{"id": "2503.01425", "title": "MeshPad: Interactive Sketch-Conditioned Artist-Reminiscent Mesh Generation and Editing", "abstract": "We introduce MeshPad, a generative approach that creates 3D meshes from sketch inputs. Building on recent advances in artist-reminiscent triangle mesh generation, our approach addresses the need for interactive mesh creation. To this end, we focus on enabling consistent edits by decomposing editing into 'deletion' of regions of a mesh, followed by 'addition' of new mesh geometry. Both operations are invoked by simple user edits of a sketch image, facilitating an iterative content creation process and enabling the construction of complex 3D meshes. Our approach is based on a triangle sequence-based mesh representation, exploiting a large Transformer model for mesh triangle addition and deletion. In order to perform edits interactively, we introduce a vertex-aligned speculative prediction strategy on top of our additive mesh generator. This speculator predicts multiple output tokens corresponding to a vertex, thus significantly reducing the computational cost of inference and accelerating the editing process, making it possible to execute each editing step in only a few seconds. Comprehensive experiments demonstrate that MeshPad outperforms state-of-the-art sketch-conditioned mesh generation methods, achieving more than 22% mesh quality improvement in Chamfer distance, and being preferred by 90% of participants in perceptual evaluations.", "date": "2025-03-03", "categories": []} {"id": "2508.15228", "title": "Collaborative Multi-Modal Coding for High-Quality 3D Generation", "abstract": "3D content inherently encompasses multi-modal characteristics and can be projected into different modalities (e.g., RGB images, RGBD, and point clouds). Each modality exhibits distinct advantages in 3D asset modeling: RGB images contain vivid 3D textures, whereas point clouds define fine-grained 3D geometries. However, most existing 3D-native generative architectures either operate predominantly within single-modality paradigms-thus overlooking the complementary benefits of multi-modality data-or restrict themselves to 3D structures, thereby limiting the scope of available training datasets. To holistically harness multi-modalities for 3D modeling, we present TriMM, the first feed-forward 3D-native generative model that learns from basic multi-modalities (e.g., RGB, RGBD, and point cloud). Specifically, 1) TriMM first introduces collaborative multi-modal coding, which integrates modality-specific features while preserving their unique representational strengths. 2) Furthermore, auxiliary 2D and 3D supervision are introduced to raise the robustness and performance of multi-modal coding. 3) Based on the embedded multi-modal code, TriMM employs a triplane latent diffusion model to generate 3D assets of superior quality, enhancing both the texture and the geometric detail. Extensive experiments on multiple well-known datasets demonstrate that TriMM, by effectively leveraging multi-modality, achieves competitive performance with models trained on large-scale datasets, despite utilizing a small amount of training data. Furthermore, we conduct additional experiments on recent RGB-D datasets, verifying the feasibility of incorporating other multi-modal datasets into 3D generation.", "date": "2025-08-21", "categories": []} {"id": "2606.20131", "title": "TriFlow: Generating Artist-Like 3D Mesh Topology via Nearest-Vertex Vector Fields", "abstract": "We present TriFlow, a new generative approach for producing compact 3D meshes with artist-like triangle topology directly from input geometry conditions such as signed distance fields. Our key insight is to represent mesh topology as a nearest-vertex vector field (NVF) defined over the surface, where each point encodes its association to the nearest triangle vertex in the local barycentric frame. We train a latent flow-matching model to synthesize this field, enabling topology generation conditioned on the input geometry. To extract a coherent mesh, we cluster surface regions using the generated NVF and guide a constrained quadric error metric (QEM) mesh simplification with topology-aware optimization. This yields output meshes that closely match the input geometry while exhibiting structured, artist-like connectivity. Experiments demonstrate that TriFlow achieves stronger generalization and significantly improved topology quality compared to state-of-the-art learning-based approaches, alongside 90% lower Chamfer Distance and an 8x speedup.", "date": "2026-06-18", "categories": []} {"id": "2606.20563", "title": "JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-Space Denoising", "abstract": "Creating 3D visual illusions, a single 3D mesh that reveals entirely different semantics from various viewing angles, is a fascinating but tough challenge. Existing optimization-based methods are slow and can produce oversaturated colors. In contrast, naive stitching approaches fail to produce geometrically coherent objects. This results in visible unnatural seams and semantic leaks. In this paper, we present a fast and training-free framework for generating text-driven 3D visual illusions. Our approach decouples the generation into two stages. First, we propose a cross-space dual-branch denoising process. This process dynamically decodes 3D latents into voxel space for CLIP-guided orientation alignment and Signed Distance Field (SDF) blending, which ensures seamless geometric fusion. Second, we introduce a view-conditioned texture synthesis module that projects and aggregates view-specific 2D diffusion priors onto the fused geometry. Extensive experiments demonstrate that our method generates highly realistic, dual-semantic 3D illusions in just 3-5 minutes. It significantly outperforms existing methods in geometric integrity, semantic recognizability, and efficiency. Project page: https://siang1105.github.io/JanusMesh.github.io/", "date": "2026-06-18", "categories": []} {"id": "2606.13376", "title": "MoVerse: Real-Time Video World Modeling with Panoramic Gaussian Scaffold", "abstract": "We present MoVerse, a real-time video world model that creates an interactively navigable scene from a single narrow-field-of-view image. This setting is challenging because the input observes only a small fraction of the environment, while interactive roaming requires a complete surrounding world, persistent geometry, controllable camera motion, and temporally coherent high-fidelity observations. MoVerse addresses this problem by separating world construction from observation rendering. It first expands the input into a gravity-aligned 360$^\\circ$ panorama with topology-aware diffusion, closing the missing field of view before 3D reasoning. It then lifts the panorama into a persistent 3D Gaussian scaffold using panoramic geometry-aware residual prediction, yielding a dense and directly renderable spatial memory. Finally, a Gaussian-conditioned video renderer translates scaffold renderings along user-specified camera trajectories into photorealistic video. To make this renderer practical for interaction, we train a bidirectional diffusion teacher for high-quality conditional rendering and distill it into a causal autoregressive student for bounded-latency streaming. This design combines the controllability and long-range consistency of explicit 3D representations with the perceptual quality of generative video models. MoVerse supports real-time scene roaming at 8~FPS on a single NVIDIA RTX~4090 GPU, demonstrating a practical path toward single-image world creation with interactive video output.", "date": "2026-06-11", "categories": []} {"id": "2506.24121", "title": "TextMesh4D: Zero-shot Text-to-4D Mesh Generation", "abstract": "Large-scale, high-quality dynamic 3D (4D) assets are essential for learning physically grounded representations, but remain costly to capture and annotate at scale. This limits the viability of supervised 4D learning and motivates zero-shot text-to-4D generation leveraging pretrained diffusion priors. To model complex dynamics, prior methods typically adopt implicit 3D representations (e.g., NeRFs or 3DGS) for their deformation capacity. However, their implicit nature provides limited control over surface topology, which hinders high-fidelity geometry and makes temporally coherent surface reconstruction challenging. To address these limitations, we explore zero-shot text-to-4D mesh generation. However, a structural mismatch arises when combining diffusion-based guidance with topology-constrained meshes: the guidance is noisy and spatially inconsistent, while meshes impose severe topological constraints, making direct vertex-level deformation unstable. In this paper, we introduce TextMesh4D, the first zero-shot framework for text-to-4D that directly generates dynamic meshes by addressing the above challenge at two complementary levels. Geometrically, we shift deformation modeling from vertices to faces via a Jacobian Deformation Field (JDF), enabling topology-aware surface reconstruction through an integrability-enforcing integration formulation. Semantically, we propose a Local-Global Semantic Regularizer (LGSR) that preserves identity over time by jointly constraining local deformation plausibility and global shape consistency. Extensive experiments demonstrate state-of-the-art temporal consistency, structural fidelity, and visual quality, while remaining efficient on a single 24GB GPU.", "date": "2025-06-30", "categories": []} {"id": "2605.07971", "title": "DVD: Discrete Voxel Diffusion for 3D Generation and Editing", "abstract": "We introduce Discrete Voxel Diffusion (DVD), a discrete diffusion framework to generate, assess, and edit sparse voxels for SLat (Structured LATent) based 3D generative pipelines. Although discrete diffusion has not generally displaced continuous diffusion in image-like generation, we show that it can be an effective first-stage prior for sparse voxel scaffolds. By treating voxel occupancy as a native discrete variable, DVD avoids continuous-to-discrete thresholding and provides a simple framework for voxel generation, uncertainty estimation, and editing. Beyond quality gains, DVD provides more interpretable generation dynamics through explicit categorical modeling. Furthermore, we leverage the predictive entropy as a robust uncertainty metric to identify ambiguous voxel regions and complicated samples, facilitating tasks such as data filtering and quality assessment. Finally, we propose a lightweight fine-tuning strategy using block-structured perturbation patterns. This approach empowers the model to inpaint and edit voxels within a single sampling round, requiring negligible auxiliary computation and no additional model evaluations. Code is available at https://github.com/TeCai/DVD.", "date": "2026-05-08", "categories": []} {"id": "2606.08402", "title": "SceneConductor: 3D Scene Generation from a Single Image with Multi-Agent Orchestration", "abstract": "Generating complete 3D scenes from a single image requires inferring globally consistent geometry, object relationships, and environmental context from inherently ambiguous visual evidence. Despite recent progress in joint layout-and-mesh generation, existing methods often rely on holistic or weakly decomposed pipelines that entangle many factors at once and demand extensive scene-level supervision, limiting their generalization to complex real-world environments. We propose a multi-agent orchestration framework that decomposes single-image 3D scene generation into three structured stages: scene initialization, environment construction, and multi-agent refinement. The initialization stage extracts image-derived object masks, builds object-level 3D representations, and predicts an initial spatial layout to form a coarse 3D scene. The environment-construction stage then leverages this initialization together with point-map geometry to build an environmental scaffold of supporting surfaces, room boundaries, materials, and illumination. Finally, in the refinement stage, a planner agent identifies structural and visual inconsistencies, applies simple corrections directly, and dispatches specialist agents for complex localized revisions that are reintegrated into the global scene. To provide reliable structural initialization while reducing reliance on scene-level annotations, we further introduce a geometry-aware layout predictor supervised by sparse geometric priors derived from point maps. Unlike fully supervised layout generators, the predictor can be trained from segmentation-level data and generalizes robustly to diverse real-world scenes. Extensive experiments on benchmark datasets show that our method consistently outperforms prior approaches in geometric accuracy, spatial consistency, and perceptual realism.", "date": "2026-06-07", "categories": []} {"id": "2606.17432", "title": "Edit3DGS: Unified Framework for Dynamic Head Editing via 2D Instruction-Guided Diffusion and 3D Gaussian Splatting", "abstract": "We present Edit3DGS, a unified framework for dynamic 3D head editing that integrates 2D instruction-guided diffusion with 3D Gaussian splatting. Unlike prior approaches that separately address frame-based edits or static 3D reconstruction, our method couples semantic controllability in the image domain with photorealistic, temporally consistent 3D representations. Given an input video, editable facial regions are masked and modified using a text-conditioned diffusion model to support fine-grained operations such as expression transformation, attribute modification, and appearance refinement. The edited frames are then aggregated through 3D Gaussian splatting to produce a coherent, high-fidelity avatar that preserves both identity and motion dynamics. To enforce consistency, Edit3DGS incorporates multi-view batch editing and lightweight inpainting strategies that recover lost expressions across timesteps. Experimental results demonstrate that our framework enables controllable, artifact-free head editing with smooth temporal transitions, offering practical applications in virtual avatars, immersive communication, film production, and interactive media.", "date": "2026-06-16", "categories": []} {"id": "2603.01371", "title": "TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity", "abstract": "Precise spatial fidelity in Image-to-3D multi-instance generation is critical for downstream real-world applications. Recent work attempts to address this by fine-tuning pre-trained Image-to-3D (I23D) models on multi-instance datasets, which incurs substantial training overhead and struggles to guarantee spatial fidelity. In fact, we observe that pre-trained I23D models already possess meaningful spatial priors, which remain underutilized as evidenced by instance entanglement issues. Motivated by this, we propose TIMI, a novel Training-free framework for Image-to-3D Multi-Instance generation that achieves high spatial fidelity. Specifically, we first introduce an Instance-aware Separation Guidance (ISG) module, which facilitates instance disentanglement during the early denoising stage. Next, to stabilize the guidance introduced by ISG, we devise a Spatial-stabilized Geometry-adaptive Update (SGU) module that promotes the preservation of the geometric characteristics of instances while maintaining their relative relationships. Extensive experiments demonstrate that our method yields better performance in terms of both global layout and distinct local instances compared to existing multi-instance methods, without requiring additional training and with faster inference speed.", "date": "2026-03-02", "categories": []} {"id": "2605.19876", "title": "Structural Energy Guidance for View-Consistent Text-to-3D Generation", "abstract": "Text-to-3D generation based on diffusion models often suffers from the Janus problem, leading to inconsistent geometry across viewpoints. This work identifies viewpoint bias in 2D diffusion priors as the main cause and proposes Structural Energy-Guided Sampling (SEGS), a training-free and plug-and-play framework to improve multi-view consistency. SEGS constructs a structural energy in the PCA subspace of U-Net features and injects its gradient into the denoising process. It can be easily integrated into SDS/VSD pipelines without retraining. Experiments show that SEGS reduces the Janus Rate by about 10% on average and improves View-CS scores across multiple baselines, including DreamFusion, Magic3D, and LucidDreamer. This method effectively alleviates viewpoint artifacts while preserving appearance fidelity, providing a flexible solution for high-quality text-to-3D content generation.", "date": "2026-05-19", "categories": []} {"id": "2606.04621", "title": "MeshFlow: Efficient Artistic Mesh Generation via MeshVAE and Flow-based Diffusion Transformer", "abstract": "We present MeshFlow, a new method for generating artist-like 3D meshes. Current mesh generators often adopt Auto-Regressive (AR) next-token prediction, a natural choice given the discrete nature of mesh topology. However, AR methods scale poorly because the inference cost is quadratic in mesh size. They also require discretizing the vertex coordinates, which introduces quantization errors. To address these challenges, we introduce a Variational Autoencoder (VAE) that, supervised with a contrastive loss, represents both continuous vertex positions and discrete connectivity in a continuous latent space. This latent space is significantly more compact than prior token-based mesh representations. We then build a 3D generator based on a Rectified Flow transformer, generating all mesh vertices and edges in parallel. Our model generates meshes 18x faster than the fastest AR generator while also achieving excellent accuracy across standard mesh-generation metrics. Homepage: https://mesh-flow.github.io/, Code: https://github.com/facebookresearch/meshflow", "date": "2026-06-03", "categories": []} {"id": "2605.21472", "title": "Stream3D: Sequential Multi-View 3D Generation via Evidential Memory", "abstract": "View-conditioned 3D generators such as SAM 3D, TRELLIS, and Hunyuan3D produce high-quality object reconstructions from a single view, but real-world visual observation often arrives as long monocular streams. Naively applying these generators to each streaming frame independently leads to severe temporal inconsistency in the generated results. To address this problem, we propose Stream3D, the first training-free streaming mechanism that turns a frozen view-conditioned 3D generator into a streaming generator with constant cross-chunk memory. Stream3D achieves this by maintaining a compact evidential memory, which selectively caches the most informative historical frames based on a proposed evidence score mechanism. As the stream progresses, the memory dynamically updates to retain a fixed number of informative frames, preventing the memory footprint from growing linearly with sequence length. This also prevents degradation over long sequences and keeps the underlying generator completely unchanged without retraining, architectural modifications, or auxiliary losses. Evaluated on both realistic and synthetic streaming benchmarks, Stream3D outperforms latent-transport baselines, including KV-cache reuse and flow-based feature editing, across both photometric and geometric metrics. More details can be found at: https://stream-3d.github.io/stream3d.github.io/.", "date": "2026-05-20", "categories": []} {"id": "2606.14168", "title": "MUSE: Agentic 3D Scene Authoring via Memory-Grounded Incremental Requirement Satisfaction", "abstract": "Text-driven 3D scene generation is a promising technique for digital content creation, embodied AI simulation, and interactive design, yet practical workflows often require refining, extending, or correcting existing scenes while preserving non-target content. Existing methods can produce realistic and structurally plausible scenes, but they generally lack editability with requirement-level state tracking, so part-level failures often lead to full-scene regeneration or manual intervention. To tackle this challenge, we formulate controllable 3D scene authoring as incremental requirement satisfaction, unifying construction and editing. In this paper, we present MUSE, a memory-grounded multi-agent framework in which an Architect compiles instructions into structured requirements, a Sculptor executes local scene operations, and an Inspector verifies each step while updating Working, Scene, and Skill Memory. To evaluate requirement-level controllability and preservation-aware editing, we introduce AuthorBench, offering 145 constrained construction cases and a 1,584-case preservation-aware editing pool paired with external structured checks. On full construction cases, MUSE improves All-Goal success from 37.9 to 80.7 and surface-constraint fulfillment from 35.0 to 92.6 over the strongest baseline. On a stratified 240-case editing test split, MUSE achieves 49.6 All-Goal success, 99.9 preservation rate, and only 0.6 unintended change rate. Beyond automated metrics, human evaluations on compared local-editing baselines support stronger alignment with user intent, and downstream navigation-proxy tests indicate stronger spatial stability. Combined with ablations validating our memory designs, these results establish MUSE as an effective framework for controllable 3D scene authoring.", "date": "2026-06-12", "categories": []} {"id": "2606.13345", "title": "JointEdit3D: Feed-Forward 3D Scene Editing in a Unified Latent Space", "abstract": "Existing 3D scene editing methods typically rely on per-scene optimization over explicit 3D representations or cascaded edit-and-reconstruct pipelines, resulting in high test-time cost, limited 3D awareness, and structural inconsistencies. To couple appearance synthesis and geometry prediction during editing, we build on a unified RGB-geometry reconstruction-generation latent space and adapt it to feed-forward 3D scene editing. The resulting framework, \\textbf{JointEdit3D}, performs asymmetric latent inpainting by observing only a single edited RGB reference latent and generating the remaining RGB views and edited geometry latent under source-scene anchoring. JointEdit3D introduces a dedicated SceneAnchor Branch to inject source-scene structure without forcing direct copying, and adopts edit/background-aware losses to balance edited-region fidelity with unedited-content preservation. To address the lack of paired resources for standardized 3D scene editing evaluation, we introduce SceneEdit3D-15K, a dataset with 15K paired editing samples and renderer-provided 3D annotations, together with SceneEdit3D-Bench, a curated 100-sample benchmark. Experiments show that JointEdit3D improves edited-region quality and 3D structural completeness over prior baselines while maintaining competitive background preservation.", "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": "2606.11446", "title": "3D-CBM: A Framework for Concept-Based Interpretability in Generative 3D Modeling", "abstract": "This research introduces a framework for incorporating Concept Bottleneck Models (CBMs) into 3D generative architectures to address the inherent 'semantic gap' in deep geometric learning. As deep models become central to 3D content creation, explainability shifts from a peripheral feature to a fundamental requirement for trust and accountability in safety-critical domains such as healthcare and manufacturing. CBMs provide an intrinsic interpretability solution by constraining latent representations to align with human-defined concepts, yet their application to unstructured 3D data remains largely unexplored.\n We design, implement, and validate a formal 3D-CBM architecture that maps raw geometric inputs, including point clouds and meshes, into a multi-tiered taxonomy of interpretable primitives and functional attributes. The framework further identifies strategic datasets, such as PartNet and ShapeNet, specialized for concept-based supervision. Experimental results from a 3D part-manipulation proof-of-concept experiment demonstrate the framework's efficacy, achieving a concept prediction accuracy of 88.8\\% and a Chamfer Distance of 0.0115. Critically, the model enables precise test-time intervention, allowing for the interactive correction of structural errors. This work establishes a foundation for semantically-steerable 3D generation and invites further exploration into collaborative human-in-the-loop design systems.", "date": "2026-06-09", "categories": []} {"id": "2606.11805", "title": "TextHOI-3D: Text-to-3D Hand-Object Interaction via Discrete Multi-View Generation and Joint Mesh Optimization", "abstract": "Text-conditioned 3D generation has progressed rapidly for images and isolated objects, but producing a hand-object mesh remains challenging: the output must preserve language semantics, cross-view consistency, object geometry, articulated hand shape, and physically plausible contact. We present TextHOI-3D, a staged framework that uses generated multi-view observations as an explicit interface between text-conditioned visual generation and geometry-aware hand-object recovery. TextHOI-3D learns a compact VQ token space for fixed-camera hand-object observations, predicts multi-view visual tokens from text with a CLIP-conditioned visual autoregressive model, and recovers a unified hand-object mesh through prior initialization, multi-view joint optimization, and anti-penetration refinement. The design separates semantic generation from geometric recovery while keeping both stages connected by a discrete multi-view representation. On HO3D-derived evaluations, the multi-view setting reduces object CD from 17.26 mm to 4.92 mm and penetration volume from 5.3721 cm^3 to 0.2193 cm^3 compared with a single-view counterpart, while improving hand errors and surface F-scores. These results support multi-view visual tokens as an effective intermediate representation for text-driven 3D hand-object mesh creation.", "date": "2026-06-10", "categories": []} {"id": "2606.12099", "title": "ISAP-3D: Identity-Slot Aligned Part-Aware 3D Generation", "abstract": "Part-aware 3D generation aims to synthesize structured objects with semantically meaningful components, yet often suffers from structural ambiguity due to identity-layout entanglement. Existing methods either infer part identity and spatial layout implicitly, which can lead to unstable part allocation (e.g., slot swapping or part merging), or rely on strong layout conditions that are difficult to obtain in practice. We attribute this ambiguity to identity-slot permutation freedom: without explicit identity-slot alignment, the correspondence between semantic parts and generation slots is not identifiable during training, allowing multiple slot assignments to fit the same supervision and leading to inconsistent decomposition. Based on this insight, we argue that stable part-aware generation requires identity-aligned one-to-one slot modelling. We therefore propose an identity-slot aligned framework, ISAP-3D, which anchors each part with semantic identity tokens and performs identity-conditioned one-to-one layout prediction, followed by layout-conditioned geometry synthesis. Structured local-global conditioning maintains identity alignment across semantic, spatial, and geometric stages. We also construct a part-level dataset with a unified semantic protocol to enable learnable and consistent identity-slot alignment. Extensive experiments demonstrate improved structural stability, controllability, and robustness over state-of-the-art part-aware generation baselines.", "date": "2026-06-10", "categories": []} {"id": "2606.09967", "title": "ABot-Earth 0.5: Generative 3D Earth Model", "abstract": "We present ABot-Earth 0.5, a generative 3D framework designed to synthesize vast, seamless 3D environments from ubiquitous, geospatially referenced satellite imagery. To achieve this, we propose a novel generative model formulated directly with the 3D Gaussian Splatting (3DGS) representation. The model is trained on a diverse corpus of existing real-world urban reconstructions, learning to generate realistic geometry and textures. At inference, it synthesizes novel 3D scenes conditioned solely on satellite imagery at a scalable rate of under 10 minutes per square kilometer, while demonstrating exceptional realism. The framework is designed for accessibility, with integrated hierarchical level-of-detail (LOD) structures that permit real-time, interactive visualization on web-based map engines. This high-fidelity simulation sandbox effectively mitigates the sim-to-real domain gap, enabling critical downstream Embodied AI applications like closed-loop UAV navigation. By providing an ultra-low-cost and high-efficiency solution, ABot-Earth 0.5 significantly lowers the technical and financial barriers to large-scale 3D reconstruction and empowers the future of global digital earth visualization.", "date": "2026-06-08", "categories": []} {"id": "2502.06819", "title": "AccioScene: Compositional 3D Scene Generation via Graph Diffusion and Interaction-driven Critics", "abstract": "This paper presents a framework for generating 3D indoor scenes from text prompts. Existing methods often formulate scene synthesis as an object layout prediction problem conditioned on a single input modality, such as a text description, room shape, or scene graph. This design can lead to object collisions and limited functional plausibility, reducing its practical applicability. To address these limitations, we introduce a multi-stage pipeline that better reflects practical scene creation scenarios. Given a text prompt describing partial scene content, our method first uses graph diffusion to produce a contextually coherent scene graph and then predicts a realistic object layout. In addition, we incorporate lightweight human-object interaction priors to encourage human-centric and functional arrangements, with explicit spatial constraints to reduce interpenetration. Our approach generates coherent 3D scenes with viable layouts that better support human interaction. Experiments on the 3D-FRONT dataset demonstrate that our method achieves competitive or state-of-the-art performance compared with existing approaches, while improving the physical plausibility of generated scenes.", "date": "2025-02-05", "categories": []} {"id": "2508.07011", "title": "HiMat: DiT-based Ultra-High Resolution SVBRDF Generation", "abstract": "Creating ultra-high-resolution spatially varying bidirectional reflectance functions (SVBRDFs) is critical for photorealistic 3D content creation, to faithfully represent fine-scale surface details required for close-up rendering. However, achieving 4K generation faces two key challenges: (1) the need to synthesize multiple reflectance maps at full resolution, which multiplies the pixel budget and imposes prohibitive memory and computational cost, and (2) the requirement to maintain strong pixel-level alignment across maps at 4K, which is particularly difficult when adapting pretrained models designed for the RGB image domain. We introduce HiMat, a diffusion-based framework tailored for efficient and diverse 4K SVBRDF generation. To address the first challenge, HiMat performs generation in a high-compression latent space via DC-AE, and employs a pretrained diffusion transformer with linear attention to improve per-map efficiency. To address the second challenge, we propose CrossStitch, a lightweight convolutional module that enforces cross-map consistency without incurring the cost of global attention. Our experiments show that HiMat achieves high-fidelity 4K SVBRDF generation with superior efficiency, structural consistency, and diversity compared to prior methods. Beyond materials, our framework also generalizes to related applications such as intrinsic decomposition.", "date": "2025-08-09", "categories": []} {"id": "2601.03256", "title": "Muses: Designing, Composing, Generating Nonexistent Fantasy 3D Creatures without Training", "abstract": "We present Muses, the first training-free method for fantastic 3D creature generation in a feed-forward paradigm. Previous methods, which rely on part-aware optimization, manual assembly, or 2D image generation, often produce unrealistic or incoherent 3D assets due to the challenges of intricate part-level manipulation and limited out-of-domain generation. In contrast, Muses leverages the 3D skeleton, a fundamental representation of biological forms, to explicitly and rationally compose diverse elements. This skeletal foundation formalizes 3D content creation as a structure-aware pipeline of design, composition, and generation. Muses begins by constructing a creatively composed 3D skeleton with coherent layout and scale through graph-constrained reasoning. This skeleton then guides a voxel-based assembly process within a structured latent space, integrating regions from different objects. Finally, image-guided appearance modeling under skeletal conditions is applied to generate a style-consistent and harmonious texture for the assembled shape. Extensive experiments establish Muses' state-of-the-art performance in terms of visual fidelity and alignment with textual descriptions, and potential on flexible 3D object editing. Project page: https://luhexiao.github.io/Muses.github.io/.", "date": "2026-01-06", "categories": []} {"id": "2606.08957", "title": "Rethinking 3D Shape Generation: Diffusion over Superquadrics", "abstract": "Diffusion models have advanced 3D shape generation, yet most methods still denoise in high-cardinality spaces (e.g., voxel/SDF grids, meshes, or point clouds), which is computationally and memory intensive and makes it difficult to scale in terms of both higher resolution and stronger controllability. We rethink the diffusion representation and propose to move diffusion from dense geometry to compact geometric primitives, representing each shape as a small set of superquadrics. Instead of operating on thousands to millions of geometric representation values, we leverage 7KB superquadric parameters (pose, size, and shape), drastically reducing diffusion-state dimensionality and per-step compute/memory. Our diffusion-over-superquadrics improves scalability by supporting broader capabilities (e.g., resolution-free point-cloud decoding, part-level editing, and constraint-based design) and achieving competitive surface-fidelity and distributional performance on standard benchmarks after point-cloud decoding, while enabling efficient generation within 0.6s per shape for most conditions.", "date": "2026-06-08", "categories": []} {"id": "2606.09187", "title": "CP4D: Compositional Physics-aware 4D Scene Generation", "abstract": "4D generation (\\textit{i.e.}, dynamic 3D generation) has recently emerged as a rapidly growing research frontier due to its powerful spatiotemporal modeling capabilities. However, despite notable advances, existing approaches typically fail to capture the underlying physical principles, producing results that are both physically inconsistent and visually implausible. To overcome this limitation, we present CP4D, a novel paradigm for photorealistic 4D scene synthesis with faithful adherence to complex physical dynamics. Drawing inspiration from the compositional nature of real-world scenes, where immutable static backgrounds coexist with dynamic, physically plausible foregrounds, CP4D reformulates 4D generation as the integration of a static 3D environment with physically grounded dynamic objects. On this basis, our framework follows a three-stage pipeline: \\textbf{1)} Firstly, we leverage pre-trained expert models to generate high-fidelity 3D representations of the environment and foreground objects respectively. \\textbf{2)} Subsequently, to produce physically plausible trajectories and realistic interactions for these objects, we propose a hybrid motion synthesis strategy that integrates priors from physical simulators with the common sense embedded in video diffusion models. \\textbf{3)} Finally, we develop an automated composition mechanism that seamlessly fuses the static environment and dynamic objects into coherent, physically consistent 4D scenes. Extensive experiments demonstrate that CP4D can generate explorable and interactive 4D scenes with high visual fidelity, strong physical plausibility, and fine-grained controllability, significantly outperforming existing methods. The project page: https://anonymous.4open.science/w/CP4D/.", "date": "2026-06-08", "categories": []} {"id": "2606.09273", "title": "EditSSC: Toward Editable Semantic Occupancy Scenes with Unconditional Diffusion Models", "abstract": "3D semantic scene generation is crucial for autonomous driving applications, yet most methods rely on complex 3D-specific architectures such as triplane encoders and adapted diffusion networks, limiting both their simplicity and their editing capabilities. We propose EditSSC, an editing-ready method for 3D semantic scene generation using 2D Bird's Eye View (BEV) representations and off-the-shelf latent diffusion network. Our approach reshapes 3D semantic occupancy grids into multi-channel BEV images and leverages the quantized autoencoder and UNet from Stable Diffusion with minimal modifications. We perform diffusion on the latents after quantization, which enables training-free editing capabilities. By exploiting class-to-code correspondences in the codebook, our method supports sketch-guided generation, inpainting, and outpainting without any retraining. On SemanticKITTI, EditSSC outperforms existing 3D-specific baselines on unconditional generation, demonstrating that well-established 2D architectures can be effectively repurposed for 3D scene generation and editing.", "date": "2026-06-08", "categories": []} {"id": "2606.09738", "title": "HDSL: A Hierarchical Domain-Specific Language for Structured 3D Indoor Scene Generation and Localized Editing with LLM Agents", "abstract": "Text-driven indoor scene generation and editing require an intermediate representation that language models can both produce and revise. Existing LLM-based systems often rely on scene graphs or global constraint lists, which are compact but underspecify local geometry and make instruction-based edits difficult to localize. We frame this problem as structured program generation and local program repair, and propose Hierarchical Descriptive Scene Language (HDSL), an XML/CSS-style domain-specific language for structured 3D indoor scenes. HDSL represents rooms, regions, objects, and support surfaces as a tree with local coordinates, making complex scenes easier to plan recursively and easier to retrieve for editing. Our pipeline uses LLM agents to generate HDSL subtrees with bounded verification, grounds non-virtual nodes through multimodal asset retrieval, and applies force-directed layout optimization to repair boundary and collision errors. For editing, Hierarchical Retrieval-Augmented Generation retrieves the relevant subtree, asks the LLM to rewrite only that local context, and merges the result back through a deterministic three-way merge. In our reproduced benchmark, HDSL improves average object coverage, text-scene alignment, and generation time over full text-to-scene baselines while remaining competitive with recent layout-only reproductions on geometry metrics; for editing, HRAG reduces token use by $5.22\\times$ and runtime by $6.19\\times$, produces valid DSL for all eight paired edits, and better preserves unrelated scene objects.", "date": "2026-06-08", "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": "2606.06887", "title": "ARAPDiffusion: ARAP Regularization for Diffusion-Based Deformable Shape Space Learning", "abstract": "This paper introduces ARAPDiffusion, a latent diffusion model to learn the underlying continuous shape space of a deformation shape collection. The key innovation is in injecting the as-rigid-as-possible (ARAP) deformation model as regularization losses into latent diffusion (LD), releasing the requirement of having abundant 3D training data for learning generative models. In contrast to the standard LD, we show how the ARAP model can be used to improve both the encoder/decoder and the LD model. The training procedure alternates between using the synthetic distribution defined by the LD model to develop a regularization loss that enhances the shape encoder/decoder and using the shape decoder to develop a regularization loss to improve the LD model. We also show the benefit of the LD paradigm in combining a representation-free LD process and an implicit shape decoder that is applicable to unorganized point clouds. The experimental results of unconditional and conditional shape generation demonstrate the advantages of ARAPDiffusion over baseline approaches.", "date": "2026-06-05", "categories": []} {"id": "2606.07024", "title": "GuideCAD: A Lightweight Multimodal Framework for 3D CAD Model Generation via Prefix Embedding", "abstract": "Multi-modal approaches used for 3D CAD generation require substantial computational resources, necessitating efficient training. To address this, we propose GuideCAD, which leverages semantically rich visual-textual representations having only a small number of trainable parameters to generate 3D CAD models. Specifically, GuideCAD uses a mapping network that converts image embeddings into prefix embeddings, enabling a pretrained large language model to integrate visual and textual information. As a result, a transformer-based decoder predicts the construction sequence using the visual-textual embeddings in order to generate the 3D CAD model. For experimental evaluation, we construct a new dataset, referred to as GuideCAD, which consists of text-image pairs. Each pair includes a text prompt that represents a 3D CAD construction sequence and its corresponding 3D CAD image. Our experimental results show that GuideCAD generates comparably high-quality 3D CAD models while using approximately four times fewer parameters and achieving twice the training efficiency compared to fine-tuning approaches. We have released the source code and dataset for our method at: https://github.com/mskimS2/GuideCAD", "date": "2026-06-05", "categories": []} {"id": "2606.07115", "title": "3DMorph: Single-Image-Guided Local 3D Shape Editing and Morphing", "abstract": "Despite recent progress in 3D generation, intuitive editing of existing shapes remains limited. Unlike images, which benefit from well-established inpainting tools, general 3D objects such as meshes still lack simple and effective methods for local shape editing. Existing approaches are often global, domain-specific, require complex user interaction, or focus on appearance (color and texture) rather than geometry. We introduce 3DMorph, a training-free framework for single-image-guided local 3D shape editing and morphing. Given an edited image showing a desired shape modification, our method automatically localizes the relevant 3D region and transfers 2D modifications to 3D while preserving unmodified areas. 3DMorph also enables intermediate shape generation between the original and edited objects, facilitating design exploration. To benchmark editing quality, we introduce Delta3D, an image-guided local 3D editing benchmark with paired ground-truth edits. Experimental results show that 3DMorph translates intuitive 2D edits into 3D, outperforming state-of-the-art generative and editing methods.", "date": "2026-06-05", "categories": []} {"id": "2606.07117", "title": "Native3D: End-to-End 3D Scene Generation via Unified Mesh-Texture Modeling and Semantic Alignment", "abstract": "This paper presents Native3D, the first end-to-end 3D scene generation framework that completely bypasses 2D intermediate representations. Traditional approaches typically require adapting 3D representations to the 2D domain to leverage pre-trained diffusion models, which inevitably introduces domain adaptation issues including geometric structural distortion and texture detail degradation. To address these limitations, we design a unified mesh-texture joint representation that simultaneously models both geometric structures and texture features through a Transformer-based scene encoder, effectively maintaining spatial relationships and visual consistency among objects within scenes. We further propose the 3D Representation Alignment Loss (3D REPA Loss), which employs an improved contrastive learning mechanism to align multi-level semantic representations in the latent space, significantly enhancing geometric and textural fidelity. Experimental results demonstrate that Native3D outperforms existing methods in both generation quality and editing flexibility, providing a novel solution for 3D scene editing.", "date": "2026-06-05", "categories": []} {"id": "2605.30819", "title": "Function2Scene: 3D Indoor Scene Layout from Functional Specifications", "abstract": "Most text-driven 3D indoor scene synthesis methods generate rooms from object-centric prompts, asking what furniture should be placed rather than how the space is used. Yet in real interior design, a layout is judged by how well it supports its occupants, e.g., their activities and physical needs. We introduce Function2Scene, a framework for generating 3D indoor layouts from functional specifications, i.e., natural-language design briefs describing who will use a room and what they need to do there. Given such a specification, our system parses occupant personas and activities, derives a customized set of functional design constraints from a taxonomy of 17 criteria spanning spatial, ergonomic, activity, and environmental considerations, and uses these constraints to guide layout generation. Rather than relying on an LLM to directly produce a final scene, Function2Scene performs iterative evaluation and refinement through a tool-augmented check-and-repair loop, combining geometric measurements, LLM-based contextual reasoning, and VLM-based visual assessment. Experiments on 30 professionally written interior-design cases show that Function2Scene produces layouts that better satisfy functional requirements than recent LLM-based scene synthesis baselines, with our results preferred in 94.3% of pairwise comparisons. Our work reframes text-driven indoor scene synthesis from placing plausible objects to designing spaces that support human use.", "date": "2026-05-29", "categories": []} {"id": "2511.16624", "title": "SAM 3D: 3Dfy Anything in Images", "abstract": "We present SAM 3D, a generative model for visually grounded 3D object reconstruction, predicting geometry, texture, and layout from a single image. SAM 3D excels in natural images, where occlusion and scene clutter are common and visual recognition cues from context play a larger role. We achieve this with a human- and model-in-the-loop pipeline for annotating object shape, texture, and pose, providing visually grounded 3D reconstruction data at unprecedented scale. We learn from this data in a modern, multi-stage training framework that combines synthetic pretraining with real-world alignment, breaking the 3D \"data barrier\". We obtain significant gains over recent work, with at least a 5:1 win rate in human preference tests on real-world objects and scenes. We will release our code and model weights, an online demo, and a new challenging benchmark for in-the-wild 3D object reconstruction.", "date": "2025-11-20", "categories": []} {"id": "2606.01649", "title": "PhyScene3D: Physically Consistent Interactive 3D Tabletop Scene Generation", "abstract": "Generating physically consistent 3D tabletop scenes is a fundamental yet underexplored problem for interactive and generalist robotic learning. The challenge stems from dense object hierarchies and irregular affordances. Here, an interactive scene denotes a physically valid, collision-free environment directly loadable into physics simulators. Existing methods, ranging from decoupled symbolic solvers to end-to-end regression models, often suffer from error propagation or overfitting to noisy supervision containing widespread physical violations. To address these limitations, we introduce PhyScene3D, a framework that reformulates generation as a Human-Mimetic Constructive Process. The proposed Cognitive Topological Reasoning Chain (CTRC) factorizes scene synthesis into a sequential, anchor-conditioned process. It employs a 3D AABB-based placement scheme that imposes a strong structural inductive bias. To address imperfect supervision and physical infeasibility, we introduce Physics-Aware Denoising Alignment (PADA). It integrates a differentiable Signed Distance Field (SDF) with Test-Time Optimization (TTO) to project generated scenes onto a physics-feasible manifold while preserving semantic intent. Experiments demonstrate that PhyScene3D outperforms state-of-the-art approaches in both semantic accuracy and physical validity, achieving a 40% reduction in scene-wise collision rate relative to the human-annotated training data.", "date": "2026-06-01", "categories": []} {"id": "2606.04108", "title": "SymTRELLIS: Symmetry-Enforced Voxel Latents for 3D Generation", "abstract": "Single-view 3D generative models have achieved impressive visual quality, yet they are not designed to satisfy structural or functional requirements, and in practice, often fall short. Symmetry is one such requirement: violations, even subtle ones, on symmetry can render a model physically unusable. We present SymTRELLIS, a method that enforces arbitrary finite point group symmetries (rotational, reflectional, and polyhedral) during the flow-based 3D generation of TRELLIS.2, without retraining the underlying VAE or flow model. Our key idea is to approximate the latent-space action of spatial transformations as a learned linear operator on voxel latents, implemented as a lightweight spatial-transform latent mapper trained on generic, non-symmetric 3D data. At generation time, we enforce symmetry by averaging predicted flow velocities across all symmetry-equivalent transformations at each ODE step, a process we call velocity symmetrization. The symmetry specification can be estimated automatically from an initial TRELLIS.2 generation or supplied by the user, enabling deliberate fold manipulation beyond what the input image suggests. On a curated benchmark of 266 strictly symmetric objects spanning 2- to 20-fold rotations and polyhedral symmetry groups, SymTRELLIS substantially reduces all symmetry error metrics compared to TRELLIS.2, Hunyuan3D-2.1, and TripoSG, while maintaining reconstruction accuracy comparable to the base model.", "date": "2026-06-02", "categories": []} {"id": "2606.04688", "title": "MeshWeaver: Sparse-Voxel-Guided Surface Weaving for Autoregressive Mesh Generation", "abstract": "Autoregressive mesh generation has gained attention by tokenizing meshes into sequences and training models in a language-modeling fashion. However, existing approaches suffer from two fundamental limitations: (i) low tokenization efficiency, which yields long token sequences and prevents scaling to high-poly meshes, and (ii) absence of geometry-aware guidance, as generation is conditioned only on global shape embeddings rather than local surface cues. We introduce MeshWeaver, an autoregressive framework that treats mesh generation as a surface weaving process by directly predicting the next vertex instead of independent coordinates. At its core is a multi-level sparse-voxel encoder that injects geometric context into the generative process in three complementary ways: providing voxel features as vertex representations, guiding token prediction via cross-attention to voxel features, and serving as a structural scaffold that constrains generation around the input surface. Our hierarchical design enables coarse-to-fine vertex prediction in a single decoding step, while tightly coupling the generative model with 3D geometry. Extensive experiments demonstrate that MeshWeaver achieves a state-of-the-art compression ratio of 18%, can generate meshes with up to 16K faces, and significantly improves geometric fidelity over prior approaches.", "date": "2026-06-03", "categories": []} {"id": "2606.05162", "title": "Controllable Dynamic 3D Shape Generation via 3D Trajectories and Text", "abstract": "We introduce T2Mo, a feed-forward framework for controllable dynamic 3D shape generation conditioned on 3D trajectories and text. Due to the inherent ambiguity of language, generating precisely intended motions using text alone remains challenging. To address this, we adopt 3D trajectories as controllable spatial guidance, specifying the exact paths along which selected points should move. By combining both, T2Mo generates object motions that spatially adhere to the given trajectories while globally reflecting the text semantics. To robustly handle trajectory inputs with arbitrary configurations, ranging from dense to sparse and unevenly distributed, we further propose a shape-grounded trajectory embedding that maps an input trajectory set into a shape-aware token set covering the entire object. We conduct extensive comparisons against text-based baselines and cascaded video-based baselines that combine trajectory-guided video generation with video-to-dynamic mesh generation. Quantitative and qualitative evaluations, along with user studies, demonstrate that our approach produces motions that more faithfully follow the given prompts with higher expressiveness while preserving motion quality.", "date": "2026-06-03", "categories": []} {"id": "2510.22491", "title": "LAMP: Data-Efficient Linear Affine Weight-Space Models for Parameter-Controlled 3D Shape Generation and Extrapolation", "abstract": "Generating high-fidelity 3D geometries under explicit parameter constraints is central to engineering design, yet current methods often require large datasets and fail to provide reliable control beyond the training distribution. We introduce LAMP, a data-efficient framework for controllable and interpretable 3D generation that aligns signed distance function (SDF) decoders by overfitting each exemplar from a shared initialization, then generates new designs by solving a parameter-constrained affine mixing problem in the aligned weight space. To improve reliability, we propose a linearity-mismatch safety metric that detects when mixed decoders leave the valid local regime. We evaluate LAMP on DrivAerNet++, BlendedNet, and additional industry-level vehicle families, including sports cars, SUVs, and convertibles. LAMP enables controlled interpolation with as few as 50 samples, safe extrapolation up to 100% beyond training ranges, and performance-guided optimization under fixed parameters, outperforming conditional autoencoder and Deep Network Interpolation (DNI) baselines in extrapolation, data efficiency, and parameter fidelity. Our results demonstrate that LAMP advances controllable, data-efficient, and safe 3D generation for design exploration, dataset generation, and performance-driven optimization.", "date": "2025-10-26", "categories": []} {"id": "2605.16813", "title": "QuadLink: Autoregressive Quad-Dominant Mesh Generation via Point-Relation Learning", "abstract": "The generation of production-ready quad-dominant meshes is a cornerstone of modern 3D content creation. Generating anisotropic quad-dominant meshes from point clouds is challenging, as existing methods are typically limited to producing either pure triangular meshes or pure quadrilateral meshes with isotropic densities. In this paper, we present QuadLink, a unified framework consisting of three stages for quad-dominant mesh generation by linking points into structured faces. QuadLink formulates polygonal mesh generation as a hybrid centroid-conditioned vertex linking model: it first predicts a unified set of anchors (vertices and face centroids), then learns centroid-conditioned links that associate vertices with face centroids, and finally assembles polygonal faces with a quad-first strategy guided by robust geometric verification strategies. This link-based formulation enables efficient generation of sparse and anisotropic quad-dominant meshes with coherent edge flow and meanwhile supporting hybrid polygonal topology. To construct training data for this model, we further introduce a Tri-to-Quad Operator that converts artistic triangle meshes into quad-dominant training data via global merge selection. Extensive experiments show that QuadLink produces production-ready quad-dominant meshes from point clouds and achieves improved geometric fidelity and topological quality compared to prior baselines. Our method natively supports hybrid polygonal topology, generalizing to arbitrary n-gon meshes without architectural changes.", "date": "2026-05-16", "categories": []} {"id": "2503.06136", "title": "GSV3D: Gaussian Splatting-based Geometric Distillation with Stable Video Diffusion for Single-Image 3D Object Generation", "abstract": "Image-based 3D generation has vast applications in robotics and gaming, where high-quality, diverse outputs and consistent 3D representations are crucial. However, existing methods have limitations: 3D diffusion models are limited by dataset scarcity and the absence of strong pre-trained priors, while 2D diffusion-based approaches struggle with geometric consistency. We propose a method that leverages 2D diffusion models' implicit 3D reasoning ability while ensuring 3D consistency via Gaussian-splatting-based geometric distillation. Specifically, the proposed Gaussian Splatting Decoder enforces 3D consistency by transforming SV3D latent outputs into an explicit 3D representation. Unlike SV3D, which only relies on implicit 2D representations for video generation, Gaussian Splatting explicitly encodes spatial and appearance attributes, enabling multi-view consistency through geometric constraints. These constraints correct view inconsistencies, ensuring robust geometric consistency. As a result, our approach simultaneously generates high-quality, multi-view-consistent images and accurate 3D models, providing a scalable solution for single-image-based 3D generation and bridging the gap between 2D Diffusion diversity and 3D structural coherence. Experimental results demonstrate state-of-the-art multi-view consistency and strong generalization across diverse datasets. The code will be made publicly available upon acceptance.", "date": "2025-03-08", "categories": []} {"id": "2605.28995", "title": "GAP3D: Generative Alignment of VLM Latents to Patch-Level Embeddings for 3D Generation", "abstract": "Recent approaches integrating vision-language models (VLMs) as prompt encoders for generative model conditioning typically rely on expensive end-to-end training or map features to compressed representations, discarding the dense spatial structure required for geometry-aware tasks like 3D asset generation. To address this, we propose GAP3D, a modular, diffusion-based approach that aligns VLM-generated latents directly to the complete, patch-level feature space of a pre-trained image encoder, enabling a frozen downstream generative model to utilize a VLM as prompt encoder while maintaining a spatially structured conditioning signal. Evaluated on 3D asset generation, our method bypasses the need for large-scale 3D data by training mainly on general-domain image-text pairs. It also exhibits emergent zero-shot capabilities for multimodal prompts, despite being trained exclusively on text input. Finally, while currently prioritizing high-level semantics over fine-grained detail, GAP3D demonstrates that the representation gap between VLM and image-encoder feature spaces can be partially bridged through diffusion-based alignment, taking the first steps towards a modular integration of foundation models through generative alignment to dense embedding spaces.", "date": "2026-05-27", "categories": []} {"id": "2606.01021", "title": "Learning Neural Deformation Representation for 4D Dynamic Shape Generation", "abstract": "Recent developments in 3D shape representation opened new possibilities for generating detailed 3D shapes. Despite these advances, there are few studies dealing with the generation of 4D dynamic shapes that have the form of 3D objects deforming over time. To bridge this gap, we focus on generating 4D dynamic shapes with an emphasis on both generation quality and efficiency in this paper. HyperDiffusion, a previous work on 4D generation, proposed a method of directly generating the weight parameters of 4D occupancy fields but suffered from low temporal consistency and slow rendering speed due to motion representation that is not separated from the shape representation of 4D occupancy fields. Therefore, we propose a new neural deformation representation and combine it with conditional neural signed distance fields to design a 4D representation architecture in which the motion latent space is disentangled from the shape latent space. The proposed deformation representation, which works by predicting skinning weights and rigid transformations for multiple parts, also has advantages over the deformation modules of existing 4D representations in understanding the structure of shapes. In addition, we design a training process of a diffusion model that utilizes the shape and motion features that are extracted by our 4D representation as data points. The results of unconditional generation, conditional generation, and motion retargeting experiments demonstrate that our method not only shows better performance than previous works in 4D dynamic shape generation but also has various potential applications.", "date": "2026-05-31", "categories": []} {"id": "2606.01057", "title": "3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code", "abstract": "Procedural 3D modeling through code is emerging as a versatile paradigm, offering deterministic, engine-ready, and precisely editable assets that neural 3D generators inherently lack. Authoring such procedural content, however, demands deep expertise in 3D software APIs, parametric design, and code-level geometric reasoning. In this paper, we propose 3DCodeBench, a systematic benchmark for evaluating vision-language model (VLM) agents for procedural 3D generation in 3D modeling software. Specifically, 3DCodeBench evaluates how effectively 12 advanced VLMs can serve as procedural 3D modelers by translating text and image references into procedural code for 3D modeling software. Recognizing that automated metrics may not fully capture the perceptual quality of 3D shapes, we build 3DCodeArena, a ranking platform based on pairwise human preferences over generated 3D outputs. From extensive evaluations and results, we observe that: (1) Failures mostly arise from API mismatches, while successful renders still suffer from disconnected or floating 3D geometric components. (2) Test-time scaling, such as higher thinking budgets and multi-turn refinement, improves performance overall. Our findings highlight a critical need for high-quality procedural coding data to advance commercial VLMs. Furthermore, effective procedural 3D modeling requires a robust execution environment that provides high-fidelity feedback for iterative refinement. We release 3DCodeBench, including the curated large-scale dataset of multimodal (text/image) prompts, procedural code, 3D object triplets, evaluation protocol, and the public 3DCodeArena platform as a foundational toolkit for exploring VLM-based procedural 3D modelers.", "date": "2026-05-31", "categories": []} {"id": "2606.02491", "title": "MORPHOS: Autoregressive 4D Generation with Temporal Structured Latents", "abstract": "We present MORPHOS, a novel autoregressive framework that generates dynamic 3D assets from videos across diverse representations, including meshes, 3D Gaussians, and radiance fields. Existing methods are typically limited to a single representation, struggle to model topological changes, or fail to maintain temporal consistency over long videos. To address these limitations, we introduce the Temporal Structured Latents (T-SLAT), a unified 4D representation that jointly encodes geometry and appearance along the temporal dimension. Leveraging T-SLAT, MORPHOS autoregressively generates dynamic 3D assets via causal attention, conditioning each frame on its preceding history to ensure temporal consistency while handling evolving topologies. We also propose a temporal-structural augmentation to mitigate error accumulation in autoregressive generation. MORPHOS achieves state-of-the-art performance in appearance and competitive results in geometry across multiple benchmarks, demonstrating superior generalization across various representations and robustness in long-horizon generation.", "date": "2026-06-01", "categories": []} {"id": "2606.02510", "title": "Not All Points Are Equal: Uncertainty-Aware 4D LiDAR Scene Synthesis", "abstract": "Constructing faithful 4D worlds from LiDAR-acquired sequences is crucial for embodied AI, yet current generative frameworks apply uniform modeling capacity across all spatial regions. This ignores that perceptual difficulty varies dramatically within a single scan: distant surfaces, occluded boundaries, and small-scale objects carry far higher uncertainty than well-observed structures. We present U4D, a new framework that explicitly leverages spatial uncertainty to guide LiDAR scene generation in a \"hard-to-easy\" schedule. U4D derives per-point uncertainty maps via Shannon Entropy from a pretrained segmentor, then applies an unconditional diffusion stage to synthesize high-entropy areas with precise geometry, followed by a conditional completion stage that fills in the remaining regions using these structures as priors. A MoST (Mixture of Spatio-Temporal) block further maintains cross-frame coherence by dynamically balancing spatial detail and temporal continuity. Extensive experiments on nuScenes and SemanticKITTI demonstrate state-of-the-art scene fidelity, temporal consistency, and downstream performance.", "date": "2026-06-01", "categories": []} {"id": "2605.29655", "title": "SuperVoxelGPT: Adaptive and Ordered 3D Tokenization for Autoregressive Shape Generation", "abstract": "Autoregressive multimodal large language models (MLLMs) enable 3D generation but struggle to scale to high-resolution shapes due to inadequate 3D tokenizations. Compact set-based representations discard deterministic spatial ordering, leading to ambiguous sequence prediction, while uniform or octree-based voxel grids preserve ordering at the cost of severe redundancy and excessively long sequences. This structural trade-off limits stable and efficient autoregressive 3D generation. We present SuperVoxelGPT, a representation-first framework that resolves this tension through adaptive and deterministically ordered supervoxel tokenization. Given a prompt, we first predict a coarse geometric saliency distribution and construct a shape-adaptive supervoxel partition using saliency-guided centroidal Voronoi tessellation, allocating fine-grained cells to complex regions and larger cells to smooth regions. Conditioned on the text and ordered supervoxel layout, we introduce a SuperVoxelVAE and fine-tune a pretrained MLLM to autoregressively generate supervoxel tokens. Experiments on Trellis-500K show that SuperVoxelGPT reduces token sequence length to 12.8% of uniform voxel tokenization while achieving state-of-the-art generation quality and an average 10$\\times$ speedup over prior methods.", "date": "2026-05-28", "categories": []} {"id": "2605.30268", "title": "PhyGenHOI: Physically-Aware 4D Generation of Dynamic Human-Object Interactions", "abstract": "We address the task of generating physically accurate and visually faithful 4D Human-Object Interaction (HOI). Given a static 3D human and target object represented as 3D Gaussian Splats (3DGS), our goal is to synthesize dynamic scenes where the human actively engages with the object through actions, such as punching or kicking, in accordance with a given input text. To this end, we introduce PhyGenHOI, a novel framework that couples generative human motion with an explicit physical object simulation. We model the human as a semantic agent driven by a Motion Diffusion Model (MDM) and the object as a physical agent simulated via the Material Point Method (MPM), utilizing 3D Gaussians as a unified, differentiable representation. We supervise their interaction through three coupled mechanisms: (1) A Windowed Attraction Loss that temporally synchronizes generative motion to intercept the object; (2) A Contact-Driven Re-simulation step that triggers physically consistent momentum transfer upon impact; and (3) A Masked Video-SDS objective that injects video-based priors to enhance contact fidelity. Experiments show PhyGenHOI generates physically consistent 4D HOI across diverse actions, humans, and objects, outperforming baselines. Project page and videos: https://omerbenishu.github.io/PhyGenHOI/", "date": "2026-05-28", "categories": []} {"id": "2501.04144", "title": "Chirpy3D: Part-Aware Multi-View Diffusion for Creative Fine-Grained Object Generation", "abstract": "Understanding and generating the fine-grained structure of objects -- such as birds with species-specific beaks, wings, and tails -- is a long-standing challenge in computer vision. We propose Chirpy3D, a part-aware multi-view diffusion framework that learns a hierarchical part latent space from unposed 2D images, using only off-the-shelf 2D part segmentation masks as spatial guidance -- without requiring any 3D data, camera poses, or manual part annotations. This latent space enables intuitive part-level swapping, interpolation, and zero-shot composition. A self-supervised feature consistency loss further encourages structural alignment across views, allowing coherent generation even with hybrid or unseen part combinations. Our core contribution is the controllable part-aware latent space and multi-view diffusion model. Downstream 3D generation is supported via any differentiable renderer such as NeRF but is orthogonal to the main framework, making Chirpy3D a flexible foundation for creative object generation in the absence of structured 3D data. Code is released at https://github.com/kamwoh/chirpy3d.", "date": "2025-01-07", "categories": []} {"id": "2508.18271", "title": "ObjFiller3D: Scaling 3D Object Inpainting to Dense Multi-View Consistency", "abstract": "3D object inpainting is commonly achieved via multi-view 2D image completion, yet independently inpainted views often suffer from cross-view inconsistencies, leading to blurred textures, geometric discontinuities, and visual artifacts in the reconstructed 3D objects. To overcome these limitations, we propose ObjFiller-3D, a novel method designed for the completion and editing of high-quality and consistent 3D objects. Instead of relying on sparse-view editing or per-view 2D inpainting, our method jointly optimizes a sequence of densely sampled views along a $360^\\circ$ trajectory, enabling global coherence across viewpoints. We design a new framework with three complementary components: a Temporal-Driven Generative Encoder for modeling dense-view dependencies, a Semantic-Aware Completion Encoder for object-level inpainting, and a Cycle-Consistent 3D Encoder that enforces global coherence through a closed-loop formulation. Our framework also supports reference-guided 3D inpainting, allowing fine-grained control over appearance. Extensive experiments on diverse datasets demonstrate that ObjFiller-3D significantly outperforms prior methods, achieving higher reconstruction fidelity (PSNR 26.6 vs.\\ 15.9 of NeRFiller) and perceptual quality (LPIPS 0.19 vs.\\ 0.25 of Instant3dit), while reducing reconstruction time from over 40 minutes to under 10 minutes. These results highlight the effectiveness and practical potential of our approach for real-world 3D editing applications. Project page: https://objfiller3d.github.io/ Code: https://github.com/objfiller3d/ObjFiller-3D .", "date": "2025-08-25", "categories": []} {"id": "2603.05425", "title": "RelaxFlow: Text-Driven Amodal 3D Generation", "abstract": "Image-to-3D generation faces inherent semantic ambiguity under occlusion, where partial observation alone is often insufficient to determine object category. In this work, we formalize text-driven amodal 3D generation, where text prompts steer the completion of unseen regions while strictly preserving input observation. Crucially, we identify that these objectives demand distinct control granularities: rigid control for the observation versus relaxed structural control for the prompt. To this end, we propose RelaxFlow, a training-free dual-branch framework that decouples control granularity via a Multi-Prior Consensus Module and a Relaxation Mechanism. Theoretically, we prove that our relaxation is equivalent to applying a low-pass filter on the generative vector field, which suppresses high-frequency instance details to isolate geometric structure that accommodates the observation. To facilitate evaluation, we introduce two diagnostic benchmarks, ExtremeOcc-3D and AmbiSem-3D. Extensive experiments demonstrate that RelaxFlow successfully steers the generation of unseen regions to match the prompt intent without compromising visual fidelity.", "date": "2026-03-05", "categories": []} {"id": "2605.26391", "title": "Garment Particles: A 2D--3D Symmetric Garment Representation for Generation and Editing", "abstract": "Practical garment design spans two modes: intuitive creation from high-level intent, such as a reference image or text description, and complex low-level editing across 2D sewing patterns and 3D draped geometry, which requires professional training to navigate their complex interdependencies. Yet existing frameworks address only part of this challenge, offering either garment generation from casual inputs or direct editing on sewing patterns. To support both ends of the spectrum, we propose Garment Particles, a 5D point-cloud representation that jointly encodes 2D sewing patterns and 3D geometry. This representation enables Garment Particles Flow (GPF), a rectified flow framework that supports intuitive generation from high-level inputs (text, images, sketches) and various editing operations on 2D sewing patterns and 3D geometries via diffusion posterior sampling. Finally, we introduce Particles-to-Pattern Flow that converts generated garment particles into curved-based patterns for simulation. We validate our model's generation ability on multiple datasets, achieving state-of-the-art garment generation results against competitive baselines. Our model also enables many garment editing scenarios, including garment interpolation, sewing pattern editing, point-cloud- and silhouette-conditioned garment generation. Our project website is at https://garment-particles.github.io .", "date": "2026-05-25", "categories": []} {"id": "2605.27351", "title": "Feedforward 3D Editing Learns from Semantic-Part Transformation", "abstract": "3D editing is a fundamental capability for scalable 3D content creation. While image editing has rapidly evolved toward large-scale feedforward generative paradigms, 3D AI generation remains dominated by training-free editing pipelines. A central challenge of feedforward 3D editing lies in the lack of high-quality paired supervision. Editable 3D assets require simultaneous preservation of geometry, multi-view consistency, structural coherence, and localized edit controllability. Existing 3D editing datasets often rely on independently generated assets, image-mediated reconstruction or narrow edit taxonomies, leading to inaccurate localization, weak preservation, blurred edit boundaries, and limited semantic consistency. In this work, we introduce a new perspective: scalable feedforward 3D editing should be learned from semantic-part transformations. Based on this insight, we propose Pxform, a high-quality 3D editing dataset with over 100K consistent before/after editing pairs across seven edit types. Instead of treating objects as unstructured shapes, our pipeline grounds edits directly in semantic 3D parts. Built upon Pxform, we further propose PartFlow, a feedforward 3D editing network that injects source-aware latent control into pretrained 3D generative priors. PartFlow introduces mask-aware velocity preservation and render-space consistency supervision to jointly improve edit fidelity and source preservation, while requiring no 3D edit mask during inference. Extensive experiments demonstrate that high-quality semantic-part supervision substantially improves scalable 3D editing, enabling PartFlow to achieve state-of-the-art performance on both geometric and appearance editing benchmarks.", "date": "2026-05-26", "categories": []} {"id": "2605.28763", "title": "CubePart: An Open-Vocabulary Part-Controllable 3D Generator", "abstract": "Interactive 3D assets used in games and simulation are typically decomposed into specific semantic parts to support animation, physics, and scripted behaviors, yet most generative 3D models produce either monolithic meshes or arbitrary part decompositions that cannot be aligned with application-specific requirements. We present CubePart, a generative framework for open-vocabulary, part-controllable 3D mesh generation that exposes part structure as an explicit inference-time control signal. Given a global text prompt and a user-defined parts schema expressed as an open-ended list of part names, our method generates a set of meshes - one per schema element - that assemble into a coherent object while respecting the specified semantic structure. To enable this capability, we introduce a scalable data pipeline to construct a large open-vocabulary, part-labeled 3D dataset, along with a two-stage generative architecture that separates global shape synthesis from part-level decoding. We demonstrate that the resulting assets can be directly integrated into game engines and driven by animation and behavior scripts without manual post-processing. Project Page: https://cubepart.github.io/", "date": "2026-05-27", "categories": []} {"id": "2605.26137", "title": "AssetGen: Deployable 3D Asset Generation at Interactive Speed", "abstract": "While 3D generation is progressing rapidly, recent work has often focused on obtaining high-resolution assets, leaving user experience and deployability as afterthoughts. We present AssetGen, a 3D generator that focuses instead on these two aspects. Given one reference image, in 30 seconds it produces a high-quality mesh with baked normals, a color texture, and a controlled polygon budget suitable for real-time rendering, including mobile use cases. The AssetGen Flash variant further reduces latency to 14 seconds for interactive and agentic creation loops. Our model generates the object geometry with a coarse-to-refine VecSet framework, which implements mesh simplification, cleaning, and normal baking on the GPU, and a fast parallel UV unwrapping. It then generates textures in a multi-view fashion, followed by backprojection and 3D inpainting. Model distillation, kernel optimization, and pipeline parallelization are co-designed to accelerate the system end-to-end. We introduce numerous automated and blind human evaluations and demonstrate competitive visual quality against leading commercial solutions in 30 seconds and preview-quality results in less than 15 seconds. The final result is a system that supports AI-assisted, deployable 3D content creation in interactive workflows.", "date": "2026-05-22", "categories": []} {"id": "2605.26149", "title": "AnySurf: Any Surface Generation with Directed Edge", "abstract": "Open surface components prevail in real industrial 3D content and support rendering, physical simulation and geometric editing. Garments serve as a typical open surface type, with numerous existing generation methods leveraging sewing patterns to generate 2D panels and stitch them into 3D shapes. Such domain-specific designs lack scalability and cannot generalize to shoes and accessories. Common field-based 3D generators prioritize watertight meshes and tend to create flawed double-layer structures on open surfaces. Though Trellis2 adopts field-free representation, its open surface results still contain normal and topology errors. We present AnySurf, a unified framework generating open, closed and hybrid 3D surfaces with accurate face orientation. Built on directed-edge enhanced Flexible Dual Grid (FDG-D), our representation retains normal direction information via oriented grid edges. We also propose ROS-FT post-training and a lightweight DE-Adapter with merely 1% extra parameters, facilitating directed edge learning while preserving original generation performance. We further construct Outfit3D dataset containing industrial garments and closed accessories. Our work transforms garment modeling into a universal 3D generation task. Experimental results demonstrate superior mesh quality and better practicality for downstream applications.", "date": "2026-05-23", "categories": []} {"id": "2605.26182", "title": "BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenization", "abstract": "Generating physically buildable brick structures from 3D shapes requires more than geometric reconstruction: the output must also satisfy discrete part constraints and structural stability. Existing brick generation methods either rely on heuristic optimization, which can break down when the target 3D shape does not admit a feasible structure under predefined constraints, or generate brick sequences without explicitly modeling the underlying 3D geometry and assembly relations. In this work, we present BrickAnything, a geometry-conditioned autoregressive framework for generating buildable brick structures from diverse 3D representations. BrickAnything uses point clouds as a unified geometric interface and predicts brick sequences that reconstruct the target shape under assembly constraints. To model structural dependencies among bricks, we introduce a structure-aware tree tokenization, which represents brick structures through local attachment relations. This formulation makes sequence generation more consistent with the physical construction process, and reduces invalid intermediate states. We further introduce preference-based alignment post-training, validity-constrained decoding and adaptive rollback to improve buildability objectives such as stability and geometric fidelity. Extensive experiments demonstrate that BrickAnything produces geometrically faithful and physically realizable brick structures, and that the proposed tokenization effectively reduces rollback and regeneration compared with conventional ordering strategies.", "date": "2026-05-25", "categories": []} {"id": "2506.19117", "title": "PrITTI: Primitive-based Generation of Controllable and Editable 3D Semantic Urban Scenes", "abstract": "Existing approaches to 3D semantic urban scene generation predominantly rely on voxel-based representations, which are bound by fixed resolution, challenging to edit, and memory-intensive in their dense form. In contrast, we advocate for a primitive-based paradigm where urban scenes are represented using compact, semantically meaningful 3D elements that are easy to manipulate and compose. To this end, we introduce PrITTI, a latent diffusion model that leverages vectorized object primitives and rasterized ground surfaces for generating diverse, controllable, and editable 3D semantic urban scenes. This hybrid representation yields a structured latent space that facilitates object- and ground-level manipulation. Experiments on KITTI-360 show that primitive-based representations unlock the full capabilities of diffusion transformers, achieving state-of-the-art 3D scene generation quality with lower memory requirements, faster inference, and greater editability than voxel-based methods. Beyond generation, PrITTI supports a range of downstream applications, including scene editing, inpainting, outpainting, and photo-realistic street-view synthesis. The source code and more results can be found at https://raniatze.github.io/pritti/.", "date": "2025-06-23", "categories": []} {"id": "2510.15264", "title": "DriveGen3D: Boosting Feed-Forward Driving Scene Generation with Efficient Video Diffusion", "abstract": "We present DriveGen3D, a novel framework for generating high-quality and highly controllable dynamic 3D driving scenes that addresses critical limitations in existing methodologies. Current approaches to driving scene synthesis either suffer from prohibitive computational demands for extended temporal generation, focus exclusively on prolonged video synthesis without 3D representation, or restrict themselves to static single-scene reconstruction. Our work bridges this methodological gap by integrating accelerated long-term video generation with large-scale dynamic scene reconstruction through multimodal conditional control. DriveGen3D introduces a unified pipeline consisting of two specialized components: FastDrive-DiT, an efficient video diffusion transformer for high-resolution, temporally coherent video synthesis under text and Bird's-Eye-View (BEV) layout guidance; and FastRecon3D, a feed-forward module that rapidly builds 3D Gaussian representations across time, ensuring spatial-temporal consistency. DriveGen3D enable the generation of long driving videos (up to $800\\times424$ at $12$ FPS) and corresponding 3D scenes, achieving state-of-the-art results while maintaining efficiency.", "date": "2025-10-17", "categories": []} {"id": "2605.24114", "title": "COSY: Compositional 3DGS Synthesis for Disentangled Human Head Editing", "abstract": "Recent 3D Gaussian Splatting (3DGS) GANs for human heads synthesize and render photorealistic 3D models in real-time and offer a vast variety in identity and appearance. However, controlling specific semantic attributes such as hair color or glasses remains challenging, as edits in the entangled latent space often induce unintended changes in identity or appearance. Although there are several methods that aim to disentangle the latent space post training by estimating directions that only modify certain features, these methods cannot guarantee complete disentanglement and often require pre-trained classifiers. In our approach, we propose a new generator architecture that synthesizes components, such as hair, skin, glasses, and torso, completely independently. This allows for changing the latent vector for one region while keeping the remaining parts fixed. Further, we achieve this separation using only sparse information such as the hair or skin color, eliminating the requirement of segmentation masks or geometric priors, often seen in prior work. To ensure matching shape and lighting conditions during editing, we allow minimal shared information via context tokens between the independent generators. These tokens even allow us to control the shape and light, without any prior annotation. Compared to existing works on GAN-based generation and editing, our method shows better disentanglement, more precise editing control, and competitive visual quality.", "date": "2026-05-22", "categories": []} {"id": "2605.25220", "title": "Multi-view Consistent 3D Gaussian Head Avatars 'without' Multi-view Generation", "abstract": "High-fidelity 3D Gaussian head avatar generation is critical for applications such as AR/VR, telepresence, and digital humans. Existing methods depend on multi-view datasets, 3D captures, or intermediate 2D view synthesis. In contrast, we learn both conditional and unconditional 3D head models from randomly sampled 2D images alone, without using multi-view data, 3D supervision, or intermediate view generation. We introduce MVCHead, a single-shot state space model that enforces multi-view consistency (MVC) directly in the 3D representation while regressing 3D Gaussians under these constraints. At its core, we propose a Hierarchical State Space (HiSS) block that progressively refines Gaussians from coarse to fine, while capturing long-range dependencies. Within each HiSS block, we modify Mamba's standard unidirectional scan with the proposed Hierarchical Bi-directional State Scan (HiBiSS) that aligns recurrence with the axes along which multi-view inconsistencies are strongest. Finally, we design an SE(3) Multi-view Critic that judges whether a set of self-renders arises from a single underlying 3D configuration, rewarding cross-view pixel alignment without observing real multi-view pairs. MVCHead achieves state-of-the-art perceptual quality, surpasses prior methods in both texture and geometric consistency, and maintains comparable shape consistency. To demonstrate scalability, we release FaceGS-10K, the first large-scale dataset of ready-to-use 3D Gaussian head assets for training and evaluation of 3D head models. Project Page and code: https://humansensinglab.github.io/MVCHead/", "date": "2026-05-24", "categories": []} {"id": "2605.26109", "title": "Helix4D: Complex 4D Mesh Generation", "abstract": "Current video-to-4D methods struggle with complex topology changes, transparent materials, thin structures, and inner surfaces. We present Helix4D, a dynamic mesh generation framework by inheriting the expressive representation of Trellis2, adapting it from image-to-3D to video-conditioned 4D generation. Our design arises from two key questions: (a) how to enable Trellis2's frame-local attention to share information across frames while preserving its pretrained quality on rare cases such as transparent objects and inner surfaces, and (b) how to inject temporal information into a purely 3D positional encoding without breaking pretrained capabilities. We address (a) with a sliding-window cross-frame attention and anchor on the first frame. The first frame is generated by the base Trellis2 model and injected into our model, letting it inherit Trellis2's quality in rare cases through cross-frame attention. We address (b) with a 4D temporal encoding that repurposes redundant low-frequency spatial RoPE bands for time, extending the encoding from 3D with no additional parameters. Extensive experiments show the effectiveness of Helix4D for high-quality dynamic mesh generation on ActionBench and our own challenging complex dynamics set.", "date": "2026-05-25", "categories": []} {"id": "2605.21572", "title": "PhysX-Omni: Unified Simulation-Ready Physical 3D Generation for Rigid, Deformable, and Articulated Objects", "abstract": "Simulation-ready physical 3D assets have emerged as a promising direction owing to their broad applicability in downstream tasks. However, most existing 3D generation methods either neglect physical properties or are limited to a single asset category, e.g., rigid, deformable, or articulated objects. To address these limitations, we introduce PhysX-Omni, a unified framework for simulation-ready physical 3D generation across diverse asset types. Specifically, we develop a novel and efficient geometry representation tailored for Vision-Language Models, which directly encodes high-resolution 3D structures without compression, significantly improving generation performance. In addition, we construct the first general simulation-ready 3D dataset, PhysXVerse, covering diverse indoor and outdoor categories. Furthermore, to comprehensively and flexibly evaluate both generative and understanding capabilities in the wild, we propose PhysX-Bench, which encompasses six key attributes: geometry, absolute scale, material, affordance, kinematics, and function description. Extensive experiments with conventional metrics and PhysX-Bench show that PhysX-Omni performs strongly in both generation and understanding. Moreover, additional studies further validate the potential of PhysX-Omni for applications in simulation-ready scene generation and robotic policy learning. We believe PhysX-Omni can significantly advance a wide range of downstream applications, particularly in embodied AI and physics-based simulation.", "date": "2026-05-20", "categories": []} {"id": "2605.20274", "title": "PolycubeNet: A Dual-latent Diffusion Model for Polycube-Based Hexahedral Mesh Generation", "abstract": "Hexahedral meshes are widely used in simulation pipelines, yet automatic generation remains challenging for complex CAD geometries. Polycube-based hexahedral meshing is a representative approach due to its regular, parameterization-friendly structure, but existing polycube construction methods often rely on intricate surface segmentation and local heuristics, which can produce artifacts or fail on difficult shapes.\n In this paper, we propose an end-to-end framework for polycube generation based on conditional diffusion models. Given an input geometry represented as a point cloud, our method directly produces a corresponding polycube point cloud, eliminating the need for explicit surface segmentation or predefined polycube templates. At the core of our approach is a dual-latent conditional diffusion architecture that confines computationally expensive self-attention operations to a fixed-capacity, low-dimensional latent space. This design effectively decouples computational complexity from the resolution of both the input geometry and the output polycube, thereby avoiding the quadratic cost typical of point cloud self-attention mechanisms while supporting flexible input and output resolutions.\n To obtain a hexahedral mesh, the generated polycube is aligned to the input shape via rigid and non-rigid point cloud registration to establish surface correspondence, followed by a polycube-to-hex pipeline. We additionally create and release a paired dataset of CAD meshes and their corresponding polycube meshes, together with the core implementation of our model. Experiments show that PolycubeNet generalizes to complex CAD models with arbitrary genus and produces high-quality polycube structures within seconds, improving robustness and efficiency over prior learning-based approaches.", "date": "2026-05-19", "categories": []} {"id": "2605.20872", "title": "CAdam: Context-Adaptive Moment Estimation for 3D Gaussian Densification in Generative Distillation", "abstract": "Adaptive densification is the engine of 3D Gaussian Splatting (3DGS). However, when transposed to the optimization-based Generative Distillation paradigm, this reconstruction-native mechanism reveals fundamental limitations, resulting in inefficient representations cluttered with redundant primitives. We diagnose this failure as a Densification Dilemma stemming from the stochastic nature of generative guidance: the standard magnitude-based accumulation indiscriminately aggregates transient noise alongside geometric signals, making it difficult to strike a balance between over-densification and under-fitting. To resolve this, we introduce Context-Adaptive Moment Estimation (CAdam), a novel framework that reinterprets densification as a statistically grounded signal verification problem. CAdam leverages the first moment of gradients to exploit the interference principle, where stochastic fluctuations cancel out via destructive interference while consistent geometric drifts accumulate via constructive interference, effectively disentangling the underlying signal from the generative noise floor. This is further augmented by a quantile-based context awareness and an intrinsic Signal-to-Noise Ratio (SNR) gating mechanism, which ensure robust adaptation across optimization stages and enable the soft termination of densification. Extensive experiments across diverse objectives (SDS, ISM, VFDS) and strong generative 3DGS backbones show that CAdam reduces Gaussian count by 85%-97% relative to standard densification while preserving overall comparable perceptual quality. These results highlight signal-aware density control as a practical way to improve memory efficiency in optimization-based generative distillation.", "date": "2026-05-20", "categories": []} {"id": "2605.21121", "title": "ROAR-3D: Routing Arbitrary Views for High-Fidelity 3D Generation", "abstract": "Single-image-to-3D generative models can now produce high-quality geometry, yet conditioning on a single view inevitably introduces ambiguity about unseen regions. Multi-view conditioning can reduce this ambiguity, but existing methods either require fixed canonical viewpoints or rely on external reconstruction modules that impose heavy training costs and limit generation quality. We observe that pretrained single-view models already possess strong 2D-to-3D grounding that can be reused for multi-view conditioning. However, a closer analysis reveals that their conditioning mechanism entangles orientation control with geometry transfer, two functions that conflict when images from different viewpoints are naively combined. Based on this analysis, we propose ROAR-3D, a lightweight method that upgrades a pretrained single-view model to accept an arbitrary number of unposed images. A token-wise view router assigns each 3D latent token to its most relevant view, implicitly establishing 2D-to-3D correspondences without explicit pose input. A dual-stream attention design preserves the pretrained primary-view behavior while routing auxiliary views through a separate path dedicated to geometric enrichment. An orientation perturbation strategy ensures the auxiliary path learns orientation-independent geometry transfer. These components introduce minimal trainable parameters and add negligible inference overhead relative to the single-view baseline. ROAR-3D achieves state-of-the-art multi-view 3D generation quality and supports test-time view scaling from 1 to 12+ views with consistent improvements.", "date": "2026-05-20", "categories": []} {"id": "2512.11234", "title": "RoomPilot: Controllable Indoor Scene Synthesis via Multimodal Semantic Parsing", "abstract": "Generating controllable indoor scenes is fundamental to applications in game development, architectural visualization, and embodied AI. However, existing approaches either support a limited input modalities or rely on implicit generation processes that hinder precise control over scene structure and semantics. To address these limitations, we introduce RoomPilot, a unified framework for controllable indoor scene synthesis from multi-modal inputs, including textual descriptions and CAD floor plans. RoomPilot maps heterogeneous inputs into an Indoor Domain-Specific Language (IDSL), which serves as a structured and interpretable semantic representation for describing indoor scenes. Built upon IDSL, RoomPilot presents a hierarchical synthesis pipeline that progressively organizes scenes at the building, room, and object levels, promoting structural coherence and functional consistency across multi-room layouts. Moreover, RoomPilot constructs a curated asset dataset with rich semantic annotations to support high-quality scene synthesis, improving visual realism and appearance consistency. Extensive experiments demonstrate effective multi-modal understanding, fine-grained controllability in scene generation, and improved physical consistency and visual fidelity, marking a significant step toward controllable 3D indoor scene synthesis. Code and model will be available.", "date": "2025-12-12", "categories": []} {"id": "2605.15186", "title": "VGGT-Edit: Feed-forward Native 3D Scene Editing with Residual Field Prediction", "abstract": "High-quality 3D scene reconstruction has recently advanced toward generalizable feed-forward architectures, enabling the generation of complex environments in a single forward pass. However, despite their strong performance in static scene perception, these models remain limited in responding to dynamic human instructions, which restricts their use in interactive applications. Existing editing methods typically rely on a 2D-lifting strategy, where individual views are edited independently and then lifted back into 3D space. This indirect pipeline often leads to blurry textures and inconsistent geometry, as 2D editors lack the spatial awareness required to preserve structure across viewpoints. To address these limitations, we propose VGGT-Edit, a feed-forward framework for text-conditioned native 3D scene editing. VGGT-Edit introduces depth-synchronized text injection to align semantic guidance with the backbone's spatial poses, ensuring stable instruction grounding. This semantic signal is then processed by a residual transformation head, which directly predicts 3D geometric displacements to deform the scene while preserving background stability. To ensure high-fidelity results, we supervise the framework with a multi-term objective function that enforces geometric accuracy and cross-view consistency. We also construct the DeltaScene Dataset, a large-scale dataset generated through an automated pipeline with 3D agreement filtering to ensure ground-truth quality. Experiments show that VGGT-Edit substantially outperforms 2D-lifting baselines, producing sharper object details, stronger multi-view consistency, and near-instant inference speed. The project page is https://chriszkxxx.github.io/VGGT-Edit/.", "date": "2026-05-14", "categories": []} {"id": "2605.17916", "title": "PanoWorld: A Generative Spatial World Model for Consistent Whole-House Panorama Synthesis", "abstract": "Generating a consistent whole-house VR tour from a floorplan and style reference requires both photorealistic panoramas and cross-view spatial coherence. Pure 2D generators produce appealing single panoramas but re-imagine geometry and materials when the viewpoint changes, whereas monolithic 3D generation becomes expensive and loses fine texture at multi-room scale. We introduce PanoWorld, a generative spatial world model that treats whole-house synthesis as autoregressive generation of node-based 360-degree panoramas, matching the discrete navigation used by real VR tour products. PanoWorld uses a floorplan-derived 3D shell as a global geometric proxy and a dynamic 3D Gaussian Splatting cache as renderable spatial memory. A feed-forward panoramic LRM designed for metric-scale multi-room 360-degree inputs lifts generated panoramas into local 3DGS updates, while Room-aware Group Attention suppresses cross-room feature interference. A topology-aware progressive caching strategy fuses these local updates without repeatedly reconstructing the full history. By decoupling shell-based geometry guidance from cache-rendered visual memory, PanoWorld preserves high-frequency 2D synthesis quality while improving cross-node layout and material consistency. The project link is https://jjrcn.github.io/PanoWorld-project-home/", "date": "2026-05-18", "categories": []} {"id": "2605.19350", "title": "CompoSE: Compositional Synthesis and Editing of 3D Shapes via Part-Aware Control", "abstract": "Creating and editing high-quality 3D content remains a central challenge in computer graphics. We address this challenge by introducing CompoSE, a novel method for Compositional Synthesis and Editing of 3D shapes via part-aware control. Our method takes as input a set of coarse geometric primitives (e.g., bounding boxes) that represent distinct object parts arranged in a particular spatial configuration, and synthesizes as output part-separated 3D objects that support localized granular (i.e., compositional) editing of individual parts. The key insight that enables our method is our use of a diffusion transformer architecture that alternates between processing each part locally and aggregating contextual information across parts globally, and features a novel conditioning technique that ensures strong adherence to the user's input. Importantly, our method learns to infer part semantics and symmetries directly from the user's coarse layout guidance, and does not require part-level text prompts. We demonstrate that our method enables powerful part-level editing capabilities, including context-aware substitution, addition, deletion, and style-preserving resizing operations. We show through extensive experiments that our method significantly outperforms existing approaches on guided synthesis, as measured by objective metrics and LLM-based evaluations.", "date": "2026-05-19", "categories": []} {"id": "2605.19587", "title": "SceneCode: Executable World Programs for Editable Indoor Scenes with Articulated Objects", "abstract": "Indoor scene synthesis underpins embodied AI, robotic manipulation, and simulation-based policy evaluation, where a useful scene must specify not only what the environment looks like, but also how its objects are structured. Existing pipelines, however, typically represent generated content as static meshes and inherit articulation only from curated asset libraries, which limits object-level controllability and prevents new interactable assets from being produced on demand. We address this gap by formulating physically interactable indoor scene synthesis as programmatic world generation, and present SceneCode, a framework that compiles a natural language prompt into an executable, code-driven indoor world rather than a collection of opaque meshes. A room-level agentic backbone first turns the prompt into a structured house layout and emits per-object AssetRequests through a planner--designer--critic loop. Each request is then routed to one of five code-generation strategies and converted into a synthesized part-wise Blender Python programs that are validated through an execution-guided repair-and-refine loop. The resulting programs are compiled into simulation-ready assets, and exported as SDF for physics simulation. A persistent scene-state registry links object requests, executable programs, rendered geometry, and simulation assets, turning scene assembly into a traceable and locally editable world-building process. We evaluate SceneCode across scene-level synthesis, object-level asset quality, human judgment, and downstream robot interaction. Results show that executable world programs improve prompt-faithful indoor scene generation and produce assets with cleaner mesh structure, and simulator-loadable articulation metadata. Project page: https://scene-code.github.io/.", "date": "2026-05-19", "categories": []} {"id": "2605.19974", "title": "SphericalDreamer: Generating Navigable Immersive 3D Worlds with Panorama Fusion", "abstract": "The generation of immersive and navigable 3D environments is increasingly prevalent with the growing adoption of virtual reality and 3D content. However, recent methods face a fundamental limitation: they cannot produce 3D worlds that simultaneously (i) are navigable over long-range spatial extents and (ii) cover the complete omnidirectional field of view ($360^\\circ$ horizontally and $180^\\circ$ vertically). To address this challenge, we introduce SphericalDreamer, a method for generating fully immersive and long-range 3D outdoor environments from textual prompts. Our approach is built on the generation of multiple panoramic images, which are subsequently lifted into 3D and fused together while maintaining visual and geometric consistency. SphericalDreamer produces highly detailed, fully immersive 3D environments, while substantially improving scale and navigability compared to prior approaches.", "date": "2026-05-19", "categories": []} {"id": "2511.18801", "title": "PartDiffuser: Part-wise 3D Mesh Generation via Discrete Diffusion", "abstract": "Existing autoregressive (AR) methods for generating artist-designed meshes struggle to balance global structural consistency with high-fidelity local details, and are susceptible to error accumulation. To address this, we propose PartDiffuser, a novel semi-autoregressive diffusion framework for point-cloud-to-mesh generation. The method first performs semantic segmentation on the mesh and then operates in a \"part-wise\" manner: it employs autoregression between parts to ensure global topology, while utilizing a parallel discrete diffusion process within each semantic part to precisely reconstruct high-frequency geometric features. PartDiffuser is based on the DiT architecture and introduces a part-aware cross-attention mechanism, using point clouds as hierarchical geometric conditioning to dynamically control the generation process, thereby effectively decoupling the global and local generation tasks. Experiments demonstrate that this method significantly outperforms state-of-the-art (SOTA) models in generating 3D meshes with rich detail, exhibiting exceptional detail representation suitable for real-world applications.", "date": "2025-11-24", "categories": []} {"id": "2512.18953", "title": "Symmetry Matters: Auditing and Symmetrizing 3D Generative Models", "abstract": "Symmetry is a strong prior present in many object categories, yet standard benchmarks for 3D generative models rarely report whether this prior is preserved. We study symmetry preservation in unconditional point cloud generation. We first audit the symmetry of generated shapes by several 3D generative models and compute a normalized symmetry score based on the Chamfer Distance (CD). We show that although current 3D generative models achieve competitive results under standard evaluation, they reveal a persistent symmetry gap when a symmetry-aware evaluation protocol is applied. To test whether this gap is merely inherited from the training data, we evaluate these models over a mirrored-objects dataset derived from ShapeNet and analyze symmetry dynamics during training. Mechanistic interpretability techniques were employed at the sampling and latent levels to further show that reflection symmetry is not reliably encoded in the learned generative process. Finally, to address this gap, we propose a data-centric symmetry-aware intervention: training generative models on a half-objects dataset and reconstructing full objects by reflection during sampling. Across multiple backbones, this intervention substantially improves geometric consistency and visual plausibility while remaining competitive under standard metrics. These findings suggest that symmetry-aware evaluation is needed alongside standard benchmarks, and incoming 3D generative models should incorporate this prior explicitly, either during training or sampling.", "date": "2025-12-22", "categories": []} {"id": "2604.28134", "title": "MeshReGen: A Unified 3D Geometry Regeneration Framework", "abstract": "We consider the problem of regenerating 3D objects from 2D images and initial 3D shapes. Most 3D generators operate in a one-shot fashion, converting text or images to a 3D object with limited controllability. We introduce instead MeshReGen, a 3D regenerator that is conditioned on an initial 3D shape. This conceptually simple formulation allows us to support numerous useful tasks, including 3D enhancement, reconstruction, and editing. MeshReGen uses a new conditioning mechanism based on VecSet, which allows the regenerator to update or improve the input geometry with consistent fine-grained details. MeshReGen learns a widely applicable regeneration prior from off-the-shelf 3D datasets via self-supervised pretext tasks and augmentations, without additional annotations. We evaluate both the geometric consistency and fine-grained quality of MeshReGen, achieving state-of-the-art performance in controllable 3D generation across several tasks.", "date": "2026-04-30", "categories": []} {"id": "2605.16355", "title": "Generative 3D Gaussians with Learned Density Control", "abstract": "We present Density-Sampled Gaussians (DeG), a novel 3D representation designed to bridge the gap between adaptive rendering primitives and scalable generative modeling. Unlike existing approaches that constrain 3D Gaussians to fixed voxel grids or arrays, DeG models Gaussian centers as samples from a learnable probability density function defined over an octree. This formulation provides a rigorous mathematical framework for adaptive density control: by jointly optimizing the spatial density and Gaussian attributes under rendering supervision, our model naturally concentrates primitives in regions of high geometric complexity. We achieve this via a new render loss contribution gradient that serves as a fully differentiable analogue to the discrete densification and pruning heuristics used in standard Gaussian Splatting. The resulting representation is highly flexible, supporting variable-resolution decoding from a single latent code by simply adjusting the sampling budget. To enable generative synthesis, we train a latent diffusion model on DeG. We identify a critical challenge in applying diffusion to unordered set-structured latents, which can significantly slow convergence, and propose VecSeq, a canonical re-indexing mechanism that anchors latent tokens to a deterministic 3D Sobol sequence. This transforms the ambiguous set-generation problem into a robust sequence modeling task. Extensive experiments demonstrate that our pipeline achieves state-of-the-art quality in single-image-to-3D generation, combining the structural adaptivity of unstructured primitives with the training stability of grid-based methods.", "date": "2026-05-08", "categories": []} {"id": "2605.16745", "title": "EVA01: Unified Native 3D Understanding and Generation via Mixture-of-Transformers", "abstract": "This paper addresses the challenge of integrating 3D meshes as a native modality within Multimodal Large Language Models (MLLMs). Diffusion-based large reconstruction models decouple semantic understanding from geometric reasoning, operating as stateless reconstructors conditioned on dense 2D pixel priors. Recent MLLM-based methods treat the 3D modality as an external output rather than a native component of the multimodal sequence, making incremental adaptations without a systematic analysis of how geometric manifolds align with MLLM feature spaces. We introduce EVA01, a unified framework that extends the modality boundary of MLLMs to natively incorporate 3D mesh understanding, generation, and context-aware editing. Built upon a Mixture-of-Transformers (MoT) architecture, EVA01 decouples the model into a pre-trained Understanding Expert ($E_{\\mathrm{und}}$) and a structurally mirrored Generation Expert ($E_{\\mathrm{gen}}$), coupled through shared global self-attention with hard modality routing. This design aligns the semantic latent space of the MLLM backbone with the geometric manifold, enabling direct transfer of multimodal priors without intermediate 2D representations. Results show that EVA01 achieves state-of-the-art native text-to-3D generation fidelity and unlocks robust long-context multi-turn geometric editing with identity preservation, a capability fundamentally inaccessible to stateless reconstruction pipelines. Our findings further offer architectural insights for integrating 2D foundation models with 3D tasks, informing the design of 3D-native multimodal systems. Project Page: https://www.seeles.ai/research/pages/EVA01", "date": "2026-05-16", "categories": []} {"id": "2605.16990", "title": "DreamEdit3D: Personalization of Multi-View Diffusion Models for 3D Editing", "abstract": "While 2D diffusion models have achieved remarkable success in identity-preserving personalization, extending this capability to 3D assets remains a significant challenge due to the complexities of multi-view consistency and spatial control. Inspired by these 2D advancements, we present a novel personalization method for text-guided 3D editing that enables compositional, object-level control through natural language. Given a 3D input, we render orthogonal views and extract object-level segmentation masks to isolate semantic components. We then learn distinct token embeddings for each component through a tailored two-phase optimization strategy: multi-view textual inversion with attention alignment, followed by full fine-tuning of multi-view diffusion model. During inference, these disentangled tokens seamlessly compose with editing prompts to generate multi-view consistent images, which are subsequently lifted into high-fidelity textured 3D meshes. Extensive evaluations across diverse editing scenarios demonstrate that our method successfully transfers the flexibility of 2D personalization to 3D, achieving state-of-the-art edit faithfulness and identity preservation compared to existing baselines.", "date": "2026-05-16", "categories": []} {"id": "2605.17102", "title": "VoxScene: Anchor-Conditioned Voxel Diffusion for Indoor Scene Arrangement", "abstract": "We present VoxScene, a novel anchor-conditioned voxel diffusion framework tailored for 3D scene synthesis. Current data-driven layout generation techniques typically rely on bounding proxies or implicit representations, which overlook volumetric structures. This geometric blindness inevitably leads to severe physical collisions and structural entanglement, particularly in densely populated environments. To overcome these limitations, we shift the paradigm to an explicit, object-centric voxel representation. Our pipeline sequentially synthesizes discrete volumetric occupancies conditioned on prior anchors and local context. By exploiting the mutually exclusive nature of discrete voxels, our approach eliminates spatial ambiguities and guarantees collision-free arrangements, even in highly complex environments. Furthermore, the synthesized high-fidelity voxel grids serve as discriminative geometric queries for downstream asset retrieval. Extensive experiments demonstrate the universality of our method, achieving state-of-the-art physical plausibility and unlocking shape diversity compared to existing layout planners.", "date": "2026-05-16", "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": "2605.18451", "title": "Code-as-Room: Generating 3D Rooms from Top-Down View Images via Agentic Code Synthesis", "abstract": "Designing realistic and functional 3D indoor rooms is essential for a wide range of applications, including interior design, virtual reality, gaming, and embodied AI. While recent MLLM-based approaches have shown great potential for 3D room synthesis from textual descriptions or reference images, text-based methods struggle to capture precise spatial information, and existing image-conditioned agents suffer from instability and infinite looping when tasked with holistic room generation from top-down views. To address these limitations, we propose Code-as-Room, an MLLM-based agentic framework equipped with a structured execution harness, which represents 3D rooms with Blender codes. Given a top-down room image, the framework parses the reference image to extract scene elements and their spatial relationships, and synthesizes executable Blender code for geometry, materials, and lighting in a principled, multi-stage pipeline. A cross-stage memory module is maintained throughout to mitigate context forgetting inherent to existing agent-based frameworks. We further introduce a dedicated benchmark for code-based 3D room synthesis, encompassing various evaluation protocols. Based on our benchmark, comprehensive comparisons against existing agent-based methods are conducted to validate the effectiveness of our proposed execution harness.", "date": "2026-05-18", "categories": []} {"id": "2601.21798", "title": "CG-MLLM: Captioning and Generating 3D content via Multi-modal Large Language Models", "abstract": "Large Language Models(LLMs) have revolutionized text generation and multimodal perception,but their capabilities in 3D content generation remain underexplored. Existing methods compromise by producing either low-resolution meshes or coarse structural proxies, failing to capture finegrained geometry natively. In this paper, we propose CG-MLLM, a novel Multi-modal Large Language Model (MLLM) capable of 3D captioning and high-resolution 3D generation in a single framework. Leveraging the Mixture-ofTransformer architecture, CG-MLLM decouples disparate modeling needs, where the Token-level Autoregressive (TokenAR) Transformer handles token-level content, and the Block-level Autoregressive (BlockAR) Transformer handles blocklevel content. By integrating a pre-trained visionlanguage backbone with a specialized 3D VAE latent space, CG-MLLM facilitates long-context interactions between standard tokens and spatial blocks within a single integrated architecture. Experimental results show that CG-MLLM significantly outperforms existing MLLMs in generating high-fidelity 3D objects, effectively bringing high-resolution 3D content creation into the mainstream LLM paradigm. Beyond generation, we further observe that learning to produce 3D content transfers back to perception, strengthening the model's image-based 3D understanding.", "date": "2026-01-29", "categories": []} {"id": "2501.12202", "title": "Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation", "abstract": "We present Hunyuan3D 2.0, an advanced large-scale 3D synthesis system for generating high-resolution textured 3D assets. This system includes two foundation components: a large-scale shape generation model -- Hunyuan3D-DiT, and a large-scale texture synthesis model -- Hunyuan3D-Paint. The shape generative model, built on a scalable flow-based diffusion transformer, aims to create geometry that properly aligns with a given condition image, laying a solid foundation for downstream applications. The texture synthesis model, benefiting from strong geometric and diffusion priors, produces high-resolution and vibrant texture maps for either generated or hand-crafted meshes. Furthermore, we build Hunyuan3D-Studio -- a versatile, user-friendly production platform that simplifies the re-creation process of 3D assets. It allows both professional and amateur users to manipulate or even animate their meshes efficiently. We systematically evaluate our models, showing that Hunyuan3D 2.0 outperforms previous state-of-the-art models, including the open-source models and closed-source models in geometry details, condition alignment, texture quality, and etc. Hunyuan3D 2.0 is publicly released in order to fill the gaps in the open-source 3D community for large-scale foundation generative models. The code and pre-trained weights of our models are available at: https://github.com/Tencent/Hunyuan3D-2", "date": "2025-01-21", "categories": []} {"id": "2505.22394", "title": "PacTure: Efficient PBR Texture Generation on Packed Views with Visual Autoregressive Models", "abstract": "We present PacTure, a novel framework for generating physically-based rendering (PBR) material textures for an untextured 3D mesh from a text description. Existing 2D generation-based texturing approaches either generate textures sequentially from different views, resulting in long inference times and globally inconsistent textures, or adopt multi-view generation with cross-view attention to enhance global consistency, which, however, limits the resolution for each view. In response to these weaknesses, we first introduce view packing, a novel technique that significantly increases the effective resolution for each view during multi-view generation, without imposing additional inference cost. Unlike UV mapping, it preserves the spatial proximity essential for image generation and maintains full compatibility with current 2D generative models. To further reduce the inferencing cost, we enable fine-grained control and multi-domain generation within the next-scale prediction autoregressive framework, creating an efficient multi-view PBR generation backbone. Extensive experiments show that PacTure outperforms state-of-the-art methods in both quality and efficiency.", "date": "2025-05-28", "categories": []} {"id": "2605.13838", "title": "R-DMesh: Video-Guided 3D Animation via Rectified Dynamic Mesh Flow", "abstract": "Video-guided 3D animation holds immense potential for content creation, offering intuitive and precise control over dynamic assets. However, practical deployment faces a critical yet frequently overlooked hurdle: the pose misalignment dilemma. In real-world scenarios, the initial pose of a user-provided static mesh rarely aligns with the starting frame of a reference video. Naively forcing a mesh to follow a mismatched trajectory inevitably leads to severe geometric distortion or animation failure. To address this, we present Rectified Dynamic Mesh (R-DMesh), a unified framework designed to generate high-fidelity 4D meshes that are ``rectified'' to align with video context. Unlike standard motion transfer approaches, our method introduces a novel VAE that explicitly disentangles the input into a conditional base mesh, relative motion trajectories, and a crucial rectification jump offset. This offset is learned to automatically transform the arbitrary pose of the input mesh to match the video's initial state before animation begins. We process these components via a Triflow Attention mechanism, which leverages vertex-wise geometric features to modulate the three orthogonal flows, ensuring physical consistency and local rigidity during the rectification and animation process. For generation, we employ a Rectified Flow-based Diffusion Transformer conditioned on pre-trained video latents, effectively transferring rich spatio-temporal priors to the 3D domain. To support this task, we construct Video-RDMesh, a large-scale dataset of over 500k dynamic mesh sequences specifically curated to simulate pose misalignment. Extensive experiments demonstrate that R-DMesh not only solves the alignment problem but also enables robust downstream applications, including pose retargeting and holistic 4D generation.", "date": "2026-05-13", "categories": []} {"id": "2605.13852", "title": "Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning", "abstract": "We often aim to generate images that are both photorealistic and 3D-consistent, adhering to precise geometry, material, and viewpoint controls. Typically, this is achieved by fine-tuning an image generator, pre-trained on billions of real images, using renders of synthetic 3D assets, where annotations for control signals are available. While this approach can learn the desired controls, it often compromises the realism of the images due to domain gap between photographs and renders. We observe that this issue largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images. To address this, we introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain. Then, the generator can be trained to gain controllability, without fitting to specific visual domain. In this way, the model can be guided to produce realistic images even when controls are applied. We enhance control transferability to the real domain by leveraging insights about roles of different layers and denoising steps in diffusion-based generators, informing new training and inference strategies that further mitigate the gap. We demonstrate the advantages of Realiz3D in tasks as text-to-multiview generation and texturing from 3D inputs, producing outputs that are 3D-consistent and photorealistic.", "date": "2026-03-25", "categories": []} {"id": "2605.13862", "title": "Seed3D 2.0: Advancing High-Fidelity Simulation-Ready 3D Content Generation", "abstract": "We present Seed3D 2.0, an advanced 3D content generation system built on Seed3D 1.0, with substantial improvements across generation fidelity, simulation-ready capabilities, and application coverage. For geometry, a coarse-to-fine two-stage pipeline decouples global structure learning from high-frequency detail recovery, while a locality-aware VAE achieves higher spatial compression and more efficient decoding. For texture and material generation, we replace the cascaded pipeline of Seed3D 1.0 with a unified PBR model that directly generates multi-view albedo and metallic-roughness maps, enhanced by Mixture-of-Experts scaling and VLM-based semantic conditioning for improved material precision and visual fidelity. Beyond single-object generation, Seed3D 2.0 introduces a simulation-ready model suite comprising scene layout planning, part-aware decomposition, and training-free articulation generation, enabling coherent scene construction and part-level physical interaction across physics and graphics engines. A large-scale human preference study against five recent commercial models shows that Seed3D 2.0 achieves consistent win rates of 69.0% to 89.9% in textured 3D asset generation. Seed3D 2.0 is available on https://exp.volcengine.com/ark/vision?_vtm_=0.0.c70961.d701978.0&mode=vision&modelId=doubao-seed3d-2-0-260328&tab=Gen3D", "date": "2026-04-22", "categories": []} {"id": "2605.14594", "title": "TOPOS: High-Fidelity and Efficient Industry-Grade 3D Head Generation", "abstract": "High-fidelity 3D head generation plays a crucial role in the film, animation and video game industries. In industrial pipelines, studios typically enforce a fixed reference topology across all head assets, as such a clean and uniform topology is a prerequisite for production-level rigging, skinning and animation. In this paper, we present TOPOS, a framework tailored for single image conditioned 3D head generation that jointly recovers geometry and appearance under such an industry-standard topology. In contrast to general 3D generative models which produce triangle meshes with inconsistent topology and numerous vertices, hindering semantic correspondence and asset-level reuse, TOPOS generates head meshes with a fixed, studio-style topology, enabling consistent vertex-level correspondence across all generated heads. To model heads under this unified topology, we proposed a novel variational autoencoder structure, termed TOPOS-VAE. Inspired by multi-model large language models (MLLMs), our TOPOS-VAE leverages the Perceiver Resampler to convert input pointclouds sampled from head meshes of diverse topologies into the target reference topology. Building upon TOPOS-VAE's structured latent space, we train a rectified flow transformer, TOPOS-DiT, to efficiently generate high-fidelity head meshes from a single image. We further present TOPOS-Texture, an end-to-end module that produces relightable UV texture maps from the same portrait image via fine-tuning a multimodal image generative model. The generated textures are spatially aligned with the underlying mesh geometry and faithfully preserve high-frequency appearance details. Extensive experiments demonstrate that TOPOS achieves state-of-the-art performance on 3D head generation, surpassing both classical face reconstruction methods and general 3D object generative models, highlighting its effectiveness for digital human creation.", "date": "2026-05-14", "categories": []} {"id": "2605.14984", "title": "Sat3DGen: Comprehensive Street-Level 3D Scene Generation from Single Satellite Image", "abstract": "Generating a street-level 3D scene from a single satellite image is a crucial yet challenging task. Current methods present a stark trade-off: geometry-colorization models achieve high geometric fidelity but are typically building-focused and lack semantic diversity. In contrast, proxy-based models use feed-forward image-to-3D frameworks to generate holistic scenes by jointly learning geometry and texture, a process that yields rich content but coarse and unstable geometry. We attribute these geometric failures to the extreme viewpoint gap and sparse, inconsistent supervision inherent in satellite-to-street data. We introduce Sat3DGen to address these fundamental challenges, which embodies a geometry-first methodology. This methodology enhances the feed-forward paradigm by integrating novel geometric constraints with a perspective-view training strategy, explicitly countering the primary sources of geometric error. This geometry-centric strategy yields a dramatic leap in both 3D accuracy and photorealism. For validation, we first constructed a new benchmark by pairing the VIGOR-OOD test set with high-resolution DSM data. On this benchmark, our method improves geometric RMSE from 6.76m to 5.20m. Crucially, this geometric leap also boosts photorealism, reducing the Fr\\'echet Inception Distance (FID) from $\\sim$40 to 19 against the leading method, Sat2Density++, despite using no extra tailored image-quality modules. We demonstrate the versatility of our high-quality 3D assets through diverse downstream applications, including semantic-map-to-3D synthesis, multi-camera video generation, large-scale meshing, and unsupervised single-image Digital Surface Model (DSM) estimation. The code has been released on https://github.com/qianmingduowan/Sat3DGen.", "date": "2026-05-14", "categories": []} {"id": "2605.15187", "title": "Articraft: An Agentic System for Scalable Articulated 3D Asset Generation", "abstract": "A bottleneck in learning to understand articulated 3D objects is the lack of large and diverse datasets. In this paper, we propose to leverage large language models (LLMs) to close this gap and generate articulated assets at scale. We reduce the problem of generating an articulated 3D asset to that of writing a program that builds it. We then introduce a new agentic system, Articraft, that writes such programs automatically. We design a programmatic interface and harness to help the LLM do so effectively. The LLM writes code against a domain-specific SDK for defining parts, composing geometry, specifying joints, and writing tests to validate the resulting assets. The harness exposes a restricted workspace and interface to the LLM, validates the resulting assets, and returns structured feedback. In this way, the LLM is not distracted by details such as authoring a URDF file or managing a complex software environment. We show that this produces higher-quality assets than both state-of-the-art articulated-asset generators and general-purpose coding agents. Using Articraft, we build Articraft-10K, a curated dataset of over 10K articulated assets spanning 245 categories, and show its utility both for training models of articulated assets and in downstream applications such as robotics simulation and virtual reality.", "date": "2026-05-14", "categories": []} {"id": "2510.11303", "title": "sketch2symm: Symmetry-aware sketch-to-shape generation via semantic bridging", "abstract": "Sketch-based 3D reconstruction remains a challenging task due to the abstract and sparse nature of sketch inputs, which often lack sufficient semantic and geometric information. To address this, we propose Sketch2Symm, a two-stage generation method that produces geometrically consistent 3D shapes from sketches. Our approach introduces semantic bridging via sketch-to-image translation to enrich sparse sketch representations, and incorporates symmetry constraints as geometric priors to leverage the structural regularity commonly found in everyday objects. Experiments on mainstream sketch datasets demonstrate that our method achieves superior performance compared to existing sketch-based reconstruction methods in terms of Chamfer Distance, Earth Mover's Distance, and F-Score, verifying the effectiveness of the proposed semantic bridging and symmetry-aware design.", "date": "2025-10-13", "categories": []} {"id": "2605.12957", "title": "GTA: Advancing Image-to-3D World Generation via Geometry Then Appearance Video Diffusion", "abstract": "Recent developments in generative models and large-scale datasets have substantially advanced 3D world generation, facilitating a broad range of domains including spatial intelligence, embodied intelligence, and autonomous driving. While achieving remarkable progress, existing approaches to 3D world generation typically prioritize appearance prediction with limited modeling of the underlying geometry, leading to issues such as unreliable scene structure estimation and degraded cross-view consistency. To address these limitations, motivated by the coarse-to-fine nature of human visual perception, we propose GTA, a novel image-to-3D world generation method following a Geometry-Then-Appearance paradigm. Specifically, given a single input image, to improve the structural fidelity of synthesized 3D scenes, GTA adopts a two-stage framework with two dedicated video diffusion models, which first generate coarse geometric structure from novel viewpoints and then synthesize fine-grained appearance conditioned on the predicted geometry. To further enhance cross-view appearance consistency, we introduce a random latent shuffle strategy during the training process, along with a test-time scaling scheme that improves perceptual quality without compromising quantitative performance. Extensive experiments have demonstrated that our proposed method consistently outperforms existing approaches in terms of fidelity, visual quality, and geometric accuracy. Moreover, GTA is shown to be effective as a general enhancement module that further improves the generation quality of existing image-to-3D world pipelines, as well as supporting multiple downstream applications and exhibiting favorable data efficiency during model training, highlighting its versatility and broad applicability. Project page: https://hanxinzhu-lab.github.io/GTA/.", "date": "2026-05-13", "categories": []} {"id": "2605.13129", "title": "Rigel3D: Rig-aware Latents for Animation-Ready 3D Asset Generation", "abstract": "Recent 3D generative models can synthesize high-quality assets, but their outputs are typically static: they lack the skeletal rigs, joint hierarchies, and skinning weights required for animation. This limits their use in games, film, simulation, virtual agents, and embodied AI, where assets must not only look plausible but also move plausibly. We introduce Rigel3D, a generative method for animation-ready 3D assets represented as rigged meshes. Unlike post-hoc auto-rigging methods that attach rigs to completed shapes, our method jointly models geometry and rig structure through coupled surface and skeleton structured latent representations. A rig-aware autoencoder decodes these representations into mesh geometry, skeleton topology, joint coordinates, and skinning weights, while a two-stage latent generative model synthesizes both surface and skeleton representations for image-conditioned generation. To support downstream animation workflows, we further introduce an open-vocabulary joint labeling module that embeds generated joints into a shared vision-language space, enabling correspondence to arbitrary retargeting templates. Experiments on large-scale rigged asset datasets demonstrate that our method generates diverse, high-quality animation-ready assets and outperforms existing rigging baselines across multiple metrics.", "date": "2026-05-13", "categories": []} {"id": "2605.13293", "title": "Img2CADSeq: Image-to-CAD Generation via Sequence-Based Diffusion", "abstract": "Boundary Representation (BRep) is the standard format for Computer-Aided Design (CAD), yet reconstructing high-quality BReps from single-view images remains challenging due to the complexity of topological constraints and operation sequences. We present Img2CADSeq, a multi-stage pipeline that overcomes these limitations by encoding CAD sequences into a three-level hierarchical codebook. Guided by an importance prioritization, this strategy values profiles over details, compressing long sequences into a stable discrete latent space. To bridge the modality gap, we leverage a coarse-to-fine point cloud intermediate, aligning 2D visual features with 3D CAD sequences via contrastive learning to condition a VQ-Diffusion model. Supported by newly introduced CAD-220K and PrintCAD datasets, our approach ensures robust industrial domain adaptation. Extensive experiments demonstrate that Img2CADSeq significantly outperforms state-of-the-art methods, producing standard STEP files that can be directly used in commercial CAD software.", "date": "2026-05-13", "categories": []} {"id": "2605.13815", "title": "OmniLiDAR: A Unified Diffusion Framework for Multi-Domain 3D LiDAR Generation", "abstract": "LiDAR scene generation is increasingly important for scalable simulation and synthetic data creation, especially under diverse sensing conditions that are costly to capture at scale. Typically, diffusion-based LiDAR generators are developed under single-domain settings, requiring separate models for different datasets or sensing conditions and hindering unified, controllable synthesis under heterogeneous distribution shifts. To this end, we present OmniLiDAR, a unified text-conditioned diffusion framework that generates LiDAR scans in a shared range-image representation across eight representative domains spanning three shift types: adverse weather, sensor-configuration changes (e.g., reduced beams), and cross-platform acquisition (vehicle, drone, and quadruped). To enable training a single model over heterogeneous domains without isolating optimization by domain, we introduce a Cross-Domain Training Strategy (CDTS) that mixes domains within each mini-batch and leverages conditioning to steer generation. We further propose Cross-Domain Feature Modeling (CDFM), which captures directional dependencies along azimuth and elevation axes to reflect the anisotropic scanning structure of range images, and Domain-Adaptive Feature Scaling (DAFS) as a lightweight modulation to account for structured domain-dependent feature shifts during denoising. In the absence of a public consolidated benchmark, we construct an 8-domain dataset by combining real-world scans with physically based weather simulation and systematic beam reduction while following official splits. Extensive experiments demonstrate strong generation fidelity and consistent gains in downstream use cases, including generative data augmentation for LiDAR semantic segmentation and 3D object detection, as well as robustness evaluation under corruptions, with consistent benefits in limited-label regimes.", "date": "2026-05-13", "categories": []} {"id": "2604.11808", "title": "Pair2Scene: Learning Local Object Relations for Procedural Scene Generation", "abstract": "Generating high-fidelity 3D indoor scenes remains a significant challenge due to data scarcity and the complexity of modeling intricate spatial relations. Current methods often struggle to scale beyond training distribution to dense scenes or rely on LLMs/VLMs that lack the ability for precise spatial reasoning. Building on top of the observation that object placement relies mainly on local dependencies instead of information-redundant global distributions, in this paper, we propose Pair2Scene, a novel procedural generation framework that integrates learned local rules with scene hierarchies and physics-based algorithms. These rules mainly capture two types of inter-object relations, namely support relations that follow physical hierarchies, and functional relations that reflect semantic links. We model these rules through a network, which estimates spatial position distributions of dependent objects conditioned on position and geometry of the anchor ones. Accordingly, we curate a dataset 3D-Pairs from existing scene data to train the model. During inference, our framework can generate scenes by recursively applying our model within a hierarchical structure, leveraging collision-aware rejection sampling to align local rules into coherent global layouts. Extensive experiments demonstrate that our framework outperforms existing methods in generating complex environments that go beyond training data while maintaining physical and semantic plausibility.", "date": "2026-04-13", "categories": []} {"id": "2605.08744", "title": "MeshFIM: Local Low-Poly Mesh Editing via Fill-in-the-Middle Autoregressive Generation", "abstract": "Autoregressive (AR) models can generate high-quality low-poly meshes from point clouds, but they still operate in an all-or-nothing manner: when a local region is unsatisfactory, the entire mesh must be regenerated, wasting computation and destroying satisfactory mesh structure elsewhere. We introduce MeshFIM, a Fill-in-the-Middle (FIM) framework that regenerates a target region of a low-poly mesh conditioned on the surrounding context. MeshFIM addresses three mesh-specific challenges: enforcing exact attachment along the exposed boundary, preserving topological order in the context, and suppressing overflow beyond the intended region. It does so with five complementary design choices: boundary vertex markers, context positional embeddings, expanded context width, context augmentation, and a low-poly geometry encoder whose gated subtraction mechanism focuses generation on the missing region by leveraging the difference between the reference surface and the existing mesh. Detailed ablation studies are presented to show the effectiveness of every introduced component. Based on MeshFIM, we demonstrate two applications: interactive brush-based editing and automatic defect repair on low-poly mesh (see Figure 1). Last but not least, experiments show that MeshFIM outperforms a range of baselines in mesh refinement, mesh repair and whole mesh generation plus stitch-back scheme.", "date": "2026-05-09", "categories": []} {"id": "2605.08824", "title": "HairGPT: Strand-as-Language Autoregressive Modeling for Realistic 3D Hairstyle Synthesis", "abstract": "Hair is a rich medium of visual and cultural expression, yet its digital modeling remains challenging due to the duality of fluidity and structure. Many existing generative approaches rely primarily on continuous diffusion fields, which entangle global topology with local texture and obscure the semantic and structural organization of hairstyles. To address this, we propose HairGPT, a strand-centric framework that treats strands as generative primitives and formulates realistic 3D hairstyle synthesis as a dual-decoupled autoregressive sequence modeling problem. Our method applies spatial decoupling across semantic scalp regions and structural decoupling along a hierarchical strand representation, progressing from global layout to fine-grained style. We further introduce a geometric tokenizer and region-aware semantic annotations to guide strand-level generation, enabling compositional editing, synthesis of rare and complex hairstyles, and adaptation to stylized domains. By aligning generative modeling with the workflow of digital grooming, HairGPT turns hair generation from opaque texture synthesis into a structured and semantically controllable authoring process, supporting robust semantic conditioning and high-fidelity results across realistic and stylized domains. Project Page: https://haiminluo.github.io/hairgpt/", "date": "2026-05-09", "categories": []} {"id": "2605.09071", "title": "Probability-Flow Distillation: Exact Wasserstein Gradient Flow for High-Fidelity 3D Generation", "abstract": "Score Distillation Sampling (SDS) and its variants have been widely used for text-to-3D generation by distilling 2D image diffusion priors. However, the standard SDS objective is prone to severe mode collapse, frequently yielding over-smoothed and over-saturated results. Although recent advancements, such as Score Distillation via Inversion (SDI), mitigate these artifacts and produce visually sharper models, they ultimately fail to faithfully capture the full target distribution. In this work, we show that the bottleneck limiting the sampling capacity of SDI stems from its reliance on the posterior mean estimator, which is mathematically equivalent to a single-step Euler approximation of the deterministic reverse DDIM trajectory. To address this, we propose a naturally motivated extension termed Probability-Flow Distillation (PFD). We establish that PFD corresponds exactly to a Wasserstein gradient flow, thereby inducing principled distribution-matching dynamics. Finally, we show that PFD can synthesize 3D assets with fine-grained, high-fidelity details and achieve improved quality compared to existing methods.", "date": "2026-05-09", "categories": []} {"id": "2605.10438", "title": "Beyond Spatial Compression: Interface-Centric Generative States for Open-World 3D Structure", "abstract": "Current 3D tokenizers largely treat representation as spatial compression: compact codes reconstruct surface geometry, but leave component ownership and attachment validity implicit. In open-world assets with intersecting components, noisy topology, and weak canonical structure, this creates a representation mismatch: local shape, component identity, and assembly relations become entangled in a latent stream and are not natively addressable during decoding. We formulate an alternative view, interface-centric generative states, in which tokenization constructs an operational state rather than a passive compressed code. The state exposes local geometry, component ownership, and attachment validity as variables that can be queried, constrained, and repaired during decoding. We instantiate this formulation with Component-Conditioned Canonical Local Tokens (C2LT-3D), factorizing representation into canonical local geometry, partition-conditioned context, and relational seam variables. Each factor targets a distinct failure mode of compression-centric tokens: pose leakage, cross-component interference, or invalid local attachment. This exposed state supports attachment validation, latent structural repair, targeted intervention, and constrained serialization without a separate post-hoc structure recovery module. Trained on single-object CAD models and evaluated zero-shot on open-world multi-component assets, C2LT-3D improves structural robustness and shows that its latent variables remain actionable under adversarial attachment settings. These results suggest that open-world 3D generative representations should be evaluated not only by reconstruction fidelity, but by whether their discrete states remain operational for assembly-level structural reasoning.", "date": "2026-05-11", "categories": []} {"id": "2605.10922", "title": "Pixal3D: Pixel-Aligned 3D Generation from Images", "abstract": "Recent advances in 3D generative models have rapidly improved image-to-3D synthesis quality, enabling higher-resolution geometry and more realistic appearance. Yet fidelity, which measures pixel-level faithfulness of the generated 3D asset to the input image, still remains a central bottleneck. We argue this stems from an implicit 2D-3D correspondence issue: most 3D-native generators synthesize shape in canonical space and inject image cues via attention, leaving pixel-to-3D associations ambiguous. To tackle this issue, we draw inspiration from 3D reconstruction and propose Pixal3D, a pixel-aligned 3D generation paradigm for high-fidelity 3D asset creation from images. Instead of generating in a canonical pose, Pixal3D directly generates 3D in a pixel-aligned way, consistent with the input view. To enable this, we introduce a pixel back-projection conditioning scheme that explicitly lifts multi-scale image features into a 3D feature volume, establishing direct pixel-to-3D correspondence without ambiguity. We show that Pixal3D is not only scalable and capable of producing high-quality 3D assets, but also substantially improves fidelity, approaching the fidelity level of reconstruction. Furthermore, Pixal3D naturally extends to multi-view generation by aggregating back-projected feature volumes across views. Finally, we show pixel-aligned generation benefits scene synthesis, and present a modular pipeline that produces high-fidelity, object-separated 3D scenes from images. Pixal3D for the first time demonstrates 3D-native pixel-aligned generation at scale, and provides a new inspiring way towards high-fidelity 3D generation of object or scene from single or multi-view images. Project page: https://ldyang694.github.io/projects/pixal3d/", "date": "2026-05-11", "categories": []} {"id": "2604.26509", "title": "3D Generation for Embodied AI and Robotic Simulation: A Survey", "abstract": "Embodied AI and robotic systems increasingly depend on scalable, diverse, and physically grounded 3D content for simulation-based training and real-world deployment. While 3D generative modeling has advanced rapidly, embodied applications impose requirements far beyond visual realism: generated objects must carry kinematic structure and material properties, scenes must support interaction and task execution, and the resulting content must bridge the gap between simulation and reality. This survey reviews 3D generation for embodied AI and organizes the literature around three roles that 3D generation plays in embodied systems. In Data Generator, 3D generation produces simulation-ready objects and assets, including articulated, physically grounded, and deformable content for downstream interaction; in Simulation Environments, it constructs interactive and task-oriented worlds, spanning structure-aware, controllable, and agentic scene generation; and in Sim2Real Bridge, it supports digital twin reconstruction, data augmentation, and synthetic demonstrations for downstream robot learning and real-world transfer. We also show that the field is shifting from visual realism toward interaction readiness, and we identify the main bottlenecks, including limited physical annotations, the gap between geometric quality and physical validity, fragmented evaluation, and the persistent sim-to-real divide, that must be addressed for 3D generation to become a dependable foundation for embodied intelligence. Our project page is at https://3dgen4robot.github.io.", "date": "2026-04-29", "categories": []} {"id": "2605.06947", "title": "Rollback-Free Stable Brick Structures Generation", "abstract": "While autoregressive models have advanced 3D generation, creating physically stable brick structures remains a challenge due to the strict requirements of gravity and interconnectivity. Existing approaches rely on external physical simulators during inference to perform rejection sampling and brick-by-brick rollbacks, which severely bottlenecks efficiency. To address this, we propose a reinforcement learning paradigm that shifts physical validity enforcement from test-time correction to training-time policy optimization. By utilizing assembly-level rewards, the model optimizes for collision avoidance, global connectivity, structural interlocking, and shape conformity. This paradigm allows the model to internalize physical priors, enabling the first rollback-free generation of stable brick structures. Experimental results demonstrate that our approach achieves state-of-the-art generation quality while accelerating inference speed by orders of magnitude. Our code and dataset are available at https://github.com/miniHuiHui/STABLE. Our models are available at https://huggingface.co/miniHui/STABLE.", "date": "2026-05-07", "categories": []} {"id": "2605.07385", "title": "Velocity-Space 3D Asset Editing", "abstract": "Editing a 3D asset locally, modifying a target region while preserving the rest, is a fundamental requirement of native 3D editing. Existing methods enforce locality through mechanisms external to the generator, such as manual 3D masks, post-hoc voxel merging, or 2D multi-view lifting. None of them intervene where the corruption actually originates: inside the ODE sampler. For a rectified-flow generator to achieve faithful local editing, its velocity field should be strong over the target editing region while vanishing on preserved content. Yet a single velocity field can hardly satisfy both requirements simultaneously, leading to three problems: (i) identity leakage that keeps the edit signal non-zero on preserved regions; (ii) no dedicated edit-amplification channel, so strengthening the edit inevitably perturbs identity; and (iii) an identity drag at the geometry and material stages, where a global condition pulls every token toward the target. We propose VS3D (Velocity-Space 3D Asset editing}), an inversion-free, training-free, and mask-free framework that addresses each problem with a targeted intervention inside the sampler. VS3D integrates three complementary modules, each corresponding to a specific stage of the editing pipeline. Reconstruction-Anchored Source Injection (RASI) absorbs identity leakage by turning the unconditional embedding into a per-step, asset-specific anchor calibrated through source reconstruction. Partial-Mean Guidance (PMG) amplifies the edit signal by contrasting high- and low-quality subsample estimates of the velocity difference, active only where a consistent edit exists. Twin-Agreement Residual injection (TAR) lets the sampler decide token by token what to preserve at the geometry and material stages.", "date": "2026-05-08", "categories": []} {"id": "2605.04412", "title": "Structured 3D Latents Are Surprisingly Powerful: Unleashing Generalizable Style with 2D Diffusion", "abstract": "3D asset generation plays a pivotal role in fields such as gaming and virtual reality, enabling the rapid synthesis of high-fidelity 3D objects from a single or multiple images. Building on this capability, enabling style-controllable generation naturally emerges as an important and desirable direction. However, existing approaches typically rely on style images that lie within or are similar to the training distribution of 3D generation models. When presented with out-of-distribution (OOD) styles, their performance degrades significantly or even fails. To address this limitation, we introduce \\textbf{DiLAST}: 2D Diffusion-based Latent Awakening for 3D Style Transfer. Specifically, we leverage a pretrained 2D diffusion model as a teacher to provide rich and generalizable style priors. By aligning rendered views with the target style under diffusion-based guidance, our method optimizes the structured 3D latent representations for stylization. We observe that this limitation stems not from insufficient model capacity, but from the underutilization of structured 3D latents, which are inherently expressive. Despite being trained on comparatively limited data, 3D generation models can leverage 2D diffusion guidance to steer denoising toward specific directions in latent space, thereby producing diverse, OOD styles. Extensive experiments across diverse data and multiple 3D generation backbones demonstrate the effectiveness and plug-and-play nature of our approach.", "date": "2026-05-06", "categories": []} {"id": "2605.05711", "title": "Closing the Loop: Unified 3D Scene Generation and Immersive Interaction via LLM-RL Coupling", "abstract": "Recent advances in large language models (LLMs) have significantly improved language-driven 3D content generation, but most existing approaches still treat scene generation and user interaction as separate processes, limiting the adaptability and immersive potential of interactive multimedia systems. This paper presents a unified framework that closes the loop between language-driven 3D scene generation and immersive user interaction. Given natural language instructions, the system first constructs structured scene representations using LLMs, and then optimizes spatial layouts via reinforcement learning under geometric and semantic constraints. The generated environments are deployed in a virtual reality setting to facilitate HRI-in-the-loop, where user interactions provide continuous feedback to align generated content with human perception and usability. By tightly coupling generation and interaction, the proposed framework enables more responsive, adaptive, and realistic multimedia experiences. Experiments on the ALFRED benchmark demonstrate state-of-the-art performance in task-based scene generation. Furthermore, qualitative results and user studies show consistent improvements in immersion, interaction quality, and task efficiency, highlighting the importance of closed-loop integration of generation and interaction for next-generation multimedia systems. Our project page can be found at https://proj-showcase.github.io/h3ds/.", "date": "2026-05-07", "categories": []} {"id": "2605.05163", "title": "PhysForge: Generating Physics-Grounded 3D Assets for Interactive Virtual World", "abstract": "Synthesizing physics-grounded 3D assets is a critical bottleneck for interactive virtual worlds and embodied AI. Existing methods predominantly focus on static geometry, overlooking the functional properties essential for interaction. We propose that interactive asset generation must be rooted in functional logic and hierarchical physics. To bridge this gap, we introduce PhysForge, a decoupled two-stage framework supported by PhysDB, a large-scale dataset of 150,000 assets with four-tier physical annotations. First, a VLM acts as a \"physical architect\" to plan a \"Hierarchical Physical Blueprint\" defining material, functional, and kinematic constraints. Second, a physics-grounded diffusion model realizes this blueprint by synthesizing high-fidelity geometry alongside precise kinematic parameters via a novel KineVoxel Injection (KVI) mechanism. Experiments demonstrate that PhysForge produces functionally plausible, simulation-ready assets, providing a robust data engine for interactive 3D content and embodied agents.", "date": "2026-05-06", "categories": []} {"id": "2401.16764", "title": "BoostDream: Efficient Refining for High-Quality Text-to-3D Generation from Multi-View Diffusion", "abstract": "Witnessing the evolution of text-to-image diffusion models, significant strides have been made in text-to-3D generation. Currently, two primary paradigms dominate the field of text-to-3D: the feed-forward generation solutions, capable of swiftly producing 3D assets but often yielding coarse results, and the Score Distillation Sampling (SDS) based solutions, known for generating high-fidelity 3D assets albeit at a slower pace. The synergistic integration of these methods holds substantial promise for advancing 3D generation techniques. In this paper, we present BoostDream, a highly efficient plug-and-play 3D refining method designed to transform coarse 3D assets into high-quality. The BoostDream framework comprises three distinct processes: (1) We introduce 3D model distillation that fits differentiable representations from the 3D assets obtained through feed-forward generation. (2) A novel multi-view SDS loss is designed, which utilizes a multi-view aware 2D diffusion model to refine the 3D assets. (3) We propose to use prompt and multi-view consistent normal maps as guidance in refinement. Our extensive experiment is conducted on different differentiable 3D representations, revealing that BoostDream excels in generating high-quality 3D assets rapidly, overcoming the Janus problem compared to conventional SDS-based methods. This breakthrough signifies a substantial advancement in both the efficiency and quality of 3D generation processes.", "date": "2024-01-30", "categories": []} {"id": "2510.04236", "title": "Scaling Sequence-to-Sequence Generative Neural Rendering", "abstract": "We present Kaleido, a family of generative models designed for photorealistic, unified object- and scene-level neural rendering. Kaleido operates on the principle that 3D can be regarded as a specialised sub-domain of video, expressed purely as a sequence-to-sequence image synthesis task. Through a systemic study of scaling sequence-to-sequence generative neural rendering, we introduce key architectural innovations that enable our model to: i) perform generative view synthesis without explicit 3D representations; ii) generate any number of 6-DoF target views conditioned on any number of reference views via a masked autoregressive framework; and iii) seamlessly unify 3D and video modelling within a single decoder-only rectified flow transformer. Within this unified framework, Kaleido leverages large-scale video data for pre-training, which significantly improves spatial consistency and reduces reliance on scarce, camera-labelled 3D datasets -- all without any architectural modifications. Kaleido sets a new state-of-the-art on a range of view synthesis benchmarks. Its zero-shot performance substantially outperforms other generative methods in few-view settings, and, for the first time, matches the quality of per-scene optimisation methods in many-view settings.", "date": "2025-10-05", "categories": []} {"id": "2604.09132", "title": "Strips as Tokens: Artist Mesh Generation with Native UV Segmentation", "abstract": "Recent advancements in autoregressive transformers have demonstrated remarkable potential for generating artist-quality meshes. However, the token ordering strategies employed by existing methods typically fail to meet professional artist standards, where coordinate-based sorting yields inefficiently long sequences, and patch-based heuristics disrupt the continuous edge flow and structural regularity essential for high-quality modeling. To address these limitations, we propose Strips as Tokens (SATO), a novel framework with a token ordering strategy inspired by triangle strips. By constructing the sequence as a connected chain of faces that explicitly encodes UV boundaries, our method naturally preserves the organized edge flow and semantic layout characteristic of artist-created meshes. A key advantage of this formulation is its unified representation, enabling the same token sequence to be decoded into either a triangle or quadrilateral mesh. This flexibility facilitates joint training on both data types: large-scale triangle data provides fundamental structural priors, while high-quality quad data enhances the geometric regularity of the outputs. Extensive experiments demonstrate that SATO consistently outperforms prior methods in terms of geometric quality, structural coherence, and UV segmentation. Project page: https://ruixu.me/html/SATO/index.html", "date": "2026-04-10", "categories": []} {"id": "2605.01743", "title": "MOC-3D: Manifold-Order Consistency for Text-to-3D Generation", "abstract": "With the burgeoning development of fields such as the Metaverse, Virtual Reality (VR), and Digital Twins, text-to-3D generation has emerged as a research hotspot in both academia and industry. Currently, optimization methods based on Score Distillation Sampling (SDS) utilizing 2D diffusion priors have become the mainstream technological paradigm in this field. However, due to the view bias of 2D priors and the mode-seeking ambiguity combined with gradient noise induced by high Classifier-Free Guidance (CFG), these methods still suffer from macro-topological inconsistency (e.g., the Janus problem) and micro-geometric discontinuity.