widely used solution for surface\nreconstruction from 3D point clouds. The latest methods infer a distance or\noccupancy field by overfitting a neural network on a single point cloud.\nHowever, these methods suffer from a slow inference due to the slow convergence\nof neural networks and the extensive calculation of distances to surface\npoints, which limits them to small scale points. To resolve the scalability\nissue in surface reconstruction, we propose GridPull to improve the efficiency\nof learning implicit representations from large scale point clouds. Our novelty\nlies in the fast inference of a discrete distance field defined on grids\nwithout using any neural components. To remedy the lack of continuousness\nbrought by neural networks, we introduce a loss function to encourage\ncontinuous distances and consistent gradients in the field during pulling\nqueries onto the surface in grids near to the surface. We use uniform grids for\na fast grid search to localize sampled queries, and organize surface points in\na tree structure to speed up the calculation of distances to the surface. We do\nnot rely on learning priors or normal supervision during optimization, and\nachieve superiority over the latest methods in terms of complexity and\naccuracy. We evaluate our method on shape and scene benchmarks, and report\nnumerical and visual comparisons with the latest methods to justify our\neffectiveness and superiority. The code is available at\nhttps://github.com/chenchao15/GridPull.\n", "date": "2023-08-25", "categories": []}
{"id": "2308.12138", "title": "Select-and-Combine (SAC): A Novel Multi-Stereo Depth Fusion Algorithm\n  for Point Cloud Generation via Efficient Local Markov Netlets", "abstract": "  Many practical systems for image-based surface reconstruction employ a\nstereo/multi-stereo paradigm, due to its ability to scale for large scenes and\nits ease of implementation for out-of-core operations. In this process,\nmultiple and abundant depth maps from stereo matching must be combined and\nfused into a single, consistent, and clean point cloud. However, the noises and\noutliers caused by stereo matching and the heterogenous geometric errors of the\nposes present a challenge for existing fusion algorithms, since they mostly\nassume Gaussian errors and predict fused results based on data from local\nspatial neighborhoods, which may inherit uncertainties from multiple depths\nresulting in lowered accuracy. In this paper, we propose a novel depth fusion\nparadigm, that instead of numerically fusing points from multiple depth maps,\nselects the best depth map per point, and combines them into a single and clean\npoint cloud. This paradigm, called select-and-combine (SAC), is achieved\nthrough modeling the point level fusion using local Markov Netlets, a\nmicro-network over point across neighboring views for depth/view selection,\nfollowed by a Netlets collapse process for point combination. The Markov\nNetlets are optimized such that they can inherently leverage spatial\nconsistencies among depth maps of neighboring views, thus they can address\nerrors beyond Gaussian ones. Our experiment results show that our approach\noutperforms existing depth fusion approaches by increasing the F1 score that\nconsiders both accuracy and completeness by 2.07% compared to the best existing\nmethod. Finally, our approach generates clearer point clouds that are 18% less\nredundant while with a higher accuracy before fusion\n", "date": "2023-08-23", "categories": []}
{"id": "2308.12139", "title": "Mesh Conflation of Oblique Photogrammetric Models using Virtual Cameras\n  and Truncated Signed Distance Field", "abstract": "  Conflating/stitching 2.5D raster digital surface models (DSM) into a large\none has been a running practice in geoscience applications, however, conflating\nfull-3D mesh models, such as those from oblique photogrammetry, is extremely\nchallenging. In this letter, we propose a novel approach to address this\nchallenge by conflating multiple full-3D oblique photogrammetric models into a\nsingle, and seamless mesh for high-resolution site modeling. Given two or more\nindividually collected and created photogrammetric meshes, we first propose to\ncreate a virtual camera field (with a panoramic field of view) to incubate\nvirtual spaces represented by Truncated Signed Distance Field (TSDF), an\nimplicit volumetric field friendly for linear 3D fusion; then we adaptively\nleverage the truncated bound of meshes in TSDF to conflate them into a single\nand accurate full 3D site model. With drone-based 3D meshes, we show that our\napproach significantly improves upon traditional methods for model conflations,\nto drive new potentials to create excessively large and accurate full 3D mesh\nmodels in support of geoscience and environmental applications.\n", "date": "2023-08-23", "categories": []}
{"id": "2304.00054", "title": "LivePose: Online 3D Reconstruction from Monocular Video with Dynamic\n  Camera Poses", "abstract": "  Dense 3D reconstruction from RGB images traditionally assumes static camera\npose estimates. This assumption has endured, even as recent works have\nincreasingly focused on real-time methods for mobile devices. However, the\nassumption of a fixed pose for each image does not hold for online execution:\nposes from real-time SLAM are dynamic and may be updated following events such\nas bundle adjustment and loop closure. This has been addressed in the RGB-D\nsetting, by de-integrating past views and re-integrating them with updated\nposes, but it remains largely untreated in the RGB-only setting. We formalize\nthis problem to define the new task of dense online reconstruction from\ndynamically-posed images. To support further research, we introduce a dataset\ncalled LivePose containing the dynamic poses from a SLAM system running on\nScanNet. We select three recent reconstruction systems and apply a framework\nbased on de-integration to adapt each one to the dynamic-pose setting. In\naddition, we propose a novel, non-linear de-integration module that learns to\nremove stale scene content. We show that responding to pose updates is critical\nfor high-quality reconstruction, and that our de-integration framework is an\neffective solution.\n", "date": "2023-03-31", "categories": []}
{"id": "2304.01480", "title": "FineRecon: Depth-aware Feed-forward Network for Detailed 3D\n  Reconstruction", "abstract": "  Recent works on 3D reconstruction from posed images have demonstrated that\ndirect inference of scene-level 3D geometry without test-time optimization is\nfeasible using deep neural networks, showing remarkable promise and high\nefficiency. However, the reconstructed geometry, typically represented as a 3D\ntruncated signed distance function (TSDF), is often coarse without fine\ngeometric details. To address this problem, we propose three effective\nsolutions for improving the fidelity of inference-based 3D reconstructions. We\nfirst present a resolution-agnostic TSDF supervision strategy to provide the\nnetwork with a more accurate learning signal during training, avoiding the\npitfalls of TSDF interpolation seen in previous work. We then introduce a depth\nguidance strategy using multi-view depth estimates to enhance the scene\nrepresentation and recover more accurate surfaces. Finally, we develop a novel\narchitecture for the final layers of the network, conditioning the output TSDF\nprediction on high-resolution image features in addition to coarse voxel\nfeatures, enabling sharper reconstruction of fine details. Our method,\nFineRecon, produces smooth and highly accurate reconstructions, showing\nsignificant improvements across multiple depth and 3D reconstruction metrics.\n", "date": "2023-04-04", "categories": []}
{"id": "2308.10337", "title": "Strata-NeRF : Neural Radiance Fields for Stratified Scenes", "abstract": "  Neural Radiance Field (NeRF) approaches learn the underlying 3D\nrepresentation of a scene and generate photo-realistic novel views with high\nfidelity. However, most proposed settings concentrate on modelling a single\nobject or a single level of a scene. However, in the real world, we may capture\na scene at multiple levels, resulting in a layered capture. For example,\ntourists usually capture a monument's exterior structure before capturing the\ninner structure. Modelling such scenes in 3D with seamless switching between\nlevels can drastically improve immersive experiences. However, most existing\ntechniques struggle in modelling such scenes. We propose Strata-NeRF, a single\nneural radiance field that implicitly captures a scene with multiple levels.\nStrata-NeRF achieves this by conditioning the NeRFs on Vector Quantized (VQ)\nlatent representations which allow sudden changes in scene structure. We\nevaluate the effectiveness of our approach in multi-layered synthetic dataset\ncomprising diverse scenes and then further validate its generalization on the\nreal-world RealEstate10K dataset. We find that Strata-NeRF effectively captures\nstratified scenes, minimizes artifacts, and synthesizes high-fidelity views\ncompared to existing approaches.\n", "date": "2023-08-20", "categories": []}
{"id": "2302.13987", "title": "UMIFormer: Mining the Correlations between Similar Tokens for Multi-View\n  3D Reconstruction", "abstract": "  In recent years, many video tasks have achieved breakthroughs by utilizing\nthe vision transformer and establishing spatial-temporal decoupling for feature\nextraction. Although multi-view 3D reconstruction also faces multiple images as\ninput, it cannot immediately inherit their success due to completely ambiguous\nassociations between unstructured views. There is not usable prior\nrelationship, which is similar to the temporally-coherence property in a video.\nTo solve this problem, we propose a novel transformer network for Unstructured\nMultiple Images (UMIFormer). It exploits transformer blocks for decoupled\nintra-view encoding and designed blocks for token rectification that mine the\ncorrelation between similar tokens from different views to achieve decoupled\ninter-view encoding. Afterward, all tokens acquired from various branches are\ncompressed into a fixed-size compact representation while preserving rich\ninformation for reconstruction by leveraging the similarities between tokens.\nWe empirically demonstrate on ShapeNet and confirm that our decoupled learning\nmethod is adaptable for unstructured multiple images. Meanwhile, the\nexperiments also verify our model outperforms existing SOTA methods by a large\nmargin. Code will be available at https://github.com/GaryZhu1996/UMIFormer.\n", "date": "2023-02-27", "categories": []}
{"id": "2303.13855", "title": "Deformable Model-Driven Neural Rendering for High-Fidelity 3D\n  Reconstruction of Human Heads Under Low-View Settings", "abstract": "  Reconstructing 3D human heads in low-view settings presents technical\nchallenges, mainly due to the pronounced risk of overfitting with limited views\nand high-frequency signals. To address this, we propose geometry decomposition\nand adopt a two-stage, coarse-to-fine training strategy, allowing for\nprogressively capturing high-frequency geometric details. We represent 3D human\nheads using the zero level-set of a combined signed distance field, comprising\na smooth template, a non-rigid deformation, and a high-frequency displacement\nfield. The template captures features that are independent of both identity and\nexpression and is co-trained with the deformation network across multiple\nindividuals with sparse and randomly selected views. The displacement field,\ncapturing individual-specific details, undergoes separate training for each\nperson. Our network training does not require 3D supervision or object masks.\nExperimental results demonstrate the effectiveness and robustness of our\ngeometry decomposition and two-stage training strategy. Our method outperforms\nexisting neural rendering approaches in terms of reconstruction accuracy and\nnovel view synthesis under low-view settings. Moreover, the pre-trained\ntemplate serves a good initialization for our model when encountering unseen\nindividuals.\n", "date": "2023-03-24", "categories": []}
{"id": "2308.07868", "title": "ObjectSDF++: Improved Object-Compositional Neural Implicit Surfaces", "abstract": "  In recent years, neural implicit surface reconstruction has emerged as a\npopular paradigm for multi-view 3D reconstruction. Unlike traditional\nmulti-view stereo approaches, the neural implicit surface-based methods\nleverage neural networks to represent 3D scenes as signed distance functions\n(SDFs). However, they tend to disregard the reconstruction of individual\nobjects within the scene, which limits their performance and practical\napplications. To address this issue, previous work ObjectSDF introduced a nice\nframework of object-composition neural implicit surfaces, which utilizes 2D\ninstance masks to supervise individual object SDFs. In this paper, we propose a\nnew framework called ObjectSDF++ to overcome the limitations of ObjectSDF.\nFirst, in contrast to ObjectSDF whose performance is primarily restricted by\nits converted semantic field, the core component of our model is an\nocclusion-aware object opacity rendering formulation that directly\nvolume-renders object opacity to be supervised with instance masks. Second, we\ndesign a novel regularization term for object distinction, which can\neffectively mitigate the issue that ObjectSDF may result in unexpected\nreconstruction in invisible regions due to the lack of constraint to prevent\ncollisions. Our extensive experiments demonstrate that our novel framework not\nonly produces superior object reconstruction results but also significantly\nimproves the quality of scene reconstruction. Code and more resources can be\nfound in \\url{https://qianyiwu.github.io/objectsdf++}\n", "date": "2023-08-15", "categories": []}
{"id": "2308.08724", "title": "Long-Range Grouping Transformer for Multi-View 3D Reconstruction", "abstract": "  Nowadays, transformer networks have demonstrated superior performance in many\ncomputer vision tasks. In a multi-view 3D reconstruction algorithm following\nthis paradigm, self-attention processing has to deal with intricate image\ntokens including massive information when facing heavy amounts of view input.\nThe curse of information content leads to the extreme difficulty of model\nlearning. To alleviate this problem, recent methods compress the token number\nrepresenting each view or discard the attention operations between the tokens\nfrom different views. Obviously, they give a negative impact on performance.\nTherefore, we propose long-range grouping attention (LGA) based on the\ndivide-and-conquer principle. Tokens from all views are grouped for separate\nattention operations. The tokens in each group are sampled from all views and\ncan provide macro representation for the resided view. The richness of feature\nlearning is guaranteed by the diversity among different groups. An effective\nand efficient encoder can be established which connects inter-view features\nusing LGA and extract intra-view features using the standard self-attention\nlayer. Moreover, a novel progressive upsampling decoder is also designed for\nvoxel generation with relatively high resolution. Hinging on the above, we\nconstruct a powerful transformer-based network, called LRGT. Experimental\nresults on ShapeNet verify our method achieves SOTA accuracy in multi-view\nreconstruction. Code will be available at\nhttps://github.com/LiyingCV/Long-Range-Grouping-Transformer.\n", "date": "2023-08-17", "categories": []}
{"id": "2308.09713", "title": "Dynamic 3D Gaussians: Tracking by Persistent Dynamic View Synthesis", "abstract": "  We present a method that simultaneously addresses the tasks of dynamic scene\nnovel-view synthesis and six degree-of-freedom (6-DOF) tracking of all dense\nscene elements. We follow an analysis-by-synthesis framework, inspired by\nrecent work that models scenes as a collection of 3D Gaussians which are\noptimized to reconstruct input images via differentiable rendering. To model\ndynamic scenes, we allow Gaussians to move and rotate over time while enforcing\nthat they have persistent color, opacity, and size. By regularizing Gaussians'\nmotion and rotation with local-rigidity constraints, we show that our Dynamic\n3D Gaussians correctly model the same area of physical space over time,\nincluding the rotation of that space. Dense 6-DOF tracking and dynamic\nreconstruction emerges naturally from persistent dynamic view synthesis,\nwithout requiring any correspondence or flow as input. We demonstrate a large\nnumber of downstream applications enabled by our representation, including\nfirst-person view synthesis, dynamic compositional scene synthesis, and 4D\nvideo editing.\n", "date": "2023-08-18", "categories": []}
{"id": "2307.04246", "title": "Convex Decomposition of Indoor Scenes", "abstract": "  We describe a method to parse a complex, cluttered indoor scene into\nprimitives which offer a parsimonious abstraction of scene structure. Our\nprimitives are simple convexes. Our method uses a learned regression procedure\nto parse a scene into a fixed number of convexes from RGBD input, and can\noptionally accept segmentations to improve the decomposition. The result is\nthen polished with a descent method which adjusts the convexes to produce a\nvery good fit, and greedily removes superfluous primitives. Because the entire\nscene is parsed, we can evaluate using traditional depth, normal, and\nsegmentation error metrics. Our evaluation procedure demonstrates that the\nerror from our primitive representation is comparable to that of predicting\ndepth from a single image.\n", "date": "2023-07-09", "categories": []}
{"id": "2308.08258", "title": "SceNeRFlow: Time-Consistent Reconstruction of General Dynamic Scenes", "abstract": "  Existing methods for the 4D reconstruction of general, non-rigidly deforming\nobjects focus on novel-view synthesis and neglect correspondences. However,\ntime consistency enables advanced downstream tasks like 3D editing, motion\nanalysis, or virtual-asset creation. We propose SceNeRFlow to reconstruct a\ngeneral, non-rigid scene in a time-consistent manner. Our dynamic-NeRF method\ntakes multi-view RGB videos and background images from static cameras with\nknown camera parameters as input. It then reconstructs the deformations of an\nestimated canonical model of the geometry and appearance in an online fashion.\nSince this canonical model is time-invariant, we obtain correspondences even\nfor long-term, long-range motions. We employ neural scene representations to\nparametrize the components of our method. Like prior dynamic-NeRF methods, we\nuse a backwards deformation model. We find non-trivial adaptations of this\nmodel necessary to handle larger motions: We decompose the deformations into a\nstrongly regularized coarse component and a weakly regularized fine component,\nwhere the coarse component also extends the deformation field into the space\nsurrounding the object, which enables tracking over time. We show\nexperimentally that, unlike prior work that only handles small motion, our\nmethod enables the reconstruction of studio-scale motions.\n", "date": "2023-08-16", "categories": []}
{"id": "2308.07837", "title": "CCD-3DR: Consistent Conditioning in Diffusion for Single-Image 3D\n  Reconstruction", "abstract": "  In this paper, we present a novel shape reconstruction method leveraging\ndiffusion model to generate 3D sparse point cloud for the object captured in a\nsingle RGB image. Recent methods typically leverage global embedding or local\nprojection-based features as the condition to guide the diffusion model.\nHowever, such strategies fail to consistently align the denoised point cloud\nwith the given image, leading to unstable conditioning and inferior\nperformance. In this paper, we present CCD-3DR, which exploits a novel centered\ndiffusion probabilistic model for consistent local feature conditioning. We\nconstrain the noise and sampled point cloud from the diffusion model into a\nsubspace where the point cloud center remains unchanged during the forward\ndiffusion process and reverse process. The stable point cloud center further\nserves as an anchor to align each point with its corresponding local\nprojection-based features. Extensive experiments on synthetic benchmark\nShapeNet-R2N2 demonstrate that CCD-3DR outperforms all competitors by a large\nmargin, with over 40% improvement. We also provide results on real-world\ndataset Pix3D to thoroughly demonstrate the potential of CCD-3DR in real-world\napplications. Codes will be released soon\n", "date": "2023-08-15", "categories": []}
{"id": "2308.06974", "title": "A One Stop 3D Target Reconstruction and multilevel Segmentation Method", "abstract": "  3D object reconstruction and multilevel segmentation are fundamental to\ncomputer vision research. Existing algorithms usually perform 3D scene\nreconstruction and target objects segmentation independently, and the\nperformance is not fully guaranteed due to the challenge of the 3D\nsegmentation. Here we propose an open-source one stop 3D target reconstruction\nand multilevel segmentation framework (OSTRA), which performs segmentation on\n2D images, tracks multiple instances with segmentation labels in the image\nsequence, and then reconstructs labelled 3D objects or multiple parts with\nMulti-View Stereo (MVS) or RGBD-based 3D reconstruction methods. We extend\nobject tracking and 3D reconstruction algorithms to support continuous\nsegmentation labels to leverage the advances in the 2D image segmentation,\nespecially the Segment-Anything Model (SAM) which uses the pretrained neural\nnetwork without additional training for new scenes, for 3D object segmentation.\nOSTRA supports most popular 3D object models including point cloud, mesh and\nvoxel, and achieves high performance for semantic segmentation, instance\nsegmentation and part segmentation on several 3D datasets. It even surpasses\nthe manual segmentation in scenes with complex structures and occlusions. Our\nmethod opens up a new avenue for reconstructing 3D targets embedded with rich\nmulti-scale segmentation information in complex scenes. OSTRA is available from\nhttps://github.com/ganlab/OSTRA.\n", "date": "2023-08-14", "categories": []}
{"id": "2308.05733", "title": "FrozenRecon: Pose-free 3D Scene Reconstruction with Frozen Depth Models", "abstract": "  3D scene reconstruction is a long-standing vision task. Existing approaches\ncan be categorized into geometry-based and learning-based methods. The former\nleverages multi-view geometry but can face catastrophic failures due to the\nreliance on accurate pixel correspondence across views. The latter was\nproffered to mitigate these issues by learning 2D or 3D representation\ndirectly. However, without a large-scale video or 3D training data, it can\nhardly generalize to diverse real-world scenarios due to the presence of tens\nof millions or even billions of optimization parameters in the deep network.\nRecently, robust monocular depth estimation models trained with large-scale\ndatasets have been proven to possess weak 3D geometry prior, but they are\ninsufficient for reconstruction due to the unknown camera parameters, the\naffine-invariant property, and inter-frame inconsistency. Here, we propose a\nnovel test-time optimization approach that can transfer the robustness of\naffine-invariant depth models such as LeReS to challenging diverse scenes while\nensuring inter-frame consistency, with only dozens of parameters to optimize\nper video frame. Specifically, our approach involves freezing the pre-trained\naffine-invariant depth model's depth predictions, rectifying them by optimizing\nthe unknown scale-shift values with a geometric consistency alignment module,\nand employing the resulting scale-consistent depth maps to robustly obtain\ncamera poses and achieve dense scene reconstruction, even in low-texture\nregions. Experiments show that our method achieves state-of-the-art\ncross-dataset reconstruction on five zero-shot testing datasets.\n", "date": "2023-08-10", "categories": []}
{"id": "2211.16762", "title": "GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided\n  Distance Representation", "abstract": "  We present a learning-based method, namely GeoUDF,to tackle the long-standing\nand challenging problem of reconstructing a discrete surface from a sparse\npoint cloud.To be specific, we propose a geometry-guided learning method for\nUDF and its gradient estimation that explicitly formulates the unsigned\ndistance of a query point as the learnable affine averaging of its distances to\nthe tangent planes of neighboring points on the surface. Besides,we model the\nlocal geometric structure of the input point clouds by explicitly learning a\nquadratic polynomial for each point. This not only facilitates upsampling the\ninput sparse point cloud but also naturally induces unoriented normal, which\nfurther augments UDF estimation. Finally, to extract triangle meshes from the\npredicted UDF we propose a customized edge-based marching cube module. We\nconduct extensive experiments and ablation studies to demonstrate the\nsignificant advantages of our method over state-of-the-art methods in terms of\nreconstruction accuracy, efficiency, and generality. The source code is\npublicly available at https://github.com/rsy6318/GeoUDF.\n", "date": "2022-11-30", "categories": []}
{"id": "2307.14570", "title": "Physically Plausible 3D Human-Scene Reconstruction from Monocular RGB\n  Image using an Adversarial Learning Approach", "abstract": "  Holistic 3D human-scene reconstruction is a crucial and emerging research\narea in robot perception. A key challenge in holistic 3D human-scene\nreconstruction is to generate a physically plausible 3D scene from a single\nmonocular RGB image. The existing research mainly proposes optimization-based\napproaches for reconstructing the scene from a sequence of RGB frames with\nexplicitly defined physical laws and constraints between different scene\nelements (humans and objects). However, it is hard to explicitly define and\nmodel every physical law in every scenario. This paper proposes using an\nimplicit feature representation of the scene elements to distinguish a\nphysically plausible alignment of humans and objects from an implausible one.\nWe propose using a graph-based holistic representation with an encoded physical\nrepresentation of the scene to analyze the human-object and object-object\ninteractions within the scene. Using this graphical representation, we\nadversarially train our model to learn the feasible alignments of the scene\nelements from the training data itself without explicitly defining the laws and\nconstraints between them. Unlike the existing inference-time optimization-based\napproaches, we use this adversarially trained model to produce a per-frame 3D\nreconstruction of the scene that abides by the physical laws and constraints.\nOur learning-based method achieves comparable 3D reconstruction quality to\nexisting optimization-based holistic human-scene reconstruction methods and\ndoes not need inference time optimization. This makes it better suited when\ncompared to existing methods, for potential use in robotic applications, such\nas robot navigation, etc.\n", "date": "2023-07-27", "categories": []}
{"id": "2307.14009", "title": "Car-Studio: Learning Car Radiance Fields from Single-View and Endless\n  In-the-wild Images", "abstract": "  Compositional neural scene graph studies have shown that radiance fields can\nbe an efficient tool in an editable autonomous driving simulator. However,\nprevious studies learned within a sequence of autonomous driving datasets,\nresulting in unsatisfactory blurring when rotating the car in the simulator. In\nthis letter, we propose a pipeline for learning unconstrained images and\nbuilding a dataset from processed images. To meet the requirements of the\nsimulator, which demands that the vehicle maintain clarity when the perspective\nchanges and that the contour remains sharp from the background to avoid\nartifacts when editing, we design a radiation field of the vehicle, a crucial\npart of the urban scene foreground. Through experiments, we demonstrate that\nour model achieves competitive performance compared to baselines. Using the\ndatasets built from in-the-wild images, our method gradually presents a\ncontrollable appearance editing function. We will release the dataset and code\non https://lty2226262.github.io/car-studio/ to facilitate further research in\nthe field.\n", "date": "2023-07-26", "categories": []}
{"id": "2212.06524", "title": "SST: Real-time End-to-end Monocular 3D Reconstruction via Sparse\n  Spatial-Temporal Guidance", "abstract": "  Real-time monocular 3D reconstruction is a challenging problem that remains\nunsolved. Although recent end-to-end methods have demonstrated promising\nresults, tiny structures and geometric boundaries are hardly captured due to\ntheir insufficient supervision neglecting spatial details and oversimplified\nfeature fusion ignoring temporal cues. To address the problems, we propose an\nend-to-end 3D reconstruction network SST, which utilizes Sparse estimated\npoints from visual SLAM system as additional Spatial guidance and fuses\nTemporal features via a novel cross-modal attention mechanism, achieving more\ndetailed reconstruction results. We propose a Local Spatial-Temporal Fusion\nmodule to exploit more informative spatial-temporal cues from multi-view color\ninformation and sparse priors, as well a Global Spatial-Temporal Fusion module\nto refine the local TSDF volumes with the world-frame model from coarse to\nfine. Extensive experiments on ScanNet and 7-Scenes demonstrate that SST\noutperforms all state-of-the-art competitors, whilst keeping a high inference\nspeed at 59 FPS, enabling real-world applications with real-time requirements.\n", "date": "2022-12-13", "categories": []}
{"id": "2307.12194", "title": "LIST: Learning Implicitly from Spatial Transformers for Single-View 3D\n  Reconstruction", "abstract": "  Accurate reconstruction of both the geometric and topological details of a 3D\nobject from a single 2D image embodies a fundamental challenge in computer\nvision. Existing explicit/implicit solutions to this problem struggle to\nrecover self-occluded geometry and/or faithfully reconstruct topological shape\nstructures. To resolve this dilemma, we introduce LIST, a novel neural\narchitecture that leverages local and global image features to accurately\nreconstruct the geometric and topological structure of a 3D object from a\nsingle image. We utilize global 2D features to predict a coarse shape of the\ntarget object and then use it as a base for higher-resolution reconstruction.\nBy leveraging both local 2D features from the image and 3D features from the\ncoarse prediction, we can predict the signed distance between an arbitrary\npoint and the target surface via an implicit predictor with great accuracy.\nFurthermore, our model does not require camera estimation or pixel alignment.\nIt provides an uninfluenced reconstruction from the input-view direction.\nThrough qualitative and quantitative analysis, we show the superiority of our\nmodel in reconstructing 3D objects from both synthetic and real-world images\nagainst the state of the art.\n", "date": "2023-07-23", "categories": []}
{"id": "2307.10776", "title": "Urban Radiance Field Representation with Deformable Neural Mesh\n  Primitives", "abstract": "  Neural Radiance Fields (NeRFs) have achieved great success in the past few\nyears. However, most current methods still require intensive resources due to\nray marching-based rendering. To construct urban-level radiance fields\nefficiently, we design Deformable Neural Mesh Primitive~(DNMP), and propose to\nparameterize the entire scene with such primitives. The DNMP is a flexible and\ncompact neural variant of classic mesh representation, which enjoys both the\nefficiency of rasterization-based rendering and the powerful neural\nrepresentation capability for photo-realistic image synthesis. Specifically, a\nDNMP consists of a set of connected deformable mesh vertices with paired vertex\nfeatures to parameterize the geometry and radiance information of a local area.\nTo constrain the degree of freedom for optimization and lower the storage\nbudgets, we enforce the shape of each primitive to be decoded from a relatively\nlow-dimensional latent space. The rendering colors are decoded from the vertex\nfeatures (interpolated with rasterization) by a view-dependent MLP. The DNMP\nprovides a new paradigm for urban-level scene representation with appealing\nproperties: $(1)$ High-quality rendering. Our method achieves leading\nperformance for novel view synthesis in urban scenarios. $(2)$ Low\ncomputational costs. Our representation enables fast rendering (2.07ms/1k\npixels) and low peak memory usage (110MB/1k pixels). We also present a\nlightweight version that can run 33$\\times$ faster than vanilla NeRFs, and\ncomparable to the highly-optimized Instant-NGP (0.61 vs 0.71ms/1k pixels).\nProject page: \\href{https://dnmp.github.io/}{https://dnmp.github.io/}.\n", "date": "2023-07-20", "categories": []}
{"id": "2301.04581", "title": "Elevation Estimation-Driven Building 3D Reconstruction from Single-View\n  Remote Sensing Imagery", "abstract": "  Building 3D reconstruction from remote sensing images has a wide range of\napplications in smart cities, photogrammetry and other fields. Methods for\nautomatic 3D urban building modeling typically employ multi-view images as\ninput to algorithms to recover point clouds and 3D models of buildings.\nHowever, such models rely heavily on multi-view images of buildings, which are\ntime-intensive and limit the applicability and practicality of the models. To\nsolve these issues, we focus on designing an efficient DSM estimation-driven\nreconstruction framework (Building3D), which aims to reconstruct 3D building\nmodels from the input single-view remote sensing image. First, we propose a\nSemantic Flow Field-guided DSM Estimation (SFFDE) network, which utilizes the\nproposed concept of elevation semantic flow to achieve the registration of\nlocal and global features. Specifically, in order to make the network semantics\nglobally aware, we propose an Elevation Semantic Globalization (ESG) module to\nrealize the semantic globalization of instances. Further, in order to alleviate\nthe semantic span of global features and original local features, we propose a\nLocal-to-Global Elevation Semantic Registration (L2G-ESR) module based on\nelevation semantic flow. Our Building3D is rooted in the SFFDE network for\nbuilding elevation prediction, synchronized with a building extraction network\nfor building masks, and then sequentially performs point cloud reconstruction,\nsurface reconstruction (or CityGML model reconstruction). On this basis, our\nBuilding3D can optionally generate CityGML models or surface mesh models of the\nbuildings. Extensive experiments on ISPRS Vaihingen and DFC2019 datasets on the\nDSM estimation task show that our SFFDE significantly improves upon\nstate-of-the-arts. Furthermore, our Building3D achieves impressive results in\nthe 3D point cloud and 3D model reconstruction process.\n", "date": "2023-01-11", "categories": []}
{"id": "2304.10406", "title": "LiDAR-NeRF: Novel LiDAR View Synthesis via Neural Radiance Fields", "abstract": "  We introduce a new task, novel view synthesis for LiDAR sensors. While\ntraditional model-based LiDAR simulators with style-transfer neural networks\ncan be applied to render novel views, they fall short of producing accurate and\nrealistic LiDAR patterns because the renderers rely on explicit 3D\nreconstruction and exploit game engines, that ignore important attributes of\nLiDAR points. We address this challenge by formulating, to the best of our\nknowledge, the first differentiable end-to-end LiDAR rendering framework,\nLiDAR-NeRF, leveraging a neural radiance field (NeRF) to facilitate the joint\nlearning of geometry and the attributes of 3D points. However, simply employing\nNeRF cannot achieve satisfactory results, as it only focuses on learning\nindividual pixels while ignoring local information, especially at low texture\nareas, resulting in poor geometry. To this end, we have taken steps to address\nthis issue by introducing a structural regularization method to preserve local\nstructural details. To evaluate the effectiveness of our approach, we establish\nan object-centric multi-view LiDAR dataset, dubbed NeRF-MVL. It contains\nobservations of objects from 9 categories seen from 360-degree viewpoints\ncaptured with multiple LiDAR sensors. Our extensive experiments on the\nscene-level KITTI-360 dataset, and on our object-level NeRF-MVL show that our\nLiDAR-NeRF surpasses the model-based algorithms significantly.\n", "date": "2023-04-20", "categories": []}
{"id": "2307.05409", "title": "3D detection of roof sections from a single satellite image and\n  application to LOD2-building reconstruction", "abstract": "  Reconstructing urban areas in 3D out of satellite raster images has been a\nlong-standing and challenging goal of both academical and industrial research.\nThe rare methods today achieving this objective at a Level Of Details $2$ rely\non procedural approaches based on geometry, and need stereo images and/or LIDAR\ndata as input. We here propose a method for urban 3D reconstruction named\nKIBS(\\textit{Keypoints Inference By Segmentation}), which comprises two novel\nfeatures: i) a full deep learning approach for the 3D detection of the roof\nsections, and ii) only one single (non-orthogonal) satellite raster image as\nmodel input. This is achieved in two steps: i) by a Mask R-CNN model performing\na 2D segmentation of the buildings' roof sections, and after blending these\nlatter segmented pixels within the RGB satellite raster image, ii) by another\nidentical Mask R-CNN model inferring the heights-to-ground of the roof\nsections' corners via panoptic segmentation, unto full 3D reconstruction of the\nbuildings and city. We demonstrate the potential of the KIBS method by\nreconstructing different urban areas in a few minutes, with a Jaccard index for\nthe 2D segmentation of individual roof sections of $88.55\\%$ and $75.21\\%$ on\nour two data sets resp., and a height's mean error of such correctly segmented\npixels for the 3D reconstruction of $1.60$ m and $2.06$ m on our two data sets\nresp., hence within the LOD2 precision range.\n", "date": "2023-07-11", "categories": []}
{"id": "2306.17643", "title": "Neural 3D Scene Reconstruction from Multiple 2D Images without 3D\n  Supervision", "abstract": "  Neural 3D scene reconstruction methods have achieved impressive performance\nwhen reconstructing complex geometry and low-textured regions in indoor scenes.\nHowever, these methods heavily rely on 3D data which is costly and\ntime-consuming to obtain in real world. In this paper, we propose a novel\nneural reconstruction method that reconstructs scenes using sparse depth under\nthe plane constraints without 3D supervision. We introduce a signed distance\nfunction field, a color field, and a probability field to represent a scene. We\noptimize these fields to reconstruct the scene by using differentiable ray\nmarching with accessible 2D images as supervision. We improve the\nreconstruction quality of complex geometry scene regions with sparse depth\nobtained by using the geometric constraints. The geometric constraints project\n3D points on the surface to similar-looking regions with similar features in\ndifferent 2D images. We impose the plane constraints to make large planes\nparallel or vertical to the indoor floor. Both two constraints help reconstruct\naccurate and smooth geometry structures of the scene. Without 3D supervision,\nour method achieves competitive performance compared with existing methods that\nuse 3D supervision on the ScanNet dataset.\n", "date": "2023-06-30", "categories": []}
{"id": "2306.16928", "title": "One-2-3-45: Any Single Image to 3D Mesh in 45 Seconds without Per-Shape\n  Optimization", "abstract": "  Single image 3D reconstruction is an important but challenging task that\nrequires extensive knowledge of our natural world. Many existing methods solve\nthis problem by optimizing a neural radiance field under the guidance of 2D\ndiffusion models but suffer from lengthy optimization time, 3D inconsistency\nresults, and poor geometry. In this work, we propose a novel method that takes\na single image of any object as input and generates a full 360-degree 3D\ntextured mesh in a single feed-forward pass. Given a single image, we first use\na view-conditioned 2D diffusion model, Zero123, to generate multi-view images\nfor the input view, and then aim to lift them up to 3D space. Since traditional\nreconstruction methods struggle with inconsistent multi-view predictions, we\nbuild our 3D reconstruction module upon an SDF-based generalizable neural\nsurface reconstruction method and propose several critical training strategies\nto enable the reconstruction of 360-degree meshes. Without costly\noptimizations, our method reconstructs 3D shapes in significantly less time\nthan existing methods. Moreover, our method favors better geometry, generates\nmore 3D consistent results, and adheres more closely to the input image. We\nevaluate our approach on both synthetic data and in-the-wild images and\ndemonstrate its superiority in terms of both mesh quality and runtime. In\naddition, our approach can seamlessly support the text-to-3D task by\nintegrating with off-the-shelf text-to-image diffusion models.\n", "date": "2023-06-29", "categories": []}
{"id": "2306.15669", "title": "Detector-Free Structure from Motion", "abstract": "  We propose a new structure-from-motion framework to recover accurate camera\nposes and point clouds from unordered images. Traditional SfM systems typically\nrely on the successful detection of repeatable keypoints across multiple views\nas the first step, which is difficult for texture-poor scenes, and poor\nkeypoint detection may break down the whole SfM system. We propose a new\ndetector-free SfM framework to draw benefits from the recent success of\ndetector-free matchers to avoid the early determination of keypoints, while\nsolving the multi-view inconsistency issue of detector-free matchers.\nSpecifically, our framework first reconstructs a coarse SfM model from\nquantized detector-free matches. Then, it refines the model by a novel\niterative refinement pipeline, which iterates between an attention-based\nmulti-view matching module to refine feature tracks and a geometry refinement\nmodule to improve the reconstruction accuracy. Experiments demonstrate that the\nproposed framework outperforms existing detector-based SfM systems on common\nbenchmark datasets. We also collect a texture-poor SfM dataset to demonstrate\nthe capability of our framework to reconstruct texture-poor scenes. Based on\nthis framework, we take $\\textit{first place}$ in Image Matching Challenge\n2023.\n", "date": "2023-06-27", "categories": []}
{"id": "2306.02851", "title": "Scene as Occupancy", "abstract": "  Human driver can easily describe the complex traffic scene by visual system.\nSuch an ability of precise perception is essential for driver's planning. To\nachieve this, a geometry-aware representation that quantizes the physical 3D\nscene into structured grid map with semantic labels per cell, termed as 3D\nOccupancy, would be desirable. Compared to the form of bounding box, a key\ninsight behind occupancy is that it could capture the fine-grained details of\ncritical obstacles in the scene, and thereby facilitate subsequent tasks. Prior\nor concurrent literature mainly concentrate on a single scene completion task,\nwhere we might argue that the potential of this occupancy representation might\nobsess broader impact. In this paper, we propose OccNet, a multi-view\nvision-centric pipeline with a cascade and temporal voxel decoder to\nreconstruct 3D occupancy. At the core of OccNet is a general occupancy\nembedding to represent 3D physical world. Such a descriptor could be applied\ntowards a wide span of driving tasks, including detection, segmentation and\nplanning. To validate the effectiveness of this new representation and our\nproposed algorithm, we propose OpenOcc, the first dense high-quality 3D\noccupancy benchmark built on top of nuScenes. Empirical experiments show that\nthere are evident performance gain across multiple tasks, e.g., motion planning\ncould witness a collision rate reduction by 15%-58%, demonstrating the\nsuperiority of our method.\n", "date": "2023-06-05", "categories": []}
{"id": "2211.16193", "title": "In-Hand 3D Object Scanning from an RGB Sequence", "abstract": "  We propose a method for in-hand 3D scanning of an unknown object with a\nmonocular camera. Our method relies on a neural implicit surface representation\nthat captures both the geometry and the appearance of the object, however, by\ncontrast with most NeRF-based methods, we do not assume that the camera-object\nrelative poses are known. Instead, we simultaneously optimize both the object\nshape and the pose trajectory. As direct optimization over all shape and pose\nparameters is prone to fail without coarse-level initialization, we propose an\nincremental approach that starts by splitting the sequence into carefully\nselected overlapping segments within which the optimization is likely to\nsucceed. We reconstruct the object shape and track its poses independently\nwithin each segment, then merge all the segments before performing a global\noptimization. We show that our method is able to reconstruct the shape and\ncolor of both textured and challenging texture-less objects, outperforms\nclassical methods that rely only on appearance features, and that its\nperformance is close to recent methods that assume known camera poses.\n", "date": "2022-11-28", "categories": []}