\n To address these challenges, we propose MOC-3D, a text-to-3D generation method based on geometric manifold and semantic view-order consistency. Built upon the ScaleDreamer framework, our method incorporates a Semantic View-Order Constraint Module and a Manifold-based Feature Continuity Module. The former aims to rectify macro-topological inconsistency, while the latter focuses on eliminating micro-geometric discontinuity. Specifically, the Semantic View-Order Constraint Module leverages the prior knowledge of CLIP to impose a Monotonicity Rank Constraint on semantic score representations across different views, thereby providing effective guidance for the global topological structure of 3D objects. Meanwhile, the Manifold-based Feature Continuity Module employs the Riemannian Metric on the Symmetric Positive Definite (SPD) manifold. By measuring the distance of feature statistical distributions in the Riemannian space, it promotes the smooth evolution and continuity of micro-textures across multi-views in a statistical sense. Under the macro-micro synergistic optimization of these two modules, our model can simultaneously improve macro-structural consistency and micro-detail continuity.", "date": "2026-05-03", "categories": []} {"id": "2605.02222", "title": "Generative Modeling with Orbit-Space Particle Flow Matching", "abstract": "We present Orbit-Space Geometric Probability Paths (OGPP), a particle-native flow-matching framework for generative modeling of particle systems. OGPP is motivated by two insights: (i) particles are defined up to permutation symmetries, so anonymous indexing inflates per-index target variance and yields curved, hard-to-learn flows; and (ii) particles live in physical space, so the flow terminal velocity has physical meaning and can encode geometric attributes, e.g., surface normals. OGPP instantiates three key components: (1) orbit-space canonicalization of the probability-path terminal endpoint, (2) particle index embeddings for role specialization, and (3) geometric probability paths with arc-length-aware terminal velocities that generate normals as a byproduct of the flow. We evaluate OGPP on minimal-surface benchmarks, where it reduces metric error by up to two orders of magnitude in a single inference step; on ShapeNet, where it matches the state of the art with 5x fewer steps and reaches airplane EMD comparable to DiT-3D with 26x fewer parameters and 5x fewer steps; and on single-shape encoding, where it produces normals and reconstructions competitive with 6D generators while operating entirely in 3D.", "date": "2026-05-04", "categories": []} {"id": "2605.02537", "title": "Orchestrating Spatial Semantics via a Zone-Graph Paradigm for Intricate Indoor Scene Generation", "abstract": "Autonomous 3D indoor scene synthesis breaks down in non-convex rooms with tightly coupled spatial constraints. Data-driven generators lack topological priors for long-horizon planning, while iterative agents fragment semantics and become geometrically brittle. We present ZoneMaestro, a unified framework that shifts the paradigm from object-centric synthesis to Zone-Graph Orchestration. By internalizing a novel zone-based logic, ZoneMaestro translates high-level semantic intent into functional zones and topological constraints, enabling robust adaptation to diverse architectural forms. To support this, we construct Zone-Scene-10K, a large-scale dataset enriched with explicit Zone-Graph annotations. We further introduce an Alternating Alignment Strategy that cycles between reasoning internalization and Zone-Aware Group Relative Policy Optimization (Z-GRPO), effectively reconciling the tension between semantic richness and geometric validity without relying on external physics engines. To rigorously evaluate spatial intelligence beyond convex primitives, we formally define the task of Intricate Spatial Orchestration and release SCALE, a stress-test benchmark for irregular indoor scenarios with complex, dense spatial relations. Extensive experiments demonstrate that ZoneMaestro resolves the density-safety dichotomy, significantly outperforming state-of-the-art baselines in both structural coherence and intent adherence.", "date": "2026-05-04", "categories": []} {"id": "2603.28980", "title": "Stepper: Stepwise Immersive Scene Generation with Multiview Panoramas", "abstract": "The synthesis of immersive 3D scenes from text is rapidly maturing, driven by novel video generative models and feed-forward 3D reconstruction, with vast potential in AR/VR and world modeling. While panoramic images have proven effective for scene initialization, existing approaches suffer from a trade-off between visual fidelity and explorability: autoregressive expansion suffers from context drift, while panoramic video generation is limited to low resolution. We present Stepper, a unified framework for text-driven immersive 3D scene synthesis that circumvents these limitations via stepwise panoramic scene expansion. Stepper leverages a novel multi-view 360{\\deg} diffusion model that enables consistent, high-resolution expansion, coupled with a geometry reconstruction pipeline that enforces geometric coherence. Trained on a new large-scale, multi-view panorama dataset, Stepper achieves state-of-the-art fidelity and structural consistency, outperforming prior approaches, thereby setting a new standard for immersive scene generation.", "date": "2026-03-30", "categories": []} {"id": "2605.00345", "title": "Pose-Aware Diffusion for 3D Generation", "abstract": "Generating pose-aligned 3D objects is challenging due to the spatial mismatches and transformation ambiguities inherent in decoupled canonical-then-rotate paradigms. To this end, we introduce Pose-Aware Diffusion (PAD), a novel end-to-end diffusion framework that synthesizes 3D geometry directly within the observation space. By unprojecting monocular depth into a partial point cloud and explicitly injecting it as a 3D geometric anchor, PAD abandons canonical assumptions to enforce rigorous spatial supervision. This native generation intrinsically resolves pose ambiguity, producing high-fidelity pose-aligned assets. Extensive experiments demonstrate that PAD achieves superior geometric alignment and image-to-3D correspondence compared to state-of-the-art methods. Additionally, PAD naturally extends to compositional 3D scene reconstruction via a simple union of independently generated objects, highlighting its robust ability to preserve precise spatial layouts.", "date": "2026-05-01", "categories": []} {"id": "2605.00632", "title": "BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code Synthesis", "abstract": "Automatic generation of executable Blender code from natural language remains challenging, with state-of-the-art LLMs producing frequent syntactic errors and geometrically inconsistent objects. We present BlenderRAG, a retrieval-augmented generation system that operates on a curated multimodal dataset of 500 expert-validated examples (text, code, image) across 50 object categories. By retrieving semantically similar examples during generation, BlenderRAG improves compilation success rates from 40.8% to 70.0% and semantic normalized alignment from 0.41 to 0.77 (CLIP similarity) across four state-of-the-art LLMs, without requiring fine-tuning or specialized hardware, making it immediately accessible for deployment. The dataset and code will be available at https://github.com/MaxRondelli/BlenderRAG.", "date": "2026-05-01", "categories": []} {"id": "2605.00664", "title": "InpaintSLat: Inpainting Structured 3D Latents via Initial Noise Optimization", "abstract": "We present a training-free approach for controllable 3D inpainting based on initial noise optimization. In the structured 3D latent diffusion framework, we observe that the underlying geometric structure is established during the early stages of the diffusion process and exhibits high sensitivity to the initial noise. Such characteristics compromise stability in tasks like inpainting and editing, where the model must ensure strict alignment with the existing context while synthesizing a new structure. In this paper, we introduce a strategy to optimize the initial noise within the structured 3D latent diffusion framework, ensuring high-fidelity 3D inpainting. Specifically, we update the initial noise by leveraging a backpropagation approximation grounded in the rectified flow model, with the spectral parameterization specially designed for robust and efficient structured 3D latent optimization. Experiments demonstrate consistent improvements in contextual consistency and prompt alignment over representative training-free inpainting baselines, establishing initial noise control as an independent dimension for 3D inpainting, orthogonal to conventional sampling trajectory manipulation.", "date": "2026-05-01", "categories": []} {"id": "2605.00781", "title": "Map2World: Segment Map Conditioned Text to 3D World Generation", "abstract": "3D world generation is essential for applications such as immersive content creation or autonomous driving simulation. Recent advances in 3D world generation have shown promising results; however, these methods are constrained by grid layouts and suffer from inconsistencies in object scale throughout the entire world. In this work, we introduce a novel framework, Map2World, that first enables 3D world generation conditioned on user-defined segment maps of arbitrary shapes and scales, ensuring global-scale consistency and flexibility across expansive environments. To further enhance the quality, we propose a detail enhancer network that generates fine details of the world. The detail enhancer enables the addition of fine-grained details without compromising overall scene coherence by incorporating global structure information. We design the entire pipeline to leverage strong priors from asset generators, achieving robust generalization across diverse domains, even under limited training data for scene generation. Extensive experiments demonstrate that our method significantly outperforms existing approaches in user-controllability, scale consistency, and content coherence, enabling users to generate 3D worlds under more complex conditions.", "date": "2026-05-01", "categories": []} {"id": "2503.01448", "title": "Generative Human Geometry Distribution", "abstract": "Realistic human geometry generation is an important yet challenging task, requiring both the preservation of fine clothing details and the accurate modeling of clothing-body interactions. To tackle this challenge, we build upon Geometry distributions, a recently proposed representation that can model a single human geometry with high fidelity using a flow matching model. However, extending a single-geometry distribution to a dataset is non-trivial and inefficient for large-scale learning. To address this, we propose a new geometry distribution model by two key techniques: (1) encoding distributions as 2D feature maps rather than network parameters, and (2) using SMPL models as the domain instead of Gaussian and refining the associated flow velocity field. We then design a generative framework adopting a two staged training paradigm analogous to state-of-the-art image and 3D generative models. In the first stage, we compress geometry distributions into a latent space using a diffusion flow model; the second stage trains another flow model on this latent space. We validate our approach on two key tasks: pose-conditioned random avatar generation and avatar-consistent novel pose synthesis. Experimental results demonstrate that our method outperforms existing state-of-the-art methods, achieving a 57% improvement in geometry quality.", "date": "2025-03-03", "categories": []} {"id": "2604.27329", "title": "SQuadGen: Generating Simple Quad Layouts via Chart Distance Fields", "abstract": "3D shapes from scanning, reconstruction, or AI-generated content often lack simple quad mesh layouts -- critical for efficient editing and modeling. Existing quad-remeshing techniques typically produce complex layouts with irregular loops, leading to tedious manual cleanup and extensive algorithm tuning. We introduce SQuadGen, a diffusion-based generative framework that leverages Chart Distance Fields (CDF) to synthesize simple quad layouts on 3D shapes. Our approach addresses two key challenges: (1) the discrete nature of mesh connectivity, which hinders learning, and (2) the scarcity of large-scale datasets with simple quad meshes. To overcome the first, we propose CDF, a continuous surface-based representation enabling effective learning and synthesis of quad layouts. To address the second, we define loop-aware simplicity metrics and construct a large-scale dataset of high-quality quad layouts recovered from public 3D repositories through a robust quad-recovery pipeline. Extensive evaluations across diverse 3D inputs show that SQuadGen consistently outperforms existing methods, producing robust, artist-friendly simple quad layouts.", "date": "2026-04-30", "categories": []} {"id": "2604.27361", "title": "CasLayout: Cascaded 3D Layout Diffusion for Indoor Scene Synthesis with Implicit Relation Modeling", "abstract": "Synthesizing realistic 3D indoor scenes remains challenging due to data scarcity and the difficulty of simultaneously enforcing global architectural constraints and local semantic consistency. Existing approaches often overlook structural boundaries or rely on fully connected relation graphs that introduce redundant generation errors. Inspired by human design cognition, we present CasLayout, a cascaded diffusion framework that decomposes the joint scene generation task into four conditional sub-stages with explicit physical and semantic roles: (1) predicting furniture quantity and categories, (2) refining object sizes and feature embeddings, (3) modeling spatial relationships in a latent space, and (4) generating Oriented Bounding Boxes (OBBs). This decoupled architecture reduces data requirements and enables flexible integration of Large Language Models (LLMs) and Vision Language Models (VLMs) for zero-shot tasks such as image-to-scene generation. To maintain physical validity within complex floor plans, we explicitly model building elements (e.g., walls, doors, and windows) as conditional constraints. Furthermore, to address the high entropy of dense relation graphs, we introduce a sparse relation graph formulation aligned with human spatial descriptions. By encoding these sparse graphs into a compact latent space using a bidirectional Variational Autoencoder (VAE), the proposed framework provides enhanced relational controllability, allowing generated layouts to better respect functional organization. Experiments demonstrate that CasLayout achieves state-of-the-art performance in fidelity and diversity while enabling improved controllability in practical applications.", "date": "2026-04-30", "categories": []} {"id": "2604.27504", "title": "REVIVE 3D: Refinement via Encoded Voluminous Inflated prior for Volume Enhancement", "abstract": "Recent generative models have shown strong performance in generating diverse 3D assets from 2D images, a fundamental research topic in computer vision and graphics. However, these models still struggle to generate voluminous 3D assets when the input is a flat image that provides limited 3D cues. We introduce REVIVE 3D, a two-stage, plug-and-play pipeline for generating voluminous 3D assets from flat images. In Stage 1, we construct an Inflated Prior by inflating the foreground silhouette to recover global volume and superimposing part-aware details to capture local structure. In Stage 2, 3D Latent Refinement injects Gaussian noise into the Inflated Prior's latent and then denoises it, using the prior's geometric cues to leverage the backbone's pretrained 3D knowledge. Furthermore, our framework supports image-conditioned 3D editing. To quantify volume and surface flatness, we propose Compactness and Normal Anisotropy. We validate Compactness and Normal Anisotropy through a user study, showing that these metrics align with human perception of volume and quality. We show that REVIVE 3D achieves state-of-the-art performance on a challenging flat image dataset, based on extensive qualitative and quantitative evaluations.", "date": "2026-04-30", "categories": []} {"id": "2604.27555", "title": "SpatialGrammar: A Domain-Specific Language for LLM-Based 3D Indoor Scene Generation", "abstract": "Automatically generating interactive 3D indoor scenes from natural language is crucial for virtual reality, gaming, and embodied AI. However, existing LLM-based approaches often suffer from spatial errors and collisions, in part because common scene representations-raw coordinates or verbose code-are difficult for models to reason about 3D spatial relationships and physical constraints. We propose SpatialGrammar, a domain-specific language that represents gravity-aligned indoor layouts as BEV grid placements with deterministic compilation to valid 3D geometry, enabling verifiable constraint checking. Building on this representation, we develop (1) SG-Agent, a closed-loop system that uses compiler feedback to iteratively refine scenes and enforce collision constraints, and (2) SG-Mini, a 104M-parameter model trained entirely on compiler-validated synthetic data. Across 159 test scenes spanning five scenarios of different complexity, SG-Agent improves spatial fidelity and physical plausibility over prior methods, while SG-Mini performs competitively against larger LLM-based baselines on single-shot generation scenarios.", "date": "2026-04-30", "categories": []} {"id": "2604.16552", "title": "Co-generation of Layout and Shape from Text via Autoregressive 3D Diffusion", "abstract": "Recent text-to-scene generation approaches largely reduced the manual efforts required to create 3D scenes. However, their focus is either to generate a scene layout or to generate objects, and few generate both. The generated scene layout is often simple even with LLM's help. Moreover, the generated scene is often inconsistent with the text input that contains non-trivial descriptions of the shape, appearance, and spatial arrangement of the objects. We present a new paradigm of sequential text-to-scene generation and propose a novel generative model for interactive scene creation. At the core is a 3D Autoregressive Diffusion model 3D-ARD+, which unifies the autoregressive generation over a multimodal token sequence and diffusion generation of next-object 3D latents. To generate the next object, the model uses one autoregressive step to generate the coarse-grained 3D latents in the scene space, conditioned on both the current seen text instructions and already synthesized 3D scene. It then uses a second step to generate the 3D latents in the smaller object space, which can be decoded into fine-grained object geometry and appearance. We curate a large dataset of 230K indoor scenes with paired text instructions for training. We evaluate 7B 3D-ARD+, on challenging scenes, and showcase the model can generate and place objects following non-trivial spatial layout and semantics prescribed by the text instructions.", "date": "2026-04-17", "categories": []} {"id": "2604.26917", "title": "AnimateAnyMesh++: A Flexible 4D Foundation Model for High-Fidelity Text-Driven Mesh Animation", "abstract": "Recent advances in 4D content generation have attracted increasing attention, yet creating high-quality animated 3D models remains challenging due to the complexity of modeling spatio-temporal distributions and the scarcity of 4D training data. We present AnimateAnyMesh++, a feed-forward framework for text-driven animation of arbitrary 3D meshes with substantial upgrades in data, architecture, and generative capability. First, we expand the DyMesh-XL dataset by mining dynamic content from Objaverse-XL, increasing the number of unique identities from 60K to 300K and substantially broadening category and motion diversity. Second, we redesign DyMeshVAE-Flex with power-law topology-aware attention and vertex-normal enhanced features, which significantly improves trajectory reconstruction, local geometry preservation, and mitigates trajectory-sticking artifacts. Third, we introduce architectural changes to both DyMeshVAE-Flex and the rectified-flow (RF) generator to support variable-length sequence training and generation, enabling longer animations while preserving reconstruction fidelity. Extensive experiments demonstrate that AnimateAnyMesh++ generates semantically accurate and temporally coherent mesh animations within seconds, surpassing prior approaches in quality and efficiency. The enlarged DyMesh-XL, the upgraded DyMeshVAE-Flex, and variable-length RF together deliver consistent gains across benchmarks and in-the-wild meshes. We will release code, models, and the expanded DyMesh-XL upon acceptance of this manuscript to facilitate research in 4D content creation.", "date": "2026-04-29", "categories": []} {"id": "2504.02316", "title": "ConsDreamer: Advancing Multi-View Consistency for Zero-Shot Text-to-3D Generation", "abstract": "Recent advances in zero-shot text-to-3D generation have revolutionized 3D content creation by enabling direct synthesis from textual descriptions. While state-of-the-art methods leverage 3D Gaussian Splatting with score distillation to enhance multi-view rendering through pre-trained text-to-image (T2I) models, they suffer from inherent prior view biases in T2I priors. These biases lead to inconsistent 3D generation, particularly manifesting as the multi-face Janus problem, where objects exhibit conflicting features across views. To address this fundamental challenge, we propose ConsDreamer, a novel method that mitigates view bias by refining both the conditional and unconditional terms in the score distillation process: (1) a View Disentanglement Module (VDM) that eliminates viewpoint biases in conditional prompts by decoupling irrelevant view components and injecting precise view control; and (2) a similarity-based partial order loss that enforces geometric consistency in the unconditional term by aligning cosine similarities with azimuth relationships. Extensive experiments demonstrate that ConsDreamer can be seamlessly integrated into various 3D representations and score distillation paradigms, effectively mitigating the multi-face Janus problem.", "date": "2025-04-03", "categories": []} {"id": "2602.05676", "title": "ShapeUP: Scalable Image-Conditioned 3D Editing", "abstract": "Recent advancements in 3D foundation models have enabled the generation of high-fidelity assets, yet precise 3D manipulation remains a significant challenge. Existing 3D editing frameworks often face a difficult trade-off between visual controllability, geometric consistency, and scalability. Specifically, optimization-based methods are prohibitively slow, multi-view 2D propagation techniques suffer from visual drift, and training-free latent manipulation methods are inherently bound by frozen priors and cannot directly benefit from scaling. In this work, we present ShapeUP, a scalable, image-conditioned 3D editing framework that formulates editing as a supervised latent-to-latent translation within a native 3D representation. This formulation allows ShapeUP to build on a pretrained 3D foundation model, leveraging its strong generative prior while adapting it to editing through supervised training. In practice, ShapeUP is trained on triplets consisting of a source 3D shape, an edited 2D image, and the corresponding edited 3D shape, and learns a direct mapping using a 3D Diffusion Transformer (DiT). This image-as-prompt approach enables fine-grained visual control over both local and global edits and achieves implicit, mask-free localization, while maintaining strict structural consistency with the original asset. Our extensive evaluations demonstrate that ShapeUP consistently outperforms current trained and training-free baselines in both identity preservation and edit fidelity, offering a robust and scalable paradigm for native 3D content creation.", "date": "2026-02-05", "categories": []} {"id": "2604.22828", "title": "MetaEarth3D: Unlocking World-scale 3D Generation with Spatially Scalable Generative Modeling", "abstract": "Recent generative AI models have achieved remarkable breakthroughs in language and visual understanding. However, although these models can generate realistic visual content, their spatial scale remains confined to bounded environments, preventing them from capturing how geographic environments evolve across thousands of kilometers or from modeling the spatial structure of the large-scale physical world. This limitation poses a critical challenge for ultra-wide-area spatial intelligence in Earth observation and simulation, revealing a deeper gap in generative AI: progress has relied primarily on scaling model parameters and training data, while overlooking spatial scale as a core dimension of intelligence. Here, motivated by this missing dimension, we investigate spatial scale as a new scaling axis in foundation models and present MetaEarth3D, the first generative foundation model capable of spatially consistent generation at the planetary scale. Taking optical Earth observation simulation as a testbed, MetaEarth3D enables the generation of multi-level, unbounded, and diverse 3D scenes spanning large-scale terrains, medium-scale cities, and fine-grained street blocks. Built upon 10 million globally distributed real-world training images, MetaEarth3D demonstrates both strong visual realism and geospatial statistical realism. Beyond generation, MetaEarth3D serves as a generative data engine for diverse virtual environments in ultra-wide spatial intelligence. We argue that this study may help empower next-generation spatial intelligence for Earth observation.", "date": "2026-04-19", "categories": []} {"id": "2604.22847", "title": "Dream-Cubed: Controllable Generative Modeling in Minecraft by Training on Billions of Cubes", "abstract": "We introduce Dream-Cubed, a large-scale dataset of Minecraft worlds at voxel resolution, and a family of models using cubes as powerful compositional units for efficient generation of interactive 3D environments. Dream-Cubed comprises tens of billions of tokens from a carefully curated mixture of procedural biome terrain and high-quality human-authored maps. We use this dataset to conduct the first large-scale study of 3D diffusion models for voxel generation, analyzing discrete and continuous diffusion formulations, data compositions, and architectural design choices. Our models operate directly in the space of blocks, enabling efficient and semantically grounded generation while supporting interactive user workflows such as inpainting and outpainting from user-authored blocks. To quantitatively evaluate our models, we adapt the FID metric to assess semantic differences between real and generated world renderings, and validate generation quality through a human preference study. We release the full dataset, code, and all our pretrained models, which we hope will provide a foundation for future research in efficient generative modeling for structured, interactive 3D environments.", "date": "2026-04-22", "categories": []} {"id": "2604.23010", "title": "GenAssets: Generating in-the-wild 3D Assets in Latent Space", "abstract": "High-quality 3D assets for traffic participants are critical for multi-sensor simulation, which is essential for the safe end-to-end development of autonomy. Building assets from in-the-wild data is key for diversity and realism, but existing neural-rendering based reconstruction methods are slow and generate assets that render well only from viewpoints close to the original observations, limiting their usefulness in simulation. Recent diffusion-based generative models build complete and diverse assets, but perform poorly on in-the-wild driving scenes, where observed actors are captured under sparse and limited fields of view, and are partially occluded. In this work, we propose a 3D latent diffusion model that learns on in-the-wild LiDAR and camera data captured by a sensor platform and generates high-quality 3D assets with complete geometry and appearance. Key to our method is a \"reconstruct-then-generate\" approach that first leverages occlusion-aware neural rendering trained over multiple scenes to build a high-quality latent space for objects, and then trains a diffusion model that operates on the latent space. We show our method outperforms existing reconstruction and generation based methods, unlocking diverse and scalable content creation for simulation.", "date": "2026-04-24", "categories": []} {"id": "2509.24004", "title": "SIE3D: Single-Image Expressive 3D Avatar Generation via Semantic Embedding and Perceptual Expression Loss", "abstract": "Generating high-fidelity 3D head avatars from a single image is challenging, as current methods lack fine-grained, intuitive control over expressions via text. This paper proposes SIE3D, a framework that generates expressive 3D avatars from a single image and descriptive text. SIE3D fuses identity features from the image with semantic embedding from text through a novel conditioning scheme, enabling detailed control. To ensure generated expressions accurately match the text, it introduces an innovative perceptual expression loss function. This loss uses a pre-trained expression classifier to regularize the generation process, guaranteeing expression accuracy. Extensive experiments show SIE3D significantly improves controllability and realism, outperforming competitive methods in identity preservation and expression fidelity on a single consumer-grade GPU. Project page: https://huang-zhiqi.github.io/SIE3D/", "date": "2025-09-28", "categories": []} {"id": "2604.17801", "title": "View-Consistent 3D Scene Editing via Dual-Path Structural Correspondense and Semantic Continuity", "abstract": "Text-driven 3D scene editing has recently attracted increasing attention. Most existing methods follow a render-edit-optimize pipeline, where multi-view images are rendered from a 3D scene, edited with 2D image editors, and then used to optimize the underlying 3D representation. However, cross-view inconsistency remains a major bottleneck. Although recent methods introduce geometric cues, cross-view interactions, or video priors to mitigate this issue, they still largely rely on inference-time synchronization and thus remain limited in robustness and generalization.In this work, we recast multi-view consistent 3D editing from a distributional perspective: 3D scene editing essentially requires a joint distribution modeling across viewpoints.Based on this insight, we propose a view-consistent 3D editing framework that explicitly introduces cross-view dependencies into the editing process. Furthermore, motivated by the observation that structural correspondence and semantic continuity rely on different cross-view cues, we introduce a dual-path consistency mechanism consisting of projection-guided structural guidance and patch-level semantic propagation for effective cross-view editing. Further, we construct a paired multi-view editing dataset that provides reliable supervision for learning cross-view consistency in edited scenes. Extensive experiments demonstrate that our method achieves superior editing performance with precise and consistent views for complex scenes.", "date": "2026-04-20", "categories": []} {"id": "2511.21978", "title": "PAT3D: Physics-Augmented Text-to-3D Scene Generation", "abstract": "We introduce PAT3D, the first physics-augmented text-to-3D scene generation framework that integrates vision-language models with physics-based simulation to produce physically plausible, simulation-ready, and intersection-free 3D scenes. Given a text prompt, PAT3D generates 3D objects, infers their spatial relations, and organizes them into a hierarchical scene tree, which is then converted into initial conditions for simulation. A differentiable rigid-body simulator ensures realistic object interactions under gravity, driving the scene toward static equilibrium without interpenetrations. To further enhance scene quality, we introduce a simulation-in-the-loop optimization procedure that guarantees physical stability and non-intersection, while improving semantic consistency with the input prompt. Experiments demonstrate that PAT3D substantially outperforms prior approaches in physical plausibility, semantic consistency, and visual quality. Beyond high-quality generation, PAT3D uniquely enables simulation-ready 3D scenes for downstream tasks such as scene editing and robotic manipulation. Code and data are available at: https://github.com/Simulation-Intelligence/PAT3D.", "date": "2025-11-26", "categories": []} {"id": "2604.21592", "title": "Sculpt4D: Generating 4D Shapes via Sparse-Attention Diffusion Transformers", "abstract": "Recent breakthroughs in 3D generative modeling have yielded remarkable progress in static shape synthesis, yet high-fidelity dynamic 4D generation remains elusive, hindered by temporal artifacts and prohibitive computational demand. We present Sculpt4D, a native 4D generative framework that seamlessly integrates efficient temporal modeling into a pretrained 3D Diffusion Transformer (Hunyuan3D 2.1), thereby mitigating the scarcity of 4D training data. At its core lies a Block Sparse Attention mechanism that preserves object identity by anchoring to the initial frame while capturing rich motion dynamics via a time-decaying sparse mask. This design faithfully models complex spatiotemporal dependencies with high fidelity, while sidestepping the quadratic overhead of full attention and reducing network total computation by 56%. Consequently, Sculpt4D establishes a new state-of-the-art in temporally coherent 4D synthesis and charts a path toward efficient and scalable 4D generation.", "date": "2026-04-23", "categories": []} {"id": "2604.19907", "title": "SceneOrchestra: Efficient Agentic 3D Scene Synthesis via Full Tool-Call Trajectory Generation", "abstract": "Recent agentic frameworks for 3D scene synthesis have advanced realism and diversity by integrating heterogeneous generation and editing tools. These tools are organized into workflows orchestrated by an off-the-shelf LLM. Current approaches typically adopt an execute-review-reflect loop: at each step, the orchestrator executes a tool, renders intermediate results for review, and then decides on the tool and its parameters for the next step. However, this design has two key limitations. First, next-step tool selection and parameter configuration are driven by heuristic rules, which can lead to suboptimal execution flows, unnecessary tool invocations, degraded output quality, and increased runtime. Second, rendering and reviewing intermediate results after each step introduces additional latency. To address these issues, we propose SceneOrchestra, a trainable orchestration framework that optimizes the tool-call execution flow and eliminates the step-by-step review loop, improving both efficiency and output quality. SceneOrchestra consists of an orchestrator and a discriminator, which we fine-tune with a two-phase training strategy. In the first phase, the orchestrator learns context-aware tool selection and complete tool-call trajectory generation, while the discriminator is trained to assess the quality of full trajectories, enabling it to select the best trajectory from multiple candidates. In the second phase, we perform interleaved training, where the discriminator adapts to the orchestrator's evolving trajectory distribution and distills its discriminative capability back into the orchestrator. At inference, we only use the orchestrator to generate and execute full tool-call trajectories from instructions, without requiring the discriminator. Extensive experiments show that our method achieves state-of-the-art scene quality while reducing runtime compared to previous work.", "date": "2026-04-21", "categories": []} {"id": "2604.20038", "title": "FluSplat: Sparse-View 3D Editing without Test-Time Optimization", "abstract": "Recent advances in text-guided image editing and 3D Gaussian Splatting (3DGS) have enabled high-quality 3D scene manipulation. However, existing pipelines rely on iterative edit-and-fit optimization at test time, alternating between 2D diffusion editing and 3D reconstruction. This process is computationally expensive, scene-specific, and prone to cross-view inconsistencies.\n We propose a feed-forward framework for cross-view consistent 3D scene editing from sparse views. Instead of enforcing consistency through iterative 3D refinement, we introduce a cross-view regularization scheme in the image domain during training. By jointly supervising multi-view edits with geometric alignment constraints, our model produces view-consistent results without per-scene optimization at inference. The edited views are then lifted into 3D via a feedforward 3DGS model, yielding a coherent 3DGS representation in a single forward pass.\n Experiments demonstrate competitive editing fidelity and substantially improved cross-view consistency compared to optimization-based methods, while reducing inference time by orders of magnitude.", "date": "2026-04-21", "categories": []} {"id": "2604.19202", "title": "SketchFaceGS: Real-Time Sketch-Driven Face Editing and Generation with Gaussian Splatting", "abstract": "3D Gaussian representations have emerged as a powerful paradigm for digital head modeling, achieving photorealistic quality with real-time rendering. However, intuitive and interactive creation or editing of 3D Gaussian head models remains challenging. Although 2D sketches provide an ideal interaction modality for fast, intuitive conceptual design, they are sparse, depth-ambiguous, and lack high-frequency appearance cues, making it difficult to infer dense, geometrically consistent 3D Gaussian structures from strokes - especially under real-time constraints. To address these challenges, we propose SketchFaceGS, the first sketch-driven framework for real-time generation and editing of photorealistic 3D Gaussian head models from 2D sketches. Our method uses a feed-forward, coarse-to-fine architecture. A Transformer-based UV feature-prediction module first reconstructs a coarse but geometrically consistent UV feature map from the input sketch, and then a 3D UV feature enhancement module refines it with high-frequency, photorealistic detail to produce a high-fidelity 3D head. For editing, we introduce a UV Mask Fusion technique combined with a layer-by-layer feature-fusion strategy, enabling precise, real-time, free-viewpoint modifications. Extensive experiments show that SketchFaceGS outperforms existing methods in both generation fidelity and editing flexibility, producing high-quality, editable 3D heads from sketches in a single forward pass.", "date": "2026-04-21", "categories": []} {"id": "2601.22858", "title": "Learning to Build Shapes by Extrusion", "abstract": "We introduce Text Encoded Extrusions (TEE), a text-based representation that expresses mesh construction as sequences of face extrusions rather than polygon lists, and a method for generating 3D meshes from TEE using a large language model (LLM). By learning extrusion sequences that assemble a mesh, similar to the way artists create meshes, our approach naturally supports arbitrary output face counts and produces manifold meshes by design, in contrast to recent mesh generative transformer based models. The learnt extrusion sequences can also be applied to existing meshes - enabling editing in addition to generation. To train our model, we decompose a library of quadrilateral meshes with non-self-intersecting face loops into constituent loops, which can be viewed as their building blocks, and finetune an LLM on the steps for reassembling the quadrilateral meshes by performing a sequence of extrusions. We demonstrate that our representation enables reconstruction, novel shape synthesis, and the addition of new features to existing meshes.", "date": "2026-01-30", "categories": []} {"id": "2604.04787", "title": "AvatarPointillist: AutoRegressive 4D Gaussian Avatarization", "abstract": "We introduce AvatarPointillist, a novel framework for generating dynamic 4D Gaussian avatars from a single portrait image. At the core of our method is a decoder-only Transformer that autoregressively generates a point cloud for 3D Gaussian Splatting. This sequential approach allows for precise, adaptive construction, dynamically adjusting point density and the total number of points based on the subject's complexity. During point generation, the AR model also jointly predicts per-point binding information, enabling realistic animation. After generation, a dedicated Gaussian decoder converts the points into complete, renderable Gaussian attributes. We demonstrate that conditioning the decoder on the latent features from the AR generator enables effective interaction between stages and markedly improves fidelity. Extensive experiments validate that AvatarPointillist produces high-quality, photorealistic, and controllable avatars. We believe this autoregressive formulation represents a new paradigm for avatar generation, and we will release our code inspire future research.", "date": "2026-04-06", "categories": []} {"id": "2604.17472", "title": "UniMesh: Unifying 3D Mesh Understanding and Generation", "abstract": "Recent advances in 3D vision have led to specialized models for either 3D understanding (e.g., shape classification, segmentation, reconstruction) or 3D generation (e.g., synthesis, completion, and editing). However, these tasks are often tackled in isolation, resulting in fragmented architectures and representations that hinder knowledge transfer and holistic scene modeling. To address these challenges, we propose UniMesh, a unified framework that jointly learns 3D generation and understanding within a single architecture. First, we introduce a novel Mesh Head that acts as a cross model interface, bridging diffusion based image generation with implicit shape decoders. Second, we develop Chain of Mesh (CoM), a geometric instantiation of iterative reasoning that enables user driven semantic mesh editing through a closed loop latent, prompting, and re generation cycle. Third, we incorporate a self reflection mechanism based on an Actor Evaluator Self reflection triad to diagnose and correct failures in high level tasks like 3D captioning. Experimental results demonstrate that UniMesh not only achieves competitive performance on standard benchmarks but also unlocks novel capabilities in iterative editing and mutual enhancement between generation and understanding. Code: https://github.com/AIGeeksGroup/UniMesh. Website: https://aigeeksgroup.github.io/UniMesh.", "date": "2026-04-19", "categories": []} {"id": "2504.10466", "title": "Art3D: Training-Free 3D Generation from Flat-Colored Illustration", "abstract": "Large-scale pre-trained image-to-3D generative models have exhibited remarkable capabilities in diverse shape generations. However, most of them struggle to synthesize plausible 3D assets when the reference image is flat-colored like hand drawings due to the lack of 3D illusion, which are often the most user-friendly input modalities in art content creation. To this end, we propose Art3D, a training-free method that can lift flat-colored 2D designs into 3D. By leveraging structural and semantic features with pre-trained 2D image generation models and a VLM-based realism evaluation, Art3D successfully enhances the three-dimensional illusion in reference images, thus simplifying the process of generating 3D from 2D, and proves adaptable to a wide range of painting styles. To benchmark the generalization performance of existing image-to-3D models on flat-colored images without 3D feeling, we collect a new dataset, Flat-2D, with over 100 samples. Experimental results demonstrate the performance and robustness of Art3D, exhibiting superior generalizable capacity and promising practical applicability. Our source code and dataset will be publicly available on our project page: https://joy-jy11.github.io/ .", "date": "2025-04-14", "categories": []} {"id": "2604.16299", "title": "Repurposing 3D Generative Model for Autoregressive Layout Generation", "abstract": "We introduce LaviGen, a framework that repurposes 3D generative models for 3D layout generation. Unlike previous methods that infer object layouts from textual descriptions, LaviGen operates directly in the native 3D space, formulating layout generation as an autoregressive process that explicitly models geometric relations and physical constraints among objects, producing coherent and physically plausible 3D scenes. To further enhance this process, we propose an adapted 3D diffusion model that integrates scene, object, and instruction information and employs a dual-guidance self-rollout distillation mechanism to improve efficiency and spatial accuracy. Extensive experiments on the LayoutVLM benchmark show LaviGen achieves superior 3D layout generation performance, with 19% higher physical plausibility than the state of the art and 65% faster computation. Our code is publicly available at https://github.com/fenghora/LaviGen.", "date": "2026-04-17", "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": "2604.14302", "title": "Geometrically Consistent Multi-View Scene Generation from Freehand Sketches", "abstract": "We tackle a new problem: generating geometrically consistent multi-view scenes from a single freehand sketch. Freehand sketches are the most geometrically impoverished input one could offer a multi-view generator. They convey scene intent through abstract strokes while introducing spatial distortions that actively conflict with any consistent 3D interpretation. No prior method attempts this; existing multi-view approaches require photographs or text, while sketch-to-3D methods need multiple views or costly per-scene optimisation.\n We address three compounding challenges; absent training data, the need for geometric reasoning from distorted 2D input, and cross-view consistency, through three mutually reinforcing contributions: (i) a curated dataset of $\\sim$9k sketch-to-multiview samples, constructed via an automated generation and filtering pipeline; (ii) Parallel Camera-Aware Attention Adapters (CA3) that inject geometric inductive biases into the video transformer; and (iii) a Sparse Correspondence Supervision Loss (CSL) derived from Structure-from-Motion reconstructions.\n Our framework synthesizes all views in a single denoising process without requiring reference images, iterative refinement, or per-scene optimization. Our approach significantly outperforms state-of-the-art two-stage baselines, improving realism (FID) by over 60% and geometric consistency (Corr-Acc) by 23%, while providing up to a 3.7$\\times$ inference speedup.", "date": "2026-04-15", "categories": []} {"id": "2604.14782", "title": "One-shot Compositional 3D Head Avatars with Deformable Hair", "abstract": "We propose a compositional method for constructing a complete 3D head avatar from a single image. Prior one-shot holistic approaches frequently fail to produce realistic hair dynamics during animation, largely due to inadequate decoupling of hair from the facial region, resulting in entangled geometry and unnatural deformations. Our method explicitly decouples hair from the face, modeling these components using distinct deformation paradigms while integrating them into a unified rendering pipeline. Furthermore, by leveraging image-to-3D lifting techniques, we preserve fine-grained textures from the input image to the greatest extent possible, effectively mitigating the common issue of high-frequency information loss in generalized models. Specifically, given a frontal portrait image, we first perform hair removal to obtain a bald image. Both the original image and the bald image are then lifted to dense, detail-rich 3D Gaussian Splatting (3DGS) representations. For the bald 3DGS, we rig it to a FLAME mesh via non-rigid registration with a prior model, enabling natural deformation that follows the mesh triangles during animation. For the hair component, we employ semantic label supervision combined with a boundary-aware reassignment strategy to extract a clean and isolated set of hair Gaussians. To control hair deformation, we introduce a cage structure that supports Position-Based Dynamics (PBD) simulation, allowing realistic and physically plausible transformations of the hair Gaussian primitives under head motion, gravity, and inertial effects. Striking qualitative results, including dynamic animations under diverse head motions, gravity effects, and expressions, showcase substantially more realistic hair behavior alongside faithfully preserved facial details, outperforming state-of-the-art one-shot methods in perceptual realism.", "date": "2026-04-16", "categories": []} {"id": "2604.14914", "title": "Beyond Prompts: Unconditional 3D Inversion for Out-of-Distribution Shapes", "abstract": "Text-driven inversion of generative models is a core paradigm for manipulating 2D or 3D content, unlocking numerous applications such as text-based editing, style transfer, or inverse problems. However, it relies on the assumption that generative models remain sensitive to natural language prompts. We demonstrate that for state-of-the-art native text-to-3D generative models, this assumption often collapses. We identify a critical failure mode where generation trajectories are drawn into latent ``sink traps'': regions where the model becomes insensitive to prompt modifications. In these regimes, changes to the input text fail to alter internal representations in a way that alters the output geometry. Crucially, we observe that this is not a limitation of the model's \\textit{geometric} expressivity; the same generative models possess the ability to produce a vast diversity of shapes but, as we demonstrate, become insensitive to out-of-distribution \\textit{text} guidance. We investigate this behavior by analyzing the sampling trajectories of the generative model, and find that complex geometries can still be represented and produced by leveraging the model's unconditional generative prior. This leads to a more robust framework for text-based 3D shape editing that bypasses latent sinks by decoupling a model's geometric representation power from its linguistic sensitivity. Our approach addresses the limitations of current 3D pipelines and enables high-fidelity semantic manipulation of out-of-distribution 3D shapes. Project webpage: https://daidedou.sorpi.fr/publication/beyondprompts", "date": "2026-04-16", "categories": []} {"id": "2604.13688", "title": "Beyond Voxel 3D Editing: Learning from 3D Masks and Self-Constructed Data", "abstract": "3D editing refers to the ability to apply local or global modifications to 3D assets. Effective 3D editing requires maintaining semantic consistency by performing localized changes according to prompts, while also preserving local invariance so that unchanged regions remain consistent with the original. However, existing approaches have significant limitations: multi-view editing methods incur losses when projecting back to 3D, while voxel-based editing is constrained in both the regions that can be modified and the scale of modifications. Moreover, the lack of sufficiently large editing datasets for training and evaluation remains a challenge. To address these challenges, we propose a Beyond Voxel 3D Editing (BVE) framework with a self-constructed large-scale dataset specifically tailored for 3D editing. Building upon this dataset, our model enhances a foundational image-to-3D generative architecture with lightweight, trainable modules, enabling efficient injection of textual semantics without the need for expensive full-model retraining. Furthermore, we introduce an annotation-free 3D masking strategy to preserve local invariance, maintaining the integrity of unchanged regions during editing. Extensive experiments demonstrate that BVE achieves superior performance in generating high-quality, text-aligned 3D assets, while faithfully retaining the visual characteristics of the original input.", "date": "2026-04-15", "categories": []} {"id": "2604.13730", "title": "ReConText3D: Replay-based Continual Text-to-3D Generation", "abstract": "Continual learning enables models to acquire new knowledge over time while retaining previously learned capabilities. However, its application to text-to-3D generation remains unexplored. We present ReConText3D, the first framework for continual text-to-3D generation. We first demonstrate that existing text-to-3D models suffer from catastrophic forgetting under incremental training. ReConText3D enables generative models to incrementally learn new 3D categories from textual descriptions while preserving the ability to synthesize previously seen assets. Our method constructs a compact and diverse replay memory through text-embedding k-Center selection, allowing representative rehearsal of prior knowledge without modifying the underlying architecture. To systematically evaluate continual text-to-3D learning, we introduce Toys4K-CL, a benchmark derived from the Toys4K dataset that provides balanced and semantically diverse class-incremental splits. Extensive experiments on the Toys4K-CL benchmark show that ReConText3D consistently outperforms all baselines across different generative backbones, maintaining high-quality generation for both old and new classes. To the best of our knowledge, this work establishes the first continual learning framework and benchmark for text-to-3D generation, opening a new direction for incremental 3D generative modeling. Project page is available at: https://mauk95.github.io/ReConText3D/.", "date": "2026-04-15", "categories": []} {"id": "2604.13905", "title": "Rethinking Image-to-3D Generation with Sparse Queries: Efficiency, Capacity, and Input-View Bias", "abstract": "We present SparseGen, a novel framework for efficient image-to-3D generation, which exhibits low input-view bias while being significantly faster. Unlike traditional approaches that rely on dense volumetric grids, triplanes, or pixel-aligned primitives, we model scenes with a compact sparse set of learned 3D anchor queries and a learned expansion operator that decodes each transformed query into a small local set of 3D Gaussian primitives. Trained under a rectified-flow reconstruction objective without 3D supervision, our model learns to allocate representation capacity where geometry and appearance matter, achieving significant reductions in memory and inference time while preserving multi-view fidelity. We introduce quantitative measures of input-view bias and utilization to show that sparse queries reduce overfitting to conditioning views while being representationally efficient. Our results argue that sparse set-latent expansion is a principled, practical alternative for efficient 3D generative modeling.", "date": "2026-04-15", "categories": []} {"id": "2604.08746", "title": "AniGen: Unified $S^3$ Fields for Animatable 3D Asset Generation", "abstract": "Animatable 3D assets, defined as geometry equipped with an articulated skeleton and skinning weights, are fundamental to interactive graphics, embodied agents, and animation production. While recent 3D generative models can synthesize visually plausible shapes from images, the results are typically static. Obtaining usable rigs via post-hoc auto-rigging is brittle and often produces skeletons that are topologically inconsistent with the generated geometry. We present AniGen, a unified framework that directly generates animate-ready 3D assets conditioned on a single image. Our key insight is to represent shape, skeleton, and skinning as mutually consistent $S^3$ Fields (Shape, Skeleton, Skin) defined over a shared spatial domain. To enable the robust learning of these fields, we introduce two technical innovations: (i) a confidence-decaying skeleton field that explicitly handles the geometric ambiguity of bone prediction at Voronoi boundaries, and (ii) a dual skin feature field that decouples skinning weights from specific joint counts, allowing a fixed-architecture network to predict rigs of arbitrary complexity. Built upon a two-stage flow-matching pipeline, AniGen first synthesizes a sparse structural scaffold and then generates dense geometry and articulation in a structured latent space. Extensive experiments demonstrate that AniGen substantially outperforms state-of-the-art sequential baselines in rig validity and animation quality, generalizing effectively to in-the-wild images across diverse categories including animals, humanoids, and machinery. Homepage: https://yihua7.github.io/AniGen-web/", "date": "2026-04-09", "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": "2412.10273", "title": "How to Spin an Object: First, Get the Shape Right", "abstract": "Image-to-3D models increasingly rely on hierarchical generation to disentangle geometry and texture. However, the design choices underlying these two-stage models--particularly the optimal choice of intermediate geometric representations--remain largely understudied. To investigate this, we introduce unPIC (undo-a-Picture), a modular framework for empirical analysis of image-to-3D pipelines. By factorizing the generation process into a multiview-geometry prior followed by an appearance decoder, unPIC enables a rigorous comparison of intermediate geometry representations. Through this framework, we identify that a specific representation, Camera-Relative Object Coordinates (CROCS), significantly outperforms alternatives such as depth maps, pretrained visual features, and other pointmap-based representations. We demonstrate that CROCS is not only easier for the first-stage geometry prior to predict, but also serves as an effective conditioning signal for ensuring 360-degree consistency during appearance decoding. Another advantage is that CROCS enables fully feedforward, direct 3D point cloud generation without requiring a separate post-hoc reconstruction step. Our unPIC formulation utilizing CROCS achieves superior novel-view quality, geometric accuracy, and multiview consistency; it outperforms leading baselines, including InstantMesh, Direct3D, CAT3D, Free3D, and EscherNet, on datasets of real-world 3D captures like Google Scanned Objects and the Digital Twin Catalog.", "date": "2024-12-13", "categories": []} {"id": "2604.02736", "title": "THOM: Generating Physically Plausible Hand-Object Meshes From Text", "abstract": "Generating photorealistic 3D hand-object interactions (HOIs) from text is important for applications like robotic grasping and AR/VR content creation. In practice, however, achieving both visual fidelity and physical plausibility remains difficult, as mesh extraction from text-generated Gaussians is inherently ill-posed and the resulting meshes are often unreliable for physics-based optimization. We present THOM, a training-free framework that generates physically plausible 3D HOI meshes directly from text prompts, without requiring template object meshes. THOM follows a two-stage pipeline: it first generates hand and object Gaussians guided by text, and then refines their interaction using physics-based optimization. To enable reliable interaction modeling, we introduce a mesh extraction method with an explicit vertex-to-Gaussian mapping, which enables topology-aware regularization. We further improve physical plausibility through contact-aware optimization and vision-language model (VLM)-guided translation refinement. Extensive experiments show that THOM produces high-quality HOIs with strong text alignment, visual realism, and interaction plausibility.", "date": "2026-04-03", "categories": []} {"id": "2604.10125", "title": "PhyMix: Towards Physically Consistent Single-Image 3D Indoor Scene Generation with Implicit--Explicit Optimization", "abstract": "Existing single-image 3D indoor scene generators often produce results that look visually plausible but fail to obey real-world physics, limiting their reliability in robotics, embodied AI, and design. To examine this gap, we introduce a unified Physics Evaluator that measures four main aspects: geometric priors, contact, stability, and deployability, which are further decomposed into nine sub-constraints, establishing the first benchmark to measure physical consistency. Based on this evaluator, our analysis shows that state-of-the-art methods remain largely physics-unaware. To overcome this limitation, we further propose a framework that integrates feedback from the Physics Evaluator into both training and inference, enhancing the physical plausibility of generated scenes. Specifically, we propose PhyMix, which is composed of two complementary components: (i) implicit alignment via Scene-GRPO, a critic-free group-relative policy optimization that leverages the Physics Evaluator as a preference signal and biases sampling towards physically feasible layouts, and (ii) explicit refinement via a plug-and-play Test-Time Optimizer (TTO) that uses differentiable evaluator signals to correct residual violations during generation. Overall, our method unifies evaluation, reward shaping, and inference-time correction, producing 3D indoor scenes that are visually faithful and physically plausible. Extensive synthetic evaluations confirm state-of-the-art performance in both visual fidelity and physical plausibility, and extensive qualitative examples in stylized and real-world images further showcase the robustness of the method. We will release codes and models upon publication.", "date": "2026-04-11", "categories": []} {"id": "2604.10772", "title": "HOG-Layout: Hierarchical 3D Scene Generation, Optimization and Editing via Vision-Language Models", "abstract": "3D layout generation and editing play a crucial role in Embodied AI and immersive VR interaction. However, manual creation requires tedious labor, while data-driven generation often lacks diversity. The emergence of large models introduces new possibilities for 3D scene synthesis. We present HOG-Layout that enables text-driven hierarchical scene generation, optimization and real-time scene editing with large language models (LLMs) and vision-language models (VLMs). HOG-Layout improves scene semantic consistency and plausibility through retrieval-augmented generation (RAG) technology, incorporates an optimization module to enhance physical consistency, and adopts a hierarchical representation to enhance inference and optimization, achieving real-time editing. Experimental results demonstrate that HOG-Layout produces more reasonable environments compared with existing baselines, while supporting fast and intuitive scene editing.", "date": "2026-04-12", "categories": []} {"id": "2604.11331", "title": "Any 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale", "abstract": "3D scene generation has long been dominated by 2D multi-view or video diffusion models. This is due not only to the lack of scene-level 3D latent representation, but also to the fact that most scene-level 3D visual data exists in the form of multi-view images or videos, which are naturally compatible with 2D diffusion architectures. Typically, these 2D-based approaches degrade 3D spatial extrapolation to 2D temporal extension, which introduces two fundamental issues: (i) representing 3D scenes via 2D views leads to significant representation redundancy, and (ii) latent space rooted in 2D inherently limits the spatial consistency of the generated 3D scenes. In this paper, we propose, for the first time, to perform 3D scene generation directly within an implicit 3D latent space to address these limitations. First, we repurpose frozen 2D representation encoders to construct our 3D Representation Autoencoder (3DRAE), which grounds view-coupled 2D semantic representations into a view-decoupled 3D latent representation. This enables representing 3D scenes observed from arbitrary numbers of views--at any resolution and aspect ratio--with fixed complexity and rich semantics. Then we introduce 3D Diffusion Transformer (3DDiT), which performs diffusion modeling in this 3D latent space, achieving remarkably efficient and spatially consistent 3D scene generation while supporting diverse conditioning configurations. Moreover, since our approach directly generates a 3D scene representation, it can be decoded to images and optional point maps along arbitrary camera trajectories without requiring per-trajectory diffusion sampling pass, which is common in 2D-based approaches.", "date": "2026-04-13", "categories": []} {"id": "2604.08760", "title": "SIC3D: Style Image Conditioned Text-to-3D Gaussian Splatting Generation", "abstract": "Recent progress in text-to-3D object generation enables the synthesis of detailed geometry from text input by leveraging 2D diffusion models and differentiable 3D representations. However, the approaches often suffer from limited controllability and texture ambiguity due to the limitation of the text modality. To address this, we present SIC3D, a controllable image-conditioned text-to-3D generation pipeline with 3D Gaussian Splatting (3DGS). There are two stages in SIC3D. The first stage generates the 3D object content from text with a text-to-3DGS generation model. The second stage transfers style from a reference image to the 3DGS. Within this stylization stage, we introduce a novel Variational Stylized Score Distillation (VSSD) loss to effectively capture both global and local texture patterns while mitigating conflicts between geometry and appearance. A scaling regularization is further applied to prevent the emergence of artifacts and preserve the pattern from the style image. Extensive experiments demonstrate that SIC3D enhances geometric fidelity and style adherence, outperforming prior approaches in both qualitative and quantitative evaluations.", "date": "2026-04-09", "categories": []} {"id": "2604.09231", "title": "Hitem3D 2.0: Multi-View Guided Native 3D Texture Generation", "abstract": "Although recent advances have improved the quality of 3D texture generation, existing methods still struggle with incomplete texture coverage, cross-view inconsistency, and misalignment between geometry and texture. To address these limitations, we propose Hitem3D 2.0, a multi-view guided native 3D texture generation framework that enhances texture quality through the integration of 2D multi-view generation priors and native 3D texture representations. Hitem3D 2.0 comprises two key components: a multi-view synthesis framework and a native 3D texture generation model. The multi-view generation is built upon a pre-trained image editing backbone and incorporates plug-and-play modules that explicitly promote geometric alignment, cross-view consistency, and illumination uniformity, thereby enabling the synthesis of high-fidelity multi-view images. Conditioned on the generated views and 3D geometry, the native 3D texture generation model projects multi-view textures onto 3D surfaces while plausibly completing textures in unseen regions. Through the integration of multi-view consistency constraints with native 3D texture modeling, Hitem3D 2.0 significantly improves texture completeness, cross-view coherence, and geometric alignment. Experimental results demonstrate that Hitem3D 2.0 outperforms existing methods in terms of texture detail, fidelity, consistency, coherence, and alignment.", "date": "2026-04-10", "categories": []} {"id": "2603.11633", "title": "MV-SAM3D: Adaptive Multi-View Fusion for Layout-Aware 3D Generation", "abstract": "Recent unified 3D generation models have made remarkable progress in producing high-quality 3D assets from a single image. Notably, layout-aware approaches such as SAM3D can reconstruct multiple objects while preserving their spatial arrangement, opening the door to practical scene-level 3D generation. However, current methods are limited to single-view input and cannot leverage complementary multi-view observations, while independently estimated object poses often lead to physically implausible layouts such as interpenetration and floating artifacts.\n We present MV-SAM3D, a training-free framework that extends layout-aware 3D generation with multi-view consistency and physical plausibility. We formulate multi-view fusion as a Multi-Diffusion process in 3D latent space and propose two adaptive weighting strategies -- attention-entropy weighting and visibility weighting -- that enable confidence-aware fusion, ensuring each viewpoint contributes according to its local observation reliability. For multi-object composition, we introduce physics-aware optimization that injects collision and contact constraints both during and after generation, yielding physically plausible object arrangements. Experiments on standard benchmarks and real-world multi-object scenes demonstrate significant improvements in reconstruction fidelity and layout plausibility, all without any additional training. Code is available at https://github.com/devinli123/MV-SAM3D.", "date": "2026-03-12", "categories": []} {"id": "2604.07273", "title": "GenLCA: 3D Diffusion for Full-Body Avatars from In-the-Wild Videos", "abstract": "We present GenLCA, a diffusion-based generative model for generating and editing photorealistic full-body avatars from text and image inputs. The generated avatars are faithful to the inputs, while supporting high-fidelity facial and full-body animations. The core idea is a novel paradigm that enables training a full-body 3D diffusion model from partially observable 2D data, allowing the training dataset to scale to millions of real-world videos. This scalability contributes to the superior photorealism and generalizability of GenLCA. Specifically, we scale up the dataset by repurposing a pretrained feed-forward avatar reconstruction model as an animatable 3D tokenizer, which encodes unstructured video frames into structured 3D tokens. However, most real-world videos only provide partial observations of body parts, resulting in excessive blurring or transparency artifacts in the 3D tokens. To address this, we propose a novel visibility-aware diffusion training strategy that replaces invalid regions with learnable tokens and computes losses only over valid regions. We then train a flow-based diffusion model on the token dataset, inherently maintaining the photorealism and animatability provided by the pretrained avatar reconstruction model. Our approach effectively enables the use of large-scale real-world video data to train a diffusion model natively in 3D. We demonstrate the efficacy of our method through diverse and high-fidelity generation and editing results, outperforming existing solutions by a large margin. The project page is available at https://onethousandwu.com/GenLCA-Page.", "date": "2026-04-08", "categories": []} {"id": "2604.07795", "title": "Image-Guided Geometric Stylization of 3D Meshes", "abstract": "Recent generative models can create visually plausible 3D representations of objects. However, the generation process often allows for implicit control signals, such as contextual descriptions, and rarely supports bold geometric distortions beyond existing data distributions. We propose a geometric stylization framework that deforms a 3D mesh, allowing it to express the style of an image. While style is inherently ambiguous, we utilize pre-trained diffusion models to extract an abstract representation of the provided image. Our coarse-to-fine stylization pipeline can drastically deform the input 3D model to express a diverse range of geometric variations while retaining the valid topology of the original mesh and part-level semantics. We also propose an approximate VAE encoder that provides efficient and reliable gradients from mesh renderings. Extensive experiments demonstrate that our method can create stylized 3D meshes that reflect unique geometric features of the pictured assets, such as expressive poses and silhouettes, thereby supporting the creation of distinctive artistic 3D creations. Project page: https://changwoonchoi.github.io/GeoStyle", "date": "2026-04-09", "categories": []} {"id": "2504.03468", "title": "D-Garment: Physically Grounded Latent Diffusion for Dynamic Garment Deformations", "abstract": "We present a method to dynamically deform 3D garments, in the form of a 3D polygon mesh, based on body shape, motion, and physical cloth material properties. Considering physical cloth properties allows to learn a physically grounded model, with the advantage of being more accurate in terms of physically inspired metrics such as strain or curvature. Existing work studies pose-dependent garment modeling to generate garment deformations from example data, and possibly data-driven dynamic cloth simulation to generate realistic garments in motion. We propose D-Garment, a learning-based approach trained on new data generated with a physics-based simulator. Compared to prior work, our 3D generative model learns garment deformations conditioned by physical material properties, which allows to model loose cloth geometry, especially for large deformations and dynamic wrinkles driven by body motion. Furthermore, the model can be efficiently fitted to observations captured using vision sensors such as 3D point clouds. We leverage the capability of diffusion models to learn flexible and powerful generative priors by modeling the 3D garment in a 2D parameter space independently from the mesh resolution. This representation allows to learn a template-specific latent diffusion model. This allows to condition global and local geometry with body and cloth material information. We quantitatively and qualitatively evaluate D-Garment on both simulations and data captured with a multi-view acquisition platform. Compared to recent baselines, our method is more realistic and accurate in terms of shape similarity and physical validity metrics. Code and data are available for research purposes at https://dumoulina.github.io/d-garment/", "date": "2025-04-04", "categories": []} {"id": "2512.15599", "title": "FlexAvatar: Learning Complete 3D Head Avatars with Partial Supervision", "abstract": "We introduce FlexAvatar, a method for creating high-quality and complete 3D head avatars from a single image. A core challenge lies in the limited availability of multi-view data and the tendency of monocular training to yield incomplete 3D head reconstructions. We identify the root cause of this issue as the entanglement between driving signal and target viewpoint when learning from monocular videos. To address this, we propose a transformer-based 3D portrait animation model with learnable data source tokens, so-called bias sinks, which enables unified training across monocular and multi-view datasets. This design leverages the strengths of both data sources during inference: strong generalization from monocular data and full 3D completeness from multi-view supervision. Furthermore, our training procedure yields a smooth latent avatar space that facilitates identity interpolation and flexible fitting to an arbitrary number of input observations. In extensive evaluations on single-view, few-shot, and monocular avatar creation tasks, we verify the efficacy of FlexAvatar. Many existing methods struggle with view extrapolation while FlexAvatar generates complete 3D head avatars with realistic facial animations. Website: https://tobias-kirschstein.github.io/flexavatar/", "date": "2025-12-17", "categories": []} {"id": "2604.01972", "title": "SDesc3D: Towards Layout-Aware 3D Indoor Scene Generation from Short Descriptions", "abstract": "3D indoor scene generation conditioned on short textual descriptions provides a promising avenue for interactive 3D environment construction without the need for labor-intensive layout specification. Despite recent progress in text-conditioned 3D scene generation, existing works suffer from poor physical plausibility and insufficient detail richness in such semantic condensation cases, largely due to their reliance on explicit semantic cues about compositional objects and their spatial relationships. This limitation highlights the need for enhanced 3D reasoning capabilities, particularly in terms of prior integration and spatial anchoring. Motivated by this, we propose SDesc3D, a short-text conditioned 3D indoor scene generation framework, that leverages multi-view structural priors and regional functionality implications to enable 3D layout reasoning under sparse textual guidance. Specifically, we introduce a Multi-view scene prior augmentation that enriches underspecified textual inputs with aggregated multi-view structural knowledge, shifting from inaccessible semantic relation cues to multi-view relational prior aggregation. Building on this, we design a Functionality-aware layout grounding, employing regional functionality grounding for implicit spatial anchors and conducting hierarchical layout reasoning to enhance scene organization and semantic plausibility. Furthermore, an Iterative reflection-rectification scheme is employed for progressive structural plausibility refinement via self-rectification. Extensive experiments show that our method outperforms existing approaches on short-text conditioned 3D indoor scene generation. Code will be publicly available.", "date": "2026-04-02", "categories": []} {"id": "2604.05070", "title": "Part-Level 3D Gaussian Vehicle Generation with Joint and Hinge Axis Estimation", "abstract": "Simulation is essential for autonomous driving, yet current frameworks often model vehicles as rigid assets and fail to capture part-level articulation. With perception algorithms increasingly leveraging dynamics such as wheel steering or door opening, realistic simulation requires animatable vehicle representations. Existing CAD-based pipelines are limited by library coverage and fixed templates, preventing faithful reconstruction of in-the-wild instances.\n We propose a generative framework that, from a single image or sparse multi-view input, synthesizes an animatable 3D Gaussian vehicle. Our method addresses two challenges: (i) large 3D asset generators are optimized for static quality but not articulation, leading to distortions at part boundaries when animated; and (ii) segmentation alone cannot provide the kinematic parameters required for motion. To overcome this, we introduce a part-edge refinement module that enforces exclusive Gaussian ownership and a kinematic reasoning head that predicts joint positions and hinge axes of movable parts. Together, these components enable faithful part-aware simulation, bridging the gap between static generation and animatable vehicle models.", "date": "2026-04-06", "categories": []} {"id": "2604.05721", "title": "GaussianGrow: Geometry-aware Gaussian Growing from 3D Point Clouds with Text Guidance", "abstract": "3D Gaussian Splatting has demonstrated superior performance in rendering efficiency and quality, yet the generation of 3D Gaussians still remains a challenge without proper geometric priors. Existing methods have explored predicting point maps as geometric references for inferring Gaussian primitives, while the unreliable estimated geometries may lead to poor generations. In this work, we introduce GaussianGrow, a novel approach that generates 3D Gaussians by learning to grow them from easily accessible 3D point clouds, naturally enforcing geometric accuracy in Gaussian generation. Specifically, we design a text-guided Gaussian growing scheme that leverages a multi-view diffusion model to synthesize consistent appearances from input point clouds for supervision. To mitigate artifacts caused by fusing neighboring views, we constrain novel views generated at non-preset camera poses identified in overlapping regions across different views. For completing the hard-to-observe regions, we propose to iteratively detect the camera pose by observing the largest un-grown regions in point clouds and inpainting them by inpainting the rendered view with a pretrained 2D diffusion model. The process continues until complete Gaussians are generated. We extensively evaluate GaussianGrow on text-guided Gaussian generation from synthetic and even real-scanned point clouds. Project Page: https://weiqi-zhang.github.io/GaussianGrow", "date": "2026-04-07", "categories": []} {"id": "2604.06113", "title": "SEM-ROVER: Semantic Voxel-Guided Diffusion for Large-Scale Driving Scene Generation", "abstract": "Scalable generation of outdoor driving scenes requires 3D representations that remain consistent across multiple viewpoints and scale to large areas. Existing solutions either rely on image or video generative models distilled to 3D space, harming the geometric coherence and restricting the rendering to training views, or are limited to small-scale 3D scene or object-centric generation. In this work, we propose a 3D generative framework based on $\\Sigma$-Voxfield grid, a discrete representation where each occupied voxel stores a fixed number of colorized surface samples. To generate this representation, we train a semantic-conditioned diffusion model that operates on local voxel neighborhoods and uses 3D positional encodings to capture spatial structure. We scale to large scenes via progressive spatial outpainting over overlapping regions. Finally, we render the generated $\\Sigma$-Voxfield grid with a deferred rendering module to obtain photorealistic images, enabling large-scale multiview-consistent 3D scene generation without per-scene optimization. Extensive experiments show that our approach can generate diverse large-scale urban outdoor scenes, renderable into photorealistic images with various sensor configurations and camera trajectories while maintaining moderate computation cost compared to existing approaches.", "date": "2026-04-07", "categories": []} {"id": "2509.07435", "title": "DreamLifting: A Plug-in Module Lifting MV Diffusion Models for 3D Asset Generation", "abstract": "The labor- and experience-intensive creation of 3D assets with physically based rendering (PBR) materials demands an autonomous 3D asset creation pipeline. However, most existing 3D generation methods focus on geometry modeling, either baking textures into simple vertex colors or leaving texture synthesis to post-processing with image diffusion models. To achieve end-to-end PBR-ready 3D asset generation, we present Lightweight Gaussian Asset Adapter (LGAA), a novel framework that unifies the modeling of geometry and PBR materials by exploiting multi-view (MV) diffusion priors from a novel perspective. The LGAA features a modular design with three components. Specifically, the LGAA Wrapper reuses and adapts network layers from MV diffusion models, which encapsulate knowledge acquired from billions of images, enabling better convergence in a data-efficient manner. To incorporate multiple diffusion priors for geometry and PBR synthesis, the LGAA Switcher aligns multiple LGAA Wrapper layers encapsulating different knowledge. Then, a tamed variational autoencoder (VAE), termed LGAA Decoder, is designed to predict 2D Gaussian Splatting (2DGS) with PBR channels. Finally, we introduce a dedicated post-processing procedure to effectively extract high-quality, relightable mesh assets from the resulting 2DGS. Extensive quantitative and qualitative experiments demonstrate the superior performance of LGAA with both text- and image-conditioned MV diffusion models. Additionally, the modular design enables flexible incorporation of multiple diffusion priors, and the knowledge-preserving scheme effectively preseves the 2D priors learned on massive image dataset, which leads to data efficient finetuning to lift the MV diffuison models for 3D generation with merely 69k multi-view instances.", "date": "2025-09-09", "categories": []} {"id": "2604.03315", "title": "StoryBlender: Inter-Shot Consistent and Editable 3D Storyboard with Spatial-temporal Dynamics", "abstract": "Storyboarding is a core skill in visual storytelling for film, animation, and games. However, automating this process requires a system to achieve two properties that current approaches rarely satisfy simultaneously: inter-shot consistency and explicit editability. While 2D diffusion-based generators produce vivid imagery, they often suffer from identity drift along with limited geometric control; conversely, traditional 3D animation workflows are consistent and editable but require expert-heavy, labor-intensive authoring. We present StoryBlender, a grounded 3D storyboard generation framework governed by a Story-centric Reflection Scheme. At its core, we propose the StoryBlender system, which is built on a three-stage pipeline: (1) Semantic-Spatial Grounding, to construct a continuity memory graph to decouple global assets from shot-specific variables for long-horizon consistency; (2) Canonical Asset Materialization, to instantiate entities in a unified coordinate space to maintain visual identity; and (3) Spatial-Temporal Dynamics, to achieve layout design and cinematic evolution through visual metrics. By orchestrating multiple agents in a hierarchical manner within a verification loop, StoryBlender iteratively self-corrects spatial hallucinations via engine-verified feedback. The resulting native 3D scenes support direct, precise editing of cameras and visual assets while preserving unwavering multi-shot continuity. Experiments demonstrate that StoryBlender significantly improves consistency and editability over both diffusion-based and 3D-grounded baselines. Code, data, and demonstration video will be available on https://engineeringai-lab.github.io/StoryBlender/", "date": "2026-04-01", "categories": []} {"id": "2604.04406", "title": "3D-Fixer: Coarse-to-Fine In-place Completion for 3D Scenes from a Single Image", "abstract": "Compositional 3D scene generation from a single view requires the simultaneous recovery of scene layout and 3D assets. Existing approaches mainly fall into two categories: feed-forward generation methods and per-instance generation methods. The former directly predict 3D assets with explicit 6DoF poses through efficient network inference, but they generalize poorly to complex scenes. The latter improve generalization through a divide-and-conquer strategy, but suffer from time-consuming pose optimization. To bridge this gap, we introduce 3D-Fixer, a novel in-place completion paradigm. Specifically, 3D-Fixer extends 3D object generative priors to generate complete 3D assets conditioned on the partially visible point cloud at the original locations, which are cropped from the fragmented geometry obtained from the geometry estimation methods. Unlike prior works that require explicit pose alignment, 3D-Fixer uses fragmented geometry as a spatial anchor to preserve layout fidelity. At its core, we propose a coarse-to-fine generation scheme to resolve boundary ambiguity under occlusion, supported by a dual-branch conditioning network and an Occlusion-Robust Feature Alignment (ORFA) strategy for stable training. Furthermore, to address the data scarcity bottleneck, we present ARSG-110K, the largest scene-level dataset to date, comprising over 110K diverse scenes and 3M annotated images with high-fidelity 3D ground truth. Extensive experiments show that 3D-Fixer achieves state-of-the-art geometric accuracy, which significantly outperforms baselines such as MIDI and Gen3DSR, while maintaining the efficiency of the diffusion process. Code and data will be publicly available at https://zx-yin.github.io/3dfixer.", "date": "2026-04-06", "categories": []} {"id": "2604.04425", "title": "HandDreamer: Zero-Shot Text to 3D Hand Model Generation using Corrective Hand Shape Guidance", "abstract": "The emergence of virtual reality has necessitated the generation of detailed and customizable 3D hand models for interaction in the virtual world. However, the current methods for 3D hand model generation are both expensive and cumbersome, offering very little customizability to the users. While recent advancements in zero-shot text-to-3D synthesis have enabled the generation of diverse and customizable 3D models using Score Distillation Sampling (SDS), they do not generalize very well to 3D hand model generation, resulting in unnatural hand structures, view-inconsistencies and loss of details. To address these limitations, we introduce HandDreamer, the first method for zero-shot 3D hand model generation from text prompts. Our findings suggest that view-inconsistencies in SDS is primarily caused due to the ambiguity in the probability landscape described by the text prompt, resulting in similar views converging to different modes of the distribution. This is particularly aggravated for hands due to the large variations in articulations and poses. To alleviate this, we propose to use MANO hand model based initialization and a hand skeleton guided diffusion process to provide a strong prior for the hand structure and to ensure view and pose consistency. Further, we propose a novel corrective hand shape guidance loss to ensure that all the views of the 3D hand model converges to view-consistent modes, without leading to geometric distortions. Extensive evaluations demonstrate the superiority of our method over the state-of-the-art methods, paving a new way forward in 3D hand model generation.", "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": "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": "2604.02799", "title": "UNICA: A Unified Neural Framework for Controllable 3D Avatars", "abstract": "Controllable 3D human avatars have found widespread applications in 3D games, the metaverse, and AR/VR scenarios. The conventional approach to creating such a 3D avatar requires a lengthy, intricate pipeline encompassing appearance modeling, motion planning, rigging, and physical simulation. In this paper, we introduce UNICA (UNIfied neural Controllable Avatar), a skeleton-free generative model that unifies all avatar control components into a single neural framework. Given keyboard inputs akin to video game controls, UNICA generates the next frame of a 3D avatar's geometry through an action-conditioned diffusion model operating on 2D position maps. A point transformer then maps the resulting geometry to 3D Gaussian Splatting for high-fidelity free-view rendering. Our approach naturally captures hair and loose clothing dynamics without manually designed physical simulation, and supports extra-long autoregressive generation. To the best of our knowledge, UNICA is the first model to unify the workflow of \"motion planning, rigging, physical simulation, and rendering\". Code is released at https://github.com/zjh21/UNICA.", "date": "2026-04-03", "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": "2604.02847", "title": "HiDiGen: Hierarchical Diffusion for B-Rep Generation with Explicit Topological Constraints", "abstract": "Boundary representation (B-rep) is the standard 3D modeling format in CAD systems, encoding both geometric primitives and topological connectivity. Despite its prevalence, deep generative modeling of valid B-rep structures remains challenging due to the intricate interplay between discrete topology and continuous geometry. In this paper, we propose HiDiGen, a hierarchical generation framework that decouples geometry modeling into two stages, each guided by explicitly modeled topological constraints. Specifically, our approach first establishes face-edge incidence relations to define a coherent topological scaffold, upon which face proxies and initial edge curves are generated. Subsequently, multiple Transformer-based diffusion modules are employed to refine the geometry by generating precise face surfaces and vertex positions, with edge-vertex adjacencies dynamically established and enforced to preserve structural consistency. This progressive geometry hierarchy enables the generation of more novel and diverse shapes, while two-stage topological modeling ensures high validity. Experimental results show that HiDiGen achieves strong performance, generating novel, diverse, and topologically sound CAD models.", "date": "2026-04-03", "categories": []} {"id": "2604.01641", "title": "LivingWorld: Interactive 4D World Generation with Environmental Dynamics", "abstract": "We introduce LivingWorld, an interactive framework for generating 4D worlds with environmental dynamics from a single image. While recent advances in 3D scene generation enable large-scale environment creation, most approaches focus primarily on reconstructing static geometry, leaving scene-scale environmental dynamics such as clouds, water, or smoke largely unexplored. Modeling such dynamics is challenging because motion must remain coherent across an expanding scene while supporting low-latency user feedback. LivingWorld addresses this challenge by progressively constructing a globally coherent motion field as the scene expands. To maintain global consistency during expansion, we introduce a geometry-aware alignment module that resolves directional and scale ambiguities across views. We further represent motion using a compact hash-based motion field, enabling efficient querying and stable propagation of dynamics throughout the scene. This representation also supports bidirectional motion propagation during rendering, producing long and temporally coherent 4D sequences without relying on expensive video-based refinement. On a single RTX 5090 GPU, generating each new scene expansion step requires 9 seconds, followed by 3 seconds for motion alignment and motion field updates, enabling interactive 4D world generation with globally coherent environmental dynamics. Video demonstrations are available at cvsp-lab.github.io/LivingWorld.", "date": "2026-04-02", "categories": []} {"id": "2604.02289", "title": "Omni123: Exploring 3D Native Foundation Models with Limited 3D Data by Unifying Text to 2D and 3D Generation", "abstract": "Recent multimodal large language models have achieved strong performance in unified text and image understanding and generation, yet extending such native capability to 3D remains challenging due to limited data. Compared to abundant 2D imagery, high-quality 3D assets are scarce, making 3D synthesis under-constrained. Existing methods often rely on indirect pipelines that edit in 2D and lift results into 3D via optimization, sacrificing geometric consistency. We present Omni123, a 3D-native foundation model that unifies text-to-2D and text-to-3D generation within a single autoregressive framework. Our key insight is that cross-modal consistency between images and 3D can serve as an implicit structural constraint. By representing text, images, and 3D as discrete tokens in a shared sequence space, the model leverages abundant 2D data as a geometric prior to improve 3D representations. We introduce an interleaved X-to-X training paradigm that coordinates diverse cross-modal tasks over heterogeneous paired datasets without requiring fully aligned text-image-3D triplets. By traversing semantic-visual-geometric cycles (e.g., text to image to 3D to image) within autoregressive sequences, the model jointly enforces semantic alignment, appearance fidelity, and multi-view geometric consistency. Experiments show that Omni123 significantly improves text-guided 3D generation and editing, demonstrating a scalable path toward multimodal 3D world models.", "date": "2026-04-02", "categories": []} {"id": "2509.16483", "title": "Octree Diffusion for Semantic Scene Generation and Completion", "abstract": "The completion, extension, and generation of 3D semantic scenes are an interrelated set of capabilities that are useful for robotic navigation and exploration. Existing approaches seek to decouple these problems and solve them one-off. Additionally, these approaches are often domain-specific, requiring separate models for different data distributions, e.g.\\ indoor vs.\\ outdoor scenes.\n To unify these techniques and provide cross-domain compatibility, we develop a single framework that can perform scene completion, extension, and generation in both indoor and outdoor scenes, which we term Octree Latent Semantic Diffusion. Our approach operates directly on an efficient dual octree graph latent representation: a hierarchical, sparse, and memory-efficient occupancy structure. This technique disentangles synthesis into two stages: (i) structure diffusion, which predicts binary split signals to construct a coarse occupancy octree, and (ii) latent semantic diffusion, which generates semantic embeddings decoded by a graph VAE into voxel-level semantic labels. To perform semantic scene completion or extension, our model leverages inference-time latent inpainting, or outpainting respectively. These inference-time methods use partial LiDAR scans or maps to condition generation, without the need for retraining or finetuning.\n We demonstrate high-quality structure, coherent semantics, and robust completion from single LiDAR scans, as well as zero-shot generalization to out-of-distribution LiDAR data. These results indicate that completion-through-generation in a dual octree graph latent space is a practical and scalable alternative to regression-based pipelines for real-world robotic perception tasks.", "date": "2025-09-20", "categories": []} {"id": "2601.16148", "title": "ActionMesh: Animated 3D Mesh Generation with Temporal 3D Diffusion", "abstract": "Generating animated 3D objects is at the heart of many applications, yet most advanced works are typically difficult to apply in practice because of their limited setup, their long runtime, or their limited quality. We introduce ActionMesh, a generative model that predicts production-ready 3D meshes \"in action\" in a feed-forward manner. Drawing inspiration from early video models, our key insight is to modify existing 3D diffusion models to include a temporal axis, resulting in a framework we dubbed \"temporal 3D diffusion\". Specifically, we first adapt the 3D diffusion stage to generate a sequence of synchronized latents representing time-varying and independent 3D shapes. Second, we design a temporal 3D autoencoder that translates a sequence of independent shapes into the corresponding deformations of a pre-defined reference shape, allowing us to build an animation. Combining these two components, ActionMesh generates animated 3D meshes from different inputs like a monocular video, a text description, or even a 3D mesh with a text prompt describing its animation. Besides, compared to previous approaches, our method is fast and produces results that are rig-free and topology consistent, hence enabling rapid iteration and seamless applications like texturing and retargeting. We evaluate our model on standard video-to-4D benchmarks (Consistent4D, Objaverse) and report state-of-the-art performances on both geometric accuracy and temporal consistency, demonstrating that our model can deliver animated 3D meshes with unprecedented speed and quality.", "date": "2026-01-22", "categories": []} {"id": "2603.09266", "title": "ForgeDreamer: Industrial Text-to-3D Generation with Multi-Expert LoRA and Cross-View Hypergraph", "abstract": "Current text-to-3D generation methods excel in natural scenes but struggle with industrial applications due to two critical limitations: domain adaptation challenges where conventional LoRA fusion causes knowledge interference across categories, and geometric reasoning deficiencies where pairwise consistency constraints fail to capture higher-order structural dependencies essential for precision manufacturing. We propose a novel framework named ForgeDreamer addressing both challenges through two key innovations. First, we introduce a Multi-Expert LoRA Ensemble mechanism that consolidates multiple category-specific LoRA models into a unified representation, achieving superior cross-category generalization while eliminating knowledge interference. Second, building on enhanced semantic understanding, we develop a Cross-View Hypergraph Geometric Enhancement approach that captures structural dependencies spanning multiple viewpoints simultaneously. These components work synergistically improved semantic understanding, enables more effective geometric reasoning, while hypergraph modeling ensures manufacturing-level consistency. Extensive experiments on a custom industrial dataset demonstrate superior semantic generalization and enhanced geometric fidelity compared to state-of-the-art approaches. Code is available at https://github.com/Junhaocai27/ForgeDreamer", "date": "2026-03-10", "categories": []} {"id": "2604.00494", "title": "ARGS: Auto-Regressive Gaussian Splatting via Parallel Progressive Next-Scale Prediction", "abstract": "Auto-regressive frameworks for next-scale prediction of 2D images have demonstrated strong potential for producing diverse and sophisticated content by progressively refining a coarse input. However, extending this paradigm to 3D object generation remains largely unexplored. In this paper, we introduce auto-regressive Gaussian splatting (ARGS), a framework for making next-scale predictions in parallel for generation according to levels of detail. We propose a Gaussian simplification strategy and reverse the simplification to guide next-scale generation. Benefiting from the use of hierarchical trees, the generation process requires only \\(\\mathcal{O}(\\log n)\\) steps, where \\(n\\) is the number of points. Furthermore, we propose a tree-based transformer to predict the tree structure auto-regressively, allowing leaf nodes to attend to their internal ancestors to enhance structural consistency. Extensive experiments demonstrate that our approach effectively generates multi-scale Gaussian representations with controllable levels of detail, visual fidelity, and a manageable time consumption budget.", "date": "2026-04-01", "categories": []} {"id": "2503.12052", "title": "A Text-to-3D Framework for Joint Generation of CG-Ready Humans and Compatible Garments", "abstract": "Creating detailed 3D human avatars with fitted garments traditionally requires specialized expertise and labor-intensive workflows. While recent advances in generative AI have enabled text-to-3D human and clothing synthesis, existing methods fall short in offering accessible, integrated pipelines for generating CG-ready 3D avatars with physically compatible outfits; here we use the term CG-ready for models following a technical aesthetic common in computer graphics (CG) and adopt standard CG polygonal meshes and strands representations (rather than neural representations like NeRF and 3DGS) that can be directly integrated into conventional CG pipelines and support downstream tasks such as physical simulation. To bridge this gap, we introduce Tailor, an integrated text-to-3D framework that generates high-fidelity, customizable 3D avatars dressed in simulation-ready garments. Tailor consists of three stages. (1) Seman tic Parsing: we employ a large language model to interpret textual descriptions and translate them into parameterized human avatars and semantically matched garment templates. (2) Geometry-Aware Garment Generation: we propose topology-preserving deformation with novel geometric losses to generate body-aligned garments under text control. (3) Consistent Texture Synthesis: we propose a novel multi-view diffusion process optimized for garment texturing, which enforces view consistency, preserves photorealistic details, and optionally supports symmetric texture generation common in garments. Through comprehensive quantitative and qualitative evaluations, we demonstrate that Tailor outperforms state-of-the-art methods in fidelity, usability, and diversity. Our code will be released for academic use. Project page: https://human-tailor.github.io", "date": "2025-03-15", "categories": []} {"id": "2603.01142", "title": "ArtLLM: Generating Articulated Assets via 3D LLM", "abstract": "Creating interactive digital environments for gaming, robotics, and simulation relies on articulated 3D objects whose functionality emerges from their part geometry and kinematic structure. However, existing approaches remain fundamentally limited: optimization-based reconstruction methods require slow, per-object joint fitting and typically handle only simple, single-joint objects, while retrieval-based methods assemble parts from a fixed library, leading to repetitive geometry and poor generalization. To address these challenges, we introduce ArtLLM, a novel framework for generating high-quality articulated assets directly from complete 3D meshes. At its core is a 3D multimodal large language model trained on a large-scale articulation dataset curated from both existing articulation datasets and procedurally generated objects. Unlike prior work, ArtLLM autoregressively predicts a variable number of parts and joints, inferring their kinematic structure in a unified manner from the object's point cloud. This articulation-aware layout then conditions a 3D generative model to synthesize high-fidelity part geometries. Experiments on the PartNet-Mobility dataset show that ArtLLM significantly outperforms state-of-the-art methods in both part layout accuracy and joint prediction, while generalizing robustly to real-world objects. Finally, we demonstrate its utility in constructing digital twins, highlighting its potential for scalable robot learning.", "date": "2026-03-01", "categories": []} {"id": "2603.18782", "title": "Points-to-3D: Structure-Aware 3D Generation with Point Cloud Priors", "abstract": "Recent progress in 3D generation has been driven largely by models conditioned on images or text, while readily available 3D priors are still underused. In many real-world scenarios, the visible-region point cloud are easy to obtain from active sensors such as LiDAR or from feed-forward predictors like VGGT, offering explicit geometric constraints that current methods fail to exploit. In this work, we introduce Points-to-3D, a diffusion-based framework that leverages point cloud priors for geometry-controllable 3D asset and scene generation. Built on a latent 3D diffusion model TRELLIS, Points-to-3D first replaces pure-noise sparse structure latent initialization with a point cloud priors tailored input formulation.A structure inpainting network, trained within the TRELLIS framework on task-specific data designed to learn global structural inpainting, is then used for inference with a staged sampling strategy (structural inpainting followed by boundary refinement), completing the global geometry while preserving the visible regions of the input priors. In practice, Points-to-3D can take either accurate point-cloud priors or VGGT-estimated point clouds from single images as input. Experiments on both objects and scene scenarios consistently demonstrate superior performance over state-of-the-art baselines in terms of rendering quality and geometric fidelity, highlighting the effectiveness of explicitly embedding point-cloud priors for achieving more accurate and structurally controllable 3D generation. Project page: https://jiatongxia.github.io/points2-3D/", "date": "2026-03-19", "categories": []} {"id": "2603.29089", "title": "WorldFlow3D: Flowing Through 3D Distributions for Unbounded World Generation", "abstract": "Unbounded 3D world generation is emerging as a foundational task for scene modeling in computer vision, graphics, and robotics. In this work, we present WorldFlow3D, a novel method capable of generating unbounded 3D worlds. Building upon a foundational property of flow matching - namely, defining a path of transport between two data distributions - we model 3D generation more generally as a problem of flowing through 3D data distributions, not limited to conditional denoising. We find that our latent-free flow approach generates causal and accurate 3D structure, and can use this as an intermediate distribution to guide the generation of more complex structure and high-quality texture - all while converging more rapidly than existing methods. We enable controllability over generated scenes with vectorized scene layout conditions for geometric structure control and visual texture control through scene attributes. We confirm the effectiveness of WorldFlow3D on both real outdoor driving scenes and synthetic indoor scenes, validating cross-domain generalizability and high-quality generation on real data distributions. We confirm favorable scene generation fidelity over approaches in all tested settings for unbounded scene generation. For more, see https://light.princeton.edu/worldflow3d.", "date": "2026-03-31", "categories": []} {"id": "2603.29387", "title": "Extend3D: Town-Scale 3D Generation", "abstract": "In this paper, we propose Extend3D, a training-free pipeline for 3D scene generation from a single image, built upon an object-centric 3D generative model. To overcome the limitations of fixed-size latent spaces in object-centric models for representing wide scenes, we extend the latent space in the $x$ and $y$ directions. Then, by dividing the extended latent space into overlapping patches, we apply the object-centric 3D generative model to each patch and couple them at each time step. Since patch-wise 3D generation with image conditioning requires strict spatial alignment between image and latent patches, we initialize the scene using a point cloud prior from a monocular depth estimator and iteratively refine occluded regions through SDEdit. We discovered that treating the incompleteness of 3D structure as noise during 3D refinement enables 3D completion via a concept, which we term under-noising. Furthermore, to address the sub-optimality of object-centric models for sub-scene generation, we optimize the extended latent during denoising, ensuring that the denoising trajectories remain consistent with the sub-scene dynamics. To this end, we introduce 3D-aware optimization objectives for improved geometric structure and texture fidelity. We demonstrate that our method yields better results than prior methods, as evidenced by human preference and quantitative experiments.", "date": "2026-03-31", "categories": []} {"id": "2603.29572", "title": "Turbo4DGen: Ultra-Fast Acceleration for 4D Generation", "abstract": "4D generation, or dynamic 3D content generation, integrates spatial, temporal, and view dimensions to model realistic dynamic scenes, playing a foundational role in advancing world models and physical AI. However, maintaining long-chain consistency across both frames and viewpoints through the unique spatio-camera-motion (SCM) attention mechanism introduces substantial computational and memory overhead, often leading to out-of-memory (OOM) failures and prohibitive generation times. To address these challenges, we propose Turbo4DGen, an ultra-fast acceleration framework for diffusion-based multi-view 4D content generation. Turbo4DGen introduces a spatiotemporal cache mechanism that persistently reuses intermediate attention across denoising steps, combined with dynamically semantic-aware attention pruning and an adaptive SCM chain bypass scheduler, to drastically reduce redundant SCM attention computation. Our experimental results show that Turbo4DGen achieves an average 9.7$\\times$ speedup without quality degradation on the ObjaverseDy and Consistent4D datasets. To the best of our knowledge, Turbo4DGen is the first dedicated acceleration framework for 4D generation.", "date": "2026-01-24", "categories": []} {"id": "2603.29592", "title": "Bioinspired123D: Generative 3D Modeling System for Bioinspired Structures", "abstract": "Generative AI has made rapid progress in text, image, and video synthesis, yet text-to-3D modeling for scientific design remains particularly challenging due to limited controllability and high computational cost. Most existing 3D generative methods rely on meshes, voxels, or point clouds which can be costly to train and difficult to control. We introduce Bioinspired123D, a lightweight and modular code-as-geometry pipeline that generates fabricable 3D structures directly through parametric programs rather than dense visual representations. At the core of Bioinspired123D is Bioinspired3D, a compact language model finetuned to translate natural language design cues into Blender Python scripts encoding smooth, biologically inspired geometries. We curate a domain-specific dataset of over 4,000 bioinspired and geometric design scripts spanning helical, cellular, and tubular motifs with parametric variability. The dataset is expanded and validated through an automated LLM-driven, Blender-based quality control pipeline. Bioinspired3D is then embedded in a graph-based agentic framework that integrates multimodal retrieval-augmented generation and a vision-language model critic to iteratively evaluate, critique, and repair generated scripts. We evaluate performance on a new benchmark for 3D geometry script generation and show that Bioinspired123D demonstrates a near fourfold improvement over its non-finetuned base model, while also outperforming substantially larger state-of-the-art language models despite using far fewer parameters and compute. By prioritizing code-as-geometry representations, Bioinspired123D enables compute-efficient, controllable, and interpretable text-to-3D generation, lowering barriers to AI driven scientific discovery in materials and structural design.", "date": "2026-02-11", "categories": []} {"id": "2509.13688", "title": "CraftMesh: High-Fidelity Generative Mesh Manipulation via Poisson Seamless Fusion", "abstract": "Controllable, high-fidelity mesh editing remains a significant challenge in 3D content creation. Existing generative methods often struggle with complex geometries and fail to produce detailed results. We propose CraftMesh, a novel framework for high-fidelity generative mesh manipulation via Poisson Seamless Fusion. Our key insight is to decompose mesh editing into a pipeline that leverages the strengths of 2D and 3D generative models: we edit a 2D reference image, then generate a region-specific 3D mesh, and seamlessly fuse it into the original model. We introduce two core techniques: Poisson Geometric Fusion, which utilizes a hybrid SDF/Mesh representation with normal blending to achieve harmonious geometric integration, and Poisson Texture Harmonization for visually consistent texture blending. Experimental results demonstrate that CraftMesh outperforms state-of-the-art methods, delivering superior global consistency and local detail in complex editing tasks.", "date": "2025-09-17", "categories": []} {"id": "2601.16771", "title": "AutoRegressive Generation with B-rep Holistic Token Sequence Representation", "abstract": "Previous representation and generation approaches for the B-rep relied on graph-based representations that disentangle geometric and topological features through decoupled computational pipelines, thereby precluding the application of sequence-based generative frameworks, such as transformer architectures that have demonstrated remarkable performance. In this paper, we propose BrepARG, the first attempt to encode B-rep's geometry and topology into a holistic token sequence representation, enabling sequence-based B-rep generation with an autoregressive architecture. Specifically, BrepARG encodes B-rep into 3 types of tokens: geometry and position tokens representing geometric features, and face index tokens representing topology. Then the holistic token sequence is constructed hierarchically, starting with constructing the geometry blocks (i.e., faces and edges) using the above tokens, followed by geometry block sequencing. Finally, we assemble the holistic sequence representation for the entire B-rep. We also construct a transformer-based autoregressive model that learns the distribution over holistic token sequences via next-token prediction, using a multi-layer decoder-only architecture with causal masking. Experiments demonstrate that BrepARG achieves state-of-the-art (SOTA) performance. BrepARG validates the feasibility of representing B-rep as holistic token sequences, opening new directions for B-rep generation.", "date": "2026-01-23", "categories": []} {"id": "2603.28126", "title": "SVGS: Single-View to 3D Object Editing via Gaussian Splatting", "abstract": "Text-driven 3D scene editing has attracted considerable interest due to its convenience and user-friendliness. However, methods that rely on implicit 3D representations, such as Neural Radiance Fields (NeRF), while effective in rendering complex scenes, are hindered by slow processing speeds and limited control over specific regions of the scene. Moreover, existing approaches, including Instruct-NeRF2NeRF and GaussianEditor, which utilize multi-view editing strategies, frequently produce inconsistent results across different views when executing text instructions. This inconsistency can adversely affect the overall performance of the model, complicating the task of balancing the consistency of editing results with editing efficiency. To address these challenges, we propose a novel method termed Single-View to 3D Object Editing via Gaussian Splatting (SVGS), which is a single-view text-driven editing technique based on 3D Gaussian Splatting (3DGS). Specifically, in response to text instructions, we introduce a single-view editing strategy grounded in multi-view diffusion models, which reconstructs 3D scenes by leveraging only those views that yield consistent editing results. Additionally, we employ sparse 3D Gaussian Splatting as the 3D representation, which significantly enhances editing efficiency. We conducted a comparative analysis of SVGS against existing baseline methods across various scene settings, and the results indicate that SVGS outperforms its counterparts in both editing capability and processing speed, representing a significant advancement in 3D editing technology. For further details, please visit our project page at: https://amateurc.github.io/svgs.github.io.", "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": "2512.09435", "title": "UniPart: Part-Level 3D Generation with Unified 3D Geom-Seg Latents", "abstract": "Part-level 3D generation is essential for applications requiring decomposable and structured 3D synthesis. However, existing methods either rely on implicit part segmentation with limited granularity control or depend on strong external segmenters trained on large annotated datasets. In this work, we observe that part awareness emerges naturally during whole-object geometry learning and propose Geom-Seg VecSet, a unified geometry-segmentation latent representation that jointly encodes object geometry and part-level structure. Building on this representation, we introduce UniPart, a two-stage latent diffusion framework for image-guided part-level 3D generation. The first stage performs joint geometry generation and latent part segmentation, while the second stage conditions part-level diffusion on both whole-object and part-specific latents. A dual-space generation scheme further enhances geometric fidelity by predicting part latents in both global and canonical spaces. Extensive experiments demonstrate that UniPart achieves superior segmentation controllability and part-level geometric quality compared with existing approaches.", "date": "2025-12-10", "categories": []} {"id": "2603.26661", "title": "GaussianGPT: Towards Autoregressive 3D Gaussian Scene Generation", "abstract": "Most recent advances in 3D generative modeling rely on diffusion or flow-matching formulations. We instead explore a fully autoregressive alternative and introduce GaussianGPT, a transformer-based model that directly generates 3D Gaussians via next-token prediction, thus facilitating full 3D scene generation. We first compress Gaussian primitives into a discrete latent grid using a sparse 3D convolutional autoencoder with vector quantization. The resulting tokens are serialized and modeled using a causal transformer with 3D rotary positional embedding, enabling sequential generation of spatial structure and appearance. Unlike diffusion-based methods that refine scenes holistically, our formulation constructs scenes step-by-step, naturally supporting completion, outpainting, controllable sampling via temperature, and flexible generation horizons. This formulation leverages the compositional inductive biases and scalability of autoregressive modeling while operating on explicit representations compatible with modern neural rendering pipelines, positioning autoregressive transformers as a complementary paradigm for controllable and context-aware 3D generation.", "date": "2026-03-27", "categories": []} {"id": "2603.24278", "title": "TopoMesh: High-Fidelity Mesh Autoencoding via Topological Unification", "abstract": "The dominant paradigm for high-fidelity 3D generation relies on a VAE-Diffusion pipeline, where the VAE's reconstruction capability sets a firm upper bound on generation quality. A fundamental challenge limiting existing VAEs is the representation mismatch between ground-truth meshes and network predictions: GT meshes have arbitrary, variable topology, while VAEs typically predict fixed-structure implicit fields (\\eg, SDF on regular grids). This inherent misalignment prevents establishing explicit mesh-level correspondences, forcing prior work to rely on indirect supervision signals such as SDF or rendering losses. Consequently, fine geometric details, particularly sharp features, are poorly preserved during reconstruction. To address this, we introduce TopoMesh, a sparse voxel-based VAE that unifies both GT and predicted meshes under a shared Dual Marching Cubes (DMC) topological framework. Specifically, we convert arbitrary input meshes into DMC-compliant representations via a remeshing algorithm that preserves sharp edges using an L$\\infty$ distance metric. Our decoder outputs meshes in the same DMC format, ensuring that both predicted and target meshes share identical topological structures. This establishes explicit correspondences at the vertex and face level, allowing us to derive explicit mesh-level supervision signals for topology, vertex positions, and face orientations with clear gradients. Our sparse VAE architecture employs this unified framework and is trained with Teacher Forcing and progressive resolution training for stable and efficient convergence. Extensive experiments demonstrate that TopoMesh significantly outperforms existing VAEs in reconstruction fidelity, achieving superior preservation of sharp features and geometric details.", "date": "2026-03-25", "categories": []} {"id": "2504.05296", "title": "Let it Snow! Animating 3D Gaussian Scenes with Dynamic Weather Effects via Physics-Guided Score Distillation", "abstract": "3D Gaussian Splatting has recently enabled fast and photorealistic reconstruction of static 3D scenes. However, dynamic editing of such scenes remains a significant challenge. We introduce a novel framework, Physics-Guided Score Distillation, to address a fundamental conflict: physics simulation provides a strong motion prior that is insufficient for photorealism , while video-based Score Distillation Sampling (SDS) alone cannot generate coherent motion for complex, multi-particle scenarios. We resolve this through a unified optimization framework where physics simulation guides Score Distillation to jointly refine the motion prior for photorealism while simultaneously optimizing appearance. Specifically, we learn a neural dynamics model that predicts particle motion and appearance, optimized end-to-end via a combined loss integrating Video-SDS for photorealism with our physics-guidance prior. This allows for photorealistic refinements while ensuring the dynamics remain plausible. Our framework enables scene-wide dynamic weather effects, including snowfall, rainfall, fog, and sandstorms, with physically plausible motion. Experiments demonstrate our physics-guided approach significantly outperforms baselines, with ablations confirming this joint refinement is essential for generating coherent, high-fidelity dynamics.", "date": "2025-04-07", "categories": []} {"id": "2511.22533", "title": "Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration", "abstract": "Diffusion models have achieved impressive generative quality across modalities like 2D images, videos, and 3D shapes, but their inference remains computationally expensive due to the iterative denoising process. While recent caching-based methods effectively reuse redundant computations to speed up 2D and video generation, directly applying these techniques to 3D diffusion models can severely disrupt geometric consistency. In 3D synthesis, even minor numerical errors in cached latent features accumulate, causing structural artifacts and topological inconsistencies. To overcome this limitation, we propose Fast3Dcache, a training-free geometry-aware caching framework that accelerates 3D diffusion inference while preserving geometric fidelity. Our method introduces a Predictive Caching Scheduler Constraint (PCSC) to dynamically determine cache quotas according to voxel stabilization patterns and a Spatiotemporal Stability Criterion (SSC) to select stable features for reuse based on velocity magnitude and acceleration criterion. Comprehensive experiments show that Fast3Dcache accelerates inference significantly, achieving up to a 27.12% speed-up and a 54.83% reduction in FLOPs, with minimal degradation in geometric quality as measured by Chamfer Distance (2.48%) and F-Score (1.95%).", "date": "2025-11-27", "categories": []} {"id": "2512.06438", "title": "AGORA: Adversarial Generation Of Real-time Animatable 3D Gaussian Head Avatars", "abstract": "The generation of high-fidelity, animatable 3D human avatars remains a core challenge in computer graphics and vision, with applications in VR, telepresence, and entertainment. Existing approaches based on implicit representations like NeRFs suffer from slow rendering and dynamic inconsistencies, while 3D Gaussian Splatting (3DGS) methods are typically limited to static head generation, lacking dynamic control. We bridge this gap by introducing AGORA, a novel framework that extends 3DGS within a generative adversarial network to produce animatable avatars. Our formulation combines spatial shape conditioning with a dual-discriminator training strategy that supervises both rendered appearance and synthetic geometry cues, improving expression fidelity and controllability. To enable practical deployment, we further introduce a simple inference-time approach that extracts Gaussian blendshapes and reuses them for animation on-device. AGORA generates avatars that are visually realistic, precisely controllable, and achieves state-of-the-art performance among animatable generative head-avatar methods. Quantitatively, we render at 560 FPS on a single GPU and 60 FPS on mobile phones, marking a significant step toward practical, high-performance digital humans. Project website: https://ramazan793.github.io/AGORA/", "date": "2025-12-06", "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": "2603.22972", "title": "WorldMesh: Generating Navigable Multi-Room 3D Scenes via Mesh-Conditioned Image Diffusion", "abstract": "Recent progress in image and video synthesis has inspired their use in advancing 3D scene generation. However, we observe that text-to-image and -video approaches struggle to maintain scene- and object-level consistency beyond a limited environment scale due to the absence of explicit geometry. We thus present a geometry-first approach that decouples this complex problem of large-scale 3D scene synthesis into its structural composition, represented as a mesh scaffold, and realistic appearance synthesis, which leverages powerful image synthesis models conditioned on the mesh scaffold. From an input text description, we first construct a mesh capturing the environment's geometry (walls, floors, etc.), and then use image synthesis, segmentation and object reconstruction to populate the mesh structure with objects in realistic layouts. This mesh scaffold is then rendered to condition image synthesis, providing a structural backbone for consistent appearance generation. This enables scalable, arbitrarily-sized 3D scenes of high object richness and diversity, combining robust 3D consistency with photorealistic detail. We believe this marks a significant step toward generating truly environment-scale, immersive 3D worlds.", "date": "2026-03-24", "categories": []} {"id": "2512.02982", "title": "U4D: Uncertainty-Aware 4D World Modeling from LiDAR Sequences", "abstract": "Modeling dynamic 3D environments from LiDAR sequences is central to building reliable 4D worlds for autonomous driving and embodied AI. Existing generative frameworks, however, often treat all spatial regions uniformly, overlooking the varying uncertainty across real-world scenes. This uniform generation leads to artifacts in complex or ambiguous regions, limiting realism and temporal stability. In this work, we present U4D, an uncertainty-aware framework for 4D LiDAR world modeling. Our approach first estimates spatial uncertainty maps from a pretrained segmentation model to localize semantically challenging regions. It then performs generation in a \"hard-to-easy\" manner through two sequential stages: (1) uncertainty-region modeling, which reconstructs high-entropy regions with fine geometric fidelity, and (2) uncertainty-conditioned completion, which synthesizes the remaining areas under learned structural priors. To further ensure temporal coherence, U4D incorporates a mixture of spatio-temporal (MoST) block that adaptively fuses spatial and temporal representations during diffusion. Extensive experiments show that U4D produces geometrically faithful and temporally consistent LiDAR sequences, advancing the reliability of 4D world modeling for autonomous perception and simulation.", "date": "2025-12-02", "categories": []} {"id": "2603.22782", "title": "Know3D: Prompting 3D Generation with Knowledge from Vision-Language Models", "abstract": "Recent advances in 3D generation have improved the fidelity and geometric details of synthesized 3D assets. However, due to the inherent ambiguity of single-view observations and the lack of robust global structural priors caused by limited 3D training data, the unseen regions generated by existing models are often stochastic and difficult to control, which may sometimes fail to align with user intentions or produce implausible geometries. In this paper, we propose Know3D, a novel framework that incorporates rich knowledge from multimodal large language models into 3D generative processes via latent hidden-state injection, enabling language-controllable generation of the back-view for 3D assets. We utilize a VLM-diffusion-based model, where the VLM is responsible for semantic understanding and guidance. The diffusion model acts as a bridge that transfers semantic knowledge from the VLM to the 3D generation model. In this way, we successfully bridge the gap between abstract textual instructions and the geometric reconstruction of unobserved regions, transforming the traditionally stochastic back-view hallucination into a semantically controllable process, demonstrating a promising direction for future 3D generation models.", "date": "2026-03-24", "categories": []} {"id": "2603.23246", "title": "GO-Renderer: Generative Object Rendering with 3D-aware Controllable Video Diffusion Models", "abstract": "Reconstructing a renderable 3D model from images is a useful but challenging task. Recent feedforward 3D reconstruction methods have demonstrated remarkable success in efficiently recovering geometry, but still cannot accurately model the complex appearances of these 3D reconstructed models. Recent diffusion-based generative models can synthesize realistic images or videos of an object using reference images without explicitly modeling its appearance, which provides a promising direction for object rendering, but lacks accurate control over the viewpoints. In this paper, we propose GO-Renderer, a unified framework integrating the reconstructed 3D proxies to guide the video generative models to achieve high-quality object rendering on arbitrary viewpoints under arbitrary lighting conditions. Our method not only enjoys the accurate viewpoint control using the reconstructed 3D proxy but also enables high-quality rendering in different lighting environments using diffusion generative models without explicitly modeling complex materials and lighting. Extensive experiments demonstrate that GO-Renderer achieves state-of-the-art performance across the object rendering tasks, including synthesizing images on new viewpoints, rendering the objects in a novel lighting environment, and inserting an object into an existing video.", "date": "2026-03-24", "categories": []} {"id": "2603.23386", "title": "SIMART: Decomposing Monolithic Meshes into Sim-ready Articulated Assets via MLLM", "abstract": "High-quality articulated 3D assets are indispensable for embodied AI and physical simulation, yet 3D generation still focuses on static meshes, leaving a gap in \"sim-ready\" interactive objects. Most recent articulated object creation methods rely on multi-stage pipelines that accumulate errors across decoupled modules. Alternatively, unified MLLMs offer a single-stage path to joint static asset understanding and sim-ready asset generation. However dense voxel-based 3D tokenization yields long 3D token sequences and high memory overhead, limiting scalability to complex articulated objects. To address this, we propose SIMART, a unified MLLM framework that jointly performs part-level decomposition and kinematic prediction. By introducing a Sparse 3D VQ-VAE, SIMART reduces token counts by 70% vs. dense voxel tokens, enabling high-fidelity multi-part assemblies. SIMART achieves state-of-the-art performance on PartNet-Mobility and in-the-wild AIGC datasets, and enables physics-based robotic simulation.", "date": "2026-03-24", "categories": []} {"id": "2412.01583", "title": "3DSceneEditor: Controllable 3D Scene Editing with Gaussian Splatting", "abstract": "The creation of 3D scenes has traditionally been both labor-intensive and costly, requiring designers to meticulously configure 3D assets and environments. Recent advancements in generative AI, including text-to-3D and image-to-3D methods, have dramatically reduced the complexity and cost of this process. However, current techniques for editing complex 3D scenes continue to rely on generally interactive multi-step, 2D-to-3D projection methods and diffusion-based techniques, which often lack precision in control and hamper interactive-rate performance. In this work, we propose ***3DSceneEditor***, a fully 3D-based paradigm for interactive-rate, precise editing of intricate 3D scenes using Gaussian Splatting. Unlike conventional methods, 3DSceneEditor operates through a streamlined 3D pipeline, enabling direct Gaussian-based manipulation for efficient, high-quality edits based on input prompts. The proposed framework (i) integrates a pre-trained instance segmentation model for semantic labeling; (ii) employs a zero-shot grounding approach with CLIP to align target objects with user prompts; and (iii) applies scene modifications, such as object addition, repositioning, recoloring, replacing, and removal--directly on Gaussians. Extensive experimental results show that 3DSceneEditor surpasses existing state-of-the-art techniques in terms of both editing precision and efficiency, establishing a new benchmark for efficient and interactive 3D scene customization.", "date": "2024-12-02", "categories": []} {"id": "2506.02459", "title": "ReSpace: Text-Driven Autoregressive 3D Indoor Scene Synthesis and Editing", "abstract": "Scene synthesis and editing has emerged as a promising direction in computer graphics. Current trained approaches for 3D indoor scene generation either oversimplify object semantics through one-hot class encodings (e.g., 'chair' or 'table'), require masked diffusion for editing, ignore room boundaries, or rely on floor plan renderings that fail to capture complex layouts. LLM-based methods enable richer semantics via natural language, but lack editing functionality, are limited to rectangular layouts, or rely on weak spatial reasoning from implicit world models. We introduce ReSpace, a generative framework for autoregressive text-driven 3D indoor scene synthesis and editing. Our approach features a compact structured scene representation with explicit room boundaries that enables asset-agnostic deployment and frames scene manipulation as a next-token prediction task, supporting object addition, removal, and swapping via natural language. We employ supervised fine-tuning with a preference alignment stage to train a specialized language model for object addition that accounts for user instructions, spatial geometry, object semantics, and scene-level composition. We further introduce a voxelization-based evaluation metric capturing fine-grained geometric violations beyond 3D bounding boxes. Experiments surpass state-of-the-art on object addition and achieve superior human-perceived quality on the application of full scene synthesis, despite not being trained on it.", "date": "2025-06-03", "categories": []} {"id": "2506.21076", "title": "PoseMaster: A Unified 3D Native Framework for Stylized Pose Generation", "abstract": "Pose stylization, which aims to synthesize stylized content aligning with target poses, serves as a fundamental task across 2D, 3D, and video domains. In the 3D realm, prevailing approaches typically rely on a cascade pipeline: first manipulating the image pose via 2D foundation models and subsequently lifting it into 3D representations. However, this paradigm limits the precision and diversity of the 3d pose stylization. To this end, we propose a novel paradigm for 3D pose stylization that unifies pose stylization and 3D generation within a cohesive framework. This integration minimizes the risk of cumulative errors and enhances the model's efficiency and effectiveness. In addition, diverging from previous works that typically utilize 2D skeleton images as guidance, we directly utilize the 3D skeleton because it can provide a more accurate representation of 3D spatial and topological relationships, which significantly enhances the model's capacity to achieve richer and more precise pose stylization. Moreover, we develop a scalable data engine to construct a large-scale dataset of ''Image-Skeleton-Mesh'' triplets, enabling the model to jointly learn identity preservation and geometric alignment. Extensive experiments demonstrate that PoseMaster significantly outperforms state-of-the-art methods in both qualitative and quantitative metrics. Owing to the strict spatial alignment between the generated 3D meshes and the conditioning skeletons, PoseMaster enables the direct creation of animatable assets when coupled with automated skinning models, highlighting its compelling potential for automated character rigging.", "date": "2025-06-26", "categories": []} {"id": "2509.15130", "title": "Taming Video Models for 3D and 4D Generation via Zero-Shot Camera Control", "abstract": "Video diffusion models have rich world priors, but their use in spatial tasks is limited by poor control, spatial-temporal inconsistent results, and entangled scene-camera dynamics. Current approaches, such as per-task fine-tuning or post-process warping, often introduce visual artifacts, fail to generalize, or incur high computational costs. We introduce WorldForge, a novel, training-free framework that operates purely at inference time to resolve these issues. Our method comprises three synergistic components. First, an intra-step refinement loop injects fine-grained motion guidance during the denoising process, iteratively correcting the output to ensure strict adherence to the target camera path. Second, an optical flow-based analysis identifies and isolates motion-related channels within the latent space. This allows our framework to selectively apply guidance, thereby decoupling motion from appearance and preserving visual fidelity. Third, a dual-path guidance strategy adaptively corrects for drift by comparing the guided generation against an unguided, reference denoising path, effectively neutralizing artifacts caused by misaligned structural inputs. Together, these components inject precise, trajectory-aligned control without model retraining, achieving accurate motion guidance and photorealistic synthesis. As a plug-and-play, model-agnostic solution, WorldForge demonstrates highly versatile generalizability. Beyond robust zero-shot 3D/4D generation, it readily empowers over a dozen diverse downstream applications, seamlessly enabling tasks like video editing, stabilization, and virtual try-on. Extensive experiments confirm state-of-the-art performance in trajectory adherence and perceptual quality, outperforming both training-dependent and inference-only baselines.", "date": "2025-09-18", "categories": []} {"id": "2512.00422", "title": "PhysGen: Physically Grounded 3D Shape Generation for Industrial Design", "abstract": "Existing generative models for 3D shapes can synthesize high-fidelity and visually plausible shapes. For certain classes of shapes that have undergone an engineering design process, the realism of the shape is tightly coupled with the underlying physical properties, e.g., aerodynamic efficiency for automobiles. Since existing methods lack knowledge of such physics, they are unable to use this knowledge to enhance the realism of shape generation. Motivated by this, we propose a unified physics-based 3D shape generation pipeline, with a focus on industrial design applications. Specifically, we introduce a new flow matching model with explicit physical guidance, consisting of an alternating update process. We iteratively perform a velocity-based update and a physics-based refinement, progressively adjusting the latent code to align with the desired 3D shapes and physical properties. We further strengthen physical validity by incorporating a physics-aware regularization term into the velocity-based update step. To support such physics-guided updates, we build a shape-and-physics variational autoencoder (SP-VAE) that jointly encodes shape and physics information into a unified latent space. The experiments on three benchmarks show that this synergistic formulation improves shape realism beyond mere visual plausibility. Our code and model weights are available at https://github.com/kasvii/PhysGen.", "date": "2025-11-29", "categories": []} {"id": "2601.04090", "title": "Gen3R: 3D Scene Generation Meets Feed-Forward Reconstruction", "abstract": "We present Gen3R, a method that bridges the strong priors of foundational reconstruction models and video diffusion models for scene-level 3D generation. We repurpose the VGGT reconstruction model to produce geometric latents by training an adapter on its tokens, which are regularized to align with the appearance latents of pre-trained video diffusion models. By jointly generating these disentangled yet aligned latents, Gen3R produces both RGB videos and corresponding 3D geometry, including camera poses, depth maps, and global point clouds. Experiments demonstrate that our approach achieves state-of-the-art results in single- and multi-image conditioned 3D scene generation. Additionally, our method can enhance the robustness of reconstruction by leveraging generative priors, demonstrating the mutual benefit of tightly coupling reconstruction and generative models.", "date": "2026-01-07", "categories": []} {"id": "2602.21499", "title": "Easy3E: Feed-Forward 3D Asset Editing via Rectified Voxel Flow", "abstract": "Existing 3D editing methods rely on computationally intensive scene-by-scene iterative optimization and suffer from multi-view inconsistency. We propose an effective and feed-forward 3D editing framework based on the TRELLIS generative backbone, capable of modifying 3D models from a single editing view. Our framework addresses two key issues: adapting training-free 2D editing to structured 3D representations, and overcoming the bottleneck of appearance fidelity in compressed 3D features. To ensure geometric consistency, we introduce Voxel FlowEdit, an edit-driven flow in the sparse voxel latent space that achieves globally consistent 3D deformation in a single pass. To restore high-fidelity details, we develop a normal-guided single to multi-view generation module as an external appearance prior, successfully recovering high-frequency textures. Experiments demonstrate that our method enables fast, globally consistent, and high-fidelity 3D model editing.", "date": "2026-02-25", "categories": []} {"id": "2603.21295", "title": "Text-Image Conditioned 3D Generation", "abstract": "High-quality 3D assets are essential for VR/AR, industrial design, and entertainment, motivating growing interest in generative models that create 3D content from user prompts. Most existing 3D generators, however, rely on a single conditioning modality: image-conditioned models achieve high visual fidelity by exploiting pixel-aligned cues but suffer from viewpoint bias when the input view is limited or ambiguous, while text-conditioned models provide broad semantic guidance yet lack low-level visual detail. This limits how users can express intent and raises a natural question: can these two modalities be combined for more flexible and faithful 3D generation? Our diagnostic study shows that even simple late fusion of text- and image-conditioned predictions outperforms single-modality models, revealing strong cross-modal complementarity. We therefore formalize Text-Image Conditioned 3D Generation, which requires joint reasoning over a visual exemplar and a textual specification. To address this task, we introduce TIGON, a minimalist dual-branch baseline with separate image- and text-conditioned backbones and lightweight cross-modal fusion. Extensive experiments show that text-image conditioning consistently improves over single-modality methods, highlighting complementary vision-language guidance as a promising direction for future 3D generation research. Project page: https://jumpat.github.io/tigon-page", "date": "2026-03-22", "categories": []} {"id": "2603.21557", "title": "From Part to Whole: 3D Generative World Model with an Adaptive Structural Hierarchy", "abstract": "Single-image 3D generation lies at the core of vision-to-graphics models in the real world. However, it remains a fundamental challenge to achieve reliable generalization across diverse semantic categories and highly variable structural complexity under sparse supervision. Existing approaches typically model objects in a monolithic manner or rely on a fixed number of parts, including recent part-aware models such as PartCrafter, which still require a labor-intensive user-specified part count. Such designs easily lead to overfitting, fragmented or missing structural components, and limited compositional generalization when encountering novel object layouts. To this end, this paper rethinks single-image 3D generation as learning an adaptive part-whole hierarchy in the flexible 3D latent space. We present a novel part-to-whole 3D generative world model that autonomously discovers latent structural slots by inferring soft and compositional masks directly from image tokens. Specifically, an adaptive slot-gating mechanism dynamically determines the slot-wise activation probabilities and smoothly consolidates redundant slots within different objects, ensuring that the emergent structure remains compact yet expressive across categories. Each distilled slot is then aligned to a learnable, class-agnostic prototype bank, enabling powerful cross-category shape sharing and denoising through universal geometric prototypes in the real world. Furthermore, a lightweight 3D denoiser is introduced to reconstruct geometry and appearance via unified diffusion objectives. Experiments show consistent gains in cross-category transfer and part-count extrapolation, and ablations confirm complementary benefits of the prototype bank for shape-prior sharing as well as slot-gating for structural adaptation.", "date": "2026-03-23", "categories": []} {"id": "2603.21978", "title": "GeoFusion-CAD: Structure-Aware Diffusion with Geometric State Space for Parametric 3D Design", "abstract": "Parametric Computer-Aided Design (CAD) is fundamental to modern 3D modeling, yet existing methods struggle to generate long command sequences, especially under complex geometric and topological dependencies. Transformer-based architectures dominate CAD sequence generation due to their strong dependency modeling, but their quadratic attention cost and limited context windowing hinder scalability to long programs. We propose GeoFusion-CAD, an end-to-end diffusion framework for scalable and structure-aware generation. Our proposal encodes CAD programs as hierarchical trees, jointly capturing geometry and topology within a state-space diffusion process. Specifically, a lightweight C-Mamba block models long-range structural dependencies through selective state transitions, enabling coherent generation across extended command sequences. To support long-sequence evaluation, we introduce DeepCAD-240, an extended benchmark that increases the sequence length ranging from 40 to 240 while preserving sketch-extrusion semantics from the ABC dataset. Extensive experiments demonstrate that GeoFusion-CAD achieves superior performance on both short and long command ranges, maintaining high geometric fidelity and topological consistency where Transformer-based models degrade. Our approach sets new state-of-the-art scores for long-sequence parametric CAD generation, establishing a scalable foundation for next-generation CAD modeling systems. Code and datasets are available at GitHub.", "date": "2026-03-23", "categories": []} {"id": "2603.22275", "title": "Repurposing Geometric Foundation Models for Multi-view Diffusion", "abstract": "While recent advances in generative latent spaces have driven substantial progress in single-image generation, the optimal latent space for novel view synthesis (NVS) remains largely unexplored. In particular, NVS requires geometrically consistent generation across viewpoints, but existing approaches typically operate in a view-independent VAE latent space. In this paper, we propose Geometric Latent Diffusion (GLD), a framework that repurposes the geometrically consistent feature space of geometric foundation models as the latent space for multi-view diffusion. We show that these features not only support high-fidelity RGB reconstruction but also encode strong cross-view geometric correspondences, providing a well-suited latent space for NVS. Our experiments demonstrate that GLD outperforms both VAE and RAE on 2D image quality and 3D consistency metrics, while accelerating training by more than 4.4x compared to the VAE latent space. Notably, GLD remains competitive with state-of-the-art methods that leverage large-scale text-to-image pretraining, despite training its diffusion model from scratch without such generative pretraining.", "date": "2026-03-23", "categories": []} {"id": "2506.09814", "title": "DreamCS: Geometry-Aware Text-to-3D Generation with Unpaired 3D Reward Supervision", "abstract": "While text-to-3D generation has attracted growing interest, existing methods often struggle to produce 3D assets that align well with human preferences. Current preference alignment techniques for 3D content typically rely on hardly-collected preference-paired multi-view 2D images to train 2D reward models, when then guide 3D generation -- leading to geometric artifacts due to their inherent 2D bias. To address these limitations, we construct 3D-MeshPref, the first large-scale unpaired 3D preference dataset, featuring diverse 3D meshes annotated by a large language model and refined by human evaluators. We then develop RewardCS, the first reward model trained directly on unpaired 3D-MeshPref data using a novel Cauchy-Schwarz divergence objective, enabling effective learning of human-aligned 3D geometric preferences without requiring paired comparisons. Building on this, we propose DreamCS, a unified framework that integrates RewardCS into text-to-3D pipelines -- enhancing both implicit and explicit 3D generation with human preference feedback. Extensive experiments show DreamCS outperforms prior methods, producing 3D assets that are both geometrically faithful and human-preferred. Code and models will be released publicly.", "date": "2025-06-11", "categories": []} {"id": "2508.19752", "title": "Fast 3D Diffusion for Scalable Granular Media Synthesis", "abstract": "Discrete Element Method (DEM) simulations of granular media are computationally intensive, particularly during initialization phases dominated by large displacements and kinetic energy. This paper presents a novel generative pipeline based on 3D diffusion models that directly synthesizes arbitrarily large granular assemblies in mechanically realistic configurations. The approach employs a two-stage pipeline. First, an unconditional diffusion model generates independent 3D voxel grids representing granular media; second, a 3D inpainting model, adapted from 2D techniques using masked inputs and repainting strategies, seamlessly stitches these grids together. The inpainting model uses the outputs of the unconditional diffusion model to learn from the context of adjacent generations and creates new regions that blend smoothly into the context region. Both models are trained on binarized 3D occupancy grids derived from a database of small-scale DEM simulations, scaling linearly with the number of output voxels. Simulations that spanned over days can now run in hours, practically enabling simulations containing more than 200k ballast particles. The pipeline remains fully compatible with existing DEM workflows as it post-processes the diffusion generated voxel grids into DEM compatible particle meshes. Being mechanically consistent on key granulometry metrics with the original DEM simulations, the pipeline is also compatible with many other applications in the field of granular media, with capability of generating both convex and non-convex particles. Showcased on two examples (railway ballast and lunar regolith), the pipeline reimagines the way initialization of granular media simulations is performed, enabling scales of generation previously unattainable with traditional DEM simulations.", "date": "2025-08-27", "categories": []} {"id": "2603.19121", "title": "CustomTex: High-fidelity Indoor Scene Texturing via Multi-Reference Customization", "abstract": "The creation of high-fidelity, customizable 3D indoor scene textures remains a significant challenge. While text-driven methods offer flexibility, they lack the precision for fine-grained, instance-level control, and often produce textures with insufficient quality, artifacts, and baked-in shading. To overcome these limitations, we introduce CustomTex, a novel framework for instance-level, high-fidelity scene texturing driven by reference images. CustomTex takes an untextured 3D scene and a set of reference images specifying the desired appearance for each object instance, and generates a unified, high-resolution texture map. The core of our method is a dual-distillation approach that separates semantic control from pixel-level enhancement. We employ semantic-level distillation, equipped with an instance cross-attention, to ensure semantic plausibility and ``reference-instance'' alignment, and pixel-level distillation to enforce high visual fidelity. Both are unified within a Variational Score Distillation (VSD) optimization framework. Experiments demonstrate that CustomTex achieves precise instance-level consistency with reference images and produces textures with superior sharpness, reduced artifacts, and minimal baked-in shading compared to state-of-the-art methods. Our work establishes a more direct and user-friendly path to high-quality, customizable 3D scene appearance editing.", "date": "2026-03-19", "categories": []} {"id": "2603.19598", "title": "FlowScene: Style-Consistent Indoor Scene Generation with Multimodal Graph Rectified Flow", "abstract": "Scene generation has extensive industrial applications, demanding both high realism and precise control over geometry and appearance. Language-driven retrieval methods compose plausible scenes from a large object database, but overlook object-level control and often fail to enforce scene-level style coherence. Graph-based formulations offer higher controllability over objects and inform holistic consistency by explicitly modeling relations, yet existing methods struggle to produce high-fidelity textured results, thereby limiting their practical utility. We present FlowScene, a tri-branch scene generative model conditioned on multimodal graphs that collaboratively generates scene layouts, object shapes, and object textures. At its core lies a tight-coupled rectified flow model that exchanges object information during generation, enabling collaborative reasoning across the graph. This enables fine-grained control of objects' shapes, textures, and relations while enforcing scene-level style coherence across structure and appearance. Extensive experiments show that FlowScene outperforms both language-conditioned and graph-conditioned baselines in terms of generation realism, style consistency, and alignment with human preferences.", "date": "2026-03-20", "categories": []} {"id": "2603.19708", "title": "WorldAgents: Can Foundation Image Models be Agents for 3D World Models?", "abstract": "Given the remarkable ability of 2D foundation image models to generate high-fidelity outputs, we investigate a fundamental question: do 2D foundation image models inherently possess 3D world model capabilities? To answer this, we systematically evaluate multiple state-of-the-art image generation models and Vision-Language Models (VLMs) on the task of 3D world synthesis. To harness and benchmark their potential implicit 3D capability, we propose an agentic framing to facilitate 3D world generation. Our approach employs a multi-agent architecture: a VLM-based director that formulates prompts to guide image synthesis, a generator that synthesizes new image views, and a VLM-backed two-step verifier that evaluates and selectively curates generated frames from both 2D image and 3D reconstruction space. Crucially, we demonstrate that our agentic approach provides coherent and robust 3D reconstruction, producing output scenes that can be explored by rendering novel views. Through extensive experiments across various foundation models, we demonstrate that 2D models do indeed encapsulate a grasp of 3D worlds. By exploiting this understanding, our method successfully synthesizes expansive, realistic, and 3D-consistent worlds.", "date": "2026-03-20", "categories": []} {"id": "2509.22592", "title": "OT-MeanFlow3D: Bridging Optimal Transport and Meanflow for Efficient 3D Point Cloud Generation", "abstract": "Flow-matching models have recently emerged as a powerful framework for continuous generative modeling, including 3D point cloud synthesis. However, their deployment is limited by the need for multiple sequential sampling steps at inference time. MeanFlow enables single-step generation and significantly accelerates inference, but often struggles to approximate the trajectories of the original multi-step flow, leading to degraded sample quality. In this work, we propose an Optimal Transport-enhanced MeanFlow framework (OT-MF3D) for efficient and accurate 3D point cloud generation and completion. By incorporating optimal transport-based sampling, our method better preserves the geometric and distributional structure of the underlying multi-step flow while retaining single-step inference. Experiments on ShapeNet show improved generation and completion quality compared to recent baselines, while reducing training and inference costs relative to conventional diffusion and flow-based models.", "date": "2025-09-26", "categories": []} {"id": "2603.19053", "title": "SwiftTailor: Efficient 3D Garment Generation with Geometry Image Representation", "abstract": "Realistic and efficient 3D garment generation remains a longstanding challenge in computer vision and digital fashion. Existing methods typically rely on large vision- language models to produce serialized representations of 2D sewing patterns, which are then transformed into simulation-ready 3D meshes using garment modeling framework such as GarmentCode. Although these approaches yield high-quality results, they often suffer from slow inference times, ranging from 30 seconds to a minute. In this work, we introduce SwiftTailor, a novel two-stage framework that unifies sewing-pattern reasoning and geometry-based mesh synthesis through a compact geometry image representation. SwiftTailor comprises two lightweight modules: PatternMaker, an efficient vision-language model that predicts sewing patterns from diverse input modalities, and GarmentSewer, an efficient dense prediction transformer that converts these patterns into a novel Garment Geometry Image, encoding the 3D surface of all garment panels in a unified UV space. The final 3D mesh is reconstructed through an efficient inverse mapping process that incorporates remeshing and dynamic stitching algorithms to directly assemble the garment, thereby amortizing the cost of physical simulation. Extensive experiments on the Multimodal GarmentCodeData demonstrate that SwiftTailor achieves state-of-the-art accuracy and visual fidelity while significantly reducing inference time. This work offers a scalable, interpretable, and high-performance solution for next-generation 3D garment generation.", "date": "2026-03-19", "categories": []} {"id": "2603.19216", "title": "DreamPartGen: Semantically Grounded Part-Level 3D Generation via Collaborative Latent Denoising", "abstract": "Understanding and generating 3D objects as compositions of meaningful parts is fundamental to human perception and reasoning. However, most text-to-3D methods overlook the semantic and functional structure of parts. While recent part-aware approaches introduce decomposition, they remain largely geometry-focused, lacking semantic grounding and failing to model how parts align with textual descriptions or their inter-part relations. We propose DreamPartGen, a framework for semantically grounded, part-aware text-to-3D generation. DreamPartGen introduces Duplex Part Latents (DPLs) that jointly model each part's geometry and appearance, and Relational Semantic Latents (RSLs) that capture inter-part dependencies derived from language. A synchronized co-denoising process enforces mutual geometric and semantic consistency, enabling coherent, interpretable, and text-aligned 3D synthesis. Across multiple benchmarks, DreamPartGen delivers state-of-the-art performance in geometric fidelity and text-shape alignment.", "date": "2026-03-19", "categories": []} {"id": "2509.03887", "title": "OccTENS: 3D Occupancy World Model via Temporal Next-Scale Prediction", "abstract": "In this paper, we propose OccTENS, a generative occupancy world model that enables controllable, high-fidelity long-term occupancy generation while maintaining computational efficiency. Different from visual generation, the occupancy world model must capture the fine-grained 3D geometry and dynamic evolution of the 3D scenes, posing great challenges for the generative models. Recent approaches based on autoregression (AR) have demonstrated the potential to predict vehicle movement and future occupancy scenes simultaneously from historical observations, but they typically suffer from \\textbf{inefficiency}, \\textbf{temporal degradation} in long-term generation and \\textbf{lack of controllability}. To holistically address these issues, we reformulate the occupancy world model as a temporal next-scale prediction (TENS) task, which decomposes the temporal sequence modeling problem into the modeling of spatial scale-by-scale generation and temporal scene-by-scene prediction. With a \\textbf{TensFormer}, OccTENS can effectively manage the temporal causality and spatial relationships of occupancy sequences in a flexible and scalable way. To enhance the pose controllability, we further propose a holistic pose aggregation strategy, which features a unified sequence modeling for occupancy and ego-motion. Experiments show that OccTENS outperforms the state-of-the-art method with both higher occupancy quality and faster inference time.", "date": "2025-09-04", "categories": []} {"id": "2509.23728", "title": "M3DLayout: A Multi-Source Dataset of 3D Indoor Layouts and Structured Descriptions for 3D Generation", "abstract": "In text-driven 3D scene generation, object layout serves as a crucial intermediate representation that bridges high-level language instructions with detailed geometric output. It not only provides a structural blueprint for ensuring physical plausibility but also supports semantic controllability and interactive editing. However, the learning capabilities of current 3D indoor layout generation models are constrained by the limited scale, diversity, and annotation quality of existing datasets. To address this, we introduce M3DLayout, a large-scale, multi-source dataset for 3D indoor layout generation. M3DLayout comprises 21,367 layouts and over 433k object instances, integrating three distinct sources: real-world scans, professional CAD designs, and procedurally generated scenes. Each layout is paired with detailed structured text describing global scene summaries, relational placements of large furniture, and fine-grained arrangements of smaller items. This diverse and richly annotated resource enables models to learn complex spatial and semantic patterns across a wide variety of indoor environments. To assess the potential of M3DLayout, we establish a benchmark using both a text-conditioned diffusion model and a text-conditioned autoregressive model. Experimental results demonstrate that our dataset provides a solid foundation for training layout generation models. Its multi-source composition enhances diversity, notably through the Inf3DLayout subset which provides rich small-object information, enabling the generation of more complex and detailed scenes. We hope that M3DLayout can serve as a valuable resource for advancing research in text-driven 3D scene synthesis. All dataset and code will be made public upon acceptance.", "date": "2025-09-28", "categories": []} {"id": "2510.15869", "title": "Skyfall-GS: Synthesizing Immersive 3D Urban Scenes from Satellite Imagery", "abstract": "Synthesizing large-scale, explorable, and geometrically accurate 3D urban scenes is a challenging yet valuable task for immersive and embodied applications. The challenge lies in the lack of large-scale and high-quality real-world 3D scans for training generalizable generative models. In this paper, we take an alternative route to create large-scale 3D scenes by leveraging readily available satellite imagery for realistic coarse geometry and open-domain diffusion models for high-quality close-up appearance synthesis. We propose Skyfall-GS, a novel hybrid framework that synthesizes immersive city-block scale 3D urban scenes by combining satellite reconstruction with diffusion refinement, eliminating the need for costly 3D annotations, and also featuring real-time, immersive 3D exploration. We tailor a curriculum-driven iterative refinement strategy to progressively enhance geometric completeness and photorealistic texture. Extensive experiments demonstrate that Skyfall-GS provides improved cross-view consistent geometry and more realistic textures compared to state-of-the-art approaches. Project page: https://skyfall-gs.jayinnn.dev/", "date": "2025-10-17", "categories": []} {"id": "2603.17735", "title": "TAPESTRY: From Geometry to Appearance via Consistent Turntable Videos", "abstract": "Automatically generating photorealistic and self-consistent appearances for untextured 3D models is a critical challenge in digital content creation. The advancement of large-scale video generation models offers a natural approach: directly synthesizing 360-degree turntable videos (TTVs), which can serve not only as high-quality dynamic previews but also as an intermediate representation to drive texture synthesis and neural rendering. However, existing general-purpose video diffusion models struggle to maintain strict geometric consistency and appearance stability across the full range of views, making their outputs ill-suited for high-quality 3D reconstruction. To this end, we introduce TAPESTRY, a framework for generating high-fidelity TTVs conditioned on explicit 3D geometry. We reframe the 3D appearance generation task as a geometry-conditioned video diffusion problem: given a 3D mesh, we first render and encode multi-modal geometric features to constrain the video generation process with pixel-level precision, thereby enabling the creation of high-quality and consistent TTVs. Building upon this, we also design a method for downstream reconstruction tasks from the TTV input, featuring a multi-stage pipeline with 3D-Aware Inpainting. By rotating the model and performing a context-aware secondary generation, this pipeline effectively completes self-occluded regions to achieve full surface coverage. The videos generated by TAPESTRY are not only high-quality dynamic previews but also serve as a reliable, 3D-aware intermediate representation that can be seamlessly back-projected into UV textures or used to supervise neural rendering methods like 3DGS. This enables the automated creation of production-ready, complete 3D assets from untextured meshes. Experimental results demonstrate that our method outperforms existing approaches in both video consistency and final reconstruction quality.", "date": "2026-03-18", "categories": []} {"id": "2603.17841", "title": "Omni-3DEdit: Generalized Versatile 3D Editing in One-Pass", "abstract": "Most instruction-driven 3D editing methods rely on 2D models to guide the explicit and iterative optimization of 3D representations. This paradigm, however, suffers from two primary drawbacks. First, it lacks a universal design of different 3D editing tasks because the explicit manipulation of 3D geometry necessitates task-dependent rules, e.g., 3D appearance editing demands inherent source 3D geometry, while 3D removal alters source geometry. Second, the iterative optimization process is highly time-consuming, often requiring thousands of invocations of 2D/3D updating. We present Omni-3DEdit, a unified, learning-based model that generalizes various 3D editing tasks implicitly. One key challenge to achieve our goal is the scarcity of paired source-edited multi-view assets for training. To address this issue, we construct a data pipeline, synthesizing a relatively rich number of high-quality paired multi-view editing samples. Subsequently, we adapt the pre-trained generative model SEVA as our backbone by concatenating source view latents along with conditional tokens in sequence space. A dual-stream LoRA module is proposed to disentangle different view cues, largely enhancing our model's representational learning capability. As a learning-based model, our model is free of the time-consuming online optimization, and it can complete various 3D editing tasks in one forward pass, reducing the inference time from tens of minutes to approximately two minutes. Extensive experiments demonstrate the effectiveness and efficiency of Omni-3DEdit.", "date": "2026-03-18", "categories": []} {"id": "2603.17995", "title": "LoST: Level of Semantics Tokenization for 3D Shapes", "abstract": "Tokenization is a fundamental technique in the generative modeling of various modalities. In particular, it plays a critical role in autoregressive (AR) models, which have recently emerged as a compelling option for 3D generation. However, optimal tokenization of 3D shapes remains an open question. State-of-the-art (SOTA) methods primarily rely on geometric level-of-detail (LoD) hierarchies, originally designed for rendering and compression. These spatial hierarchies are often token-inefficient and lack semantic coherence for AR modeling. We propose Level-of-Semantics Tokenization (LoST), which orders tokens by semantic salience, such that early prefixes decode into complete, plausible shapes that possess principal semantics, while subsequent tokens refine instance-specific geometric and semantic details. To train LoST, we introduce Relational Inter-Distance Alignment (RIDA), a novel 3D semantic alignment loss that aligns the relational structure of the 3D shape latent space with that of the semantic DINO feature space. Experiments show that LoST achieves SOTA reconstruction, surpassing previous LoD-based 3D shape tokenizers by large margins on both geometric and semantic reconstruction metrics. Moreover, LoST achieves efficient, high-quality AR 3D generation and enables downstream tasks like semantic retrieval, while using only 0.1%-10% of the tokens needed by prior AR models.", "date": "2026-03-18", "categories": []} {"id": "2506.23205", "title": "BridgeShape: Latent Diffusion Schr\\\"odinger Bridge for 3D Shape Completion", "abstract": "Existing diffusion-based 3D shape completion methods typically use a conditional paradigm, injecting incomplete shape information into the denoising network via deep feature interactions (e.g., concatenation, cross-attention) to guide sampling toward complete shapes, often represented by voxel-based distance functions. However, these approaches fail to explicitly model the optimal global transport path, leading to suboptimal completions. Moreover, performing diffusion directly in voxel space imposes resolution constraints, limiting the generation of fine-grained geometric details. To address these challenges, we propose BridgeShape, a novel framework for 3D shape completion via latent diffusion Schr\\\"odinger bridge. The key innovations lie in two aspects: (i) BridgeShape formulates shape completion as an optimal transport problem, explicitly modeling the transition between incomplete and complete shapes to ensure a globally coherent transformation. (ii) We introduce a Depth-Enhanced Vector Quantized Variational Autoencoder (VQ-VAE) to encode 3D shapes into a compact latent space, leveraging self-projected multi-view depth information enriched with strong DINOv2 features to enhance geometric structural perception. By operating in a compact yet structurally informative latent space, BridgeShape effectively mitigates resolution constraints and enables more efficient and high-fidelity 3D shape completion. BridgeShape achieves state-of-the-art performance on large-scale 3D shape completion benchmarks, demonstrating superior fidelity at higher resolutions and for unseen object classes.", "date": "2025-06-29", "categories": []} {"id": "2512.04761", "title": "Order Matters: 3D Shape Generation from Sequential VR Sketches", "abstract": "VR sketching lets users explore and iterate on ideas directly in 3D, offering a faster and more intuitive alternative to conventional CAD tools. However, existing sketch-to-shape models ignore the temporal ordering of strokes, discarding crucial cues about structure and design intent. We introduce VRSketch2Shape, the first framework and multi-category dataset for generating 3D shapes from sequential VR sketches. Our contributions are threefold: (i) an automated pipeline that generates sequential VR sketches from arbitrary shapes, (ii) a dataset of over 20k synthetic and 900 hand-drawn sketch-shape pairs across four categories, and (iii) an order-aware sketch encoder coupled with a diffusion-based 3D generator. Our approach yields higher geometric fidelity than prior work, generalizes effectively from synthetic to real sketches with minimal supervision, and performs well even on partial sketches. All data and models will be released open-source at https://chenyizi086.github.io/VRSketch2Shape_website.", "date": "2025-12-04", "categories": []} {"id": "2603.16085", "title": "Interact3D: Compositional 3D Generation of Interactive Objects", "abstract": "Recent breakthroughs in 3D generation have enabled the synthesis of high-fidelity individual assets. However, generating 3D compositional objects from single images--particularly under occlusions--remains challenging. Existing methods often degrade geometric details in hidden regions and fail to preserve the underlying object-object spatial relationships (OOR). We present a novel framework Interact3D designed to generate physically plausible interacting 3D compositional objects. Our approach first leverages advanced generative priors to curate high-quality individual assets with a unified 3D guidance scene. To physically compose these assets, we then introduce a robust two-stage composition pipeline. Based on the 3D guidance scene, the primary object is anchored through precise global-to-local geometric alignment (registration), while subsequent geometries are integrated using a differentiable Signed Distance Field (SDF)-based optimization that explicitly penalizes geometry intersections. To reduce challenging collisions, we further deploy a closed-loop, agentic refinement strategy. A Vision-Language Model (VLM) autonomously analyzes multi-view renderings of the composed scene, formulates targeted corrective prompts, and guides an image editing module to iteratively self-correct the generation pipeline. Extensive experiments demonstrate that Interact3D successfully produces promising collsion-aware compositions with improved geometric fidelity and consistent spatial relationships.", "date": "2026-03-17", "categories": []} {"id": "2603.16099", "title": "OneWorld: Taming Scene Generation with 3D Unified Representation Autoencoder", "abstract": "Existing diffusion-based 3D scene generation methods primarily operate in 2D image/video latent spaces, which makes maintaining cross-view appearance and geometric consistency inherently challenging. To bridge this gap, we present OneWorld, a framework that performs diffusion directly within a coherent 3D representation space. Central to our approach is the 3D Unified Representation Autoencoder (3D-URAE); it leverages pretrained 3D foundation models and augments their geometry-centric nature by injecting appearance and distilling semantics into a unified 3D latent space. Furthermore, we introduce token-level Cross-View-Correspondence (CVC) consistency loss to explicitly enforce structural alignment across views, and propose Manifold-Drift Forcing (MDF) to mitigate train-inference exposure bias and shape a robust 3D manifold by mixing drifted and original representations. Comprehensive experiments demonstrate that OneWorld generates high-quality 3D scenes with superior cross-view consistency compared to state-of-the-art 2D-based methods. Our code will be available at https://github.com/SensenGao/OneWorld.", "date": "2026-03-17", "categories": []} {"id": "2603.16566", "title": "VideoMatGen: PBR Materials through Joint Generative Modeling", "abstract": "We present a method for generating physically-based materials for 3D shapes based on a video diffusion transformer architecture. Our method is conditioned on input geometry and a text description, and jointly models multiple material properties (base color, roughness, metallicity, height map) to form physically plausible materials. We further introduce a custom variational auto-encoder which encodes multiple material modalities into a compact latent space, which enables joint generation of multiple modalities without increasing the number of tokens. Our pipeline generates high-quality materials for 3D shapes given a text prompt, compatible with common content creation tools.", "date": "2026-03-17", "categories": []} {"id": "2603.16871", "title": "WorldCam: Interactive Autoregressive 3D Gaming Worlds with Camera Pose as a Unifying Geometric Representation", "abstract": "Recent advances in video diffusion transformers have enabled interactive gaming world models that allow users to explore generated environments over extended horizons. However, existing approaches struggle with precise action control and long-horizon 3D consistency. Most prior works treat user actions as abstract conditioning signals, overlooking the fundamental geometric coupling between actions and the 3D world, whereby actions induce relative camera motions that accumulate into a global camera pose within a 3D world. In this paper, we establish camera pose as a unifying geometric representation to jointly ground immediate action control and long-term 3D consistency. First, we define a physics-based continuous action space and represent user inputs in the Lie algebra to derive precise 6-DoF camera poses, which are injected into the generative model via a camera embedder to ensure accurate action alignment. Second, we use global camera poses as spatial indices to retrieve relevant past observations, enabling geometrically consistent revisiting of locations during long-horizon navigation. To support this research, we introduce a large-scale dataset comprising 3,000 minutes of authentic human gameplay annotated with camera trajectories and textual descriptions. Extensive experiments show that our approach substantially outperforms state-of-the-art interactive gaming world models in action controllability, long-horizon visual quality, and 3D spatial consistency.", "date": "2026-03-17", "categories": []} {"id": "2503.09160", "title": "WonderVerse: Extendable 3D Scene Generation with Video Generative Models", "abstract": "We introduce \\textit{WonderVerse}, a simple but effective framework for generating extendable 3D scenes. Unlike existing methods that rely on iterative depth estimation and image inpainting, often leading to geometric distortions and inconsistencies, WonderVerse leverages the powerful world-level priors embedded within video generative foundation models to create highly immersive and geometrically coherent 3D environments. Furthermore, we propose a new technique for controllable 3D scene extension to substantially increase the scale of the generated environments. Besides, we introduce a novel abnormal sequence detection module that utilizes camera trajectory to address geometric inconsistency in the generated videos. Finally, WonderVerse is compatible with various 3D reconstruction methods, allowing both efficient and high-quality generation. Extensive experiments on 3D scene generation demonstrate that our WonderVerse, with an elegant and simple pipeline, delivers extendable and highly-realistic 3D scenes, markedly outperforming existing works that rely on more complex architectures.", "date": "2025-03-12", "categories": []} {"id": "2511.21945", "title": "GENA3D: Generative Amodal 3D Modeling by Bridging 2D Priors and 3D Coherence", "abstract": "Generating complete 3D objects under partial occlusions (i.e., amodal scenarios) is a practically important yet challenging problem, as large portions of object geometry are unobserved in real-world scenarios. Existing approaches either operate directly in 3D, which ensures geometric consistency but often lacks generative expressiveness, or rely on 2D amodal completion, which provides strong appearance priors but does not guarantee reliable 3D structure. This raises a key question: how can we achieve both generative plausibility and geometric coherence in amodal 3D modeling? To answer this question, we introduce GENA3D (GENarative Amodal 3D), a framework that integrates learned 2D generative priors with explicit 3D geometric reasoning within a conditional 3D generation paradigm. The 2D priors enable the model to plausibly infer diverse occluded content, while the 3D representation enforces multi-view consistency and spatial validity. Our design incorporates a novel View-Wise Cross-Attention for multi-view alignment and a Stereo-Conditioned Cross-Attention to anchor generative predictions in 3D relationships. By combining generative imagination with structural constraints, GENA3D generates complete and coherent 3D objects from limited observations without sacrificing geometric fidelity. Experiments demonstrate that our method outperforms existing approaches in both synthetic and real-world amodal scenarios, highlighting the effectiveness of bridging 2D priors and 3D coherence in generating plausible and geometrically consistent 3D structures in complex environments.", "date": "2025-11-26", "categories": []} {"id": "2603.13910", "title": "Scene Generation at Absolute Scale: Utilizing Semantic and Geometric Guidance From Text for Accurate and Interpretable 3D Indoor Scene Generation", "abstract": "We present GuidedSceneGen, a text-to-3D generation framework that produces metrically accurate, globally consistent, and semantically interpretable indoor scenes. Unlike prior text-driven methods that often suffer from geometric drift or scale ambiguity, our approach maintains an absolute world coordinate frame throughout the entire generation process. Starting from a textual scene description, we predict a global 3D layout encoding both semantic and geometric structure, which serves as a guiding proxy for downstream stages. A semantics- and depth-conditioned panoramic diffusion model then synthesizes 360{\\deg} imagery aligned with the global layout, substantially improving spatial coherence. To explore unobserved regions, we employ a video diffusion model guided by optimized camera trajectories that balances coverage and collision avoidance, achieving up to 10x faster sampling compared to exhaustive path exploration. The generated views are fused using 3D Gaussian Splatting, yielding a consistent and fully navigable 3D scene in absolute scale. GuidedSceneGen enables accurate transfer of object poses and semantic labels from layout to reconstruction, and supports progressive scene expansion without re-alignment. Quantitative results and a user study demonstrate greater 3D consistency and layout plausibility compared to recent panoramic text-to-3D baselines.", "date": "2026-03-14", "categories": []} {"id": "2603.14021", "title": "EI-Part: Explode for Completion and Implode for Refinement", "abstract": "Part-level 3D generation is crucial for various downstream applications, including gaming, film production, and industrial design. However, decomposing a 3D shape into geometrically plausible and meaningful components remains a significant challenge. Previous part-based generation methods often struggle to produce well-constructed parts, exhibiting poor structural coherence, geometric implausibility, inaccuracy, or inefficiency.\n To address these challenges, we introduce EI-Part, a novel framework specifically designed to generate high-quality 3D shapes with components, characterized by strong structural coherence, geometric plausibility, geometric fidelity, and generation efficiency. We propose utilizing distinct representations at different stages: an Explode state for part completion and an Implode state for geometry refinement. This strategy fully leverages spatial resolution, enabling flexible part completion and fine geometric detail generation. To maintain structural coherence between parts, a self-attention mechanism is incorporated in both exploded and imploded states, facilitating effective information perception and feature fusion among components during generation.\n Extensive experiments on multiple benchmarks demonstrate that EI-Part efficiently produces semantically meaningful and structurally coherent parts with fine-grained geometric details, achieving state-of-the-art performance in part-level 3D generation.\n Project page: https://cvhadessun.github.io/EI-Part/", "date": "2026-03-14", "categories": []} {"id": "2603.14152", "title": "SK-Adapter: Skeleton-Based Structural Control for Native 3D Generation", "abstract": "Native 3D generative models have achieved remarkable fidelity and speed, yet they suffer from a critical limitation: inability to prescribe precise structural articulations, where precise structural control within the native 3D space remains underexplored. This paper proposes SK-Adapter, a simple and yet highly efficient and effective framework that unlocks precise skeletal manipulation for native 3D generation. Moving beyond text or image prompts, which can be ambiguous for precise structure, we treat the 3D skeleton as a first-class control signal. SK-Adapter is a lightweight structural adapter network that encodes joint coordinates and topology into learnable tokens, which are injected into the frozen 3D generation backbone via cross-attention. This smart design allows the model to not only effectively \"attend\" to specific 3D structural constraints but also preserve its original generative priors. To bridge the data gap, we contribute Objaverse-TMS dataset, a large-scale dataset of 24k text-mesh-skeleton pairs. Extensive experiments confirm that our method achieves robust structural control while preserving the geometry and texture quality of the foundation model, significantly outperforming existing baselines. Furthermore, we extend this capability to local 3D editing, enabling the region specific editing of existing assets with skeletal guidance, which is unattainable by previous methods. Project Page: https://sk-adapter.github.io/", "date": "2026-03-14", "categories": []} {"id": "2603.14781", "title": "High-Fidelity 3D Facial Avatar Synthesis with Controllable Fine-Grained Expressions", "abstract": "Facial expression editing methods can be mainly categorized into two types based on their architectures: 2D-based and 3D-based methods. The former lacks 3D face modeling capabilities, making it difficult to edit 3D factors effectively. The latter has demonstrated superior performance in generating high-quality and view-consistent renderings using single-view 2D face images. Although these methods have successfully used animatable models to control facial expressions, they still have limitations in achieving precise control over fine-grained expressions. To address this issue, in this paper, we propose a novel approach by simultaneously refining both the latent code of a pretrained 3D-Aware GAN model for texture editing and the expression code of the driven 3DMM model for mesh editing. Specifically, we introduce a Dual Mappers module, comprising Texture Mapper and Emotion Mapper, to learn the transformations of the given latent code for textures and the expression code for meshes, respectively. To optimize the Dual Mappers, we propose a Text-Guided Optimization method, leveraging a CLIP-based objective function with expression text prompts as targets, while integrating a SubSpace Projection mechanism to project the text embedding to the expression subspace such that we can have more precise control over fine-grained expressions. Extensive experiments and comparative analyses demonstrate the effectiveness and superiority of our proposed method.", "date": "2026-03-16", "categories": []} {"id": "2603.15436", "title": "MV2UV: Generating High-quality UV Texture Maps with Multiview Prompts", "abstract": "Generating high-quality textures for 3D assets is a challenging task. Existing multiview texture generation methods suffer from the multiview inconsistency and missing textures on unseen parts, while UV inpainting texture methods do not generalize well due to insufficient UV data and cannot well utilize 2D image diffusion priors. In this paper, we propose a new method called MV2UV that combines 2D generative priors from multiview generation and the inpainting ability of UV refinement to get high-quality texture maps. Our key idea is to adopt a UV space generative model that simultaneously inpaints unseen parts of multiview images while resolving the inconsistency of multiview images. Experiments show that our method enables a better texture generation quality than existing methods, especially in unseen occluded and multiview-inconsistent parts.", "date": "2026-03-16", "categories": []} {"id": "2412.16253", "title": "ExCellGen: Fast, Controllable, Photorealistic 3D Scene Generation from a Single Real-World Exemplar", "abstract": "Photorealistic 3D scene generation is challenging due to the scarcity of large-scale, high-quality real-world 3D datasets and complex workflows requiring specialized expertise for manual modeling. These constraints often result in slow iteration cycles, where each modification demands substantial effort, ultimately stifling creativity. We propose a fast, exemplar-driven framework for generating 3D scenes from a single casual input, such as handheld video or drone footage. Our method first leverages 3D Gaussian Splatting (3DGS) to robustly reconstruct input scenes with a high-quality 3D appearance model. We then train a per-scene Generative Cellular Automaton (GCA) to produce a sparse volume of featurized voxels, effectively amortizing scene generation while enabling controllability. A subsequent patch-based remapping step composites the complete scene from the exemplar's initial 3D Gaussian splats, successfully recovering the appearance statistics of the input scene. The entire pipeline can be trained in less than 10 minutes for each exemplar and generates scenes in 0.5-2 seconds. Our method enables interactive creation with full user control, and we showcase complex 3D generation results from real-world exemplars within a self-contained interactive GUI.", "date": "2024-12-20", "categories": []} {"id": "2512.05343", "title": "SpaceControl: Introducing Test-Time Spatial Control to 3D Generative Modeling", "abstract": "Generative methods for 3D assets have recently achieved remarkable progress, yet providing intuitive and precise control over the object geometry remains a key challenge. Existing approaches predominantly rely on text or image prompts, which often fall short in geometric specificity: language can be ambiguous, and images are difficult to manipulate. In this work, we introduce SpaceControl, a training-free test-time method for explicit spatial control of 3D asset generation. Our approach accepts a wide range of geometric inputs, from coarse primitives to detailed meshes, and integrates seamlessly with modern generative models without requiring any additional training. A control parameter lets users trade off between geometric fidelity and output realism. Extensive quantitative evaluation and user studies demonstrate that SpaceControl outperforms both training-based and optimization-based baselines in geometric faithfulness while preserving high visual quality. Finally, we present an interactive interface for real-time superquadric editing and direct 3D asset generation, enabling seamless use in creative workflows. Project page: https://spacecontrol3d.github.io/.", "date": "2025-12-05", "categories": []} {"id": "2603.12734", "title": "VecMol: Vector-Field Representations for 3D Molecule Generation", "abstract": "Generative modeling of three-dimensional (3D) molecules is a fundamental yet challenging problem in drug discovery and materials science. Existing approaches typically represent molecules as 3D graphs and co-generate discrete atom types with continuous atomic coordinates, leading to intrinsic learning difficulties such as heterogeneous modality entanglement and geometry-chemistry coherence constraints. We propose VecMol, a paradigm-shifting framework that reimagines molecular representation by modeling 3D molecules as continuous vector fields over Euclidean space, where vectors point toward nearby atoms and implicitly encode molecular structure. The vector field is parameterized by a neural field and generated using a latent diffusion model, avoiding explicit graph generation and decoupling structure learning from discrete atom instantiation. Experiments on the QM9 and GEOM-Drugs benchmarks validate the feasibility of this novel approach, suggesting vector-field-based representations as a promising new direction for 3D molecular generation.", "date": "2026-03-13", "categories": []} {"id": "2603.13098", "title": "SldprtNet: A Large-Scale Multimodal Dataset for CAD Generation in Language-Driven 3D Design", "abstract": "We introduce SldprtNet, a large-scale dataset comprising over 242,000 industrial parts, designed for semantic-driven CAD modeling, geometric deep learning, and the training and fine-tuning of multimodal models for 3D design. The dataset provides 3D models in both .step and .sldprt formats to support diverse training and testing. To enable parametric modeling and facilitate dataset scalability, we developed supporting tools, an encoder and a decoder, which support 13 types of CAD commands and enable lossless transformation between 3D models and a structured text representation. Additionally, each sample is paired with a composite image created by merging seven rendered views from different viewpoints of the 3D model, effectively reducing input token length and accelerating inference. By combining this image with the parameterized text output from the encoder, we employ the lightweight multimodal language model Qwen2.5-VL-7B to generate a natural language description of each part's appearance and functionality. To ensure accuracy, we manually verified and aligned the generated descriptions, rendered images, and 3D models. These descriptions, along with the parameterized modeling scripts, rendered images, and 3D model files, are fully aligned to construct SldprtNet. To assess its effectiveness, we fine-tuned baseline models on a dataset subset, comparing image-plus-text inputs with text-only inputs. Results confirm the necessity and value of multimodal datasets for CAD generation. It features carefully selected real-world industrial parts, supporting tools for scalable dataset expansion, diverse modalities, and ensured diversity in model complexity and geometric features, making it a comprehensive multimodal dataset built for semantic-driven CAD modeling and cross-modal learning.", "date": "2026-03-13", "categories": []} {"id": "2603.12126", "title": "Hoi3DGen: Generating High-Quality Human-Object-Interactions in 3D", "abstract": "Modeling and generating 3D human-object interactions from text is crucial for applications in AR, XR, and gaming. Existing approaches often rely on score distillation from text-to-image models, but their results suffer from the Janus problem and do not follow text prompts faithfully due to the scarcity of high-quality interaction data. We introduce Hoi3DGen, a framework that generates high-quality textured meshes of human-object interaction that follow the input interaction descriptions precisely. We first curate realistic and high-quality interaction data leveraging multimodal large language models, and then create a full text-to-3D pipeline, which achieves orders-of-magnitude improvements in interaction fidelity. Our method surpasses baselines by 4-15x in text consistency and 3-7x in 3D model quality, exhibiting strong generalization to diverse categories and interaction types, while maintaining high-quality 3D generation.", "date": "2026-03-12", "categories": []} {"id": "2603.12238", "title": "SceneAssistant: A Visual Feedback Agent for Open-Vocabulary 3D Scene Generation", "abstract": "Text-to-3D scene generation from natural language is highly desirable for digital content creation. However, existing methods are largely domain-restricted or reliant on predefined spatial relationships, limiting their capacity for unconstrained, open-vocabulary 3D scene synthesis. In this paper, we introduce SceneAssistant, a visual-feedback-driven agent designed for open-vocabulary 3D scene generation. Our framework leverages modern 3D object generation model along with the spatial reasoning and planning capabilities of Vision-Language Models (VLMs). To enable open-vocabulary scene composition, we provide the VLMs with a comprehensive set of atomic operations (e.g., Scale, Rotate, FocusOn). At each interaction step, the VLM receives rendered visual feedback and takes actions accordingly, iteratively refining the scene to achieve more coherent spatial arrangements and better alignment with the input text. Experimental results demonstrate that our method can generate diverse, open-vocabulary, and high-quality 3D scenes. Both qualitative analysis and quantitative human evaluations demonstrate the superiority of our approach over existing methods. Furthermore, our method allows users to instruct the agent to edit existing scenes based on natural language commands. Our code is available at https://github.com/ROUJINN/SceneAssistant", "date": "2026-03-12", "categories": []} {"id": "2504.14373", "title": "SEGA: Drivable 3D Gaussian Head Avatar from a Single Image", "abstract": "Creating photorealistic 3D head avatars from limited input has become increasingly important for applications in virtual reality, telepresence, and digital entertainment. While recent advances like neural rendering and 3D Gaussian splatting have enabled high-quality digital human avatar creation and animation, most methods rely on multiple images or multi-view inputs, limiting their practicality for real-world use. In this paper, we propose SEGA, a novel approach for Single-imagE-based 3D drivable Gaussian head Avatar creation that combines generalized prior models with a new hierarchical UV-space Gaussian Splatting framework. SEGA seamlessly combines priors derived from large-scale 2D datasets with 3D priors learned from multi-view, multi-expression, and multi-ID data, achieving robust generalization to unseen identities while ensuring 3D consistency across novel viewpoints and expressions. We further present a hierarchical UV-space Gaussian Splatting framework that leverages FLAME-based structural priors and employs a dual-branch architecture to disentangle dynamic and static facial components effectively. The dynamic branch encodes expression-driven fine details, while the static branch focuses on expression-invariant regions, enabling efficient parameter inference and precomputation. This design maximizes the utility of limited 3D data and achieves real-time performance for animation and rendering. Additionally, SEGA performs person-specific fine-tuning to further enhance the fidelity and realism of the generated avatars. Experiments show our method outperforms state-of-the-art approaches in generalization ability, identity preservation, and expression realism, advancing one-shot avatar creation for practical applications.", "date": "2025-04-19", "categories": []} {"id": "2510.13702", "title": "MVCustom: Multi-View Customized Diffusion via Geometric Latent Rendering and Completion", "abstract": "Multi-view generation with camera pose control and prompt-based customization are both essential elements for achieving controllable generative models. However, existing multi-view generation models do not support customization with geometric consistency, whereas customization models lack explicit viewpoint control, making them challenging to unify. Motivated by these gaps, we introduce a novel task, multi-view customization, which aims to jointly achieve multi-view camera pose control and customization. Due to the scarcity of training data in customization, existing multi-view generation models, which inherently rely on large-scale datasets, struggle to generalize to diverse prompts. To address this, we propose MVCustom, a novel diffusion-based framework explicitly designed to achieve both multi-view consistency and customization fidelity. In the training stage, MVCustom learns the subject's identity and geometry using a feature-field representation, incorporating the text-to-video diffusion backbone enhanced with dense spatio-temporal attention, which leverages temporal coherence for multi-view consistency. In the inference stage, we introduce two novel techniques: depth-aware feature rendering explicitly enforces geometric consistency, and consistent-aware latent completion ensures accurate perspective alignment of the customized subject and surrounding backgrounds. Extensive experiments demonstrate that MVCustom achieves the most balanced and consistent competitive performance across multi-view consistency, customization fidelity, demonstrating effective solution of multi-objective generation task.", "date": "2025-10-15", "categories": []} {"id": "2502.01949", "title": "LAYOUTDREAMER: Physics-guided Layout for Text-to-3D Compositional Scene Generation", "abstract": "Recently, the field of text-guided 3D scene generation has garnered significant attention. High-quality generation that aligns with physical realism and high controllability is crucial for practical 3D scene applications. However, existing methods face fundamental limitations: (i) difficulty capturing complex relationships between multiple objects described in the text, (ii) inability to generate physically plausible scene layouts, and (iii) lack of controllability and extensibility in compositional scenes. In this paper, we introduce LayoutDreamer, a framework that leverages 3D Gaussian Splatting (3DGS) to facilitate high-quality, physically consistent compositional scene generation guided by text. Specifically, given a text prompt, we convert it into a directed scene graph and adaptively adjust the density and layout of the initial compositional 3D Gaussians. Subsequently, dynamic camera adjustments are made based on the training focal point to ensure entity-level generation quality. Finally, by extracting directed dependencies from the scene graph, we tailor physical and layout energy to ensure both realism and flexibility. Comprehensive experiments demonstrate that LayoutDreamer outperforms other compositional scene generation quality and semantic alignment methods. Specifically, it achieves state-of-the-art (SOTA) performance in the multiple objects generation metric of T3Bench.", "date": "2025-02-04", "categories": []} {"id": "2409.16990", "title": "Single Image, Any Face: Generalisable 3D Face Generation", "abstract": "The creation of 3D human face avatars from a single unconstrained image is a fundamental task that underlies numerous real-world vision and graphics applications. Despite the significant progress made in generative models, existing methods are either less suited in design for human faces or fail to generalise from the restrictive training domain to unconstrained facial images. To address these limitations, we propose a novel model, Gen3D-Face, which generates 3D human faces with unconstrained single image input within a multi-view consistent diffusion framework. Given a specific input image, our model first produces multi-view images, followed by neural surface construction. To incorporate face geometry information while preserving generalisation to in-the-wild inputs, we estimate a subject-specific mesh directly from the input image, enabling training and evaluation without ground-truth 3D supervision. Importantly, we introduce a multi-view joint generation scheme to enhance the appearance consistency among different views. To the best of our knowledge, this is the first attempt and benchmark for creating photorealistic 3D human face avatars from single images for generic human subject across domains. Extensive experiments demonstrate the efficacy and superiority of our method over previous alternatives for out-of-domain single image 3D face generation and the top ranking competition for the in-domain setting.", "date": "2024-09-25", "categories": []} {"id": "2503.11978", "title": "Snapmoji: Instant Generation of Animatable Dual-Stylized Avatars", "abstract": "Despite the increasing popularity of avatar systems such as Snapchat Bitmojis, existing production avatar platforms face several limitations, such as a limited number of predefined assets, tedious customization processes, and inefficient rendering requirements. Addressing these shortcomings, we introduce Snapmoji, an avatar generation system that instantly creates 3D avatars, and enables customization in a process we call dual-stylization. Snapmoji first maps a selfie of a user to a primary avatar (e.g., Bitmoji style) using a new technique we name Gaussian Domain Adaptation (GDA), then applies a secondary style (e.g., skeleton, yarn, toy) to the primary avatar, all while preserving the user's identity. The generated 3D avatars can then be rendered an animated on mobile devices at 30-40 FPS.", "date": "2025-03-15", "categories": []} {"id": "2508.11952", "title": "UniUGG: Unified 3D Understanding and Generation via Geometric-Semantic Encoding", "abstract": "Despite the impressive progress on understanding and generating images shown by the recent unified architectures, the integration of 3D tasks remains challenging and largely unexplored. In this paper, we introduce UniUGG, the first unified understanding and generation framework for 3D modalities. Our unified framework employs an LLM to comprehend and decode sentences and 3D representations. At its core, we propose a spatial decoder leveraging a latent diffusion model to generate high-quality 3D representations. This allows for the generation and imagination of 3D scenes based on a reference image and an arbitrary view transformation, while remaining supports for spatial visual question answering (VQA) tasks. Additionally, we propose a geometric-semantic learning strategy to pretrain the vision encoder. This design jointly captures the input's semantic and geometric cues, enhancing both spatial understanding and generation. Extensive experimental results demonstrate the superiority of our method in visual representation, spatial understanding, and 3D generation.", "date": "2025-08-16", "categories": []} {"id": "2511.18734", "title": "Yo'City: Personalized and Boundless 3D Realistic City Scene Generation via Self-Critic Expansion", "abstract": "Realistic 3D city generation is fundamental to a wide range of applications, including virtual reality and digital twins. However, most existing methods rely on training a single diffusion model, which limits their ability to generate personalized and boundless city-scale scenes. In this paper, we present Yo'City, a novel agentic framework that enables user-customized and infinitely expandable 3D city generation by leveraging the reasoning and compositional capabilities of off-the-shelf large models. Specifically, Yo'City first conceptualizes the city through a top-down planning strategy that defines a hierarchical \"City-District-Grid\" structure. The Global Planner determines the overall layout and potential functional districts, while the Local Designer further refines each district with detailed grid-level descriptions. Subsequently, the grid-level 3D generation is achieved through a \"produce-refine-evaluate\" isometric image synthesis loop, followed by image-to-3D generation. To simulate continuous city evolution, Yo'City further introduces a user-interactive, relationship-guided expansion mechanism, which performs scene graph-based distance- and semantics-aware layout optimization, ensuring spatially coherent city growth. To comprehensively evaluate our method, we construct a diverse benchmark dataset and design six multi-dimensional metrics that assess generation quality from the perspectives of semantics, geometry, texture, and layout. Extensive experiments demonstrate that Yo'City consistently outperforms existing state-of-the-art methods across all evaluation aspects.", "date": "2025-11-24", "categories": []} {"id": "2603.08228", "title": "GarmentPainter: Efficient 3D Garment Texture Synthesis with Character-Guided Diffusion Model", "abstract": "Generating high-fidelity, 3D-consistent garment textures remains a challenging problem due to the inherent complexities of garment structures and the stringent requirement for detailed, globally consistent texture synthesis. Existing approaches either rely on 2D-based diffusion models, which inherently struggle with 3D consistency, require expensive multi-step optimization or depend on strict spatial alignment between 2D reference images and 3D meshes, which limits their flexibility and scalability. In this work, we introduce GarmentPainter, a simple yet efficient framework for synthesizing high-quality, 3D-aware garment textures in UV space. Our method leverages a UV position map as the 3D structural guidance, ensuring texture consistency across the garment surface during texture generation. To enhance control and adaptability, we introduce a type selection module, enabling fine-grained texture generation for specific garment components based on a character reference image, without requiring alignment between the reference image and the 3D mesh. GarmentPainter efficiently integrates all guidance signals into the input of a diffusion model in a spatially aligned manner, without modifying the underlying UNet architecture. Extensive experiments demonstrate that GarmentPainter achieves state-of-the-art performance in terms of visual fidelity, 3D consistency, and computational efficiency, outperforming existing methods in both qualitative and quantitative evaluations.", "date": "2026-03-09", "categories": []} {"id": "2511.11368", "title": "LaxMotion: Rethinking Supervision Granularity for 3D Human Motion Generation", "abstract": "Recent 3D human motion generation models demonstrate remarkable reconstruction accuracy yet struggle to generalize beyond training distributions. This limitation arises partly from the use of precise 3D supervision, which encourages models to fit fixed coordinate patterns instead of learning the essential 3D structure and motion semantic cues required for robust generalization. To overcome this limitation, we propose LaxMotion, a framework that synthesizes realistic 3D motions without direct 3D pose supervision. Instead of regressing toward exact coordinates, LaxMotion learns 3D motion as a consistent explanation of global trajectories and monocular 2D kinematic cues. We introduce a structured motion factorization together with a reformulated training paradigm under relaxed observability. This design is further supported by relaxed regularization objectives that enforce view consistent alignment, orientation coherence, and structural stability. Under this relaxed supervision paradigm, LaxMotion generates diverse, temporally coherent, and semantically aligned 3D motions, achieving performance comparable to or surpassing fully 3D supervised methods. These results indicate that shifting supervision from exact coordinate matching to structural consistency promotes stronger reasoning and improved generalization, offering a scalable and data efficient paradigm for 3D motion generation.", "date": "2025-11-14", "categories": []} {"id": "2512.08535", "title": "Photo3D: Advancing Photorealistic 3D Generation through Structure-Aligned Detail Enhancement", "abstract": "Although recent 3D-native generators have made great progress in synthesizing reliable geometry, they still fall short in achieving realistic appearances. A key obstacle lies in the lack of diverse and high-quality real-world 3D assets with rich texture details, since capturing such data is intrinsically difficult due to the diverse scales of scenes, non-rigid motions of objects, and the limited precision of 3D scanners. We introduce Photo3D, a framework for advancing photorealistic 3D generation, which is driven by the image data generated by the GPT-4o-Image model. Considering that the generated images can distort 3D structures due to their lack of multi-view consistency, we design a structure-aligned multi-view synthesis pipeline and construct a detail-enhanced multi-view dataset paired with 3D geometry. Building on it, we present a realistic detail enhancement scheme that leverages perceptual feature adaptation and semantic structure matching to enforce appearance consistency with realistic details while preserving the structural consistency with the 3D-native geometry. Our scheme is general to different 3D-native generators, and we present dedicated training strategies to facilitate the optimization of geometry-texture coupled and decoupled 3D-native generation paradigms. Experiments demonstrate that Photo3D generalizes well across diverse 3D-native generation paradigms and achieves state-of-the-art photorealistic 3D generation performance.", "date": "2025-12-09", "categories": []} {"id": "2603.05607", "title": "DreamCAD: Scaling Multi-modal CAD Generation using Differentiable Parametric Surfaces", "abstract": "Computer-Aided Design (CAD) relies on structured and editable geometric representations, yet existing generative methods are constrained by small annotated datasets with explicit design histories or boundary representation (BRep) labels. Meanwhile, millions of unannotated 3D meshes remain untapped, limiting progress in scalable CAD generation. To address this, we propose DreamCAD, a multi-modal generative framework that directly produces editable BReps from point-level supervision, without CAD-specific annotations. DreamCAD represents each BRep as a set of parametric patches (e.g., B\\'ezier surfaces) and uses a differentiable tessellation method to generate meshes. This enables large-scale training on 3D datasets while reconstructing connected and editable surfaces. Furthermore, we introduce CADCap-1M, the largest CAD captioning dataset to date, with 1M+ descriptions generated using GPT-5 for advancing text-to-CAD research. DreamCAD achieves state-of-the-art performance on ABC and Objaverse benchmarks across text, image, and point modalities, improving geometric fidelity and surpassing 75% user preference. Code and dataset will be publicly available.", "date": "2026-03-05", "categories": []} {"id": "2603.05845", "title": "Cog2Gen3D: Sculpturing 3D Semantic-Geometric Cognition for 3D Generation", "abstract": "Generative models have achieved success in producing semantically plausible 2D images, but it remains challenging in 3D generation due to the absence of spatial geometry constraints. Typically, existing methods utilize geometric features as conditions to enhance spatial awareness. However, these methods can only model relative relationships and are prone to scale inconsistency of absolute geometry. Thus, we argue that semantic information and absolute geometry empower 3D cognition, thereby enabling controllable 3D generation for the physical world. In this work, we propose Cog2Gen3D, a 3D cognition-guided diffusion framework for 3D generation. Our model is guided by three key designs: 1) Cognitive Feature Embeddings. We encode different modalities into semantic and geometric representations and further extract logical representations. 2) 3D Latent Cognition Graph. We structure different representations into dual-stream semantic-geometric graphs and fuse them via common-based cross-attention to obtain a 3D cognition graph. 3) Cognition-Guided Latent Diffusion. We leverage the fused 3D cognition graph as the condition to guide the latent diffusion process for 3D Gaussian generation. Under this unified framework, the 3D cognition graph ensures the physical plausibility and structural rationality of 3D generation. Moreover, we construct a validation subset based on the Marble World Labs. Extensive experiments demonstrate that our Cog2Gen3D significantly outperforms existing methods in both semantic fidelity and geometric plausibility.", "date": "2026-03-06", "categories": []} {"id": "2603.05908", "title": "Pano3DComposer: Feed-Forward Compositional 3D Scene Generation from Single Panoramic Image", "abstract": "Current compositional image-to-3D scene generation approaches construct 3D scenes by time-consuming iterative layout optimization or inflexible joint object-layout generation. Moreover, most methods rely on limited field-of-view perspective images, hindering the creation of complete 360-degree environments. To address these limitations, we design Pano3DComposer, an efficient feed-forward framework for panoramic images. To decouple object generation from layout estimation, we propose a plug-and-play Object-World Transformation Predictor. This module converts the 3D objects generated by off-the-shelf image-to-3D models from local to world coordinates. To achieve this, we adapt the VGGT architecture to Alignment-VGGT by using target object crop, multi-view object renderings and camera parameters to predict the transformation. The predictor is trained using pseudo-geometric supervision to address the shape discrepancy between generated and ground-truth objects. For input images from unseen domains, we further introduce a Coarse-to-Fine (C2F) alignment mechanism for Pano3DComposer that iteratively refines geometric consistency with feedback of scene rendering. Our method achieves superior geometric accuracy for image/text-to-3D tasks on synthetic and real-world datasets. It can generate a high-fidelity 3D scene in approximately 20 seconds on an RTX 4090 GPU. Project page: https://qiuzidian.github.io/pano3dcomposer-page/.", "date": "2026-03-06", "categories": []} {"id": "2603.06357", "title": "LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents", "abstract": "In this paper, we introduce LATO, a novel topology-preserving latent representation that enables scalable, flow matching-based synthesis of explicit 3D meshes. LATO represents a mesh as a Vertex Displacement Field (VDF) anchored on surface, incorporating a sparse voxel Variational Autoencoder (VAE) to compress this explicit signal into a structured, topology-aware voxel latent. To decapsulate the mesh, the VAE decoder progressively subdivides and prunes latent voxels to instantiate precise vertex locations. In the end, a dedicated connection head queries the voxel latent to predict edge connectivity between vertex pairs directly, allowing mesh topology to be recovered without isosurface extraction or heuristic meshing. For generative modeling, LATO adopts a two-stage flow matching process, first synthesizing the structure voxels and subsequently refining the voxel-wise topology features. Compared to prior isosurface/triangle-based diffusion models and autoregressive generation approaches, LATO generates meshes with complex geometry, well-formed topology while being highly efficient in inference.", "date": "2026-03-06", "categories": []} {"id": "2510.13454", "title": "Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator", "abstract": "The rapid progress of large, pretrained models for both visual content generation and 3D reconstruction opens up new possibilities for text-to-3D generation. Intuitively, one could obtain a formidable 3D scene generator if one were able to combine the power of a modern latent text-to-video model as \"generator\" with the geometric abilities of a recent (feedforward) 3D reconstruction system as \"decoder\". We introduce VIST3A, a general framework that does just that, addressing two main challenges. First, the two components must be joined in a way that preserves the rich knowledge encoded in their weights. We revisit model stitching, i.e., we identify the layer in the 3D decoder that best matches the latent representation produced by the text-to-video generator and stitch the two parts together. That operation requires only a small dataset and no labels. Second, the text-to-video generator must be aligned with the stitched 3D decoder, to ensure that the generated latents are decodable into consistent, perceptually convincing 3D scene geometry. To that end, we adapt direct reward finetuning, a popular technique for human preference alignment. We evaluate the proposed VIST3A approach with different video generators and 3D reconstruction models. All tested pairings markedly improve over prior text-to-3D models that output Gaussian splats. Moreover, by choosing a suitable 3D base model, VIST3A also enables high-quality text-to-pointmap generation.", "date": "2025-10-15", "categories": []} {"id": "2601.00204", "title": "MorphAny3D: Unleashing the Power of Structured Latent in 3D Morphing", "abstract": "3D morphing remains challenging due to the difficulty of generating semantically consistent and temporally smooth deformations, especially across categories. We present MorphAny3D, a training-free framework that leverages Structured Latent (SLAT) representations for high-quality 3D morphing. Our key insight is that intelligently blending source and target SLAT features within the attention mechanisms of 3D generators naturally produces plausible morphing sequences. To this end, we introduce Morphing Cross-Attention (MCA), which fuses source and target information for structural coherence, and Temporal-Fused Self-Attention (TFSA), which enhances temporal consistency by incorporating features from preceding frames. An orientation correction strategy further mitigates the pose ambiguity within the morphing steps. Extensive experiments show that our method generates state-of-the-art morphing sequences, even for challenging cross-category cases. MorphAny3D further supports advanced applications such as decoupled morphing and 3D style transfer, and can be generalized to other SLAT-based generative models. Project page: https://xiaokunsun.github.io/MorphAny3D.github.io/.", "date": "2026-01-01", "categories": []} {"id": "2603.03602", "title": "DM-CFO: A Diffusion Model for Compositional 3D Tooth Generation with Collision-Free Optimization", "abstract": "The automatic design of a 3D tooth model plays a crucial role in dental digitization. However, current approaches face challenges in compositional 3D tooth generation because both the layouts and shapes of missing teeth need to be optimized.In addition, collision conflicts are often omitted in 3D Gaussian-based compositional 3D generation, where objects may intersect with each other due to the absence of explicit geometric information on the object surfaces. Motivated by graph generation through diffusion models and collision detection using 3D Gaussians, we propose an approach named DM-CFO for compositional tooth generation, where the layout of missing teeth is progressively restored during the denoising phase under both text and graph constraints. Then, the Gaussian parameters of each layout-guided tooth and the entire jaw are alternately updated using score distillation sampling (SDS). Furthermore, a regularization term based on the distances between the 3D Gaussians of neighboring teeth and the anchor tooth is introduced to penalize tooth intersections. Experimental results on three tooth-design datasets demonstrate that our approach significantly improves the multiview consistency and realism of the generated teeth compared with existing methods. Project page: https://amateurc.github.io/CF-3DTeeth/.", "date": "2026-03-04", "categories": []} {"id": "2603.04307", "title": "Dual Diffusion Models for Multi-modal Guided 3D Avatar Generation", "abstract": "Generating high-fidelity 3D avatars from text or image prompts is highly sought after in virtual reality and human-computer interaction. However, existing text-driven methods often rely on iterative Score Distillation Sampling (SDS) or CLIP optimization, which struggle with fine-grained semantic control and suffer from excessively slow inference. Meanwhile, image-driven approaches are severely bottlenecked by the scarcity and high acquisition cost of high-quality 3D facial scans, limiting model generalization. To address these challenges, we first construct a novel, large-scale dataset comprising over 100,000 pairs across four modalities: fine-grained textual descriptions, in-the-wild face images, high-quality light-normalized texture UV maps, and 3D geometric shapes. Leveraging this comprehensive dataset, we propose PromptAvatar, a framework featuring dual diffusion models. Specifically, it integrates a Texture Diffusion Model (TDM) that supports flexible multi-condition guidance from text and/or image prompts, alongside a Geometry Diffusion Model (GDM) guided by text prompts. By learning the direct mapping from multi-modal prompts to 3D representations, PromptAvatar eliminates the need for time-consuming iterative optimization, successfully generating high-fidelity, shading-free 3D avatars in under 10 seconds. Extensive quantitative and qualitative experiments demonstrate that our method significantly outperforms existing state-of-the-art approaches in generation quality, fine-grained detail alignment, and computational efficiency.", "date": "2026-03-04", "categories": []} {"id": "2603.01515", "title": "FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation", "abstract": "Autoregressive models for 3D mesh generation suffer from a fundamental limitation: they flatten meshes into long vertex-coordinate sequences. This results in prohibitive computational costs, hindering the efficient synthesis of high-fidelity geometry. We argue this bottleneck stems from operating at the wrong semantic level. We introduce FACE, a novel Autoregressive Autoencoder (ARAE) framework that reconceptualizes the task by generating meshes at the face level. Our one-face-one-token strategy treats each triangle face, the fundamental building block of a mesh, as a single, unified token. This simple yet powerful design reduces the sequence length by a factor of nine, leading to an unprecedented compression ratio of 0.11, halving the previous state-of-the-art. This dramatic efficiency gain does not compromise quality; by pairing our face-level decoder with a powerful VecSet encoder, FACE achieves state-of-the-art reconstruction quality on standard benchmarks. The versatility of the learned latent space is further demonstrated by training a latent diffusion model that achieves high-fidelity, single-image-to-mesh generation. FACE provides a simple, scalable, and powerful paradigm that lowers the barrier to high-quality structured 3D content creation.", "date": "2026-03-02", "categories": []} {"id": "2603.03143", "title": "Geometry-Guided Reinforcement Learning for Multi-view Consistent 3D Scene Editing", "abstract": "Leveraging the priors of 2D diffusion models for 3D editing has emerged as a promising paradigm. However, maintaining multi-view consistency in edited results remains challenging, and the extreme scarcity of 3D-consistent editing paired data renders supervised fine-tuning (SFT), the most effective training strategy for editing tasks, infeasible. In this paper, we observe that, while generating multi-view consistent 3D content is highly challenging, verifying 3D consistency is tractable, naturally positioning reinforcement learning (RL) as a feasible solution. Motivated by this, we propose \\textbf{RL3DEdit}, a single-pass framework driven by RL optimization with novel rewards derived from the 3D foundation model, VGGT. Specifically, we leverage VGGT's robust priors learned from massive real-world data, feed the edited images, and utilize the output confidence maps and pose estimation errors as reward signals, effectively anchoring the 2D editing priors onto a 3D-consistent manifold via RL. Extensive experiments demonstrate that RL3DEdit achieves stable multi-view consistency and outperforms state-of-the-art methods in editing quality with high efficiency. To promote the development of 3D editing, we will release the code and model.", "date": "2026-03-03", "categories": []} {"id": "2505.17702", "title": "Seek-CAD: A Self-refined Generative Modeling for 3D Parametric CAD Using Local Inference via DeepSeek", "abstract": "The advent of Computer-Aided Design (CAD) generative modeling will significantly transform the design of industrial products. The recent research endeavor has extended into the realm of Large Language Models (LLMs). In contrast to fine-tuning methods, training-free approaches typically utilize the advanced closed-source LLMs, thereby offering enhanced flexibility and efficiency in the development of AI agents for generating CAD parametric models. However, the substantial cost and limitations of local deployment of the top-tier closed-source LLMs pose challenges in practical applications. The Seek-CAD is the pioneer exploration of locally deployed open-source inference LLM DeepSeek-R1 for CAD parametric model generation with a training-free methodology. This study is the first investigation to incorporate both visual and Chain-of-Thought (CoT) feedback within the self-refinement mechanism for generating CAD models. Specifically, the initial generated parametric CAD model is rendered into a sequence of step-wise perspective images, which are subsequently processed by a Vision Language Model (VLM) alongside the corresponding CoTs derived from DeepSeek-R1 to assess the CAD model generation. Then, the feedback is utilized by DeepSeek-R1 to refine the initial generated model for the next round of generation. Moreover, we present an innovative 3D CAD model dataset structured around the SSR (Sketch, Sketch-based feature, and Refinements) triple design paradigm. This dataset encompasses a wide range of CAD commands, thereby aligning effectively with industrial application requirements and proving suitable for the generation of LLMs. Extensive experiments validate the effectiveness of Seek-CAD under various metrics.", "date": "2025-05-23", "categories": []} {"id": "2507.02477", "title": "Topology-Preserved Auto-regressive Mesh Generation in the Manner of Weaving Silk", "abstract": "Existing auto-regressive mesh generation approaches suffer from ineffective topology preservation, which is crucial for practical applications. This limitation stems from previous mesh tokenization methods treating meshes as simple collections of equivalent triangles, lacking awareness of the overall topological structure during generation. To address this issue, we propose a novel mesh tokenization algorithm that provides a canonical topological framework through vertex layering and ordering, ensuring critical geometric properties including manifoldness, watertightness, face normal consistency, and part awareness in the generated meshes. Measured by Compression Ratio and Bits-per-face, we also achieved state-of-the-art compression efficiency. Furthermore, we introduce an online non-manifold data processing algorithm and a training resampling strategy to expand the scale of trainable dataset and avoid costly manual data curation. Experimental results demonstrate the effectiveness of our approach, showcasing not only intricate mesh generation but also significantly improved geometric integrity.", "date": "2025-07-03", "categories": []} {"id": "2509.05285", "title": "Improved 3D Scene Stylization via Text-Guided Generative Image Editing with Region-Based Control", "abstract": "Recent advances in text-driven 3D scene editing and stylization, which leverage the powerful capabilities of 2D generative models, have demonstrated promising outcomes. However, challenges remain in ensuring high-quality stylization and view consistency simultaneously. Moreover, applying style consistently to different regions or objects in the scene with semantic correspondence is a challenging task. To address these limitations, we introduce techniques that enhance the quality of 3D stylization while maintaining view consistency and providing optional region-controlled style transfer. Our method achieves stylization by re-training an initial 3D representation using stylized multi-view 2D images of the source views. Therefore, ensuring both style consistency and view consistency of stylized multi-view images is crucial. We achieve this by extending the style-aligned depth-conditioned view generation framework, replacing the fully shared attention mechanism with a single reference-based attention-sharing mechanism, which effectively aligns style across different viewpoints. Additionally, inspired by recent 3D inpainting methods, we utilize a grid of multiple depth maps as a single-image reference to further strengthen view consistency among stylized images. Finally, we propose Multi-Region Importance-Weighted Sliced Wasserstein Distance Loss, allowing styles to be applied to distinct image regions using segmentation masks from off-the-shelf models. We demonstrate that this optional feature enhances the faithfulness of style transfer and enables the mixing of different styles across distinct regions of the scene. Experimental evaluations, both qualitative and quantitative, demonstrate that our pipeline effectively improves the results of text-driven 3D stylization. Project Page: https://haruolabs.github.io/improved-gs-style-page/", "date": "2025-09-04", "categories": []} {"id": "2509.21420", "title": "QuadGPT: Native Quadrilateral Mesh Generation with Autoregressive Models", "abstract": "The generation of quadrilateral-dominant meshes is a cornerstone of professional 3D content creation. However, existing generative models generate quad meshes by first generating triangle meshes and then merging triangles into quadrilaterals with some specific rules, which typically produces quad meshes with poor topology. In this paper, we introduce QuadGPT, the first autoregressive framework for generating quadrilateral meshes in an end-to-end manner. QuadGPT formulates this as a sequence prediction paradigm, distinguished by two key innovations: a unified tokenization method to handle mixed topologies of triangles and quadrilaterals, and a specialized Reinforcement Learning fine-tuning method tDPO for better generation quality. Extensive experiments demonstrate that QuadGPT significantly surpasses previous triangle-to-quad conversion pipelines in both geometric accuracy and topological quality. Our work establishes a new benchmark for native quad-mesh generation and showcases the power of combining large-scale autoregressive models with topology-aware RL refinement for creating structured 3D assets.", "date": "2025-09-25", "categories": []} {"id": "2602.18873", "title": "BiMotion: B-spline Motion for Text-guided Dynamic 3D Character Generation", "abstract": "Text-guided dynamic 3D character generation has advanced rapidly, yet producing high-quality motion that faithfully reflects rich textual descriptions remains challenging. Existing methods tend to generate limited sub-actions or incoherent motion due to fixed-length temporal inputs and discrete frame-wise representations that fail to capture rich motion semantics. We address these limitations by representing motion with continuous differentiable B-spline curves, enabling more effective motion generation without modifying the capabilities of the underlying generative model. Specifically, our closed-form, Laplacian-regularized B-spline solver efficiently compresses variable-length motion sequences into compact representations with a fixed number of control points. Further, we introduce a normal-fusion strategy for input shape adherence along with correspondence-aware and local-rigidity losses for motion-restoration quality. To train our model, we collate BIMO, a new dataset containing diverse variable-length 3D motion sequences with rich, high-quality text annotations. Extensive evaluations show that our feed-forward framework BiMotion generates more expressive, higher-quality, and better prompt-aligned motions than existing state-of-the-art methods, while also achieving faster generation. Our project page is at: https://wangmiaowei.github.io/BiMotion.github.io/.", "date": "2026-02-21", "categories": []} {"id": "2603.00439", "title": "Mamba-CAD: State Space Model For 3D Computer-Aided Design Generative Modeling", "abstract": "Computer-Aided Design (CAD) generative modeling has a strong and long-term application in the industry. Recently, the parametric CAD sequence as the design logic of an object has been widely mined by sequence models. However, the industrial CAD models, especially in component objects, are fine-grained and complex, requiring a longer parametric CAD sequence to define. To address the problem, we introduce Mamba-CAD, a self-supervised generative modeling for complex CAD models in the industry, which can model on a longer parametric CAD sequence. Specifically, we first design an encoder-decoder framework based on a Mamba architecture and pair it with a CAD reconstruction task for pre-training to model the latent representation of CAD models; and then we utilize the learned representation to guide a generative adversarial network to produce the fake representation of CAD models, which would be finally recovered into parametric CAD sequences via the decoder of MambaCAD. To train Mamba-CAD, we further create a new dataset consisting of 77,078 CAD models with longer parametric CAD sequences. Comprehensive experiments are conducted to demonstrate the effectiveness of our model under various evaluation metrics, especially in the generation length of valid parametric CAD sequences. The code and dataset can be achieved from https://github.com/Sunny-Hack/Code-for-Mamba-CAD-AAAI-2025-.", "date": "2026-02-28", "categories": []} {"id": "2603.00526", "title": "Mesh-Pro: Asynchronous Advantage-guided Ranking Preference Optimization for Artist-style Quadrilateral Mesh Generation", "abstract": "Reinforcement learning (RL) has demonstrated remarkable success in text and image generation, yet its potential in 3D generation remains largely unexplored. Existing attempts typically rely on offline direct preference optimization (DPO) method, which suffers from low training efficiency and limited generalization. In this work, we aim to enhance both the training efficiency and generation quality of RL in 3D mesh generation. Specifically, (1) we design the first asynchronous online RL framework tailored for 3D mesh generation post-training efficiency improvement, which is 3.75$\\times$ faster than synchronous RL. (2) We propose Advantage-guided Ranking Preference Optimization (ARPO), a novel RL algorithm that achieves a better trade-off between training efficiency and generalization than current RL algorithms designed for 3D mesh generation, such as DPO and group relative policy optimization (GRPO). (3) Based on asynchronous ARPO, we propose Mesh-Pro, which additionally introduces a novel diagonal-aware mixed triangular-quadrilateral tokenization for mesh representation and a ray-based reward for geometric integrity. Mesh-Pro achieves state-of-the-art performance on artistic and dense meshes.", "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": "2603.01594", "title": "Preference Score Distillation: Leveraging 2D Rewards to Align Text-to-3D Generation with Human Preference", "abstract": "Human preference alignment presents a critical yet underexplored challenge for diffusion models in text-to-3D generation. Existing solutions typically require task-specific fine-tuning, posing significant hurdles in data-scarce 3D domains. To address this, we propose Preference Score Distillation (PSD), an optimization-based framework that leverages pretrained 2D reward models for human-aligned text-to-3D synthesis without 3D training data. Our key insight stems from the incompatibility of pixel-level gradients: due to the absence of noisy samples during reward model training, direct application of 2D reward gradients disturbs the denoising process. Noticing that similar issue occurs in the naive classifier guidance in conditioned diffusion models, we fundamentally rethink preference alignment as a classifier-free guidance (CFG)-style mechanism through our implicit reward model. Furthermore, recognizing that frozen pretrained diffusion models constrain performance, we introduce an adaptive strategy to co-optimize preference scores and negative text embeddings. By incorporating CFG during optimization, online refinement of negative text embeddings dynamically enhances alignment. To our knowledge, we are the first to bridge human preference alignment with CFG theory under score distillation framework. Experiments demonstrate the superiority of PSD in aesthetic metrics, seamless integration with diverse pipelines, and strong extensibility.", "date": "2026-03-02", "categories": []} {"id": "2503.21449", "title": "Towards Generating Realistic 3D Semantic Training Data for Autonomous Driving", "abstract": "Semantic scene understanding is crucial for robotics and computer vision applications. In autonomous driving, 3D semantic segmentation plays an important role for enabling safe navigation. Despite significant advances in the field, the complexity of collecting and annotating 3D data is a bottleneck in this developments. To overcome that data annotation limitation, synthetic simulated data has been used to generate annotated data on demand. There is still, however, a domain gap between real and simulated data. More recently, diffusion models have been in the spotlight, enabling close-to-real data synthesis. Those generative models have been recently applied to the 3D data domain for generating scene-scale data with semantic annotations. Still, those methods either rely on image projection or decoupled models trained with different resolutions in a coarse-to-fine manner. Such intermediary representations impact the generated data quality due to errors added in those transformations. In this work, we propose a novel approach able to generate 3D semantic scene-scale data without relying on any projection or decoupled trained multi-resolution models, achieving more realistic semantic scene data generation compared to previous state-of-the-art methods. Besides improving 3D semantic scene-scale data synthesis, we thoroughly evaluate the use of the synthetic scene samples as labeled data to train a semantic segmentation network. In our experiments, we show that using the synthetic annotated data generated by our method as training data together with the real semantic segmentation labels, leads to an improvement in the semantic segmentation model performance. Our results show the potential of generated scene-scale point clouds to generate more training data to extend existing datasets, reducing the data annotation effort. Our code is available at https://github.com/PRBonn/3DiSS.", "date": "2025-03-27", "categories": []} {"id": "2602.05362", "title": "Imagine a City: CityGenAgent for Procedural 3D City Generation", "abstract": "The automated generation of interactive 3D cities is a critical challenge with broad applications in autonomous driving, virtual reality, and embodied intelligence. While recent advances in generative models and procedural techniques have improved the realism of city generation, existing methods often struggle with high-fidelity asset creation, controllability, and manipulation. In this work, we introduce CityGenAgent, a natural language-driven framework for hierarchical procedural generation of high-quality 3D cities. Our approach decomposes city generation into two interpretable components, Block Program and Building Program. To ensure structural correctness and semantic alignment, we adopt a two-stage learning strategy: (1) Supervised Fine-Tuning (SFT). We train BlockGen and BuildingGen to generate valid programs that adhere to schema constraints, including non-self-intersecting polygons and complete fields; (2) Reinforcement Learning (RL). We design Spatial Alignment Reward to enhance spatial reasoning ability and Visual Consistency Reward to bridge the gap between textual descriptions and the visual modality. Benefiting from the programs and the models' generalization, CityGenAgent supports natural language editing and manipulation. Comprehensive evaluations demonstrate superior semantic alignment, visual quality, and controllability compared to existing methods, establishing a robust foundation for scalable 3D city generation.", "date": "2026-02-05", "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": "2602.22596", "title": "BetterScene: 3D Scene Synthesis with Representation-Aligned Generative Model", "abstract": "We present BetterScene, an approach to enhance novel view synthesis (NVS) quality for diverse real-world scenes using extremely sparse, unconstrained photos. BetterScene leverages the production-ready Stable Video Diffusion (SVD) model pretrained on billions of frames as a strong backbone, aiming to mitigate artifacts and recover view-consistent details at inference time. Conventional methods have developed similar diffusion-based solutions to address these challenges of novel view synthesis. Despite significant improvements, these methods typically rely on off-the-shelf pretrained diffusion priors and fine-tune only the UNet module while keeping other components frozen, which still leads to inconsistent details and artifacts even when incorporating geometry-aware regularizations like depth or semantic conditions. To address this, we investigate the latent space of the diffusion model and introduce two components: (1) temporal equivariance regularization and (2) vision foundation model-aligned representation, both applied to the variational autoencoder (VAE) module within the SVD pipeline. BetterScene integrates a feed-forward 3D Gaussian Splatting (3DGS) model to render features as inputs for the SVD enhancer and generate continuous, artifact-free, consistent novel views. We evaluate on the challenging DL3DV-10K dataset and demonstrate superior performance compared to state-of-the-art methods.", "date": "2026-02-26", "categories": []} {"id": "2602.22785", "title": "SceneTransporter: Optimal Transport-Guided Compositional Latent Diffusion for Single-Image Structured 3D Scene Generation", "abstract": "We introduce SceneTransporter, an end-to-end framework for structured 3D scene generation from a single image. While existing methods generate part-level 3D objects, they often fail to organize these parts into distinct instances in open-world scenes. Through a debiased clustering probe, we reveal a critical insight: this failure stems from the lack of structural constraints within the model's internal assignment mechanism. Based on this finding, we reframe the task of structured 3D scene generation as a global correlation assignment problem. To solve this, SceneTransporter formulates and solves an entropic Optimal Transport (OT) objective within the denoising loop of the compositional DiT model. This formulation imposes two powerful structural constraints. First, the resulting transport plan gates cross-attention to enforce an exclusive, one-to-one routing of image patches to part-level 3D latents, preventing entanglement. Second, the competitive nature of the transport encourages the grouping of similar patches, a process that is further regularized by an edge-based cost, to form coherent objects and prevent fragmentation. Extensive experiments show that SceneTransporter outperforms existing methods on open-world scene generation, significantly improving instance-level coherence and geometric fidelity. Code and models will be publicly available at https://2019epwl.github.io/SceneTransporter/.", "date": "2026-02-26", "categories": []} {"id": "2408.09126", "title": "DreamBarbie: Text to Barbie-Style 3D Avatars", "abstract": "To integrate digital humans into everyday life, there is a strong demand for generating high-quality, fine-grained disentangled 3D avatars that support expressive animation and simulation capabilities, ideally from low-cost textual inputs. Although text-driven 3D avatar generation has made significant progress by leveraging 2D generative priors, existing methods still struggle to fulfill all these requirements simultaneously. To address this challenge, we propose DreamBarbie, a novel text-driven framework for generating animatable 3D avatars with separable shoes, accessories, and simulation-ready garments, truly capturing the iconic ``Barbie doll'' aesthetic. The core of our framework lies in an expressive 3D representation combined with appropriate modeling constraints. Unlike prior methods, we use G-Shell to uniformly model watertight components (e.g., bodies, shoes) and non-watertight garments. By reformulating boundaries as Euclidean field intersections instead of manifold geodesics, we propose an SDF-based initialization and a hole regularization loss that together achieve a 100x speedup and stable open topology without image input. These disentangled 3D representations are then optimized by specialized expert diffusion models tailored to each domain, ensuring high-fidelity outputs. To mitigate geometric artifacts and texture conflicts when combining different expert models, we further propose several effective geometric losses and strategies. Extensive experiments demonstrate that DreamBarbie outperforms existing methods in both dressed human and outfit generation. Our framework further enables diverse applications, including apparel combination, editing, expressive animation, and physical simulation. Project page: https://xiaokunsun.github.io/DreamBarbie.github.io/.", "date": "2024-08-17", "categories": []} {"id": "2602.20666", "title": "BoxSplitGen: A Generative Model for 3D Part Bounding Boxes in Varying Granularity", "abstract": "Human creativity follows a perceptual process, moving from abstract ideas to finer details during creation. While 3D generative models have advanced dramatically, models specifically designed to assist human imagination in 3D creation -- particularly for detailing abstractions from coarse to fine -- have not been explored. We propose a framework that enables intuitive and interactive 3D shape generation by iteratively splitting bounding boxes to refine the set of bounding boxes. The main technical components of our framework are two generative models: the box-splitting generative model and the box-to-shape generative model. The first model, named BoxSplitGen, generates a collection of 3D part bounding boxes with varying granularity by iteratively splitting coarse bounding boxes. It utilizes part bounding boxes created through agglomerative merging and learns the reverse of the merging process -- the splitting sequences. The model consists of two main components: the first learns the categorical distribution of the box to be split, and the second learns the distribution of the two new boxes, given the set of boxes and the indication of which box to split. The second model, the box-to-shape generative model, is trained by leveraging the 3D shape priors learned by an existing 3D diffusion model while adapting the model to incorporate bounding box conditioning. In our experiments, we demonstrate that the box-splitting generative model outperforms token prediction models and the inpainting approach with an unconditional diffusion model. Also, we show that our box-to-shape model, based on a state-of-the-art 3D diffusion model, provides superior results compared to a previous model.", "date": "2026-02-24", "categories": []} {"id": "2602.10116", "title": "SAGE: Scalable Agentic 3D Scene Generation for Embodied AI", "abstract": "Real-world data collection for embodied agents remains costly and unsafe, calling for scalable, realistic, and simulator-ready 3D environments. However, existing scene-generation systems often rely on rule-based or task-specific pipelines, yielding artifacts and physically invalid scenes. We present SAGE, an agentic framework that, given a user-specified embodied task (e.g., \"pick up a bowl and place it on the table\"), understands the intent and automatically generates simulation-ready environments at scale. The agent couples multiple generators for layout and object composition with critics that evaluate semantic plausibility, visual realism, and physical stability. Through iterative reasoning and adaptive tool selection, it self-refines the scenes until meeting user intent and physical validity. The resulting environments are realistic, diverse, and directly deployable in modern simulators for policy training. Policies trained purely on this data exhibit clear scaling trends and generalize to unseen objects and layouts, demonstrating the promise of simulation-driven scaling for embodied AI. Code, demos, and the SAGE-10k dataset can be found on the project page here: https://research.nvidia.com/labs/dir/sage/.", "date": "2026-02-10", "categories": []} {"id": "2602.18830", "title": "Spatial-Temporal State Propagation Autoregressive Model for 4D Object Generation", "abstract": "Generating high-quality 4D objects with spatial-temporal consistency is still formidable. Existing diffusion-based methods often struggle with spatial-temporal inconsistency, as they fail to leverage outputs from all previous timesteps to guide the generation at the current timestep. Therefore, we propose a Spatial-Temporal State Propagation AutoRegressive Model (4DSTAR), which generates 4D objects maintaining temporal-spatial consistency. 4DSTAR formulates the generation problem as the prediction of tokens that represent the 4D object. It consists of two key components: (1) The dynamic spatial-temporal state propagation autoregressive model (STAR) is proposed, which achieves spatial-temporal consistent generation. Unlike standard autoregressive models, STAR divides prediction tokens into groups based on timesteps. It models long-term dependencies by propagating spatial-temporal states from previous groups and utilizes these dependencies to guide generation at the next timestep. To this end, a spatial-temporal container is proposed, which dynamically updating the effective spatial-temporal state features from all historical groups, then updated features serve as conditional features to guide the prediction of the next token group. (2) The 4D VQ-VAE is proposed, which implicitly encodes the 4D structure into discrete space and decodes the discrete tokens predicted by STAR into temporally coherent dynamic 3D Gaussians. Experiments demonstrate that 4DSTAR generates spatial-temporal consistent 4D objects, and achieves performance competitive with diffusion models.", "date": "2026-02-21", "categories": []} {"id": "2602.19089", "title": "Ani3DHuman: Photorealistic 3D Human Animation with Self-guided Stochastic Sampling", "abstract": "Current 3D human animation methods struggle to achieve photorealism: kinematics-based approaches lack non-rigid dynamics (e.g., clothing dynamics), while methods that leverage video diffusion priors can synthesize non-rigid motion but suffer from quality artifacts and identity loss. To overcome these limitations, we present Ani3DHuman, a framework that marries kinematics-based animation with video diffusion priors. We first introduce a layered motion representation that disentangles rigid motion from residual non-rigid motion. Rigid motion is generated by a kinematic method, which then produces a coarse rendering to guide the video diffusion model in generating video sequences that restore the residual non-rigid motion. However, this restoration task, based on diffusion sampling, is highly challenging, as the initial renderings are out-of-distribution, causing standard deterministic ODE samplers to fail. Therefore, we propose a novel self-guided stochastic sampling method, which effectively addresses the out-of-distribution problem by combining stochastic sampling (for photorealistic quality) with self-guidance (for identity fidelity). These restored videos provide high-quality supervision, enabling the optimization of the residual non-rigid motion field. Extensive experiments demonstrate that \\MethodName can generate photorealistic 3D human animation, outperforming existing methods. Code is available in https://github.com/qiisun/ani3dhuman.", "date": "2026-02-22", "categories": []} {"id": "2602.19542", "title": "Vinedresser3D: Agentic Text-guided 3D Editing", "abstract": "Text-guided 3D editing aims to modify existing 3D assets using natural-language instructions. Current methods struggle to jointly understand complex prompts, automatically localize edits in 3D, and preserve unedited content. We introduce Vinedresser3D, an agentic framework for high-quality text-guided 3D editing that operates directly in the latent space of a native 3D generative model. Given a 3D asset and an editing prompt, Vinedresser3D uses a multimodal large language model to infer rich descriptions of the original asset, identify the edit region and edit type (addition, modification, deletion), and generate decomposed structural and appearance-level text guidance. The agent then selects an informative view and applies an image editing model to obtain visual guidance. Finally, an inversion-based rectified-flow inpainting pipeline with an interleaved sampling module performs editing in the 3D latent space, enforcing prompt alignment while maintaining 3D coherence and unedited regions. Experiments on diverse 3D edits demonstrate that Vinedresser3D outperforms prior baselines in both automatic metrics and human preference studies, while enabling precise, coherent, and mask-free 3D editing.", "date": "2026-02-23", "categories": []} {"id": "2508.11603", "title": "CoreEditor: Correspondence-constrained Diffusion for Consistent 3D Editing", "abstract": "Text-driven 3D editing seeks to modify 3D scenes according to textual descriptions, and most existing approaches tackle this by adapting pre-trained 2D image editors to multi-view inputs. However, without explicit control over multi-view information exchange, they often fail to maintain cross-view consistency, leading to insufficient edits and blurry details. We introduce CoreEditor, a novel framework for consistent text-to-3D editing. The key innovation is a correspondence-constrained attention mechanism that enforces precise interactions between pixels expected to remain consistent throughout the diffusion denoising process. Beyond relying solely on geometric alignment, we further incorporate semantic similarity estimated during denoising, enabling more reliable correspondence modeling and robust multi-view editing. In addition, we design a selective editing pipeline that allows users to choose preferred results from multiple candidates, offering greater flexibility and user control. Extensive experiments show that CoreEditor produces high-quality, 3D-consistent edits with sharper details, significantly outperforming prior methods.", "date": "2025-08-15", "categories": []} {"id": "2602.17033", "title": "PartRAG: Retrieval-Augmented Part-Level 3D Generation and Editing", "abstract": "Single-image 3D generation with part-level structure remains challenging: learned priors struggle to cover the long tail of part geometries and maintain multi-view consistency, and existing systems provide limited support for precise, localized edits. We present PartRAG, a retrieval-augmented framework that integrates an external part database with a diffusion transformer to couple generation with an editable representation. To overcome the first challenge, we introduce a Hierarchical Contrastive Retrieval module that aligns dense image patches with 3D part latents at both part and object granularity, retrieving from a curated bank of 1,236 part-annotated assets to inject diverse, physically plausible exemplars into denoising. To overcome the second challenge, we add a masked, part-level editor that operates in a shared canonical space, enabling swaps, attribute refinements, and compositional updates without regenerating the whole object while preserving non-target parts and multi-view consistency. PartRAG achieves competitive results on Objaverse, ShapeNet, and ABO-reducing Chamfer Distance from 0.1726 to 0.1528 and raising F-Score from 0.7472 to 0.844 on Objaverse-with inference of 38s and interactive edits in 5-8s. Qualitatively, PartRAG produces sharper part boundaries, better thin-structure fidelity, and robust behavior on articulated objects. Code: https://github.com/AIGeeksGroup/PartRAG. Website: https://aigeeksgroup.github.io/PartRAG.", "date": "2026-02-19", "categories": []} {"id": "2506.20367", "title": "DreamAnywhere: Object-Centric Panoramic 3D Scene Generation", "abstract": "Recent advances in text-to-3D scene generation have demonstrated significant potential to transform content creation across multiple industries. Although the research community has made impressive progress in addressing the challenges of this complex task, existing methods often generate environments that are only front-facing, lack visual fidelity, exhibit limited scene understanding, and are typically fine-tuned for either indoor or outdoor settings. In this work, we address these issues and propose DreamAnywhere, a modular system for the fast generation and prototyping of 3D scenes. Our system synthesizes a 360{\\deg} panoramic image from text, decomposes it into background and objects, constructs a complete 3D representation through hybrid inpainting, and lifts object masks to detailed 3D objects that are placed in the virtual environment. DreamAnywhere supports immersive navigation and intuitive object-level editing, making it ideal for scene exploration, visual mock-ups, and rapid prototyping -- all with minimal manual modeling. These features make our system particularly suitable for low-budget movie production, enabling quick iteration on scene layout and visual tone without the overhead of traditional 3D workflows. Our modular pipeline is highly customizable as it allows components to be replaced independently. Compared to current state-of-the-art text and image-based 3D scene generation approaches, DreamAnywhere shows significant improvements in coherence in novel view synthesis and achieves competitive image quality, demonstrating its effectiveness across diverse and challenging scenarios. A comprehensive user study demonstrates a clear preference for our method over existing approaches, validating both its technical robustness and practical usefulness.", "date": "2025-06-25", "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": "2412.11586", "title": "StrandHead: Text to Hair-Disentangled 3D Head Avatars Using Human-Centric Priors", "abstract": "While haircut indicates distinct personality, existing avatar generation methods fail to model practical hair due to the data limitation or entangled representation. We propose StrandHead, a novel text-driven method capable of generating 3D hair strands and disentangled head avatars with strand-level attributes. Instead of using large-scale hair-text paired data for supervision, we demonstrate that realistic hair strands can be generated from prompts by distilling 2D generative models pre-trained on human mesh data. To this end, we propose a meshing approach guided by strand geometry to guarantee the gradient flow from the distillation objective to the neural strand representation. The optimization is then regularized by statistically significant haircut features, leading to stable updating of strands against unreasonable drifting. These employed 2D/3D human-centric priors contribute to text-aligned and realistic 3D strand generation. Extensive experiments show that StrandHead achieves the state-of-the-art performance on text to strand generation and disentangled 3D head avatar modeling. The generated 3D hair can be applied on avatars for strand-level editing, as well as implemented in the graphics engine for physical simulation or other applications. Project page: https://xiaokunsun.github.io/StrandHead.github.io/.", "date": "2024-12-16", "categories": []} {"id": "2602.09713", "title": "Stroke3D: Lifting 2D strokes into rigged 3D model via latent diffusion models", "abstract": "Rigged 3D assets are fundamental to 3D deformation and animation. However, existing 3D generation methods face challenges in generating animatable geometry, while rigging techniques lack fine-grained structural control over skeleton creation. To address these limitations, we introduce Stroke3D, a novel framework that directly generates rigged meshes from user inputs: 2D drawn strokes and a descriptive text prompt. Our approach pioneers a two-stage pipeline that separates the generation into: 1) Controllable Skeleton Generation, we employ the Skeletal Graph VAE (Sk-VAE) to encode the skeleton's graph structure into a latent space, where the Skeletal Graph DiT (Sk-DiT) generates a skeletal embedding. The generation process is conditioned on both the text for semantics and the 2D strokes for explicit structural control, with the VAE's decoder reconstructing the final high-quality 3D skeleton; and 2) Enhanced Mesh Synthesis via TextuRig and SKA-DPO, where we then synthesize a textured mesh conditioned on the generated skeleton. For this stage, we first enhance an existing skeleton-to-mesh model by augmenting its training data with TextuRig: a dataset of textured and rigged meshes with captions, curated from Objaverse-XL. Additionally, we employ a preference optimization strategy, SKA-DPO, guided by a skeleton-mesh alignment score, to further improve geometric fidelity. Together, our framework enables a more intuitive workflow for creating ready to animate 3D content. To the best of our knowledge, our work is the first to generate rigged 3D meshes conditioned on user-drawn 2D strokes. Extensive experiments demonstrate that Stroke3D produces plausible skeletons and high-quality meshes.", "date": "2026-02-10", "categories": []} {"id": "2602.12157", "title": "TexSpot: 3D Texture Enhancement with Spatially-uniform Point Latent Representation", "abstract": "High-quality 3D texture generation remains a fundamental challenge due to the view-inconsistency inherent in current mainstream multi-view diffusion pipelines. Existing representations either rely on UV maps, which suffer from distortion during unwrapping, or point-based methods, which tightly couple texture fidelity to geometric density that limits high-resolution texture generation. To address these limitations, we introduce TexSpot, a diffusion-based texture enhancement framework. At its core is Texlet, a novel 3D texture representation that merges the geometric expressiveness of point-based 3D textures with the compactness of UV-based representation. Each Texlet latent vector encodes a local texture patch via a 2D encoder and is further aggregated using a 3D encoder to incorporate global shape context. A cascaded 3D-to-2D decoder reconstructs high-quality texture patches, enabling the Texlet space learning. Leveraging this representation, we train a diffusion transformer conditioned on Texlets to refine and enhance textures produced by multi-view diffusion methods. Extensive experiments demonstrate that TexSpot significantly improves visual fidelity, geometric consistency, and robustness over existing state-of-the-art 3D texture generation and enhancement approaches. Project page: https://texlet-arch.github.io/TexSpot-page.", "date": "2026-02-12", "categories": []} {"id": "2602.13818", "title": "VAR-3D: View-aware Auto-Regressive Model for Text-to-3D Generation via a 3D Tokenizer", "abstract": "Recent advances in auto-regressive transformers have achieved remarkable success in generative modeling. However, text-to-3D generation remains challenging, primarily due to bottlenecks in learning discrete 3D representations. Specifically, existing approaches often suffer from information loss during encoding, causing representational distortion before the quantization process. This effect is further amplified by vector quantization, ultimately degrading the geometric coherence of text-conditioned 3D shapes. Moreover, the conventional two-stage training paradigm induces an objective mismatch between reconstruction and text-conditioned auto-regressive generation. To address these issues, we propose View-aware Auto-Regressive 3D (VAR-3D), which intergrates a view-aware 3D Vector Quantized-Variational AutoEncoder (VQ-VAE) to convert the complex geometric structure of 3D models into discrete tokens. Additionally, we introduce a rendering-supervised training strategy that couples discrete token prediction with visual reconstruction, encouraging the generative process to better preserve visual fidelity and structural consistency relative to the input text. Experiments demonstrate that VAR-3D significantly outperforms existing methods in both generation quality and text-3D alignment.", "date": "2026-02-14", "categories": []} {"id": "2602.14965", "title": "PAct: Part-Decomposed Single-View Articulated Object Generation", "abstract": "Articulated objects are central to interactive 3D applications, including embodied AI, robotics, and VR/AR, where functional part decomposition and kinematic motion are essential. Yet producing high-fidelity articulated assets remains difficult to scale because it requires reliable part decomposition and kinematic rigging. Existing approaches largely fall into two paradigms: optimization-based reconstruction or distillation, which can be accurate but often takes tens of minutes to hours per instance, and inference-time methods that rely on template or part retrieval, producing plausible results that may not match the specific structure and appearance in the input observation. We introduce a part-centric generative framework for articulated object creation that synthesizes part geometry, composition, and articulation under explicit part-aware conditioning. Our representation models an object as a set of movable parts, each encoded by latent tokens augmented with part identity and articulation cues. Conditioned on a single image, the model generates articulated 3D assets that preserve instance-level correspondence while maintaining valid part structure and motion. The resulting approach avoids per-instance optimization, enables fast feed-forward inference, and supports controllable assembly and articulation, which are important for embodied interaction. Experiments on common articulated categories (e.g., drawers and doors) show improved input consistency, part accuracy, and articulation plausibility over optimization-based and retrieval-driven baselines, while substantially reducing inference time.", "date": "2026-02-16", "categories": []} {"id": "2601.07692", "title": "R3DPA: Leveraging 3D Representation Alignment and RGB Pretrained Priors for LiDAR Scene Generation", "abstract": "LiDAR scene synthesis is an emerging solution to scarcity in 3D data for robotic tasks such as autonomous driving. Recent approaches employ diffusion or flow matching models to generate realistic scenes, but 3D data remains limited compared to RGB datasets with millions of samples. We introduce R3DPA, the first LiDAR scene generation method to unlock image-pretrained priors for LiDAR point clouds, and leverage self-supervised 3D representations for state-of-the-art results. Specifically, we (i) align intermediate features of our generative model with self-supervised 3D features, which substantially improves generation quality; (ii) transfer knowledge from large-scale image-pretrained generative models to LiDAR generation, mitigating limited LiDAR datasets; and (iii) enable point cloud control at inference for object inpainting and scene mixing with solely an unconditional model. On the KITTI-360 benchmark R3DPA achieves state of the art performance. Code and pretrained models are available at https://github.com/valeoai/R3DPA.", "date": "2026-01-12", "categories": []} {"id": "2602.11693", "title": "OMEGA-Avatar: One-shot Modeling of 360{\\deg} Gaussian Avatars", "abstract": "Creating high-fidelity, animatable 3D avatars from a single image remains a formidable challenge. We identified three desirable attributes of avatar generation: 1) the method should be feed-forward, 2) model a 360{\\deg} full-head, and 3) should be animation-ready. However, current work addresses only two of the three points simultaneously. To address these limitations, we propose OMEGA-Avatar, the first feed-forward framework that simultaneously generates a generalizable, 360{\\deg}-complete, and animatable 3D Gaussian head from a single image. Starting from a feed-forward and animatable framework, we address the 360{\\deg} full-head avatar generation problem with two novel components. First, to overcome poor hair modeling in full-head avatar generation, we introduce a semantic-aware mesh deformation module that integrates multi-view normals to optimize a FLAME head with hair while preserving its topology structure. Second, to enable effective feed-forward decoding of full-head features, we propose a multi-view feature splatting module that constructs a shared canonical UV representation from features across multiple views through differentiable bilinear splatting, hierarchical UV mapping, and visibility-aware fusion. This approach preserves both global structural coherence and local high-frequency details across all viewpoints, ensuring 360{\\deg} consistency without per-instance optimization. Extensive experiments demonstrate that OMEGA-Avatar achieves state-of-the-art performance, significantly outperforming existing baselines in 360{\\deg} full-head completeness while robustly preserving identity across different viewpoints.", "date": "2026-02-12", "categories": []} {"id": "2602.11875", "title": "DiffPlace: Street View Generation via Place-Controllable Diffusion Model Enhancing Place Recognition", "abstract": "Generative models have advanced significantly in realistic image synthesis, with diffusion models excelling in quality and stability. Recent multi-view diffusion models improve 3D-aware street view generation, but they struggle to produce place-aware and background-consistent urban scenes from text, BEV maps, and object bounding boxes. This limits their effectiveness in generating realistic samples for place recognition tasks. To address these challenges, we propose DiffPlace, a novel framework that introduces a place-ID controller to enable place-controllable multi-view image generation. The place-ID controller employs linear projection, perceiver transformer, and contrastive learning to map place-ID embeddings into a fixed CLIP space, allowing the model to synthesize images with consistent background buildings while flexibly modifying foreground objects and weather conditions. Extensive experiments, including quantitative comparisons and augmented training evaluations, demonstrate that DiffPlace outperforms existing methods in both generation quality and training support for visual place recognition. Our results highlight the potential of generative models in enhancing scene-level and place-aware synthesis, providing a valuable approach for improving place recognition in autonomous driving", "date": "2026-02-12", "categories": []} {"id": "2602.12100", "title": "AssetFormer: Modular 3D Assets Generation with Autoregressive Transformer", "abstract": "The digital industry demands high-quality, diverse modular 3D assets, especially for user-generated content~(UGC). In this work, we introduce AssetFormer, an autoregressive Transformer-based model designed to generate modular 3D assets from textual descriptions. Our pilot study leverages real-world modular assets collected from online platforms. AssetFormer tackles the challenge of creating assets composed of primitives that adhere to constrained design parameters for various applications. By innovatively adapting module sequencing and decoding techniques inspired by language models, our approach enhances asset generation quality through autoregressive modeling. Initial results indicate the effectiveness of AssetFormer in streamlining asset creation for professional development and UGC scenarios. This work presents a flexible framework extendable to various types of modular 3D assets, contributing to the broader field of 3D content generation. The code is available at https://github.com/Advocate99/AssetFormer.", "date": "2026-02-12", "categories": []} {"id": "2512.12768", "title": "CoRe3D: Collaborative Reasoning as a Foundation for 3D Intelligence", "abstract": "Recent advances in large multimodal models suggest that explicit reasoning mechanisms play a critical role in improving model reliability, interpretability, and cross-modal alignment. While such reasoning-centric approaches have been proven effective in language and vision tasks, their extension to 3D remains underdeveloped. CoRe3D introduces a unified 3D understanding and generation reasoning framework that jointly operates over semantic and spatial abstractions, enabling high-level intent inferred from language to directly guide low-level 3D content formation. Central to this design is a spatially grounded reasoning representation that decomposes 3D latent space into localized regions, allowing the model to reason over geometry in a compositional and procedural manner. By tightly coupling semantic chain-of-thought inference with structured spatial reasoning, CoRe3D produces 3D outputs that exhibit strong local consistency and faithful alignment with linguistic descriptions.", "date": "2025-12-14", "categories": []} {"id": "2602.09432", "title": "SceneReVis: A Self-Reflective Vision-Grounded Framework for 3D Indoor Scene Synthesis via Multi-turn RL", "abstract": "Current one-pass 3D scene synthesis methods often suffer from spatial hallucinations, such as collisions, due to a lack of deliberative reasoning. To bridge this gap, we introduce SceneReVis, a vision-grounded self-reflection framework that employs an iterative ``diagnose-and-act'' loop to explicitly intercept and resolve spatial conflicts using multi-modal feedback. To support this step-wise paradigm, we construct SceneChain-12k, a large-scale dataset of causal construction trajectories derived through a novel reverse engineering pipeline. We further propose a two-stage training recipe that transitions from Supervised Fine-Tuning to Agentic Reinforcement Learning, evolving the model into an active spatial planner. Extensive experiments demonstrate that SceneReVis achieves state-of-the-art performance in high-fidelity generation and goal-oriented optimization, with robust generalization to long-tail domains.", "date": "2026-02-10", "categories": []} {"id": "2602.07198", "title": "Condition Matters in Full-head 3D GANs", "abstract": "Conditioning is crucial for stable training of full-head 3D GANs. Without any conditioning signal, the model suffers from severe mode collapse, making it impractical to training. However, a series of previous full-head 3D GANs conventionally choose the view angle as the conditioning input, which leads to a bias in the learned 3D full-head space along the conditional view direction. This is evident in the significant differences in generation quality and diversity between the conditional view and non-conditional views of the generated 3D heads, resulting in global incoherence across different head regions. In this work, we propose to use view-invariant semantic feature as the conditioning input, thereby decoupling the generative capability of 3D heads from the viewing direction. To construct a view-invariant semantic condition for each training image, we create a novel synthesized head image dataset. We leverage FLUX.1 Kontext to extend existing high-quality frontal face datasets to a wide range of view angles. The image clip feature extracted from the frontal view is then used as a shared semantic condition across all views in the extended images, ensuring semantic alignment while eliminating directional bias. This also allows supervision from different views of the same subject to be consolidated under a shared semantic condition, which accelerates training and enhances the global coherence of the generated 3D heads. Moreover, as GANs often experience slower improvements in diversity once the generator learns a few modes that successfully fool the discriminator, our semantic conditioning encourages the generator to follow the true semantic distribution, thereby promoting continuous learning and diverse generation. Extensive experiments on full-head synthesis and single-view GAN inversion demonstrate that our method achieves significantly higher fidelity, diversity, and generalizability.", "date": "2026-02-06", "categories": []} {"id": "2602.08046", "title": "Enhanced Mixture 3D CGAN for Completion and Generation of 3D Objects", "abstract": "The generation and completion of 3D objects represent a transformative challenge in computer vision. Generative Adversarial Networks (GANs) have recently demonstrated strong potential in synthesizing realistic visual data. However, they often struggle to capture complex and diverse data distributions, particularly in scenarios involving incomplete inputs or significant missing regions. These challenges arise mainly from the high computational requirements and the difficulty of modeling heterogeneous and structurally intricate data, which restrict their applicability in real-world settings. Mixture of Experts (MoE) models have emerged as a promising solution to these limitations. By dynamically selecting and activating the most relevant expert sub-networks for a given input, MoEs improve both performance and efficiency. In this paper, we investigate the integration of Deep 3D Convolutional GANs (CGANs) with a MoE framework to generate high-quality 3D models and reconstruct incomplete or damaged objects. The proposed architecture incorporates multiple generators, each specialized to capture distinct modalities within the dataset. Furthermore, an auxiliary loss-free dynamic capacity constraint (DCC) mechanism is introduced to guide the selection of categorical generators, ensuring a balance between specialization, training stability, and computational efficiency, which is critical for 3D voxel processing. We evaluated the model's ability to generate and complete shapes with missing regions of varying sizes and compared its performance with state-of-the-art approaches. Both quantitative and qualitative results confirm the effectiveness of the proposed MoE-DCGAN in handling complex 3D data.", "date": "2026-02-08", "categories": []} {"id": "2602.08198", "title": "PEGAsus: 3D Personalization of Geometry and Appearance", "abstract": "We present PEGAsus, a new framework capable of generating Personalized 3D shapes by learning shape concepts at both Geometry and Appearance levels. First, we formulate 3D shape personalization as extracting reusable, category-agnostic geometric and appearance attributes from reference shapes, and composing these attributes with text to generate novel shapes. Second, we design a progressive optimization strategy to learn shape concepts at both the geometry and appearance levels, decoupling the shape concept learning process. Third, we extend our approach to region-wise concept learning, enabling flexible concept extraction, with context-aware and context-free losses. Extensive experimental results show that PEGAsus is able to effectively extract attributes from a wide range of reference shapes and then flexibly compose these concepts with text to synthesize new shapes. This enables fine-grained control over shape generation and supports the creation of diverse, personalized results, even in challenging cross-category scenarios. Both quantitative and qualitative experiments demonstrate that our approach outperforms existing state-of-the-art solutions.", "date": "2026-02-09", "categories": []} {"id": "2408.04567", "title": "Sketch2Scene: Automatic Generation of Interactive 3D Game Scenes from User's Casual Sketches", "abstract": "3D Content Generation is at the heart of many computer graphics applications, including video gaming, film-making, virtual and augmented reality, etc. This paper proposes a novel deep-learning based approach for automatically generating interactive and playable 3D game scenes, all from the user's casual prompts such as a hand-drawn sketch. Sketch-based input offers a natural, and convenient way to convey the user's design intention in the content creation process. To circumvent the data-deficient challenge in learning (i.e. the lack of large training data of 3D scenes), our method leverages a pre-trained 2D denoising diffusion model to generate a 2D image of the scene as the conceptual guidance. In this process, we adopt the isometric projection mode to factor out unknown camera poses while obtaining the scene layout. From the generated isometric image, we use a pre-trained image understanding method to segment the image into meaningful parts, such as off-ground objects, trees, and buildings, and extract the 2D scene layout. These segments and layouts are subsequently fed into a procedural content generation (PCG) engine, such as a 3D video game engine like Unity or Unreal, to create the 3D scene. The resulting 3D scene can be seamlessly integrated into a game development environment and is readily playable. Extensive tests demonstrate that our method can efficiently generate high-quality and interactive 3D game scenes with layouts that closely follow the user's intention.", "date": "2024-08-08", "categories": []} {"id": "2504.15259", "title": "Bringing Diversity from Diffusion Models to Semantic-Guided Face Asset Generation", "abstract": "Digital modeling and reconstruction of human faces serve various applications. However, its availability is often hindered by the requirements of data capturing devices, manual labor, and suitable actors. This situation restricts the diversity, expressiveness, and control over the resulting models. This work aims to demonstrate that a semantically controllable generative network can provide enhanced control over the digital face modeling process. To enhance diversity beyond the limited human faces scanned in a controlled setting, we introduce a novel data generation pipeline that creates a high-quality 3D face database using a pre-trained diffusion model. Our proposed normalization module converts synthesized data from the diffusion model into high-quality scanned data. Using the 44,000 face models we obtained, we further developed an efficient GAN-based generator. This generator accepts semantic attributes as input, and generates geometry and albedo. It also allows continuous post-editing of attributes in the latent space. Our asset refinement component subsequently creates physically-based facial assets. We introduce a comprehensive system designed for creating and editing high-quality face assets. Our proposed model has undergone extensive experiment, comparison and evaluation. We also integrate everything into a web-based interactive tool. We aim to make this tool publicly available with the release of the paper.", "date": "2025-04-21", "categories": []} {"id": "2602.03907", "title": "HY3D-Bench: Generation of 3D Assets", "abstract": "While recent advances in neural representations and generative models have revolutionized 3D content creation, the field remains constrained by significant data processing bottlenecks. To address this, we introduce HY3D-Bench, an open-source ecosystem designed to establish a unified, high-quality foundation for 3D generation. Our contributions are threefold: (1) We curate a library of 250k high-fidelity 3D objects distilled from large-scale repositories, employing a rigorous pipeline to deliver training-ready artifacts, including watertight meshes and multi-view renderings; (2) We introduce structured part-level decomposition, providing the granularity essential for fine-grained perception and controllable editing; and (3) We bridge real-world distribution gaps via a scalable AIGC synthesis pipeline, contributing 125k synthetic assets to enhance diversity in long-tail categories. Validated empirically through the training of Hunyuan3D-2.1-Small, HY3D-Bench democratizes access to robust data resources, aiming to catalyze innovation across 3D perception, robotics, and digital content creation.", "date": "2026-02-03", "categories": []} {"id": "2602.04271", "title": "SkeletonGaussian: Editable 4D Generation through Gaussian Skeletonization", "abstract": "4D generation has made remarkable progress in synthesizing dynamic 3D objects from input text, images, or videos. However, existing methods often represent motion as an implicit deformation field, which limits direct control and editability. To address this issue, we propose SkeletonGaussian, a novel framework for generating editable dynamic 3D Gaussians from monocular video input. Our approach introduces a hierarchical articulated representation that decomposes motion into sparse rigid motion explicitly driven by a skeleton and fine-grained non-rigid motion. Concretely, we extract a robust skeleton and drive rigid motion via linear blend skinning, followed by a hexplane-based refinement for non-rigid deformations, enhancing interpretability and editability. Experimental results demonstrate that SkeletonGaussian surpasses existing methods in generation quality while enabling intuitive motion editing, establishing a new paradigm for editable 4D generation. Project page: https://wusar.github.io/projects/skeletongaussian/", "date": "2026-02-04", "categories": []} {"id": "2602.02974", "title": "SceneLinker: Compositional 3D Scene Generation via Semantic Scene Graph from RGB Sequences", "abstract": "We introduce SceneLinker, a novel framework that generates compositional 3D scenes via semantic scene graph from RGB sequences. To adaptively experience Mixed Reality (MR) content based on each user's space, it is essential to generate a 3D scene that reflects the real-world layout by compactly capturing the semantic cues of the surroundings. Prior works struggled to fully capture the contextual relationship between objects or mainly focused on synthesizing diverse shapes, making it challenging to generate 3D scenes aligned with object arrangements. We address these challenges by designing a graph network with cross-check feature attention for scene graph prediction and constructing a graph-variational autoencoder (graph-VAE), which consists of a joint shape and layout block for 3D scene generation. Experiments on the 3RScan/3DSSG and SG-FRONT datasets demonstrate that our approach outperforms state-of-the-art methods in both quantitative and qualitative evaluations, even in complex indoor environments and under challenging scene graph constraints. Our work enables users to generate consistent 3D spaces from their physical environments via scene graphs, allowing them to create spatial MR content. Project page is https://scenelinker2026.github.io.", "date": "2026-02-03", "categories": []} {"id": "2602.03533", "title": "PnP-U3D: Plug-and-Play 3D Framework Bridging Autoregression and Diffusion for Unified Understanding and Generation", "abstract": "The rapid progress of large multimodal models has inspired efforts toward unified frameworks that couple understanding and generation. While such paradigms have shown remarkable success in 2D, extending them to 3D remains largely underexplored. Existing attempts to unify 3D tasks under a single autoregressive (AR) paradigm lead to significant performance degradation due to forced signal quantization and prohibitive training cost. Our key insight is that the essential challenge lies not in enforcing a unified autoregressive paradigm, but in enabling effective information interaction between generation and understanding while minimally compromising their inherent capabilities and leveraging pretrained models to reduce training cost. Guided by this perspective, we present the first unified framework for 3D understanding and generation that combines autoregression with diffusion. Specifically, we adopt an autoregressive next-token prediction paradigm for 3D understanding, and a continuous diffusion paradigm for 3D generation. A lightweight transformer bridges the feature space of large language models and the conditional space of 3D diffusion models, enabling effective cross-modal information exchange while preserving the priors learned by standalone models. Extensive experiments demonstrate that our framework achieves state-of-the-art performance across diverse 3D understanding and generation benchmarks, while also excelling in 3D editing tasks. These results highlight the potential of unified AR+diffusion models as a promising direction for building more general-purpose 3D intelligence.", "date": "2026-02-03", "categories": []} {"id": "2403.19652", "title": "InterDreamer: Zero-Shot Text to 3D Dynamic Human-Object Interaction", "abstract": "Text-conditioned human motion generation has experienced significant advancements with diffusion models trained on extensive motion capture data and corresponding textual annotations. However, extending such success to 3D dynamic human-object interaction (HOI) generation faces notable challenges, primarily due to the lack of large-scale interaction data and comprehensive descriptions that align with these interactions. This paper takes the initiative and showcases the potential of generating human-object interactions without direct training on text-interaction pair data. Our key insight in achieving this is that interaction semantics and dynamics can be decoupled. Being unable to learn interaction semantics through supervised training, we instead leverage pre-trained large models, synergizing knowledge from a large language model and a text-to-motion model. While such knowledge offers high-level control over interaction semantics, it cannot grasp the intricacies of low-level interaction dynamics. To overcome this issue, we further introduce a world model designed to comprehend simple physics, modeling how human actions influence object motion. By integrating these components, our novel framework, InterDreamer, is able to generate text-aligned 3D HOI sequences in a zero-shot manner. We apply InterDreamer to the BEHAVE and CHAIRS datasets, and our comprehensive experimental analysis demonstrates its capability to generate realistic and coherent interaction sequences that seamlessly align with the text directives.", "date": "2024-03-28", "categories": []} {"id": "2602.00463", "title": "PSGS: Text-driven Panorama Sliding Scene Generation via Gaussian Splatting", "abstract": "Generating realistic 3D scenes from text is crucial for immersive applications like VR, AR, and gaming. While text-driven approaches promise efficiency, existing methods suffer from limited 3D-text data and inconsistent multi-view stitching, resulting in overly simplistic scenes. To address this, we propose PSGS, a two-stage framework for high-fidelity panoramic scene generation. First, a novel two-layer optimization architecture generates semantically coherent panoramas: a layout reasoning layer parses text into structured spatial relationships, while a self-optimization layer refines visual details via iterative MLLM feedback. Second, our panorama sliding mechanism initializes globally consistent 3D Gaussian Splatting point clouds by strategically sampling overlapping perspectives. By incorporating depth and semantic coherence losses during training, we greatly improve the quality and detail fidelity of rendered scenes. Our experiments demonstrate that PSGS outperforms existing methods in panorama generation and produces more appealing 3D scenes, offering a robust solution for scalable immersive content creation.", "date": "2026-01-31", "categories": []} {"id": "2601.13837", "title": "FastGHA: Generalized Few-Shot 3D Gaussian Head Avatars with Real-Time Animation", "abstract": "Despite recent progress in 3D Gaussian-based head avatar modeling, efficiently generating high fidelity avatars remains a challenge. Current methods typically rely on extensive multi-view capture setups or monocular videos with per-identity optimization during inference, limiting their scalability and ease of use on unseen subjects. To overcome these efficiency drawbacks, we propose FastGHA, a feed-forward method to generate high-quality Gaussian head avatars from only a few input images while supporting real-time animation. Our approach directly learns a per-pixel Gaussian representation from the input images, and aggregates multi-view information using a transformer-based encoder that fuses image features from both DINOv3 and Stable Diffusion VAE. For real-time animation, we extend the explicit Gaussian representations with per-Gaussian features and introduce a lightweight MLP-based dynamic network to predict 3D Gaussian deformations from expression codes. Furthermore, to enhance geometric smoothness of the 3D head, we employ point maps from a pre-trained large reconstruction model as geometry supervision. Experiments show that our approach significantly outperforms existing methods in both rendering quality and inference efficiency, while supporting real-time dynamic avatar animation.", "date": "2026-01-20", "categories": []} {"id": "2401.08742", "title": "Efficient4D: Fast Dynamic 3D Object Generation from a Single-view Video", "abstract": "Generating dynamic 3D object from a single-view video is challenging due to the lack of 4D labeled data. An intuitive approach is to extend previous image-to-3D pipelines by transferring off-the-shelf image generation models such as score distillation sampling.However, this approach would be slow and expensive to scale due to the need for back-propagating the information-limited supervision signals through a large pretrained model. To address this, we propose an efficient video-to-4D object generation framework called Efficient4D. It generates high-quality spacetime-consistent images under different camera views, and then uses them as labeled data to directly reconstruct the 4D content through a 4D Gaussian splatting model. Importantly, our method can achieve real-time rendering under continuous camera trajectories. To enable robust reconstruction under sparse views, we introduce inconsistency-aware confidence-weighted loss design, along with a lightly weighted score distillation loss. Extensive experiments on both synthetic and real videos show that Efficient4D offers a remarkable 10-fold increase in speed when compared to prior art alternatives while preserving the quality of novel view synthesis. For example, Efficient4D takes only 10 minutes to model a dynamic object, vs 120 minutes by the previous art model Consistent4D.", "date": "2024-01-16", "categories": []} {"id": "2601.21314", "title": "HiFi-Mesh: High-Fidelity Efficient 3D Mesh Generation via Compact Autoregressive Dependence", "abstract": "High-fidelity 3D meshes can be tokenized into one-dimension (1D) sequences and directly modeled using autoregressive approaches for faces and vertices. However, existing methods suffer from insufficient resource utilization, resulting in slow inference and the ability to handle only small-scale sequences, which severely constrains the expressible structural details. We introduce the Latent Autoregressive Network (LANE), which incorporates compact autoregressive dependencies in the generation process, achieving a $6\\times$ improvement in maximum generatable sequence length compared to existing methods. To further accelerate inference, we propose the Adaptive Computation Graph Reconfiguration (AdaGraph) strategy, which effectively overcomes the efficiency bottleneck of traditional serial inference through spatiotemporal decoupling in the generation process. Experimental validation demonstrates that LANE achieves superior performance across generation speed, structural detail, and geometric consistency, providing an effective solution for high-quality 3D mesh generation.", "date": "2026-01-29", "categories": []} {"id": "2601.19785", "title": "GeoDiff3D: Self-Supervised 3D Scene Generation with Geometry-Constrained 2D Diffusion Guidance", "abstract": "3D scene generation is a core technology for gaming, film/VFX, and VR/AR. Growing demand for rapid iteration, high-fidelity detail, and accessible content creation has further increased interest in this area. Existing methods broadly follow two paradigms - indirect 2D-to-3D reconstruction and direct 3D generation - but both are limited by weak structural modeling and heavy reliance on large-scale ground-truth supervision, often producing structural artifacts, geometric inconsistencies, and degraded high-frequency details in complex scenes. We propose GeoDiff3D, an efficient self-supervised framework that uses coarse geometry as a structural anchor and a geometry-constrained 2D diffusion model to provide texture-rich reference images. Importantly, GeoDiff3D does not require strict multi-view consistency of the diffusion-generated references and remains robust to the resulting noisy, inconsistent guidance. We further introduce voxel-aligned 3D feature aggregation and dual self-supervision to maintain scene coherence and fine details while substantially reducing dependence on labeled data. GeoDiff3D also trains with low computational cost and enables fast, high-quality 3D scene generation. Extensive experiments on challenging scenes show improved generalization and generation quality over existing baselines, offering a practical solution for accessible and efficient 3D scene construction.", "date": "2026-01-27", "categories": []} {"id": "2601.20425", "title": "Quartet of Diffusions: Structure-Aware Point Cloud Generation through Part and Symmetry Guidance", "abstract": "We introduce the Quartet of Diffusions, a structure-aware point cloud generation framework that explicitly models part composition and symmetry. Unlike prior methods that treat shape generation as a holistic process or only support part composition, our approach leverages four coordinated diffusion models to learn distributions of global shape latents, symmetries, semantic parts, and their spatial assembly. This structured pipeline ensures guaranteed symmetry, coherent part placement, and diverse, high-quality outputs. By disentangling the generative process into interpretable components, our method supports fine-grained control over shape attributes, enabling targeted manipulation of individual parts while preserving global consistency. A central global latent further reinforces structural coherence across assembled parts. Our experiments show that the Quartet achieves state-of-the-art performance. To our best knowledge, this is the first 3D point cloud generation framework that fully integrates and enforces both symmetry and part priors throughout the generative process.", "date": "2026-01-28", "categories": []} {"id": "2601.19048", "title": "NuiWorld: Exploring a Scalable Framework for End-to-End Controllable World Generation", "abstract": "World generation is a fundamental capability for applications like video games, simulation, and robotics. However, existing approaches face three main obstacles: controllability, scalability, and efficiency. End-to-end scene generation models have been limited by data scarcity. While object-centric generation approaches rely on fixed resolution representations, degrading fidelity for larger scenes. Training-free approaches, while flexible, are often slow and computationally expensive at inference time. We present NuiWorld, a framework that attempts to address these challenges. To overcome data scarcity, we propose a generative bootstrapping strategy that starts from a few input images. Leveraging recent 3D reconstruction and expandable scene generation techniques, we synthesize scenes of varying sizes and layouts, producing enough data to train an end-to-end model. Furthermore, our framework enables controllability through pseudo sketch labels, and demonstrates a degree of generalization to previously unseen sketches. Our approach represents scenes as a collection of variable scene chunks, which are compressed into a flattened vector-set representation. This significantly reduces the token length for large scenes, enabling consistent geometric fidelity across scenes sizes while improving training and inference efficiency.", "date": "2026-01-27", "categories": []} {"id": "2601.02103", "title": "HeadLighter: Disentangling Illumination in Generative 3D Gaussian Heads via Lightstage Captures", "abstract": "Recent 3D-aware head generative models based on 3D Gaussian Splatting achieve real-time, photorealistic and view-consistent head synthesis. However, a fundamental limitation persists: the deep entanglement of illumination and intrinsic appearance prevents controllable relighting. Existing disentanglement methods rely on strong assumptions to enable weakly supervised learning, which restricts their capacity for complex illumination. To address this challenge, we introduce HeadLighter, a novel supervised framework that learns a physically plausible decomposition of appearance and illumination in head generative models. Specifically, we design a dual-branch architecture that separately models lighting-invariant head attributes and physically grounded rendering components. A progressive disentanglement training is employed to gradually inject head appearance priors into the generative architecture, supervised by multi-view images captured under controlled light conditions with a light stage setup. We further introduce a distillation strategy to generate high-quality normals for realistic rendering. Experiments demonstrate that our method preserves high-quality generation and real-time rendering, while simultaneously supporting explicit lighting and viewpoint editing. We will publicly release our code and dataset.", "date": "2026-01-05", "categories": []} {"id": "2601.16532", "title": "AnchoredDream: Zero-Shot 360{\\deg} Indoor Scene Generation from a Single View via Geometric Grounding", "abstract": "Single-view indoor scene generation plays a crucial role in a range of real-world applications. However, generating a complete 360{\\deg} scene from a single image remains a highly ill-posed and challenging problem. Recent approaches have made progress by leveraging diffusion models and depth estimation networks, yet they still struggle to maintain appearance consistency and geometric plausibility under large viewpoint changes, limiting their effectiveness in full-scene generation. To address this, we propose AnchoredDream, a novel zero-shot pipeline that anchors 360{\\deg} scene generation on high-fidelity geometry via an appearance-geometry mutual boosting mechanism. Given a single-view image, our method first performs appearance-guided geometry generation to construct a reliable 3D scene layout. Then, we progressively generate the complete scene through a series of modules: warp-and-inpaint, warp-and-refine, post-optimization, and a novel Grouting Block, which ensures seamless transitions between the input view and generated regions. Extensive experiments demonstrate that AnchoredDream outperforms existing methods by a large margin in both appearance consistency and geometric plausibility--all in a zero-shot manner. Our results highlight the potential of geometric grounding for high-quality, zero-shot single-view scene generation.", "date": "2026-01-23", "categories": []} {"id": "2601.17733", "title": "Flatten The Complex: Joint B-Rep Generation via Compositional $k$-Cell Particles", "abstract": "Boundary Representation (B-Rep) is the widely adopted standard\n in Computer-Aided Design (CAD) and manufacturing. However, generative modeling of B-Reps remains a formidable challenge due to their inherent heterogeneity as geometric cell complexes, which entangles topology with geometry across cells of varying orders (i.e., $k$-cells such as vertices, edges, faces). Previous methods typically rely on cascaded sequences to handle this hierarchy, which fails to fully exploit the geometric relationships between cells, such as adjacency and sharing, limiting context awareness and error recovery. To fill this gap, we introduce a novel paradigm that reformulates B-Reps into sets of compositional $k$-cell particles. Our approach encodes each topological entity as a composition of particles, where adjacent cells share identical latents at their interfaces, thereby promoting geometric coupling along shared boundaries. By decoupling the rigid hierarchy, our representation unifies vertices, edges, and faces, enabling the joint generation of topology and geometry with global context awareness.\n We synthesize these particle sets using a multi-modal flow matching framework to handle unconditional generation as well as precise conditional tasks, such as 3D reconstruction from single-view or point cloud. Furthermore, the explicit and localized nature of our representation naturally extends to downstream tasks like local in-painting and enables the direct synthesis of non-manifold structures (e.g., wireframes). Extensive experiments demonstrate that our method produces high-fidelity CAD models with superior validity and editability compared to state-of-the-art methods.", "date": "2026-01-25", "categories": []} {"id": "2601.16955", "title": "3D Molecule Generation from Rigid Motifs via SE(3) Flows", "abstract": "Three-dimensional molecular structure generation is typically performed at the level of individual atoms, yet molecular graph generation techniques often consider fragments as their structural units. Building on the advances in frame-based protein structure generation, we extend these fragmentation ideas to 3D, treating general molecules as sets of rigid-body motifs. Utilising this representation, we employ SE(3)-equivariant generative modelling for de novo 3D molecule generation from rigid motifs. In our evaluations, we observe comparable or superior results to state-of-the-art across benchmarks, surpassing it in atom stability on GEOM-Drugs, while yielding a 2x to 10x reduction in generation steps and offering 3.5x compression in molecular representations compared to the standard atom-based methods.", "date": "2026-01-23", "categories": []} {"id": "2505.17590", "title": "CGS-GAN: 3D Consistent Gaussian Splatting GANs for High Resolution Human Head Synthesis", "abstract": "Recently, 3D GANs based on 3D Gaussian splatting have been proposed for high quality synthesis of human heads. However, existing methods stabilize training and enhance rendering quality from steep viewpoints by conditioning the random latent vector on the current camera position. This compromises 3D consistency, as we observe significant identity changes when re-synthesizing the 3D head with each camera shift. Conversely, fixing the camera to a single viewpoint yields high-quality renderings for that perspective but results in poor performance for novel views. Removing view-conditioning typically destabilizes GAN training, often causing the training to collapse. In response to these challenges, we introduce CGS-GAN, a novel 3D Gaussian Splatting GAN framework that enables stable training and high-quality 3D-consistent synthesis of human heads without relying on view-conditioning. To ensure training stability, we introduce a multi-view regularization technique that enhances generator convergence with minimal computational overhead. Additionally, we adapt the conditional loss used in existing 3D Gaussian splatting GANs and propose a generator architecture designed to not only stabilize training but also facilitate efficient rendering and straightforward scaling, enabling output resolutions up to $2048^2$. To evaluate the capabilities of CGS-GAN, we curate a new dataset derived from FFHQ. This dataset enables very high resolutions, focuses on larger portions of the human head, reduces view-dependent artifacts for improved 3D consistency, and excludes images where subjects are obscured by hands or other objects. As a result, our approach achieves very high rendering quality, supported by competitive FID scores, while ensuring consistent 3D scene generation. Check our our project page here: https://fraunhoferhhi.github.io/cgs-gan/", "date": "2025-05-23", "categories": []} {"id": "2601.16192", "title": "360Anything: Geometry-Free Lifting of Images and Videos to 360{\\deg}", "abstract": "Lifting perspective images and videos to 360{\\deg} panoramas enables immersive 3D world generation. Existing approaches often rely on explicit geometric alignment between the perspective and the equirectangular projection (ERP) space. Yet, this requires known camera metadata, obscuring the application to in-the-wild data where such calibration is typically absent or noisy. We propose 360Anything, a geometry-free framework built upon pre-trained diffusion transformers. By treating the perspective input and the panorama target simply as token sequences, 360Anything learns the perspective-to-equirectangular mapping in a purely data-driven way, eliminating the need for camera information. Our approach achieves state-of-the-art performance on both image and video perspective-to-360{\\deg} generation, outperforming prior works that use ground-truth camera information. We also trace the root cause of the seam artifacts at ERP boundaries to zero-padding in the VAE encoder, and introduce Circular Latent Encoding to facilitate seamless generation. Finally, we show competitive results in zero-shot camera FoV and orientation estimation benchmarks, demonstrating 360Anything's deep geometric understanding and broader utility in computer vision tasks. Additional results are available at https://360anything.github.io/.", "date": "2026-01-22", "categories": []} {"id": "2601.15221", "title": "ScenDi: 3D-to-2D Scene Diffusion Cascades for Urban Generation", "abstract": "Recent advancements in 3D object generation using diffusion models have achieved remarkable success, but generating realistic 3D urban scenes remains challenging. Existing methods relying solely on 3D diffusion models tend to suffer a degradation in appearance details, while those utilizing only 2D diffusion models typically compromise camera controllability. To overcome this limitation, we propose ScenDi, a method for urban scene generation that integrates both 3D and 2D diffusion models. We first train a 3D latent diffusion model to generate 3D Gaussians, enabling the rendering of images at a relatively low resolution. To enable controllable synthesis, this 3DGS generation process can be optionally conditioned by specifying inputs such as 3d bounding boxes, road maps, or text prompts. Then, we train a 2D video diffusion model to enhance appearance details conditioned on rendered images from the 3D Gaussians. By leveraging the coarse 3D scene as guidance for 2D video diffusion, ScenDi generates desired scenes based on input conditions and successfully adheres to accurate camera trajectories. Experiments on two challenging real-world datasets, Waymo and KITTI-360, demonstrate the effectiveness of our approach.", "date": "2026-01-21", "categories": []} {"id": "2510.13245", "title": "CymbaDiff: Structured Spatial Diffusion for Sketch-based 3D Semantic Urban Scene Generation", "abstract": "Outdoor 3D semantic scene generation produces realistic and semantically rich environments for applications such as urban simulation and autonomous driving. However, advances in this direction are constrained by the absence of publicly available, well-annotated datasets. We introduce SketchSem3D, the first large-scale benchmark for generating 3D outdoor semantic scenes from abstract freehand sketches and pseudo-labeled annotations of satellite images. SketchSem3D includes two subsets, Sketch-based SemanticKITTI and Sketch-based KITTI-360 (containing LiDAR voxels along with their corresponding sketches and annotated satellite images), to enable standardized, rigorous, and diverse evaluations. We also propose Cylinder Mamba Diffusion (CymbaDiff) that significantly enhances spatial coherence in outdoor 3D scene generation. CymbaDiff imposes structured spatial ordering, explicitly captures cylindrical continuity and vertical hierarchy, and preserves both physical neighborhood relationships and global context within the generated scenes. Extensive experiments on SketchSem3D demonstrate that CymbaDiff achieves superior semantic consistency, spatial realism, and cross-dataset generalization. The code and dataset will be available at https://github.com/Lillian-research-hub/CymbaDiff", "date": "2025-10-15", "categories": []} {"id": "2511.11074", "title": "Evaluating Latent Generative Paradigms for High-Fidelity 3D Shape Completion from a Single Depth Image", "abstract": "While generative models have seen significant adoption across a wide range of data modalities, including 3D data, a consensus on which model is best suited for which task has yet to be reached. Further, conditional information such as text and images to steer the generation process are frequently employed, whereas others, like partial 3D data, have not been thoroughly evaluated. In this work, we compare two of the most promising generative models--Denoising Diffusion Probabilistic Models and Autoregressive Causal Transformers--which we adapt for the tasks of generative shape modeling and completion. We conduct a thorough quantitative evaluation and comparison of both tasks, including a baseline discriminative model and an extensive ablation study. Our results show that (1) the diffusion model with continuous latents outperforms both the discriminative model and the autoregressive approach and delivers state-of-the-art performance on multi-modal shape completion from a single, noisy depth image under realistic conditions and (2) when compared on the same discrete latent space, the autoregressive model can match or exceed diffusion performance on these tasks.", "date": "2025-11-14", "categories": []} {"id": "2601.12234", "title": "Proc3D: Procedural 3D Generation and Parametric Editing of 3D Shapes with Large Language Models", "abstract": "Generating 3D models has traditionally been a complex task requiring specialized expertise. While recent advances in generative AI have sought to automate this process, existing methods produce non-editable representation, such as meshes or point clouds, limiting their adaptability for iterative design. In this paper, we introduce Proc3D, a system designed to generate editable 3D models while enabling real-time modifications. At its core, Proc3D introduces procedural compact graph (PCG), a graph representation of 3D models, that encodes the algorithmic rules and structures necessary for generating the model. This representation exposes key parameters, allowing intuitive manual adjustments via sliders and checkboxes, as well as real-time, automated modifications through natural language prompts using Large Language Models (LLMs). We demonstrate Proc3D's capabilities using two generative approaches: GPT-4o with in-context learning (ICL) and a fine-tuned LLAMA-3 model. Experimental results show that Proc3D outperforms existing methods in editing efficiency, achieving more than 400x speedup over conventional approaches that require full regeneration for each modification. Additionally, Proc3D improves ULIP scores by 28%, a metric that evaluates the alignment between generated 3D models and text prompts. By enabling text-aligned 3D model generation along with precise, real-time parametric edits, Proc3D facilitates highly accurate text-based image editing applications.", "date": "2026-01-18", "categories": []} {"id": "2601.12391", "title": "Class-Partitioned VQ-VAE and Latent Flow Matching for Point Cloud Scene Generation", "abstract": "Most 3D scene generation methods are limited to only generating object bounding box parameters while newer diffusion methods also generate class labels and latent features. Using object size or latent feature, they then retrieve objects from a predefined database. For complex scenes of varied, multi-categorical objects, diffusion-based latents cannot be effectively decoded by current autoencoders into the correct point cloud objects which agree with target classes. We introduce a Class-Partitioned Vector Quantized Variational Autoencoder (CPVQ-VAE) that is trained to effectively decode object latent features, by employing a pioneering $\\textit{class-partitioned codebook}$ where codevectors are labeled by class. To address the problem of $\\textit{codebook collapse}$, we propose a $\\textit{class-aware}$ running average update which reinitializes dead codevectors within each partition. During inference, object features and class labels, both generated by a Latent-space Flow Matching Model (LFMM) designed specifically for scene generation, are consumed by the CPVQ-VAE. The CPVQ-VAE's class-aware inverse look-up then maps generated latents to codebook entries that are decoded to class-specific point cloud shapes. Thereby, we achieve pure point cloud generation without relying on an external objects database for retrieval. Extensive experiments reveal that our method reliably recovers plausible point cloud scenes, with up to 70.4% and 72.3% reduction in Chamfer and Point2Mesh errors on complex living room scenes.", "date": "2026-01-18", "categories": []} {"id": "2601.13371", "title": "Spherical Geometry Diffusion: Generating High-quality 3D Face Geometry via Sphere-anchored Representations", "abstract": "A fundamental challenge in text-to-3D face generation is achieving high-quality geometry. The core difficulty lies in the arbitrary and intricate distribution of vertices in 3D space, making it challenging for existing models to establish clean connectivity and resulting in suboptimal geometry. To address this, our core insight is to simplify the underlying geometric structure by constraining the distribution onto a simple and regular manifold, a topological sphere. Building on this, we first propose the Spherical Geometry Representation, a novel face representation that anchors geometric signals to uniform spherical coordinates. This guarantees a regular point distribution, from which the mesh connectivity can be robustly reconstructed. Critically, this canonical sphere can be seamlessly unwrapped into a 2D map, creating a perfect synergy with powerful 2D generative models. We then introduce Spherical Geometry Diffusion, a conditional diffusion framework built upon this 2D map. It enables diverse and controllable generation by jointly modeling geometry and texture, where the geometry explicitly conditions the texture synthesis process. Our method's effectiveness is demonstrated through its success in a wide range of tasks: text-to-3D generation, face reconstruction, and text-based 3D editing. Extensive experiments show that our approach substantially outperforms existing methods in geometric quality, textual fidelity, and inference efficiency.", "date": "2026-01-19", "categories": []} {"id": "2601.13445", "title": "BladeSDF : Unconditional and Conditional Generative Modeling of Representative Blade Geometries Using Signed Distance Functions", "abstract": "Generative AI has emerged as a transformative paradigm in engineering design, enabling automated synthesis and reconstruction of complex 3D geometries while preserving feasibility and performance relevance. This paper introduces a domain-specific implicit generative framework for turbine blade geometry using DeepSDF, addressing critical gaps in performance-aware modeling and manufacturable design generation. The proposed method leverages a continuous signed distance function (SDF) representation to reconstruct and generate smooth, watertight geometries with quantified accuracy. It establishes an interpretable, near-Gaussian latent space that aligns with blade-relevant parameters, such as taper and chord ratios, enabling controlled exploration and unconditional synthesis through interpolation and Gaussian sampling. In addition, a compact neural network maps engineering descriptors, such as maximum directional strains, to latent codes, facilitating the generation of performance-informed geometry. The framework achieves high reconstruction fidelity, with surface distance errors concentrated within $1\\%$ of the maximum blade dimension, and demonstrates robust generalization to unseen designs. By integrating constraints, objectives, and performance metrics, this approach advances beyond traditional 2D-guided or unconstrained 3D pipelines, offering a practical and interpretable solution for data-driven turbine blade modeling and concept generation.", "date": "2026-01-19", "categories": []} {"id": "2601.14103", "title": "Interp3D: Correspondence-aware Interpolation for Generative Textured 3D Morphing", "abstract": "Textured 3D morphing seeks to generate smooth and plausible transitions between two 3D assets, preserving both structural coherence and fine-grained appearance. This ability is crucial not only for advancing 3D generation research but also for practical applications in animation, editing, and digital content creation. Existing approaches either operate directly on geometry, limiting them to shape-only morphing while neglecting textures, or extend 2D interpolation strategies into 3D, which often causes semantic ambiguity, structural misalignment, and texture blurring. These challenges underscore the necessity to jointly preserve geometric consistency, texture alignment, and robustness throughout the transition process. To address this, we propose Interp3D, a novel training-free framework for textured 3D morphing. It harnesses generative priors and adopts a progressive alignment principle to ensure both geometric fidelity and texture coherence. Starting from semantically aligned interpolation in condition space, Interp3D enforces structural consistency via SLAT (Structured Latent)-guided structure interpolation, and finally transfers appearance details through fine-grained texture fusion. For comprehensive evaluations, we construct a dedicated dataset, Interp3DData, with graded difficulty levels and assess generation results from fidelity, transition smoothness, and plausibility. Both quantitative metrics and human studies demonstrate the significant advantages of our proposed approach over previous methods. Source code is available at https://github.com/xiaolul2/Interp3D.", "date": "2026-01-20", "categories": []} {"id": "2601.05810", "title": "SceneFoundry: Generating Interactive Infinite 3D Worlds", "abstract": "The ability to automatically generate large-scale, interactive, and physically realistic 3D environments is crucial for advancing robotic learning and embodied intelligence. However, existing generative approaches often fail to capture the functional complexity of real-world interiors, particularly those containing articulated objects with movable parts essential for manipulation and navigation. This paper presents SceneFoundry, a language-guided diffusion framework that generates apartment-scale 3D worlds with functionally articulated furniture and semantically diverse layouts for robotic training. From natural language prompts, an LLM module controls floor layout generation, while diffusion-based posterior sampling efficiently populates the scene with articulated assets from large-scale 3D repositories. To ensure physical usability, SceneFoundry employs differentiable guidance functions to regulate object quantity, prevent articulation collisions, and maintain sufficient walkable space for robotic navigation. Extensive experiments demonstrate that our framework generates structurally valid, semantically coherent, and functionally interactive environments across diverse scene types and conditions, enabling scalable embodied AI research. project page: https://anc891203.github.io/SceneFoundry-Demo/", "date": "2026-01-09", "categories": []} {"id": "2601.11514", "title": "ShapeR: Robust Conditional 3D Shape Generation from Casual Captures", "abstract": "Recent advances in 3D shape generation have achieved impressive results, but most existing methods rely on clean, unoccluded, and well-segmented inputs. Such conditions are rarely met in real-world scenarios. We present ShapeR, a novel approach for conditional 3D object shape generation from casually captured sequences. Given an image sequence, we leverage off-the-shelf visual-inertial SLAM, 3D detection algorithms, and vision-language models to extract, for each object, a set of sparse SLAM points, posed multi-view images, and machine-generated captions. A rectified flow transformer trained to effectively condition on these modalities then generates high-fidelity metric 3D shapes. To ensure robustness to the challenges of casually captured data, we employ a range of techniques including on-the-fly compositional augmentations, a curriculum training scheme spanning object- and scene-level datasets, and strategies to handle background clutter. Additionally, we introduce a new evaluation benchmark comprising 178 in-the-wild objects across 7 real-world scenes with geometry annotations. Experiments show that ShapeR significantly outperforms existing approaches in this challenging setting, achieving an improvement of 2.7x in Chamfer distance compared to state of the art.", "date": "2026-01-16", "categories": []} {"id": "2503.08516", "title": "High-Quality 3D Head Reconstruction from Any Single Portrait Image", "abstract": "In this work, we introduce a novel high-fidelity 3D head reconstruction method from a single portrait image, regardless of perspective, expression, or accessories. Despite significant efforts in adapting 2D generative models for novel view synthesis and 3D optimization, most methods struggle to produce high-quality 3D portraits. The lack of crucial information, such as identity, expression, hair, and accessories, limits these approaches in generating realistic 3D head models. To address these challenges, we construct a new high-quality dataset containing 227 sequences of digital human portraits captured from 96 different perspectives, totalling 21,792 frames, featuring diverse expressions and accessories. To further improve performance, we integrate identity and expression information into the multi-view diffusion process to enhance facial consistency across views. Specifically, we apply identity- and expression-aware guidance and supervision to extract accurate facial representations, which guide the model and enforce objective functions to ensure high identity and expression consistency during generation. Finally, we generate an orbital video around the portrait consisting of 96 multi-view frames, which can be used for 3D portrait model reconstruction. Our method demonstrates robust performance across challenging scenarios, including side-face angles and complex accessories", "date": "2025-03-11", "categories": []} {"id": "2508.19188", "title": "FastMesh: Efficient Artistic Mesh Generation via Component Decoupling", "abstract": "Recent mesh generation approaches typically tokenize triangle meshes into sequences of tokens and train autoregressive models to generate these tokens sequentially. Despite substantial progress, such token sequences inevitably reuse vertices multiple times to fully represent manifold meshes, as each vertex is shared by multiple faces. This redundancy leads to excessively long token sequences and inefficient generation processes. In this paper, we propose an efficient framework that generates artistic meshes by treating vertices and faces separately, significantly reducing redundancy. We employ an autoregressive model solely for vertex generation, decreasing the token count to approximately 23% of that required by the most compact existing tokenizer. Next, we leverage a bidirectional transformer to complete the mesh in a single step by capturing inter-vertex relationships and constructing the adjacency matrix that defines the mesh faces. To further improve the generation quality, we introduce a fidelity enhancer to refine vertex positioning into more natural arrangements and propose a post-processing framework to remove undesirable edge connections. Experimental results show that our method achieves more than 8x faster speed on mesh generation compared to state-of-the-art approaches, while producing higher mesh quality.", "date": "2025-08-26", "categories": []} {"id": "2509.14981", "title": "SPATIALGEN: Layout-guided 3D Indoor Scene Generation", "abstract": "Creating high-fidelity 3D models of indoor environments is essential for applications in design, virtual reality, and robotics. However, manual 3D modeling remains time-consuming and labor-intensive. While recent advances in generative AI have enabled automated scene synthesis, existing methods often face challenges in balancing visual quality, diversity, semantic consistency, and user control. A major bottleneck is the lack of a large-scale, high-quality dataset tailored to this task. To address this gap, we introduce a comprehensive synthetic dataset, featuring 12,328 structured annotated scenes with 57,431 rooms, and 4.7M photorealistic 2D renderings. Leveraging this dataset, we present SpatialGen, a novel multi-view multi-modal diffusion model that generates realistic and semantically consistent 3D indoor scenes. Given a 3D layout and a reference image (derived from a text prompt), our model synthesizes appearance (color image), geometry (scene coordinate map), and semantic (semantic segmentation map) from arbitrary viewpoints, while preserving spatial consistency across modalities. SpatialGen consistently generates superior results to previous methods in our experiments. We are open-sourcing our data and models to empower the community and advance the field of indoor scene understanding and generation.", "date": "2025-09-18", "categories": []} {"id": "2601.08015", "title": "Decoder Generates Manufacturable Structures: A Framework for 3D-Printable Object Synthesis", "abstract": "This paper presents a novel decoder-based approach for generating manufacturable 3D structures optimized for additive manufacturing. We introduce a deep learning framework that decodes latent representations into geometrically valid, printable objects while respecting manufacturing constraints such as overhang angles, wall thickness, and structural integrity. The methodology demonstrates that neural decoders can learn complex mapping functions from abstract representations to valid 3D geometries, producing parts with significantly improved manufacturability compared to naive generation approaches. We validate the approach on diverse object categories and demonstrate practical 3D printing of decoder-generated structures.", "date": "2026-01-07", "categories": []} {"id": "2601.08179", "title": "Instruction-Driven 3D Facial Expression Generation and Transition", "abstract": "A 3D avatar typically has one of six cardinal facial expressions. To simulate realistic emotional variation, we should be able to render a facial transition between two arbitrary expressions. This study presents a new framework for instruction-driven facial expression generation that produces a 3D face and, starting from an image of the face, transforms the facial expression from one designated facial expression to another. The Instruction-driven Facial Expression Decomposer (IFED) module is introduced to facilitate multimodal data learning and capture the correlation between textual descriptions and facial expression features. Subsequently, we propose the Instruction to Facial Expression Transition (I2FET) method, which leverages IFED and a vertex reconstruction loss function to refine the semantic comprehension of latent vectors, thus generating a facial expression sequence according to the given instruction. Lastly, we present the Facial Expression Transition model to generate smooth transitions between facial expressions. Extensive evaluation suggests that the proposed model outperforms state-of-the-art methods on the CK+ and CelebV-HQ datasets. The results show that our framework can generate facial expression trajectories according to text instruction. Considering that text prompts allow us to make diverse descriptions of human emotional states, the repertoire of facial expressions and the transitions between them can be expanded greatly. We expect our framework to find various practical applications More information about our project can be found at https://vohoanganh.github.io/tg3dfet/", "date": "2026-01-13", "categories": []} {"id": "2601.07218", "title": "SceneNAT: Masked Generative Modeling for Language-Guided Indoor Scene Synthesis", "abstract": "We present SceneNAT, a single-stage masked non-autoregressive Transformer that synthesizes complete 3D indoor scenes from natural language instructions through only a few parallel decoding passes, offering improved performance and efficiency compared to prior state-of-the-art approaches. SceneNAT is trained via masked modeling over fully discretized representations of both semantic and spatial attributes. By applying a masking strategy at both the attribute level and the instance level, the model can better capture intra-object and inter-object structure. To boost relational reasoning, SceneNAT employs a dedicated triplet predictor for modeling the scene's layout and object relationships by mapping a set of learnable relation queries to a sparse set of symbolic triplets (subject, predicate, object). Extensive experiments on the 3D-FRONT dataset demonstrate that SceneNAT achieves superior performance compared to state-of-the-art autoregressive and diffusion baselines in both semantic compliance and spatial arrangement accuracy, while operating with substantially lower computational cost.", "date": "2026-01-12", "categories": []} {"id": "2601.07660", "title": "StdGEN++: A Comprehensive System for Semantic-Decomposed 3D Character Generation", "abstract": "We present StdGEN++, a novel and comprehensive system for generating high-fidelity, semantically decomposed 3D characters from diverse inputs. Existing 3D generative methods often produce monolithic meshes that lack the structural flexibility required by industrial pipelines in gaming and animation. Addressing this gap, StdGEN++ is built upon a Dual-branch Semantic-aware Large Reconstruction Model (Dual-Branch S-LRM), which jointly reconstructs geometry, color, and per-component semantics in a feed-forward manner. To achieve production-level fidelity, we introduce a novel semantic surface extraction formalism compatible with hybrid implicit fields. This mechanism is accelerated by a coarse-to-fine proposal scheme, which significantly reduces memory footprint and enables high-resolution mesh generation. Furthermore, we propose a video-diffusion-based texture decomposition module that disentangles appearance into editable layers (e.g., separated iris and skin), resolving semantic confusion in facial regions. Experiments demonstrate that StdGEN++ achieves state-of-the-art performance, significantly outperforming existing methods in geometric accuracy and semantic disentanglement. Crucially, the resulting structural independence unlocks advanced downstream capabilities, including non-destructive editing, physics-compliant animation, and gaze tracking, making it a robust solution for automated character asset production.", "date": "2026-01-12", "categories": []} {"id": "2411.19037", "title": "3D-WAG: Hierarchical Wavelet-Guided Autoregressive Generation for High-Fidelity 3D Shapes", "abstract": "Autoregressive (AR) models have achieved remarkable success in natural language and image generation, but their application to 3D shape modeling remains largely unexplored. Unlike diffusion models, AR models enable more efficient and controllable generation with faster inference times, making them especially suitable for data-intensive domains. Traditional 3D generative models using AR approaches often rely on ``next-token\" predictions at the voxel or point level. While effective for certain applications, these methods can be restrictive and computationally expensive when dealing with large-scale 3D data. To tackle these challenges, we introduce 3D-WAG, an AR model for 3D implicit distance fields that can perform unconditional shape generation, class-conditioned and also text-conditioned shape generation. Our key idea is to encode shapes as multi-scale wavelet token maps and use a Transformer to predict the ``next higher-resolution token map\" in an autoregressive manner. By redefining 3D AR generation task as ``next-scale\" prediction, we reduce the computational cost of generation compared to traditional ``next-token\" prediction models, while preserving essential geometric details of 3D shapes in a more structured and hierarchical manner. We evaluate 3D-WAG to showcase its benefit by quantitative and qualitative comparisons with state-of-the-art methods on widely used benchmarks. Our results show 3D-WAG achieves superior performance in key metrics like Coverage and MMD, generating high-fidelity 3D shapes that closely match the real data distribution.", "date": "2024-11-28", "categories": []} {"id": "2601.05722", "title": "Rotate Your Character: Revisiting Video Diffusion Models for High-Quality 3D Character Generation", "abstract": "Generating high-quality 3D characters from single images remains a significant challenge in digital content creation, particularly due to complex body poses and self-occlusion. In this paper, we present RCM (Rotate your Character Model), an advanced image-to-video diffusion framework tailored for high-quality novel view synthesis (NVS) and 3D character generation. Compared to existing diffusion-based approaches, RCM offers several key advantages: (1) transferring characters with any complex poses into a canonical pose, enabling consistent novel view synthesis across the entire viewing orbit, (2) high-resolution orbital video generation at 1024x1024 resolution, (3) controllable observation positions given different initial camera poses, and (4) multi-view conditioning supporting up to 4 input images, accommodating diverse user scenarios. Extensive experiments demonstrate that RCM outperforms state-of-the-art methods in both novel view synthesis and 3D generation quality.", "date": "2026-01-09", "categories": []} {"id": "2512.13683", "title": "I-Scene: 3D Instance Models are Implicit Generalizable Spatial Learners", "abstract": "Generalization remains the central challenge for interactive 3D scene generation. Existing learning-based approaches ground spatial understanding in limited scene dataset, restricting generalization to new layouts. We instead reprogram a pre-trained 3D instance generator to act as a scene level learner, replacing dataset-bounded supervision with model-centric spatial supervision. This reprogramming unlocks the generator transferable spatial knowledge, enabling generalization to unseen layouts and novel object compositions. Remarkably, spatial reasoning still emerges even when the training scenes are randomly composed objects. This demonstrates that the generator's transferable scene prior provides a rich learning signal for inferring proximity, support, and symmetry from purely geometric cues. Replacing widely used canonical space, we instantiate this insight with a view-centric formulation of the scene space, yielding a fully feed-forward, generalizable scene generator that learns spatial relations directly from the instance model. Quantitative and qualitative results show that a 3D instance generator is an implicit spatial learner and reasoner, pointing toward foundation models for interactive 3D scene understanding and generation. Project page: https://luling06.github.io/I-Scene-project/", "date": "2025-12-15", "categories": []} {"id": "2601.04194", "title": "Choreographing a World of Dynamic Objects", "abstract": "Dynamic objects in our physical 4D (3D + time) world are constantly evolving, deforming, and interacting with other objects, leading to diverse 4D scene dynamics. In this paper, we present a universal generative pipeline, CHORD, for CHOReographing Dynamic objects and scenes and synthesizing this type of phenomena. Traditional rule-based graphics pipelines to create these dynamics are based on category-specific heuristics, yet are labor-intensive and not scalable. Recent learning-based methods typically demand large-scale datasets, which may not cover all object categories in interest. Our approach instead inherits the universality from the video generative models by proposing a distillation-based pipeline to extract the rich Lagrangian motion information hidden in the Eulerian representations of 2D videos. Our method is universal, versatile, and category-agnostic. We demonstrate its effectiveness by conducting experiments to generate a diverse range of multi-body 4D dynamics, show its advantage compared to existing methods, and demonstrate its applicability in generating robotics manipulation policies. Project page: https://yanzhelyu.github.io/chord", "date": "2026-01-07", "categories": []} {"id": "2508.15169", "title": "MeSS: City Mesh-Guided Outdoor Scene Generation with Cross-View Consistent Diffusion", "abstract": "Mesh models have become increasingly accessible for numerous cities; however, the lack of realistic textures restricts their application in virtual urban navigation and autonomous driving. To address this, this paper proposes MeSS (Meshbased Scene Synthesis) for generating high-quality, styleconsistent outdoor scenes with city mesh models serving as the geometric prior. While image and video diffusion models can leverage spatial layouts (such as depth maps or HD maps) as control conditions to generate street-level perspective views, they are not directly applicable to 3D scene generation. Video diffusion models excel at synthesizing consistent view sequences that depict scenes but often struggle to adhere to predefined camera paths or align accurately with rendered control videos. In contrast, image diffusion models, though unable to guarantee cross-view visual consistency, can produce more geometry-aligned results when combined with ControlNet. Building on this insight, our approach enhances image diffusion models by improving cross-view consistency. The pipeline comprises three key stages: first, we generate geometrically consistent sparse views using Cascaded Outpainting ControlNets; second, we propagate denser intermediate views via a component dubbed AGInpaint; and third, we globally eliminate visual inconsistencies (e.g., varying exposure) using the GCAlign module. Concurrently with generation, a 3D Gaussian Splatting (3DGS) scene is reconstructed by initializing Gaussian balls on the mesh surface. Our method outperforms existing approaches in both geometric alignment and generation quality. Once synthesized, the scene can be rendered in diverse styles through relighting and style transfer techniques. project page: https://albertchen98.github.io/mess/", "date": "2025-08-21", "categories": []} {"id": "2511.22425", "title": "Wukong's 72 Transformations: High-fidelity Textured 3D Morphing via Flow Models", "abstract": "We present WUKONG, a novel training-free framework for high-fidelity textured 3D morphing that takes a pair of source and target prompts (image or text) as input. Unlike conventional methods -- which rely on manual correspondence matching and deformation trajectory estimation (limiting generalization and requiring costly preprocessing) -- WUKONG leverages the generative prior of flow-based transformers to produce high-fidelity 3D transitions with rich texture details. To ensure smooth shape transitions, we exploit the inherent continuity of flow-based generative processes and formulate morphing as an optimal transport barycenter problem. We further introduce a sequential initialization strategy to prevent abrupt geometric distortions and preserve identity coherence. For faithful texture preservation, we propose a similarity-guided semantic consistency mechanism that selectively retains high-frequency details and enables precise control over blending dynamics. This empowers WUKONG to support both global texture transitions and identity-preserving texture morphing, catering to diverse generation needs. Extensive quantitative and qualitative evaluations demonstrate that WUKONG significantly outperforms state-of-the-art methods, achieving superior results across diverse geometry and texture variations.", "date": "2025-11-27", "categories": []} {"id": "2507.14921", "title": "Stereo-GS: Multi-View Stereo Vision Model for Generalizable 3D Gaussian Splatting Reconstruction", "abstract": "Generalizable 3D Gaussian Splatting reconstruction showcases advanced Image-to-3D content creation but requires substantial computational resources and large datasets, posing challenges to training models from scratch. Current methods usually entangle the prediction of 3D Gaussian geometry and appearance, which rely heavily on data-driven priors and result in slow regression speeds. To address this, we propose \\method, a disentangled framework for efficient 3D Gaussian prediction. Our method extracts features from local image pairs using a stereo vision backbone and fuses them via global attention blocks. Dedicated point and Gaussian prediction heads generate multi-view point-maps for geometry and Gaussian features for appearance, combined as GS-maps to represent the 3DGS object. A refinement network enhances these GS-maps for high-quality reconstruction. Unlike existing methods that depend on camera parameters, our approach achieves pose-free 3D reconstruction, improving robustness and practicality. By reducing resource demands while maintaining high-quality outputs, \\method provides an efficient, scalable solution for real-world 3D content generation.", "date": "2025-07-20", "categories": []} {"id": "2601.00368", "title": "Mask-Conditioned Voxel Diffusion for Joint Geometry and Color Inpainting", "abstract": "We present a lightweight two-stage framework for joint geometry and color inpainting of damaged 3D objects, motivated by the digital restoration of cultural heritage artifacts. The pipeline separates damage localization from reconstruction. In the first stage, a 2D convolutional network predicts damage masks on RGB slices extracted from a voxelized object, and these predictions are aggregated into a volumetric mask. In the second stage, a diffusion-based 3D U-Net performs mask-conditioned inpainting directly on voxel grids, reconstructing geometry and color while preserving observed regions. The model jointly predicts occupancy and color using a composite objective that combines occupancy reconstruction with masked color reconstruction and perceptual regularization. We evaluate the approach on a curated set of textured artifacts with synthetically generated damage using standard geometric and color metrics. Compared to symmetry-based baselines, our method produces more complete geometry and more coherent color reconstructions at a fixed 32^3 resolution. Overall, the results indicate that explicit mask conditioning is a practical way to guide volumetric diffusion models for joint 3D geometry and color inpainting.", "date": "2026-01-01", "categories": []} {"id": "2512.25071", "title": "Edit3r: Instant 3D Scene Editing from Sparse Unposed Images", "abstract": "We present Edit3r, a feed-forward framework that reconstructs and edits 3D scenes in a single pass from unposed, view-inconsistent, instruction-edited images. Unlike prior methods requiring per-scene optimization, Edit3r directly predicts instruction-aligned 3D edits, enabling fast and photorealistic rendering without optimization or pose estimation. A key challenge in training such a model lies in the absence of multi-view consistent edited images for supervision. We address this with (i) a SAM2-based recoloring strategy that generates reliable, cross-view-consistent supervision, and (ii) an asymmetric input strategy that pairs a recolored reference view with raw auxiliary views, encouraging the network to fuse and align disparate observations. At inference, our model effectively handles images edited by 2D methods such as InstructPix2Pix, despite not being exposed to such edits during training. For large-scale quantitative evaluation, we introduce DL3DV-Edit-Bench, a benchmark built on the DL3DV test split, featuring 20 diverse scenes, 4 edit types and 100 edits in total. Comprehensive quantitative and qualitative results show that Edit3r achieves superior semantic alignment and enhanced 3D consistency compared to recent baselines, while operating at significantly higher inference speed, making it promising for real-time 3D editing applications.", "date": "2025-12-31", "categories": []} {"id": "2412.19413", "title": "Multi-scale Latent Point Consistency Models for 3D Shape Generation", "abstract": "Consistency Models (CMs) have significantly accelerated the sampling process in diffusion models, yielding impressive results in synthesizing high-resolution images. To explore and extend these advancements to point-cloud-based 3D shape generation, we propose a novel Multi-scale Latent Point Consistency Model (MLPCM). Our MLPCM follows a latent diffusion framework and introduces hierarchical levels of latent representations, ranging from point-level to super-point levels, each corresponding to a different spatial resolution. We design a multi-scale latent integration module along with 3D spatial attention to effectively denoise the point-level latent representations conditioned on those from multiple super-point levels. Additionally, we propose a latent consistency model, learned through consistency distillation, that compresses the prior into a one-step generator. This significantly improves sampling efficiency while preserving the performance of the original teacher model. Extensive experiments on standard benchmarks ShapeNet and ShapeNet-Vol demonstrate that MLPCM achieves a 100x speedup in the generation process, while surpassing state-of-the-art diffusion models in terms of both shape quality and diversity.", "date": "2024-12-27", "categories": []} {"id": "2512.23628", "title": "Memorization in 3D Shape Generation: An Empirical Study", "abstract": "Generative models are increasingly used in 3D vision to synthesize novel shapes, yet it remains unclear whether their generation relies on memorizing training shapes. Understanding their memorization could help prevent training data leakage and improve the diversity of generated results. In this paper, we design an evaluation framework to quantify memorization in 3D generative models and study the influence of different data and modeling designs on memorization. We first apply our framework to quantify memorization in existing methods. Next, through controlled experiments with a latent vector-set (Vecset) diffusion model, we find that, on the data side, memorization depends on data modality, and increases with data diversity and finer-grained conditioning; on the modeling side, it peaks at a moderate guidance scale and can be mitigated by longer Vecsets and simple rotation augmentation. Together, our framework and analysis provide an empirical understanding of memorization in 3D generative models and suggest simple yet effective strategies to reduce it without degrading generation quality. Our code is available at https://github.com/zlab-princeton/3d_mem.", "date": "2025-12-29", "categories": []} {"id": "2505.14537", "title": "Personalize Your Gaussian: Consistent 3D Scene Personalization from a Single Image", "abstract": "Personalizing 3D scenes from a single reference image enables intuitive user-guided editing, which requires achieving both multi-view consistency across perspectives and referential consistency with the input image. However, these goals are particularly challenging due to the viewpoint bias caused by the limited perspective provided in a single image. Lacking the mechanisms to effectively expand reference information beyond the original view, existing methods of image-conditioned 3DGS personalization often suffer from this viewpoint bias and struggle to produce consistent results. Therefore, in this paper, we present Consistent Personalization for 3D Gaussian Splatting (CP-GS), a framework that progressively propagates the single-view reference appearance to novel perspectives. In particular, CP-GS integrates pre-trained image-to-3D generation and iterative LoRA fine-tuning to extract and extend the reference appearance, and finally produces faithful multi-view guidance images and the personalized 3DGS outputs through a view-consistent generation process guided by geometric cues. Extensive experiments on real-world scenes show that our CP-GS effectively mitigates the viewpoint bias, achieving high-quality personalization that significantly outperforms existing methods.", "date": "2025-05-20", "categories": []} {"id": "2512.21185", "title": "UltraShape 1.0: High-Fidelity 3D Shape Generation via Scalable Geometric Refinement", "abstract": "In this report, we introduce UltraShape 1.0, a scalable 3D diffusion framework for high-fidelity 3D geometry generation. The proposed approach adopts a two-stage generation pipeline: a coarse global structure is first synthesized and then refined to produce detailed, high-quality geometry. To support reliable 3D generation, we develop a comprehensive data processing pipeline that includes a novel watertight processing method and high-quality data filtering. This pipeline improves the geometric quality of publicly available 3D datasets by removing low-quality samples, filling holes, and thickening thin structures, while preserving fine-grained geometric details. To enable fine-grained geometry refinement, we decouple spatial localization from geometric detail synthesis in the diffusion process. We achieve this by performing voxel-based refinement at fixed spatial locations, where voxel queries derived from coarse geometry provide explicit positional anchors encoded via RoPE, allowing the diffusion model to focus on synthesizing local geometric details within a reduced, structured solution space. Our model is trained exclusively on publicly available 3D datasets, achieving strong geometric quality despite limited training resources. Extensive evaluations demonstrate that UltraShape 1.0 performs competitively with existing open-source methods in both data processing quality and geometry generation. All code and trained models will be released to support future research.", "date": "2025-12-24", "categories": []} {"id": "2512.20105", "title": "LiDARDraft: Generating LiDAR Point Cloud from Versatile Inputs", "abstract": "Generating realistic and diverse LiDAR point clouds is crucial for autonomous driving simulation. Although previous methods achieve LiDAR point cloud generation from user inputs, they struggle to attain high-quality results while enabling versatile controllability, due to the imbalance between the complex distribution of LiDAR point clouds and the simple control signals. To address the limitation, we propose LiDARDraft, which utilizes the 3D layout to build a bridge between versatile conditional signals and LiDAR point clouds. The 3D layout can be trivially generated from various user inputs such as textual descriptions and images. Specifically, we represent text, images, and point clouds as unified 3D layouts, which are further transformed into semantic and depth control signals. Then, we employ a rangemap-based ControlNet to guide LiDAR point cloud generation. This pixel-level alignment approach demonstrates excellent performance in controllable LiDAR point clouds generation, enabling \"simulation from scratch\", allowing self-driving environments to be created from arbitrary textual descriptions, images and sketches.", "date": "2025-12-23", "categories": []} {"id": "2504.11734", "title": "Recent Advances in 3D Object and Scene Generation: A Survey", "abstract": "In recent years, the demand for 3D content has grown exponentially with the intelligent upgrade of interactive media, extended reality (XR), and Metaverse industries. In order to overcome the limitations of traditional manual modeling approaches, such as labor-intensive workflows and prolonged production cycles, revolutionary advances have been achieved through the convergence of novel 3D representation paradigms and artificial intelligence generative technologies. In this survey, we conduct a systematic review of the cutting-edge achievements in static 3D object and scene generation, as well as establish a comprehensive technical framework through systematic categorization. We start our analysis with mainstream 3D object representations. Subsequently, we delve into the technical pathways of 3D object generation based on four mainstream deep generative models: Variational Autoencoders, Generative Adversarial Networks, Autoregressive Models, and Diffusion Models. Regarding scene generation, we focus on three dominant paradigms: layout-guided generation, lifting based on 2D priors, and rule-driven modeling. Finally, we critically examine persistent challenges in 3D generation and propose potential research directions for future investigation. This survey aims to provide readers with a structured understanding of state-of-the-art 3D generation technologies while inspiring researchers to undertake more exploration in this domain.", "date": "2025-04-16", "categories": []} {"id": "2505.04262", "title": "Bridging Geometry-Coherent Text-to-3D Generation with Multi-View Diffusion Priors and Gaussian Splatting", "abstract": "Score Distillation Sampling (SDS) leverages pretrained 2D diffusion models to advance text-to-3D generation but neglects multi-view correlations, being prone to geometric inconsistencies and multi-face artifacts in the generated 3D content. In this work, we propose Coupled Score Distillation (CSD), a framework that couples multi-view joint distribution priors to ensure geometrically consistent 3D generation while enabling the stable and direct optimization of 3D Gaussian Splatting. Specifically, by reformulating the optimization as a multi-view joint optimization problem, we derive an effective optimization rule that effectively couples multi-view priors to guide optimization across different viewpoints while preserving the diversity of generated 3D assets. Additionally, we propose a framework that directly optimizes 3D Gaussian Splatting (3D-GS) with random initialization to generate geometrically consistent 3D content. We further employ a deformable tetrahedral grid, initialized from 3D-GS and refined through CSD, to produce high-quality, refined meshes. Quantitative and qualitative experimental results demonstrate the efficiency and competitive quality of our approach.", "date": "2025-05-07", "categories": []} {"id": "2506.22678", "title": "3D Shape Generation: A Survey", "abstract": "Recent advances in deep learning have significantly transformed the field of 3D shape generation, enabling the synthesis of complex, diverse, and semantically meaningful 3D objects. This survey provides a comprehensive overview of the current state-of-the-art in 3D shape generation, organizing the discussion around three core components: shape representations, generative modeling approaches, and evaluation protocols. We begin by categorizing 3D representations into explicit, implicit, and hybrid setups, highlighting their structural properties, advantages, and limitations. Next, we review a wide range of generation methods, focusing on feedforward architectures. We further summarize commonly used datasets and evaluation metrics that assess fidelity, diversity, and realism of generated shapes. Finally, we identify open challenges and outline future research directions that could drive progress in controllable, efficient, and high-quality 3D shape generation. This survey aims to serve as a valuable reference for researchers and practitioners seeking a structured and in-depth understanding of this rapidly evolving field.", "date": "2025-06-27", "categories": []} {"id": "2512.19560", "title": "BabyFlow: 3D modeling of realistic and expressive infant faces", "abstract": "Early detection of developmental disorders can be aided by analyzing infant craniofacial morphology, but modeling infant faces is challenging due to limited data and frequent spontaneous expressions. We introduce BabyFlow, a generative AI model that disentangles facial identity and expression, enabling independent control over both. Using normalizing flows, BabyFlow learns flexible, probabilistic representations that capture the complex, non-linear variability of expressive infant faces without restrictive linear assumptions. To address scarce and uncontrolled expressive data, we perform cross-age expression transfer, adapting expressions from adult 3D scans to enrich infant datasets with realistic and systematic expressive variants. As a result, BabyFlow improves 3D reconstruction accuracy, particularly in highly expressive regions such as the mouth, eyes, and nose, and supports synthesis and modification of infant expressions while preserving identity. Additionally, by integrating with diffusion models, BabyFlow generates high-fidelity 2D infant images with consistent 3D geometry, providing powerful tools for data augmentation and early facial analysis.", "date": "2025-12-22", "categories": []} {"id": "2403.12019", "title": "LN3DIFF++: Scalable Latent Neural Fields Diffusion for Speedy 3D Generation", "abstract": "The field of neural rendering has witnessed significant progress with advancements in generative models and differentiable rendering techniques. Though 2D diffusion has achieved success, a unified 3D diffusion pipeline remains unsettled. This paper introduces a novel framework called LN3Diff++ to address this gap and enable fast, high-quality, and generic conditional 3D generation. Our approach harnesses a 3D-aware architecture and variational autoencoder (VAE) to encode the input image into a structured, compact, and 3D latent space. The latent is decoded by a transformer-based decoder into a high-capacity 3D neural field. Through training a diffusion model on this 3D-aware latent space, our method achieves state-of-the-art performance on ShapeNet for 3D generation and demonstrates superior performance in monocular 3D reconstruction and conditional 3D generation across various datasets. Moreover, it surpasses existing 3D diffusion methods in terms of inference speed, requiring no per-instance optimization. Our proposed LN3Diff presents a significant advancement in 3D generative modeling and holds promise for various applications in 3D vision and graphics tasks.", "date": "2024-03-18", "categories": []} {"id": "2508.05899", "title": "HOLODECK 2.0: Vision-Language-Guided 3D World Generation with Editing", "abstract": "3D scene generation plays a crucial role in gaming, artistic creation, virtual reality, and many other domains. However, current 3D scene design still relies heavily on extensive manual effort from creators, and existing automated methods struggle to generate open-domain scenes or support flexible editing. To address those challenges, we introduce HOLODECK 2.0, an advanced vision-language-guided framework for 3D world generation with support for interactive scene editing based on human feedback. HOLODECK 2.0 can generate diverse and stylistically rich 3D scenes (e.g., realistic, cartoon, anime, and cyberpunk styles) that exhibit high semantic fidelity to fine-grained input descriptions, suitable for both indoor and open-domain environments. HOLODECK 2.0 leverages vision-language models (VLMs) to identify and parse the objects required in a scene and generates corresponding high-quality assets via state-of-the-art 3D generative models. Then, HOLODECK 2.0 iteratively applies spatial constraints derived from the VLMs to achieve semantically coherent and physically plausible layouts. Both human and model evaluations demonstrate that HOLODECK 2.0 effectively generates high-quality scenes closely aligned with detailed textual descriptions, consistently outperforming baselines across indoor and open-domain scenarios. Additionally, HOLODECK 2.0 provides editing capabilities that flexibly adapt to human feedback, supporting layout refinement and style-consistent object edits. Finally, we present a practical application of HOLODECK 2.0 in procedural game modeling to generate visually rich and immersive environments that can boost efficiency in game design.", "date": "2025-08-07", "categories": []} {"id": "2512.17302", "title": "MatLat: Material Latent Space for PBR Texture Generation", "abstract": "We propose a generative framework for producing high-quality PBR textures on a given 3D mesh. As large-scale PBR texture datasets are scarce, our approach focuses on effectively leveraging the embedding space and diffusion priors of pretrained latent image generative models while learning a material latent space, MatLat, through targeted fine-tuning. Unlike prior methods that freeze the embedding network and thus lead to distribution shifts when encoding additional PBR channels and hinder subsequent diffusion training, we fine-tune the pretrained VAE so that new material channels can be incorporated with minimal latent distribution deviation. We further show that correspondence-aware attention alone is insufficient for cross-view consistency unless the latent-to-image mapping preserves locality. To enforce this locality, we introduce a regularization in the VAE fine-tuning that crops latent patches, decodes them, and aligns the corresponding image regions to maintain strong pixel-latent spatial correspondence. Ablation studies and comparison with previous baselines demonstrate that our framework improves PBR texture fidelity and that each component is critical for achieving state-of-the-art performance.", "date": "2025-12-19", "categories": []} {"id": "2505.04050", "title": "TerraFusion: Joint Generation of Terrain Geometry and Texture Using Latent Diffusion Models", "abstract": "3D terrain models are essential in fields such as video game development and film production. Since surface color often correlates with terrain geometry, capturing this relationship is crucial to achieving realism. However, most existing methods generate either a heightmap or a texture, without sufficiently accounting for the inherent correlation. In this paper, we propose a method that jointly generates terrain heightmaps and textures using a latent diffusion model. First, we train the model in an unsupervised manner to randomly generate paired heightmaps and textures. Then, we perform supervised learning of an external adapter to enable user control via hand-drawn sketches. Experiments show that our approach allows intuitive terrain generation while preserving the correlation between heightmaps and textures.", "date": "2025-05-07", "categories": []} {"id": "2512.07345", "title": "Debiasing Diffusion Priors via 3D Attention for Consistent Gaussian Splatting", "abstract": "Versatile 3D tasks (e.g., generation or editing) that distill from Text-to-Image (T2I) diffusion models have attracted significant research interest for not relying on extensive 3D training data. However, T2I models exhibit limitations resulting from prior view bias, which produces conflicting appearances between different views of an object. This bias causes subject-words to preferentially activate prior view features during cross-attention (CA) computation, regardless of the target view condition. To overcome this limitation, we conduct a comprehensive mathematical analysis to reveal the root cause of the prior view bias in T2I models. Moreover, we find different UNet layers show different effects of prior view in CA. Therefore, we propose a novel framework, TD-Attn, which addresses multi-view inconsistency via two key components: (1) the 3D-Aware Attention Guidance Module (3D-AAG) constructs a view-consistent 3D attention Gaussian for subject-words to enforce spatial consistency across attention-focused regions, thereby compensating for the limited spatial information in 2D individual view CA maps; (2) the Hierarchical Attention Modulation Module (HAM) utilizes a Semantic Guidance Tree (SGT) to direct the Semantic Response Profiler (SRP) in localizing and modulating CA layers that are highly responsive to view conditions, where the enhanced CA maps further support the construction of more consistent 3D attention Gaussians. Notably, HAM facilitates semantic-specific interventions, enabling controllable and precise 3D editing. Extensive experiments firmly establish that TD-Attn has the potential to serve as a universal plugin, significantly enhancing multi-view consistency across 3D tasks.", "date": "2025-12-08", "categories": []} {"id": "2512.14284", "title": "SS4D: Native 4D Generative Model via Structured Spacetime Latents", "abstract": "We present SS4D, a native 4D generative model that synthesizes dynamic 3D objects directly from monocular video. Unlike prior approaches that construct 4D representations by optimizing over 3D or video generative models, we train a generator directly on 4D data, achieving high fidelity, temporal coherence, and structural consistency. At the core of our method is a compressed set of structured spacetime latents. Specifically, (1) To address the scarcity of 4D training data, we build on a pre-trained single-image-to-3D model, preserving strong spatial consistency. (2) Temporal consistency is enforced by introducing dedicated temporal layers that reason across frames. (3) To support efficient training and inference over long video sequences, we compress the latent sequence along the temporal axis using factorized 4D convolutions and temporal downsampling blocks. In addition, we employ a carefully designed training strategy to enhance robustness against occlusion", "date": "2025-12-16", "categories": []} {"id": "2512.14677", "title": "VASA-3D: Lifelike Audio-Driven Gaussian Head Avatars from a Single Image", "abstract": "We propose VASA-3D, an audio-driven, single-shot 3D head avatar generator. This research tackles two major challenges: capturing the subtle expression details present in real human faces, and reconstructing an intricate 3D head avatar from a single portrait image. To accurately model expression details, VASA-3D leverages the motion latent of VASA-1, a method that yields exceptional realism and vividness in 2D talking heads. A critical element of our work is translating this motion latent to 3D, which is accomplished by devising a 3D head model that is conditioned on the motion latent. Customization of this model to a single image is achieved through an optimization framework that employs numerous video frames of the reference head synthesized from the input image. The optimization takes various training losses robust to artifacts and limited pose coverage in the generated training data. Our experiment shows that VASA-3D produces realistic 3D talking heads that cannot be achieved by prior art, and it supports the online generation of 512x512 free-viewpoint videos at up to 75 FPS, facilitating more immersive engagements with lifelike 3D avatars.", "date": "2025-12-16", "categories": []} {"id": "2512.14692", "title": "Native and Compact Structured Latents for 3D Generation", "abstract": "Recent advancements in 3D generative modeling have significantly improved the generation realism, yet the field is still hampered by existing representations, which struggle to capture assets with complex topologies and detailed appearance. This paper present an approach for learning a structured latent representation from native 3D data to address this challenge. At its core is a new sparse voxel structure called O-Voxel, an omni-voxel representation that encodes both geometry and appearance. O-Voxel can robustly model arbitrary topology, including open, non-manifold, and fully-enclosed surfaces, while capturing comprehensive surface attributes beyond texture color, such as physically-based rendering parameters. Based on O-Voxel, we design a Sparse Compression VAE which provides a high spatial compression rate and a compact latent space. We train large-scale flow-matching models comprising 4B parameters for 3D generation using diverse public 3D asset datasets. Despite their scale, inference remains highly efficient. Meanwhile, the geometry and material quality of our generated assets far exceed those of existing models. We believe our approach offers a significant advancement in 3D generative modeling.", "date": "2025-12-16", "categories": []} {"id": "2511.19004", "title": "A Self-Conditioned Representation Guided Diffusion Model for Realistic Text-to-LiDAR Scene Generation", "abstract": "Text-to-LiDAR generation can customize 3D data with rich structures and diverse scenes for downstream tasks. However, the scarcity of Text-LiDAR pairs often causes insufficient training priors, generating overly smooth 3D scenes. Moreover, low-quality text descriptions may degrade generation quality and controllability. In this paper, we propose a Text-to-LiDAR Diffusion Model for scene generation, named T2LDM, with a Self-Conditioned Representation Guidance (SCRG). Specifically, SCRG, by aligning to the real representations, provides the soft supervision with reconstruction details for the Denoising Network (DN) in training, while decoupled in inference. In this way, T2LDM can perceive rich geometric structures from data distribution, generating detailed objects in scenes. Meanwhile, we construct a content-composable Text-LiDAR benchmark, T2nuScenes, along with a controllability metric. Based on this, we analyze the effects of different text prompts for LiDAR generation quality and controllability, providing practical prompt paradigms and insights. Furthermore, a directional position prior is designed to mitigate street distortion, further improving scene fidelity. Additionally, by learning a conditional encoder via frozen DN, T2LDM can support multiple conditional tasks, including Sparse-to-Dense, Dense-to-Sparse, and Semantic-to-LiDAR generation. Extensive experiments in unconditional and conditional generation demonstrate that T2LDM outperforms existing methods, achieving state-of-the-art scene generation.", "date": "2025-11-24", "categories": []} {"id": "2512.12395", "title": "ArtGen: Conditional Generative Modeling of Articulated Objects in Arbitrary Part-Level States", "abstract": "Generating articulated assets is crucial for robotics, digital twins, and embodied intelligence. Existing generative models often rely on single-view inputs representing closed states, resulting in ambiguous or unrealistic kinematic structures due to the entanglement between geometric shape and joint dynamics. To address these challenges, we introduce ArtGen, a conditional diffusion-based framework capable of generating articulated 3D objects with accurate geometry and coherent kinematics from single-view images or text descriptions at arbitrary part-level states. Specifically, ArtGen employs cross-state Monte Carlo sampling to explicitly enforce global kinematic consistency, reducing structural-motion entanglement. Additionally, we integrate a Chain-of-Thought reasoning module to infer robust structural priors, such as part semantics, joint types, and connectivity, guiding a sparse-expert Diffusion Transformer to specialize in diverse kinematic interactions. Furthermore, a compositional 3D-VAE latent prior enhanced with local-global attention effectively captures fine-grained geometry and global part-level relationships. Extensive experiments on the PartNet-Mobility benchmark demonstrate that ArtGen significantly outperforms state-of-the-art methods.", "date": "2025-12-13", "categories": []} {"id": "2512.12534", "title": "Animus3D: Text-driven 3D Animation via Motion Score Distillation", "abstract": "We present Animus3D, a text-driven 3D animation framework that generates motion field given a static 3D asset and text prompt. Previous methods mostly leverage the vanilla Score Distillation Sampling (SDS) objective to distill motion from pretrained text-to-video diffusion, leading to animations with minimal movement or noticeable jitter. To address this, our approach introduces a novel SDS alternative, Motion Score Distillation (MSD). Specifically, we introduce a LoRA-enhanced video diffusion model that defines a static source distribution rather than pure noise as in SDS, while another inversion-based noise estimation technique ensures appearance preservation when guiding motion. To further improve motion fidelity, we incorporate explicit temporal and spatial regularization terms that mitigate geometric distortions across time and space. Additionally, we propose a motion refinement module to upscale the temporal resolution and enhance fine-grained details, overcoming the fixed-resolution constraints of the underlying video model. Extensive experiments demonstrate that Animus3D successfully animates static 3D assets from diverse text prompts, generating significantly more substantial and detailed motion than state-of-the-art baselines while maintaining high visual integrity. Code will be released at https://qiisun.github.io/animus3d_page.", "date": "2025-12-14", "categories": []} {"id": "2512.13678", "title": "Feedforward 3D Editing via Text-Steerable Image-to-3D", "abstract": "Recent progress in image-to-3D has opened up immense possibilities for design, AR/VR, and robotics. However, to use AI-generated 3D assets in real applications, a critical requirement is the capability to edit them easily. We present a feedforward method, Steer3D, to add text steerability to image-to-3D models, which enables editing of generated 3D assets with language. Our approach is inspired by ControlNet, which we adapt to image-to-3D generation to enable text steering directly in a forward pass. We build a scalable data engine for automatic data generation, and develop a two-stage training recipe based on flow-matching training and Direct Preference Optimization (DPO). Compared to competing methods, Steer3D more faithfully follows the language instruction and maintains better consistency with the original 3D asset, while being 2.4x to 28.5x faster. Steer3D demonstrates that it is possible to add a new modality (text) to steer the generation of pretrained image-to-3D generative models with 100k data. Project website: https://glab-caltech.github.io/steer3d/", "date": "2025-12-15", "categories": []} {"id": "2509.00269", "title": "3D-LATTE: Latent Space 3D Editing from Textual Instructions", "abstract": "Despite the recent success of multi-view diffusion models for text/image-based 3D asset generation, instruction-based editing of 3D assets lacks surprisingly far behind the quality of generation models. The main reason is that recent approaches using 2D priors suffer from view-inconsistent editing signals. Going beyond 2D prior distillation methods and multi-view editing strategies, we propose a training-free editing method that operates within the latent space of a native 3D diffusion model, allowing us to directly manipulate 3D geometry. We guide the edit synthesis by blending 3D attention maps from the generation with the source object. Coupled with geometry-aware regularization guidance, a spectral modulation strategy in the Fourier domain and a refinement step for 3D enhancement, our method outperforms previous 3D editing methods enabling high-fidelity and precise edits across a wide range of shapes and semantic manipulations. Our project webpage is https://mparelli.github.io/3d-latte", "date": "2025-08-29", "categories": []} {"id": "2512.10991", "title": "MolSculpt: Sculpting 3D Molecular Geometries from Chemical Syntax", "abstract": "Generating precise 3D molecular geometries is crucial for drug discovery and material science. While prior efforts leverage 1D representations like SELFIES to ensure molecular validity, they fail to fully exploit the rich chemical knowledge entangled within 1D models, leading to a disconnect between 1D syntactic generation and 3D geometric realization. To bridge this gap, we propose MolSculpt, a novel framework that \"sculpts\" 3D molecular geometries from chemical syntax. MolSculpt is built upon a frozen 1D molecular foundation model and a 3D molecular diffusion model. We introduce a set of learnable queries to extract inherent chemical knowledge from the foundation model, and a trainable projector then injects this cross-modal information into the conditioning space of the diffusion model to guide the 3D geometry generation. In this way, our model deeply integrates 1D latent chemical knowledge into the 3D generation process through end-to-end optimization. Experiments demonstrate that MolSculpt achieves state-of-the-art (SOTA) performance in \\textit{de novo} 3D molecule generation and conditional 3D molecule generation, showing superior 3D fidelity and stability on both the GEOM-DRUGS and QM9 datasets. Code is available at https://github.com/SakuraTroyChen/MolSculpt.", "date": "2025-12-09", "categories": []} {"id": "2512.11199", "title": "CADKnitter: Compositional CAD Generation from Text and Geometry Guidance", "abstract": "Crafting computer-aided design (CAD) models has long been a painstaking and time-intensive task, demanding both precision and expertise from designers. With the emergence of 3D generation, this task has undergone a transformative impact, shifting not only from visual fidelity to functional utility but also enabling editable CAD designs. Prior works have achieved early success in single-part CAD generation, which is not well-suited for real-world applications, as multiple parts need to be assembled under semantic and geometric constraints. In this paper, we propose CADKnitter, a compositional CAD generation framework with a geometry-guided diffusion sampling strategy. CADKnitter is able to generate a complementary CAD part that follows both the geometric constraints of the given CAD model and the semantic constraints of the desired design text prompt. We also curate a dataset, so-called KnitCAD, containing over 310,000 samples of CAD models, along with textual prompts and assembly metadata that provide semantic and geometric constraints. Intensive experiments demonstrate that our proposed method outperforms other state-of-the-art baselines by a clear margin.", "date": "2025-12-12", "categories": []} {"id": "2312.08591", "title": "Joint2Human: High-quality 3D Human Generation via Compact Spherical Embedding of 3D Joints", "abstract": "3D human generation is increasingly significant in various applications. However, the direct use of 2D generative methods in 3D generation often results in losing local details, while methods that reconstruct geometry from generated images struggle with global view consistency. In this work, we introduce Joint2Human, a novel method that leverages 2D diffusion models to generate detailed 3D human geometry directly, ensuring both global structure and local details. To achieve this, we employ the Fourier occupancy field (FOF) representation, enabling the direct generation of 3D shapes as preliminary results with 2D generative models. With the proposed high-frequency enhancer and the multi-view recarving strategy, our method can seamlessly integrate the details from different views into a uniform global shape. To better utilize the 3D human prior and enhance control over the generated geometry, we introduce a compact spherical embedding of 3D joints. This allows for an effective guidance of pose during the generation process. Additionally, our method can generate 3D humans guided by textual inputs. Our experimental results demonstrate the capability of our method to ensure global structure, local details, high resolution, and low computational cost simultaneously. More results and the code can be found on our project page at http://cic.tju.edu.cn/faculty/likun/projects/Joint2Human.", "date": "2023-12-14", "categories": []} {"id": "2512.05991", "title": "EmoDiffTalk:Emotion-aware Diffusion for Editable 3D Gaussian Talking Head", "abstract": "Recent photo-realistic 3D talking head via 3D Gaussian Splatting still has significant shortcoming in emotional expression manipulation, especially for fine-grained and expansive dynamics emotional editing using multi-modal control. This paper introduces a new editable 3D Gaussian talking head, i.e. EmoDiffTalk. Our key idea is a novel Emotion-aware Gaussian Diffusion, which includes an action unit (AU) prompt Gaussian diffusion process for fine-grained facial animator, and moreover an accurate text-to-AU emotion controller to provide accurate and expansive dynamic emotional editing using text input. Experiments on public EmoTalk3D and RenderMe-360 datasets demonstrate superior emotional subtlety, lip-sync fidelity, and controllability of our EmoDiffTalk over previous works, establishing a principled pathway toward high-quality, diffusion-driven, multimodal editable 3D talking-head synthesis. To our best knowledge, our EmoDiffTalk is one of the first few 3D Gaussian Splatting talking-head generation framework, especially supporting continuous, multimodal emotional editing within the AU-based expression space.", "date": "2025-11-30", "categories": []} {"id": "2512.10860", "title": "SWiT-4D: Sliding-Window Transformer for Lossless and Parameter-Free Temporal 4D Generation", "abstract": "Despite significant progress in 4D content generation, the conversion of monocular videos into high-quality animated 3D assets with explicit 4D meshes remains considerably challenging. The scarcity of large-scale, naturally captured 4D mesh datasets further limits the ability to train generalizable video-to-4D models from scratch in a purely data-driven manner. Meanwhile, advances in image-to-3D generation, supported by extensive datasets, offer powerful prior models that can be leveraged. To better utilize these priors while minimizing reliance on 4D supervision, we introduce SWiT-4D, a Sliding-Window Transformer for lossless, parameter-free temporal 4D mesh generation. SWiT-4D integrates seamlessly with any Diffusion Transformer (DiT)-based image-to-3D generator, adding spatial-temporal modeling across video frames while preserving the original single-image forward process, enabling 4D mesh reconstruction from videos of arbitrary length. To recover global translation, we further introduce an optimization-based trajectory module tailored for static-camera monocular videos. SWiT-4D demonstrates strong data efficiency: with only a single short (<10s) video for fine-tuning, it achieves high-fidelity geometry and stable temporal consistency, indicating practical deployability under extremely limited 4D supervision. Comprehensive experiments on both in-domain zoo-test sets and challenging out-of-domain benchmarks (C4D, Objaverse, and in-the-wild videos) show that SWiT-4D consistently outperforms existing baselines in temporal smoothness. Project page: https://animotionlab.github.io/SWIT4D/", "date": "2025-12-11", "categories": []} {"id": "2512.10949", "title": "Are We Ready for RL in Text-to-3D Generation? A Progressive Investigation", "abstract": "Reinforcement learning (RL), earlier proven to be effective in large language and multi-modal models, has been successfully extended to enhance 2D image generation recently. However, applying RL to 3D generation remains largely unexplored due to the higher spatial complexity of 3D objects, which require globally consistent geometry and fine-grained local textures. This makes 3D generation significantly sensitive to reward designs and RL algorithms. To address these challenges, we conduct the first systematic study of RL for text-to-3D autoregressive generation across several dimensions. (1) Reward designs: We evaluate reward dimensions and model choices, showing that alignment with human preference is crucial, and that general multi-modal models provide robust signal for 3D attributes. (2) RL algorithms: We study GRPO variants, highlighting the effectiveness of token-level optimization, and further investigate the scaling of training data and iterations. (3) Text-to-3D Benchmarks: Since existing benchmarks fail to measure implicit reasoning abilities in 3D generation models, we introduce MME-3DR. (4) Advanced RL paradigms: Motivated by the natural hierarchy of 3D generation, we propose Hi-GRPO, which optimizes the global-to-local hierarchical 3D generation through dedicated reward ensembles. Based on these insights, we develop AR3D-R1, the first RL-enhanced text-to-3D model, expert from coarse shape to texture refinement. We hope this study provides insights into RL-driven reasoning for 3D generation. Code is released at https://github.com/Ivan-Tang-3D/3DGen-R1.", "date": "2025-12-11", "categories": []} {"id": "2512.10957", "title": "SceneMaker: Open-set 3D Scene Generation with Decoupled De-occlusion and Pose Estimation Model", "abstract": "We propose a decoupled 3D scene generation framework called SceneMaker in this work. Due to the lack of sufficient open-set de-occlusion and pose estimation priors, existing methods struggle to simultaneously produce high-quality geometry and accurate poses under severe occlusion and open-set settings. To address these issues, we first decouple the de-occlusion model from 3D object generation, and enhance it by leveraging image datasets and collected de-occlusion datasets for much more diverse open-set occlusion patterns. Then, we propose a unified pose estimation model that integrates global and local mechanisms for both self-attention and cross-attention to improve accuracy. Besides, we construct an open-set 3D scene dataset to further extend the generalization of the pose estimation model. Comprehensive experiments demonstrate the superiority of our decoupled framework on both indoor and open-set scenes. Our codes and datasets is released at https://idea-research.github.io/SceneMaker/.", "date": "2025-12-11", "categories": []} {"id": "2512.09164", "title": "WonderZoom: Multi-Scale 3D World Generation", "abstract": "We present WonderZoom, a novel approach to generating 3D scenes with contents across multiple spatial scales from a single image. Existing 3D world generation models remain limited to single-scale synthesis and cannot produce coherent scene contents at varying granularities. The fundamental challenge is the lack of a scale-aware 3D representation capable of generating and rendering content with largely different spatial sizes. WonderZoom addresses this through two key innovations: (1) scale-adaptive Gaussian surfels for generating and real-time rendering of multi-scale 3D scenes, and (2) a progressive detail synthesizer that iteratively generates finer-scale 3D contents. Our approach enables users to \"zoom into\" a 3D region and auto-regressively synthesize previously non-existent fine details from landscapes to microscopic features. Experiments demonstrate that WonderZoom significantly outperforms state-of-the-art video and 3D models in both quality and alignment, enabling multi-scale 3D world creation from a single image. We show video results and an interactive viewer of generated multi-scale 3D worlds in https://wonderzoom.github.io/", "date": "2025-12-09", "categories": []} {"id": "2512.09617", "title": "FROMAT: Multiview Material Appearance Transfer via Few-Shot Self-Attention Adaptation", "abstract": "Multiview diffusion models have rapidly emerged as a powerful tool for content creation with spatial consistency across viewpoints, offering rich visual realism without requiring explicit geometry and appearance representation. However, compared to meshes or radiance fields, existing multiview diffusion models offer limited appearance manipulation, particularly in terms of material, texture, or style.