{"id": "2304.10950", "title": "Factored Neural Representation for Scene Understanding", "abstract": "  A long-standing goal in scene understanding is to obtain interpretable and\neditable representations that can be directly constructed from a raw monocular\nRGB-D video, without requiring specialized hardware setup or priors. The\nproblem is significantly more challenging in the presence of multiple moving\nand/or deforming objects. Traditional methods have approached the setup with a\nmix of simplifications, scene priors, pretrained templates, or known\ndeformation models. The advent of neural representations, especially neural\nimplicit representations and radiance fields, opens the possibility of\nend-to-end optimization to collectively capture geometry, appearance, and\nobject motion. However, current approaches produce global scene encoding,\nassume multiview capture with limited or no motion in the scenes, and do not\nfacilitate easy manipulation beyond novel view synthesis. In this work, we\nintroduce a factored neural scene representation that can directly be learned\nfrom a monocular RGB-D video to produce object-level neural presentations with\nan explicit encoding of object movement (e.g., rigid trajectory) and/or\ndeformations (e.g., nonrigid movement). We evaluate ours against a set of\nneural approaches on both synthetic and real data to demonstrate that the\nrepresentation is efficient, interpretable, and editable (e.g., change object\ntrajectory). Code and data are available at\nhttp://geometry.cs.ucl.ac.uk/projects/2023/factorednerf .\n", "date": "2023-04-21", "categories": []}
{"id": "2305.02756", "title": "Floaters No More: Radiance Field Gradient Scaling for Improved\n  Near-Camera Training", "abstract": "  NeRF acquisition typically requires careful choice of near planes for the\ndifferent cameras or suffers from background collapse, creating floating\nartifacts on the edges of the captured scene. The key insight of this work is\nthat background collapse is caused by a higher density of samples in regions\nnear cameras. As a result of this sampling imbalance, near-camera volumes\nreceive significantly more gradients, leading to incorrect density buildup. We\npropose a gradient scaling approach to counter-balance this sampling imbalance,\nremoving the need for near planes, while preventing background collapse. Our\nmethod can be implemented in a few lines, does not induce any significant\noverhead, and is compatible with most NeRF implementations.\n", "date": "2023-05-04", "categories": []}
{"id": "2306.06300", "title": "NERFBK: A High-Quality Benchmark for NERF-Based 3D Reconstruction", "abstract": "  This paper introduces a new real and synthetic dataset called NeRFBK\nspecifically designed for testing and comparing NeRF-based 3D reconstruction\nalgorithms. High-quality 3D reconstruction has significant potential in various\nfields, and advancements in image-based algorithms make it essential to\nevaluate new advanced techniques. However, gathering diverse data with precise\nground truth is challenging and may not encompass all relevant applications.\nThe NeRFBK dataset addresses this issue by providing multi-scale, indoor and\noutdoor datasets with high-resolution images and videos and camera parameters\nfor testing and comparing NeRF-based algorithms. This paper presents the design\nand creation of the NeRFBK benchmark, various examples and application\nscenarios, and highlights its potential for advancing the field of 3D\nreconstruction.\n", "date": "2023-06-09", "categories": []}
{"id": "2306.08671", "title": "Learning to Predict Scene-Level Implicit 3D from Posed RGBD Data", "abstract": "  We introduce a method that can learn to predict scene-level implicit\nfunctions for 3D reconstruction from posed RGBD data. At test time, our system\nmaps a previously unseen RGB image to a 3D reconstruction of a scene via\nimplicit functions. While implicit functions for 3D reconstruction have often\nbeen tied to meshes, we show that we can train one using only a set of posed\nRGBD images. This setting may help 3D reconstruction unlock the sea of\naccelerometer+RGBD data that is coming with new phones. Our system, D2-DRDF,\ncan match and sometimes outperform current methods that use mesh supervision\nand shows better robustness to sparse data.\n", "date": "2023-06-14", "categories": []}
{"id": "2306.03092", "title": "Neuralangelo: High-Fidelity Neural Surface Reconstruction", "abstract": "  Neural surface reconstruction has been shown to be powerful for recovering\ndense 3D surfaces via image-based neural rendering. However, current methods\nstruggle to recover detailed structures of real-world scenes. To address the\nissue, we present Neuralangelo, which combines the representation power of\nmulti-resolution 3D hash grids with neural surface rendering. Two key\ningredients enable our approach: (1) numerical gradients for computing\nhigher-order derivatives as a smoothing operation and (2) coarse-to-fine\noptimization on the hash grids controlling different levels of details. Even\nwithout auxiliary inputs such as depth, Neuralangelo can effectively recover\ndense 3D surface structures from multi-view images with fidelity significantly\nsurpassing previous methods, enabling detailed large-scale scene reconstruction\nfrom RGB video captures.\n", "date": "2023-06-05", "categories": []}
{"id": "2301.09060", "title": "3D Reconstruction of Non-cooperative Resident Space Objects using\n  Instant NGP-accelerated NeRF and D-NeRF", "abstract": "  The proliferation of non-cooperative resident space objects (RSOs) in orbit\nhas spurred the demand for active space debris removal, on-orbit servicing\n(OOS), classification, and functionality identification of these RSOs. Recent\nadvances in computer vision have enabled high-definition 3D modeling of objects\nbased on a set of 2D images captured from different viewing angles. This work\nadapts Instant NeRF and D-NeRF, variations of the neural radiance field (NeRF)\nalgorithm to the problem of mapping RSOs in orbit for the purposes of\nfunctionality identification and assisting with OOS. The algorithms are\nevaluated for 3D reconstruction quality and hardware requirements using\ndatasets of images of a spacecraft mock-up taken under two different lighting\nand motion conditions at the Orbital Robotic Interaction, On-Orbit Servicing\nand Navigation (ORION) Laboratory at Florida Institute of Technology. Instant\nNeRF is shown to learn high-fidelity 3D models with a computational cost that\ncould feasibly be trained on on-board computers.\n", "date": "2023-01-22", "categories": []}
{"id": "2306.06360", "title": "3D reconstruction using Structure for Motion", "abstract": "  We are working towards 3D reconstruction of indoor spaces using a pair of HDR\ncameras in a stereo vision configuration mounted on an indoor mobile floor\nrobot that captures various textures and spatial features as 2D images and this\ndata is simultaneously utilized as a feed to our algorithm which will allow us\nto visualize the depth map.\n", "date": "2023-06-10", "categories": []}
{"id": "2212.09948", "title": "MM-3DScene: 3D Scene Understanding by Customizing Masked Modeling with\n  Informative-Preserved Reconstruction and Self-Distilled Consistency", "abstract": "  Masked Modeling (MM) has demonstrated widespread success in various vision\nchallenges, by reconstructing masked visual patches. Yet, applying MM for\nlarge-scale 3D scenes remains an open problem due to the data sparsity and\nscene complexity. The conventional random masking paradigm used in 2D images\noften causes a high risk of ambiguity when recovering the masked region of 3D\nscenes. To this end, we propose a novel informative-preserved reconstruction,\nwhich explores local statistics to discover and preserve the representative\nstructured points, effectively enhancing the pretext masking task for 3D scene\nunderstanding. Integrated with a progressive reconstruction manner, our method\ncan concentrate on modeling regional geometry and enjoy less ambiguity for\nmasked reconstruction. Besides, such scenes with progressive masking ratios can\nalso serve to self-distill their intrinsic spatial consistency, requiring to\nlearn the consistent representations from unmasked areas. By elegantly\ncombining informative-preserved reconstruction on masked areas and consistency\nself-distillation from unmasked areas, a unified framework called MM-3DScene is\nyielded. We conduct comprehensive experiments on a host of downstream tasks.\nThe consistent improvement (e.g., +6.1 mAP@0.5 on object detection and +2.2%\nmIoU on semantic segmentation) demonstrates the superiority of our approach.\n", "date": "2022-12-20", "categories": []}
{"id": "2305.19590", "title": "Neural Kernel Surface Reconstruction", "abstract": "  We present a novel method for reconstructing a 3D implicit surface from a\nlarge-scale, sparse, and noisy point cloud. Our approach builds upon the\nrecently introduced Neural Kernel Fields (NKF) representation. It enjoys\nsimilar generalization capabilities to NKF, while simultaneously addressing its\nmain limitations: (a) We can scale to large scenes through compactly supported\nkernel functions, which enable the use of memory-efficient sparse linear\nsolvers. (b) We are robust to noise, through a gradient fitting solve. (c) We\nminimize training requirements, enabling us to learn from any dataset of dense\noriented points, and even mix training data consisting of objects and scenes at\ndifferent scales. Our method is capable of reconstructing millions of points in\na few seconds, and handling very large scenes in an out-of-core fashion. We\nachieve state-of-the-art results on reconstruction benchmarks consisting of\nsingle objects, indoor scenes, and outdoor scenes.\n", "date": "2023-05-31", "categories": []}
{"id": "2212.14871", "title": "Equivariant Light Field Convolution and Transformer", "abstract": "  3D reconstruction and novel view rendering can greatly benefit from geometric\npriors when the input views are not sufficient in terms of coverage and\ninter-view baselines. Deep learning of geometric priors from 2D images often\nrequires each image to be represented in a $2D$ canonical frame and the prior\nto be learned in a given or learned $3D$ canonical frame. In this paper, given\nonly the relative poses of the cameras, we show how to learn priors from\nmultiple views equivariant to coordinate frame transformations by proposing an\n$SE(3)$-equivariant convolution and transformer in the space of rays in 3D.\nThis enables the creation of a light field that remains equivariant to the\nchoice of coordinate frame. The light field as defined in our work, refers both\nto the radiance field and the feature field defined on the ray space. We model\nthe ray space, the domain of the light field, as a homogeneous space of $SE(3)$\nand introduce the $SE(3)$-equivariant convolution in ray space. Depending on\nthe output domain of the convolution, we present convolution-based\n$SE(3)$-equivariant maps from ray space to ray space and to $\\mathbb{R}^3$. Our\nmathematical framework allows us to go beyond convolution to\n$SE(3)$-equivariant attention in the ray space. We demonstrate how to tailor\nand adapt the equivariant convolution and transformer in the tasks of\nequivariant neural rendering and $3D$ reconstruction from multiple views. We\ndemonstrate $SE(3)$-equivariance by obtaining robust results in roto-translated\ndatasets without performing transformation augmentation.\n", "date": "2022-12-30", "categories": []}
{"id": "2306.04988", "title": "StreetSurf: Extending Multi-view Implicit Surface Reconstruction to\n  Street Views", "abstract": "  We present a novel multi-view implicit surface reconstruction technique,\ntermed StreetSurf, that is readily applicable to street view images in\nwidely-used autonomous driving datasets, such as Waymo-perception sequences,\nwithout necessarily requiring LiDAR data. As neural rendering research expands\nrapidly, its integration into street views has started to draw interests.\nExisting approaches on street views either mainly focus on novel view synthesis\nwith little exploration of the scene geometry, or rely heavily on dense LiDAR\ndata when investigating reconstruction. Neither of them investigates multi-view\nimplicit surface reconstruction, especially under settings without LiDAR data.\nOur method extends prior object-centric neural surface reconstruction\ntechniques to address the unique challenges posed by the unbounded street views\nthat are captured with non-object-centric, long and narrow camera trajectories.\nWe delimit the unbounded space into three parts, close-range, distant-view and\nsky, with aligned cuboid boundaries, and adapt cuboid/hyper-cuboid hash-grids\nalong with road-surface initialization scheme for finer and disentangled\nrepresentation. To further address the geometric errors arising from\ntextureless regions and insufficient viewing angles, we adopt geometric priors\nthat are estimated using general purpose monocular models. Coupled with our\nimplementation of efficient and fine-grained multi-stage ray marching strategy,\nwe achieve state of the art reconstruction quality in both geometry and\nappearance within only one to two hours of training time with a single RTX3090\nGPU for each street view sequence. Furthermore, we demonstrate that the\nreconstructed implicit surfaces have rich potential for various downstream\ntasks, including ray tracing and LiDAR simulation.\n", "date": "2023-06-08", "categories": []}
{"id": "2306.05303", "title": "Enhance-NeRF: Multiple Performance Evaluation for Neural Radiance Fields", "abstract": "  The quality of three-dimensional reconstruction is a key factor affecting the\neffectiveness of its application in areas such as virtual reality (VR) and\naugmented reality (AR) technologies. Neural Radiance Fields (NeRF) can generate\nrealistic images from any viewpoint. It simultaneously reconstructs the shape,\nlighting, and materials of objects, and without surface defects, which breaks\ndown the barrier between virtuality and reality. The potential spatial\ncorrespondences displayed by NeRF between reconstructed scenes and real-world\nscenes offer a wide range of practical applications possibilities. Despite\nsignificant progress in 3D reconstruction since NeRF were introduced, there\nremains considerable room for exploration and experimentation. NeRF-based\nmodels are susceptible to interference issues caused by colored \"fog\" noise.\nAdditionally, they frequently encounter instabilities and failures while\nattempting to reconstruct unbounded scenes. Moreover, the model takes a\nsignificant amount of time to converge, making it even more challenging to use\nin such scenarios. Our approach, coined Enhance-NeRF, which adopts joint color\nto balance low and high reflectivity objects display, utilizes a decoding\narchitecture with prior knowledge to improve recognition, and employs\nmulti-layer performance evaluation mechanisms to enhance learning capacity. It\nachieves reconstruction of outdoor scenes within one hour under single-card\ncondition. Based on experimental results, Enhance-NeRF partially enhances\nfitness capability and provides some support to outdoor scene reconstruction.\nThe Enhance-NeRF method can be used as a plug-and-play component, making it\neasy to integrate with other NeRF-based models. The code is available at:\nhttps://github.com/TANQIanQ/Enhance-NeRF\n", "date": "2023-06-08", "categories": []}
{"id": "2306.03747", "title": "Towards Scalable Multi-View Reconstruction of Geometry and Materials", "abstract": "  In this paper, we propose a novel method for joint recovery of camera pose,\nobject geometry and spatially-varying Bidirectional Reflectance Distribution\nFunction (svBRDF) of 3D scenes that exceed object-scale and hence cannot be\ncaptured with stationary light stages. The input are high-resolution RGB-D\nimages captured by a mobile, hand-held capture system with point lights for\nactive illumination. Compared to previous works that jointly estimate geometry\nand materials from a hand-held scanner, we formulate this problem using a\nsingle objective function that can be minimized using off-the-shelf\ngradient-based solvers. To facilitate scalability to large numbers of\nobservation views and optimization variables, we introduce a distributed\noptimization algorithm that reconstructs 2.5D keyframe-based representations of\nthe scene. A novel multi-view consistency regularizer effectively synchronizes\nneighboring keyframes such that the local optimization results allow for\nseamless integration into a globally consistent 3D model. We provide a study on\nthe importance of each component in our formulation and show that our method\ncompares favorably to baselines. We further demonstrate that our method\naccurately reconstructs various objects and materials and allows for expansion\nto spatially larger scenes. We believe that this work represents a significant\nstep towards making geometry and material estimation from hand-held scanners\nscalable.\n", "date": "2023-06-06", "categories": []}
{"id": "2210.07181", "title": "MonoNeRF: Learning Generalizable NeRFs from Monocular Videos without\n  Camera Pose", "abstract": "  We propose a generalizable neural radiance fields - MonoNeRF, that can be\ntrained on large-scale monocular videos of moving in static scenes without any\nground-truth annotations of depth and camera poses. MonoNeRF follows an\nAutoencoder-based architecture, where the encoder estimates the monocular depth\nand the camera pose, and the decoder constructs a Multiplane NeRF\nrepresentation based on the depth encoder feature, and renders the input frames\nwith the estimated camera. The learning is supervised by the reconstruction\nerror. Once the model is learned, it can be applied to multiple applications\nincluding depth estimation, camera pose estimation, and single-image novel view\nsynthesis. More qualitative results are available at:\nhttps://oasisyang.github.io/mononerf .\n", "date": "2022-10-13", "categories": []}
{"id": "2303.04341", "title": "Neural Vector Fields: Implicit Representation by Explicit Learning", "abstract": "  Deep neural networks (DNNs) are widely applied for nowadays 3D surface\nreconstruction tasks and such methods can be further divided into two\ncategories, which respectively warp templates explicitly by moving vertices or\nrepresent 3D surfaces implicitly as signed or unsigned distance functions.\nTaking advantage of both advanced explicit learning process and powerful\nrepresentation ability of implicit functions, we propose a novel 3D\nrepresentation method, Neural Vector Fields (NVF). It not only adopts the\nexplicit learning process to manipulate meshes directly, but also leverages the\nimplicit representation of unsigned distance functions (UDFs) to break the\nbarriers in resolution and topology. Specifically, our method first predicts\nthe displacements from queries towards the surface and models the shapes as\n\\textit{Vector Fields}. Rather than relying on network differentiation to\nobtain direction fields as most existing UDF-based methods, the produced vector\nfields encode the distance and direction fields both and mitigate the ambiguity\nat \"ridge\" points, such that the calculation of direction fields is\nstraightforward and differentiation-free. The differentiation-free\ncharacteristic enables us to further learn a shape codebook via Vector\nQuantization, which encodes the cross-object priors, accelerates the training\nprocedure, and boosts model generalization on cross-category reconstruction.\nThe extensive experiments on surface reconstruction benchmarks indicate that\nour method outperforms those state-of-the-art methods in different evaluation\nscenarios including watertight vs non-watertight shapes, category-specific vs\ncategory-agnostic reconstruction, category-unseen reconstruction, and\ncross-domain reconstruction. Our code is released at\nhttps://github.com/Wi-sc/NVF.\n", "date": "2023-03-08", "categories": []}
{"id": "2306.00075", "title": "CAROM Air -- Vehicle Localization and Traffic Scene Reconstruction from\n  Aerial Videos", "abstract": "  Road traffic scene reconstruction from videos has been desirable by road\nsafety regulators, city planners, researchers, and autonomous driving\ntechnology developers. However, it is expensive and unnecessary to cover every\nmile of the road with cameras mounted on the road infrastructure. This paper\npresents a method that can process aerial videos to vehicle trajectory data so\nthat a traffic scene can be automatically reconstructed and accurately\nre-simulated using computers. On average, the vehicle localization error is\nabout 0.1 m to 0.3 m using a consumer-grade drone flying at 120 meters. This\nproject also compiles a dataset of 50 reconstructed road traffic scenes from\nabout 100 hours of aerial videos to enable various downstream traffic analysis\napplications and facilitate further road traffic related research. The dataset\nis available at https://github.com/duolu/CAROM.\n", "date": "2023-05-31", "categories": []}
{"id": "2306.00180", "title": "FlowCam: Training Generalizable 3D Radiance Fields without Camera Poses\n  via Pixel-Aligned Scene Flow", "abstract": "  Reconstruction of 3D neural fields from posed images has emerged as a\npromising method for self-supervised representation learning. The key challenge\npreventing the deployment of these 3D scene learners on large-scale video data\nis their dependence on precise camera poses from structure-from-motion, which\nis prohibitively expensive to run at scale. We propose a method that jointly\nreconstructs camera poses and 3D neural scene representations online and in a\nsingle forward pass. We estimate poses by first lifting frame-to-frame optical\nflow to 3D scene flow via differentiable rendering, preserving locality and\nshift-equivariance of the image processing backbone. SE(3) camera pose\nestimation is then performed via a weighted least-squares fit to the scene flow\nfield. This formulation enables us to jointly supervise pose estimation and a\ngeneralizable neural scene representation via re-rendering the input video, and\nthus, train end-to-end and fully self-supervised on real-world video datasets.\nWe demonstrate that our method performs robustly on diverse, real-world video,\nnotably on sequences traditionally challenging to optimization-based pose\nestimation techniques.\n", "date": "2023-05-31", "categories": []}
{"id": "2305.17916", "title": "Volume Feature Rendering for Fast Neural Radiance Field Reconstruction", "abstract": "  Neural radiance fields (NeRFs) are able to synthesize realistic novel views\nfrom multi-view images captured from distinct positions and perspectives. In\nNeRF's rendering pipeline, neural networks are used to represent a scene\nindependently or transform queried learnable feature vector of a point to the\nexpected color or density. With the aid of geometry guides either in occupancy\ngrids or proposal networks, the number of neural network evaluations can be\nreduced from hundreds to dozens in the standard volume rendering framework.\nInstead of rendering yielded color after neural network evaluation, we propose\nto render the queried feature vectors of a ray first and then transform the\nrendered feature vector to the final pixel color by a neural network. This\nfundamental change to the standard volume rendering framework requires only one\nsingle neural network evaluation to render a pixel, which substantially lowers\nthe high computational complexity of the rendering framework attributed to a\nlarge number of neural network evaluations. Consequently, we can use a\ncomparably larger neural network to achieve a better rendering quality while\nmaintaining the same training and rendering time costs. Our model achieves the\nstate-of-the-art rendering quality on both synthetic and real-world datasets\nwhile requiring a training time of several minutes.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.17858", "title": "FastMESH: Fast Surface Reconstruction by Hexagonal Mesh-based Neural\n  Rendering", "abstract": "  Despite the promising results of multi-view reconstruction, the recent neural\nrendering-based methods, such as implicit surface rendering (IDR) and volume\nrendering (NeuS), not only incur a heavy computational burden on training but\nalso have the difficulties in disentangling the geometric and appearance.\nAlthough having achieved faster training speed than implicit representation and\nhash coding, the explicit voxel-based method obtains the inferior results on\nrecovering surface. To address these challenges, we propose an effective\nmesh-based neural rendering approach, named FastMESH, which only samples at the\nintersection of ray and mesh. A coarse-to-fine scheme is introduced to\nefficiently extract the initial mesh by space carving. More importantly, we\nsuggest a hexagonal mesh model to preserve surface regularity by constraining\nthe second-order derivatives of vertices, where only low level of positional\nencoding is engaged for neural rendering. The experiments demonstrate that our\napproach achieves the state-of-the-art results on both reconstruction and novel\nview synthesis. Besides, we obtain 10-fold acceleration on training comparing\nto the implicit representation-based methods.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.18163", "title": "Compact Real-time Radiance Fields with Neural Codebook", "abstract": "  Reconstructing neural radiance fields with explicit volumetric\nrepresentations, demonstrated by Plenoxels, has shown remarkable advantages on\ntraining and rendering efficiency, while grid-based representations typically\ninduce considerable overhead for storage and transmission. In this work, we\npresent a simple and effective framework for pursuing compact radiance fields\nfrom the perspective of compression methodology. By exploiting intrinsic\nproperties exhibiting in grid models, a non-uniform compression stem is\ndeveloped to significantly reduce model complexity and a novel parameterized\nmodule, named Neural Codebook, is introduced for better encoding high-frequency\ndetails specific to per-scene models via a fast optimization. Our approach can\nachieve over 40 $\\times$ reduction on grid model storage with competitive\nrendering quality. In addition, the method can achieve real-time rendering\nspeed with 180 fps, realizing significant advantage on storage cost compared to\nreal-time rendering methods.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.14918", "title": "Incremental Dense Reconstruction from Monocular Video with Guided Sparse\n  Feature Volume Fusion", "abstract": "  Incrementally recovering 3D dense structures from monocular videos is of\nparamount importance since it enables various robotics and AR applications.\nFeature volumes have recently been shown to enable efficient and accurate\nincremental dense reconstruction without the need to first estimate depth, but\nthey are not able to achieve as high of a resolution as depth-based methods due\nto the large memory consumption of high-resolution feature volumes. This letter\nproposes a real-time feature volume-based dense reconstruction method that\npredicts TSDF (Truncated Signed Distance Function) values from a novel\nsparsified deep feature volume, which is able to achieve higher resolutions\nthan previous feature volume-based methods, and is favorable in large-scale\noutdoor scenarios where the majority of voxels are empty. An uncertainty-aware\nmulti-view stereo (MVS) network is leveraged to infer initial voxel locations\nof the physical surface in a sparse feature volume. Then for refining the\nrecovered 3D geometry, deep features are attentively aggregated from multiview\nimages at potential surface locations, and temporally fused. Besides achieving\nhigher resolutions than before, our method is shown to produce more complete\nreconstructions with finer detail in many cases. Extensive evaluations on both\npublic and self-collected datasets demonstrate a very competitive real-time\nreconstruction result for our method compared to state-of-the-art\nreconstruction methods in both indoor and outdoor settings.\n", "date": "2023-05-24", "categories": []}
{"id": "2305.13220", "title": "Fast Monocular Scene Reconstruction with Global-Sparse Local-Dense Grids", "abstract": "  Indoor scene reconstruction from monocular images has long been sought after\nby augmented reality and robotics developers. Recent advances in neural field\nrepresentations and monocular priors have led to remarkable results in\nscene-level surface reconstructions. The reliance on Multilayer Perceptrons\n(MLP), however, significantly limits speed in training and rendering. In this\nwork, we propose to directly use signed distance function (SDF) in sparse voxel\nblock grids for fast and accurate scene reconstruction without MLPs. Our\nglobally sparse and locally dense data structure exploits surfaces' spatial\nsparsity, enables cache-friendly queries, and allows direct extensions to\nmulti-modal data such as color and semantic labels. To apply this\nrepresentation to monocular scene reconstruction, we develop a scale\ncalibration algorithm for fast geometric initialization from monocular depth\npriors. We apply differentiable volume rendering from this initialization to\nrefine details with fast convergence. We also introduce efficient\nhigh-dimensional Continuous Random Fields (CRFs) to further exploit the\nsemantic-geometry consistency between scene objects. Experiments show that our\napproach is 10x faster in training and 100x faster in rendering while achieving\ncomparable accuracy to state-of-the-art neural implicit methods.\n", "date": "2023-05-22", "categories": []}
{"id": "2302.04495", "title": "3D reconstruction from spherical images: A review of techniques,\n  applications, and prospects", "abstract": "  3D reconstruction plays an increasingly important role in modern\nphotogrammetric systems. Conventional satellite or aerial-based remote sensing\n(RS) platforms can provide the necessary data sources for the 3D reconstruction\nof large-scale landforms and cities. Even with low-altitude UAVs (Unmanned\nAerial Vehicles), 3D reconstruction in complicated situations, such as urban\ncanyons and indoor scenes, is challenging due to frequent tracking failures\nbetween camera frames and high data collection costs. Recently, spherical\nimages have been extensively used due to the capability of recording\nsurrounding environments from one camera exposure. In contrast to perspective\nimages with limited FOV (Field of View), spherical images can cover the whole\nscene with full horizontal and vertical FOV and facilitate camera tracking and\ndata acquisition in these complex scenes. With the rapid evolution and\nextensive use of professional and consumer-grade spherical cameras, spherical\nimages show great potential for the 3D modeling of urban and indoor scenes.\nClassical 3D reconstruction pipelines, however, cannot be directly used for\nspherical images. Besides, there exist few software packages that are designed\nfor the 3D reconstruction of spherical images. As a result, this research\nprovides a thorough survey of the state-of-the-art for 3D reconstruction of\nspherical images in terms of data acquisition, feature detection and matching,\nimage orientation, and dense matching as well as presenting promising\napplications and discussing potential prospects. We anticipate that this study\noffers insightful clues to direct future research.\n", "date": "2023-02-09", "categories": []}
{"id": "2305.11031", "title": "ConsistentNeRF: Enhancing Neural Radiance Fields with 3D Consistency for\n  Sparse View Synthesis", "abstract": "  Neural Radiance Fields (NeRF) has demonstrated remarkable 3D reconstruction\ncapabilities with dense view images. However, its performance significantly\ndeteriorates under sparse view settings. We observe that learning the 3D\nconsistency of pixels among different views is crucial for improving\nreconstruction quality in such cases. In this paper, we propose ConsistentNeRF,\na method that leverages depth information to regularize both multi-view and\nsingle-view 3D consistency among pixels. Specifically, ConsistentNeRF employs\ndepth-derived geometry information and a depth-invariant loss to concentrate on\npixels that exhibit 3D correspondence and maintain consistent depth\nrelationships. Extensive experiments on recent representative works reveal that\nour approach can considerably enhance model performance in sparse view\nconditions, achieving improvements of up to 94% in PSNR, 76% in SSIM, and 31%\nin LPIPS compared to the vanilla baselines across various benchmarks, including\nDTU, NeRF Synthetic, and LLFF.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.11102", "title": "Progressive Learning of 3D Reconstruction Network from 2D GAN Data", "abstract": "  This paper presents a method to reconstruct high-quality textured 3D models\nfrom single images. Current methods rely on datasets with expensive\nannotations; multi-view images and their camera parameters. Our method relies\non GAN generated multi-view image datasets which have a negligible annotation\ncost. However, they are not strictly multi-view consistent and sometimes GANs\noutput distorted images. This results in degraded reconstruction qualities. In\nthis work, to overcome these limitations of generated datasets, we have two\nmain contributions which lead us to achieve state-of-the-art results on\nchallenging objects: 1) A robust multi-stage learning scheme that gradually\nrelies more on the models own predictions when calculating losses, 2) A novel\nadversarial learning pipeline with online pseudo-ground truth generations to\nachieve fine details. Our work provides a bridge from 2D supervisions of GAN\nmodels to 3D reconstruction models and removes the expensive annotation\nefforts. We show significant improvements over previous methods whether they\nwere trained on GAN generated multi-view images or on real images with\nexpensive annotations. Please visit our web-page for 3D visuals:\nhttps://research.nvidia.com/labs/adlr/progressive-3d-learning\n", "date": "2023-05-18", "categories": []}
{"id": "2211.02299", "title": "GARNet: Global-Aware Multi-View 3D Reconstruction Network and the\n  Cost-Performance Tradeoff", "abstract": "  Deep learning technology has made great progress in multi-view 3D\nreconstruction tasks. At present, most mainstream solutions establish the\nmapping between views and shape of an object by assembling the networks of 2D\nencoder and 3D decoder as the basic structure while they adopt different\napproaches to obtain aggregation of features from several views. Among them,\nthe methods using attention-based fusion perform better and more stable than\nthe others, however, they still have an obvious shortcoming -- the strong\nindependence of each view during predicting the weights for merging leads to a\nlack of adaption of the global state. In this paper, we propose a global-aware\nattention-based fusion approach that builds the correlation between each branch\nand the global to provide a comprehensive foundation for weights inference. In\norder to enhance the ability of the network, we introduce a novel loss function\nto supervise the shape overall and propose a dynamic two-stage training\nstrategy that can effectively adapt to all reconstructors with attention-based\nfusion. Experiments on ShapeNet verify that our method outperforms existing\nSOTA methods while the amount of parameters is far less than the same type of\nalgorithm, Pix2Vox++. Furthermore, we propose a view-reduction method based on\nmaximizing diversity and discuss the cost-performance tradeoff of our model to\nachieve a better performance when facing heavy input amount and limited\ncomputational cost.\n", "date": "2022-11-04", "categories": []}
{"id": "2303.09952", "title": "Single-view Neural Radiance Fields with Depth Teacher", "abstract": "  Neural Radiance Fields (NeRF) have been proposed for photorealistic novel\nview rendering. However, it requires many different views of one scene for\ntraining. Moreover, it has poor generalizations to new scenes and requires\nretraining or fine-tuning on each scene. In this paper, we develop a new NeRF\nmodel for novel view synthesis using only a single image as input. We propose\nto combine the (coarse) planar rendering and the (fine) volume rendering to\nachieve higher rendering quality and better generalizations. We also design a\ndepth teacher net that predicts dense pseudo depth maps to supervise the joint\nrendering mechanism and boost the learning of consistent 3D geometry. We\nevaluate our method on three challenging datasets. It outperforms\nstate-of-the-art single-view NeRFs by achieving 5$\\sim$20\\% improvements in\nPSNR and reducing 20$\\sim$50\\% of the errors in the depth rendering. It also\nshows excellent generalization abilities to unseen data without the need to\nfine-tune on each new scene.\n", "date": "2023-03-17", "categories": []}
{"id": "2305.06278", "title": "A Multi-modal Garden Dataset and Hybrid 3D Dense Reconstruction\n  Framework Based on Panoramic Stereo Images for a Trimming Robot", "abstract": "  Recovering an outdoor environment's surface mesh is vital for an agricultural\nrobot during task planning and remote visualization. Our proposed solution is\nbased on a newly-designed panoramic stereo camera along with a hybrid novel\nsoftware framework that consists of three fusion modules. The panoramic stereo\ncamera with a pentagon shape consists of 5 stereo vision camera pairs to stream\nsynchronized panoramic stereo images for the following three fusion modules. In\nthe disparity fusion module, rectified stereo images produce the initial\ndisparity maps using multiple stereo vision algorithms. Then, these initial\ndisparity maps, along with the intensity images, are input into a disparity\nfusion network to produce refined disparity maps. Next, the refined disparity\nmaps are converted into full-view point clouds or single-view point clouds for\nthe pose fusion module. The pose fusion module adopts a two-stage\nglobal-coarse-to-local-fine strategy. In the first stage, each pair of\nfull-view point clouds is registered by a global point cloud matching algorithm\nto estimate the transformation for a global pose graph's edge, which\neffectively implements loop closure. In the second stage, a local point cloud\nmatching algorithm is used to match single-view point clouds in different\nnodes. Next, we locally refine the poses of all corresponding edges in the\nglobal pose graph using three proposed rules, thus constructing a refined pose\ngraph. The refined pose graph is optimized to produce a global pose trajectory\nfor volumetric fusion. In the volumetric fusion module, the global poses of all\nthe nodes are used to integrate the single-view point clouds into the volume to\nproduce the mesh of the whole garden. The proposed framework and its three\nfusion modules are tested on a real outdoor garden dataset to show the\nsuperiority of the performance.\n", "date": "2023-05-10", "categories": []}
{"id": "2305.06314", "title": "Scan2LoD3: Reconstructing semantic 3D building models at LoD3 using ray\n  casting and Bayesian networks", "abstract": "  Reconstructing semantic 3D building models at the level of detail (LoD) 3 is\na long-standing challenge. Unlike mesh-based models, they require watertight\ngeometry and object-wise semantics at the fa\\c{c}ade level. The principal\nchallenge of such demanding semantic 3D reconstruction is reliable\nfa\\c{c}ade-level semantic segmentation of 3D input data. We present a novel\nmethod, called Scan2LoD3, that accurately reconstructs semantic LoD3 building\nmodels by improving fa\\c{c}ade-level semantic 3D segmentation. To this end, we\nleverage laser physics and 3D building model priors to probabilistically\nidentify model conflicts. These probabilistic physical conflicts propose\nlocations of model openings: Their final semantics and shapes are inferred in a\nBayesian network fusing multimodal probabilistic maps of conflicts, 3D point\nclouds, and 2D images. To fulfill demanding LoD3 requirements, we use the\nestimated shapes to cut openings in 3D building priors and fit semantic 3D\nobjects from a library of fa\\c{c}ade objects. Extensive experiments on the TUM\ncity campus datasets demonstrate the superior performance of the proposed\nScan2LoD3 over the state-of-the-art methods in fa\\c{c}ade-level detection,\nsemantic segmentation, and LoD3 building model reconstruction. We believe our\nmethod can foster the development of probability-driven semantic 3D\nreconstruction at LoD3 since not only the high-definition reconstruction but\nalso reconstruction confidence becomes pivotal for various applications such as\nautonomous driving and urban simulations.\n", "date": "2023-05-10", "categories": []}
{"id": "2305.05594", "title": "PET-NeuS: Positional Encoding Tri-Planes for Neural Surfaces", "abstract": "  A signed distance function (SDF) parametrized by an MLP is a common\ningredient of neural surface reconstruction. We build on the successful recent\nmethod NeuS to extend it by three new components. The first component is to\nborrow the tri-plane representation from EG3D and represent signed distance\nfields as a mixture of tri-planes and MLPs instead of representing it with MLPs\nonly. Using tri-planes leads to a more expressive data structure but will also\nintroduce noise in the reconstructed surface. The second component is to use a\nnew type of positional encoding with learnable weights to combat noise in the\nreconstruction process. We divide the features in the tri-plane into multiple\nfrequency scales and modulate them with sin and cos functions of different\nfrequencies. The third component is to use learnable convolution operations on\nthe tri-plane features using self-attention convolution to produce features\nwith different frequency bands. The experiments show that PET-NeuS achieves\nhigh-fidelity surface reconstruction on standard datasets. Following previous\nwork and using the Chamfer metric as the most important way to measure surface\nreconstruction quality, we are able to improve upon the NeuS baseline by 57% on\nNerf-synthetic (0.84 compared to 1.97) and by 15.5% on DTU (0.71 compared to\n0.84). The qualitative evaluation reveals how our method can better control the\ninterference of high-frequency noise. Code available at\n\\url{https://github.com/yiqun-wang/PET-NeuS}.\n", "date": "2023-05-09", "categories": []}
{"id": "2212.04666", "title": "Neural Volume Super-Resolution", "abstract": "  Neural volumetric representations have become a widely adopted model for\nradiance fields in 3D scenes. These representations are fully implicit or\nhybrid function approximators of the instantaneous volumetric radiance in a\nscene, which are typically learned from multi-view captures of the scene. We\ninvestigate the new task of neural volume super-resolution - rendering\nhigh-resolution views corresponding to a scene captured at low resolution. To\nthis end, we propose a neural super-resolution network that operates directly\non the volumetric representation of the scene. This approach allows us to\nexploit an advantage of operating in the volumetric domain, namely the ability\nto guarantee consistent super-resolution across different viewing directions.\nTo realize our method, we devise a novel 3D representation that hinges on\nmultiple 2D feature planes. This allows us to super-resolve the 3D scene\nrepresentation by applying 2D convolutional networks on the 2D feature planes.\nWe validate the proposed method by super-resolving multi-view consistent views\non a diverse set of unseen 3D scenes, confirming qualitative and quantitatively\nfavorable quality over existing approaches.\n", "date": "2022-12-09", "categories": []}
{"id": "2305.04296", "title": "HashCC: Lightweight Method to Improve the Quality of the Camera-less\n  NeRF Scene Generation", "abstract": "  Neural Radiance Fields has become a prominent method of scene generation via\nview synthesis. A critical requirement for the original algorithm to learn\nmeaningful scene representation is camera pose information for each image in a\ndata set. Current approaches try to circumnavigate this assumption with\nmoderate success, by learning approximate camera positions alongside learning\nneural representations of a scene. This requires complicated camera models,\ncausing a long and complicated training process, or results in a lack of\ntexture and sharp details in rendered scenes. In this work we introduce Hash\nColor Correction (HashCC) -- a lightweight method for improving Neural Radiance\nFields rendered image quality, applicable also in situations where camera\npositions for a given set of images are unknown.\n", "date": "2023-05-07", "categories": []}
{"id": "2206.08368", "title": "Unbiased 4D: Monocular 4D Reconstruction with a Neural Deformation Model", "abstract": "  Capturing general deforming scenes from monocular RGB video is crucial for\nmany computer graphics and vision applications. However, current approaches\nsuffer from drawbacks such as struggling with large scene deformations,\ninaccurate shape completion or requiring 2D point tracks. In contrast, our\nmethod, Ub4D, handles large deformations, performs shape completion in occluded\nregions, and can operate on monocular RGB videos directly by using\ndifferentiable volume rendering. This technique includes three new in the\ncontext of non-rigid 3D reconstruction components, i.e., 1) A coordinate-based\nand implicit neural representation for non-rigid scenes, which in conjunction\nwith differentiable volume rendering enables an unbiased reconstruction of\ndynamic scenes, 2) a proof that extends the unbiased formulation of volume\nrendering to dynamic scenes, and 3) a novel dynamic scene flow loss, which\nenables the reconstruction of larger deformations by leveraging the coarse\nestimates of other methods. Results on our new dataset, which will be made\npublicly available, demonstrate a clear improvement over the state of the art\nin terms of surface reconstruction accuracy and robustness to large\ndeformations.\n", "date": "2022-06-16", "categories": []}