\n In this paper, we present a lightweight adaptation technique for appearance transfer in multiview diffusion models. Our method learns to combine object identity from an input image with appearance cues rendered in a separate reference image, producing multi-view-consistent output that reflects the desired materials, textures, or styles. This allows explicit specification of appearance parameters at generation time while preserving the underlying object geometry and view coherence. We leverage three diffusion denoising processes responsible for generating the original object, the reference, and the target images, and perform reverse sampling to aggregate a small subset of layer-wise self-attention features from the object and the reference to influence the target generation. Our method requires only a few training examples to introduce appearance awareness to pretrained multiview models. The experiments show that our method provides a simple yet effective way toward multiview generation with diverse appearance, advocating the adoption of implicit generative 3D representations in practice.", "date": "2025-12-10", "categories": []} {"id": "2506.18331", "title": "End-to-End Fine-Tuning of 3D Texture Generation using Differentiable Rewards", "abstract": "While recent 3D generative models can produce high-quality texture images, they often fail to capture human preferences or meet task-specific requirements. Moreover, a core challenge in the 3D texture generation domain is that most existing approaches rely on repeated calls to 2D text-to-image generative models, which lack an inherent understanding of the 3D structure of the input 3D mesh object. To alleviate these issues, we propose an end-to-end differentiable, reinforcement-learning-free framework that embeds human feedback, expressed as differentiable reward functions, directly into the 3D texture synthesis pipeline. By back-propagating preference signals through both geometric and appearance modules of the proposed framework, our method generates textures that respect the 3D geometry structure and align with desired criteria. To demonstrate its versatility, we introduce three novel geometry-aware reward functions, which offer a more controllable and interpretable pathway for creating high-quality 3D content from natural language. By conducting qualitative, quantitative, and user-preference evaluations against state-of-the-art methods, we demonstrate that our proposed strategy consistently outperforms existing approaches. Our implementation code is publicly available at: https://github.com/AHHHZ975/Differentiable-Texture-Learning", "date": "2025-06-23", "categories": []} {"id": "2508.15769", "title": "SceneGen: Single-Image 3D Scene Generation in One Feedforward Pass", "abstract": "3D content generation has recently attracted significant research interest, driven by its critical applications in VR/AR and embodied AI. In this work, we tackle the challenging task of synthesizing multiple 3D assets within a single scene image. Concretely, our contributions are fourfold: (i) we present SceneGen, a novel framework that takes a scene image and corresponding object masks as input, simultaneously producing multiple 3D assets with geometry and texture. Notably, SceneGen operates with no need for extra optimization or asset retrieval; (ii) we introduce a novel feature aggregation module that integrates local and global scene information from visual and geometric encoders within the feature extraction module. Coupled with a position head, this enables the generation of 3D assets and their relative spatial positions in a single feedforward pass; (iii) we demonstrate SceneGen's direct extensibility to multi-image input scenarios. Despite being trained solely on single-image inputs, our architecture yields improved generation performance when multiple images are provided; and (iv) extensive quantitative and qualitative evaluations confirm the efficiency and robustness of our approach. We believe this paradigm offers a novel solution for high-quality 3D content generation, potentially advancing its practical applications in downstream tasks. The code and model will be publicly available at: https://mengmouxu.github.io/SceneGen.", "date": "2025-08-21", "categories": []} {"id": "2512.07514", "title": "MeshRipple: Structured Autoregressive Generation of Artist-Meshes", "abstract": "Meshes serve as a primary representation for 3D assets. Autoregressive mesh generators serialize faces into sequences and train on truncated segments with sliding-window inference to cope with memory limits. However, this mismatch breaks long-range geometric dependencies, producing holes and fragmented components. To address this critical limitation, we introduce MeshRipple, which expands a mesh outward from an active generation frontier, akin to a ripple on a surface. MeshRipple rests on three key innovations: a frontier-aware BFS tokenization that aligns the generation order with surface topology; an expansive prediction strategy that maintains coherent, connected surface growth; and a sparse-attention global memory that provides an effectively unbounded receptive field to resolve long-range topological dependencies. This integrated design enables MeshRipple to generate meshes with high surface fidelity and topological completeness, outperforming strong recent baselines.", "date": "2025-12-08", "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": "2506.13558", "title": "X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability", "abstract": "Diffusion models are advancing autonomous driving by enabling realistic data synthesis, predictive end-to-end planning, and closed-loop simulation, with a primary focus on temporally consistent generation. However, large-scale 3D scene generation requiring spatial coherence remains underexplored. In this paper, we present X-Scene, a novel framework for large-scale driving scene generation that achieves geometric intricacy, appearance fidelity, and flexible controllability. Specifically, X-Scene supports multi-granular control, including low-level layout conditioning driven by user input or text for detailed scene composition, and high-level semantic guidance informed by user intent and LLM-enriched prompts for efficient customization. To enhance geometric and visual fidelity, we introduce a unified pipeline that sequentially generates 3D semantic occupancy and corresponding multi-view images and videos, ensuring alignment and temporal consistency across modalities. We further extend local regions into large-scale scenes via consistency-aware outpainting, which extrapolates occupancy and images from previously generated areas to maintain spatial and visual coherence. The resulting scenes are lifted into high-quality 3DGS representations, supporting diverse applications such as simulation and scene exploration. Extensive experiments demonstrate that X-Scene substantially advances controllability and fidelity in large-scale scene generation, empowering data generation and simulation for autonomous driving.", "date": "2025-06-16", "categories": []} {"id": "2510.23880", "title": "TRELLISWorld: Training-Free World Generation from Object Generators", "abstract": "Text-driven 3D scene generation holds promise for a wide range of applications, from virtual prototyping to AR/VR and simulation. However, existing methods are often constrained to single-object generation, require domain-specific training, or lack support for full 360-degree viewability. In this work, we present a training-free approach to 3D scene synthesis by repurposing general-purpose text-to-3D object diffusion models as modular tile generators. We reformulate scene generation as a multi-tile denoising problem, where overlapping 3D regions are independently generated and seamlessly blended via weighted averaging. This enables scalable synthesis of large, coherent scenes while preserving local semantic control. Our method eliminates the need for scene-level datasets or retraining, relies on minimal heuristics, and inherits the generalization capabilities of object-level priors. We demonstrate that our approach supports diverse scene layouts, efficient generation, and flexible editing, establishing a simple yet powerful foundation for general-purpose, language-driven 3D scene construction.", "date": "2025-10-27", "categories": []} {"id": "2511.20415", "title": "MajutsuCity: Language-driven Aesthetic-adaptive City Generation with Controllable 3D Assets and Layouts", "abstract": "Generating realistic 3D cities is fundamental to world models, virtual reality, and game development, where an ideal urban scene must satisfy both stylistic diversity, fine-grained, and controllability. However, existing methods struggle to balance the creative flexibility offered by text-based generation with the object-level editability enabled by explicit structural representations. We introduce MajutsuCity, a natural language-driven and aesthetically adaptive framework for synthesizing structurally consistent and stylistically diverse 3D urban scenes. MajutsuCity represents a city as a composition of controllable layouts, assets, and materials, and operates through a four-stage pipeline. To extend controllability beyond initial generation, we further integrate MajutsuAgent, an interactive language-grounded editing agent} that supports five object-level operations. To support photorealistic and customizable scene synthesis, we also construct MajutsuDataset, a high-quality multimodal dataset} containing 2D semantic layouts and height maps, diverse 3D building assets, and curated PBR materials and skyboxes, each accompanied by detailed annotations. Meanwhile, we develop a practical set of evaluation metrics, covering key dimensions such as structural consistency, scene complexity, material fidelity, and lighting atmosphere. Extensive experiments demonstrate MajutsuCity reduces layout FID by 83.7% compared with CityDreamer and by 20.1% over CityCraft. Our method ranks first across all AQS and RDR scores, outperforming existing methods by a clear margin. These results confirm MajutsuCity as a new state-of-the-art in geometric fidelity, stylistic adaptability, and semantic controllability for 3D city generation. We expect our framework can inspire new avenues of research in 3D city generation. Our project page: https://longhz140516.github.io/MajutsuCity/.", "date": "2025-11-25", "categories": []} {"id": "2512.06158", "title": "Tracking-Guided 4D Generation: Foundation-Tracker Motion Priors for 3D Model Animation", "abstract": "Generating dynamic 4D objects from sparse inputs is difficult because it demands joint preservation of appearance and motion coherence across views and time while suppressing artifacts and temporal drift. We hypothesize that the view discrepancy arises from supervision limited to pixel- or latent-space video-diffusion losses, which lack explicitly temporally aware, feature-level tracking guidance. We present \\emph{Track4DGen}, a two-stage framework that couples a multi-view video diffusion model with a foundation point tracker and a hybrid 4D Gaussian Splatting (4D-GS) reconstructor. The central idea is to explicitly inject tracker-derived motion priors into intermediate feature representations for both multi-view video generation and 4D-GS. In Stage One, we enforce dense, feature-level point correspondences inside the diffusion generator, producing temporally consistent features that curb appearance drift and enhance cross-view coherence. In Stage Two, we reconstruct a dynamic 4D-GS using a hybrid motion encoding that concatenates co-located diffusion features (carrying Stage-One tracking priors) with Hex-plane features, and augment them with 4D Spherical Harmonics for higher-fidelity dynamics modeling. \\emph{Track4DGen} surpasses baselines on both multi-view video generation and 4D generation benchmarks, yielding temporally stable, text-editable 4D assets. Lastly, we curate \\emph{Sketchfab28}, a high-quality dataset for benchmarking object-centric 4D generation and fostering future research.", "date": "2025-12-05", "categories": []} {"id": "2512.07459", "title": "Human Geometry Distribution for 3D Animation Generation", "abstract": "Generating realistic human geometry animations remains a challenging task, as it requires modeling natural clothing dynamics with fine-grained geometric details under limited data. To address these challenges, we propose two novel designs. First, we propose a compact distribution-based latent representation that enables efficient and high-quality geometry generation. We improve upon previous work by establishing a more uniform mapping between SMPL and avatar geometries. Second, we introduce a generative animation model that fully exploits the diversity of limited motion data. We focus on short-term transitions while maintaining long-term consistency through an identity-conditioned design. These two designs formulate our method as a two-stage framework: the first stage learns a latent space, while the second learns to generate animations within this latent space. We conducted experiments on both our latent space and animation model. We demonstrate that our latent space produces high-fidelity human geometry surpassing previous methods ($90\\%$ lower Chamfer Dist.). The animation model synthesizes diverse animations with detailed and natural dynamics ($2.2 \\times$ higher user study score), achieving the best results across all evaluation metrics.", "date": "2025-12-08", "categories": []} {"id": "2512.07628", "title": "MoCA: Mixture-of-Components Attention for Scalable Compositional 3D Generation", "abstract": "Compositionality is critical for 3D object and scene generation, but existing part-aware 3D generation methods suffer from poor scalability due to quadratic global attention costs when increasing the number of components. In this work, we present MoCA, a compositional 3D generative model with two key designs: (1) importance-based component routing that selects top-k relevant components for sparse global attention, and (2) unimportant components compression that preserve contextual priors of unselected components while reducing computational complexity of global attention. With these designs, MoCA enables efficient, fine-grained compositional 3D asset creation with scalable number of components. Extensive experiments show MoCA outperforms baselines on both compositional object and scene generation tasks. Project page: https://lizhiqi49.github.io/MoCA", "date": "2025-12-08", "categories": []} {"id": "2503.16848", "title": "HSM: Hierarchical Scene Motifs for Multi-Scale Indoor Scene Generation", "abstract": "Despite advances in indoor 3D scene layout generation, synthesizing scenes with dense object arrangements remains challenging. Existing methods focus on large furniture while neglecting smaller objects, resulting in unrealistically empty scenes. Those that place small objects typically do not honor arrangement specifications, resulting in largely random placement not following the text description. We present Hierarchical Scene Motifs (HSM): a hierarchical framework for indoor scene generation with dense object arrangements across spatial scales. Indoor scenes are inherently hierarchical, with surfaces supporting objects at different scales, from large furniture on floors to smaller objects on tables and shelves. HSM embraces this hierarchy and exploits recurring cross-scale spatial patterns to generate complex and realistic scenes in a unified manner. Our experiments show that HSM outperforms existing methods by generating scenes that better conform to user input across room types and spatial configurations. Project website is available at https://3dlg-hcvc.github.io/hsm .", "date": "2025-03-21", "categories": []} {"id": "2512.01204", "title": "TabletopGen: Instance-Level Interactive 3D Tabletop Scene Generation from Text or Single Image", "abstract": "Generating high-fidelity, physically interactive 3D simulated tabletop scenes is essential for embodied AI -- especially for robotic manipulation policy learning and data synthesis. However, current text- or image-driven 3D scene generation methods mainly focus on large-scale scenes, struggling to capture the high-density layouts and complex spatial relations that characterize tabletop scenes. To address these challenges, we propose TabletopGen, a training-free, fully automatic framework that generates diverse, instance-level interactive 3D tabletop scenes. TabletopGen accepts a reference image as input, which can be synthesized by a text-to-image model to enhance scene diversity. We then perform instance segmentation and completion on the reference to obtain per-instance images. Each instance is reconstructed into a 3D model followed by canonical coordinate alignment. The aligned 3D models then undergo pose and scale estimation before being assembled into a collision-free, simulation-ready tabletop scene. A key component of our framework is a novel pose and scale alignment approach that decouples the complex spatial reasoning into two stages: a Differentiable Rotation Optimizer for precise rotation recovery and a Top-view Spatial Alignment mechanism for robust translation and scale estimation, enabling accurate 3D reconstruction from 2D reference. Extensive experiments and user studies show that TabletopGen achieves state-of-the-art performance, markedly surpassing existing methods in visual fidelity, layout accuracy, and physical plausibility, capable of generating realistic tabletop scenes with rich stylistic and spatial diversity. Our code will be publicly available.", "date": "2025-12-01", "categories": []} {"id": "2512.04248", "title": "MVRoom: Controllable 3D Indoor Scene Generation with Multi-View Diffusion Models", "abstract": "We introduce MVRoom, a controllable novel view synthesis (NVS) pipeline for 3D indoor scenes that uses multi-view diffusion conditioned on a coarse 3D layout. MVRoom employs a two-stage design in which the 3D layout is used throughout to enforce multi-view consistency. The first stage employs novel representations to effectively bridge the 3D layout and consistent image-based condition signals for multi-view generation. The second stage performs image-conditioned multi-view generation, incorporating a layout-aware epipolar attention mechanism to enhance multi-view consistency during the diffusion process. Additionally, we introduce an iterative framework that generates 3D scenes with varying numbers of objects and scene complexities by recursively performing multi-view generation (MVRoom), supporting text-to-scene generation. Experimental results demonstrate that our approach achieves high-fidelity and controllable 3D scene generation for NVS, outperforming state-of-the-art baseline methods both quantitatively and qualitatively. Ablation studies further validate the effectiveness of key components within our generation pipeline.", "date": "2025-12-03", "categories": []} {"id": "2512.04786", "title": "LaFiTe: A Generative Latent Field for 3D Native Texturing", "abstract": "Generating high-fidelity, seamless textures directly on 3D surfaces, what we term 3D-native texturing, remains a fundamental open challenge, with the potential to overcome long-standing limitations of UV-based and multi-view projection methods. However, existing native approaches are constrained by the absence of a powerful and versatile latent representation, which severely limits the fidelity and generality of their generated textures. We identify this representation gap as the principal barrier to further progress. We introduce LaFiTe, a framework that addresses this challenge by learning to generate textures as a 3D generative sparse latent color field. At its core, LaFiTe employs a variational autoencoder (VAE) to encode complex surface appearance into a sparse, structured latent space, which is subsequently decoded into a continuous color field. This representation achieves unprecedented fidelity, exceeding state-of-the-art methods by >10 dB PSNR in reconstruction, by effectively disentangling texture appearance from mesh topology and UV parameterization. Building upon this strong representation, a conditional rectified-flow model synthesizes high-quality, coherent textures across diverse styles and geometries. Extensive experiments demonstrate that LaFiTe not only sets a new benchmark for 3D-native texturing but also enables flexible downstream applications such as material synthesis and texture super-resolution, paving the way for the next generation of 3D content creation workflows.", "date": "2025-12-04", "categories": []} {"id": "2512.04830", "title": "FreeGen: Feed-Forward Reconstruction-Generation Co-Training for Free-Viewpoint Driving Scene Synthesis", "abstract": "Closed-loop simulation and scalable pre-training for autonomous driving require synthesizing free-viewpoint driving scenes. However, existing datasets and generative pipelines rarely provide consistent off-trajectory observations, limiting large-scale evaluation and training. While recent generative models demonstrate strong visual realism, they struggle to jointly achieve interpolation consistency and extrapolation realism without per-scene optimization. To address this, we propose FreeGen, a feed-forward reconstruction-generation co-training framework for free-viewpoint driving scene synthesis. The reconstruction model provides stable geometric representations to ensure interpolation consistency, while the generation model performs geometry-aware enhancement to improve realism at unseen viewpoints. Through co-training, generative priors are distilled into the reconstruction model to improve off-trajectory rendering, and the refined geometry in turn offers stronger structural guidance for generation. Experiments demonstrate that FreeGen achieves state-of-the-art performance for free-viewpoint driving scene synthesis.", "date": "2025-12-04", "categories": []} {"id": "2512.05044", "title": "Joint 3D Geometry Reconstruction and Motion Generation for 4D Synthesis from a Single Image", "abstract": "Generating interactive and dynamic 4D scenes from a single static image remains a core challenge. Most existing generate-then-reconstruct and reconstruct-then-generate methods decouple geometry from motion, causing spatiotemporal inconsistencies and poor generalization. To address these, we extend the reconstruct-then-generate framework to jointly perform Motion generation and geometric Reconstruction for 4D Synthesis (MoRe4D). We first introduce TrajScene-60K, a large-scale dataset of 60,000 video samples with dense point trajectories, addressing the scarcity of high-quality 4D scene data. Based on this, we propose a diffusion-based 4D Scene Trajectory Generator (4D-STraG) to jointly generate geometrically consistent and motion-plausible 4D point trajectories. To leverage single-view priors, we design a depth-guided motion normalization strategy and a motion-aware module for effective geometry and dynamics integration. We then propose a 4D View Synthesis Module (4D-ViSM) to render videos with arbitrary camera trajectories from 4D point track representations. Experiments show that MoRe4D generates high-quality 4D scenes with multi-view consistency and rich dynamic details from a single image. Code: https://github.com/Zhangyr2022/MoRe4D.", "date": "2025-12-04", "categories": []} {"id": "2410.18084", "title": "DynamicCity: Large-Scale 4D Occupancy Generation from Dynamic Scenes", "abstract": "Urban scene generation has been developing rapidly recently. However, existing methods primarily focus on generating static and single-frame scenes, overlooking the inherently dynamic nature of real-world driving environments. In this work, we introduce DynamicCity, a novel 4D occupancy generation framework capable of generating large-scale, high-quality dynamic 4D scenes with semantics. DynamicCity mainly consists of two key models. 1) A VAE model for learning HexPlane as the compact 4D representation. Instead of using naive averaging operations, DynamicCity employs a novel Projection Module to effectively compress 4D features into six 2D feature maps for HexPlane construction, which significantly enhances HexPlane fitting quality (up to 12.56 mIoU gain). Furthermore, we utilize an Expansion & Squeeze Strategy to reconstruct 3D feature volumes in parallel, which improves both network training efficiency and reconstruction accuracy than naively querying each 3D point (up to 7.05 mIoU gain, 2.06x training speedup, and 70.84% memory reduction). 2) A DiT-based diffusion model for HexPlane generation. To make HexPlane feasible for DiT generation, a Padded Rollout Operation is proposed to reorganize all six feature planes of the HexPlane as a squared 2D feature map. In particular, various conditions could be introduced in the diffusion or sampling process, supporting versatile 4D generation applications, such as trajectory- and command-driven generation, inpainting, and layout-conditioned generation. Extensive experiments on the CarlaSC and Waymo datasets demonstrate that DynamicCity significantly outperforms existing state-of-the-art 4D occupancy generation methods across multiple metrics. The code and models have been released to facilitate future research.", "date": "2024-10-23", "categories": []} {"id": "2412.09997", "title": "GT23D-Bench: A Comprehensive General Text-to-3D Generation Benchmark", "abstract": "Text-to-3D (T23D) generation has emerged as a crucial visual generation task, aiming at synthesizing 3D content from textual descriptions. Studies of this task are currently shifting from per-scene T23D, which requires optimization of the model for every content generated, to General T23D (GT23D), which requires only one pre-trained model to generate different content without re-optimization, for more generalized and efficient 3D generation. Despite notable advancements, GT23D is severely bottlenecked by two interconnected challenges: the lack of high-quality, large-scale training data and the prevalence of evaluation metrics that overlook intrinsic 3D properties. Existing datasets often suffer from incomplete annotations, noisy organization, and inconsistent quality, while current evaluations rely heavily on 2D image-text similarity or scoring, failing to thoroughly assess 3D geometric integrity and semantic relevance. To address these fundamental gaps, we introduce GT23D-Bench, the first comprehensive benchmark specifically designed for GT23D training and evaluation. We first construct a high-quality dataset of 400K 3D assets, featuring diverse visual annotations (70M+ visual samples) and multi-granularity hierarchical captions (1M+ descriptions) to foster robust semantic learning. Second, we propose a comprehensive evaluation suite with 10 metrics assessing both text-3D alignment and 3D visual quality at multiple levels. Crucially, we demonstrate through rigorous experiments that our proposed metrics exhibit significantly higher correlation with human judgment compared to existing methods. Our in-depth analysis of eight leading GT23D models using this benchmark provides the community with critical insights into current model capabilities and their shared failure modes. GT23D-Bench will be publicly available to facilitate rigorous and reproducible research.", "date": "2024-12-13", "categories": []} {"id": "2512.03052", "title": "LATTICE: Democratize High-Fidelity 3D Generation at Scale", "abstract": "We present LATTICE, a new framework for high-fidelity 3D asset generation that bridges the quality and scalability gap between 3D and 2D generative models. While 2D image synthesis benefits from fixed spatial grids and well-established transformer architectures, 3D generation remains fundamentally more challenging due to the need to predict both spatial structure and detailed geometric surfaces from scratch. These challenges are exacerbated by the computational complexity of existing 3D representations and the lack of structured and scalable 3D asset encoding schemes. To address this, we propose VoxSet, a semi-structured representation that compresses 3D assets into a compact set of latent vectors anchored to a coarse voxel grid, enabling efficient and position-aware generation. VoxSet retains the simplicity and compression advantages of prior VecSet methods while introducing explicit structure into the latent space, allowing positional embeddings to guide generation and enabling strong token-level test-time scaling. Built upon this representation, LATTICE adopts a two-stage pipeline: first generating a sparse voxelized geometry anchor, then producing detailed geometry using a rectified flow transformer. Our method is simple at its core, but supports arbitrary resolution decoding, low-cost training, and flexible inference schemes, achieving state-of-the-art performance on various aspects, and offering a significant step toward scalable, high-quality 3D asset creation.", "date": "2025-11-24", "categories": []} {"id": "2512.03566", "title": "GAOT: Generating Articulated Objects Through Text-Guided Diffusion Models", "abstract": "Articulated object generation has seen increasing advancements, yet existing models often lack the ability to be conditioned on text prompts. To address the significant gap between textual descriptions and 3D articulated object representations, we propose GAOT, a three-phase framework that generates articulated objects from text prompts, leveraging diffusion models and hypergraph learning in a three-step process. First, we fine-tune a point cloud generation model to produce a coarse representation of objects from text prompts. Given the inherent connection between articulated objects and graph structures, we design a hypergraph-based learning method to refine these coarse representations, representing object parts as graph vertices. Finally, leveraging a diffusion model, the joints of articulated objects-represented as graph edges-are generated based on the object parts. Extensive qualitative and quantitative experiments on the PartNet-Mobility dataset demonstrate the effectiveness of our approach, achieving superior performance over previous methods.", "date": "2025-12-03", "categories": []} {"id": "2512.03683", "title": "GaussianBlender: Instant Stylization of 3D Gaussians with Disentangled Latent Spaces", "abstract": "3D stylization is central to game development, virtual reality, and digital arts, where the demand for diverse assets calls for scalable methods that support fast, high-fidelity manipulation. Existing text-to-3D stylization methods typically distill from 2D image editors, requiring time-intensive per-asset optimization and exhibiting multi-view inconsistency due to the limitations of current text-to-image models, which makes them impractical for large-scale production. In this paper, we introduce GaussianBlender, a pioneering feed-forward framework for text-driven 3D stylization that performs edits instantly at inference. Our method learns structured, disentangled latent spaces with controlled information sharing for geometry and appearance from spatially-grouped 3D Gaussians. A latent diffusion model then applies text-conditioned edits on these learned representations. Comprehensive evaluations show that GaussianBlender not only delivers instant, high-fidelity, geometry-preserving, multi-view consistent stylization, but also surpasses methods that require per-instance test-time optimization - unlocking practical, democratized 3D stylization at scale.", "date": "2025-12-03", "categories": []} {"id": "2502.01846", "title": "UVGS: Reimagining Unstructured 3D Gaussian Splatting using UV Mapping", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated superior quality in modeling 3D objects and scenes. However, generating 3DGS remains challenging due to their discrete, unstructured, and permutation-invariant nature. In this work, we present a simple yet effective method to overcome these challenges. We utilize spherical mapping to transform 3DGS into a structured 2D representation, termed UVGS. UVGS can be viewed as multi-channel images, with feature dimensions as a concatenation of Gaussian attributes such as position, scale, color, opacity, and rotation. We further find that these heterogeneous features can be compressed into a lower-dimensional (e.g., 3-channel) shared feature space using a carefully designed multi-branch network. The compressed UVGS can be treated as typical RGB images. Remarkably, we discover that typical VAEs trained with latent diffusion models can directly generalize to this new representation without additional training. Our novel representation makes it effortless to leverage foundational 2D models, such as diffusion models, to directly model 3DGS. Additionally, one can simply increase the 2D UV resolution to accommodate more Gaussians, making UVGS a scalable solution compared to typical 3D backbones. This approach immediately unlocks various novel generation applications of 3DGS by inherently utilizing the already developed superior 2D generation capabilities. In our experiments, we demonstrate various unconditional, conditional generation, and inpainting applications of 3DGS based on diffusion models, which were previously non-trivial.", "date": "2025-02-03", "categories": []} {"id": "2512.02993", "title": "TEXTRIX: Latent Attribute Grid for Native Texture Generation and Beyond", "abstract": "Prevailing 3D texture generation methods, which often rely on multi-view fusion, are frequently hindered by inter-view inconsistencies and incomplete coverage of complex surfaces, limiting the fidelity and completeness of the generated content. To overcome these challenges, we introduce TEXTRIX, a native 3D attribute generation framework for high-fidelity texture synthesis and downstream applications such as precise 3D part segmentation. Our approach constructs a latent 3D attribute grid and leverages a Diffusion Transformer equipped with sparse attention, enabling direct coloring of 3D models in volumetric space and fundamentally avoiding the limitations of multi-view fusion. Built upon this native representation, the framework naturally extends to high-precision 3D segmentation by training the same architecture to predict semantic attributes on the grid. Extensive experiments demonstrate state-of-the-art performance on both tasks, producing seamless, high-fidelity textures and accurate 3D part segmentation with precise boundaries.", "date": "2025-12-02", "categories": []} {"id": "2512.03018", "title": "AutoBrep: Autoregressive B-Rep Generation with Unified Topology and Geometry", "abstract": "The boundary representation (B-Rep) is the standard data structure used in Computer-Aided Design (CAD) for defining solid models. Despite recent progress, directly generating B-Reps end-to-end with precise geometry and watertight topology remains a challenge. This paper presents AutoBrep, a novel Transformer model that autoregressively generates B-Reps with high quality and validity. AutoBrep employs a unified tokenization scheme that encodes both geometric and topological characteristics of a B-Rep model as a sequence of discrete tokens. Geometric primitives (i.e., surfaces and curves) are encoded as latent geometry tokens, and their structural relationships are defined as special topological reference tokens. Sequence order in AutoBrep naturally follows a breadth first traversal of the B-Rep face adjacency graph. At inference time, neighboring faces and edges along with their topological structure are progressively generated. Extensive experiments demonstrate the advantages of our unified representation when coupled with next-token prediction for B-Rep generation. AutoBrep outperforms baselines with better quality and watertightness. It is also highly scalable to complex solids with good fidelity and inference speed. We further show that autocompleting B-Reps is natively supported through our unified tokenization, enabling user-controllable CAD generation with minimal changes. Code is available at https://github.com/AutodeskAILab/AutoBrep.", "date": "2025-12-02", "categories": []} {"id": "2512.03045", "title": "CAMEO: Correspondence-Attention Alignment for Multi-View Diffusion Models", "abstract": "Multi-view diffusion models have recently emerged as a powerful paradigm for novel view synthesis, yet the underlying mechanism that enables their view-consistency remains unclear. In this work, we first verify that the attention maps of these models acquire geometric correspondence throughout training, attending to the geometrically corresponding regions across reference and target views for view-consistent generation. However, this correspondence signal remains incomplete, with its accuracy degrading under large viewpoint changes. Building on these findings, we introduce CAMEO, a simple yet effective training technique that directly supervises attention maps using geometric correspondence to enhance both the training efficiency and generation quality of multi-view diffusion models. Notably, supervising a single attention layer is sufficient to guide the model toward learning precise correspondences, thereby preserving the geometry and structure of reference images, accelerating convergence, and improving novel view synthesis performance. CAMEO reduces the number of training iterations required for convergence by half while achieving superior performance at the same iteration counts. We further demonstrate that CAMEO is model-agnostic and can be applied to any multi-view diffusion model.", "date": "2025-12-02", "categories": []} {"id": "2505.18151", "title": "WonderPlay: Dynamic 3D Scene Generation from a Single Image and Actions", "abstract": "WonderPlay is a novel framework integrating physics simulation with video generation for generating action-conditioned dynamic 3D scenes from a single image. While prior works are restricted to rigid body or simple elastic dynamics, WonderPlay features a hybrid generative simulator to synthesize a wide range of 3D dynamics. The hybrid generative simulator first uses a physics solver to simulate coarse 3D dynamics, which subsequently conditions a video generator to produce a video with finer, more realistic motion. The generated video is then used to update the simulated dynamic 3D scene, closing the loop between the physics solver and the video generator. This approach enables intuitive user control to be combined with the accurate dynamics of physics-based simulators and the expressivity of diffusion-based video generators. Experimental results demonstrate that WonderPlay enables users to interact with various scenes of diverse content, including cloth, sand, snow, liquid, smoke, elastic, and rigid bodies -- all using a single image input. Code will be made public. Project website: https://kyleleey.github.io/WonderPlay/", "date": "2025-05-23", "categories": []} {"id": "2505.22643", "title": "SPIRAL: Semantic-Aware Progressive LiDAR Scene Generation and Understanding", "abstract": "Leveraging recent diffusion models, LiDAR-based large-scale 3D scene generation has achieved great success. While recent voxel-based approaches can generate both geometric structures and semantic labels, existing range-view methods are limited to producing unlabeled LiDAR scenes. Relying on pretrained segmentation models to predict the semantic maps often results in suboptimal cross-modal consistency. To address this limitation while preserving the advantages of range-view representations, such as computational efficiency and simplified network design, we propose Spiral, a novel range-view LiDAR diffusion model that simultaneously generates depth, reflectance images, and semantic maps. Furthermore, we introduce novel semantic-aware metrics to evaluate the quality of the generated labeled range-view data. Experiments on the SemanticKITTI and nuScenes datasets demonstrate that Spiral achieves state-of-the-art performance with the smallest parameter size, outperforming two-step methods that combine the generative and segmentation models. Additionally, we validate that range images generated by Spiral can be effectively used for synthetic data augmentation in the downstream segmentation training, significantly reducing the labeling effort on LiDAR data.", "date": "2025-05-28", "categories": []} {"id": "2511.21309", "title": "CaliTex: Geometry-Calibrated Attention for View-Coherent 3D Texture Generation", "abstract": "Despite major advances brought by diffusion-based models, current 3D texture generation systems remain hindered by cross-view inconsistency -- textures that appear convincing from one viewpoint often fail to align across others. We find that this issue arises from attention ambiguity, where unstructured full attention is applied indiscriminately across tokens and modalities, causing geometric confusion and unstable appearance-structure coupling. To address this, we introduce CaliTex, a framework of geometry-calibrated attention that explicitly aligns attention with 3D structure. It introduces two modules: Part-Aligned Attention that enforces spatial alignment across semantically matched parts, and Condition-Routed Attention which routes appearance information through geometry-conditioned pathways to maintain spatial fidelity. Coupled with a two-stage diffusion transformer, CaliTex makes geometric coherence an inherent behavior of the network rather than a byproduct of optimization. Empirically, CaliTex produces seamless and view-consistent textures and outperforms both open-source and commercial baselines.", "date": "2025-11-26", "categories": []} {"id": "2511.23172", "title": "Fast Multi-view Consistent 3D Editing with Video Priors", "abstract": "Text-driven 3D editing enables user-friendly 3D object or scene editing with text instructions. Due to the lack of multi-view consistency priors, existing methods typically resort to employing 2D generation or editing models to process each view individually, followed by iterative 2D-3D-2D updating. However, these methods are not only time-consuming but also prone to over-smoothed results because the different editing signals gathered from different views are averaged during the iterative process. In this paper, we propose generative Video Prior based 3D Editing (ViP3DE) to employ the temporal consistency priors from pre-trained video generation models for multi-view consistent 3D editing in a single forward pass. Our key insight is to condition the video generation model on a single edited view to generate other consistent edited views for 3D updating directly, thereby bypassing the iterative editing paradigm. Since 3D updating requires edited views to be paired with specific camera poses, we propose motion-preserved noise blending for the video model to generate edited views at predefined camera poses. In addition, we introduce geometry-aware denoising to further enhance multi-view consistency by integrating 3D geometric priors into video models. Extensive experiments demonstrate that our proposed ViP3DE can achieve high-quality 3D editing results even within a single forward pass, significantly outperforming existing methods in both editing quality and speed.", "date": "2025-11-28", "categories": []} {"id": "2512.00413", "title": "SplatFont3D: Structure-Aware Text-to-3D Artistic Font Generation with Part-Level Style Control", "abstract": "Artistic font generation (AFG) can assist human designers in creating innovative artistic fonts. However, most previous studies primarily focus on 2D artistic fonts in flat design, leaving personalized 3D-AFG largely underexplored. 3D-AFG not only enables applications in immersive 3D environments such as video games and animations, but also may enhance 2D-AFG by rendering 2D fonts of novel views. Moreover, unlike general 3D objects, 3D fonts exhibit precise semantics with strong structural constraints and also demand fine-grained part-level style control. To address these challenges, we propose SplatFont3D, a novel structure-aware text-to-3D AFG framework with 3D Gaussian splatting, which enables the creation of 3D artistic fonts from diverse style text prompts with precise part-level style control. Specifically, we first introduce a Glyph2Cloud module, which progressively enhances both the shapes and styles of 2D glyphs (or components) and produces their corresponding 3D point clouds for Gaussian initialization. The initialized 3D Gaussians are further optimized through interaction with a pretrained 2D diffusion model using score distillation sampling. To enable part-level control, we present a dynamic component assignment strategy that exploits the geometric priors of 3D Gaussians to partition components, while alleviating drift-induced entanglement during 3D Gaussian optimization. Our SplatFont3D provides more explicit and effective part-level style control than NeRF, attaining faster rendering efficiency. Experiments show that our SplatFont3D outperforms existing 3D models for 3D-AFG in style-text consistency, visual quality, and rendering efficiency.", "date": "2025-11-29", "categories": []} {"id": "2512.00493", "title": "CC-FMO: Camera-Conditioned Zero-Shot Single Image to 3D Scene Generation with Foundation Model Orchestration", "abstract": "High-quality 3D scene generation from a single image is crucial for AR/VR and embodied AI applications. Early approaches struggle to generalize due to reliance on specialized models trained on curated small datasets. While recent advancements in large-scale 3D foundation models have significantly enhanced instance-level generation, coherent scene generation remains a challenge, where performance is limited by inaccurate per-object pose estimations and spatial inconsistency. To this end, this paper introduces CC-FMO, a zero-shot, camera-conditioned pipeline for single-image to 3D scene generation that jointly conforms to the object layout in input image and preserves instance fidelity. CC-FMO employs a hybrid instance generator that combines semantics-aware vector-set representation with detail-rich structured latent representation, yielding object geometries that are both semantically plausible and high-quality. Furthermore, CC-FMO enables the application of foundational pose estimation models in the scene generation task via a simple yet effective camera-conditioned scale-solving algorithm, to enforce scene-level coherence. Extensive experiments demonstrate that CC-FMO consistently generates high-fidelity camera-aligned compositional scenes, outperforming all state-of-the-art methods.", "date": "2025-11-29", "categories": []} {"id": "2507.12465", "title": "PhysX-3D: Physical-Grounded 3D Asset Generation", "abstract": "3D modeling is moving from virtual to physical. Existing 3D generation primarily emphasizes geometries and textures while neglecting physical-grounded modeling. Consequently, despite the rapid development of 3D generative models, the synthesized 3D assets often overlook rich and important physical properties, hampering their real-world application in physical domains like simulation and embodied AI. As an initial attempt to address this challenge, we propose \\textbf{PhysX-3D}, an end-to-end paradigm for physical-grounded 3D asset generation. 1) To bridge the critical gap in physics-annotated 3D datasets, we present PhysXNet - the first physics-grounded 3D dataset systematically annotated across five foundational dimensions: absolute scale, material, affordance, kinematics, and function description. In particular, we devise a scalable human-in-the-loop annotation pipeline based on vision-language models, which enables efficient creation of physics-first assets from raw 3D assets.2) Furthermore, we propose \\textbf{PhysXGen}, a feed-forward framework for physics-grounded image-to-3D asset generation, injecting physical knowledge into the pre-trained 3D structural space. Specifically, PhysXGen employs a dual-branch architecture to explicitly model the latent correlations between 3D structures and physical properties, thereby producing 3D assets with plausible physical predictions while preserving the native geometry quality. Extensive experiments validate the superior performance and promising generalization capability of our framework. All the code, data, and models will be released to facilitate future research in generative physical AI.", "date": "2025-07-16", "categories": []} {"id": "2508.07769", "title": "Dream4D: Lifting Camera-Controlled I2V towards Spatiotemporally Consistent 4D Generation", "abstract": "The synthesis of spatiotemporally coherent 4D content presents fundamental challenges in computer vision, requiring simultaneous modeling of high-fidelity spatial representations and physically plausible temporal dynamics. Current approaches often struggle to maintain view consistency while handling complex scene dynamics, particularly in large-scale environments with multiple interacting elements. This work introduces Dream4D, a novel framework that bridges this gap through a synergy of controllable video generation and neural 4D reconstruction. Our approach seamlessly combines a two-stage architecture: it first predicts optimal camera trajectories from a single image using few-shot learning, then generates geometrically consistent multi-view sequences via a specialized pose-conditioned diffusion process, which are finally converted into a persistent 4D representation. This framework is the first to leverage both rich temporal priors from video diffusion models and geometric awareness of the reconstruction models, which significantly facilitates 4D generation and shows higher quality (e.g., mPSNR, mSSIM) over existing methods.", "date": "2025-08-11", "categories": []} {"id": "2511.21887", "title": "UniArt: Unified 3D Representation for Generating 3D Articulated Objects with Open-Set Articulation", "abstract": "Articulated 3D objects play a vital role in realistic simulation and embodied robotics, yet manually constructing such assets remains costly and difficult to scale. In this paper, we present UniArt, a diffusion-based framework that directly synthesizes fully articulated 3D objects from a single image in an end-to-end manner. Unlike prior multi-stage techniques, UniArt establishes a unified latent representation that jointly encodes geometry, texture, part segmentation, and kinematic parameters. We introduce a reversible joint-to-voxel embedding, which spatially aligns articulation features with volumetric geometry, enabling the model to learn coherent motion behaviors alongside structural formation. Furthermore, we formulate articulation type prediction as an open-set problem, removing the need for fixed joint semantics and allowing generalization to novel joint categories and unseen object types. Experiments on the PartNet-Mobility benchmark demonstrate that UniArt achieves state-of-the-art mesh quality and articulation accuracy.", "date": "2025-11-26", "categories": []} {"id": "2511.22171", "title": "BrepGPT: Autoregressive B-rep Generation with Voronoi Half-Patch", "abstract": "Boundary representation (B-rep) is the de facto standard for CAD model representation in modern industrial design. The intricate coupling between geometric and topological elements in B-rep structures has forced existing generative methods to rely on cascaded multi-stage networks, resulting in error accumulation and computational inefficiency. We present BrepGPT, a single-stage autoregressive framework for B-rep generation. Our key innovation lies in the Voronoi Half-Patch (VHP) representation, which decomposes B-reps into unified local units by assigning geometry to nearest half-edges and sampling their next pointers. Unlike hierarchical representations that require multiple distinct encodings for different structural levels, our VHP representation facilitates unifying geometric attributes and topological relations in a single, coherent format. We further leverage dual VQ-VAEs to encode both vertex topology and Voronoi Half-Patches into vertex-based tokens, achieving a more compact sequential encoding. A decoder-only Transformer is then trained to autoregressively predict these tokens, which are subsequently mapped to vertex-based features and decoded into complete B-rep models. Experiments demonstrate that BrepGPT achieves state-of-the-art performance in unconditional B-rep generation. The framework also exhibits versatility in various applications, including conditional generation from category labels, point clouds, text descriptions, and images, as well as B-rep autocompletion and interpolation.", "date": "2025-11-27", "categories": []} {"id": "2511.22194", "title": "Controllable 3D Object Generation with Single Image Prompt", "abstract": "Recently, the impressive generative capabilities of diffusion models have been demonstrated, producing images with remarkable fidelity. Particularly, existing methods for the 3D object generation tasks, which is one of the fastest-growing segments in computer vision, pre-dominantly use text-to-image diffusion models with textual inversion which train a pseudo text prompt to describe the given image. In practice, various text-to-image generative models employ textual inversion to learn concepts or styles of target object in the pseudo text prompt embedding space, thereby generating sophisticated outputs. However, textual inversion requires additional training time and lacks control ability. To tackle this issues, we propose two innovative methods: (1) using an off-the-shelf image adapter that generates 3D objects without textual inversion, offering enhanced control over conditions such as depth, pose, and text. (2) a depth conditioned warmup strategy to enhance 3D consistency. In experimental results, ours show qualitatively and quantitatively comparable performance and improved 3D consistency to the existing text-inversion-based alternatives. Furthermore, we conduct a user study to assess (i) how well results match the input image and (ii) whether 3D consistency is maintained. User study results show that our model outperforms the alternatives, validating the effectiveness of our approaches. Our code is available at GitHub repository:https://github.com/Seooooooogi/Control3D_IP/", "date": "2025-11-27", "categories": []} {"id": "2511.22357", "title": "AnchorFlow: Training-Free 3D Editing via Latent Anchor-Aligned Flows", "abstract": "Training-free 3D editing aims to modify 3D shapes based on human instructions without model finetuning. It plays a crucial role in 3D content creation. However, existing approaches often struggle to produce strong or geometrically stable edits, largely due to inconsistent latent anchors introduced by timestep-dependent noise during diffusion sampling. To address these limitations, we introduce AnchorFlow, which is built upon the principle of latent anchor consistency. Specifically, AnchorFlow establishes a global latent anchor shared between the source and target trajectories, and enforces coherence using a relaxed anchor-alignment loss together with an anchor-aligned update rule. This design ensures that transformations remain stable and semantically faithful throughout the editing process. By stabilizing the latent reference space, AnchorFlow enables more pronounced semantic modifications. Moreover, AnchorFlow is mask-free. Without mask supervision, it effectively preserves geometric fidelity. Experiments on the Eval3DEdit benchmark show that AnchorFlow consistently delivers semantically aligned and structurally robust edits across diverse editing types. Code is at https://github.com/ZhenglinZhou/AnchorFlow.", "date": "2025-11-27", "categories": []} {"id": "2511.22456", "title": "ITS3D: Inference-Time Scaling for Text-Guided 3D Diffusion Models", "abstract": "We explore inference-time scaling in text-guided 3D diffusion models to enhance generative quality without additional training. To this end, we introduce ITS3D, a framework that formulates the task as an optimization problem to identify the most effective Gaussian noise input. The framework is driven by a verifier-guided search algorithm, where the search algorithm iteratively refines noise candidates based on verifier feedback. To address the inherent challenges of 3D generation, we introduce three techniques for improved stability, efficiency, and exploration capability. 1) Gaussian normalization is applied to stabilize the search process. It corrects distribution shifts when noise candidates deviate from a standard Gaussian distribution during iterative updates. 2) The high-dimensional nature of the 3D search space increases computational complexity. To mitigate this, a singular value decomposition-based compression technique is employed to reduce dimensionality while preserving effective search directions. 3) To further prevent convergence to suboptimal local minima, a singular space reset mechanism dynamically updates the search space based on diversity measures. Extensive experiments demonstrate that ITS3D enhances text-to-3D generation quality, which shows the potential of computationally efficient search methods in generative processes. The source code is available at https://github.com/ZhenglinZhou/ITS3D.", "date": "2025-11-27", "categories": []} {"id": "2511.23051", "title": "GOATex: Geometry & Occlusion-Aware Texturing", "abstract": "We present GOATex, a diffusion-based method for 3D mesh texturing that generates high-quality textures for both exterior and interior surfaces. While existing methods perform well on visible regions, they inherently lack mechanisms to handle occluded interiors, resulting in incomplete textures and visible seams. To address this, we introduce an occlusion-aware texturing framework based on the concept of hit levels, which quantify the relative depth of mesh faces via multi-view ray casting. This allows us to partition mesh faces into ordered visibility layers, from outermost to innermost. We then apply a two-stage visibility control strategy that progressively reveals interior regions with structural coherence, followed by texturing each layer using a pretrained diffusion model. To seamlessly merge textures obtained across layers, we propose a soft UV-space blending technique that weighs each texture's contribution based on view-dependent visibility confidence. Empirical results demonstrate that GOATex consistently outperforms existing methods, producing seamless, high-fidelity textures across both visible and occluded surfaces. Unlike prior works, GOATex operates entirely without costly fine-tuning of a pretrained diffusion model and allows separate prompting for exterior and interior mesh regions, enabling fine-grained control over layered appearances. For more qualitative results, please visit our project page: https://goatex3d.github.io/.", "date": "2025-11-28", "categories": []} {"id": "2511.23191", "title": "GeoWorld: Unlocking the Potential of Geometry Models to Facilitate High-Fidelity 3D Scene Generation", "abstract": "Previous works leveraging video models for image-to-3D scene generation tend to suffer from geometric distortions and blurry content. In this paper, we renovate the pipeline of image-to-3D scene generation by unlocking the potential of geometry models and present our GeoWorld. Instead of exploiting geometric information obtained from a single-frame input, we propose to first generate consecutive video frames and then take advantage of the geometry model to provide full-frame geometry features, which contain richer information than single-frame depth maps or camera embeddings used in previous methods, and use these geometry features as geometrical conditions to aid the video generation model. To enhance the consistency of geometric structures, we further propose a geometry alignment loss to provide the model with real-world geometric constraints and a geometry adaptation module to ensure the effective utilization of geometry features. Extensive experiments show that our GeoWorld can generate high-fidelity 3D scenes from a single image and a given camera trajectory, outperforming prior methods both qualitatively and quantitatively. Project Page: https://peaes.github.io/GeoWorld/.", "date": "2025-11-28", "categories": []} {"id": "2412.06698", "title": "Gen-3Diffusion: Realistic Image-to-3D Generation via 2D & 3D Diffusion Synergy", "abstract": "Creating realistic 3D objects and clothed avatars from a single RGB image is an attractive yet challenging problem. Due to its ill-posed nature, recent works leverage powerful prior from 2D diffusion models pretrained on large datasets. Although 2D diffusion models demonstrate strong generalization capability, they cannot guarantee the generated multi-view images are 3D consistent. In this paper, we propose Gen-3Diffusion: Realistic Image-to-3D Generation via 2D & 3D Diffusion Synergy. We leverage a pre-trained 2D diffusion model and a 3D diffusion model via our elegantly designed process that synchronizes two diffusion models at both training and sampling time. The synergy between the 2D and 3D diffusion models brings two major advantages: 1) 2D helps 3D in generalization: the pretrained 2D model has strong generalization ability to unseen images, providing strong shape priors for the 3D diffusion model; 2) 3D helps 2D in multi-view consistency: the 3D diffusion model enhances the 3D consistency of 2D multi-view sampling process, resulting in more accurate multi-view generation. We validate our idea through extensive experiments in image-based objects and clothed avatar generation tasks. Results show that our method generates realistic 3D objects and avatars with high-fidelity geometry and texture. Extensive ablations also validate our design choices and demonstrate the strong generalization ability to diverse clothing and compositional shapes. Our code and pretrained models will be publicly released on https://yuxuan-xue.com/gen-3diffusion.", "date": "2024-12-09", "categories": []} {"id": "2511.09298", "title": "DensiCrafter: Physically-Constrained Generation and Fabrication of Self-Supporting Hollow Structures", "abstract": "The rise of 3D generative models has enabled automatic 3D geometry and texture synthesis from multimodal inputs (e.g., text or images). However, these methods often ignore physical constraints and manufacturability considerations. In this work, we address the challenge of producing 3D designs that are both lightweight and self-supporting. We present DensiCrafter, a framework for generating lightweight, self-supporting 3D hollow structures by optimizing the density field. Starting from coarse voxel grids produced by Trellis, we interpret these as continuous density fields to optimize and introduce three differentiable, physically constrained, and simulation-free loss terms. Additionally, a mass regularization penalizes unnecessary material, while a restricted optimization domain preserves the outer surface. Our method seamlessly integrates with pretrained Trellis-based models (e.g., Trellis, DSO) without any architectural changes. In extensive evaluations, we achieve up to 43% reduction in material mass on the text-to-3D task. Compared to state-of-the-art baselines, our method could improve the stability and maintain high geometric fidelity. Real-world 3D-printing experiments confirm that our hollow designs can be reliably fabricated and could be self-supporting.", "date": "2025-11-12", "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": "2506.08640", "title": "Orientation Matters: Making 3D Generative Models Orientation-Aligned", "abstract": "Humans intuitively perceive object shape and orientation from a single image, guided by strong priors about canonical poses. However, existing 3D generative models often produce misaligned results due to inconsistent training data, limiting their usability in downstream tasks. To address this gap, we introduce the task of orientation-aligned 3D object generation: producing 3D objects from single images with consistent orientations across categories. To facilitate this, we construct Objaverse-OA, a dataset of 14,832 orientation-aligned 3D models spanning 1,008 categories. Leveraging Objaverse-OA, we fine-tune two representative 3D generative models based on multi-view diffusion and 3D variational autoencoder frameworks to produce aligned objects that generalize well to unseen objects across various categories. Experimental results demonstrate the superiority of our method over post-hoc alignment approaches. Furthermore, we showcase downstream applications enabled by our aligned object generation, including zero-shot object orientation estimation via analysis-by-synthesis and efficient arrow-based object rotation manipulation.", "date": "2025-06-10", "categories": []} {"id": "2506.23150", "title": "AlignCVC: Aligning Cross-View Consistency for Single-Image-to-3D Generation", "abstract": "Single-image-to-3D models typically follow a sequential generation and reconstruction workflow. However, intermediate multi-view images synthesized by pre-trained generation models often lack cross-view consistency (CVC), significantly degrading 3D reconstruction performance. While recent methods attempt to refine CVC by feeding reconstruction results back into the multi-view generator, these approaches struggle with noisy and unstable reconstruction outputs that limit effective CVC improvement. We introduce AlignCVC, a novel framework that fundamentally re-frames single-image-to-3D generation through distribution alignment rather than relying on strict regression losses. Our key insight is to align both generated and reconstructed multi-view distributions toward the ground-truth multi-view distribution, establishing a principled foundation for improved CVC. Observing that generated images exhibit weak CVC while reconstructed images display strong CVC due to explicit rendering, we propose a soft-hard alignment strategy with distinct objectives for generation and reconstruction models. This approach not only enhances generation quality but also dramatically accelerates inference to as few as 4 steps. As a plug-and-play paradigm, our method, namely AlignCVC, seamlessly integrates various multi-view generation models with 3D reconstruction models. Extensive experiments demonstrate the effectiveness and efficiency of AlignCVC for single-image-to-3D generation.", "date": "2025-06-29", "categories": []} {"id": "2511.19512", "title": "Single Image to High-Quality 3D Object via Latent Features", "abstract": "3D assets are essential in the digital age. While automatic 3D generation, such as image-to-3d, has made significant strides in recent years, it often struggles to achieve fast, detailed, and high-fidelity generation simultaneously. In this work, we introduce LatentDreamer, a novel framework for generating 3D objects from single images. The key to our approach is a pre-trained variational autoencoder that maps 3D geometries to latent features, which greatly reducing the difficulty of 3D generation. Starting from latent features, the pipeline of LatentDreamer generates coarse geometries, refined geometries, and realistic textures sequentially. The 3D objects generated by LatentDreamer exhibit high fidelity to the input images, and the entire generation process can be completed within a short time (typically in 70 seconds). Extensive experiments show that with only a small amount of training, LatentDreamer demonstrates competitive performance compared to contemporary approachs.", "date": "2025-11-24", "categories": []} {"id": "2511.20624", "title": "ShapeGen: Towards High-Quality 3D Shape Synthesis", "abstract": "Inspired by generative paradigms in image and video, 3D shape generation has made notable progress, enabling the rapid synthesis of high-fidelity 3D assets from a single image. However, current methods still face challenges, including the lack of intricate details, overly smoothed surfaces, and fragmented thin-shell structures. These limitations leave the generated 3D assets still one step short of meeting the standards favored by artists. In this paper, we present ShapeGen, which achieves high-quality image-to-3D shape generation through 3D representation and supervision improvements, resolution scaling up, and the advantages of linear transformers. These advancements allow the generated assets to be seamlessly integrated into 3D pipelines, facilitating their widespread adoption across various applications. Through extensive experiments, we validate the impact of these improvements on overall performance. Ultimately, thanks to the synergistic effects of these enhancements, ShapeGen achieves a significant leap in image-to-3D generation, establishing a new state-of-the-art performance.", "date": "2025-11-25", "categories": []} {"id": "2509.00062", "title": "Scaffold Diffusion: Sparse Multi-Category Voxel Structure Generation with Discrete Diffusion", "abstract": "Generating realistic sparse multi-category 3D voxel structures is difficult due to the cubic memory scaling of voxel structures and moreover the significant class imbalance caused by sparsity. We introduce Scaffold Diffusion, a generative model designed for sparse multi-category 3D voxel structures. By treating voxels as tokens, Scaffold Diffusion uses a discrete diffusion language model to generate 3D voxel structures. We show that discrete diffusion language models can be extended beyond inherently sequential domains such as text to generate spatially coherent 3D structures. We evaluate on Minecraft house structures from the 3D-Craft dataset and demonstrate that, unlike prior baselines and an auto-regressive formulation, Scaffold Diffusion produces realistic and coherent structures even when trained on data with over 98% sparsity. We provide an interactive viewer where readers can visualize generated samples and the generation process: https://scaffold.deepexploration.org/", "date": "2025-08-26", "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": "2511.17883", "title": "ArticFlow: Generative Simulation of Articulated Mechanisms", "abstract": "Recent advances in generative models have produced strong results for static 3D shapes, whereas articulated 3D generation remains challenging due to action-dependent deformations and limited datasets. We introduce ArticFlow, a two-stage flow matching framework that learns a controllable velocity field from noise to target point sets under explicit action control. ArticFlow couples (i) a latent flow that transports noise to a shape-prior code and (ii) a point flow that transports points conditioned on the action and the shape prior, enabling a single model to represent diverse articulated categories and generalize across actions. On MuJoCo Menagerie, ArticFlow functions both as a generative model and as a neural simulator: it predicts action-conditioned kinematics from a compact prior and synthesizes novel morphologies via latent interpolation. Compared with object-specific simulators and an action-conditioned variant of static point-cloud generators, ArticFlow achieves higher kinematic accuracy and better shape quality. Results show that action-conditioned flow matching is a practical route to controllable and high-quality articulated mechanism generation.", "date": "2025-11-22", "categories": []} {"id": "2511.18005", "title": "RAISECity: A Multimodal Agent Framework for Reality-Aligned 3D World Generation at City-Scale", "abstract": "City-scale 3D generation is of great importance for the development of embodied intelligence and world models. Existing methods, however, face significant challenges regarding quality, fidelity, and scalability in 3D world generation. Thus, we propose RAISECity, a \\textbf{R}eality-\\textbf{A}ligned \\textbf{I}ntelligent \\textbf{S}ynthesis \\textbf{E}ngine that creates detailed, \\textbf{C}ity-scale 3D worlds. We introduce an agentic framework that leverages diverse multimodal foundation tools to acquire real-world knowledge, maintain robust intermediate representations, and construct complex 3D scenes. This agentic design, featuring dynamic data processing, iterative self-reflection and refinement, and the invocation of advanced multimodal tools, minimizes cumulative errors and enhances overall performance. Extensive quantitative experiments and qualitative analyses validate the superior performance of RAISECity in real-world alignment, shape precision, texture fidelity, and aesthetics level, achieving over a 90% win-rate against existing baselines for overall perceptual quality. This combination of 3D quality, reality alignment, scalability, and seamless compatibility with computer graphics pipelines makes RAISECity a promising foundation for applications in immersive media, embodied intelligence, and world models.", "date": "2025-11-22", "categories": []} {"id": "2511.18900", "title": "MatMart: Material Reconstruction of 3D Objects via Diffusion", "abstract": "Applying diffusion models to physically-based material estimation and generation has recently gained prominence. In this paper, we propose \\ttt, a novel material reconstruction framework for 3D objects, offering the following advantages. First, \\ttt\\ adopts a two-stage reconstruction, starting with accurate material prediction from inputs and followed by prior-guided material generation for unobserved views, yielding high-fidelity results. Second, by utilizing progressive inference alongside the proposed view-material cross-attention (VMCA), \\ttt\\ enables reconstruction from an arbitrary number of input images, demonstrating strong scalability and flexibility. Finally, \\ttt\\ achieves both material prediction and generation capabilities through end-to-end optimization of a single diffusion model, without relying on additional pre-trained models, thereby exhibiting enhanced stability across various types of objects. Extensive experiments demonstrate that \\ttt\\ achieves superior performance in material reconstruction compared to existing methods.", "date": "2025-11-24", "categories": []} {"id": "2511.16807", "title": "Mesh RAG: Retrieval Augmentation for Autoregressive Mesh Generation", "abstract": "3D meshes are a critical building block for applications ranging from industrial design and gaming to simulation and robotics. Traditionally, meshes are crafted manually by artists, a process that is time-intensive and difficult to scale. To automate and accelerate this asset creation, autoregressive models have emerged as a powerful paradigm for artistic mesh generation. However, current methods to enhance quality typically rely on larger models or longer sequences that result in longer generation time, and their inherent sequential nature imposes a severe quality-speed trade-off. This sequential dependency also significantly complicates incremental editing. To overcome these limitations, we propose Mesh RAG, a novel, training-free, plug-and-play framework for autoregressive mesh generation models. Inspired by RAG for language models, our approach augments the generation process by leveraging point cloud segmentation, spatial transformation, and point cloud registration to retrieve, generate, and integrate mesh components. This retrieval-based approach decouples generation from its strict sequential dependency, facilitating efficient and parallelizable inference. We demonstrate the wide applicability of Mesh RAG across various foundational autoregressive mesh generation models, showing it significantly enhances mesh quality, accelerates generation speed compared to sequential part prediction, and enables incremental editing, all without model retraining.", "date": "2025-11-20", "categories": []} {"id": "2511.16825", "title": "WorldGen: From Text to Traversable and Interactive 3D Worlds", "abstract": "We introduce WorldGen, a system that enables the automatic creation of large-scale, interactive 3D worlds directly from text prompts. Our approach transforms natural language descriptions into traversable, fully textured environments that can be immediately explored or edited within standard game engines. By combining LLM-driven scene layout reasoning, procedural generation, diffusion-based 3D generation, and object-aware scene decomposition, WorldGen bridges the gap between creative intent and functional virtual spaces, allowing creators to design coherent, navigable worlds without manual modeling or specialized 3D expertise. The system is fully modular and supports fine-grained control over layout, scale, and style, producing worlds that are geometrically consistent, visually rich, and efficient to render in real time. This work represents a step towards accessible, generative world-building at scale, advancing the frontier of 3D generative AI for applications in gaming, simulation, and immersive social environments.", "date": "2025-11-20", "categories": []} {"id": "2511.17048", "title": "RoomPlanner: Explicit Layout Planner for Easier LLM-Driven 3D Room Generation", "abstract": "In this paper, we propose RoomPlanner, the first fully automatic 3D room generation framework for painlessly creating realistic indoor scenes with only short text as input. Without any manual layout design or panoramic image guidance, our framework can generate explicit layout criteria for rational spatial placement. We begin by introducing a hierarchical structure of language-driven agent planners that can automatically parse short and ambiguous prompts into detailed scene descriptions. These descriptions include raw spatial and semantic attributes for each object and the background, which are then used to initialize 3D point clouds. To position objects within bounded environments, we implement two arrangement constraints that iteratively optimize spatial arrangements, ensuring a collision-free and accessible layout solution. In the final rendering stage, we propose a novel AnyReach Sampling strategy for camera trajectory, along with the Interval Timestep Flow Sampling (ITFS) strategy, to efficiently optimize the coarse 3D Gaussian scene representation. These approaches help reduce the total generation time to under 30 minutes. Extensive experiments demonstrate that our method can produce geometrically rational 3D indoor scenes, surpassing prior approaches in both rendering speed and visual quality while preserving editability. The code will be available soon.", "date": "2025-11-21", "categories": []} {"id": "2511.17501", "title": "Native 3D Editing with Full Attention", "abstract": "Instruction-guided 3D editing is a rapidly emerging field with the potential to broaden access to 3D content creation. However, existing methods face critical limitations: optimization-based approaches are prohibitively slow, while feed-forward approaches relying on multi-view 2D editing often suffer from inconsistent geometry and degraded visual quality. To address these issues, we propose a novel native 3D editing framework that directly manipulates 3D representations in a single, efficient feed-forward pass. Specifically, we create a large-scale, multi-modal dataset for instruction-guided 3D editing, covering diverse addition, deletion, and modification tasks. This dataset is meticulously curated to ensure that edited objects faithfully adhere to the instructional changes while preserving the consistency of unedited regions with the source object. Building upon this dataset, we explore two distinct conditioning strategies for our model: a conventional cross-attention mechanism and a novel 3D token concatenation approach. Our results demonstrate that token concatenation is more parameter-efficient and achieves superior performance. Extensive evaluations show that our method outperforms existing 2D-lifting approaches, setting a new benchmark in generation quality, 3D consistency, and instruction fidelity.", "date": "2025-11-21", "categories": []} {"id": "2412.17957", "title": "ArchComplete: Autoregressive 3D Architectural Design Generation with Hierarchical Diffusion-Based Upsampling", "abstract": "Recent advances in 3D generative models have shown promising results but often fall short in capturing the complexity of architectural geometries and topologies and fine geometric details at high resolutions. To tackle this, we present ArchComplete, a two-stage voxel-based 3D generative pipeline consisting of a vector-quantised model, whose composition is modelled with an autoregressive transformer for generating coarse shapes, followed by a hierarchical upsampling strategy for further enrichment with fine structures and details. Key to our pipeline is (i) learning a contextually rich codebook of local patch embeddings, optimised alongside a 2.5D perceptual loss that captures global spatial correspondence of projections onto three axis-aligned orthogonal planes, and (ii) redefining upsampling as a set of conditional diffusion models learning from a hierarchy of randomly cropped coarse-to-fine local volumetric patches. Trained on our introduced dataset of 3D house models with fully modelled exterior and interior, ArchComplete autoregressively generates models at the resolution of $64^{3}$ and progressively refines them up to $512^{3}$, with voxel sizes as small as $ \\approx 9\\text{cm}$. ArchComplete solves a variety of tasks, including genetic interpolation and variation, unconditional synthesis, shape and plan-drawing completion, as well as geometric detailisation, while achieving state-of-the-art performance in quality, diversity, and computational efficiency.", "date": "2024-12-23", "categories": []} {"id": "2511.16317", "title": "NaTex: Seamless Texture Generation as Latent Color Diffusion", "abstract": "We present NaTex, a native texture generation framework that predicts texture color directly in 3D space. In contrast to previous approaches that rely on baking 2D multi-view images synthesized by geometry-conditioned Multi-View Diffusion models (MVDs), NaTex avoids several inherent limitations of the MVD pipeline. These include difficulties in handling occluded regions that require inpainting, achieving precise mesh-texture alignment along boundaries, and maintaining cross-view consistency and coherence in both content and color intensity. NaTex features a novel paradigm that addresses the aforementioned issues by viewing texture as a dense color point cloud. Driven by this idea, we propose latent color diffusion, which comprises a geometry-awared color point cloud VAE and a multi-control diffusion transformer (DiT), entirely trained from scratch using 3D data, for texture reconstruction and generation. To enable precise alignment, we introduce native geometry control that conditions the DiT on direct 3D spatial information via positional embeddings and geometry latents. We co-design the VAE-DiT architecture, where the geometry latents are extracted via a dedicated geometry branch tightly coupled with the color VAE, providing fine-grained surface guidance that maintains strong correspondence with the texture. With these designs, NaTex demonstrates strong performance, significantly outperforming previous methods in texture coherence and alignment. Moreover, NaTex also exhibits strong generalization capabilities, either training-free or with simple tuning, for various downstream applications, e.g., material generation, texture refinement, and part segmentation and texturing.", "date": "2025-11-20", "categories": []} {"id": "2511.16642", "title": "TRIM: Scalable 3D Gaussian Diffusion Inference with Temporal and Spatial Trimming", "abstract": "Recent advances in 3D Gaussian diffusion models suffer from time-intensive denoising and post-denoising processing due to the massive number of Gaussian primitives, resulting in slow generation and limited scalability along sampling trajectories. To improve the efficiency of 3D diffusion models, we propose $\\textbf{TRIM}$ ($\\textbf{T}$rajectory $\\textbf{R}$eduction and $\\textbf{I}$nstance $\\textbf{M}$ask denoising), a post-training approach that incorporates both temporal and spatial trimming strategies, to accelerate inference without compromising output quality while supporting the inference-time scaling for Gaussian diffusion models. Instead of scaling denoising trajectories in a costly end-to-end manner, we develop a lightweight selector model to evaluate latent Gaussian primitives derived from multiple sampled noises, enabling early trajectory reduction by selecting candidates with high-quality potential. Furthermore, we introduce instance mask denoising to prune learnable Gaussian primitives by filtering out redundant background regions, reducing inference computation at each denoising step. Extensive experiments and analysis demonstrate that TRIM significantly improves both the efficiency and quality of 3D generation. Source code is available at $\\href{https://github.com/zeyuanyin/TRIM}{link}$.", "date": "2025-11-20", "categories": []} {"id": "2511.16662", "title": "TriDiff-4D: Fast 4D Generation through Diffusion-based Triplane Re-posing", "abstract": "With the increasing demand for 3D animation, generating high-fidelity, controllable 4D avatars from textual descriptions remains a significant challenge. Despite notable efforts in 4D generative modeling, existing methods exhibit fundamental limitations that impede their broader applicability, including temporal and geometric inconsistencies, perceptual artifacts, motion irregularities, high computational costs, and limited control over dynamics. To address these challenges, we propose TriDiff-4D, a novel 4D generative pipeline that employs diffusion-based triplane re-posing to produce high-quality, temporally coherent 4D avatars. Our model adopts an auto-regressive strategy to generate 4D sequences of arbitrary length, synthesizing each 3D frame with a single diffusion process. By explicitly learning 3D structure and motion priors from large-scale 3D and motion datasets, TriDiff-4D enables skeleton-driven 4D generation that excels in temporal consistency, motion accuracy, computational efficiency, and visual fidelity. Specifically, TriDiff-4D first generates a canonical 3D avatar and a corresponding motion sequence from a text prompt, then uses a second diffusion model to animate the avatar according to the motion sequence, supporting arbitrarily long 4D generation. Experimental results demonstrate that TriDiff-4D significantly outperforms existing methods, reducing generation time from hours to seconds by eliminating the optimization process, while substantially improving the generation of complex motions with high-fidelity appearance and accurate 3D geometry.", "date": "2025-11-20", "categories": []} {"id": "2511.01767", "title": "Wonder3D++: Cross-domain Diffusion for High-fidelity 3D Generation from a Single Image", "abstract": "In this work, we introduce \\textbf{Wonder3D++}, a novel method for efficiently generating high-fidelity textured meshes from single-view images. Recent methods based on Score Distillation Sampling (SDS) have shown the potential to recover 3D geometry from 2D diffusion priors, but they typically suffer from time-consuming per-shape optimization and inconsistent geometry. In contrast, certain works directly produce 3D information via fast network inferences, but their results are often of low quality and lack geometric details. To holistically improve the quality, consistency, and efficiency of single-view reconstruction tasks, we propose a cross-domain diffusion model that generates multi-view normal maps and the corresponding color images. To ensure the consistency of generation, we employ a multi-view cross-domain attention mechanism that facilitates information exchange across views and modalities. Lastly, we introduce a cascaded 3D mesh extraction algorithm that drives high-quality surfaces from the multi-view 2D representations in only about $3$ minute in a coarse-to-fine manner. Our extensive evaluations demonstrate that our method achieves high-quality reconstruction results, robust generalization, and good efficiency compared to prior works. Code available at https://github.com/xxlong0/Wonder3D/tree/Wonder3D_Plus.", "date": "2025-11-03", "categories": []} {"id": "2511.14870", "title": "B-Rep Distance Functions (BR-DF): How to Represent a B-Rep Model by Volumetric Distance Functions?", "abstract": "This paper presents a novel geometric representation for CAD Boundary Representation (B-Rep) based on volumetric distance functions, dubbed B-Rep Distance Functions (BR-DF). BR-DF encodes the surface mesh geometry of a CAD model as signed distance function (SDF). B-Rep vertices, edges, faces and their topology information are encoded as per-face unsigned distance functions (UDFs). An extension of the Marching Cubes algorithm converts BR-DF directly into watertight CAD B-Rep model (strictly speaking a faceted B-Rep model). A surprising characteristic of BR-DF is that this conversion process never fails. Leveraging the volumetric nature of BR-DF, we propose a multi-branch latent diffusion with 3D U-Net backbone for jointly generating the SDF and per-face UDFs of a BR-DF model. Our approach achieves comparable CAD generation performance against SOTA methods while reaching the unprecedented 100% success rate in producing (faceted) B-Rep models.", "date": "2025-11-18", "categories": []} {"id": "2511.14884", "title": "GeoSceneGraph: Geometric Scene Graph Diffusion Model for Text-guided 3D Indoor Scene Synthesis", "abstract": "Methods that synthesize indoor 3D scenes from text prompts have wide-ranging applications in film production, interior design, video games, virtual reality, and synthetic data generation for training embodied agents. Existing approaches typically either train generative models from scratch or leverage vision-language models (VLMs). While VLMs achieve strong performance, particularly for complex or open-ended prompts, smaller task-specific models remain necessary for deployment on resource-constrained devices such as extended reality (XR) glasses or mobile phones. However, many generative approaches that train from scratch overlook the inherent graph structure of indoor scenes, which can limit scene coherence and realism. Conversely, methods that incorporate scene graphs either demand a user-provided semantic graph, which is generally inconvenient and restrictive, or rely on ground-truth relationship annotations, limiting their capacity to capture more varied object interactions. To address these challenges, we introduce GeoSceneGraph, a method that synthesizes 3D scenes from text prompts by leveraging the graph structure and geometric symmetries of 3D scenes, without relying on predefined relationship classes. Despite not using ground-truth relationships, GeoSceneGraph achieves performance comparable to methods that do. Our model is built on equivariant graph neural networks (EGNNs), but existing EGNN approaches are typically limited to low-dimensional conditioning and are not designed to handle complex modalities such as text. We propose a simple and effective strategy for conditioning EGNNs on text features, and we validate our design through ablation studies.", "date": "2025-11-18", "categories": []} {"id": "2511.15618", "title": "FlashMesh: Faster and Better Autoregressive Mesh Synthesis via Structured Speculation", "abstract": "Autoregressive models can generate high-quality 3D meshes by sequentially producing vertices and faces, but their token-by-token decoding results in slow inference, limiting practical use in interactive and large-scale applications. We present FlashMesh, a fast and high-fidelity mesh generation framework that rethinks autoregressive decoding through a predict-correct-verify paradigm. The key insight is that mesh tokens exhibit strong structural and geometric correlations that enable confident multi-token speculation. FlashMesh leverages this by introducing a speculative decoding scheme tailored to the commonly used hourglass transformer architecture, enabling parallel prediction across face, point, and coordinate levels. Extensive experiments show that FlashMesh achieves up to a 2 x speedup over standard autoregressive models while also improving generation fidelity. Our results demonstrate that structural priors in mesh data can be systematically harnessed to accelerate and enhance autoregressive generation.", "date": "2025-11-19", "categories": []} {"id": "2511.10040", "title": "LoG3D: Ultra-High-Resolution 3D Shape Modeling via Local-to-Global Partitioning", "abstract": "Generating high-fidelity 3D contents remains a fundamental challenge due to the complexity of representing arbitrary topologies-such as open surfaces and intricate internal structures-while preserving geometric details. Prevailing methods based on signed distance fields (SDFs) are hampered by costly watertight preprocessing and struggle with non-manifold geometries, while point-cloud representations often suffer from sampling artifacts and surface discontinuities. To overcome these limitations, we propose a novel 3D variational autoencoder (VAE) framework built upon unsigned distance fields (UDFs)-a more robust and computationally efficient representation that naturally handles complex and incomplete shapes. Our core innovation is a local-to-global (LoG) architecture that processes the UDF by partitioning it into uniform subvolumes, termed UBlocks. This architecture couples 3D convolutions for capturing local detail with sparse transformers for enforcing global coherence. A Pad-Average strategy further ensures smooth transitions at subvolume boundaries during reconstruction. This modular design enables seamless scaling to ultra-high resolutions up to $2048^3$-a regime previously unattainable for 3D VAEs. Experiments demonstrate state-of-the-art performance in both reconstruction accuracy and generative quality, yielding superior surface smoothness and geometric flexibility.", "date": "2025-11-13", "categories": []} {"id": "2511.14271", "title": "Let Language Constrain Geometry: Vision-Language Models as Semantic and Spatial Critics for 3D Generation", "abstract": "Text-to-3D generation has advanced rapidly, yet state-of-the-art models, encompassing both optimization-based and feed-forward architectures, still face two fundamental limitations. First, they struggle with coarse semantic alignment, often failing to capture fine-grained prompt details. Second, they lack robust 3D spatial understanding, leading to geometric inconsistencies and catastrophic failures in part assembly and spatial relationships. To address these challenges, we propose VLM3D, a general framework that repurposes large vision-language models (VLMs) as powerful, differentiable semantic and spatial critics. Our core contribution is a dual-query critic signal derived from the VLM's Yes or No log-odds, which assesses both semantic fidelity and geometric coherence. We demonstrate the generality of this guidance signal across two distinct paradigms: (1) As a reward objective for optimization-based pipelines, VLM3D significantly outperforms existing methods on standard benchmarks. (2) As a test-time guidance module for feed-forward pipelines, it actively steers the iterative sampling process of SOTA native 3D models to correct severe spatial errors. VLM3D establishes a principled and generalizable path to inject the VLM's rich, language-grounded understanding of both semantics and space into diverse 3D generative pipelines.", "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": "2511.06194", "title": "NURBGen: High-Fidelity Text-to-CAD Generation through LLM-Driven NURBS Modeling", "abstract": "Generating editable 3D CAD models from natural language remains challenging, as existing text-to-CAD systems either produce meshes or rely on scarce design-history data. We present NURBGen, the first framework to generate high-fidelity 3D CAD models directly from text using Non-Uniform Rational B-Splines (NURBS). To achieve this, we fine-tune a large language model (LLM) to translate free-form texts into JSON representations containing NURBS surface parameters (\\textit{i.e}, control points, knot vectors, degrees, and rational weights) which can be directly converted into BRep format using Python. We further propose a hybrid representation that combines untrimmed NURBS with analytic primitives to handle trimmed surfaces and degenerate regions more robustly, while reducing token complexity. Additionally, we introduce partABC, a curated subset of the ABC dataset consisting of individual CAD components, annotated with detailed captions using an automated annotation pipeline. NURBGen demonstrates strong performance on diverse prompts, surpassing prior methods in geometric fidelity and dimensional accuracy, as confirmed by expert evaluations. Code and dataset will be released publicly.", "date": "2025-11-09", "categories": []} {"id": "2511.11692", "title": "AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation", "abstract": "Optimization-based text-to-3D methods distill guidance from 2D generative models via Score Distillation Sampling (SDS), but implicitly treat this guidance as static. This work shows that ignoring source dynamics yields inconsistent trajectories that suppress or merge semantic cues, leading to \"semantic over-smoothing\" artifacts. As such, we reformulate text-to-3D optimization as mapping a dynamically evolving source distribution to a fixed target distribution. We cast the problem into a dual-conditioned latent space, conditioned on both the text prompt and the intermediately rendered image. Given this joint setup, we observe that the image condition naturally anchors the current source distribution. Building on this insight, we introduce AnchorDS, an improved score distillation mechanism that provides state-anchored guidance with image conditions and stabilizes generation. We further penalize erroneous source estimates and design a lightweight filter strategy and fine-tuning strategy that refines the anchor with negligible overhead. AnchorDS produces finer-grained detail, more natural colours, and stronger semantic consistency, particularly for complex prompts, while maintaining efficiency. Extensive experiments show that our method surpasses previous methods in both quality and efficiency.", "date": "2025-11-12", "categories": []} {"id": "2511.11710", "title": "Target-Balanced Score Distillation", "abstract": "Score Distillation Sampling (SDS) enables 3D asset generation by distilling priors from pretrained 2D text-to-image diffusion models, but vanilla SDS suffers from over-saturation and over-smoothing. To mitigate this issue, recent variants have incorporated negative prompts. However, these methods face a critical trade-off: limited texture optimization, or significant texture gains with shape distortion. In this work, we first conduct a systematic analysis and reveal that this trade-off is fundamentally governed by the utilization of the negative prompts, where Target Negative Prompts (TNP) that embed target information in the negative prompts dramatically enhancing texture realism and fidelity but inducing shape distortions. Informed by this key insight, we introduce the Target-Balanced Score Distillation (TBSD). It formulates generation as a multi-objective optimization problem and introduces an adaptive strategy that effectively resolves the aforementioned trade-off. Extensive experiments demonstrate that TBSD significantly outperforms existing state-of-the-art methods, yielding 3D assets with high-fidelity textures and geometrically accurate shape.", "date": "2025-11-12", "categories": []} {"id": "2511.12202", "title": "LSS3D: Learnable Spatial Shifting for Consistent and High-Quality 3D Generation from Single-Image", "abstract": "Recently, multi-view diffusion-based 3D generation methods have gained significant attention. However, these methods often suffer from shape and texture misalignment across generated multi-view images, leading to low-quality 3D generation results, such as incomplete geometric details and textural ghosting. Some methods are mainly optimized for the frontal perspective and exhibit poor robustness to oblique perspective inputs. In this paper, to tackle the above challenges, we propose a high-quality image-to-3D approach, named LSS3D, with learnable spatial shifting to explicitly and effectively handle the multiview inconsistencies and non-frontal input view. Specifically, we assign learnable spatial shifting parameters to each view, and adjust each view towards a spatially consistent target, guided by the reconstructed mesh, resulting in high-quality 3D generation with more complete geometric details and clean textures. Besides, we include the input view as an extra constraint for the optimization, further enhancing robustness to non-frontal input angles, especially for elevated viewpoint inputs. We also provide a comprehensive quantitative evaluation pipeline that can contribute to the community in performance comparisons. Extensive experiments demonstrate that our method consistently achieves leading results in both geometric and texture evaluation metrics across more flexible input viewpoints.", "date": "2025-11-15", "categories": []} {"id": "2511.13309", "title": "DriveLiDAR4D: Sequential and Controllable LiDAR Scene Generation for Autonomous Driving", "abstract": "The generation of realistic LiDAR point clouds plays a crucial role in the development and evaluation of autonomous driving systems. Although recent methods for 3D LiDAR point cloud generation have shown significant improvements, they still face notable limitations, including the lack of sequential generation capabilities and the inability to produce accurately positioned foreground objects and realistic backgrounds. These shortcomings hinder their practical applicability. In this paper, we introduce DriveLiDAR4D, a novel LiDAR generation pipeline consisting of multimodal conditions and a novel sequential noise prediction model LiDAR4DNet, capable of producing temporally consistent LiDAR scenes with highly controllable foreground objects and realistic backgrounds. To the best of our knowledge, this is the first work to address the sequential generation of LiDAR scenes with full scene manipulation capability in an end-to-end manner. We evaluated DriveLiDAR4D on the nuScenes and KITTI datasets, where we achieved an FRD score of 743.13 and an FVD score of 16.96 on the nuScenes dataset, surpassing the current state-of-the-art (SOTA) method, UniScene, with an performance boost of 37.2% in FRD and 24.1% in FVD, respectively.", "date": "2025-11-17", "categories": []} {"id": "2511.13648", "title": "PhysX-Anything: Simulation-Ready Physical 3D Assets from Single Image", "abstract": "3D modeling is shifting from static visual representations toward physical, articulated assets that can be directly used in simulation and interaction. However, most existing 3D generation methods overlook key physical and articulation properties, thereby limiting their utility in embodied AI. To bridge this gap, we introduce PhysX-Anything, the first simulation-ready physical 3D generative framework that, given a single in-the-wild image, produces high-quality sim-ready 3D assets with explicit geometry, articulation, and physical attributes. Specifically, we propose the first VLM-based physical 3D generative model, along with a new 3D representation that efficiently tokenizes geometry. It reduces the number of tokens by 193x, enabling explicit geometry learning within standard VLM token budgets without introducing any special tokens during fine-tuning and significantly improving generative quality. In addition, to overcome the limited diversity of existing physical 3D datasets, we construct a new dataset, PhysX-Mobility, which expands the object categories in prior physical 3D datasets by over 2x and includes more than 2K common real-world objects with rich physical annotations. Extensive experiments on PhysX-Mobility and in-the-wild images demonstrate that PhysX-Anything delivers strong generative performance and robust generalization. Furthermore, simulation-based experiments in a MuJoCo-style environment validate that our sim-ready assets can be directly used for contact-rich robotic policy learning. We believe PhysX-Anything can substantially empower a broad range of downstream applications, especially in embodied AI and physics-based simulation.", "date": "2025-11-17", "categories": []} {"id": "2410.23931", "title": "Manipulating Vehicle 3D Shapes through Latent Space Editing", "abstract": "Although 3D object editing has the potential to significantly influence various industries, recent research in 3D generation and editing has primarily focused on converting text and images into 3D models, often overlooking the need for fine-grained control over the editing of existing 3D objects. This paper introduces a framework that employs a pre-trained regressor, enabling continuous, precise, attribute-specific modifications to both the stylistic and geometric attributes of vehicle 3D models. Our method not only preserves the inherent identity of vehicle 3D objects, but also supports multi-attribute editing, allowing for extensive customization without compromising the model's structural integrity. Experimental results demonstrate the efficacy of our approach in achieving detailed edits on various vehicle 3D models.", "date": "2024-10-31", "categories": []} {"id": "2509.19995", "title": "MeshMosaic: Scaling Artist Mesh Generation via Local-to-Global Assembly", "abstract": "Scaling artist-designed meshes to high triangle numbers remains challenging for autoregressive generative models. Existing transformer-based methods suffer from long-sequence bottlenecks and limited quantization resolution, primarily due to the large number of tokens required and constrained quantization granularity. These issues prevent faithful reproduction of fine geometric details and structured density patterns. We introduce MeshMosaic, a novel local-to-global framework for artist mesh generation that scales to over 100K triangles--substantially surpassing prior methods, which typically handle only around 8K faces. MeshMosaic first segments shapes into patches, generating each patch autoregressively and leveraging shared boundary conditions to promote coherence, symmetry, and seamless connectivity between neighboring regions. This strategy enhances scalability to high-resolution meshes by quantizing patches individually, resulting in more symmetrical and organized mesh density and structure. Extensive experiments across multiple public datasets demonstrate that MeshMosaic significantly outperforms state-of-the-art methods in both geometric fidelity and user preference, supporting superior detail representation and practical mesh generation for real-world applications.", "date": "2025-09-24", "categories": []} {"id": "2511.11470", "title": "Sat2RealCity: Geometry-Aware and Appearance-Controllable 3D Urban Generation from Satellite Imagery", "abstract": "Recent advances in generative modeling have substantially enhanced 3D urban generation, enabling applications in digital twins, virtual cities, and large-scale simulations. However, existing methods face two key challenges: (1) the need for large-scale 3D city assets for supervised training, which are difficult and costly to obtain, and (2) reliance on semantic or height maps, which are used exclusively for generating buildings in virtual worlds and lack connection to real-world appearance, limiting the realism and generalizability of generated cities. To address these limitations, we propose Sat2RealCity, a geometry-aware and appearance-controllable framework for 3D urban generation from real-world satellite imagery. Unlike previous city-level generation methods, Sat2RealCity builds generation upon individual building entities, enabling the use of rich priors and pretrained knowledge from 3D object generation while substantially reducing dependence on large-scale 3D city assets. Specifically, (1) we introduce the OSM-based spatial priors strategy to achieve interpretable geometric generation from spatial topology to building instances; (2) we design an appearance-guided controllable modeling mechanism for fine-grained appearance realism and style control; and (3) we construct an MLLM-powered semantic-guided generation pipeline, bridging semantic interpretation and geometric reconstruction. Extensive quantitative and qualitative experiments demonstrate that Sat2RealCity significantly surpasses existing baselines in structural consistency and appearance realism, establishing a strong foundation for real-world aligned 3D urban content creation. The code will be released soon.", "date": "2025-11-14", "categories": []} {"id": "2505.05469", "title": "Generating Physically Stable and Buildable Brick Structures from Text", "abstract": "We introduce BrickGPT, the first approach for generating physically stable interconnecting brick assembly models from text prompts. To achieve this, we construct a large-scale, physically stable dataset of brick structures, along with their associated captions, and train an autoregressive large language model to predict the next brick to add via next-token prediction. To improve the stability of the resulting designs, we employ an efficient validity check and physics-aware rollback during autoregressive inference, which prunes infeasible token predictions using physics laws and assembly constraints. Our experiments show that BrickGPT produces stable, diverse, and aesthetically pleasing brick structures that align closely with the input text prompts. We also develop a text-based brick texturing method to generate colored and textured designs. We show that our designs can be assembled manually by humans and automatically by robotic arms. We release our new dataset, StableText2Brick, containing over 47,000 brick structures of over 28,000 unique 3D objects accompanied by detailed captions, along with our code and models at the project website: https://avalovelace1.github.io/BrickGPT/.", "date": "2025-05-08", "categories": []} {"id": "2509.26091", "title": "Text-to-Scene with Large Reasoning Models", "abstract": "Prompt-driven scene synthesis allows users to generate complete 3D environments from textual descriptions. Current text-to-scene methods often struggle with complex geometries and object transformations, and tend to show weak adherence to complex instructions. We address these limitations by introducing Reason-3D, a text-to-scene model powered by large reasoning models (LRMs). Reason-3D integrates object retrieval using captions covering physical, functional, and contextual attributes. Reason-3D then places the selected objects based on implicit and explicit layout constraints, and refines their positions with collision-aware spatial reasoning. Evaluated on instructions ranging from simple to complex indoor configurations, Reason-3D significantly outperforms previous methods in human-rated visual fidelity, adherence to constraints, and asset retrieval quality. Beyond its contribution to the field of text-to-scene generation, our work showcases the advanced spatial reasoning abilities of modern LRMs. Additionally, we release the codebase to further the research in object retrieval and placement with LRMs.", "date": "2025-09-30", "categories": []} {"id": "2504.03536", "title": "HumanDreamer-X: Photorealistic Single-image Human Avatars Reconstruction via Gaussian Restoration", "abstract": "Single-image human reconstruction is vital for digital human modeling applications but remains an extremely challenging task. Current approaches rely on generative models to synthesize multi-view images for subsequent 3D reconstruction and animation. However, directly generating multiple views from a single human image suffers from geometric inconsistencies, resulting in issues like fragmented or blurred limbs in the reconstructed models. To tackle these limitations, we introduce \\textbf{HumanDreamer-X}, a novel framework that integrates multi-view human generation and reconstruction into a unified pipeline, which significantly enhances the geometric consistency and visual fidelity of the reconstructed 3D models. In this framework, 3D Gaussian Splatting serves as an explicit 3D representation to provide initial geometry and appearance priority. Building upon this foundation, \\textbf{HumanFixer} is trained to restore 3DGS renderings, which guarantee photorealistic results. Furthermore, we delve into the inherent challenges associated with attention mechanisms in multi-view human generation, and propose an attention modulation strategy that effectively enhances geometric details identity consistency across multi-view. Experimental results demonstrate that our approach markedly improves generation and reconstruction PSNR quality metrics by 16.45% and 12.65%, respectively, achieving a PSNR of up to 25.62 dB, while also showing generalization capabilities on in-the-wild data and applicability to various human reconstruction backbone models.", "date": "2025-04-04", "categories": []} {"id": "2511.08823", "title": "DT-NVS: Diffusion Transformers for Novel View Synthesis", "abstract": "Generating novel views of a natural scene, e.g., every-day scenes both indoors and outdoors, from a single view is an under-explored problem, even though it is an organic extension to the object-centric novel view synthesis. Existing diffusion-based approaches focus rather on small camera movements in real scenes or only consider unnatural object-centric scenes, limiting their potential applications in real-world settings. In this paper we move away from these constrained regimes and propose a 3D diffusion model trained with image-only losses on a large-scale dataset of real-world, multi-category, unaligned, and casually acquired videos of everyday scenes. We propose DT-NVS, a 3D-aware diffusion model for generalized novel view synthesis that exploits a transformer-based architecture backbone. We make significant contributions to transformer and self-attention architectures to translate images to 3d representations, and novel camera conditioning strategies to allow training on real-world unaligned datasets. In addition, we introduce a novel training paradigm swapping the role of reference frame between the conditioning image and the sampled noisy input. We evaluate our approach on the 3D task of generalized novel view synthesis from a single input image and show improvements over state-of-the-art 3D aware diffusion models and deterministic approaches, while generating diverse outputs.", "date": "2025-11-11", "categories": []} {"id": "2511.08203", "title": "Twist and Compute: The Cost of Pose in 3D Generative Diffusion", "abstract": "Despite their impressive results, large-scale image-to-3D generative models remain opaque in their inductive biases. We identify a significant limitation in image-conditioned 3D generative models: a strong canonical view bias. Through controlled experiments using simple 2D rotations, we show that the state-of-the-art Hunyuan3D 2.0 model can struggle to generalize across viewpoints, with performance degrading under rotated inputs. We show that this failure can be mitigated by a lightweight CNN that detects and corrects input orientation, restoring model performance without modifying the generative backbone. Our findings raise an important open question: Is scale enough, or should we pursue modular, symmetry-aware designs?", "date": "2025-11-11", "categories": []} {"id": "2511.05308", "title": "Rethinking Metrics and Diffusion Architecture for 3D Point Cloud Generation", "abstract": "As 3D point clouds become a cornerstone of modern technology, the need for sophisticated generative models and reliable evaluation metrics has grown exponentially. In this work, we first expose that some commonly used metrics for evaluating generated point clouds, particularly those based on Chamfer Distance (CD), lack robustness against defects and fail to capture geometric fidelity and local shape consistency when used as quality indicators. We further show that introducing samples alignment prior to distance calculation and replacing CD with Density-Aware Chamfer Distance (DCD) are simple yet essential steps to ensure the consistency and robustness of point cloud generative model evaluation metrics. While existing metrics primarily focus on directly comparing 3D Euclidean coordinates, we present a novel metric, named Surface Normal Concordance (SNC), which approximates surface similarity by comparing estimated point normals. This new metric, when combined with traditional ones, provides a more comprehensive evaluation of the quality of generated samples. Finally, leveraging recent advancements in transformer-based models for point cloud analysis, such as serialized patch attention , we propose a new architecture for generating high-fidelity 3D structures, the Diffusion Point Transformer. We perform extensive experiments and comparisons on the ShapeNet dataset, showing that our model outperforms previous solutions, particularly in terms of quality of generated point clouds, achieving new state-of-the-art. Code available at https://github.com/matteo-bastico/DiffusionPointTransformer.", "date": "2025-11-07", "categories": []} {"id": "2511.05609", "title": "Walking the Schr\\\"odinger Bridge: A Direct Trajectory for Text-to-3D Generation", "abstract": "Recent advancements in optimization-based text-to-3D generation heavily rely on distilling knowledge from pre-trained text-to-image diffusion models using techniques like Score Distillation Sampling (SDS), which often introduce artifacts such as over-saturation and over-smoothing into the generated 3D assets. In this paper, we address this essential problem by formulating the generation process as learning an optimal, direct transport trajectory between the distribution of the current rendering and the desired target distribution, thereby enabling high-quality generation with smaller Classifier-free Guidance (CFG) values. At first, we theoretically establish SDS as a simplified instance of the Schr\\\"odinger Bridge framework. We prove that SDS employs the reverse process of an Schr\\\"odinger Bridge, which, under specific conditions (e.g., a Gaussian noise as one end), collapses to SDS's score function of the pre-trained diffusion model. Based upon this, we introduce Trajectory-Centric Distillation (TraCe), a novel text-to-3D generation framework, which reformulates the mathematically trackable framework of Schr\\\"odinger Bridge to explicitly construct a diffusion bridge from the current rendering to its text-conditioned, denoised target, and trains a LoRA-adapted model on this trajectory's score dynamics for robust 3D optimization. Comprehensive experiments demonstrate that TraCe consistently achieves superior quality and fidelity to state-of-the-art techniques.", "date": "2025-11-06", "categories": []} {"id": "2511.07067", "title": "RaLD: Generating High-Resolution 3D Radar Point Clouds with Latent Diffusion", "abstract": "Millimeter-wave radar offers a promising sensing modality for autonomous systems thanks to its robustness in adverse conditions and low cost. However, its utility is significantly limited by the sparsity and low resolution of radar point clouds, which poses challenges for tasks requiring dense and accurate 3D perception. Despite that recent efforts have shown great potential by exploring generative approaches to address this issue, they often rely on dense voxel representations that are inefficient and struggle to preserve structural detail. To fill this gap, we make the key observation that latent diffusion models (LDMs), though successful in other modalities, have not been effectively leveraged for radar-based 3D generation due to a lack of compatible representations and conditioning strategies. We introduce RaLD, a framework that bridges this gap by integrating scene-level frustum-based LiDAR autoencoding, order-invariant latent representations, and direct radar spectrum conditioning. These insights lead to a more compact and expressive generation process. Experiments show that RaLD produces dense and accurate 3D point clouds from raw radar spectrums, offering a promising solution for robust perception in challenging environments.", "date": "2025-11-10", "categories": []} {"id": "2511.07142", "title": "ProcGen3D: Learning Neural Procedural Graph Representations for Image-to-3D Reconstruction", "abstract": "We introduce ProcGen3D, a new approach for 3D content creation by generating procedural graph abstractions of 3D objects, which can then be decoded into rich, complex 3D assets. Inspired by the prevalent use of procedural generators in production 3D applications, we propose a sequentialized, graph-based procedural graph representation for 3D assets. We use this to learn to approximate the landscape of a procedural generator for image-based 3D reconstruction. We employ edge-based tokenization to encode the procedural graphs, and train a transformer prior to predict the next token conditioned on an input RGB image. Crucially, to enable better alignment of our generated outputs to an input image, we incorporate Monte Carlo Tree Search (MCTS) guided sampling into our generation process, steering output procedural graphs towards more image-faithful reconstructions. Our approach is applicable across a variety of objects that can be synthesized with procedural generators. Extensive experiments on cacti, trees, and bridges show that our neural procedural graph generation outperforms both state-of-the-art generative 3D methods and domain-specific modeling techniques. Furthermore, this enables improved generalization on real-world input images, despite training only on synthetic data.", "date": "2025-11-10", "categories": []} {"id": "2503.13816", "title": "MOSAIC: Generating Consistent, Privacy-Preserving Scenes from Multiple Depth Views in Multi-Room Environments", "abstract": "We introduce a diffusion-based approach for generating privacy-preserving digital twins of multi-room indoor environments from depth images only. Central to our approach is a novel Multi-view Overlapped Scene Alignment with Implicit Consistency (MOSAIC) model that explicitly considers cross-view dependencies within the same scene in the probabilistic sense. MOSAIC operates through a multi-channel inference-time optimization that avoids error accumulation common in sequential or single-room constraints in panorama-based approaches. MOSAIC scales to complex scenes with zero extra training and provably reduces the variance during denoising process when more overlapping views are added, leading to improved generation quality. Experiments show that MOSAIC outperforms state-of-the-art baselines on image fidelity metrics in reconstructing complex multi-room environments. Resources and code are at https://mosaic-cmubig.github.io", "date": "2025-03-18", "categories": []} {"id": "2504.01619", "title": "3DBonsai: Structure-Aware Bonsai Modeling Using Conditioned 3D Gaussian Splatting", "abstract": "Recent advancements in text-to-3D generation have shown remarkable results by leveraging 3D priors in combination with 2D diffusion. However, previous methods utilize 3D priors that lack detailed and complex structural information, limiting them to generating simple objects and presenting challenges for creating intricate structures such as bonsai. In this paper, we propose 3DBonsai, a novel text-to-3D framework for generating 3D bonsai with complex structures. Technically, we first design a trainable 3D space colonization algorithm to produce bonsai structures, which are then enhanced through random sampling and point cloud augmentation to serve as the 3D Gaussian priors. We introduce two bonsai generation pipelines with distinct structural levels: fine structure conditioned generation, which initializes 3D Gaussians using a 3D structure prior to produce detailed and complex bonsai, and coarse structure conditioned generation, which employs a multi-view structure consistency module to align 2D and 3D structures. Moreover, we have compiled a unified 2D and 3D Chinese-style bonsai dataset. Our experimental results demonstrate that 3DBonsai significantly outperforms existing methods, providing a new benchmark for structure-aware 3D bonsai generation.", "date": "2025-04-02", "categories": []} {"id": "2510.25765", "title": "FreeArt3D: Training-Free Articulated Object Generation using 3D Diffusion", "abstract": "Articulated 3D objects are central to many applications in robotics, AR/VR, and animation. Recent approaches to modeling such objects either rely on optimization-based reconstruction pipelines that require dense-view supervision or on feed-forward generative models that produce coarse geometric approximations and often overlook surface texture. In contrast, open-world 3D generation of static objects has achieved remarkable success, especially with the advent of native 3D diffusion models such as Trellis. However, extending these methods to articulated objects by training native 3D diffusion models poses significant challenges. In this work, we present FreeArt3D, a training-free framework for articulated 3D object generation. Instead of training a new model on limited articulated data, FreeArt3D repurposes a pre-trained static 3D diffusion model (e.g., Trellis) as a powerful shape prior. It extends Score Distillation Sampling (SDS) into the 3D-to-4D domain by treating articulation as an additional generative dimension. Given a few images captured in different articulation states, FreeArt3D jointly optimizes the object's geometry, texture, and articulation parameters without requiring task-specific training or access to large-scale articulated datasets. Our method generates high-fidelity geometry and textures, accurately predicts underlying kinematic structures, and generalizes well across diverse object categories. Despite following a per-instance optimization paradigm, FreeArt3D completes in minutes and significantly outperforms prior state-of-the-art approaches in both quality and versatility. Please check our website for more details: https://czzzzh.github.io/FreeArt3D", "date": "2025-10-29", "categories": []} {"id": "2510.08271", "title": "SViM3D: Stable Video Material Diffusion for Single Image 3D Generation", "abstract": "We present Stable Video Materials 3D (SViM3D), a framework to predict multi-view consistent physically based rendering (PBR) materials, given a single image. Recently, video diffusion models have been successfully used to reconstruct 3D objects from a single image efficiently. However, reflectance is still represented by simple material models or needs to be estimated in additional steps to enable relighting and controlled appearance edits. We extend a latent video diffusion model to output spatially varying PBR parameters and surface normals jointly with each generated view based on explicit camera control. This unique setup allows for relighting and generating a 3D asset using our model as neural prior. We introduce various mechanisms to this pipeline that improve quality in this ill-posed setting. We show state-of-the-art relighting and novel view synthesis performance on multiple object-centric datasets. Our method generalizes to diverse inputs, enabling the generation of relightable 3D assets useful in AR/VR, movies, games and other visual media.", "date": "2025-10-09", "categories": []} {"id": "2510.04539", "title": "C3Editor: Achieving Controllable Consistency in 2D Model for 3D Editing", "abstract": "Existing 2D-lifting-based 3D editing methods often encounter challenges related to inconsistency, stemming from the lack of view-consistent 2D editing models and the difficulty of ensuring consistent editing across multiple views. To address these issues, we propose C3Editor, a controllable and consistent 2D-lifting-based 3D editing framework. Given an original 3D representation and a text-based editing prompt, our method selectively establishes a view-consistent 2D editing model to achieve superior 3D editing results. The process begins with the controlled selection of a ground truth (GT) view and its corresponding edited image as the optimization target, allowing for user-defined manual edits. Next, we fine-tune the 2D editing model within the GT view and across multiple views to align with the GT-edited image while ensuring multi-view consistency. To meet the distinct requirements of GT view fitting and multi-view consistency, we introduce separate LoRA modules for targeted fine-tuning. Our approach delivers more consistent and controllable 2D and 3D editing results than existing 2D-lifting-based methods, outperforming them in both qualitative and quantitative evaluations.", "date": "2025-10-06", "categories": []} {"id": "2510.27148", "title": "HiGS: Hierarchical Generative Scene Framework for Multi-Step Associative Semantic Spatial Composition", "abstract": "Three-dimensional scene generation holds significant potential in gaming, film, and virtual reality. However, most existing methods adopt a single-step generation process, making it difficult to balance scene complexity with minimal user input. Inspired by the human cognitive process in scene modeling, which progresses from global to local, focuses on key elements, and completes the scene through semantic association, we propose HiGS, a hierarchical generative framework for multi-step associative semantic spatial composition. HiGS enables users to iteratively expand scenes by selecting key semantic objects, offering fine-grained control over regions of interest while the model completes peripheral areas automatically. To support structured and coherent generation, we introduce the Progressive Hierarchical Spatial-Semantic Graph (PHiSSG), which dynamically organizes spatial relationships and semantic dependencies across the evolving scene structure. PHiSSG ensures spatial and geometric consistency throughout the generation process by maintaining a one-to-one mapping between graph nodes and generated objects and supporting recursive layout optimization. Experiments demonstrate that HiGS outperforms single-stage methods in layout plausibility, style consistency, and user preference, offering a controllable and extensible paradigm for efficient 3D scene construction.", "date": "2025-10-31", "categories": []} {"id": "2510.27497", "title": "InertialAR: Autoregressive 3D Molecule Generation with Inertial Frames", "abstract": "Transformer-based autoregressive models have emerged as a unifying paradigm across modalities such as text and images, but their extension to 3D molecule generation remains underexplored. The gap stems from two fundamental challenges: (1) tokenizing molecules into a canonical 1D sequence of tokens that is invariant to both SE(3) transformations and atom index permutations, and (2) designing an architecture capable of modeling hybrid atom-based tokens that couple discrete atom types with continuous 3D coordinates. To address these challenges, we introduce InertialAR. InertialAR devises a canonical tokenization that aligns molecules to their inertial frames and reorders atoms to ensure SE(3) and permutation invariance. Moreover, InertialAR equips the attention mechanism with geometric awareness via geometric rotary positional encoding (GeoRoPE). In addition, it utilizes a hierarchical autoregressive paradigm to predict the next atom-based token, predicting the atom type first and then its 3D coordinates via Diffusion loss. Experimentally, InertialAR achieves state-of-the-art performance on 7 of the 10 evaluation metrics for unconditional molecule generation across QM9, GEOM-Drugs, and B3LYP. Moreover, it significantly outperforms strong baselines in controllable generation for targeted chemical functionality, attaining state-of-the-art results across all 5 metrics.", "date": "2025-10-31", "categories": []} {"id": "2507.08772", "title": "From One to More: Contextual Part Latents for 3D Generation", "abstract": "Recent advances in 3D generation have transitioned from multi-view 2D rendering approaches to 3D-native latent diffusion frameworks that exploit geometric priors in ground truth data. Despite progress, three key limitations persist: (1) Single-latent representations fail to capture complex multi-part geometries, causing detail degradation; (2) Holistic latent coding neglects part independence and interrelationships critical for compositional design; (3) Global conditioning mechanisms lack fine-grained controllability. Inspired by human 3D design workflows, we propose CoPart - a part-aware diffusion framework that decomposes 3D objects into contextual part latents for coherent multi-part generation. This paradigm offers three advantages: i) Reduces encoding complexity through part decomposition; ii) Enables explicit part relationship modeling; iii) Supports part-level conditioning. We further develop a mutual guidance strategy to fine-tune pre-trained diffusion models for joint part latent denoising, ensuring both geometric coherence and foundation model priors. To enable large-scale training, we construct Partverse - a novel 3D part dataset derived from Objaverse through automated mesh segmentation and human-verified annotations. Extensive experiments demonstrate CoPart's superior capabilities in part-level editing, articulated object generation, and scene composition with unprecedented controllability.", "date": "2025-07-11", "categories": []} {"id": "2510.26140", "title": "FullPart: Generating each 3D Part at Full Resolution", "abstract": "Part-based 3D generation holds great potential for various applications. Previous part generators that represent parts using implicit vector-set tokens often suffer from insufficient geometric details. Another line of work adopts an explicit voxel representation but shares a global voxel grid among all parts; this often causes small parts to occupy too few voxels, leading to degraded quality. In this paper, we propose FullPart, a novel framework that combines both implicit and explicit paradigms. It first derives the bounding box layout through an implicit box vector-set diffusion process, a task that implicit diffusion handles effectively since box tokens contain little geometric detail. Then, it generates detailed parts, each within its own fixed full-resolution voxel grid. Instead of sharing a global low-resolution space, each part in our method - even small ones - is generated at full resolution, enabling the synthesis of intricate details. We further introduce a center-point encoding strategy to address the misalignment issue when exchanging information between parts of different actual sizes, thereby maintaining global coherence. Moreover, to tackle the scarcity of reliable part data, we present PartVerse-XL, the largest human-annotated 3D part dataset to date with 40K objects and 320K parts. Extensive experiments demonstrate that FullPart achieves state-of-the-art results in 3D part generation. We will release all code, data, and model to benefit future research in 3D part generation.", "date": "2025-10-30", "categories": []} {"id": "2510.26800", "title": "OmniX: From Unified Panoramic Generation and Perception to Graphics-Ready 3D Scenes", "abstract": "There are two prevalent ways to constructing 3D scenes: procedural generation and 2D lifting. Among them, panorama-based 2D lifting has emerged as a promising technique, leveraging powerful 2D generative priors to produce immersive, realistic, and diverse 3D environments. In this work, we advance this technique to generate graphics-ready 3D scenes suitable for physically based rendering (PBR), relighting, and simulation. Our key insight is to repurpose 2D generative models for panoramic perception of geometry, textures, and PBR materials. Unlike existing 2D lifting approaches that emphasize appearance generation and ignore the perception of intrinsic properties, we present OmniX, a versatile and unified framework. Based on a lightweight and efficient cross-modal adapter structure, OmniX reuses 2D generative priors for a broad range of panoramic vision tasks, including panoramic perception, generation, and completion. Furthermore, we construct a large-scale synthetic panorama dataset containing high-quality multimodal panoramas from diverse indoor and outdoor scenes. Extensive experiments demonstrate the effectiveness of our model in panoramic visual perception and graphics-ready 3D scene generation, opening new possibilities for immersive and physically realistic virtual world generation.", "date": "2025-10-30", "categories": []} {"id": "2510.25319", "title": "4-Doodle: Text to 3D Sketches that Move!", "abstract": "We present a novel task: text-to-3D sketch animation, which aims to bring freeform sketches to life in dynamic 3D space. Unlike prior works focused on photorealistic content generation, we target sparse, stylized, and view-consistent 3D vector sketches, a lightweight and interpretable medium well-suited for visual communication and prototyping. However, this task is very challenging: (i) no paired dataset exists for text and 3D (or 4D) sketches; (ii) sketches require structural abstraction that is difficult to model with conventional 3D representations like NeRFs or point clouds; and (iii) animating such sketches demands temporal coherence and multi-view consistency, which current pipelines do not address. Therefore, we propose 4-Doodle, the first training-free framework for generating dynamic 3D sketches from text. It leverages pretrained image and video diffusion models through a dual-space distillation scheme: one space captures multi-view-consistent geometry using differentiable B\\'ezier curves, while the other encodes motion dynamics via temporally-aware priors. Unlike prior work (e.g., DreamFusion), which optimizes from a single view per step, our multi-view optimization ensures structural alignment and avoids view ambiguity, critical for sparse sketches. Furthermore, we introduce a structure-aware motion module that separates shape-preserving trajectories from deformation-aware changes, enabling expressive motion such as flipping, rotation, and articulated movement. Extensive experiments show that our method produces temporally realistic and structurally stable 3D sketch animations, outperforming existing baselines in both fidelity and controllability. We hope this work serves as a step toward more intuitive and accessible 4D content creation.", "date": "2025-10-29", "categories": []} {"id": "2510.23929", "title": "TurboPortrait3D: Single-step diffusion-based fast portrait novel-view synthesis", "abstract": "We introduce TurboPortrait3D: a method for low-latency novel-view synthesis of human portraits. Our approach builds on the observation that existing image-to-3D models for portrait generation, while capable of producing renderable 3D representations, are prone to visual artifacts, often lack of detail, and tend to fail at fully preserving the identity of the subject. On the other hand, image diffusion models excel at generating high-quality images, but besides being computationally expensive, are not grounded in 3D and thus are not directly capable of producing multi-view consistent outputs. In this work, we demonstrate that image-space diffusion models can be used to significantly enhance the quality of existing image-to-avatar methods, while maintaining 3D-awareness and running with low-latency. Our method takes a single frontal image of a subject as input, and applies a feedforward image-to-avatar generation pipeline to obtain an initial 3D representation and corresponding noisy renders. These noisy renders are then fed to a single-step diffusion model which is conditioned on input image(s), and is specifically trained to refine the renders in a multi-view consistent way. Moreover, we introduce a novel effective training strategy that includes pre-training on a large corpus of synthetic multi-view data, followed by fine-tuning on high-quality real images. We demonstrate that our approach both qualitatively and quantitatively outperforms current state-of-the-art for portrait novel-view synthesis, while being efficient in time.", "date": "2025-10-27", "categories": []} {"id": "2509.20414", "title": "SceneWeaver: All-in-One 3D Scene Synthesis with an Extensible and Self-Reflective Agent", "abstract": "Indoor scene synthesis has become increasingly important with the rise of Embodied AI, which requires 3D environments that are not only visually realistic but also physically plausible and functionally diverse. While recent approaches have advanced visual fidelity, they often remain constrained to fixed scene categories, lack sufficient object-level detail and physical consistency, and struggle to align with complex user instructions. In this work, we present SceneWeaver, a reflective agentic framework that unifies diverse scene synthesis paradigms through tool-based iterative refinement. At its core, SceneWeaver employs a language model-based planner to select from a suite of extensible scene generation tools, ranging from data-driven generative models to visual- and LLM-based methods, guided by self-evaluation of physical plausibility, visual realism, and semantic alignment with user input. This closed-loop reason-act-reflect design enables the agent to identify semantic inconsistencies, invoke targeted tools, and update the environment over successive iterations. Extensive experiments on both common and open-vocabulary room types demonstrate that SceneWeaver not only outperforms prior methods on physical, visual, and semantic metrics, but also generalizes effectively to complex scenes with diverse instructions, marking a step toward general-purpose 3D environment generation. Project website: https://scene-weaver.github.io/.", "date": "2025-09-24", "categories": []} {"id": "2510.21264", "title": "Topology Sculptor, Shape Refiner: Discrete Diffusion Model for High-Fidelity 3D Meshes Generation", "abstract": "In this paper, we introduce Topology Sculptor, Shape Refiner (TSSR), a novel method for generating high-quality, artist-style 3D meshes based on Discrete Diffusion Models (DDMs). Our primary motivation for TSSR is to achieve highly accurate token prediction while enabling parallel generation, a significant advantage over sequential autoregressive methods. By allowing TSSR to \"see\" all mesh tokens concurrently, we unlock a new level of efficiency and control. We leverage this parallel generation capability through three key innovations: 1) Decoupled Training and Hybrid Inference, which distinctly separates the DDM-based generation into a topology sculpting stage and a subsequent shape refinement stage. This strategic decoupling enables TSSR to effectively capture both intricate local topology and overarching global shape. 2) An Improved Hourglass Architecture, featuring bidirectional attention enriched by face-vertex-sequence level Rotational Positional Embeddings (RoPE), thereby capturing richer contextual information across the mesh structure. 3) A novel Connection Loss, which acts as a topological constraint to further enhance the realism and fidelity of the generated meshes. Extensive experiments on complex datasets demonstrate that TSSR generates high-quality 3D artist-style meshes, capable of achieving up to 10,000 faces at a remarkable spatial resolution of $1024^3$. The code will be released at: https://github.com/psky1111/Tencent-TSSR.", "date": "2025-10-24", "categories": []} {"id": "2510.22337", "title": "GeoDiffusion: A Training-Free Framework for Accurate 3D Geometric Conditioning in Image Generation", "abstract": "Precise geometric control in image generation is essential for engineering \\& product design and creative industries to control 3D object features accurately in image space. Traditional 3D editing approaches are time-consuming and demand specialized skills, while current image-based generative methods lack accuracy in geometric conditioning. To address these challenges, we propose GeoDiffusion, a training-free framework for accurate and efficient geometric conditioning of 3D features in image generation. GeoDiffusion employs a class-specific 3D object as a geometric prior to define keypoints and parametric correlations in 3D space. We ensure viewpoint consistency through a rendered image of a reference 3D object, followed by style transfer to meet user-defined appearance specifications. At the core of our framework is GeoDrag, improving accuracy and speed of drag-based image editing on geometry guidance tasks and general instructions on DragBench. Our results demonstrate that GeoDiffusion enables precise geometric modifications across various iterative design workflows.", "date": "2025-10-25", "categories": []} {"id": "2510.23605", "title": "Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture Infilling", "abstract": "Current 3D/4D generation methods are usually optimized for photorealism, efficiency, and aesthetics. However, they often fail to preserve the semantic identity of the subject across different viewpoints. Adapting generation methods with one or few images of a specific subject (also known as Personalization or Subject-driven generation) allows generating visual content that align with the identity of the subject. However, personalized 3D/4D generation is still largely underexplored. In this work, we introduce TIRE (Track, Inpaint, REsplat), a novel method for subject-driven 3D/4D generation. It takes an initial 3D asset produced by an existing 3D generative model as input and uses video tracking to identify the regions that need to be modified. Then, we adopt a subject-driven 2D inpainting model for progressively infilling the identified regions. Finally, we resplat the modified 2D multi-view observations back to 3D while still maintaining consistency. Extensive experiments demonstrate that our approach significantly improves identity preservation in 3D/4D generation compared to state-of-the-art methods. Our project website is available at https://zsh2000.github.io/track-inpaint-resplat.github.io/.", "date": "2025-10-27", "categories": []} {"id": "2510.19944", "title": "Seed3D 1.0: From Images to High-Fidelity Simulation-Ready 3D Assets", "abstract": "Developing embodied AI agents requires scalable training environments that balance content diversity with physics accuracy. World simulators provide such environments but face distinct limitations: video-based methods generate diverse content but lack real-time physics feedback for interactive learning, while physics-based engines provide accurate dynamics but face scalability limitations from costly manual asset creation. We present Seed3D 1.0, a foundation model that generates simulation-ready 3D assets from single images, addressing the scalability challenge while maintaining physics rigor. Unlike existing 3D generation models, our system produces assets with accurate geometry, well-aligned textures, and realistic physically-based materials. These assets can be directly integrated into physics engines with minimal configuration, enabling deployment in robotic manipulation and simulation training. Beyond individual objects, the system scales to complete scene generation through assembling objects into coherent environments. By enabling scalable simulation-ready content creation, Seed3D 1.0 provides a foundation for advancing physics-based world simulators. Seed3D 1.0 is now available on https://console.volcengine.com/ark/region:ark+cn-beijing/experience/vision?modelId=doubao-seed3d-1-0-250928&tab=Gen3D", "date": "2025-10-22", "categories": []} {"id": "2510.21432", "title": "ArtiLatent: Realistic Articulated 3D Object Generation via Structured Latents", "abstract": "We propose ArtiLatent, a generative framework that synthesizes human-made 3D objects with fine-grained geometry, accurate articulation, and realistic appearance. Our approach jointly models part geometry and articulation dynamics by embedding sparse voxel representations and associated articulation properties, including joint type, axis, origin, range, and part category, into a unified latent space via a variational autoencoder. A latent diffusion model is then trained over this space to enable diverse yet physically plausible sampling. To reconstruct photorealistic 3D shapes, we introduce an articulation-aware Gaussian decoder that accounts for articulation-dependent visibility changes (e.g., revealing the interior of a drawer when opened). By conditioning appearance decoding on articulation state, our method assigns plausible texture features to regions that are typically occluded in static poses, significantly improving visual realism across articulation configurations. Extensive experiments on furniture-like objects from PartNet-Mobility and ACD datasets demonstrate that ArtiLatent outperforms existing approaches in geometric consistency and appearance fidelity. Our framework provides a scalable solution for articulated 3D object synthesis and manipulation.", "date": "2025-10-24", "categories": []} {"id": "2510.21682", "title": "WorldGrow: Generating Infinite 3D World", "abstract": "We tackle the challenge of generating the infinitely extendable 3D world -- large, continuous environments with coherent geometry and realistic appearance. Existing methods face key challenges: 2D-lifting approaches suffer from geometric and appearance inconsistencies across views, 3D implicit representations are hard to scale up, and current 3D foundation models are mostly object-centric, limiting their applicability to scene-level generation. Our key insight is leveraging strong generation priors from pre-trained 3D models for structured scene block generation. To this end, we propose WorldGrow, a hierarchical framework for unbounded 3D scene synthesis. Our method features three core components: (1) a data curation pipeline that extracts high-quality scene blocks for training, making the 3D structured latent representations suitable for scene generation; (2) a 3D block inpainting mechanism that enables context-aware scene extension; and (3) a coarse-to-fine generation strategy that ensures both global layout plausibility and local geometric/textural fidelity. Evaluated on the large-scale 3D-FRONT dataset, WorldGrow achieves SOTA performance in geometry reconstruction, while uniquely supporting infinite scene generation with photorealistic and structurally consistent outputs. These results highlight its capability for constructing large-scale virtual environments and potential for building future world models.", "date": "2025-10-24", "categories": []} {"id": "2505.16761", "title": "Mesh-RFT: Enhancing Mesh Generation via Fine-grained Reinforcement Fine-Tuning", "abstract": "Existing pretrained models for 3D mesh generation often suffer from data biases and produce low-quality results, while global reinforcement learning (RL) methods rely on object-level rewards that struggle to capture local structure details. To address these challenges, we present Mesh-RFT, a novel fine-grained reinforcement fine-tuning framework that employs Masked Direct Preference Optimization (M-DPO) to enable localized refinement via quality-aware face masking. To facilitate efficient quality evaluation, we introduce an objective topology-aware scoring system to evaluate geometric integrity and topological regularity at both object and face levels through two metrics: Boundary Edge Ratio (BER) and Topology Score (TS). By integrating these metrics into a fine-grained RL strategy, Mesh-RFT becomes the first method to optimize mesh quality at the granularity of individual faces, resolving localized errors while preserving global coherence. Experiment results show that our M-DPO approach reduces Hausdorff Distance (HD) by 24.6% and improves Topology Score (TS) by 3.8% over pre-trained models, while outperforming global DPO methods with a 17.4% HD reduction and 4.9% TS gain. These results demonstrate Mesh-RFT's ability to improve geometric integrity and topological regularity, achieving new state-of-the-art performance in production-ready mesh generation. Project Page: https://hitcslj.github.io/mesh-rft/.", "date": "2025-05-22", "categories": []} {"id": "2506.05341", "title": "Direct Numerical Layout Generation for 3D Indoor Scene Synthesis via Spatial Reasoning", "abstract": "Realistic 3D indoor scene synthesis is vital for embodied AI and digital content creation. It can be naturally divided into two subtasks: object generation and layout generation. While recent generative models have significantly advanced object-level quality and controllability, layout generation remains challenging due to limited datasets. Existing methods either overfit to these datasets or rely on predefined constraints to optimize numerical layout that sacrifice flexibility. As a result, they fail to generate scenes that are both open-vocabulary and aligned with fine-grained user instructions. We introduce DirectLayout, a framework that directly generates numerical 3D layouts from text descriptions using generalizable spatial reasoning of large language models (LLMs). DirectLayout decomposes the generation into three stages: producing a Bird's-Eye View (BEV) layout, lifting it into 3D space, and refining object placements. To enable explicit spatial reasoning and help the model grasp basic principles of object placement, we employ Chain-of-Thought (CoT) Activation based on the 3D-Front dataset. Additionally, we design CoT-Grounded Generative Layout Reward to enhance generalization and spatial planning. During inference, DirectLayout addresses asset-layout mismatches via Iterative Asset-Layout Alignment through in-context learning. Extensive experiments demonstrate that DirectLayout achieves impressive semantic consistency, generalization and physical plausibility.", "date": "2025-06-05", "categories": []} {"id": "2510.18287", "title": "Efficient Few-shot Identity Preserving Attribute Editing for 3D-aware Deep Generative Models", "abstract": "Identity preserving editing of faces is a generative task that enables modifying the illumination, adding/removing eyeglasses, face aging, editing hairstyles, modifying expression etc., while preserving the identity of the face. Recent progress in 2D generative models have enabled photorealistic editing of faces using simple techniques leveraging the compositionality in GANs. However, identity preserving editing for 3D faces with a given set of attributes is a challenging task as the generative model must reason about view consistency from multiple poses and render a realistic 3D face. Further, 3D portrait editing requires large-scale attribute labelled datasets and presents a trade-off between editability in low-resolution and inflexibility to editing in high resolution. In this work, we aim to alleviate some of the constraints in editing 3D faces by identifying latent space directions that correspond to photorealistic edits. To address this, we present a method that builds on recent advancements in 3D-aware deep generative models and 2D portrait editing techniques to perform efficient few-shot identity preserving attribute editing for 3D-aware generative models. We aim to show from experimental results that using just ten or fewer labelled images of an attribute is sufficient to estimate edit directions in the latent space that correspond to 3D-aware attribute editing. In this work, we leverage an existing face dataset with masks to obtain the synthetic images for few attribute examples required for estimating the edit directions. Further, to demonstrate the linearity of edits, we investigate one-shot stylization by performing sequential editing and use the (2D) Attribute Style Manipulation (ASM) technique to investigate a continuous style manifold for 3D consistent identity preserving face aging. Code and results are available at: https://vishal-vinod.github.io/gmpi-edit/", "date": "2025-10-21", "categories": []} {"id": "2506.21152", "title": "Geometry and Perception Guided Gaussians for Multiview-consistent 3D Generation from a Single Image", "abstract": "Generating realistic 3D objects from single-view images requires natural appearance, 3D consistency, and the ability to capture multiple plausible interpretations of unseen regions. Existing approaches often rely on fine-tuning pretrained 2D diffusion models or directly generating 3D information through fast network inference or 3D Gaussian Splatting, but their results generally suffer from poor multiview consistency and lack geometric detail. To tackle these issues, we present a novel method that seamlessly integrates geometry and perception information without requiring additional model training to reconstruct detailed 3D objects from a single image. Specifically, we incorporate geometry and perception priors to initialize the Gaussian branches and guide their parameter optimization. The geometry prior captures the rough 3D shapes, while the perception prior utilizes the 2D pretrained diffusion model to enhance multiview information. Subsequently, we introduce a stable Score Distillation Sampling for fine-grained prior distillation to ensure effective knowledge transfer. The model is further enhanced by a reprojection-based strategy that enforces depth consistency. Experimental results show that we outperform existing methods on novel view synthesis and 3D reconstruction, demonstrating robust and consistent 3D object generation.", "date": "2025-06-26", "categories": []} {"id": "2510.16136", "title": "GuideFlow3D: Optimization-Guided Rectified Flow For Appearance Transfer", "abstract": "Transferring appearance to 3D assets using different representations of the appearance object - such as images or text - has garnered interest due to its wide range of applications in industries like gaming, augmented reality, and digital content creation. However, state-of-the-art methods still fail when the geometry between the input and appearance objects is significantly different. A straightforward approach is to directly apply a 3D generative model, but we show that this ultimately fails to produce appealing results. Instead, we propose a principled approach inspired by universal guidance. Given a pretrained rectified flow model conditioned on image or text, our training-free method interacts with the sampling process by periodically adding guidance. This guidance can be modeled as a differentiable loss function, and we experiment with two different types of guidance including part-aware losses for appearance and self-similarity. Our experiments show that our approach successfully transfers texture and geometric details to the input 3D asset, outperforming baselines both qualitatively and quantitatively. We also show that traditional metrics are not suitable for evaluating the task due to their inability of focusing on local details and comparing dissimilar inputs, in absence of ground truth data. We thus evaluate appearance transfer quality with a GPT-based system objectively ranking outputs, ensuring robust and human-like assessment, as further confirmed by our user study. Beyond showcased scenarios, our method is general and could be extended to different types of diffusion models and guidance functions.", "date": "2025-10-17", "categories": []} {"id": "2510.17603", "title": "ShapeCraft: LLM Agents for Structured, Textured and Interactive 3D Modeling", "abstract": "3D generation from natural language offers significant potential to reduce expert manual modeling efforts and enhance accessibility to 3D assets. However, existing methods often yield unstructured meshes and exhibit poor interactivity, making them impractical for artistic workflows. To address these limitations, we represent 3D assets as shape programs and introduce ShapeCraft, a novel multi-agent framework for text-to-3D generation. At its core, we propose a Graph-based Procedural Shape (GPS) representation that decomposes complex natural language into a structured graph of sub-tasks, thereby facilitating accurate LLM comprehension and interpretation of spatial relationships and semantic shape details. Specifically, LLM agents hierarchically parse user input to initialize GPS, then iteratively refine procedural modeling and painting to produce structured, textured, and interactive 3D assets. Qualitative and quantitative experiments demonstrate ShapeCraft's superior performance in generating geometrically accurate and semantically rich 3D assets compared to existing LLM-based agents. We further show the versatility of ShapeCraft through examples of animated and user-customized editing, highlighting its potential for broader interactive applications.", "date": "2025-10-20", "categories": []} {"id": "2503.14445", "title": "Bolt3D: Generating 3D Scenes in Seconds", "abstract": "We present a latent diffusion model for fast feed-forward 3D scene generation. Given one or more images, our model Bolt3D directly samples a 3D scene representation in less than seven seconds on a single GPU. We achieve this by leveraging powerful and scalable existing 2D diffusion network architectures to produce consistent high-fidelity 3D scene representations. To train this model, we create a large-scale multiview-consistent dataset of 3D geometry and appearance by applying state-of-the-art dense 3D reconstruction techniques to existing multiview image datasets. Compared to prior multiview generative models that require per-scene optimization for 3D reconstruction, Bolt3D reduces the inference cost by a factor of up to 300 times.", "date": "2025-03-18", "categories": []} {"id": "2510.15019", "title": "NANO3D: A Training-Free Approach for Efficient 3D Editing Without Masks", "abstract": "3D object editing is essential for interactive content creation in gaming, animation, and robotics, yet current approaches remain inefficient, inconsistent, and often fail to preserve unedited regions. Most methods rely on editing multi-view renderings followed by reconstruction, which introduces artifacts and limits practicality. To address these challenges, we propose Nano3D, a training-free framework for precise and coherent 3D object editing without masks. Nano3D integrates FlowEdit into TRELLIS to perform localized edits guided by front-view renderings, and further introduces region-aware merging strategies, Voxel/Slat-Merge, which adaptively preserve structural fidelity by ensuring consistency between edited and unedited areas. Experiments demonstrate that Nano3D achieves superior 3D consistency and visual quality compared with existing methods. Based on this framework, we construct the first large-scale 3D editing datasets Nano3D-Edit-100k, which contains over 100,000 high-quality 3D editing pairs. This work addresses long-standing challenges in both algorithm design and data availability, significantly improving the generality and reliability of 3D editing, and laying the groundwork for the development of feed-forward 3D editing models. Project Page:https://jamesyjl.github.io/Nano3D", "date": "2025-10-16", "categories": []} {"id": "2510.15564", "title": "Imaginarium: Vision-guided High-Quality 3D Scene Layout Generation", "abstract": "Generating artistic and coherent 3D scene layouts is crucial in digital content creation. Traditional optimization-based methods are often constrained by cumbersome manual rules, while deep generative models face challenges in producing content with richness and diversity. Furthermore, approaches that utilize large language models frequently lack robustness and fail to accurately capture complex spatial relationships. To address these challenges, this paper presents a novel vision-guided 3D layout generation system. We first construct a high-quality asset library containing 2,037 scene assets and 147 3D scene layouts. Subsequently, we employ an image generation model to expand prompt representations into images, fine-tuning it to align with our asset library. We then develop a robust image parsing module to recover the 3D layout of scenes based on visual semantics and geometric information. Finally, we optimize the scene layout using scene graphs and overall visual semantics to ensure logical coherence and alignment with the images. Extensive user testing demonstrates that our algorithm significantly outperforms existing methods in terms of layout richness and quality. The code and dataset will be available at https://github.com/HiHiAllen/Imaginarium.", "date": "2025-10-17", "categories": []} {"id": "2501.15008", "title": "HuGDiffusion: Generalizable Single-Image Human Rendering via 3D Gaussian Diffusion", "abstract": "We present HuGDiffusion, a generalizable 3D Gaussian splatting (3DGS) learning pipeline to achieve novel view synthesis (NVS) of human characters from single-view input images. Existing approaches typically require monocular videos or calibrated multi-view images as inputs, whose applicability could be weakened in real-world scenarios with arbitrary and/or unknown camera poses. In this paper, we aim to generate the set of 3DGS attributes via a diffusion-based framework conditioned on human priors extracted from a single image. Specifically, we begin with carefully integrated human-centric feature extraction procedures to deduce informative conditioning signals. Based on our empirical observations that jointly learning the whole 3DGS attributes is challenging to optimize, we design a multi-stage generation strategy to obtain different types of 3DGS attributes. To facilitate the training process, we investigate constructing proxy ground-truth 3D Gaussian attributes as high-quality attribute-level supervision signals. Through extensive experiments, our HuGDiffusion shows significant performance improvements over the state-of-the-art methods. Our code will be made publicly available.", "date": "2025-01-25", "categories": []} {"id": "2509.23402", "title": "WorldSplat: Gaussian-Centric Feed-Forward 4D Scene Generation for Autonomous Driving", "abstract": "Recent advances in driving-scene generation and reconstruction have demonstrated significant potential for enhancing autonomous driving systems by producing scalable and controllable training data. Existing generation methods primarily focus on synthesizing diverse and high-fidelity driving videos; however, due to limited 3D consistency and sparse viewpoint coverage, they struggle to support convenient and high-quality novel-view synthesis (NVS). Conversely, recent 3D/4D reconstruction approaches have significantly improved NVS for real-world driving scenes, yet inherently lack generative capabilities. To overcome this dilemma between scene generation and reconstruction, we propose WorldSplat, a novel feed-forward framework for 4D driving-scene generation. Our approach effectively generates consistent multi-track videos through two key steps: (i) We introduce a 4D-aware latent diffusion model integrating multi-modal information to produce pixel-aligned 4D Gaussians in a feed-forward manner. (ii) Subsequently, we refine the novel view videos rendered from these Gaussians using a enhanced video diffusion model. Extensive experiments conducted on benchmark datasets demonstrate that WorldSplat effectively generates high-fidelity, temporally and spatially consistent multi-track novel view driving videos. Project: https://wm-research.github.io/worldsplat/", "date": "2025-09-27", "categories": []} {"id": "2510.14977", "title": "Terra: Explorable Native 3D World Model with Point Latents", "abstract": "World models have garnered increasing attention for comprehensive modeling of the real world. However, most existing methods still rely on pixel-aligned representations as the basis for world evolution, neglecting the inherent 3D nature of the physical world. This could undermine the 3D consistency and diminish the modeling efficiency of world models. In this paper, we present Terra, a native 3D world model that represents and generates explorable environments in an intrinsic 3D latent space. Specifically, we propose a novel point-to-Gaussian variational autoencoder (P2G-VAE) that encodes 3D inputs into a latent point representation, which is subsequently decoded as 3D Gaussian primitives to jointly model geometry and appearance. We then introduce a sparse point flow matching network (SPFlow) for generating the latent point representation, which simultaneously denoises the positions and features of the point latents. Our Terra enables exact multi-view consistency with native 3D representation and architecture, and supports flexible rendering from any viewpoint with only a single generation process. Furthermore, Terra achieves explorable world modeling through progressive generation in the point latent space. We conduct extensive experiments on the challenging indoor scenes from ScanNet v2. Terra achieves state-of-the-art performance in both reconstruction and generation with high 3D consistency.", "date": "2025-10-16", "categories": []} {"id": "2510.13652", "title": "EditCast3D: Single-Frame-Guided 3D Editing with Video Propagation and View Selection", "abstract": "Recent advances in foundation models have driven remarkable progress in image editing, yet their extension to 3D editing remains underexplored. A natural approach is to replace the image editing modules in existing workflows with foundation models. However, their heavy computational demands and the restrictions and costs of closed-source APIs make plugging these models into existing iterative editing strategies impractical. To address this limitation, we propose EditCast3D, a pipeline that employs video generation foundation models to propagate edits from a single first frame across the entire dataset prior to reconstruction. While editing propagation enables dataset-level editing via video models, its consistency remains suboptimal for 3D reconstruction, where multi-view alignment is essential. To overcome this, EditCast3D introduces a view selection strategy that explicitly identifies consistent and reconstruction-friendly views and adopts feedforward reconstruction without requiring costly refinement. In combination, the pipeline both minimizes reliance on expensive image editing and mitigates prompt ambiguities that arise when applying foundation models independently across images. We evaluate EditCast3D on commonly used 3D editing datasets and compare it against state-of-the-art 3D editing baselines, demonstrating superior editing quality and high efficiency. These results establish EditCast3D as a scalable and general paradigm for integrating foundation models into 3D editing pipelines. The code is available at https://github.com/UNITES-Lab/EditCast3D", "date": "2025-10-11", "categories": []} {"id": "2510.13678", "title": "FlashWorld: High-quality 3D Scene Generation within Seconds", "abstract": "We propose FlashWorld, a generative model that produces 3D scenes from a single image or text prompt in seconds, 10~100$\\times$ faster than previous works while possessing superior rendering quality. Our approach shifts from the conventional multi-view-oriented (MV-oriented) paradigm, which generates multi-view images for subsequent 3D reconstruction, to a 3D-oriented approach where the model directly produces 3D Gaussian representations during multi-view generation. While ensuring 3D consistency, 3D-oriented method typically suffers poor visual quality. FlashWorld includes a dual-mode pre-training phase followed by a cross-mode post-training phase, effectively integrating the strengths of both paradigms. Specifically, leveraging the prior from a video diffusion model, we first pre-train a dual-mode multi-view diffusion model, which jointly supports MV-oriented and 3D-oriented generation modes. To bridge the quality gap in 3D-oriented generation, we further propose a cross-mode post-training distillation by matching distribution from consistent 3D-oriented mode to high-quality MV-oriented mode. This not only enhances visual quality while maintaining 3D consistency, but also reduces the required denoising steps for inference. Also, we propose a strategy to leverage massive single-view images and text prompts during this process to enhance the model's generalization to out-of-distribution inputs. Extensive experiments demonstrate the superiority and efficiency of our method.", "date": "2025-10-15", "categories": []} {"id": "2504.01483", "title": "GarmageNet: A Multimodal Generative Framework for Sewing Pattern Design and Generic Garment Modeling", "abstract": "Realistic digital garment modeling remains a labor-intensive task due to the intricate process of translating 2D sewing patterns into high-fidelity, simulation-ready 3D garments. We introduce GarmageNet, a unified generative framework that automates the creation of 2D sewing patterns, the construction of sewing relationships, and the synthesis of 3D garment initializations compatible with physics-based simulation. Central to our approach is Garmage, a novel garment representation that encodes each panel as a structured geometry image, effectively bridging the semantic and geometric gap between 2D structural patterns and 3D garment geometries. Followed by GarmageNet, a latent diffusion transformer to synthesize panel-wise geometry images and GarmageJigsaw, a neural module for predicting point-to-point sewing connections along panel contours. To support training and evaluation, we build GarmageSet, a large-scale dataset comprising 14,801 professionally designed garments with detailed structural and style annotations. Our method demonstrates versatility and efficacy across multiple application scenarios, including scalable garment generation from multi-modal design concepts (text prompts, sketches, photographs), automatic modeling from raw flat sewing patterns, pattern recovery from unstructured point clouds, and progressive garment editing using conventional instructions, laying the foundation for fully automated, production-ready pipelines in digital fashion. Project page: https://style3d.github.io/garmagenet/.", "date": "2025-04-02", "categories": []} {"id": "2510.12785", "title": "MVP4D: Multi-View Portrait Video Diffusion for Animatable 4D Avatars", "abstract": "Digital human avatars aim to simulate the dynamic appearance of humans in virtual environments, enabling immersive experiences across gaming, film, virtual reality, and more. However, the conventional process for creating and animating photorealistic human avatars is expensive and time-consuming, requiring large camera capture rigs and significant manual effort from professional 3D artists. With the advent of capable image and video generation models, recent methods enable automatic rendering of realistic animated avatars from a single casually captured reference image of a target subject. While these techniques significantly lower barriers to avatar creation and offer compelling realism, they lack constraints provided by multi-view information or an explicit 3D representation. So, image quality and realism degrade when rendered from viewpoints that deviate strongly from the reference image. Here, we build a video model that generates animatable multi-view videos of digital humans based on a single reference image and target expressions. Our model, MVP4D, is based on a state-of-the-art pre-trained video diffusion model and generates hundreds of frames simultaneously from viewpoints varying by up to 360 degrees around a target subject. We show how to distill the outputs of this model into a 4D avatar that can be rendered in real-time. Our approach significantly improves the realism, temporal consistency, and 3D consistency of generated avatars compared to previous methods.", "date": "2025-10-14", "categories": []} {"id": "2411.14384", "title": "Baking Gaussian Splatting into Diffusion Denoiser for Fast and Scalable Single-stage Image-to-3D Generation and Reconstruction", "abstract": "Existing feedforward image-to-3D methods mainly rely on 2D multi-view diffusion models that cannot guarantee 3D consistency. These methods easily collapse when changing the prompt view direction and mainly handle object-centric cases. In this paper, we propose a novel single-stage 3D diffusion model, DiffusionGS, for object generation and scene reconstruction from a single view. DiffusionGS directly outputs 3D Gaussian point clouds at each timestep to enforce view consistency and allow the model to generate robustly given prompt views of any directions, beyond object-centric inputs. Plus, to improve the capability and generality of DiffusionGS, we scale up 3D training data by developing a scene-object mixed training strategy. Experiments show that DiffusionGS yields improvements of 2.20 dB/23.25 and 1.34 dB/19.16 in PSNR/FID for objects and scenes than the state-of-the-art methods, without depth estimator. Plus, our method enjoys over 5$\\times$ faster speed ($\\sim$6s on an A100 GPU). Our Project page at https://caiyuanhao1998.github.io/project/DiffusionGS/ shows the video and interactive results. The code and models are publicly available at https://github.com/caiyuanhao1998/Open-DiffusionGS", "date": "2024-11-21", "categories": []} {"id": "2510.10292", "title": "From Programs to Poses: Factored Real-World Scene Generation via Learned Program Libraries", "abstract": "Real-world scenes, such as those in ScanNet, are difficult to capture, with highly limited data available. Generating realistic scenes with varied object poses remains an open and challenging task. In this work, we propose FactoredScenes, a framework that synthesizes realistic 3D scenes by leveraging the underlying structure of rooms while learning the variation of object poses from lived-in scenes. We introduce a factored representation that decomposes scenes into hierarchically organized concepts of room programs and object poses. To encode structure, FactoredScenes learns a library of functions capturing reusable layout patterns from which scenes are drawn, then uses large language models to generate high-level programs, regularized by the learned library. To represent scene variations, FactoredScenes learns a program-conditioned model to hierarchically predict object poses, and retrieves and places 3D objects in a scene. We show that FactoredScenes generates realistic, real-world rooms that are difficult to distinguish from real ScanNet scenes.", "date": "2025-10-11", "categories": []} {"id": "2510.11631", "title": "EvoCAD: Evolutionary CAD Code Generation with Vision Language Models", "abstract": "Combining large language models with evolutionary computation algorithms represents a promising research direction leveraging the remarkable generative and in-context learning capabilities of LLMs with the strengths of evolutionary algorithms. In this work, we present EvoCAD, a method for generating computer-aided design (CAD) objects through their symbolic representations using vision language models and evolutionary optimization. Our method samples multiple CAD objects, which are then optimized using an evolutionary approach with vision language and reasoning language models. We assess our method using GPT-4V and GPT-4o, evaluating it on the CADPrompt benchmark dataset and comparing it to prior methods. Additionally, we introduce two new metrics based on topological properties defined by the Euler characteristic, which capture a form of semantic similarity between 3D objects. Our results demonstrate that EvoCAD outperforms previous approaches on multiple metrics, particularly in generating topologically correct objects, which can be efficiently evaluated using our two novel metrics that complement existing spatial metrics.", "date": "2025-10-13", "categories": []} {"id": "2510.11650", "title": "InfiniHuman: Infinite 3D Human Creation with Precise Control", "abstract": "Generating realistic and controllable 3D human avatars is a long-standing challenge, particularly when covering broad attribute ranges such as ethnicity, age, clothing styles, and detailed body shapes. Capturing and annotating large-scale human datasets for training generative models is prohibitively expensive and limited in scale and diversity. The central question we address in this paper is: Can existing foundation models be distilled to generate theoretically unbounded, richly annotated 3D human data? We introduce InfiniHuman, a framework that synergistically distills these models to produce richly annotated human data at minimal cost and with theoretically unlimited scalability. We propose InfiniHumanData, a fully automatic pipeline that leverages vision-language and image generation models to create a large-scale multi-modal dataset. User study shows our automatically generated identities are undistinguishable from scan renderings. InfiniHumanData contains 111K identities spanning unprecedented diversity. Each identity is annotated with multi-granularity text descriptions, multi-view RGB images, detailed clothing images, and SMPL body-shape parameters. Building on this dataset, we propose InfiniHumanGen, a diffusion-based generative pipeline conditioned on text, body shape, and clothing assets. InfiniHumanGen enables fast, realistic, and precisely controllable avatar generation. Extensive experiments demonstrate significant improvements over state-of-the-art methods in visual quality, generation speed, and controllability. Our approach enables high-quality avatar generation with fine-grained control at effectively unbounded scale through a practical and affordable solution. We will publicly release the automatic data generation pipeline, the comprehensive InfiniHumanData dataset, and the InfiniHumanGen models at https://yuxuan-xue.com/infini-human.", "date": "2025-10-13", "categories": []} {"id": "2409.02851", "title": "Human-VDM: Learning Single-Image 3D Human Gaussian Splatting from Video Diffusion Models", "abstract": "Generating lifelike 3D humans from a single RGB image remains a challenging task in computer vision, as it requires accurate modeling of geometry, high-quality texture, and plausible unseen parts. Existing methods typically use multi-view diffusion models for 3D generation, but they often face inconsistent view issues, which hinder high-quality 3D human generation. To address this, we propose Human-VDM, a novel method for generating 3D human from a single RGB image using Video Diffusion Models. Human-VDM provides temporally consistent views for 3D human generation using Gaussian Splatting. It consists of three modules: a view-consistent human video diffusion module, a video augmentation module, and a Gaussian Splatting module. First, a single image is fed into a human video diffusion module to generate a coherent human video. Next, the video augmentation module applies super-resolution and video interpolation to enhance the textures and geometric smoothness of the generated video. Finally, the 3D Human Gaussian Splatting module learns lifelike humans under the guidance of these high-resolution and view-consistent images. Experiments demonstrate that Human-VDM achieves high-quality 3D human from a single image, outperforming state-of-the-art methods in both generation quality and quantity. Project page: https://human-vdm.github.io/Human-VDM/", "date": "2024-09-04", "categories": []} {"id": "2510.08656", "title": "A 3D Generation Framework from Cross Modality to Parameterized Primitive", "abstract": "Recent advancements in AI-driven 3D model generation have leveraged cross modality, yet generating models with smooth surfaces and minimizing storage overhead remain challenges. This paper introduces a novel multi-stage framework for generating 3D models composed of parameterized primitives, guided by textual and image inputs. In the framework, A model generation algorithm based on parameterized primitives, is proposed, which can identifies the shape features of the model constituent elements, and replace the elements with parameterized primitives with high quality surface. In addition, a corresponding model storage method is proposed, it can ensure the original surface quality of the model, while retaining only the parameters of parameterized primitives. Experiments on virtual scene dataset and real scene dataset demonstrate the effectiveness of our method, achieving a Chamfer Distance of 0.003092, a VIoU of 0.545, a F1-Score of 0.9139 and a NC of 0.8369, with primitive parameter files approximately 6KB in size. Our approach is particularly suitable for rapid prototyping of simple models.", "date": "2025-10-09", "categories": []} {"id": "2503.16278", "title": "Unified Cross-Scale 3D Generation and Understanding via Autoregressive Modeling", "abstract": "3D structure modeling is essential across scales, enabling applications from fluid simulation and 3D reconstruction to protein folding and molecular docking. Yet, despite shared 3D spatial patterns, current approaches remain fragmented, with models narrowly specialized for specific domains and unable to generalize across tasks or scales. We propose Uni-3DAR, a unified autoregressive framework for cross-scale 3D generation and understanding. At its core is a coarse-to-fine tokenizer based on octree data structures, which compresses diverse 3D structures into compact 1D token sequences. We further propose a two-level subtree compression strategy, which reduces the octree token sequence by up to 8x. To address the challenge of dynamically varying token positions introduced by compression, we introduce a masked next-token prediction strategy that ensures accurate positional modeling, significantly boosting model performance. Extensive experiments across multiple 3D generation and understanding tasks, including small molecules, proteins, polymers, crystals, and macroscopic 3D objects, validate its effectiveness and versatility. Notably, Uni-3DAR surpasses previous state-of-the-art diffusion models by a substantial margin, achieving up to 256\\% relative improvement while delivering inference speeds up to 21.8x faster.", "date": "2025-03-20", "categories": []} {"id": "2510.06967", "title": "Generating Surface for Text-to-3D using 2D Gaussian Splatting", "abstract": "Recent advancements in Text-to-3D modeling have shown significant potential for the creation of 3D content. However, due to the complex geometric shapes of objects in the natural world, generating 3D content remains a challenging task. Current methods either leverage 2D diffusion priors to recover 3D geometry, or train the model directly based on specific 3D representations. In this paper, we propose a novel method named DirectGaussian, which focuses on generating the surfaces of 3D objects represented by surfels. In DirectGaussian, we utilize conditional text generation models and the surface of a 3D object is rendered by 2D Gaussian splatting with multi-view normal and texture priors. For multi-view geometric consistency problems, DirectGaussian incorporates curvature constraints on the generated surface during optimization process. Through extensive experiments, we demonstrate that our framework is capable of achieving diverse and high-fidelity 3D content creation.", "date": "2025-10-08", "categories": []} {"id": "2510.06208", "title": "ShapeGen4D: Towards High Quality 4D Shape Generation from Videos", "abstract": "Video-conditioned 4D shape generation aims to recover time-varying 3D geometry and view-consistent appearance directly from an input video. In this work, we introduce a native video-to-4D shape generation framework that synthesizes a single dynamic 3D representation end-to-end from the video. Our framework introduces three key components based on large-scale pre-trained 3D models: (i) a temporal attention that conditions generation on all frames while producing a time-indexed dynamic representation; (ii) a time-aware point sampling and 4D latent anchoring that promote temporally consistent geometry and texture; and (iii) noise sharing across frames to enhance temporal stability. Our method accurately captures non-rigid motion, volume changes, and even topological transitions without per-frame optimization. Across diverse in-the-wild videos, our method improves robustness and perceptual fidelity and reduces failure modes compared with the baselines.", "date": "2025-10-07", "categories": []} {"id": "2503.11981", "title": "DecompDreamer: A Composition-Aware Curriculum for Structured 3D Asset Generation", "abstract": "Current text-to-3D methods excel at generating single objects but falter on compositional prompts. We argue this failure is fundamental to their optimization schedules, as simultaneous or iterative heuristics predictably collapse under a combinatorial explosion of conflicting gradients, leading to entangled geometry or catastrophic divergence. In this paper, we reframe the core challenge of compositional generation as one of optimization scheduling. We introduce DecompDreamer, a framework built on a novel staged optimization strategy that functions as an implicit curriculum. Our method first establishes a coherent structural scaffold by prioritizing inter-object relationships before shifting to the high-fidelity refinement of individual components. This temporal decoupling of competing objectives provides a robust solution to gradient conflict. Qualitative and quantitative evaluations on diverse compositional prompts demonstrate that DecompDreamer outperforms state-of-the-art methods in fidelity, disentanglement, and spatial coherence.", "date": "2025-03-15", "categories": []} {"id": "2505.15765", "title": "Constructing a 3D Scene from a Single Image", "abstract": "Acquiring detailed 3D scenes typically demands costly equipment, multi-view data, or labor-intensive modeling. Therefore, a lightweight alternative, generating complex 3D scenes from a single top-down image, plays an essential role in real-world applications. While recent 3D generative models have achieved remarkable results at the object level, their extension to full-scene generation often leads to inconsistent geometry, layout hallucinations, and low-quality meshes. In this work, we introduce SceneFuse-3D, a training-free framework designed to synthesize coherent 3D scenes from a single top-down view. Our method is grounded in two principles: region-based generation to improve image-to-3D alignment and resolution, and spatial-aware 3D inpainting to ensure global scene coherence and high-quality geometry generation. Specifically, we decompose the input image into overlapping regions and generate each using a pretrained 3D object generator, followed by a masked rectified flow inpainting process that fills in missing geometry while maintaining structural continuity. This modular design allows us to overcome resolution bottlenecks and preserve spatial structure without requiring 3D supervision or fine-tuning. Extensive experiments across diverse scenes show that SceneFuse-3D outperforms state-of-the-art baselines, including Trellis, Hunyuan3D-2, TripoSG, and LGM, in terms of geometry quality, spatial coherence, and texture fidelity. Our results demonstrate that high-quality coherent 3D scene-level asset generation is achievable from a single top-down image using a principled, training-free pipeline.", "date": "2025-05-21", "categories": []} {"id": "2510.02994", "title": "Towards Scalable and Consistent 3D Editing", "abstract": "3D editing - the task of locally modifying the geometry or appearance of a 3D asset - has wide applications in immersive content creation, digital entertainment, and AR/VR. However, unlike 2D editing, it remains challenging due to the need for cross-view consistency, structural fidelity, and fine-grained controllability. Existing approaches are often slow, prone to geometric distortions, or dependent on manual and accurate 3D masks that are error-prone and impractical. To address these challenges, we advance both the data and model fronts. On the data side, we introduce 3DEditVerse, the largest paired 3D editing benchmark to date, comprising 116,309 high-quality training pairs and 1,500 curated test pairs. Built through complementary pipelines of pose-driven geometric edits and foundation model-guided appearance edits, 3DEditVerse ensures edit locality, multi-view consistency, and semantic alignment. On the model side, we propose 3DEditFormer, a 3D-structure-preserving conditional transformer. By enhancing image-to-3D generation with dual-guidance attention and time-adaptive gating, 3DEditFormer disentangles editable regions from preserved structure, enabling precise and consistent edits without requiring auxiliary 3D masks. Extensive experiments demonstrate that our framework outperforms state-of-the-art baselines both quantitatively and qualitatively, establishing a new standard for practical and scalable 3D editing. Dataset and code will be released. Project: https://www.lv-lab.org/3DEditFormer/", "date": "2025-10-03", "categories": []} {"id": "2510.01183", "title": "EvoWorld: Evolving Panoramic World Generation with Explicit 3D Memory", "abstract": "Humans possess a remarkable ability to mentally explore and replay 3D environments they have previously experienced. Inspired by this mental process, we present EvoWorld: a world model that bridges panoramic video generation with evolving 3D memory to enable spatially consistent long-horizon exploration. Given a single panoramic image as input, EvoWorld first generates future video frames by leveraging a video generator with fine-grained view control, then evolves the scene's 3D reconstruction using a feedforward plug-and-play transformer, and finally synthesizes futures by conditioning on geometric reprojections from this evolving explicit 3D memory. Unlike prior state-of-the-arts that synthesize videos only, our key insight lies in exploiting this evolving 3D reconstruction as explicit spatial guidance for the video generation process, projecting the reconstructed geometry onto target viewpoints to provide rich spatial cues that significantly enhance both visual realism and geometric consistency. To evaluate long-range exploration capabilities, we introduce the first comprehensive benchmark spanning synthetic outdoor environments, Habitat indoor scenes, and challenging real-world scenarios, with particular emphasis on loop-closure detection and spatial coherence over extended trajectories. Extensive experiments demonstrate that our evolving 3D memory substantially improves visual fidelity and maintains spatial scene coherence compared to existing approaches, representing a significant advance toward long-horizon spatially consistent world modeling.", "date": "2025-10-01", "categories": []} {"id": "2412.21206", "title": "PERSE: Personalized 3D Generative Avatars from A Single Portrait", "abstract": "We present PERSE, a method for building a personalized 3D generative avatar from a reference portrait. Our avatar enables facial attribute editing in a continuous and disentangled latent space to control each facial attribute, while preserving the individual's identity. To achieve this, our method begins by synthesizing large-scale synthetic 2D video datasets, where each video contains consistent changes in facial expression and viewpoint, along with variations in a specific facial attribute from the original input. We propose a novel pipeline to produce high-quality, photorealistic 2D videos with facial attribute editing. Leveraging this synthetic attribute dataset, we present a personalized avatar creation method based on 3D Gaussian Splatting, learning a continuous and disentangled latent space for intuitive facial attribute manipulation. To enforce smooth transitions in this latent space, we introduce a latent space regularization technique by using interpolated 2D faces as supervision. Compared to previous approaches, we demonstrate that PERSE generates high-quality avatars with interpolated attributes while preserving the identity of the reference individual.", "date": "2024-12-30", "categories": []} {"id": "2504.08366", "title": "In-2-4D: Inbetweening from Two Single-View Images to 4D Generation", "abstract": "We pose a new problem, In-2-4D, for generative 4D (i.e., 3D + motion) inbetweening to interpolate two single-view images. In contrast to video/4D generation from only text or a single image, our interpolative task can leverage more precise motion control to better constrain the generation. Given two monocular RGB images representing the start and end states of an object in motion, our goal is to generate and reconstruct the motion in 4D, without making assumptions on the object category, motion type, length, or complexity. To handle such arbitrary and diverse motions, we utilize a foundational video interpolation model for motion prediction. However, large frame-to-frame motion gaps can lead to ambiguous interpretations. To this end, we employ a hierarchical approach to identify keyframes that are visually close to the input states while exhibiting significant motions, then generate smooth fragments between them. For each fragment, we construct a 3D representation of the keyframe using Gaussian Splatting (3DGS). The temporal frames within the fragment guide the motion, enabling their transformation into dynamic 3DGS through a deformation field. To improve temporal consistency and refine the 3D motion, we expand the self-attention of multi-view diffusion across timesteps and apply rigid transformation regularization. Finally, we merge the independently generated 3D motion segments by interpolating boundary deformation fields and optimizing them to align with the guiding video, ensuring smooth and flicker-free transitions. Through extensive qualitative and quantitive experiments as well as a user study, we demonstrate the effectiveness of our method and design choices.", "date": "2025-04-11", "categories": []} {"id": "2505.20431", "title": "ART-DECO: Arbitrary Text Guidance for 3D Detailizer Construction", "abstract": "We introduce a 3D detailizer, a neural model which can instantaneously (in <1s) transform a coarse 3D shape proxy into a high-quality asset with detailed geometry and texture as guided by an input text prompt. Our model is trained using the text prompt, which defines the shape class and characterizes the appearance and fine-grained style of the generated details. The coarse 3D proxy, which can be easily varied and adjusted (e.g., via user editing), provides structure control over the final shape. Importantly, our detailizer is not optimized for a single shape; it is the result of distilling a generative model, so that it can be reused, without retraining, to generate any number of shapes, with varied structures, whose local details all share a consistent style and appearance. Our detailizer training utilizes a pretrained multi-view image diffusion model, with text conditioning, to distill the foundational knowledge therein into our detailizer via Score Distillation Sampling (SDS). To improve SDS and enable our detailizer architecture to learn generalizable features over complex structures, we train our model in two training stages to generate shapes with increasing structural complexity. Through extensive experiments, we show that our method generates shapes of superior quality and details compared to existing text-to-3D models under varied structure control. Our detailizer can refine a coarse shape in less than a second, making it possible to interactively author and adjust 3D shapes. Furthermore, the user-imposed structure control can lead to creative, and hence out-of-distribution, 3D asset generations that are beyond the current capabilities of leading text-to-3D generative models. We demonstrate an interactive 3D modeling workflow our method enables, and its strong generalizability over styles, structures, and object categories.", "date": "2025-05-26", "categories": []} {"id": "2509.23646", "title": "Sparse-Up: Learnable Sparse Upsampling for 3D Generation with High-Fidelity Textures", "abstract": "The creation of high-fidelity 3D assets is often hindered by a 'pixel-level pain point': the loss of high-frequency details. Existing methods often trade off one aspect for another: either sacrificing cross-view consistency, resulting in torn or drifting textures, or remaining trapped by the resolution ceiling of explicit voxels, forfeiting fine texture detail. In this work, we propose Sparse-Up, a memory-efficient, high-fidelity texture modeling framework that effectively preserves high-frequency details. We use sparse voxels to guide texture reconstruction and ensure multi-view consistency, while leveraging surface anchoring and view-domain partitioning to break through resolution constraints. Surface anchoring employs a learnable upsampling strategy to constrain voxels to the mesh surface, eliminating over 70% of redundant voxels present in traditional voxel upsampling. View-domain partitioning introduces an image patch-guided voxel partitioning scheme, supervising and back-propagating gradients only on visible local patches. Through these two strategies, we can significantly reduce memory consumption during high-resolution voxel training without sacrificing geometric consistency, while preserving high-frequency details in textures.", "date": "2025-09-28", "categories": []} {"id": "2509.23709", "title": "StrucADT: Generating Structure-controlled 3D Point Clouds with Adjacency Diffusion Transformer", "abstract": "In the field of 3D point cloud generation, numerous 3D generative models have demonstrated the ability to generate diverse and realistic 3D shapes. However, the majority of these approaches struggle to generate controllable 3D point cloud shapes that meet user-specific requirements, hindering the large-scale application of 3D point cloud generation. To address the challenge of lacking control in 3D point cloud generation, we are the first to propose controlling the generation of point clouds by shape structures that comprise part existences and part adjacency relationships. We manually annotate the adjacency relationships between the segmented parts of point cloud shapes, thereby constructing a StructureGraph representation. Based on this StructureGraph representation, we introduce StrucADT, a novel structure-controllable point cloud generation model, which consists of StructureGraphNet module to extract structure-aware latent features, cCNF Prior module to learn the distribution of the latent features controlled by the part adjacency, and Diffusion Transformer module conditioned on the latent features and part adjacency to generate structure-consistent point cloud shapes. Experimental results demonstrate that our structure-controllable 3D point cloud generation method produces high-quality and diverse point cloud shapes, enabling the generation of controllable point clouds based on user-specified shape structures and achieving state-of-the-art performance in controllable point cloud generation on the ShapeNet dataset.", "date": "2025-09-28", "categories": []} {"id": "2509.23804", "title": "Controllable Generation of Large-Scale 3D Urban Layouts with Semantic and Structural Guidance", "abstract": "Urban modeling is essential for city planning, scene synthesis, and gaming. Existing image-based methods generate diverse layouts but often lack geometric continuity and scalability, while graph-based methods capture structural relations yet overlook parcel semantics. We present a controllable framework for large-scale 3D vector urban layout generation, conditioned on both geometry and semantics. By fusing geometric and semantic attributes, introducing edge weights, and embedding building height in the graph, our method extends 2D layouts to realistic 3D structures. It also enables users to directly control the output by modifying semantic attributes. Experiments show that it produces valid, large-scale urban models, offering an effective tool for data-driven planning and design.", "date": "2025-09-28", "categories": []} {"id": "2509.24410", "title": "RapidMV: Leveraging Spatio-Angular Representations for Efficient and Consistent Text-to-Multi-View Synthesis", "abstract": "Generating synthetic multi-view images from a text prompt is an essential bridge to generating synthetic 3D assets. In this work, we introduce RapidMV, a novel text-to-multi-view generative model that can produce 32 multi-view synthetic images in just around 5 seconds. In essence, we propose a novel spatio-angular latent space, encoding both the spatial appearance and angular viewpoint deviations into a single latent for improved efficiency and multi-view consistency. We achieve effective training of RapidMV by strategically decomposing our training process into multiple steps. We demonstrate that RapidMV outperforms existing methods in terms of consistency and latency, with competitive quality and text-image alignment.", "date": "2025-09-29", "categories": []} {"id": "2509.24441", "title": "NeoWorld: Neural Simulation of Explorable Virtual Worlds via Progressive 3D Unfolding", "abstract": "We introduce NeoWorld, a deep learning framework for generating interactive 3D virtual worlds from a single input image. Inspired by the on-demand worldbuilding concept in the science fiction novel Simulacron-3 (1964), our system constructs expansive environments where only the regions actively explored by the user are rendered with high visual realism through object-centric 3D representations. Unlike previous approaches that rely on global world generation or 2D hallucination, NeoWorld models key foreground objects in full 3D, while synthesizing backgrounds and non-interacted regions in 2D to ensure efficiency. This hybrid scene structure, implemented with cutting-edge representation learning and object-to-3D techniques, enables flexible viewpoint manipulation and physically plausible scene animation, allowing users to control object appearance and dynamics using natural language commands. As users interact with the environment, the virtual world progressively unfolds with increasing 3D detail, delivering a dynamic, immersive, and visually coherent exploration experience. NeoWorld significantly outperforms existing 2D and depth-layered 2.5D methods on the WorldScore benchmark.", "date": "2025-09-29", "categories": []} {"id": "2509.25079", "title": "UniLat3D: Geometry-Appearance Unified Latents for Single-Stage 3D Generation", "abstract": "High-fidelity 3D asset generation is crucial for various industries. While recent 3D pretrained models show strong capability in producing realistic content, most are built upon diffusion models and follow a two-stage pipeline that first generates geometry and then synthesizes appearance. Such a decoupled design tends to produce geometry-texture misalignment and non-negligible cost. In this paper, we propose UniLat3D, a unified framework that encodes geometry and appearance in a single latent space, enabling direct single-stage generation. Our key contribution is a geometry-appearance Unified VAE, which compresses high-resolution sparse features into a compact latent representation -- UniLat. UniLat integrates structural and visual information into a dense low-resolution latent, which can be efficiently decoded into diverse 3D formats, e.g., 3D Gaussians and meshes. Based on this unified representation, we train a single flow-matching model to map Gaussian noise directly into UniLat, eliminating redundant stages. Trained solely on public datasets, UniLat3D produces high-quality 3D assets in seconds from a single image, achieving superior appearance fidelity and geometric quality. More demos \\& code are available at https://unilat3d.github.io/", "date": "2025-09-29", "categories": []} {"id": "2412.07385", "title": "LOGen: Toward Lidar Object Generation by Point Diffusion", "abstract": "The generation of LiDAR scans is a growing topic with diverse applications to autonomous driving. However, scan generation remains challenging, especially when compared to the rapid advancement of image and 3D object generation. We consider the task of LiDAR object generation, requiring models to produce 3D objects as viewed by a LiDAR scan. It focuses LiDAR scan generation on a key aspect of scenes, the objects, while also benefiting from advancements in 3D object generative methods. We introduce a novel diffusion-based model to produce LiDAR point clouds of dataset objects, including intensity, and with an extensive control of the generation via conditioning information. Our experiments on nuScenes and KITTI-360 show the quality of our generations measured with new 3D metrics developed to suit LiDAR objects. The code is available at https://github.com/valeoai/LOGen.", "date": "2024-12-10", "categories": []} {"id": "2509.21888", "title": "Drag4D: Align Your Motion with Text-Driven 3D Scene Generation", "abstract": "We introduce Drag4D, an interactive framework that integrates object motion control within text-driven 3D scene generation. This framework enables users to define 3D trajectories for the 3D objects generated from a single image, seamlessly integrating them into a high-quality 3D background. Our Drag4D pipeline consists of three stages. First, we enhance text-to-3D background generation by applying 2D Gaussian Splatting with panoramic images and inpainted novel views, resulting in dense and visually complete 3D reconstructions. In the second stage, given a reference image of the target object, we introduce a 3D copy-and-paste approach: the target instance is extracted in a full 3D mesh using an off-the-shelf image-to-3D model and seamlessly composited into the generated 3D scene. The object mesh is then positioned within the 3D scene via our physics-aware object position learning, ensuring precise spatial alignment. Lastly, the spatially aligned object is temporally animated along a user-defined 3D trajectory. To mitigate motion hallucination and ensure view-consistent temporal alignment, we develop a part-augmented, motion-conditioned video diffusion model that processes multiview image pairs together with their projected 2D trajectories. We demonstrate the effectiveness of our unified architecture through evaluations at each stage and in the final results, showcasing the harmonized alignment of user-controlled object motion within a high-quality 3D background.", "date": "2025-09-26", "categories": []} {"id": "2509.22112", "title": "Large Material Gaussian Model for Relightable 3D Generation", "abstract": "The increasing demand for 3D assets across various industries necessitates efficient and automated methods for 3D content creation. Leveraging 3D Gaussian Splatting, recent large reconstruction models (LRMs) have demonstrated the ability to efficiently achieve high-quality 3D rendering by integrating multiview diffusion for generation and scalable transformers for reconstruction. However, existing models fail to produce the material properties of assets, which is crucial for realistic rendering in diverse lighting environments. In this paper, we introduce the Large Material Gaussian Model (MGM), a novel framework designed to generate high-quality 3D content with Physically Based Rendering (PBR) materials, ie, albedo, roughness, and metallic properties, rather than merely producing RGB textures with uncontrolled light baking. Specifically, we first fine-tune a new multiview material diffusion model conditioned on input depth and normal maps. Utilizing the generated multiview PBR images, we explore a Gaussian material representation that not only aligns with 2D Gaussian Splatting but also models each channel of the PBR materials. The reconstructed point clouds can then be rendered to acquire PBR attributes, enabling dynamic relighting by applying various ambient light maps. Extensive experiments demonstrate that the materials produced by our method not only exhibit greater visual appeal compared to baseline methods but also enhance material modeling, thereby enabling practical downstream rendering applications.", "date": "2025-09-26", "categories": []} {"id": "2504.02045", "title": "Generating 360{\\deg} Video is What You Need For a 3D Scene", "abstract": "Generating 3D scenes is still a challenging task due to the lack of readily available scene data. Most existing methods only produce partial scenes and provide limited navigational freedom. We introduce a practical and scalable solution that uses 360{\\deg} video as an intermediate scene representation, capturing the full-scene context and ensuring consistent visual content throughout the generation. We propose WorldPrompter, a generative pipeline that synthesizes traversable 3D scenes from text prompts. WorldPrompter incorporates a conditional 360{\\deg} panoramic video generator, capable of producing a 128-frame video that simulates a person walking through and capturing a virtual environment. The resulting video is then reconstructed as Gaussian splats by a fast feedforward 3D reconstructor, enabling a true walkable experience within the 3D scene. Experiments demonstrate that our panoramic video generation model, trained with a mix of image and video data, achieves convincing spatial and temporal consistency for static scenes. This is validated by an average COLMAP matching rate of 94.6\\%, allowing for high-quality panoramic Gaussian splat reconstruction and improved navigation throughout the scene. Qualitative and quantitative results also show it outperforms the state-of-the-art 360{\\deg} video generators and 3D scene generation models.", "date": "2025-04-02", "categories": []} {"id": "2509.20824", "title": "ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction", "abstract": "Directly generating 3D meshes, the default representation for 3D shapes in the graphics industry, using auto-regressive (AR) models has become popular these days, thanks to their sharpness, compactness in the generated results, and ability to represent various types of surfaces. However, AR mesh generative models typically construct meshes face by face in lexicographic order, which does not effectively capture the underlying geometry in a manner consistent with human perception. Inspired by 2D models that progressively refine images, such as the prevailing next-scale prediction AR models, we propose generating meshes auto-regressively in a progressive coarse-to-fine manner. Specifically, we view mesh simplification algorithms, which gradually merge mesh faces to build simpler meshes, as a natural fine-to-coarse process. Therefore, we generalize meshes to simplicial complexes and develop a transformer-based AR model to approximate the reverse process of simplification in the order of level of detail, constructing meshes initially from a single point and gradually adding geometric details through local remeshing, where the topology is not predefined and is alterable. Our experiments show that this novel progressive mesh generation approach not only provides intuitive control over generation quality and time consumption by early stopping the auto-regressive process but also enables applications such as mesh refinement and editing.", "date": "2025-09-25", "categories": []} {"id": "2509.21114", "title": "CHARM: Control-point-based 3D Anime Hairstyle Auto-Regressive Modeling", "abstract": "We present CHARM, a novel parametric representation and generative framework for anime hairstyle modeling. While traditional hair modeling methods focus on realistic hair using strand-based or volumetric representations, anime hairstyle exhibits highly stylized, piecewise-structured geometry that challenges existing techniques. Existing works often rely on dense mesh modeling or hand-crafted spline curves, making them inefficient for editing and unsuitable for scalable learning. CHARM introduces a compact, invertible control-point-based parameterization, where a sequence of control points represents each hair card, and each point is encoded with only five geometric parameters. This efficient and accurate representation supports both artist-friendly design and learning-based generation. Built upon this representation, CHARM introduces an autoregressive generative framework that effectively generates anime hairstyles from input images or point clouds. By interpreting anime hairstyles as a sequential \"hair language\", our autoregressive transformer captures both local geometry and global hairstyle topology, resulting in high-fidelity anime hairstyle creation. To facilitate both training and evaluation of anime hairstyle generation, we construct AnimeHair, a large-scale dataset of 37K high-quality anime hairstyles with separated hair cards and processed mesh data. Extensive experiments demonstrate state-of-the-art performance of CHARM in both reconstruction accuracy and generation quality, offering an expressive and scalable solution for anime hairstyle modeling. Project page: https://hyzcluster.github.io/charm/", "date": "2025-09-25", "categories": []} {"id": "2509.21245", "title": "Hunyuan3D-Omni: A Unified Framework for Controllable Generation of 3D Assets", "abstract": "Recent advances in 3D-native generative models have accelerated asset creation for games, film, and design. However, most methods still rely primarily on image or text conditioning and lack fine-grained, cross-modal controls, which limits controllability and practical adoption. To address this gap, we present Hunyuan3D-Omni, a unified framework for fine-grained, controllable 3D asset generation built on Hunyuan3D 2.1. In addition to images, Hunyuan3D-Omni accepts point clouds, voxels, bounding boxes, and skeletal pose priors as conditioning signals, enabling precise control over geometry, topology, and pose. Instead of separate heads for each modality, our model unifies all signals in a single cross-modal architecture. We train with a progressive, difficulty-aware sampling strategy that selects one control modality per example and biases sampling toward harder signals (e.g., skeletal pose) while downweighting easier ones (e.g., point clouds), encouraging robust multi-modal fusion and graceful handling of missing inputs. Experiments show that these additional controls improve generation accuracy, enable geometry-aware transformations, and increase robustness for production workflows.", "date": "2025-09-25", "categories": []} {"id": "2310.03602", "title": "Ctrl-Room: Controllable Text-to-3D Room Meshes Generation with Layout Constraints", "abstract": "Text-driven 3D indoor scene generation is useful for gaming, the film industry, and AR/VR applications. However, existing methods cannot faithfully capture the room layout, nor do they allow flexible editing of individual objects in the room. To address these problems, we present Ctrl-Room, which can generate convincing 3D rooms with designer-style layouts and high-fidelity textures from just a text prompt. Moreover, Ctrl-Room enables versatile interactive editing operations such as resizing or moving individual furniture items. Our key insight is to separate the modeling of layouts and appearance. Our proposed method consists of two stages: a Layout Generation Stage and an Appearance Generation Stage. The Layout Generation Stage trains a text-conditional diffusion model to learn the layout distribution with our holistic scene code parameterization. Next, the Appearance Generation Stage employs a fine-tuned ControlNet to produce a vivid panoramic image of the room guided by the 3D scene layout and text prompt. We thus achieve a high-quality 3D room generation with convincing layouts and lively textures. Benefiting from the scene code parameterization, we can easily edit the generated room model through our mask-guided editing module, without expensive edit-specific training. Extensive experiments on the Structured3D dataset demonstrate that our method outperforms existing methods in producing more reasonable, view-consistent, and editable 3D rooms from natural language prompts.", "date": "2023-10-05", "categories": []} {"id": "2509.08643", "title": "X-Part: high fidelity and structure coherent shape decomposition", "abstract": "Generating 3D shapes at part level is pivotal for downstream applications such as mesh retopology, UV mapping, and 3D printing. However, existing part-based generation methods often lack sufficient controllability and suffer from poor semantically meaningful decomposition. To this end, we introduce X-Part, a controllable generative model designed to decompose a holistic 3D object into semantically meaningful and structurally coherent parts with high geometric fidelity. X-Part exploits the bounding box as prompts for the part generation and injects point-wise semantic features for meaningful decomposition. Furthermore, we design an editable pipeline for interactive part generation. Extensive experimental results show that X-Part achieves state-of-the-art performance in part-level shape generation. This work establishes a new paradigm for creating production-ready, editable, and structurally sound 3D assets. Codes will be released for public research.", "date": "2025-09-10", "categories": []} {"id": "2509.20251", "title": "4D Driving Scene Generation With Stereo Forcing", "abstract": "Current generative models struggle to synthesize dynamic 4D driving scenes that simultaneously support temporal extrapolation and spatial novel view synthesis (NVS) without per-scene optimization. Bridging generation and novel view synthesis remains a major challenge. We present PhiGenesis, a unified framework for 4D scene generation that extends video generation techniques with geometric and temporal consistency. Given multi-view image sequences and camera parameters, PhiGenesis produces temporally continuous 4D Gaussian splatting representations along target 3D trajectories. In its first stage, PhiGenesis leverages a pre-trained video VAE with a novel range-view adapter to enable feed-forward 4D reconstruction from multi-view images. This architecture supports single-frame or video inputs and outputs complete 4D scenes including geometry, semantics, and motion. In the second stage, PhiGenesis introduces a geometric-guided video diffusion model, using rendered historical 4D scenes as priors to generate future views conditioned on trajectories. To address geometric exposure bias in novel views, we propose Stereo Forcing, a novel conditioning strategy that integrates geometric uncertainty during denoising. This method enhances temporal coherence by dynamically adjusting generative influence based on uncertainty-aware perturbations. Our experimental results demonstrate that our method achieves state-of-the-art performance in both appearance and geometric reconstruction, temporal generation and novel view synthesis (NVS) tasks, while simultaneously delivering competitive performance in downstream evaluations. Homepage is at \\href{https://jiangxb98.github.io/PhiGensis}{PhiGensis}.", "date": "2025-09-24", "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": "2504.21476", "title": "GarmentDiffusion: 3D Garment Sewing Pattern Generation with Multimodal Diffusion Transformers", "abstract": "Garment sewing patterns are fundamental design elements that bridge the gap between design concepts and practical manufacturing. The generative modeling of sewing patterns is crucial for creating diversified garments. However, existing approaches are limited either by reliance on a single input modality or by suboptimal generation efficiency. In this work, we present GarmentDiffusion, a new generative model capable of producing centimeter-precise, vectorized 3D sewing patterns from multimodal inputs (text, image, and incomplete sewing pattern). Our method efficiently encodes 3D sewing pattern parameters into compact edge token representations, achieving a sequence length that is 10 times shorter than that of the autoregressive SewingGPT in DressCode. By employing a diffusion transformer, we simultaneously denoise all edge tokens along the temporal axis, while maintaining a constant number of denoising steps regardless of dataset-specific edge and panel statistics. With all combination of designs of our model, the sewing pattern generation speed is accelerated by 100 times compared to SewingGPT. We achieve new state-of-the-art results on DressCodeData, as well as on the largest sewing pattern dataset, namely GarmentCodeData. The project website is available at https://shenfu-research.github.io/Garment-Diffusion/.", "date": "2025-04-30", "categories": []} {"id": "2509.16768", "title": "MMPart: Harnessing Multi-Modal Large Language Models for Part-Aware 3D Generation", "abstract": "Generative 3D modeling has advanced rapidly, driven by applications in VR/AR, metaverse, and robotics. However, most methods represent the target object as a closed mesh devoid of any structural information, limiting editing, animation, and semantic understanding. Part-aware 3D generation addresses this problem by decomposing objects into meaningful components, but existing pipelines face challenges: in existing methods, the user has no control over which objects are separated and how model imagine the occluded parts in isolation phase. In this paper, we introduce MMPart, an innovative framework for generating part-aware 3D models from a single image. We first use a VLM to generate a set of prompts based on the input image and user descriptions. In the next step, a generative model generates isolated images of each object based on the initial image and the previous step's prompts as supervisor (which control the pose and guide model how imagine previously occluded areas). Each of those images then enters the multi-view generation stage, where a number of consistent images from different views are generated. Finally, a reconstruction model converts each of these multi-view images into a 3D model.", "date": "2025-09-20", "categories": []} {"id": "2509.17206", "title": "Guided and Unguided Conditional Diffusion Mechanisms for Structured and Semantically-Aware 3D Point Cloud Generation", "abstract": "Generating realistic 3D point clouds is a fundamental problem in computer vision with applications in remote sensing, robotics, and digital object modeling. Existing generative approaches primarily capture geometry, and when semantics are considered, they are typically imposed post hoc through external segmentation or clustering rather than integrated into the generative process itself. We propose a diffusion-based framework that embeds per-point semantic conditioning directly within generation. Each point is associated with a conditional variable corresponding to its semantic label, which guides the diffusion dynamics and enables the joint synthesis of geometry and semantics. This design produces point clouds that are both structurally coherent and segmentation-aware, with object parts explicitly represented during synthesis. Through a comparative analysis of guided and unguided diffusion processes, we demonstrate the significant impact of conditional variables on diffusion dynamics and generation quality. Extensive experiments validate the efficacy of our approach, producing detailed and accurate 3D point clouds tailored to specific parts and features.", "date": "2025-09-21", "categories": []} {"id": "2509.15249", "title": "Causal Reasoning Elicits Controllable 3D Scene Generation", "abstract": "Existing 3D scene generation methods often struggle to model the complex logical dependencies and physical constraints between objects, limiting their ability to adapt to dynamic and realistic environments. We propose CausalStruct, a novel framework that embeds causal reasoning into 3D scene generation. Utilizing large language models (LLMs), We construct causal graphs where nodes represent objects and attributes, while edges encode causal dependencies and physical constraints. CausalStruct iteratively refines the scene layout by enforcing causal order to determine the placement order of objects and applies causal intervention to adjust the spatial configuration according to physics-driven constraints, ensuring consistency with textual descriptions and real-world dynamics. The refined scene causal graph informs subsequent optimization steps, employing a Proportional-Integral-Derivative(PID) controller to iteratively tune object scales and positions. Our method uses text or images to guide object placement and layout in 3D scenes, with 3D Gaussian Splatting and Score Distillation Sampling improving shape accuracy and rendering stability. Extensive experiments show that CausalStruct generates 3D scenes with enhanced logical coherence, realistic spatial interactions, and robust adaptability.", "date": "2025-09-18", "categories": []} {"id": "2509.15772", "title": "Vision-Language Models as Differentiable Semantic and Spatial Rewards for Text-to-3D Generation", "abstract": "Score Distillation Sampling (SDS) enables high-quality text-to-3D generation by supervising 3D models through the denoising of multi-view 2D renderings, using a pretrained text-to-image diffusion model to align with the input prompt and ensure 3D consistency. However, existing SDS-based methods face two fundamental limitations: (1) their reliance on CLIP-style text encoders leads to coarse semantic alignment and struggles with fine-grained prompts; and (2) 2D diffusion priors lack explicit 3D spatial constraints, resulting in geometric inconsistencies and inaccurate object relationships in multi-object scenes. To address these challenges, we propose VLM3D, a novel text-to-3D generation framework that integrates large vision-language models (VLMs) into the SDS pipeline as differentiable semantic and spatial priors. Unlike standard text-to-image diffusion priors, VLMs leverage rich language-grounded supervision that enables fine-grained prompt alignment. Moreover, their inherent vision language modeling provides strong spatial understanding, which significantly enhances 3D consistency for single-object generation and improves relational reasoning in multi-object scenes. We instantiate VLM3D based on the open-source Qwen2.5-VL model and evaluate it on the GPTeval3D benchmark. Experiments across diverse objects and complex scenes show that VLM3D significantly outperforms prior SDS-based methods in semantic fidelity, geometric coherence, and spatial correctness.", "date": "2025-09-19", "categories": []} {"id": "2508.05631", "title": "GAP: Gaussianize Any Point Clouds with Text Guidance", "abstract": "3D Gaussian Splatting (3DGS) has demonstrated its advantages in achieving fast and high-quality rendering. As point clouds serve as a widely-used and easily accessible form of 3D representation, bridging the gap between point clouds and Gaussians becomes increasingly important. Recent studies have explored how to convert the colored points into Gaussians, but directly generating Gaussians from colorless 3D point clouds remains an unsolved challenge. In this paper, we propose GAP, a novel approach that gaussianizes raw point clouds into high-fidelity 3D Gaussians with text guidance. Our key idea is to design a multi-view optimization framework that leverages a depth-aware image diffusion model to synthesize consistent appearances across different viewpoints. To ensure geometric accuracy, we introduce a surface-anchoring mechanism that effectively constrains Gaussians to lie on the surfaces of 3D shapes during optimization. Furthermore, GAP incorporates a diffuse-based inpainting strategy that specifically targets at completing hard-to-observe regions. We evaluate GAP on the Point-to-Gaussian generation task across varying complexity levels, from synthetic point clouds to challenging real-world scans, and even large-scale scenes. Project Page: https://weiqi-zhang.github.io/GAP.", "date": "2025-08-07", "categories": []} {"id": "2509.13301", "title": "StyleSculptor: Zero-Shot Style-Controllable 3D Asset Generation with Texture-Geometry Dual Guidance", "abstract": "Creating 3D assets that follow the texture and geometry style of existing ones is often desirable or even inevitable in practical applications like video gaming and virtual reality. While impressive progress has been made in generating 3D objects from text or images, creating style-controllable 3D assets remains a complex and challenging problem. In this work, we propose StyleSculptor, a novel training-free approach for generating style-guided 3D assets from a content image and one or more style images. Unlike previous works, StyleSculptor achieves style-guided 3D generation in a zero-shot manner, enabling fine-grained 3D style control that captures the texture, geometry, or both styles of user-provided style images. At the core of StyleSculptor is a novel Style Disentangled Attention (SD-Attn) module, which establishes a dynamic interaction between the input content image and style image for style-guided 3D asset generation via a cross-3D attention mechanism, enabling stable feature fusion and effective style-guided generation. To alleviate semantic content leakage, we also introduce a style-disentangled feature selection strategy within the SD-Attn module, which leverages the variance of 3D feature patches to disentangle style- and content-significant channels, allowing selective feature injection within the attention framework. With SD-Attn, the network can dynamically compute texture-, geometry-, or both-guided features to steer the 3D generation process. Built upon this, we further propose the Style Guided Control (SGC) mechanism, which enables exclusive geometry- or texture-only stylization, as well as adjustable style intensity control. Extensive experiments demonstrate that StyleSculptor outperforms existing baseline methods in producing high-fidelity 3D assets.", "date": "2025-09-16", "categories": []} {"id": "2506.01799", "title": "WorldExplorer: Towards Generating Fully Navigable 3D Scenes", "abstract": "Generating 3D worlds from text is a highly anticipated goal in computer vision. Existing works are limited by the degree of exploration they allow inside of a scene, i.e., produce streched-out and noisy artifacts when moving beyond central or panoramic perspectives. To this end, we propose WorldExplorer, a novel method based on autoregressive video trajectory generation, which builds fully navigable 3D scenes with consistent visual quality across a wide range of viewpoints. We initialize our scenes by creating multi-view consistent images corresponding to a 360 degree panorama. Then, we expand it by leveraging video diffusion models in an iterative scene generation pipeline. Concretely, we generate multiple videos along short, pre-defined trajectories, that explore the scene in depth, including motion around objects. Our novel scene memory conditions each video on the most relevant prior views, while a collision-detection mechanism prevents degenerate results, like moving into objects. Finally, we fuse all generated views into a unified 3D representation via 3D Gaussian Splatting optimization. Compared to prior approaches, WorldExplorer produces high-quality scenes that remain stable under large camera motion, enabling for the first time realistic and unrestricted exploration. We believe this marks a significant step toward generating immersive and truly explorable virtual 3D environments.", "date": "2025-06-02", "categories": []} {"id": "2509.12501", "title": "Artist-Created Mesh Generation from Raw Observation", "abstract": "We present an end-to-end framework for generating artist-style meshes from noisy or incomplete point clouds, such as those captured by real-world sensors like LiDAR or mobile RGB-D cameras. Artist-created meshes are crucial for commercial graphics pipelines due to their compatibility with animation and texturing tools and their efficiency in rendering. However, existing approaches often assume clean, complete inputs or rely on complex multi-stage pipelines, limiting their applicability in real-world scenarios. To address this, we propose an end-to-end method that refines the input point cloud and directly produces high-quality, artist-style meshes. At the core of our approach is a novel reformulation of 3D point cloud refinement as a 2D inpainting task, enabling the use of powerful generative models. Preliminary results on the ShapeNet dataset demonstrate the promise of our framework in producing clean, complete meshes.", "date": "2025-09-15", "categories": []} {"id": "2509.12721", "title": "SPGen: Spherical Projection as Consistent and Flexible Representation for Single Image 3D Shape Generation", "abstract": "Existing single-view 3D generative models typically adopt multiview diffusion priors to reconstruct object surfaces, yet they remain prone to inter-view inconsistencies and are unable to faithfully represent complex internal structure or nontrivial topologies. In particular, we encode geometry information by projecting it onto a bounding sphere and unwrapping it into a compact and structural multi-layer 2D Spherical Projection (SP) representation. Operating solely in the image domain, SPGen offers three key advantages simultaneously: (1) Consistency. The injective SP mapping encodes surface geometry with a single viewpoint which naturally eliminates view inconsistency and ambiguity; (2) Flexibility. Multi-layer SP maps represent nested internal structures and support direct lifting to watertight or open 3D surfaces; (3) Efficiency. The image-domain formulation allows the direct inheritance of powerful 2D diffusion priors and enables efficient finetuning with limited computational resources. Extensive experiments demonstrate that SPGen significantly outperforms existing baselines in geometric quality and computational efficiency.", "date": "2025-09-16", "categories": []} {"id": "2509.12815", "title": "Hunyuan3D Studio: End-to-End AI Pipeline for Game-Ready 3D Asset Generation", "abstract": "The creation of high-quality 3D assets, a cornerstone of modern game development, has long been characterized by labor-intensive and specialized workflows. This paper presents Hunyuan3D Studio, an end-to-end AI-powered content creation platform designed to revolutionize the game production pipeline by automating and streamlining the generation of game-ready 3D assets. At its core, Hunyuan3D Studio integrates a suite of advanced neural modules (such as Part-level 3D Generation, Polygon Generation, Semantic UV, etc.) into a cohesive and user-friendly system. This unified framework allows for the rapid transformation of a single concept image or textual description into a fully-realized, production-quality 3D model complete with optimized geometry and high-fidelity PBR textures. We demonstrate that assets generated by Hunyuan3D Studio are not only visually compelling but also adhere to the stringent technical requirements of contemporary game engines, significantly reducing iteration time and lowering the barrier to entry for 3D content creation. By providing a seamless bridge from creative intent to technical asset, Hunyuan3D Studio represents a significant leap forward for AI-assisted workflows in game development and interactive media.", "date": "2025-09-16", "categories": []} {"id": "2412.08641", "title": "3D Mesh Editing using Masked LRMs", "abstract": "We present a novel approach to shape editing, building on recent progress in 3D reconstruction from multi-view images. We formulate shape editing as a conditional reconstruction problem, where the model must reconstruct the input shape with the exception of a specified 3D region, in which the geometry should be generated from the conditional signal. To this end, we train a conditional Large Reconstruction Model (LRM) for masked reconstruction, using multi-view consistent masks rendered from a randomly generated 3D occlusion, and using one clean viewpoint as the conditional signal. During inference, we manually define a 3D region to edit and provide an edited image from a canonical viewpoint to fill that region. We demonstrate that, in just a single forward pass, our method not only preserves the input geometry in the unmasked region through reconstruction capabilities on par with SoTA, but is also expressive enough to perform a variety of mesh edits from a single image guidance that past works struggle with, while being 2-10x faster than the top-performing prior work.", "date": "2024-12-11", "categories": []} {"id": "2509.11624", "title": "A Controllable 3D Deepfake Generation Framework with Gaussian Splatting", "abstract": "We propose a novel 3D deepfake generation framework based on 3D Gaussian Splatting that enables realistic, identity-preserving face swapping and reenactment in a fully controllable 3D space. Compared to conventional 2D deepfake approaches that suffer from geometric inconsistencies and limited generalization to novel view, our method combines a parametric head model with dynamic Gaussian representations to support multi-view consistent rendering, precise expression control, and seamless background integration. To address editing challenges in point-based representations, we explicitly separate the head and background Gaussians and use pre-trained 2D guidance to optimize the facial region across views. We further introduce a repair module to enhance visual consistency under extreme poses and expressions. Experiments on NeRSemble and additional evaluation videos demonstrate that our method achieves comparable performance to state-of-the-art 2D approaches in identity preservation, as well as pose and expression consistency, while significantly outperforming them in multi-view rendering quality and 3D consistency. Our approach bridges the gap between 3D modeling and deepfake synthesis, enabling new directions for scene-aware, controllable, and immersive visual forgeries, revealing the threat that emerging 3D Gaussian Splatting technique could be used for manipulation attacks.", "date": "2025-09-15", "categories": []} {"id": "2509.11959", "title": "Learning to Generate 4D LiDAR Sequences", "abstract": "While generative world models have advanced video and occupancy-based data synthesis, LiDAR generation remains underexplored despite its importance for accurate 3D perception. Extending generation to 4D LiDAR data introduces challenges in controllability, temporal stability, and evaluation. We present LiDARCrafter, a unified framework that converts free-form language into editable LiDAR sequences. Instructions are parsed into ego-centric scene graphs, which a tri-branch diffusion model transforms into object layouts, trajectories, and shapes. A range-image diffusion model generates the initial scan, and an autoregressive module extends it into a temporally coherent sequence. The explicit layout design further supports object-level editing, such as insertion or relocation. To enable fair assessment, we provide EvalSuite, a benchmark spanning scene-, object-, and sequence-level metrics. On nuScenes, LiDARCrafter achieves state-of-the-art fidelity, controllability, and temporal consistency, offering a foundation for LiDAR-based simulation and data augmentation.", "date": "2025-09-15", "categories": []} {"id": "2505.13791", "title": "Scalable Autoregressive 3D Molecule Generation", "abstract": "Generative models of 3D molecular structure play a rapidly growing role in the design and simulation of molecules. Diffusion models currently dominate the space of 3D molecule generation, while autoregressive models have trailed behind. In this work, we present Quetzal, a simple but scalable autoregressive model that builds molecules atom-by-atom in 3D. Treating each molecule as an ordered sequence of atoms, Quetzal combines a causal transformer that predicts the next atom's discrete type with a smaller Diffusion MLP that models the continuous next-position distribution. Compared to existing autoregressive baselines, Quetzal achieves substantial improvements in generation quality and is competitive with the performance of state-of-the-art diffusion models. In addition, by reducing the number of expensive forward passes through a dense transformer, Quetzal enables significantly faster generation speed, as well as exact divergence-based likelihood computation. Finally, without any architectural changes, Quetzal natively handles variable-size tasks like hydrogen decoration and scaffold completion. We hope that our work motivates a perspective on scalability and generality for generative modelling of 3D molecules.", "date": "2025-05-20", "categories": []} {"id": "2508.18597", "title": "SemLayoutDiff: Semantic Layout Generation with Diffusion Model for Indoor Scene Synthesis", "abstract": "We present SemLayoutDiff, a unified model for synthesizing diverse 3D indoor scenes across multiple room types. The model introduces a scene layout representation combining a top-down semantic map and attributes for each object. Unlike prior approaches, which cannot condition on architectural constraints, SemLayoutDiff employs a categorical diffusion model capable of conditioning scene synthesis explicitly on room masks. It first generates a coherent semantic map, followed by a cross-attention-based network to predict furniture placements that respect the synthesized layout. Our method also accounts for architectural elements such as doors and windows, ensuring that generated furniture arrangements remain practical and unobstructed. Experiments on the 3D-FRONT dataset show that SemLayoutDiff produces spatially coherent, realistic, and varied scenes, outperforming previous methods.", "date": "2025-08-26", "categories": []} {"id": "2509.05263", "title": "LatticeWorld: A Multimodal Large Language Model-Empowered Framework for Interactive Complex World Generation", "abstract": "Recent research has been increasingly focusing on developing 3D world models that simulate complex real-world scenarios. World models have found broad applications across various domains, including embodied AI, autonomous driving, entertainment, etc. A more realistic simulation with accurate physics will effectively narrow the sim-to-real gap and allow us to gather rich information about the real world conveniently. While traditional manual modeling has enabled the creation of virtual 3D scenes, modern approaches have leveraged advanced machine learning algorithms for 3D world generation, with most recent advances focusing on generative methods that can create virtual worlds based on user instructions. This work explores such a research direction by proposing LatticeWorld, a simple yet effective 3D world generation framework that streamlines the industrial production pipeline of 3D environments. LatticeWorld leverages lightweight LLMs (LLaMA-2-7B) alongside the industry-grade rendering engine (e.g., Unreal Engine 5) to generate a dynamic environment. Our proposed framework accepts textual descriptions and visual instructions as multimodal inputs and creates large-scale 3D interactive worlds with dynamic agents, featuring competitive multi-agent interaction, high-fidelity physics simulation, and real-time rendering. We conduct comprehensive experiments to evaluate LatticeWorld, showing that it achieves superior accuracy in scene layout generation and visual fidelity. Moreover, LatticeWorld achieves over a $90\\times$ increase in industrial production efficiency while maintaining high creative quality compared with traditional manual production methods. Our demo video is available at https://youtu.be/8VWZXpERR18", "date": "2025-09-05", "categories": []} {"id": "2501.04631", "title": "Disentangled Clothed Avatar Generation with Layered Representation", "abstract": "Clothed avatar generation has wide applications in virtual and augmented reality, filmmaking, and more. Previous methods have achieved success in generating diverse digital avatars, however, generating avatars with disentangled components (\\eg, body, hair, and clothes) has long been a challenge. In this paper, we propose LayerAvatar, the first feed-forward diffusion-based method for generating component-disentangled clothed avatars. To achieve this, we first propose a layered UV feature plane representation, where components are distributed in different layers of the Gaussian-based UV feature plane with corresponding semantic labels. This representation supports high-resolution and real-time rendering, as well as expressive animation including controllable gestures and facial expressions. Based on the well-designed representation, we train a single-stage diffusion model and introduce constrain terms to address the severe occlusion problem of the innermost human body layer. Extensive experiments demonstrate the impressive performances of our method in generating disentangled clothed avatars, and we further explore its applications in component transfer. The project page is available at: https://olivia23333.github.io/LayerAvatar/", "date": "2025-01-08", "categories": []} {"id": "2509.05131", "title": "A Scalable Attention-Based Approach for Image-to-3D Texture Mapping", "abstract": "High-quality textures are critical for realistic 3D content creation, yet existing generative methods are slow, rely on UV maps, and often fail to remain faithful to a reference image. To address these challenges, we propose a transformer-based framework that predicts a 3D texture field directly from a single image and a mesh, eliminating the need for UV mapping and differentiable rendering, and enabling faster texture generation. Our method integrates a triplane representation with depth-based backprojection losses, enabling efficient training and faster inference. Once trained, it generates high-fidelity textures in a single forward pass, requiring only 0.2s per shape. Extensive qualitative, quantitative, and user preference evaluations demonstrate that our method outperforms state-of-the-art baselines on single-image texture reconstruction in terms of both fidelity to the input image and perceptual quality, highlighting its practicality for scalable, high-quality, and controllable 3D content creation.", "date": "2025-09-05", "categories": []} {"id": "2508.17832", "title": "HLG: Comprehensive 3D Room Construction via Hierarchical Layout Generation", "abstract": "Realistic 3D indoor scene generation is crucial for virtual reality, interior design, embodied intelligence, and scene understanding. While existing methods have made progress in coarse-scale furniture arrangement, they struggle to capture fine-grained object placements, limiting the realism and utility of generated environments. This gap hinders immersive virtual experiences and detailed scene comprehension for embodied AI applications. To address these issues, we propose Hierarchical Layout Generation (HLG), a novel method for fine-grained 3D scene generation. HLG is the first to adopt a coarse-to-fine hierarchical approach, refining scene layouts from large-scale furniture placement to intricate object arrangements. Specifically, our fine-grained layout alignment module constructs a hierarchical layout through vertical and horizontal decoupling, effectively decomposing complex 3D indoor scenes into multiple levels of granularity. Additionally, our trainable layout optimization network addresses placement issues, such as incorrect positioning, orientation errors, and object intersections, ensuring structurally coherent and physically plausible scene generation. We demonstrate the effectiveness of our approach through extensive experiments, showing superior performance in generating realistic indoor scenes compared to existing methods. This work advances the field of scene generation and opens new possibilities for applications requiring detailed 3D environments. We will release our code upon publication to encourage future research.", "date": "2025-08-25", "categories": []} {"id": "2509.04406", "title": "Few-step Flow for 3D Generation via Marginal-Data Transport Distillation", "abstract": "Flow-based 3D generation models typically require dozens of sampling steps during inference. Though few-step distillation methods, particularly Consistency Models (CMs), have achieved substantial advancements in accelerating 2D diffusion models, they remain under-explored for more complex 3D generation tasks. In this study, we propose a novel framework, MDT-dist, for few-step 3D flow distillation. Our approach is built upon a primary objective: distilling the pretrained model to learn the Marginal-Data Transport. Directly learning this objective needs to integrate the velocity fields, while this integral is intractable to be implemented. Therefore, we propose two optimizable objectives, Velocity Matching (VM) and Velocity Distillation (VD), to equivalently convert the optimization target from the transport level to the velocity and the distribution level respectively. Velocity Matching (VM) learns to stably match the velocity fields between the student and the teacher, but inevitably provides biased gradient estimates. Velocity Distillation (VD) further enhances the optimization process by leveraging the learned velocity fields to perform probability density distillation. When evaluated on the pioneer 3D generation framework TRELLIS, our method reduces sampling steps of each flow transformer from 25 to 1 or 2, achieving 0.68s (1 step x 2) and 0.94s (2 steps x 2) latency with 9.0x and 6.5x speedup on A800, while preserving high visual and geometric fidelity. Extensive experiments demonstrate that our method significantly outperforms existing CM distillation methods, and enables TRELLIS to achieve superior performance in few-step 3D generation.", "date": "2025-09-04", "categories": []} {"id": "2509.02466", "title": "TeRA: Rethinking Text-guided Realistic 3D Avatar Generation", "abstract": "In this paper, we rethink text-to-avatar generative models by proposing TeRA, a more efficient and effective framework than the previous SDS-based models and general large 3D generative models. Our approach employs a two-stage training strategy for learning a native 3D avatar generative model. Initially, we distill a decoder to derive a structured latent space from a large human reconstruction model. Subsequently, a text-controlled latent diffusion model is trained to generate photorealistic 3D human avatars within this latent space. TeRA enhances the model performance by eliminating slow iterative optimization and enables text-based partial customization through a structured 3D human representation. Experiments have proven our approach's superiority over previous text-to-avatar generative models in subjective and objective evaluation.", "date": "2025-09-02", "categories": []} {"id": "2403.12470", "title": "SC-Diff: 3D Shape Completion with Latent Diffusion Models", "abstract": "We present a novel 3D shape completion framework that unifies multimodal conditioning, leveraging both 2D images and 3D partial scans through a latent diffusion model. Shapes are represented as Truncated Signed Distance Functions (TSDFs) and encoded into a discrete latent space jointly supervised by 2D and 3D cues, enabling efficient high-resolution processing while reducing GPU memory usage by 30\\% compared to state-of-the-art methods. Our approach guides the generation process with flexible multimodal conditioning, ensuring consistent integration of 2D and 3D information from encoding to reconstruction. Our training strategy simulates realistic partial observations, avoiding assumptions about input structure and improving robustness in real-world scenarios. Leveraging our efficient latent space and multimodal conditioning, our model generalizes across object categories, outperforming class-specific models by 12\\% and class-agnostic models by 47\\% in $l_1$ reconstruction error, while producing more diverse, realistic, and high-fidelity completions than prior approaches.", "date": "2024-03-19", "categories": []} {"id": "2407.01074", "title": "Multimodal Conditional 3D Face Geometry Generation", "abstract": "We present a new method for multimodal conditional 3D face geometry generation that allows user-friendly control over the output identity and expression via a number of different conditioning signals. Within a single model, we demonstrate 3D faces generated from artistic sketches, portrait photos, Canny edges, FLAME face model parameters, 2D face landmarks, or text prompts. Our approach is based on a diffusion process that generates 3D geometry in a 2D parameterized UV domain. Geometry generation passes each conditioning signal through a set of cross-attention layers (IP-Adapter), one set for each user-defined conditioning signal. The result is an easy-to-use 3D face generation tool that produces topology-consistent, high-quality geometry with fine-grain user control.", "date": "2024-07-01", "categories": []} {"id": "2501.08983", "title": "Compositional Generative Model of Unbounded 4D Cities", "abstract": "3D scene generation has garnered growing attention in recent years and has made significant progress. Generating 4D cities is more challenging than 3D scenes due to the presence of structurally complex, visually diverse objects like buildings and vehicles, and heightened human sensitivity to distortions in urban environments. To tackle these issues, we propose CityDreamer4D, a compositional generative model specifically tailored for generating unbounded 4D cities. Our main insights are 1) 4D city generation should separate dynamic objects (e.g., vehicles) from static scenes (e.g., buildings and roads), and 2) all objects in the 4D scene should be composed of different types of neural fields for buildings, vehicles, and background stuff. Specifically, we propose Traffic Scenario Generator and Unbounded Layout Generator to produce dynamic traffic scenarios and static city layouts using a highly compact BEV representation. Objects in 4D cities are generated by combining stuff-oriented and instance-oriented neural fields for background stuff, buildings, and vehicles. To suit the distinct characteristics of background stuff and instances, the neural fields employ customized generative hash grids and periodic positional embeddings as scene parameterizations. Furthermore, we offer a comprehensive suite of datasets for city generation, including OSM, GoogleEarth, and CityTopia. The OSM dataset provides a variety of real-world city layouts, while the Google Earth and CityTopia datasets deliver large-scale, high-quality city imagery complete with 3D instance annotations. Leveraging its compositional design, CityDreamer4D supports a range of downstream applications, such as instance editing, city stylization, and urban simulation, while delivering state-of-the-art performance in generating realistic 4D cities.", "date": "2025-01-15", "categories": []} {"id": "2501.10462", "title": "BloomScene: Lightweight Structured 3D Gaussian Splatting for Crossmodal Scene Generation", "abstract": "With the widespread use of virtual reality applications, 3D scene generation has become a new challenging research frontier. 3D scenes have highly complex structures and need to ensure that the output is dense, coherent, and contains all necessary structures. Many current 3D scene generation methods rely on pre-trained text-to-image diffusion models and monocular depth estimators. However, the generated scenes occupy large amounts of storage space and often lack effective regularisation methods, leading to geometric distortions. To this end, we propose BloomScene, a lightweight structured 3D Gaussian splatting for crossmodal scene generation, which creates diverse and high-quality 3D scenes from text or image inputs. Specifically, a crossmodal progressive scene generation framework is proposed to generate coherent scenes utilizing incremental point cloud reconstruction and 3D Gaussian splatting. Additionally, we propose a hierarchical depth prior-based regularization mechanism that utilizes multi-level constraints on depth accuracy and smoothness to enhance the realism and continuity of the generated scenes. Ultimately, we propose a structured context-guided compression mechanism that exploits structured hash grids to model the context of unorganized anchor attributes, which significantly eliminates structural redundancy and reduces storage overhead. Comprehensive experiments across multiple scenes demonstrate the significant potential and advantages of our framework compared with several baselines.", "date": "2025-01-15", "categories": []} {"id": "2503.16375", "title": "NuiScene: Exploring Efficient Generation of Unbounded Outdoor Scenes", "abstract": "In this paper, we explore the task of generating expansive outdoor scenes, ranging from castles to high-rises. Unlike indoor scene generation, which has been a primary focus of prior work, outdoor scene generation presents unique challenges, including wide variations in scene heights and the need for a method capable of rapidly producing large landscapes. To address this, we propose an efficient approach that encodes scene chunks as uniform vector sets, offering better compression and performance than the spatially structured latents used in prior methods. Furthermore, we train an explicit outpainting model for unbounded generation, which improves coherence compared to prior resampling-based inpainting schemes while also speeding up generation by eliminating extra diffusion steps. To facilitate this task, we curate NuiScene43, a small but high-quality set of scenes, preprocessed for joint training. Notably, when trained on scenes of varying styles, our model can blend different environments, such as rural houses and city skyscrapers, within the same scene, highlighting the potential of our curation process to leverage heterogeneous scenes for joint training.", "date": "2025-03-20", "categories": []} {"id": "2509.01134", "title": "RealMat: Realistic Materials with Diffusion and Reinforcement Learning", "abstract": "Generative models for high-quality materials are particularly desirable to make 3D content authoring more accessible. However, the majority of material generation methods are trained on synthetic data. Synthetic data provides precise supervision for material maps, which is convenient but also tends to create a significant visual gap with real-world materials. Alternatively, recent work used a small dataset of real flash photographs to guarantee realism, however such data is limited in scale and diversity. To address these limitations, we propose RealMat, a diffusion-based material generator that leverages realistic priors, including a text-to-image model and a dataset of realistic material photos under natural lighting. In RealMat, we first finetune a pretrained Stable Diffusion XL (SDXL) with synthetic material maps arranged in $2 \\times 2$ grids. This way, our model inherits some realism of SDXL while learning the data distribution of the synthetic material grids. Still, this creates a realism gap, with some generated materials appearing synthetic. We propose to further finetune our model through reinforcement learning (RL), encouraging the generation of realistic materials. We develop a realism reward function for any material image under natural lighting, by collecting a large-scale dataset of realistic material images. We show that this approach increases generated materials' realism compared to our base model and related work.", "date": "2025-09-01", "categories": []} {"id": "2509.01492", "title": "A Continuous-Time Consistency Model for 3D Point Cloud Generation", "abstract": "Fast and accurate 3D shape generation from point clouds is essential for applications in robotics, AR/VR, and digital content creation. We introduce ConTiCoM-3D, a continuous-time consistency model that synthesizes 3D shapes directly in point space, without discretized diffusion steps, pre-trained teacher models, or latent-space encodings. The method integrates a TrigFlow-inspired continuous noise schedule with a Chamfer Distance-based geometric loss, enabling stable training on high-dimensional point sets while avoiding expensive Jacobian-vector products. This design supports efficient one- to two-step inference with high geometric fidelity. In contrast to previous approaches that rely on iterative denoising or latent decoders, ConTiCoM-3D employs a time-conditioned neural network operating entirely in continuous time, thereby achieving fast generation. Experiments on the ShapeNet benchmark show that ConTiCoM-3D matches or outperforms state-of-the-art diffusion and latent consistency models in both quality and efficiency, establishing it as a practical framework for scalable 3D shape generation.", "date": "2025-09-01", "categories": []} {"id": "2509.02357", "title": "Category-Aware 3D Object Composition with Disentangled Texture and Shape Multi-view Diffusion", "abstract": "In this paper, we tackle a new task of 3D object synthesis, where a 3D model is composited with another object category to create a novel 3D model. However, most existing text/image/3D-to-3D methods struggle to effectively integrate multiple content sources, often resulting in inconsistent textures and inaccurate shapes. To overcome these challenges, we propose a straightforward yet powerful approach, category+3D-to-3D (C33D), for generating novel and structurally coherent 3D models. Our method begins by rendering multi-view images and normal maps from the input 3D model, then generating a novel 2D object using adaptive text-image harmony (ATIH) with the front-view image and a text description from another object category as inputs. To ensure texture consistency, we introduce texture multi-view diffusion, which refines the textures of the remaining multi-view RGB images based on the novel 2D object. For enhanced shape accuracy, we propose shape multi-view diffusion to improve the 2D shapes of both the multi-view RGB images and the normal maps, also conditioned on the novel 2D object. Finally, these outputs are used to reconstruct a complete and novel 3D model. Extensive experiments demonstrate the effectiveness of our method, yielding impressive 3D creations, such as shark(3D)-crocodile(text) in the first row of Fig. 1. A project page is available at: https://xzr52.github.io/C33D/", "date": "2025-09-02", "categories": []} {"id": "2509.02474", "title": "Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework", "abstract": "Following rapid advancements in text and image generation, research has increasingly shifted towards 3D generation. Unlike the well-established pixel-based representation in images, 3D representations remain diverse and fragmented, encompassing a wide variety of approaches such as voxel grids, neural radiance fields, signed distance functions, point clouds, or octrees, each offering distinct advantages and limitations. In this work, we present a unified evaluation framework designed to assess the performance of 3D representations in reconstruction and generation. We compare these representations based on multiple criteria: quality, computational efficiency, and generalization performance. Beyond standard model benchmarking, our experiments aim to derive best practices over all steps involved in the 3D generation pipeline, including preprocessing, mesh reconstruction, compression with autoencoders, and generation. Our findings highlight that reconstruction errors significantly impact overall performance, underscoring the need to evaluate generation and reconstruction jointly. We provide insights that can inform the selection of suitable 3D models for various applications, facilitating the development of more robust and application-specific solutions in 3D generation. The code for our framework is available at https://github.com/isl-org/unifi3d.", "date": "2025-09-02", "categories": []} {"id": "2503.22677", "title": "DSO: Aligning 3D Generators with Simulation Feedback for Physical Soundness", "abstract": "Most 3D object generators prioritize aesthetic quality, often neglecting the physical constraints necessary for practical applications. One such constraint is that a 3D object should be self-supporting, i.e., remain balanced under gravity. Previous approaches to generating stable 3D objects relied on differentiable physics simulators to optimize geometry at test time, which is slow, unstable, and prone to local optima. Inspired by the literature on aligning generative models with external feedback, we propose Direct Simulation Optimization (DSO). This framework leverages feedback from a (non-differentiable) simulator to increase the likelihood that the 3D generator directly outputs stable 3D objects. We construct a dataset of 3D objects labeled with stability scores obtained from the physics simulator. This dataset enables fine-tuning of the 3D generator using the stability score as an alignment metric, via direct preference optimization (DPO) or direct reward optimization (DRO) - a novel objective we introduce to align diffusion models without requiring pairwise preferences. Our experiments demonstrate that the fine-tuned feed-forward generator, using either the DPO or DRO objective, is significantly faster and more likely to produce stable objects than test-time optimization. Notably, the DSO framework functions even without any ground-truth 3D objects for training, allowing the 3D generator to self-improve by automatically collecting simulation feedback on its own outputs.", "date": "2025-03-28", "categories": []} {"id": "2504.08419", "title": "GeoTexBuild: 3D Building Model Generation from Map Footprints", "abstract": "We introduce GeoTexBuild, a modular generative framework for creating 3D building models from footprints derived from site planning or map designs. The system is designed for architects and city planners, offering a seamless solution that directly converts map features into 3D buildings. The proposed framework employs a three-stage process comprising height map generation, geometry reconstruction, and appearance stylization, culminating in building models with detailed geometry and appearance attributes. By integrating customized ControlNet, Neural style field (NSF), and Multi-view diffusion model, we explore effective methods for controlling both geometric and visual attributes during the generation process. Our approach eliminates the problem of structural variations in a single facade image in existing 3D generation techniques for buildings. Experimental results at each stage validate the capability of GeoTexBuild to generate detailed and accurate building models from footprints.", "date": "2025-04-11", "categories": []} {"id": "2508.20470", "title": "Droplet3D: Commonsense Priors from Videos Facilitate 3D Generation", "abstract": "Scaling laws have validated the success and promise of large-data-trained models in creative generation across text, image, and video domains. However, this paradigm faces data scarcity in the 3D domain, as there is far less of it available on the internet compared to the aforementioned modalities. Fortunately, there exist adequate videos that inherently contain commonsense priors, offering an alternative supervisory signal to mitigate the generalization bottleneck caused by limited native 3D data. On the one hand, videos capturing multiple views of an object or scene provide a spatial consistency prior for 3D generation. On the other hand, the rich semantic information contained within the videos enables the generated content to be more faithful to the text prompts and semantically plausible. This paper explores how to apply the video modality in 3D asset generation, spanning datasets to models. We introduce Droplet3D-4M, the first large-scale video dataset with multi-view level annotations, and train Droplet3D, a generative model supporting both image and dense text input. Extensive experiments validate the effectiveness of our approach, demonstrating its ability to produce spatially consistent and semantically plausible content. Moreover, in contrast to the prevailing 3D solutions, our approach exhibits the potential for extension to scene-level applications. This indicates that the commonsense priors from the videos significantly facilitate 3D creation. We have open-sourced all resources including the dataset, code, technical framework, and model weights: https://dropletx.github.io/.", "date": "2025-08-28", "categories": []} {"id": "2405.18525", "title": "REPARO: Compositional 3D Assets Generation with Differentiable 3D Layout Alignment", "abstract": "Traditional image-to-3D models often struggle with scenes containing multiple objects due to biases and occlusion complexities. To address this challenge, we present REPARO, a novel approach for compositional 3D asset generation from single images. REPARO employs a two-step process: first, it extracts individual objects from the scene and reconstructs their 3D meshes using off-the-shelf image-to-3D models; then, it optimizes the layout of these meshes through differentiable rendering techniques, ensuring coherent scene composition. By integrating optimal transport-based long-range appearance loss term and high-level semantic loss term in the differentiable rendering, REPARO can effectively recover the layout of 3D assets. The proposed method can significantly enhance object independence, detail accuracy, and overall scene coherence. Extensive evaluation of multi-object scenes demonstrates that our REPARO offers a comprehensive approach to address the complexities of multi-object 3D scene generation from single images.", "date": "2024-05-28", "categories": []} {"id": "2505.04831", "title": "Steerable Scene Generation with Post Training and Inference-Time Search", "abstract": "Training robots in simulation requires diverse 3D scenes that reflect the specific challenges of downstream tasks. However, scenes that satisfy strict task requirements, such as high-clutter environments with plausible spatial arrangement, are rare and costly to curate manually. Instead, we generate large-scale scene data using procedural models that approximate realistic environments for robotic manipulation, and adapt it to task-specific goals. We do this by training a unified diffusion-based generative model that predicts which objects to place from a fixed asset library, along with their SE(3) poses. This model serves as a flexible scene prior that can be adapted using reinforcement learning-based post training, conditional generation, or inference-time search, steering generation toward downstream objectives even when they differ from the original data distribution. Our method enables goal-directed scene synthesis that respects physical feasibility and scales across scene types. We introduce a novel MCTS-based inference-time search strategy for diffusion models, enforce feasibility via projection and simulation, and release a dataset of over 44 million SE(3) scenes spanning five diverse environments. Website with videos, code, data, and model weights: https://steerable-scene-generation.github.io/", "date": "2025-05-07", "categories": []} {"id": "2508.18531", "title": "SAT-SKYLINES: 3D Building Generation from Satellite Imagery and Coarse Geometric Priors", "abstract": "We present SatSkylines, a 3D building generation approach that takes satellite imagery and coarse geometric priors. Without proper geometric guidance, existing image-based 3D generation methods struggle to recover accurate building structures from the top-down views of satellite images alone. On the other hand, 3D detailization methods tend to rely heavily on highly detailed voxel inputs and fail to produce satisfying results from simple priors such as cuboids. To address these issues, our key idea is to model the transformation from interpolated noisy coarse priors to detailed geometries, enabling flexible geometric control without additional computational cost. We have further developed Skylines-50K, a large-scale dataset of over 50,000 unique and stylized 3D building assets in order to support the generations of detailed building models. Extensive evaluations indicate the effectiveness of our model and strong generalization ability.", "date": "2025-08-25", "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": "2508.19247", "title": "VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space", "abstract": "3D local editing of specified regions is crucial for game industry and robot interaction. Recent methods typically edit rendered multi-view images and then reconstruct 3D models, but they face challenges in precisely preserving unedited regions and overall coherence. Inspired by structured 3D generative models, we propose VoxHammer, a novel training-free approach that performs precise and coherent editing in 3D latent space. Given a 3D model, VoxHammer first predicts its inversion trajectory and obtains its inverted latents and key-value tokens at each timestep. Subsequently, in the denoising and editing phase, we replace the denoising features of preserved regions with the corresponding inverted latents and cached key-value tokens. By retaining these contextual features, this approach ensures consistent reconstruction of preserved areas and coherent integration of edited parts. To evaluate the consistency of preserved regions, we constructed Edit3D-Bench, a human-annotated dataset comprising hundreds of samples, each with carefully labeled 3D editing regions. Experiments demonstrate that VoxHammer significantly outperforms existing methods in terms of both 3D consistency of preserved regions and overall quality. Our method holds promise for synthesizing high-quality edited paired data, thereby laying the data foundation for in-context 3D generation. See our project page at https://huanngzh.github.io/VoxHammer-Page/.", "date": "2025-08-26", "categories": []} {"id": "2501.13087", "title": "Orchid: Image Latent Diffusion for Joint Appearance and Geometry Generation", "abstract": "We introduce Orchid, a unified latent diffusion model that learns a joint appearance-geometry prior to generate color, depth, and surface normal images in a single diffusion process. This unified approach is more efficient and coherent than current pipelines that use separate models for appearance and geometry. Orchid is versatile - it directly generates color, depth, and normal images from text, supports joint monocular depth and normal estimation with color-conditioned finetuning, and seamlessly inpaints large 3D regions by sampling from the joint distribution. It leverages a novel Variational Autoencoder (VAE) that jointly encodes RGB, relative depth, and surface normals into a shared latent space, combined with a latent diffusion model that denoises these latents. Our extensive experiments demonstrate that Orchid delivers competitive performance against SOTA task-specific methods for geometry prediction, even surpassing them in normal-prediction accuracy and depth-normal consistency. It also inpaints color-depth-normal images jointly, with more qualitative realism than existing multi-step methods.", "date": "2025-01-22", "categories": []} {"id": "2508.16917", "title": "Structural Energy-Guided Sampling for View-Consistent Text-to-3D", "abstract": "Text-to-3D generation often suffers from the Janus problem, where objects look correct from the front but collapse into duplicated or distorted geometry from other angles. We attribute this failure to viewpoint bias in 2D diffusion priors, which propagates into 3D optimization. To address this, we propose Structural Energy-Guided Sampling (SEGS), a training-free, plug-and-play framework that enforces multi-view consistency entirely at sampling time. SEGS defines a structural energy in a PCA subspace of intermediate U-Net features and injects its gradients into the denoising trajectory, steering geometry toward the intended viewpoint while preserving appearance fidelity. Integrated seamlessly into SDS/VSD pipelines, SEGS significantly reduces Janus artifacts, achieving improved geometric alignment and viewpoint consistency without retraining or weight modification.", "date": "2025-08-23", "categories": []} {"id": "2508.16932", "title": "Align 3D Representation and Text Embedding for 3D Content Personalization", "abstract": "Recent advances in NeRF and 3DGS have significantly enhanced the efficiency and quality of 3D content synthesis. However, efficient personalization of generated 3D content remains a critical challenge. Current 3D personalization approaches predominantly rely on knowledge distillation-based methods, which require computationally expensive retraining procedures. To address this challenge, we propose \\textbf{Invert3D}, a novel framework for convenient 3D content personalization. Nowadays, vision-language models such as CLIP enable direct image personalization through aligned vision-text embedding spaces. However, the inherent structural differences between 3D content and 2D images preclude direct application of these techniques to 3D personalization. Our approach bridges this gap by establishing alignment between 3D representations and text embedding spaces. Specifically, we develop a camera-conditioned 3D-to-text inverse mechanism that projects 3D contents into a 3D embedding aligned with text embeddings. This alignment enables efficient manipulation and personalization of 3D content through natural language prompts, eliminating the need for computationally retraining procedures. Extensive experiments demonstrate that Invert3D achieves effective personalization of 3D content. Our work is available at: https://github.com/qsong2001/Invert3D.", "date": "2025-08-23", "categories": []} {"id": "2408.00083", "title": "Localized Gaussian Splatting Editing with Contextual Awareness", "abstract": "Recent text-guided generation of individual 3D object has achieved great success using diffusion priors. However, these methods are not suitable for object insertion and replacement tasks as they do not consider the background, leading to illumination mismatches within the environment. To bridge the gap, we introduce an illumination-aware 3D scene editing pipeline for 3D Gaussian Splatting (3DGS) representation. Our key observation is that inpainting by the state-of-the-art conditional 2D diffusion model is consistent with background in lighting. To leverage the prior knowledge from the well-trained diffusion models for 3D object generation, our approach employs a coarse-to-fine objection optimization pipeline with inpainted views. In the first coarse step, we achieve image-to-3D lifting given an ideal inpainted view. The process employs 3D-aware diffusion prior from a view-conditioned diffusion model, which preserves illumination present in the conditioning image. To acquire an ideal inpainted image, we introduce an Anchor View Proposal (AVP) algorithm to find a single view that best represents the scene illumination in target region. In the second Texture Enhancement step, we introduce a novel Depth-guided Inpainting Score Distillation Sampling (DI-SDS), which enhances geometry and texture details with the inpainting diffusion prior, beyond the scope of the 3D-aware diffusion prior knowledge in the first coarse step. DI-SDS not only provides fine-grained texture enhancement, but also urges optimization to respect scene lighting. Our approach efficiently achieves local editing with global illumination consistency without explicitly modeling light transport. We demonstrate robustness of our method by evaluating editing in real scenes containing explicit highlight and shadows, and compare against the state-of-the-art text-to-3D editing methods.", "date": "2024-07-31", "categories": []} {"id": "2412.11183", "title": "OccScene: Semantic Occupancy-based Cross-task Mutual Learning for 3D Scene Generation", "abstract": "Recent diffusion models have demonstrated remarkable performance in both 3D scene generation and perception tasks. Nevertheless, existing methods typically separate these two processes, acting as a data augmenter to generate synthetic data for downstream perception tasks. In this work, we propose OccScene, a novel mutual learning paradigm that integrates fine-grained 3D perception and high-quality generation in a unified framework, achieving a cross-task win-win effect. OccScene generates new and consistent 3D realistic scenes only depending on text prompts, guided with semantic occupancy in a joint-training diffusion framework. To align the occupancy with the diffusion latent, a Mamba-based Dual Alignment module is introduced to incorporate fine-grained semantics and geometry as perception priors. Within OccScene, the perception module can be effectively improved with customized and diverse generated scenes, while the perception priors in return enhance the generation performance for mutual benefits. Extensive experiments show that OccScene achieves realistic 3D scene generation in broad indoor and outdoor scenarios, while concurrently boosting the perception models to achieve substantial performance improvements in the 3D perception task of semantic occupancy prediction.", "date": "2024-12-15", "categories": []} {"id": "2503.10660", "title": "Text-to-3D Generation using Jensen-Shannon Score Distillation", "abstract": "Score distillation sampling is an effective technique to generate 3D models from text prompts, utilizing pre-trained large-scale text-to-image diffusion models as guidance. However, the produced 3D assets tend to be over-saturating, over-smoothing, with limited diversity. These issues are results from a reverse Kullback-Leibler (KL) divergence objective, which makes the optimization unstable and results in mode-seeking behavior. In this paper, we derive a bounded score distillation objective based on Jensen-Shannon divergence (JSD), which stabilizes the optimization process and produces high-quality 3D generation. JSD can match well generated and target distribution, therefore mitigating mode seeking. We provide a practical implementation of JSD by utilizing the theory of generative adversarial networks to define an approximate objective function for the generator, assuming the discriminator is well trained. By assuming the discriminator following a log-odds classifier, we propose a minority sampling algorithm to estimate the gradients of our proposed objective, providing a practical implementation for JSD. We conduct both theoretical and empirical studies to validate our method. Experimental results on T3Bench demonstrate that our method can produce high-quality and diversified 3D assets.", "date": "2025-03-08", "categories": []} {"id": "2508.16577", "title": "MV-RAG: Retrieval Augmented Multiview Diffusion", "abstract": "Text-to-3D generation approaches have advanced significantly by leveraging pretrained 2D diffusion priors, producing high-quality and 3D-consistent outputs. However, they often fail to produce out-of-domain (OOD) or rare concepts, yielding inconsistent or inaccurate results. To this end, we propose MV-RAG, a novel text-to-3D pipeline that first retrieves relevant 2D images from a large in-the-wild 2D database and then conditions a multiview diffusion model on these images to synthesize consistent and accurate multiview outputs. Training such a retrieval-conditioned model is achieved via a novel hybrid strategy bridging structured multiview data and diverse 2D image collections. This involves training on multiview data using augmented conditioning views that simulate retrieval variance for view-specific reconstruction, alongside training on sets of retrieved real-world 2D images using a distinctive held-out view prediction objective: the model predicts the held-out view from the other views to infer 3D consistency from 2D data. To facilitate a rigorous OOD evaluation, we introduce a new collection of challenging OOD prompts. Experiments against state-of-the-art text-to-3D, image-to-3D, and personalization baselines show that our approach significantly improves 3D consistency, photorealism, and text adherence for OOD/rare concepts, while maintaining competitive performance on standard benchmarks.", "date": "2025-08-22", "categories": []} {"id": "2508.14811", "title": "Tinker: Diffusion's Gift to 3D--Multi-View Consistent Editing From Sparse Inputs without Per-Scene Optimization", "abstract": "We introduce Tinker, a versatile framework for high-fidelity 3D editing that operates in both one-shot and few-shot regimes without any per-scene finetuning. Unlike prior techniques that demand extensive per-scene optimization to ensure multi-view consistency or to produce dozens of consistent edited input views, Tinker delivers robust, multi-view consistent edits from as few as one or two images. This capability stems from repurposing pretrained diffusion models, which unlocks their latent 3D awareness. To drive research in this space, we curate the first large-scale multi-view editing dataset and data pipeline, spanning diverse scenes and styles. Building on this dataset, we develop our framework capable of generating multi-view consistent edited views without per-scene training, which consists of two novel components: (1) Referring multi-view editor: Enables precise, reference-driven edits that remain coherent across all viewpoints. (2) Any-view-to-video synthesizer: Leverages spatial-temporal priors from video diffusion to perform high-quality scene completion and novel-view generation even from sparse inputs. Through extensive experiments, Tinker significantly reduces the barrier to generalizable 3D content creation, achieving state-of-the-art performance on editing, novel-view synthesis, and rendering enhancement tasks. We believe that Tinker represents a key step towards truly scalable, zero-shot 3D editing. Project webpage: https://aim-uofa.github.io/Tinker", "date": "2025-08-20", "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": "2502.00360", "title": "Shape from Semantics: 3D Shape Generation from Multi-View Semantics", "abstract": "Existing 3D reconstruction methods utilize guidances such as 2D images, 3D point clouds, shape contours and single semantics to recover the 3D surface, which limits the creative exploration of 3D modeling. In this paper, we propose a novel 3D modeling task called ``Shape from Semantics'', which aims to create 3D models whose geometry and appearance are consistent with the given text semantics when viewed from different views. The reconstructed 3D models incorporate more than one semantic elements and are easy for observers to distinguish. We adopt generative models as priors and disentangle the connection between geometry and appearance to solve this challenging problem. Specifically, we propose Local Geometry-Aware Distillation (LGAD), a strategy that employs multi-view normal-depth diffusion priors to complete partial geometries, ensuring realistic shape generation. We also integrate view-adaptive guidance scales to enable smooth semantic transitions across views. For appearance modeling, we adopt physically based rendering to generate high-quality material properties, which are subsequently baked into fabricable meshes. Extensive experimental results demonstrate that our method can generate meshes with well-structured, intricately detailed geometries, coherent textures, and smooth transitions, resulting in visually appealing 3D shape designs. Project page: https://shapefromsemantics.github.io", "date": "2025-02-01", "categories": []} {"id": "2508.13154", "title": "4DNeX: Feed-Forward 4D Generative Modeling Made Easy", "abstract": "We present 4DNeX, the first feed-forward framework for generating 4D (i.e., dynamic 3D) scene representations from a single image. In contrast to existing methods that rely on computationally intensive optimization or require multi-frame video inputs, 4DNeX enables efficient, end-to-end image-to-4D generation by fine-tuning a pretrained video diffusion model. Specifically, 1) to alleviate the scarcity of 4D data, we construct 4DNeX-10M, a large-scale dataset with high-quality 4D annotations generated using advanced reconstruction approaches. 2) we introduce a unified 6D video representation that jointly models RGB and XYZ sequences, facilitating structured learning of both appearance and geometry. 3) we propose a set of simple yet effective adaptation strategies to repurpose pretrained video diffusion models for 4D modeling. 4DNeX produces high-quality dynamic point clouds that enable novel-view video synthesis. Extensive experiments demonstrate that 4DNeX outperforms existing 4D generation methods in efficiency and generalizability, offering a scalable solution for image-to-4D modeling and laying the foundation for generative 4D world models that simulate dynamic scene evolution.", "date": "2025-08-18", "categories": []} {"id": "2508.11106", "title": "HierOctFusion: Multi-scale Octree-based 3D Shape Generation via Part-Whole-Hierarchy Message Passing", "abstract": "3D content generation remains a fundamental yet challenging task due to the inherent structural complexity of 3D data. While recent octree-based diffusion models offer a promising balance between efficiency and quality through hierarchical generation, they often overlook two key insights: 1) existing methods typically model 3D objects as holistic entities, ignoring their semantic part hierarchies and limiting generalization; and 2) holistic high-resolution modeling is computationally expensive, whereas real-world objects are inherently sparse and hierarchical, making them well-suited for layered generation. Motivated by these observations, we propose HierOctFusion, a part-aware multi-scale octree diffusion model that enhances hierarchical feature interaction for generating fine-grained and sparse object structures. Furthermore, we introduce a cross-attention conditioning mechanism that injects part-level information into the generation process, enabling semantic features to propagate effectively across hierarchical levels from parts to the whole. Additionally, we construct a 3D dataset with part category annotations using a pre-trained segmentation model to facilitate training and evaluation. Experiments demonstrate that HierOctFusion achieves superior shape quality and efficiency compared to prior methods.", "date": "2025-08-14", "categories": []} {"id": "2508.11203", "title": "StyleMM: Stylized 3D Morphable Face Model via Text-Driven Aligned Image Translation", "abstract": "We introduce StyleMM, a novel framework that can construct a stylized 3D Morphable Model (3DMM) based on user-defined text descriptions specifying a target style. Building upon a pre-trained mesh deformation network and a texture generator for original 3DMM-based realistic human faces, our approach fine-tunes these models using stylized facial images generated via text-guided image-to-image (i2i) translation with a diffusion model, which serve as stylization targets for the rendered mesh. To prevent undesired changes in identity, facial alignment, or expressions during i2i translation, we introduce a stylization method that explicitly preserves the facial attributes of the source image. By maintaining these critical attributes during image stylization, the proposed approach ensures consistent 3D style transfer across the 3DMM parameter space through image-based training. Once trained, StyleMM enables feed-forward generation of stylized face meshes with explicit control over shape, expression, and texture parameters, producing meshes with consistent vertex connectivity and animatability. Quantitative and qualitative evaluations demonstrate that our approach outperforms state-of-the-art methods in terms of identity-level facial diversity and stylization capability. The code and videos are available at [kwanyun.github.io/stylemm_page](kwanyun.github.io/stylemm_page).", "date": "2025-08-15", "categories": []} {"id": "2412.02287", "title": "Improving Viewpoint Consistency in 3D Generation via Structure Feature and CLIP Guidance", "abstract": "Despite recent advances in text-to-3D generation techniques, current methods often suffer from geometric inconsistencies, commonly referred to as the Janus Problem. This paper identifies the root cause of the Janus Problem: viewpoint generation bias in diffusion models, which creates a significant gap between the actual generated viewpoint and the expected one required for optimizing the 3D model. To address this issue, we propose a tuning-free approach called the Attention and CLIP Guidance (ACG) mechanism. ACG enhances desired viewpoints by adaptively controlling cross-attention maps, employs CLIP-based view-text similarities to filter out erroneous viewpoints, and uses a coarse-to-fine optimization strategy with staged prompts to progressively refine 3D generation. Extensive experiments demonstrate that our method significantly reduces the Janus Problem without compromising generation speed, establishing ACG as an efficient, plug-and-play component for existing text-to-3D frameworks.", "date": "2024-12-03", "categories": []} {"id": "2501.14317", "title": "Nautilus: Locality-aware Autoencoder for Scalable Mesh Generation", "abstract": "Triangle meshes are fundamental to 3D applications, enabling efficient modification and rasterization while maintaining compatibility with standard rendering pipelines. However, current automatic mesh generation methods typically rely on intermediate representations that lack the continuous surface quality inherent to meshes. Converting these representations into meshes produces dense, suboptimal outputs. Although recent autoregressive approaches demonstrate promise in directly modeling mesh vertices and faces, they are constrained by the limitation in face count, scalability, and structural fidelity. To address these challenges, we propose Nautilus, a locality-aware autoencoder for artist-like mesh generation that leverages the local properties of manifold meshes to achieve structural fidelity and efficient representation. Our approach introduces a novel tokenization algorithm that preserves face proximity relationships and compresses sequence length through locally shared vertices and edges, enabling the generation of meshes with an unprecedented scale of up to 5,000 faces. Furthermore, we develop a Dual-stream Point Conditioner that provides multi-scale geometric guidance, ensuring global consistency and local structural fidelity by capturing fine-grained geometric features. Extensive experiments demonstrate that Nautilus significantly outperforms state-of-the-art methods in both fidelity and scalability. The project page is at https://nautilusmeshgen.github.io.", "date": "2025-01-24", "categories": []} {"id": "2504.14975", "title": "Cyc3D: Fine-grained Controllable 3D Generation via Cycle Consistency Regularization", "abstract": "Despite the remarkable progress of 3D generation, achieving controllability, i.e., ensuring consistency between generated 3D content and input conditions like edge and depth, remains a significant challenge. Existing methods often struggle to maintain accurate alignment, leading to noticeable discrepancies. To address this issue, we propose \\name{}, a new framework that enhances controllable 3D generation by explicitly encouraging cyclic consistency between the second-order 3D content, generated based on extracted signals from the first-order generation, and its original input controls. Specifically, we employ an efficient feed-forward backbone that can generate a 3D object from an input condition and a text prompt. Given an initial viewpoint and a control signal, a novel view is rendered from the generated 3D content, from which the extracted condition is used to regenerate the 3D content. This re-generated output is then rendered back to the initial viewpoint, followed by another round of control signal extraction, forming a cyclic process with two consistency constraints. \\emph{View consistency} ensures coherence between the two generated 3D objects, measured by semantic similarity to accommodate generative diversity. \\emph{Condition consistency} aligns the final extracted signal with the original input control, preserving structural or geometric details throughout the process. Extensive experiments on popular benchmarks demonstrate that \\name{} significantly improves controllability, especially for fine-grained details, outperforming existing methods across various conditions (e.g., +14.17\\% PSNR for edge, +6.26\\% PSNR for sketch).", "date": "2025-04-21", "categories": []} {"id": "2504.19860", "title": "CoherenDream: Boosting Holistic Text Coherence in 3D Generation via Multimodal Large Language Models Feedback", "abstract": "Score Distillation Sampling (SDS) has achieved remarkable success in text-to-3D content generation. However, SDS-based methods struggle to maintain semantic fidelity for user prompts, particularly when involving multiple objects with intricate interactions. While existing approaches often address 3D consistency through multiview diffusion model fine-tuning on 3D datasets, this strategy inadvertently exacerbates text-3D alignment degradation. The limitation stems from SDS's inherent accumulation of view-independent biases during optimization, which progressively diverges from the ideal text alignment direction. To alleviate this limitation, we propose a novel SDS objective, dubbed as Textual Coherent Score Distillation (TCSD), which integrates alignment feedback from multimodal large language models (MLLMs). Our TCSD leverages cross-modal understanding capabilities of MLLMs to assess and guide the text-3D correspondence during the optimization. We further develop 3DLLaVA-CRITIC - a fine-tuned MLLM specialized for evaluating multiview text alignment in 3D generations. Additionally, we introduce an LLM-layout initialization that significantly accelerates optimization convergence through semantic-aware spatial configuration. Our framework, CoherenDream, achieves consistent improvement across multiple metrics on TIFA subset.As the first study to incorporate MLLMs into SDS optimization, we also conduct extensive ablation studies to explore optimal MLLM adaptations for 3D generation tasks.", "date": "2025-04-28", "categories": []} {"id": "2507.21809", "title": "HunyuanWorld 1.0: Generating Immersive, Explorable, and Interactive 3D Worlds from Words or Pixels", "abstract": "Creating immersive and playable 3D worlds from texts or images remains a fundamental challenge in computer vision and graphics. Existing world generation approaches typically fall into two categories: video-based methods that offer rich diversity but lack 3D consistency and rendering efficiency, and 3D-based methods that provide geometric consistency but struggle with limited training data and memory-inefficient representations. To address these limitations, we present HunyuanWorld 1.0, a novel framework that combines the best of both worlds for generating immersive, explorable, and interactive 3D scenes from text and image conditions. Our approach features three key advantages: 1) 360{\\deg} immersive experiences via panoramic world proxies; 2) mesh export capabilities for seamless compatibility with existing computer graphics pipelines; 3) disentangled object representations for augmented interactivity. The core of our framework is a semantically layered 3D mesh representation that leverages panoramic images as 360{\\deg} world proxies for semantic-aware world decomposition and reconstruction, enabling the generation of diverse 3D worlds. Extensive experiments demonstrate that our method achieves state-of-the-art performance in generating coherent, explorable, and interactive 3D worlds while enabling versatile applications in virtual reality, physical simulation, game development, and interactive content creation.", "date": "2025-07-29", "categories": []} {"id": "2508.09973", "title": "PERSONA: Personalized Whole-Body 3D Avatar with Pose-Driven Deformations from a Single Image", "abstract": "Two major approaches exist for creating animatable human avatars. The first, a 3D-based approach, optimizes a NeRF- or 3DGS-based avatar from videos of a single person, achieving personalization through a disentangled identity representation. However, modeling pose-driven deformations, such as non-rigid cloth deformations, requires numerous pose-rich videos, which are costly and impractical to capture in daily life. The second, a diffusion-based approach, learns pose-driven deformations from large-scale in-the-wild videos but struggles with identity preservation and pose-dependent identity entanglement. We present PERSONA, a framework that combines the strengths of both approaches to obtain a personalized 3D human avatar with pose-driven deformations from a single image. PERSONA leverages a diffusion-based approach to generate pose-rich videos from the input image and optimizes a 3D avatar based on them. To ensure high authenticity and sharp renderings across diverse poses, we introduce balanced sampling and geometry-weighted optimization. Balanced sampling oversamples the input image to mitigate identity shifts in diffusion-generated training videos. Geometry-weighted optimization prioritizes geometry constraints over image loss, preserving rendering quality in diverse poses.", "date": "2025-08-13", "categories": []} {"id": "2411.16185", "title": "Fancy123: One Image to High-Quality 3D Mesh Generation via Plug-and-Play Deformation", "abstract": "Generating 3D meshes from a single image is an important but ill-posed task. Existing methods mainly adopt 2D multiview diffusion models to generate intermediate multiview images, and use the Large Reconstruction Model (LRM) to create the final meshes. However, the multiview images exhibit local inconsistencies, and the meshes often lack fidelity to the input image or look blurry. We propose Fancy123, featuring two enhancement modules and an unprojection operation to address the above three issues, respectively. The appearance enhancement module deforms the 2D multiview images to realign misaligned pixels for better multiview consistency. The fidelity enhancement module deforms the 3D mesh to match the input image. The unprojection of the input image and deformed multiview images onto LRM's generated mesh ensures high clarity, discarding LRM's predicted blurry-looking mesh colors. Extensive qualitative and quantitative experiments verify Fancy123's SoTA performance with significant improvement. Also, the two enhancement modules are plug-and-play and work at inference time, allowing seamless integration into various existing single-image-to-3D methods. Code at: https://github.com/YuQiao0303/Fancy123", "date": "2024-11-25", "categories": []} {"id": "2508.08821", "title": "3DFroMLLM: 3D Prototype Generation only from Pretrained Multimodal LLMs", "abstract": "Recent Multi-Modal Large Language Models (MLLMs) have demonstrated strong capabilities in learning joint representations from text and images. However, their spatial reasoning remains limited. We introduce 3DFroMLLM, a novel framework that enables the generation of 3D object prototypes directly from MLLMs, including geometry and part labels. Our pipeline is agentic, comprising a designer, coder, and visual inspector operating in a refinement loop. Notably, our approach requires no additional training data or detailed user instructions. Building on prior work in 2D generation, we demonstrate that rendered images produced by our framework can be effectively used for image classification pretraining tasks and outperforms previous methods by 15%. As a compelling real-world use case, we show that the generated prototypes can be leveraged to improve fine-grained vision-language models by using the rendered, part-labeled prototypes to fine-tune CLIP for part segmentation and achieving a 55% accuracy improvement without relying on any additional human-labeled data.", "date": "2025-08-12", "categories": []} {"id": "2508.09062", "title": "VertexRegen: Mesh Generation with Continuous Level of Detail", "abstract": "We introduce VertexRegen, a novel mesh generation framework that enables generation at a continuous level of detail. Existing autoregressive methods generate meshes in a partial-to-complete manner and thus intermediate steps of generation represent incomplete structures. VertexRegen takes inspiration from progressive meshes and reformulates the process as the reversal of edge collapse, i.e. vertex split, learned through a generative model. Experimental results demonstrate that VertexRegen produces meshes of comparable quality to state-of-the-art methods while uniquely offering anytime generation with the flexibility to halt at any step to yield valid meshes with varying levels of detail.", "date": "2025-08-12", "categories": []} {"id": "2412.16919", "title": "TAR3D: Creating High-Quality 3D Assets via Next-Part Prediction", "abstract": "We present TAR3D, a novel framework that consists of a 3D-aware Vector Quantized-Variational AutoEncoder (VQ-VAE) and a Generative Pre-trained Transformer (GPT) to generate high-quality 3D assets. The core insight of this work is to migrate the multimodal unification and promising learning capabilities of the next-token prediction paradigm to conditional 3D object generation. To achieve this, the 3D VQ-VAE first encodes a wide range of 3D shapes into a compact triplane latent space and utilizes a set of discrete representations from a trainable codebook to reconstruct fine-grained geometries under the supervision of query point occupancy. Then, the 3D GPT, equipped with a custom triplane position embedding called TriPE, predicts the codebook index sequence with prefilling prompt tokens in an autoregressive manner so that the composition of 3D geometries can be modeled part by part. Extensive experiments on ShapeNet and Objaverse demonstrate that TAR3D can achieve superior generation quality over existing methods in text-to-3D and image-to-3D tasks", "date": "2024-12-22", "categories": []} {"id": "2503.12929", "title": "AR-1-to-3: Single Image to Consistent 3D Object Generation via Next-View Prediction", "abstract": "Novel view synthesis (NVS) is a cornerstone for image-to-3d creation. However, existing works still struggle to maintain consistency between the generated views and the input views, especially when there is a significant camera pose difference, leading to poor-quality 3D geometries and textures. We attribute this issue to their treatment of all target views with equal priority according to our empirical observation that the target views closer to the input views exhibit higher fidelity. With this inspiration, we propose AR-1-to-3, a novel next-view prediction paradigm based on diffusion models that first generates views close to the input views, which are then utilized as contextual information to progressively synthesize farther views. To encode the generated view subsequences as local and global conditions for the next-view prediction, we accordingly develop a stacked local feature encoding strategy (Stacked-LE) and an LSTM-based global feature encoding strategy (LSTM-GE). Extensive experiments demonstrate that our method significantly improves the consistency between the generated views and the input views, producing high-fidelity 3D assets.", "date": "2025-03-17", "categories": []} {"id": "2504.12684", "title": "SOPHY: Learning to Generate Simulation-Ready Objects with Physical Materials", "abstract": "We present SOPHY, a generative model for 3D physics-aware shape synthesis. Unlike existing 3D generative models that focus solely on static geometry or 4D models that produce physics-agnostic animations, our method jointly synthesizes shape, texture, and material properties related to physics-grounded dynamics, making the generated objects ready for simulations and interactive, dynamic environments. To train our model, we introduce a dataset of 3D objects annotated with detailed physical material attributes, along with an efficient pipeline for material annotation. Our method enables applications such as text-driven generation of interactive, physics-aware 3D objects and single-image reconstruction of physically plausible shapes. Furthermore, our experiments show that jointly modeling shape and material properties enhances the realism and fidelity of the generated shapes, improving performance on both generative geometry and physical plausibility.", "date": "2025-04-17", "categories": []} {"id": "2508.07700", "title": "Make Your MoVe: Make Your 3D Contents by Adapting Multi-View Diffusion Models to External Editing", "abstract": "As 3D generation techniques continue to flourish, the demand for generating personalized content is rapidly rising. Users increasingly seek to apply various editing methods to polish generated 3D content, aiming to enhance its color, style, and lighting without compromising the underlying geometry. However, most existing editing tools focus on the 2D domain, and directly feeding their results into 3D generation methods (like multi-view diffusion models) will introduce information loss, degrading the quality of the final 3D assets. In this paper, we propose a tuning-free, plug-and-play scheme that aligns edited assets with their original geometry in a single inference run. Central to our approach is a geometry preservation module that guides the edited multi-view generation with original input normal latents. Besides, an injection switcher is proposed to deliberately control the supervision extent of the original normals, ensuring the alignment between the edited color and normal views. Extensive experiments show that our method consistently improves both the multi-view consistency and mesh quality of edited 3D assets, across multiple combinations of multi-view diffusion models and editing methods.", "date": "2025-08-11", "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": "2508.08228", "title": "LL3M: Large Language 3D Modelers", "abstract": "We present LL3M, a multi-agent system that leverages pretrained large language models (LLMs) to generate 3D assets by writing interpretable Python code in Blender. We break away from the typical generative approach that learns from a collection of 3D data. Instead, we reformulate shape generation as a code-writing task, enabling greater modularity, editability, and integration with artist workflows. Given a text prompt, LL3M coordinates a team of specialized LLM agents to plan, retrieve, write, debug, and refine Blender scripts that generate and edit geometry and appearance. The generated code works as a high-level, interpretable, human-readable, well-documented representation of scenes and objects, making full use of sophisticated Blender constructs (e.g. B-meshes, geometry modifiers, shader nodes) for diverse, unconstrained shapes, materials, and scenes. This code presents many avenues for further agent and human editing and experimentation via code tweaks or procedural parameters. This medium naturally enables a co-creative loop in our system: agents can automatically self-critique using code and visuals, while iterative user instructions provide an intuitive way to refine assets. A shared code context across agents enables awareness of previous attempts, and a retrieval-augmented generation knowledge base built from Blender API documentation, BlenderRAG, equips agents with examples, types, and functions empowering advanced modeling operations and code correctness. We demonstrate the effectiveness of LL3M across diverse shape categories, style and material edits, and user-driven refinements. Our experiments showcase the power of code as a generative and interpretable medium for 3D asset creation. Our project page is at https://threedle.github.io/ll3m.", "date": "2025-08-11", "categories": []} {"id": "2502.20045", "title": "Text2VDM: Text to Vector Displacement Maps for Expressive and Interactive 3D Sculpting", "abstract": "Professional 3D asset creation often requires diverse sculpting brushes to add surface details and geometric structures. Despite recent progress in 3D generation, producing reusable sculpting brushes compatible with artists' workflows remains an open and challenging problem. These sculpting brushes are typically represented as vector displacement maps (VDMs), which existing models cannot easily generate compared to natural images. This paper presents Text2VDM, a novel framework for text-to-VDM brush generation through the deformation of a dense planar mesh guided by score distillation sampling (SDS). The original SDS loss is designed for generating full objects and struggles with generating desirable sub-object structures from scratch in brush generation. We refer to this issue as semantic coupling, which we address by introducing weighted blending of prompt tokens to SDS, resulting in a more accurate target distribution and semantic guidance. Experiments demonstrate that Text2VDM can generate diverse, high-quality VDM brushes for sculpting surface details and geometric structures. Our generated brushes can be seamlessly integrated into mainstream modeling software, enabling various applications such as mesh stylization and real-time interactive modeling.", "date": "2025-02-27", "categories": []} {"id": "2503.15877", "title": "Repurposing 2D Diffusion Models with Gaussian Atlas for 3D Generation", "abstract": "Recent advances in text-to-image diffusion models have been driven by the increasing availability of paired 2D data. However, the development of 3D diffusion models has been hindered by the scarcity of high-quality 3D data, resulting in less competitive performance compared to their 2D counterparts. To address this challenge, we propose repurposing pre-trained 2D diffusion models for 3D object generation. We introduce Gaussian Atlas, a novel representation that utilizes dense 2D grids, enabling the fine-tuning of 2D diffusion models to generate 3D Gaussians. Our approach demonstrates successful transfer learning from a pre-trained 2D diffusion model to a 2D manifold flattened from 3D structures. To support model training, we compile GaussianVerse, a large-scale dataset comprising 205K high-quality 3D Gaussian fittings of various 3D objects. Our experimental results show that text-to-image diffusion models can be effectively adapted for 3D content generation, bridging the gap between 2D and 3D modeling.", "date": "2025-03-20", "categories": []} {"id": "2504.20830", "title": "CMT: A Cascade MAR with Topology Predictor for Multimodal Conditional CAD Generation", "abstract": "While accurate and user-friendly Computer-Aided Design (CAD) is crucial for industrial design and manufacturing, existing methods still struggle to achieve this due to their over-simplified representations or architectures incapable of supporting multimodal design requirements. In this paper, we attempt to tackle this problem from both methods and datasets aspects. First, we propose a cascade MAR with topology predictor (CMT), the first multimodal framework for CAD generation based on Boundary Representation (B-Rep). Specifically, the cascade MAR can effectively capture the ``edge-counters-surface'' priors that are essential in B-Reps, while the topology predictor directly estimates topology in B-Reps from the compact tokens in MAR. Second, to facilitate large-scale training, we develop a large-scale multimodal CAD dataset, mmABC, which includes over 1.3 million B-Rep models with multimodal annotations, including point clouds, text descriptions, and multi-view images. Extensive experiments show the superior of CMT in both conditional and unconditional CAD generation tasks. For example, we improve Coverage and Valid ratio by +10.68% and +10.3%, respectively, compared to state-of-the-art methods on ABC in unconditional generation. CMT also improves +4.01 Chamfer on image conditioned CAD generation on mmABC.", "date": "2025-04-29", "categories": []} {"id": "2507.23777", "title": "XSpecMesh: Quality-Preserving Auto-Regressive Mesh Generation Acceleration via Multi-Head Speculative Decoding", "abstract": "Current auto-regressive models can generate high-quality, topologically precise meshes; however, they necessitate thousands-or even tens of thousands-of next-token predictions during inference, resulting in substantial latency. We introduce XSpecMesh, a quality-preserving acceleration method for auto-regressive mesh generation models. XSpecMesh employs a lightweight, multi-head speculative decoding scheme to predict multiple tokens in parallel within a single forward pass, thereby accelerating inference. We further propose a verification and resampling strategy: the backbone model verifies each predicted token and resamples any tokens that do not meet the quality criteria. In addition, we propose a distillation strategy that trains the lightweight decoding heads by distilling from the backbone model, encouraging their prediction distributions to align and improving the success rate of speculative predictions. Extensive experiments demonstrate that our method achieves a 1.7x speedup without sacrificing generation quality. Our code will be released.", "date": "2025-07-31", "categories": []} {"id": "2508.03925", "title": "Point-Based Shape Representation Generation with a Correspondence-Preserving Diffusion Model", "abstract": "We propose a diffusion model designed to generate point-based shape representations with correspondences. Traditional statistical shape models have considered point correspondences extensively, but current deep learning methods do not take them into account, focusing on unordered point clouds instead. Current deep generative models for point clouds do not address generating shapes with point correspondences between generated shapes. This work aims to formulate a diffusion model that is capable of generating realistic point-based shape representations, which preserve point correspondences that are present in the training data. Using shape representation data with correspondences derived from Open Access Series of Imaging Studies 3 (OASIS-3), we demonstrate that our correspondence-preserving model effectively generates point-based hippocampal shape representations that are highly realistic compared to existing methods. We further demonstrate the applications of our generative model by downstream tasks, such as conditional generation of healthy and AD subjects and predicting morphological changes of disease progression by counterfactual generation.", "date": "2025-08-05", "categories": []} {"id": "2508.04467", "title": "4DVD: Cascaded Dense-view Video Diffusion Model for High-quality 4D Content Generation", "abstract": "Given the high complexity of directly generating high-dimensional data such as 4D, we present 4DVD, a cascaded video diffusion model that generates 4D content in a decoupled manner. Unlike previous multi-view video methods that directly model 3D space and temporal features simultaneously with stacked cross view/temporal attention modules, 4DVD decouples this into two subtasks: coarse multi-view layout generation and structure-aware conditional generation, and effectively unifies them. Specifically, given a monocular video, 4DVD first predicts the dense view content of its layout with superior cross-view and temporal consistency. Based on the produced layout priors, a structure-aware spatio-temporal generation branch is developed, combining these coarse structural priors with the exquisite appearance content of input monocular video to generate final high-quality dense-view videos. Benefit from this, explicit 4D representation~(such as 4D Gaussian) can be optimized accurately, enabling wider practical application. To train 4DVD, we collect a dynamic 3D object dataset, called D-Objaverse, from the Objaverse benchmark and render 16 videos with 21 frames for each object. Extensive experiments demonstrate our state-of-the-art performance on both novel view synthesis and 4D generation. Our project page is https://4dvd.github.io/", "date": "2025-08-06", "categories": []} {"id": "2411.04954", "title": "CAD-MLLM: Unifying Multimodality-Conditioned CAD Generation With MLLM", "abstract": "This paper aims to design a unified Computer-Aided Design (CAD) generation system that can easily generate CAD models based on the user's inputs in the form of textual description, images, point clouds, or even a combination of them. Towards this goal, we introduce the CAD-MLLM, the first system capable of generating parametric CAD models conditioned on the multimodal input. Specifically, within the CAD-MLLM framework, we leverage the command sequences of CAD models and then employ advanced large language models (LLMs) to align the feature space across these diverse multi-modalities data and CAD models' vectorized representations. To facilitate the model training, we design a comprehensive data construction and annotation pipeline that equips each CAD model with corresponding multimodal data. Our resulting dataset, named Omni-CAD, is the first multimodal CAD dataset that contains textual description, multi-view images, points, and command sequence for each CAD model. It contains approximately 450K instances and their CAD construction sequences. To thoroughly evaluate the quality of our generated CAD models, we go beyond current evaluation metrics that focus on reconstruction quality by introducing additional metrics that assess topology quality and surface enclosure extent. Extensive experimental results demonstrate that CAD-MLLM significantly outperforms existing conditional generative methods and remains highly robust to noises and missing points. The project page and more visualizations can be found at: https://cad-mllm.github.io/", "date": "2024-11-07", "categories": []} {"id": "2508.01242", "title": "MeshLLM: Empowering Large Language Models to Progressively Understand and Generate 3D Mesh", "abstract": "We present MeshLLM, a novel framework that leverages large language models (LLMs) to understand and generate text-serialized 3D meshes. Our approach addresses key limitations in existing methods, including the limited dataset scale when catering to LLMs' token length and the loss of 3D structural information during mesh serialization. We introduce a Primitive-Mesh decomposition strategy, which divides 3D meshes into structurally meaningful subunits. This enables the creation of a large-scale dataset with 1500k+ samples, almost 50 times larger than previous methods, which aligns better with the LLM scaling law principles. Furthermore, we propose inferring face connectivity from vertices and local mesh assembly training strategies, significantly enhancing the LLMs' ability to capture mesh topology and spatial structures. Experiments show that MeshLLM outperforms the state-of-the-art LLaMA-Mesh in both mesh generation quality and shape understanding, highlighting its great potential in processing text-serialized 3D meshes.", "date": "2025-08-02", "categories": []} {"id": "2508.03691", "title": "La La LiDAR: Large-Scale Layout Generation from LiDAR Data", "abstract": "Controllable generation of realistic LiDAR scenes is crucial for applications such as autonomous driving and robotics. While recent diffusion-based models achieve high-fidelity LiDAR generation, they lack explicit control over foreground objects and spatial relationships, limiting their usefulness for scenario simulation and safety validation. To address these limitations, we propose Large-scale Layout-guided LiDAR generation model (\"La La LiDAR\"), a novel layout-guided generative framework that introduces semantic-enhanced scene graph diffusion with relation-aware contextual conditioning for structured LiDAR layout generation, followed by foreground-aware control injection for complete scene generation. This enables customizable control over object placement while ensuring spatial and semantic consistency. To support our structured LiDAR generation, we introduce Waymo-SG and nuScenes-SG, two large-scale LiDAR scene graph datasets, along with new evaluation metrics for layout synthesis. Extensive experiments demonstrate that La La LiDAR achieves state-of-the-art performance in both LiDAR generation and downstream perception tasks, establishing a new benchmark for controllable 3D scene generation.", "date": "2025-08-05", "categories": []} {"id": "2502.06957", "title": "GAS: Generative Avatar Synthesis from a Single Image", "abstract": "We present a unified and generalizable framework for synthesizing view-consistent and temporally coherent avatars from a single image, addressing the challenging task of single-image avatar generation. Existing diffusion-based methods often condition on sparse human templates (e.g., depth or normal maps), which leads to multi-view and temporal inconsistencies due to the mismatch between these signals and the true appearance of the subject. Our approach bridges this gap by combining the reconstruction power of regression-based 3D human reconstruction with the generative capabilities of a diffusion model. In a first step, an initial 3D reconstructed human through a generalized NeRF provides comprehensive conditioning, ensuring high-quality synthesis faithful to the reference appearance and structure. Subsequently, the derived geometry and appearance from the generalized NeRF serve as input to a video-based diffusion model. This strategic integration is pivotal for enforcing both multi-view and temporal consistency throughout the avatar's generation. Empirical results underscore the superior generalization ability of our proposed method, demonstrating its effectiveness across diverse in-domain and out-of-domain in-the-wild datasets.", "date": "2025-02-10", "categories": []} {"id": "2504.02817", "title": "Efficient Autoregressive Shape Generation via Octree-Based Adaptive Tokenization", "abstract": "Many 3D generative models rely on variational autoencoders (VAEs) to learn compact shape representations. However, existing methods encode all shapes into a fixed-size token, disregarding the inherent variations in scale and complexity across 3D data. This leads to inefficient latent representations that can compromise downstream generation. We address this challenge by introducing Octree-based Adaptive Tokenization, a novel framework that adjusts the dimension of latent representations according to shape complexity. Our approach constructs an adaptive octree structure guided by a quadric-error-based subdivision criterion and allocates a shape latent vector to each octree cell using a query-based transformer. Building upon this tokenization, we develop an octree-based autoregressive generative model that effectively leverages these variable-sized representations in shape generation. Extensive experiments demonstrate that our approach reduces token counts by 50% compared to fixed-size methods while maintaining comparable visual quality. When using a similar token length, our method produces significantly higher-quality shapes. When incorporated with our downstream generative model, our method creates more detailed and diverse 3D content than existing approaches.", "date": "2025-04-03", "categories": []} {"id": "2507.03257", "title": "LACONIC: A 3D Layout Adapter for Controllable Image Creation", "abstract": "Existing generative approaches for guided image synthesis of multi-object scenes typically rely on 2D controls in the image or text space. As a result, these methods struggle to maintain and respect consistent three-dimensional geometric structure, underlying the scene. In this paper, we propose a novel conditioning approach, training method and adapter network that can be plugged into pretrained text-to-image diffusion models. Our approach provides a way to endow such models with 3D-awareness, while leveraging their rich prior knowledge. Our method supports camera control, conditioning on explicit 3D geometries and, for the first time, accounts for the entire context of a scene, i.e., both on and off-screen items, to synthesize plausible and semantically rich images. Despite its multi-modal nature, our model is lightweight, requires a reasonable number of data for supervised learning and shows remarkable generalization power. We also introduce methods for intuitive and consistent image editing and restyling, e.g., by positioning, rotating or resizing individual objects in a scene. Our method integrates well within various image creation workflows and enables a richer set of applications compared to previous approaches.", "date": "2025-07-04", "categories": []} {"id": "2507.04183", "title": "Voyaging into Perpetual Dynamic Scenes from a Single View", "abstract": "The problem of generating a perpetual dynamic scene from a single view is an important problem with widespread applications in augmented and virtual reality, and robotics. However, since dynamic scenes regularly change over time, a key challenge is to ensure that different generated views be consistent with the underlying 3D motions. Prior work learns such consistency by training on multiple views, but the generated scene regions often interpolate between training views and fail to generate perpetual views. To address this issue, we propose DynamicVoyager, which reformulates dynamic scene generation as a scene outpainting problem with new dynamic content. As 2D outpainting models struggle at generating 3D consistent motions from a single 2D view, we enrich 2D pixels with information from their 3D rays that facilitates learning of 3D motion consistency. More specifically, we first map the single-view video input to a dynamic point cloud using the estimated video depths. We then render a partial video of the point cloud from a novel view and outpaint the missing regions using ray information (e.g., the distance from a ray to the point cloud) to generate 3D consistent motions. Next, we use the outpainted video to update the point cloud, which is used for outpainting the scene from future novel views. Moreover, we can control the generated content with the input text prompt. Experiments show that our model can generate perpetual scenes with consistent motions along fly-through cameras. Project page: https://tianfr.github.io/DynamicVoyager.", "date": "2025-07-05", "categories": []} {"id": "2507.05256", "title": "SegmentDreamer: Towards High-fidelity Text-to-3D Synthesis with Segmented Consistency Trajectory Distillation", "abstract": "Recent advancements in text-to-3D generation improve the visual quality of Score Distillation Sampling (SDS) and its variants by directly connecting Consistency Distillation (CD) to score distillation. However, due to the imbalance between self-consistency and cross-consistency, these CD-based methods inherently suffer from improper conditional guidance, leading to sub-optimal generation results. To address this issue, we present SegmentDreamer, a novel framework designed to fully unleash the potential of consistency models for high-fidelity text-to-3D generation. Specifically, we reformulate SDS through the proposed Segmented Consistency Trajectory Distillation (SCTD), effectively mitigating the imbalance issues by explicitly defining the relationship between self- and cross-consistency. Moreover, SCTD partitions the Probability Flow Ordinary Differential Equation (PF-ODE) trajectory into multiple sub-trajectories and ensures consistency within each segment, which can theoretically provide a significantly tighter upper bound on distillation error. Additionally, we propose a distillation pipeline for a more swift and stable generation. Extensive experiments demonstrate that our SegmentDreamer outperforms state-of-the-art methods in visual quality, enabling high-fidelity 3D asset creation through 3D Gaussian Splatting (3DGS).", "date": "2025-07-07", "categories": []} {"id": "2508.01464", "title": "Can3Tok: Canonical 3D Tokenization and Latent Modeling of Scene-Level 3D Gaussians", "abstract": "3D generation has made significant progress, however, it still largely remains at the object-level. Feedforward 3D scene-level generation has been rarely explored due to the lack of models capable of scaling-up latent representation learning on 3D scene-level data. Unlike object-level generative models, which are trained on well-labeled 3D data in a bounded canonical space, scene-level generations with 3D scenes represented by 3D Gaussian Splatting (3DGS) are unbounded and exhibit scale inconsistency across different scenes, making unified latent representation learning for generative purposes extremely challenging. In this paper, we introduce Can3Tok, the first 3D scene-level variational autoencoder (VAE) capable of encoding a large number of Gaussian primitives into a low-dimensional latent embedding, which effectively captures both semantic and spatial information of the inputs. Beyond model design, we propose a general pipeline for 3D scene data processing to address scale inconsistency issue. We validate our method on the recent scene-level 3D dataset DL3DV-10K, where we found that only Can3Tok successfully generalizes to novel 3D scenes, while compared methods fail to converge on even a few hundred scene inputs during training and exhibit zero generalization ability during inference. Finally, we demonstrate image-to-3DGS and text-to-3DGS generation as our applications to demonstrate its ability to facilitate downstream generation tasks.", "date": "2025-08-02", "categories": []} {"id": "2508.01650", "title": "StrandDesigner: Towards Practical Strand Generation with Sketch Guidance", "abstract": "Realistic hair strand generation is crucial for applications like computer graphics and virtual reality. While diffusion models can generate hairstyles from text or images, these inputs lack precision and user-friendliness. Instead, we propose the first sketch-based strand generation model, which offers finer control while remaining user-friendly. Our framework tackles key challenges, such as modeling complex strand interactions and diverse sketch patterns, through two main innovations: a learnable strand upsampling strategy that encodes 3D strands into multi-scale latent spaces, and a multi-scale adaptive conditioning mechanism using a transformer with diffusion heads to ensure consistency across granularity levels. Experiments on several benchmark datasets show our method outperforms existing approaches in realism and precision. Qualitative results further confirm its effectiveness. Code will be released at [GitHub](https://github.com/fighting-Zhang/StrandDesigner).", "date": "2025-08-03", "categories": []} {"id": "2508.01684", "title": "DisCo3D: Distilling Multi-View Consistency for 3D Scene Editing", "abstract": "While diffusion models have demonstrated remarkable progress in 2D image generation and editing, extending these capabilities to 3D editing remains challenging, particularly in maintaining multi-view consistency. Classical approaches typically update 3D representations through iterative refinement based on a single editing view. However, these methods often suffer from slow convergence and blurry artifacts caused by cross-view inconsistencies. Recent methods improve efficiency by propagating 2D editing attention features, yet still exhibit fine-grained inconsistencies and failure modes in complex scenes due to insufficient constraints. To address this, we propose \\textbf{DisCo3D}, a novel framework that distills 3D consistency priors into a 2D editor. Our method first fine-tunes a 3D generator using multi-view inputs for scene adaptation, then trains a 2D editor through consistency distillation. The edited multi-view outputs are finally optimized into 3D representations via Gaussian Splatting. Experimental results show DisCo3D achieves stable multi-view consistency and outperforms state-of-the-art methods in editing quality.", "date": "2025-08-03", "categories": []} {"id": "2508.00558", "title": "Guiding Diffusion-Based Articulated Object Generation by Partial Point Cloud Alignment and Physical Plausibility Constraints", "abstract": "Articulated objects are an important type of interactable objects in everyday environments. In this paper, we propose PhysNAP, a novel diffusion model-based approach for generating articulated objects that aligns them with partial point clouds and improves their physical plausibility. The model represents part shapes by signed distance functions (SDFs). We guide the reverse diffusion process using a point cloud alignment loss computed using the predicted SDFs. Additionally, we impose non-penetration and mobility constraints based on the part SDFs for guiding the model to generate more physically plausible objects. We also make our diffusion approach category-aware to further improve point cloud alignment if category information is available. We evaluate the generative ability and constraint consistency of samples generated with PhysNAP using the PartNet-Mobility dataset. We also compare it with an unguided baseline diffusion model and demonstrate that PhysNAP can improve constraint consistency and provides a tradeoff with generative ability.", "date": "2025-08-01", "categories": []} {"id": "2507.17745", "title": "Ultra3D: Efficient and High-Fidelity 3D Generation with Part Attention", "abstract": "Recent advances in sparse voxel representations have significantly improved the quality of 3D content generation, enabling high-resolution modeling with fine-grained geometry. However, existing frameworks suffer from severe computational inefficiencies due to the quadratic complexity of attention mechanisms in their two-stage diffusion pipelines. In this work, we propose Ultra3D, an efficient 3D generation framework that significantly accelerates sparse voxel modeling without compromising quality. Our method leverages the compact VecSet representation to efficiently generate a coarse object layout in the first stage, reducing token count and accelerating voxel coordinate prediction. To refine per-voxel latent features in the second stage, we introduce Part Attention, a geometry-aware localized attention mechanism that restricts attention computation within semantically consistent part regions. This design preserves structural continuity while avoiding unnecessary global attention, achieving up to 6.7x speed-up in latent generation. To support this mechanism, we construct a scalable part annotation pipeline that converts raw meshes into part-labeled sparse voxels. Extensive experiments demonstrate that Ultra3D supports high-resolution 3D generation at 1024 resolution and achieves state-of-the-art performance in both visual fidelity and user preference.", "date": "2025-07-23", "categories": []} {"id": "2507.18371", "title": "MVG4D: Image Matrix-Based Multi-View and Motion Generation for 4D Content Creation from a Single Image", "abstract": "Advances in generative modeling have significantly enhanced digital content creation, extending from 2D images to complex 3D and 4D scenes. Despite substantial progress, producing high-fidelity and temporally consistent dynamic 4D content remains a challenge. In this paper, we propose MVG4D, a novel framework that generates dynamic 4D content from a single still image by combining multi-view synthesis with 4D Gaussian Splatting (4D GS). At its core, MVG4D employs an image matrix module that synthesizes temporally coherent and spatially diverse multi-view images, providing rich supervisory signals for downstream 3D and 4D reconstruction. These multi-view images are used to optimize a 3D Gaussian point cloud, which is further extended into the temporal domain via a lightweight deformation network. Our method effectively enhances temporal consistency, geometric fidelity, and visual realism, addressing key challenges in motion discontinuity and background degradation that affect prior 4D GS-based methods. Extensive experiments on the Objaverse dataset demonstrate that MVG4D outperforms state-of-the-art baselines in CLIP-I, PSNR, FVD, and time efficiency. Notably, it reduces flickering artifacts and sharpens structural details across views and time, enabling more immersive AR/VR experiences. MVG4D sets a new direction for efficient and controllable 4D generation from minimal inputs.", "date": "2025-07-24", "categories": []} {"id": "2507.23785", "title": "Gaussian Variation Field Diffusion for High-fidelity Video-to-4D Synthesis", "abstract": "In this paper, we present a novel framework for video-to-4D generation that creates high-quality dynamic 3D content from single video inputs. Direct 4D diffusion modeling is extremely challenging due to costly data construction and the high-dimensional nature of jointly representing 3D shape, appearance, and motion. We address these challenges by introducing a Direct 4DMesh-to-GS Variation Field VAE that directly encodes canonical Gaussian Splats (GS) and their temporal variations from 3D animation data without per-instance fitting, and compresses high-dimensional animations into a compact latent space. Building upon this efficient representation, we train a Gaussian Variation Field diffusion model with temporal-aware Diffusion Transformer conditioned on input videos and canonical GS. Trained on carefully-curated animatable 3D objects from the Objaverse dataset, our model demonstrates superior generation quality compared to existing methods. It also exhibits remarkable generalization to in-the-wild video inputs despite being trained exclusively on synthetic data, paving the way for generating high-quality animated 3D content. Project page: https://gvfdiffusion.github.io/.", "date": "2025-07-31", "categories": []} {"id": "2503.10289", "title": "MaterialMVP: Illumination-Invariant Material Generation via Multi-view PBR Diffusion", "abstract": "Physically-based rendering (PBR) has become a cornerstone in modern computer graphics, enabling realistic material representation and lighting interactions in 3D scenes. In this paper, we present MaterialMVP, a novel end-to-end model for generating PBR textures from 3D meshes and image prompts, addressing key challenges in multi-view material synthesis. Our approach leverages Reference Attention to extract and encode informative latent from the input reference images, enabling intuitive and controllable texture generation. We also introduce a Consistency-Regularized Training strategy to enforce stability across varying viewpoints and illumination conditions, ensuring illumination-invariant and geometrically consistent results. Additionally, we propose Dual-Channel Material Generation, which separately optimizes albedo and metallic-roughness (MR) textures while maintaining precise spatial alignment with the input images through Multi-Channel Aligned Attention. Learnable material embeddings are further integrated to capture the distinct properties of albedo and MR. Experimental results demonstrate that our model generates PBR textures with realistic behavior across diverse lighting scenarios, outperforming existing methods in both consistency and quality for scalable 3D asset creation.", "date": "2025-03-13", "categories": []} {"id": "2503.12024", "title": "SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering", "abstract": "Recent progress in 3D/4D scene generation emphasizes the importance of physical alignment throughout video generation and scene reconstruction. However, existing methods improve the alignment separately at each stage, making it difficult to manage subtle misalignments arising from another stage. Here, we present SteerX, a zero-shot inference-time steering method that unifies scene reconstruction into the generation process, tilting data distributions toward better geometric alignment. To this end, we introduce two geometric reward functions for 3D/4D scene generation by using pose-free feed-forward scene reconstruction models. Through extensive experiments, we demonstrate the effectiveness of SteerX in improving 3D/4D scene generation.", "date": "2025-03-15", "categories": []} {"id": "2507.13985", "title": "DreamScene: 3D Gaussian-based End-to-end Text-to-3D Scene Generation", "abstract": "Generating 3D scenes from natural language holds great promise for applications in gaming, film, and design. However, existing methods struggle with automation, 3D consistency, and fine-grained control. We present DreamScene, an end-to-end framework for high-quality and editable 3D scene generation from text or dialogue. DreamScene begins with a scene planning module, where a GPT-4 agent infers object semantics and spatial constraints to construct a hybrid graph. A graph-based placement algorithm then produces a structured, collision-free layout. Based on this layout, Formation Pattern Sampling (FPS) generates object geometry using multi-timestep sampling and reconstructive optimization, enabling fast and realistic synthesis. To ensure global consistent, DreamScene employs a progressive camera sampling strategy tailored to both indoor and outdoor settings. Finally, the system supports fine-grained scene editing, including object movement, appearance changes, and 4D dynamic motion. Experiments demonstrate that DreamScene surpasses prior methods in quality, consistency, and flexibility, offering a practical solution for open-domain 3D content creation. Code and demos are available at https://jahnsonblack.github.io/DreamScene-Full/.", "date": "2025-07-18", "categories": []} {"id": "2412.00623", "title": "A Lesson in Splats: Teacher-Guided Diffusion for 3D Gaussian Splats Generation with 2D Supervision", "abstract": "We present a novel framework for training 3D image-conditioned diffusion models using only 2D supervision. Recovering 3D structure from 2D images is inherently ill-posed due to the ambiguity of possible reconstructions, making generative models a natural choice. However, most existing 3D generative models rely on full 3D supervision, which is impractical due to the scarcity of large-scale 3D datasets. To address this, we propose leveraging sparse-view supervision as a scalable alternative. While recent reconstruction models use sparse-view supervision with differentiable rendering to lift 2D images to 3D, they are predominantly deterministic, failing to capture the diverse set of plausible solutions and producing blurry predictions in uncertain regions. A key challenge in training 3D diffusion models with 2D supervision is that the standard training paradigm requires both the denoising process and supervision to be in the same modality. We address this by decoupling the noisy samples being denoised from the supervision signal, allowing the former to remain in 3D while the latter is provided in 2D. Our approach leverages suboptimal predictions from a deterministic image-to-3D model-acting as a \"teacher\"-to generate noisy 3D inputs, enabling effective 3D diffusion training without requiring full 3D ground truth. We validate our framework on both object-level and scene-level datasets, using two different 3D Gaussian Splat (3DGS) teachers. Our results show that our approach consistently improves upon these deterministic teachers, demonstrating its effectiveness in scalable and high-fidelity 3D generative modeling. See our project page at https://lesson-in-splats.github.io/", "date": "2024-12-01", "categories": []} {"id": "2412.06334", "title": "TriDi: Trilateral Diffusion of 3D Humans, Objects, and Interactions", "abstract": "Modeling 3D human-object interaction (HOI) is a problem of great interest for computer vision and a key enabler for virtual and mixed-reality applications. Existing methods work in a one-way direction: some recover plausible human interactions conditioned on a 3D object; others recover the object pose conditioned on a human pose. Instead, we provide the first unified model - TriDi which works in any direction. Concretely, we generate Human, Object, and Interaction modalities simultaneously with a new three-way diffusion process, allowing to model seven distributions with one network. We implement TriDi as a transformer attending to the various modalities' tokens, thereby discovering conditional relations between them. The user can control the interaction either as a text description of HOI or a contact map. We embed these two representations into a shared latent space, combining the practicality of text descriptions with the expressiveness of contact maps. Using a single network, TriDi unifies all the special cases of prior work and extends to new ones, modeling a family of seven distributions. Remarkably, despite using a single model, TriDi generated samples surpass one-way specialized baselines on GRAB and BEHAVE in terms of both qualitative and quantitative metrics, and demonstrating better diversity. We show the applicability of TriDi to scene population, generating objects for human-contact datasets, and generalization to unseen object geometry. The project page is available at: https://virtualhumans.mpi-inf.mpg.de/tridi.", "date": "2024-12-09", "categories": []} {"id": "2503.08737", "title": "Representing 3D Shapes With 64 Latent Vectors for 3D Diffusion Models", "abstract": "Constructing a compressed latent space through a variational autoencoder (VAE) is the key for efficient 3D diffusion models. This paper introduces COD-VAE that encodes 3D shapes into a COmpact set of 1D latent vectors without sacrificing quality. COD-VAE introduces a two-stage autoencoder scheme to improve compression and decoding efficiency. First, our encoder block progressively compresses point clouds into compact latent vectors via intermediate point patches. Second, our triplane-based decoder reconstructs dense triplanes from latent vectors instead of directly decoding neural fields, significantly reducing computational overhead of neural fields decoding. Finally, we propose uncertainty-guided token pruning, which allocates resources adaptively by skipping computations in simpler regions and improves the decoder efficiency. Experimental results demonstrate that COD-VAE achieves 16x compression compared to the baseline while maintaining quality. This enables 20.8x speedup in generation, highlighting that a large number of latent vectors is not a prerequisite for high-quality reconstruction and generation. The code is available at https://github.com/join16/COD-VAE.", "date": "2025-03-11", "categories": []} {"id": "2503.21745", "title": "3DGen-Bench: Comprehensive Benchmark Suite for 3D Generative Models", "abstract": "3D generation is experiencing rapid advancements, while the development of 3D evaluation has not kept pace. How to keep automatic evaluation equitably aligned with human perception has become a well-recognized challenge. Recent advances in the field of language and image generation have explored human preferences and showcased respectable fitting ability. However, the 3D domain still lacks such a comprehensive preference dataset over generative models. To mitigate this absence, we develop 3DGen-Arena, an integrated platform in a battle manner. Then, we carefully design diverse text and image prompts and leverage the arena platform to gather human preferences from both public users and expert annotators, resulting in a large-scale multi-dimension human preference dataset 3DGen-Bench. Using this dataset, we further train a CLIP-based scoring model, 3DGen-Score, and a MLLM-based automatic evaluator, 3DGen-Eval. These two models innovatively unify the quality evaluation of text-to-3D and image-to-3D generation, and jointly form our automated evaluation system with their respective strengths. Extensive experiments demonstrate the efficacy of our scoring model in predicting human preferences, exhibiting a superior correlation with human ranks compared to existing metrics. We believe that our 3DGen-Bench dataset and automated evaluation system will foster a more equitable evaluation in the field of 3D generation, further promoting the development of 3D generative models and their downstream applications. Project page is available at https://zyh482.github.io/3DGen-Bench/.", "date": "2025-03-27", "categories": []} {"id": "2405.14475", "title": "MagicDrive3D: Controllable 3D Generation for Any-View Rendering in Street Scenes", "abstract": "Controllable generative models for images and videos have seen significant success, yet 3D scene generation, especially in unbounded scenarios like autonomous driving, remains underdeveloped. Existing methods lack flexible controllability and often rely on dense view data collection in controlled environments, limiting their generalizability across common datasets (e.g., nuScenes). In this paper, we introduce MagicDrive3D, a novel framework for controllable 3D street scene generation that combines video-based view synthesis with 3D representation (3DGS) generation. It supports multi-condition control, including road maps, 3D objects, and text descriptions. Unlike previous approaches that require 3D representation before training, MagicDrive3D first trains a multi-view video generation model to synthesize diverse street views. This method utilizes routinely collected autonomous driving data, reducing data acquisition challenges and enriching 3D scene generation. In the 3DGS generation step, we introduce Fault-Tolerant Gaussian Splatting to address minor errors and use monocular depth for better initialization, alongside appearance modeling to manage exposure discrepancies across viewpoints. Experiments show that MagicDrive3D generates diverse, high-quality 3D driving scenes, supports any-view rendering, and enhances downstream tasks like BEV segmentation, demonstrating its potential for autonomous driving simulation and beyond.", "date": "2024-05-23", "categories": []} {"id": "2507.19058", "title": "ScenePainter: Semantically Consistent Perpetual 3D Scene Generation with Concept Relation Alignment", "abstract": "Perpetual 3D scene generation aims to produce long-range and coherent 3D view sequences, which is applicable for long-term video synthesis and 3D scene reconstruction. Existing methods follow a \"navigate-and-imagine\" fashion and rely on outpainting for successive view expansion. However, the generated view sequences suffer from semantic drift issue derived from the accumulated deviation of the outpainting module. To tackle this challenge, we propose ScenePainter, a new framework for semantically consistent 3D scene generation, which aligns the outpainter's scene-specific prior with the comprehension of the current scene. To be specific, we introduce a hierarchical graph structure dubbed SceneConceptGraph to construct relations among multi-level scene concepts, which directs the outpainter for consistent novel views and can be dynamically refined to enhance diversity. Extensive experiments demonstrate that our framework overcomes the semantic drift issue and generates more consistent and immersive 3D view sequences. Project Page: https://xiac20.github.io/ScenePainter/.", "date": "2025-07-25", "categories": []} {"id": "2507.19481", "title": "HairCUP: Hair Compositional Universal Prior for 3D Gaussian Avatars", "abstract": "We present a universal prior model for 3D head avatars with explicit hair compositionality. Existing approaches to build generalizable priors for 3D head avatars often adopt a holistic modeling approach, treating the face and hair as an inseparable entity. This overlooks the inherent compositionality of the human head, making it difficult for the model to naturally disentangle face and hair representations, especially when the dataset is limited. Furthermore, such holistic models struggle to support applications like 3D face and hairstyle swapping in a flexible and controllable manner. To address these challenges, we introduce a prior model that explicitly accounts for the compositionality of face and hair, learning their latent spaces separately. A key enabler of this approach is our synthetic hairless data creation pipeline, which removes hair from studio-captured datasets using estimated hairless geometry and texture derived from a diffusion prior. By leveraging a paired dataset of hair and hairless captures, we train disentangled prior models for face and hair, incorporating compositionality as an inductive bias to facilitate effective separation. Our model's inherent compositionality enables seamless transfer of face and hair components between avatars while preserving identity. Additionally, we demonstrate that our model can be fine-tuned in a few-shot manner using monocular captures to create high-fidelity, hair-compositional 3D head avatars for unseen subjects. These capabilities highlight the practical applicability of our approach in real-world scenarios, paving the way for flexible and expressive 3D avatar generation.", "date": "2025-07-25", "categories": []} {"id": "2507.16535", "title": "EarthCrafter: Scalable 3D Earth Generation via Dual-Sparse Latent Diffusion", "abstract": "Despite the remarkable developments achieved by recent 3D generation works, scaling these methods to geographic extents, such as modeling thousands of square kilometers of Earth's surface, remains an open challenge. We address this through a dual innovation in data infrastructure and model architecture. First, we introduce Aerial-Earth3D, the largest 3D aerial dataset to date, consisting of 50k curated scenes (each measuring 600m x 600m) captured across the U.S. mainland, comprising 45M multi-view Google Earth frames. Each scene provides pose-annotated multi-view images, depth maps, normals, semantic segmentation, and camera poses, with explicit quality control to ensure terrain diversity. Building on this foundation, we propose EarthCrafter, a tailored framework for large-scale 3D Earth generation via sparse-decoupled latent diffusion. Our architecture separates structural and textural generation: 1) Dual sparse 3D-VAEs compress high-resolution geometric voxels and textural 2D Gaussian Splats (2DGS) into compact latent spaces, largely alleviating the costly computation suffering from vast geographic scales while preserving critical information. 