{"id": "2209.15619", "title": "Point normal orientation and surface reconstruction by incorporating\n  isovalue constraints to Poisson equation", "abstract": "  Oriented normals are common pre-requisites for many geometric algorithms\nbased on point clouds, such as Poisson surface reconstruction. However, it is\nnot trivial to obtain a consistent orientation. In this work, we bridge\norientation and reconstruction in the implicit space and propose a novel\napproach to orient point cloud normals by incorporating isovalue constraints to\nthe Poisson equation. In implicit surface reconstruction, the reconstructed\nshape is represented as an isosurface of an implicit function defined in the\nambient space. Therefore, when such a surface is reconstructed from a set of\nsample points, the implicit function values at the points should be close to\nthe isovalue corresponding to the surface. Based on this observation and the\nPoisson equation, we propose an optimization formulation that combines isovalue\nconstraints with local consistency requirements for normals. We optimize\nnormals and implicit functions simultaneously and solve for a globally\nconsistent orientation. Thanks to the sparsity of the linear system, our method\ncan work on an average laptop with reasonable computational time. Experiments\nshow that our method can achieve high performance in non-uniform and noisy data\nand manage varying sampling densities, artifacts, multiple connected\ncomponents, and nested surfaces. The source code is available at\n\\url{https://github.com/Submanifold/IsoConstraints}.\n", "date": "2022-09-30", "categories": []}
{"id": "2304.14377", "title": "Co-SLAM: Joint Coordinate and Sparse Parametric Encodings for Neural\n  Real-Time SLAM", "abstract": "  We present Co-SLAM, a neural RGB-D SLAM system based on a hybrid\nrepresentation, that performs robust camera tracking and high-fidelity surface\nreconstruction in real time. Co-SLAM represents the scene as a multi-resolution\nhash-grid to exploit its high convergence speed and ability to represent\nhigh-frequency local features. In addition, Co-SLAM incorporates one-blob\nencoding, to encourage surface coherence and completion in unobserved areas.\nThis joint parametric-coordinate encoding enables real-time and robust\nperformance by bringing the best of both worlds: fast convergence and surface\nhole filling. Moreover, our ray sampling strategy allows Co-SLAM to perform\nglobal bundle adjustment over all keyframes instead of requiring keyframe\nselection to maintain a small number of active keyframes as competing neural\nSLAM approaches do. Experimental results show that Co-SLAM runs at 10-17Hz and\nachieves state-of-the-art scene reconstruction results, and competitive\ntracking performance in various datasets and benchmarks (ScanNet, TUM, Replica,\nSynthetic RGBD). Project page: https://hengyiwang.github.io/projects/CoSLAM\n", "date": "2023-04-27", "categories": []}
{"id": "2304.12624", "title": "Shape-Net: Room Layout Estimation from Panoramic Images Robust to\n  Occlusion using Knowledge Distillation with 3D Shapes as Additional Inputs", "abstract": "  Estimating the layout of a room from a single-shot panoramic image is\nimportant in virtual/augmented reality and furniture layout simulation. This\ninvolves identifying three-dimensional (3D) geometry, such as the location of\ncorners and boundaries, and performing 3D reconstruction. However, occlusion is\na common issue that can negatively impact room layout estimation, and this has\nnot been thoroughly studied to date. It is possible to obtain 3D shape\ninformation of rooms as drawings of buildings and coordinates of corners from\nimage datasets, thus we propose providing both 2D panoramic and 3D information\nto a model to effectively deal with occlusion. However, simply feeding 3D\ninformation to a model is not sufficient to utilize the shape information for\nan occluded area. Therefore, we improve the model by introducing 3D\nIntersection over Union (IoU) loss to effectively use 3D information. In some\ncases, drawings are not available or the construction deviates from a drawing.\nConsidering such practical cases, we propose a method for distilling knowledge\nfrom a model trained with both images and 3D information to a model that takes\nonly images as input. The proposed model, which is called Shape-Net, achieves\nstate-of-the-art (SOTA) performance on benchmark datasets. We also confirmed\nits effectiveness in dealing with occlusion through significantly improved\naccuracy on images with occlusion compared with existing models.\n", "date": "2023-04-25", "categories": []}
{"id": "2304.10687", "title": "VisFusion: Visibility-aware Online 3D Scene Reconstruction from Videos", "abstract": "  We propose VisFusion, a visibility-aware online 3D scene reconstruction\napproach from posed monocular videos. In particular, we aim to reconstruct the\nscene from volumetric features. Unlike previous reconstruction methods which\naggregate features for each voxel from input views without considering its\nvisibility, we aim to improve the feature fusion by explicitly inferring its\nvisibility from a similarity matrix, computed from its projected features in\neach image pair. Following previous works, our model is a coarse-to-fine\npipeline including a volume sparsification process. Different from their works\nwhich sparsify voxels globally with a fixed occupancy threshold, we perform the\nsparsification on a local feature volume along each visual ray to preserve at\nleast one voxel per ray for more fine details. The sparse local volume is then\nfused with a global one for online reconstruction. We further propose to\npredict TSDF in a coarse-to-fine manner by learning its residuals across scales\nleading to better TSDF predictions. Experimental results on benchmarks show\nthat our method can achieve superior performance with more scene details. Code\nis available at: https://github.com/huiyu-gao/VisFusion\n", "date": "2023-04-21", "categories": []}
{"id": "2304.10261", "title": "Anything-3D: Towards Single-view Anything Reconstruction in the Wild", "abstract": "  3D reconstruction from a single-RGB image in unconstrained real-world\nscenarios presents numerous challenges due to the inherent diversity and\ncomplexity of objects and environments. In this paper, we introduce\nAnything-3D, a methodical framework that ingeniously combines a series of\nvisual-language models and the Segment-Anything object segmentation model to\nelevate objects to 3D, yielding a reliable and versatile system for single-view\nconditioned 3D reconstruction task. Our approach employs a BLIP model to\ngenerate textural descriptions, utilizes the Segment-Anything model for the\neffective extraction of objects of interest, and leverages a text-to-image\ndiffusion model to lift object into a neural radiance field. Demonstrating its\nability to produce accurate and detailed 3D reconstructions for a wide array of\nobjects, \\emph{Anything-3D\\footnotemark[2]} shows promise in addressing the\nlimitations of existing methodologies. Through comprehensive experiments and\nevaluations on various datasets, we showcase the merits of our approach,\nunderscoring its potential to contribute meaningfully to the field of 3D\nreconstruction. Demos and code will be available at\n\\href{https://github.com/Anything-of-anything/Anything-3D}{https://github.com/Anything-of-anything/Anything-3D}.\n", "date": "2023-04-19", "categories": []}
{"id": "2301.07668", "title": "Behind the Scenes: Density Fields for Single View Reconstruction", "abstract": "  Inferring a meaningful geometric scene representation from a single image is\na fundamental problem in computer vision. Approaches based on traditional depth\nmap prediction can only reason about areas that are visible in the image.\nCurrently, neural radiance fields (NeRFs) can capture true 3D including color,\nbut are too complex to be generated from a single image. As an alternative, we\npropose to predict implicit density fields. A density field maps every location\nin the frustum of the input image to volumetric density. By directly sampling\ncolor from the available views instead of storing color in the density field,\nour scene representation becomes significantly less complex compared to NeRFs,\nand a neural network can predict it in a single forward pass. The prediction\nnetwork is trained through self-supervision from only video data. Our\nformulation allows volume rendering to perform both depth prediction and novel\nview synthesis. Through experiments, we show that our method is able to predict\nmeaningful geometry for regions that are occluded in the input image.\nAdditionally, we demonstrate the potential of our approach on three datasets\nfor depth prediction and novel-view synthesis.\n", "date": "2023-01-18", "categories": []}
{"id": "2304.09371", "title": "3 Dimensional Dense Reconstruction: A Review of Algorithms and Dataset", "abstract": "  3D dense reconstruction refers to the process of obtaining the complete shape\nand texture features of 3D objects from 2D planar images. 3D reconstruction is\nan important and extensively studied problem, but it is far from being solved.\nThis work systematically introduces classical methods of 3D dense\nreconstruction based on geometric and optical models, as well as methods based\non deep learning. It also introduces datasets for deep learning and the\nperformance and advantages and disadvantages demonstrated by deep learning\nmethods on these datasets.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.08971", "title": "SurfelNeRF: Neural Surfel Radiance Fields for Online Photorealistic\n  Reconstruction of Indoor Scenes", "abstract": "  Online reconstructing and rendering of large-scale indoor scenes is a\nlong-standing challenge. SLAM-based methods can reconstruct 3D scene geometry\nprogressively in real time but can not render photorealistic results. While\nNeRF-based methods produce promising novel view synthesis results, their long\noffline optimization time and lack of geometric constraints pose challenges to\nefficiently handling online input. Inspired by the complementary advantages of\nclassical 3D reconstruction and NeRF, we thus investigate marrying explicit\ngeometric representation with NeRF rendering to achieve efficient online\nreconstruction and high-quality rendering. We introduce SurfelNeRF, a variant\nof neural radiance field which employs a flexible and scalable neural surfel\nrepresentation to store geometric attributes and extracted appearance features\nfrom input images. We further extend the conventional surfel-based fusion\nscheme to progressively integrate incoming input frames into the reconstructed\nglobal neural scene representation. In addition, we propose a highly-efficient\ndifferentiable rasterization scheme for rendering neural surfel radiance\nfields, which helps SurfelNeRF achieve $10\\times$ speedups in training and\ninference time, respectively. Experimental results show that our method\nachieves the state-of-the-art 23.82 PSNR and 29.58 PSNR on ScanNet in\nfeedforward inference and per-scene optimization settings, respectively.\n", "date": "2023-04-18", "categories": []}
{"id": "2212.07388", "title": "NoPe-NeRF: Optimising Neural Radiance Field with No Pose Prior", "abstract": "  Training a Neural Radiance Field (NeRF) without pre-computed camera poses is\nchallenging. Recent advances in this direction demonstrate the possibility of\njointly optimising a NeRF and camera poses in forward-facing scenes. However,\nthese methods still face difficulties during dramatic camera movement. We\ntackle this challenging problem by incorporating undistorted monocular depth\npriors. These priors are generated by correcting scale and shift parameters\nduring training, with which we are then able to constrain the relative poses\nbetween consecutive frames. This constraint is achieved using our proposed\nnovel loss functions. Experiments on real-world indoor and outdoor scenes show\nthat our method can handle challenging camera trajectories and outperforms\nexisting methods in terms of novel view rendering quality and pose estimation\naccuracy. Our project page is https://nope-nerf.active.vision.\n", "date": "2022-12-14", "categories": []}
{"id": "2209.15511", "title": "Sphere-Guided Training of Neural Implicit Surfaces", "abstract": "  In recent years, neural distance functions trained via volumetric ray\nmarching have been widely adopted for multi-view 3D reconstruction. These\nmethods, however, apply the ray marching procedure for the entire scene volume,\nleading to reduced sampling efficiency and, as a result, lower reconstruction\nquality in the areas of high-frequency details. In this work, we address this\nproblem via joint training of the implicit function and our new coarse\nsphere-based surface reconstruction. We use the coarse representation to\nefficiently exclude the empty volume of the scene from the volumetric ray\nmarching procedure without additional forward passes of the neural surface\nnetwork, which leads to an increased fidelity of the reconstructions compared\nto the base systems. We evaluate our approach by incorporating it into the\ntraining procedures of several implicit surface modeling methods and observe\nuniform improvements across both synthetic and real-world datasets. Our\ncodebase can be accessed via the project page:\nhttps://andreeadogaru.github.io/SphereGuided\n", "date": "2022-09-30", "categories": []}
{"id": "2211.11610", "title": "Tensor4D : Efficient Neural 4D Decomposition for High-fidelity Dynamic\n  Reconstruction and Rendering", "abstract": "  We present Tensor4D, an efficient yet effective approach to dynamic scene\nmodeling. The key of our solution is an efficient 4D tensor decomposition\nmethod so that the dynamic scene can be directly represented as a 4D\nspatio-temporal tensor. To tackle the accompanying memory issue, we decompose\nthe 4D tensor hierarchically by projecting it first into three time-aware\nvolumes and then nine compact feature planes. In this way, spatial information\nover time can be simultaneously captured in a compact and memory-efficient\nmanner. When applying Tensor4D for dynamic scene reconstruction and rendering,\nwe further factorize the 4D fields to different scales in the sense that\nstructural motions and dynamic detailed changes can be learned from coarse to\nfine. The effectiveness of our method is validated on both synthetic and\nreal-world scenes. Extensive experiments show that our method is able to\nachieve high-quality dynamic reconstruction and rendering from sparse-view\ncamera rigs or even a monocular camera. The code and dataset will be released\nat https://liuyebin.com/tensor4d/tensor4d.html.\n", "date": "2022-11-21", "categories": []}
{"id": "2304.06178", "title": "Dynamic Voxel Grid Optimization for High-Fidelity RGB-D Supervised\n  Surface Reconstruction", "abstract": "  Direct optimization of interpolated features on multi-resolution voxel grids\nhas emerged as a more efficient alternative to MLP-like modules. However, this\napproach is constrained by higher memory expenses and limited representation\ncapabilities. In this paper, we introduce a novel dynamic grid optimization\nmethod for high-fidelity 3D surface reconstruction that incorporates both RGB\nand depth observations. Rather than treating each voxel equally, we optimize\nthe process by dynamically modifying the grid and assigning more finer-scale\nvoxels to regions with higher complexity, allowing us to capture more intricate\ndetails. Furthermore, we develop a scheme to quantify the dynamic subdivision\nof voxel grid during optimization without requiring any priors. The proposed\napproach is able to generate high-quality 3D reconstructions with fine details\non both synthetic and real-world data, while maintaining computational\nefficiency, which is substantially faster than the baseline method NeuralRGBD.\n", "date": "2023-04-12", "categories": []}
{"id": "2210.13634", "title": "Vitruvio: 3D Building Meshes via Single Perspective Sketches", "abstract": "  Today's architectural engineering and construction (AEC) software require a\nlearning curve to generate a three-dimension building representation. This\nlimits the ability to quickly validate the volumetric implications of an\ninitial design idea communicated via a single sketch. Allowing designers to\ntranslate a single sketch to a 3D building will enable owners to instantly\nvisualize 3D project information without the cognitive load required. If\nprevious state-of-the-art (SOTA) data-driven methods for single view\nreconstruction (SVR) showed outstanding results in the reconstruction process\nfrom a single image or sketch, they lacked specific applications, analysis, and\nexperiments in the AEC. Therefore, this research addresses this gap,\nintroducing the first deep learning method focused only on buildings that aim\nto convert a single sketch to a 3D building mesh: Vitruvio. Vitruvio adapts\nOccupancy Network for SVR tasks on a specific building dataset (Manhattan 1K).\nThis adaptation brings two main improvements. First, it accelerates the\ninference process by more than 26% (from 0.5s to 0.37s). Second, it increases\nthe reconstruction accuracy (measured by the Chamfer Distance) by 18%. During\nthis adaptation in the AEC domain, we evaluate the effect of the building\norientation in the learning procedure since it constitutes an important design\nfactor. While aligning all the buildings to a canonical pose improved the\noverall quantitative metrics, it did not capture fine-grain details in more\ncomplex building shapes (as shown in our qualitative analysis). Finally,\nVitruvio outputs a 3D-printable building mesh with arbitrary topology and genus\nfrom a single perspective sketch, providing a step forward to allow owners and\ndesigners to communicate 3D information via a 2D, effective, intuitive, and\nuniversal communication medium: the sketch.\n", "date": "2022-10-24", "categories": []}
{"id": "2304.03999", "title": "Analysis of Sampling Strategies for Implicit 3D Reconstruction", "abstract": "  In the training process of the implicit 3D reconstruction network, the choice\nof spatial query points' sampling strategy affects the final performance of the\nmodel. Different works have differences in the selection of sampling\nstrategies, not only in the spatial distribution of query points but also in\nthe order of magnitude difference in the density of query points. For how to\nselect the sampling strategy of query points, current works are more akin to an\nenumerating operation to find the optimal solution, which seriously affects\nwork efficiency. In this work, we explored the relationship between sampling\nstrategy and network final performance through classification analysis and\nexperimental comparison from three aspects: the relationship between network\ntype and sampling strategy, the relationship between implicit function and\nsampling strategy, and the impact of sampling density on model performance. In\naddition, we also proposed two methods, linear sampling and distance mask, to\nimprove the sampling strategy of query points, making it more general and\nrobust.\n", "date": "2023-04-08", "categories": []}
{"id": "2304.05296", "title": "EvAC3D: From Event-based Apparent Contours to 3D Models via Continuous\n  Visual Hulls", "abstract": "  3D reconstruction from multiple views is a successful computer vision field\nwith multiple deployments in applications. State of the art is based on\ntraditional RGB frames that enable optimization of photo-consistency cross\nviews. In this paper, we study the problem of 3D reconstruction from\nevent-cameras, motivated by the advantages of event-based cameras in terms of\nlow power and latency as well as by the biological evidence that eyes in nature\ncapture the same data and still perceive well 3D shape. The foundation of our\nhypothesis that 3D reconstruction is feasible using events lies in the\ninformation contained in the occluding contours and in the continuous scene\nacquisition with events. We propose Apparent Contour Events (ACE), a novel\nevent-based representation that defines the geometry of the apparent contour of\nan object. We represent ACE by a spatially and temporally continuous implicit\nfunction defined in the event x-y-t space. Furthermore, we design a novel\ncontinuous Voxel Carving algorithm enabled by the high temporal resolution of\nthe Apparent Contour Events. To evaluate the performance of the method, we\ncollect MOEC-3D, a 3D event dataset of a set of common real-world objects. We\ndemonstrate the ability of EvAC3D to reconstruct high-fidelity mesh surfaces\nfrom real event sequences while allowing the refinement of the 3D\nreconstruction for each individual event.\n", "date": "2023-04-11", "categories": []}
{"id": "2207.11232", "title": "Neural Groundplans: Persistent Neural Scene Representations from a\n  Single Image", "abstract": "  We present a method to map 2D image observations of a scene to a persistent\n3D scene representation, enabling novel view synthesis and disentangled\nrepresentation of the movable and immovable components of the scene. Motivated\nby the bird's-eye-view (BEV) representation commonly used in vision and\nrobotics, we propose conditional neural groundplans, ground-aligned 2D feature\ngrids, as persistent and memory-efficient scene representations. Our method is\ntrained self-supervised from unlabeled multi-view observations using\ndifferentiable rendering, and learns to complete geometry and appearance of\noccluded regions. In addition, we show that we can leverage multi-view videos\nat training time to learn to separately reconstruct static and movable\ncomponents of the scene from a single image at test time. The ability to\nseparately reconstruct movable objects enables a variety of downstream tasks\nusing simple heuristics, such as extraction of object-centric 3D\nrepresentations, novel view synthesis, instance-level segmentation, 3D bounding\nbox prediction, and scene editing. This highlights the value of neural\ngroundplans as a backbone for efficient 3D scene understanding models.\n", "date": "2022-07-22", "categories": []}
{"id": "2303.05703", "title": "MovingParts: Motion-based 3D Part Discovery in Dynamic Radiance Field", "abstract": "  We present MovingParts, a NeRF-based method for dynamic scene reconstruction\nand part discovery. We consider motion as an important cue for identifying\nparts, that all particles on the same part share the common motion pattern.\nFrom the perspective of fluid simulation, existing deformation-based methods\nfor dynamic NeRF can be seen as parameterizing the scene motion under the\nEulerian view, i.e., focusing on specific locations in space through which the\nfluid flows as time passes. However, it is intractable to extract the motion of\nconstituting objects or parts using the Eulerian view representation. In this\nwork, we introduce the dual Lagrangian view and enforce representations under\nthe Eulerian/Lagrangian views to be cycle-consistent. Under the Lagrangian\nview, we parameterize the scene motion by tracking the trajectory of particles\non objects. The Lagrangian view makes it convenient to discover parts by\nfactorizing the scene motion as a composition of part-level rigid motions.\nExperimentally, our method can achieve fast and high-quality dynamic scene\nreconstruction from even a single moving camera, and the induced part-based\nrepresentation allows direct applications of part tracking, animation, 3D scene\nediting, etc.\n", "date": "2023-03-10", "categories": []}
{"id": "2304.02704", "title": "Real-Time Dense 3D Mapping of Underwater Environments", "abstract": "  This paper addresses real-time dense 3D reconstruction for a\nresource-constrained Autonomous Underwater Vehicle (AUV). Underwater\nvision-guided operations are among the most challenging as they combine 3D\nmotion in the presence of external forces, limited visibility, and absence of\nglobal positioning. Obstacle avoidance and effective path planning require\nonline dense reconstructions of the environment. Autonomous operation is\ncentral to environmental monitoring, marine archaeology, resource utilization,\nand underwater cave exploration. To address this problem, we propose to use\nSVIn2, a robust VIO method, together with a real-time 3D reconstruction\npipeline. We provide extensive evaluation on four challenging underwater\ndatasets. Our pipeline produces comparable reconstruction with that of COLMAP,\nthe state-of-the-art offline 3D reconstruction method, at high frame rates on a\nsingle CPU.\n", "date": "2023-04-05", "categories": []}
{"id": "2212.08067", "title": "VolRecon: Volume Rendering of Signed Ray Distance Functions for\n  Generalizable Multi-View Reconstruction", "abstract": "  The success of the Neural Radiance Fields (NeRF) in novel view synthesis has\ninspired researchers to propose neural implicit scene reconstruction. However,\nmost existing neural implicit reconstruction methods optimize per-scene\nparameters and therefore lack generalizability to new scenes. We introduce\nVolRecon, a novel generalizable implicit reconstruction method with Signed Ray\nDistance Function (SRDF). To reconstruct the scene with fine details and little\nnoise, VolRecon combines projection features aggregated from multi-view\nfeatures, and volume features interpolated from a coarse global feature volume.\nUsing a ray transformer, we compute SRDF values of sampled points on a ray and\nthen render color and depth. On DTU dataset, VolRecon outperforms SparseNeuS by\nabout 30% in sparse view reconstruction and achieves comparable accuracy as\nMVSNet in full view reconstruction. Furthermore, our approach exhibits good\ngeneralization performance on the large-scale ETH3D benchmark.\n", "date": "2022-12-15", "categories": []}
{"id": "2304.00330", "title": "Multi-view reconstruction of bullet time effect based on improved NSFF\n  model", "abstract": "  Bullet time is a type of visual effect commonly used in film, television and\ngames that makes time seem to slow down or stop while still preserving dynamic\ndetails in the scene. It usually requires multiple sets of cameras to move\nslowly with the subject and is synthesized using post-production techniques,\nwhich is costly and one-time. The dynamic scene perspective reconstruction\ntechnology based on neural rendering field can be used to solve this\nrequirement, but most of the current methods are poor in reconstruction\naccuracy due to the blurred input image and overfitting of dynamic and static\nregions. Based on the NSFF algorithm, this paper reconstructed the common time\nspecial effects scenes in movies and television from a new perspective. To\nimprove the accuracy of the reconstructed images, fuzzy kernel was added to the\nnetwork for reconstruction and analysis of the fuzzy process, and the clear\nperspective after analysis was input into the NSFF to improve the accuracy. By\nusing the optical flow prediction information to suppress the dynamic network\ntimely, the network is forced to improve the reconstruction effect of dynamic\nand static networks independently, and the ability to understand and\nreconstruct dynamic and static scenes is improved. To solve the overfitting\nproblem of dynamic and static scenes, a new dynamic and static cross entropy\nloss is designed. Experimental results show that compared with original NSFF\nand other new perspective reconstruction algorithms of dynamic scenes, the\nimproved NSFF-RFCT improves the reconstruction accuracy and enhances the\nunderstanding ability of dynamic and static scenes.\n", "date": "2023-04-01", "categories": []}
{"id": "2303.17088", "title": "Depth-NeuS: Neural Implicit Surfaces Learning for Multi-view\n  Reconstruction Based on Depth Information Optimization", "abstract": "  Recently, methods for neural surface representation and rendering, for\nexample NeuS, have shown that learning neural implicit surfaces through volume\nrendering is becoming increasingly popular and making good progress. However,\nthese methods still face some challenges. Existing methods lack a direct\nrepresentation of depth information, which makes object reconstruction\nunrestricted by geometric features, resulting in poor reconstruction of objects\nwith texture and color features. This is because existing methods only use\nsurface normals to represent implicit surfaces without using depth information.\nTherefore, these methods cannot model the detailed surface features of objects\nwell. To address this problem, we propose a neural implicit surface learning\nmethod called Depth-NeuS based on depth information optimization for multi-view\nreconstruction. In this paper, we introduce depth loss to explicitly constrain\nSDF regression and introduce geometric consistency loss to optimize for\nlow-texture areas. Specific experiments show that Depth-NeuS outperforms\nexisting technologies in multiple scenarios and achieves high-quality surface\nreconstruction in multiple scenarios.\n", "date": "2023-03-30", "categories": []}
{"id": "2303.07634", "title": "I$^2$-SDF: Intrinsic Indoor Scene Reconstruction and Editing via\n  Raytracing in Neural SDFs", "abstract": "  In this work, we present I$^2$-SDF, a new method for intrinsic indoor scene\nreconstruction and editing using differentiable Monte Carlo raytracing on\nneural signed distance fields (SDFs). Our holistic neural SDF-based framework\njointly recovers the underlying shapes, incident radiance and materials from\nmulti-view images. We introduce a novel bubble loss for fine-grained small\nobjects and error-guided adaptive sampling scheme to largely improve the\nreconstruction quality on large-scale indoor scenes. Further, we propose to\ndecompose the neural radiance field into spatially-varying material of the\nscene as a neural field through surface-based, differentiable Monte Carlo\nraytracing and emitter semantic segmentations, which enables physically based\nand photorealistic scene relighting and editing applications. Through a number\nof qualitative and quantitative experiments, we demonstrate the superior\nquality of our method on indoor scene reconstruction, novel view synthesis, and\nscene editing compared to state-of-the-art baselines.\n", "date": "2023-03-14", "categories": []}
{"id": "2303.16293", "title": "SnakeVoxFormer: Transformer-based Single Image\\\\Voxel Reconstruction\n  with Run Length Encoding", "abstract": "  Deep learning-based 3D object reconstruction has achieved unprecedented\nresults. Among those, the transformer deep neural model showed outstanding\nperformance in many applications of computer vision. We introduce\nSnakeVoxFormer, a novel, 3D object reconstruction in voxel space from a single\nimage using the transformer. The input to SnakeVoxFormer is a 2D image, and the\nresult is a 3D voxel model. The key novelty of our approach is in using the\nrun-length encoding that traverses (like a snake) the voxel space and encodes\nwide spatial differences into a 1D structure that is suitable for transformer\nencoding. We then use dictionary encoding to convert the discovered RLE blocks\ninto tokens that are used for the transformer. The 1D representation is a\nlossless 3D shape data compression method that converts to 1D data that use\nonly about 1% of the original data size. We show how different voxel traversing\nstrategies affect the effect of encoding and reconstruction. We compare our\nmethod with the state-of-the-art for 3D voxel reconstruction from images and\nour method improves the state-of-the-art methods by at least 2.8% and up to\n19.8%.\n", "date": "2023-03-28", "categories": []}
{"id": "2303.16447", "title": "Multi-View Azimuth Stereo via Tangent Space Consistency", "abstract": "  We present a method for 3D reconstruction only using calibrated multi-view\nsurface azimuth maps. Our method, multi-view azimuth stereo, is effective for\ntextureless or specular surfaces, which are difficult for conventional\nmulti-view stereo methods. We introduce the concept of tangent space\nconsistency: Multi-view azimuth observations of a surface point should be\nlifted to the same tangent space. Leveraging this consistency, we recover the\nshape by optimizing a neural implicit surface representation. Our method\nharnesses the robust azimuth estimation capabilities of photometric stereo\nmethods or polarization imaging while bypassing potentially complex zenith\nangle estimation. Experiments using azimuth maps from various sources validate\nthe accurate shape recovery with our method, even without zenith angles.\n", "date": "2023-03-29", "categories": []}
{"id": "2211.12562", "title": "PermutoSDF: Fast Multi-View Reconstruction with Implicit Surfaces using\n  Permutohedral Lattices", "abstract": "  Neural radiance-density field methods have become increasingly popular for\nthe task of novel-view rendering. Their recent extension to hash-based\npositional encoding ensures fast training and inference with visually pleasing\nresults. However, density-based methods struggle with recovering accurate\nsurface geometry. Hybrid methods alleviate this issue by optimizing the density\nbased on an underlying SDF. However, current SDF methods are overly smooth and\nmiss fine geometric details. In this work, we combine the strengths of these\ntwo lines of work in a novel hash-based implicit surface representation. We\npropose improvements to the two areas by replacing the voxel hash encoding with\na permutohedral lattice which optimizes faster, especially for higher\ndimensions. We additionally propose a regularization scheme which is crucial\nfor recovering high-frequency geometric detail. We evaluate our method on\nmultiple datasets and show that we can recover geometric detail at the level of\npores and wrinkles while using only RGB images for supervision. Furthermore,\nusing sphere tracing we can render novel views at 30 fps on an RTX 3090. Code\nis publicly available at: https://radualexandru.github.io/permuto_sdf\n", "date": "2022-11-22", "categories": []}
{"id": "2210.15664", "title": "State of the Art in Dense Monocular Non-Rigid 3D Reconstruction", "abstract": "  3D reconstruction of deformable (or non-rigid) scenes from a set of monocular\n2D image observations is a long-standing and actively researched area of\ncomputer vision and graphics. It is an ill-posed inverse problem, since --\nwithout additional prior assumptions -- it permits infinitely many solutions\nleading to accurate projection to the input 2D images. Non-rigid reconstruction\nis a foundational building block for downstream applications like robotics,\nAR/VR, or visual content creation. The key advantage of using monocular cameras\nis their omnipresence and availability to the end users as well as their ease\nof use compared to more sophisticated camera set-ups such as stereo or\nmulti-view systems. This survey focuses on state-of-the-art methods for dense\nnon-rigid 3D reconstruction of various deformable objects and composite scenes\nfrom monocular videos or sets of monocular views. It reviews the fundamentals\nof 3D reconstruction and deformation modeling from 2D image observations. We\nthen start from general methods -- that handle arbitrary scenes and make only a\nfew prior assumptions -- and proceed towards techniques making stronger\nassumptions about the observed objects and types of deformations (e.g. human\nfaces, bodies, hands, and animals). A significant part of this STAR is also\ndevoted to classification and a high-level comparison of the methods, as well\nas an overview of the datasets for training and evaluation of the discussed\ntechniques. We conclude by discussing open challenges in the field and the\nsocial aspects associated with the usage of the reviewed methods.\n", "date": "2022-10-27", "categories": []}
{"id": "2211.12285", "title": "Exact-NeRF: An Exploration of a Precise Volumetric Parameterization for\n  Neural Radiance Fields", "abstract": "  Neural Radiance Fields (NeRF) have attracted significant attention due to\ntheir ability to synthesize novel scene views with great accuracy. However,\ninherent to their underlying formulation, the sampling of points along a ray\nwith zero width may result in ambiguous representations that lead to further\nrendering artifacts such as aliasing in the final scene. To address this issue,\nthe recent variant mip-NeRF proposes an Integrated Positional Encoding (IPE)\nbased on a conical view frustum. Although this is expressed with an integral\nformulation, mip-NeRF instead approximates this integral as the expected value\nof a multivariate Gaussian distribution. This approximation is reliable for\nshort frustums but degrades with highly elongated regions, which arises when\ndealing with distant scene objects under a larger depth of field. In this\npaper, we explore the use of an exact approach for calculating the IPE by using\na pyramid-based integral formulation instead of an approximated conical-based\none. We denote this formulation as Exact-NeRF and contribute the first approach\nto offer a precise analytical solution to the IPE within the NeRF domain. Our\nexploratory work illustrates that such an exact formulation Exact-NeRF matches\nthe accuracy of mip-NeRF and furthermore provides a natural extension to more\nchallenging scenarios without further modification, such as in the case of\nunbounded scenes. Our contribution aims to both address the hitherto unexplored\nissues of frustum approximation in earlier NeRF work and additionally provide\ninsight into the potential future consideration of analytical solutions in\nfuture NeRF extensions.\n", "date": "2022-11-22", "categories": []}
{"id": "2301.10241", "title": "K-Planes: Explicit Radiance Fields in Space, Time, and Appearance", "abstract": "  We introduce k-planes, a white-box model for radiance fields in arbitrary\ndimensions. Our model uses d choose 2 planes to represent a d-dimensional\nscene, providing a seamless way to go from static (d=3) to dynamic (d=4)\nscenes. This planar factorization makes adding dimension-specific priors easy,\ne.g. temporal smoothness and multi-resolution spatial structure, and induces a\nnatural decomposition of static and dynamic components of a scene. We use a\nlinear feature decoder with a learned color basis that yields similar\nperformance as a nonlinear black-box MLP decoder. Across a range of synthetic\nand real, static and dynamic, fixed and varying appearance scenes, k-planes\nyields competitive and often state-of-the-art reconstruction fidelity with low\nmemory usage, achieving 1000x compression over a full 4D grid, and fast\noptimization with a pure PyTorch implementation. For video results and code,\nplease see https://sarafridov.github.io/K-Planes.\n", "date": "2023-01-24", "categories": []}
{"id": "2302.12251", "title": "VoxFormer: Sparse Voxel Transformer for Camera-based 3D Semantic Scene\n  Completion", "abstract": "  Humans can easily imagine the complete 3D geometry of occluded objects and\nscenes. This appealing ability is vital for recognition and understanding. To\nenable such capability in AI systems, we propose VoxFormer, a Transformer-based\nsemantic scene completion framework that can output complete 3D volumetric\nsemantics from only 2D images. Our framework adopts a two-stage design where we\nstart from a sparse set of visible and occupied voxel queries from depth\nestimation, followed by a densification stage that generates dense 3D voxels\nfrom the sparse ones. A key idea of this design is that the visual features on\n2D images correspond only to the visible scene structures rather than the\noccluded or empty spaces. Therefore, starting with the featurization and\nprediction of the visible structures is more reliable. Once we obtain the set\nof sparse queries, we apply a masked autoencoder design to propagate the\ninformation to all the voxels by self-attention. Experiments on SemanticKITTI\nshow that VoxFormer outperforms the state of the art with a relative\nimprovement of 20.0% in geometry and 18.1% in semantics and reduces GPU memory\nduring training to less than 16GB. Our code is available on\nhttps://github.com/NVlabs/VoxFormer.\n", "date": "2023-02-23", "categories": []}
{"id": "2302.13543", "title": "BLiRF: Bandlimited Radiance Fields for Dynamic Scene Modeling", "abstract": "  Reasoning the 3D structure of a non-rigid dynamic scene from a single moving\ncamera is an under-constrained problem. Inspired by the remarkable progress of\nneural radiance fields (NeRFs) in photo-realistic novel view synthesis of\nstatic scenes, extensions have been proposed for dynamic settings. These\nmethods heavily rely on neural priors in order to regularize the problem. In\nthis work, we take a step back and reinvestigate how current implementations\nmay entail deleterious effects, including limited expressiveness, entanglement\nof light and density fields, and sub-optimal motion localization. As a remedy,\nwe advocate for a bridge between classic non-rigid-structure-from-motion\n(\\nrsfm) and NeRF, enabling the well-studied priors of the former to constrain\nthe latter. To this end, we propose a framework that factorizes time and space\nby formulating a scene as a composition of bandlimited, high-dimensional\nsignals. We demonstrate compelling results across complex dynamic scenes that\ninvolve changes in lighting, texture and long-range dynamics.\n", "date": "2023-02-27", "categories": []}
{"id": "2303.14243", "title": "DyLiN: Making Light Field Networks Dynamic", "abstract": "  Light Field Networks, the re-formulations of radiance fields to oriented\nrays, are magnitudes faster than their coordinate network counterparts, and\nprovide higher fidelity with respect to representing 3D structures from 2D\nobservations. They would be well suited for generic scene representation and\nmanipulation, but suffer from one problem: they are limited to holistic and\nstatic scenes. In this paper, we propose the Dynamic Light Field Network\n(DyLiN) method that can handle non-rigid deformations, including topological\nchanges. We learn a deformation field from input rays to canonical rays, and\nlift them into a higher dimensional space to handle discontinuities. We further\nintroduce CoDyLiN, which augments DyLiN with controllable attribute inputs. We\ntrain both models via knowledge distillation from pretrained dynamic radiance\nfields. We evaluated DyLiN using both synthetic and real world datasets that\ninclude various non-rigid deformations. DyLiN qualitatively outperformed and\nquantitatively matched state-of-the-art methods in terms of visual fidelity,\nwhile being 25 - 71x computationally faster. We also tested CoDyLiN on\nattribute annotated data and it surpassed its teacher model. Project page:\nhttps://dylin2023.github.io .\n", "date": "2023-03-24", "categories": []}
{"id": "2303.14478", "title": "DBARF: Deep Bundle-Adjusting Generalizable Neural Radiance Fields", "abstract": "  Recent works such as BARF and GARF can bundle adjust camera poses with neural\nradiance fields (NeRF) which is based on coordinate-MLPs. Despite the\nimpressive results, these methods cannot be applied to Generalizable NeRFs\n(GeNeRFs) which require image feature extractions that are often based on more\ncomplicated 3D CNN or transformer architectures. In this work, we first analyze\nthe difficulties of jointly optimizing camera poses with GeNeRFs, and then\nfurther propose our DBARF to tackle these issues. Our DBARF which bundle\nadjusts camera poses by taking a cost feature map as an implicit cost function\ncan be jointly trained with GeNeRFs in a self-supervised manner. Unlike BARF\nand its follow-up works, which can only be applied to per-scene optimized NeRFs\nand need accurate initial camera poses with the exception of forward-facing\nscenes, our method can generalize across scenes and does not require any good\ninitialization. Experiments show the effectiveness and generalization ability\nof our DBARF when evaluated on real-world datasets. Our code is available at\n\\url{https://aibluefisher.github.io/dbarf}.\n", "date": "2023-03-25", "categories": []}
{"id": "2303.14505", "title": "Unsupervised Inference of Signed Distance Functions from Single Sparse\n  Point Clouds without Learning Priors", "abstract": "  It is vital to infer signed distance functions (SDFs) from 3D point clouds.\nThe latest methods rely on generalizing the priors learned from large scale\nsupervision. However, the learned priors do not generalize well to various\ngeometric variations that are unseen during training, especially for extremely\nsparse point clouds. To resolve this issue, we present a neural network to\ndirectly infer SDFs from single sparse point clouds without using signed\ndistance supervision, learned priors or even normals. Our insight here is to\nlearn surface parameterization and SDFs inference in an end-to-end manner. To\nmake up the sparsity, we leverage parameterized surfaces as a coarse surface\nsampler to provide many coarse surface estimations in training iterations,\naccording to which we mine supervision and our thin plate splines (TPS) based\nnetwork infers SDFs as smooth functions in a statistical way. Our method\nsignificantly improves the generalization ability and accuracy in unseen point\nclouds. Our experimental results show our advantages over the state-of-the-art\nmethods in surface reconstruction for sparse point clouds under synthetic\ndatasets and real scans.The code is available at\n\\url{https://github.com/chenchao15/NeuralTPS}.\n", "date": "2023-03-25", "categories": []}
{"id": "2303.14569", "title": "VisCo Grids: Surface Reconstruction with Viscosity and Coarea Grids", "abstract": "  Surface reconstruction has been seeing a lot of progress lately by utilizing\nImplicit Neural Representations (INRs). Despite their success, INRs often\nintroduce hard to control inductive bias (i.e., the solution surface can\nexhibit unexplainable behaviours), have costly inference, and are slow to\ntrain. The goal of this work is to show that replacing neural networks with\nsimple grid functions, along with two novel geometric priors achieve comparable\nresults to INRs, with instant inference, and improved training times. To that\nend we introduce VisCo Grids: a grid-based surface reconstruction method\nincorporating Viscosity and Coarea priors. Intuitively, the Viscosity prior\nreplaces the smoothness inductive bias of INRs, while the Coarea favors a\nminimal area solution. Experimenting with VisCo Grids on a standard\nreconstruction baseline provided comparable results to the best performing INRs\non this dataset.\n", "date": "2023-03-25", "categories": []}