2) We propose condition-aware flow matching models trained on mixed inputs (semantics, images, or neither) to flexibly model latent geometry and texture features independently. Extensive experiments demonstrate that EarthCrafter performs substantially better in extremely large-scale generation. The framework further supports versatile applications, from semantic-guided urban layout generation to unconditional terrain synthesis, while maintaining geographic plausibility through our rich data priors from Aerial-Earth3D. Our project page is available at https://whiteinblue.github.io/earthcrafter/", "date": "2025-07-22", "categories": []} {"id": "2503.15475", "title": "Cube: A Roblox View of 3D Intelligence", "abstract": "Foundation models trained on vast amounts of data have demonstrated remarkable reasoning and generation capabilities in the domains of text, images, audio and video. Our goal at Roblox is to build such a foundation model for 3D intelligence, a model that can support developers in producing all aspects of a Roblox experience, from generating 3D objects and scenes to rigging characters for animation to producing programmatic scripts describing object behaviors. We discuss three key design requirements for such a 3D foundation model and then present our first step towards building such a model. We expect that 3D geometric shapes will be a core data type and describe our solution for 3D shape tokenizer. We show how our tokenization scheme can be used in applications for text-to-shape generation, shape-to-text generation and text-to-scene generation. We demonstrate how these applications can collaborate with existing large language models (LLMs) to perform scene analysis and reasoning. We conclude with a discussion outlining our path to building a fully unified foundation model for 3D intelligence.", "date": "2025-03-19", "categories": []} {"id": "2507.13346", "title": "AutoPartGen: Autogressive 3D Part Generation and Discovery", "abstract": "We introduce AutoPartGen, a model that generates objects composed of 3D parts in an autoregressive manner. This model can take as input an image of an object, 2D masks of the object's parts, or an existing 3D object, and generate a corresponding compositional 3D reconstruction. Our approach builds upon 3DShape2VecSet, a recent latent 3D representation with powerful geometric expressiveness. We observe that this latent space exhibits strong compositional properties, making it particularly well-suited for part-based generation tasks. Specifically, AutoPartGen generates object parts autoregressively, predicting one part at a time while conditioning on previously generated parts and additional inputs, such as 2D images, masks, or 3D objects. This process continues until the model decides that all parts have been generated, thus determining automatically the type and number of parts. The resulting parts can be seamlessly assembled into coherent objects or scenes without requiring additional optimization. We evaluate both the overall 3D generation capabilities and the part-level generation quality of AutoPartGen, demonstrating that it achieves state-of-the-art performance in 3D part generation.", "date": "2025-07-17", "categories": []} {"id": "2507.15212", "title": "MeshMamba: State Space Models for Articulated 3D Mesh Generation and Reconstruction", "abstract": "In this paper, we introduce MeshMamba, a neural network model for learning 3D articulated mesh models by employing the recently proposed Mamba State Space Models (Mamba-SSMs). MeshMamba is efficient and scalable in handling a large number of input tokens, enabling the generation and reconstruction of body mesh models with more than 10,000 vertices, capturing clothing and hand geometries. The key to effectively learning MeshMamba is the serialization technique of mesh vertices into orderings that are easily processed by Mamba. This is achieved by sorting the vertices based on body part annotations or the 3D vertex locations of a template mesh, such that the ordering respects the structure of articulated shapes. Based on MeshMamba, we design 1) MambaDiff3D, a denoising diffusion model for generating 3D articulated meshes and 2) Mamba-HMR, a 3D human mesh recovery model that reconstructs a human body shape and pose from a single image. Experimental results showed that MambaDiff3D can generate dense 3D human meshes in clothes, with grasping hands, etc., and outperforms previous approaches in the 3D human shape generation task. Additionally, Mamba-HMR extends the capabilities of previous non-parametric human mesh recovery approaches, which were limited to handling body-only poses using around 500 vertex tokens, to the whole-body setting with face and hands, while achieving competitive performance in (near) real-time.", "date": "2025-07-21", "categories": []} {"id": "2507.15399", "title": "Blended Point Cloud Diffusion for Localized Text-guided Shape Editing", "abstract": "Natural language offers a highly intuitive interface for enabling localized fine-grained edits of 3D shapes. However, prior works face challenges in preserving global coherence while locally modifying the input 3D shape. In this work, we introduce an inpainting-based framework for editing shapes represented as point clouds. Our approach leverages foundation 3D diffusion models for achieving localized shape edits, adding structural guidance in the form of a partial conditional shape, ensuring that other regions correctly preserve the shape's identity. Furthermore, to encourage identity preservation also within the local edited region, we propose an inference-time coordinate blending algorithm which balances reconstruction of the full shape with inpainting at a progression of noise levels during the inference process. Our coordinate blending algorithm seamlessly blends the original shape with its edited version, enabling a fine-grained editing of 3D shapes, all while circumventing the need for computationally expensive and often inaccurate inversion. Extensive experiments show that our method outperforms alternative techniques across a wide range of metrics that evaluate both fidelity to the original shape and also adherence to the textual description.", "date": "2025-07-21", "categories": []} {"id": "2507.15606", "title": "CylinderPlane: Nested Cylinder Representation for 3D-aware Image Generation", "abstract": "While the proposal of the Tri-plane representation has advanced the development of the 3D-aware image generative models, problems rooted in its inherent structure, such as multi-face artifacts caused by sharing the same features in symmetric regions, limit its ability to generate 360$^\\circ$ view images. In this paper, we propose CylinderPlane, a novel implicit representation based on Cylindrical Coordinate System, to eliminate the feature ambiguity issue and ensure multi-view consistency in 360$^\\circ$. Different from the inevitable feature entanglement in Cartesian coordinate-based Tri-plane representation, the cylindrical coordinate system explicitly separates features at different angles, allowing our cylindrical representation possible to achieve high-quality, artifacts-free 360$^\\circ$ image synthesis. We further introduce the nested cylinder representation that composites multiple cylinders at different scales, thereby enabling the model more adaptable to complex geometry and varying resolutions. The combination of cylinders with different resolutions can effectively capture more critical locations and multi-scale features, greatly facilitates fine detail learning and robustness to different resolutions. Moreover, our representation is agnostic to implicit rendering methods and can be easily integrated into any neural rendering pipeline. Extensive experiments on both synthetic dataset and unstructured in-the-wild images demonstrate that our proposed representation achieves superior performance over previous methods.", "date": "2025-07-21", "categories": []} {"id": "2412.03558", "title": "MIDI: Multi-Instance Diffusion for Single Image to 3D Scene Generation", "abstract": "This paper introduces MIDI, a novel paradigm for compositional 3D scene generation from a single image. Unlike existing methods that rely on reconstruction or retrieval techniques or recent approaches that employ multi-stage object-by-object generation, MIDI extends pre-trained image-to-3D object generation models to multi-instance diffusion models, enabling the simultaneous generation of multiple 3D instances with accurate spatial relationships and high generalizability. At its core, MIDI incorporates a novel multi-instance attention mechanism, that effectively captures inter-object interactions and spatial coherence directly within the generation process, without the need for complex multi-step processes. The method utilizes partial object images and global scene context as inputs, directly modeling object completion during 3D generation. During training, we effectively supervise the interactions between 3D instances using a limited amount of scene-level data, while incorporating single-object data for regularization, thereby maintaining the pre-trained generalization ability. MIDI demonstrates state-of-the-art performance in image-to-scene generation, validated through evaluations on synthetic data, real-world scene data, and stylized scene images generated by text-to-image diffusion models.", "date": "2024-12-04", "categories": []} {"id": "2506.18527", "title": "Auto-Regressively Generating Multi-View Consistent Images", "abstract": "Generating multi-view images from human instructions is crucial for 3D content creation. The primary challenges involve maintaining consistency across multiple views and effectively synthesizing shapes and textures under diverse conditions. In this paper, we propose the Multi-View Auto-Regressive (\\textbf{MV-AR}) method, which leverages an auto-regressive model to progressively generate consistent multi-view images from arbitrary prompts. Firstly, the next-token-prediction capability of the AR model significantly enhances its effectiveness in facilitating progressive multi-view synthesis. When generating widely-separated views, MV-AR can utilize all its preceding views to extract effective reference information. Subsequently, we propose a unified model that accommodates various prompts via architecture designing and training strategies. To address multiple conditions, we introduce condition injection modules for text, camera pose, image, and shape. To manage multi-modal conditions simultaneously, a progressive training strategy is employed. This strategy initially adopts the text-to-multi-view (t2mv) model as a baseline to enhance the development of a comprehensive X-to-multi-view (X2mv) model through the randomly dropping and combining conditions. Finally, to alleviate the overfitting problem caused by limited high-quality data, we propose the ``Shuffle View\" data augmentation technique, thus significantly expanding the training data by several magnitudes. Experiments demonstrate the performance and versatility of our MV-AR, which consistently generates consistent multi-view images across a range of conditions and performs on par with leading diffusion-based multi-view image generation models. The code and models are released at https://github.com/MILab-PKU/MVAR.", "date": "2025-06-23", "categories": []} {"id": "2507.09748", "title": "Advancing Text-to-3D Generation with Linearized Lookahead Variational Score Distillation", "abstract": "Text-to-3D generation based on score distillation of pre-trained 2D diffusion models has gained increasing interest, with variational score distillation (VSD) as a remarkable example. VSD proves that vanilla score distillation can be improved by introducing an extra score-based model, which characterizes the distribution of images rendered from 3D models, to correct the distillation gradient. Despite the theoretical foundations, VSD, in practice, is likely to suffer from slow and sometimes ill-posed convergence. In this paper, we perform an in-depth investigation of the interplay between the introduced score model and the 3D model, and find that there exists a mismatching problem between LoRA and 3D distributions in practical implementation. We can simply adjust their optimization order to improve the generation quality. By doing so, the score model looks ahead to the current 3D state and hence yields more reasonable corrections. Nevertheless, naive lookahead VSD may suffer from unstable training in practice due to the potential over-fitting. To address this, we propose to use a linearized variant of the model for score distillation, giving rise to the Linearized Lookahead Variational Score Distillation ($L^2$-VSD). $L^2$-VSD can be realized efficiently with forward-mode autodiff functionalities of existing deep learning libraries. Extensive experiments validate the efficacy of $L^2$-VSD, revealing its clear superiority over prior score distillation-based methods. We also show that our method can be seamlessly incorporated into any other VSD-based text-to-3D framework.", "date": "2025-07-13", "categories": []} {"id": "2307.14770", "title": "3DPortraitGAN: Learning One-Quarter Headshot 3D GANs from a Single-View Portrait Dataset with Diverse Body Poses", "abstract": "3D-aware face generators are typically trained on 2D real-life face image datasets that primarily consist of near-frontal face data, and as such, they are unable to construct one-quarter headshot 3D portraits with complete head, neck, and shoulder geometry. Two reasons account for this issue: First, existing facial recognition methods struggle with extracting facial data captured from large camera angles or back views. Second, it is challenging to learn a distribution of 3D portraits covering the one-quarter headshot region from single-view data due to significant geometric deformation caused by diverse body poses. To this end, we first create the dataset 360{\\deg}-Portrait-HQ (360{\\deg}PHQ for short) which consists of high-quality single-view real portraits annotated with a variety of camera parameters (the yaw angles span the entire 360{\\deg} range) and body poses. We then propose 3DPortraitGAN, the first 3D-aware one-quarter headshot portrait generator that learns a canonical 3D avatar distribution from the 360{\\deg}PHQ dataset with body pose self-learning. Our model can generate view-consistent portrait images from all camera angles with a canonical one-quarter headshot 3D representation. Our experiments show that the proposed framework can accurately predict portrait body poses and generate view-consistent, realistic portrait images with complete geometry from all camera angles.", "date": "2023-07-27", "categories": []} {"id": "2507.05763", "title": "DreamArt: Generating Interactable Articulated Objects from a Single Image", "abstract": "Generating articulated objects, such as laptops and microwaves, is a crucial yet challenging task with extensive applications in Embodied AI and AR/VR. Current image-to-3D methods primarily focus on surface geometry and texture, neglecting part decomposition and articulation modeling. Meanwhile, neural reconstruction approaches (e.g., NeRF or Gaussian Splatting) rely on dense multi-view or interaction data, limiting their scalability. In this paper, we introduce DreamArt, a novel framework for generating high-fidelity, interactable articulated assets from single-view images. DreamArt employs a three-stage pipeline: firstly, it reconstructs part-segmented and complete 3D object meshes through a combination of image-to-3D generation, mask-prompted 3D segmentation, and part amodal completion. Second, we fine-tune a video diffusion model to capture part-level articulation priors, leveraging movable part masks as prompt and amodal images to mitigate ambiguities caused by occlusion. Finally, DreamArt optimizes the articulation motion, represented by a dual quaternion, and conducts global texture refinement and repainting to ensure coherent, high-quality textures across all parts. Experimental results demonstrate that DreamArt effectively generates high-quality articulated objects, possessing accurate part shape, high appearance fidelity, and plausible articulation, thereby providing a scalable solution for articulated asset generation. Our project page is available at https://dream-art-0.github.io/DreamArt/.", "date": "2025-07-08", "categories": []} {"id": "2507.06165", "title": "OmniPart: Part-Aware 3D Generation with Semantic Decoupling and Structural Cohesion", "abstract": "The creation of 3D assets with explicit, editable part structures is crucial for advancing interactive applications, yet most generative methods produce only monolithic shapes, limiting their utility. We introduce OmniPart, a novel framework for part-aware 3D object generation designed to achieve high semantic decoupling among components while maintaining robust structural cohesion. OmniPart uniquely decouples this complex task into two synergistic stages: (1) an autoregressive structure planning module generates a controllable, variable-length sequence of 3D part bounding boxes, critically guided by flexible 2D part masks that allow for intuitive control over part decomposition without requiring direct correspondences or semantic labels; and (2) a spatially-conditioned rectified flow model, efficiently adapted from a pre-trained holistic 3D generator, synthesizes all 3D parts simultaneously and consistently within the planned layout. Our approach supports user-defined part granularity, precise localization, and enables diverse downstream applications. Extensive experiments demonstrate that OmniPart achieves state-of-the-art performance, paving the way for more interpretable, editable, and versatile 3D content.", "date": "2025-07-08", "categories": []} {"id": "2407.14419", "title": "HOTS3D: Hyper-Spherical Optimal Transport for Semantic Alignment of Text-to-3D Generation", "abstract": "Recent CLIP-guided 3D generation methods have achieved promising results but struggle with generating faithful 3D shapes that conform with input text due to the gap between text and image embeddings. To this end, this paper proposes HOTS3D which makes the first attempt to effectively bridge this gap by aligning text features to the image features with spherical optimal transport(SOT). However, in high-dimensional situations, solving the SOT remains a challenge. To obtain the SOT map for high-dimensional features obtained from CLIP encoding of two modalities, we mathematically formulate and derive the solution based on Villani's theorem, which can directly align two hyper-sphere distributions without manifold exponential maps. Furthermore, we implement it by leveraging input convex neural networks (ICNNs) for the optimal Kantorovich potential. With the optimally mapped features, a diffusion-based generator is utilized to decode them into 3D shapes. Extensive quantitative and qualitative comparisons with state-of-the-art methods demonstrate the superiority of HOTS3D for text-to-3D generation, especially in the consistency with text semantics.", "date": "2024-07-19", "categories": []} {"id": "2411.12089", "title": "FruitNinja: 3D Object Interior Texture Generation with Gaussian Splatting", "abstract": "In the real world, objects reveal internal textures when sliced or cut, yet this behavior is not well-studied in 3D generation tasks today. For example, slicing a virtual 3D watermelon should reveal flesh and seeds. Given that no available dataset captures an object's full internal structure and collecting data from all slices is impractical, generative methods become the obvious approach. However, current 3D generation and inpainting methods often focus on visible appearance and overlook internal textures. To bridge this gap, we introduce FruitNinja, the first method to generate internal textures for 3D objects undergoing geometric and topological changes. Our approach produces objects via 3D Gaussian Splatting (3DGS) with both surface and interior textures synthesized, enabling real-time slicing and rendering without additional optimization. FruitNinja leverages a pre-trained diffusion model to progressively inpaint cross-sectional views and applies voxel-grid-based smoothing to achieve cohesive textures throughout the object. Our OpaqueAtom GS strategy overcomes 3DGS limitations by employing densely distributed opaque Gaussians, avoiding biases toward larger particles that destabilize training and sharp color transitions for fine-grained textures. Experimental results show that FruitNinja substantially outperforms existing approaches, showcasing unmatched visual quality in real-time rendered internal views across arbitrary geometry manipulations.", "date": "2024-11-18", "categories": []} {"id": "2505.07003", "title": "CMD: Controllable Multiview Diffusion for 3D Editing and Progressive Generation", "abstract": "Recently, 3D generation methods have shown their powerful ability to automate 3D model creation. However, most 3D generation methods only rely on an input image or a text prompt to generate a 3D model, which lacks the control of each component of the generated 3D model. Any modifications of the input image lead to an entire regeneration of the 3D models. In this paper, we introduce a new method called CMD that generates a 3D model from an input image while enabling flexible local editing of each component of the 3D model. In CMD, we formulate the 3D generation as a conditional multiview diffusion model, which takes the existing or known parts as conditions and generates the edited or added components. This conditional multiview diffusion model not only allows the generation of 3D models part by part but also enables local editing of 3D models according to the local revision of the input image without changing other 3D parts. Extensive experiments are conducted to demonstrate that CMD decomposes a complex 3D generation task into multiple components, improving the generation quality. Meanwhile, CMD enables efficient and flexible local editing of a 3D model by just editing one rendered image.", "date": "2025-05-11", "categories": []} {"id": "2505.20129", "title": "Agentic 3D Scene Generation with Spatially Contextualized VLMs", "abstract": "Despite recent advances in multimodal content generation enabled by vision-language models (VLMs), their ability to reason about and generate structured 3D scenes remains largely underexplored. This limitation constrains their utility in spatially grounded tasks such as embodied AI, immersive simulations, and interactive 3D applications. We introduce a new paradigm that enables VLMs to generate, understand, and edit complex 3D environments by injecting a continually evolving spatial context. Constructed from multimodal input, this context consists of three components: a scene portrait that provides a high-level semantic blueprint, a semantically labeled point cloud capturing object-level geometry, and a scene hypergraph that encodes rich spatial relationships, including unary, binary, and higher-order constraints. Together, these components provide the VLM with a structured, geometry-aware working memory that integrates its inherent multimodal reasoning capabilities with structured 3D understanding for effective spatial reasoning. Building on this foundation, we develop an agentic 3D scene generation pipeline in which the VLM iteratively reads from and updates the spatial context. The pipeline features high-quality asset generation with geometric restoration, environment setup with automatic verification, and ergonomic adjustment guided by the scene hypergraph. Experiments show that our framework can handle diverse and challenging inputs, achieving a level of generalization not observed in prior work. Further results demonstrate that injecting spatial context enables VLMs to perform downstream tasks such as interactive scene editing and path planning, suggesting strong potential for spatially intelligent systems in computer graphics, 3D vision, and embodied applications. Project page: https://spatctxvlm.github.io/project_page/.", "date": "2025-05-26", "categories": []} {"id": "2507.04285", "title": "SeqTex: Generate Mesh Textures in Video Sequence", "abstract": "Training native 3D texture generative models remains a fundamental yet challenging problem, largely due to the limited availability of large-scale, high-quality 3D texture datasets. This scarcity hinders generalization to real-world scenarios. To address this, most existing methods finetune foundation image generative models to exploit their learned visual priors. However, these approaches typically generate only multi-view images and rely on post-processing to produce UV texture maps -- an essential representation in modern graphics pipelines. Such two-stage pipelines often suffer from error accumulation and spatial inconsistencies across the 3D surface. In this paper, we introduce SeqTex, a novel end-to-end framework that leverages the visual knowledge encoded in pretrained video foundation models to directly generate complete UV texture maps. Unlike previous methods that model the distribution of UV textures in isolation, SeqTex reformulates the task as a sequence generation problem, enabling the model to learn the joint distribution of multi-view renderings and UV textures. This design effectively transfers the consistent image-space priors from video foundation models into the UV domain. To further enhance performance, we propose several architectural innovations: a decoupled multi-view and UV branch design, geometry-informed attention to guide cross-domain feature alignment, and adaptive token resolution to preserve fine texture details while maintaining computational efficiency. Together, these components allow SeqTex to fully utilize pretrained video priors and synthesize high-fidelity UV texture maps without the need for post-processing. Extensive experiments show that SeqTex achieves state-of-the-art performance on both image-conditioned and text-conditioned 3D texture generation tasks, with superior 3D consistency, texture-geometry alignment, and real-world generalization.", "date": "2025-07-06", "categories": []} {"id": "2507.04403", "title": "Sat2City: 3D City Generation from A Single Satellite Image with Cascaded Latent Diffusion", "abstract": "Recent advancements in generative models have enabled 3D urban scene generation from satellite imagery, unlocking promising applications in gaming, digital twins, and beyond. However, most existing methods rely heavily on neural rendering techniques, which hinder their ability to produce detailed 3D structures on a broader scale, largely due to the inherent structural ambiguity derived from relatively limited 2D observations. To address this challenge, we propose Sat2City, a novel framework that synergizes the representational capacity of sparse voxel grids with latent diffusion models, tailored specifically for our novel 3D city dataset. Our approach is enabled by three key components: (1) A cascaded latent diffusion framework that progressively recovers 3D city structures from satellite imagery, (2) a Re-Hash operation at its Variational Autoencoder (VAE) bottleneck to compute multi-scale feature grids for stable appearance optimization and (3) an inverse sampling strategy enabling implicit supervision for smooth appearance transitioning.To overcome the challenge of collecting real-world city-scale 3D models with high-quality geometry and appearance, we introduce a dataset of synthesized large-scale 3D cities paired with satellite-view height maps. Validated on this dataset, our framework generates detailed 3D structures from a single satellite image, achieving superior fidelity compared to existing city generation models.", "date": "2025-07-06", "categories": []} {"id": "2507.04765", "title": "GraphBrep: Learning B-Rep in Graph Structure for Efficient CAD Generation", "abstract": "Direct B-Rep generation is increasingly important in CAD workflows, eliminating costly modeling sequence data and supporting complex features. A key challenge is modeling joint distribution of the misaligned geometry and topology. Existing methods tend to implicitly embed topology into the geometric features of edges. Although this integration ensures feature alignment, it also causes edge geometry to carry more redundant structural information compared to the original B-Rep, leading to significantly higher computational cost. To reduce redundancy, we propose GraphBrep, a B-Rep generation model that explicitly represents and learns compact topology. Following the original structure of B-Rep, we construct an undirected weighted graph to represent surface topology. A graph diffusion model is employed to learn topology conditioned on surface features, serving as the basis for determining connectivity between primitive surfaces. The explicit representation ensures a compact data structure, effectively reducing computational cost during both training and inference. Experiments on two large-scale unconditional datasets and one category-conditional dataset demonstrate the proposed method significantly reduces training and inference times (up to 31.3% and 56.3% for given datasets, respectively) while maintaining high-quality CAD generation compared with SOTA.", "date": "2025-07-07", "categories": []} {"id": "2507.02299", "title": "DreamComposer++: Empowering Diffusion Models with Multi-View Conditions for 3D Content Generation", "abstract": "Recent advancements in leveraging pre-trained 2D diffusion models achieve the generation of high-quality novel views from a single in-the-wild image. However, existing works face challenges in producing controllable novel views due to the lack of information from multiple views. In this paper, we present DreamComposer++, a flexible and scalable framework designed to improve current view-aware diffusion models by incorporating multi-view conditions. Specifically, DreamComposer++ utilizes a view-aware 3D lifting module to extract 3D representations of an object from various views. These representations are then aggregated and rendered into the latent features of target view through the multi-view feature fusion module. Finally, the obtained features of target view are integrated into pre-trained image or video diffusion models for novel view synthesis. Experimental results demonstrate that DreamComposer++ seamlessly integrates with cutting-edge view-aware diffusion models and enhances their abilities to generate controllable novel views from multi-view conditions. This advancement facilitates controllable 3D object reconstruction and enables a wide range of applications.", "date": "2025-07-03", "categories": []} {"id": "2502.07007", "title": "Grounding Creativity in Physics: A Brief Survey of Physical Priors in AIGC", "abstract": "Recent advancements in AI-generated content have significantly improved the realism of 3D and 4D generation. However, most existing methods prioritize appearance consistency while neglecting underlying physical principles, leading to artifacts such as unrealistic deformations, unstable dynamics, and implausible objects interactions. Incorporating physics priors into generative models has become a crucial research direction to enhance structural integrity and motion realism. This survey provides a review of physics-aware generative methods, systematically analyzing how physical constraints are integrated into 3D and 4D generation. First, we examine recent works in incorporating physical priors into static and dynamic 3D generation, categorizing methods based on representation types, including vision-based, NeRF-based, and Gaussian Splatting-based approaches. Second, we explore emerging techniques in 4D generation, focusing on methods that model temporal dynamics with physical simulations. Finally, we conduct a comparative analysis of major methods, highlighting their strengths, limitations, and suitability for different materials and motion dynamics. By presenting an in-depth analysis of physics-grounded AIGC, this survey aims to bridge the gap between generative models and physical realism, providing insights that inspire future research in physically consistent content generation.", "date": "2025-02-10", "categories": []} {"id": "2507.00006", "title": "MVGBench: Comprehensive Benchmark for Multi-view Generation Models", "abstract": "We propose MVGBench, a comprehensive benchmark for multi-view image generation models (MVGs) that evaluates 3D consistency in geometry and texture, image quality, and semantics (using vision language models). Recently, MVGs have been the main driving force in 3D object creation. However, existing metrics compare generated images against ground truth target views, which is not suitable for generative tasks where multiple solutions exist while differing from ground truth. Furthermore, different MVGs are trained on different view angles, synthetic data and specific lightings -- robustness to these factors and generalization to real data are rarely evaluated thoroughly. Without a rigorous evaluation protocol, it is also unclear what design choices contribute to the progress of MVGs. MVGBench evaluates three different aspects: best setup performance, generalization to real data and robustness. Instead of comparing against ground truth, we introduce a novel 3D self-consistency metric which compares 3D reconstructions from disjoint generated multi-views. We systematically compare 12 existing MVGs on 4 different curated real and synthetic datasets. With our analysis, we identify important limitations of existing methods specially in terms of robustness and generalization, and we find the most critical design choices. Using the discovered best practices, we propose ViFiGen, a method that outperforms all evaluated MVGs on 3D consistency. Our code, model, and benchmark suite will be publicly released.", "date": "2025-06-11", "categories": []} {"id": "2507.00429", "title": "DiGA3D: Coarse-to-Fine Diffusional Propagation of Geometry and Appearance for Versatile 3D Inpainting", "abstract": "Developing a unified pipeline that enables users to remove, re-texture, or replace objects in a versatile manner is crucial for text-guided 3D inpainting. However, there are still challenges in performing multiple 3D inpainting tasks within a unified framework: 1) Single reference inpainting methods lack robustness when dealing with views that are far from the reference view. 2) Appearance inconsistency arises when independently inpainting multi-view images with 2D diffusion priors; 3) Geometry inconsistency limits performance when there are significant geometric changes in the inpainting regions. To tackle these challenges, we introduce DiGA3D, a novel and versatile 3D inpainting pipeline that leverages diffusion models to propagate consistent appearance and geometry in a coarse-to-fine manner. First, DiGA3D develops a robust strategy for selecting multiple reference views to reduce errors during propagation. Next, DiGA3D designs an Attention Feature Propagation (AFP) mechanism that propagates attention features from the selected reference views to other views via diffusion models to maintain appearance consistency. Furthermore, DiGA3D introduces a Texture-Geometry Score Distillation Sampling (TG-SDS) loss to further improve the geometric consistency of inpainted 3D scenes. Extensive experiments on multiple 3D inpainting tasks demonstrate the effectiveness of our method. The project page is available at https://rorisis.github.io/DiGA3D/.", "date": "2025-07-01", "categories": []} {"id": "2506.10186", "title": "Scalable Non-Equivariant 3D Molecule Generation via Rotational Alignment", "abstract": "Equivariant diffusion models have achieved impressive performance in 3D molecule generation. These models incorporate Euclidean symmetries of 3D molecules by utilizing an SE(3)-equivariant denoising network. However, specialized equivariant architectures limit the scalability and efficiency of diffusion models. In this paper, we propose an approach that relaxes such equivariance constraints. Specifically, our approach learns a sample-dependent SO(3) transformation for each molecule to construct an aligned latent space. A non-equivariant diffusion model is then trained over the aligned representations. Experimental results demonstrate that our approach performs significantly better than previously reported non-equivariant models. It yields sample quality comparable to state-of-the-art equivariant diffusion models and offers improved training and sampling efficiency. Our code is available at https://github.com/skeletondyh/RADM", "date": "2025-06-11", "categories": []} {"id": "2506.10507", "title": "Edit360: 2D Image Edits to 3D Assets from Any Angle", "abstract": "Recent advances in diffusion models have significantly improved image generation and editing, but extending these capabilities to 3D assets remains challenging, especially for fine-grained edits that require multi-view consistency. Existing methods typically restrict editing to predetermined viewing angles, severely limiting their flexibility and practical applications. We introduce Edit360, a tuning-free framework that extends 2D modifications to multi-view consistent 3D editing. Built upon video diffusion models, Edit360 enables user-specific editing from arbitrary viewpoints while ensuring structural coherence across all views. The framework selects anchor views for 2D modifications and propagates edits across the entire 360-degree range. To achieve this, Edit360 introduces a novel Anchor-View Editing Propagation mechanism, which effectively aligns and merges multi-view information within the latent and attention spaces of diffusion models. The resulting edited multi-view sequences facilitate the reconstruction of high-quality 3D assets, enabling customizable 3D content creation.", "date": "2025-06-12", "categories": []} {"id": "2506.21544", "title": "DeOcc-1-to-3: 3D De-Occlusion from a Single Image via Self-Supervised Multi-View Diffusion", "abstract": "Reconstructing 3D objects from a single image remains challenging, especially under real-world occlusions. While recent diffusion-based view synthesis models can generate consistent novel views from a single RGB image, they typically assume fully visible inputs and fail when parts of the object are occluded, resulting in degraded 3D reconstruction quality. We propose DeOcc-1-to-3, an end-to-end framework for occlusion-aware multi-view generation that synthesizes six structurally consistent novel views directly from a single occluded image, enabling reliable 3D reconstruction without prior inpainting or manual annotations. Our self-supervised training pipeline leverages occluded-unoccluded image pairs and pseudo-ground-truth views to teach the model structure-aware completion and view consistency. Without modifying the original architecture, we fully fine-tune the view synthesis model to jointly learn completion and multi-view generation. Additionally, we introduce the first benchmark for occlusion-aware reconstruction, covering diverse occlusion levels, object categories, and masking patterns, providing a standardized protocol for future evaluation.", "date": "2025-06-26", "categories": []} {"id": "2506.22833", "title": "SemFaceEdit: Semantic Face Editing on Generative Radiance Manifolds", "abstract": "Despite multiple view consistency offered by 3D-aware GAN techniques, the resulting images often lack the capacity for localized editing. In response, generative radiance manifolds emerge as an efficient approach for constrained point sampling within volumes, effectively reducing computational demands and enabling the learning of fine details. This work introduces SemFaceEdit, a novel method that streamlines the appearance and geometric editing process by generating semantic fields on generative radiance manifolds. Utilizing latent codes, our method effectively disentangles the geometry and appearance associated with different facial semantics within the generated image. In contrast to existing methods that can change the appearance of the entire radiance field, our method enables the precise editing of particular facial semantics while preserving the integrity of other regions. Our network comprises two key modules: the Geometry module, which generates semantic radiance and occupancy fields, and the Appearance module, which is responsible for predicting RGB radiance. We jointly train both modules in adversarial settings to learn semantic-aware geometry and appearance descriptors. The appearance descriptors are then conditioned on their respective semantic latent codes by the Appearance Module, facilitating disentanglement and enhanced control. Our experiments highlight SemFaceEdit's superior performance in semantic field-based editing, particularly in achieving improved radiance field disentanglement.", "date": "2025-06-28", "categories": []} {"id": "2502.08377", "title": "Not All Frame Features Are Equal: Video-to-4D Generation via Decoupling Dynamic-Static Features", "abstract": "Recently, the generation of dynamic 3D objects from a video has shown impressive results. Existing methods directly optimize Gaussians using whole information in frames. However, when dynamic regions are interwoven with static regions within frames, particularly if the static regions account for a large proportion, existing methods often overlook information in dynamic regions and are prone to overfitting on static regions. This leads to producing results with blurry textures. We consider that decoupling dynamic-static features to enhance dynamic representations can alleviate this issue. Thus, we propose a dynamic-static feature decoupling module (DSFD). Along temporal axes, it regards the regions of current frame features that possess significant differences relative to reference frame features as dynamic features. Conversely, the remaining parts are the static features. Then, we acquire decoupled features driven by dynamic features and current frame features. Moreover, to further enhance the dynamic representation of decoupled features from different viewpoints and ensure accurate motion prediction, we design a temporal-spatial similarity fusion module (TSSF). Along spatial axes, it adaptively selects similar information of dynamic regions. Hinging on the above, we construct a novel approach, DS4D. Experimental results verify our method achieves state-of-the-art (SOTA) results in video-to-4D. In addition, the experiments on a real-world scenario dataset demonstrate its effectiveness on the 4D scene. Our code will be publicly available.", "date": "2025-02-12", "categories": []} {"id": "2506.22291", "title": "RoomCraft: Controllable and Complete 3D Indoor Scene Generation", "abstract": "Generating realistic 3D indoor scenes from user inputs remains a challenging problem in computer vision and graphics, requiring careful balance of geometric consistency, spatial relationships, and visual realism. While neural generation methods often produce repetitive elements due to limited global spatial reasoning, procedural approaches can leverage constraints for controllable generation but struggle with multi-constraint scenarios. When constraints become numerous, object collisions frequently occur, forcing the removal of furniture items and compromising layout completeness.\n To address these limitations, we propose RoomCraft, a multi-stage pipeline that converts real images, sketches, or text descriptions into coherent 3D indoor scenes. Our approach combines a scene generation pipeline with a constraint-driven optimization framework. The pipeline first extracts high-level scene information from user inputs and organizes it into a structured format containing room type, furniture items, and spatial relations. It then constructs a spatial relationship network to represent furniture arrangements and generates an optimized placement sequence using a heuristic-based depth-first search (HDFS) algorithm to ensure layout coherence. To handle complex multi-constraint scenarios, we introduce a unified constraint representation that processes both formal specifications and natural language inputs, enabling flexible constraint-oriented adjustments through a comprehensive action space design. Additionally, we propose a Conflict-Aware Positioning Strategy (CAPS) that dynamically adjusts placement weights to minimize furniture collisions and ensure layout completeness.\n Extensive experiments demonstrate that RoomCraft significantly outperforms existing methods in generating realistic, semantically coherent, and visually appealing room layouts across diverse input modalities.", "date": "2025-06-27", "categories": []} {"id": "2412.03934", "title": "InfiniCube: Unbounded and Controllable Dynamic 3D Driving Scene Generation with World-Guided Video Models", "abstract": "We present InfiniCube, a scalable method for generating unbounded dynamic 3D driving scenes with high fidelity and controllability. Previous methods for scene generation either suffer from limited scales or lack geometric and appearance consistency along generated sequences. In contrast, we leverage the recent advancements in scalable 3D representation and video models to achieve large dynamic scene generation that allows flexible controls through HD maps, vehicle bounding boxes, and text descriptions. First, we construct a map-conditioned sparse-voxel-based 3D generative model to unleash its power for unbounded voxel world generation. Then, we re-purpose a video model and ground it on the voxel world through a set of carefully designed pixel-aligned guidance buffers, synthesizing a consistent appearance. Finally, we propose a fast feed-forward approach that employs both voxel and pixel branches to lift the dynamic videos to dynamic 3D Gaussians with controllable objects. Our method can generate controllable and realistic 3D driving scenes, and extensive experiments validate the effectiveness and superiority of our model.", "date": "2024-12-05", "categories": []} {"id": "2506.20875", "title": "3DGH: 3D Head Generation with Composable Hair and Face", "abstract": "We present 3DGH, an unconditional generative model for 3D human heads with composable hair and face components. Unlike previous work that entangles the modeling of hair and face, we propose to separate them using a novel data representation with template-based 3D Gaussian Splatting, in which deformable hair geometry is introduced to capture the geometric variations across different hairstyles. Based on this data representation, we design a 3D GAN-based architecture with dual generators and employ a cross-attention mechanism to model the inherent correlation between hair and face. The model is trained on synthetic renderings using carefully designed objectives to stabilize training and facilitate hair-face separation. We conduct extensive experiments to validate the design choice of 3DGH, and evaluate it both qualitatively and quantitatively by comparing with several state-of-the-art 3D GAN methods, demonstrating its effectiveness in unconditional full-head image synthesis and composable 3D hairstyle editing. More details will be available on our project page: https://c-he.github.io/projects/3dgh/.", "date": "2025-06-25", "categories": []} {"id": "2506.20946", "title": "Consistent Zero-shot 3D Texture Synthesis Using Geometry-aware Diffusion and Temporal Video Models", "abstract": "Current texture synthesis methods, which generate textures from fixed viewpoints, suffer from inconsistencies due to the lack of global context and geometric understanding. Meanwhile, recent advancements in video generation models have demonstrated remarkable success in achieving temporally consistent videos. In this paper, we introduce VideoTex, a novel framework for seamless texture synthesis that leverages video generation models to address both spatial and temporal inconsistencies in 3D textures. Our approach incorporates geometry-aware conditions, enabling precise utilization of 3D mesh structures. Additionally, we propose a structure-wise UV diffusion strategy, which enhances the generation of occluded areas by preserving semantic information, resulting in smoother and more coherent textures. VideoTex not only achieves smoother transitions across UV boundaries but also ensures high-quality, temporally stable textures across video frames. Extensive experiments demonstrate that VideoTex outperforms existing methods in texture fidelity, seam blending, and stability, paving the way for dynamic real-time applications that demand both visual quality and temporal coherence.", "date": "2025-06-26", "categories": []} {"id": "2506.20590", "title": "WonderFree: Enhancing Novel View Quality and Cross-View Consistency for 3D Scene Exploration", "abstract": "Interactive 3D scene generation from a single image has gained significant attention due to its potential to create immersive virtual worlds. However, a key challenge in current 3D generation methods is the limited explorability, which cannot render high-quality images during larger maneuvers beyond the original viewpoint, particularly when attempting to move forward into unseen areas. To address this challenge, we propose WonderFree, the first model that enables users to interactively generate 3D worlds with the freedom to explore from arbitrary angles and directions. Specifically, we decouple this challenge into two key subproblems: novel view quality, which addresses visual artifacts and floating issues in novel views, and cross-view consistency, which ensures spatial consistency across different viewpoints. To enhance rendering quality in novel views, we introduce WorldRestorer, a data-driven video restoration model designed to eliminate floaters and artifacts. In addition, a data collection pipeline is presented to automatically gather training data for WorldRestorer, ensuring it can handle scenes with varying styles needed for 3D scene generation. Furthermore, to improve cross-view consistency, we propose ConsistView, a multi-view joint restoration mechanism that simultaneously restores multiple perspectives while maintaining spatiotemporal coherence. Experimental results demonstrate that WonderFree not only enhances rendering quality across diverse viewpoints but also significantly improves global coherence and consistency. These improvements are confirmed by CLIP-based metrics and a user study showing a 77.20% preference for WonderFree over WonderWorld enabling a seamless and immersive 3D exploration experience. The code, model, and data will be publicly available.", "date": "2025-06-25", "categories": []} {"id": "2506.20601", "title": "Video Perception Models for 3D Scene Synthesis", "abstract": "Traditionally, 3D scene synthesis requires expert knowledge and significant manual effort. Automating this process could greatly benefit fields such as architectural design, robotics simulation, virtual reality, and gaming. Recent approaches to 3D scene synthesis often rely on the commonsense reasoning of large language models (LLMs) or strong visual priors of modern image generation models. However, current LLMs demonstrate limited 3D spatial reasoning ability, which restricts their ability to generate realistic and coherent 3D scenes. Meanwhile, image generation-based methods often suffer from constraints in viewpoint selection and multi-view inconsistencies. In this work, we present Video Perception models for 3D Scene synthesis (VIPScene), a novel framework that exploits the encoded commonsense knowledge of the 3D physical world in video generation models to ensure coherent scene layouts and consistent object placements across views. VIPScene accepts both text and image prompts and seamlessly integrates video generation, feedforward 3D reconstruction, and open-vocabulary perception models to semantically and geometrically analyze each object in a scene. This enables flexible scene synthesis with high realism and structural consistency. For more precise analysis, we further introduce First-Person View Score (FPVScore) for coherence and plausibility evaluation, utilizing continuous first-person perspective to capitalize on the reasoning ability of multimodal large language models. Extensive experiments show that VIPScene significantly outperforms existing methods and generalizes well across diverse scenarios. The code will be released.", "date": "2025-06-25", "categories": []} {"id": "2506.20652", "title": "EditP23: 3D Editing via Propagation of Image Prompts to Multi-View", "abstract": "We present EditP23, a method for mask-free 3D editing that propagates 2D image edits to multi-view representations in a 3D-consistent manner. In contrast to traditional approaches that rely on text-based prompting or explicit spatial masks, EditP23 enables intuitive edits by conditioning on a pair of images: an original view and its user-edited counterpart. These image prompts are used to guide an edit-aware flow in the latent space of a pre-trained multi-view diffusion model, allowing the edit to be coherently propagated across views. Our method operates in a feed-forward manner, without optimization, and preserves the identity of the original object, in both structure and appearance. We demonstrate its effectiveness across a range of object categories and editing scenarios, achieving high fidelity to the source while requiring no manual masks.", "date": "2025-06-25", "categories": []} {"id": "2506.17707", "title": "Programmable-Room: Interactive Textured 3D Room Meshes Generation Empowered by Large Language Models", "abstract": "We present Programmable-Room, a framework which interactively generates and edits a 3D room mesh, given natural language instructions. For precise control of a room's each attribute, we decompose the challenging task into simpler steps such as creating plausible 3D coordinates for room meshes, generating panorama images for the texture, constructing 3D meshes by integrating the coordinates and panorama texture images, and arranging furniture. To support the various decomposed tasks with a unified framework, we incorporate visual programming (VP). VP is a method that utilizes a large language model (LLM) to write a Python-like program which is an ordered list of necessary modules for the various tasks given in natural language. We develop most of the modules. Especially, for the texture generating module, we utilize a pretrained large-scale diffusion model to generate panorama images conditioned on text and visual prompts (i.e., layout, depth, and semantic map) simultaneously. Specifically, we enhance the panorama image generation quality by optimizing the training objective with a 1D representation of a panorama scene obtained from bidirectional LSTM. We demonstrate Programmable-Room's flexibility in generating and editing 3D room meshes, and prove our framework's superiority to an existing model quantitatively and qualitatively. Project page is available in https://jihyun0510.github.io/Programmable_Room_Page/.", "date": "2025-06-21", "categories": []} {"id": "2506.18839", "title": "4Real-Video-V2: Fused View-Time Attention and Feedforward Reconstruction for 4D Scene Generation", "abstract": "We propose the first framework capable of computing a 4D spatio-temporal grid of video frames and 3D Gaussian particles for each time step using a feed-forward architecture. Our architecture has two main components, a 4D video model and a 4D reconstruction model. In the first part, we analyze current 4D video diffusion architectures that perform spatial and temporal attention either sequentially or in parallel within a two-stream design. We highlight the limitations of existing approaches and introduce a novel fused architecture that performs spatial and temporal attention within a single layer. The key to our method is a sparse attention pattern, where tokens attend to others in the same frame, at the same timestamp, or from the same viewpoint. In the second part, we extend existing 3D reconstruction algorithms by introducing a Gaussian head, a camera token replacement algorithm, and additional dynamic layers and training. Overall, we establish a new state of the art for 4D generation, improving both visual quality and reconstruction capability.", "date": "2025-06-18", "categories": []} {"id": "2506.16504", "title": "Hunyuan3D 2.5: Towards High-Fidelity 3D Assets Generation with Ultimate Details", "abstract": "In this report, we present Hunyuan3D 2.5, a robust suite of 3D diffusion models aimed at generating high-fidelity and detailed textured 3D assets. Hunyuan3D 2.5 follows two-stages pipeline of its previous version Hunyuan3D 2.0, while demonstrating substantial advancements in both shape and texture generation. In terms of shape generation, we introduce a new shape foundation model -- LATTICE, which is trained with scaled high-quality datasets, model-size, and compute. Our largest model reaches 10B parameters and generates sharp and detailed 3D shape with precise image-3D following while keeping mesh surface clean and smooth, significantly closing the gap between generated and handcrafted 3D shapes. In terms of texture generation, it is upgraded with phyiscal-based rendering (PBR) via a novel multi-view architecture extended from Hunyuan3D 2.0 Paint model. Our extensive evaluation shows that Hunyuan3D 2.5 significantly outperforms previous methods in both shape and end-to-end texture generation.", "date": "2025-06-19", "categories": []} {"id": "2506.17074", "title": "Assembler: Scalable 3D Part Assembly via Anchor Point Diffusion", "abstract": "We present Assembler, a scalable and generalizable framework for 3D part assembly that reconstructs complete objects from input part meshes and a reference image. Unlike prior approaches that mostly rely on deterministic part pose prediction and category-specific training, Assembler is designed to handle diverse, in-the-wild objects with varying part counts, geometries, and structures. It addresses the core challenges of scaling to general 3D part assembly through innovations in task formulation, representation, and data. First, Assembler casts part assembly as a generative problem and employs diffusion models to sample plausible configurations, effectively capturing ambiguities arising from symmetry, repeated parts, and multiple valid assemblies. Second, we introduce a novel shape-centric representation based on sparse anchor point clouds, enabling scalable generation in Euclidean space rather than SE(3) pose prediction. Third, we construct a large-scale dataset of over 320K diverse part-object assemblies using a synthesis and filtering pipeline built on existing 3D shape repositories. Assembler achieves state-of-the-art performance on PartNet and is the first to demonstrate high-quality assembly for complex, real-world objects. Based on Assembler, we further introduce an interesting part-aware 3D modeling system that generates high-resolution, editable objects from images, demonstrating potential for interactive and compositional design. Project page: https://assembler3d.github.io", "date": "2025-06-20", "categories": []} {"id": "2506.17206", "title": "DreamCube: 3D Panorama Generation via Multi-plane Synchronization", "abstract": "3D panorama synthesis is a promising yet challenging task that demands high-quality and diverse visual appearance and geometry of the generated omnidirectional content. Existing methods leverage rich image priors from pre-trained 2D foundation models to circumvent the scarcity of 3D panoramic data, but the incompatibility between 3D panoramas and 2D single views limits their effectiveness. In this work, we demonstrate that by applying multi-plane synchronization to the operators from 2D foundation models, their capabilities can be seamlessly extended to the omnidirectional domain. Based on this design, we further introduce DreamCube, a multi-plane RGB-D diffusion model for 3D panorama generation, which maximizes the reuse of 2D foundation model priors to achieve diverse appearances and accurate geometry while maintaining multi-view consistency. Extensive experiments demonstrate the effectiveness of our approach in panoramic image generation, panoramic depth estimation, and 3D scene generation.", "date": "2025-06-20", "categories": []} {"id": "2506.15442", "title": "Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material", "abstract": "3D AI-generated content (AIGC) is a passionate field that has significantly accelerated the creation of 3D models in gaming, film, and design. Despite the development of several groundbreaking models that have revolutionized 3D generation, the field remains largely accessible only to researchers, developers, and designers due to the complexities involved in collecting, processing, and training 3D models. To address these challenges, we introduce Hunyuan3D 2.1 as a case study in this tutorial. This tutorial offers a comprehensive, step-by-step guide on processing 3D data, training a 3D generative model, and evaluating its performance using Hunyuan3D 2.1, an advanced system for producing high-resolution, textured 3D assets. The system comprises two core components: the Hunyuan3D-DiT for shape generation and the Hunyuan3D-Paint for texture synthesis. We will explore the entire workflow, including data preparation, model architecture, training strategies, evaluation metrics, and deployment. By the conclusion of this tutorial, you will have the knowledge to finetune or develop a robust 3D generative model suitable for applications in gaming, virtual reality, and industrial design.", "date": "2025-06-18", "categories": []} {"id": "2506.15684", "title": "Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards", "abstract": "Generating high-quality and photorealistic 3D assets remains a longstanding challenge in 3D vision and computer graphics. Although state-of-the-art generative models, such as diffusion models, have made significant progress in 3D generation, they often fall short of human-designed content due to limited ability to follow instructions, align with human preferences, or produce realistic textures, geometries, and physical attributes. In this paper, we introduce Nabla-R2D3, a highly effective and sample-efficient reinforcement learning alignment framework for 3D-native diffusion models using 2D rewards. Built upon the recently proposed Nabla-GFlowNet method, which matches the score function to reward gradients in a principled manner for reward finetuning, our Nabla-R2D3 enables effective adaptation of 3D diffusion models using only 2D reward signals. Extensive experiments show that, unlike vanilla finetuning baselines which either struggle to converge or suffer from reward hacking, Nabla-R2D3 consistently achieves higher rewards and reduced prior forgetting within a few finetuning steps.", "date": "2025-06-18", "categories": []} {"id": "2506.14015", "title": "Disentangling 3D from Large Vision-Language Models for Controlled Portrait Generation", "abstract": "We consider the problem of disentangling 3D from large vision-language models, which we show on generative 3D portraits. This allows free-form text control of appearance attributes like age, hair style, and glasses, and 3D geometry control of face expression and camera pose. In this setting, we assume we use a pre-trained large vision-language model (LVLM; CLIP) to generate from a smaller 2D dataset with no additional paired labels and with a pre-defined 3D morphable model (FLAME). First, we disentangle using canonicalization to a 2D reference frame from a deformable neural 3D triplane representation. But another form of entanglement arises from the significant noise in the LVLM's embedding space that describes irrelevant features. This damages output quality and diversity, but we overcome this with a Jacobian regularization that can be computed efficiently with a stochastic approximator. Compared to existing methods, our approach produces portraits with added text and 3D control, where portraits remain consistent when either control is changed. Broadly, this approach lets creators control 3D generators on their own 2D face data without needing resources to label large data or train large models.", "date": "2025-06-16", "categories": []} {"id": "2403.13238", "title": "Learning Coherent Matrixized Representation in Latent Space for Volumetric 4D Generation", "abstract": "Directly learning to model 4D content, including shape, color, and motion, is challenging. Existing methods rely on pose priors for motion control, resulting in limited motion diversity and continuity in details. To address this, we propose a framework that generates volumetric 4D sequences, where 3D shapes are animated under given conditions (text-image guidance) with dynamic evolution in shape and color across spatial and temporal dimensions, allowing for free navigation and rendering from any direction. We first use a coherent 3D shape and color modeling to encode the shape and color of each detailed 3D geometry frame into a latent space. Then we propose a matrixized 4D sequence representation allowing efficient diffusion model operation. Finally, we introduce spatio-temporal diffusion for 4D volumetric generation under given images and text prompts. Extensive experiments on the ShapeNet, 3DBiCar, DeformingThings4D and Objaverse datasets for several tasks demonstrate that our method effectively learns to generate high quality 3D shapes with consistent color and coherent mesh animations, improving over the current methods. Our code will be publicly available.", "date": "2024-03-20", "categories": []} {"id": "2506.10600", "title": "EmbodiedGen: Towards a Generative 3D World Engine for Embodied Intelligence", "abstract": "Constructing a physically realistic and accurately scaled simulated 3D world is crucial for the training and evaluation of embodied intelligence tasks. The diversity, realism, low cost accessibility and affordability of 3D data assets are critical for achieving generalization and scalability in embodied AI. However, most current embodied intelligence tasks still rely heavily on traditional 3D computer graphics assets manually created and annotated, which suffer from high production costs and limited realism. These limitations significantly hinder the scalability of data driven approaches. We present EmbodiedGen, a foundational platform for interactive 3D world generation. It enables the scalable generation of high-quality, controllable and photorealistic 3D assets with accurate physical properties and real-world scale in the Unified Robotics Description Format (URDF) at low cost. These assets can be directly imported into various physics simulation engines for fine-grained physical control, supporting downstream tasks in training and evaluation. EmbodiedGen is an easy-to-use, full-featured toolkit composed of six key modules: Image-to-3D, Text-to-3D, Texture Generation, Articulated Object Generation, Scene Generation and Layout Generation. EmbodiedGen generates diverse and interactive 3D worlds composed of generative 3D assets, leveraging generative AI to address the challenges of generalization and evaluation to the needs of embodied intelligence related research. Code is available at https://horizonrobotics.github.io/robot_lab/embodied_gen/index.html.", "date": "2025-06-12", "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": "2506.13594", "title": "Dive3D: Diverse Distillation-based Text-to-3D Generation via Score Implicit Matching", "abstract": "Distilling pre-trained 2D diffusion models into 3D assets has driven remarkable advances in text-to-3D synthesis. However, existing methods typically rely on Score Distillation Sampling (SDS) loss, which involves asymmetric KL divergence--a formulation that inherently favors mode-seeking behavior and limits generation diversity. In this paper, we introduce Dive3D, a novel text-to-3D generation framework that replaces KL-based objectives with Score Implicit Matching (SIM) loss, a score-based objective that effectively mitigates mode collapse. Furthermore, Dive3D integrates both diffusion distillation and reward-guided optimization under a unified divergence perspective. Such reformulation, together with SIM loss, yields significantly more diverse 3D outputs while improving text alignment, human preference, and overall visual fidelity. We validate Dive3D across various 2D-to-3D prompts and find that it consistently outperforms prior methods in qualitative assessments, including diversity, photorealism, and aesthetic appeal. We further evaluate its performance on the GPTEval3D benchmark, comparing against nine state-of-the-art baselines. Dive3D also achieves strong results on quantitative metrics, including text-asset alignment, 3D plausibility, text-geometry consistency, texture quality, and geometric detail.", "date": "2025-06-16", "categories": []} {"id": "2506.11148", "title": "LLM-to-Phy3D: Physically Conform Online 3D Object Generation with LLMs", "abstract": "The emergence of generative artificial intelligence (GenAI) and large language models (LLMs) has revolutionized the landscape of digital content creation in different modalities. However, its potential use in Physical AI for engineering design, where the production of physically viable artifacts is paramount, remains vastly underexplored. The absence of physical knowledge in existing LLM-to-3D models often results in outputs detached from real-world physical constraints. To address this gap, we introduce LLM-to-Phy3D, a physically conform online 3D object generation that enables existing LLM-to-3D models to produce physically conforming 3D objects on the fly. LLM-to-Phy3D introduces a novel online black-box refinement loop that empowers large language models (LLMs) through synergistic visual and physics-based evaluations. By delivering directional feedback in an iterative refinement process, LLM-to-Phy3D actively drives the discovery of prompts that yield 3D artifacts with enhanced physical performance and greater geometric novelty relative to reference objects, marking a substantial contribution to AI-driven generative design. Systematic evaluations of LLM-to-Phy3D, supported by ablation studies in vehicle design optimization, reveal various LLM improvements gained by 4.5% to 106.7% in producing physically conform target domain 3D designs over conventional LLM-to-3D models. The encouraging results suggest the potential general use of LLM-to-Phy3D in Physical AI for scientific and engineering applications.", "date": "2025-06-11", "categories": []} {"id": "2505.14521", "title": "Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modeling", "abstract": "High-fidelity 3D object synthesis remains significantly more challenging than 2D image generation due to the unstructured nature of mesh data and the cubic complexity of dense volumetric grids. Existing two-stage pipelines-compressing meshes with a VAE (using either 2D or 3D supervision), followed by latent diffusion sampling-often suffer from severe detail loss caused by inefficient representations and modality mismatches introduced in VAE. We introduce Sparc3D, a unified framework that combines a sparse deformable marching cubes representation Sparcubes with a novel encoder Sparconv-VAE. Sparcubes converts raw meshes into high-resolution ($1024^3$) surfaces with arbitrary topology by scattering signed distance and deformation fields onto a sparse cube, allowing differentiable optimization. Sparconv-VAE is the first modality-consistent variational autoencoder built entirely upon sparse convolutional networks, enabling efficient and near-lossless 3D reconstruction suitable for high-resolution generative modeling through latent diffusion. Sparc3D achieves state-of-the-art reconstruction fidelity on challenging inputs, including open surfaces, disconnected components, and intricate geometry. It preserves fine-grained shape details, reduces training and inference cost, and integrates naturally with latent diffusion models for scalable, high-resolution 3D generation.", "date": "2025-05-20", "categories": []} {"id": "2506.10612", "title": "TexTailor: Customized Text-aligned Texturing via Effective Resampling", "abstract": "We present TexTailor, a novel method for generating consistent object textures from textual descriptions. Existing text-to-texture synthesis approaches utilize depth-aware diffusion models to progressively generate images and synthesize textures across predefined multiple viewpoints. However, these approaches lead to a gradual shift in texture properties across viewpoints due to (1) insufficient integration of previously synthesized textures at each viewpoint during the diffusion process and (2) the autoregressive nature of the texture synthesis process. Moreover, the predefined selection of camera positions, which does not account for the object's geometry, limits the effective use of texture information synthesized from different viewpoints, ultimately degrading overall texture consistency. In TexTailor, we address these issues by (1) applying a resampling scheme that repeatedly integrates information from previously synthesized textures within the diffusion process, and (2) fine-tuning a depth-aware diffusion model on these resampled textures. During this process, we observed that using only a few training images restricts the model's original ability to generate high-fidelity images aligned with the conditioning, and therefore propose an performance preservation loss to mitigate this issue. Additionally, we improve the synthesis of view-consistent textures by adaptively adjusting camera positions based on the object's geometry. Experiments on a subset of the Objaverse dataset and the ShapeNet car dataset demonstrate that TexTailor outperforms state-of-the-art methods in synthesizing view-consistent textures. The source code for TexTailor is available at https://github.com/Adios42/Textailor", "date": "2025-06-12", "categories": []} {"id": "2506.09980", "title": "Efficient Part-level 3D Object Generation via Dual Volume Packing", "abstract": "Recent progress in 3D object generation has greatly improved both the quality and efficiency. However, most existing methods generate a single mesh with all parts fused together, which limits the ability to edit or manipulate individual parts. A key challenge is that different objects may have a varying number of parts. To address this, we propose a new end-to-end framework for part-level 3D object generation. Given a single input image, our method generates high-quality 3D objects with an arbitrary number of complete and semantically meaningful parts. We introduce a dual volume packing strategy that organizes all parts into two complementary volumes, allowing for the creation of complete and interleaved parts that assemble into the final object. Experiments show that our model achieves better quality, diversity, and generalization than previous image-based part-level generation methods.", "date": "2025-06-11", "categories": []} {"id": "2506.09982", "title": "AnimateAnyMesh: A Feed-Forward 4D Foundation Model for Text-Driven Universal Mesh Animation", "abstract": "Recent advances in 4D content generation have attracted increasing attention, yet creating high-quality animated 3D models remains challenging due to the complexity of modeling spatio-temporal distributions and the scarcity of 4D training data. In this paper, we present AnimateAnyMesh, the first feed-forward framework that enables efficient text-driven animation of arbitrary 3D meshes. Our approach leverages a novel DyMeshVAE architecture that effectively compresses and reconstructs dynamic mesh sequences by disentangling spatial and temporal features while preserving local topological structures. To enable high-quality text-conditional generation, we employ a Rectified Flow-based training strategy in the compressed latent space. Additionally, we contribute the DyMesh Dataset, containing over 4M diverse dynamic mesh sequences with text annotations. Experimental results demonstrate that our method generates semantically accurate and temporally coherent mesh animations in a few seconds, significantly outperforming existing approaches in both quality and efficiency. Our work marks a substantial step forward in making 4D content creation more accessible and practical. All the data, code, and models will be open-released.", "date": "2025-06-11", "categories": []} {"id": "2505.04481", "title": "CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation", "abstract": "Recently, Large Language Models (LLMs) have achieved significant success, prompting increased interest in expanding their generative capabilities beyond general text into domain-specific areas. This study investigates the generation of parametric sequences for computer-aided design (CAD) models using LLMs. This endeavor represents an initial step towards creating parametric 3D shapes with LLMs, as CAD model parameters directly correlate with shapes in three-dimensional space. Despite the formidable generative capacities of LLMs, this task remains challenging, as these models neither encounter parametric sequences during their pretraining phase nor possess direct awareness of 3D structures. To address this, we present CAD-Llama, a framework designed to enhance pretrained LLMs for generating parametric 3D CAD models. Specifically, we develop a hierarchical annotation pipeline and a code-like format to translate parametric 3D CAD command sequences into Structured Parametric CAD Code (SPCC), incorporating hierarchical semantic descriptions. Furthermore, we propose an adaptive pretraining approach utilizing SPCC, followed by an instruction tuning process aligned with CAD-specific guidelines. This methodology aims to equip LLMs with the spatial knowledge inherent in parametric sequences. Experimental results demonstrate that our framework significantly outperforms prior autoregressive methods and existing LLM baselines.", "date": "2025-05-07", "categories": []} {"id": "2408.14732", "title": "OctFusion: Octree-based Diffusion Models for 3D Shape Generation", "abstract": "Diffusion models have emerged as a popular method for 3D generation. However, it is still challenging for diffusion models to efficiently generate diverse and high-quality 3D shapes. In this paper, we introduce OctFusion, which can generate 3D shapes with arbitrary resolutions in 2.5 seconds on a single Nvidia 4090 GPU, and the extracted meshes are guaranteed to be continuous and manifold. The key components of OctFusion are the octree-based latent representation and the accompanying diffusion models. The representation combines the benefits of both implicit neural representations and explicit spatial octrees and is learned with an octree-based variational autoencoder. The proposed diffusion model is a unified multi-scale U-Net that enables weights and computation sharing across different octree levels and avoids the complexity of widely used cascaded diffusion schemes. We verify the effectiveness of OctFusion on the ShapeNet and Objaverse datasets and achieve state-of-the-art performances on shape generation tasks. We demonstrate that OctFusion is extendable and flexible by generating high-quality color fields for textured mesh generation and high-quality 3D shapes conditioned on text prompts, sketches, or category labels. Our code and pre-trained models are available at https://github.com/octree-nn/octfusion.", "date": "2024-08-27", "categories": []} {"id": "2412.11596", "title": "MeshArt: Generating Articulated Meshes with Structure-Guided Transformers", "abstract": "Articulated 3D object generation is fundamental for creating realistic, functional, and interactable virtual assets which are not simply static. We introduce MeshArt, a hierarchical transformer-based approach to generate articulated 3D meshes with clean, compact geometry, reminiscent of human-crafted 3D models. We approach articulated mesh generation in a part-by-part fashion across two stages. First, we generate a high-level articulation-aware object structure; then, based on this structural information, we synthesize each part's mesh faces. Key to our approach is modeling both articulation structures and part meshes as sequences of quantized triangle embeddings, leading to a unified hierarchical framework with transformers for autoregressive generation. Object part structures are first generated as their bounding primitives and articulation modes; a second transformer, guided by these articulation structures, then generates each part's mesh triangles. To ensure coherency among generated parts, we introduce structure-guided conditioning that also incorporates local part mesh connectivity. MeshArt shows significant improvements over state of the art, with 57.1% improvement in structure coverage and a 209-point improvement in mesh generation FID.", "date": "2024-12-16", "categories": []} {"id": "2506.07091", "title": "SceneLCM: End-to-End Layout-Guided Interactive Indoor Scene Generation with Latent Consistency Model", "abstract": "Our project page: https://scutyklin.github.io/SceneLCM/. Automated generation of complex, interactive indoor scenes tailored to user prompt remains a formidable challenge. While existing methods achieve indoor scene synthesis, they struggle with rigid editing constraints, physical incoherence, excessive human effort, single-room limitations, and suboptimal material quality. To address these limitations, we propose SceneLCM, an end-to-end framework that synergizes Large Language Model (LLM) for layout design with Latent Consistency Model(LCM) for scene optimization. Our approach decomposes scene generation into four modular pipelines: (1) Layout Generation. We employ LLM-guided 3D spatial reasoning to convert textual descriptions into parametric blueprints(3D layout). And an iterative programmatic validation mechanism iteratively refines layout parameters through LLM-mediated dialogue loops; (2) Furniture Generation. SceneLCM employs Consistency Trajectory Sampling(CTS), a consistency distillation sampling loss guided by LCM, to form fast, semantically rich, and high-quality representations. We also offer two theoretical justification to demonstrate that our CTS loss is equivalent to consistency loss and its distillation error is bounded by the truncation error of the Euler solver; (3) Environment Optimization. We use a multiresolution texture field to encode the appearance of the scene, and optimize via CTS loss. To maintain cross-geometric texture coherence, we introduce a normal-aware cross-attention decoder to predict RGB by cross-attending to the anchors locations in geometrically heterogeneous instance. (4)Physically Editing. SceneLCM supports physically editing by integrating physical simulation, achieved persistent physical realism. Extensive experiments validate SceneLCM's superiority over state-of-the-art techniques, showing its wide-ranging potential for diverse applications.", "date": "2025-06-08", "categories": []} {"id": "2506.07698", "title": "NOVA3D: Normal Aligned Video Diffusion Model for Single Image to 3D Generation", "abstract": "3D AI-generated content (AIGC) has made it increasingly accessible for anyone to become a 3D content creator. While recent methods leverage Score Distillation Sampling to distill 3D objects from pretrained image diffusion models, they often suffer from inadequate 3D priors, leading to insufficient multi-view consistency. In this work, we introduce NOVA3D, an innovative single-image-to-3D generation framework. Our key insight lies in leveraging strong 3D priors from a pretrained video diffusion model and integrating geometric information during multi-view video fine-tuning. To facilitate information exchange between color and geometric domains, we propose the Geometry-Temporal Alignment (GTA) attention mechanism, thereby improving generalization and multi-view consistency. Moreover, we introduce the de-conflict geometry fusion algorithm, which improves texture fidelity by addressing multi-view inaccuracies and resolving discrepancies in pose alignment. Extensive experiments validate the superiority of NOVA3D over existing baselines.", "date": "2025-06-09", "categories": []} {"id": "2412.05278", "title": "Birth and Death of a Rose", "abstract": "We study the problem of generating temporal object intrinsics -- temporally evolving sequences of object geometry, reflectance, and texture, such as a blooming rose -- from pre-trained 2D foundation models. Unlike conventional 3D modeling and animation techniques that require extensive manual effort and expertise, we introduce a method that generates such assets with signals distilled from pre-trained 2D diffusion models. To ensure the temporal consistency of object intrinsics, we propose Neural Templates for temporal-state-guided distillation, derived automatically from image features from self-supervised learning. Our method can generate high-quality temporal object intrinsics for several natural phenomena and enable the sampling and controllable rendering of these dynamic objects from any viewpoint, under any environmental lighting conditions, at any time of their lifespan. Project website: https://chen-geng.com/rose4d", "date": "2024-12-06", "categories": []} {"id": "2412.09625", "title": "Illusion3D: 3D Multiview Illusion with 2D Diffusion Priors", "abstract": "Automatically generating multiview illusions is a compelling challenge, where a single piece of visual content offers distinct interpretations from different viewing perspectives. Traditional methods, such as shadow art and wire art, create interesting 3D illusions but are limited to simple visual outputs (i.e., figure-ground or line drawing), restricting their artistic expressiveness and practical versatility. Recent diffusion-based illusion generation methods can generate more intricate designs but are confined to 2D images. In this work, we present a simple yet effective approach for creating 3D multiview illusions based on user-provided text prompts or images. Our method leverages a pre-trained text-to-image diffusion model to optimize the textures and geometry of neural 3D representations through differentiable rendering. When viewed from multiple angles, this produces different interpretations. We develop several techniques to improve the quality of the generated 3D multiview illusions. We demonstrate the effectiveness of our approach through extensive experiments and showcase illusion generation with diverse 3D forms.", "date": "2024-12-12", "categories": []} {"id": "2506.05449", "title": "AI-powered Contextual 3D Environment Generation: A Systematic Review", "abstract": "The generation of high-quality 3D environments is crucial for industries such as gaming, virtual reality, and cinema, yet remains resource-intensive due to the reliance on manual processes. This study performs a systematic review of existing generative AI techniques for 3D scene generation, analyzing their characteristics, strengths, limitations, and potential for improvement. By examining state-of-the-art approaches, it presents key challenges such as scene authenticity and the influence of textual inputs. Special attention is given to how AI can blend different stylistic domains while maintaining coherence, the impact of training data on output quality, and the limitations of current models. In addition, this review surveys existing evaluation metrics for assessing realism and explores how industry professionals incorporate AI into their workflows. The findings of this study aim to provide a comprehensive understanding of the current landscape and serve as a foundation for future research on AI-driven 3D content generation. Key findings include that advanced generative architectures enable high-quality 3D content creation at a high computational cost, effective multi-modal integration techniques like cross-attention and latent space alignment facilitate text-to-3D tasks, and the quality and diversity of training data combined with comprehensive evaluation metrics are critical to achieving scalable, robust 3D scene generation.", "date": "2025-06-05", "categories": []} {"id": "2506.05573", "title": "PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers", "abstract": "We introduce PartCrafter, the first structured 3D generative model that jointly synthesizes multiple semantically meaningful and geometrically distinct 3D meshes from a single RGB image. Unlike existing methods that either produce monolithic 3D shapes or follow two-stage pipelines, i.e., first segmenting an image and then reconstructing each segment, PartCrafter adopts a unified, compositional generation architecture that does not rely on pre-segmented inputs. Conditioned on a single image, it simultaneously denoises multiple 3D parts, enabling end-to-end part-aware generation of both individual objects and complex multi-object scenes. PartCrafter builds upon a pretrained 3D mesh diffusion transformer (DiT) trained on whole objects, inheriting the pretrained weights, encoder, and decoder, and introduces two key innovations: (1) A compositional latent space, where each 3D part is represented by a set of disentangled latent tokens; (2) A hierarchical attention mechanism that enables structured information flow both within individual parts and across all parts, ensuring global coherence while preserving part-level detail during generation. To support part-level supervision, we curate a new dataset by mining part-level annotations from large-scale 3D object datasets. Experiments show that PartCrafter outperforms existing approaches in generating decomposable 3D meshes, including parts that are not directly visible in input images, demonstrating the strength of part-aware generative priors for 3D understanding and synthesis. Code and training data will be released.", "date": "2025-06-05", "categories": []} {"id": "2506.04351", "title": "HuGeDiff: 3D Human Generation via Diffusion with Gaussian Splatting", "abstract": "3D human generation is an important problem with a wide range of applications in computer vision and graphics. Despite recent progress in generative AI such as diffusion models or rendering methods like Neural Radiance Fields or Gaussian Splatting, controlling the generation of accurate 3D humans from text prompts remains an open challenge. Current methods struggle with fine detail, accurate rendering of hands and faces, human realism, and controlability over appearance. The lack of diversity, realism, and annotation in human image data also remains a challenge, hindering the development of a foundational 3D human model. We present a weakly supervised pipeline that tries to address these challenges. In the first step, we generate a photorealistic human image dataset with controllable attributes such as appearance, race, gender, etc using a state-of-the-art image diffusion model. Next, we propose an efficient mapping approach from image features to 3D point clouds using a transformer-based architecture. Finally, we close the loop by training a point-cloud diffusion model that is conditioned on the same text prompts used to generate the original samples. We demonstrate orders-of-magnitude speed-ups in 3D human generation compared to the state-of-the-art approaches, along with significantly improved text-prompt alignment, realism, and rendering quality. We will make the code and dataset available.", "date": "2025-06-04", "categories": []} {"id": "2506.04606", "title": "SmartAvatar: Text- and Image-Guided Human Avatar Generation with VLM AI Agents", "abstract": "SmartAvatar is a vision-language-agent-driven framework for generating fully rigged, animation-ready 3D human avatars from a single photo or textual prompt. While diffusion-based methods have made progress in general 3D object generation, they continue to struggle with precise control over human identity, body shape, and animation readiness. In contrast, SmartAvatar leverages the commonsense reasoning capabilities of large vision-language models (VLMs) in combination with off-the-shelf parametric human generators to deliver high-quality, customizable avatars. A key innovation is an autonomous verification loop, where the agent renders draft avatars, evaluates facial similarity, anatomical plausibility, and prompt alignment, and iteratively adjusts generation parameters for convergence. This interactive, AI-guided refinement process promotes fine-grained control over both facial and body features, enabling users to iteratively refine their avatars via natural-language conversations. Unlike diffusion models that rely on static pre-trained datasets and offer limited flexibility, SmartAvatar brings users into the modeling loop and ensures continuous improvement through an LLM-driven procedural generation and verification system. The generated avatars are fully rigged and support pose manipulation with consistent identity and appearance, making them suitable for downstream animation and interactive applications. Quantitative benchmarks and user studies demonstrate that SmartAvatar outperforms recent text- and image-driven avatar generation systems in terms of reconstructed mesh quality, identity fidelity, attribute accuracy, and animation readiness, making it a versatile tool for realistic, customizable avatar creation on consumer-grade hardware.", "date": "2025-06-05", "categories": []} {"id": "2506.04225", "title": "Voyager: Long-Range and World-Consistent Video Diffusion for Explorable 3D Scene Generation", "abstract": "Real-world applications like video gaming and virtual reality often demand the ability to model 3D scenes that users can explore along custom camera trajectories. While significant progress has been made in generating 3D objects from text or images, creating long-range, 3D-consistent, explorable 3D scenes remains a complex and challenging problem. In this work, we present Voyager, a novel video diffusion framework that generates world-consistent 3D point-cloud sequences from a single image with user-defined camera path. Unlike existing approaches, Voyager achieves end-to-end scene generation and reconstruction with inherent consistency across frames, eliminating the need for 3D reconstruction pipelines (e.g., structure-from-motion or multi-view stereo). Our method integrates three key components: 1) World-Consistent Video Diffusion: A unified architecture that jointly generates aligned RGB and depth video sequences, conditioned on existing world observation to ensure global coherence 2) Long-Range World Exploration: An efficient world cache with point culling and an auto-regressive inference with smooth video sampling for iterative scene extension with context-aware consistency, and 3) Scalable Data Engine: A video reconstruction pipeline that automates camera pose estimation and metric depth prediction for arbitrary videos, enabling large-scale, diverse training data curation without manual 3D annotations. Collectively, these designs result in a clear improvement over existing methods in visual quality and geometric accuracy, with versatile applications.", "date": "2025-06-04", "categories": []} {"id": "2505.04185", "title": "S3D: Sketch-Driven 3D Model Generation", "abstract": "Generating high-quality 3D models from 2D sketches is a challenging task due to the inherent ambiguity and sparsity of sketch data. In this paper, we present S3D, a novel framework that converts simple hand-drawn sketches into detailed 3D models. Our method utilizes a U-Net-based encoder-decoder architecture to convert sketches into face segmentation masks, which are then used to generate a 3D representation that can be rendered from novel views. To ensure robust consistency between the sketch domain and the 3D output, we introduce a novel style-alignment loss that aligns the U-Net bottleneck features with the initial encoder outputs of the 3D generation module, significantly enhancing reconstruction fidelity. To further enhance the network's robustness, we apply augmentation techniques to the sketch dataset. This streamlined framework demonstrates the effectiveness of S3D in generating high-quality 3D models from sketch inputs. The source code for this project is publicly available at https://github.com/hailsong/S3D.", "date": "2025-05-07", "categories": []} {"id": "2506.00512", "title": "Pro3D-Editor : A Progressive-Views Perspective for Consistent and Precise 3D Editing", "abstract": "Text-guided 3D editing aims to precisely edit semantically relevant local 3D regions, which has significant potential for various practical applications ranging from 3D games to film production. Existing methods typically follow a view-indiscriminate paradigm: editing 2D views indiscriminately and projecting them back into 3D space. However, they overlook the different cross-view interdependencies, resulting in inconsistent multi-view editing. In this study, we argue that ideal consistent 3D editing can be achieved through a \\textit{progressive-views paradigm}, which propagates editing semantics from the editing-salient view to other editing-sparse views. Specifically, we propose \\textit{Pro3D-Editor}, a novel framework, which mainly includes Primary-view Sampler, Key-view Render, and Full-view Refiner. Primary-view Sampler dynamically samples and edits the most editing-salient view as the primary view. Key-view Render accurately propagates editing semantics from the primary view to other key views through its Mixture-of-View-Experts Low-Rank Adaption (MoVE-LoRA). Full-view Refiner edits and refines the 3D object based on the edited multi-views. Extensive experiments demonstrate that our method outperforms existing methods in editing accuracy and spatial consistency.", "date": "2025-05-31", "categories": []} {"id": "2506.02620", "title": "FlexPainter: Flexible and Multi-View Consistent Texture Generation", "abstract": "Texture map production is an important part of 3D modeling and determines the rendering quality. Recently, diffusion-based methods have opened a new way for texture generation. However, restricted control flexibility and limited prompt modalities may prevent creators from producing desired results. Furthermore, inconsistencies between generated multi-view images often lead to poor texture generation quality. To address these issues, we introduce \\textbf{FlexPainter}, a novel texture generation pipeline that enables flexible multi-modal conditional guidance and achieves highly consistent texture generation. A shared conditional embedding space is constructed to perform flexible aggregation between different input modalities. Utilizing such embedding space, we present an image-based CFG method to decompose structural and style information, achieving reference image-based stylization. Leveraging the 3D knowledge within the image diffusion prior, we first generate multi-view images simultaneously using a grid representation to enhance global understanding. Meanwhile, we propose a view synchronization and adaptive weighting module during diffusion sampling to further ensure local consistency. Finally, a 3D-aware texture completion model combined with a texture enhancement model is used to generate seamless, high-resolution texture maps. Comprehensive experiments demonstrate that our framework significantly outperforms state-of-the-art methods in both flexibility and generation quality.", "date": "2025-06-03", "categories": []} {"id": "2506.02781", "title": "FreeScene: Mixed Graph Diffusion for 3D Scene Synthesis from Free Prompts", "abstract": "Controllability plays a crucial role in the practical applications of 3D indoor scene synthesis. Existing works either allow rough language-based control, that is convenient but lacks fine-grained scene customization, or employ graph based control, which offers better controllability but demands considerable knowledge for the cumbersome graph design process. To address these challenges, we present FreeScene, a user-friendly framework that enables both convenient and effective control for indoor scene synthesis.Specifically, FreeScene supports free-form user inputs including text description and/or reference images, allowing users to express versatile design intentions. The user inputs are adequately analyzed and integrated into a graph representation by a VLM-based Graph Designer. We then propose MG-DiT, a Mixed Graph Diffusion Transformer, which performs graph-aware denoising to enhance scene generation. Our MG-DiT not only excels at preserving graph structure but also offers broad applicability to various tasks, including, but not limited to, text-to-scene, graph-to-scene, and rearrangement, all within a single model. Extensive experiments demonstrate that FreeScene provides an efficient and user-friendly solution that unifies text-based and graph based scene synthesis, outperforming state-of-the-art methods in terms of both generation quality and controllability in a range of applications.", "date": "2025-06-03", "categories": []} {"id": "2410.00890", "title": "Flex3D: Feed-Forward 3D Generation with Flexible Reconstruction Model and Input View Curation", "abstract": "Generating high-quality 3D content from text, single images, or sparse view images remains a challenging task with broad applications. Existing methods typically employ multi-view diffusion models to synthesize multi-view images, followed by a feed-forward process for 3D reconstruction. However, these approaches are often constrained by a small and fixed number of input views, limiting their ability to capture diverse viewpoints and, even worse, leading to suboptimal generation results if the synthesized views are of poor quality. To address these limitations, we propose Flex3D, a novel two-stage framework capable of leveraging an arbitrary number of high-quality input views. The first stage consists of a candidate view generation and curation pipeline. We employ a fine-tuned multi-view image diffusion model and a video diffusion model to generate a pool of candidate views, enabling a rich representation of the target 3D object. Subsequently, a view selection pipeline filters these views based on quality and consistency, ensuring that only the high-quality and reliable views are used for reconstruction. In the second stage, the curated views are fed into a Flexible Reconstruction Model (FlexRM), built upon a transformer architecture that can effectively process an arbitrary number of inputs. FlemRM directly outputs 3D Gaussian points leveraging a tri-plane representation, enabling efficient and detailed 3D generation. Through extensive exploration of design and training strategies, we optimize FlexRM to achieve superior performance in both reconstruction and generation tasks. Our results demonstrate that Flex3D achieves state-of-the-art performance, with a user study winning rate of over 92% in 3D generation tasks when compared to several of the latest feed-forward 3D generative models.", "date": "2024-10-01", "categories": []} {"id": "2506.00742", "title": "ArtiScene: Language-Driven Artistic 3D Scene Generation Through Image Intermediary", "abstract": "Designing 3D scenes is traditionally a challenging task that demands both artistic expertise and proficiency with complex software. Recent advances in text-to-3D generation have greatly simplified this process by letting users create scenes based on simple text descriptions. However, as these methods generally require extra training or in-context learning, their performance is often hindered by the limited availability of high-quality 3D data. In contrast, modern text-to-image models learned from web-scale images can generate scenes with diverse, reliable spatial layouts and consistent, visually appealing styles. Our key insight is that instead of learning directly from 3D scenes, we can leverage generated 2D images as an intermediary to guide 3D synthesis. In light of this, we introduce ArtiScene, a training-free automated pipeline for scene design that integrates the flexibility of free-form text-to-image generation with the diversity and reliability of 2D intermediary layouts.\n First, we generate 2D images from a scene description, then extract the shape and appearance of objects to create 3D models. These models are assembled into the final scene using geometry, position, and pose information derived from the same intermediary image. Being generalizable to a wide range of scenes and styles, ArtiScene outperforms state-of-the-art benchmarks by a large margin in layout and aesthetic quality by quantitative metrics. It also averages a 74.89% winning rate in extensive user studies and 95.07% in GPT-4o evaluation. Project page: https://artiscene-cvpr.github.io/", "date": "2025-05-31", "categories": []} {"id": "2506.01853", "title": "ShapeLLM-Omni: A Native Multimodal LLM for 3D Generation and Understanding", "abstract": "Recently, the powerful text-to-image capabilities of ChatGPT-4o have led to growing appreciation for native multimodal large language models. However, its multimodal capabilities remain confined to images and text. Yet beyond images, the ability to understand and generate 3D content is equally crucial. To address this gap, we propose ShapeLLM-Omni-a native 3D large language model capable of understanding and generating 3D assets and text in any sequence. First, we train a 3D vector-quantized variational autoencoder (VQVAE), which maps 3D objects into a discrete latent space to achieve efficient and accurate shape representation and reconstruction. Building upon the 3D-aware discrete tokens, we innovatively construct a large-scale continuous training dataset named 3D-Alpaca, encompassing generation, comprehension, and editing, thus providing rich resources for future research and training. Finally, by performing instruction-based training of the Qwen-2.5-vl-7B-Instruct model on the 3D-Alpaca dataset. Our work provides an effective attempt at extending multimodal models with basic 3D capabilities, which contributes to future research in 3D-native AI. Project page: https://github.com/JAMESYJL/ShapeLLM-Omni", "date": "2025-06-02", "categories": []} {"id": "2405.14979", "title": "CraftsMan3D: High-fidelity Mesh Generation with 3D Native Generation and Interactive Geometry Refiner", "abstract": "We present a novel generative 3D modeling system, coined CraftsMan, which can generate high-fidelity 3D geometries with highly varied shapes, regular mesh topologies, and detailed surfaces, and, notably, allows for refining the geometry in an interactive manner. Despite the significant advancements in 3D generation, existing methods still struggle with lengthy optimization processes, irregular mesh topologies, noisy surfaces, and difficulties in accommodating user edits, consequently impeding their widespread adoption and implementation in 3D modeling software. Our work is inspired by the craftsman, who usually roughs out the holistic figure of the work first and elaborates the surface details subsequently. Specifically, we employ a 3D native diffusion model, which operates on latent space learned from latent set-based 3D representations, to generate coarse geometries with regular mesh topology in seconds. In particular, this process takes as input a text prompt or a reference image and leverages a powerful multi-view (MV) diffusion model to generate multiple views of the coarse geometry, which are fed into our MV-conditioned 3D diffusion model for generating the 3D geometry, significantly improving robustness and generalizability. Following that, a normal-based geometry refiner is used to significantly enhance the surface details. This refinement can be performed automatically, or interactively with user-supplied edits. Extensive experiments demonstrate that our method achieves high efficacy in producing superior-quality 3D assets compared to existing methods. HomePage: https://craftsman3d.github.io/, Code: https://github.com/wyysf-98/CraftsMan", "date": "2024-05-23", "categories": []} {"id": "2412.01506", "title": "Structured 3D Latents for Scalable and Versatile 3D Generation", "abstract": "We introduce a novel 3D generation method for versatile and high-quality 3D asset creation. The cornerstone is a unified Structured LATent (SLAT) representation which allows decoding to different output formats, such as Radiance Fields, 3D Gaussians, and meshes. This is achieved by integrating a sparsely-populated 3D grid with dense multiview visual features extracted from a powerful vision foundation model, comprehensively capturing both structural (geometry) and textural (appearance) information while maintaining flexibility during decoding. We employ rectified flow transformers tailored for SLAT as our 3D generation models and train models with up to 2 billion parameters on a large 3D asset dataset of 500K diverse objects. Our model generates high-quality results with text or image conditions, significantly surpassing existing methods, including recent ones at similar scales. We showcase flexible output format selection and local 3D editing capabilities which were not offered by previous models. Code, model, and data will be released.", "date": "2024-12-02", "categories": []} {"id": "2505.24245", "title": "LTM3D: Bridging Token Spaces for Conditional 3D Generation with Auto-Regressive Diffusion Framework", "abstract": "We present LTM3D, a Latent Token space Modeling framework for conditional 3D shape generation that integrates the strengths of diffusion and auto-regressive (AR) models. While diffusion-based methods effectively model continuous latent spaces and AR models excel at capturing inter-token dependencies, combining these paradigms for 3D shape generation remains a challenge. To address this, LTM3D features a Conditional Distribution Modeling backbone, leveraging a masked autoencoder and a diffusion model to enhance token dependency learning. Additionally, we introduce Prefix Learning, which aligns condition tokens with shape latent tokens during generation, improving flexibility across modalities. We further propose a Latent Token Reconstruction module with Reconstruction-Guided Sampling to reduce uncertainty and enhance structural fidelity in generated shapes. Our approach operates in token space, enabling support for multiple 3D representations, including signed distance fields, point clouds, meshes, and 3D Gaussian Splatting. Extensive experiments on image- and text-conditioned shape generation tasks demonstrate that LTM3D outperforms existing methods in prompt fidelity and structural accuracy while offering a generalizable framework for multi-modal, multi-representation 3D generation.", "date": "2025-05-30", "categories": []} {"id": "2505.24877", "title": "AdaHuman: Animatable Detailed 3D Human Generation with Compositional Multiview Diffusion", "abstract": "Existing methods for image-to-3D avatar generation struggle to produce highly detailed, animation-ready avatars suitable for real-world applications. We introduce AdaHuman, a novel framework that generates high-fidelity animatable 3D avatars from a single in-the-wild image. AdaHuman incorporates two key innovations: (1) A pose-conditioned 3D joint diffusion model that synthesizes consistent multi-view images in arbitrary poses alongside corresponding 3D Gaussian Splats (3DGS) reconstruction at each diffusion step; (2) A compositional 3DGS refinement module that enhances the details of local body parts through image-to-image refinement and seamlessly integrates them using a novel crop-aware camera ray map, producing a cohesive detailed 3D avatar. These components allow AdaHuman to generate highly realistic standardized A-pose avatars with minimal self-occlusion, enabling rigging and animation with any input motion. Extensive evaluation on public benchmarks and in-the-wild images demonstrates that AdaHuman significantly outperforms state-of-the-art methods in both avatar reconstruction and reposing. Code and models will be publicly available for research purposes.", "date": "2025-05-30", "categories": []} {"id": "2505.23253", "title": "UniTEX: Universal High Fidelity Generative Texturing for 3D Shapes", "abstract": "We present UniTEX, a novel two-stage 3D texture generation framework to create high-quality, consistent textures for 3D assets. Existing approaches predominantly rely on UV-based inpainting to refine textures after reprojecting the generated multi-view images onto the 3D shapes, which introduces challenges related to topological ambiguity. To address this, we propose to bypass the limitations of UV mapping by operating directly in a unified 3D functional space. Specifically, we first propose that lifts texture generation into 3D space via Texture Functions (TFs)--a continuous, volumetric representation that maps any 3D point to a texture value based solely on surface proximity, independent of mesh topology. Then, we propose to predict these TFs directly from images and geometry inputs using a transformer-based Large Texturing Model (LTM). To further enhance texture quality and leverage powerful 2D priors, we develop an advanced LoRA-based strategy for efficiently adapting large-scale Diffusion Transformers (DiTs) for high-quality multi-view texture synthesis as our first stage. Extensive experiments demonstrate that UniTEX achieves superior visual quality and texture integrity compared to existing approaches, offering a generalizable and scalable solution for automated 3D texture generation. Code will available in: https://github.com/YixunLiang/UniTEX.", "date": "2025-05-29", "categories": []} {"id": "2505.23287", "title": "GenCAD-Self-Repairing: Feasibility Enhancement for 3D CAD Generation", "abstract": "With the advancement of generative AI, research on its application to 3D model generation has gained traction, particularly in automating the creation of Computer-Aided Design (CAD) files from images. GenCAD is a notable model in this domain, leveraging an autoregressive transformer-based architecture with a contrastive learning framework to generate CAD programs.\n However, a major limitation of GenCAD is its inability to consistently produce feasible boundary representations (B-reps), with approximately 10% of generated designs being infeasible. To address this, we propose GenCAD-Self-Repairing, a framework that enhances the feasibility of generative CAD models through diffusion guidance and a self-repairing pipeline. This framework integrates a guided diffusion denoising process in the latent space and a regression-based correction mechanism to refine infeasible CAD command sequences while preserving geometric accuracy. Our approach successfully converted two-thirds of infeasible designs in the baseline method into feasible ones, significantly improving the feasibility rate while simultaneously maintaining a reasonable level of geometric accuracy between the point clouds of ground truth models and generated models.\n By significantly improving the feasibility rate of generating CAD models, our approach helps expand the availability of high-quality training data and enhances the applicability of AI-driven CAD generation in manufacturing, architecture, and product design.", "date": "2025-05-29", "categories": []} {"id": "2505.20460", "title": "DIPO: Dual-State Images Controlled Articulated Object Generation Powered by Diverse Data", "abstract": "We present DIPO, a novel framework for the controllable generation of articulated 3D objects from a pair of images: one depicting the object in a resting state and the other in an articulated state. Compared to the single-image approach, our dual-image input imposes only a modest overhead for data collection, but at the same time provides important motion information, which is a reliable guide for predicting kinematic relationships between parts. Specifically, we propose a dual-image diffusion model that captures relationships between the image pair to generate part layouts and joint parameters. In addition, we introduce a Chain-of-Thought (CoT) based graph reasoner that explicitly infers part connectivity relationships. To further improve robustness and generalization on complex articulated objects, we develop a fully automated dataset expansion pipeline, name LEGO-Art, that enriches the diversity and complexity of PartNet-Mobility dataset. We propose PM-X, a large-scale dataset of complex articulated 3D objects, accompanied by rendered images, URDF annotations, and textual descriptions. Extensive experiments demonstrate that DIPO significantly outperforms existing baselines in both the resting state and the articulated state, while the proposed PM-X dataset further enhances generalization to diverse and structurally complex articulated objects. Our code and dataset will be released to the community upon publication.", "date": "2025-05-26", "categories": []} {"id": "2505.21050", "title": "Advancing high-fidelity 3D and Texture Generation with 2.5D latents", "abstract": "Despite the availability of large-scale 3D datasets and advancements in 3D generative models, the complexity and uneven quality of 3D geometry and texture data continue to hinder the performance of 3D generation techniques. In most existing approaches, 3D geometry and texture are generated in separate stages using different models and non-unified representations, frequently leading to unsatisfactory coherence between geometry and texture. To address these challenges, we propose a novel framework for joint generation of 3D geometry and texture. Specifically, we focus in generate a versatile 2.5D representations that can be seamlessly transformed between 2D and 3D. Our approach begins by integrating multiview RGB, normal, and coordinate images into a unified representation, termed as 2.5D latents. Next, we adapt pre-trained 2D foundation models for high-fidelity 2.5D generation, utilizing both text and image conditions. Finally, we introduce a lightweight 2.5D-to-3D refiner-decoder framework that efficiently generates detailed 3D representations from 2.5D images. Extensive experiments demonstrate that our model not only excels in generating high-quality 3D objects with coherent structure and color from text and image inputs but also significantly outperforms existing methods in geometry-conditioned texture generation.", "date": "2025-05-27", "categories": []} {"id": "2505.19868", "title": "Harnessing the Power of Training-Free Techniques in Text-to-2D Generation for Text-to-3D Generation via Score Distillation Sampling", "abstract": "Recent studies show that simple training-free techniques can dramatically improve the quality of text-to-2D generation outputs, e.g. Classifier-Free Guidance (CFG) or FreeU. However, these training-free techniques have been underexplored in the lens of Score Distillation Sampling (SDS), which is a popular and effective technique to leverage the power of pretrained text-to-2D diffusion models for various tasks. In this paper, we aim to shed light on the effect such training-free techniques have on SDS, via a particular application of text-to-3D generation via 2D lifting. We present our findings, which show that varying the scales of CFG presents a trade-off between object size and surface smoothness, while varying the scales of FreeU presents a trade-off between texture details and geometric errors. Based on these findings, we provide insights into how we can effectively harness training-free techniques for SDS, via a strategic scaling of such techniques in a dynamic manner with respect to the timestep or optimization iteration step. We show that using our proposed scheme strikes a favorable balance between texture details and surface smoothness in text-to-3D generations, while preserving the size of the output and mitigating the occurrence of geometric defects.", "date": "2025-05-26", "categories": []} {"id": "2505.17412", "title": "Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention", "abstract": "Generating high-resolution 3D shapes using volumetric representations such as Signed Distance Functions (SDFs) presents substantial computational and memory challenges. We introduce Direct3D-S2, a scalable 3D generation framework based on sparse volumes that achieves superior output quality with dramatically reduced training costs. Our key innovation is the Spatial Sparse Attention (SSA) mechanism, which greatly enhances the efficiency of Diffusion Transformer (DiT) computations on sparse volumetric data. SSA allows the model to effectively process large token sets within sparse volumes, substantially reducing computational overhead and achieving a 3.9x speedup in the forward pass and a 9.6x speedup in the backward pass. Our framework also includes a variational autoencoder (VAE) that maintains a consistent sparse volumetric format across input, latent, and output stages. Compared to previous methods with heterogeneous representations in 3D VAE, this unified design significantly improves training efficiency and stability. Our model is trained on public available datasets, and experiments demonstrate that Direct3D-S2 not only surpasses state-of-the-art methods in generation quality and efficiency, but also enables training at 1024 resolution using only 8 GPUs, a task typically requiring at least 32 GPUs for volumetric representations at 256 resolution, thus making gigascale 3D generation both practical and accessible. Project page: https://www.neural4d.com/research/direct3d-s2.", "date": "2025-05-23", "categories": []} {"id": "2505.19492", "title": "ViewCraft3D: High-Fidelity and View-Consistent 3D Vector Graphics Synthesis", "abstract": "3D vector graphics play a crucial role in various applications including 3D shape retrieval, conceptual design, and virtual reality interactions due to their ability to capture essential structural information with minimal representation. While recent approaches have shown promise in generating 3D vector graphics, they often suffer from lengthy processing times and struggle to maintain view consistency. To address these limitations, we propose ViewCraft3D (VC3D), an efficient method that leverages 3D priors to generate 3D vector graphics. Specifically, our approach begins with 3D object analysis, employs a geometric extraction algorithm to fit 3D vector graphics to the underlying structure, and applies view-consistent refinement process to enhance visual quality. Our comprehensive experiments demonstrate that VC3D outperforms previous methods in both qualitative and quantitative evaluations, while significantly reducing computational overhead. The resulting 3D sketches maintain view consistency and effectively capture the essential characteristics of the original objects.", "date": "2025-05-26", "categories": []} {"id": "2410.07658", "title": "SeMv-3D: Towards Concurrency of Semantic and Multi-view Consistency in General Text-to-3D Generation", "abstract": "General Text-to-3D (GT23D) generation is crucial for creating diverse 3D content across objects and scenes, yet it faces two key challenges: 1) ensuring semantic consistency between input text and generated 3D models, and 2) maintaining multi-view consistency across different perspectives within 3D. Existing approaches typically address only one of these challenges, often leading to suboptimal results in semantic fidelity and structural coherence. To overcome these limitations, we propose SeMv-3D, a novel framework that jointly enhances semantic alignment and multi-view consistency in GT23D generation. At its core, we introduce Triplane Prior Learning (TPL), which effectively learns triplane priors by capturing spatial correspondences across three orthogonal planes using a dedicated Orthogonal Attention mechanism, thereby ensuring geometric consistency across viewpoints. Additionally, we present Prior-based Semantic Aligning in Triplanes (SAT), which enables consistent any-view synthesis by leveraging attention-based feature alignment to reinforce the correspondence between textual semantics and triplane representations. Extensive experiments demonstrate that our method sets a new state-of-the-art in multi-view consistency, while maintaining competitive performance in semantic consistency compared to methods focused solely on semantic alignment. These results emphasize the remarkable ability of our approach to effectively balance and excel in both dimensions, establishing a new benchmark in the field.", "date": "2024-10-10", "categories": []} {"id": "2502.03449", "title": "Dress-1-to-3: Single Image to Simulation-Ready 3D Outfit with Diffusion Prior and Differentiable Physics", "abstract": "Recent advances in large models have significantly advanced image-to-3D reconstruction. However, the generated models are often fused into a single piece, limiting their applicability in downstream tasks. This paper focuses on 3D garment generation, a key area for applications like virtual try-on with dynamic garment animations, which require garments to be separable and simulation-ready. We introduce Dress-1-to-3, a novel pipeline that reconstructs physics-plausible, simulation-ready separated garments with sewing patterns and humans from an in-the-wild image. Starting with the image, our approach combines a pre-trained image-to-sewing pattern generation model for creating coarse sewing patterns with a pre-trained multi-view diffusion model to produce multi-view images. The sewing pattern is further refined using a differentiable garment simulator based on the generated multi-view images. Versatile experiments demonstrate that our optimization approach substantially enhances the geometric alignment of the reconstructed 3D garments and humans with the input image. Furthermore, by integrating a texture generation module and a human motion generation module, we produce customized physics-plausible and realistic dynamic garment demonstrations. Project page: https://dress-1-to-3.github.io/", "date": "2025-02-05", "categories": []} {"id": "2503.07578", "title": "Denoising Score Distillation: From Noisy Diffusion Pretraining to One-Step High-Quality Generation", "abstract": "Diffusion models have achieved remarkable success in generating high-resolution, realistic images across diverse natural distributions. However, their performance heavily relies on high-quality training data, making it challenging to learn meaningful distributions from corrupted samples. This limitation restricts their applicability in scientific domains where clean data is scarce or costly to obtain. In this work, we introduce denoising score distillation (DSD), a surprisingly effective and novel approach for training high-quality generative models from low-quality data. DSD first pretrains a diffusion model exclusively on noisy, corrupted samples and then distills it into a one-step generator capable of producing refined, clean outputs. While score distillation is traditionally viewed as a method to accelerate diffusion models, we show that it can also significantly enhance sample quality, particularly when starting from a degraded teacher model. Across varying noise levels and datasets, DSD consistently improves generative performancewe summarize our empirical evidence in Fig. 1. Furthermore, we provide theoretical insights showing that, in a linear model setting, DSD identifies the eigenspace of the clean data distributions covariance matrix, implicitly regularizing the generator. This perspective reframes score distillation as not only a tool for efficiency but also a mechanism for improving generative models, particularly in low-quality data settings.", "date": "2025-03-10", "categories": []} {"id": "2505.15528", "title": "PlantDreamer: Achieving Realistic 3D Plant Models with Diffusion-Guided Gaussian Splatting", "abstract": "Recent years have seen substantial improvements in the ability to generate synthetic 3D objects using AI. However, generating complex 3D objects, such as plants, remains a considerable challenge. Current generative 3D models struggle with plant generation compared to general objects, limiting their usability in plant analysis tools, which require fine detail and accurate geometry. We introduce PlantDreamer, a novel approach to 3D synthetic plant generation, which can achieve greater levels of realism for complex plant geometry and textures than available text-to-3D models. To achieve this, our new generation pipeline leverages a depth ControlNet, fine-tuned Low-Rank Adaptation and an adaptable Gaussian culling algorithm, which directly improve textural realism and geometric integrity of generated 3D plant models. Additionally, PlantDreamer enables both purely synthetic plant generation, by leveraging L-System-generated meshes, and the enhancement of real-world plant point clouds by converting them into 3D Gaussian Splats. We evaluate our approach by comparing its outputs with state-of-the-art text-to-3D models, demonstrating that PlantDreamer outperforms existing methods in producing high-fidelity synthetic plants. Our results indicate that our approach not only advances synthetic plant generation, but also facilitates the upgrading of legacy point cloud datasets, making it a valuable tool for 3D phenotyping applications.", "date": "2025-05-21", "categories": []} {"id": "2505.13573", "title": "FreeMesh: Boosting Mesh Generation with Coordinates Merging", "abstract": "The next-coordinate prediction paradigm has emerged as the de facto standard in current auto-regressive mesh generation methods. Despite their effectiveness, there is no efficient measurement for the various tokenizers that serialize meshes into sequences. In this paper, we introduce a new metric Per-Token-Mesh-Entropy (PTME) to evaluate the existing mesh tokenizers theoretically without any training. Building upon PTME, we propose a plug-and-play tokenization technique called coordinate merging. It further improves the compression ratios of existing tokenizers by rearranging and merging the most frequent patterns of coordinates. Through experiments on various tokenization methods like MeshXL, MeshAnything V2, and Edgerunner, we further validate the performance of our method. We hope that the proposed PTME and coordinate merging can enhance the existing mesh tokenizers and guide the further development of native mesh generation.", "date": "2025-05-19", "categories": []} {"id": "2410.15891", "title": "TexPro: Text-guided PBR Texturing with Procedural Material Modeling", "abstract": "In this paper, we present TexPro, a novel method for high-fidelity material generation for input 3D meshes given text prompts. Unlike existing text-conditioned texture generation methods that typically generate RGB textures with baked lighting, TexPro is able to produce diverse texture maps via procedural material modeling, which enables physically-based rendering, relighting, and additional benefits inherent to procedural materials. Specifically, we first generate multi-view reference images given the input textual prompt by employing the latest text-to-image model. We then derive texture maps through rendering-based optimization with recent differentiable procedural materials. To this end, we design several techniques to handle the misalignment between the generated multi-view images and 3D meshes, and introduce a novel material agent that enhances material classification and matching by exploring both part-level understanding and object-aware material reasoning. Experiments demonstrate the superiority of the proposed method over existing SOTAs, and its capability of relighting.", "date": "2024-10-21", "categories": []} {"id": "2505.12635", "title": "MVPainter: Accurate and Detailed 3D Texture Generation via Multi-View Diffusion with Geometric Control", "abstract": "Recently, significant advances have been made in 3D object generation. Building upon the generated geometry, current pipelines typically employ image diffusion models to generate multi-view RGB images, followed by UV texture reconstruction through texture baking. While 3D geometry generation has improved significantly, supported by multiple open-source frameworks, 3D texture generation remains underexplored. In this work, we systematically investigate 3D texture generation through the lens of three core dimensions: reference-texture alignment, geometry-texture consistency, and local texture quality. To tackle these issues, we propose MVPainter, which employs data filtering and augmentation strategies to enhance texture fidelity and detail, and introduces ControlNet-based geometric conditioning to improve texture-geometry alignment. Furthermore, we extract physically-based rendering (PBR) attributes from the generated views to produce PBR meshes suitable for real-world rendering applications. MVPainter achieves state-of-the-art results across all three dimensions, as demonstrated by human-aligned evaluations. To facilitate further research and reproducibility, we also release our full pipeline as an open-source system, including data construction, model architecture, and evaluation tools.", "date": "2025-05-19", "categories": []} {"id": "2505.09998", "title": "From Air to Wear: Personalized 3D Digital Fashion with AR/VR Immersive 3D Sketching", "abstract": "In the era of immersive consumer electronics, such as AR/VR headsets and smart devices, people increasingly seek ways to express their identity through virtual fashion. However, existing 3D garment design tools remain inaccessible to everyday users due to steep technical barriers and limited data. In this work, we introduce a 3D sketch-driven 3D garment generation framework that empowers ordinary users - even those without design experience - to create high-quality digital clothing through simple 3D sketches in AR/VR environments. By combining a conditional diffusion model, a sketch encoder trained in a shared latent space, and an adaptive curriculum learning strategy, our system interprets imprecise, free-hand input and produces realistic, personalized garments. To address the scarcity of training data, we also introduce KO3DClothes, a new dataset of paired 3D garments and user-created sketches. Extensive experiments and user studies confirm that our method significantly outperforms existing baselines in both fidelity and usability, demonstrating its promise for democratized fashion design on next-generation consumer platforms.", "date": "2025-05-15", "categories": []} {"id": "2207.14425", "title": "3D Cartoon Face Generation with Controllable Expressions from a Single GAN Image", "abstract": "In this paper, we investigate an open research task of generating 3D cartoon face shapes from single 2D GAN generated human faces and without 3D supervision, where we can also manipulate the facial expressions of the 3D shapes. To this end, we discover the semantic meanings of StyleGAN latent space, such that we are able to produce face images of various expressions, poses, and lighting conditions by controlling the latent codes. Specifically, we first finetune the pretrained StyleGAN face model on the cartoon datasets. By feeding the same latent codes to face and cartoon generation models, we aim to realize the translation from 2D human face images to cartoon styled avatars. We then discover semantic directions of the GAN latent space, in an attempt to change the facial expressions while preserving the original identity. As we do not have any 3D annotations for cartoon faces, we manipulate the latent codes to generate images with different poses and lighting conditions, such that we can reconstruct the 3D cartoon face shapes. We validate the efficacy of our method on three cartoon datasets qualitatively and quantitatively.", "date": "2022-07-29", "categories": []} {"id": "2505.09140", "title": "TopoDiT-3D: Topology-Aware Diffusion Transformer with Bottleneck Structure for 3D Point Cloud Generation", "abstract": "Recent advancements in Diffusion Transformer (DiT) models have significantly improved 3D point cloud generation. However, existing methods primarily focus on local feature extraction while overlooking global topological information, such as voids, which are crucial for maintaining shape consistency and capturing complex geometries. To address this limitation, we propose TopoDiT-3D, a Topology-Aware Diffusion Transformer with a bottleneck structure for 3D point cloud generation. Specifically, we design the bottleneck structure utilizing Perceiver Resampler, which not only offers a mode to integrate topological information extracted through persistent homology into feature learning, but also adaptively filters out redundant local features to improve training efficiency. Experimental results demonstrate that TopoDiT-3D outperforms state-of-the-art models in visual quality, diversity, and training efficiency. Furthermore, TopoDiT-3D demonstrates the importance of rich topological information for 3D point cloud generation and its synergy with conventional local feature learning. Videos and code are available at https://github.com/Zechao-Guan/TopoDiT-3D.", "date": "2025-05-14", "categories": []} {"id": "2312.06198", "title": "Optimized View and Geometry Distillation from Multi-view Diffuser", "abstract": "Generating multi-view images from a single input view using image-conditioned diffusion models is a recent advancement and has shown considerable potential. However, issues such as the lack of consistency in synthesized views and over-smoothing in extracted geometry persist. Previous methods integrate multi-view consistency modules or impose additional supervisory to enhance view consistency while compromising on the flexibility of camera positioning and limiting the versatility of view synthesis. In this study, we consider the radiance field optimized during geometry extraction as a more rigid consistency prior, compared to volume and ray aggregation used in previous works. We further identify and rectify a critical bias in the traditional radiance field optimization process through score distillation from a multi-view diffuser. We introduce an Unbiased Score Distillation (USD) that utilizes unconditioned noises from a 2D diffusion model, greatly refining the radiance field fidelity. We leverage the rendered views from the optimized radiance field as the basis and develop a two-step specialization process of a 2D diffusion model, which is adept at conducting object-specific denoising and generating high-quality multi-view images. Finally, we recover faithful geometry and texture directly from the refined multi-view images. Empirical evaluations demonstrate that our optimized geometry and view distillation technique generates comparable results to the state-of-the-art models trained on extensive datasets, all while maintaining freedom in camera positioning. Please see our project page at https://youjiazhang.github.io/USD/.", "date": "2023-12-11", "categories": []} {"id": "2505.08531", "title": "Building-Block Aware Generative Modeling for 3D Crystals of Metal Organic Frameworks", "abstract": "Metal-organic frameworks (MOFs) marry inorganic nodes, organic edges, and topological nets into programmable porous crystals, yet their astronomical design space defies brute-force synthesis. Generative modeling holds ultimate promise, but existing models either recycle known building blocks or are restricted to small unit cells. We introduce Building-Block-Aware MOF Diffusion (BBA MOF Diffusion), an SE(3)-equivariant diffusion model that learns 3D all-atom representations of individual building blocks, encoding crystallographic topological nets explicitly. Trained on the CoRE-MOF database, BBA MOF Diffusion readily samples MOFs with unit cells containing 1000 atoms with great geometric validity, novelty, and diversity mirroring experimental databases. Its native building-block representation produces unprecedented metal nodes and organic edges, expanding accessible chemical space by orders of magnitude. One high-scoring [Zn(1,4-TDC)(EtOH)2] MOF predicted by the model was synthesized, where powder X-ray diffraction, thermogravimetric analysis, and N2 sorption confirm its structural fidelity. BBA-Diff thus furnishes a practical pathway to synthesizable and high-performing MOFs.", "date": "2025-05-13", "categories": []} {"id": "2504.14257", "title": "HoLa: B-Rep Generation using a Holistic Latent Representation", "abstract": "We introduce a novel representation for learning and generating Computer-Aided Design (CAD) models in the form of $\\textit{boundary representations}$ (B-Reps). Our representation unifies the continuous geometric properties of B-Rep primitives in different orders (e.g., surfaces and curves) and their discrete topological relations in a $\\textit{holistic latent}$ (HoLa) space. This is based on the simple observation that the topological connection between two surfaces is intrinsically tied to the geometry of their intersecting curve. Such a prior allows us to reformulate topology learning in B-Reps as a geometric reconstruction problem in Euclidean space. Specifically, we eliminate the presence of curves, vertices, and all the topological connections in the latent space by learning to distinguish and derive curve geometries from a pair of surface primitives via a neural intersection network. To this end, our holistic latent space is only defined on surfaces but encodes a full B-Rep model, including the geometry of surfaces, curves, vertices, and their topological relations. Our compact and holistic latent space facilitates the design of a first diffusion-based generator to take on a large variety of inputs including point clouds, single/multi-view images, 2D sketches, and text prompts. Our method significantly reduces ambiguities, redundancies, and incoherences among the generated B-Rep primitives, as well as training complexities inherent in prior multi-step B-Rep learning pipelines, while achieving greatly improved validity rate over current state of the art: 82% vs. $\\approx$50%.", "date": "2025-04-19", "categories": []} {"id": "2505.07747", "title": "Step1X-3D: Towards High-Fidelity and Controllable Generation of Textured 3D Assets", "abstract": "While generative artificial intelligence has advanced significantly across text, image, audio, and video domains, 3D generation remains comparatively underdeveloped due to fundamental challenges such as data scarcity, algorithmic limitations, and ecosystem fragmentation. To this end, we present Step1X-3D, an open framework addressing these challenges through: (1) a rigorous data curation pipeline processing >5M assets to create a 2M high-quality dataset with standardized geometric and textural properties; (2) a two-stage 3D-native architecture combining a hybrid VAE-DiT geometry generator with an diffusion-based texture synthesis module; and (3) the full open-source release of models, training code, and adaptation modules. For geometry generation, the hybrid VAE-DiT component produces TSDF representations by employing perceiver-based latent encoding with sharp edge sampling for detail preservation. The diffusion-based texture synthesis module then ensures cross-view consistency through geometric conditioning and latent-space synchronization. Benchmark results demonstrate state-of-the-art performance that exceeds existing open-source methods, while also achieving competitive quality with proprietary solutions. Notably, the framework uniquely bridges the 2D and 3D generation paradigms by supporting direct transfer of 2D control techniques~(e.g., LoRA) to 3D synthesis. By simultaneously advancing data quality, algorithmic fidelity, and reproducibility, Step1X-3D aims to establish new standards for open research in controllable 3D asset generation.", "date": "2025-05-12", "categories": []} {"id": "2505.05505", "title": "Apply Hierarchical-Chain-of-Generation to Complex Attributes Text-to-3D Generation", "abstract": "Recent text-to-3D models can render high-quality assets, yet they still stumble on objects with complex attributes. The key obstacles are: (1) existing text-to-3D approaches typically lift text-to-image models to extract semantics via text encoders, while the text encoder exhibits limited comprehension ability for long descriptions, leading to deviated cross-attention focus, subsequently wrong attribute binding in generated results. (2) Occluded object parts demand a disciplined generation order and explicit part disentanglement. Though some works introduce manual efforts to alleviate the above issues, their quality is unstable and highly reliant on manual information. To tackle above problems, we propose a automated method Hierarchical-Chain-of-Generation (HCoG). It leverages a large language model to decompose the long description into blocks representing different object parts, and orders them from inside out according to occlusions, forming a hierarchical chain. Within each block we first coarsely create components, then precisely bind attributes via target-region localization and corresponding 3D Gaussian kernel optimization. Between blocks, we introduce Gaussian Extension and Label Elimination to seamlessly generate new parts by extending new Gaussian kernels, re-assigning semantic labels, and eliminating unnecessary kernels, ensuring that only relevant parts are added without disrupting previously optimized parts. Experiments confirm that HCoG yields structurally coherent, attribute-faithful 3D objects with complex attributes. The code is available at https://github.com/Wakals/GASCOL .", "date": "2025-05-07", "categories": []} {"id": "2505.06166", "title": "DiffLocks: Generating 3D Hair from a Single Image using Diffusion Models", "abstract": "We address the task of generating 3D hair geometry from a single image, which is challenging due to the diversity of hairstyles and the lack of paired image-to-3D hair data. Previous methods are primarily trained on synthetic data and cope with the limited amount of such data by using low-dimensional intermediate representations, such as guide strands and scalp-level embeddings, that require post-processing to decode, upsample, and add realism. These approaches fail to reconstruct detailed hair, struggle with curly hair, or are limited to handling only a few hairstyles. To overcome these limitations, we propose DiffLocks, a novel framework that enables detailed reconstruction of a wide variety of hairstyles directly from a single image. First, we address the lack of 3D hair data by automating the creation of the largest synthetic hair dataset to date, containing 40K hairstyles. Second, we leverage the synthetic hair dataset to learn an image-conditioned diffusion-transfomer model that generates accurate 3D strands from a single frontal image. By using a pretrained image backbone, our method generalizes to in-the-wild images despite being trained only on synthetic data. Our diffusion model predicts a scalp texture map in which any point in the map contains the latent code for an individual hair strand. These codes are directly decoded to 3D strands without post-processing techniques. Representing individual strands, instead of guide strands, enables the transformer to model the detailed spatial structure of complex hairstyles. With this, DiffLocks can recover highly curled hair, like afro hairstyles, from a single image for the first time. Data and code is available at https://radualexandru.github.io/difflocks/", "date": "2025-05-09", "categories": []} {"id": "2410.09049", "title": "SceneCraft: Layout-Guided 3D Scene Generation", "abstract": " The creation of complex 3D scenes tailored to user specifications has been a\ntedious and challenging task with traditional 3D modeling tools. Although some\npioneering methods have achieved automatic text-to-3D generation, they are\ngenerally limited to small-scale scenes with restricted control over the shape\nand texture. We introduce SceneCraft, a novel method for generating detailed\nindoor scenes that adhere to textual descriptions and spatial layout\npreferences provided by users. Central to our method is a rendering-based\ntechnique, which converts 3D semantic layouts into multi-view 2D proxy maps.\nFurthermore, we design a semantic and depth conditioned diffusion model to\ngenerate multi-view images, which are used to learn a neural radiance field\n(NeRF) as the final scene representation. Without the constraints of panorama\nimage generation, we surpass previous methods in supporting complicated indoor\nspace generation beyond a single room, even as complicated as a whole\nmulti-bedroom apartment with irregular shapes and layouts. Through experimental\nanalysis, we demonstrate that our method significantly outperforms existing\napproaches in complex indoor scene generation with diverse textures, consistent\ngeometry, and realistic visual quality. Code and more results are available at:\nhttps://orangesodahub.github.io/SceneCraft\n", "date": "2024-10-11", "categories": []} {"id": "2505.05474", "title": "3D Scene Generation: A Survey", "abstract": " 3D scene generation seeks to synthesize spatially structured, semantically\nmeaningful, and photorealistic environments for applications such as immersive\nmedia, robotics, autonomous driving, and embodied AI. Early methods based on\nprocedural rules offered scalability but limited diversity. Recent advances in\ndeep generative models (e.g., GANs, diffusion models) and 3D representations\n(e.g., NeRF, 3D Gaussians) have enabled the learning of real-world scene\ndistributions, improving fidelity, diversity, and view consistency. Recent\nadvances like diffusion models bridge 3D scene synthesis and photorealism by\nreframing generation as image or video synthesis problems. This survey provides\na systematic overview of state-of-the-art approaches, organizing them into four\nparadigms: procedural generation, neural 3D-based generation, image-based\ngeneration, and video-based generation. We analyze their technical foundations,\ntrade-offs, and representative results, and review commonly used datasets,\nevaluation protocols, and downstream applications. We conclude by discussing\nkey challenges in generation capacity, 3D representation, data and annotations,\nand evaluation, and outline promising directions including higher fidelity,\nphysics-aware and interactive generation, and unified perception-generation\nmodels. This review organizes recent advances in 3D scene generation and\nhighlights promising directions at the intersection of generative AI, 3D\nvision, and embodied intelligence. To track ongoing developments, we maintain\nan up-to-date project page:\nhttps://github.com/hzxie/Awesome-3D-Scene-Generation.\n", "date": "2025-05-08", "categories": []} {"id": "2505.05475", "title": "SVAD: From Single Image to 3D Avatar via Synthetic Data Generation with\n Video Diffusion and Data Augmentation", "abstract": " Creating high-quality animatable 3D human avatars from a single image remains\na significant challenge in computer vision due to the inherent difficulty of\nreconstructing complete 3D information from a single viewpoint. Current\napproaches face a clear limitation: 3D Gaussian Splatting (3DGS) methods\nproduce high-quality results but require multiple views or video sequences,\nwhile video diffusion models can generate animations from single images but\nstruggle with consistency and identity preservation. We present SVAD, a novel\napproach that addresses these limitations by leveraging complementary strengths\nof existing techniques. Our method generates synthetic training data through\nvideo diffusion, enhances it with identity preservation and image restoration\nmodules, and utilizes this refined data to train 3DGS avatars. Comprehensive\nevaluations demonstrate that SVAD outperforms state-of-the-art (SOTA)\nsingle-image methods in maintaining identity consistency and fine details\nacross novel poses and viewpoints, while enabling real-time rendering\ncapabilities. Through our data augmentation pipeline, we overcome the\ndependency on dense monocular or multi-view training data typically required by\ntraditional 3DGS approaches. Extensive quantitative, qualitative comparisons\nshow our method achieves superior performance across multiple metrics against\nbaseline models. By effectively combining the generative power of diffusion\nmodels with both the high-quality results and rendering efficiency of 3DGS, our\nwork establishes a new approach for high-fidelity avatar generation from a\nsingle image input.\n", "date": "2025-05-08", "categories": []} {"id": "2505.04622", "title": "PrimitiveAnything: Human-Crafted 3D Primitive Assembly Generation with\n Auto-Regressive Transformer", "abstract": " Shape primitive abstraction, which decomposes complex 3D shapes into simple\ngeometric elements, plays a crucial role in human visual cognition and has\nbroad applications in computer vision and graphics. While recent advances in 3D\ncontent generation have shown remarkable progress, existing primitive\nabstraction methods either rely on geometric optimization with limited semantic\nunderstanding or learn from small-scale, category-specific datasets, struggling\nto generalize across diverse shape categories. We present PrimitiveAnything, a\nnovel framework that reformulates shape primitive abstraction as a primitive\nassembly generation task. PrimitiveAnything includes a shape-conditioned\nprimitive transformer for auto-regressive generation and an ambiguity-free\nparameterization scheme to represent multiple types of primitives in a unified\nmanner. The proposed framework directly learns the process of primitive\nassembly from large-scale human-crafted abstractions, enabling it to capture\nhow humans decompose complex shapes into primitive elements. Through extensive\nexperiments, we demonstrate that PrimitiveAnything can generate high-quality\nprimitive assemblies that better align with human perception while maintaining\ngeometric fidelity across diverse shape categories. It benefits various 3D\napplications and shows potential for enabling primitive-based user-generated\ncontent (UGC) in games. Project page: https://primitiveanything.github.io\n", "date": "2025-05-07", "categories": []} {"id": "2504.19174", "title": "CLR-Wire: Towards Continuous Latent Representations for 3D Curve\n Wireframe Generation", "abstract": " We introduce CLR-Wire, a novel framework for 3D curve-based wireframe\ngeneration that integrates geometry and topology into a unified Continuous\nLatent Representation. Unlike conventional methods that decouple vertices,\nedges, and faces, CLR-Wire encodes curves as Neural Parametric Curves along\nwith their topological connectivity into a continuous and fixed-length latent\nspace using an attention-driven variational autoencoder (VAE). This unified\napproach facilitates joint learning and generation of both geometry and\ntopology. To generate wireframes, we employ a flow matching model to\nprogressively map Gaussian noise to these latents, which are subsequently\ndecoded into complete 3D wireframes. Our method provides fine-grained modeling\nof complex shapes and irregular topologies, and supports both unconditional\ngeneration and generation conditioned on point cloud or image inputs.\nExperimental results demonstrate that, compared with state-of-the-art\ngenerative approaches, our method achieves substantial improvements in\naccuracy, novelty, and diversity, offering an efficient and comprehensive\nsolution for CAD design, geometric reconstruction, and 3D content creation.\n", "date": "2025-04-27", "categories": []} {"id": "2505.01888", "title": "Rethinking Score Distilling Sampling for 3D Editing and Generation", "abstract": " Score Distillation Sampling (SDS) has emerged as a prominent method for\ntext-to-3D generation by leveraging the strengths of 2D diffusion models.\nHowever, SDS is limited to generation tasks and lacks the capability to edit\nexisting 3D assets. Conversely, variants of SDS that introduce editing\ncapabilities often can not generate new 3D assets effectively. In this work, we\nobserve that the processes of generation and editing within SDS and its\nvariants have unified underlying gradient terms. Building on this insight, we\npropose Unified Distillation Sampling (UDS), a method that seamlessly\nintegrates both the generation and editing of 3D assets. Essentially, UDS\nrefines the gradient terms used in vanilla SDS methods, unifying them to\nsupport both tasks. Extensive experiments demonstrate that UDS not only\noutperforms baseline methods in generating 3D assets with richer details but\nalso excels in editing tasks, thereby bridging the gap between 3D generation\nand editing. The code is available on: https://github.com/xingy038/UDS.\n", "date": "2025-05-03", "categories": []} {"id": "2505.02836", "title": "Scenethesis: A Language and Vision Agentic Framework for 3D Scene\n Generation", "abstract": " Synthesizing interactive 3D scenes from text is essential for gaming, virtual\nreality, and embodied AI. However, existing methods face several challenges.\nLearning-based approaches depend on small-scale indoor datasets, limiting the\nscene diversity and layout complexity. While large language models (LLMs) can\nleverage diverse text-domain knowledge, they struggle with spatial realism,\noften producing unnatural object placements that fail to respect common sense.\nOur key insight is that vision perception can bridge this gap by providing\nrealistic spatial guidance that LLMs lack. To this end, we introduce\nScenethesis, a training-free agentic framework that integrates LLM-based scene\nplanning with vision-guided layout refinement. Given a text prompt, Scenethesis\nfirst employs an LLM to draft a coarse layout. A vision module then refines it\nby generating an image guidance and extracting scene structure to capture\ninter-object relations. Next, an optimization module iteratively enforces\naccurate pose alignment and physical plausibility, preventing artifacts like\nobject penetration and instability. Finally, a judge module verifies spatial\ncoherence. Comprehensive experiments show that Scenethesis generates diverse,\nrealistic, and physically plausible 3D interactive scenes, making it valuable\nfor virtual content creation, simulation environments, and embodied AI\nresearch.\n", "date": "2025-05-05", "categories": []} {"id": "2405.00998", "title": "Part-aware Shape Generation with Latent 3D Diffusion of Neural Voxel\n Fields", "abstract": " This paper presents a novel latent 3D diffusion model for the generation of\nneural voxel fields, aiming to achieve accurate part-aware structures. Compared\nto existing methods, there are two key designs to ensure high-quality and\naccurate part-aware generation. On one hand, we introduce a latent 3D diffusion\nprocess for neural voxel fields, enabling generation at significantly higher\nresolutions that can accurately capture rich textural and geometric details. On\nthe other hand, a part-aware shape decoder is introduced to integrate the part\ncodes into the neural voxel fields, guiding the accurate part decomposition and\nproducing high-quality rendering results. Through extensive experimentation and\ncomparisons with state-of-the-art methods, we evaluate our approach across four\ndifferent classes of data. The results demonstrate the superior generative\ncapabilities of our proposed method in part-aware shape generation,\noutperforming existing state-of-the-art methods.\n", "date": "2024-05-02", "categories": []} {"id": "2312.12419", "title": "Scene-Conditional 3D Object Stylization and Composition", "abstract": " Recently, 3D generative models have made impressive progress, enabling the\ngeneration of almost arbitrary 3D assets from text or image inputs. However,\nthese approaches generate objects in isolation without any consideration for\nthe scene where they will eventually be placed. In this paper, we propose a\nframework that allows for the stylization of an existing 3D asset to fit into a\ngiven 2D scene, and additionally produce a photorealistic composition as if the\nasset was placed within the environment. This not only opens up a new level of\ncontrol for object stylization, for example, the same assets can be stylized to\nreflect changes in the environment, such as summer to winter or fantasy versus\nfuturistic settings-but also makes the object-scene composition more\ncontrollable. We achieve this by combining modeling and optimizing the object's\ntexture and environmental lighting through differentiable ray tracing with\nimage priors from pre-trained text-to-image diffusion models. We demonstrate\nthat our method is applicable to a wide variety of indoor and outdoor scenes\nand arbitrary objects. Project page:\nhttps://jensenzhoujh.github.io/scene-cond-3d/.\n", "date": "2023-12-19", "categories": []} {"id": "2406.03184", "title": "Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion", "abstract": " Existing single image-to-3D creation methods typically involve a two-stage\nprocess, first generating multi-view images, and then using these images for 3D\nreconstruction. However, training these two stages separately leads to\nsignificant data bias in the inference phase, thus affecting the quality of\nreconstructed results. We introduce a unified 3D generation framework, named\nOuroboros3D, which integrates diffusion-based multi-view image generation and\n3D reconstruction into a recursive diffusion process. In our framework, these\ntwo modules are jointly trained through a self-conditioning mechanism, allowing\nthem to adapt to each other's characteristics for robust inference. During the\nmulti-view denoising process, the multi-view diffusion model uses the 3D-aware\nmaps rendered by the reconstruction module at the previous timestep as\nadditional conditions. The recursive diffusion framework with 3D-aware feedback\nunites the entire process and improves geometric consistency.Experiments show\nthat our framework outperforms separation of these two stages and existing\nmethods that combine them at the inference phase. Project page:\nhttps://costwen.github.io/Ouroboros3D/\n", "date": "2024-06-05", "categories": []} {"id": "2409.08215", "title": "LT3SD: Latent Trees for 3D Scene Diffusion", "abstract": " We present LT3SD, a novel latent diffusion model for large-scale 3D scene\ngeneration. Recent advances in diffusion models have shown impressive results\nin 3D object generation, but are limited in spatial extent and quality when\nextended to 3D scenes. To generate complex and diverse 3D scene structures, we\nintroduce a latent tree representation to effectively encode both\nlower-frequency geometry and higher-frequency detail in a coarse-to-fine\nhierarchy. We can then learn a generative diffusion process in this latent 3D\nscene space, modeling the latent components of a scene at each resolution\nlevel. To synthesize large-scale scenes with varying sizes, we train our\ndiffusion model on scene patches and synthesize arbitrary-sized output 3D\nscenes through shared diffusion generation across multiple scene patches.\nThrough extensive experiments, we demonstrate the efficacy and benefits of\nLT3SD for large-scale, high-quality unconditional 3D scene generation and for\nprobabilistic completion for partial scene observations.\n", "date": "2024-09-12", "categories": []} {"id": "2403.09326", "title": "HeadEvolver: Text to Head Avatars via Expressive and\n Attribute-Preserving Mesh Deformation", "abstract": " Current text-to-avatar methods often rely on implicit representations (e.g.,\nNeRF, SDF, and DMTet), leading to 3D content that artists cannot easily edit\nand animate in graphics software. This paper introduces a novel framework for\ngenerating stylized head avatars from text guidance, which leverages locally\nlearnable mesh deformation and 2D diffusion priors to achieve high-quality\ndigital assets for attribute-preserving manipulation. Given a template mesh,\nour method represents mesh deformation with per-face Jacobians and adaptively\nmodulates local deformation using a learnable vector field. This vector field\nenables anisotropic scaling while preserving the rotation of vertices, which\ncan better express identity and geometric details. We employ landmark- and\ncontour-based regularization terms to balance the expressiveness and\nplausibility of generated avatars from multiple views without relying on any\nspecific shape prior. Our framework can generate realistic shapes and textures\nthat can be further edited via text, while supporting seamless editing using\nthe preserved attributes from the template mesh, such as 3DMM parameters,\nblendshapes, and UV coordinates. Extensive experiments demonstrate that our\nframework can generate diverse and expressive head avatars with high-quality\nmeshes that artists can easily manipulate in graphics software, facilitating\ndownstream applications such as efficient asset creation and animation with\npreserved attributes.\n", "date": "2024-03-14", "categories": []} {"id": "2403.18816", "title": "Garment3DGen: 3D Garment Stylization and Texture Generation", "abstract": " We introduce Garment3DGen a new method to synthesize 3D garment assets from a\nbase mesh given a single input image as guidance. Our proposed approach allows\nusers to generate 3D textured clothes based on both real and synthetic images,\nsuch as those generated by text prompts. The generated assets can be directly\ndraped and simulated on human bodies. We leverage the recent progress of\nimage-to-3D diffusion methods to generate 3D garment geometries. However, since\nthese geometries cannot be utilized directly for downstream tasks, we propose\nto use them as pseudo ground-truth and set up a mesh deformation optimization\nprocedure that deforms a base template mesh to match the generated 3D target.