{"id": "2303.14587", "title": "PAniC-3D: Stylized Single-view 3D Reconstruction from Portraits of Anime\n  Characters", "abstract": "  We propose PAniC-3D, a system to reconstruct stylized 3D character heads\ndirectly from illustrated (p)ortraits of (ani)me (c)haracters. Our anime-style\ndomain poses unique challenges to single-view reconstruction; compared to\nnatural images of human heads, character portrait illustrations have hair and\naccessories with more complex and diverse geometry, and are shaded with\nnon-photorealistic contour lines. In addition, there is a lack of both 3D model\nand portrait illustration data suitable to train and evaluate this ambiguous\nstylized reconstruction task. Facing these challenges, our proposed PAniC-3D\narchitecture crosses the illustration-to-3D domain gap with a line-filling\nmodel, and represents sophisticated geometries with a volumetric radiance\nfield. We train our system with two large new datasets (11.2k Vroid 3D models,\n1k Vtuber portrait illustrations), and evaluate on a novel AnimeRecon benchmark\nof illustration-to-3D pairs. PAniC-3D significantly outperforms baseline\nmethods, and provides data to establish the task of stylized reconstruction\nfrom portrait illustrations.\n", "date": "2023-03-25", "categories": []}
{"id": "2206.11896", "title": "EventNeRF: Neural Radiance Fields from a Single Colour Event Camera", "abstract": "  Asynchronously operating event cameras find many applications due to their\nhigh dynamic range, vanishingly low motion blur, low latency and low data\nbandwidth. The field saw remarkable progress during the last few years, and\nexisting event-based 3D reconstruction approaches recover sparse point clouds\nof the scene. However, such sparsity is a limiting factor in many cases,\nespecially in computer vision and graphics, that has not been addressed\nsatisfactorily so far. Accordingly, this paper proposes the first approach for\n3D-consistent, dense and photorealistic novel view synthesis using just a\nsingle colour event stream as input. At its core is a neural radiance field\ntrained entirely in a self-supervised manner from events while preserving the\noriginal resolution of the colour event channels. Next, our ray sampling\nstrategy is tailored to events and allows for data-efficient training. At test,\nour method produces results in the RGB space at unprecedented quality. We\nevaluate our method qualitatively and numerically on several challenging\nsynthetic and real scenes and show that it produces significantly denser and\nmore visually appealing renderings than the existing methods. We also\ndemonstrate robustness in challenging scenarios with fast motion and under low\nlighting conditions. We release the newly recorded dataset and our source code\nto facilitate the research field, see https://4dqv.mpi-inf.mpg.de/EventNeRF.\n", "date": "2022-06-23", "categories": []}
{"id": "2211.04041", "title": "ParticleNeRF: A Particle-Based Encoding for Online Neural Radiance\n  Fields", "abstract": "  While existing Neural Radiance Fields (NeRFs) for dynamic scenes are offline\nmethods with an emphasis on visual fidelity, our paper addresses the online use\ncase that prioritises real-time adaptability. We present ParticleNeRF, a new\napproach that dynamically adapts to changes in the scene geometry by learning\nan up-to-date representation online, every 200ms. ParticleNeRF achieves this\nusing a novel particle-based parametric encoding. We couple features to\nparticles in space and backpropagate the photometric reconstruction loss into\nthe particles' position gradients, which are then interpreted as velocity\nvectors. Governed by a lightweight physics system to handle collisions, this\nlets the features move freely with the changing scene geometry. We demonstrate\nParticleNeRF on various dynamic scenes containing translating, rotating,\narticulated, and deformable objects. ParticleNeRF is the first online dynamic\nNeRF and achieves fast adaptability with better visual fidelity than\nbrute-force online InstantNGP and other baseline approaches on dynamic scenes\nwith online constraints. Videos of our system can be found at our project\nwebsite https://sites.google.com/view/particlenerf.\n", "date": "2022-11-08", "categories": []}
{"id": "2303.12408", "title": "Balanced Spherical Grid for Egocentric View Synthesis", "abstract": "  We present EgoNeRF, a practical solution to reconstruct large-scale\nreal-world environments for VR assets. Given a few seconds of casually captured\n360 video, EgoNeRF can efficiently build neural radiance fields which enable\nhigh-quality rendering from novel viewpoints. Motivated by the recent\nacceleration of NeRF using feature grids, we adopt spherical coordinate instead\nof conventional Cartesian coordinate. Cartesian feature grid is inefficient to\nrepresent large-scale unbounded scenes because it has a spatially uniform\nresolution, regardless of distance from viewers. The spherical parameterization\nbetter aligns with the rays of egocentric images, and yet enables factorization\nfor performance enhancement. However, the na\\\"ive spherical grid suffers from\nirregularities at two poles, and also cannot represent unbounded scenes. To\navoid singularities near poles, we combine two balanced grids, which results in\na quasi-uniform angular grid. We also partition the radial grid exponentially\nand place an environment map at infinity to represent unbounded scenes.\nFurthermore, with our resampling technique for grid-based methods, we can\nincrease the number of valid samples to train NeRF volume. We extensively\nevaluate our method in our newly introduced synthetic and real-world egocentric\n360 video datasets, and it consistently achieves state-of-the-art performance.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.13582", "title": "SCADE: NeRFs from Space Carving with Ambiguity-Aware Depth Estimates", "abstract": "  Neural radiance fields (NeRFs) have enabled high fidelity 3D reconstruction\nfrom multiple 2D input views. However, a well-known drawback of NeRFs is the\nless-than-ideal performance under a small number of views, due to insufficient\nconstraints enforced by volumetric rendering. To address this issue, we\nintroduce SCADE, a novel technique that improves NeRF reconstruction quality on\nsparse, unconstrained input views for in-the-wild indoor scenes. To constrain\nNeRF reconstruction, we leverage geometric priors in the form of per-view depth\nestimates produced with state-of-the-art monocular depth estimation models,\nwhich can generalize across scenes. A key challenge is that monocular depth\nestimation is an ill-posed problem, with inherent ambiguities. To handle this\nissue, we propose a new method that learns to predict, for each view, a\ncontinuous, multimodal distribution of depth estimates using conditional\nImplicit Maximum Likelihood Estimation (cIMLE). In order to disambiguate\nexploiting multiple views, we introduce an original space carving loss that\nguides the NeRF representation to fuse multiple hypothesized depth maps from\neach view and distill from them a common geometry that is consistent with all\nviews. Experiments show that our approach enables higher fidelity novel view\nsynthesis from sparse views. Our project page can be found at\nhttps://scade-spacecarving-nerfs.github.io .\n", "date": "2023-03-23", "categories": []}
{"id": "2303.13805", "title": "Seeing Through the Glass: Neural 3D Reconstruction of Object Inside a\n  Transparent Container", "abstract": "  In this paper, we define a new problem of recovering the 3D geometry of an\nobject confined in a transparent enclosure. We also propose a novel method for\nsolving this challenging problem. Transparent enclosures pose challenges of\nmultiple light reflections and refractions at the interface between different\npropagation media e.g. air or glass. These multiple reflections and refractions\ncause serious image distortions which invalidate the single viewpoint\nassumption. Hence the 3D geometry of such objects cannot be reliably\nreconstructed using existing methods, such as traditional structure from motion\nor modern neural reconstruction methods. We solve this problem by explicitly\nmodeling the scene as two distinct sub-spaces, inside and outside the\ntransparent enclosure. We use an existing neural reconstruction method (NeuS)\nthat implicitly represents the geometry and appearance of the inner subspace.\nIn order to account for complex light interactions, we develop a hybrid\nrendering strategy that combines volume rendering with ray tracing. We then\nrecover the underlying geometry and appearance of the model by minimizing the\ndifference between the real and hybrid rendered images. We evaluate our method\non both synthetic and real data. Experiment results show that our method\noutperforms the state-of-the-art (SOTA) methods. Codes and data will be\navailable at https://github.com/hirotong/ReNeuS\n", "date": "2023-03-24", "categories": []}
{"id": "2303.12012", "title": "NeAT: Learning Neural Implicit Surfaces with Arbitrary Topologies from\n  Multi-view Images", "abstract": "  Recent progress in neural implicit functions has set new state-of-the-art in\nreconstructing high-fidelity 3D shapes from a collection of images. However,\nthese approaches are limited to closed surfaces as they require the surface to\nbe represented by a signed distance field. In this paper, we propose NeAT, a\nnew neural rendering framework that can learn implicit surfaces with arbitrary\ntopologies from multi-view images. In particular, NeAT represents the 3D\nsurface as a level set of a signed distance function (SDF) with a validity\nbranch for estimating the surface existence probability at the query positions.\nWe also develop a novel neural volume rendering method, which uses SDF and\nvalidity to calculate the volume opacity and avoids rendering points with low\nvalidity. NeAT supports easy field-to-mesh conversion using the classic\nMarching Cubes algorithm. Extensive experiments on DTU, MGN, and Deep Fashion\n3D datasets indicate that our approach is able to faithfully reconstruct both\nwatertight and non-watertight surfaces. In particular, NeAT significantly\noutperforms the state-of-the-art methods in the task of open surface\nreconstruction both quantitatively and qualitatively.\n", "date": "2023-03-21", "categories": []}
{"id": "2211.11674", "title": "Shape, Pose, and Appearance from a Single Image via Bootstrapped\n  Radiance Field Inversion", "abstract": "  Neural Radiance Fields (NeRF) coupled with GANs represent a promising\ndirection in the area of 3D reconstruction from a single view, owing to their\nability to efficiently model arbitrary topologies. Recent work in this area,\nhowever, has mostly focused on synthetic datasets where exact ground-truth\nposes are known, and has overlooked pose estimation, which is important for\ncertain downstream applications such as augmented reality (AR) and robotics. We\nintroduce a principled end-to-end reconstruction framework for natural images,\nwhere accurate ground-truth poses are not available. Our approach recovers an\nSDF-parameterized 3D shape, pose, and appearance from a single image of an\nobject, without exploiting multiple views during training. More specifically,\nwe leverage an unconditional 3D-aware generator, to which we apply a hybrid\ninversion scheme where a model produces a first guess of the solution which is\nthen refined via optimization. Our framework can de-render an image in as few\nas 10 steps, enabling its use in practical scenarios. We demonstrate\nstate-of-the-art results on a variety of real and synthetic benchmarks.\n", "date": "2022-11-21", "categories": []}
{"id": "2212.05993", "title": "RGBD2: Generative Scene Synthesis via Incremental View Inpainting using\n  RGBD Diffusion Models", "abstract": "  We address the challenge of recovering an underlying scene geometry and\ncolors from a sparse set of RGBD view observations. In this work, we present a\nnew solution termed RGBD$^2$ that sequentially generates novel RGBD views along\na camera trajectory, and the scene geometry is simply the fusion result of\nthese views. More specifically, we maintain an intermediate surface mesh used\nfor rendering new RGBD views, which subsequently becomes complete by an\ninpainting network; each rendered RGBD view is later back-projected as a\npartial surface and is supplemented into the intermediate mesh. The use of\nintermediate mesh and camera projection helps solve the tough problem of\nmulti-view inconsistency. We practically implement the RGBD inpainting network\nas a versatile RGBD diffusion model, which is previously used for 2D generative\nmodeling; we make a modification to its reverse diffusion process to enable our\nuse. We evaluate our approach on the task of 3D scene synthesis from sparse\nRGBD inputs; extensive experiments on the ScanNet dataset demonstrate the\nsuperiority of our approach over existing ones. Project page:\nhttps://jblei.site/proj/rgbd-diffusion.\n", "date": "2022-12-12", "categories": []}
{"id": "2303.09758", "title": "Hierarchical Prior Mining for Non-local Multi-View Stereo", "abstract": "  As a fundamental problem in computer vision, multi-view stereo (MVS) aims at\nrecovering the 3D geometry of a target from a set of 2D images. Recent advances\nin MVS have shown that it is important to perceive non-local structured\ninformation for recovering geometry in low-textured areas. In this work, we\npropose a Hierarchical Prior Mining for Non-local Multi-View Stereo (HPM-MVS).\nThe key characteristics are the following techniques that exploit non-local\ninformation to assist MVS: 1) A Non-local Extensible Sampling Pattern (NESP),\nwhich is able to adaptively change the size of sampled areas without becoming\nsnared in locally optimal solutions. 2) A new approach to leverage non-local\nreliable points and construct a planar prior model based on K-Nearest Neighbor\n(KNN), to obtain potential hypotheses for the regions where prior construction\nis challenging. 3) A Hierarchical Prior Mining (HPM) framework, which is used\nto mine extensive non-local prior information at different scales to assist 3D\nmodel recovery, this strategy can achieve a considerable balance between the\nreconstruction of details and low-textured areas. Experimental results on the\nETH3D and Tanks \\& Temples have verified the superior performance and strong\ngeneralization capability of our method. Our code will be released.\n", "date": "2023-03-17", "categories": []}
{"id": "2303.09977", "title": "Semantic Scene Completion with Cleaner Self", "abstract": "  Semantic Scene Completion (SSC) transforms an image of single-view depth\nand/or RGB 2D pixels into 3D voxels, each of whose semantic labels are\npredicted. SSC is a well-known ill-posed problem as the prediction model has to\n\"imagine\" what is behind the visible surface, which is usually represented by\nTruncated Signed Distance Function (TSDF). Due to the sensory imperfection of\nthe depth camera, most existing methods based on the noisy TSDF estimated from\ndepth values suffer from 1) incomplete volumetric predictions and 2) confused\nsemantic labels. To this end, we use the ground-truth 3D voxels to generate a\nperfect visible surface, called TSDF-CAD, and then train a \"cleaner\" SSC model.\nAs the model is noise-free, it is expected to focus more on the \"imagination\"\nof unseen voxels. Then, we propose to distill the intermediate \"cleaner\"\nknowledge into another model with noisy TSDF input. In particular, we use the\n3D occupancy feature and the semantic relations of the \"cleaner self\" to\nsupervise the counterparts of the \"noisy self\" to respectively address the\nabove two incorrect predictions. Experimental results validate that our method\nimproves the noisy counterparts with 3.1% IoU and 2.2% mIoU for measuring scene\ncompletion and SSC, and also achieves new state-of-the-art accuracy on the\npopular NYU dataset.\n", "date": "2023-03-17", "categories": []}
{"id": "2211.12254", "title": "SPIn-NeRF: Multiview Segmentation and Perceptual Inpainting with Neural\n  Radiance Fields", "abstract": "  Neural Radiance Fields (NeRFs) have emerged as a popular approach for novel\nview synthesis. While NeRFs are quickly being adapted for a wider set of\napplications, intuitively editing NeRF scenes is still an open challenge. One\nimportant editing task is the removal of unwanted objects from a 3D scene, such\nthat the replaced region is visually plausible and consistent with its context.\nWe refer to this task as 3D inpainting. In 3D, solutions must be both\nconsistent across multiple views and geometrically valid. In this paper, we\npropose a novel 3D inpainting method that addresses these challenges. Given a\nsmall set of posed images and sparse annotations in a single input image, our\nframework first rapidly obtains a 3D segmentation mask for a target object.\nUsing the mask, a perceptual optimizationbased approach is then introduced that\nleverages learned 2D image inpainters, distilling their information into 3D\nspace, while ensuring view consistency. We also address the lack of a diverse\nbenchmark for evaluating 3D scene inpainting methods by introducing a dataset\ncomprised of challenging real-world scenes. In particular, our dataset contains\nviews of the same scene with and without a target object, enabling more\nprincipled benchmarking of the 3D inpainting task. We first demonstrate the\nsuperiority of our approach on multiview segmentation, comparing to NeRFbased\nmethods and 2D segmentation approaches. We then evaluate on the task of 3D\ninpainting, establishing state-ofthe-art performance against other NeRF\nmanipulation algorithms, as well as a strong 2D image inpainter baseline.\nProject Page: https://spinnerf3d.github.io\n", "date": "2022-11-22", "categories": []}
{"id": "2303.09152", "title": "Learning a Room with the Occ-SDF Hybrid: Signed Distance Function\n  Mingled with Occupancy Aids Scene Representation", "abstract": "  Implicit neural rendering, which uses signed distance function (SDF)\nrepresentation with geometric priors (such as depth or surface normal), has led\nto impressive progress in the surface reconstruction of large-scale scenes.\nHowever, applying this method to reconstruct a room-level scene from images may\nmiss structures in low-intensity areas or small and thin objects. We conducted\nexperiments on three datasets to identify limitations of the original color\nrendering loss and priors-embedded SDF scene representation.\n  We found that the color rendering loss results in optimization bias against\nlow-intensity areas, causing gradient vanishing and leaving these areas\nunoptimized. To address this issue, we propose a feature-based color rendering\nloss that utilizes non-zero feature values to bring back optimization signals.\nAdditionally, the SDF representation can be influenced by objects along a ray\npath, disrupting the monotonic change of SDF values when a single object is\npresent. To counteract this, we explore using the occupancy representation,\nwhich encodes each point separately and is unaffected by objects along a\nquerying ray. Our experimental results demonstrate that the joint forces of the\nfeature-based rendering loss and Occ-SDF hybrid representation scheme can\nprovide high-quality reconstruction results, especially in challenging\nroom-level scenarios. The code would be released.\n", "date": "2023-03-16", "categories": []}
{"id": "2303.09431", "title": "NeRFMeshing: Distilling Neural Radiance Fields into\n  Geometrically-Accurate 3D Meshes", "abstract": "  With the introduction of Neural Radiance Fields (NeRFs), novel view synthesis\nhas recently made a big leap forward. At the core, NeRF proposes that each 3D\npoint can emit radiance, allowing to conduct view synthesis using\ndifferentiable volumetric rendering. While neural radiance fields can\naccurately represent 3D scenes for computing the image rendering, 3D meshes are\nstill the main scene representation supported by most computer graphics and\nsimulation pipelines, enabling tasks such as real time rendering and\nphysics-based simulations. Obtaining 3D meshes from neural radiance fields\nstill remains an open challenge since NeRFs are optimized for view synthesis,\nnot enforcing an accurate underlying geometry on the radiance field. We thus\npropose a novel compact and flexible architecture that enables easy 3D surface\nreconstruction from any NeRF-driven approach. Upon having trained the radiance\nfield, we distill the volumetric 3D representation into a Signed Surface\nApproximation Network, allowing easy extraction of the 3D mesh and appearance.\nOur final 3D mesh is physically accurate and can be rendered in real time on an\narray of devices.\n", "date": "2023-03-16", "categories": []}
{"id": "2303.06418", "title": "Rethinking the Multi-view Stereo from the Perspective of Rendering-based\n  Augmentation", "abstract": "  GigaMVS presents several challenges to existing Multi-View Stereo (MVS)\nalgorithms for its large scale, complex occlusions, and gigapixel images. To\naddress these problems, we first apply one of the state-of-the-art\nlearning-based MVS methods, --MVSFormer, to overcome intractable scenarios such\nas textureless and reflections regions suffered by traditional PatchMatch\nmethods, but it fails in a few large scenes' reconstructions. Moreover,\ntraditional PatchMatch algorithms such as ACMMP, OpenMVS, and RealityCapture\nare leveraged to further improve the completeness in large scenes. Furthermore,\nto unify both advantages of deep learning methods and the traditional\nPatchMatch, we propose to render depth and color images to further fine-tune\nthe MVSFormer model. Notably, we find that the MVS method could produce much\nbetter predictions through rendered images due to the coincident illumination,\nwhich we believe is significant for the MVS community. Thus, MVSFormer is\ncapable of generalizing to large-scale scenes and complementarily solves the\ntextureless reconstruction problem. Finally, we have assembled all point clouds\nmentioned above \\textit{except ones from RealityCapture} and ranked Top-1 on\nthe competitive GigaReconstruction.\n", "date": "2023-03-11", "categories": []}
{"id": "2303.06884", "title": "SCPNet: Semantic Scene Completion on Point Cloud", "abstract": "  Training deep models for semantic scene completion (SSC) is challenging due\nto the sparse and incomplete input, a large quantity of objects of diverse\nscales as well as the inherent label noise for moving objects. To address the\nabove-mentioned problems, we propose the following three solutions: 1)\nRedesigning the completion sub-network. We design a novel completion\nsub-network, which consists of several Multi-Path Blocks (MPBs) to aggregate\nmulti-scale features and is free from the lossy downsampling operations. 2)\nDistilling rich knowledge from the multi-frame model. We design a novel\nknowledge distillation objective, dubbed Dense-to-Sparse Knowledge Distillation\n(DSKD). It transfers the dense, relation-based semantic knowledge from the\nmulti-frame teacher to the single-frame student, significantly improving the\nrepresentation learning of the single-frame model. 3) Completion label\nrectification. We propose a simple yet effective label rectification strategy,\nwhich uses off-the-shelf panoptic segmentation labels to remove the traces of\ndynamic objects in completion labels, greatly improving the performance of deep\nmodels especially for those moving objects. Extensive experiments are conducted\nin two public SSC benchmarks, i.e., SemanticKITTI and SemanticPOSS. Our SCPNet\nranks 1st on SemanticKITTI semantic scene completion challenge and surpasses\nthe competitive S3CNet by 7.2 mIoU. SCPNet also outperforms previous completion\nalgorithms on the SemanticPOSS dataset. Besides, our method also achieves\ncompetitive results on SemanticKITTI semantic segmentation tasks, showing that\nknowledge learned in the scene completion is beneficial to the segmentation\ntask.\n", "date": "2023-03-13", "categories": []}
{"id": "2208.07227", "title": "DM-NeRF: 3D Scene Geometry Decomposition and Manipulation from 2D Images", "abstract": "  In this paper, we study the problem of 3D scene geometry decomposition and\nmanipulation from 2D views. By leveraging the recent implicit neural\nrepresentation techniques, particularly the appealing neural radiance fields,\nwe introduce an object field component to learn unique codes for all individual\nobjects in 3D space only from 2D supervision. The key to this component is a\nseries of carefully designed loss functions to enable every 3D point,\nespecially in non-occupied space, to be effectively optimized even without 3D\nlabels. In addition, we introduce an inverse query algorithm to freely\nmanipulate any specified 3D object shape in the learned scene representation.\nNotably, our manipulation algorithm can explicitly tackle key issues such as\nobject collisions and visual occlusions. Our method, called DM-NeRF, is among\nthe first to simultaneously reconstruct, decompose, manipulate and render\ncomplex 3D scenes in a single pipeline. Extensive experiments on three datasets\nclearly show that our method can accurately decompose all 3D objects from 2D\nviews, allowing any interested object to be freely manipulated in 3D space such\nas translation, rotation, size adjustment, and deformation.\n", "date": "2022-08-15", "categories": []}
{"id": "2303.03361", "title": "Nerflets: Local Radiance Fields for Efficient Structure-Aware 3D Scene\n  Representation from 2D Supervision", "abstract": "  We address efficient and structure-aware 3D scene representation from images.\nNerflets are our key contribution -- a set of local neural radiance fields that\ntogether represent a scene. Each nerflet maintains its own spatial position,\norientation, and extent, within which it contributes to panoptic, density, and\nradiance reconstructions. By leveraging only photometric and inferred panoptic\nimage supervision, we can directly and jointly optimize the parameters of a set\nof nerflets so as to form a decomposed representation of the scene, where each\nobject instance is represented by a group of nerflets. During experiments with\nindoor and outdoor environments, we find that nerflets: (1) fit and approximate\nthe scene more efficiently than traditional global NeRFs, (2) allow the\nextraction of panoptic and photometric renderings from arbitrary views, and (3)\nenable tasks rare for NeRFs, such as 3D panoptic segmentation and interactive\nediting.\n", "date": "2023-03-06", "categories": []}
{"id": "2303.05937", "title": "Structural Multiplane Image: Bridging Neural View Synthesis and 3D\n  Reconstruction", "abstract": "  The Multiplane Image (MPI), containing a set of fronto-parallel RGBA layers,\nis an effective and efficient representation for view synthesis from sparse\ninputs. Yet, its fixed structure limits the performance, especially for\nsurfaces imaged at oblique angles. We introduce the Structural MPI (S-MPI),\nwhere the plane structure approximates 3D scenes concisely. Conveying RGBA\ncontexts with geometrically-faithful structures, the S-MPI directly bridges\nview synthesis and 3D reconstruction. It can not only overcome the critical\nlimitations of MPI, i.e., discretization artifacts from sloped surfaces and\nabuse of redundant layers, and can also acquire planar 3D reconstruction.\nDespite the intuition and demand of applying S-MPI, great challenges are\nintroduced, e.g., high-fidelity approximation for both RGBA layers and plane\nposes, multi-view consistency, non-planar regions modeling, and efficient\nrendering with intersected planes. Accordingly, we propose a transformer-based\nnetwork based on a segmentation model. It predicts compact and expressive S-MPI\nlayers with their corresponding masks, poses, and RGBA contexts. Non-planar\nregions are inclusively handled as a special case in our unified framework.\nMulti-view consistency is ensured by sharing global proxy embeddings, which\nencode plane-level features covering the complete 3D scenes with aligned\ncoordinates. Intensive experiments show that our method outperforms both\nprevious state-of-the-art MPI-based view synthesis methods and planar\nreconstruction methods.\n", "date": "2023-03-10", "categories": []}
{"id": "2211.12046", "title": "DP-NeRF: Deblurred Neural Radiance Field with Physical Scene Priors", "abstract": "  Neural Radiance Field (NeRF) has exhibited outstanding three-dimensional (3D)\nreconstruction quality via the novel view synthesis from multi-view images and\npaired calibrated camera parameters. However, previous NeRF-based systems have\nbeen demonstrated under strictly controlled settings, with little attention\npaid to less ideal scenarios, including with the presence of noise such as\nexposure, illumination changes, and blur. In particular, though blur frequently\noccurs in real situations, NeRF that can handle blurred images has received\nlittle attention. The few studies that have investigated NeRF for blurred\nimages have not considered geometric and appearance consistency in 3D space,\nwhich is one of the most important factors in 3D reconstruction. This leads to\ninconsistency and the degradation of the perceptual quality of the constructed\nscene. Hence, this paper proposes a DP-NeRF, a novel clean NeRF framework for\nblurred images, which is constrained with two physical priors. These priors are\nderived from the actual blurring process during image acquisition by the\ncamera. DP-NeRF proposes rigid blurring kernel to impose 3D consistency\nutilizing the physical priors and adaptive weight proposal to refine the color\ncomposition error in consideration of the relationship between depth and blur.\nWe present extensive experimental results for synthetic and real scenes with\ntwo types of blur: camera motion blur and defocus blur. The results demonstrate\nthat DP-NeRF successfully improves the perceptual quality of the constructed\nNeRF ensuring 3D geometric and appearance consistency. We further demonstrate\nthe effectiveness of our model with comprehensive ablation analysis.\n", "date": "2022-11-22", "categories": []}
{"id": "2301.13510", "title": "3D Former: Monocular Scene Reconstruction with 3D SDF Transformers", "abstract": "  Monocular scene reconstruction from posed images is challenging due to the\ncomplexity of a large environment. Recent volumetric methods learn to directly\npredict the TSDF volume and have demonstrated promising results in this task.\nHowever, most methods focus on how to extract and fuse the 2D features to a 3D\nfeature volume, but none of them improve the way how the 3D volume is\naggregated. In this work, we propose an SDF transformer network, which replaces\nthe role of 3D CNN for better 3D feature aggregation. To reduce the explosive\ncomputation complexity of the 3D multi-head attention, we propose a sparse\nwindow attention module, where the attention is only calculated between the\nnon-empty voxels within a local window. Then a top-down-bottom-up 3D attention\nnetwork is built for 3D feature aggregation, where a dilate-attention structure\nis proposed to prevent geometry degeneration, and two global modules are\nemployed to equip with global receptive fields. The experiments on multiple\ndatasets show that this 3D transformer network generates a more accurate and\ncomplete reconstruction, which outperforms previous methods by a large margin.\nRemarkably, the mesh accuracy is improved by 41.8%, and the mesh completeness\nis improved by 25.3% on the ScanNet dataset. Project page:\nhttps://weihaosky.github.io/sdfformer.\n", "date": "2023-01-31", "categories": []}
{"id": "2206.12455", "title": "Ev-NeRF: Event Based Neural Radiance Field", "abstract": "  We present Ev-NeRF, a Neural Radiance Field derived from event data. While\nevent cameras can measure subtle brightness changes in high frame rates, the\nmeasurements in low lighting or extreme motion suffer from significant domain\ndiscrepancy with complex noise. As a result, the performance of event-based\nvision tasks does not transfer to challenging environments, where the event\ncameras are expected to thrive over normal cameras. We find that the multi-view\nconsistency of NeRF provides a powerful self-supervision signal for eliminating\nthe spurious measurements and extracting the consistent underlying structure\ndespite highly noisy input. Instead of posed images of the original NeRF, the\ninput to Ev-NeRF is the event measurements accompanied by the movements of the\nsensors. Using the loss function that reflects the measurement model of the\nsensor, Ev-NeRF creates an integrated neural volume that summarizes the\nunstructured and sparse data points captured for about 2-4 seconds. The\ngenerated neural volume can also produce intensity images from novel views with\nreasonable depth estimates, which can serve as a high-quality input to various\nvision-based tasks. Our results show that Ev-NeRF achieves competitive\nperformance for intensity image reconstruction under extreme noise conditions\nand high-dynamic-range imaging.\n", "date": "2022-06-24", "categories": []}
{"id": "2303.02978", "title": "System for 3D Acquisition and 3D Reconstruction using Structured Light\n  for Sewer Line Inspection", "abstract": "  The assessment of sewer pipe systems is a highly important, but at the same\ntime cumbersome and error-prone task. We introduce an innovative system based\non single-shot structured light modules that facilitates the detection and\nclassification of spatial defects like jutting intrusions, spallings, or\nmisaligned joints. This system creates highly accurate 3D measurements with\nsub-millimeter resolution of pipe surfaces and fuses them into a holistic 3D\nmodel. The benefit of such a holistic 3D model is twofold: on the one hand, it\nfacilitates the accurate manual sewer pipe assessment, on the other, it\nsimplifies the detection of defects in downstream automatic systems as it\nendows the input with highly accurate depth information. In this work, we\nprovide an extensive overview of the system and give valuable insights into our\ndesign choices.\n", "date": "2023-03-06", "categories": []}
{"id": "2209.00082", "title": "Multi-View Reconstruction using Signed Ray Distance Functions (SRDF)", "abstract": "  In this paper, we investigate a new optimization framework for multi-view 3D\nshape reconstructions. Recent differentiable rendering approaches have provided\nbreakthrough performances with implicit shape representations though they can\nstill lack precision in the estimated geometries. On the other hand multi-view\nstereo methods can yield pixel wise geometric accuracy with local depth\npredictions along viewing rays. Our approach bridges the gap between the two\nstrategies with a novel volumetric shape representation that is implicit but\nparameterized with pixel depths to better materialize the shape surface with\nconsistent signed distances along viewing rays. The approach retains\npixel-accuracy while benefiting from volumetric integration in the\noptimization. To this aim, depths are optimized by evaluating, at each 3D\nlocation within the volumetric discretization, the agreement between the depth\nprediction consistency and the photometric consistency for the corresponding\npixels. The optimization is agnostic to the associated photo-consistency term\nwhich can vary from a median-based baseline to more elaborate criteria learned\nfunctions. Our experiments demonstrate the benefit of the volumetric\nintegration with depth predictions. They also show that our approach\noutperforms existing approaches over standard 3D benchmarks with better\ngeometry estimations.\n", "date": "2022-08-31", "categories": []}
{"id": "2302.14340", "title": "HelixSurf: A Robust and Efficient Neural Implicit Surface Learning of\n  Indoor Scenes with Iterative Intertwined Regularization", "abstract": "  Recovery of an underlying scene geometry from multiview images stands as a\nlong-time challenge in computer vision research. The recent promise leverages\nneural implicit surface learning and differentiable volume rendering, and\nachieves both the recovery of scene geometry and synthesis of novel views,\nwhere deep priors of neural models are used as an inductive smoothness bias.\nWhile promising for object-level surfaces, these methods suffer when coping\nwith complex scene surfaces. In the meanwhile, traditional multi-view stereo\ncan recover the geometry of scenes with rich textures, by globally optimizing\nthe local, pixel-wise correspondences across multiple views. We are thus\nmotivated to make use of the complementary benefits from the two strategies,\nand propose a method termed Helix-shaped neural implicit Surface learning or\nHelixSurf; HelixSurf uses the intermediate prediction from one strategy as the\nguidance to regularize the learning of the other one, and conducts such\nintertwined regularization iteratively during the learning process. We also\npropose an efficient scheme for differentiable volume rendering in HelixSurf.\nExperiments on surface reconstruction of indoor scenes show that our method\ncompares favorably with existing methods and is orders of magnitude faster,\neven when some of existing methods are assisted with auxiliary training data.\nThe source code is available at https://github.com/Gorilla-Lab-SCUT/HelixSurf.\n", "date": "2023-02-28", "categories": []}
{"id": "2303.00050", "title": "Dynamic Multi-View Scene Reconstruction Using Neural Implicit Surface", "abstract": "  Reconstructing general dynamic scenes is important for many computer vision\nand graphics applications. Recent works represent the dynamic scene with neural\nradiance fields for photorealistic view synthesis, while their surface geometry\nis under-constrained and noisy. Other works introduce surface constraints to\nthe implicit neural representation to disentangle the ambiguity of geometry and\nappearance field for static scene reconstruction. To bridge the gap between\nrendering dynamic scenes and recovering static surface geometry, we propose a\ntemplate-free method to reconstruct surface geometry and appearance using\nneural implicit representations from multi-view videos. We leverage\ntopology-aware deformation and the signed distance field to learn complex\ndynamic surfaces via differentiable volume rendering without scene-specific\nprior knowledge like template models. Furthermore, we propose a novel\nmask-based ray selection strategy to significantly boost the optimization on\nchallenging time-varying regions. Experiments on different multi-view video\ndatasets demonstrate that our method achieves high-fidelity surface\nreconstruction as well as photorealistic novel view synthesis.\n", "date": "2023-02-28", "categories": []}
{"id": "2303.00236", "title": "P$^2$SDF for Neural Indoor Scene Reconstruction", "abstract": "  Given only a set of images, neural implicit surface representation has shown\nits capability in 3D surface reconstruction. However, as the nature of\nper-scene optimization is based on the volumetric rendering of color, previous\nneural implicit surface reconstruction methods usually fail in low-textured\nregions, including the floors, walls, etc., which commonly exist for indoor\nscenes. Being aware of the fact that these low-textured regions usually\ncorrespond to planes, without introducing additional ground-truth supervisory\nsignals or making additional assumptions about the room layout, we propose to\nleverage a novel Pseudo Plane-regularized Signed Distance Field (P$^2$SDF) for\nindoor scene reconstruction. Specifically, we consider adjacent pixels with\nsimilar colors to be on the same pseudo planes. The plane parameters are then\nestimated on the fly during training by an efficient and effective two-step\nscheme. Then the signed distances of the points on the planes are regularized\nby the estimated plane parameters in the training phase. As the unsupervised\nplane segments are usually noisy and inaccurate, we propose to assign different\nweights to the sampled points on the plane in plane estimation as well as the\nregularization loss. The weights come by fusing the plane segments from\ndifferent views. As the sampled rays in the planar regions are redundant,\nleading to inefficient training, we further propose a keypoint-guided rays\nsampling strategy that attends to the informative textured regions with large\ncolor variations, and the implicit network gets a better reconstruction,\ncompared with the original uniform ray sampling strategy. Experiments show that\nour P$^2$SDF achieves competitive reconstruction performance in Manhattan\nscenes. Further, as we do not introduce any additional room layout assumption,\nour P$^2$SDF generalizes well to the reconstruction of non-Manhattan scenes.\n", "date": "2023-03-01", "categories": []}
{"id": "2303.00749", "title": "S-NeRF: Neural Radiance Fields for Street Views", "abstract": "  Neural Radiance Fields (NeRFs) aim to synthesize novel views of objects and\nscenes, given the object-centric camera views with large overlaps. However, we\nconjugate that this paradigm does not fit the nature of the street views that\nare collected by many self-driving cars from the large-scale unbounded scenes.\nAlso, the onboard cameras perceive scenes without much overlapping. Thus,\nexisting NeRFs often produce blurs, 'floaters' and other artifacts on\nstreet-view synthesis. In this paper, we propose a new street-view NeRF\n(S-NeRF) that considers novel view synthesis of both the large-scale background\nscenes and the foreground moving vehicles jointly. Specifically, we improve the\nscene parameterization function and the camera poses for learning better neural\nrepresentations from street views. We also use the the noisy and sparse LiDAR\npoints to boost the training and learn a robust geometry and reprojection based\nconfidence to address the depth outliers. Moreover, we extend our S-NeRF for\nreconstructing moving vehicles that is impracticable for conventional NeRFs.\nThorough experiments on the large-scale driving datasets (e.g., nuScenes and\nWaymo) demonstrate that our method beats the state-of-the-art rivals by\nreducing 7% to 40% of the mean-squared error in the street-view synthesis and a\n45% PSNR gain for the moving vehicles rendering.\n", "date": "2023-03-01", "categories": []}
{"id": "2302.14363", "title": "Efficient Implicit Neural Reconstruction Using LiDAR", "abstract": "  Modeling scene geometry using implicit neural representation has revealed its\nadvantages in accuracy, flexibility, and low memory usage. Previous approaches\nhave demonstrated impressive results using color or depth images but still have\ndifficulty handling poor light conditions and large-scale scenes. Methods\ntaking global point cloud as input require accurate registration and ground\ntruth coordinate labels, which limits their application scenarios. In this\npaper, we propose a new method that uses sparse LiDAR point clouds and rough\nodometry to reconstruct fine-grained implicit occupancy field efficiently\nwithin a few minutes. We introduce a new loss function that supervises directly\nin 3D space without 2D rendering, avoiding information loss. We also manage to\nrefine poses of input frames in an end-to-end manner, creating consistent\ngeometry without global point cloud registration. As far as we know, our method\nis the first to reconstruct implicit scene representation from LiDAR-only\ninput. Experiments on synthetic and real-world datasets, including indoor and\noutdoor scenes, prove that our method is effective, efficient, and accurate,\nobtaining comparable results with existing methods using dense input.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.12883", "title": "3D Surface Reconstruction in the Wild by Deforming Shape Priors from\n  Synthetic Data", "abstract": "  Reconstructing the underlying 3D surface of an object from a single image is\na challenging problem that has received extensive attention from the computer\nvision community. Many learning-based approaches tackle this problem by\nlearning a 3D shape prior from either ground truth 3D data or multi-view\nobservations. To achieve state-of-the-art results, these methods assume that\nthe objects are specified with respect to a fixed canonical coordinate frame,\nwhere instances of the same category are perfectly aligned. In this work, we\npresent a new method for joint category-specific 3D reconstruction and object\npose estimation from a single image. We show that one can leverage shape priors\nlearned on purely synthetic 3D data together with a point cloud pose\ncanonicalization method to achieve high-quality 3D reconstruction in the wild.\nGiven a single depth image at test time, we first transform this partial point\ncloud into a learned canonical frame. Then, we use a neural deformation field\nto reconstruct the 3D surface of the object. Finally, we jointly optimize\nobject pose and 3D shape to fit the partial depth observation. Our approach\nachieves state-of-the-art reconstruction performance across several real-world\ndatasets, even when trained only on synthetic data. We further show that our\nmethod generalizes to different input modalities, from dense depth images to\nsparse and noisy LIDAR scans.\n", "date": "2023-02-24", "categories": []}