\nCarefully designed losses allow the base mesh to freely deform towards the\ndesired target, yet preserve mesh quality and topology such that they can be\nsimulated. Finally, we generate high-fidelity texture maps that are globally\nand locally consistent and faithfully capture the input guidance, allowing us\nto render the generated 3D assets. With Garment3DGen users can generate the\nsimulation-ready 3D garment of their choice without the need of artist\nintervention. We present a plethora of quantitative and qualitative comparisons\non various assets and demonstrate that Garment3DGen unlocks key applications\nranging from sketch-to-simulated garments or interacting with the garments in\nVR. Code is publicly available.\n", "date": "2024-03-27", "categories": []} {"id": "2504.21836", "title": "3D Stylization via Large Reconstruction Model", "abstract": " With the growing success of text or image guided 3D generators, users demand\nmore control over the generation process, appearance stylization being one of\nthem. Given a reference image, this requires adapting the appearance of a\ngenerated 3D asset to reflect the visual style of the reference while\nmaintaining visual consistency from multiple viewpoints. To tackle this\nproblem, we draw inspiration from the success of 2D stylization methods that\nleverage the attention mechanisms in large image generation models to capture\nand transfer visual style. In particular, we probe if large reconstruction\nmodels, commonly used in the context of 3D generation, has a similar\ncapability. We discover that the certain attention blocks in these models\ncapture the appearance specific features. By injecting features from a visual\nstyle image to such blocks, we develop a simple yet effective 3D appearance\nstylization method. Our method does not require training or test time\noptimization. Through both quantitative and qualitative evaluations, we\ndemonstrate that our approach achieves superior results in terms of 3D\nappearance stylization, significantly improving efficiency while maintaining\nhigh-quality visual outcomes.\n", "date": "2025-04-30", "categories": []} {"id": "2412.18565", "title": "3DEnhancer: Consistent Multi-View Diffusion for 3D Enhancement", "abstract": " Despite advances in neural rendering, due to the scarcity of high-quality 3D\ndatasets and the inherent limitations of multi-view diffusion models, view\nsynthesis and 3D model generation are restricted to low resolutions with\nsuboptimal multi-view consistency. In this study, we present a novel 3D\nenhancement pipeline, dubbed 3DEnhancer, which employs a multi-view latent\ndiffusion model to enhance coarse 3D inputs while preserving multi-view\nconsistency. Our method includes a pose-aware encoder and a diffusion-based\ndenoiser to refine low-quality multi-view images, along with data augmentation\nand a multi-view attention module with epipolar aggregation to maintain\nconsistent, high-quality 3D outputs across views. Unlike existing video-based\napproaches, our model supports seamless multi-view enhancement with improved\ncoherence across diverse viewing angles. Extensive evaluations show that\n3DEnhancer significantly outperforms existing methods, boosting both multi-view\nenhancement and per-instance 3D optimization tasks.\n", "date": "2024-12-24", "categories": []} {"id": "2504.20409", "title": "GarmentX: Autoregressive Parametric Representations for High-Fidelity 3D\n Garment Generation", "abstract": " This work presents GarmentX, a novel framework for generating diverse,\nhigh-fidelity, and wearable 3D garments from a single input image. Traditional\ngarment reconstruction methods directly predict 2D pattern edges and their\nconnectivity, an overly unconstrained approach that often leads to severe\nself-intersections and physically implausible garment structures. In contrast,\nGarmentX introduces a structured and editable parametric representation\ncompatible with GarmentCode, ensuring that the decoded sewing patterns always\nform valid, simulation-ready 3D garments while allowing for intuitive\nmodifications of garment shape and style. To achieve this, we employ a masked\nautoregressive model that sequentially predicts garment parameters, leveraging\nautoregressive modeling for structured generation while mitigating\ninconsistencies in direct pattern prediction. Additionally, we introduce\nGarmentX dataset, a large-scale dataset of 378,682 garment parameter-image\npairs, constructed through an automatic data generation pipeline that\nsynthesizes diverse and high-quality garment images conditioned on parametric\ngarment representations. Through integrating our method with GarmentX dataset,\nwe achieve state-of-the-art performance in geometric fidelity and input image\nalignment, significantly outperforming prior approaches. We will release\nGarmentX dataset upon publication.\n", "date": "2025-04-29", "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": "2504.19478", "title": "CasaGPT: Cuboid Arrangement and Scene Assembly for Interior Design", "abstract": " We present a novel approach for indoor scene synthesis, which learns to\narrange decomposed cuboid primitives to represent 3D objects within a scene.\nUnlike conventional methods that use bounding boxes to determine the placement\nand scale of 3D objects, our approach leverages cuboids as a straightforward\nyet highly effective alternative for modeling objects. This allows for compact\nscene generation while minimizing object intersections. Our approach, coined\nCasaGPT for Cuboid Arrangement and Scene Assembly, employs an autoregressive\nmodel to sequentially arrange cuboids, producing physically plausible scenes.\nBy applying rejection sampling during the fine-tuning stage to filter out\nscenes with object collisions, our model further reduces intersections and\nenhances scene quality. Additionally, we introduce a refined dataset,\n3DFRONT-NC, which eliminates significant noise presented in the original\ndataset, 3D-FRONT. Extensive experiments on the 3D-FRONT dataset as well as our\ndataset demonstrate that our approach consistently outperforms the\nstate-of-the-art methods, enhancing the realism of generated scenes, and\nproviding a promising direction for 3D scene synthesis.\n", "date": "2025-04-28", "categories": []} {"id": "2504.18318", "title": "STP4D: Spatio-Temporal-Prompt Consistent Modeling for Text-to-4D\n Gaussian Splatting", "abstract": " Text-to-4D generation is rapidly developing and widely applied in various\nscenarios. However, existing methods often fail to incorporate adequate\nspatio-temporal modeling and prompt alignment within a unified framework,\nresulting in temporal inconsistencies, geometric distortions, or low-quality 4D\ncontent that deviates from the provided texts. Therefore, we propose STP4D, a\nnovel approach that aims to integrate comprehensive spatio-temporal-prompt\nconsistency modeling for high-quality text-to-4D generation. Specifically,\nSTP4D employs three carefully designed modules: Time-varying Prompt Embedding,\nGeometric Information Enhancement, and Temporal Extension Deformation, which\ncollaborate to accomplish this goal. Furthermore, STP4D is among the first\nmethods to exploit the Diffusion model to generate 4D Gaussians, combining the\nfine-grained modeling capabilities and the real-time rendering process of 4DGS\nwith the rapid inference speed of the Diffusion model. Extensive experiments\ndemonstrate that STP4D excels in generating high-fidelity 4D content with\nexceptional efficiency (approximately 4.6s per asset), surpassing existing\nmethods in both quality and speed.\n", "date": "2025-04-25", "categories": []} {"id": "2504.15835", "title": "Text-based Animatable 3D Avatars with Morphable Model Alignment", "abstract": " The generation of high-quality, animatable 3D head avatars from text has\nenormous potential in content creation applications such as games, movies, and\nembodied virtual assistants. Current text-to-3D generation methods typically\ncombine parametric head models with 2D diffusion models using score\ndistillation sampling to produce 3D-consistent results. However, they struggle\nto synthesize realistic details and suffer from misalignments between the\nappearance and the driving parametric model, resulting in unnatural animation\nresults. We discovered that these limitations stem from ambiguities in the 2D\ndiffusion predictions during 3D avatar distillation, specifically: i) the\navatar's appearance and geometry is underconstrained by the text input, and ii)\nthe semantic alignment between the predictions and the parametric head model is\ninsufficient because the diffusion model alone cannot incorporate information\nfrom the parametric model. In this work, we propose a novel framework,\nAnimPortrait3D, for text-based realistic animatable 3DGS avatar generation with\nmorphable model alignment, and introduce two key strategies to address these\nchallenges. First, we tackle appearance and geometry ambiguities by utilizing\nprior information from a pretrained text-to-3D model to initialize a 3D avatar\nwith robust appearance, geometry, and rigging relationships to the morphable\nmodel. Second, we refine the initial 3D avatar for dynamic expressions using a\nControlNet that is conditioned on semantic and normal maps of the morphable\nmodel to ensure accurate alignment. As a result, our method outperforms\nexisting approaches in terms of synthesis quality, alignment, and animation\nfidelity. Our experiments show that the proposed method advances the state of\nthe art in text-based, animatable 3D head avatar generation.\n", "date": "2025-04-22", "categories": []} {"id": "2503.20519", "title": "MAR-3D: Progressive Masked Auto-regressor for High-Resolution 3D\n Generation", "abstract": " Recent advances in auto-regressive transformers have revolutionized\ngenerative modeling across different domains, from language processing to\nvisual generation, demonstrating remarkable capabilities. However, applying\nthese advances to 3D generation presents three key challenges: the unordered\nnature of 3D data conflicts with sequential next-token prediction paradigm,\nconventional vector quantization approaches incur substantial compression loss\nwhen applied to 3D meshes, and the lack of efficient scaling strategies for\nhigher resolution latent prediction. To address these challenges, we introduce\nMAR-3D, which integrates a pyramid variational autoencoder with a cascaded\nmasked auto-regressive transformer (Cascaded MAR) for progressive latent\nupscaling in the continuous space. Our architecture employs random masking\nduring training and auto-regressive denoising in random order during inference,\nnaturally accommodating the unordered property of 3D latent tokens.\nAdditionally, we propose a cascaded training strategy with condition\naugmentation that enables efficiently up-scale the latent token resolution with\nfast convergence. Extensive experiments demonstrate that MAR-3D not only\nachieves superior performance and generalization capabilities compared to\nexisting methods but also exhibits enhanced scaling capabilities compared to\njoint distribution modeling approaches (e.g., diffusion transformers).\n", "date": "2025-03-26", "categories": []} {"id": "2412.11520", "title": "EditSplat: Multi-View Fusion and Attention-Guided Optimization for\n View-Consistent 3D Scene Editing with 3D Gaussian Splatting", "abstract": " Recent advancements in 3D editing have highlighted the potential of\ntext-driven methods in real-time, user-friendly AR/VR applications. However,\ncurrent methods rely on 2D diffusion models without adequately considering\nmulti-view information, resulting in multi-view inconsistency. While 3D\nGaussian Splatting (3DGS) significantly improves rendering quality and speed,\nits 3D editing process encounters difficulties with inefficient optimization,\nas pre-trained Gaussians retain excessive source information, hindering\noptimization. To address these limitations, we propose EditSplat, a novel\ntext-driven 3D scene editing framework that integrates Multi-view Fusion\nGuidance (MFG) and Attention-Guided Trimming (AGT). Our MFG ensures multi-view\nconsistency by incorporating essential multi-view information into the\ndiffusion process, leveraging classifier-free guidance from the text-to-image\ndiffusion model and the geometric structure inherent to 3DGS. Additionally, our\nAGT utilizes the explicit representation of 3DGS to selectively prune and\noptimize 3D Gaussians, enhancing optimization efficiency and enabling precise,\nsemantically rich local editing. Through extensive qualitative and quantitative\nevaluations, EditSplat achieves state-of-the-art performance, establishing a\nnew benchmark for text-driven 3D scene editing.\n", "date": "2024-12-16", "categories": []} {"id": "2311.17510", "title": "StructRe: Rewriting for Structured Shape Modeling", "abstract": " Man-made 3D shapes are naturally organized in parts and hierarchies; such\nstructures provide important constraints for shape reconstruction and\ngeneration. Modeling shape structures is difficult, because there can be\nmultiple hierarchies for a given shape, causing ambiguity, and across different\ncategories the shape structures are correlated with semantics, limiting\ngeneralization. We present StructRe, a structure rewriting system, as a novel\napproach to structured shape modeling. Given a 3D object represented by points\nand components, StructRe can rewrite it upward into more concise structures, or\ndownward into more detailed structures; by iterating the rewriting process,\nhierarchies are obtained. Such a localized rewriting process enables\nprobabilistic modeling of ambiguous structures and robust generalization across\nobject categories. We train StructRe on PartNet data and show its\ngeneralization to cross-category and multiple object hierarchies, and test its\nextension to ShapeNet. We also demonstrate the benefits of probabilistic and\ngeneralizable structure modeling for shape reconstruction, generation and\nediting tasks.\n", "date": "2023-11-29", "categories": []} {"id": "2504.13072", "title": "HiScene: Creating Hierarchical 3D Scenes with Isometric View Generation", "abstract": " Scene-level 3D generation represents a critical frontier in multimedia and\ncomputer graphics, yet existing approaches either suffer from limited object\ncategories or lack editing flexibility for interactive applications. In this\npaper, we present HiScene, a novel hierarchical framework that bridges the gap\nbetween 2D image generation and 3D object generation and delivers high-fidelity\nscenes with compositional identities and aesthetic scene content. Our key\ninsight is treating scenes as hierarchical \"objects\" under isometric views,\nwhere a room functions as a complex object that can be further decomposed into\nmanipulatable items. This hierarchical approach enables us to generate 3D\ncontent that aligns with 2D representations while maintaining compositional\nstructure. To ensure completeness and spatial alignment of each decomposed\ninstance, we develop a video-diffusion-based amodal completion technique that\neffectively handles occlusions and shadows between objects, and introduce shape\nprior injection to ensure spatial coherence within the scene. Experimental\nresults demonstrate that our method produces more natural object arrangements\nand complete object instances suitable for interactive applications, while\nmaintaining physical plausibility and alignment with user inputs.\n", "date": "2025-04-17", "categories": []} {"id": "2411.16443", "title": "SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting\n Synthesis", "abstract": " Text-based generation and editing of 3D scenes hold significant potential for\nstreamlining content creation through intuitive user interactions. While recent\nadvances leverage 3D Gaussian Splatting (3DGS) for high-fidelity and real-time\nrendering, existing methods are often specialized and task-focused, lacking a\nunified framework for both generation and editing. In this paper, we introduce\nSplatFlow, a comprehensive framework that addresses this gap by enabling direct\n3DGS generation and editing. SplatFlow comprises two main components: a\nmulti-view rectified flow (RF) model and a Gaussian Splatting Decoder\n(GSDecoder). The multi-view RF model operates in latent space, generating\nmulti-view images, depths, and camera poses simultaneously, conditioned on text\nprompts, thus addressing challenges like diverse scene scales and complex\ncamera trajectories in real-world settings. Then, the GSDecoder efficiently\ntranslates these latent outputs into 3DGS representations through a\nfeed-forward 3DGS method. Leveraging training-free inversion and inpainting\ntechniques, SplatFlow enables seamless 3DGS editing and supports a broad range\nof 3D tasks-including object editing, novel view synthesis, and camera pose\nestimation-within a unified framework without requiring additional complex\npipelines. We validate SplatFlow's capabilities on the MVImgNet and DL3DV-7K\ndatasets, demonstrating its versatility and effectiveness in various 3D\ngeneration, editing, and inpainting-based tasks.\n", "date": "2024-11-25", "categories": []} {"id": "2412.06780", "title": "Diverse Score Distillation", "abstract": " Score distillation of 2D diffusion models has proven to be a powerful\nmechanism to guide 3D optimization, for example enabling text-based 3D\ngeneration or single-view reconstruction. A common limitation of existing score\ndistillation formulations, however, is that the outputs of the (mode-seeking)\noptimization are limited in diversity despite the underlying diffusion model\nbeing capable of generating diverse samples. In this work, inspired by the\nsampling process in denoising diffusion, we propose a score formulation that\nguides the optimization to follow generation paths defined by random initial\nseeds, thus ensuring diversity. We then present an approximation to adopt this\nformulation for scenarios where the optimization may not precisely follow the\ngeneration paths (\\eg a 3D representation whose renderings evolve in a\nco-dependent manner). We showcase the applications of our `Diverse Score\nDistillation' (DSD) formulation across tasks such as 2D optimization,\ntext-based 3D inference, and single-view reconstruction. We also empirically\nvalidate DSD against prior score distillation formulations and show that it\nsignificantly improves sample diversity while preserving fidelity.\n", "date": "2024-12-09", "categories": []} {"id": "2504.10001", "title": "GaussVideoDreamer: 3D Scene Generation with Video Diffusion and\n Inconsistency-Aware Gaussian Splatting", "abstract": " Single-image 3D scene reconstruction presents significant challenges due to\nits inherently ill-posed nature and limited input constraints. Recent advances\nhave explored two promising directions: multiview generative models that train\non 3D consistent datasets but struggle with out-of-distribution generalization,\nand 3D scene inpainting and completion frameworks that suffer from cross-view\ninconsistency and suboptimal error handling, as they depend exclusively on\ndepth data or 3D smoothness, which ultimately degrades output quality and\ncomputational performance. Building upon these approaches, we present\nGaussVideoDreamer, which advances generative multimedia approaches by bridging\nthe gap between image, video, and 3D generation, integrating their strengths\nthrough two key innovations: (1) A progressive video inpainting strategy that\nharnesses temporal coherence for improved multiview consistency and faster\nconvergence. (2) A 3D Gaussian Splatting consistency mask to guide the video\ndiffusion with 3D consistent multiview evidence. Our pipeline combines three\ncore components: a geometry-aware initialization protocol, Inconsistency-Aware\nGaussian Splatting, and a progressive video inpainting strategy. Experimental\nresults demonstrate that our approach achieves 32% higher LLaVA-IQA scores and\nat least 2x speedup compared to existing methods while maintaining robust\nperformance across diverse scenes.\n", "date": "2025-04-14", "categories": []} {"id": "2504.09975", "title": "OctGPT: Octree-based Multiscale Autoregressive Models for 3D Shape\n Generation", "abstract": " Autoregressive models have achieved remarkable success across various\ndomains, yet their performance in 3D shape generation lags significantly behind\nthat of diffusion models. In this paper, we introduce OctGPT, a novel\nmultiscale autoregressive model for 3D shape generation that dramatically\nimproves the efficiency and performance of prior 3D autoregressive approaches,\nwhile rivaling or surpassing state-of-the-art diffusion models. Our method\nemploys a serialized octree representation to efficiently capture the\nhierarchical and spatial structures of 3D shapes. Coarse geometry is encoded\nvia octree structures, while fine-grained details are represented by binary\ntokens generated using a vector quantized variational autoencoder (VQVAE),\ntransforming 3D shapes into compact multiscale binary sequences suitable for\nautoregressive prediction. To address the computational challenges of handling\nlong sequences, we incorporate octree-based transformers enhanced with 3D\nrotary positional encodings, scale-specific embeddings, and token-parallel\ngeneration schemes. These innovations reduce training time by 13 folds and\ngeneration time by 69 folds, enabling the efficient training of high-resolution\n3D shapes, e.g.,$1024^3$, on just four NVIDIA 4090 GPUs only within days.\nOctGPT showcases exceptional versatility across various tasks, including text-,\nsketch-, and image-conditioned generation, as well as scene-level synthesis\ninvolving multiple objects. Extensive experiments demonstrate that OctGPT\naccelerates convergence and improves generation quality over prior\nautoregressive methods, offering a new paradigm for high-quality, scalable 3D\ncontent creation. Our code and trained models are available at\nhttps://github.com/octree-nn/octgpt.\n", "date": "2025-04-14", "categories": []} {"id": "2404.17569", "title": "MaPa: Text-driven Photorealistic Material Painting for 3D Shapes", "abstract": " This paper aims to generate materials for 3D meshes from text descriptions.\nUnlike existing methods that synthesize texture maps, we propose to generate\nsegment-wise procedural material graphs as the appearance representation, which\nsupports high-quality rendering and provides substantial flexibility in\nediting. Instead of relying on extensive paired data, i.e., 3D meshes with\nmaterial graphs and corresponding text descriptions, to train a material graph\ngenerative model, we propose to leverage the pre-trained 2D diffusion model as\na bridge to connect the text and material graphs. Specifically, our approach\ndecomposes a shape into a set of segments and designs a segment-controlled\ndiffusion model to synthesize 2D images that are aligned with mesh parts. Based\non generated images, we initialize parameters of material graphs and fine-tune\nthem through the differentiable rendering module to produce materials in\naccordance with the textual description. Extensive experiments demonstrate the\nsuperior performance of our framework in photorealism, resolution, and\neditability over existing methods. Project page: https://zju3dv.github.io/MaPa\n", "date": "2024-04-26", "categories": []} {"id": "2501.05763", "title": "StarGen: A Spatiotemporal Autoregression Framework with Video Diffusion\n Model for Scalable and Controllable Scene Generation", "abstract": " Recent advances in large reconstruction and generative models have\nsignificantly improved scene reconstruction and novel view generation. However,\ndue to compute limitations, each inference with these large models is confined\nto a small area, making long-range consistent scene generation challenging. To\naddress this, we propose StarGen, a novel framework that employs a pre-trained\nvideo diffusion model in an autoregressive manner for long-range scene\ngeneration. The generation of each video clip is conditioned on the 3D warping\nof spatially adjacent images and the temporally overlapping image from\npreviously generated clips, improving spatiotemporal consistency in long-range\nscene generation with precise pose control. The spatiotemporal condition is\ncompatible with various input conditions, facilitating diverse tasks, including\nsparse view interpolation, perpetual view generation, and layout-conditioned\ncity generation. Quantitative and qualitative evaluations demonstrate StarGen's\nsuperior scalability, fidelity, and pose accuracy compared to state-of-the-art\nmethods. Project page: https://zju3dv.github.io/StarGen.\n", "date": "2025-01-10", "categories": []} {"id": "2504.07334", "title": "Objaverse++: Curated 3D Object Dataset with Quality Annotations", "abstract": " This paper presents Objaverse++, a curated subset of Objaverse enhanced with\ndetailed attribute annotations by human experts. Recent advances in 3D content\ngeneration have been driven by large-scale datasets such as Objaverse, which\ncontains over 800,000 3D objects collected from the Internet. Although\nObjaverse represents the largest available 3D asset collection, its utility is\nlimited by the predominance of low-quality models. To address this limitation,\nwe manually annotate 10,000 3D objects with detailed attributes, including\naesthetic quality scores, texture color classifications, multi-object\ncomposition flags, transparency characteristics, etc. Then, we trained a neural\nnetwork capable of annotating the tags for the rest of the Objaverse dataset.\nThrough experiments and a user study on generation results, we demonstrate that\nmodels pre-trained on our quality-focused subset achieve better performance\nthan those trained on the larger dataset of Objaverse in image-to-3D generation\ntasks. In addition, by comparing multiple subsets of training data filtered by\nour tags, our results show that the higher the data quality, the faster the\ntraining loss converges. These findings suggest that careful curation and rich\nannotation can compensate for the raw dataset size, potentially offering a more\nefficient path to develop 3D generative models. We release our enhanced dataset\nof approximately 500,000 curated 3D models to facilitate further research on\nvarious downstream tasks in 3D computer vision. In the near future, we aim to\nextend our annotations to cover the entire Objaverse dataset.\n", "date": "2025-04-09", "categories": []} {"id": "2504.09328", "title": "Text To 3D Object Generation For Scalable Room Assembly", "abstract": " Modern machine learning models for scene understanding, such as depth\nestimation and object tracking, rely on large, high-quality datasets that mimic\nreal-world deployment scenarios. To address data scarcity, we propose an\nend-to-end system for synthetic data generation for scalable, high-quality, and\ncustomizable 3D indoor scenes. By integrating and adapting text-to-image and\nmulti-view diffusion models with Neural Radiance Field-based meshing, this\nsystem generates highfidelity 3D object assets from text prompts and\nincorporates them into pre-defined floor plans using a rendering tool. By\nintroducing novel loss functions and training strategies into existing methods,\nthe system supports on-demand scene generation, aiming to alleviate the\nscarcity of current available data, generally manually crafted by artists. This\nsystem advances the role of synthetic data in addressing machine learning\ntraining limitations, enabling more robust and generalizable models for\nreal-world applications.\n", "date": "2025-04-12", "categories": []} {"id": "2504.10316", "title": "ESCT3D: Efficient and Selectively Controllable Text-Driven 3D Content\n Generation with Gaussian Splatting", "abstract": " In recent years, significant advancements have been made in text-driven 3D\ncontent generation. However, several challenges remain. In practical\napplications, users often provide extremely simple text inputs while expecting\nhigh-quality 3D content. Generating optimal results from such minimal text is a\ndifficult task due to the strong dependency of text-to-3D models on the quality\nof input prompts. Moreover, the generation process exhibits high variability,\nmaking it difficult to control. Consequently, multiple iterations are typically\nrequired to produce content that meets user expectations, reducing generation\nefficiency. To address this issue, we propose GPT-4V for self-optimization,\nwhich significantly enhances the efficiency of generating satisfactory content\nin a single attempt. Furthermore, the controllability of text-to-3D generation\nmethods has not been fully explored. Our approach enables users to not only\nprovide textual descriptions but also specify additional conditions, such as\nstyle, edges, scribbles, poses, or combinations of multiple conditions,\nallowing for more precise control over the generated 3D content. Additionally,\nduring training, we effectively integrate multi-view information, including\nmulti-view depth, masks, features, and images, to address the common Janus\nproblem in 3D content generation. Extensive experiments demonstrate that our\nmethod achieves robust generalization, facilitating the efficient and\ncontrollable generation of high-quality 3D content.\n", "date": "2025-04-14", "categories": []} {"id": "2409.16938", "title": "Generative Object Insertion in Gaussian Splatting with a Multi-View\n Diffusion Model", "abstract": " Generating and inserting new objects into 3D content is a compelling approach\nfor achieving versatile scene recreation. Existing methods, which rely on SDS\noptimization or single-view inpainting, often struggle to produce high-quality\nresults. To address this, we propose a novel method for object insertion in 3D\ncontent represented by Gaussian Splatting. Our approach introduces a multi-view\ndiffusion model, dubbed MVInpainter, which is built upon a pre-trained stable\nvideo diffusion model to facilitate view-consistent object inpainting. Within\nMVInpainter, we incorporate a ControlNet-based conditional injection module to\nenable controlled and more predictable multi-view generation. After generating\nthe multi-view inpainted results, we further propose a mask-aware 3D\nreconstruction technique to refine Gaussian Splatting reconstruction from these\nsparse inpainted views. By leveraging these fabricate techniques, our approach\nyields diverse results, ensures view-consistent and harmonious insertions, and\nproduces better object quality. Extensive experiments demonstrate that our\napproach outperforms existing methods.\n", "date": "2024-09-25", "categories": []} {"id": "2504.08100", "title": "ContrastiveGaussian: High-Fidelity 3D Generation with Contrastive\n Learning and Gaussian Splatting", "abstract": " Creating 3D content from single-view images is a challenging problem that has\nattracted considerable attention in recent years. Current approaches typically\nutilize score distillation sampling (SDS) from pre-trained 2D diffusion models\nto generate multi-view 3D representations. Although some methods have made\nnotable progress by balancing generation speed and model quality, their\nperformance is often limited by the visual inconsistencies of the diffusion\nmodel outputs. In this work, we propose ContrastiveGaussian, which integrates\ncontrastive learning into the generative process. By using a perceptual loss,\nwe effectively differentiate between positive and negative samples, leveraging\nthe visual inconsistencies to improve 3D generation quality. To further enhance\nsample differentiation and improve contrastive learning, we incorporate a\nsuper-resolution model and introduce another Quantity-Aware Triplet Loss to\naddress varying sample distributions during training. Our experiments\ndemonstrate that our approach achieves superior texture fidelity and improved\ngeometric consistency.\n", "date": "2025-04-10", "categories": []} {"id": "2411.08033", "title": "GaussianAnything: Interactive Point Cloud Flow Matching For 3D Object\n Generation", "abstract": " While 3D content generation has advanced significantly, existing methods\nstill face challenges with input formats, latent space design, and output\nrepresentations. This paper introduces a novel 3D generation framework that\naddresses these challenges, offering scalable, high-quality 3D generation with\nan interactive Point Cloud-structured Latent space. Our framework employs a\nVariational Autoencoder (VAE) with multi-view posed RGB-D(epth)-N(ormal)\nrenderings as input, using a unique latent space design that preserves 3D shape\ninformation, and incorporates a cascaded latent flow-based model for improved\nshape-texture disentanglement. The proposed method, GaussianAnything, supports\nmulti-modal conditional 3D generation, allowing for point cloud, caption, and\nsingle image inputs. Notably, the newly proposed latent space naturally enables\ngeometry-texture disentanglement, thus allowing 3D-aware editing. Experimental\nresults demonstrate the effectiveness of our approach on multiple datasets,\noutperforming existing native 3D methods in both text- and image-conditioned 3D\ngeneration.\n", "date": "2024-11-12", "categories": []} {"id": "2408.13055", "title": "Atlas Gaussians Diffusion for 3D Generation", "abstract": " Using the latent diffusion model has proven effective in developing novel 3D\ngeneration techniques. To harness the latent diffusion model, a key challenge\nis designing a high-fidelity and efficient representation that links the latent\nspace and the 3D space. In this paper, we introduce Atlas Gaussians, a novel\nrepresentation for feed-forward native 3D generation. Atlas Gaussians represent\na shape as the union of local patches, and each patch can decode 3D Gaussians.\nWe parameterize a patch as a sequence of feature vectors and design a learnable\nfunction to decode 3D Gaussians from the feature vectors. In this process, we\nincorporate UV-based sampling, enabling the generation of a sufficiently large,\nand theoretically infinite, number of 3D Gaussian points. The large amount of\n3D Gaussians enables the generation of high-quality details. Moreover, due to\nlocal awareness of the representation, the transformer-based decoding procedure\noperates on a patch level, ensuring efficiency. We train a variational\nautoencoder to learn the Atlas Gaussians representation, and then apply a\nlatent diffusion model on its latent space for learning 3D Generation.\nExperiments show that our approach outperforms the prior arts of feed-forward\nnative 3D generation. Project page:\nhttps://yanghtr.github.io/projects/atlas_gaussians.\n", "date": "2024-08-23", "categories": []} {"id": "2409.16294", "title": "GenCAD: Image-Conditioned Computer-Aided Design Generation with\n Transformer-Based Contrastive Representation and Diffusion Priors", "abstract": " The creation of manufacturable and editable 3D shapes through Computer-Aided\nDesign (CAD) remains a highly manual and time-consuming task, hampered by the\ncomplex topology of boundary representations of 3D solids and unintuitive\ndesign tools. While most work in the 3D shape generation literature focuses on\nrepresentations like meshes, voxels, or point clouds, practical engineering\napplications demand the modifiability and manufacturability of CAD models and\nthe ability for multi-modal conditional CAD model generation. This paper\nintroduces GenCAD, a generative model that employs autoregressive transformers\nwith a contrastive learning framework and latent diffusion models to transform\nimage inputs into parametric CAD command sequences, resulting in editable 3D\nshape representations. Extensive evaluations demonstrate that GenCAD\nsignificantly outperforms existing state-of-the-art methods in terms of the\nunconditional and conditional generations of CAD models. Additionally, the\ncontrastive learning framework of GenCAD facilitates the retrieval of CAD\nmodels using image queries from large CAD databases, which is a critical\nchallenge within the CAD community. Our results provide a significant step\nforward in highlighting the potential of generative models to expedite the\nentire design-to-production pipeline and seamlessly integrate different design\nmodalities.\n", "date": "2024-09-08", "categories": []} {"id": "2504.06982", "title": "SIGMAN:Scaling 3D Human Gaussian Generation with Millions of Assets", "abstract": " 3D human digitization has long been a highly pursued yet challenging task.\nExisting methods aim to generate high-quality 3D digital humans from single or\nmultiple views, but remain primarily constrained by current paradigms and the\nscarcity of 3D human assets. Specifically, recent approaches fall into several\nparadigms: optimization-based and feed-forward (both single-view regression and\nmulti-view generation with reconstruction). However, they are limited by slow\nspeed, low quality, cascade reasoning, and ambiguity in mapping low-dimensional\nplanes to high-dimensional space due to occlusion and invisibility,\nrespectively. Furthermore, existing 3D human assets remain small-scale,\ninsufficient for large-scale training. To address these challenges, we propose\na latent space generation paradigm for 3D human digitization, which involves\ncompressing multi-view images into Gaussians via a UV-structured VAE, along\nwith DiT-based conditional generation, we transform the ill-posed\nlow-to-high-dimensional mapping problem into a learnable distribution shift,\nwhich also supports end-to-end inference. In addition, we employ the multi-view\noptimization approach combined with synthetic data to construct the HGS-1M\ndataset, which contains $1$ million 3D Gaussian assets to support the\nlarge-scale training. Experimental results demonstrate that our paradigm,\npowered by large-scale training, produces high-quality 3D human Gaussians with\nintricate textures, facial details, and loose clothing deformation.\n", "date": "2025-04-09", "categories": []} {"id": "2504.05720", "title": "QEMesh: Employing A Quadric Error Metrics-Based Representation for Mesh\n Generation", "abstract": " Mesh generation plays a crucial role in 3D content creation, as mesh is\nwidely used in various industrial applications. Recent works have achieved\nimpressive results but still face several issues, such as unrealistic patterns\nor pits on surfaces, thin parts missing, and incomplete structures. Most of\nthese problems stem from the choice of shape representation or the capabilities\nof the generative network. To alleviate these, we extend PoNQ, a Quadric Error\nMetrics (QEM)-based representation, and propose a novel model, QEMesh, for\nhigh-quality mesh generation. PoNQ divides the shape surface into tiny patches,\neach represented by a point with its normal and QEM matrix, which preserves\nfine local geometry information. In our QEMesh, we regard these elements as\ngenerable parameters and design a unique latent diffusion model containing a\nnovel multi-decoder VAE for PoNQ parameters generation. Given the latent code\ngenerated by the diffusion model, three parameter decoders produce several PoNQ\nparameters within each voxel cell, and an occupancy decoder predicts which\nvoxel cells containing parameters to form the final shape. Extensive\nevaluations demonstrate that our method generates results with watertight\nsurfaces and is comparable to state-of-the-art methods in several main metrics.\n", "date": "2025-04-08", "categories": []} {"id": "2410.15391", "title": "Layout-your-3D: Controllable and Precise 3D Generation with 2D Blueprint", "abstract": " We present Layout-Your-3D, a framework that allows controllable and\ncompositional 3D generation from text prompts. Existing text-to-3D methods\noften struggle to generate assets with plausible object interactions or require\ntedious optimization processes. To address these challenges, our approach\nleverages 2D layouts as a blueprint to facilitate precise and plausible control\nover 3D generation. Starting with a 2D layout provided by a user or generated\nfrom a text description, we first create a coarse 3D scene using a carefully\ndesigned initialization process based on efficient reconstruction models. To\nenforce coherent global 3D layouts and enhance the quality of instance\nappearances, we propose a collision-aware layout optimization process followed\nby instance-wise refinement. Experimental results demonstrate that\nLayout-Your-3D yields more reasonable and visually appealing compositional 3D\nassets while significantly reducing the time required for each prompt.\nAdditionally, Layout-Your-3D can be easily applicable to downstream tasks, such\nas 3D editing and object insertion. Our project page is available\nat:https://colezwhy.github.io/layoutyour3d/\n", "date": "2024-10-20", "categories": []} {"id": "2504.04153", "title": "Video4DGen: Enhancing Video and 4D Generation through Mutual\n Optimization", "abstract": " The advancement of 4D (i.e., sequential 3D) generation opens up new\npossibilities for lifelike experiences in various applications, where users can\nexplore dynamic objects or characters from any viewpoint. Meanwhile, video\ngenerative models are receiving particular attention given their ability to\nproduce realistic and imaginative frames. These models are also observed to\nexhibit strong 3D consistency, indicating the potential to act as world\nsimulators. In this work, we present Video4DGen, a novel framework that excels\nin generating 4D representations from single or multiple generated videos as\nwell as generating 4D-guided videos. This framework is pivotal for creating\nhigh-fidelity virtual contents that maintain both spatial and temporal\ncoherence. The 4D outputs generated by Video4DGen are represented using our\nproposed Dynamic Gaussian Surfels (DGS), which optimizes time-varying warping\nfunctions to transform Gaussian surfels (surface elements) from a static state\nto a dynamically warped state. We design warped-state geometric regularization\nand refinements on Gaussian surfels, to preserve the structural integrity and\nfine-grained appearance details. To perform 4D generation from multiple videos\nand capture representation across spatial, temporal, and pose dimensions, we\ndesign multi-video alignment, root pose optimization, and pose-guided frame\nsampling strategies. The leveraging of continuous warping fields also enables a\nprecise depiction of pose, motion, and deformation over per-video frames.\nFurther, to improve the overall fidelity from the observation of all camera\nposes, Video4DGen performs novel-view video generation guided by the 4D\ncontent, with the proposed confidence-filtered DGS to enhance the quality of\ngenerated sequences. With the ability of 4D and video generation, Video4DGen\noffers a powerful tool for applications in virtual reality, animation, and\nbeyond.\n", "date": "2025-04-05", "categories": []} {"id": "2504.04454", "title": "PRISM: Probabilistic Representation for Integrated Shape Modeling and\n Generation", "abstract": " Despite the advancements in 3D full-shape generation, accurately modeling\ncomplex geometries and semantics of shape parts remains a significant\nchallenge, particularly for shapes with varying numbers of parts. Current\nmethods struggle to effectively integrate the contextual and structural\ninformation of 3D shapes into their generative processes. We address these\nlimitations with PRISM, a novel compositional approach for 3D shape generation\nthat integrates categorical diffusion models with Statistical Shape Models\n(SSM) and Gaussian Mixture Models (GMM). Our method employs compositional SSMs\nto capture part-level geometric variations and uses GMM to represent part\nsemantics in a continuous space. This integration enables both high fidelity\nand diversity in generated shapes while preserving structural coherence.\nThrough extensive experiments on shape generation and manipulation tasks, we\ndemonstrate that our approach significantly outperforms previous methods in\nboth quality and controllability of part-level operations. Our code will be\nmade publicly available.\n", "date": "2025-04-06", "categories": []} {"id": "2504.05152", "title": "PanoDreamer: Consistent Text to 360-Degree Scene Generation", "abstract": " Automatically generating a complete 3D scene from a text description, a\nreference image, or both has significant applications in fields like virtual\nreality and gaming. However, current methods often generate low-quality\ntextures and inconsistent 3D structures. This is especially true when\nextrapolating significantly beyond the field of view of the reference image. To\naddress these challenges, we propose PanoDreamer, a novel framework for\nconsistent, 3D scene generation with flexible text and image control. Our\napproach employs a large language model and a warp-refine pipeline, first\ngenerating an initial set of images and then compositing them into a 360-degree\npanorama. This panorama is then lifted into 3D to form an initial point cloud.\nWe then use several approaches to generate additional images, from different\nviewpoints, that are consistent with the initial point cloud and expand/refine\nthe initial point cloud. Given the resulting set of images, we utilize 3D\nGaussian Splatting to create the final 3D scene, which can then be rendered\nfrom different viewpoints. Experiments demonstrate the effectiveness of\nPanoDreamer in generating high-quality, geometrically consistent 3D scenes.\n", "date": "2025-04-07", "categories": []} {"id": "2412.07237", "title": "ArtFormer: Controllable Generation of Diverse 3D Articulated Objects", "abstract": " This paper presents a novel framework for modeling and conditional generation\nof 3D articulated objects. Troubled by flexibility-quality tradeoffs, existing\nmethods are often limited to using predefined structures or retrieving shapes\nfrom static datasets. To address these challenges, we parameterize an\narticulated object as a tree of tokens and employ a transformer to generate\nboth the object's high-level geometry code and its kinematic relations.\nSubsequently, each sub-part's geometry is further decoded using a\nsigned-distance-function (SDF) shape prior, facilitating the synthesis of\nhigh-quality 3D shapes. Our approach enables the generation of diverse objects\nwith high-quality geometry and varying number of parts. Comprehensive\nexperiments on conditional generation from text descriptions demonstrate the\neffectiveness and flexibility of our method.\n", "date": "2024-12-10", "categories": []} {"id": "2504.00457", "title": "Distilling Multi-view Diffusion Models into 3D Generators", "abstract": " We introduce DD3G, a formulation that Distills a multi-view Diffusion model\n(MV-DM) into a 3D Generator using gaussian splatting. DD3G compresses and\nintegrates extensive visual and spatial geometric knowledge from the MV-DM by\nsimulating its ordinary differential equation (ODE) trajectory, ensuring the\ndistilled generator generalizes better than those trained solely on 3D data.\nUnlike previous amortized optimization approaches, we align the MV-DM and 3D\ngenerator representation spaces to transfer the teacher's probabilistic flow to\nthe student, thus avoiding inconsistencies in optimization objectives caused by\nprobabilistic sampling. The introduction of probabilistic flow and the coupling\nof various attributes in 3D Gaussians introduce challenges in the generation\nprocess. To tackle this, we propose PEPD, a generator consisting of Pattern\nExtraction and Progressive Decoding phases, which enables efficient fusion of\nprobabilistic flow and converts a single image into 3D Gaussians within 0.06\nseconds. Furthermore, to reduce knowledge loss and overcome sparse-view\nsupervision, we design a joint optimization objective that ensures the quality\nof generated samples through explicit supervision and implicit verification.\nLeveraging existing 2D generation models, we compile 120k high-quality RGBA\nimages for distillation. Experiments on synthetic and public datasets\ndemonstrate the effectiveness of our method. Our project is available at:\nhttps://qinbaigao.github.io/DD3G_project/\n", "date": "2025-04-01", "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": "2504.02261", "title": "WonderTurbo: Generating Interactive 3D World in 0.72 Seconds", "abstract": " Interactive 3D generation is gaining momentum and capturing extensive\nattention for its potential to create immersive virtual experiences. However, a\ncritical challenge in current 3D generation technologies lies in achieving\nreal-time interactivity. To address this issue, we introduce WonderTurbo, the\nfirst real-time interactive 3D scene generation framework capable of generating\nnovel perspectives of 3D scenes within 0.72 seconds. Specifically, WonderTurbo\naccelerates both geometric and appearance modeling in 3D scene generation. In\nterms of geometry, we propose StepSplat, an innovative method that constructs\nefficient 3D geometric representations through dynamic updates, each taking\nonly 0.26 seconds. Additionally, we design QuickDepth, a lightweight depth\ncompletion module that provides consistent depth input for StepSplat, further\nenhancing geometric accuracy. For appearance modeling, we develop FastPaint, a\n2-steps diffusion model tailored for instant inpainting, which focuses on\nmaintaining spatial appearance consistency. Experimental results demonstrate\nthat WonderTurbo achieves a remarkable 15X speedup compared to baseline\nmethods, while preserving excellent spatial consistency and delivering\nhigh-quality output.\n", "date": "2025-04-03", "categories": []} {"id": "2504.02337", "title": "LPA3D: 3D Room-Level Scene Generation from In-the-Wild Images", "abstract": " Generating realistic, room-level indoor scenes with semantically plausible\nand detailed appearances from in-the-wild images is crucial for various\napplications in VR, AR, and robotics. The success of NeRF-based generative\nmethods indicates a promising direction to address this challenge. However,\nunlike their success at the object level, existing scene-level generative\nmethods require additional information, such as multiple views, depth images,\nor semantic guidance, rather than relying solely on RGB images. This is because\nNeRF-based methods necessitate prior knowledge of camera poses, which is\nchallenging to approximate for indoor scenes due to the complexity of defining\nalignment and the difficulty of globally estimating poses from a single image,\ngiven the unseen parts behind the camera. To address this challenge, we\nredefine global poses within the framework of Local-Pose-Alignment (LPA) -- an\nanchor-based multi-local-coordinate system that uses a selected number of\nanchors as the roots of these coordinates. Building on this foundation, we\nintroduce LPA-GAN, a novel NeRF-based generative approach that incorporates\nspecific modifications to estimate the priors of camera poses under LPA. It\nalso co-optimizes the pose predictor and scene generation processes. Our\nablation study and comparisons with straightforward extensions of NeRF-based\nobject generative methods demonstrate the effectiveness of our approach.\nFurthermore, visual comparisons with other techniques reveal that our method\nachieves superior view-to-view consistency and semantic normality.\n", "date": "2025-04-03", "categories": []} {"id": "2504.02762", "title": "MD-ProjTex: Texturing 3D Shapes with Multi-Diffusion Projection", "abstract": " We introduce MD-ProjTex, a method for fast and consistent text-guided texture\ngeneration for 3D shapes using pretrained text-to-image diffusion models. At\nthe core of our approach is a multi-view consistency mechanism in UV space,\nwhich ensures coherent textures across different viewpoints. Specifically,\nMD-ProjTex fuses noise predictions from multiple views at each diffusion step\nand jointly updates the per-view denoising directions to maintain 3D\nconsistency. In contrast to existing state-of-the-art methods that rely on\noptimization or sequential view synthesis, MD-ProjTex is computationally more\nefficient and achieves better quantitative and qualitative results.\n", "date": "2025-04-03", "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": "2404.09227", "title": "DreamScape: 3D Scene Creation via Gaussian Splatting joint Correlation\n Modeling", "abstract": " Recent advances in text-to-3D creation integrate the potent prior of\nDiffusion Models from text-to-image generation into 3D domain. Nevertheless,\ngenerating 3D scenes with multiple objects remains challenging. Therefore, we\npresent DreamScape, a method for generating 3D scenes from text. Utilizing\nGaussian Splatting for 3D representation, DreamScape introduces 3D Gaussian\nGuide that encodes semantic primitives, spatial transformations and\nrelationships from text using LLMs, enabling local-to-global optimization.\nProgressive scale control is tailored during local object generation,\naddressing training instability issue arising from simple blending in the\nglobal optimization stage. Collision relationships between objects are modeled\nat the global level to mitigate biases in LLMs priors, ensuring physical\ncorrectness. Additionally, to generate pervasive objects like rain and snow\ndistributed extensively across the scene, we design specialized sparse\ninitialization and densification strategy. Experiments demonstrate that\nDreamScape achieves state-of-the-art performance, enabling high-fidelity,\ncontrollable 3D scene generation.\n", "date": "2024-04-14", "categories": []} {"id": "2504.01512", "title": "High-fidelity 3D Object Generation from Single Image with RGBN-Volume\n Gaussian Reconstruction Model", "abstract": " Recently single-view 3D generation via Gaussian splatting has emerged and\ndeveloped quickly. They learn 3D Gaussians from 2D RGB images generated from\npre-trained multi-view diffusion (MVD) models, and have shown a promising\navenue for 3D generation through a single image. Despite the current progress,\nthese methods still suffer from the inconsistency jointly caused by the\ngeometric ambiguity in the 2D images, and the lack of structure of 3D\nGaussians, leading to distorted and blurry 3D object generation. In this paper,\nwe propose to fix these issues by GS-RGBN, a new RGBN-volume Gaussian\nReconstruction Model designed to generate high-fidelity 3D objects from\nsingle-view images. Our key insight is a structured 3D representation can\nsimultaneously mitigate the afore-mentioned two issues. To this end, we propose\na novel hybrid Voxel-Gaussian representation, where a 3D voxel representation\ncontains explicit 3D geometric information, eliminating the geometric ambiguity\nfrom 2D images. It also structures Gaussians during learning so that the\noptimization tends to find better local optima. Our 3D voxel representation is\nobtained by a fusion module that aligns RGB features and surface normal\nfeatures, both of which can be estimated from 2D images. Extensive experiments\ndemonstrate the superiority of our methods over prior works in terms of\nhigh-quality reconstruction results, robust generalization, and good\nefficiency.\n", "date": "2025-04-02", "categories": []} {"id": "2410.04738", "title": "Diffusion Models in 3D Vision: A Survey", "abstract": " In recent years, 3D vision has become a crucial field within computer vision,\npowering a wide range of applications such as autonomous driving, robotics,\naugmented reality, and medical imaging. This field relies on accurate\nperception, understanding, and reconstruction of 3D scenes from 2D images or\ntext data sources. Diffusion models, originally designed for 2D generative\ntasks, offer the potential for more flexible, probabilistic methods that can\nbetter capture the variability and uncertainty present in real-world 3D data.\nIn this paper, we review the state-of-the-art methods that use diffusion models\nfor 3D visual tasks, including but not limited to 3D object generation, shape\ncompletion, point-cloud reconstruction, and scene construction. We provide an\nin-depth discussion of the underlying mathematical principles of diffusion\nmodels, outlining their forward and reverse processes, as well as the various\narchitectural advancements that enable these models to work with 3D datasets.\nWe also discuss the key challenges in applying diffusion models to 3D vision,\nsuch as handling occlusions and varying point densities, and the computational\ndemands of high-dimensional data. Finally, we discuss potential solutions,\nincluding improving computational efficiency, enhancing multimodal fusion, and\nexploring the use of large-scale pretraining for better generalization across\n3D tasks. This paper serves as a foundation for future exploration and\ndevelopment in this rapidly evolving field.\n", "date": "2024-10-07", "categories": []} {"id": "2411.16820", "title": "DetailGen3D: Generative 3D Geometry Enhancement via Data-Dependent Flow", "abstract": " Modern 3D generation methods can rapidly create shapes from sparse or single\nviews, but their outputs often lack geometric detail due to computational\nconstraints. We present DetailGen3D, a generative approach specifically\ndesigned to enhance these generated 3D shapes. Our key insight is to model the\ncoarse-to-fine transformation directly through data-dependent flows in latent\nspace, avoiding the computational overhead of large-scale 3D generative models.\nWe introduce a token matching strategy that ensures accurate spatial\ncorrespondence during refinement, enabling local detail synthesis while\npreserving global structure. By carefully designing our training data to match\nthe characteristics of synthesized coarse shapes, our method can effectively\nenhance shapes produced by various 3D generation and reconstruction approaches,\nfrom single-view to sparse multi-view inputs. Extensive experiments demonstrate\nthat DetailGen3D achieves high-fidelity geometric detail synthesis while\nmaintaining efficiency in training.\n", "date": "2024-11-25", "categories": []} {"id": "2501.03397", "title": "DoubleDiffusion: Combining Heat Diffusion with Denoising Diffusion for\n Texture Generation on 3D Meshes", "abstract": " This paper addresses the problem of generating textures for 3D mesh assets.\nExisting approaches often rely on image diffusion models to generate multi-view\nimage observations, which are then transformed onto the mesh surface to produce\na single texture. However, due to the gap between multi-view images and 3D\nspace, such process is susceptible to arange of issues such as geometric\ninconsistencies, visibility occlusion, and baking artifacts. To overcome this\nproblem, we propose a novel approach that directly generates texture on 3D\nmeshes. Our approach leverages heat dissipation diffusion, which serves as an\nefficient operator that propagates features on the geometric surface of a mesh,\nwhile remaining insensitive to the specific layout of the wireframe. By\nintegrating this technique into a generative diffusion pipeline, we\nsignificantly improve the efficiency of texture generation compared to existing\ntexture generation methods. We term our approach DoubleDiffusion, as it\ncombines heat dissipation diffusion with denoising diffusion to enable native\ngenerative learning on 3D mesh surfaces.\n", "date": "2025-01-06", "categories": []} {"id": "2412.16778", "title": "RoomPainter: View-Integrated Diffusion for Consistent Indoor Scene\n Texturing", "abstract": " Indoor scene texture synthesis has garnered significant interest due to its\nimportant potential applications in virtual reality, digital media and creative\narts. Existing diffusion-model-based researches either rely on per-view\ninpainting techniques, which are plagued by severe cross-view inconsistencies\nand conspicuous seams, or adopt optimization-based approaches that involve\nsubstantial computational overhead. In this work, we present RoomPainter, a\nframework that seamlessly integrates efficiency and consistency to achieve\nhigh-fidelity texturing of indoor scenes. The core of RoomPainter features a\nzero-shot technique that effectively adapts a 2D diffusion model for\n3D-consistent texture synthesis, along with a two-stage generation strategy\nthat ensures both global and local consistency. Specifically, we introduce\nAttention-Guided Multi-View Integrated Sampling (MVIS) combined with a\nneighbor-integrated attention mechanism for zero-shot texture map generation.\nUsing the MVIS, we firstly generate texture map for the entire room to ensure\nglobal consistency, then adopt its variant, namely Attention-Guided Multi-View\nIntegrated Repaint Sampling\n (MVRS) to repaint individual instances within the room, thereby further\nenhancing local consistency and addressing the occlusion problem. Experiments\ndemonstrate that RoomPainter achieves superior performance for indoor scene\ntexture synthesis in visual quality, global consistency and generation\nefficiency.\n", "date": "2024-12-21", "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": "2503.22236", "title": "Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal\n Bridging", "abstract": " With the growing demand for high-fidelity 3D models from 2D images, existing\nmethods still face significant challenges in accurately reproducing\nfine-grained geometric details due to limitations in domain gaps and inherent\nambiguities in RGB images. To address these issues, we propose Hi3DGen, a novel\nframework for generating high-fidelity 3D geometry from images via normal\nbridging. Hi3DGen consists of three key components: (1) an image-to-normal\nestimator that decouples the low-high frequency image pattern with noise\ninjection and dual-stream training to achieve generalizable, stable, and sharp\nestimation; (2) a normal-to-geometry learning approach that uses\nnormal-regularized latent diffusion learning to enhance 3D geometry generation\nfidelity; and (3) a 3D data synthesis pipeline that constructs a high-quality\ndataset to support training. Extensive experiments demonstrate the\neffectiveness and superiority of our framework in generating rich geometric\ndetails, outperforming state-of-the-art methods in terms of fidelity. Our work\nprovides a new direction for high-fidelity 3D geometry generation from images\nby leveraging normal maps as an intermediate representation.\n", "date": "2025-03-28", "categories": []} {"id": "2503.23022", "title": "MeshCraft: Exploring Efficient and Controllable Mesh Generation with\n Flow-based DiTs", "abstract": " In the domain of 3D content creation, achieving optimal mesh topology through\nAI models has long been a pursuit for 3D artists. Previous methods, such as\nMeshGPT, have explored the generation of ready-to-use 3D objects via mesh\nauto-regressive techniques. While these methods produce visually impressive\nresults, their reliance on token-by-token predictions in the auto-regressive\nprocess leads to several significant limitations. These include extremely slow\ngeneration speeds and an uncontrollable number of mesh faces. In this paper, we\nintroduce MeshCraft, a novel framework for efficient and controllable mesh\ngeneration, which leverages continuous spatial diffusion to generate discrete\ntriangle faces. Specifically, MeshCraft consists of two core components: 1) a\ntransformer-based VAE that encodes raw meshes into continuous face-level tokens\nand decodes them back to the original meshes, and 2) a flow-based diffusion\ntransformer conditioned on the number of faces, enabling the generation of\nhigh-quality 3D meshes with a predefined number of faces. By utilizing the\ndiffusion model for the simultaneous generation of the entire mesh topology,\nMeshCraft achieves high-fidelity mesh generation at significantly faster speeds\ncompared to auto-regressive methods. Specifically, MeshCraft can generate an\n800-face mesh in just 3.2 seconds (35$\\times$ faster than existing baselines).\nExtensive experiments demonstrate that MeshCraft outperforms state-of-the-art\ntechniques in both qualitative and quantitative evaluations on ShapeNet dataset\nand demonstrates superior performance on Objaverse dataset. Moreover, it\nintegrates seamlessly with existing conditional guidance strategies, showcasing\nits potential to relieve artists from the time-consuming manual work involved\nin mesh creation.\n", "date": "2025-03-29", "categories": []} {"id": "2412.03889", "title": "CRAFT: Designing Creative and Functional 3D Objects", "abstract": " For designing a wide range of everyday objects, the design process should be\naware of both the human body and the underlying semantics of the design\nspecification. However, these two objectives present significant challenges to\nthe current AI-based designing tools. In this work, we present a method to\nsynthesize body-aware 3D objects from a base mesh given an input body geometry\nand either text or image as guidance. The generated objects can be simulated on\nvirtual characters, or fabricated for real-world use. We propose to use a mesh\ndeformation procedure that optimizes for both semantic alignment as well as\ncontact and penetration losses. Using our method, users can generate both\nvirtual or real-world objects from text, image, or sketch, without the need for\nmanual artist intervention. We present both qualitative and quantitative\nresults on various object categories, demonstrating the effectiveness of our\napproach.\n", "date": "2024-12-05", "categories": []} {"id": "2503.21830", "title": "Shape Generation via Weight Space Learning", "abstract": " Foundation models for 3D shape generation have recently shown a remarkable\ncapacity to encode rich geometric priors across both global and local\ndimensions. However, leveraging these priors for downstream tasks can be\nchallenging as real-world data are often scarce or noisy, and traditional\nfine-tuning can lead to catastrophic forgetting. In this work, we treat the\nweight space of a large 3D shape-generative model as a data modality that can\nbe explored directly. We hypothesize that submanifolds within this\nhigh-dimensional weight space can modulate topological properties or\nfine-grained part features separately, demonstrating early-stage evidence via\ntwo experiments. First, we observe a sharp phase transition in global\nconnectivity when interpolating in conditioning space, suggesting that small\nchanges in weight space can drastically alter topology. Second, we show that\nlow-dimensional reparameterizations yield controlled local geometry changes\neven with very limited data. These results highlight the potential of weight\nspace learning to unlock new approaches for 3D shape generation and specialized\nfine-tuning.\n", "date": "2025-03-26", "categories": []} {"id": "2502.06608", "title": "TripoSG: High-Fidelity 3D Shape Synthesis using Large-Scale Rectified\n Flow Models", "abstract": " Recent advancements in diffusion techniques have propelled image and video\ngeneration to unprecedented levels of quality, significantly accelerating the\ndeployment and application of generative AI. However, 3D shape generation\ntechnology has so far lagged behind, constrained by limitations in 3D data\nscale, complexity of 3D data processing, and insufficient exploration of\nadvanced techniques in the 3D domain. Current approaches to 3D shape generation\nface substantial challenges in terms of output quality, generalization\ncapability, and alignment with input conditions. We present TripoSG, a new\nstreamlined shape diffusion paradigm capable of generating high-fidelity 3D\nmeshes with precise correspondence to input images. Specifically, we propose:\n1) A large-scale rectified flow transformer for 3D shape generation, achieving\nstate-of-the-art fidelity through training on extensive, high-quality data. 2)\nA hybrid supervised training strategy combining SDF, normal, and eikonal losses\nfor 3D VAE, achieving high-quality 3D reconstruction performance. 3) A data\nprocessing pipeline to generate 2 million high-quality 3D samples, highlighting\nthe crucial rules for data quality and quantity in training 3D generative\nmodels. Through comprehensive experiments, we have validated the effectiveness\nof each component in our new framework. The seamless integration of these parts\nhas enabled TripoSG to achieve state-of-the-art performance in 3D shape\ngeneration. The resulting 3D shapes exhibit enhanced detail due to\nhigh-resolution capabilities and demonstrate exceptional fidelity to input\nimages. Moreover, TripoSG demonstrates improved versatility in generating 3D\nmodels from diverse image styles and contents, showcasing strong generalization\ncapabilities. To foster progress and innovation in the field of 3D generation,\nwe will make our model publicly available.\n", "date": "2025-02-10", "categories": []} {"id": "2503.21694", "title": "Progressive Rendering Distillation: Adapting Stable Diffusion for\n Instant Text-to-Mesh Generation without 3D Data", "abstract": " It is highly desirable to obtain a model that can generate high-quality 3D\nmeshes from text prompts in just seconds. While recent attempts have adapted\npre-trained text-to-image diffusion models, such as Stable Diffusion (SD), into\ngenerators of 3D representations (e.g., Triplane), they often suffer from poor\nquality due to the lack of sufficient high-quality 3D training data. Aiming at\novercoming the data shortage, we propose a novel training scheme, termed as\nProgressive Rendering Distillation (PRD), eliminating the need for 3D\nground-truths by distilling multi-view diffusion models and adapting SD into a\nnative 3D generator. In each iteration of training, PRD uses the U-Net to\nprogressively denoise the latent from random noise for a few steps, and in each\nstep it decodes the denoised latent into 3D output. Multi-view diffusion\nmodels, including MVDream and RichDreamer, are used in joint with SD to distill\ntext-consistent textures and geometries into the 3D outputs through score\ndistillation. Since PRD supports training without 3D ground-truths, we can\neasily scale up the training data and improve generation quality for\nchallenging text prompts with creative concepts. Meanwhile, PRD can accelerate\nthe inference speed of the generation model in just a few steps. With PRD, we\ntrain a Triplane generator, namely TriplaneTurbo, which adds only $2.5\\%$\ntrainable parameters to adapt SD for Triplane generation. TriplaneTurbo\noutperforms previous text-to-3D generators in both efficiency and quality.\nSpecifically, it can produce high-quality 3D meshes in 1.2 seconds and\ngeneralize well for challenging text input. The code is available at\nhttps://github.com/theEricMa/TriplaneTurbo.\n", "date": "2025-03-27", "categories": []} {"id": "2503.21732", "title": "SparseFlex: High-Resolution and Arbitrary-Topology 3D Shape Modeling", "abstract": " Creating high-fidelity 3D meshes with arbitrary topology, including open\nsurfaces and complex interiors, remains a significant challenge. Existing\nimplicit field methods often require costly and detail-degrading watertight\nconversion, while other approaches struggle with high resolutions. This paper\nintroduces SparseFlex, a novel sparse-structured isosurface representation that\nenables differentiable mesh reconstruction at resolutions up to $1024^3$\ndirectly from rendering losses. SparseFlex combines the accuracy of Flexicubes\nwith a sparse voxel structure, focusing computation on surface-adjacent regions\nand efficiently handling open surfaces. Crucially, we introduce a frustum-aware\nsectional voxel training strategy that activates only relevant voxels during\nrendering, dramatically reducing memory consumption and enabling\nhigh-resolution training. This also allows, for the first time, the\nreconstruction of mesh interiors using only rendering supervision. Building\nupon this, we demonstrate a complete shape modeling pipeline by training a\nvariational autoencoder (VAE) and a rectified flow transformer for high-quality\n3D shape generation. Our experiments show state-of-the-art reconstruction\naccuracy, with a ~82% reduction in Chamfer Distance and a ~88% increase in\nF-score compared to previous methods, and demonstrate the generation of\nhigh-resolution, detailed 3D shapes with arbitrary topology. By enabling\nhigh-resolution, differentiable mesh reconstruction and generation with\nrendering losses, SparseFlex significantly advances the state-of-the-art in 3D\nshape representation and modeling.\n", "date": "2025-03-27", "categories": []} {"id": "2411.17945", "title": "MARVEL-40M+: Multi-Level Visual Elaboration for High-Fidelity Text-to-3D\n Content Creation", "abstract": " Generating high-fidelity 3D content from text prompts remains a significant\nchallenge in computer vision due to the limited size, diversity, and annotation\ndepth of the existing datasets. To address this, we introduce MARVEL-40M+, an\nextensive dataset with 40 million text annotations for over 8.9 million 3D\nassets aggregated from seven major 3D datasets. Our contribution is a novel\nmulti-stage annotation pipeline that integrates open-source pretrained\nmulti-view VLMs and LLMs to automatically produce multi-level descriptions,\nranging from detailed (150-200 words) to concise semantic tags (10-20 words).\nThis structure supports both fine-grained 3D reconstruction and rapid\nprototyping. Furthermore, we incorporate human metadata from source datasets\ninto our annotation pipeline to add domain-specific information in our\nannotation and reduce VLM hallucinations. Additionally, we develop MARVEL-FX3D,\na two-stage text-to-3D pipeline. We fine-tune Stable Diffusion with our\nannotations and use a pretrained image-to-3D network to generate 3D textured\nmeshes within 15s. Extensive evaluations show that MARVEL-40M+ significantly\noutperforms existing datasets in annotation quality and linguistic diversity,\nachieving win rates of 72.41% by GPT-4 and 73.40% by human evaluators. Project\npage is available at https://sankalpsinha-cmos.github.io/MARVEL/.\n", "date": "2024-11-26", "categories": []} {"id": "2502.05874", "title": "MMGDreamer: Mixed-Modality Graph for Geometry-Controllable 3D Indoor\n Scene Generation", "abstract": " Controllable 3D scene generation has extensive applications in virtual\nreality and interior design, where the generated scenes should exhibit high\nlevels of realism and controllability in terms of geometry. Scene graphs\nprovide a suitable data representation that facilitates these applications.\nHowever, current graph-based methods for scene generation are constrained to\ntext-based inputs and exhibit insufficient adaptability to flexible user\ninputs, hindering the ability to precisely control object geometry. To address\nthis issue, we propose MMGDreamer, a dual-branch diffusion model for scene\ngeneration that incorporates a novel Mixed-Modality Graph, visual enhancement\nmodule, and relation predictor. The mixed-modality graph allows object nodes to\nintegrate textual and visual modalities, with optional relationships between\nnodes. It enhances adaptability to flexible user inputs and enables meticulous\ncontrol over the geometry of objects in the generated scenes. The visual\nenhancement module enriches the visual fidelity of text-only nodes by\nconstructing visual representations using text embeddings. Furthermore, our\nrelation predictor leverages node representations to infer absent relationships\nbetween nodes, resulting in more coherent scene layouts. Extensive experimental\nresults demonstrate that MMGDreamer exhibits superior control of object\ngeometry, achieving state-of-the-art scene generation performance. Project\npage: https://yangzhifeio.github.io/project/MMGDreamer.\n", "date": "2025-02-09", "categories": []} {"id": "2503.16302", "title": "Unleashing Vecset Diffusion Model for Fast Shape Generation", "abstract": " 3D shape generation has greatly flourished through the development of\nso-called \"native\" 3D diffusion, particularly through the Vecset Diffusion\nModel (VDM). While recent advancements have shown promising results in\ngenerating high-resolution 3D shapes, VDM still struggles with high-speed\ngeneration. Challenges exist because of difficulties not only in accelerating\ndiffusion sampling but also VAE decoding in VDM, areas under-explored in\nprevious works. To address these challenges, we present FlashVDM, a systematic\nframework for accelerating both VAE and DiT in VDM. For DiT, FlashVDM enables\nflexible diffusion sampling with as few as 5 inference steps and comparable\nquality, which is made possible by stabilizing consistency distillation with\nour newly introduced Progressive Flow Distillation. For VAE, we introduce a\nlightning vecset decoder equipped with Adaptive KV Selection, Hierarchical\nVolume Decoding, and Efficient Network Design. By exploiting the locality of\nthe vecset and the sparsity of shape surface in the volume, our decoder\ndrastically lowers FLOPs, minimizing the overall decoding overhead. We apply\nFlashVDM to Hunyuan3D-2 to obtain Hunyuan3D-2 Turbo. Through systematic\nevaluation, we show that our model significantly outperforms existing fast 3D\ngeneration methods, achieving comparable performance to the state-of-the-art\nwhile reducing inference time by over 45x for reconstruction and 32x for\ngeneration. Code and models are available at\nhttps://github.com/Tencent/FlashVDM.\n", "date": "2025-03-20", "categories": []} {"id": "2503.20784", "title": "FB-4D: Spatial-Temporal Coherent Dynamic 3D Content Generation with\n Feature Banks", "abstract": " With the rapid advancements in diffusion models and 3D generation techniques,\ndynamic 3D content generation has become a crucial research area. However,\nachieving high-fidelity 4D (dynamic 3D) generation with strong spatial-temporal\nconsistency remains a challenging task. Inspired by recent findings that\npretrained diffusion features capture rich correspondences, we propose FB-4D, a\nnovel 4D generation framework that integrates a Feature Bank mechanism to\nenhance both spatial and temporal consistency in generated frames. In FB-4D, we\nstore features extracted from previous frames and fuse them into the process of\ngenerating subsequent frames, ensuring consistent characteristics across both\ntime and multiple views. To ensure a compact representation, the Feature Bank\nis updated by a proposed dynamic merging mechanism. Leveraging this Feature\nBank, we demonstrate for the first time that generating additional reference\nsequences through multiple autoregressive iterations can continuously improve\ngeneration performance. Experimental results show that FB-4D significantly\noutperforms existing methods in terms of rendering quality, spatial-temporal\nconsistency, and robustness. It surpasses all multi-view generation tuning-free\napproaches by a large margin and achieves performance on par with\ntraining-based methods.\n", "date": "2025-03-26", "categories": []} {"id": "2406.09394", "title": "WonderWorld: Interactive 3D Scene Generation from a Single Image", "abstract": " We present WonderWorld, a novel framework for interactive 3D scene generation\nthat enables users to interactively specify scene contents and layout and see\nthe created scenes in low latency. The major challenge lies in achieving fast\ngeneration of 3D scenes. Existing scene generation approaches fall short of\nspeed as they often require (1) progressively generating many views and depth\nmaps, and (2) time-consuming optimization of the scene geometry\nrepresentations. We introduce the Fast Layered Gaussian Surfels (FLAGS) as our\nscene representation and an algorithm to generate it from a single view. Our\napproach does not need multiple views, and it leverages a geometry-based\ninitialization that significantly reduces optimization time. Another challenge\nis generating coherent geometry that allows all scenes to be connected. We\nintroduce the guided depth diffusion that allows partial conditioning of depth\nestimation. WonderWorld generates connected and diverse 3D scenes in less than\n10 seconds on a single A6000 GPU, enabling real-time user interaction and\nexploration. We demonstrate the potential of WonderWorld for user-driven\ncontent creation and exploration in virtual environments. We release full code\nand software for reproducibility. Project website:\nhttps://kovenyu.com/WonderWorld/.\n", "date": "2024-06-13", "categories": []} {"id": "2411.19292", "title": "UrbanCAD: Towards Highly Controllable and Photorealistic 3D Vehicles for\n Urban Scene Simulation", "abstract": " Photorealistic 3D vehicle models with high controllability are essential for\nautonomous driving simulation and data augmentation. While handcrafted CAD\nmodels provide flexible controllability, free CAD libraries often lack the\nhigh-quality materials necessary for photorealistic rendering. Conversely,\nreconstructed 3D models offer high-fidelity rendering but lack controllability.\nIn this work, we introduce UrbanCAD, a framework that generates highly\ncontrollable and photorealistic 3D vehicle digital twins from a single urban\nimage, leveraging a large collection of free 3D CAD models and handcrafted\nmaterials. To achieve this, we propose a novel pipeline that follows a\nretrieval-optimization manner, adapting to observational data while preserving\nfine-grained expert-designed priors for both geometry and material. This\nenables vehicles' realistic 360-degree rendering, background insertion,\nmaterial transfer, relighting, and component manipulation. Furthermore, given\nmulti-view background perspective and fisheye images, we approximate\nenvironment lighting using fisheye images and reconstruct the background with\n3DGS, enabling the photorealistic insertion of optimized CAD models into\nrendered novel view backgrounds. Experimental results demonstrate that UrbanCAD\noutperforms baselines in terms of photorealism. Additionally, we show that\nvarious perception models maintain their accuracy when evaluated on UrbanCAD\nwith in-distribution configurations but degrade when applied to realistic\nout-of-distribution data generated by our method. This suggests that UrbanCAD\nis a significant advancement in creating photorealistic, safety-critical\ndriving scenarios for downstream applications.\n", "date": "2024-11-28", "categories": []} {"id": "2502.08977", "title": "Text-driven 3D Human Generation via Contrastive Preference Optimization", "abstract": " Recent advances in Score Distillation Sampling (SDS) have improved 3D human\ngeneration from textual descriptions. However, existing methods still face\nchallenges in accurately aligning 3D models with long and complex textual\ninputs. To address this challenge, we propose a novel framework that introduces\ncontrastive preferences, where human-level preference models, guided by both\npositive and negative prompts, assist SDS for improved alignment. Specifically,\nwe design a preference optimization module that integrates multiple models to\ncomprehensively capture the full range of textual features. Furthermore, we\nintroduce a negation preference module to mitigate over-optimization of\nirrelevant details by leveraging static-dynamic negation prompts, effectively\npreventing ``reward hacking\". Extensive experiments demonstrate that our method\nachieves state-of-the-art results, significantly enhancing texture realism and\nvisual alignment with textual descriptions, particularly for long and complex\ninputs.\n", "date": "2025-02-13", "categories": []} {"id": "2503.15851", "title": "Zero-1-to-A: Zero-Shot One Image to Animatable Head Avatars Using Video\n Diffusion", "abstract": " Animatable head avatar generation typically requires extensive data for\ntraining. To reduce the data requirements, a natural solution is to leverage\nexisting data-free static avatar generation methods, such as pre-trained\ndiffusion models with score distillation sampling (SDS), which align avatars\nwith pseudo ground-truth outputs from the diffusion model. However, directly\ndistilling 4D avatars from video diffusion often leads to over-smooth results\ndue to spatial and temporal inconsistencies in the generated video. To address\nthis issue, we propose Zero-1-to-A, a robust method that synthesizes a spatial\nand temporal consistency dataset for 4D avatar reconstruction using the video\ndiffusion model. Specifically, Zero-1-to-A iteratively constructs video\ndatasets and optimizes animatable avatars in a progressive manner, ensuring\nthat avatar quality increases smoothly and consistently throughout the learning\nprocess. This progressive learning involves two stages: (1) Spatial Consistency\nLearning fixes expressions and learns from front-to-side views, and (2)\nTemporal Consistency Learning fixes views and learns from relaxed to\nexaggerated expressions, generating 4D avatars in a simple-to-complex manner.\nExtensive experiments demonstrate that Zero-1-to-A improves fidelity, animation\nquality, and rendering speed compared to existing diffusion-based methods,\nproviding a solution for lifelike avatar creation. Code is publicly available\nat: https://github.com/ZhenglinZhou/Zero-1-to-A.\n", "date": "2025-03-20", "categories": []} {"id": "2503.16396", "title": "SV4D 2.0: Enhancing Spatio-Temporal Consistency in Multi-View Video\n Diffusion for High-Quality 4D Generation", "abstract": " We present Stable Video 4D 2.0 (SV4D 2.0), a multi-view video diffusion model\nfor dynamic 3D asset generation. Compared to its predecessor SV4D, SV4D 2.0 is\nmore robust to occlusions and large motion, generalizes better to real-world\nvideos, and produces higher-quality outputs in terms of detail sharpness and\nspatio-temporal consistency. We achieve this by introducing key improvements in\nmultiple aspects: 1) network architecture: eliminating the dependency of\nreference multi-views and designing blending mechanism for 3D and frame\nattention, 2) data: enhancing quality and quantity of training data, 3)\ntraining strategy: adopting progressive 3D-4D training for better\ngeneralization, and 4) 4D optimization: handling 3D inconsistency and large\nmotion via 2-stage refinement and progressive frame sampling. Extensive\nexperiments demonstrate significant performance gain by SV4D 2.0 both visually\nand quantitatively, achieving better detail (-14\\% LPIPS) and 4D consistency\n(-44\\% FV4D) in novel-view video synthesis and 4D optimization (-12\\% LPIPS and\n-24\\% FV4D) compared to SV4D. Project page: https://sv4d20.github.io.\n", "date": "2025-03-20", "categories": []} {"id": "2503.18155", "title": "Decorum: A Language-Based Approach For Style-Conditioned Synthesis of\n Indoor 3D Scenes", "abstract": " 3D indoor scene generation is an important problem for the design of digital\nand real-world environments. To automate this process, a scene generation model\nshould be able to not only generate plausible scene layouts, but also take into\nconsideration visual features and style preferences. Existing methods for this\ntask exhibit very limited control over these attributes, only allowing text\ninputs in the form of simple object-level descriptions or pairwise spatial\nrelationships. Our proposed method Decorum enables users to control the scene\ngeneration process with natural language by adopting language-based\nrepresentations at each stage. This enables us to harness recent advancements\nin Large Language Models (LLMs) to model language-to-language mappings. In\naddition, we show that using a text-based representation allows us to select\nfurniture for our scenes using a novel object retrieval method based on\nmultimodal LLMs. Evaluations on the benchmark 3D-FRONT dataset show that our\nmethods achieve improvements over existing work in text-conditioned scene\nsynthesis and object retrieval.\n", "date": "2025-03-23", "categories": []} {"id": "2503.18476", "title": "Global-Local Tree Search in VLMs for 3D Indoor Scene Generation", "abstract": " Large Vision-Language Models (VLMs), such as GPT-4, have achieved remarkable\nsuccess across various fields. However, there are few studies on 3D indoor\nscene generation with VLMs. This paper considers this task as a planning\nproblem subject to spatial and layout common sense constraints. To solve the\nproblem with a VLM, we propose a new global-local tree search algorithm.\nGlobally, the method places each object sequentially and explores multiple\nplacements during each placement process, where the problem space is\nrepresented as a tree. To reduce the depth of the tree, we decompose the scene\nstructure hierarchically, i.e. room level, region level, floor object level,\nand supported object level. The algorithm independently generates the floor\nobjects in different regions and supported objects placed on different floor\nobjects. Locally, we also decompose the sub-task, the placement of each object,\ninto multiple steps. The algorithm searches the tree of problem space. To\nleverage the VLM model to produce positions of objects, we discretize the\ntop-down view space as a dense grid and fill each cell with diverse emojis to\nmake to cells distinct. We prompt the VLM with the emoji grid and the VLM\nproduces a reasonable location for the object by describing the position with\nthe name of emojis. The quantitative and qualitative experimental results\nillustrate our approach generates more plausible 3D scenes than\nstate-of-the-art approaches. Our source code is available at\nhttps://github.com/dw-dengwei/TreeSearchGen .\n", "date": "2025-03-24", "categories": []} {"id": "2503.19011", "title": "RomanTex: Decoupling 3D-aware Rotary Positional Embedded Multi-Attention\n Network for Texture Synthesis", "abstract": " Painting textures for existing geometries is a critical yet labor-intensive\nprocess in 3D asset generation. Recent advancements in text-to-image (T2I)\nmodels have led to significant progress in texture generation. Most existing\nresearch approaches this task by first generating images in 2D spaces using\nimage diffusion models, followed by a texture baking process to achieve UV\ntexture. However, these methods often struggle to produce high-quality textures\ndue to inconsistencies among the generated multi-view images, resulting in\nseams and ghosting artifacts. In contrast, 3D-based texture synthesis methods\naim to address these inconsistencies, but they often neglect 2D diffusion model\npriors, making them challenging to apply to real-world objects To overcome\nthese limitations, we propose RomanTex, a multiview-based texture generation\nframework that integrates a multi-attention network with an underlying 3D\nrepresentation, facilitated by our novel 3D-aware Rotary Positional Embedding.\nAdditionally, we incorporate a decoupling characteristic in the multi-attention\nblock to enhance the model's robustness in image-to-texture task, enabling\nsemantically-correct back-view synthesis. Furthermore, we introduce a\ngeometry-related Classifier-Free Guidance (CFG) mechanism to further improve\nthe alignment with both geometries and images. Quantitative and qualitative\nevaluations, along with comprehensive user studies, demonstrate that our method\nachieves state-of-the-art results in texture quality and consistency.\n", "date": "2025-03-24", "categories": []} {"id": "2411.16856", "title": "SAR3D: Autoregressive 3D Object Generation and Understanding via\n Multi-scale 3D VQVAE", "abstract": " Autoregressive models have demonstrated remarkable success across various\nfields, from large language models (LLMs) to large multimodal models (LMMs) and\n2D content generation, moving closer to artificial general intelligence (AGI).\nDespite these advances, applying autoregressive approaches to 3D object\ngeneration and understanding remains largely unexplored. This paper introduces\nScale AutoRegressive 3D (SAR3D), a novel framework that leverages a multi-scale\n3D vector-quantized variational autoencoder (VQVAE) to tokenize 3D objects for\nefficient autoregressive generation and detailed understanding. By predicting\nthe next scale in a multi-scale latent representation instead of the next\nsingle token, SAR3D reduces generation time significantly, achieving fast 3D\nobject generation in just 0.82 seconds on an A6000 GPU. Additionally, given the\ntokens enriched with hierarchical 3D-aware information, we finetune a\npretrained LLM on them, enabling multimodal comprehension of 3D content. Our\nexperiments show that SAR3D surpasses current 3D generation methods in both\nspeed and quality and allows LLMs to interpret and caption 3D models\ncomprehensively.\n", "date": "2024-11-25", "categories": []} {"id": "2411.19654", "title": "TexGaussian: Generating High-quality PBR Material via Octree-based 3D\n Gaussian Splatting", "abstract": " Physically Based Rendering (PBR) materials play a crucial role in modern\ngraphics, enabling photorealistic rendering across diverse environment maps.\nDeveloping an effective and efficient algorithm that is capable of\nautomatically generating high-quality PBR materials rather than RGB texture for\n3D meshes can significantly streamline the 3D content creation. Most existing\nmethods leverage pre-trained 2D diffusion models for multi-view image\nsynthesis, which often leads to severe inconsistency between the generated\ntextures and input 3D meshes. This paper presents TexGaussian, a novel method\nthat uses octant-aligned 3D Gaussian Splatting for rapid PBR material\ngeneration. Specifically, we place each 3D Gaussian on the finest leaf node of\nthe octree built from the input 3D mesh to render the multi-view images not\nonly for the albedo map but also for roughness and metallic. Moreover, our\nmodel is trained in a regression manner instead of diffusion denoising, capable\nof generating the PBR material for a 3D mesh in a single feed-forward process.\nExtensive experiments on publicly available benchmarks demonstrate that our\nmethod synthesizes more visually pleasing PBR materials and runs faster than\nprevious methods in both unconditional and text-conditional scenarios,\nexhibiting better consistency with the given geometry. Our code and trained\nmodels are available at https://3d-aigc.github.io/TexGaussian.\n", "date": "2024-11-29", "categories": []} {"id": "2412.17808", "title": "Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders", "abstract": " Recent 3D content generation pipelines commonly employ Variational\nAutoencoders (VAEs) to encode shapes into compact latent representations for\ndiffusion-based generation. However, the widely adopted uniform point sampling\nstrategy in Shape VAE training often leads to a significant loss of geometric\ndetails, limiting the quality of shape reconstruction and downstream generation\ntasks. We present Dora-VAE, a novel approach that enhances VAE reconstruction\nthrough our proposed sharp edge sampling strategy and a dual cross-attention\nmechanism. By identifying and prioritizing regions with high geometric\ncomplexity during training, our method significantly improves the preservation\nof fine-grained shape features. Such sampling strategy and the dual attention\nmechanism enable the VAE to focus on crucial geometric details that are\ntypically missed by uniform sampling approaches. To systematically evaluate VAE\nreconstruction quality, we additionally propose Dora-bench, a benchmark that\nquantifies shape complexity through the density of sharp edges, introducing a\nnew metric focused on reconstruction accuracy at these salient geometric\nfeatures. Extensive experiments on the Dora-bench demonstrate that Dora-VAE\nachieves comparable reconstruction quality to the state-of-the-art dense\nXCube-VAE while requiring a latent space at least 8$\\times$ smaller (1,280 vs.\n> 10,000 codes).\n", "date": "2024-12-23", "categories": []} {"id": "2502.20732", "title": "CADDreamer: CAD Object Generation from Single-view Images", "abstract": " Diffusion-based 3D generation has made remarkable progress in recent years.\nHowever, existing 3D generative models often produce overly dense and\nunstructured meshes, which stand in stark contrast to the compact, structured,\nand sharply-edged Computer-Aided Design (CAD) models crafted by human\ndesigners. To address this gap, we introduce CADDreamer, a novel approach for\ngenerating boundary representations (B-rep) of CAD objects from a single image.\nCADDreamer employs a primitive-aware multi-view diffusion model that captures\nboth local geometric details and high-level structural semantics during the\ngeneration process. By encoding primitive semantics into the color domain, the\nmethod leverages the strong priors of pre-trained diffusion models to align\nwith well-defined primitives. This enables the inference of multi-view normal\nmaps and semantic maps from a single image, facilitating the reconstruction of\na mesh with primitive labels. Furthermore, we introduce geometric optimization\ntechniques and topology-preserving extraction methods to mitigate noise and\ndistortion in the generated primitives. These enhancements result in a complete\nand seamless B-rep of the CAD model. Experimental results demonstrate that our\nmethod effectively recovers high-quality CAD objects from single-view images.\nCompared to existing 3D generation techniques, the B-rep models produced by\nCADDreamer are compact in representation, clear in structure, sharp in edges,\nand watertight in topology.\n", "date": "2025-02-28", "categories": []} {"id": "2503.16653", "title": "iFlame: Interleaving Full and Linear Attention for Efficient Mesh\n Generation", "abstract": " This paper propose iFlame, a novel transformer-based network architecture for\nmesh generation. While attention-based models have demonstrated remarkable\nperformance in mesh generation, their quadratic computational complexity limits\nscalability, particularly for high-resolution 3D data. Conversely, linear\nattention mechanisms offer lower computational costs but often struggle to\ncapture long-range dependencies, resulting in suboptimal outcomes. To address\nthis trade-off, we propose an interleaving autoregressive mesh generation\nframework that combines the efficiency of linear attention with the expressive\npower of full attention mechanisms. To further enhance efficiency and leverage\nthe inherent structure of mesh representations, we integrate this interleaving\napproach into an hourglass architecture, which significantly boosts efficiency.\nOur approach reduces training time while achieving performance comparable to\npure attention-based models. To improve inference efficiency, we implemented a\ncaching algorithm that almost doubles the speed and reduces the KV cache size\nby seven-eighths compared to the original Transformer. We evaluate our\nframework on ShapeNet and Objaverse, demonstrating its ability to generate\nhigh-quality 3D meshes efficiently. Our results indicate that the proposed\ninterleaving framework effectively balances computational efficiency and\ngenerative performance, making it a practical solution for mesh generation. The\ntraining takes only 2 days with 4 GPUs on 39k data with a maximum of 4k faces\non Objaverse.\n", "date": "2025-03-20", "categories": []} {"id": "2503.18461", "title": "MuMA: 3D PBR Texturing via Multi-Channel Multi-View Generation and\n Agentic Post-Processing", "abstract": " Current methods for 3D generation still fall short in physically based\nrendering (PBR) texturing, primarily due to limited data and challenges in\nmodeling multi-channel materials. In this work, we propose MuMA, a method for\n3D PBR texturing through Multi-channel Multi-view generation and Agentic\npost-processing. Our approach features two key innovations: 1) We opt to model\nshaded and albedo appearance channels, where the shaded channels enables the\nintegration intrinsic decomposition modules for material properties. 2)\nLeveraging multimodal large language models, we emulate artists' techniques for\nmaterial assessment and selection. Experiments demonstrate that MuMA achieves\nsuperior results in visual quality and material fidelity compared to existing\nmethods.\n", "date": "2025-03-24", "categories": []} {"id": "2412.06699", "title": "You See it, You Got it: Learning 3D Creation on Pose-Free Videos at\n Scale", "abstract": " Recent 3D generation models typically rely on limited-scale 3D `gold-labels'\nor 2D diffusion priors for 3D content creation. However, their performance is\nupper-bounded by constrained 3D priors due to the lack of scalable learning\nparadigms. In this work, we present See3D, a visual-conditional multi-view\ndiffusion model trained on large-scale Internet videos for open-world 3D\ncreation. The model aims to Get 3D knowledge by solely Seeing the visual\ncontents from the vast and rapidly growing video data -- You See it, You Got\nit. To achieve this, we first scale up the training data using a proposed data\ncuration pipeline that automatically filters out multi-view inconsistencies and\ninsufficient observations from source videos. This results in a high-quality,\nrichly diverse, large-scale dataset of multi-view images, termed WebVi3D,\ncontaining 320M frames from 16M video clips. Nevertheless, learning generic 3D\npriors from videos without explicit 3D geometry or camera pose annotations is\nnontrivial, and annotating poses for web-scale videos is prohibitively\nexpensive. To eliminate the need for pose conditions, we introduce an\ninnovative visual-condition - a purely 2D-inductive visual signal generated by\nadding time-dependent noise to the masked video data. Finally, we introduce a\nnovel visual-conditional 3D generation framework by integrating See3D into a\nwarping-based pipeline for high-fidelity 3D generation. Our numerical and\nvisual comparisons on single and sparse reconstruction benchmarks show that\nSee3D, trained on cost-effective and scalable video data, achieves notable\nzero-shot and open-world generation capabilities, markedly outperforming models\ntrained on costly and constrained 3D datasets. Please refer to our project page\nat: https://vision.baai.ac.cn/see3d\n", "date": "2024-12-09", "categories": []} {"id": "2503.01370", "title": "Kiss3DGen: Repurposing Image Diffusion Models for 3D Asset Generation", "abstract": " Diffusion models have achieved great success in generating 2D images.\nHowever, the quality and generalizability of 3D content generation remain\nlimited. State-of-the-art methods often require large-scale 3D assets for\ntraining, which are challenging to collect. In this work, we introduce\nKiss3DGen (Keep It Simple and Straightforward in 3D Generation), an efficient\nframework for generating, editing, and enhancing 3D objects by repurposing a\nwell-trained 2D image diffusion model for 3D generation. Specifically, we\nfine-tune a diffusion model to generate ''3D Bundle Image'', a tiled\nrepresentation composed of multi-view images and their corresponding normal\nmaps. The normal maps are then used to reconstruct a 3D mesh, and the\nmulti-view images provide texture mapping, resulting in a complete 3D model.\nThis simple method effectively transforms the 3D generation problem into a 2D\nimage generation task, maximizing the utilization of knowledge in pretrained\ndiffusion models. Furthermore, we demonstrate that our Kiss3DGen model is\ncompatible with various diffusion model techniques, enabling advanced features\nsuch as 3D editing, mesh and texture enhancement, etc. Through extensive\nexperiments, we demonstrate the effectiveness of our approach, showcasing its\nability to produce high-quality 3D models efficiently.\n", "date": "2025-03-03", "categories": []} {"id": "2503.16474", "title": "From Voices to Worlds: Developing an AI-Powered Framework for 3D Object\n Generation in Augmented Reality", "abstract": " This paper presents Matrix, an advanced AI-powered framework designed for\nreal-time 3D object generation in Augmented Reality (AR) environments. By\nintegrating a cutting-edge text-to-3D generative AI model, multilingual\nspeech-to-text translation, and large language models (LLMs), the system\nenables seamless user interactions through spoken commands. The framework\nprocesses speech inputs, generates 3D objects, and provides object\nrecommendations based on contextual understanding, enhancing AR experiences. A\nkey feature of this framework is its ability to optimize 3D models by reducing\nmesh complexity, resulting in significantly smaller file sizes and faster\nprocessing on resource-constrained AR devices. Our approach addresses the\nchallenges of high GPU usage, large model output sizes, and real-time system\nresponsiveness, ensuring a smoother user experience. Moreover, the system is\nequipped with a pre-generated object repository, further reducing GPU load and\nimproving efficiency. We demonstrate the practical applications of this\nframework in various fields such as education, design, and accessibility, and\ndiscuss future enhancements including image-to-3D conversion, environmental\nobject detection, and multimodal support. The open-source nature of the\nframework promotes ongoing innovation and its utility across diverse\nindustries.\n", "date": "2025-03-04", "categories": []} {"id": "2503.16611", "title": "A Recipe for Generating 3D Worlds From a Single Image", "abstract": " We introduce a recipe for generating immersive 3D worlds from a single image\nby framing the task as an in-context learning problem for 2D inpainting models.\nThis approach requires minimal training and uses existing generative models.\nOur process involves two steps: generating coherent panoramas using a\npre-trained diffusion model and lifting these into 3D with a metric depth\nestimator. We then fill unobserved regions by conditioning the inpainting model\non rendered point clouds, requiring minimal fine-tuning. Tested on both\nsynthetic and real images, our method produces high-quality 3D environments\nsuitable for VR display. By explicitly modeling the 3D structure of the\ngenerated environment from the start, our approach consistently outperforms\nstate-of-the-art, video synthesis-based methods along multiple quantitative\nimage quality metrics. Project Page: https://katjaschwarz.github.io/worlds/\n", "date": "2025-03-20", "categories": []} {"id": "2312.13016", "title": "DiffPortrait3D: Controllable Diffusion for Zero-Shot Portrait View\n Synthesis", "abstract": " We present DiffPortrait3D, a conditional diffusion model that is capable of\nsynthesizing 3D-consistent photo-realistic novel views from as few as a single\nin-the-wild portrait. Specifically, given a single RGB input, we aim to\nsynthesize plausible but consistent facial details rendered from novel camera\nviews with retained both identity and facial expression. In lieu of\ntime-consuming optimization and fine-tuning, our zero-shot method generalizes\nwell to arbitrary face portraits with unposed camera views, extreme facial\nexpressions, and diverse artistic depictions. At its core, we leverage the\ngenerative prior of 2D diffusion models pre-trained on large-scale image\ndatasets as our rendering backbone, while the denoising is guided with\ndisentangled attentive control of appearance and camera pose. To achieve this,\nwe first inject the appearance context from the reference image into the\nself-attention layers of the frozen UNets. The rendering view is then\nmanipulated with a novel conditional control module that interprets the camera\npose by watching a condition image of a crossed subject from the same view.\nFurthermore, we insert a trainable cross-view attention module to enhance view\nconsistency, which is further strengthened with a novel 3D-aware noise\ngeneration process during inference. We demonstrate state-of-the-art results\nboth qualitatively and quantitatively on our challenging in-the-wild and\nmulti-view benchmarks.\n", "date": "2023-12-20", "categories": []} {"id": "2410.16499", "title": "SINGAPO: Single Image Controlled Generation of Articulated Parts in\n Objects", "abstract": " We address the challenge of creating 3D assets for household articulated\nobjects from a single image. Prior work on articulated object creation either\nrequires multi-view multi-state input, or only allows coarse control over the\ngeneration process. These limitations hinder the scalability and practicality\nfor articulated object modeling. In this work, we propose a method to generate\narticulated objects from a single image. Observing the object in resting state\nfrom an arbitrary view, our method generates an articulated object that is\nvisually consistent with the input image. To capture the ambiguity in part\nshape and motion posed by a single view of the object, we design a diffusion\nmodel that learns the plausible variations of objects in terms of geometry and\nkinematics. To tackle the complexity of generating structured data with\nattributes in multiple domains, we design a pipeline that produces articulated\nobjects from high-level structure to geometric details in a coarse-to-fine\nmanner, where we use a part connectivity graph and part abstraction as proxies.\nOur experiments show that our method outperforms the state-of-the-art in\narticulated object creation by a large margin in terms of the generated object\nrealism, resemblance to the input image, and reconstruction quality.\n", "date": "2024-10-21", "categories": []} {"id": "2412.05279", "title": "Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories", "abstract": " Recent advancements in text-based diffusion models have accelerated progress\nin 3D reconstruction and text-based 3D editing. Although existing 3D editing\nmethods excel at modifying color, texture, and style, they struggle with\nextensive geometric or appearance changes, thus limiting their applications. To\nthis end, we propose Perturb-and-Revise, which makes possible a variety of NeRF\nediting. First, we perturb the NeRF parameters with random initializations to\ncreate a versatile initialization. The level of perturbation is determined\nautomatically through analysis of the local loss landscape. Then, we revise the\nedited NeRF via generative trajectories. Combined with the generative process,\nwe impose identity-preserving gradients to refine the edited NeRF. Extensive\nexperiments demonstrate that Perturb-and-Revise facilitates flexible,\neffective, and consistent editing of color, appearance, and geometry in 3D. For\n360{\\deg} results, please visit our project page:\nhttps://susunghong.github.io/Perturb-and-Revise.\n", "date": "2024-12-06", "categories": []} {"id": "2503.15562", "title": "Shap-MeD", "abstract": " We present Shap-MeD, a text-to-3D object generative model specialized in the\nbiomedical domain. The objective of this study is to develop an assistant that\nfacilitates the 3D modeling of medical objects, thereby reducing development\ntime. 3D modeling in medicine has various applications, including surgical\nprocedure simulation and planning, the design of personalized prosthetic\nimplants, medical education, the creation of anatomical models, and the\ndevelopment of research prototypes. To achieve this, we leverage Shap-e, an\nopen-source text-to-3D generative model developed by OpenAI, and fine-tune it\nusing a dataset of biomedical objects. Our model achieved a mean squared error\n(MSE) of 0.089 in latent generation on the evaluation set, compared to Shap-e's\nMSE of 0.147. Additionally, we conducted a qualitative evaluation, comparing\nour model with others in the generation of biomedical objects. Our results\nindicate that Shap-MeD demonstrates higher structural accuracy in biomedical\nobject generation.\n", "date": "2025-03-19", "categories": []} {"id": "2503.15667", "title": "DiffPortrait360: Consistent Portrait Diffusion for 360 View Synthesis", "abstract": " Generating high-quality 360-degree views of human heads from single-view\nimages is essential for enabling accessible immersive telepresence applications\nand scalable personalized content creation. While cutting-edge methods for full\nhead generation are limited to modeling realistic human heads, the latest\ndiffusion-based approaches for style-omniscient head synthesis can produce only\nfrontal views and struggle with view consistency, preventing their conversion\ninto true 3D models for rendering from arbitrary angles. We introduce a novel\napproach that generates fully consistent 360-degree head views, accommodating\nhuman, stylized, and anthropomorphic forms, including accessories like glasses\nand hats. Our method builds on the DiffPortrait3D framework, incorporating a\ncustom ControlNet for back-of-head detail generation and a dual appearance\nmodule to ensure global front-back consistency. By training on continuous view\nsequences and integrating a back reference image, our approach achieves robust,\nlocally continuous view synthesis. Our model can be used to produce\nhigh-quality neural radiance fields (NeRFs) for real-time, free-viewpoint\nrendering, outperforming state-of-the-art methods in object synthesis and\n360-degree head generation for very challenging input portraits.\n", "date": "2025-03-19", "categories": []} {"id": "2503.15855", "title": "VideoRFSplat: Direct Scene-Level Text-to-3D Gaussian Splatting\n Generation with Flexible Pose and Multi-View Joint Modeling", "abstract": " We propose VideoRFSplat, a direct text-to-3D model leveraging a video\ngeneration model to generate realistic 3D Gaussian Splatting (3DGS) for\nunbounded real-world scenes. To generate diverse camera poses and unbounded\nspatial extent of real-world scenes, while ensuring generalization to arbitrary\ntext prompts, previous methods fine-tune 2D generative models to jointly model\ncamera poses and multi-view images. However, these methods suffer from\ninstability when extending 2D generative models to joint modeling due to the\nmodality gap, which necessitates additional models to stabilize training and\ninference. In this work, we propose an architecture and a sampling strategy to\njointly model multi-view images and camera poses when fine-tuning a video\ngeneration model. Our core idea is a dual-stream architecture that attaches a\ndedicated pose generation model alongside a pre-trained video generation model\nvia communication blocks, generating multi-view images and camera poses through\nseparate streams. This design reduces interference between the pose and image\nmodalities. Additionally, we propose an asynchronous sampling strategy that\ndenoises camera poses faster than multi-view images, allowing rapidly denoised\nposes to condition multi-view generation, reducing mutual ambiguity and\nenhancing cross-modal consistency. Trained on multiple large-scale real-world\ndatasets (RealEstate10K, MVImgNet, DL3DV-10K, ACID), VideoRFSplat outperforms\nexisting text-to-3D direct generation methods that heavily depend on post-hoc\nrefinement via score distillation sampling, achieving superior results without\nsuch refinement.\n", "date": "2025-03-20", "categories": []} {"id": "2503.15975", "title": "Acc3D: Accelerating Single Image to 3D Diffusion Models via Edge\n Consistency Guided Score Distillation", "abstract": " We present Acc3D to tackle the challenge of accelerating the diffusion\nprocess to generate 3D models from single images. To derive high-quality\nreconstructions through few-step inferences, we emphasize the critical issue of\nregularizing the learning of score function in states of random noise. To this\nend, we propose edge consistency, i.e., consistent predictions across the high\nsignal-to-noise ratio region, to enhance a pre-trained diffusion model,\nenabling a distillation-based refinement of the endpoint score function.\nBuilding on those distilled diffusion models, we propose an adversarial\naugmentation strategy to further enrich the generation detail and boost overall\ngeneration quality. The two modules complement each other, mutually reinforcing\nto elevate generative performance. Extensive experiments demonstrate that our\nAcc3D not only achieves over a $20\\times$ increase in computational efficiency\nbut also yields notable quality improvements, compared to the\nstate-of-the-arts.\n", "date": "2025-03-20", "categories": []} {"id": "2503.16420", "title": "SynCity: Training-Free Generation of 3D Worlds", "abstract": " We address the challenge of generating 3D worlds from textual descriptions.\nWe propose SynCity, a training- and optimization-free approach, which leverages\nthe geometric precision of pre-trained 3D generative models and the artistic\nversatility of 2D image generators to create large, high-quality 3D spaces.\nWhile most 3D generative models are object-centric and cannot generate\nlarge-scale worlds, we show how 3D and 2D generators can be combined to\ngenerate ever-expanding scenes. Through a tile-based approach, we allow\nfine-grained control over the layout and the appearance of scenes. The world is\ngenerated tile-by-tile, and each new tile is generated within its world-context\nand then fused with the scene. SynCity generates compelling and immersive\nscenes that are rich in detail and diversity.\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": "2503.15265", "title": "DeepMesh: Auto-Regressive Artist-mesh Creation with Reinforcement\n Learning", "abstract": " Triangle meshes play a crucial role in 3D applications for efficient\nmanipulation and rendering. While auto-regressive methods generate structured\nmeshes by predicting discrete vertex tokens, they are often constrained by\nlimited face counts and mesh incompleteness. To address these challenges, we\npropose DeepMesh, a framework that optimizes mesh generation through two key\ninnovations: (1) an efficient pre-training strategy incorporating a novel\ntokenization algorithm, along with improvements in data curation and\nprocessing, and (2) the introduction of Reinforcement Learning (RL) into 3D\nmesh generation to achieve human preference alignment via Direct Preference\nOptimization (DPO). We design a scoring standard that combines human evaluation\nwith 3D metrics to collect preference pairs for DPO, ensuring both visual\nappeal and geometric accuracy. Conditioned on point clouds and images, DeepMesh\ngenerates meshes with intricate details and precise topology, outperforming\nstate-of-the-art methods in both precision and quality. Project page:\nhttps://zhaorw02.github.io/DeepMesh/\n", "date": "2025-03-19", "categories": []} {"id": "2311.12891", "title": "Text-Guided Texturing by Synchronized Multi-View Diffusion", "abstract": " This paper introduces a novel approach to synthesize texture to dress up a\ngiven 3D object, given a text prompt. Based on the pretrained text-to-image\n(T2I) diffusion model, existing methods usually employ a project-and-inpaint\napproach, in which a view of the given object is first generated and warped to\nanother view for inpainting. But it tends to generate inconsistent texture due\nto the asynchronous diffusion of multiple views. We believe such asynchronous\ndiffusion and insufficient information sharing among views are the root causes\nof the inconsistent artifact. In this paper, we propose a synchronized\nmulti-view diffusion approach that allows the diffusion processes from\ndifferent views to reach a consensus of the generated content early in the\nprocess, and hence ensures the texture consistency. To synchronize the\ndiffusion, we share the denoised content among different views in each\ndenoising step, specifically blending the latent content in the texture domain\nfrom views with overlap. Our method demonstrates superior performance in\ngenerating consistent, seamless, highly detailed textures, comparing to\nstate-of-the-art methods.\n", "date": "2023-11-21", "categories": []} {"id": "2411.15205", "title": "DAGSM: Disentangled Avatar Generation with GS-enhanced Mesh", "abstract": " Text-driven avatar generation has gained significant attention owing to its\nconvenience. However, existing methods typically model the human body with all\ngarments as a single 3D model, limiting its usability, such as clothing\nreplacement, and reducing user control over the generation process. To overcome\nthe limitations above, we propose DAGSM, a novel pipeline that generates\ndisentangled human bodies and garments from the given text prompts.\nSpecifically, we model each part (e.g., body, upper/lower clothes) of the\nclothed human as one GS-enhanced mesh (GSM), which is a traditional mesh\nattached with 2D Gaussians to better handle complicated textures (e.g., woolen,\ntranslucent clothes) and produce realistic cloth animations. During the\ngeneration, we first create the unclothed body, followed by a sequence of\nindividual cloth generation based on the body, where we introduce a\nsemantic-based algorithm to achieve better human-cloth and garment-garment\nseparation. To improve texture quality, we propose a view-consistent texture\nrefinement module, including a cross-view attention mechanism for texture style\nconsistency and an incident-angle-weighted denoising (IAW-DE) strategy to\nupdate the appearance. Extensive experiments have demonstrated that DAGSM\ngenerates high-quality disentangled avatars, supports clothing replacement and\nrealistic animation, and outperforms the baselines in visual quality.\n", "date": "2024-11-20", "categories": []} {"id": "2503.14002", "title": "MeshFleet: Filtered and Annotated 3D Vehicle Dataset for Domain Specific\n Generative Modeling", "abstract": " Generative models have recently made remarkable progress in the field of 3D\nobjects. However, their practical application in fields like engineering\nremains limited since they fail to deliver the accuracy, quality, and\ncontrollability needed for domain-specific tasks. Fine-tuning large generative\nmodels is a promising perspective for making these models available in these\nfields. Creating high-quality, domain-specific 3D datasets is crucial for\nfine-tuning large generative models, yet the data filtering and annotation\nprocess remains a significant bottleneck. We present MeshFleet, a filtered and\nannotated 3D vehicle dataset extracted from Objaverse-XL, the most extensive\npublicly available collection of 3D objects. Our approach proposes a pipeline\nfor automated data filtering based on a quality classifier. This classifier is\ntrained on a manually labeled subset of Objaverse, incorporating DINOv2 and\nSigLIP embeddings, refined through caption-based analysis and uncertainty\nestimation. We demonstrate the efficacy of our filtering method through a\ncomparative analysis against caption and image aesthetic score-based techniques\nand fine-tuning experiments with SV3D, highlighting the importance of targeted\ndata selection for domain-specific 3D generative modeling.\n", "date": "2025-03-18", "categories": []} {"id": "2405.12663", "title": "LAGA: Layered 3D Avatar Generation and Customization via Gaussian\n Splatting", "abstract": " Creating and customizing a 3D clothed avatar from textual descriptions is a\ncritical and challenging task. Traditional methods often treat the human body\nand clothing as inseparable, limiting users' ability to freely mix and match\ngarments. In response to this limitation, we present LAyered Gaussian Avatar\n(LAGA), a carefully designed framework enabling the creation of high-fidelity\ndecomposable avatars with diverse garments. By decoupling garments from avatar,\nour framework empowers users to conviniently edit avatars at the garment level.\nOur approach begins by modeling the avatar using a set of Gaussian points\norganized in a layered structure, where each layer corresponds to a specific\ngarment or the human body itself. To generate high-quality garments for each\nlayer, we introduce a coarse-to-fine strategy for diverse garment generation\nand a novel dual-SDS loss function to maintain coherence between the generated\ngarments and avatar components, including the human body and other garments.\nMoreover, we introduce three regularization losses to guide the movement of\nGaussians for garment transfer, allowing garments to be freely transferred to\nvarious avatars. Extensive experimentation demonstrates that our approach\nsurpasses existing methods in the generation of 3D clothed humans.\n", "date": "2024-05-21", "categories": []} {"id": "2406.15020", "title": "A3D: Does Diffusion Dream about 3D Alignment?", "abstract": " We tackle the problem of text-driven 3D generation from a geometry alignment\nperspective. Given a set of text prompts, we aim to generate a collection of\nobjects with semantically corresponding parts aligned across them. Recent\nmethods based on Score Distillation have succeeded in distilling the knowledge\nfrom 2D diffusion models to high-quality representations of the 3D objects.\nThese methods handle multiple text queries separately, and therefore the\nresulting objects have a high variability in object pose and structure.\nHowever, in some applications, such as 3D asset design, it may be desirable to\nobtain a set of objects aligned with each other. In order to achieve the\nalignment of the corresponding parts of the generated objects, we propose to\nembed these objects into a common latent space and optimize the continuous\ntransitions between these objects. We enforce two kinds of properties of these\ntransitions: smoothness of the transition and plausibility of the intermediate\nobjects along the transition. We demonstrate that both of these properties are\nessential for good alignment. We provide several practical scenarios that\nbenefit from alignment between the objects, including 3D editing and object\nhybridization, and experimentally demonstrate the effectiveness of our method.\nhttps://voyleg.github.io/a3d/\n", "date": "2024-06-21", "categories": []} {"id": "2409.12957", "title": "3DTopia-XL: Scaling High-quality 3D Asset Generation via Primitive\n Diffusion", "abstract": " The increasing demand for high-quality 3D assets across various industries\nnecessitates efficient and automated 3D content creation. Despite recent\nadvancements in 3D generative models, existing methods still face challenges\nwith optimization speed, geometric fidelity, and the lack of assets for\nphysically based rendering (PBR). In this paper, we introduce 3DTopia-XL, a\nscalable native 3D generative model designed to overcome these limitations.\n3DTopia-XL leverages a novel primitive-based 3D representation, PrimX, which\nencodes detailed shape, albedo, and material field into a compact tensorial\nformat, facilitating the modeling of high-resolution geometry with PBR assets.\nOn top of the novel representation, we propose a generative framework based on\nDiffusion Transformer (DiT), which comprises 1) Primitive Patch Compression, 2)\nand Latent Primitive Diffusion. 3DTopia-XL learns to generate high-quality 3D\nassets from textual or visual inputs. We conduct extensive qualitative and\nquantitative experiments to demonstrate that 3DTopia-XL significantly\noutperforms existing methods in generating high-quality 3D assets with\nfine-grained textures and materials, efficiently bridging the quality gap\nbetween generative models and real-world applications.\n", "date": "2024-09-19", "categories": []} {"id": "2502.02187", "title": "ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel\n Diffusion", "abstract": " This paper proposes ShapeShifter, a new 3D generative model that learns to\nsynthesize shape variations based on a single reference model. While generative\nmethods for 3D objects have recently attracted much attention, current\ntechniques often lack geometric details and/or require long training times and\nlarge resources. Our approach remedies these issues by combining sparse voxel\ngrids and point, normal, and color sampling within a multiscale neural\narchitecture that can be trained efficiently and in parallel. We show that our\nresulting variations better capture the fine details of their original input\nand can handle more general types of surfaces than previous SDF-based methods.\nMoreover, we offer interactive generation of 3D shape variants, allowing more\nhuman control in the design loop if needed.\n", "date": "2025-02-04", "categories": []} {"id": "2503.00605", "title": "GenVDM: Generating Vector Displacement Maps From a Single Image", "abstract": " We introduce the first method for generating Vector Displacement Maps (VDMs):\nparameterized, detailed geometric stamps commonly used in 3D modeling. Given a\nsingle input image, our method first generates multi-view normal maps and then\nreconstructs a VDM from the normals via a novel reconstruction pipeline. We\nalso propose an efficient algorithm for extracting VDMs from 3D objects, and\npresent the first academic VDM dataset. Compared to existing 3D generative\nmodels focusing on complete shapes, we focus on generating parts that can be\nseamlessly attached to shape surfaces. The method gives artists rich control\nover adding geometric details to a 3D shape. Experiments demonstrate that our\napproach outperforms existing baselines. Generating VDMs offers additional\nbenefits, such as using 2D image editing to customize and refine 3D details.\n", "date": "2025-03-01", "categories": []} {"id": "2503.11958", "title": "CHOrD: Generation of Collision-Free, House-Scale, and Organized Digital\n Twins for 3D Indoor Scenes with Controllable Floor Plans and Optimal Layouts", "abstract": " We introduce CHOrD, a novel framework for scalable synthesis of 3D indoor\nscenes, designed to create house-scale, collision-free, and hierarchically\nstructured indoor digital twins. In contrast to existing methods that directly\nsynthesize the scene layout as a scene graph or object list, CHOrD incorporates\na 2D image-based intermediate layout representation, enabling effective\nprevention of collision artifacts by successfully capturing them as\nout-of-distribution (OOD) scenarios during generation. Furthermore, unlike\nexisting methods, CHOrD is capable of generating scene layouts that adhere to\ncomplex floor plans with multi-modal controls, enabling the creation of\ncoherent, house-wide layouts robust to both geometric and semantic variations\nin room structures. Additionally, we propose a novel dataset with expanded\ncoverage of household items and room configurations, as well as significantly\nimproved data quality. CHOrD demonstrates state-of-the-art performance on both\nthe 3D-FRONT and our proposed datasets, delivering photorealistic, spatially\ncoherent indoor scene synthesis adaptable to arbitrary floor plan variations.\n", "date": "2025-03-15", "categories": []} {"id": "2503.12383", "title": "VRsketch2Gaussian: 3D VR Sketch Guided 3D Object Generation with\n Gaussian Splatting", "abstract": " We propose VRSketch2Gaussian, a first VR sketch-guided, multi-modal, native\n3D object generation framework that incorporates a 3D Gaussian Splatting\nrepresentation. As part of our work, we introduce VRSS, the first large-scale\npaired dataset containing VR sketches, text, images, and 3DGS, bridging the gap\nin multi-modal VR sketch-based generation. Our approach features the following\nkey innovations: 1) Sketch-CLIP feature alignment. We propose a two-stage\nalignment strategy that bridges the domain gap between sparse VR sketch\nembeddings and rich CLIP embeddings, facilitating both VR sketch-based\nretrieval and generation tasks. 2) Fine-Grained multi-modal conditioning. We\ndisentangle the 3D generation process by using explicit VR sketches for\ngeometric conditioning and text descriptions for appearance control. To\nfacilitate this, we propose a generalizable VR sketch encoder that effectively\naligns different modalities. 3) Efficient and high-fidelity 3D native\ngeneration. Our method leverages a 3D-native generation approach that enables\nfast and texture-rich 3D object synthesis. Experiments conducted on our VRSS\ndataset demonstrate that our method achieves high-quality, multi-modal VR\nsketch-based 3D generation. We believe our VRSS dataset and VRsketch2Gaussian\nmethod will be beneficial for the 3D generation community.\n", "date": "2025-03-16", "categories": []} {"id": "2503.12834", "title": "PASTA: Part-Aware Sketch-to-3D Shape Generation with Text-Aligned Prior", "abstract": " A fundamental challenge in conditional 3D shape generation is to minimize the\ninformation loss and maximize the intention of user input. Existing approaches\nhave predominantly focused on two types of isolated conditional signals, i.e.,\nuser sketches and text descriptions, each of which does not offer flexible\ncontrol of the generated shape. In this paper, we introduce PASTA, the flexible\napproach that seamlessly integrates a user sketch and a text description for 3D\nshape generation. The key idea is to use text embeddings from a vision-language\nmodel to enrich the semantic representation of sketches. Specifically, these\ntext-derived priors specify the part components of the object, compensating for\nmissing visual cues from ambiguous sketches. In addition, we introduce ISG-Net\nwhich employs two types of graph convolutional networks: IndivGCN, which\nprocesses fine-grained details, and PartGCN, which aggregates these details\ninto parts and refines the structure of objects. Extensive experiments\ndemonstrate that PASTA outperforms existing methods in part-level editing and\nachieves state-of-the-art results in sketch-to-3D shape generation.\n", "date": "2025-03-17", "categories": []} {"id": "2503.13110", "title": "DTGBrepGen: A Novel B-rep Generative Model through Decoupling Topology\n and Geometry", "abstract": " Boundary representation (B-rep) of geometric models is a fundamental format\nin Computer-Aided Design (CAD). However, automatically generating valid and\nhigh-quality B-rep models remains challenging due to the complex\ninterdependence between the topology and geometry of the models. Existing\nmethods tend to prioritize geometric representation while giving insufficient\nattention to topological constraints, making it difficult to maintain\nstructural validity and geometric accuracy. In this paper, we propose\nDTGBrepGen, a novel topology-geometry decoupled framework for B-rep generation\nthat explicitly addresses both aspects. Our approach first generates valid\ntopological structures through a two-stage process that independently models\nedge-face and edge-vertex adjacency relationships. Subsequently, we employ\nTransformer-based diffusion models for sequential geometry generation,\nprogressively generating vertex coordinates, followed by edge geometries and\nface geometries which are represented as B-splines. Extensive experiments on\ndiverse CAD datasets show that DTGBrepGen significantly outperforms existing\nmethods in both topological validity and geometric accuracy, achieving higher\nvalidity rates and producing more diverse and realistic B-reps. Our code is\npublicly available at https://github.com/jinli99/DTGBrepGen.\n", "date": "2025-03-17", "categories": []} {"id": "2503.13272", "title": "Generative Gaussian Splatting: Generating 3D Scenes with Video Diffusion\n Priors", "abstract": " Synthesizing consistent and photorealistic 3D scenes is an open problem in\ncomputer vision. Video diffusion models generate impressive videos but cannot\ndirectly synthesize 3D representations, i.e., lack 3D consistency in the\ngenerated sequences. In addition, directly training generative 3D models is\nchallenging due to a lack of 3D training data at scale. In this work, we\npresent Generative Gaussian Splatting (GGS) -- a novel approach that integrates\na 3D representation with a pre-trained latent video diffusion model.\nSpecifically, our model synthesizes a feature field parameterized via 3D\nGaussian primitives. The feature field is then either rendered to feature maps\nand decoded into multi-view images, or directly upsampled into a 3D radiance\nfield. We evaluate our approach on two common benchmark datasets for scene\nsynthesis, RealEstate10K and ScanNet+, and find that our proposed GGS model\nsignificantly improves both the 3D consistency of the generated multi-view\nimages, and the quality of the generated 3D scenes over all relevant baselines.\nCompared to a similar model without 3D representation, GGS improves FID on the\ngenerated 3D scenes by ~20% on both RealEstate10K and ScanNet+. Project page:\nhttps://katjaschwarz.github.io/ggs/\n", "date": "2025-03-17", "categories": []} {"id": "2503.13439", "title": "Amodal3R: Amodal 3D Reconstruction from Occluded 2D Images", "abstract": " Most image-based 3D object reconstructors assume that objects are fully\nvisible, ignoring occlusions that commonly occur in real-world scenarios. In\nthis paper, we introduce Amodal3R, a conditional 3D generative model designed\nto reconstruct 3D objects from partial observations. We start from a\n\"foundation\" 3D generative model and extend it to recover plausible 3D geometry\nand appearance from occluded objects. We introduce a mask-weighted multi-head\ncross-attention mechanism followed by an occlusion-aware attention layer that\nexplicitly leverages occlusion priors to guide the reconstruction process. We\ndemonstrate that, by training solely on synthetic data, Amodal3R learns to\nrecover full 3D objects even in the presence of occlusions in real scenes. It\nsubstantially outperforms existing methods that independently perform 2D amodal\ncompletion followed by 3D reconstruction, thereby establishing a new benchmark\nfor occlusion-aware 3D reconstruction.\n", "date": "2025-03-17", "categories": []} {"id": "2503.11143", "title": "GaussianIP: Identity-Preserving Realistic 3D Human Generation via\n Human-Centric Diffusion Prior", "abstract": " Text-guided 3D human generation has advanced with the development of\nefficient 3D representations and 2D-lifting methods like Score Distillation\nSampling (SDS). However, current methods suffer from prolonged training times\nand often produce results that lack fine facial and garment details. In this\npaper, we propose GaussianIP, an effective two-stage framework for generating\nidentity-preserving realistic 3D humans from text and image prompts. Our core\ninsight is to leverage human-centric knowledge to facilitate the generation\nprocess. In stage 1, we propose a novel Adaptive Human Distillation Sampling\n(AHDS) method to rapidly generate a 3D human that maintains high identity\nconsistency with the image prompt and achieves a realistic appearance. Compared\nto traditional SDS methods, AHDS better aligns with the human-centric\ngeneration process, enhancing visual quality with notably fewer training steps.\nTo further improve the visual quality of the face and clothes regions, we\ndesign a View-Consistent Refinement (VCR) strategy in stage 2. Specifically, it\nproduces detail-enhanced results of the multi-view images from stage 1\niteratively, ensuring the 3D texture consistency across views via mutual\nattention and distance-guided attention fusion. Then a polished version of the\n3D human can be achieved by directly perform reconstruction with the refined\nimages. Extensive experiments demonstrate that GaussianIP outperforms existing\nmethods in both visual quality and training efficiency, particularly in\ngenerating identity-preserving results. Our code is available at:\nhttps://github.com/silence-tang/GaussianIP.\n", "date": "2025-03-14", "categories": []} {"id": "2503.11368", "title": "PBR3DGen: A VLM-guided Mesh Generation with High-quality PBR Texture", "abstract": " Generating high-quality physically based rendering (PBR) materials is\nimportant to achieve realistic rendering in the downstream tasks, yet it\nremains challenging due to the intertwined effects of materials and lighting.\nWhile existing methods have made breakthroughs by incorporating material\ndecomposition in the 3D generation pipeline, they tend to bake highlights into\nalbedo and ignore spatially varying properties of metallicity and roughness. In\nthis work, we present PBR3DGen, a two-stage mesh generation method with\nhigh-quality PBR materials that integrates the novel multi-view PBR material\nestimation model and a 3D PBR mesh reconstruction model. Specifically, PBR3DGen\nleverages vision language models (VLM) to guide multi-view diffusion, precisely\ncapturing the spatial distribution and inherent attributes of\nreflective-metalness material. Additionally, we incorporate view-dependent\nillumination-aware conditions as pixel-aware priors to enhance spatially\nvarying material properties. Furthermore, our reconstruction model reconstructs\nhigh-quality mesh with PBR materials. Experimental results demonstrate that\nPBR3DGen significantly outperforms existing methods, achieving new\nstate-of-the-art results for PBR estimation and mesh generation. More results\nand visualization can be found on our project page:\nhttps://pbr3dgen1218.github.io/.\n", "date": "2025-03-14", "categories": []} {"id": "2503.11601", "title": "Advancing 3D Gaussian Splatting Editing with Complementary and Consensus\n Information", "abstract": " We present a novel framework for enhancing the visual fidelity and\nconsistency of text-guided 3D Gaussian Splatting (3DGS) editing. Existing\nediting approaches face two critical challenges: inconsistent geometric\nreconstructions across multiple viewpoints, particularly in challenging camera\npositions, and ineffective utilization of depth information during image\nmanipulation, resulting in over-texture artifacts and degraded object\nboundaries. To address these limitations, we introduce: 1) A complementary\ninformation mutual learning network that enhances depth map estimation from\n3DGS, enabling precise depth-conditioned 3D editing while preserving geometric\nstructures. 2) A wavelet consensus attention mechanism that effectively aligns\nlatent codes during the diffusion denoising process, ensuring multi-view\nconsistency in the edited results. Through extensive experimentation, our\nmethod demonstrates superior performance in rendering quality and view\nconsistency compared to state-of-the-art approaches. The results validate our\nframework as an effective solution for text-guided editing of 3D scenes.\n", "date": "2025-03-14", "categories": []} {"id": "2503.11629", "title": "TreeMeshGPT: Artistic Mesh Generation with Autoregressive Tree\n Sequencing", "abstract": " We introduce TreeMeshGPT, an autoregressive Transformer designed to generate\nhigh-quality artistic meshes aligned with input point clouds. Instead of the\nconventional next-token prediction in autoregressive Transformer, we propose a\nnovel Autoregressive Tree Sequencing where the next input token is retrieved\nfrom a dynamically growing tree structure that is built upon the triangle\nadjacency of faces within the mesh. Our sequencing enables the mesh to extend\nlocally from the last generated triangular face at each step, and therefore\nreduces training difficulty and improves mesh quality. Our approach represents\neach triangular face with two tokens, achieving a compression rate of\napproximately 22% compared to the naive face tokenization. This efficient\ntokenization enables our model to generate highly detailed artistic meshes with\nstrong point cloud conditioning, surpassing previous methods in both capacity\nand fidelity. Furthermore, our method generates mesh with strong normal\norientation constraints, minimizing flipped normals commonly encountered in\nprevious methods. Our experiments show that TreeMeshGPT enhances the mesh\ngeneration quality with refined details and normal orientation consistency.\n", "date": "2025-03-14", "categories": []} {"id": "2503.09601", "title": "RewardSDS: Aligning Score Distillation via Reward-Weighted Sampling", "abstract": " Score Distillation Sampling (SDS) has emerged as an effective technique for\nleveraging 2D diffusion priors for tasks such as text-to-3D generation. While\npowerful, SDS struggles with achieving fine-grained alignment to user intent.\nTo overcome this, we introduce RewardSDS, a novel approach that weights noise\nsamples based on alignment scores from a reward model, producing a weighted SDS\nloss. This loss prioritizes gradients from noise samples that yield aligned\nhigh-reward output. Our approach is broadly applicable and can extend SDS-based\nmethods. In particular, we demonstrate its applicability to Variational Score\nDistillation (VSD) by introducing RewardVSD. We evaluate RewardSDS and\nRewardVSD on text-to-image, 2D editing, and text-to-3D generation tasks,\nshowing significant improvements over SDS and VSD on a diverse set of metrics\nmeasuring generation quality and alignment to desired reward models, enabling\nstate-of-the-art performance. Project page is available at\nhttps://itaychachy.github.io/reward-sds/.\n", "date": "2025-03-12", "categories": []} {"id": "2503.10403", "title": "Hyper3D: Efficient 3D Representation via Hybrid Triplane and Octree\n Feature for Enhanced 3D Shape Variational Auto-Encoders", "abstract": " Recent 3D content generation pipelines often leverage Variational\nAutoencoders (VAEs) to encode shapes into compact latent representations,\nfacilitating diffusion-based generation. Efficiently compressing 3D shapes\nwhile preserving intricate geometric details remains a key challenge. Existing\n3D shape VAEs often employ uniform point sampling and 1D/2D latent\nrepresentations, such as vector sets or triplanes, leading to significant\ngeometric detail loss due to inadequate surface coverage and the absence of\nexplicit 3D representations in the latent space. Although recent work explores\n3D latent representations, their large scale hinders high-resolution encoding\nand efficient training. Given these challenges, we introduce Hyper3D, which\nenhances VAE reconstruction through efficient 3D representation that integrates\nhybrid triplane and octree features. First, we adopt an octree-based feature\nrepresentation to embed mesh information into the network, mitigating the\nlimitations of uniform point sampling in capturing geometric distributions\nalong the mesh surface. Furthermore, we propose a hybrid latent space\nrepresentation that integrates a high-resolution triplane with a low-resolution\n3D grid. This design not only compensates for the lack of explicit 3D\nrepresentations but also leverages a triplane to preserve high-resolution\ndetails. Experimental results demonstrate that Hyper3D outperforms traditional\nrepresentations by reconstructing 3D shapes with higher fidelity and finer\ndetails, making it well-suited for 3D generation pipelines.\n", "date": "2025-03-13", "categories": []} {"id": "2412.05929", "title": "Text-to-3D Generation by 2D Editing", "abstract": " Distilling 3D representations from pretrained 2D diffusion models is\nessential for 3D creative applications across gaming, film, and interior\ndesign. Current SDS-based methods are hindered by inefficient information\ndistillation from diffusion models, which prevents the creation of\nphotorealistic 3D contents. In this paper, we first reevaluate the SDS approach\nby analyzing its fundamental nature as a basic image editing process that\ncommonly results in over-saturation, over-smoothing, lack of rich content and\ndiversity due to the poor-quality single-step denoising. In light of this, we\nthen propose a novel method called 3D Generation by Editing (GE3D). Each\niteration of GE3D utilizes a 2D editing framework that combines a noising\ntrajectory to preserve the information of the input image, alongside a\ntext-guided denoising trajectory. We optimize the process by aligning the\nlatents across both trajectories. This approach fully exploits pretrained\ndiffusion models to distill multi-granularity information through multiple\ndenoising steps, resulting in photorealistic 3D outputs. Both theoretical and\nexperimental results confirm the effectiveness of our approach, which not only\nadvances 3D generation technology but also establishes a novel connection\nbetween 3D generation and 2D editing. This could potentially inspire further\nresearch in the field. Code and demos are released at\nhttps://jahnsonblack.github.io/GE3D/.\n", "date": "2024-12-08", "categories": []} {"id": "2404.07199", "title": "RealmDreamer: Text-Driven 3D Scene Generation with Inpainting and Depth\n Diffusion", "abstract": " We introduce RealmDreamer, a technique for generating forward-facing 3D\nscenes from text descriptions. Our method optimizes a 3D Gaussian Splatting\nrepresentation to match complex text prompts using pretrained diffusion models.\nOur key insight is to leverage 2D inpainting diffusion models conditioned on an\ninitial scene estimate to provide low variance supervision for unknown regions\nduring 3D distillation. In conjunction, we imbue high-fidelity geometry with\ngeometric distillation from a depth diffusion model, conditioned on samples\nfrom the inpainting model. We find that the initialization of the optimization\nis crucial, and provide a principled methodology for doing so. Notably, our\ntechnique doesn't require video or multi-view data and can synthesize various\nhigh-quality 3D scenes in different styles with complex layouts. Further, the\ngenerality of our method allows 3D synthesis from a single image. As measured\nby a comprehensive user study, our method outperforms all existing approaches,\npreferred by 88-95%. Project Page: https://realmdreamer.github.io/\n", "date": "2024-04-10", "categories": []} {"id": "2501.06714", "title": "F3D-Gaus: Feed-forward 3D-aware Generation on ImageNet with\n Cycle-Aggregative Gaussian Splatting", "abstract": " This paper tackles the problem of generalizable 3D-aware generation from\nmonocular datasets, e.g., ImageNet. The key challenge of this task is learning\na robust 3D-aware representation without multi-view or dynamic data, while\nensuring consistent texture and geometry across different viewpoints. Although\nsome baseline methods are capable of 3D-aware generation, the quality of the\ngenerated images still lags behind state-of-the-art 2D generation approaches,\nwhich excel in producing high-quality, detailed images. To address this severe\nlimitation, we propose a novel feed-forward pipeline based on pixel-aligned\nGaussian Splatting, coined as F3D-Gaus, which can produce more realistic and\nreliable 3D renderings from monocular inputs. In addition, we introduce a\nself-supervised cycle-aggregative constraint to enforce cross-view consistency\nin the learned 3D representation. This training strategy naturally allows\naggregation of multiple aligned Gaussian primitives and significantly\nalleviates the interpolation limitations inherent in single-view pixel-aligned\nGaussian Splatting. Furthermore, we incorporate video model priors to perform\ngeometry-aware refinement, enhancing the generation of fine details in\nwide-viewpoint scenarios and improving the model's capability to capture\nintricate 3D textures. Extensive experiments demonstrate that our approach not\nonly achieves high-quality, multi-view consistent 3D-aware generation from\nmonocular datasets, but also significantly improves training and inference\nefficiency.\n", "date": "2025-01-12", "categories": []} {"id": "2503.08165", "title": "Multimodal Generation of Animatable 3D Human Models with AvatarForge", "abstract": " We introduce AvatarForge, a framework for generating animatable 3D human\navatars from text or image inputs using AI-driven procedural generation. While\ndiffusion-based methods have made strides in general 3D object generation, they\nstruggle with high-quality, customizable human avatars due to the complexity\nand diversity of human body shapes, poses, exacerbated by the scarcity of\nhigh-quality data. Additionally, animating these avatars remains a significant\nchallenge for existing methods. AvatarForge overcomes these limitations by\ncombining LLM-based commonsense reasoning with off-the-shelf 3D human\ngenerators, enabling fine-grained control over body and facial details. Unlike\ndiffusion models which often rely on pre-trained datasets lacking precise\ncontrol over individual human features, AvatarForge offers a more flexible\napproach, bringing humans into the iterative design and modeling loop, with its\nauto-verification system allowing for continuous refinement of the generated\navatars, and thus promoting high accuracy and customization. Our evaluations\nshow that AvatarForge outperforms state-of-the-art methods in both text- and\nimage-to-avatar generation, making it a versatile tool for artistic creation\nand animation.\n", "date": "2025-03-11", "categories": []} {"id": "2503.08664", "title": "MEAT: Multiview Diffusion Model for Human Generation on Megapixels with\n Mesh Attention", "abstract": " Multiview diffusion models have shown considerable success in image-to-3D\ngeneration for general objects. However, when applied to human data, existing\nmethods have yet to deliver promising results, largely due to the challenges of\nscaling multiview attention to higher resolutions. In this paper, we explore\nhuman multiview diffusion models at the megapixel level and introduce a\nsolution called mesh attention to enable training at 1024x1024 resolution.\nUsing a clothed human mesh as a central coarse geometric representation, the\nproposed mesh attention leverages rasterization and projection to establish\ndirect cross-view coordinate correspondences. This approach significantly\nreduces the complexity of multiview attention while maintaining cross-view\nconsistency. Building on this foundation, we devise a mesh attention block and\ncombine it with keypoint conditioning to create our human-specific multiview\ndiffusion model, MEAT. In addition, we present valuable insights into applying\nmultiview human motion videos for diffusion training, addressing the\nlongstanding issue of data scarcity. Extensive experiments show that MEAT\neffectively generates dense, consistent multiview human images at the megapixel\nlevel, outperforming existing multiview diffusion methods.\n", "date": "2025-03-11", "categories": []} {"id": "2405.15305", "title": "Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve\n Rendering", "abstract": " 3D sketches are widely used for visually representing the 3D shape and\nstructure of objects or scenes. However, the creation of 3D sketch often\nrequires users to possess professional artistic skills. Existing research\nefforts primarily focus on enhancing the ability of interactive sketch\ngeneration in 3D virtual systems. In this work, we propose Diff3DS, a novel\ndifferentiable rendering framework for generating view-consistent 3D sketch by\noptimizing 3D parametric curves under various supervisions. Specifically, we\nperform perspective projection to render the 3D rational B\\'ezier curves into\n2D curves, which are subsequently converted to a 2D raster image via our\ncustomized differentiable rasterizer. Our framework bridges the domains of 3D\nsketch and raster image, achieving end-toend optimization of 3D sketch through\ngradients computed in the 2D image domain. Our Diff3DS can enable a series of\nnovel 3D sketch generation tasks, including textto-3D sketch and image-to-3D\nsketch, supported by the popular distillation-based supervision, such as Score\nDistillation Sampling (SDS). Extensive experiments have yielded promising\nresults and demonstrated the potential of our framework. Project page is at\nhttps://yiboz2001.github.io/Diff3DS/.\n", "date": "2024-05-24", "categories": []} {"id": "2503.06485", "title": "A Mesh Is Worth 512 Numbers: Spectral-domain Diffusion Modeling for\n High-dimension Shape Generation", "abstract": " Recent advancements in learning latent codes derived from high-dimensional\nshapes have demonstrated impressive outcomes in 3D generative modeling.\nTraditionally, these approaches employ a trained autoencoder to acquire a\ncontinuous implicit representation of source shapes, which can be\ncomputationally expensive. This paper introduces a novel framework,\nspectral-domain diffusion for high-quality shape generation SpoDify, that\nutilizes singular value decomposition (SVD) for shape encoding. The resulting\neigenvectors can be stored for subsequent decoding, while generative modeling\nis performed on the eigenfeatures. This approach efficiently encodes complex\nmeshes into continuous implicit representations, such as encoding a 15k-vertex\nmesh to a 512-dimensional latent code without learning. Our method exhibits\nsignificant advantages in scenarios with limited samples or GPU resources. In\nmesh generation tasks, our approach produces high-quality shapes that are\ncomparable to state-of-the-art methods.\n", "date": "2025-03-09", "categories": []} {"id": "2503.06784", "title": "Infinite Leagues Under the Sea: Photorealistic 3D Underwater Terrain\n Generation by Latent Fractal Diffusion Models", "abstract": " This paper tackles the problem of generating representations of underwater 3D\nterrain. Off-the-shelf generative models, trained on Internet-scale data but\nnot on specialized underwater images, exhibit downgraded realism, as images of\nthe seafloor are relatively uncommon. To this end, we introduce DreamSea, a\ngenerative model to generate hyper-realistic underwater scenes. DreamSea is\ntrained on real-world image databases collected from underwater robot surveys.\nImages from these surveys contain massive real seafloor observations and\ncovering large areas, but are prone to noise and artifacts from the real world.\nWe extract 3D geometry and semantics from the data with visual foundation\nmodels, and train a diffusion model that generates realistic seafloor images in\nRGBD channels, conditioned on novel fractal distribution-based latent\nembeddings. We then fuse the generated images into a 3D map, building a 3DGS\nmodel supervised by 2D diffusion priors which allows photorealistic novel view\nrendering. DreamSea is rigorously evaluated, demonstrating the ability to\nrobustly generate large-scale underwater scenes that are consistent, diverse,\nand photorealistic. Our work drives impact in multiple domains, spanning\nfilming, gaming, and robot simulation.\n", "date": "2025-03-09", "categories": []} {"id": "2503.06900", "title": "DirectTriGS: Triplane-based Gaussian Splatting Field Representation for\n 3D Generation", "abstract": " We present DirectTriGS, a novel framework designed for 3D object generation\nwith Gaussian Splatting (GS). GS-based rendering for 3D content has gained\nconsiderable attention recently. However, there has been limited exploration in\ndirectly generating 3D Gaussians compared to traditional generative modeling\napproaches. The main challenge lies in the complex data structure of GS\nrepresented by discrete point clouds with multiple channels. To overcome this\nchallenge, we propose employing the triplane representation, which allows us to\nrepresent Gaussian Splatting as an image-like continuous field. This\nrepresentation effectively encodes both the geometry and texture information,\nenabling smooth transformation back to Gaussian point clouds and rendering into\nimages by a TriRenderer, with only 2D supervisions. The proposed TriRenderer is\nfully differentiable, so that the rendering loss can supervise both texture and\ngeometry encoding. Furthermore, the triplane representation can be compressed\nusing a Variational Autoencoder (VAE), which can subsequently be utilized in\nlatent diffusion to generate 3D objects. The experiments demonstrate that the\nproposed generation framework can produce high-quality 3D object geometry and\nrendering results in the text-to-3D task.\n", "date": "2025-03-10", "categories": []} {"id": "2503.07152", "title": "Controllable 3D Outdoor Scene Generation via Scene Graphs", "abstract": " Three-dimensional scene generation is crucial in computer vision, with\napplications spanning autonomous driving, gaming and the metaverse. Current\nmethods either lack user control or rely on imprecise, non-intuitive\nconditions. In this work, we propose a method that uses, scene graphs, an\naccessible, user friendly control format to generate outdoor 3D scenes. We\ndevelop an interactive system that transforms a sparse scene graph into a dense\nBEV (Bird's Eye View) Embedding Map, which guides a conditional diffusion model\nto generate 3D scenes that match the scene graph description. During inference,\nusers can easily create or modify scene graphs to generate large-scale outdoor\nscenes. We create a large-scale dataset with paired scene graphs and 3D\nsemantic scenes to train the BEV embedding and diffusion models. Experimental\nresults show that our approach consistently produces high-quality 3D urban\nscenes closely aligned with the input scene graphs. To the best of our\nknowledge, this is the first approach to generate 3D outdoor scenes conditioned\non scene graphs.\n", "date": "2025-03-10", "categories": []} {"id": "2411.05738", "title": "StdGEN: Semantic-Decomposed 3D Character Generation from Single Images", "abstract": " We present StdGEN, an innovative pipeline for generating semantically\ndecomposed high-quality 3D characters from single images, enabling broad\napplications in virtual reality, gaming, and filmmaking, etc. Unlike previous\nmethods which struggle with limited decomposability, unsatisfactory quality,\nand long optimization times, StdGEN features decomposability, effectiveness and\nefficiency; i.e., it generates intricately detailed 3D characters with\nseparated semantic components such as the body, clothes, and hair, in three\nminutes. At the core of StdGEN is our proposed Semantic-aware Large\nReconstruction Model (S-LRM), a transformer-based generalizable model that\njointly reconstructs geometry, color and semantics from multi-view images in a\nfeed-forward manner. A differentiable multi-layer semantic surface extraction\nscheme is introduced to acquire meshes from hybrid implicit fields\nreconstructed by our S-LRM. Additionally, a specialized efficient multi-view\ndiffusion model and an iterative multi-layer surface refinement module are\nintegrated into the pipeline to facilitate high-quality, decomposable 3D\ncharacter generation. Extensive experiments demonstrate our state-of-the-art\nperformance in 3D anime character generation, surpassing existing baselines by\na significant margin in geometry, texture and decomposability. StdGEN offers\nready-to-use semantic-decomposed 3D characters and enables flexible\ncustomization for a wide range of applications. Project page:\nhttps://stdgen.github.io\n", "date": "2024-11-08", "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": "2503.02595", "title": "StageDesigner: Artistic Stage Generation for Scenography via Theater\n Scripts", "abstract": " In this work, we introduce StageDesigner, the first comprehensive framework\nfor artistic stage generation using large language models combined with\nlayout-controlled diffusion models. Given the professional requirements of\nstage scenography, StageDesigner simulates the workflows of seasoned artists to\ngenerate immersive 3D stage scenes. Specifically, our approach is divided into\nthree primary modules: Script Analysis, which extracts thematic and spatial\ncues from input scripts; Foreground Generation, which constructs and arranges\nessential 3D objects; and Background Generation, which produces a harmonious\nbackground aligned with the narrative atmosphere and maintains spatial\ncoherence by managing occlusions between foreground and background elements.\nFurthermore, we introduce the StagePro-V1 dataset, a dedicated dataset with 276\nunique stage scenes spanning different historical styles and annotated with\nscripts, images, and detailed 3D layouts, specifically tailored for this task.\nFinally, evaluations using both standard and newly proposed metrics, along with\nextensive user studies, demonstrate the effectiveness of StageDesigner. Project\ncan be found at: https://deadsmither5.github.io/2025/01/03/StageDesigner/\n", "date": "2025-03-04", "categories": []} {"id": "2312.04465", "title": "FitDiff: Robust monocular 3D facial shape and reflectance estimation\n using Diffusion Models", "abstract": " The remarkable progress in 3D face reconstruction has resulted in high-detail\nand photorealistic facial representations. Recently, Diffusion Models have\nrevolutionized the capabilities of generative methods by surpassing the\nperformance of GANs. In this work, we present FitDiff, a diffusion-based 3D\nfacial avatar generative model. Leveraging diffusion principles, our model\naccurately generates relightable facial avatars, utilizing an identity\nembedding extracted from an \"in-the-wild\" 2D facial image. The introduced\nmulti-modal diffusion model is the first to concurrently output facial\nreflectance maps (diffuse and specular albedo and normals) and shapes,\nshowcasing great generalization capabilities. It is solely trained on an\nannotated subset of a public facial dataset, paired with 3D reconstructions. We\nrevisit the typical 3D facial fitting approach by guiding a reverse diffusion\nprocess using perceptual and face recognition losses. Being the first 3D LDM\nconditioned on face recognition embeddings, FitDiff reconstructs relightable\nhuman avatars, that can be used as-is in common rendering engines, starting\nonly from an unconstrained facial image, and achieving state-of-the-art\nperformance.\n", "date": "2023-12-07", "categories": []} {"id": "2405.15252", "title": "Accelerating 3D Molecule Generation via Jointly Geometric Optimal\n Transport", "abstract": " This paper proposes a new 3D molecule generation framework, called GOAT, for\nfast and effective 3D molecule generation based on the flow-matching optimal\ntransport objective. Specifically, we formulate a geometric transport formula\nfor measuring the cost of mapping multi-modal features (e.g., continuous atom\ncoordinates and categorical atom types) between a base distribution and a\ntarget data distribution. Our formula is solved within a joint, equivariant,\nand smooth representation space. This is achieved by transforming the\nmulti-modal features into a continuous latent space with equivariant networks.\nIn addition, we find that identifying optimal distributional coupling is\nnecessary for fast and effective transport between any two distributions. We\nfurther propose a mechanism for estimating and purifying optimal coupling to\ntrain the flow model with optimal transport. By doing so, GOAT can turn\narbitrary distribution couplings into new deterministic couplings, leading to\nan estimated optimal transport plan for fast 3D molecule generation. The\npurification filters out the subpar molecules to ensure the ultimate generation\nquality. We theoretically and empirically prove that the proposed optimal\ncoupling estimation and purification yield transport plan with non-increasing\ncost. Finally, extensive experiments show that GOAT enjoys the efficiency of\nsolving geometric optimal transport, leading to a double speedup compared to\nthe sub-optimal method while achieving the best generation quality regarding\nvalidity, uniqueness, and novelty. The code is available at\nhttps://github.com/WanyuGroup/ICLR2025-GOAT.\n", "date": "2024-05-24", "categories": []} {"id": "2410.10010", "title": "InterMask: 3D Human Interaction Generation via Collaborative Masked\n Modeling", "abstract": " Generating realistic 3D human-human interactions from textual descriptions\nremains a challenging task. Existing approaches, typically based on diffusion\nmodels, often produce results lacking realism and fidelity. In this work, we\nintroduce InterMask, a novel framework for generating human interactions using\ncollaborative masked modeling in discrete space. InterMask first employs a\nVQ-VAE to transform each motion sequence into a 2D discrete motion token map.\nUnlike traditional 1D VQ token maps, it better preserves fine-grained\nspatio-temporal details and promotes spatial awareness within each token.\nBuilding on this representation, InterMask utilizes a generative masked\nmodeling framework to collaboratively model the tokens of two interacting\nindividuals. This is achieved by employing a transformer architecture\nspecifically designed to capture complex spatio-temporal inter-dependencies.\nDuring training, it randomly masks the motion tokens of both individuals and\nlearns to predict them. For inference, starting from fully masked sequences, it\nprogressively fills in the tokens for both individuals. With its enhanced\nmotion representation, dedicated architecture, and effective learning strategy,\nInterMask achieves state-of-the-art results, producing high-fidelity and\ndiverse human interactions. It outperforms previous methods, achieving an FID\nof $5.154$ (vs $5.535$ of in2IN) on the InterHuman dataset and $0.399$ (vs\n$5.207$ of InterGen) on the InterX dataset. Additionally, InterMask seamlessly\nsupports reaction generation without the need for model redesign or\nfine-tuning.\n", "date": "2024-10-13", "categories": []} {"id": "2411.18135", "title": "ModeDreamer: Mode Guiding Score Distillation for Text-to-3D Generation\n using Reference Image Prompts", "abstract": " Existing Score Distillation Sampling (SDS)-based methods have driven\nsignificant progress in text-to-3D generation. However, 3D models produced by\nSDS-based methods tend to exhibit over-smoothing and low-quality outputs. These\nissues arise from the mode-seeking behavior of current methods, where the\nscores used to update the model oscillate between multiple modes, resulting in\nunstable optimization and diminished output quality. To address this problem,\nwe introduce a novel image prompt score distillation loss named ISD, which\nemploys a reference image to direct text-to-3D optimization toward a specific\nmode. Our ISD loss can be implemented by using IP-Adapter, a lightweight\nadapter for integrating image prompt capability to a text-to-image diffusion\nmodel, as a mode-selection module. A variant of this adapter, when not being\nprompted by a reference image, can serve as an efficient control variate to\nreduce variance in score estimates, thereby enhancing both output quality and\noptimization stability. Our experiments demonstrate that the ISD loss\nconsistently achieves visually coherent, high-quality outputs and improves\noptimization speed compared to prior text-to-3D methods, as demonstrated\nthrough both qualitative and quantitative evaluations on the T3Bench benchmark\nsuite.\n", "date": "2024-11-27", "categories": []} {"id": "2501.15445", "title": "StochSync: Stochastic Diffusion Synchronization for Image Generation in\n Arbitrary Spaces", "abstract": " We propose a zero-shot method for generating images in arbitrary spaces\n(e.g., a sphere for 360{\\deg} panoramas and a mesh surface for texture) using a\npretrained image diffusion model. The zero-shot generation of various visual\ncontent using a pretrained image diffusion model has been explored mainly in\ntwo directions. First, Diffusion Synchronization-performing reverse diffusion\nprocesses jointly across different projected spaces while synchronizing them in\nthe target space-generates high-quality outputs when enough conditioning is\nprovided, but it struggles in its absence. Second, Score Distillation\nSampling-gradually updating the target space data through gradient\ndescent-results in better coherence but often lacks detail. In this paper, we\nreveal for the first time the interconnection between these two methods while\nhighlighting their differences. To this end, we propose StochSync, a novel\napproach that combines the strengths of both, enabling effective performance\nwith weak conditioning. Our experiments demonstrate that StochSync provides the\nbest performance in 360{\\deg} panorama generation (where image conditioning is\nnot given), outperforming previous finetuning-based methods, and also delivers\ncomparable results in 3D mesh texturing (where depth conditioning is provided)\nwith previous methods.\n", "date": "2025-01-26", "categories": []} {"id": "2503.00928", "title": "Revisiting CAD Model Generation by Learning Raster Sketch", "abstract": " The integration of deep generative networks into generating Computer-Aided\nDesign (CAD) models has garnered increasing attention over recent years.\nTraditional methods often rely on discrete sequences of parametric line/curve\nsegments to represent sketches. Differently, we introduce RECAD, a novel\nframework that generates Raster sketches and 3D Extrusions for CAD models.\nRepresenting sketches as raster images offers several advantages over discrete\nsequences: 1) it breaks the limitations on the types and numbers of\nlines/curves, providing enhanced geometric representation capabilities; 2) it\nenables interpolation within a continuous latent space; and 3) it allows for\nmore intuitive user control over the output. Technically, RECAD employs two\ndiffusion networks: the first network generates extrusion boxes conditioned on\nthe number and types of extrusions, while the second network produces sketch\nimages conditioned on these extrusion boxes. By combining these two networks,\nRECAD effectively generates sketch-and-extrude CAD models, offering a more\nrobust and intuitive approach to CAD model generation. Experimental results\nindicate that RECAD achieves strong performance in unconditional generation,\nwhile also demonstrating effectiveness in conditional generation and output\nediting.\n", "date": "2025-03-02", "categories": []} {"id": "2407.17470", "title": "SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View\n Consistency", "abstract": " We present Stable Video 4D (SV4D), a latent video diffusion model for\nmulti-frame and multi-view consistent dynamic 3D content generation. Unlike\nprevious methods that rely on separately trained generative models for video\ngeneration and novel view synthesis, we design a unified diffusion model to\ngenerate novel view videos of dynamic 3D objects. Specifically, given a\nmonocular reference video, SV4D generates novel views for each video frame that\nare temporally consistent. We then use the generated novel view videos to\noptimize an implicit 4D representation (dynamic NeRF) efficiently, without the\nneed for cumbersome SDS-based optimization used in most prior works. To train\nour unified novel view video generation model, we curate a dynamic 3D object\ndataset from the existing Objaverse dataset. Extensive experimental results on\nmultiple datasets and user studies demonstrate SV4D's state-of-the-art\nperformance on novel-view video synthesis as well as 4D generation compared to\nprior works.\n", "date": "2024-07-24", "categories": []} {"id": "2405.00915", "title": "EchoScene: Indoor Scene Generation via Information Echo over Scene Graph\n Diffusion", "abstract": " We present EchoScene, an interactive and controllable generative model that\ngenerates 3D indoor scenes on scene graphs. EchoScene leverages a dual-branch\ndiffusion model that dynamically adapts to scene graphs. Existing methods\nstruggle to handle scene graphs due to varying numbers of nodes, multiple edge\ncombinations, and manipulator-induced node-edge operations. EchoScene overcomes\nthis by associating each node with a denoising process and enables\ncollaborative information exchange, enhancing controllable and consistent\ngeneration aware of global constraints. This is achieved through an information\necho scheme in both shape and layout branches. At every denoising step, all\nprocesses share their denoising data with an information exchange unit that\ncombines these updates using graph convolution. The scheme ensures that the\ndenoising processes are influenced by a holistic understanding of the scene\ngraph, facilitating the generation of globally coherent scenes. The resulting\nscenes can be manipulated during inference by editing the input scene graph and\nsampling the noise in the diffusion model. Extensive experiments validate our\napproach, which maintains scene controllability and surpasses previous methods\nin generation fidelity. Moreover, the generated scenes are of high quality and\nthus directly compatible with off-the-shelf texture generation. Code and\ntrained models are open-sourced.\n", "date": "2024-05-02", "categories": []} {"id": "2406.06526", "title": "Generative Gaussian Splatting for Unbounded 3D City Generation", "abstract": " 3D city generation with NeRF-based methods shows promising generation results\nbut is computationally inefficient. Recently 3D Gaussian Splatting (3D-GS) has\nemerged as a highly efficient alternative for object-level 3D generation.\nHowever, adapting 3D-GS from finite-scale 3D objects and humans to\ninfinite-scale 3D cities is non-trivial. Unbounded 3D city generation entails\nsignificant storage overhead (out-of-memory issues), arising from the need to\nexpand points to billions, often demanding hundreds of Gigabytes of VRAM for a\ncity scene spanning 10km^2. In this paper, we propose GaussianCity, a\ngenerative Gaussian Splatting framework dedicated to efficiently synthesizing\nunbounded 3D cities with a single feed-forward pass. Our key insights are\ntwo-fold: 1) Compact 3D Scene Representation: We introduce BEV-Point as a\nhighly compact intermediate representation, ensuring that the growth in VRAM\nusage for unbounded scenes remains constant, thus enabling unbounded city\ngeneration. 2) Spatial-aware Gaussian Attribute Decoder: We present\nspatial-aware BEV-Point decoder to produce 3D Gaussian attributes, which\nleverages Point Serializer to integrate the structural and contextual\ncharacteristics of BEV points. Extensive experiments demonstrate that\nGaussianCity achieves state-of-the-art results in both drone-view and\nstreet-view 3D city generation. Notably, compared to CityDreamer, GaussianCity\nexhibits superior performance with a speedup of 60 times (10.72 FPS v.s. 0.18\nFPS).\n", "date": "2024-06-10", "categories": []} {"id": "2502.09278", "title": "ConsistentDreamer: View-Consistent Meshes Through Balanced Multi-View\n Gaussian Optimization", "abstract": " Recent advances in diffusion models have significantly improved 3D\ngeneration, enabling the use of assets generated from an image for embodied AI\nsimulations. However, the one-to-many nature of the image-to-3D problem limits\ntheir use due to inconsistent content and quality across views. Previous models\noptimize a 3D model by sampling views from a view-conditioned diffusion prior,\nbut diffusion models cannot guarantee view consistency. Instead, we present\nConsistentDreamer, where we first generate a set of fixed multi-view prior\nimages and sample random views between them with another diffusion model\nthrough a score distillation sampling (SDS) loss. Thereby, we limit the\ndiscrepancies between the views guided by the SDS loss and ensure a consistent\nrough shape. In each iteration, we also use our generated multi-view prior\nimages for fine-detail reconstruction. To balance between the rough shape and\nthe fine-detail optimizations, we introduce dynamic task-dependent weights\nbased on homoscedastic uncertainty, updated automatically in each iteration.\nAdditionally, we employ opacity, depth distortion, and normal alignment losses\nto refine the surface for mesh extraction. Our method ensures better view\nconsistency and visual quality compared to the state-of-the-art.\n", "date": "2025-02-13", "categories": []} {"id": "2502.14247", "title": "Pandora3D: A Comprehensive Framework for High-Quality 3D Shape and\n Texture Generation", "abstract": " This report presents a comprehensive framework for generating high-quality 3D\nshapes and textures from diverse input prompts, including single images,\nmulti-view images, and text descriptions. The framework consists of 3D shape\ngeneration and texture generation. (1). The 3D shape generation pipeline\nemploys a Variational Autoencoder (VAE) to encode implicit 3D geometries into a\nlatent space and a diffusion network to generate latents conditioned on input\nprompts, with modifications to enhance model capacity. An alternative\nArtist-Created Mesh (AM) generation approach is also explored, yielding\npromising results for simpler geometries. (2). Texture generation involves a\nmulti-stage process starting with frontal images generation followed by\nmulti-view images generation, RGB-to-PBR texture conversion, and\nhigh-resolution multi-view texture refinement. A consistency scheduler is\nplugged into every stage, to enforce pixel-wise consistency among multi-view\ntextures during inference, ensuring seamless integration.\n The pipeline demonstrates effective handling of diverse input formats,\nleveraging advanced neural architectures and novel methodologies to produce\nhigh-quality 3D content. This report details the system architecture,\nexperimental results, and potential future directions to improve and expand the\nframework. The source code and pretrained weights are released at:\nhttps://github.com/Tencent/Tencent-XR-3DGen.\n", "date": "2025-02-20", "categories": []} {"id": "2502.16475", "title": "Dragen3D: Multiview Geometry Consistent 3D Gaussian Generation with\n Drag-Based Control", "abstract": " Single-image 3D generation has emerged as a prominent research topic, playing\na vital role in virtual reality, 3D modeling, and digital content creation.\nHowever, existing methods face challenges such as a lack of multi-view\ngeometric consistency and limited controllability during the generation\nprocess, which significantly restrict their usability. % To tackle these\nchallenges, we introduce Dragen3D, a novel approach that achieves geometrically\nconsistent and controllable 3D generation leveraging 3D Gaussian Splatting\n(3DGS). We introduce the Anchor-Gaussian Variational Autoencoder (Anchor-GS\nVAE), which encodes a point cloud and a single image into anchor latents and\ndecode these latents into 3DGS, enabling efficient latent-space generation. To\nenable multi-view geometry consistent and controllable generation, we propose a\nSeed-Point-Driven strategy: first generate sparse seed points as a coarse\ngeometry representation, then map them to anchor latents via the Seed-Anchor\nMapping Module. Geometric consistency is ensured by the easily learned sparse\nseed points, and users can intuitively drag the seed points to deform the final\n3DGS geometry, with changes propagated through the anchor latents. To the best\nof our knowledge, we are the first to achieve geometrically controllable 3D\nGaussian generation and editing without relying on 2D diffusion priors,\ndelivering comparable 3D generation quality to state-of-the-art methods.