{"id": "2302.13540", "title": "OccDepth: A Depth-Aware Method for 3D Semantic Scene Completion", "abstract": "  3D Semantic Scene Completion (SSC) can provide dense geometric and semantic\nscene representations, which can be applied in the field of autonomous driving\nand robotic systems. It is challenging to estimate the complete geometry and\nsemantics of a scene solely from visual images, and accurate depth information\nis crucial for restoring 3D geometry. In this paper, we propose the first\nstereo SSC method named OccDepth, which fully exploits implicit depth\ninformation from stereo images (or RGBD images) to help the recovery of 3D\ngeometric structures. The Stereo Soft Feature Assignment (Stereo-SFA) module is\nproposed to better fuse 3D depth-aware features by implicitly learning the\ncorrelation between stereo images. In particular, when the input are RGBD\nimage, a virtual stereo images can be generated through original RGB image and\ndepth map. Besides, the Occupancy Aware Depth (OAD) module is used to obtain\ngeometry-aware 3D features by knowledge distillation using pre-trained depth\nmodels. In addition, a reformed TartanAir benchmark, named SemanticTartanAir,\nis provided in this paper for further testing our OccDepth method on SSC task.\nCompared with the state-of-the-art RGB-inferred SSC method, extensive\nexperiments on SemanticKITTI show that our OccDepth method achieves superior\nperformance with improving +4.82% mIoU, of which +2.49% mIoU comes from stereo\nimages and +2.33% mIoU comes from our proposed depth-aware method. Our code and\ntrained models are available at https://github.com/megvii-research/OccDepth.\n", "date": "2023-02-27", "categories": []}
{"id": "2302.14052", "title": "LODE: Locally Conditioned Eikonal Implicit Scene Completion from Sparse\n  LiDAR", "abstract": "  Scene completion refers to obtaining dense scene representation from an\nincomplete perception of complex 3D scenes. This helps robots detect\nmulti-scale obstacles and analyse object occlusions in scenarios such as\nautonomous driving. Recent advances show that implicit representation learning\ncan be leveraged for continuous scene completion and achieved through physical\nconstraints like Eikonal equations. However, former Eikonal completion methods\nonly demonstrate results on watertight meshes at a scale of tens of meshes.\nNone of them are successfully done for non-watertight LiDAR point clouds of\nopen large scenes at a scale of thousands of scenes. In this paper, we propose\na novel Eikonal formulation that conditions the implicit representation on\nlocalized shape priors which function as dense boundary value constraints, and\ndemonstrate it works on SemanticKITTI and SemanticPOSS. It can also be extended\nto semantic Eikonal scene completion with only small modifications to the\nnetwork architecture. With extensive quantitative and qualitative results, we\ndemonstrate the benefits and drawbacks of existing Eikonal methods, which\nnaturally leads to the new locally conditioned formulation. Notably, we improve\nIoU from 31.7% to 51.2% on SemanticKITTI and from 40.5% to 48.7% on\nSemanticPOSS. We extensively ablate our methods and demonstrate that the\nproposed formulation is robust to a wide spectrum of implementation\nhyper-parameters. Codes and models are publicly available at\nhttps://github.com/AIR-DISCOVER/LODE.\n", "date": "2023-02-27", "categories": []}
{"id": "2302.10668", "title": "$PC^2$: Projection-Conditioned Point Cloud Diffusion for Single-Image 3D\n  Reconstruction", "abstract": "  Reconstructing the 3D shape of an object from a single RGB image is a\nlong-standing and highly challenging problem in computer vision. In this paper,\nwe propose a novel method for single-image 3D reconstruction which generates a\nsparse point cloud via a conditional denoising diffusion process. Our method\ntakes as input a single RGB image along with its camera pose and gradually\ndenoises a set of 3D points, whose positions are initially sampled randomly\nfrom a three-dimensional Gaussian distribution, into the shape of an object.\nThe key to our method is a geometrically-consistent conditioning process which\nwe call projection conditioning: at each step in the diffusion process, we\nproject local image features onto the partially-denoised point cloud from the\ngiven camera pose. This projection conditioning process enables us to generate\nhigh-resolution sparse geometries that are well-aligned with the input image,\nand can additionally be used to predict point colors after shape\nreconstruction. Moreover, due to the probabilistic nature of the diffusion\nprocess, our method is naturally capable of generating multiple different\nshapes consistent with a single input image. In contrast to prior work, our\napproach not only performs well on synthetic benchmarks, but also gives large\nqualitative improvements on complex real-world data.\n", "date": "2023-02-21", "categories": []}
{"id": "2210.02299", "title": "SHINE-Mapping: Large-Scale 3D Mapping Using Sparse Hierarchical Implicit\n  Neural Representations", "abstract": "  Accurate mapping of large-scale environments is an essential building block\nof most outdoor autonomous systems. Challenges of traditional mapping methods\ninclude the balance between memory consumption and mapping accuracy. This paper\naddresses the problem of achieving large-scale 3D reconstruction using implicit\nrepresentations built from 3D LiDAR measurements. We learn and store implicit\nfeatures through an octree-based, hierarchical structure, which is sparse and\nextensible. The implicit features can be turned into signed distance values\nthrough a shallow neural network. We leverage binary cross entropy loss to\noptimize the local features with the 3D measurements as supervision. Based on\nour implicit representation, we design an incremental mapping system with\nregularization to tackle the issue of forgetting in continual learning. Our\nexperiments show that our 3D reconstructions are more accurate, complete, and\nmemory-efficient than current state-of-the-art 3D mapping methods.\n", "date": "2022-10-05", "categories": []}
{"id": "2212.00792", "title": "SparseFusion: Distilling View-conditioned Diffusion for 3D\n  Reconstruction", "abstract": "  We propose SparseFusion, a sparse view 3D reconstruction approach that\nunifies recent advances in neural rendering and probabilistic image generation.\nExisting approaches typically build on neural rendering with re-projected\nfeatures but fail to generate unseen regions or handle uncertainty under large\nviewpoint changes. Alternate methods treat this as a (probabilistic) 2D\nsynthesis task, and while they can generate plausible 2D images, they do not\ninfer a consistent underlying 3D. However, we find that this trade-off between\n3D consistency and probabilistic image generation does not need to exist. In\nfact, we show that geometric consistency and generative inference can be\ncomplementary in a mode-seeking behavior. By distilling a 3D consistent scene\nrepresentation from a view-conditioned latent diffusion model, we are able to\nrecover a plausible 3D representation whose renderings are both accurate and\nrealistic. We evaluate our approach across 51 categories in the CO3D dataset\nand show that it outperforms existing methods, in both distortion and\nperception metrics, for sparse-view novel view synthesis.\n", "date": "2022-12-01", "categories": []}
{"id": "2302.06793", "title": "HR-NeuS: Recovering High-Frequency Surface Geometry via Neural Implicit\n  Surfaces", "abstract": "  Recent advances in neural implicit surfaces for multi-view 3D reconstruction\nprimarily focus on improving large-scale surface reconstruction accuracy, but\noften produce over-smoothed geometries that lack fine surface details. To\naddress this, we present High-Resolution NeuS (HR-NeuS), a novel neural\nimplicit surface reconstruction method that recovers high-frequency surface\ngeometry while maintaining large-scale reconstruction accuracy. We achieve this\nby utilizing (i) multi-resolution hash grid encoding rather than positional\nencoding at high frequencies, which boosts our model's expressiveness of local\ngeometry details; (ii) a coarse-to-fine algorithmic framework that selectively\napplies surface regularization to coarse geometry without smoothing away fine\ndetails; (iii) a coarse-to-fine grid annealing strategy to train the network.\nWe demonstrate through experiments on DTU and BlendedMVS datasets that our\napproach produces 3D geometries that are qualitatively more detailed and\nquantitatively of similar accuracy compared to previous approaches.\n", "date": "2023-02-14", "categories": []}
{"id": "2302.05573", "title": "3D Colored Shape Reconstruction from a Single RGB Image through\n  Diffusion", "abstract": "  We propose a novel 3d colored shape reconstruction method from a single RGB\nimage through diffusion model. Diffusion models have shown great development\npotentials for high-quality 3D shape generation. However, most existing work\nbased on diffusion models only focus on geometric shape generation, they cannot\neither accomplish 3D reconstruction from a single image, or produce 3D\ngeometric shape with color information. In this work, we propose to reconstruct\na 3D colored shape from a single RGB image through a novel conditional\ndiffusion model. The reverse process of the proposed diffusion model is\nconsisted of three modules, shape prediction module, color prediction module\nand NeRF-like rendering module. In shape prediction module, the reference RGB\nimage is first encoded into a high-level shape feature and then the shape\nfeature is utilized as a condition to predict the reverse geometric noise in\ndiffusion model. Then the color of each 3D point updated in shape prediction\nmodule is predicted by color prediction module. Finally, a NeRF-like rendering\nmodule is designed to render the colored point cloud predicted by the former\ntwo modules to 2D image space to guide the training conditioned only on a\nreference image. As far as the authors know, the proposed method is the first\ndiffusion model for 3D colored shape reconstruction from a single RGB image.\nExperimental results demonstrate that the proposed method achieves competitive\nperformance on colored 3D shape reconstruction, and the ablation study\nvalidates the positive role of the color prediction module in improving the\nreconstruction quality of 3D geometric point cloud.\n", "date": "2023-02-11", "categories": []}
{"id": "2207.10985", "title": "NeurAR: Neural Uncertainty for Autonomous 3D Reconstruction with\n  Implicit Neural Representations", "abstract": "  Implicit neural representations have shown compelling results in offline 3D\nreconstruction and also recently demonstrated the potential for online SLAM\nsystems. However, applying them to autonomous 3D reconstruction, where a robot\nis required to explore a scene and plan a view path for the reconstruction, has\nnot been studied. In this paper, we explore for the first time the possibility\nof using implicit neural representations for autonomous 3D scene reconstruction\nby addressing two key challenges: 1) seeking a criterion to measure the quality\nof the candidate viewpoints for the view planning based on the new\nrepresentations, and 2) learning the criterion from data that can generalize to\ndifferent scenes instead of a hand-crafting one. To solve the challenges,\nfirstly, a proxy of Peak Signal-to-Noise Ratio (PSNR) is proposed to quantify a\nviewpoint quality; secondly, the proxy is optimized jointly with the parameters\nof an implicit neural network for the scene. With the proposed view quality\ncriterion from neural networks (termed as Neural Uncertainty), we can then\napply implicit representations to autonomous 3D reconstruction. Our method\ndemonstrates significant improvements on various metrics for the rendered image\nquality and the geometry quality of the reconstructed 3D models when compared\nwith variants using TSDF or reconstruction without view planning. Project\nwebpage https://kingteeloki-ran.github.io/NeurAR/\n", "date": "2022-07-22", "categories": []}
{"id": "2301.13244", "title": "Mono-STAR: Mono-camera Scene-level Tracking and Reconstruction", "abstract": "  We present Mono-STAR, the first real-time 3D reconstruction system that\nsimultaneously supports semantic fusion, fast motion tracking, non-rigid object\ndeformation, and topological change under a unified framework. The proposed\nsystem solves a new optimization problem incorporating optical-flow-based 2D\nconstraints to deal with fast motion and a novel semantic-aware deformation\ngraph (SAD-graph) for handling topology change. We test the proposed system\nunder various challenging scenes and demonstrate that it significantly\noutperforms existing state-of-the-art methods.\n", "date": "2023-01-30", "categories": []}
{"id": "2208.10925", "title": "Vox-Surf: Voxel-based Implicit Surface Representation", "abstract": "  Virtual content creation and interaction play an important role in modern 3D\napplications such as AR and VR. Recovering detailed 3D models from real scenes\ncan significantly expand the scope of its applications and has been studied for\ndecades in the computer vision and computer graphics community. We propose\nVox-Surf, a voxel-based implicit surface representation. Our Vox-Surf divides\nthe space into finite bounded voxels. Each voxel stores geometry and appearance\ninformation in its corner vertices. Vox-Surf is suitable for almost any\nscenario thanks to sparsity inherited from voxel representation and can be\neasily trained from multiple view images. We leverage the progressive training\nprocedure to extract important voxels gradually for further optimization so\nthat only valid voxels are preserved, which greatly reduces the number of\nsampling points and increases rendering speed.The fine voxels can also be\nconsidered as the bounding volume for collision detection.The experiments show\nthat Vox-Surf representation can learn delicate surface details and accurate\ncolor with less memory and faster rendering speed than other methods.We also\nshow that Vox-Surf can be more practical in scene editing and AR applications.\n", "date": "2022-08-21", "categories": []}
{"id": "2301.11522", "title": "A Comparison of Tiny-nerf versus Spatial Representations for 3d\n  Reconstruction", "abstract": "  Neural rendering has emerged as a powerful paradigm for synthesizing images,\noffering many benefits over classical rendering by using neural networks to\nreconstruct surfaces, represent shapes, and synthesize novel views, either for\nobjects or scenes. In this neural rendering, the environment is encoded into a\nneural network. We believe that these new representations can be used to codify\nthe scene for a mobile robot. Therefore, in this work, we perform a comparison\nbetween a trending neural rendering, called tiny-NeRF, and other volume\nrepresentations that are commonly used as maps in robotics, such as voxel maps,\npoint clouds, and triangular meshes. The target is to know the advantages and\ndisadvantages of neural representations in the robotics context. The comparison\nis made in terms of spatial complexity and processing time to obtain a model.\nExperiments show that tiny-NeRF requires three times less memory space compared\nto other representations. In terms of processing time, tiny-NeRF takes about\nsix times more to compute the model.\n", "date": "2023-01-27", "categories": []}
{"id": "2210.15059", "title": "Automated Reconstruction of 3D Open Surfaces from Sparse Point Clouds", "abstract": "  Real-world 3D data may contain intricate details defined by salient surface\ngaps. Automated reconstruction of these open surfaces (e.g., non-watertight\nmeshes) is a challenging problem for environment synthesis in mixed reality\napplications. Current learning-based implicit techniques can achieve high\nfidelity on closed-surface reconstruction. However, their dependence on the\ndistinction between the inside and outside of a surface makes them incapable of\nreconstructing open surfaces. Recently, a new class of implicit functions have\nshown promise in reconstructing open surfaces by regressing an unsigned\ndistance field. Yet, these methods rely on a discretized representation of the\nraw data, which loses important surface details and can lead to outliers in the\nreconstruction. We propose IPVNet, a learning-based implicit model that\npredicts the unsigned distance between a surface and a query point in 3D space\nby leveraging both raw point cloud data and its discretized voxel counterpart.\nExperiments on synthetic and real-world public datasets demonstrates that\nIPVNet outperforms the state of the art while producing far fewer outliers in\nthe reconstruction.\n", "date": "2022-10-26", "categories": []}
{"id": "2301.08247", "title": "Multiview Compressive Coding for 3D Reconstruction", "abstract": "  A central goal of visual recognition is to understand objects and scenes from\na single image. 2D recognition has witnessed tremendous progress thanks to\nlarge-scale learning and general-purpose representations. Comparatively, 3D\nposes new challenges stemming from occlusions not depicted in the image. Prior\nworks try to overcome these by inferring from multiple views or rely on scarce\nCAD models and category-specific priors which hinder scaling to novel settings.\nIn this work, we explore single-view 3D reconstruction by learning\ngeneralizable representations inspired by advances in self-supervised learning.\nWe introduce a simple framework that operates on 3D points of single objects or\nwhole scenes coupled with category-agnostic large-scale training from diverse\nRGB-D videos. Our model, Multiview Compressive Coding (MCC), learns to compress\nthe input appearance and geometry to predict the 3D structure by querying a\n3D-aware decoder. MCC's generality and efficiency allow it to learn from\nlarge-scale and diverse data sources with strong generalization to novel\nobjects imagined by DALL$\\cdot$E 2 or captured in-the-wild with an iPhone.\n", "date": "2023-01-19", "categories": []}
{"id": "2301.04860", "title": "Edge Preserving Implicit Surface Representation of Point Clouds", "abstract": "  Learning implicit surface directly from raw data recently has become a very\nattractive representation method for 3D reconstruction tasks due to its\nexcellent performance. However, as the raw data quality deteriorates, the\nimplicit functions often lead to unsatisfactory reconstruction results. To this\nend, we propose a novel edge-preserving implicit surface reconstruction method,\nwhich mainly consists of a differentiable Laplican regularizer and a dynamic\nedge sampling strategy. Among them, the differential Laplican regularizer can\neffectively alleviate the implicit surface unsmoothness caused by the point\ncloud quality deteriorates; Meanwhile, in order to reduce the excessive\nsmoothing at the edge regions of implicit suface, we proposed a dynamic edge\nextract strategy for sampling near the sharp edge of point cloud, which can\neffectively avoid the Laplacian regularizer from smoothing all regions.\nFinally, we combine them with a simple regularization term for robust implicit\nsurface reconstruction. Compared with the state-of-the-art methods,\nexperimental results show that our method significantly improves the quality of\n3D reconstruction results. Moreover, we demonstrate through several experiments\nthat our method can be conveniently and effectively applied to some point cloud\nanalysis tasks, including point cloud edge feature extraction, normal\nestimation,etc.\n", "date": "2023-01-12", "categories": []}
{"id": "2301.01057", "title": "BS3D: Building-scale 3D Reconstruction from RGB-D Images", "abstract": "  Various datasets have been proposed for simultaneous localization and mapping\n(SLAM) and related problems. Existing datasets often include small\nenvironments, have incomplete ground truth, or lack important sensor data, such\nas depth and infrared images. We propose an easy-to-use framework for acquiring\nbuilding-scale 3D reconstruction using a consumer depth camera. Unlike complex\nand expensive acquisition setups, our system enables crowd-sourcing, which can\ngreatly benefit data-hungry algorithms. Compared to similar systems, we utilize\nraw depth maps for odometry computation and loop closure refinement which\nresults in better reconstructions. We acquire a building-scale 3D dataset\n(BS3D) and demonstrate its value by training an improved monocular depth\nestimation model. As a unique experiment, we benchmark visual-inertial odometry\nmethods using both color and active infrared images.\n", "date": "2023-01-03", "categories": []}
{"id": "2212.14772", "title": "A Combined Approach Toward Consistent Reconstructions of Indoor Spaces\n  Based on 6D RGB-D Odometry and KinectFusion", "abstract": "  We propose a 6D RGB-D odometry approach that finds the relative camera pose\nbetween consecutive RGB-D frames by keypoint extraction and feature matching\nboth on the RGB and depth image planes. Furthermore, we feed the estimated pose\nto the highly accurate KinectFusion algorithm, which uses a fast ICP (Iterative\nClosest Point) to fine-tune the frame-to-frame relative pose and fuse the depth\ndata into a global implicit surface. We evaluate our method on a publicly\navailable RGB-D SLAM benchmark dataset by Sturm et al. The experimental results\nshow that our proposed reconstruction method solely based on visual odometry\nand KinectFusion outperforms the state-of-the-art RGB-D SLAM system accuracy.\nMoreover, our algorithm outputs a ready-to-use polygon mesh (highly suitable\nfor creating 3D virtual worlds) without any postprocessing steps.\n", "date": "2022-12-25", "categories": []}
{"id": "2212.12721", "title": "Polarimetric Multi-View Inverse Rendering", "abstract": "  A polarization camera has great potential for 3D reconstruction since the\nangle of polarization (AoP) and the degree of polarization (DoP) of reflected\nlight are related to an object's surface normal. In this paper, we propose a\nnovel 3D reconstruction method called Polarimetric Multi-View Inverse Rendering\n(Polarimetric MVIR) that effectively exploits geometric, photometric, and\npolarimetric cues extracted from input multi-view color-polarization images. We\nfirst estimate camera poses and an initial 3D model by geometric reconstruction\nwith a standard structure-from-motion and multi-view stereo pipeline. We then\nrefine the initial model by optimizing photometric rendering errors and\npolarimetric errors using multi-view RGB, AoP, and DoP images, where we propose\na novel polarimetric cost function that enables an effective constraint on the\nestimated surface normal of each vertex, while considering four possible\nambiguous azimuth angles revealed from the AoP measurement. The weight for the\npolarimetric cost is effectively determined based on the DoP measurement, which\nis regarded as the reliability of polarimetric information. Experimental\nresults using both synthetic and real data demonstrate that our Polarimetric\nMVIR can reconstruct a detailed 3D shape without assuming a specific surface\nmaterial and lighting condition.\n", "date": "2022-12-24", "categories": []}
{"id": "2212.09802", "title": "Panoptic Lifting for 3D Scene Understanding with Neural Fields", "abstract": "  We propose Panoptic Lifting, a novel approach for learning panoptic 3D\nvolumetric representations from images of in-the-wild scenes. Once trained, our\nmodel can render color images together with 3D-consistent panoptic segmentation\nfrom novel viewpoints.\n  Unlike existing approaches which use 3D input directly or indirectly, our\nmethod requires only machine-generated 2D panoptic segmentation masks inferred\nfrom a pre-trained network. Our core contribution is a panoptic lifting scheme\nbased on a neural field representation that generates a unified and multi-view\nconsistent, 3D panoptic representation of the scene. To account for\ninconsistencies of 2D instance identifiers across views, we solve a linear\nassignment with a cost based on the model's current predictions and the\nmachine-generated segmentation masks, thus enabling us to lift 2D instances to\n3D in a consistent way. We further propose and ablate contributions that make\nour method more robust to noisy, machine-generated labels, including test-time\naugmentations for confidence estimates, segment consistency loss, bounded\nsegmentation fields, and gradient stopping.\n  Experimental results validate our approach on the challenging Hypersim,\nReplica, and ScanNet datasets, improving by 8.4, 13.8, and 10.6% in scene-level\nPQ over state of the art.\n", "date": "2022-12-19", "categories": []}
{"id": "2212.04096", "title": "ALTO: Alternating Latent Topologies for Implicit 3D Reconstruction", "abstract": "  This work introduces alternating latent topologies (ALTO) for high-fidelity\nreconstruction of implicit 3D surfaces from noisy point clouds. Previous work\nidentifies that the spatial arrangement of latent encodings is important to\nrecover detail. One school of thought is to encode a latent vector for each\npoint (point latents). Another school of thought is to project point latents\ninto a grid (grid latents) which could be a voxel grid or triplane grid. Each\nschool of thought has tradeoffs. Grid latents are coarse and lose\nhigh-frequency detail. In contrast, point latents preserve detail. However,\npoint latents are more difficult to decode into a surface, and quality and\nruntime suffer. In this paper, we propose ALTO to sequentially alternate\nbetween geometric representations, before converging to an easy-to-decode\nlatent. We find that this preserves spatial expressiveness and makes decoding\nlightweight. We validate ALTO on implicit 3D recovery and observe not only a\nperformance improvement over the state-of-the-art, but a runtime improvement of\n3-10$\\times$. Project website at https://visual.ee.ucla.edu/alto.htm/.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.04386", "title": "Multi-View Mesh Reconstruction with Neural Deferred Shading", "abstract": "  We propose an analysis-by-synthesis method for fast multi-view 3D\nreconstruction of opaque objects with arbitrary materials and illumination.\nState-of-the-art methods use both neural surface representations and neural\nrendering. While flexible, neural surface representations are a significant\nbottleneck in optimization runtime. Instead, we represent surfaces as triangle\nmeshes and build a differentiable rendering pipeline around triangle\nrasterization and neural shading. The renderer is used in a gradient descent\noptimization where both a triangle mesh and a neural shader are jointly\noptimized to reproduce the multi-view images. We evaluate our method on a\npublic 3D reconstruction dataset and show that it can match the reconstruction\naccuracy of traditional baselines and neural approaches while surpassing them\nin optimization runtime. Additionally, we investigate the shader and find that\nit learns an interpretable representation of appearance, enabling applications\nsuch as 3D material editing.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.03267", "title": "NeRDi: Single-View NeRF Synthesis with Language-Guided Diffusion as\n  General Image Priors", "abstract": "  2D-to-3D reconstruction is an ill-posed problem, yet humans are good at\nsolving this problem due to their prior knowledge of the 3D world developed\nover years. Driven by this observation, we propose NeRDi, a single-view NeRF\nsynthesis framework with general image priors from 2D diffusion models.\nFormulating single-view reconstruction as an image-conditioned 3D generation\nproblem, we optimize the NeRF representations by minimizing a diffusion loss on\nits arbitrary view renderings with a pretrained image diffusion model under the\ninput-view constraint. We leverage off-the-shelf vision-language models and\nintroduce a two-section language guidance as conditioning inputs to the\ndiffusion model. This is essentially helpful for improving multiview content\ncoherence as it narrows down the general image prior conditioned on the\nsemantic and visual features of the single-view input image. Additionally, we\nintroduce a geometric loss based on estimated depth maps to regularize the\nunderlying 3D geometry of the NeRF. Experimental results on the DTU MVS dataset\nshow that our method can synthesize novel views with higher quality even\ncompared to existing methods trained on this dataset. We also demonstrate our\ngeneralizability in zero-shot NeRF synthesis for in-the-wild images.\n", "date": "2022-12-06", "categories": []}
{"id": "2212.02184", "title": "3D-LatentMapper: View Agnostic Single-View Reconstruction of 3D Shapes", "abstract": "  Computer graphics, 3D computer vision and robotics communities have produced\nmultiple approaches to represent and generate 3D shapes, as well as a vast\nnumber of use cases. However, single-view reconstruction remains a challenging\ntopic that can unlock various interesting use cases such as interactive design.\nIn this work, we propose a novel framework that leverages the intermediate\nlatent spaces of Vision Transformer (ViT) and a joint image-text\nrepresentational model, CLIP, for fast and efficient Single View Reconstruction\n(SVR). More specifically, we propose a novel mapping network architecture that\nlearns a mapping between deep features extracted from ViT and CLIP, and the\nlatent space of a base 3D generative model. Unlike previous work, our method\nenables view-agnostic reconstruction of 3D shapes, even in the presence of\nlarge occlusions. We use the ShapeNetV2 dataset and perform extensive\nexperiments with comparisons to SOTA methods to demonstrate our method's\neffectiveness.\n", "date": "2022-12-05", "categories": []}
{"id": "2211.16266", "title": "PatchMatch-Stereo-Panorama, a fast dense reconstruction from 360{\\deg}\n  video images", "abstract": "  This work proposes a new method for real-time dense 3d reconstruction for\ncommon 360{\\deg} action cams, which can be mounted on small scouting UAVs\nduring USAR missions. The proposed method extends a feature based Visual\nmonocular SLAM (OpenVSLAM, based on the popular ORB-SLAM) for robust long-term\nlocalization on equirectangular video input by adding an additional\ndensification thread that computes dense correspondences for any given keyframe\nwith respect to a local keyframe-neighboorhood using a\nPatchMatch-Stereo-approach. While PatchMatch-Stereo-types of algorithms are\nconsidered state of the art for large scale Mutli-View-Stereo they had not been\nadapted so far for real-time dense 3d reconstruction tasks. This work describes\na new massively parallel variant of the PatchMatch-Stereo-algorithm that\ndiffers from current approaches in two ways: First it supports the\nequirectangular camera model while other solutions are limited to the pinhole\ncamera model. Second it is optimized for low latency while keeping a high level\nof completeness and accuracy. To achieve this it operates only on small\nsequences of keyframes, but employs techniques to compensate for the potential\nloss of accuracy due to the limited number of frames. Results demonstrate that\ndense 3d reconstruction is possible on a consumer grade laptop with a recent\nmobile GPU and that it is possible with improved accuracy and completeness over\ncommon offline-MVS solutions with comparable quality settings.\n", "date": "2022-11-29", "categories": []}
{"id": "2210.06853", "title": "NeuralRoom: Geometry-Constrained Neural Implicit Surfaces for Indoor\n  Scene Reconstruction", "abstract": "  We present a novel neural surface reconstruction method called NeuralRoom for\nreconstructing room-sized indoor scenes directly from a set of 2D images.\nRecently, implicit neural representations have become a promising way to\nreconstruct surfaces from multiview images due to their high-quality results\nand simplicity. However, implicit neural representations usually cannot\nreconstruct indoor scenes well because they suffer severe shape-radiance\nambiguity. We assume that the indoor scene consists of texture-rich and flat\ntexture-less regions. In texture-rich regions, the multiview stereo can obtain\naccurate results. In the flat area, normal estimation networks usually obtain a\ngood normal estimation. Based on the above observations, we reduce the possible\nspatial variation range of implicit neural surfaces by reliable geometric\npriors to alleviate shape-radiance ambiguity. Specifically, we use multiview\nstereo results to limit the NeuralRoom optimization space and then use reliable\ngeometric priors to guide NeuralRoom training. Then the NeuralRoom would\nproduce a neural scene representation that can render an image consistent with\nthe input training images. In addition, we propose a smoothing method called\nperturbation-residual restrictions to improve the accuracy and completeness of\nthe flat region, which assumes that the sampling points in a local surface\nshould have the same normal and similar distance to the observation center.\nExperiments on the ScanNet dataset show that our method can reconstruct the\ntexture-less area of indoor scenes while maintaining the accuracy of detail. We\nalso apply NeuralRoom to more advanced multiview reconstruction algorithms and\nsignificantly improve their reconstruction quality.\n", "date": "2022-10-13", "categories": []}
{"id": "2211.14157", "title": "Learning 3D Scene Priors with 2D Supervision", "abstract": "  Holistic 3D scene understanding entails estimation of both layout\nconfiguration and object geometry in a 3D environment. Recent works have shown\nadvances in 3D scene estimation from various input modalities (e.g., images, 3D\nscans), by leveraging 3D supervision (e.g., 3D bounding boxes or CAD models),\nfor which collection at scale is expensive and often intractable. To address\nthis shortcoming, we propose a new method to learn 3D scene priors of layout\nand shape without requiring any 3D ground truth. Instead, we rely on 2D\nsupervision from multi-view RGB images. Our method represents a 3D scene as a\nlatent vector, from which we can progressively decode to a sequence of objects\ncharacterized by their class categories, 3D bounding boxes, and meshes. With\nour trained autoregressive decoder representing the scene prior, our method\nfacilitates many downstream applications, including scene synthesis,\ninterpolation, and single-view reconstruction. Experiments on 3D-FRONT and\nScanNet show that our method outperforms state of the art in single-view\nreconstruction, and achieves state-of-the-art results in scene synthesis\nagainst baselines which require for 3D supervision.\n", "date": "2022-11-25", "categories": []}
{"id": "2211.14173", "title": "NeuralUDF: Learning Unsigned Distance Fields for Multi-view\n  Reconstruction of Surfaces with Arbitrary Topologies", "abstract": "  We present a novel method, called NeuralUDF, for reconstructing surfaces with\narbitrary topologies from 2D images via volume rendering. Recent advances in\nneural rendering based reconstruction have achieved compelling results.\nHowever, these methods are limited to objects with closed surfaces since they\nadopt Signed Distance Function (SDF) as surface representation which requires\nthe target shape to be divided into inside and outside. In this paper, we\npropose to represent surfaces as the Unsigned Distance Function (UDF) and\ndevelop a new volume rendering scheme to learn the neural UDF representation.\nSpecifically, a new density function that correlates the property of UDF with\nthe volume rendering scheme is introduced for robust optimization of the UDF\nfields. Experiments on the DTU and DeepFashion3D datasets show that our method\nnot only enables high-quality reconstruction of non-closed shapes with complex\ntypologies, but also achieves comparable performance to the SDF based methods\non the reconstruction of closed surfaces.\n", "date": "2022-11-25", "categories": []}
{"id": "2211.14249", "title": "Neural Poisson: Indicator Functions for Neural Fields", "abstract": "  Implicit neural field generating signed distance field representations (SDFs)\nof 3D shapes have shown remarkable progress in 3D shape reconstruction and\ngeneration. We introduce a new paradigm for neural field representations of 3D\nscenes; rather than characterizing surfaces as SDFs, we propose a\nPoisson-inspired characterization for surfaces as indicator functions optimized\nby neural fields. Crucially, for reconstruction of real scan data, the\nindicator function representation enables simple and effective constraints\nbased on common range sensing inputs, which indicate empty space based on line\nof sight. Such empty space information is intrinsic to the scanning process,\nand incorporating this knowledge enables more accurate surface reconstruction.\nWe show that our approach demonstrates state-of-the-art reconstruction\nperformance on both synthetic and real scanned 3D scene data, with 9.5%\nimprovement in Chamfer distance over state of the art.\n", "date": "2022-11-25", "categories": []}
{"id": "2211.12758", "title": "PANeRF: Pseudo-view Augmentation for Improved Neural Radiance Fields\n  Based on Few-shot Inputs", "abstract": "  The method of neural radiance fields (NeRF) has been developed in recent\nyears, and this technology has promising applications for synthesizing novel\nviews of complex scenes. However, NeRF requires dense input views, typically\nnumbering in the hundreds, for generating high-quality images. With a decrease\nin the number of input views, the rendering quality of NeRF for unseen\nviewpoints tends to degenerate drastically. To overcome this challenge, we\npropose pseudo-view augmentation of NeRF, a scheme that expands a sufficient\namount of data by considering the geometry of few-shot inputs. We first\ninitialized the NeRF network by leveraging the expanded pseudo-views, which\nefficiently minimizes uncertainty when rendering unseen views. Subsequently, we\nfine-tuned the network by utilizing sparse-view inputs containing precise\ngeometry and color information. Through experiments under various settings, we\nverified that our model faithfully synthesizes novel-view images of superior\nquality and outperforms existing methods for multi-view datasets.\n", "date": "2022-11-23", "categories": []}
{"id": "2211.11836", "title": "Towards Live 3D Reconstruction from Wearable Video: An Evaluation of\n  V-SLAM, NeRF, and Videogrammetry Techniques", "abstract": "  Mixed reality (MR) is a key technology which promises to change the future of\nwarfare. An MR hybrid of physical outdoor environments and virtual military\ntraining will enable engagements with long distance enemies, both real and\nsimulated. To enable this technology, a large-scale 3D model of a physical\nenvironment must be maintained based on live sensor observations. 3D\nreconstruction algorithms should utilize the low cost and pervasiveness of\nvideo camera sensors, from both overhead and soldier-level perspectives.\nMapping speed and 3D quality can be balanced to enable live MR training in\ndynamic environments. Given these requirements, we survey several 3D\nreconstruction algorithms for large-scale mapping for military applications\ngiven only live video. We measure 3D reconstruction performance from common\nstructure from motion, visual-SLAM, and photogrammetry techniques. This\nincludes the open source algorithms COLMAP, ORB-SLAM3, and NeRF using\nInstant-NGP. We utilize the autonomous driving academic benchmark KITTI, which\nincludes both dashboard camera video and lidar produced 3D ground truth. With\nthe KITTI data, our primary contribution is a quantitative evaluation of 3D\nreconstruction computational speed when considering live video.\n", "date": "2022-11-21", "categories": []}
{"id": "2211.11971", "title": "Multi-View Neural Surface Reconstruction with Structured Light", "abstract": "  Three-dimensional (3D) object reconstruction based on differentiable\nrendering (DR) is an active research topic in computer vision. DR-based methods\nminimize the difference between the rendered and target images by optimizing\nboth the shape and appearance and realizing a high visual reproductivity.\nHowever, most approaches perform poorly for textureless objects because of the\ngeometrical ambiguity, which means that multiple shapes can have the same\nrendered result in such objects. To overcome this problem, we introduce active\nsensing with structured light (SL) into multi-view 3D object reconstruction\nbased on DR to learn the unknown geometry and appearance of arbitrary scenes\nand camera poses. More specifically, our framework leverages the\ncorrespondences between pixels in different views calculated by structured\nlight as an additional constraint in the DR-based optimization of implicit\nsurface, color representations, and camera poses. Because camera poses can be\noptimized simultaneously, our method realizes high reconstruction accuracy in\nthe textureless region and reduces efforts for camera pose calibration, which\nis required for conventional SL-based methods. Experiment results on both\nsynthetic and real data demonstrate that our system outperforms conventional\nDR- and SL-based methods in a high-quality surface reconstruction, particularly\nfor challenging objects with textureless or shiny surfaces.\n", "date": "2022-11-22", "categories": []}
{"id": "2206.10885", "title": "KiloNeuS: A Versatile Neural Implicit Surface Representation for\n  Real-Time Rendering", "abstract": "  NeRF-based techniques fit wide and deep multi-layer perceptrons (MLPs) to a\ncontinuous radiance field that can be rendered from any unseen viewpoint.\nHowever, the lack of surface and normals definition and high rendering times\nlimit their usage in typical computer graphics applications. Such limitations\nhave recently been overcome separately, but solving them together remains an\nopen problem. We present KiloNeuS, a neural representation reconstructing an\nimplicit surface represented as a signed distance function (SDF) from\nmulti-view images and enabling real-time rendering by partitioning the space\ninto thousands of tiny MLPs fast to inference. As we learn the implicit surface\nlocally using independent models, resulting in a globally coherent geometry is\nnon-trivial and needs to be addressed during training. We evaluate rendering\nperformance on a GPU-accelerated ray-caster with in-shader neural network\ninference, resulting in an average of 46 FPS at high resolution, proving a\nsatisfying tradeoff between storage costs and rendering quality. In fact, our\nevaluation for rendering quality and surface recovery shows that KiloNeuS\noutperforms its single-MLP counterpart. Finally, to exhibit the versatility of\nKiloNeuS, we integrate it into an interactive path-tracer taking full advantage\nof its surface normals. We consider our work a crucial first step toward\nreal-time rendering of implicit neural representations under global\nillumination.\n", "date": "2022-06-22", "categories": []}
{"id": "2211.11320", "title": "Recovering Fine Details for Neural Implicit Surface Reconstruction", "abstract": "  Recent works on implicit neural representations have made significant\nstrides. Learning implicit neural surfaces using volume rendering has gained\npopularity in multi-view reconstruction without 3D supervision. However,\naccurately recovering fine details is still challenging, due to the underlying\nambiguity of geometry and appearance representation. In this paper, we present\nD-NeuS, a volume rendering-base neural implicit surface reconstruction method\ncapable to recover fine geometry details, which extends NeuS by two additional\nloss functions targeting enhanced reconstruction quality. First, we encourage\nthe rendered surface points from alpha compositing to have zero signed distance\nvalues, alleviating the geometry bias arising from transforming SDF to density\nfor volume rendering. Second, we impose multi-view feature consistency on the\nsurface points, derived by interpolating SDF zero-crossings from sampled points\nalong rays. Extensive quantitative and qualitative results demonstrate that our\nmethod reconstructs high-accuracy surfaces with details, and outperforms the\nstate of the art.\n", "date": "2022-11-21", "categories": []}
{"id": "2208.06787", "title": "HDR-Plenoxels: Self-Calibrating High Dynamic Range Radiance Fields", "abstract": "  We propose high dynamic range (HDR) radiance fields, HDR-Plenoxels, that\nlearn a plenoptic function of 3D HDR radiance fields, geometry information, and\nvarying camera settings inherent in 2D low dynamic range (LDR) images. Our\nvoxel-based volume rendering pipeline reconstructs HDR radiance fields with\nonly multi-view LDR images taken from varying camera settings in an end-to-end\nmanner and has a fast convergence speed. To deal with various cameras in\nreal-world scenarios, we introduce a tone mapping module that models the\ndigital in-camera imaging pipeline (ISP) and disentangles radiometric settings.\nOur tone mapping module allows us to render by controlling the radiometric\nsettings of each novel view. Finally, we build a multi-view dataset with\nvarying camera conditions, which fits our problem setting. Our experiments show\nthat HDR-Plenoxels can express detail and high-quality HDR novel views from\nonly LDR images with various cameras.\n", "date": "2022-08-14", "categories": []}
{"id": "2211.09682", "title": "AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware\n  Training", "abstract": "  Neural Radiance Fields (NeRFs) are a powerful representation for modeling a\n3D scene as a continuous function. Though NeRF is able to render complex 3D\nscenes with view-dependent effects, few efforts have been devoted to exploring\nits limits in a high-resolution setting. Specifically, existing NeRF-based\nmethods face several limitations when reconstructing high-resolution real\nscenes, including a very large number of parameters, misaligned input data, and\noverly smooth details. In this work, we conduct the first pilot study on\ntraining NeRF with high-resolution data and propose the corresponding\nsolutions: 1) marrying the multilayer perceptron (MLP) with convolutional\nlayers which can encode more neighborhood information while reducing the total\nnumber of parameters; 2) a novel training strategy to address misalignment\ncaused by moving objects or small camera calibration errors; and 3) a\nhigh-frequency aware loss. Our approach is nearly free without introducing\nobvious training/testing costs, while experiments on different datasets\ndemonstrate that it can recover more high-frequency details compared with the\ncurrent state-of-the-art NeRF models. Project page:\n\\url{https://yifanjiang.net/alignerf.}\n", "date": "2022-11-17", "categories": []}
{"id": "2211.04753", "title": "ReFu: Refine and Fuse the Unobserved View for Detail-Preserving\n  Single-Image 3D Human Reconstruction", "abstract": "  Single-image 3D human reconstruction aims to reconstruct the 3D textured\nsurface of the human body given a single image. While implicit function-based\nmethods recently achieved reasonable reconstruction performance, they still\nbear limitations showing degraded quality in both surface geometry and texture\nfrom an unobserved view. In response, to generate a realistic textured surface,\nwe propose ReFu, a coarse-to-fine approach that refines the projected backside\nview image and fuses the refined image to predict the final human body. To\nsuppress the diffused occupancy that causes noise in projection images and\nreconstructed meshes, we propose to train occupancy probability by\nsimultaneously utilizing 2D and 3D supervisions with occupancy-based volume\nrendering. We also introduce a refinement architecture that generates\ndetail-preserving backside-view images with front-to-back warping. Extensive\nexperiments demonstrate that our method achieves state-of-the-art performance\nin 3D human reconstruction from a single image, showing enhanced geometry and\ntexture quality from an unobserved view.\n", "date": "2022-11-09", "categories": []}
{"id": "2208.11948", "title": "Learning to Construct 3D Building Wireframes from 3D Line Clouds", "abstract": "  Line clouds, though under-investigated in the previous work, potentially\nencode more compact structural information of buildings than point clouds\nextracted from multi-view images. In this work, we propose the first network to\nprocess line clouds for building wireframe abstraction. The network takes a\nline cloud as input , i.e., a nonstructural and unordered set of 3D line\nsegments extracted from multi-view images, and outputs a 3D wireframe of the\nunderlying building, which consists of a sparse set of 3D junctions connected\nby line segments. We observe that a line patch, i.e., a group of neighboring\nline segments, encodes sufficient contour information to predict the existence\nand even the 3D position of a potential junction, as well as the likelihood of\nconnectivity between two query junctions. We therefore introduce a two-layer\nLine-Patch Transformer to extract junctions and connectivities from sampled\nline patches to form a 3D building wireframe model. We also introduce a\nsynthetic dataset of multi-view images with ground-truth 3D wireframe. We\nextensively justify that our reconstructed 3D wireframe models significantly\nimprove upon multiple baseline building reconstruction methods. The code and\ndata can be found at https://github.com/Luo1Cheng/LC2WF.\n", "date": "2022-08-25", "categories": []}
{"id": "2210.12202", "title": "High-Quality RGB-D Reconstruction via Multi-View Uncalibrated\n  Photometric Stereo and Gradient-SDF", "abstract": "  Fine-detailed reconstructions are in high demand in many applications.\nHowever, most of the existing RGB-D reconstruction methods rely on\npre-calculated accurate camera poses to recover the detailed surface geometry,\nwhere the representation of a surface needs to be adapted when optimizing\ndifferent quantities. In this paper, we present a novel multi-view RGB-D based\nreconstruction method that tackles camera pose, lighting, albedo, and surface\nnormal estimation via the utilization of a gradient signed distance field\n(gradient-SDF). The proposed method formulates the image rendering process\nusing specific physically-based model(s) and optimizes the surface's quantities\non the actual surface using its volumetric representation, as opposed to other\nworks which estimate surface quantities only near the actual surface. To\nvalidate our method, we investigate two physically-based image formation models\nfor natural light and point light source applications. The experimental results\non synthetic and real-world datasets demonstrate that the proposed method can\nrecover high-quality geometry of the surface more faithfully than the\nstate-of-the-art and further improves the accuracy of estimated camera poses.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.12268", "title": "An Exploration of Neural Radiance Field Scene Reconstruction: Synthetic,\n  Real-world and Dynamic Scenes", "abstract": "  This project presents an exploration into 3D scene reconstruction of\nsynthetic and real-world scenes using Neural Radiance Field (NeRF) approaches.\nWe primarily take advantage of the reduction in training and rendering time of\nneural graphic primitives multi-resolution hash encoding, to reconstruct static\nvideo game scenes and real-world scenes, comparing and observing reconstruction\ndetail and limitations. Additionally, we explore dynamic scene reconstruction\nusing Neural Radiance Fields for Dynamic Scenes(D-NeRF). Finally, we extend the\nimplementation of D-NeRF, originally constrained to handle synthetic scenes to\nalso handle real-world dynamic scenes.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.13041", "title": "Learning Neural Radiance Fields from Multi-View Geometry", "abstract": "  We present a framework, called MVG-NeRF, that combines classical Multi-View\nGeometry algorithms and Neural Radiance Fields (NeRF) for image-based 3D\nreconstruction. NeRF has revolutionized the field of implicit 3D\nrepresentations, mainly due to a differentiable volumetric rendering\nformulation that enables high-quality and geometry-aware novel view synthesis.\nHowever, the underlying geometry of the scene is not explicitly constrained\nduring training, thus leading to noisy and incorrect results when extracting a\nmesh with marching cubes. To this end, we propose to leverage pixelwise depths\nand normals from a classical 3D reconstruction pipeline as geometric priors to\nguide NeRF optimization. Such priors are used as pseudo-ground truth during\ntraining in order to improve the quality of the estimated underlying surface.\nMoreover, each pixel is weighted by a confidence value based on the\nforward-backward reprojection error for additional robustness. Experimental\nresults on real-world data demonstrate the effectiveness of this approach in\nobtaining clean 3D meshes from images, while maintaining competitive\nperformances in novel view synthesis.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.11170", "title": "Coordinates Are NOT Lonely -- Codebook Prior Helps Implicit Neural 3D\n  Representations", "abstract": "  Implicit neural 3D representation has achieved impressive results in surface\nor scene reconstruction and novel view synthesis, which typically uses the\ncoordinate-based multi-layer perceptrons (MLPs) to learn a continuous scene\nrepresentation. However, existing approaches, such as Neural Radiance Field\n(NeRF) and its variants, usually require dense input views (i.e. 50-150) to\nobtain decent results. To relive the over-dependence on massive calibrated\nimages and enrich the coordinate-based feature representation, we explore\ninjecting the prior information into the coordinate-based network and introduce\na novel coordinate-based model, CoCo-INR, for implicit neural 3D\nrepresentation. The cores of our method are two attention modules: codebook\nattention and coordinate attention. The former extracts the useful prototypes\ncontaining rich geometry and appearance information from the prior codebook,\nand the latter propagates such prior information into each coordinate and\nenriches its feature representation for a scene or object surface. With the\nhelp of the prior information, our method can render 3D views with more\nphoto-realistic appearance and geometries than the current methods using fewer\ncalibrated images available. Experiments on various scene reconstruction\ndatasets, including DTU and BlendedMVS, and the full 3D head reconstruction\ndataset, H3DS, demonstrate the robustness under fewer input views and fine\ndetail-preserving capability of our proposed method.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.11419", "title": "GPR-Net: Multi-view Layout Estimation via a Geometry-aware Panorama\n  Registration Network", "abstract": "  Reconstructing 3D layouts from multiple $360^{\\circ}$ panoramas has received\nincreasing attention recently as estimating a complete layout of a large-scale\nand complex room from a single panorama is very difficult. The state-of-the-art\nmethod, called PSMNet, introduces the first learning-based framework that\njointly estimates the room layout and registration given a pair of panoramas.\nHowever, PSMNet relies on an approximate (i.e., \"noisy\") registration as input.\nObtaining this input requires a solution for wide baseline registration which\nis a challenging problem. In this work, we present a complete multi-view\npanoramic layout estimation framework that jointly learns panorama registration\nand layout estimation given a pair of panoramas without relying on a pose\nprior. The major improvement over PSMNet comes from a novel Geometry-aware\nPanorama Registration Network or GPR-Net that effectively tackles the wide\nbaseline registration problem by exploiting the layout geometry and computing\nfine-grained correspondences on the layout boundaries, instead of the global\npixel-space. Our architecture consists of two parts. First, given two\npanoramas, we adopt a vision transformer to learn a set of 1D horizon features\nsampled on the panorama. These 1D horizon features encode the depths of\nindividual layout boundary samples and the correspondence and covisibility maps\nbetween layout boundaries. We then exploit a non-linear registration module to\nconvert these 1D horizon features into a set of corresponding 2D boundary\npoints on the layout. Finally, we estimate the final relative camera pose via\nRANSAC and obtain the complete layout simply by taking the union of registered\nlayouts. Experimental results indicate that our method achieves\nstate-of-the-art performance in both panorama registration and layout\nestimation on a large-scale indoor panorama dataset ZInD.\n", "date": "2022-10-20", "categories": []}
{"id": "2207.00182", "title": "Recovering Detail in 3D Shapes Using Disparity Maps", "abstract": "  We present a fine-tuning method to improve the appearance of 3D geometries\nreconstructed from single images. We leverage advances in monocular depth\nestimation to obtain disparity maps and present a novel approach to\ntransforming 2D normalized disparity maps into 3D point clouds by using shape\npriors to solve an optimization on the relevant camera parameters. After\ncreating a 3D point cloud from disparity, we introduce a method to combine the\nnew point cloud with existing information to form a more faithful and detailed\nfinal geometry. We demonstrate the efficacy of our approach with multiple\nexperiments on both synthetic and real images.\n", "date": "2022-07-01", "categories": []}
{"id": "2206.13597", "title": "NeuRIS: Neural Reconstruction of Indoor Scenes Using Normal Priors", "abstract": "  Reconstructing 3D indoor scenes from 2D images is an important task in many\ncomputer vision and graphics applications. A main challenge in this task is\nthat large texture-less areas in typical indoor scenes make existing methods\nstruggle to produce satisfactory reconstruction results. We propose a new\nmethod, named NeuRIS, for high quality reconstruction of indoor scenes. The key\nidea of NeuRIS is to integrate estimated normal of indoor scenes as a prior in\na neural rendering framework for reconstructing large texture-less shapes and,\nimportantly, to do this in an adaptive manner to also enable the reconstruction\nof irregular shapes with fine details. Specifically, we evaluate the\nfaithfulness of the normal priors on-the-fly by checking the multi-view\nconsistency of reconstruction during the optimization process. Only the normal\npriors accepted as faithful will be utilized for 3D reconstruction, which\ntypically happens in the regions of smooth shapes possibly with weak texture.\nHowever, for those regions with small objects or thin structures, for which the\nnormal priors are usually unreliable, we will only rely on visual features of\nthe input images, since such regions typically contain relatively rich visual\nfeatures (e.g., shade changes and boundary contours). Extensive experiments\nshow that NeuRIS significantly outperforms the state-of-the-art methods in\nterms of reconstruction quality.\n", "date": "2022-06-27", "categories": []}
{"id": "2210.01276", "title": "Probabilistic Volumetric Fusion for Dense Monocular SLAM", "abstract": "  We present a novel method to reconstruct 3D scenes from images by leveraging\ndeep dense monocular SLAM and fast uncertainty propagation. The proposed\napproach is able to 3D reconstruct scenes densely, accurately, and in real-time\nwhile being robust to extremely noisy depth estimates coming from dense\nmonocular SLAM. Differently from previous approaches, that either use ad-hoc\ndepth filters, or that estimate the depth uncertainty from RGB-D cameras'\nsensor models, our probabilistic depth uncertainty derives directly from the\ninformation matrix of the underlying bundle adjustment problem in SLAM. We show\nthat the resulting depth uncertainty provides an excellent signal to weight the\ndepth-maps for volumetric fusion. Without our depth uncertainty, the resulting\nmesh is noisy and with artifacts, while our approach generates an accurate 3D\nmesh with significantly fewer artifacts. We provide results on the challenging\nEuroc dataset, and show that our approach achieves 92% better accuracy than\ndirectly fusing depths from monocular SLAM, and up to 90% improvements compared\nto the best competing approach.\n", "date": "2022-10-03", "categories": []}
{"id": "2210.08936", "title": "S$^3$-NeRF: Neural Reflectance Field from Shading and Shadow under a\n  Single Viewpoint", "abstract": "  In this paper, we address the \"dual problem\" of multi-view scene\nreconstruction in which we utilize single-view images captured under different\npoint lights to learn a neural scene representation. Different from existing\nsingle-view methods which can only recover a 2.5D scene representation (i.e., a\nnormal / depth map for the visible surface), our method learns a neural\nreflectance field to represent the 3D geometry and BRDFs of a scene. Instead of\nrelying on multi-view photo-consistency, our method exploits two\ninformation-rich monocular cues, namely shading and shadow, to infer scene\ngeometry. Experiments on multiple challenging datasets show that our method is\ncapable of recovering 3D geometry, including both visible and invisible parts,\nof a scene from single-view images. Thanks to the neural reflectance field\nrepresentation, our method is robust to depth discontinuities. It supports\napplications like novel-view synthesis and relighting. Our code and model can\nbe found at https://ywq.github.io/s3nerf.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.09148", "title": "Pruning-based Topology Refinement of 3D Mesh using a 2D Alpha Mask", "abstract": "  Image-based 3D reconstruction has increasingly stunning results over the past\nfew years with the latest improvements in computer vision and graphics.\nGeometry and topology are two fundamental concepts when dealing with 3D mesh\nstructures. But the latest often remains a side issue in the 3D mesh-based\nreconstruction literature. Indeed, performing per-vertex elementary\ndisplacements over a 3D sphere mesh only impacts its geometry and leaves the\ntopological structure unchanged and fixed. Whereas few attempts propose to\nupdate the geometry and the topology, all need to lean on costly 3D\nground-truth to determine the faces/edges to prune. We present in this work a\nmethod that aims to refine the topology of any 3D mesh through a face-pruning\nstrategy that extensively relies upon 2D alpha masks and camera pose\ninformation. Our solution leverages a differentiable renderer that renders each\nface as a 2D soft map. Its pixel intensity reflects the probability of being\ncovered during the rendering process by such a face. Based on the 2D soft-masks\navailable, our method is thus able to quickly highlight all the incorrectly\nrendered faces for a given viewpoint. Because our module is agnostic to the\nnetwork that produces the 3D mesh, it can be easily plugged into any\nself-supervised image-based (either synthetic or natural) 3D reconstruction\npipeline to get complex meshes with a non-spherical topology.\n", "date": "2022-10-17", "categories": []}
{"id": "2206.15258", "title": "Neural Surface Reconstruction of Dynamic Scenes with Monocular RGB-D\n  Camera", "abstract": "  We propose Neural-DynamicReconstruction (NDR), a template-free method to\nrecover high-fidelity geometry and motions of a dynamic scene from a monocular\nRGB-D camera. In NDR, we adopt the neural implicit function for surface\nrepresentation and rendering such that the captured color and depth can be\nfully utilized to jointly optimize the surface and deformations. To represent\nand constrain the non-rigid deformations, we propose a novel neural invertible\ndeforming network such that the cycle consistency between arbitrary two frames\nis automatically satisfied. Considering that the surface topology of dynamic\nscene might change over time, we employ a topology-aware strategy to construct\nthe topology-variant correspondence for the fused frames. NDR also further\nrefines the camera poses in a global optimization manner. Experiments on public\ndatasets and our collected dataset demonstrate that NDR outperforms existing\nmonocular dynamic reconstruction methods.\n", "date": "2022-06-30", "categories": []}
{"id": "2206.00665", "title": "MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface\n  Reconstruction", "abstract": "  In recent years, neural implicit surface reconstruction methods have become\npopular for multi-view 3D reconstruction. In contrast to traditional multi-view\nstereo methods, these approaches tend to produce smoother and more complete\nreconstructions due to the inductive smoothness bias of neural networks.\nState-of-the-art neural implicit methods allow for high-quality reconstructions\nof simple scenes from many input views. Yet, their performance drops\nsignificantly for larger and more complex scenes and scenes captured from\nsparse viewpoints. This is caused primarily by the inherent ambiguity in the\nRGB reconstruction loss that does not provide enough constraints, in particular\nin less-observed and textureless areas. Motivated by recent advances in the\narea of monocular geometry prediction, we systematically explore the utility\nthese cues provide for improving neural implicit surface reconstruction. We\ndemonstrate that depth and normal cues, predicted by general-purpose monocular\nestimators, significantly improve reconstruction quality and optimization time.\nFurther, we analyse and investigate multiple design choices for representing\nneural implicit surfaces, ranging from monolithic MLP models over single-grid\nto multi-resolution grid representations. We observe that geometric monocular\npriors improve performance both for small-scale single-object as well as\nlarge-scale multi-object scenes, independent of the choice of representation.\n", "date": "2022-06-01", "categories": []}
{"id": "2210.05891", "title": "Point Cloud Scene Completion with Joint Color and Semantic Estimation\n  from Single RGB-D Image", "abstract": "  We present a deep reinforcement learning method of progressive view\ninpainting for colored semantic point cloud scene completion under volume\nguidance, achieving high-quality scene reconstruction from only a single RGB-D\nimage with severe occlusion. Our approach is end-to-end, consisting of three\nmodules: 3D scene volume reconstruction, 2D RGB-D and segmentation image\ninpainting, and multi-view selection for completion. Given a single RGB-D\nimage, our method first predicts its semantic segmentation map and goes through\nthe 3D volume branch to obtain a volumetric scene reconstruction as a guide to\nthe next view inpainting step, which attempts to make up the missing\ninformation; the third step involves projecting the volume under the same view\nof the input, concatenating them to complete the current view RGB-D and\nsegmentation map, and integrating all RGB-D and segmentation maps into the\npoint cloud. Since the occluded areas are unavailable, we resort to a A3C\nnetwork to glance around and pick the next best view for large hole completion\nprogressively until a scene is adequately reconstructed while guaranteeing\nvalidity. All steps are learned jointly to achieve robust and consistent\nresults. We perform qualitative and quantitative evaluations with extensive\nexperiments on the 3D-FUTURE data, obtaining better results than\nstate-of-the-arts.\n", "date": "2022-10-12", "categories": []}
{"id": "2210.03815", "title": "Scene-level Tracking and Reconstruction without Object Priors", "abstract": "  We present the first real-time system capable of tracking and reconstructing,\nindividually, every visible object in a given scene, without any form of prior\non the rigidness of the objects, texture existence, or object category. In\ncontrast with previous methods such as Co-Fusion and MaskFusion that first\nsegment the scene into individual objects and then process each object\nindependently, the proposed method dynamically segments the non-rigid scene as\npart of the tracking and reconstruction process. When new measurements indicate\ntopology change, reconstructed models are updated in real-time to reflect that\nchange. Our proposed system can provide the live geometry and deformation of\nall visible objects in a novel scene in real-time, which makes it possible to\nbe integrated seamlessly into numerous existing robotics applications that rely\non object models for grasping and manipulation. The capabilities of the\nproposed system are demonstrated in challenging scenes that contain multiple\nrigid and non-rigid objects.\n", "date": "2022-10-07", "categories": []}
{"id": "2210.04265", "title": "3D Reconstruction of Sculptures from Single Images via Unsupervised\n  Domain Adaptation on Implicit Models", "abstract": "  Acquiring the virtual equivalent of exhibits, such as sculptures, in virtual\nreality (VR) museums, can be labour-intensive and sometimes infeasible. Deep\nlearning based 3D reconstruction approaches allow us to recover 3D shapes from\n2D observations, among which single-view-based approaches can reduce the need\nfor human intervention and specialised equipment in acquiring 3D sculptures for\nVR museums. However, there exist two challenges when attempting to use the\nwell-researched human reconstruction methods: limited data availability and\ndomain shift. Considering sculptures are usually related to humans, we propose\nour unsupervised 3D domain adaptation method for adapting a single-view 3D\nimplicit reconstruction model from the source (real-world humans) to the target\n(sculptures) domain. We have compared the generated shapes with other methods\nand conducted ablation studies as well as a user study to demonstrate the\neffectiveness of our adaptation method. We also deploy our results in a VR\napplication.\n", "date": "2022-10-09", "categories": []}
{"id": "2210.04572", "title": "Floorplan-Aware Camera Poses Refinement", "abstract": "  Processing large indoor scenes is a challenging task, as scan registration\nand camera trajectory estimation methods accumulate errors across time. As a\nresult, the quality of reconstructed scans is insufficient for some\napplications, such as visual-based localization and navigation, where the\ncorrect position of walls is crucial.\n  For many indoor scenes, there exists an image of a technical floorplan that\ncontains information about the geometry and main structural elements of the\nscene, such as walls, partitions, and doors. We argue that such a floorplan is\na useful source of spatial information, which can guide a 3D model\noptimization.\n  The standard RGB-D 3D reconstruction pipeline consists of a tracking module\napplied to an RGB-D sequence and a bundle adjustment (BA) module that takes the\nposed RGB-D sequence and corrects the camera poses to improve consistency. We\npropose a novel optimization algorithm expanding conventional BA that leverages\nthe prior knowledge about the scene structure in the form of a floorplan. Our\nexperiments on the Redwood dataset and our self-captured data demonstrate that\nutilizing floorplan improves accuracy of 3D reconstructions.\n", "date": "2022-10-10", "categories": []}
{"id": "2210.01548", "title": "Neural Implicit Surface Reconstruction from Noisy Camera Observations", "abstract": "  Representing 3D objects and scenes with neural radiance fields has become\nvery popular over the last years. Recently, surface-based representations have\nbeen proposed, that allow to reconstruct 3D objects from simple photographs.\nHowever, most current techniques require an accurate camera calibration, i.e.\ncamera parameters corresponding to each image, which is often a difficult task\nto do in real-life situations. To this end, we propose a method for learning 3D\nsurfaces from noisy camera parameters. We show that we can learn camera\nparameters together with learning the surface representation, and demonstrate\ngood quality 3D surface reconstruction even with noisy camera observations.\n", "date": "2022-10-02", "categories": []}
{"id": "2209.14265", "title": "360FusionNeRF: Panoramic Neural Radiance Fields with Joint Guidance", "abstract": "  We present a method to synthesize novel views from a single $360^\\circ$\npanorama image based on the neural radiance field (NeRF). Prior studies in a\nsimilar setting rely on the neighborhood interpolation capability of\nmulti-layer perceptions to complete missing regions caused by occlusion, which\nleads to artifacts in their predictions. We propose 360FusionNeRF, a\nsemi-supervised learning framework where we introduce geometric supervision and\nsemantic consistency to guide the progressive training process. Firstly, the\ninput image is re-projected to $360^\\circ$ images, and auxiliary depth maps are\nextracted at other camera positions. The depth supervision, in addition to the\nNeRF color guidance, improves the geometry of the synthesized views.\nAdditionally, we introduce a semantic consistency loss that encourages\nrealistic renderings of novel views. We extract these semantic features using a\npre-trained visual encoder such as CLIP, a Vision Transformer trained on\nhundreds of millions of diverse 2D photographs mined from the web with natural\nlanguage supervision. Experiments indicate that our proposed method can produce\nplausible completions of unobserved regions while preserving the features of\nthe scene. When trained across various scenes, 360FusionNeRF consistently\nachieves the state-of-the-art performance when transferring to synthetic\nStructured3D dataset (PSNR~5%, SSIM~3% LPIPS~13%), real-world Matterport3D\ndataset (PSNR~3%, SSIM~3% LPIPS~9%) and Replica360 dataset (PSNR~8%, SSIM~2%\nLPIPS~18%).\n", "date": "2022-09-28", "categories": []}
{"id": "2210.00978", "title": "Uncertainty-Driven Active Vision for Implicit Scene Reconstruction", "abstract": "  Multi-view implicit scene reconstruction methods have become increasingly\npopular due to their ability to represent complex scene details. Recent efforts\nhave been devoted to improving the representation of input information and to\nreducing the number of views required to obtain high quality reconstructions.\nYet, perhaps surprisingly, the study of which views to select to maximally\nimprove scene understanding remains largely unexplored. We propose an\nuncertainty-driven active vision approach for implicit scene reconstruction,\nwhich leverages occupancy uncertainty accumulated across the scene using volume\nrendering to select the next view to acquire. To this end, we develop an\noccupancy-based reconstruction method which accurately represents scenes using\neither 2D or 3D supervision. We evaluate our proposed approach on the ABC\ndataset and the in the wild CO3D dataset, and show that: (1) we are able to\nobtain high quality state-of-the-art occupancy reconstructions; (2) our\nperspective conditioned uncertainty definition is effective to drive\nimprovements in next best view selection and outperforms strong baseline\napproaches; and (3) we can further improve shape understanding by performing a\ngradient-based search on the view selection candidates. Overall, our results\nhighlight the importance of view selection for implicit scene reconstruction,\nmaking it a promising avenue to explore further.\n", "date": "2022-10-03", "categories": []}
{"id": "2209.15153", "title": "MonoNeuralFusion: Online Monocular Neural 3D Reconstruction with\n  Geometric Priors", "abstract": "  High-fidelity 3D scene reconstruction from monocular videos continues to be\nchallenging, especially for complete and fine-grained geometry reconstruction.\nThe previous 3D reconstruction approaches with neural implicit representations\nhave shown a promising ability for complete scene reconstruction, while their\nresults are often over-smooth and lack enough geometric details. This paper\nintroduces a novel neural implicit scene representation with volume rendering\nfor high-fidelity online 3D scene reconstruction from monocular videos. For\nfine-grained reconstruction, our key insight is to incorporate geometric priors\ninto both the neural implicit scene representation and neural volume rendering,\nthus leading to an effective geometry learning mechanism based on volume\nrendering optimization. Benefiting from this, we present MonoNeuralFusion to\nperform the online neural 3D reconstruction from monocular videos, by which the\n3D scene geometry is efficiently generated and optimized during the on-the-fly\n3D monocular scanning. The extensive comparisons with state-of-the-art\napproaches show that our MonoNeuralFusion consistently generates much better\ncomplete and fine-grained reconstruction results, both quantitatively and\nqualitatively.\n", "date": "2022-09-30", "categories": []}
{"id": "2209.15383", "title": "Semi-Supervised Single-View 3D Reconstruction via Prototype Shape Priors", "abstract": "  The performance of existing single-view 3D reconstruction methods heavily\nrelies on large-scale 3D annotations. However, such annotations are tedious and\nexpensive to collect. Semi-supervised learning serves as an alternative way to\nmitigate the need for manual labels, but remains unexplored in 3D\nreconstruction. Inspired by the recent success of semi-supervised image\nclassification tasks, we propose SSP3D, a semi-supervised framework for 3D\nreconstruction. In particular, we introduce an attention-guided prototype shape\nprior module for guiding realistic object reconstruction. We further introduce\na discriminator-guided module to incentivize better shape generation, as well\nas a regularizer to tolerate noisy training samples. On the ShapeNet benchmark,\nthe proposed approach outperforms previous supervised methods by clear margins\nunder various labeling ratios, (i.e., 1%, 5% , 10% and 20%). Moreover, our\napproach also performs well when transferring to real-world Pix3D datasets\nunder labeling ratios of 10%. We also demonstrate our method could transfer to\nnovel categories with few novel supervised data. Experiments on the popular\nShapeNet dataset show that our method outperforms the zero-shot baseline by\nover 12% and we also perform rigorous ablations and analysis to validate our\napproach.\n", "date": "2022-09-30", "categories": []}
{"id": "2209.13433", "title": "OmniNeRF: Hybriding Omnidirectional Distance and Radiance fields for\n  Neural Surface Reconstruction", "abstract": "  3D reconstruction from images has wide applications in Virtual Reality and\nAutomatic Driving, where the precision requirement is very high.\nGround-breaking research in the neural radiance field (NeRF) by utilizing\nMulti-Layer Perceptions has dramatically improved the representation quality of\n3D objects. Some later studies improved NeRF by building truncated signed\ndistance fields (TSDFs) but still suffer from the problem of blurred surfaces\nin 3D reconstruction. In this work, this surface ambiguity is addressed by\nproposing a novel way of 3D shape representation, OmniNeRF. It is based on\ntraining a hybrid implicit field of Omni-directional Distance Field (ODF) and\nneural radiance field, replacing the apparent density in NeRF with\nomnidirectional information. Moreover, we introduce additional supervision on\nthe depth map to further improve reconstruction quality. The proposed method\nhas been proven to effectively deal with NeRF defects at the edges of the\nsurface reconstruction, providing higher quality 3D scene reconstruction\nresults.\n", "date": "2022-09-27", "categories": []}
{"id": "2209.11677", "title": "PNeRF: Probabilistic Neural Scene Representations for Uncertain 3D\n  Visual Mapping", "abstract": "  Recently neural scene representations have provided very impressive results\nfor representing 3D scenes visually, however, their study and progress have\nmainly been limited to visualization of virtual models in computer graphics or\nscene reconstruction in computer vision without explicitly accounting for\nsensor and pose uncertainty. Using this novel scene representation in robotics\napplications, however, would require accounting for this uncertainty in the\nneural map. The aim of this paper is therefore to propose a novel method for\ntraining {\\em probabilistic neural scene representations} with uncertain\ntraining data that could enable the inclusion of these representations in\nrobotics applications. Acquiring images using cameras or depth sensors contains\ninherent uncertainty, and furthermore, the camera poses used for learning a 3D\nmodel are also imperfect. If these measurements are used for training without\naccounting for their uncertainty, then the resulting models are non-optimal,\nand the resulting scene representations are likely to contain artifacts such as\nblur and un-even geometry. In this work, the problem of uncertainty integration\nto the learning process is investigated by focusing on training with uncertain\ninformation in a probabilistic manner. The proposed method involves explicitly\naugmenting the training likelihood with an uncertainty term such that the\nlearnt probability distribution of the network is minimized with respect to the\ntraining uncertainty. It will be shown that this leads to more accurate image\nrendering quality, in addition to more precise and consistent geometry.\nValidation has been carried out on both synthetic and real datasets showing\nthat the proposed approach outperforms state-of-the-art methods. The results\nshow notably that the proposed method is capable of rendering novel\nhigh-quality views even when the training data is limited.\n", "date": "2022-09-23", "categories": []}
{"id": "2206.07850", "title": "HF-NeuS: Improved Surface Reconstruction Using High-Frequency Details", "abstract": "  Neural rendering can be used to reconstruct implicit representations of\nshapes without 3D supervision. However, current neural surface reconstruction\nmethods have difficulty learning high-frequency geometry details, so the\nreconstructed shapes are often over-smoothed. We develop HF-NeuS, a novel\nmethod to improve the quality of surface reconstruction in neural rendering. We\nfollow recent work to model surfaces as signed distance functions (SDFs).\nFirst, we offer a derivation to analyze the relationship between the SDF, the\nvolume density, the transparency function, and the weighting function used in\nthe volume rendering equation and propose to model transparency as transformed\nSDF. Second, we observe that attempting to jointly encode high-frequency and\nlow-frequency components in a single SDF leads to unstable optimization. We\npropose to decompose the SDF into a base function and a displacement function\nwith a coarse-to-fine strategy to gradually increase the high-frequency\ndetails. Finally, we design an adaptive optimization strategy that makes the\ntraining process focus on improving those regions near the surface where the\nSDFs have artifacts. Our qualitative and quantitative results show that our\nmethod can reconstruct fine-grained surface details and obtain better surface\nreconstruction quality than the current state of the art. Code available at\nhttps://github.com/yiqun-wang/HFS.\n", "date": "2022-06-15", "categories": []}
{"id": "2209.10029", "title": "Fast-Image2Point: Towards Real-Time Point Cloud Reconstruction of a\n  Single Image using 3D Supervision", "abstract": "  A key question in the problem of 3D reconstruction is how to train a machine\nor a robot to model 3D objects. Many tasks like navigation in real-time systems\nsuch as autonomous vehicles directly depend on this problem. These systems\nusually have limited computational power. Despite considerable progress in 3D\nreconstruction systems in recent years, applying them to real-time systems such\nas navigation systems in autonomous vehicles is still challenging due to the\nhigh complexity and computational demand of the existing methods. This study\naddresses current problems in reconstructing objects displayed in a single-view\nimage in a faster (real-time) fashion. To this end, a simple yet powerful deep\nneural framework is developed. The proposed framework consists of two\ncomponents: the feature extractor module and the 3D generator module. We use\npoint cloud representation for the output of our reconstruction module. The\nShapeNet dataset is utilized to compare the method with the existing results in\nterms of computation time and accuracy. Simulations demonstrate the superior\nperformance of the proposed method.\n  Index Terms-Real-time 3D reconstruction, single-view reconstruction,\nsupervised learning, deep neural network\n", "date": "2022-09-20", "categories": []}
{"id": "2206.14735", "title": "GO-Surf: Neural Feature Grid Optimization for Fast, High-Fidelity RGB-D\n  Surface Reconstruction", "abstract": "  We present GO-Surf, a direct feature grid optimization method for accurate\nand fast surface reconstruction from RGB-D sequences. We model the underlying\nscene with a learned hierarchical feature voxel grid that encapsulates\nmulti-level geometric and appearance local information. Feature vectors are\ndirectly optimized such that after being tri-linearly interpolated, decoded by\ntwo shallow MLPs into signed distance and radiance values, and rendered via\nsurface volume rendering, the discrepancy between synthesized and observed\nRGB/depth values is minimized. Our supervision signals -- RGB, depth and\napproximate SDF -- can be obtained directly from input images without any need\nfor fusion or post-processing. We formulate a novel SDF gradient regularization\nterm that encourages surface smoothness and hole filling while maintaining high\nfrequency details. GO-Surf can optimize sequences of $1$-$2$K frames in\n$15$-$45$ minutes, a speedup of $\\times60$ over NeuralRGB-D, the most related\napproach based on an MLP representation, while maintaining on par performance\non standard benchmarks. Project page: https://jingwenwang95.github.io/go_surf/\n", "date": "2022-06-29", "categories": []}
{"id": "2209.08221", "title": "Neural Implicit Surface Reconstruction using Imaging Sonar", "abstract": "  We present a technique for dense 3D reconstruction of objects using an\nimaging sonar, also known as forward-looking sonar (FLS). Compared to previous\nmethods that model the scene geometry as point clouds or volumetric grids, we\nrepresent the geometry as a neural implicit function. Additionally, given such\na representation, we use a differentiable volumetric renderer that models the\npropagation of acoustic waves to synthesize imaging sonar measurements. We\nperform experiments on real and synthetic datasets and show that our algorithm\nreconstructs high-fidelity surface geometry from multi-view FLS images at much\nhigher quality than was possible with previous techniques and without suffering\nfrom their associated memory overhead.\n", "date": "2022-09-17", "categories": []}
{"id": "2209.08409", "title": "Uncertainty Guided Policy for Active Robotic 3D Reconstruction using\n  Neural Radiance Fields", "abstract": "  In this paper, we tackle the problem of active robotic 3D reconstruction of\nan object. In particular, we study how a mobile robot with an arm-held camera\ncan select a favorable number of views to recover an object's 3D shape\nefficiently. Contrary to the existing solution to this problem, we leverage the\npopular neural radiance fields-based object representation, which has recently\nshown impressive results for various computer vision tasks. However, it is not\nstraightforward to directly reason about an object's explicit 3D geometric\ndetails using such a representation, making the next-best-view selection\nproblem for dense 3D reconstruction challenging. This paper introduces a\nray-based volumetric uncertainty estimator, which computes the entropy of the\nweight distribution of the color samples along each ray of the object's\nimplicit neural representation. We show that it is possible to infer the\nuncertainty of the underlying 3D geometry given a novel view with the proposed\nestimator. We then present a next-best-view selection policy guided by the\nray-based volumetric uncertainty in neural radiance fields-based\nrepresentations. Encouraging experimental results on synthetic and real-world\ndata suggest that the approach presented in this paper can enable a new\nresearch direction of using an implicit 3D object representation for the\nnext-best-view problem in robot vision applications, distinguishing our\napproach from the existing approaches that rely on explicit 3D geometric\nmodeling.\n", "date": "2022-09-17", "categories": []}
{"id": "2209.08546", "title": "ActiveNeRF: Learning where to See with Uncertainty Estimation", "abstract": "  Recently, Neural Radiance Fields (NeRF) has shown promising performances on\nreconstructing 3D scenes and synthesizing novel views from a sparse set of 2D\nimages. Albeit effective, the performance of NeRF is highly influenced by the\nquality of training samples. With limited posed images from the scene, NeRF\nfails to generalize well to novel views and may collapse to trivial solutions\nin unobserved regions. This makes NeRF impractical under resource-constrained\nscenarios. In this paper, we present a novel learning framework, ActiveNeRF,\naiming to model a 3D scene with a constrained input budget. Specifically, we\nfirst incorporate uncertainty estimation into a NeRF model, which ensures\nrobustness under few observations and provides an interpretation of how NeRF\nunderstands the scene. On this basis, we propose to supplement the existing\ntraining set with newly captured samples based on an active learning scheme. By\nevaluating the reduction of uncertainty given new inputs, we select the samples\nthat bring the most information gain. In this way, the quality of novel view\nsynthesis can be improved with minimal additional resources. Extensive\nexperiments validate the performance of our model on both realistic and\nsynthetic scenes, especially with scarcer training data. Code will be released\nat \\url{https://github.com/LeapLabTHU/ActiveNeRF}.\n", "date": "2022-09-18", "categories": []}
{"id": "2209.06926", "title": "End-to-End Multi-View Structure-from-Motion with Hypercorrelation\n  Volumes", "abstract": "  Image-based 3D reconstruction is one of the most important tasks in Computer\nVision with many solutions proposed over the last few decades. The objective is\nto extract metric information i.e. the geometry of scene objects directly from\nimages. These can then be used in a wide range of applications such as film,\ngames, virtual reality, etc. Recently, deep learning techniques have been\nproposed to tackle this problem. They rely on training on vast amounts of data\nto learn to associate features between images through deep convolutional neural\nnetworks and have been shown to outperform traditional procedural techniques.\nIn this paper, we improve on the state-of-the-art two-view\nstructure-from-motion(SfM) approach of [11] by incorporating 4D correlation\nvolume for more accurate feature matching and reconstruction. Furthermore, we\nextend it to the general multi-view case and evaluate it on the complex\nbenchmark dataset DTU [4]. Quantitative evaluations and comparisons with\nstate-of-the-art multi-view 3D reconstruction methods demonstrate its\nsuperiority in terms of the accuracy of reconstructions.\n", "date": "2022-09-14", "categories": []}
{"id": "2208.07003", "title": "Adaptive Joint Optimization for 3D Reconstruction with Differentiable\n  Rendering", "abstract": "  Due to inevitable noises introduced during scanning and quantization, 3D\nreconstruction via RGB-D sensors suffers from errors both in geometry and\ntexture, leading to artifacts such as camera drifting, mesh distortion, texture\nghosting, and blurriness. Given an imperfect reconstructed 3D model, most\nprevious methods have focused on the refinement of either geometry, texture, or\ncamera pose. Or different optimization schemes and objectives for optimizing\neach component have been used in previous joint optimization methods, forming a\ncomplicated system. In this paper, we propose a novel optimization approach\nbased on differentiable rendering, which integrates the optimization of camera\npose, geometry, and texture into a unified framework by enforcing consistency\nbetween the rendered results and the corresponding RGB-D inputs. Based on the\nunified framework, we introduce a joint optimization approach to fully exploit\nthe inter-relationships between geometry, texture, and camera pose, and\ndescribe an adaptive interleaving strategy to improve optimization stability\nand efficiency. Using differentiable rendering, an image-level adversarial loss\nis applied to further improve the 3D model, making it more photorealistic.\nExperiments on synthetic and real data using quantitative and qualitative\nevaluation demonstrated the superiority of our approach in recovering both\nfine-scale geometry and high-fidelity texture.Code is available at\nhttps://adjointopti.github.io/adjoin.github.io/.\n", "date": "2022-08-15", "categories": []}
{"id": "2208.13241", "title": "Towards Accurate Reconstruction of 3D Scene Shape from A Single\n  Monocular Image", "abstract": "  Despite significant progress made in the past few years, challenges remain\nfor depth estimation using a single monocular image. First, it is nontrivial to\ntrain a metric-depth prediction model that can generalize well to diverse\nscenes mainly due to limited training data. Thus, researchers have built\nlarge-scale relative depth datasets that are much easier to collect. However,\nexisting relative depth estimation models often fail to recover accurate 3D\nscene shapes due to the unknown depth shift caused by training with the\nrelative depth data. We tackle this problem here and attempt to estimate\naccurate scene shapes by training on large-scale relative depth data, and\nestimating the depth shift. To do so, we propose a two-stage framework that\nfirst predicts depth up to an unknown scale and shift from a single monocular\nimage, and then exploits 3D point cloud data to predict the depth shift and the\ncamera's focal length that allow us to recover 3D scene shapes. As the two\nmodules are trained separately, we do not need strictly paired training data.\nIn addition, we propose an image-level normalized regression loss and a\nnormal-based geometry loss to improve training with relative depth annotation.\nWe test our depth model on nine unseen datasets and achieve state-of-the-art\nperformance on zero-shot evaluation. Code is available at: https://git.io/Depth\n", "date": "2022-08-28", "categories": []}
{"id": "2208.14743", "title": "SimpleRecon: 3D Reconstruction Without 3D Convolutions", "abstract": "  Traditionally, 3D indoor scene reconstruction from posed images happens in\ntwo phases: per-image depth estimation, followed by depth merging and surface\nreconstruction. Recently, a family of methods have emerged that perform\nreconstruction directly in final 3D volumetric feature space. While these\nmethods have shown impressive reconstruction results, they rely on expensive 3D\nconvolutional layers, limiting their application in resource-constrained\nenvironments. In this work, we instead go back to the traditional route, and\nshow how focusing on high quality multi-view depth prediction leads to highly\naccurate 3D reconstructions using simple off-the-shelf depth fusion. We propose\na simple state-of-the-art multi-view depth estimator with two main\ncontributions: 1) a carefully-designed 2D CNN which utilizes strong image\npriors alongside a plane-sweep feature volume and geometric losses, combined\nwith 2) the integration of keyframe and geometric metadata into the cost volume\nwhich allows informed depth plane scoring. Our method achieves a significant\nlead over the current state-of-the-art for depth estimation and close or better\nfor 3D reconstruction on ScanNet and 7-Scenes, yet still allows for online\nreal-time low-memory reconstruction. Code, models and results are available at\nhttps://nianticlabs.github.io/simplerecon\n", "date": "2022-08-31", "categories": []}
{"id": "2208.08728", "title": "Neural Capture of Animatable 3D Human from Monocular Video", "abstract": "  We present a novel paradigm of building an animatable 3D human representation\nfrom a monocular video input, such that it can be rendered in any unseen poses\nand views. Our method is based on a dynamic Neural Radiance Field (NeRF) rigged\nby a mesh-based parametric 3D human model serving as a geometry proxy. Previous\nmethods usually rely on multi-view videos or accurate 3D geometry information\nas additional inputs; besides, most methods suffer from degraded quality when\ngeneralized to unseen poses. We identify that the key to generalization is a\ngood input embedding for querying dynamic NeRF: A good input embedding should\ndefine an injective mapping in the full volumetric space, guided by surface\nmesh deformation under pose variation. Based on this observation, we propose to\nembed the input query with its relationship to local surface regions spanned by\na set of geodesic nearest neighbors on mesh vertices. By including both\nposition and relative distance information, our embedding defines a\ndistance-preserved deformation mapping and generalizes well to unseen poses. To\nreduce the dependency on additional inputs, we first initialize per-frame 3D\nmeshes using off-the-shelf tools and then propose a pipeline to jointly\noptimize NeRF and refine the initial mesh. Extensive experiments show our\nmethod can synthesize plausible human rendering results under unseen poses and\nviews.\n", "date": "2022-08-18", "categories": []}
{"id": "2208.08049", "title": "PDRF: Progressively Deblurring Radiance Field for Fast and Robust Scene\n  Reconstruction from Blurry Images", "abstract": "  We present Progressively Deblurring Radiance Field (PDRF), a novel approach\nto efficiently reconstruct high quality radiance fields from blurry images.\nWhile current State-of-The-Art (SoTA) scene reconstruction methods achieve\nphoto-realistic rendering results from clean source views, their performances\nsuffer when the source views are affected by blur, which is commonly observed\nfor images in the wild. Previous deblurring methods either do not account for\n3D geometry, or are computationally intense. To addresses these issues, PDRF, a\nprogressively deblurring scheme in radiance field modeling, accurately models\nblur by incorporating 3D scene context. PDRF further uses an efficient\nimportance sampling scheme, which results in fast scene optimization.\nSpecifically, PDRF proposes a Coarse Ray Renderer to quickly estimate voxel\ndensity and feature; a Fine Voxel Renderer is then used to achieve high quality\nray tracing. We perform extensive experiments and show that PDRF is 15X faster\nthan previous SoTA while achieving better performance on both synthetic and\nreal scenes.\n", "date": "2022-08-17", "categories": []}
{"id": "2208.06335", "title": "OmniVoxel: A Fast and Precise Reconstruction Method of Omnidirectional\n  Neural Radiance Field", "abstract": "  This paper proposes a method to reconstruct the neural radiance field with\nequirectangular omnidirectional images. Implicit neural scene representation\nwith a radiance field can reconstruct the 3D shape of a scene continuously\nwithin a limited spatial area. However, training a fully implicit\nrepresentation on commercial PC hardware requires a lot of time and computing\nresources (15 $\\sim$ 20 hours per scene). Therefore, we propose a method to\naccelerate this process significantly (20 $\\sim$ 40 minutes per scene). Instead\nof using a fully implicit representation of rays for radiance field\nreconstruction, we adopt feature voxels that contain density and color features\nin tensors. Considering omnidirectional equirectangular input and the camera\nlayout, we use spherical voxelization for representation instead of cubic\nrepresentation. Our voxelization method could balance the reconstruction\nquality of the inner scene and outer scene. In addition, we adopt the\naxis-aligned positional encoding method on the color features to increase the\ntotal image quality. Our method achieves satisfying empirical performance on\nsynthetic datasets with random camera poses. Moreover, we test our method with\nreal scenes which contain complex geometries and also achieve state-of-the-art\nperformance. Our code and complete dataset will be released at the same time as\nthe paper publication.\n", "date": "2022-08-12", "categories": []}
{"id": "2207.08656", "title": "Towards High-Fidelity Single-view Holistic Reconstruction of Indoor\n  Scenes", "abstract": "  We present a new framework to reconstruct holistic 3D indoor scenes including\nboth room background and indoor objects from single-view images. Existing\nmethods can only produce 3D shapes of indoor objects with limited geometry\nquality because of the heavy occlusion of indoor scenes. To solve this, we\npropose an instance-aligned implicit function (InstPIFu) for detailed object\nreconstruction. Combining with instance-aligned attention module, our method is\nempowered to decouple mixed local features toward the occluded instances.\nAdditionally, unlike previous methods that simply represents the room\nbackground as a 3D bounding box, depth map or a set of planes, we recover the\nfine geometry of the background via implicit representation. Extensive\nexperiments on the SUN RGB-D, Pix3D, 3D-FUTURE, and 3D-FRONT datasets\ndemonstrate that our method outperforms existing approaches in both background\nand foreground object reconstruction. Our code and model will be made publicly\navailable.\n", "date": "2022-07-18", "categories": []}
{"id": "2208.04307", "title": "PlaneFormers: From Sparse View Planes to 3D Reconstruction", "abstract": "  We present an approach for the planar surface reconstruction of a scene from\nimages with limited overlap. This reconstruction task is challenging since it\nrequires jointly reasoning about single image 3D reconstruction, correspondence\nbetween images, and the relative camera pose between images. Past work has\nproposed optimization-based approaches. We introduce a simpler approach, the\nPlaneFormer, that uses a transformer applied to 3D-aware plane tokens to\nperform 3D reasoning. Our experiments show that our approach is substantially\nmore effective than prior work, and that several 3D-specific design decisions\nare crucial for its success.\n", "date": "2022-08-08", "categories": []}
{"id": "2206.05737", "title": "SparseNeuS: Fast Generalizable Neural Surface Reconstruction from Sparse\n  Views", "abstract": "  We introduce SparseNeuS, a novel neural rendering based method for the task\nof surface reconstruction from multi-view images. This task becomes more\ndifficult when only sparse images are provided as input, a scenario where\nexisting neural reconstruction approaches usually produce incomplete or\ndistorted results. Moreover, their inability of generalizing to unseen new\nscenes impedes their application in practice. Contrarily, SparseNeuS can\ngeneralize to new scenes and work well with sparse images (as few as 2 or 3).\nSparseNeuS adopts signed distance function (SDF) as the surface representation,\nand learns generalizable priors from image features by introducing geometry\nencoding volumes for generic surface prediction. Moreover, several strategies\nare introduced to effectively leverage sparse views for high-quality\nreconstruction, including 1) a multi-level geometry reasoning framework to\nrecover the surfaces in a coarse-to-fine manner; 2) a multi-scale color\nblending scheme for more reliable color prediction; 3) a consistency-aware\nfine-tuning scheme to control the inconsistent regions caused by occlusion and\nnoise. Extensive experiments demonstrate that our approach not only outperforms\nthe state-of-the-art methods, but also exhibits good efficiency,\ngeneralizability, and flexibility.\n", "date": "2022-06-12", "categories": []}
{"id": "2207.09686", "title": "Object-Compositional Neural Implicit Surfaces", "abstract": "  The neural implicit representation has shown its effectiveness in novel view\nsynthesis and high-quality 3D reconstruction from multi-view images. However,\nmost approaches focus on holistic scene representation yet ignore individual\nobjects inside it, thus limiting potential downstream applications. In order to\nlearn object-compositional representation, a few works incorporate the 2D\nsemantic map as a cue in training to grasp the difference between objects. But\nthey neglect the strong connections between object geometry and instance\nsemantic information, which leads to inaccurate modeling of individual\ninstance. This paper proposes a novel framework, ObjectSDF, to build an\nobject-compositional neural implicit representation with high fidelity in 3D\nreconstruction and object representation. Observing the ambiguity of\nconventional volume rendering pipelines, we model the scene by combining the\nSigned Distance Functions (SDF) of individual object to exert explicit surface\nconstraint. The key in distinguishing different instances is to revisit the\nstrong association between an individual object's SDF and semantic label.\nParticularly, we convert the semantic information to a function of object SDF\nand develop a unified and compact representation for scene and objects.\nExperimental results show the superiority of ObjectSDF framework in\nrepresenting both the holistic object-compositional scene and the individual\ninstances. Code can be found at https://qianyiwu.github.io/objectsdf/\n", "date": "2022-07-20", "categories": []}
{"id": "2208.00183", "title": "Few-shot Single-view 3D Reconstruction with Memory Prior Contrastive\n  Network", "abstract": "  3D reconstruction of novel categories based on few-shot learning is appealing\nin real-world applications and attracts increasing research interests. Previous\napproaches mainly focus on how to design shape prior models for different\ncategories. Their performance on unseen categories is not very competitive. In\nthis paper, we present a Memory Prior Contrastive Network (MPCN) that can store\nshape prior knowledge in a few-shot learning based 3D reconstruction framework.\nWith the shape memory, a multi-head attention module is proposed to capture\ndifferent parts of a candidate shape prior and fuse these parts together to\nguide 3D reconstruction of novel categories. Besides, we introduce a 3D-aware\ncontrastive learning method, which can not only complement the retrieval\naccuracy of memory network, but also better organize image features for\ndownstream tasks. Compared with previous few-shot 3D reconstruction methods,\nMPCN can handle the inter-class variability without category annotations.\nExperimental results on a benchmark synthetic dataset and the Pascal3D+\nreal-world dataset show that our model outperforms the current state-of-the-art\nmethods significantly.\n", "date": "2022-07-30", "categories": []}
{"id": "2207.14287", "title": "Depth Field Networks for Generalizable Multi-view Scene Representation", "abstract": "  Modern 3D computer vision leverages learning to boost geometric reasoning,\nmapping image data to classical structures such as cost volumes or epipolar\nconstraints to improve matching. These architectures are specialized according\nto the particular problem, and thus require significant task-specific tuning,\noften leading to poor domain generalization performance. Recently, generalist\nTransformer architectures have achieved impressive results in tasks such as\noptical flow and depth estimation by encoding geometric priors as inputs rather\nthan as enforced constraints. In this paper, we extend this idea and propose to\nlearn an implicit, multi-view consistent scene representation, introducing a\nseries of 3D data augmentation techniques as a geometric inductive prior to\nincrease view diversity. We also show that introducing view synthesis as an\nauxiliary task further improves depth estimation. Our Depth Field Networks\n(DeFiNe) achieve state-of-the-art results in stereo and video depth estimation\nwithout explicit geometric constraints, and improve on zero-shot domain\ngeneralization by a wide margin.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.13464", "title": "Towards the Probabilistic Fusion of Learned Priors into Standard\n  Pipelines for 3D Reconstruction", "abstract": "  The best way to combine the results of deep learning with standard 3D\nreconstruction pipelines remains an open problem. While systems that pass the\noutput of traditional multi-view stereo approaches to a network for\nregularisation or refinement currently seem to get the best results, it may be\npreferable to treat deep neural networks as separate components whose results\ncan be probabilistically fused into geometry-based systems. Unfortunately, the\nerror models required to do this type of fusion are not well understood, with\nmany different approaches being put forward. Recently, a few systems have\nachieved good results by having their networks predict probability\ndistributions rather than single values. We propose using this approach to fuse\na learned single-view depth prior into a standard 3D reconstruction system.\n  Our system is capable of incrementally producing dense depth maps for a set\nof keyframes. We train a deep neural network to predict discrete, nonparametric\nprobability distributions for the depth of each pixel from a single image. We\nthen fuse this \"probability volume\" with another probability volume based on\nthe photometric consistency between subsequent frames and the keyframe image.\nWe argue that combining the probability volumes from these two sources will\nresult in a volume that is better conditioned. To extract depth maps from the\nvolume, we minimise a cost function that includes a regularisation term based\non network predicted surface normals and occlusion boundaries. Through a series\nof experiments, we demonstrate that each of these components improves the\noverall performance of the system.\n", "date": "2022-07-27", "categories": []}
{"id": "2207.10183", "title": "2D GANs Meet Unsupervised Single-view 3D Reconstruction", "abstract": "  Recent research has shown that controllable image generation based on\npre-trained GANs can benefit a wide range of computer vision tasks. However,\nless attention has been devoted to 3D vision tasks. In light of this, we\npropose a novel image-conditioned neural implicit field, which can leverage 2D\nsupervisions from GAN-generated multi-view images and perform the single-view\nreconstruction of generic objects. Firstly, a novel offline StyleGAN-based\ngenerator is presented to generate plausible pseudo images with full control\nover the viewpoint. Then, we propose to utilize a neural implicit function,\nalong with a differentiable renderer to learn 3D geometry from pseudo images\nwith object masks and rough pose initializations. To further detect the\nunreliable supervisions, we introduce a novel uncertainty module to predict\nuncertainty maps, which remedy the negative effect of uncertain regions in\npseudo images, leading to a better reconstruction performance. The\neffectiveness of our approach is demonstrated through superior single-view 3D\nreconstruction results of generic objects.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.08434", "title": "Efficient View Clustering and Selection for City-Scale 3D Reconstruction", "abstract": "  Image datasets have been steadily growing in size, harming the feasibility\nand efficiency of large-scale 3D reconstruction methods. In this paper, a novel\napproach for scaling Multi-View Stereo (MVS) algorithms up to arbitrarily large\ncollections of images is proposed. Specifically, the problem of reconstructing\nthe 3D model of an entire city is targeted, starting from a set of videos\nacquired by a moving vehicle equipped with several high-resolution cameras.\nInitially, the presented method exploits an approximately uniform distribution\nof poses and geometry and builds a set of overlapping clusters. Then, an\nInteger Linear Programming (ILP) problem is formulated for each cluster to\nselect an optimal subset of views that guarantees both visibility and\nmatchability. Finally, local point clouds for each cluster are separately\ncomputed and merged. Since clustering is independent from pairwise visibility\ninformation, the proposed algorithm runs faster than existing literature and\nallows for a massive parallelization. Extensive testing on urban data are\ndiscussed to show the effectiveness and the scalability of this approach.\n", "date": "2022-07-18", "categories": []}
{"id": "2207.08439", "title": "Revisiting PatchMatch Multi-View Stereo for Urban 3D Reconstruction", "abstract": "  In this paper, a complete pipeline for image-based 3D reconstruction of urban\nscenarios is proposed, based on PatchMatch Multi-View Stereo (MVS). Input\nimages are firstly fed into an off-the-shelf visual SLAM system to extract\ncamera poses and sparse keypoints, which are used to initialize PatchMatch\noptimization. Then, pixelwise depths and normals are iteratively computed in a\nmulti-scale framework with a novel depth-normal consistency loss term and a\nglobal refinement algorithm to balance the inherently local nature of\nPatchMatch. Finally, a large-scale point cloud is generated by back-projecting\nmulti-view consistent estimates in 3D. The proposed approach is carefully\nevaluated against both classical MVS algorithms and monocular depth networks on\nthe KITTI dataset, showing state of the art performances.\n", "date": "2022-07-18", "categories": []}
{"id": "2207.04622", "title": "Edge-preserving Near-light Photometric Stereo with Neural Surfaces", "abstract": "  This paper presents a near-light photometric stereo method that faithfully\npreserves sharp depth edges in the 3D reconstruction. Unlike previous methods\nthat rely on finite differentiation for approximating depth partial derivatives\nand surface normals, we introduce an analytically differentiable neural surface\nin near-light photometric stereo for avoiding differentiation errors at sharp\ndepth edges, where the depth is represented as a neural function of the image\ncoordinates. By further formulating the Lambertian albedo as a dependent\nvariable resulting from the surface normal and depth, our method is\ninsusceptible to inaccurate depth initialization. Experiments on both synthetic\nand real-world scenes demonstrate the effectiveness of our method for detailed\nshape recovery with edge preservation.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.00757", "title": "PhotoScene: Photorealistic Material and Lighting Transfer for Indoor\n  Scenes", "abstract": "  Most indoor 3D scene reconstruction methods focus on recovering 3D geometry\nand scene layout. In this work, we go beyond this to propose PhotoScene, a\nframework that takes input image(s) of a scene along with approximately aligned\nCAD geometry (either reconstructed automatically or manually specified) and\nbuilds a photorealistic digital twin with high-quality materials and similar\nlighting. We model scene materials using procedural material graphs; such\ngraphs represent photorealistic and resolution-independent materials. We\noptimize the parameters of these graphs and their texture scale and rotation,\nas well as the scene lighting to best match the input image via a\ndifferentiable rendering layer. We evaluate our technique on objects and layout\nreconstructions from ScanNet, SUN RGB-D and stock photographs, and demonstrate\nthat our method reconstructs high-quality, fully relightable 3D scenes that can\nbe re-rendered under arbitrary viewpoints, zooms and lighting.\n", "date": "2022-07-02", "categories": []}
{"id": "2206.15255", "title": "Neural Rendering for Stereo 3D Reconstruction of Deformable Tissues in\n  Robotic Surgery", "abstract": "  Reconstruction of the soft tissues in robotic surgery from endoscopic stereo\nvideos is important for many applications such as intra-operative navigation\nand image-guided robotic surgery automation. Previous works on this task mainly\nrely on SLAM-based approaches, which struggle to handle complex surgical\nscenes. Inspired by recent progress in neural rendering, we present a novel\nframework for deformable tissue reconstruction from binocular captures in\nrobotic surgery under the single-viewpoint setting. Our framework adopts\ndynamic neural radiance fields to represent deformable surgical scenes in MLPs\nand optimize shapes and deformations in a learning-based manner. In addition to\nnon-rigid deformations, tool occlusion and poor 3D clues from a single\nviewpoint are also particular challenges in soft tissue reconstruction. To\novercome these difficulties, we present a series of strategies of tool\nmask-guided ray casting, stereo depth-cueing ray marching and stereo\ndepth-supervised optimization. With experiments on DaVinci robotic surgery\nvideos, our method significantly outperforms the current state-of-the-art\nreconstruction method for handling various complex non-rigid deformations. To\nour best knowledge, this is the first work leveraging neural rendering for\nsurgical scene 3D reconstruction with remarkable potential demonstrated. Code\nis available at: https://github.com/med-air/EndoNeRF.\n", "date": "2022-06-30", "categories": []}
{"id": "2206.08712", "title": "An Algorithm for the SE(3)-Transformation on Neural Implicit Maps for\n  Remapping Functions", "abstract": "  Implicit representations are widely used for object reconstruction due to\ntheir efficiency and flexibility. In 2021, a novel structure named neural\nimplicit map has been invented for incremental reconstruction. A neural\nimplicit map alleviates the problem of inefficient memory cost of previous\nonline 3D dense reconstruction while producing better quality. % However, the\nneural implicit map suffers the limitation that it does not support remapping\nas the frames of scans are encoded into a deep prior after generating the\nneural implicit map. This means, that neither this generation process is\ninvertible, nor a deep prior is transformable. The non-remappable property\nmakes it not possible to apply loop-closure techniques. % We present a neural\nimplicit map based transformation algorithm to fill this gap. As our neural\nimplicit map is transformable, our model supports remapping for this special\nmap of latent features. % Experiments show that our remapping module is capable\nto well-transform neural implicit maps to new poses. Embedded into a SLAM\nframework, our mapping model is able to tackle the remapping of loop closures\nand demonstrates high-quality surface reconstruction. % Our implementation is\navailable at github\\footnote{\\url{https://github.com/Jarrome/IMT_Mapping}} for\nthe research community.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.08749", "title": "From a few Accurate 2D Correspondences to 3D Point Clouds", "abstract": "  Key points, correspondences, projection matrices, point clouds and dense\nclouds are the skeletons in image-based 3D reconstruction, of which point\nclouds have the important role in generating a realistic and natural model for\na 3D reconstructed object. To achieve a good 3D reconstruction, the point\nclouds must be almost everywhere in the surface of the object. In this article,\nwith a main purpose to build the point clouds covering the entire surface of\nthe object, we propose a new feature named a geodesic feature or geo-feature.\nBased on the new geo-feature, if there are several (given) initial world points\non the object's surface along with all accurately estimated projection\nmatrices, some new world points on the geodesics connecting any two of these\ngiven world points will be reconstructed. Then the regions on the surface\nbordering by these initial world points will be covered by the point clouds.\nThus, if the initial world points are around the surface, the point clouds will\ncover the entire surface.\n  This article proposes a new method to estimate the world points and\nprojection matrices from their correspondences. This method derives the\nclosed-form and iterative solutions for the world points and projection\nmatrices and proves that when the number of world points is less than seven and\nthe number of images is at least five, the proposed solutions are global\noptimal. We propose an algorithm named World points from their Correspondences\n(WPfC) to estimate the world points and projection matrices from their\ncorrespondences, and another algorithm named Creating Point Clouds (CrPC) to\ncreate the point clouds from the world points and projection matrices given by\nthe first algorithm.\n", "date": "2022-06-13", "categories": []}
{"id": "2206.07710", "title": "PlanarRecon: Real-time 3D Plane Detection and Reconstruction from Posed\n  Monocular Videos", "abstract": "  We present PlanarRecon -- a novel framework for globally coherent detection\nand reconstruction of 3D planes from a posed monocular video. Unlike previous\nworks that detect planes in 2D from a single image, PlanarRecon incrementally\ndetects planes in 3D for each video fragment, which consists of a set of key\nframes, from a volumetric representation of the scene using neural networks. A\nlearning-based tracking and fusion module is designed to merge planes from\nprevious fragments to form a coherent global plane reconstruction. Such design\nallows PlanarRecon to integrate observations from multiple views within each\nfragment and temporal information across different ones, resulting in an\naccurate and coherent reconstruction of the scene abstraction with\nlow-polygonal geometry. Experiments show that the proposed approach achieves\nstate-of-the-art performances on the ScanNet dataset while being real-time.\n", "date": "2022-06-15", "categories": []}
{"id": "2206.06533", "title": "3D scene reconstruction from monocular spherical video with motion\n  parallax", "abstract": "  In this paper, we describe a method to capture nearly entirely spherical (360\ndegree) depth information using two adjacent frames from a single spherical\nvideo with motion parallax. After illustrating a spherical depth information\nretrieval using two spherical cameras, we demonstrate monocular spherical\nstereo by using stabilized first-person video footage. Experiments demonstrated\nthat the depth information was retrieved on up to 97% of the entire sphere in\nsolid angle. At a speed of 30 km/h, we were able to estimate the depth of an\nobject located over 30 m from the camera. We also reconstructed the 3D\nstructures (point cloud) using the obtained depth data and confirmed the\nstructures can be clearly observed. We can apply this method to 3D structure\nretrieval of surrounding environments such as 1) previsualization, location\nhunting/planning of a film, 2) real scene/computer graphics synthesis and 3)\nmotion capture. Thanks to its simplicity, this method can be applied to various\nvideos. As there is no pre-condition other than to be a 360 video with motion\nparallax, we can use any 360 videos including those on the Internet to\nreconstruct the surrounding environments. The cameras can be lightweight enough\nto be mounted on a drone. We also demonstrated such applications.\n", "date": "2022-06-14", "categories": []}
{"id": "2206.07028", "title": "Learning 3D Object Shape and Layout without 3D Supervision", "abstract": "  A 3D scene consists of a set of objects, each with a shape and a layout\ngiving their position in space. Understanding 3D scenes from 2D images is an\nimportant goal, with applications in robotics and graphics. While there have\nbeen recent advances in predicting 3D shape and layout from a single image,\nmost approaches rely on 3D ground truth for training which is expensive to\ncollect at scale. We overcome these limitations and propose a method that\nlearns to predict 3D shape and layout for objects without any ground truth\nshape or layout information: instead we rely on multi-view images with 2D\nsupervision which can more easily be collected at scale. Through extensive\nexperiments on 3D Warehouse, Hypersim, and ScanNet we demonstrate that our\napproach scales to large datasets of realistic images, and compares favorably\nto methods relying on 3D ground truth. On Hypersim and ScanNet where reliable\n3D ground truth is not available, our approach outperforms supervised\napproaches trained on smaller and less diverse datasets.\n", "date": "2022-06-14", "categories": []}
{"id": "2206.06340", "title": "SNeS: Learning Probably Symmetric Neural Surfaces from Incomplete Data", "abstract": "  We present a method for the accurate 3D reconstruction of partly-symmetric\nobjects. We build on the strengths of recent advances in neural reconstruction\nand rendering such as Neural Radiance Fields (NeRF). A major shortcoming of\nsuch approaches is that they fail to reconstruct any part of the object which\nis not clearly visible in the training image, which is often the case for\nin-the-wild images and videos. When evidence is lacking, structural priors such\nas symmetry can be used to complete the missing information. However,\nexploiting such priors in neural rendering is highly non-trivial: while\ngeometry and non-reflective materials may be symmetric, shadows and reflections\nfrom the ambient scene are not symmetric in general. To address this, we apply\na soft symmetry constraint to the 3D geometry and material properties, having\nfactored appearance into lighting, albedo colour and reflectivity. We evaluate\nour method on the recently introduced CO3D dataset, focusing on the car\ncategory due to the challenge of reconstructing highly-reflective materials. We\nshow that it can reconstruct unobserved regions with high fidelity and render\nhigh-quality novel view images.\n", "date": "2022-06-13", "categories": []}
{"id": "2206.04669", "title": "Beyond RGB: Scene-Property Synthesis with Neural Radiance Fields", "abstract": "  Comprehensive 3D scene understanding, both geometrically and semantically, is\nimportant for real-world applications such as robot perception. Most of the\nexisting work has focused on developing data-driven discriminative models for\nscene understanding. This paper provides a new approach to scene understanding,\nfrom a synthesis model perspective, by leveraging the recent progress on\nimplicit 3D representation and neural rendering. Building upon the great\nsuccess of Neural Radiance Fields (NeRFs), we introduce Scene-Property\nSynthesis with NeRF (SS-NeRF) that is able to not only render photo-realistic\nRGB images from novel viewpoints, but also render various accurate scene\nproperties (e.g., appearance, geometry, and semantics). By doing so, we\nfacilitate addressing a variety of scene understanding tasks under a unified\nframework, including semantic segmentation, surface normal estimation,\nreshading, keypoint detection, and edge detection. Our SS-NeRF framework can be\na powerful tool for bridging generative learning and discriminative learning,\nand thus be beneficial to the investigation of a wide range of interesting\nproblems, such as studying task relationships within a synthesis paradigm,\ntransferring knowledge to novel tasks, facilitating downstream discriminative\ntasks as ways of data augmentation, and serving as auto-labeller for data\ncreation.\n", "date": "2022-06-09", "categories": []}
{"id": "2206.02840", "title": "Spatial Acoustic Projection for 3D Imaging Sonar Reconstruction", "abstract": "  In this work we present a novel method for reconstructing 3D surfaces using a\nmulti-beam imaging sonar. We integrate the intensities measured by the sonar\nfrom different viewpoints for fixed cell positions in a 3D grid. For each cell\nwe integrate a feature vector that holds the mean intensity for a discretized\nrange of viewpoints. Based on the feature vectors and independent sparse range\nmeasurements that act as ground truth information, we train convolutional\nneural networks that allow us to predict the signed distance and direction to\nthe nearest surface for each cell. The predicted signed distances can be\nprojected into a truncated signed distance field (TSDF) along the predicted\ndirections. Utilizing the marching cubes algorithm, a polygon mesh can be\nrendered from the TSDF. Our method allows a dense 3D reconstruction from a\nlimited set of viewpoints and was evaluated on three real-world datasets.\n", "date": "2022-06-06", "categories": []}
{"id": "2206.02776", "title": "Volumetric Disentanglement for 3D Scene Manipulation", "abstract": "  Recently, advances in differential volumetric rendering enabled significant\nbreakthroughs in the photo-realistic and fine-detailed reconstruction of\ncomplex 3D scenes, which is key for many virtual reality applications. However,\nin the context of augmented reality, one may also wish to effect semantic\nmanipulations or augmentations of objects within a scene. To this end, we\npropose a volumetric framework for (i) disentangling or separating, the\nvolumetric representation of a given foreground object from the background, and\n(ii) semantically manipulating the foreground object, as well as the\nbackground. Our framework takes as input a set of 2D masks specifying the\ndesired foreground object for training views, together with the associated 2D\nviews and poses, and produces a foreground-background disentanglement that\nrespects the surrounding illumination, reflections, and partial occlusions,\nwhich can be applied to both training and novel views. Our method enables the\nseparate control of pixel color and depth as well as 3D similarity\ntransformations of both the foreground and background objects. We subsequently\ndemonstrate the applicability of our framework on a number of downstream\nmanipulation tasks including object camouflage, non-negative 3D object\ninpainting, 3D object translation, 3D object inpainting, and 3D text-based\nobject manipulation. Full results are given in our project webpage at\nhttps://sagiebenaim.github.io/volumetric-disentanglement/\n", "date": "2022-06-06", "categories": []}
{"id": "2206.00491", "title": "Semantic Room Wireframe Detection from a Single View", "abstract": "  Reconstruction of indoor surfaces with limited texture information or with\nrepeated textures, a situation common in walls and ceilings, may be difficult\nwith a monocular Structure from Motion system. We propose a Semantic Room\nWireframe Detection task to predict a Semantic Wireframe from a single\nperspective image. Such predictions may be used with shape priors to estimate\nthe Room Layout and aid reconstruction. To train and test the proposed\nalgorithm we create a new set of annotations from the simulated Structured3D\ndataset. We show qualitatively that the SRW-Net handles complex room geometries\nbetter than previous Room Layout Estimation algorithms while quantitatively\nout-performing the baseline in non-semantic Wireframe Detection.\n", "date": "2022-06-01", "categories": []}
{"id": "2606.21244", "title": "ACE-GS: Acing the Trade-off with Accurate, Compact and Efficient 3D Gaussian Splatting", "abstract": "3D Gaussian Splatting achieves exceptional real-time rendering, but its substantial computational and storage demands hinder widespread deployment. Existing accelerated paradigms often aggressively prune primitives for rapid convergence, causing severe loss of high-frequency details. To address this, we tackle the fundamental problem of achieving both exceptional rendering quality and ultra-fast reconstruction speed. In this paper, we propose ACE-GS, a progressive optimization framework tailored for accurate, compressed, and efficient scene representation. We realize that precise primitive management is the key to breaking this trade-off. Therefore, we first design a momentum consistency-guided densification strategy, strictly constraining primitive growth onto authentic geometric manifolds to avoid computational waste while significantly accelerating convergence. Building upon this efficient initialization, we deploy a statistical sensitivity-driven sparsification mechanism to precisely prune redundant primitives, yielding a further compressed footprint. Finally, to thoroughly compensate for the risk of micro-structure loss caused by the aforementioned strict primitive control, we introduce a cross-dimensional residual frequency compensation scheme that explicitly back-injects high-frequency error energy into primitive attributes, perfectly restoring sharp geometric details. Extensive experiments validate our superiority. While maintaining a highly compact scene representation, our system achieves up to 3.7 times training acceleration against the rapid framework Speedy-Splat. Requiring only 3 to 5 minutes to converge, ACE-GS secures the highest structural similarity and achieves a peak PSNR improvement of up to 0.89 dB over the original 3DGS, establishing a new benchmark for ultra-fast and high-fidelity novel view synthesis.", "date": "2026-06-19", "categories": []}
{"id": "2606.21373", "title": "FLM-Occ: Feed-forward Likelihood Maximization for Efficient Indoor Occupancy Prediction", "abstract": "Recent indoor occupancy prediction methods adopt Gaussian primitives as a sparse 3D representation for computational efficiency. However, their training relies on voxel classification, which imposes only local constraints and lacks global supervision on the distribution of the primitives. Therefore, they inevitably predict spurious primitives in empty regions, undermining both representational and computational efficiency. To address this, we propose Feed-forward Likelihood Maximization (FLM), a novel framework that reformulates occupancy prediction as voxel distribution estimation. In FLM, a network is trained to predict a mixture model that maximizes the likelihood over ground-truth occupied voxels in a feed-forward manner. To enable end-to-end training of networks and voxelization of a standard mixture model, we define mixture weights as normalized primitive volumes to implicitly enforce simplex constraints and derive novel voxelization formulas. Based on FLM, our FLM-Occ, a novel method that is capable of relocating randomly initialized primitives over long distances to model a scene. On Occ-ScanNet, FLM-Occ achieves superior accuracy using only 32 superquadrics, 2.7% of the prior SoTA, while running 3.7 times faster.", "date": "2026-06-19", "categories": []}
{"id": "2606.21596", "title": "$\\phi$-Scene: Physically Grounded Image-to-3D Scene Reconstruction", "abstract": "Reconstructing compositional 3D scenes from a single image is a fundamental challenge in 3D world modeling. Recent methods can recover high-fidelity, complete 3D objects and predict plausible scene arrangements, but most still treat scene reconstruction primarily as a visual and geometric prediction problem. Their outputs may therefore contain floating objects, interpenetrations, or unstable-contact artifacts, limiting their physical validity and downstream usability in simulation, robotics, and interactive environments. We present $\\phi$-Scene, a physically grounded approach to open-vocabulary and compositional image-to-3D scene reconstruction. The key premise is that a reconstructed scene should not be treated merely as a set of objects with predicted poses, but as a stable physical system. Accordingly, $\\phi$-Scene formulates reconstruction as topology-driven physical assembly: it infers how objects support one another, orders them accordingly, and progressively settles each object against its already stabilized support context. For each object in topological order, SDF-based optimization first resolves penetrations against the pre-settled support context, and rigid-body simulation then settles the object into a stable contact configuration under real-world physical constraints. Experiments on 3D-Front show that $\\phi$-Scene achieves the strongest overall performance among out-of-domain methods and remains highly competitive with in-domain baselines. Human and VLM evaluations further show strong preference for $\\phi$-Scene in visual quality, reference alignment, and physical plausibility. Finally, dedicated physical plausibility metrics covering static contact quality and dynamic stability demonstrate that $\\phi$-Scene substantially reduces penetration artifacts while producing much lower post-simulation drift, indicating more stable and physically grounded 3D scenes.", "date": "2026-06-19", "categories": []}
{"id": "2606.22197", "title": "Multi4D: High-Fidelity Dynamic Gaussian Splatting via Multi-Level Competitive Allocation", "abstract": "Dynamic 3D Gaussian splatting faces a fundamental tension between motion consistency and visual fidelity. Deformation-based approaches preserve temporal correspondence but suffer from motion over-factorization, oversmoothing high-frequency dynamics. In contrast, 4D-primitive methods capture fine visual details yet incur temporal overparameterization, breaking object identity and leading to severe storage overhead. To resolve this, we introduce Multi4D, a framework for high-fidelity dynamic Gaussian Splatting based on multi-level competitive allocation. Instead of a monolithic representation, we distribute modeling capacity across three structured levels: static structure, persistent dynamic geometry, and transient appearance primitives. Through shared rasterization and residual-driven optimization, these levels dynamically compete to explain photometric error, enabling adaptive specialization without pre-assigned decomposition. This allocation preserves long-term motion consistency while capturing fine dynamic detail, achieving state-of-the-art rendering quality and real-time performance with significantly fewer dynamic primitives. Furthermore, because our representation explicitly tracks compact persistent Gaussians over time, semantic features can be embedded afterward, enabling Multi4D to achieve state-of-the-art 4D segmentation accuracy with an order-of-magnitude speedup. Project page: https://batfacewayne.github.io/Multi4D.io/", "date": "2026-06-20", "categories": []}
{"id": "2606.22787", "title": "Visual Geometry Transformer in the Wild: Distractor-Free 3D Reconstruction", "abstract": "Current end-to-end multi-view 3D reconstruction methods achieve impressive results, but rely on a restrictive static assumption: the scenes is entire distractor-free with perfect cross-view geometry. This reliance on idealized inputs causes even the most advanced methods to fail in real-world settings, where transient distractors and occlusions present. To address this, we propose Visual Geometry Transformer in the Wild (VGTW), an end-to-end framework for robust reconstruction from inconsistent views. At its core, we isolate and suppress distractor-affected regions while preserving the consistent components across views. Specifically, we introduce a Distractor-aware Training (DAT) strategy that separates clean features from distractor-contaminated ones in the attention mechanism while enforcing feature consistency across images. To enable this, we train the model with an auxiliary mask prediction head, using supervision from a new dataset we collected with pixel-level distractor masks. The resulting VGTW model is a feed-forward network that directly outputs clean, distractor-free point clouds. Remarkably, it requires no additional 3D supervision, remains computationally efficient, and is compatible with existing pipelines. Extensive experiments validate our approach, demonstrating state-of-the-art performance and robust generalization in diverse, real-world scenarios.", "date": "2026-06-22", "categories": []}
{"id": "2606.23031", "title": "DrivingVoxels: Compositional Sparse Voxel Rasterization for Dynamic Driving Scene Reconstruction", "abstract": "Reconstructing dynamic urban scenes remains challenging due to the unbounded nature of driving environments and the presence of multiple dynamic objects. Currently, potentially faster sparse voxel methods are mainly designed for static scenarios. On the other hand, dynamic approaches based on 3D Gaussian Splatting, despite their high-fidelity, are often time-consuming for driving scenarios and exhibit uncontrollable memory growth in large scenes. To address these limitations, we present DrivingVoxels, a compositional sparse voxel rendering framework for dynamic driving scenes. Our method jointly rasterizes sparse voxels from multiple independent octrees within a single rendering pass. Each rigid dynamic object is represented by an octree defined in its local coordinate frame, while a separate static octree models the stationary background. DrivingVoxels adopts a fully explicit, neural-free representation together with a LiDAR-guided structural initialization that efficiently captures scene geometry. We evaluate our framework on the PandaSet benchmark, demonstrating that DrivingVoxels performs on par on perceptual metrics and better on structural metrics for NVS and reconstruction while requiring shorter training times than previous 3DGS-base methods to an efficient optimization workflow anchored by a strong LiDAR prior.", "date": "2026-06-22", "categories": []}
{"id": "2606.23212", "title": "Temporally Aware Densification for Dynamic 3D Gaussian Splatting", "abstract": "Despite modeling temporal motion, dynamic 3D Gaussian Splatting (3DGS) methods still inherit a static densification strategy that is ill-suited for dynamic scenes. This neglect of temporal behavior leads to under-reconstructed and blurry dynamic regions, as short-lived Gaussians receive sparse supervision and fail to densify effectively. We propose a Visibility-Aware Densification (VAD) framework that integrates temporal visibility into the densification process, ensuring that Gaussians are refined based on their actual temporal presence. A Temporally-Adaptive Thresholding (TAT) mechanism further adjusts each Gaussian's densification threshold according to its temporal lifespan, promoting balanced refinement of both static and dynamic regions. Finally, a Temporal Offset Warping (TOW) design enhances deformation capacity around temporal centers, extending the lifespan of highly dynamic Gaussians and facilitating more effective densification. Our approach achieves substantial improvements in the visual quality of dynamic regions, outperforming existing methods across three dynamic multi-view benchmark datasets. Moreover, the proposed VAD module generalizes across diverse dynamic 3DGS methods, consistently improving dynamic reconstruction as a plug-and-play component.", "date": "2026-06-22", "categories": []}
{"id": "2606.23298", "title": "Ocean4D: Generative Underwater 4D Reconstruction via Medium-Aware Video Diffusion", "abstract": "Underwater 4D reconstruction remains challenging due to the coupling between degraded light transport in participating media and dynamic water variations. Most existing Methods are developed under in-air assumptions and do not explicitly account for underwater absorption and backscatter. Additionally, near-static assumptions make these approaches sensitive to drifting particles and dynamic distractors , leading to unstable geometry and inconsistent cross-view results. To address these issues, we propose a generative framework for underwater 4D reconstruction, named Ocean4D, which is built on two complementary components. Specifically, 4D-GCC constructs 4D geometrically consistent conditioning with improved cross-frame coverage, while the Medium-Aware Block performs implicit medium-aware denoising in the latent diffusion process to stabilize underwater appearance under absorption and scattering. Given a monocular video and target cameras, our method generates videos along the target trajectories while preserving global structure and cross-view consistency. Extensive experiments on both dynamic and static underwater benchmarks demonstrate state-of-the-art performance on underwater reconstruction.", "date": "2026-06-22", "categories": []}
{"id": "2606.24180", "title": "Deep Learning Approaches for 3D Medical Scene Completion: From Geometric Modeling to Generative Paradigms", "abstract": "Three-dimensional scene completion has evolved as a major problem in computer vision and robotics, and its applications are diverse, including autonomous navigation and augmented reality. In this study, a systematic review has been conducted to compile the research contributions made in the last ten years, i.e., 2016 to 2026, which has revolutionized the field from the voxel semantic completion paradigm represented by SSCNet to the latest paradigm that combines generative diffusion priors with real-time rendering using a Gaussian splatting technique. The evolution in representation paradigms, such as voxel grids, point learning, implicit neural fields, transformer networks, diffusion networks, and the latest paradigm based on rendering-aware 3D Gaussian primitives, has been discussed in this study. A comprehensive analysis has been carried out on the contributions made in the last ten years, and a taxonomy has been developed to provide a clear idea about the contributions made in the field. The study has also discussed the research contributions made in the field, along with the challenges that still need to be addressed. Finally, the study has presented a research agenda that will provide a clear idea about the directions that can be followed in the development of the next-generation system", "date": "2026-06-23", "categories": []}
{"id": "2606.24628", "title": "ArtiTwinSplat: Interactable Digital Twin Reconstruction via Gaussian Splatting from RGB-D videos", "abstract": "Deploying robots in unstructured real-world environments needs accurate, interactive models of the objects. Constructing these models at scale remains a critical bottleneck for robotic system integration. We present ArtiTwinSplat, a framework that automatically constructs articulated, photo-realistic digital twins of objects directly from RGB-D videos, requiring no CAD models, simulation assets, or manual annotations. Our method is built on 3D Gaussian Splatting that preserve geometric fidelity and photometric realism, coupled with an unsupervised articulation discovery pipeline that recovers part structure and joint kinematics from observed motion alone. With tracking and optimization stages our method provides stable, queryable digital twins that support real-time rendering, viewpoint control, and interactive manipulation. Unlike prior methods confined to simulation, ArtiTwinSplat operates directly on real-world observations and produces twins that are immediately usable by downstream robot planning and learning systems. This method offers a practical, scalable pathway toward digital twin construction, lowering the integration barrier for articulated object manipulation in embodied AI and human-robot collaboration contexts.", "date": "2026-06-23", "categories": []}
{"id": "2606.25430", "title": "PRISM: Feed-Forward Single-Image 3D Reconstruction via Geometric Warp-Residual Modeling", "abstract": "Reconstructing 3D scenes from a single image is a fundamental challenge in computer vision, with broad applications in virtual reality, robotics, and content creation. Recent methods achieve outstanding performance by leveraging camera-controlled video diffusion models, but rely on iterative diffusion sampling, which greatly limits their practical deployment. We observe that geometric forward warping alone can cover the majority of a target view directly from the input image, with only a compact residual left for the encoder to correct. Motivated by this observation, we propose PRISM, a feed-forward framework that decomposes multi-view latent prediction into a parameter-free geometric prior and a learned residual correction, with no diffusion sampling required at inference. To enable generalization from purely synthetic training data, we devise a two-stage training strategy combining latents supervised distillation for geometric generalization and perceptual fine-tuning for appearance quality optimization. Extensive experiments on three benchmarks demonstrate that PRISM achieves competitive reconstruction quality compared with diffusion-based methods, while reducing inference time dramatically to only 36 seconds per scene.", "date": "2026-06-24", "categories": []}
{"id": "2606.27071", "title": "PanoImager: Geometry-Guided Novel View Synthesis and Reconstruction from Sparse Panoramic Views", "abstract": "Panoramic sensing offers wide field-of-view coverage, yet 3D reconstruction from sparse panoramas remains challenging under rotation-dominant, weak-parallax motion. In such regimes, SfM/SLAM initialization is often ill-conditioned and unreliable. We present PanoImager, an SfM-free framework that combines feed-forward pose/depth priors, geometry-conditioned diffusion view completion, and depth-guided 3DGS optimization. Given only a few panoramic images, PanoImager decomposes them into local perspective views, synthesizes auxiliary observations to enrich sparse evidence, and stabilizes Gaussian optimization for improved cross-view consistency. Experiments on multiple benchmarks show improved stability under extreme sparsity, suggesting PanoImager as an offline/background component for map refinement when SfM/SLAM fails to initialize.", "date": "2026-06-25", "categories": []}
{"id": "2606.27223", "title": "SatSplatDiff: Geometry-preserving generative refinement for high-fidelity satellite Gaussian Splatting", "abstract": "Gaussian Splatting has been recently explored for satellite 3D reconstruction, demonstrating flexibility and efficiency in representing radiometrically diverse satellite scenes. However, the limited top viewpoint of satellite imagery results in insufficient supervision on building facades, leaving surface holes and degraded visual fidelity. Generative refinement, which leverages pretrained generative priors to iteratively refine and update the rendered images used as supervision targets, has recently been investigated to improve the visual fidelity of Gaussian-rendered images. However, since these models refine each view independently, the resulting images can generate hallucinations and break photo-consistency, leading to geometric degradation. To address these limitations, we propose SatSplatDiff, which aims to minimize geometric degradation prevalent in generative refinement. Building on photogrammetric DSM initialization and 2DGS-based shadow casting established in our prior work SatSplat, we first introduce monocular depth supervision and multi-scale geometric refinement to establish a geometrically accurate and well-regularized surface representation. We then apply shadow-guided generative refinement, where geometrically calculated shadow maps guide the Gaussians to maintain consistency with the underlying geometry, improving visual fidelity while reducing geometric degradation. Extensive evaluations on the IARPA2016 and DFC2019 datasets demonstrate state-of-the-art performance, reducing geometric MAE by up to 18% and improving visual fidelity (FID-CLIP) by 28-45% over existing baselines. Our method delivers up to 5x resolution enhancement with minimal hallucination and sensor-consistent appearance, demonstrating seamless cross-tile consistency and strong scalability for large-scale reconstruction. Source code is available at https://github.com/GDAOSU/SatSplatDiff", "date": "2026-06-25", "categories": []}
{"id": "2606.27509", "title": "Structured-Li-GS: Structured 3D Gaussians Splatting with LiDAR Incorporation and Spatial Constraints", "abstract": "In this study, we develop a Structured framework for Gaussian Splatting (3DGS) with LiDAR integration (Structured-Li-GS). It is a lightweight Gaussian Splatting pipeline that leverages LiDAR-inertial-visual SLAM. Structured-Li-GS achieves high-quality 3D reconstructions with fewer Gaussians by training on accurate, dense, colorized point clouds. Gaussian primitives are anchored using sub-sampled point clouds, and their ellipsoidal parameters are initialized from local surface geometry. Our training strategy integrates a comprehensive set of loss terms, including photometric, flattening, offset, depth, and normal losses, guided by the dense point cloud, enabling accurate reconstruction without Gaussian densification. This approach produces up-to-scale, high-fidelity results with a moderate model size. For experimental validation, we develop a custom hardware-synchronized LiDAR-camera handheld scanner. Experiments on both benchmark datasets and our real-world in-house dataset demonstrate that Structured-Li-GS surpasses state-of-the-art methods while using fewer Gaussians.", "date": "2026-06-25", "categories": []}
{"id": "2606.27720", "title": "Scene and Human in One World: Reconstruction in a Feedforward Pass", "abstract": "Reconstructing humans in dynamic scenes from moving monocular cameras remains challenging due to scale ambiguity, human-scene misalignment, and occlusion interference. Rather than treating human mesh recovery and scene reconstruction as separate tasks, we believe that accurate human-scene reconstruction requires the two tasks to mutually inform each other: parametric human models offer semantic structure and metric-scale priors, while scene geometry provides spatial context for human localization and alignment. Built on this insight, we introduce SHOW, a mask-promptable human mesh recovery framework that couples feed-forward 3D scene reconstruction with Human Mesh Recovery in a unified metric space. SHOW injects human semantics and scale priors from parametric human models into normalized point-map prediction, enabling metric-scale scene reconstruction from inherently scale-ambiguous monocular input. In turn, the recovered scene geometry constrains human mesh estimation, encouraging spatially consistent human placement and improved human-scene alignment. To handle complex multi-person and cluttered scenes, SHOW further incorporates a promptable masking mechanism that enables flexible target-human selection while suppressing background distractions and occlusion interference. Through joint training, the model learns both human-aware geometric features and geometry-constrained human features, producing aligned metric-scale reconstructions from monocular human-centric videos. Extensive experiments demonstrate that SHOW improves metric-scale consistency, human-scene alignment, and reconstruction accuracy under challenging camera motion, occlusion, and cluttered backgrounds.", "date": "2026-06-26", "categories": []}
{"id": "2606.28060", "title": "ReScene: Structured Indoor Scene Reconstruction from Multi-View Captures", "abstract": "Constructing simulation-ready 3D scenes from multi-view captures is a key bottleneck for Embodied Artificial Intelligence, as downstream tasks require object-level structure, explicit inter-object relations, and physical plausibility. Existing approaches either rely on specialized capture hardware, suffer from single-view bias in object reconstruction, or yield layouts that are geometrically reasonable but physically inconsistent. We identify that the problem is not single-object reconstruction but cross-view relation fusion and physically plausible scene assembly. To address this challenge, we present ReScene, a framework that threads multi-view geometry throughout the pipeline as a unifying prior. Our method consists of two main components: HierView prioritizes reconstruction views based on semantic consistency and 3D coverage completeness, replacing the largest-mask heuristic that conflates image occupancy with object coverage; and Relation-Aware Assembly fuses multi-frame relation predictions from a vision-language model with geometric and room-shell priors into a confidence-weighted scene graph, enabling physically consistent scene assembly. ReScene sets a new state of the art across geometry, rendering, and perceptual quality on a set of ScanNet scenes, achieving a 17% reduction in Chamfer Distance and 26% in LPIPS over the strongest prior baseline, while running up to 10x faster than prior multi-view methods. Based on the reconstructed scenes, we also generate an embodied visual question answering dataset, on which fine-tuned Qwen-VL approaches the performance of strong closed-source models on several spatial reasoning tasks.", "date": "2026-06-26", "categories": []}
{"id": "2606.28215", "title": "HAT-4D: Lifting Monocular Video for 4D Multi-Object Interactions via Human-Agent Collaboration", "abstract": "Extracting dynamic 4D object interactions from massive, in-the-wild monocular videos offers a highly efficient data collection pathway for scaling Embodied AI and training VLAs. However, existing monocular 4D reconstruction methods primarily focus on isolated objects, often failing under the severe occlusions and complex dynamics inherent in multi-object interactions. To bridge this gap, we propose HAT-4D, the first agentic framework designed to reconstruct the 3D geometry, temporal dynamics, and physical interactions of multiple objects from a single video. By integrating VLMs with a multi-level human-in-the-loop feedback mechanism, HAT-4D efficiently resolves depth ambiguities and interaction-induced occlusions during 3D generation and 4D propagation, yielding physically plausible assets without relying on expensive multicamera rigs. As a scalable data engine, HAT-4D facilitates the creation of MVOIK-4D, an open-world benchmark for monocular 4D interaction reconstruction, accompanied by a novel multi-dimensional evaluation protocol focused on physical plausibility and temporal consistency. Extensive experiments demonstrate that HAT-4D achieves SOTA performance on most evaluation metrics, while maintaining competitive semantic alignment. Ablation studies show that introducing a small amount of human feedback improves interaction reconstruction. Moreover, the data produced by HAT-4D effectively improves baseline performance when used for fine-tuning. Our data and code are available at https://lijiaxin0111.github.io/HAT4D/", "date": "2026-06-26", "categories": []}
{"id": "2606.28321", "title": "StructSplat: Generalizable 3D Gaussian Splatting from Uncalibrated Sparse Views", "abstract": "We present StructSplat, a feed-forward and generalizable 3D Gaussian reconstruction framework that operates directly on uncalibrated images without requiring camera parameters. Existing methods either rely on per-scene optimization or assume known camera poses, and often entangle geometry and appearance within a unified backbone, limiting reconstruction fidelity and generalization. Our key idea is to adopt a structured representation that organizes geometry, semantic, and texture cues with explicit roles in the reconstruction process. Specifically, we introduce a pixel-aligned feature injection mechanism to enable accurate texture modeling from 2D observations, incorporate semantic-aware priors to improve global consistency, and design a camera alignment strategy to prevent information leakage and improve generalization. Experiments show that our method significantly outperforms prior approaches on challenging benchmarks. On DL3DV, our method achieves 28.045 PSNR, surpassing AnySplat (22.377) by +5.67 dB. In cross-dataset evaluation, our method achieves +1.94 dB over AnySplat on ACID and +1.72 dB on RealEstate10K. Project page: https://structsplat.github.io Code: https://github.com/J-C-Zhao/StructSplat", "date": "2026-06-26", "categories": []}
{"id": "2606.28581", "title": "SatSplat: Geometrically-Accurate Gaussian Splatting for Satellite Imagery", "abstract": "High-resolution satellite imagery demands 3D reconstruction methods that deliver both speed and geometric accuracy. Recent adaptations of 3D Gaussian Splatting (3DGS) to satellite imagery demonstrate strong efficiency, but reconstruction quality often degrades under diverse illumination across multi-date, high-altitude acquisitions (with small intersection angles), limiting applicability to remote sensing and vision tasks. We present SatSplat, the first framework to adapt 2D Gaussian Splatting (2DGS) to satellite photogrammetry, with online camera adjustment. We approximate satellite cameras with an affine model and learn a minimal delta parameterization for in-splat camera refinement from dense observations. The formulation is implemented with a 2DGS scene representation. To handle time-varying shadows and illumination changes, we integrate geometric shadow mapping and per-camera color correction during training. Across the evaluated DFC2019 and IARPA2016 benchmark sites, SatSplat achieves strong geometric accuracy while significantly outperforming prior 3DGS-based baselines. On our processed DFC2019 benchmark, SatSplat reduces mean absolute error by 11.93% and peak video memory by 31% relative to the previous state of the art. Our approach enables large-scale digital surface modeling with practical computational efficiency. The project page is available at https://gdaosu.github.io/satsplat/.", "date": "2026-06-26", "categories": []}
{"id": "2606.28820", "title": "CoGS: Compositional Dynamic Human-Object Scenes Gaussian Splatting from Monocular Video", "abstract": "Reconstructing dynamic human--object interaction scenes from monocular video is difficult because the human, manipulated object, and background obey different motion models while sharing the same pixels. Existing dynamic radiance-field and Gaussian-splatting methods often entangle these components, causing object motion to leak into the human or static scene, and monocular human reconstruction remains underconstrained in regions that are rarely observed. We present CoGS, a compositional Gaussian-splatting framework for monocular human--object scene reconstruction. CoGS decomposes the video into three coordinated branches: an articulated human initialized from a complete canonical prior, a rigid object field driven by an estimated object trajectory, and a static scene field regularized by weak scene-only planar primitives when available. A six-stage optimization schedule first stabilizes the human and object independently, then fuses them with the scene under full-image supervision, visibility-aware human anchoring, object silhouette and motion constraints, and delayed scene regularization. This design keeps each component responsible for its own geometry and motion while allowing photometric evidence to correct the final composite. Experiments on HOSNeRF and NeuMan show that CoGS improves both human--object interaction reconstruction and in-the-wild human--scene rendering, achieving stronger fidelity and perceptual quality across full-frame and human-focused evaluations. Code will be released upon publication.", "date": "2026-06-27", "categories": []}
{"id": "2606.28826", "title": "RefGlass-GS: A UAV-Enabled Fusion Framework for Photorealistic, Semantic and Interactive Digitization of Reflective Glass Facades via Gaussian Splatting", "abstract": "Existing digitization of buildings with reflective glass facades suffers from geometric reconstruction distortion, unrealistic view-dependent texture rendering, and difficulties in object-based semantic enhancement. Therefore, we propose RefGlass-GS, a fusion framework that enables end-to-end UAV-based photorealistic, semantic, and interactive digitization of reflective glass facades. The contributions include: (1) proposing an individual glass panel segmentation method based on maximum a posteriori estimation with structural regularities, robust to severe reflection and background interference; (2) formulating a UAV viewpoint planning optimization function that maximizes the coverage of view-dependent appearance for sufficient data capture; (3) developing an optimized Gaussian Splatting framework with a Reflection MLP, a novel deferred shading function, and two enhanced regularization terms for effective modeling of high-frequency near-field reflections; (4) introducing a standardized data organization paradigm for structuring GS-based representations into object-based models, facilitating interactive facility management on digital twin platforms. Experiments on real-world reflective glass facade scenes validate the effectiveness and superiority of the proposed method. Specifically, the glass panel segmentation achieves an improvement of 0.1927 in mIoU over SOTA methods, and only our method enables instance-level panel extraction. The UAV view planning improves novel view synthesis for reflective facades by 13.15 dB in PSNR compared to commercially used nap-of-the-object planning methods. The RefGlass-GS modeling outperforms SOTA Gaussian Splatting approaches for reflective scenes with an average improvement of 5.08 dB in PSNR.", "date": "2026-06-27", "categories": []}
{"id": "2606.28828", "title": "Ground4D: Consistency-Aware 4D Reconstruction from Monocular Video", "abstract": "Learning a 4D scene representation from a single monocular video that supports dynamic novel-view synthesis while maintaining faithful geometry over time remains challenging. Dynamic Gaussian Splatting achieves strong rendering performance through photometric optimization, yet does not explicitly enforce multi-view geometric consistency. In contrast, 3D foundation models recover coherent scene geometry and camera motion, but their point-based outputs are not designed for photorealistic rendering. We propose Ground4D, a geometry-grounded framework built on two stages. First, we perform geometry initialization via 3D foundation models, leveraging VGGT in a training-free manner to reconstruct multi-view-consistent 3D geometry and camera poses from monocular video. The recovered geometry provides a structured and reliable initialization for dynamic Gaussian representations. Second, we conduct geometry-consistency-aware refinement via dynamic Gaussian Splatting, optimizing the representation through differentiable rendering while maintaining multi-view geometric consistency across both observed and synthesized viewpoints. Furthermore, Ground4D inherently models the continuous 4D dynamics of the scene, naturally supporting rendering at arbitrary timestamps. By integrating foundation-level geometric priors into dynamic Gaussian optimization, Ground4D achieves stronger reconstruction fidelity and rendering performance, underscoring the role of geometry-grounded constraints in robust 4D scene modeling.", "date": "2026-06-27", "categories": []}
{"id": "2606.28840", "title": "DLGStream: Dynamic Language-embedded Guassian Splatting for Open-vocabulary Enabled Free-viewpoint Video Streaming", "abstract": "3D Gaussian Splatting~(3DGS) has emerged as a promising paradigm for reconstructing streamable free-viewpoint video~(FVV) from multi-view videos. However, 3DGS-based FVVs typically lack user interaction and editing capabilities, which diminishes the immersive experience. Recent research has integrated language features from CLIP into 3DGS via distillation, enabling open-vocabulary queries and supporting many downstream applications. Nevertheless, the stringent requirements of FVV, low frame size and high FPS, make current language Gaussian representations unsuitable for language-embedded FVV. In this paper, we propose DLGStream, a novel language-embedded FVV representation that streams time-varying language features alongside Gaussian attributes to support 4D environment interaction, scene editing, and spatial intelligence. Specifically, we propose a dual-opacity dynamic language Gaussian representation, which maintains two opacity attributes for color and language features to deal with performance degradation that occurs when colors and features are jointly optimized. Furthermore, we introduce an interpolation-based deformation field to reduce temporal redundancy. This deformation field can also be used for 4D frame interpolation, boosting FVV sequences from low to high FPS. Experimental results demonstrate that DLGStream achieves superior performance in both on open-vocabulary segmentation and reconstruction quality with an average frame size of merely 43 KB. The code is available on \\href{https://github.com/kkkzh/DLGStream}{https://github.com/kkkzh/DLGStream}.", "date": "2026-06-27", "categories": []}
{"id": "2606.29303", "title": "Occlusion-Robust Multi-Object Decoupling for Physics-Based Robotic Interaction", "abstract": "We propose a mask-free method for lossless multi-object 3D reconstruction from sparse and occluded real-world views, enabling physically plausible robotic interaction via Material Point Method (MPM) simulation. Our key insight is that object coupling stems from occlusion and limited viewpoints, which we address by formulating multi-object decoupling as a sparse-view reconstruction problem. Using 3D Gaussian Splatting as base representation, we first obtain coarse instance partitions with a SAM2-trained segmentation field. Rather than relying on masks, we reconstruct fragmented geometries by leveraging a joint Score Distillation Sampling (SDS) process, which integrates reference-view supervision with novel-view synthesis guided by 2D and 3D diffusion priors to enforce both texture fidelity and 3D consistency. Furthermore, we incorporate geometry-aware priors such as intra-object and inter-object similarity to regularize geometric reasoning. Experimental results demonstrate that our method produces complete, simulation-ready 3D objects without requiring manual masks, enabling realistic dynamic interactions on both synthetic, robotic and real-world datasets.", "date": "2026-06-28", "categories": []}
{"id": "2606.29513", "title": "Scenes as Objects, Not Primitives: Instance-Structured 3D Tokenization from Unposed Views", "abstract": "A 3D scene is understood through its objects, not the primitives that compose them. Yet feed-forward reconstruction methods output dense, unstructured sets of points or Gaussians, leaving object-level structure to be recovered after the fact. We propose a feed-forward framework that decomposes a scene into instance-structured 3D token groups directly from unposed multi-view images -- compact object-centric units from which reconstruction, segmentation, and manipulation all follow. Each token group pairs an instance token capturing entity-level identity with anchor tokens that encode local geometry and appearance, which are decoded into a set of 3D Gaussians. This two-level factorization decouples object identity from local appearance, making object instances a native interface of the representation rather than a derived product. The token groups are learned through differentiable rendering with joint reconstruction and segmentation supervision, requiring no 3D annotations. Our feed-forward model surpasses per-scene optimization baselines in class-agnostic instance segmentation while remaining competitive in novel view synthesis. Beyond these metrics, the same token groups directly unlock instance-level scene editing -- removing, translating, or inserting objects by operating on their groups -- as well as efficient open-vocabulary 3D instance retrieval, where retrieval complexity scales with the number of instances rather than primitives.", "date": "2026-06-28", "categories": []}
{"id": "2606.29794", "title": "UniTriSplat: A Unified 3D Gaussian Splatting Framework with Uniform Spherical Rasterization for Universal Cameras", "abstract": "Existing 3D Gaussian Splatting (3DGS) frameworks rely on camera-specific rasterization, suffering from inconsistent solid-angle sampling and degraded performance across heterogeneous camera models (e.g., perspective, fisheye, omnidirectional). To address this limitation, we propose UniTriSplat, a unified 3DGS framework for universal cameras that reformulates Gaussian splatting on the unit sphere via HEALPix discretization. Leveraging the equal-area property of HEALPix, we construct a spherical sampling grid aligned with the angular resolution of input images. We derive the forward rendering and gradient propagation of Gaussians directly in the spherical radian domain, yielding uniform optimization behavior from narrow-FoV images to full 360-degree panoramas. To enhance perceptual reconstruction quality, we additionally introduce a HEALPix-aware SSIM loss that respects spherical neighborhood structure. Extensive experiments across diverse camera models demonstrate that UniTriSplat consistently improves cross-camera generalization while preserving geometric fidelity and rendering quality.", "date": "2026-06-29", "categories": []}
{"id": "2606.29976", "title": "Learning Efficient 4D Gaussian Representations from Monocular Videos with Flow Splatting", "abstract": "Reconstructing dynamic 3D scenes from monocular videos is challenging due to scene complexity and temporal dynamics. With the advancement of 3D Gaussian Splatting in novel view synthesis, existing methods extend 3D Gaussians to 4D domain with deformation fields, trajectories or spatiotemporal 4D volumes to model scene element deformation. However, these methods suffer from long training time, low rendering speed or high memory consumption for per-frame reconstruction of 4D volumes, without fully exploiting dense dynamic information. To address this issue, we propose Flow Splatting, which constructs the velocity field and enables the conventional splatting technique to render optical flow from the velocity field to supervise dynamics learning process from monocular videos. Specifically, we extend 4D volumes with time varying means and covariance to represent complex dynamics. Then, we construct and approximate the velocity field naturally based on this representations. While conventional volume rendering techniques support to render color fields, we extend the volume rendering strategy to splat the velocity field by considering the influence of camera motions. We conduct experiments on various benchmarks to demonstrate the efficiency and effectiveness of our method. Compared to the state-of-the-art methods, our model achieves better image quality with less time consumption and higher rendering speed.", "date": "2026-06-29", "categories": []}
{"id": "2606.30014", "title": "Shell-Supervised Gaussian Splatting for Urban Real-to-Sim Reconstruction", "abstract": "Real-to-sim reconstruction for embodied AI requires geometry that is useful for collision reasoning, navigation, and agent-environment interaction, not only photorealistic novel-view synthesis. However, close-range urban facades are difficult for video-to-3D reconstruction: glass, reflections, repeated windows, and weak texture can produce visually plausible renderings with unstable surface geometry. We introduce shell-supervised Gaussian Splatting, a reconstruction-stage framework that uses an external facade structural shell as lightweight geometric supervision for video-driven Gaussian reconstruction. The method aligns an exterior shell to the video reconstruction frame, renders per-view depth, camera-space normal, and valid-mask maps, and applies these cues through mask-gated losses during Gaussian optimization. This design preserves RGB-driven appearance while regularizing only visible shell-supported facade regions. Experiments on anonymized close-range urban facade scenes show improved facade orientation and visible-surface point-cloud consistency over photo-only, monocular-cue, and surface-oriented Gaussian baselines, while maintaining comparable held-out rendering quality.", "date": "2026-06-29", "categories": []}
{"id": "2606.30047", "title": "Argus: Metric Panoramic 3D Reconstruction for Indoor Scenes", "abstract": "Metric feed-forward 3D reconstruction for panoramic data remains under-explored due to the lack of large-scale panoramic RGB-D training data. We present Realsee3D, a hybrid dataset of 10K indoor scenes (1K real, 9K synthetic) with 299K panoramic viewpoints and precise metric annotations, and Argus, a feed-forward network trained on it for metric panoramic 3D reconstruction. In the sparse unordered capture setting of Realsee3D, a poorly chosen coordinate anchor can cause global pose drift. Argus addresses this with a learned covisibility module that selects the geometrically optimal reference view to anchor the metric world frame. To further improve multi-task learning, we decompose the bidirectional pixel-to-world mapping into interpretable sub-steps with per-step supervision and cross-coordinate joint constraints, reinforcing geometric consistency across prediction branches. On the Realsee3D benchmark, Argus achieves state-of-the-art metric performance in camera pose estimation, depth estimation, and point cloud reconstruction. Project page: https://argus-paper.realsee.ai.", "date": "2026-06-29", "categories": []}
{"id": "2606.30352", "title": "FastPano3D: Feed-Forward Indoor Panoramic 3D Reconstruction from a Single Image", "abstract": "Recent advances in 3D scene reconstruction have highlighted the intricate trade-offs among rendering quality, inference efficiency, and data dependency. To address the challenge of rapidly reconstructing detailed 3D indoor scenes from minimal input, we introduce FastPano3D, an end-to-end framework that directly generates renderable 3D Gaussian representations from a single panoramic image. Unlike perspective-based methods, panoramic images inherently suffer from equirectangular projection distortions and spatially non-uniform feature distributions, making direct feed-forward Gaussian generation particularly challenging. In contrast to existing Gaussian Splatting based methods that rely on multi-view supervision or per-scene optimization, FastPano3D employs a lightweight feature encoder, adaptive Gaussian sampling, and a point-cloud-guided refinement strategy to achieve efficient and accurate scene generation without any test-time optimization. Our approach reconstructs high-fidelity 3D scenes within seconds, achieving up to 156 times faster inference than prior state-of-the-art methods such as Pano2Room, while using only half the parameters. Extensive experiments demonstrate that FastPano3D delivers rendering quality comparable to NeRF- and 3DGS-based reconstructions, establishing a new benchmark for rapid, single-view 3D scene inference.", "date": "2026-06-29", "categories": []}
{"id": "2606.30809", "title": "GaussLite: Online Task-Conditioned 3D Gaussian Splatting for Real-Time Robotic Mapping", "abstract": "Existing 3D Gaussian Splatting (3DGS) systems distribute representation capacity uniformly across a scene, ignoring the fact that many downstream robotic tasks engage only a fraction of the reconstructed geometry. This causes valuable onboard compute to be allocated towards optimizing irrelevant parts of the scene, either limiting online capacity or under-optimizing the most relevant parts of the scene. We introduce GaussLite, a task-driven 3DGS mapping system that conditions its representation density on a natural-language task specification. Given a posed RGB-D stream and a task such as \"prepare to pick up the object on the desk,\" GaussLite uses a one-shot LLM parser to extract target and anchor objects, which are grounded per-frame by an open-vocabulary detector and segmented to produce per-pixel relevance masks in real time. The mapper allocates seeding density, gradient flow and scaling by task relevance. At matched Gaussian budget and real-time mapping at 4 Hz on resource-constrained hardware, GaussLite outperforms baselines on ROI PSNR on the Replica Dataset by an average +2.72 dB and on a real-hardware demonstration in indoor and outdoor settings by +2.23 dB. We further show that two task-specialized agents' maps can be fused into a single shared map via per-voxel voting on active-optimization counts in real time, outperforming concatenation by +3.42 dB while only sharing an average 7.08% of the map.", "date": "2026-06-29", "categories": []}
{"id": "2606.31086", "title": "CasaMaestro: Multi-View Panoramas for House-Scale 3D Reconstruction", "abstract": "The rise of home-deployed embodied AI systems is driving a growing need for fast, metric 3D reconstruction of residential spaces to support navigation, interaction, and long-horizon task execution. However, the commonly used pinhole-camera 3D reconstruction pipelines struggle to model large indoor residences efficiently due to their limited field of view, to which achieving full coverage across multiple rooms often requires thousands of images and incurs drift from long chains of incremental alignment. In this work, we present CasaMaestro (Spanish words meaning ``house'' and ``master''), a feedforward model that can take only twenty to fifty sparse multi-view indoor panoramas as input and directly predicts metric depth along with camera poses, allowing fast point-cloud reconstruction of the entire house with full coverage. CasaMaestro is the first model that supports house-scale reconstruction with multi-view panoramas. Experiments show that CasaMaestro can robustly provide high quality results in both real-world and synthetic scenes, which can serve as a strong foundation for acquiring house-scale 3D indoor assets to be applied in close-loop simulation.", "date": "2026-06-30", "categories": []}
{"id": "2606.31388", "title": "One Video, One World: Turning Monocular Video into Physical 4D Scenes", "abstract": "We introduce \\textbf{OVOW}, the first training-free system that reconstructs \\emph{instance-level, simulation-ready} 4D mesh scenes from a single monocular video. Recent 4D reconstruction achieves impressive rendering quality, but its outputs (\\eg, implicit fields, Gaussian primitives, or point clouds) lack the watertight topology, instance separation, and standardized physical interfaces required by physics simulators and embodied AI. OVOW closes this gap with a four-stage pipeline: a vision-language model discovers, labels, and motion-classifies all instances; category-aware reconstruction yields per-instance meshes for rigid objects and topology-consistent mesh sequences for deformable ones; an iterative render-match-optimize procedure recovers metric scale and 6-DoF pose trajectories; and physics-grounded assembly enforces ground contact and inter-object support. Crucially, we model all motion, rigid and non-rigid, through direct vertex deformation without category-specific priors or skeleton rigging, producing watertight mesh scenes ready for downstream physics simulation and editing. We further establish the first benchmark for \\emph{structured Video-to-4D} evaluation, with metrics for geometric correctness, instance separation, and physical plausibility beyond visual fidelity; the same pipeline doubles as a scalable engine for \\emph{synthesizing} paired video-to-4D simulation data for future 4D world models and embodied AI. Across two synthetic benchmarks (static and 4D), OVOW attains the best overall layout and geometry accuracy and the lowest photometric and semantic error among all baselines, and on monocular video runs one to two orders of magnitude faster than the baselines, while downstream physics simulation confirms its physical stability.", "date": "2026-06-30", "categories": []}
{"id": "2606.31556", "title": "AugSplat: Radiance Field-Informed Gaussian Splatting for Sparse-View Settings", "abstract": "Generating high-quality novel views at real-time frame rates remains a central challenge in 3D vision, particularly in sparse-view scenarios. Neural radiance fields have demonstrated robust reconstruction from limited observations, but their reliance on volumetric rendering leads to high computational cost and slow inference. In contrast, Gaussian Splatting methods achieve real-time rendering through rasterization, but their optimization is highly sensitive to the quality of the initial geometry. This sensitivity becomes especially problematic in sparse-view settings, where limited observations often lead to incomplete or noisy point-cloud reconstructions. In this work, we present AugSplat, a simple framework for improving Gaussian Splatting in sparse-view regimes using radiance-field-based view augmentation. We first train a radiance field on the sparse input views and use it to synthesize additional images from nearby novel viewpoints, increasing the effective view-space coverage available for supervision. These synthetic views are then used as auxiliary supervision during Gaussian Splatting optimization. We study two variants: Staged AugSplat, which uses synthetic views for an initial optimization phase before switching to real images, and Dual AugSplat, which jointly trains on real and synthetic views with a decaying synthetic loss weight. Experiments on sparse-view mip-NeRF 360 scenes show that AugSplat improves reconstruction quality over standard Gaussian Splatting. Staged AugSplat achieves the strongest average performance, while Dual AugSplat provides a closely performing formulation that keeps real-image supervision active throughout training, and both variants preserve real-time rendering at inference.", "date": "2026-06-30", "categories": []}