\n", "date": "2025-02-23", "categories": []} {"id": "2408.13252", "title": "LayerPano3D: Layered 3D Panorama for Hyper-Immersive Scene Generation", "abstract": " 3D immersive scene generation is a challenging yet critical task in computer\nvision and graphics. A desired virtual 3D scene should 1) exhibit\nomnidirectional view consistency, and 2) allow for free exploration in complex\nscene hierarchies. Existing methods either rely on successive scene expansion\nvia inpainting or employ panorama representation to represent large FOV scene\nenvironments. However, the generated scene suffers from semantic drift during\nexpansion and is unable to handle occlusion among scene hierarchies. To tackle\nthese challenges, we introduce Layerpano3D, a novel framework for full-view,\nexplorable panoramic 3D scene generation from a single text prompt. Our key\ninsight is to decompose a reference 2D panorama into multiple layers at\ndifferent depth levels, where each layer reveals the unseen space from the\nreference views via diffusion prior. Layerpano3D comprises multiple dedicated\ndesigns: 1) We introduce a new panorama dataset Upright360, comprising 9k\nhigh-quality and upright panorama images, and finetune the advanced Flux model\non Upright360 for high-quality, upright and consistent panorama generation. 2)\nWe pioneer the Layered 3D Panorama as underlying representation to manage\ncomplex scene hierarchies and lift it into 3D Gaussians to splat detailed\n360-degree omnidirectional scenes with unconstrained viewing paths. Extensive\nexperiments demonstrate that our framework generates state-of-the-art 3D\npanoramic scene in both full view consistency and immersive exploratory\nexperience. We believe that Layerpano3D holds promise for advancing 3D\npanoramic scene creation with numerous applications.\n", "date": "2024-08-23", "categories": []} {"id": "2410.18974", "title": "3D-Adapter: Geometry-Consistent Multi-View Diffusion for High-Quality 3D\n Generation", "abstract": " Multi-view image diffusion models have significantly advanced open-domain 3D\nobject generation. However, most existing models rely on 2D network\narchitectures that lack inherent 3D biases, resulting in compromised geometric\nconsistency. To address this challenge, we introduce 3D-Adapter, a plug-in\nmodule designed to infuse 3D geometry awareness into pretrained image diffusion\nmodels. Central to our approach is the idea of 3D feedback augmentation: for\neach denoising step in the sampling loop, 3D-Adapter decodes intermediate\nmulti-view features into a coherent 3D representation, then re-encodes the\nrendered RGBD views to augment the pretrained base model through feature\naddition. We study two variants of 3D-Adapter: a fast feed-forward version\nbased on Gaussian splatting and a versatile training-free version utilizing\nneural fields and meshes. Our extensive experiments demonstrate that 3D-Adapter\nnot only greatly enhances the geometry quality of text-to-multi-view models\nsuch as Instant3D and Zero123++, but also enables high-quality 3D generation\nusing the plain text-to-image Stable Diffusion. Furthermore, we showcase the\nbroad application potential of 3D-Adapter by presenting high quality results in\ntext-to-3D, image-to-3D, text-to-texture, and text-to-avatar tasks.\n", "date": "2024-10-24", "categories": []} {"id": "2502.14316", "title": "Textured 3D Regenerative Morphing with 3D Diffusion Prior", "abstract": " Textured 3D morphing creates smooth and plausible interpolation sequences\nbetween two 3D objects, focusing on transitions in both shape and texture. This\nis important for creative applications like visual effects in filmmaking.\nPrevious methods rely on establishing point-to-point correspondences and\ndetermining smooth deformation trajectories, which inherently restrict them to\nshape-only morphing on untextured, topologically aligned datasets. This\nrestriction leads to labor-intensive preprocessing and poor generalization. To\novercome these challenges, we propose a method for 3D regenerative morphing\nusing a 3D diffusion prior. Unlike previous methods that depend on explicit\ncorrespondences and deformations, our method eliminates the additional need for\nobtaining correspondence and uses the 3D diffusion prior to generate morphing.\nSpecifically, we introduce a 3D diffusion model and interpolate the source and\ntarget information at three levels: initial noise, model parameters, and\ncondition features. We then explore an Attention Fusion strategy to generate\nmore smooth morphing sequences. To further improve the plausibility of semantic\ninterpolation and the generated 3D surfaces, we propose two strategies: (a)\nToken Reordering, where we match approximate tokens based on semantic analysis\nto guide implicit correspondences in the denoising process of the diffusion\nmodel, and (b) Low-Frequency Enhancement, where we enhance low-frequency\nsignals in the tokens to improve the quality of generated surfaces.\nExperimental results show that our method achieves superior smoothness and\nplausibility in 3D morphing across diverse cross-category object pairs,\noffering a novel regenerative method for 3D morphing with textured\nrepresentations.\n", "date": "2025-02-20", "categories": []} {"id": "2312.11535", "title": "High-Quality 3D Creation from A Single Image Using Subject-Specific\n Knowledge Prior", "abstract": " In this paper, we address the critical bottleneck in robotics caused by the\nscarcity of diverse 3D data by presenting a novel two-stage approach for\ngenerating high-quality 3D models from a single image. This method is motivated\nby the need to efficiently expand 3D asset creation, particularly for robotics\ndatasets, where the variety of object types is currently limited compared to\ngeneral image datasets. Unlike previous methods that primarily rely on general\ndiffusion priors, which often struggle to align with the reference image, our\napproach leverages subject-specific prior knowledge. By incorporating\nsubject-specific priors in both geometry and texture, we ensure precise\nalignment between the generated 3D content and the reference object.\nSpecifically, we introduce a shading mode-aware prior into the NeRF\noptimization process, enhancing the geometry and refining texture in the coarse\noutputs to achieve superior quality. Extensive experiments demonstrate that our\nmethod significantly outperforms prior approaches.\n", "date": "2023-12-15", "categories": []} {"id": "2502.12640", "title": "RecDreamer: Consistent Text-to-3D Generation via Uniform Score\n Distillation", "abstract": " Current text-to-3D generation methods based on score distillation often\nsuffer from geometric inconsistencies, leading to repeated patterns across\ndifferent poses of 3D assets. This issue, known as the Multi-Face Janus\nproblem, arises because existing methods struggle to maintain consistency\nacross varying poses and are biased toward a canonical pose. While recent work\nhas improved pose control and approximation, these efforts are still limited by\nthis inherent bias, which skews the guidance during generation. To address\nthis, we propose a solution called RecDreamer, which reshapes the underlying\ndata distribution to achieve a more consistent pose representation. The core\nidea behind our method is to rectify the prior distribution, ensuring that pose\nvariation is uniformly distributed rather than biased toward a canonical form.\nBy modifying the prescribed distribution through an auxiliary function, we can\nreconstruct the density of the distribution to ensure compliance with specific\nmarginal constraints. In particular, we ensure that the marginal distribution\nof poses follows a uniform distribution, thereby eliminating the biases\nintroduced by the prior knowledge. We incorporate this rectified data\ndistribution into existing score distillation algorithms, a process we refer to\nas uniform score distillation. To efficiently compute the posterior\ndistribution required for the auxiliary function, RecDreamer introduces a\ntraining-free classifier that estimates pose categories in a plug-and-play\nmanner. Additionally, we utilize various approximation techniques for noisy\nstates, significantly improving system performance. Our experimental results\ndemonstrate that RecDreamer effectively mitigates the Multi-Face Janus problem,\nleading to more consistent 3D asset generation across different poses.\n", "date": "2025-02-18", "categories": []} {"id": "2502.10675", "title": "Hierarchically-Structured Open-Vocabulary Indoor Scene Synthesis with\n Pre-trained Large Language Model", "abstract": " Indoor scene synthesis aims to automatically produce plausible, realistic and\ndiverse 3D indoor scenes, especially given arbitrary user requirements.\nRecently, the promising generalization ability of pre-trained large language\nmodels (LLM) assist in open-vocabulary indoor scene synthesis. However, the\nchallenge lies in converting the LLM-generated outputs into reasonable and\nphysically feasible scene layouts. In this paper, we propose to generate\nhierarchically structured scene descriptions with LLM and then compute the\nscene layouts. Specifically, we train a hierarchy-aware network to infer the\nfine-grained relative positions between objects and design a divide-and-conquer\noptimization to solve for scene layouts. The advantages of using hierarchically\nstructured scene representation are two-fold. First, the hierarchical structure\nprovides a rough grounding for object arrangement, which alleviates\ncontradictory placements with dense relations and enhances the generalization\nability of the network to infer fine-grained placements. Second, it naturally\nsupports the divide-and-conquer optimization, by first arranging the sub-scenes\nand then the entire scene, to more effectively solve for a feasible layout. We\nconduct extensive comparison experiments and ablation studies with both\nqualitative and quantitative evaluations to validate the effectiveness of our\nkey designs with the hierarchically structured scene representation. Our\napproach can generate more reasonable scene layouts while better aligned with\nthe user requirements and LLM descriptions. We also present open-vocabulary\nscene synthesis and interactive scene design results to show the strength of\nour approach in the applications.\n", "date": "2025-02-15", "categories": []} {"id": "2502.11390", "title": "MARS: Mesh AutoRegressive Model for 3D Shape Detailization", "abstract": " State-of-the-art methods for mesh detailization predominantly utilize\nGenerative Adversarial Networks (GANs) to generate detailed meshes from coarse\nones. These methods typically learn a specific style code for each category or\nsimilar categories without enforcing geometry supervision across different\nLevels of Detail (LODs). Consequently, such methods often fail to generalize\nacross a broader range of categories and cannot ensure shape consistency\nthroughout the detailization process. In this paper, we introduce MARS, a novel\napproach for 3D shape detailization. Our method capitalizes on a novel\nmulti-LOD, multi-category mesh representation to learn shape-consistent mesh\nrepresentations in latent space across different LODs. We further propose a\nmesh autoregressive model capable of generating such latent representations\nthrough next-LOD token prediction. This approach significantly enhances the\nrealism of the generated shapes. Extensive experiments conducted on the\nchallenging 3D Shape Detailization benchmark demonstrate that our proposed MARS\nmodel achieves state-of-the-art performance, surpassing existing methods in\nboth qualitative and quantitative assessments. Notably, the model's capability\nto generate fine-grained details while preserving the overall shape integrity\nis particularly commendable.\n", "date": "2025-02-17", "categories": []} {"id": "2502.11586", "title": "Syllables to Scenes: Literary-Guided Free-Viewpoint 3D Scene Synthesis\n from Japanese Haiku", "abstract": " In the era of the metaverse, where immersive technologies redefine human\nexperiences, translating abstract literary concepts into navigable 3D\nenvironments presents a fundamental challenge in preserving semantic and\nemotional fidelity. This research introduces HaikuVerse, a novel framework for\ntransforming poetic abstraction into spatial representation, with Japanese\nHaiku serving as an ideal test case due to its sophisticated encapsulation of\nprofound emotions and imagery within minimal text. While existing text-to-3D\nmethods struggle with nuanced interpretations, we present a literary-guided\napproach that synergizes traditional poetry analysis with advanced generative\ntechnologies. Our framework centers on two key innovations: (1) Hierarchical\nLiterary-Criticism Theory Grounded Parsing (H-LCTGP), which captures both\nexplicit imagery and implicit emotional resonance through structured semantic\ndecomposition, and (2) Progressive Dimensional Synthesis (PDS), a multi-stage\npipeline that systematically transforms poetic elements into coherent 3D scenes\nthrough sequential diffusion processes, geometric optimization, and real-time\nenhancement. Extensive experiments demonstrate that HaikuVerse significantly\noutperforms conventional text-to-3D approaches in both literary fidelity and\nvisual quality, establishing a new paradigm for preserving cultural heritage in\nimmersive digital spaces. Project website at:\nhttps://syllables-to-scenes.github.io/\n", "date": "2025-02-17", "categories": []} {"id": "2502.11642", "title": "GaussianMotion: End-to-End Learning of Animatable Gaussian Avatars with\n Pose Guidance from Text", "abstract": " In this paper, we introduce GaussianMotion, a novel human rendering model\nthat generates fully animatable scenes aligned with textual descriptions using\nGaussian Splatting. Although existing methods achieve reasonable text-to-3D\ngeneration of human bodies using various 3D representations, they often face\nlimitations in fidelity and efficiency, or primarily focus on static models\nwith limited pose control. In contrast, our method generates fully animatable\n3D avatars by combining deformable 3D Gaussian Splatting with text-to-3D score\ndistillation, achieving high fidelity and efficient rendering for arbitrary\nposes. By densely generating diverse random poses during optimization, our\ndeformable 3D human model learns to capture a wide range of natural motions\ndistilled from a pose-conditioned diffusion model in an end-to-end manner.\nFurthermore, we propose Adaptive Score Distillation that effectively balances\nrealistic detail and smoothness to achieve optimal 3D results. Experimental\nresults demonstrate that our approach outperforms existing baselines by\nproducing high-quality textures in both static and animated results, and by\ngenerating diverse 3D human models from various textual inputs.\n", "date": "2025-02-17", "categories": []} {"id": "2502.11697", "title": "MVTokenFlow: High-quality 4D Content Generation using Multiview Token\n Flow", "abstract": " In this paper, we present MVTokenFlow for high-quality 4D content creation\nfrom monocular videos. Recent advancements in generative models such as video\ndiffusion models and multiview diffusion models enable us to create videos or\n3D models. However, extending these generative models for dynamic 4D content\ncreation is still a challenging task that requires the generated content to be\nconsistent spatially and temporally. To address this challenge, MVTokenFlow\nutilizes the multiview diffusion model to generate multiview images on\ndifferent timesteps, which attains spatial consistency across different\nviewpoints and allows us to reconstruct a reasonable coarse 4D field. Then,\nMVTokenFlow further regenerates all the multiview images using the rendered 2D\nflows as guidance. The 2D flows effectively associate pixels from different\ntimesteps and improve the temporal consistency by reusing tokens in the\nregeneration process. Finally, the regenerated images are spatiotemporally\nconsistent and utilized to refine the coarse 4D field to get a high-quality 4D\nfield. Experiments demonstrate the effectiveness of our design and show\nsignificantly improved quality than baseline methods.\n", "date": "2025-02-17", "categories": []} {"id": "2406.18581", "title": "Dream-in-Style: Text-to-3D Generation Using Stylized Score Distillation", "abstract": " We present a method to generate 3D objects in styles. Our method takes a text\nprompt and a style reference image as input and reconstructs a neural radiance\nfield to synthesize a 3D model with the content aligning with the text prompt\nand the style following the reference image. To simultaneously generate the 3D\nobject and perform style transfer in one go, we propose a stylized score\ndistillation loss to guide a text-to-3D optimization process to output visually\nplausible geometry and appearance. Our stylized score distillation is based on\na combination of an original pretrained text-to-image model and its modified\nsibling with the key and value features of self-attention layers manipulated to\ninject styles from the reference image. Comparisons with state-of-the-art\nmethods demonstrated the strong visual performance of our method, further\nsupported by the quantitative results from our user study.\n", "date": "2024-06-05", "categories": []} {"id": "2407.11394", "title": "DreamCatalyst: Fast and High-Quality 3D Editing via Controlling\n Editability and Identity Preservation", "abstract": " Score distillation sampling (SDS) has emerged as an effective framework in\ntext-driven 3D editing tasks, leveraging diffusion models for 3D-consistent\nediting. However, existing SDS-based 3D editing methods suffer from long\ntraining times and produce low-quality results. We identify that the root cause\nof this performance degradation is \\textit{their conflict with the sampling\ndynamics of diffusion models}. Addressing this conflict allows us to treat SDS\nas a diffusion reverse process for 3D editing via sampling from data space. In\ncontrast, existing methods naively distill the score function using diffusion\nmodels. From these insights, we propose DreamCatalyst, a novel framework that\nconsiders these sampling dynamics in the SDS framework. Specifically, we devise\nthe optimization process of our DreamCatalyst to approximate the diffusion\nreverse process in editing tasks, thereby aligning with diffusion sampling\ndynamics. As a result, DreamCatalyst successfully reduces training time and\nimproves editing quality. Our method offers two modes: (1) a fast mode that\nedits Neural Radiance Fields (NeRF) scenes approximately 23 times faster than\ncurrent state-of-the-art NeRF editing methods, and (2) a high-quality mode that\nproduces superior results about 8 times faster than these methods. Notably, our\nhigh-quality mode outperforms current state-of-the-art NeRF editing methods in\nterms of both speed and quality. DreamCatalyst also surpasses the\nstate-of-the-art 3D Gaussian Splatting (3DGS) editing methods, establishing\nitself as an effective and model-agnostic 3D editing solution. See more\nextensive results on our project page: https://dream-catalyst.github.io.\n", "date": "2024-07-16", "categories": []} {"id": "2408.14899", "title": "MeshUp: Multi-Target Mesh Deformation via Blended Score Distillation", "abstract": " We propose MeshUp, a technique that deforms a 3D mesh towards multiple target\nconcepts, and intuitively controls the region where each concept is expressed.\nConveniently, the concepts can be defined as either text queries, e.g., \"a dog\"\nand \"a turtle,\" or inspirational images, and the local regions can be selected\nas any number of vertices on the mesh. We can effectively control the influence\nof the concepts and mix them together using a novel score distillation\napproach, referred to as the Blended Score Distillation (BSD). BSD operates on\neach attention layer of the denoising U-Net of a diffusion model as it extracts\nand injects the per-objective activations into a unified denoising pipeline\nfrom which the deformation gradients are calculated. To localize the expression\nof these activations, we create a probabilistic Region of Interest (ROI) map on\nthe surface of the mesh, and turn it into 3D-consistent masks that we use to\ncontrol the expression of these activations. We demonstrate the effectiveness\nof BSD empirically and show that it can deform various meshes towards multiple\nobjectives. Our project page is at https://threedle.github.io/MeshUp.\n", "date": "2024-08-27", "categories": []} {"id": "2409.20562", "title": "SpaceMesh: A Continuous Representation for Learning Manifold Surface\n Meshes", "abstract": " Meshes are ubiquitous in visual computing and simulation, yet most existing\nmachine learning techniques represent meshes only indirectly, e.g. as the level\nset of a scalar field or deformation of a template, or as a disordered triangle\nsoup lacking local structure. This work presents a scheme to directly generate\nmanifold, polygonal meshes of complex connectivity as the output of a neural\nnetwork. Our key innovation is to define a continuous latent connectivity space\nat each mesh vertex, which implies the discrete mesh. In particular, our vertex\nembeddings generate cyclic neighbor relationships in a halfedge mesh\nrepresentation, which gives a guarantee of edge-manifoldness and the ability to\nrepresent general polygonal meshes. This representation is well-suited to\nmachine learning and stochastic optimization, without restriction on\nconnectivity or topology. We first explore the basic properties of this\nrepresentation, then use it to fit distributions of meshes from large datasets.\nThe resulting models generate diverse meshes with tessellation structure\nlearned from the dataset population, with concise details and high-quality mesh\nelements. In applications, this approach not only yields high-quality outputs\nfrom generative models, but also enables directly learning challenging geometry\nprocessing tasks such as mesh repair.\n", "date": "2024-09-30", "categories": []} {"id": "2502.07785", "title": "Pippo: High-Resolution Multi-View Humans from a Single Image", "abstract": " We present Pippo, a generative model capable of producing 1K resolution dense\nturnaround videos of a person from a single casually clicked photo. Pippo is a\nmulti-view diffusion transformer and does not require any additional inputs -\ne.g., a fitted parametric model or camera parameters of the input image. We\npre-train Pippo on 3B human images without captions, and conduct multi-view\nmid-training and post-training on studio captured humans. During mid-training,\nto quickly absorb the studio dataset, we denoise several (up to 48) views at\nlow-resolution, and encode target cameras coarsely using a shallow MLP. During\npost-training, we denoise fewer views at high-resolution and use pixel-aligned\ncontrols (e.g., Spatial anchor and Plucker rays) to enable 3D consistent\ngenerations. At inference, we propose an attention biasing technique that\nallows Pippo to simultaneously generate greater than 5 times as many views as\nseen during training. Finally, we also introduce an improved metric to evaluate\n3D consistency of multi-view generations, and show that Pippo outperforms\nexisting works on multi-view human generation from a single image.\n", "date": "2025-02-11", "categories": []} {"id": "2403.09981", "title": "Controllable Text-to-3D Generation via Surface-Aligned Gaussian\n Splatting", "abstract": " While text-to-3D and image-to-3D generation tasks have received considerable\nattention, one important but under-explored field between them is controllable\ntext-to-3D generation, which we mainly focus on in this work. To address this\ntask, 1) we introduce Multi-view ControlNet (MVControl), a novel neural network\narchitecture designed to enhance existing pre-trained multi-view diffusion\nmodels by integrating additional input conditions, such as edge, depth, normal,\nand scribble maps. Our innovation lies in the introduction of a conditioning\nmodule that controls the base diffusion model using both local and global\nembeddings, which are computed from the input condition images and camera\nposes. Once trained, MVControl is able to offer 3D diffusion guidance for\noptimization-based 3D generation. And, 2) we propose an efficient multi-stage\n3D generation pipeline that leverages the benefits of recent large\nreconstruction models and score distillation algorithm. Building upon our\nMVControl architecture, we employ a unique hybrid diffusion guidance method to\ndirect the optimization process. In pursuit of efficiency, we adopt 3D\nGaussians as our representation instead of the commonly used implicit\nrepresentations. We also pioneer the use of SuGaR, a hybrid representation that\nbinds Gaussians to mesh triangle faces. This approach alleviates the issue of\npoor geometry in 3D Gaussians and enables the direct sculpting of fine-grained\ngeometry on the mesh. Extensive experiments demonstrate that our method\nachieves robust generalization and enables the controllable generation of\nhigh-quality 3D content. Project page: https://lizhiqi49.github.io/MVControl/.\n", "date": "2024-03-15", "categories": []} {"id": "2502.06392", "title": "TANGLED: Generating 3D Hair Strands from Images with Arbitrary Styles\n and Viewpoints", "abstract": " Hairstyles are intricate and culturally significant with various geometries,\ntextures, and structures. Existing text or image-guided generation methods fail\nto handle the richness and complexity of diverse styles. We present TANGLED, a\nnovel approach for 3D hair strand generation that accommodates diverse image\ninputs across styles, viewpoints, and quantities of input views. TANGLED\nemploys a three-step pipeline. First, our MultiHair Dataset provides 457\ndiverse hairstyles annotated with 74 attributes, emphasizing complex and\nculturally significant styles to improve model generalization. Second, we\npropose a diffusion framework conditioned on multi-view linearts that can\ncapture topological cues (e.g., strand density and parting lines) while\nfiltering out noise. By leveraging a latent diffusion model with\ncross-attention on lineart features, our method achieves flexible and robust 3D\nhair generation across diverse input conditions. Third, a parametric\npost-processing module enforces braid-specific constraints to maintain\ncoherence in complex structures. This framework not only advances hairstyle\nrealism and diversity but also enables culturally inclusive digital avatars and\nnovel applications like sketch-based 3D strand editing for animation and\naugmented reality.\n", "date": "2025-02-10", "categories": []} {"id": "2502.04370", "title": "DreamDPO: Aligning Text-to-3D Generation with Human Preferences via\n Direct Preference Optimization", "abstract": " Text-to-3D generation automates 3D content creation from textual\ndescriptions, which offers transformative potential across various fields.\nHowever, existing methods often struggle to align generated content with human\npreferences, limiting their applicability and flexibility. To address these\nlimitations, in this paper, we propose DreamDPO, an optimization-based\nframework that integrates human preferences into the 3D generation process,\nthrough direct preference optimization. Practically, DreamDPO first constructs\npairwise examples, then compare their alignment with human preferences using\nreward or large multimodal models, and lastly optimizes the 3D representation\nwith a preference-driven loss function. By leveraging pairwise comparison to\nreflect preferences, DreamDPO reduces reliance on precise pointwise quality\nevaluations while enabling fine-grained controllability through\npreference-guided optimization. Experiments demonstrate that DreamDPO achieves\ncompetitive results, and provides higher-quality and more controllable 3D\ncontent compared to existing methods. The code and models will be open-sourced.\n", "date": "2025-02-05", "categories": []} {"id": "2502.04762", "title": "Autoregressive Generation of Static and Growing Trees", "abstract": " We propose a transformer architecture and training strategy for tree\ngeneration. The architecture processes data at multiple resolutions and has an\nhourglass shape, with middle layers processing fewer tokens than outer layers.\nSimilar to convolutional networks, we introduce longer range skip connections\nto completent this multi-resolution approach. The key advantage of this\narchitecture is the faster processing speed and lower memory consumption. We\nare therefore able to process more complex trees than would be possible with a\nvanilla transformer architecture. Furthermore, we extend this approach to\nperform image-to-tree and point-cloud-to-tree conditional generation and to\nsimulate the tree growth processes, generating 4D trees. Empirical results\nvalidate our approach in terms of speed, memory consumption, and generation\nquality.\n", "date": "2025-02-07", "categories": []} {"id": "2309.16110", "title": "Learning Effective NeRFs and SDFs Representations with 3D Generative\n Adversarial Networks for 3D Object Generation", "abstract": " We present a solution for 3D object generation of ICCV 2023 OmniObject3D\nChallenge. In recent years, 3D object generation has made great process and\nachieved promising results, but it remains a challenging task due to the\ndifficulty of generating complex, textured, and high-fidelity results. To\nresolve this problem, we study learning effective NeRFs and SDFs\nrepresentations with 3D Generative Adversarial Networks (GANs) for 3D object\ngeneration. Specifically, inspired by recent works, we use the efficient\ngeometry-aware 3D GANs as the backbone incorporating with label embedding and\ncolor mapping, which enables to train the model on different taxonomies\nsimultaneously. Then, through a decoder, we aggregate the resulting features to\ngenerate Neural Radiance Fields (NeRFs) based representations for rendering\nhigh-fidelity synthetic images. Meanwhile, we optimize Signed Distance\nFunctions (SDFs) to effectively represent objects with 3D meshes. Besides, we\nobserve that this model can be effectively trained with only a few images of\neach object from a variety of classes, instead of using a great number of\nimages per object or training one model per class. With this pipeline, we can\noptimize an effective model for 3D object generation. This solution is among\nthe top 3 in the ICCV 2023 OmniObject3D Challenge.\n", "date": "2023-09-28", "categories": []} {"id": "2408.13724", "title": "PhysPart: Physically Plausible Part Completion for Interactable Objects", "abstract": " Interactable objects are ubiquitous in our daily lives. Recent advances in 3D\ngenerative models make it possible to automate the modeling of these objects,\nbenefiting a range of applications from 3D printing to the creation of robot\nsimulation environments. However, while significant progress has been made in\nmodeling 3D shapes and appearances, modeling object physics, particularly for\ninteractable objects, remains challenging due to the physical constraints\nimposed by inter-part motions. In this paper, we tackle the problem of\nphysically plausible part completion for interactable objects, aiming to\ngenerate 3D parts that not only fit precisely into the object but also allow\nsmooth part motions. To this end, we propose a diffusion-based part generation\nmodel that utilizes geometric conditioning through classifier-free guidance and\nformulates physical constraints as a set of stability and mobility losses to\nguide the sampling process. Additionally, we demonstrate the generation of\ndependent parts, paving the way toward sequential part generation for objects\nwith complex part-whole hierarchies. Experimentally, we introduce a new metric\nfor measuring physical plausibility based on motion success rates. Our model\noutperforms existing baselines over shape and physical metrics, especially\nthose that do not adequately model physical constraints. We also demonstrate\nour applications in 3D printing, robot manipulation, and sequential part\ngeneration, showing our strength in realistic tasks with the demand for high\nphysical plausibility.\n", "date": "2024-08-25", "categories": []} {"id": "2502.00708", "title": "PhiP-G: Physics-Guided Text-to-3D Compositional Scene Generation", "abstract": " Text-to-3D asset generation has achieved significant optimization under the\nsupervision of 2D diffusion priors. However, when dealing with compositional\nscenes, existing methods encounter several challenges: 1). failure to ensure\nthat composite scene layouts comply with physical laws; 2). difficulty in\naccurately capturing the assets and relationships described in complex scene\ndescriptions; 3). limited autonomous asset generation capabilities among layout\napproaches leveraging large language models (LLMs). To avoid these compromises,\nwe propose a novel framework for compositional scene generation, PhiP-G, which\nseamlessly integrates generation techniques with layout guidance based on a\nworld model. Leveraging LLM-based agents, PhiP-G analyzes the complex scene\ndescription to generate a scene graph, and integrating a multimodal 2D\ngeneration agent and a 3D Gaussian generation method for targeted assets\ncreation. For the stage of layout, PhiP-G employs a physical pool with adhesion\ncapabilities and a visual supervision agent, forming a world model for layout\nprediction and planning. Extensive experiments demonstrate that PhiP-G\nsignificantly enhances the generation quality and physical rationality of the\ncompositional scenes. Notably, PhiP-G attains state-of-the-art (SOTA)\nperformance in CLIP scores, achieves parity with the leading methods in\ngeneration quality as measured by the T$^3$Bench, and improves efficiency by\n24x.\n", "date": "2025-02-02", "categories": []} {"id": "2312.14140", "title": "HeadCraft: Modeling High-Detail Shape Variations for Animated 3DMMs", "abstract": " Current advances in human head modeling allow the generation of\nplausible-looking 3D head models via neural representations, such as NeRFs and\nSDFs. Nevertheless, constructing complete high-fidelity head models with\nexplicitly controlled animation remains an issue. Furthermore, completing the\nhead geometry based on a partial observation, e.g., coming from a depth sensor,\nwhile preserving a high level of detail is often problematic for the existing\nmethods. We introduce a generative model for detailed 3D head meshes on top of\nan articulated 3DMM, simultaneously allowing explicit animation and high-detail\npreservation. Our method is trained in two stages. First, we register a\nparametric head model with vertex displacements to each mesh of the recently\nintroduced NPHM dataset of accurate 3D head scans. The estimated displacements\nare baked into a hand-crafted UV layout. Second, we train a StyleGAN model to\ngeneralize over the UV maps of displacements, which we later refer to as\nHeadCraft. The decomposition of the parametric model and high-quality vertex\ndisplacements allows us to animate the model and modify the regions\nsemantically. We demonstrate the results of unconditional sampling, fitting to\na scan and editing. The project page is available at\nhttps://seva100.github.io/headcraft.\n", "date": "2023-12-21", "categories": []} {"id": "2501.18982", "title": "OmniPhysGS: 3D Constitutive Gaussians for General Physics-Based Dynamics\n Generation", "abstract": " Recently, significant advancements have been made in the reconstruction and\ngeneration of 3D assets, including static cases and those with physical\ninteractions. To recover the physical properties of 3D assets, existing methods\ntypically assume that all materials belong to a specific predefined category\n(e.g., elasticity). However, such assumptions ignore the complex composition of\nmultiple heterogeneous objects in real scenarios and tend to render less\nphysically plausible animation given a wider range of objects. We propose\nOmniPhysGS for synthesizing a physics-based 3D dynamic scene composed of more\ngeneral objects. A key design of OmniPhysGS is treating each 3D asset as a\ncollection of constitutive 3D Gaussians. For each Gaussian, its physical\nmaterial is represented by an ensemble of 12 physical domain-expert sub-models\n(rubber, metal, honey, water, etc.), which greatly enhances the flexibility of\nthe proposed model. In the implementation, we define a scene by user-specified\nprompts and supervise the estimation of material weighting factors via a\npretrained video diffusion model. Comprehensive experiments demonstrate that\nOmniPhysGS achieves more general and realistic physical dynamics across a\nbroader spectrum of materials, including elastic, viscoelastic, plastic, and\nfluid substances, as well as interactions between different materials. Our\nmethod surpasses existing methods by approximately 3% to 16% in metrics of\nvisual quality and text alignment.\n", "date": "2025-01-31", "categories": []} {"id": "2501.16764", "title": "DiffSplat: Repurposing Image Diffusion Models for Scalable Gaussian\n Splat Generation", "abstract": " Recent advancements in 3D content generation from text or a single image\nstruggle with limited high-quality 3D datasets and inconsistency from 2D\nmulti-view generation. We introduce DiffSplat, a novel 3D generative framework\nthat natively generates 3D Gaussian splats by taming large-scale text-to-image\ndiffusion models. It differs from previous 3D generative models by effectively\nutilizing web-scale 2D priors while maintaining 3D consistency in a unified\nmodel. To bootstrap the training, a lightweight reconstruction model is\nproposed to instantly produce multi-view Gaussian splat grids for scalable\ndataset curation. In conjunction with the regular diffusion loss on these\ngrids, a 3D rendering loss is introduced to facilitate 3D coherence across\narbitrary views. The compatibility with image diffusion models enables seamless\nadaptions of numerous techniques for image generation to the 3D realm.\nExtensive experiments reveal the superiority of DiffSplat in text- and\nimage-conditioned generation tasks and downstream applications. Thorough\nablation studies validate the efficacy of each critical design choice and\nprovide insights into the underlying mechanism.\n", "date": "2025-01-28", "categories": []} {"id": "2501.17162", "title": "CubeDiff: Repurposing Diffusion-Based Image Models for Panorama\n Generation", "abstract": " We introduce a novel method for generating 360{\\deg} panoramas from text\nprompts or images. Our approach leverages recent advances in 3D generation by\nemploying multi-view diffusion models to jointly synthesize the six faces of a\ncubemap. Unlike previous methods that rely on processing equirectangular\nprojections or autoregressive generation, our method treats each face as a\nstandard perspective image, simplifying the generation process and enabling the\nuse of existing multi-view diffusion models. We demonstrate that these models\ncan be adapted to produce high-quality cubemaps without requiring\ncorrespondence-aware attention layers. Our model allows for fine-grained text\ncontrol, generates high resolution panorama images and generalizes well beyond\nits training set, whilst achieving state-of-the-art results, both qualitatively\nand quantitatively. Project page: https://cubediff.github.io/\n", "date": "2025-01-28", "categories": []} {"id": "2412.07766", "title": "Make-A-Texture: Fast Shape-Aware Texture Generation in 3 Seconds", "abstract": " We present Make-A-Texture, a new framework that efficiently synthesizes\nhigh-resolution texture maps from textual prompts for given 3D geometries. Our\napproach progressively generates textures that are consistent across multiple\nviewpoints with a depth-aware inpainting diffusion model, in an optimized\nsequence of viewpoints determined by an automatic view selection algorithm.\n A significant feature of our method is its remarkable efficiency, achieving a\nfull texture generation within an end-to-end runtime of just 3.07 seconds on a\nsingle NVIDIA H100 GPU, significantly outperforming existing methods. Such an\nacceleration is achieved by optimizations in the diffusion model and a\nspecialized backprojection method. Moreover, our method reduces the artifacts\nin the backprojection phase, by selectively masking out non-frontal faces, and\ninternal faces of open-surfaced objects.\n Experimental results demonstrate that Make-A-Texture matches or exceeds the\nquality of other state-of-the-art methods. Our work significantly improves the\napplicability and practicality of texture generation models for real-world 3D\ncontent creation, including interactive creation and text-guided texture\nediting.\n", "date": "2024-12-10", "categories": []} {"id": "2501.16177", "title": "BAG: Body-Aligned 3D Wearable Asset Generation", "abstract": " While recent advancements have shown remarkable progress in general 3D shape\ngeneration models, the challenge of leveraging these approaches to\nautomatically generate wearable 3D assets remains unexplored. To this end, we\npresent BAG, a Body-aligned Asset Generation method to output 3D wearable asset\nthat can be automatically dressed on given 3D human bodies. This is achived by\ncontrolling the 3D generation process using human body shape and pose\ninformation. Specifically, we first build a general single-image to consistent\nmultiview image diffusion model, and train it on the large Objaverse dataset to\nachieve diversity and generalizability. Then we train a Controlnet to guide the\nmultiview generator to produce body-aligned multiview images. The control\nsignal utilizes the multiview 2D projections of the target human body, where\npixel values represent the XYZ coordinates of the body surface in a canonical\nspace. The body-conditioned multiview diffusion generates body-aligned\nmultiview images, which are then fed into a native 3D diffusion model to\nproduce the 3D shape of the asset. Finally, by recovering the similarity\ntransformation using multiview silhouette supervision and addressing asset-body\npenetration with physics simulators, the 3D asset can be accurately fitted onto\nthe target human body. Experimental results demonstrate significant advantages\nover existing methods in terms of image prompt-following capability, shape\ndiversity, and shape quality. Our project page is available at\nhttps://bag-3d.github.io/.\n", "date": "2025-01-27", "categories": []} {"id": "2501.13969", "title": "InsTex: Indoor Scenes Stylized Texture Synthesis", "abstract": " Generating high-quality textures for 3D scenes is crucial for applications in\ninterior design, gaming, and augmented/virtual reality (AR/VR). Although recent\nadvancements in 3D generative models have enhanced content creation,\nsignificant challenges remain in achieving broad generalization and maintaining\nstyle consistency across multiple viewpoints. Current methods, such as 2D\ndiffusion models adapted for 3D texturing, suffer from lengthy processing times\nand visual artifacts, while approaches driven by 3D data often fail to\ngeneralize effectively. To overcome these challenges, we introduce InsTex, a\ntwo-stage architecture designed to generate high-quality, style-consistent\ntextures for 3D indoor scenes. InsTex utilizes depth-to-image diffusion priors\nin a coarse-to-fine pipeline, first generating multi-view images with a\npre-trained 2D diffusion model and subsequently refining the textures for\nconsistency. Our method supports both textual and visual prompts, achieving\nstate-of-the-art results in visual quality and quantitative metrics, and\ndemonstrates its effectiveness across various 3D texturing applications.\n", "date": "2025-01-22", "categories": []} {"id": "2411.02293", "title": "Hunyuan3D 1.0: A Unified Framework for Text-to-3D and Image-to-3D\n Generation", "abstract": " While 3D generative models have greatly improved artists' workflows, the\nexisting diffusion models for 3D generation suffer from slow generation and\npoor generalization. To address this issue, we propose a two-stage approach\nnamed Hunyuan3D 1.0 including a lite version and a standard version, that both\nsupport text- and image-conditioned generation. In the first stage, we employ a\nmulti-view diffusion model that efficiently generates multi-view RGB in\napproximately 4 seconds. These multi-view images capture rich details of the 3D\nasset from different viewpoints, relaxing the tasks from single-view to\nmulti-view reconstruction. In the second stage, we introduce a feed-forward\nreconstruction model that rapidly and faithfully reconstructs the 3D asset\ngiven the generated multi-view images in approximately 7 seconds. The\nreconstruction network learns to handle noises and in-consistency introduced by\nthe multi-view diffusion and leverages the available information from the\ncondition image to efficiently recover the 3D structure. Our framework involves\nthe text-to-image model, i.e., Hunyuan-DiT, making it a unified framework to\nsupport both text- and image-conditioned 3D generation. Our standard version\nhas 3x more parameters than our lite and other existing model. Our Hunyuan3D\n1.0 achieves an impressive balance between speed and quality, significantly\nreducing generation time while maintaining the quality and diversity of the\nproduced assets.\n", "date": "2024-11-04", "categories": []} {"id": "2501.13449", "title": "MultiDreamer3D: Multi-concept 3D Customization with Concept-Aware\n Diffusion Guidance", "abstract": " While single-concept customization has been studied in 3D, multi-concept\ncustomization remains largely unexplored. To address this, we propose\nMultiDreamer3D that can generate coherent multi-concept 3D content in a\ndivide-and-conquer manner. First, we generate 3D bounding boxes using an\nLLM-based layout controller. Next, a selective point cloud generator creates\ncoarse point clouds for each concept. These point clouds are placed in the 3D\nbounding boxes and initialized into 3D Gaussian Splatting with concept labels,\nenabling precise identification of concept attributions in 2D projections.\nFinally, we refine 3D Gaussians via concept-aware interval score matching,\nguided by concept-aware diffusion. Our experimental results show that\nMultiDreamer3D not only ensures object presence and preserves the distinct\nidentities of each concept but also successfully handles complex cases such as\nproperty change or interaction. To the best of our knowledge, we are the first\nto address the multi-concept customization in 3D.\n", "date": "2025-01-23", "categories": []} {"id": "2408.05938", "title": "Deep Geometric Moments Promote Shape Consistency in Text-to-3D\n Generation", "abstract": " To address the data scarcity associated with 3D assets, 2D-lifting techniques\nsuch as Score Distillation Sampling (SDS) have become a widely adopted practice\nin text-to-3D generation pipelines. However, the diffusion models used in these\ntechniques are prone to viewpoint bias and thus lead to geometric\ninconsistencies such as the Janus problem. To counter this, we introduce MT3D,\na text-to-3D generative model that leverages a high-fidelity 3D object to\novercome viewpoint bias and explicitly infuse geometric understanding into the\ngeneration pipeline. Firstly, we employ depth maps derived from a high-quality\n3D model as control signals to guarantee that the generated 2D images preserve\nthe fundamental shape and structure, thereby reducing the inherent viewpoint\nbias. Next, we utilize deep geometric moments to ensure geometric consistency\nin the 3D representation explicitly. By incorporating geometric details from a\n3D asset, MT3D enables the creation of diverse and geometrically consistent\nobjects, thereby improving the quality and usability of our 3D representations.\nProject page and code: https://moment-3d.github.io/\n", "date": "2024-08-12", "categories": []} {"id": "2501.12393", "title": "Towards Affordance-Aware Articulation Synthesis for Rigged Objects", "abstract": " Rigged objects are commonly used in artist pipelines, as they can flexibly\nadapt to different scenes and postures. However, articulating the rigs into\nrealistic affordance-aware postures (e.g., following the context, respecting\nthe physics and the personalities of the object) remains time-consuming and\nheavily relies on human labor from experienced artists. In this paper, we\ntackle the novel problem and design A3Syn. With a given context, such as the\nenvironment mesh and a text prompt of the desired posture, A3Syn synthesizes\narticulation parameters for arbitrary and open-domain rigged objects obtained\nfrom the Internet. The task is incredibly challenging due to the lack of\ntraining data, and we do not make any topological assumptions about the\nopen-domain rigs. We propose using 2D inpainting diffusion model and several\ncontrol techniques to synthesize in-context affordance information. Then, we\ndevelop an efficient bone correspondence alignment using a combination of\ndifferentiable rendering and semantic correspondence. A3Syn has stable\nconvergence, completes in minutes, and synthesizes plausible affordance on\ndifferent combinations of in-the-wild object rigs and scenes.\n", "date": "2025-01-21", "categories": []} {"id": "2501.09433", "title": "CaPa: Carve-n-Paint Synthesis for Efficient 4K Textured Mesh Generation", "abstract": " The synthesis of high-quality 3D assets from textual or visual inputs has\nbecome a central objective in modern generative modeling. Despite the\nproliferation of 3D generation algorithms, they frequently grapple with\nchallenges such as multi-view inconsistency, slow generation times, low\nfidelity, and surface reconstruction problems. While some studies have\naddressed some of these issues, a comprehensive solution remains elusive. In\nthis paper, we introduce \\textbf{CaPa}, a carve-and-paint framework that\ngenerates high-fidelity 3D assets efficiently. CaPa employs a two-stage\nprocess, decoupling geometry generation from texture synthesis. Initially, a 3D\nlatent diffusion model generates geometry guided by multi-view inputs, ensuring\nstructural consistency across perspectives. Subsequently, leveraging a novel,\nmodel-agnostic Spatially Decoupled Attention, the framework synthesizes\nhigh-resolution textures (up to 4K) for a given geometry. Furthermore, we\npropose a 3D-aware occlusion inpainting algorithm that fills untextured\nregions, resulting in cohesive results across the entire model. This pipeline\ngenerates high-quality 3D assets in less than 30 seconds, providing\nready-to-use outputs for commercial applications. Experimental results\ndemonstrate that CaPa excels in both texture fidelity and geometric stability,\nestablishing a new standard for practical, scalable 3D asset generation.\n", "date": "2025-01-16", "categories": []} {"id": "2501.07870", "title": "Make-A-Character 2: Animatable 3D Character Generation From a Single\n Image", "abstract": " This report introduces Make-A-Character 2, an advanced system for generating\nhigh-quality 3D characters from single portrait photographs, ideal for game\ndevelopment and digital human applications. Make-A-Character 2 builds upon its\npredecessor by incorporating several significant improvements for image-based\nhead generation. We utilize the IC-Light method to correct non-ideal\nillumination in input photos and apply neural network-based color correction to\nharmonize skin tones between the photos and game engine renders. We also employ\nthe Hierarchical Representation Network to capture high-frequency facial\nstructures and conduct adaptive skeleton calibration for accurate and\nexpressive facial animations. The entire image-to-3D-character generation\nprocess takes less than 2 minutes. Furthermore, we leverage transformer\narchitecture to generate co-speech facial and gesture actions, enabling\nreal-time conversation with the generated character. These technologies have\nbeen integrated into our conversational AI avatar products.\n", "date": "2025-01-14", "categories": []} {"id": "2408.06019", "title": "HeadGAP: Few-Shot 3D Head Avatar via Generalizable Gaussian Priors", "abstract": " In this paper, we present a novel 3D head avatar creation approach capable of\ngeneralizing from few-shot in-the-wild data with high-fidelity and animatable\nrobustness. Given the underconstrained nature of this problem, incorporating\nprior knowledge is essential. Therefore, we propose a framework comprising\nprior learning and avatar creation phases. The prior learning phase leverages\n3D head priors derived from a large-scale multi-view dynamic dataset, and the\navatar creation phase applies these priors for few-shot personalization. Our\napproach effectively captures these priors by utilizing a Gaussian\nSplatting-based auto-decoder network with part-based dynamic modeling. Our\nmethod employs identity-shared encoding with personalized latent codes for\nindividual identities to learn the attributes of Gaussian primitives. During\nthe avatar creation phase, we achieve fast head avatar personalization by\nleveraging inversion and fine-tuning strategies. Extensive experiments\ndemonstrate that our model effectively exploits head priors and successfully\ngeneralizes them to few-shot personalization, achieving photo-realistic\nrendering quality, multi-view consistency, and stable animation.\n", "date": "2024-08-12", "categories": []} {"id": "2501.05379", "title": "Arc2Avatar: Generating Expressive 3D Avatars from a Single Image via ID\n Guidance", "abstract": " Inspired by the effectiveness of 3D Gaussian Splatting (3DGS) in\nreconstructing detailed 3D scenes within multi-view setups and the emergence of\nlarge 2D human foundation models, we introduce Arc2Avatar, the first SDS-based\nmethod utilizing a human face foundation model as guidance with just a single\nimage as input. To achieve that, we extend such a model for diverse-view human\nhead generation by fine-tuning on synthetic data and modifying its\nconditioning. Our avatars maintain a dense correspondence with a human face\nmesh template, allowing blendshape-based expression generation. This is\nachieved through a modified 3DGS approach, connectivity regularizers, and a\nstrategic initialization tailored for our task. Additionally, we propose an\noptional efficient SDS-based correction step to refine the blendshape\nexpressions, enhancing realism and diversity. Experiments demonstrate that\nArc2Avatar achieves state-of-the-art realism and identity preservation,\neffectively addressing color issues by allowing the use of very low guidance,\nenabled by our strong identity prior and initialization strategy, without\ncompromising detail. Please visit https://arc2avatar.github.io for more\nresources.\n", "date": "2025-01-09", "categories": []} {"id": "2403.09236", "title": "Hyper-3DG: Text-to-3D Gaussian Generation via Hypergraph", "abstract": " Text-to-3D generation represents an exciting field that has seen rapid\nadvancements, facilitating the transformation of textual descriptions into\ndetailed 3D models. However, current progress often neglects the intricate\nhigh-order correlation of geometry and texture within 3D objects, leading to\nchallenges such as over-smoothness, over-saturation and the Janus problem. In\nthis work, we propose a method named ``3D Gaussian Generation via Hypergraph\n(Hyper-3DG)'', designed to capture the sophisticated high-order correlations\npresent within 3D objects. Our framework is anchored by a well-established\nmainflow and an essential module, named ``Geometry and Texture Hypergraph\nRefiner (HGRefiner)''. This module not only refines the representation of 3D\nGaussians but also accelerates the update process of these 3D Gaussians by\nconducting the Patch-3DGS Hypergraph Learning on both explicit attributes and\nlatent visual features. Our framework allows for the production of finely\ngenerated 3D objects within a cohesive optimization, effectively circumventing\ndegradation. Extensive experimentation has shown that our proposed method\nsignificantly enhances the quality of 3D generation while incurring no\nadditional computational overhead for the underlying framework. (Project code:\nhttps://github.com/yjhboy/Hyper3DG)\n", "date": "2024-03-14", "categories": []} {"id": "2404.06429", "title": "Magic-Boost: Boost 3D Generation with Multi-View Conditioned Diffusion", "abstract": " Benefiting from the rapid development of 2D diffusion models, 3D content\ngeneration has witnessed significant progress. One promising solution is to\nfinetune the pre-trained 2D diffusion models to produce multi-view images and\nthen reconstruct them into 3D assets via feed-forward sparse-view\nreconstruction models. However, limited by the 3D inconsistency in the\ngenerated multi-view images and the low reconstruction resolution of the\nfeed-forward reconstruction models, the generated 3d assets are still limited\nto incorrect geometries and blurry textures. To address this problem, we\npresent a multi-view based refine method, named Magic-Boost, to further refine\nthe generation results. In detail, we first propose a novel multi-view\nconditioned diffusion model which extracts 3d prior from the synthesized\nmulti-view images to synthesize high-fidelity novel view images and then\nintroduce a novel iterative-update strategy to adopt it to provide precise\nguidance to refine the coarse generated results through a fast optimization\nprocess. Conditioned on the strong 3d priors extracted from the synthesized\nmulti-view images, Magic-Boost is capable of providing precise optimization\nguidance that well aligns with the coarse generated 3D assets, enriching the\nlocal detail in both geometry and texture within a short time ($\\sim15$min).\nExtensive experiments show Magic-Boost greatly enhances the coarse generated\ninputs, generates high-quality 3D assets with rich geometric and textural\ndetails. (Project Page: https://magic-research.github.io/magic-boost/)\n", "date": "2024-04-09", "categories": []} {"id": "2501.05427", "title": "Zero-1-to-G: Taming Pretrained 2D Diffusion Model for Direct 3D\n Generation", "abstract": " Recent advances in 2D image generation have achieved remarkable\nquality,largely driven by the capacity of diffusion models and the availability\nof large-scale datasets. However, direct 3D generation is still constrained by\nthe scarcity and lower fidelity of 3D datasets. In this paper, we introduce\nZero-1-to-G, a novel approach that addresses this problem by enabling direct\nsingle-view generation on Gaussian splats using pretrained 2D diffusion models.\nOur key insight is that Gaussian splats, a 3D representation, can be decomposed\ninto multi-view images encoding different attributes. This reframes the\nchallenging task of direct 3D generation within a 2D diffusion framework,\nallowing us to leverage the rich priors of pretrained 2D diffusion models. To\nincorporate 3D awareness, we introduce cross-view and cross-attribute attention\nlayers, which capture complex correlations and enforce 3D consistency across\ngenerated splats. This makes Zero-1-to-G the first direct image-to-3D\ngenerative model to effectively utilize pretrained 2D diffusion priors,\nenabling efficient training and improved generalization to unseen objects.\nExtensive experiments on both synthetic and in-the-wild datasets demonstrate\nsuperior performance in 3D object generation, offering a new approach to\nhigh-quality 3D generation.\n", "date": "2025-01-09", "categories": []} {"id": "2501.05445", "title": "Consistent Flow Distillation for Text-to-3D Generation", "abstract": " Score Distillation Sampling (SDS) has made significant strides in distilling\nimage-generative models for 3D generation. However, its\nmaximum-likelihood-seeking behavior often leads to degraded visual quality and\ndiversity, limiting its effectiveness in 3D applications. In this work, we\npropose Consistent Flow Distillation (CFD), which addresses these limitations.\nWe begin by leveraging the gradient of the diffusion ODE or SDE sampling\nprocess to guide the 3D generation. From the gradient-based sampling\nperspective, we find that the consistency of 2D image flows across different\nviewpoints is important for high-quality 3D generation. To achieve this, we\nintroduce multi-view consistent Gaussian noise on the 3D object, which can be\nrendered from various viewpoints to compute the flow gradient. Our experiments\ndemonstrate that CFD, through consistent flows, significantly outperforms\nprevious methods in text-to-3D generation.\n", "date": "2025-01-09", "categories": []} {"id": "2501.04374", "title": "Instructive3D: Editing Large Reconstruction Models with Text\n Instructions", "abstract": " Transformer based methods have enabled users to create, modify, and\ncomprehend text and image data. Recently proposed Large Reconstruction Models\n(LRMs) further extend this by providing the ability to generate high-quality 3D\nmodels with the help of a single object image. These models, however, lack the\nability to manipulate or edit the finer details, such as adding standard design\npatterns or changing the color and reflectance of the generated objects, thus\nlacking fine-grained control that may be very helpful in domains such as\naugmented reality, animation and gaming. Naively training LRMs for this purpose\nwould require generating precisely edited images and 3D object pairs, which is\ncomputationally expensive. In this paper, we propose Instructive3D, a novel LRM\nbased model that integrates generation and fine-grained editing, through user\ntext prompts, of 3D objects into a single model. We accomplish this by adding\nan adapter that performs a diffusion process conditioned on a text prompt\nspecifying edits in the triplane latent space representation of 3D object\nmodels. Our method does not require the generation of edited 3D objects.\nAdditionally, Instructive3D allows us to perform geometrically consistent\nmodifications, as the edits done through user-defined text prompts are applied\nto the triplane latent representation thus enhancing the versatility and\nprecision of 3D objects generated. We compare the objects generated by\nInstructive3D and a baseline that first generates the 3D object meshes using a\nstandard LRM model and then edits these 3D objects using text prompts when\nimages are provided from the Objaverse LVIS dataset. We find that Instructive3D\nproduces qualitatively superior 3D objects with the properties specified by the\nedit prompts.\n", "date": "2025-01-08", "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": "2408.02993", "title": "DreamLCM: Towards High-Quality Text-to-3D Generation via Latent\n Consistency Model", "abstract": " Recently, the text-to-3D task has developed rapidly due to the appearance of\nthe SDS method. However, the SDS method always generates 3D objects with poor\nquality due to the over-smooth issue. This issue is attributed to two factors:\n1) the DDPM single-step inference produces poor guidance gradients; 2) the\nrandomness from the input noises and timesteps averages the details of the 3D\ncontents. In this paper, to address the issue, we propose DreamLCM which\nincorporates the Latent Consistency Model (LCM). DreamLCM leverages the\npowerful image generation capabilities inherent in LCM, enabling generating\nconsistent and high-quality guidance, i.e., predicted noises or images. Powered\nby the improved guidance, the proposed method can provide accurate and detailed\ngradients to optimize the target 3D models. In addition, we propose two\nstrategies to enhance the generation quality further. Firstly, we propose a\nguidance calibration strategy, utilizing Euler Solver to calibrate the guidance\ndistribution to accelerate 3D models to converge. Secondly, we propose a dual\ntimestep strategy, increasing the consistency of guidance and optimizing 3D\nmodels from geometry to appearance in DreamLCM. Experiments show that DreamLCM\nachieves state-of-the-art results in both generation quality and training\nefficiency. The code is available at https://github.com/1YimingZhong/DreamLCM.\n", "date": "2024-08-06", "categories": []} {"id": "2412.17561", "title": "S-INF: Towards Realistic Indoor Scene Synthesis via Scene Implicit\n Neural Field", "abstract": " Learning-based methods have become increasingly popular in 3D indoor scene\nsynthesis (ISS), showing superior performance over traditional\noptimization-based approaches. These learning-based methods typically model\ndistributions on simple yet explicit scene representations using generative\nmodels. However, due to the oversimplified explicit representations that\noverlook detailed information and the lack of guidance from multimodal\nrelationships within the scene, most learning-based methods struggle to\ngenerate indoor scenes with realistic object arrangements and styles. In this\npaper, we introduce a new method, Scene Implicit Neural Field (S-INF), for\nindoor scene synthesis, aiming to learn meaningful representations of\nmultimodal relationships, to enhance the realism of indoor scene synthesis.\nS-INF assumes that the scene layout is often related to the object-detailed\ninformation. It disentangles the multimodal relationships into scene layout\nrelationships and detailed object relationships, fusing them later through\nimplicit neural fields (INFs). By learning specialized scene layout\nrelationships and projecting them into S-INF, we achieve a realistic generation\nof scene layout. Additionally, S-INF captures dense and detailed object\nrelationships through differentiable rendering, ensuring stylistic consistency\nacross objects. Through extensive experiments on the benchmark 3D-FRONT\ndataset, we demonstrate that our method consistently achieves state-of-the-art\nperformance under different types of ISS.\n", "date": "2024-12-23", "categories": []} {"id": "2501.00651", "title": "Taming Feed-forward Reconstruction Models as Latent Encoders for 3D\n Generative Models", "abstract": " Recent AI-based 3D content creation has largely evolved along two paths:\nfeed-forward image-to-3D reconstruction approaches and 3D generative models\ntrained with 2D or 3D supervision. In this work, we show that existing\nfeed-forward reconstruction methods can serve as effective latent encoders for\ntraining 3D generative models, thereby bridging these two paradigms. By reusing\npowerful pre-trained reconstruction models, we avoid computationally expensive\nencoder network training and obtain rich 3D latent features for generative\nmodeling for free. However, the latent spaces of reconstruction models are not\nwell-suited for generative modeling due to their unstructured nature. To enable\nflow-based model training on these latent features, we develop post-processing\npipelines, including protocols to standardize the features and spatial\nweighting to concentrate on important regions. We further incorporate a 2D\nimage space perceptual rendering loss to handle the high-dimensional latent\nspaces. Finally, we propose a multi-stream transformer-based rectified flow\narchitecture to achieve linear scaling and high-quality text-conditioned 3D\ngeneration. Our framework leverages the advancements of feed-forward\nreconstruction models to enhance the scalability of 3D generative modeling,\nachieving both high computational efficiency and state-of-the-art performance\nin text-to-3D generation.\n", "date": "2024-12-31", "categories": []} {"id": "2501.02519", "title": "Layout2Scene: 3D Semantic Layout Guided Scene Generation via Geometry\n and Appearance Diffusion Priors", "abstract": " 3D scene generation conditioned on text prompts has significantly progressed\ndue to the development of 2D diffusion generation models. However, the textual\ndescription of 3D scenes is inherently inaccurate and lacks fine-grained\ncontrol during training, leading to implausible scene generation. As an\nintuitive and feasible solution, the 3D layout allows for precise specification\nof object locations within the scene. To this end, we present a text-to-scene\ngeneration method (namely, Layout2Scene) using additional semantic layout as\nthe prompt to inject precise control of 3D object positions. Specifically, we\nfirst introduce a scene hybrid representation to decouple objects and\nbackgrounds, which is initialized via a pre-trained text-to-3D model. Then, we\npropose a two-stage scheme to optimize the geometry and appearance of the\ninitialized scene separately. To fully leverage 2D diffusion priors in geometry\nand appearance generation, we introduce a semantic-guided geometry diffusion\nmodel and a semantic-geometry guided diffusion model which are finetuned on a\nscene dataset. Extensive experiments demonstrate that our method can generate\nmore plausible and realistic scenes as compared to state-of-the-art approaches.\nFurthermore, the generated scene allows for flexible yet precise editing,\nthereby facilitating multiple downstream applications.\n", "date": "2025-01-05", "categories": []} {"id": "2501.01722", "title": "AR4D: Autoregressive 4D Generation from Monocular Videos", "abstract": " Recent advancements in generative models have ignited substantial interest in\ndynamic 3D content creation (\\ie, 4D generation). Existing approaches primarily\nrely on Score Distillation Sampling (SDS) to infer novel-view videos, typically\nleading to issues such as limited diversity, spatial-temporal inconsistency and\npoor prompt alignment, due to the inherent randomness of SDS. To tackle these\nproblems, we propose AR4D, a novel paradigm for SDS-free 4D generation.\nSpecifically, our paradigm consists of three stages. To begin with, for a\nmonocular video that is either generated or captured, we first utilize\npre-trained expert models to create a 3D representation of the first frame,\nwhich is further fine-tuned to serve as the canonical space. Subsequently,\nmotivated by the fact that videos happen naturally in an autoregressive manner,\nwe propose to generate each frame's 3D representation based on its previous\nframe's representation, as this autoregressive generation manner can facilitate\nmore accurate geometry and motion estimation. Meanwhile, to prevent overfitting\nduring this process, we introduce a progressive view sampling strategy,\nutilizing priors from pre-trained large-scale 3D reconstruction models. To\navoid appearance drift introduced by autoregressive generation, we further\nincorporate a refinement stage based on a global deformation field and the\ngeometry of each frame's 3D representation. Extensive experiments have\ndemonstrated that AR4D can achieve state-of-the-art 4D generation without SDS,\ndelivering greater diversity, improved spatial-temporal consistency, and better\nalignment with input prompts.\n", "date": "2025-01-03", "categories": []} {"id": "2412.21117", "title": "Prometheus: 3D-Aware Latent Diffusion Models for Feed-Forward Text-to-3D\n Scene Generation", "abstract": " In this work, we introduce Prometheus, a 3D-aware latent diffusion model for\ntext-to-3D generation at both object and scene levels in seconds. We formulate\n3D scene generation as multi-view, feed-forward, pixel-aligned 3D Gaussian\ngeneration within the latent diffusion paradigm. To ensure generalizability, we\nbuild our model upon pre-trained text-to-image generation model with only\nminimal adjustments, and further train it using a large number of images from\nboth single-view and multi-view datasets. Furthermore, we introduce an RGB-D\nlatent space into 3D Gaussian generation to disentangle appearance and geometry\ninformation, enabling efficient feed-forward generation of 3D Gaussians with\nbetter fidelity and geometry. Extensive experimental results demonstrate the\neffectiveness of our method in both feed-forward 3D Gaussian reconstruction and\ntext-to-3D generation. Project page:\nhttps://freemty.github.io/project-prometheus/\n", "date": "2024-12-30", "categories": []} {"id": "2412.18608", "title": "PartGen: Part-level 3D Generation and Reconstruction with Multi-View\n Diffusion Models", "abstract": " Text- or image-to-3D generators and 3D scanners can now produce 3D assets\nwith high-quality shapes and textures. These assets typically consist of a\nsingle, fused representation, like an implicit neural field, a Gaussian\nmixture, or a mesh, without any useful structure. However, most applications\nand creative workflows require assets to be made of several meaningful parts\nthat can be manipulated independently. To address this gap, we introduce\nPartGen, a novel approach that generates 3D objects composed of meaningful\nparts starting from text, an image, or an unstructured 3D object. First, given\nmultiple views of a 3D object, generated or rendered, a multi-view diffusion\nmodel extracts a set of plausible and view-consistent part segmentations,\ndividing the object into parts. Then, a second multi-view diffusion model takes\neach part separately, fills in the occlusions, and uses those completed views\nfor 3D reconstruction by feeding them to a 3D reconstruction network. This\ncompletion process considers the context of the entire object to ensure that\nthe parts integrate cohesively. The generative completion model can make up for\nthe information missing due to occlusions; in extreme cases, it can hallucinate\nentirely invisible parts based on the input 3D asset. We evaluate our method on\ngenerated and real 3D assets and show that it outperforms segmentation and\npart-extraction baselines by a large margin. We also showcase downstream\napplications such as 3D part editing.\n", "date": "2024-12-24", "categories": []} {"id": "2412.20470", "title": "JADE: Joint-aware Latent Diffusion for 3D Human Generative Modeling", "abstract": " Generative modeling of 3D human bodies have been studied extensively in\ncomputer vision. The core is to design a compact latent representation that is\nboth expressive and semantically interpretable, yet existing approaches\nstruggle to achieve both requirements. In this work, we introduce JADE, a\ngenerative framework that learns the variations of human shapes with\nfined-grained control. Our key insight is a joint-aware latent representation\nthat decomposes human bodies into skeleton structures, modeled by joint\npositions, and local surface geometries, characterized by features attached to\neach joint. This disentangled latent space design enables geometric and\nsemantic interpretation, facilitating users with flexible controllability. To\ngenerate coherent and plausible human shapes under our proposed decomposition,\nwe also present a cascaded pipeline where two diffusions are employed to model\nthe distribution of skeleton structures and local surface geometries\nrespectively. Extensive experiments are conducted on public datasets, where we\ndemonstrate the effectiveness of JADE framework in multiple tasks in terms of\nautoencoding reconstruction accuracy, editing controllability and generation\nquality compared with existing methods.\n", "date": "2024-12-29", "categories": []} {"id": "2412.20473", "title": "Toward Scene Graph and Layout Guided Complex 3D Scene Generation", "abstract": " Recent advancements in object-centric text-to-3D generation have shown\nimpressive results. However, generating complex 3D scenes remains an open\nchallenge due to the intricate relations between objects. Moreover, existing\nmethods are largely based on score distillation sampling (SDS), which\nconstrains the ability to manipulate multiobjects with specific interactions.\nAddressing these critical yet underexplored issues, we present a novel\nframework of Scene Graph and Layout Guided 3D Scene Generation (GraLa3D). Given\na text prompt describing a complex 3D scene, GraLa3D utilizes LLM to model the\nscene using a scene graph representation with layout bounding box information.\nGraLa3D uniquely constructs the scene graph with single-object nodes and\ncomposite super-nodes. In addition to constraining 3D generation within the\ndesirable layout, a major contribution lies in the modeling of interactions\nbetween objects in a super-node, while alleviating appearance leakage across\nobjects within such nodes. Our experiments confirm that GraLa3D overcomes the\nabove limitations and generates complex 3D scenes closely aligned with text\nprompts.\n", "date": "2024-12-29", "categories": []} {"id": "2412.19149", "title": "Generating Editable Head Avatars with 3D Gaussian GANs", "abstract": " Generating animatable and editable 3D head avatars is essential for various\napplications in computer vision and graphics. Traditional 3D-aware generative\nadversarial networks (GANs), often using implicit fields like Neural Radiance\nFields (NeRF), achieve photorealistic and view-consistent 3D head synthesis.\nHowever, these methods face limitations in deformation flexibility and\neditability, hindering the creation of lifelike and easily modifiable 3D heads.\nWe propose a novel approach that enhances the editability and animation control\nof 3D head avatars by incorporating 3D Gaussian Splatting (3DGS) as an explicit\n3D representation. This method enables easier illumination control and improved\neditability. Central to our approach is the Editable Gaussian Head (EG-Head)\nmodel, which combines a 3D Morphable Model (3DMM) with texture maps, allowing\nprecise expression control and flexible texture editing for accurate animation\nwhile preserving identity. To capture complex non-facial geometries like hair,\nwe use an auxiliary set of 3DGS and tri-plane features. Extensive experiments\ndemonstrate that our approach delivers high-quality 3D-aware synthesis with\nstate-of-the-art controllability. Our code and models are available at\nhttps://github.com/liguohao96/EGG3D.\n", "date": "2024-12-26", "categories": []} {"id": "2402.06149", "title": "HeadStudio: Text to Animatable Head Avatars with 3D Gaussian Splatting", "abstract": " Creating digital avatars from textual prompts has long been a desirable yet\nchallenging task. Despite the promising results achieved with 2D diffusion\npriors, current methods struggle to create high-quality and consistent animated\navatars efficiently. Previous animatable head models like FLAME have difficulty\nin accurately representing detailed texture and geometry. Additionally,\nhigh-quality 3D static representations face challenges in semantically driving\nwith dynamic priors. In this paper, we introduce \\textbf{HeadStudio}, a novel\nframework that utilizes 3D Gaussian splatting to generate realistic and\nanimatable avatars from text prompts. Firstly, we associate 3D Gaussians with\nanimatable head prior model, facilitating semantic animation on high-quality 3D\nrepresentations. To ensure consistent animation, we further enhance the\noptimization from initialization, distillation, and regularization to jointly\nlearn the shape, texture, and animation. Extensive experiments demonstrate the\nefficacy of HeadStudio in generating animatable avatars from textual prompts,\nexhibiting appealing appearances. The avatars are capable of rendering\nhigh-quality real-time ($\\geq 40$ fps) novel views at a resolution of 1024.\nMoreover, These avatars can be smoothly driven by real-world speech and video.\nWe hope that HeadStudio can enhance digital avatar creation and gain popularity\nin the community. Code is at: https://github.com/ZhenglinZhou/HeadStudio.\n", "date": "2024-02-09", "categories": []} {"id": "2406.16710", "title": "ID-Sculpt: ID-aware 3D Head Generation from Single In-the-wild Portrait\n Image", "abstract": " While recent works have achieved great success on image-to-3D object\ngeneration, high quality and fidelity 3D head generation from a single image\nremains a great challenge. Previous text-based methods for generating 3D heads\nwere limited by text descriptions and image-based methods struggled to produce\nhigh-quality head geometry. To handle this challenging problem, we propose a\nnovel framework, ID-Sculpt, to generate high-quality 3D heads while preserving\ntheir identities. Our work incorporates the identity information of the\nportrait image into three parts: 1) geometry initialization, 2) geometry\nsculpting, and 3) texture generation stages. Given a reference portrait image,\nwe first align the identity features with text features to realize ID-aware\nguidance enhancement, which contains the control signals representing the face\ninformation. We then use the canny map, ID features of the portrait image, and\na pre-trained text-to-normal/depth diffusion model to generate ID-aware\ngeometry supervision, and 3D-GAN inversion is employed to generate ID-aware\ngeometry initialization. Furthermore, with the ability to inject identity\ninformation into 3D head generation, we use ID-aware guidance to calculate\nID-aware Score Distillation (ISD) for geometry sculpting. For texture\ngeneration, we adopt the ID Consistent Texture Inpainting and Refinement which\nprogressively expands the view for texture inpainting to obtain an\ninitialization UV texture map. We then use the ID-aware guidance to provide\nimage-level supervision for noisy multi-view images to obtain a refined texture\nmap. Extensive experiments demonstrate that we can generate high-quality 3D\nheads with accurate geometry and texture from a single in-the-wild portrait\nimage.\n", "date": "2024-06-24", "categories": []} {"id": "2412.16717", "title": "GANFusion: Feed-Forward Text-to-3D with Diffusion in GAN Space", "abstract": " We train a feed-forward text-to-3D diffusion generator for human characters\nusing only single-view 2D data for supervision. Existing 3D generative models\ncannot yet match the fidelity of image or video generative models.\nState-of-the-art 3D generators are either trained with explicit 3D supervision\nand are thus limited by the volume and diversity of existing 3D data.\nMeanwhile, generators that can be trained with only 2D data as supervision\ntypically produce coarser results, cannot be text-conditioned, or must revert\nto test-time optimization. We observe that GAN- and diffusion-based generators\nhave complementary qualities: GANs can be trained efficiently with 2D\nsupervision to produce high-quality 3D objects but are hard to condition on\ntext. In contrast, denoising diffusion models can be conditioned efficiently\nbut tend to be hard to train with only 2D supervision. We introduce GANFusion,\nwhich starts by generating unconditional triplane features for 3D data using a\nGAN architecture trained with only single-view 2D data. We then generate random\nsamples from the GAN, caption them, and train a text-conditioned diffusion\nmodel that directly learns to sample from the space of good triplane features\nthat can be decoded into 3D objects.\n", "date": "2024-12-21", "categories": []} {"id": "2412.17715", "title": "GaussianPainter: Painting Point Cloud into 3D Gaussians with Normal\n Guidance", "abstract": " In this paper, we present GaussianPainter, the first method to paint a point\ncloud into 3D Gaussians given a reference image. GaussianPainter introduces an\ninnovative feed-forward approach to overcome the limitations of time-consuming\ntest-time optimization in 3D Gaussian splatting. Our method addresses a\ncritical challenge in the field: the non-uniqueness problem inherent in the\nlarge parameter space of 3D Gaussian splatting. This space, encompassing\nrotation, anisotropic scales, and spherical harmonic coefficients, introduces\nthe challenge of rendering similar images from substantially different Gaussian\nfields. As a result, feed-forward networks face instability when attempting to\ndirectly predict high-quality Gaussian fields, struggling to converge on\nconsistent parameters for a given output. To address this issue, we propose to\nestimate a surface normal for each point to determine its Gaussian rotation.\nThis strategy enables the network to effectively predict the remaining Gaussian\nparameters in the constrained space. We further enhance our approach with an\nappearance injection module, incorporating reference image appearance into\nGaussian fields via a multiscale triplane representation. Our method\nsuccessfully balances efficiency and fidelity in 3D Gaussian generation,\nachieving high-quality, diverse, and robust 3D content creation from point\nclouds in a single forward pass.\n", "date": "2024-12-23", "categories": []} {"id": "2412.15491", "title": "GCA-3D: Towards Generalized and Consistent Domain Adaptation of 3D\n Generators", "abstract": " Recently, 3D generative domain adaptation has emerged to adapt the\npre-trained generator to other domains without collecting massive datasets and\ncamera pose distributions. Typically, they leverage large-scale pre-trained\ntext-to-image diffusion models to synthesize images for the target domain and\nthen fine-tune the 3D model. However, they suffer from the tedious pipeline of\ndata generation, which inevitably introduces pose bias between the source\ndomain and synthetic dataset. Furthermore, they are not generalized to support\none-shot image-guided domain adaptation, which is more challenging due to the\nmore severe pose bias and additional identity bias introduced by the single\nimage reference. To address these issues, we propose GCA-3D, a generalized and\nconsistent 3D domain adaptation method without the intricate pipeline of data\ngeneration. Different from previous pipeline methods, we introduce multi-modal\ndepth-aware score distillation sampling loss to efficiently adapt 3D generative\nmodels in a non-adversarial manner. This multi-modal loss enables GCA-3D in\nboth text prompt and one-shot image prompt adaptation. Besides, it leverages\nper-instance depth maps from the volume rendering module to mitigate the\noverfitting problem and retain the diversity of results. To enhance the pose\nand identity consistency, we further propose a hierarchical spatial consistency\nloss to align the spatial structure between the generated images in the source\nand target domain. Experiments demonstrate that GCA-3D outperforms previous\nmethods in terms of efficiency, generalization, pose accuracy, and identity\nconsistency.\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": "2412.09545", "title": "SimAvatar: Simulation-Ready Avatars with Layered Hair and Clothing", "abstract": " We introduce SimAvatar, a framework designed to generate simulation-ready\nclothed 3D human avatars from a text prompt. Current text-driven human avatar\ngeneration methods either model hair, clothing, and the human body using a\nunified geometry or produce hair and garments that are not easily adaptable for\nsimulation within existing simulation pipelines. The primary challenge lies in\nrepresenting the hair and garment geometry in a way that allows leveraging\nestablished prior knowledge from foundational image diffusion models (e.g.,\nStable Diffusion) while being simulation-ready using either physics or neural\nsimulators. To address this task, we propose a two-stage framework that\ncombines the flexibility of 3D Gaussians with simulation-ready hair strands and\ngarment meshes. Specifically, we first employ three text-conditioned 3D\ngenerative models to generate garment mesh, body shape and hair strands from\nthe given text prompt. To leverage prior knowledge from foundational diffusion\nmodels, we attach 3D Gaussians to the body mesh, garment mesh, as well as hair\nstrands and learn the avatar appearance through optimization. To drive the\navatar given a pose sequence, we first apply physics simulators onto the\ngarment meshes and hair strands. We then transfer the motion onto 3D Gaussians\nthrough carefully designed mechanisms for each body part. As a result, our\nsynthesized avatars have vivid texture and realistic dynamic motion. To the\nbest of our knowledge, our method is the first to produce highly realistic,\nfully simulation-ready 3D avatars, surpassing the capabilities of current\napproaches.\n", "date": "2024-12-12", "categories": []} {"id": "2412.14214", "title": "GraphicsDreamer: Image to 3D Generation with Physical Consistency", "abstract": " Recently, the surge of efficient and automated 3D AI-generated content (AIGC)\nmethods has increasingly illuminated the path of transforming human imagination\ninto complex 3D structures. However, the automated generation of 3D content is\nstill significantly lags in industrial application. This gap exists because 3D\nmodeling demands high-quality assets with sharp geometry, exquisite topology,\nand physically based rendering (PBR), among other criteria. To narrow the\ndisparity between generated results and artists' expectations, we introduce\nGraphicsDreamer, a method for creating highly usable 3D meshes from single\nimages. To better capture the geometry and material details, we integrate the\nPBR lighting equation into our cross-domain diffusion model, concurrently\npredicting multi-view color, normal, depth images, and PBR materials. In the\ngeometry fusion stage, we continue to enforce the PBR constraints, ensuring\nthat the generated 3D objects possess reliable texture details, supporting\nrealistic relighting. Furthermore, our method incorporates topology\noptimization and fast UV unwrapping capabilities, allowing the 3D products to\nbe seamlessly imported into graphics engines. Extensive experiments demonstrate\nthat our model can produce high quality 3D assets in a reasonable time cost\ncompared to previous methods.\n", "date": "2024-12-18", "categories": []} {"id": "2412.14464", "title": "LiftRefine: Progressively Refined View Synthesis from 3D Lifting with\n Volume-Triplane Representations", "abstract": " We propose a new view synthesis method via synthesizing a 3D neural field\nfrom both single or few-view input images. To address the ill-posed nature of\nthe image-to-3D generation problem, we devise a two-stage method that involves\na reconstruction model and a diffusion model for view synthesis. Our\nreconstruction model first lifts one or more input images to the 3D space from\na volume as the coarse-scale 3D representation followed by a tri-plane as the\nfine-scale 3D representation. To mitigate the ambiguity in occluded regions,\nour diffusion model then hallucinates missing details in the rendered images\nfrom tri-planes. We then introduce a new progressive refinement technique that\niteratively applies the reconstruction and diffusion model to gradually\nsynthesize novel views, boosting the overall quality of the 3D representations\nand their rendering. Empirical evaluation demonstrates the superiority of our\nmethod over state-of-the-art methods on the synthetic SRN-Car dataset, the\nin-the-wild CO3D dataset, and large-scale Objaverse dataset while achieving\nboth sampling efficacy and multi-view consistency.\n", "date": "2024-12-19", "categories": []} {"id": "2412.15086", "title": "Learning Disentangled Equivariant Representation for Explicitly\n Controllable 3D Molecule Generation", "abstract": " We consider the conditional generation of 3D drug-like molecules with\n\\textit{explicit control} over molecular properties such as drug-like\nproperties (e.g., Quantitative Estimate of Druglikeness or Synthetic\nAccessibility score) and effectively binding to specific protein sites. To\ntackle this problem, we propose an E(3)-equivariant Wasserstein autoencoder and\nfactorize the latent space of our generative model into two disentangled\naspects: molecular properties and the remaining structural context of 3D\nmolecules. Our model ensures explicit control over these molecular attributes\nwhile maintaining equivariance of coordinate representation and invariance of\ndata likelihood. Furthermore, we introduce a novel alignment-based coordinate\nloss to adapt equivariant networks for auto-regressive de-novo 3D molecule\ngeneration from scratch. Extensive experiments validate our model's\neffectiveness on property-guided and context-guided molecule generation, both\nfor de-novo 3D molecule design and structure-based drug discovery against\nprotein targets.\n", "date": "2024-12-19", "categories": []} {"id": "2404.04363", "title": "Idea23D: Collaborative LMM Agents Enable 3D Model Generation from\n Interleaved Multimodal Inputs", "abstract": " With the success of 2D diffusion models, 2D AIGC content has already\ntransformed our lives. Recently, this success has been extended to 3D AIGC,\nwith state-of-the-art methods generating textured 3D models from single images\nor text. However, we argue that current 3D AIGC methods still do not fully\nunleash human creativity. We often imagine 3D content made from multimodal\ninputs, such as what it would look like if my pet bunny were eating a doughnut\non the table. In this paper, we explore a novel 3D AIGC approach: generating 3D\ncontent from IDEAs. An IDEA is a multimodal input composed of text, image, and\n3D models. To our knowledge, this challenging and exciting 3D AIGC setting has\nnot been studied before. We propose the new framework Idea23D, which combines\nthree agents based on large multimodal models (LMMs) and existing algorithmic\ntools. These three LMM-based agents are tasked with prompt generation, model\nselection, and feedback reflection. They collaborate and critique each other in\na fully automated loop, without human intervention. The framework then\ngenerates a text prompt to create 3D models that align closely with the input\nIDEAs. We demonstrate impressive 3D AIGC results that surpass previous methods.\nTo comprehensively assess the 3D AIGC capabilities of Idea23D, we introduce the\nEval3DAIGC-198 dataset, containing 198 multimodal inputs for 3D generation\ntasks. This dataset evaluates the alignment between generated 3D content and\ninput IDEAs. Our user study and quantitative results show that Idea23D\nsignificantly improves the success rate and accuracy of 3D generation, with\nexcellent compatibility across various LMM, Text-to-Image, and Image-to-3D\nmodels. Code and dataset are available at \\url{https://idea23d.github.io/}.\n", "date": "2024-04-05", "categories": []} {"id": "2412.14148", "title": "MCMat: Multiview-Consistent and Physically Accurate PBR Material\n Generation", "abstract": " Existing 2D methods utilize UNet-based diffusion models to generate\nmulti-view physically-based rendering (PBR) maps but struggle with multi-view\ninconsistency, while some 3D methods directly generate UV maps, encountering\ngeneralization issues due to the limited 3D data. To address these problems, we\npropose a two-stage approach, including multi-view generation and UV materials\nrefinement. In the generation stage, we adopt a Diffusion Transformer (DiT)\nmodel to generate PBR materials, where both the specially designed multi-branch\nDiT and reference-based DiT blocks adopt a global attention mechanism to\npromote feature interaction and fusion between different views, thereby\nimproving multi-view consistency. In addition, we adopt a PBR-based diffusion\nloss to ensure that the generated materials align with realistic physical\nprinciples. In the refinement stage, we propose a material-refined DiT that\nperforms inpainting in empty areas and enhances details in UV space. Except for\nthe normal condition, this refinement also takes the material map from the\ngeneration stage as an additional condition to reduce the learning difficulty\nand improve generalization. Extensive experiments show that our method achieves\nstate-of-the-art performance in texturing 3D objects with PBR materials and\nprovides significant advantages for graphics relighting applications. Project\nPage: https://lingtengqiu.github.io/2024/MCMat/\n", "date": "2024-12-18", "categories": []} {"id": "2403.15698", "title": "SceneX: Procedural Controllable Large-scale Scene Generation", "abstract": " Developing comprehensive explicit world models is crucial for understanding\nand simulating real-world scenarios. Recently, Procedural Controllable\nGeneration (PCG) has gained significant attention in large-scale scene\ngeneration by enabling the creation of scalable, high-quality assets. However,\nPCG faces challenges such as limited modular diversity, high expertise\nrequirements, and challenges in managing the diverse elements and structures in\ncomplex scenes. In this paper, we introduce a large-scale scene generation\nframework, SceneX, which can automatically produce high-quality procedural\nmodels according to designers' textual descriptions. Specifically, the proposed\nmethod comprises two components, PCGHub and PCGPlanner. The former encompasses\nan extensive collection of accessible procedural assets and thousands of\nhand-craft API documents to perform as a standard protocol for PCG controller.\nThe latter aims to generate executable actions for Blender to produce\ncontrollable and precise 3D assets guided by the user's instructions. Extensive\nexperiments demonstrated the capability of our method in controllable\nlarge-scale scene generation, including nature scenes and unbounded cities, as\nwell as scene editing such as asset placement and season translation.\n", "date": "2024-03-23", "categories": []} {"id": "2406.06367", "title": "MVGamba: Unify 3D Content Generation as State Space Sequence Modeling", "abstract": " Recent 3D large reconstruction models (LRMs) can generate high-quality 3D\ncontent in sub-seconds by integrating multi-view diffusion models with scalable\nmulti-view reconstructors. Current works further leverage 3D Gaussian Splatting\nas 3D representation for improved visual quality and rendering efficiency.\nHowever, we observe that existing Gaussian reconstruction models often suffer\nfrom multi-view inconsistency and blurred textures. We attribute this to the\ncompromise of multi-view information propagation in favor of adopting powerful\nyet computationally intensive architectures (e.g., Transformers). To address\nthis issue, we introduce MVGamba, a general and lightweight Gaussian\nreconstruction model featuring a multi-view Gaussian reconstructor based on the\nRNN-like State Space Model (SSM). Our Gaussian reconstructor propagates causal\ncontext containing multi-view information for cross-view self-refinement while\ngenerating a long sequence of Gaussians for fine-detail modeling with linear\ncomplexity. With off-the-shelf multi-view diffusion models integrated, MVGamba\nunifies 3D generation tasks from a single image, sparse images, or text\nprompts. Extensive experiments demonstrate that MVGamba outperforms\nstate-of-the-art baselines in all 3D content generation scenarios with\napproximately only $0.1\\times$ of the model size.\n", "date": "2024-06-10", "categories": []} {"id": "2406.08475", "title": "Human-3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent\n Diffusion Models", "abstract": " Creating realistic avatars from a single RGB image is an attractive yet\nchallenging problem. Due to its ill-posed nature, recent works leverage\npowerful prior from 2D diffusion models pretrained on large datasets. Although\n2D diffusion models demonstrate strong generalization capability, they cannot\nprovide multi-view shape priors with guaranteed 3D consistency. We propose\nHuman 3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent\nDiffusion. Our key insight is that 2D multi-view diffusion and 3D\nreconstruction models provide complementary information for each other, and by\ncoupling them in a tight manner, we can fully leverage the potential of both\nmodels. We introduce a novel image-conditioned generative 3D Gaussian Splats\nreconstruction model that leverages the priors from 2D multi-view diffusion\nmodels, and provides an explicit 3D representation, which further guides the 2D\nreverse sampling process to have better 3D consistency. Experiments show that\nour proposed framework outperforms state-of-the-art methods and enables the\ncreation of realistic avatars from a single RGB image, achieving high-fidelity\nin both geometry and appearance. Extensive ablations also validate the efficacy\nof our design, (1) multi-view 2D priors conditioning in generative 3D\nreconstruction and (2) consistency refinement of sampling trajectory via the\nexplicit 3D representation. Our code and models will be released on\nhttps://yuxuan-xue.com/human-3diffusion.\n", "date": "2024-06-12", "categories": []} {"id": "2412.11599", "title": "3D$^2$-Actor: Learning Pose-Conditioned 3D-Aware Denoiser for Realistic\n Gaussian Avatar Modeling", "abstract": " Advancements in neural implicit representations and differentiable rendering\nhave markedly improved the ability to learn animatable 3D avatars from sparse\nmulti-view RGB videos. However, current methods that map observation space to\ncanonical space often face challenges in capturing pose-dependent details and\ngeneralizing to novel poses. While diffusion models have demonstrated\nremarkable zero-shot capabilities in 2D image generation, their potential for\ncreating animatable 3D avatars from 2D inputs remains underexplored. In this\nwork, we introduce 3D$^2$-Actor, a novel approach featuring a pose-conditioned\n3D-aware human modeling pipeline that integrates iterative 2D denoising and 3D\nrectifying steps. The 2D denoiser, guided by pose cues, generates detailed\nmulti-view images that provide the rich feature set necessary for high-fidelity\n3D reconstruction and pose rendering. Complementing this, our Gaussian-based 3D\nrectifier renders images with enhanced 3D consistency through a two-stage\nprojection strategy and a novel local coordinate representation. Additionally,\nwe propose an innovative sampling strategy to ensure smooth temporal continuity\nacross frames in video synthesis. Our method effectively addresses the\nlimitations of traditional numerical solutions in handling ill-posed mappings,\nproducing realistic and animatable 3D human avatars. Experimental results\ndemonstrate that 3D$^2$-Actor excels in high-fidelity avatar modeling and\nrobustly generalizes to novel poses. Code is available at:\nhttps://github.com/silence-tang/GaussianActor.\n", "date": "2024-12-16", "categories": []} {"id": "2412.09648", "title": "DSplats: 3D Generation by Denoising Splats-Based Multiview Diffusion\n Models", "abstract": " Generating high-quality 3D content requires models capable of learning robust\ndistributions of complex scenes and the real-world objects within them. Recent\nGaussian-based 3D reconstruction techniques have achieved impressive results in\nrecovering high-fidelity 3D assets from sparse input images by predicting 3D\nGaussians in a feed-forward manner. However, these techniques often lack the\nextensive priors and expressiveness offered by Diffusion Models. On the other\nhand, 2D Diffusion Models, which have been successfully applied to denoise\nmultiview images, show potential for generating a wide range of photorealistic\n3D outputs but still fall short on explicit 3D priors and consistency. In this\nwork, we aim to bridge these two approaches by introducing DSplats, a novel\nmethod that directly denoises multiview images using Gaussian Splat-based\nReconstructors to produce a diverse array of realistic 3D assets. To harness\nthe extensive priors of 2D Diffusion Models, we incorporate a pretrained Latent\nDiffusion Model into the reconstructor backbone to predict a set of 3D\nGaussians. Additionally, the explicit 3D representation embedded in the\ndenoising network provides a strong inductive bias, ensuring geometrically\nconsistent novel view generation. Our qualitative and quantitative experiments\ndemonstrate that DSplats not only produces high-quality, spatially consistent\noutputs, but also sets a new standard in single-image to 3D reconstruction.\nWhen evaluated on the Google Scanned Objects dataset, DSplats achieves a PSNR\nof 20.38, an SSIM of 0.842, and an LPIPS of 0.109.\n", "date": "2024-12-11", "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": "2412.09008", "title": "MS2Mesh-XR: Multi-modal Sketch-to-Mesh Generation in XR Environments", "abstract": " We present MS2Mesh-XR, a novel multi-modal sketch-to-mesh generation pipeline\nthat enables users to create realistic 3D objects in extended reality (XR)\nenvironments using hand-drawn sketches assisted by voice inputs. In specific,\nusers can intuitively sketch objects using natural hand movements in mid-air\nwithin a virtual environment. By integrating voice inputs, we devise ControlNet\nto infer realistic images based on the drawn sketches and interpreted text\nprompts. Users can then review and select their preferred image, which is\nsubsequently reconstructed into a detailed 3D mesh using the Convolutional\nReconstruction Model. In particular, our proposed pipeline can generate a\nhigh-quality 3D mesh in less than 20 seconds, allowing for immersive\nvisualization and manipulation in run-time XR scenes. We demonstrate the\npracticability of our pipeline through two use cases in XR settings. By\nleveraging natural user inputs and cutting-edge generative AI capabilities, our\napproach can significantly facilitate XR-based creative production and enhance\nuser experiences. Our code and demo will be available at:\nhttps://yueqiu0911.github.io/MS2Mesh-XR/\n", "date": "2024-12-12", "categories": []} {"id": "2412.09548", "title": "Meshtron: High-Fidelity, Artist-Like 3D Mesh Generation at Scale", "abstract": " Meshes are fundamental representations of 3D surfaces. However, creating\nhigh-quality meshes is a labor-intensive task that requires significant time\nand expertise in 3D modeling. While a delicate object often requires over\n$10^4$ faces to be accurately modeled, recent attempts at generating\nartist-like meshes are limited to $1.6$K faces and heavy discretization of\nvertex coordinates. Hence, scaling both the maximum face count and vertex\ncoordinate resolution is crucial to producing high-quality meshes of realistic,\ncomplex 3D objects. We present Meshtron, a novel autoregressive mesh generation\nmodel able to generate meshes with up to 64K faces at 1024-level coordinate\nresolution --over an order of magnitude higher face count and $8{\\times}$\nhigher coordinate resolution than current state-of-the-art methods. Meshtron's\nscalability is driven by four key components: (1) an hourglass neural\narchitecture, (2) truncated sequence training, (3) sliding window inference,\n(4) a robust sampling strategy that enforces the order of mesh sequences. This\nresults in over $50{\\%}$ less training memory, $2.5{\\times}$ faster throughput,\nand better consistency than existing works. Meshtron generates meshes of\ndetailed, complex 3D objects at unprecedented levels of resolution and\nfidelity, closely resembling those created by professional artists, and opening\nthe door to more realistic generation of detailed 3D assets for animation,\ngaming, and virtual environments.\n", "date": "2024-12-12", "categories": []} {"id": "2412.07984", "title": "Diffusion-Based Attention Warping for Consistent 3D Scene Editing", "abstract": " We present a novel method for 3D scene editing using diffusion models,\ndesigned to ensure view consistency and realism across perspectives. Our\napproach leverages attention features extracted from a single reference image\nto define the intended edits. These features are warped across multiple views\nby aligning them with scene geometry derived from Gaussian splatting depth\nestimates. Injecting these warped features into other viewpoints enables\ncoherent propagation of edits, achieving high fidelity and spatial alignment in\n3D space. Extensive evaluations demonstrate the effectiveness of our method in\ngenerating versatile edits of 3D scenes, significantly advancing the\ncapabilities of scene manipulation compared to the existing methods. Project\npage: \\url{https://attention-warp.github.io}\n", "date": "2024-12-10", "categories": []} {"id": "2405.21066", "title": "Mixed Diffusion for 3D Indoor Scene Synthesis", "abstract": " Generating realistic 3D scenes is an area of growing interest in computer\nvision and robotics. However, creating high-quality, diverse synthetic 3D\ncontent often requires expert intervention, making it costly and complex.\nRecently, efforts to automate this process with learning techniques,\nparticularly diffusion models, have shown significant improvements in tasks\nlike furniture rearrangement. However, applying diffusion models to\nfloor-conditioned indoor scene synthesis remains under-explored. This task is\nespecially challenging as it requires arranging objects in continuous space\nwhile selecting from discrete object categories, posing unique difficulties for\nconventional diffusion methods. To bridge this gap, we present MiDiffusion, a\nnovel mixed discrete-continuous diffusion model designed to synthesize\nplausible 3D indoor scenes given a floor plan and pre-arranged objects. We\nrepresent a scene layout by a 2D floor plan and a set of objects, each defined\nby category, location, size, and orientation. Our approach uniquely applies\nstructured corruption across mixed discrete semantic and continuous geometric\ndomains, resulting in a better-conditioned problem for denoising. Evaluated on\nthe 3D-FRONT dataset, MiDiffusion outperforms state-of-the-art autoregressive\nand diffusion models in floor-conditioned 3D scene synthesis. Additionally, it\neffectively handles partial object constraints via a corruption-and-masking\nstrategy without task-specific training, demonstrating advantages in scene\ncompletion and furniture arrangement tasks.\n", "date": "2024-05-31", "categories": []} {"id": "2407.17572", "title": "CityX: Controllable Procedural Content Generation for Unbounded 3D\n Cities", "abstract": " Urban areas, as the primary human habitat in modern civilization, accommodate\na broad spectrum of social activities. With the surge of embodied intelligence,\nrecent years have witnessed an increasing presence of physical agents in urban\nareas, such as autonomous vehicles and delivery robots. As a result,\npractitioners significantly value crafting authentic, simulation-ready 3D\ncities to facilitate the training and verification of such agents. However,\nthis task is quite challenging. Current generative methods fall short in either\ndiversity, controllability, or fidelity. In this work, we resort to the\nprocedural content generation (PCG) technique for high-fidelity generation. It\nassembles superior assets according to empirical rules, ultimately leading to\nindustrial-grade outcomes. To ensure diverse and self contained creation, we\ndesign a management protocol to accommodate extensive PCG plugins with distinct\nfunctions and interfaces. Based on this unified PCG library, we develop a\nmulti-agent framework to transform multi-modal instructions, including OSM,\nsemantic maps, and satellite images, into executable programs. The programs\ncoordinate relevant plugins to construct the 3D city consistent with the\ncontrol condition. A visual feedback scheme is introduced to further refine the\ninitial outcomes. Our method, named CityX, demonstrates its superiority in\ncreating diverse, controllable, and realistic 3D urban scenes. The synthetic\nscenes can be seamlessly deployed as a real-time simulator and an infinite data\ngenerator for embodied intelligence research. Our project page:\nhttps://cityx-lab.github.io.\n", "date": "2024-07-24", "categories": []} {"id": "2412.05560", "title": "Text-to-3D Gaussian Splatting with Physics-Grounded Motion Generation", "abstract": " Text-to-3D generation is a valuable technology in virtual reality and digital\ncontent creation. While recent works have pushed the boundaries of text-to-3D\ngeneration, producing high-fidelity 3D objects with inefficient prompts and\nsimulating their physics-grounded motion accurately still remain unsolved\nchallenges. To address these challenges, we present an innovative framework\nthat utilizes the Large Language Model (LLM)-refined prompts and diffusion\npriors-guided Gaussian Splatting (GS) for generating 3D models with accurate\nappearances and geometric structures. We also incorporate a continuum\nmechanics-based deformation map and color regularization to synthesize vivid\nphysics-grounded motion for the generated 3D Gaussians, adhering to the\nconservation of mass and momentum. By integrating text-to-3D generation with\nphysics-grounded motion synthesis, our framework renders photo-realistic 3D\nobjects that exhibit physics-aware motion, accurately reflecting the behaviors\nof the objects under various forces and constraints across different materials.\nExtensive experiments demonstrate that our approach achieves high-quality 3D\ngenerations with realistic physics-grounded motion.\n", "date": "2024-12-07", "categories": []} {"id": "2412.06592", "title": "PrEditor3D: Fast and Precise 3D Shape Editing", "abstract": " We propose a training-free approach to 3D editing that enables the editing of\na single shape within a few minutes. The edited 3D mesh aligns well with the\nprompts, and remains identical for regions that are not intended to be altered.\nTo this end, we first project the 3D object onto 4-view images and perform\nsynchronized multi-view image editing along with user-guided text prompts and\nuser-provided rough masks. However, the targeted regions to be edited are\nambiguous due to projection from 3D to 2D. To ensure precise editing only in\nintended regions, we develop a 3D segmentation pipeline that detects edited\nareas in 3D space, followed by a merging algorithm to seamlessly integrate\nedited 3D regions with the original input. Extensive experiments demonstrate\nthe superiority of our method over previous approaches, enabling fast,\nhigh-quality editing while preserving unintended regions.\n", "date": "2024-12-09", "categories": []} {"id": "2412.06614", "title": "MVReward: Better Aligning and Evaluating Multi-View Diffusion Models\n with Human Preferences", "abstract": " Recent years have witnessed remarkable progress in 3D content generation.\nHowever, corresponding evaluation methods struggle to keep pace. Automatic\napproaches have proven challenging to align with human preferences, and the\nmixed comparison of text- and image-driven methods often leads to unfair\nevaluations. In this paper, we present a comprehensive framework to better\nalign and evaluate multi-view diffusion models with human preferences. To begin\nwith, we first collect and filter a standardized image prompt set from\nDALL$\\cdot$E and Objaverse, which we then use to generate multi-view assets\nwith several multi-view diffusion models. Through a systematic ranking pipeline\non these assets, we obtain a human annotation dataset with 16k expert pairwise\ncomparisons and train a reward model, coined MVReward, to effectively encode\nhuman preferences. With MVReward, image-driven 3D methods can be evaluated\nagainst each other in a more fair and transparent manner. Building on this, we\nfurther propose Multi-View Preference Learning (MVP), a plug-and-play\nmulti-view diffusion tuning strategy. Extensive experiments demonstrate that\nMVReward can serve as a reliable metric and MVP consistently enhances the\nalignment of multi-view diffusion models with human preferences.\n", "date": "2024-12-09", "categories": []} {"id": "2412.06785", "title": "Tactile DreamFusion: Exploiting Tactile Sensing for 3D Generation", "abstract": " 3D generation methods have shown visually compelling results powered by\ndiffusion image priors. However, they often fail to produce realistic geometric\ndetails, resulting in overly smooth surfaces or geometric details inaccurately\nbaked in albedo maps. To address this, we introduce a new method that\nincorporates touch as an additional modality to improve the geometric details\nof generated 3D assets. We design a lightweight 3D texture field to synthesize\nvisual and tactile textures, guided by 2D diffusion model priors on both visual\nand tactile domains. We condition the visual texture generation on\nhigh-resolution tactile normals and guide the patch-based tactile texture\nrefinement with a customized TextureDreambooth. We further present a multi-part\ngeneration pipeline that enables us to synthesize different textures across\nvarious regions. To our knowledge, we are the first to leverage high-resolution\ntactile sensing to enhance geometric details for 3D generation tasks. We\nevaluate our method in both text-to-3D and image-to-3D settings. Our\nexperiments demonstrate that our method provides customized and realistic fine\ngeometric textures while maintaining accurate alignment between two modalities\nof vision and touch.\n", "date": "2024-12-09", "categories": []} {"id": "2412.00091", "title": "Graph Canvas for Controllable 3D Scene Generation", "abstract": " Spatial intelligence is foundational to AI systems that interact with the\nphysical world, particularly in 3D scene generation and spatial comprehension.\nCurrent methodologies for 3D scene generation often rely heavily on predefined\ndatasets, and struggle to adapt dynamically to changing spatial relationships.\nIn this paper, we introduce GraphCanvas3D, a programmable, extensible, and\nadaptable framework for controllable 3D scene generation. Leveraging in-context\nlearning, GraphCanvas3D enables dynamic adaptability without the need for\nretraining, supporting flexible and customizable scene creation. Our framework\nemploys hierarchical, graph-driven scene descriptions, representing spatial\nelements as graph nodes and establishing coherent relationships among objects\nin 3D environments. Unlike conventional approaches, which are constrained in\nadaptability and often require predefined input masks or retraining for\nmodifications, GraphCanvas3D allows for seamless object manipulation and scene\nadjustments on the fly. Additionally, GraphCanvas3D supports 4D scene\ngeneration, incorporating temporal dynamics to model changes over time.\nExperimental results and user studies demonstrate that GraphCanvas3D enhances\nusability, flexibility, and adaptability for scene generation. Our code and\nmodels are available on the project website:\nhttps://github.com/ILGLJ/Graph-Canvas.\n", "date": "2024-11-27", "categories": []} {"id": "2407.14979", "title": "RGB2Point: 3D Point Cloud Generation from Single RGB Images", "abstract": " We introduce RGB2Point, an unposed single-view RGB image to a 3D point cloud\ngeneration based on Transformer. RGB2Point takes an input image of an object\nand generates a dense 3D point cloud. Contrary to prior works based on CNN\nlayers and diffusion denoising approaches, we use pre-trained Transformer\nlayers that are fast and generate high-quality point clouds with consistent\nquality over available categories. Our generated point clouds demonstrate high\nquality on a real-world dataset, as evidenced by improved Chamfer distance\n(51.15%) and Earth Mover's distance (45.96%) metrics compared to the current\nstate-of-the-art. Additionally, our approach shows a better quality on a\nsynthetic dataset, achieving better Chamfer distance (39.26%), Earth Mover's\ndistance (26.95%), and F-score (47.16%). Moreover, our method produces 63.1%\nmore consistent high-quality results across various object categories compared\nto prior works. Furthermore, RGB2Point is computationally efficient, requiring\nonly 2.3GB of VRAM to reconstruct a 3D point cloud from a single RGB image, and\nour implementation generates the results 15,133x faster than a SOTA\ndiffusion-based model.\n", "date": "2024-07-20", "categories": []} {"id": "2412.04470", "title": "Turbo3D: Ultra-fast Text-to-3D Generation", "abstract": " We present Turbo3D, an ultra-fast text-to-3D system capable of generating\nhigh-quality Gaussian splatting assets in under one second. Turbo3D employs a\nrapid 4-step, 4-view diffusion generator and an efficient feed-forward Gaussian\nreconstructor, both operating in latent space. The 4-step, 4-view generator is\na student model distilled through a novel Dual-Teacher approach, which\nencourages the student to learn view consistency from a multi-view teacher and\nphoto-realism from a single-view teacher. By shifting the Gaussian\nreconstructor's inputs from pixel space to latent space, we eliminate the extra\nimage decoding time and halve the transformer sequence length for maximum\nefficiency. Our method demonstrates superior 3D generation results compared to\nprevious baselines, while operating in a fraction of their runtime.\n", "date": "2024-12-05", "categories": []} {"id": "2312.17225", "title": "4DGen: Grounded 4D Content Generation with Spatial-temporal Consistency", "abstract": " Aided by text-to-image and text-to-video diffusion models, existing 4D\ncontent creation pipelines utilize score distillation sampling to optimize the\nentire dynamic 3D scene. However, as these pipelines generate 4D content from\ntext or image inputs directly, they are constrained by limited motion\ncapabilities and depend on unreliable prompt engineering for desired results.\nTo address these problems, this work introduces \\textbf{4DGen}, a novel\nframework for grounded 4D content creation. We identify monocular video\nsequences as a key component in constructing the 4D content. Our pipeline\nfacilitates controllable 4D generation, enabling users to specify the motion\nvia monocular video or adopt image-to-video generations, thus offering superior\ncontrol over content creation. Furthermore, we construct our 4D representation\nusing dynamic 3D Gaussians, which permits efficient, high-resolution\nsupervision through rendering during training, thereby facilitating\nhigh-quality 4D generation. Additionally, we employ spatial-temporal pseudo\nlabels on anchor frames, along with seamless consistency priors implemented\nthrough 3D-aware score distillation sampling and smoothness regularizations.\nCompared to existing video-to-4D baselines, our approach yields superior\nresults in faithfully reconstructing input signals and realistically inferring\nrenderings from novel viewpoints and timesteps. More importantly, compared to\nprevious image-to-4D and text-to-4D works, 4DGen supports grounded generation,\noffering users enhanced control and improved motion generation capabilities, a\nfeature difficult to achieve with previous methods. Project page:\nhttps://vita-group.github.io/4DGen/\n", "date": "2023-12-28", "categories": []} {"id": "2412.03571", "title": "Style3D: Attention-guided Multi-view Style Transfer for 3D Object\n Generation", "abstract": " We present Style3D, a novel approach for generating stylized 3D objects from\na content image and a style image. Unlike most previous methods that require\ncase- or style-specific training, Style3D supports instant 3D object\nstylization. Our key insight is that 3D object stylization can be decomposed\ninto two interconnected processes: multi-view dual-feature alignment and\nsparse-view spatial reconstruction. We introduce MultiFusion Attention, an\nattention-guided technique to achieve multi-view stylization from the\ncontent-style pair. Specifically, the query features from the content image\npreserve geometric consistency across multiple views, while the key and value\nfeatures from the style image are used to guide the stylistic transfer. This\ndual-feature alignment ensures that spatial coherence and stylistic fidelity\nare maintained across multi-view images. Finally, a large 3D reconstruction\nmodel is introduced to generate coherent stylized 3D objects. By establishing\nan interplay between structural and stylistic features across multiple views,\nour approach enables a holistic 3D stylization process. Extensive experiments\ndemonstrate that Style3D offers a more flexible and scalable solution for\ngenerating style-consistent 3D assets, surpassing existing methods in both\ncomputational efficiency and visual quality.\n", "date": "2024-12-04", "categories": []} {"id": "2412.01801", "title": "SceneFactor: Factored Latent 3D Diffusion for Controllable 3D Scene\n Generation", "abstract": " We present SceneFactor, a diffusion-based approach for large-scale 3D scene\ngeneration that enables controllable generation and effortless editing.\nSceneFactor enables text-guided 3D scene synthesis through our factored\ndiffusion formulation, leveraging latent semantic and geometric manifolds for\ngeneration of arbitrary-sized 3D scenes. While text input enables easy,\ncontrollable generation, text guidance remains imprecise for intuitive,\nlocalized editing and manipulation of the generated 3D scenes. Our factored\nsemantic diffusion generates a proxy semantic space composed of semantic 3D\nboxes that enables controllable editing of generated scenes by adding,\nremoving, changing the size of the semantic 3D proxy boxes that guides\nhigh-fidelity, consistent 3D geometric editing. Extensive experiments\ndemonstrate that our approach enables high-fidelity 3D scene synthesis with\neffective controllable editing through our factored diffusion approach.\n", "date": "2024-12-02", "categories": []} {"id": "2412.02202", "title": "3D representation in 512-Byte:Variational tokenizer is the key for\n autoregressive 3D generation", "abstract": " Autoregressive transformers have revolutionized high-fidelity image\ngeneration. One crucial ingredient lies in the tokenizer, which compresses\nhigh-resolution image patches into manageable discrete tokens with a scanning\nor hierarchical order suitable for large language models. Extending these\ntokenizers to 3D generation, however, presents a significant challenge: unlike\nimage patches that naturally exhibit spatial sequence and multi-scale\nrelationships, 3D data lacks an inherent order, making it difficult to compress\ninto fewer tokens while preserving structural details. To address this, we\nintroduce the Variational Tokenizer (VAT), which transforms unordered 3D data\ninto compact latent tokens with an implicit hierarchy, suited for efficient and\nhigh-fidelity coarse-to-fine autoregressive modeling. VAT begins with an\nin-context transformer, which compress numerous unordered 3D features into a\nreduced token set with minimal information loss. This latent space is then\nmapped to a Gaussian distribution for residual quantization, with token counts\nprogressively increasing across scales. In this way, tokens at different scales\nnaturally establish the interconnections by allocating themselves into\ndifferent subspaces within the same Gaussian distribution, facilitating\ndiscrete modeling of token relationships across scales. During the decoding\nphase, a high-resolution triplane is utilized to convert these compact latent\ntokens into detailed 3D shapes. Extensive experiments demonstrate that VAT\nenables scalable and efficient 3D generation, outperforming existing methods in\nquality, efficiency, and generalization. Remarkably, VAT achieves up to a 250x\ncompression, reducing a 1MB mesh to just 3.9KB with a 96% F-score, and can\nfurther compress to 256 int8 tokens, achieving a 2000x reduction while\nmaintaining a 92% F-score.\n", "date": "2024-12-03", "categories": []} {"id": "2412.02631", "title": "Sharp-It: A Multi-view to Multi-view Diffusion Model for 3D Synthesis\n and Manipulation", "abstract": " Advancements in text-to-image diffusion models have led to significant\nprogress in fast 3D content creation. One common approach is to generate a set\nof multi-view images of an object, and then reconstruct it into a 3D model.\nHowever, this approach bypasses the use of a native 3D representation of the\nobject and is hence prone to geometric artifacts and limited in controllability\nand manipulation capabilities. An alternative approach involves native 3D\ngenerative models that directly produce 3D representations. These models,\nhowever, are typically limited in their resolution, resulting in lower quality\n3D objects. In this work, we bridge the quality gap between methods that\ndirectly generate 3D representations and ones that reconstruct 3D objects from\nmulti-view images. We introduce a multi-view to multi-view diffusion model\ncalled Sharp-It, which takes a 3D consistent set of multi-view images rendered\nfrom a low-quality object and enriches its geometric details and texture. The\ndiffusion model operates on the multi-view set in parallel, in the sense that\nit shares features across the generated views. A high-quality 3D model can then\nbe reconstructed from the enriched multi-view set. By leveraging the advantages\nof both 2D and 3D approaches, our method offers an efficient and controllable\nmethod for high-quality 3D content creation. We demonstrate that Sharp-It\nenables various 3D applications, such as fast synthesis, editing, and\ncontrolled generation, while attaining high-quality assets.\n", "date": "2024-12-03", "categories": []} {"id": "2408.02555", "title": "MeshAnything V2: Artist-Created Mesh Generation With Adjacent Mesh\n Tokenization", "abstract": " Meshes are the de facto 3D representation in the industry but are\nlabor-intensive to produce. Recently, a line of research has focused on\nautoregressively generating meshes. This approach processes meshes into a\nsequence composed of vertices and then generates them vertex by vertex, similar\nto how a language model generates text. These methods have achieved some\nsuccess but still struggle to generate complex meshes. One primary reason for\nthis limitation is their inefficient tokenization methods. To address this\nissue, we introduce MeshAnything V2, an advanced mesh generation model designed\nto create Artist-Created Meshes that align precisely with specified shapes. A\nkey innovation behind MeshAnything V2 is our novel Adjacent Mesh Tokenization\n(AMT) method. Unlike traditional approaches that represent each face using\nthree vertices, AMT optimizes this by employing a single vertex wherever\nfeasible, effectively reducing the token sequence length by about half on\naverage. This not only streamlines the tokenization process but also results in\nmore compact and well-structured sequences, enhancing the efficiency of mesh\ngeneration. With these improvements, MeshAnything V2 effectively doubles the\nface limit compared to previous models, delivering superior performance without\nincreasing computational costs. We will make our code and models publicly\navailable. Project Page: https://buaacyw.github.io/meshanything-v2/\n", "date": "2024-08-05", "categories": []} {"id": "2412.00157", "title": "AerialGo: Walking-through City View Generation from Aerial Perspectives", "abstract": " High-quality 3D urban reconstruction is essential for applications in urban\nplanning, navigation, and AR/VR. However, capturing detailed ground-level data\nacross cities is both labor-intensive and raises significant privacy concerns\nrelated to sensitive information, such as vehicle plates, faces, and other\npersonal identifiers. To address these challenges, we propose AerialGo, a novel\nframework that generates realistic walking-through city views from aerial\nimages, leveraging multi-view diffusion models to achieve scalable,\nphotorealistic urban reconstructions without direct ground-level data\ncollection. By conditioning ground-view synthesis on accessible aerial data,\nAerialGo bypasses the privacy risks inherent in ground-level imagery. To\nsupport the model training, we introduce AerialGo dataset, a large-scale\ndataset containing diverse aerial and ground-view images, paired with camera\nand depth information, designed to support generative urban reconstruction.\nExperiments show that AerialGo significantly enhances ground-level realism and\nstructural coherence, providing a privacy-conscious, scalable solution for\ncity-scale 3D modeling.\n", "date": "2024-11-29", "categories": []} {"id": "2412.00518", "title": "Instant3dit: Multiview Inpainting for Fast Editing of 3D Objects", "abstract": " We propose a generative technique to edit 3D shapes, represented as meshes,\nNeRFs, or Gaussian Splats, in approximately 3 seconds, without the need for\nrunning an SDS type of optimization. Our key insight is to cast 3D editing as a\nmultiview image inpainting problem, as this representation is generic and can\nbe mapped back to any 3D representation using the bank of available Large\nReconstruction Models. We explore different fine-tuning strategies to obtain\nboth multiview generation and inpainting capabilities within the same diffusion\nmodel. In particular, the design of the inpainting mask is an important factor\nof training an inpainting model, and we propose several masking strategies to\nmimic the types of edits a user would perform on a 3D shape. Our approach takes\n3D generative editing from hours to seconds and produces higher-quality results\ncompared to previous works.\n", "date": "2024-11-30", "categories": []} {"id": "2412.00754", "title": "CtrlNeRF: The Generative Neural Radiation Fields for the Controllable\n Synthesis of High-fidelity 3D-Aware Images", "abstract": " The neural radiance field (NERF) advocates learning the continuous\nrepresentation of 3D geometry through a multilayer perceptron (MLP). By\nintegrating this into a generative model, the generative neural radiance field\n(GRAF) is capable of producing images from random noise z without 3D\nsupervision. In practice, the shape and appearance are modeled by z_s and z_a,\nrespectively, to manipulate them separately during inference. However, it is\nchallenging to represent multiple scenes using a solitary MLP and precisely\ncontrol the generation of 3D geometry in terms of shape and appearance. In this\npaper, we introduce a controllable generative model (i.e. \\textbf{CtrlNeRF})\nthat uses a single MLP network to represent multiple scenes with shared\nweights. Consequently, we manipulated the shape and appearance codes to realize\nthe controllable generation of high-fidelity images with 3D consistency.\nMoreover, the model enables the synthesis of novel views that do not exist in\nthe training sets via camera pose alteration and feature interpolation.\nExtensive experiments were conducted to demonstrate its superiority in 3D-aware\nimage generation compared to its counterparts.\n", "date": "2024-12-01", "categories": []} {"id": "2412.01792", "title": "CTRL-D: Controllable Dynamic 3D Scene Editing with Personalized 2D\n Diffusion", "abstract": " Recent advances in 3D representations, such as Neural Radiance Fields and 3D\nGaussian Splatting, have greatly improved realistic scene modeling and\nnovel-view synthesis. However, achieving controllable and consistent editing in\ndynamic 3D scenes remains a significant challenge. Previous work is largely\nconstrained by its editing backbones, resulting in inconsistent edits and\nlimited controllability. In our work, we introduce a novel framework that first\nfine-tunes the InstructPix2Pix model, followed by a two-stage optimization of\nthe scene based on deformable 3D Gaussians. Our fine-tuning enables the model\nto \"learn\" the editing ability from a single edited reference image,\ntransforming the complex task of dynamic scene editing into a simple 2D image\nediting process. By directly learning editing regions and styles from the\nreference, our approach enables consistent and precise local edits without the\nneed for tracking desired editing regions, effectively addressing key\nchallenges in dynamic scene editing. Then, our two-stage optimization\nprogressively edits the trained dynamic scene, using a designed edited image\nbuffer to accelerate convergence and improve temporal consistency. Compared to\nstate-of-the-art methods, our approach offers more flexible and controllable\nlocal scene editing, achieving high-quality and consistent results.\n", "date": "2024-12-02", "categories": []} {"id": "2412.01821", "title": "World-consistent Video Diffusion with Explicit 3D Modeling", "abstract": " Recent advancements in diffusion models have set new benchmarks in image and\nvideo generation, enabling realistic visual synthesis across single- and\nmulti-frame contexts. However, these models still struggle with efficiently and\nexplicitly generating 3D-consistent content. To address this, we propose\nWorld-consistent Video Diffusion (WVD), a novel framework that incorporates\nexplicit 3D supervision using XYZ images, which encode global 3D coordinates\nfor each image pixel. More specifically, we train a diffusion transformer to\nlearn the joint distribution of RGB and XYZ frames. This approach supports\nmulti-task adaptability via a flexible inpainting strategy. For example, WVD\ncan estimate XYZ frames from ground-truth RGB or generate novel RGB frames\nusing XYZ projections along a specified camera trajectory. In doing so, WVD\nunifies tasks like single-image-to-3D generation, multi-view stereo, and\ncamera-controlled video generation. Our approach demonstrates competitive\nperformance across multiple benchmarks, providing a scalable solution for\n3D-consistent video and image generation with a single pretrained model.\n", "date": "2024-12-02", "categories": []} {"id": "2404.10625", "title": "Gaussian Splatting Decoder for 3D-aware Generative Adversarial Networks", "abstract": " NeRF-based 3D-aware Generative Adversarial Networks (GANs) like EG3D or\nGIRAFFE have shown very high rendering quality under large representational\nvariety. However, rendering with Neural Radiance Fields poses challenges for 3D\napplications: First, the significant computational demands of NeRF rendering\npreclude its use on low-power devices, such as mobiles and VR/AR headsets.\nSecond, implicit representations based on neural networks are difficult to\nincorporate into explicit 3D scenes, such as VR environments or video games. 3D\nGaussian Splatting (3DGS) overcomes these limitations by providing an explicit\n3D representation that can be rendered efficiently at high frame rates. In this\nwork, we present a novel approach that combines the high rendering quality of\nNeRF-based 3D-aware GANs with the flexibility and computational advantages of\n3DGS. By training a decoder that maps implicit NeRF representations to explicit\n3D Gaussian Splatting attributes, we can integrate the representational\ndiversity and quality of 3D GANs into the ecosystem of 3D Gaussian Splatting\nfor the first time. Additionally, our approach allows for a high resolution GAN\ninversion and real-time GAN editing with 3D Gaussian Splatting scenes. Project\npage: florian-barthel.github.io/gaussian_decoder\n", "date": "2024-04-16", "categories": []} {"id": "2404.18929", "title": "DGE: Direct Gaussian 3D Editing by Consistent Multi-view Editing", "abstract": " We consider the problem of editing 3D objects and scenes based on open-ended\nlanguage instructions. A common approach to this problem is to use a 2D image\ngenerator or editor to guide the 3D editing process, obviating the need for 3D\ndata. However, this process is often inefficient due to the need for iterative\nupdates of costly 3D representations, such as neural radiance fields, either\nthrough individual view edits or score distillation sampling. A major\ndisadvantage of this approach is the slow convergence caused by aggregating\ninconsistent information across views, as the guidance from 2D models is not\nmulti-view consistent. We thus introduce the Direct Gaussian Editor (DGE), a\nmethod that addresses these issues in two stages. First, we modify a given\nhigh-quality image editor like InstructPix2Pix to be multi-view consistent. To\ndo so, we propose a training-free approach that integrates cues from the 3D\ngeometry of the underlying scene. Second, given a multi-view consistent edited\nsequence of images, we directly and efficiently optimize the 3D representation,\nwhich is based on 3D Gaussian Splatting. Because it avoids incremental and\niterative edits, DGE is significantly more accurate and efficient than existing\napproaches and offers additional benefits, such as enabling selective editing\nof parts of the scene.\n", "date": "2024-04-29", "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": "2404.00345", "title": "MaGRITTe: Manipulative and Generative 3D Realization from Image, Topview\n and Text", "abstract": " The generation of 3D scenes from user-specified conditions offers a promising\navenue for alleviating the production burden in 3D applications. Previous\nstudies required significant effort to realize the desired scene, owing to\nlimited control conditions. We propose a method for controlling and generating\n3D scenes under multimodal conditions using partial images, layout information\nrepresented in the top view, and text prompts. Combining these conditions to\ngenerate a 3D scene involves the following significant difficulties: (1) the\ncreation of large datasets, (2) reflection on the interaction of multimodal\nconditions, and (3) domain dependence of the layout conditions. We decompose\nthe process of 3D scene generation into 2D image generation from the given\nconditions and 3D scene generation from 2D images. 2D image generation is\nachieved by fine-tuning a pretrained text-to-image model with a small\nartificial dataset of partial images and layouts, and 3D scene generation is\nachieved by layout-conditioned depth estimation and neural radiance fields\n(NeRF), thereby avoiding the creation of large datasets. The use of a common\nrepresentation of spatial information using 360-degree images allows for the\nconsideration of multimodal condition interactions and reduces the domain\ndependence of the layout control. The experimental results qualitatively and\nquantitatively demonstrated that the proposed method can generate 3D scenes in\ndiverse domains, from indoor to outdoor, according to multimodal conditions.\n", "date": "2024-03-30", "categories": []} {"id": "2405.11616", "title": "Era3D: High-Resolution Multiview Diffusion using Efficient Row-wise\n Attention", "abstract": " In this paper, we introduce Era3D, a novel multiview diffusion method that\ngenerates high-resolution multiview images from a single-view image. Despite\nsignificant advancements in multiview generation, existing methods still suffer\nfrom camera prior mismatch, inefficacy, and low resolution, resulting in\npoor-quality multiview images. Specifically, these methods assume that the\ninput images should comply with a predefined camera type, e.g. a perspective\ncamera with a fixed focal length, leading to distorted shapes when the\nassumption fails. Moreover, the full-image or dense multiview attention they\nemploy leads to an exponential explosion of computational complexity as image\nresolution increases, resulting in prohibitively expensive training costs. To\nbridge the gap between assumption and reality, Era3D first proposes a\ndiffusion-based camera prediction module to estimate the focal length and\nelevation of the input image, which allows our method to generate images\nwithout shape distortions. Furthermore, a simple but efficient attention layer,\nnamed row-wise attention, is used to enforce epipolar priors in the multiview\ndiffusion, facilitating efficient cross-view information fusion. Consequently,\ncompared with state-of-the-art methods, Era3D generates high-quality multiview\nimages with up to a 512*512 resolution while reducing computation complexity by\n12x times. Comprehensive experiments demonstrate that Era3D can reconstruct\nhigh-quality and detailed 3D meshes from diverse single-view input images,\nsignificantly outperforming baseline multiview diffusion methods. Project page:\nhttps://penghtyx.github.io/Era3D/.\n", "date": "2024-05-19", "categories": []} {"id": "2411.18548", "title": "PhyCAGE: Physically Plausible Compositional 3D Asset Generation from a\n Single Image", "abstract": " We present PhyCAGE, the first approach for physically plausible compositional\n3D asset generation from a single image. Given an input image, we first\ngenerate consistent multi-view images for components of the assets. These\nimages are then fitted with 3D Gaussian Splatting representations. To ensure\nthat the Gaussians representing objects are physically compatible with each\nother, we introduce a Physical Simulation-Enhanced Score Distillation Sampling\n(PSE-SDS) technique to further optimize the positions of the Gaussians. It is\nachieved by setting the gradient of the SDS loss as the initial velocity of the\nphysical simulation, allowing the simulator to act as a physics-guided\noptimizer that progressively corrects the Gaussians' positions to a physically\ncompatible state. Experimental results demonstrate that the proposed method can\ngenerate physically plausible compositional 3D assets given a single image.\n", "date": "2024-11-27", "categories": []} {"id": "2311.11221", "title": "GaussianDiffusion: 3D Gaussian Splatting for Denoising Diffusion\n Probabilistic Models with Structured Noise", "abstract": " Text-to-3D, known for its efficient generation methods and expansive creative\npotential, has garnered significant attention in the AIGC domain. However, the\npixel-wise rendering of NeRF and its ray marching light sampling constrain the\nrendering speed, impacting its utility in downstream industrial applications.\nGaussian Splatting has recently shown a trend of replacing the traditional\npointwise sampling technique commonly used in NeRF-based methodologies, and it\nis changing various aspects of 3D reconstruction. This paper introduces a novel\ntext to 3D content generation framework, Gaussian Diffusion, based on Gaussian\nSplatting and produces more realistic renderings. The challenge of achieving\nmulti-view consistency in 3D generation significantly impedes modeling\ncomplexity and accuracy. Taking inspiration from SJC, we explore employing\nmulti-view noise distributions to perturb images generated by 3D Gaussian\nSplatting, aiming to rectify inconsistencies in multi-view geometry. We\ningeniously devise an efficient method to generate noise that produces Gaussian\nnoise from diverse viewpoints, all originating from a shared noise source.\nFurthermore, vanilla 3D Gaussian-based generation tends to trap models in local\nminima, causing artifacts like floaters, burrs, or proliferative elements. To\nmitigate these issues, we propose the variational Gaussian Splatting technique\nto enhance the quality and stability of 3D appearance. To our knowledge, our\napproach represents the first comprehensive utilization of Gaussian Diffusion\nacross the entire spectrum of 3D content generation processes.\n", "date": "2023-11-19", "categories": []} {"id": "2411.15706", "title": "Fixing the Perspective: A Critical Examination of Zero-1-to-3", "abstract": " Novel view synthesis is a fundamental challenge in image-to-3D generation,\nrequiring the generation of target view images from a set of conditioning\nimages and their relative poses. While recent approaches like Zero-1-to-3 have\ndemonstrated promising results using conditional latent diffusion models, they\nface significant challenges in generating consistent and accurate novel views,\nparticularly when handling multiple conditioning images. In this work, we\nconduct a thorough investigation of Zero-1-to-3's cross-attention mechanism\nwithin the Spatial Transformer of the diffusion 2D-conditional UNet. Our\nanalysis reveals a critical discrepancy between Zero-1-to-3's theoretical\nframework and its implementation, specifically in the processing of\nimage-conditional context. We propose two significant improvements: (1) a\ncorrected implementation that enables effective utilization of the\ncross-attention mechanism, and (2) an enhanced architecture that can leverage\nmultiple conditional views simultaneously. Our theoretical analysis and\npreliminary results suggest potential improvements in novel view synthesis\nconsistency and accuracy.\n", "date": "2024-11-24", "categories": []} {"id": "2411.16080", "title": "Boosting 3D Object Generation through PBR Materials", "abstract": " Automatic 3D content creation has gained increasing attention recently, due\nto its potential in various applications such as video games, film industry,\nand AR/VR. Recent advancements in diffusion models and multimodal models have\nnotably improved the quality and efficiency of 3D object generation given a\nsingle RGB image. However, 3D objects generated even by state-of-the-art\nmethods are still unsatisfactory compared to human-created assets. Considering\nonly textures instead of materials makes these methods encounter challenges in\nphoto-realistic rendering, relighting, and flexible appearance editing. And\nthey also suffer from severe misalignment between geometry and high-frequency\ntexture details. In this work, we propose a novel approach to boost the quality\nof generated 3D objects from the perspective of Physics-Based Rendering (PBR)\nmaterials. By analyzing the components of PBR materials, we choose to consider\nalbedo, roughness, metalness, and bump maps. For albedo and bump maps, we\nleverage Stable Diffusion fine-tuned on synthetic data to extract these values,\nwith novel usages of these fine-tuned models to obtain 3D consistent albedo UV\nand bump UV for generated objects. In terms of roughness and metalness maps, we\nadopt a semi-automatic process to provide room for interactive adjustment,\nwhich we believe is more practical. Extensive experiments demonstrate that our\nmodel is generally beneficial for various state-of-the-art generation methods,\nsignificantly boosting the quality and realism of their generated 3D objects,\nwith natural relighting effects and substantially improved geometry.\n", "date": "2024-11-25", "categories": []} {"id": "2411.14715", "title": "Any-to-3D Generation via Hybrid Diffusion Supervision", "abstract": " Recent progress in 3D object generation has been fueled by the strong priors\noffered by diffusion models. However, existing models are tailored to specific\ntasks, accommodating only one modality at a time and necessitating retraining\nto change modalities. Given an image-to-3D model and a text prompt, a naive\napproach is to convert text prompts to images and then use the image-to-3D\nmodel for generation. This approach is both time-consuming and labor-intensive,\nresulting in unavoidable information loss during modality conversion. To\naddress this, we introduce XBind, a unified framework for any-to-3D generation\nusing cross-modal pre-alignment techniques. XBind integrates an\nmultimodal-aligned encoder with pre-trained diffusion models to generate 3D\nobjects from any modalities, including text, images, and audio. We subsequently\npresent a novel loss function, termed Modality Similarity (MS) Loss, which\naligns the embeddings of the modality prompts and the rendered images,\nfacilitating improved alignment of the 3D objects with multiple modalities.\nAdditionally, Hybrid Diffusion Supervision combined with a Three-Phase\nOptimization process improves the quality of the generated 3D objects.\nExtensive experiments showcase XBind's broad generation capabilities in\nany-to-3D scenarios. To our knowledge, this is the first method to generate 3D\nobjects from any modality prompts. Project page:\nhttps://zeroooooooow1440.github.io/.\n", "date": "2024-11-22", "categories": []} {"id": "2411.14740", "title": "TEXGen: a Generative Diffusion Model for Mesh Textures", "abstract": " While high-quality texture maps are essential for realistic 3D asset\nrendering, few studies have explored learning directly in the texture space,\nespecially on large-scale datasets. In this work, we depart from the\nconventional approach of relying on pre-trained 2D diffusion models for\ntest-time optimization of 3D textures. Instead, we focus on the fundamental\nproblem of learning in the UV texture space itself. For the first time, we\ntrain a large diffusion model capable of directly generating high-resolution\ntexture maps in a feed-forward manner. To facilitate efficient learning in\nhigh-resolution UV spaces, we propose a scalable network architecture that\ninterleaves convolutions on UV maps with attention layers on point clouds.\nLeveraging this architectural design, we train a 700 million parameter\ndiffusion model that can generate UV texture maps guided by text prompts and\nsingle-view images. Once trained, our model naturally supports various extended\napplications, including text-guided texture inpainting, sparse-view texture\ncompletion, and text-driven texture synthesis. Project page is at\nhttp://cvmi-lab.github.io/TEXGen/.\n", "date": "2024-11-22", "categories": []} {"id": "2411.15138", "title": "Material Anything: Generating Materials for Any 3D Object via Diffusion", "abstract": " We present Material Anything, a fully-automated, unified diffusion framework\ndesigned to generate physically-based materials for 3D objects. Unlike existing\nmethods that rely on complex pipelines or case-specific optimizations, Material\nAnything offers a robust, end-to-end solution adaptable to objects under\ndiverse lighting conditions. Our approach leverages a pre-trained image\ndiffusion model, enhanced with a triple-head architecture and rendering loss to\nimprove stability and material quality. Additionally, we introduce confidence\nmasks as a dynamic switcher within the diffusion model, enabling it to\neffectively handle both textured and texture-less objects across varying\nlighting conditions. By employing a progressive material generation strategy\nguided by these confidence masks, along with a UV-space material refiner, our\nmethod ensures consistent, UV-ready material outputs. Extensive experiments\ndemonstrate our approach outperforms existing methods across a wide range of\nobject categories and lighting conditions.\n", "date": "2024-11-22", "categories": []} {"id": "2410.09467", "title": "Enhancing Single Image to 3D Generation using Gaussian Splatting and\n Hybrid Diffusion Priors", "abstract": " 3D object generation from a single image involves estimating the full 3D\ngeometry and texture of unseen views from an unposed RGB image captured in the\nwild. Accurately reconstructing an object's complete 3D structure and texture\nhas numerous applications in real-world scenarios, including robotic\nmanipulation, grasping, 3D scene understanding, and AR/VR. Recent advancements\nin 3D object generation have introduced techniques that reconstruct an object's\n3D shape and texture by optimizing the efficient representation of Gaussian\nSplatting, guided by pre-trained 2D or 3D diffusion models. However, a notable\ndisparity exists between the training datasets of these models, leading to\ndistinct differences in their outputs. While 2D models generate highly detailed\nvisuals, they lack cross-view consistency in geometry and texture. In contrast,\n3D models ensure consistency across different views but often result in overly\nsmooth textures. We propose bridging the gap between 2D and 3D diffusion models\nto address this limitation by integrating a two-stage frequency-based\ndistillation loss with Gaussian Splatting. Specifically, we leverage geometric\npriors in the low-frequency spectrum from a 3D diffusion model to maintain\nconsistent geometry and use a 2D diffusion model to refine the fidelity and\ntexture in the high-frequency spectrum of the generated 3D structure, resulting\nin more detailed and fine-grained outcomes. Our approach enhances geometric\nconsistency and visual quality, outperforming the current SOTA. Additionally,\nwe demonstrate the easy adaptability of our method for efficient object pose\nestimation and tracking.\n", "date": "2024-10-12", "categories": []} {"id": "2405.19957", "title": "PLA4D: Pixel-Level Alignments for Text-to-4D Gaussian Splatting", "abstract": " Previous text-to-4D methods have leveraged multiple Score Distillation\nSampling (SDS) techniques, combining motion priors from video-based diffusion\nmodels (DMs) with geometric priors from multiview DMs to implicitly guide 4D\nrenderings. However, differences in these priors result in conflicting gradient\ndirections during optimization, causing trade-offs between motion fidelity and\ngeometry accuracy, and requiring substantial optimization time to reconcile the\nmodels. In this paper, we introduce \\textbf{P}ixel-\\textbf{L}evel\n\\textbf{A}lignment for text-driven \\textbf{4D} Gaussian splatting (PLA4D) to\nresolve this motion-geometry conflict. PLA4D provides an anchor reference,\ni.e., text-generated video, to align the rendering process conditioned by\ndifferent DMs in pixel space. For static alignment, our approach introduces a\nfocal alignment method and Gaussian-Mesh contrastive learning to iteratively\nadjust focal lengths and provide explicit geometric priors at each timestep. At\nthe dynamic level, a motion alignment technique and T-MV refinement method are\nemployed to enforce both pose alignment and motion continuity across unknown\nviewpoints, ensuring intrinsic geometric consistency across views. With such\npixel-level multi-DM alignment, our PLA4D framework is able to generate 4D\nobjects with superior geometric, motion, and semantic consistency. Fully\nimplemented with open-source tools, PLA4D offers an efficient and accessible\nsolution for high-quality 4D digital content creation with significantly\nreduced generation time.\n", "date": "2024-05-30", "categories": []} {"id": "2408.08000", "title": "MVInpainter: Learning Multi-View Consistent Inpainting to Bridge 2D and\n 3D Editing", "abstract": " Novel View Synthesis (NVS) and 3D generation have recently achieved prominent\nimprovements. However, these works mainly focus on confined categories or\nsynthetic 3D assets, which are discouraged from generalizing to challenging\nin-the-wild scenes and fail to be employed with 2D synthesis directly.\nMoreover, these methods heavily depended on camera poses, limiting their\nreal-world applications. To overcome these issues, we propose MVInpainter,\nre-formulating the 3D editing as a multi-view 2D inpainting task. Specifically,\nMVInpainter partially inpaints multi-view images with the reference guidance\nrather than intractably generating an entirely novel view from scratch, which\nlargely simplifies the difficulty of in-the-wild NVS and leverages unmasked\nclues instead of explicit pose conditions. To ensure cross-view consistency,\nMVInpainter is enhanced by video priors from motion components and appearance\nguidance from concatenated reference key&value attention. Furthermore,\nMVInpainter incorporates slot attention to aggregate high-level optical flow\nfeatures from unmasked regions to control the camera movement with pose-free\ntraining and inference. Sufficient scene-level experiments on both\nobject-centric and forward-facing datasets verify the effectiveness of\nMVInpainter, including diverse tasks, such as multi-view object removal,\nsynthesis, insertion, and replacement. The project page is\nhttps://ewrfcas.github.io/MVInpainter/.\n", "date": "2024-08-15", "categories": []} {"id": "2411.12197", "title": "MTFusion: Reconstructing Any 3D Object from Single Image Using\n Multi-word Textual Inversion", "abstract": " Reconstructing 3D models from single-view images is a long-standing problem\nin computer vision. The latest advances for single-image 3D reconstruction\nextract a textual description from the input image and further utilize it to\nsynthesize 3D models. However, existing methods focus on capturing a single key\nattribute of the image (e.g., object type, artistic style) and fail to consider\nthe multi-perspective information required for accurate 3D reconstruction, such\nas object shape and material properties. Besides, the reliance on Neural\nRadiance Fields hinders their ability to reconstruct intricate surfaces and\ntexture details. In this work, we propose MTFusion, which leverages both image\ndata and textual descriptions for high-fidelity 3D reconstruction. Our approach\nconsists of two stages. First, we adopt a novel multi-word textual inversion\ntechnique to extract a detailed text description capturing the image's\ncharacteristics. Then, we use this description and the image to generate a 3D\nmodel with FlexiCubes. Additionally, MTFusion enhances FlexiCubes by employing\na special decoder network for Signed Distance Functions, leading to faster\ntraining and finer surface representation. Extensive evaluations demonstrate\nthat our MTFusion surpasses existing image-to-3D methods on a wide range of\nsynthetic and real-world images. Furthermore, the ablation study proves the\neffectiveness of our network designs.\n", "date": "2024-11-19", "categories": []} {"id": "2411.12290", "title": "SSEditor: Controllable Mask-to-Scene Generation with Diffusion Model", "abstract": " Recent advancements in 3D diffusion-based semantic scene generation have\ngained attention. However, existing methods rely on unconditional generation\nand require multiple resampling steps when editing scenes, which significantly\nlimits their controllability and flexibility. To this end, we propose SSEditor,\na controllable Semantic Scene Editor that can generate specified target\ncategories without multiple-step resampling. SSEditor employs a two-stage\ndiffusion-based framework: (1) a 3D scene autoencoder is trained to obtain\nlatent triplane features, and (2) a mask-conditional diffusion model is trained\nfor customizable 3D semantic scene generation. In the second stage, we\nintroduce a geometric-semantic fusion module that enhance the model's ability\nto learn geometric and semantic information. This ensures that objects are\ngenerated with correct positions, sizes, and categories. Extensive experiments\non SemanticKITTI and CarlaSC demonstrate that SSEditor outperforms previous\napproaches in terms of controllability and flexibility in target generation, as\nwell as the quality of semantic scene generation and reconstruction. More\nimportantly, experiments on the unseen Occ-3D Waymo dataset show that SSEditor\nis capable of generating novel urban scenes, enabling the rapid construction of\n3D scenes.\n", "date": "2024-11-19", "categories": []} {"id": "2403.11273", "title": "BrightDreamer: Generic 3D Gaussian Generative Framework for Fast\n Text-to-3D Synthesis", "abstract": " Text-to-3D synthesis has recently seen intriguing advances by combining the\ntext-to-image priors with 3D representation methods, e.g., 3D Gaussian\nSplatting (3D GS), via Score Distillation Sampling (SDS). However, a hurdle of\nexisting methods is the low efficiency, per-prompt optimization for a single 3D\nobject. Therefore, it is imperative for a paradigm shift from per-prompt\noptimization to feed-forward generation for any unseen text prompts, which yet\nremains challenging. An obstacle is how to directly generate a set of millions\nof 3D Gaussians to represent a 3D object. This paper presents BrightDreamer, an\nend-to-end feed-forward approach that can achieve generalizable and fast (77\nms) text-to-3D generation. Our key idea is to formulate the generation process\nas estimating the 3D deformation from an anchor shape with predefined\npositions. For this, we first propose a Text-guided Shape Deformation (TSD)\nnetwork to predict the deformed shape and its new positions, used as the\ncenters (one attribute) of 3D Gaussians. To estimate the other four attributes\n(i.e., scaling, rotation, opacity, and SH), we then design a novel Text-guided\nTriplane Generator (TTG) to generate a triplane representation for a 3D object.\nThe center of each Gaussian enables us to transform the spatial feature into\nthe four attributes. The generated 3D Gaussians can be finally rendered at 705\nframes per second. Extensive experiments demonstrate the superiority of our\nmethod over existing methods. Also, BrightDreamer possesses a strong semantic\nunderstanding capability even for complex text prompts. The code is available\nin the project page.\n", "date": "2024-03-17", "categories": []} {"id": "2405.18515", "title": "Atlas3D: Physically Constrained Self-Supporting Text-to-3D for\n Simulation and Fabrication", "abstract": " Existing diffusion-based text-to-3D generation methods primarily focus on\nproducing visually realistic shapes and appearances, often neglecting the\nphysical constraints necessary for downstream tasks. Generated models\nfrequently fail to maintain balance when placed in physics-based simulations or\n3D printed. This balance is crucial for satisfying user design intentions in\ninteractive gaming, embodied AI, and robotics, where stable models are needed\nfor reliable interaction. Additionally, stable models ensure that 3D-printed\nobjects, such as figurines for home decoration, can stand on their own without\nrequiring additional supports. To fill this gap, we introduce Atlas3D, an\nautomatic and easy-to-implement method that enhances existing Score\nDistillation Sampling (SDS)-based text-to-3D tools. Atlas3D ensures the\ngeneration of self-supporting 3D models that adhere to physical laws of\nstability under gravity, contact, and friction. Our approach combines a novel\ndifferentiable simulation-based loss function with physically inspired\nregularization, serving as either a refinement or a post-processing module for\nexisting frameworks. We verify Atlas3D's efficacy through extensive generation\ntasks and validate the resulting 3D models in both simulated and real-world\nenvironments.\n", "date": "2024-05-28", "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": "2411.10947", "title": "Direct and Explicit 3D Generation from a Single Image", "abstract": " Current image-to-3D approaches suffer from high computational costs and lack\nscalability for high-resolution outputs. In contrast, we introduce a novel\nframework to directly generate explicit surface geometry and texture using\nmulti-view 2D depth and RGB images along with 3D Gaussian features using a\nrepurposed Stable Diffusion model. We introduce a depth branch into U-Net for\nefficient and high quality multi-view, cross-domain generation and incorporate\nepipolar attention into the latent-to-pixel decoder for pixel-level multi-view\nconsistency. By back-projecting the generated depth pixels into 3D space, we\ncreate a structured 3D representation that can be either rendered via Gaussian\nsplatting or extracted to high-quality meshes, thereby leveraging additional\nnovel view synthesis loss to further improve our performance. Extensive\nexperiments demonstrate that our method surpasses existing baselines in\ngeometry and texture quality while achieving significantly faster generation\ntime.\n", "date": "2024-11-17", "categories": []} {"id": "2411.11475", "title": "MVLight: Relightable Text-to-3D Generation via Light-conditioned\n Multi-View Diffusion", "abstract": " Recent advancements in text-to-3D generation, building on the success of\nhigh-performance text-to-image generative models, have made it possible to\ncreate imaginative and richly textured 3D objects from textual descriptions.\nHowever, a key challenge remains in effectively decoupling light-independent\nand lighting-dependent components to enhance the quality of generated 3D models\nand their relighting performance. In this paper, we present MVLight, a novel\nlight-conditioned multi-view diffusion model that explicitly integrates\nlighting conditions directly into the generation process. This enables the\nmodel to synthesize high-quality images that faithfully reflect the specified\nlighting environment across multiple camera views. By leveraging this\ncapability to Score Distillation Sampling (SDS), we can effectively synthesize\n3D models with improved geometric precision and relighting capabilities. We\nvalidate the effectiveness of MVLight through extensive experiments and a user\nstudy.\n", "date": "2024-11-18", "categories": []} {"id": "2308.11408", "title": "MatFuse: Controllable Material Generation with Diffusion Models", "abstract": " Creating high-quality materials in computer graphics is a challenging and\ntime-consuming task, which requires great expertise. To simplify this process,\nwe introduce MatFuse, a unified approach that harnesses the generative power of\ndiffusion models for creation and editing of 3D materials. Our method\nintegrates multiple sources of conditioning, including color palettes,\nsketches, text, and pictures, enhancing creative possibilities and granting\nfine-grained control over material synthesis. Additionally, MatFuse enables\nmap-level material editing capabilities through latent manipulation by means of\na multi-encoder compression model which learns a disentangled latent\nrepresentation for each map. We demonstrate the effectiveness of MatFuse under\nmultiple conditioning settings and explore the potential of material editing.\nFinally, we assess the quality of the generated materials both quantitatively\nin terms of CLIP-IQA and FID scores and qualitatively by conducting a user\nstudy. Source code for training MatFuse and supplemental materials are publicly\navailable at https://gvecchio.com/matfuse.\n", "date": "2023-08-22", "categories": []} {"id": "2406.02968", "title": "GSGAN: Adversarial Learning for Hierarchical Generation of 3D Gaussian\n Splats", "abstract": " Most advances in 3D Generative Adversarial Networks (3D GANs) largely depend\non ray casting-based volume rendering, which incurs demanding rendering costs.\nOne promising alternative is rasterization-based 3D Gaussian Splatting (3D-GS),\nproviding a much faster rendering speed and explicit 3D representation. In this\npaper, we exploit Gaussian as a 3D representation for 3D GANs by leveraging its\nefficient and explicit characteristics. However, in an adversarial framework,\nwe observe that a na\\\"ive generator architecture suffers from training\ninstability and lacks the capability to adjust the scale of Gaussians. This\nleads to model divergence and visual artifacts due to the absence of proper\nguidance for initialized positions of Gaussians and densification to manage\ntheir scales adaptively. To address these issues, we introduce a generator\narchitecture with a hierarchical multi-scale Gaussian representation that\neffectively regularizes the position and scale of generated Gaussians.\nSpecifically, we design a hierarchy of Gaussians where finer-level Gaussians\nare parameterized by their coarser-level counterparts; the position of\nfiner-level Gaussians would be located near their coarser-level counterparts,\nand the scale would monotonically decrease as the level becomes finer, modeling\nboth coarse and fine details of the 3D scene. Experimental results demonstrate\nthat ours achieves a significantly faster rendering speed (x100) compared to\nstate-of-the-art 3D consistent GANs with comparable 3D generation capability.\nProject page: https://hse1032.github.io/gsgan.\n", "date": "2024-06-05", "categories": []} {"id": "2411.09823", "title": "Architect: Generating Vivid and Interactive 3D Scenes with Hierarchical\n 2D Inpainting", "abstract": " Creating large-scale interactive 3D environments is essential for the\ndevelopment of Robotics and Embodied AI research. Current methods, including\nmanual design, procedural generation, diffusion-based scene generation, and\nlarge language model (LLM) guided scene design, are hindered by limitations\nsuch as excessive human effort, reliance on predefined rules or training\ndatasets, and limited 3D spatial reasoning ability. Since pre-trained 2D image\ngenerative models better capture scene and object configuration than LLMs, we\naddress these challenges by introducing Architect, a generative framework that\ncreates complex and realistic 3D embodied environments leveraging\ndiffusion-based 2D image inpainting. In detail, we utilize foundation visual\nperception models to obtain each generated object from the image and leverage\npre-trained depth estimation models to lift the generated 2D image to 3D space.\nOur pipeline is further extended to a hierarchical and iterative inpainting\nprocess to continuously generate placement of large furniture and small objects\nto enrich the scene. This iterative structure brings the flexibility for our\nmethod to generate or refine scenes from various starting points, such as text,\nfloor plans, or pre-arranged environments.\n", "date": "2024-11-14", "categories": []} {"id": "2411.10369", "title": "Towards High-Fidelity 3D Portrait Generation with Rich Details by\n Cross-View Prior-Aware Diffusion", "abstract": " Recent diffusion-based Single-image 3D portrait generation methods typically\nemploy 2D diffusion models to provide multi-view knowledge, which is then\ndistilled into 3D representations. However, these methods usually struggle to\nproduce high-fidelity 3D models, frequently yielding excessively blurred\ntextures. We attribute this issue to the insufficient consideration of\ncross-view consistency during the diffusion process, resulting in significant\ndisparities between different views and ultimately leading to blurred 3D\nrepresentations. In this paper, we address this issue by comprehensively\nexploiting multi-view priors in both the conditioning and diffusion procedures\nto produce consistent, detail-rich portraits. From the conditioning standpoint,\nwe propose a Hybrid Priors Diffsion model, which explicitly and implicitly\nincorporates multi-view priors as conditions to enhance the status consistency\nof the generated multi-view portraits. From the diffusion perspective,\nconsidering the significant impact of the diffusion noise distribution on\ndetailed texture generation, we propose a Multi-View Noise Resamplig Strategy\nintegrated within the optimization process leveraging cross-view priors to\nenhance representation consistency. Extensive experiments demonstrate that our\nmethod can produce 3D portraits with accurate geometry and rich details from a\nsingle image. The project page is at\n\\url{https://haoran-wei.github.io/Portrait-Diffusion}.\n", "date": "2024-11-15", "categories": []} {"id": "2411.09595", "title": "LLaMA-Mesh: Unifying 3D Mesh Generation with Language Models", "abstract": " This work explores expanding the capabilities of large language models (LLMs)\npretrained on text to generate 3D meshes within a unified model. This offers\nkey advantages of (1) leveraging spatial knowledge already embedded in LLMs,\nderived from textual sources like 3D tutorials, and (2) enabling conversational\n3D generation and mesh understanding. A primary challenge is effectively\ntokenizing 3D mesh data into discrete tokens that LLMs can process seamlessly.\nTo address this, we introduce LLaMA-Mesh, a novel approach that represents the\nvertex coordinates and face definitions of 3D meshes as plain text, allowing\ndirect integration with LLMs without expanding the vocabulary. We construct a\nsupervised fine-tuning (SFT) dataset enabling pretrained LLMs to (1) generate\n3D meshes from text prompts, (2) produce interleaved text and 3D mesh outputs\nas required, and (3) understand and interpret 3D meshes. Our work is the first\nto demonstrate that LLMs can be fine-tuned to acquire complex spatial knowledge\nfor 3D mesh generation in a text-based format, effectively unifying the 3D and\ntext modalities. LLaMA-Mesh achieves mesh generation quality on par with models\ntrained from scratch while maintaining strong text generation performance.\n", "date": "2024-11-14", "categories": []} {"id": "2411.08017", "title": "Wavelet Latent Diffusion (Wala): Billion-Parameter 3D Generative Model\n with Compact Wavelet Encodings", "abstract": " Large-scale 3D generative models require substantial computational resources\nyet often fall short in capturing fine details and complex geometries at high\nresolutions. We attribute this limitation to the inefficiency of current\nrepresentations, which lack the compactness required to model the generative\nmodels effectively. To address this, we introduce a novel approach called\nWavelet Latent Diffusion, or WaLa, that encodes 3D shapes into wavelet-based,\ncompact latent encodings. Specifically, we compress a $256^3$ signed distance\nfield into a $12^3 \\times 4$ latent grid, achieving an impressive 2427x\ncompression ratio with minimal loss of detail. This high level of compression\nallows our method to efficiently train large-scale generative networks without\nincreasing the inference time. Our models, both conditional and unconditional,\ncontain approximately one billion parameters and successfully generate\nhigh-quality 3D shapes at $256^3$ resolution. Moreover, WaLa offers rapid\ninference, producing shapes within two to four seconds depending on the\ncondition, despite the model's scale. We demonstrate state-of-the-art\nperformance across multiple datasets, with significant improvements in\ngeneration quality, diversity, and computational efficiency. We open-source our\ncode and, to the best of our knowledge, release the largest pretrained 3D\ngenerative models across different modalities.\n", "date": "2024-11-12", "categories": []} {"id": "2411.06206", "title": "Text2CAD: Text to 3D CAD Generation via Technical Drawings", "abstract": " The generation of industrial Computer-Aided Design (CAD) models from user\nrequests and specifications is crucial to enhancing efficiency in modern\nmanufacturing. Traditional methods of CAD generation rely heavily on manual\ninputs and struggle with complex or non-standard designs, making them less\nsuited for dynamic industrial needs. To overcome these challenges, we introduce\nText2CAD, a novel framework that employs stable diffusion models tailored to\nautomate the generation process and efficiently bridge the gap between user\nspecifications in text and functional CAD models. This approach directly\ntranslates the user's textural descriptions into detailed isometric images,\nwhich are then precisely converted into orthographic views, e.g., top, front,\nand side, providing sufficient information to reconstruct 3D CAD models. This\nprocess not only streamlines the creation of CAD models from textual\ndescriptions but also ensures that the resulting models uphold physical and\ndimensional consistency essential for practical engineering applications. Our\nexperimental results show that Text2CAD effectively generates technical\ndrawings that are accurately translated into high-quality 3D CAD models,\nshowing substantial potential to revolutionize CAD automation in response to\nuser demands.\n", "date": "2024-11-09", "categories": []} {"id": "2411.07025", "title": "Scaling Mesh Generation via Compressive Tokenization", "abstract": " We propose a compressive yet effective mesh representation, Blocked and\nPatchified Tokenization (BPT), facilitating the generation of meshes exceeding\n8k faces. BPT compresses mesh sequences by employing block-wise indexing and\npatch aggregation, reducing their length by approximately 75\\% compared to the\noriginal sequences. This compression milestone unlocks the potential to utilize\nmesh data with significantly more faces, thereby enhancing detail richness and\nimproving generation robustness. Empowered with the BPT, we have built a\nfoundation mesh generative model training on scaled mesh data to support\nflexible control for point clouds and images. Our model demonstrates the\ncapability to generate meshes with intricate details and accurate topology,\nachieving SoTA performance on mesh generation and reaching the level for direct\nproduct usage.\n", "date": "2024-11-11", "categories": []} {"id": "2411.07135", "title": "Edify 3D: Scalable High-Quality 3D Asset Generation", "abstract": " We introduce Edify 3D, an advanced solution designed for high-quality 3D\nasset generation. Our method first synthesizes RGB and surface normal images of\nthe described object at multiple viewpoints using a diffusion model. The\nmulti-view observations are then used to reconstruct the shape, texture, and\nPBR materials of the object. Our method can generate high-quality 3D assets\nwith detailed geometry, clean shape topologies, high-resolution textures, and\nmaterials within 2 minutes of runtime.\n", "date": "2024-11-11", "categories": []} {"id": "2411.04928", "title": "DimensionX: Create Any 3D and 4D Scenes from a Single Image with\n Controllable Video Diffusion", "abstract": " In this paper, we introduce \\textbf{DimensionX}, a framework designed to\ngenerate photorealistic 3D and 4D scenes from just a single image with video\ndiffusion. Our approach begins with the insight that both the spatial structure\nof a 3D scene and the temporal evolution of a 4D scene can be effectively\nrepresented through sequences of video frames. While recent video diffusion\nmodels have shown remarkable success in producing vivid visuals, they face\nlimitations in directly recovering 3D/4D scenes due to limited spatial and\ntemporal controllability during generation. To overcome this, we propose\nST-Director, which decouples spatial and temporal factors in video diffusion by\nlearning dimension-aware LoRAs from dimension-variant data. This controllable\nvideo diffusion approach enables precise manipulation of spatial structure and\ntemporal dynamics, allowing us to reconstruct both 3D and 4D representations\nfrom sequential frames with the combination of spatial and temporal dimensions.\nAdditionally, to bridge the gap between generated videos and real-world scenes,\nwe introduce a trajectory-aware mechanism for 3D generation and an\nidentity-preserving denoising strategy for 4D generation. Extensive experiments\non various real-world and synthetic datasets demonstrate that DimensionX\nachieves superior results in controllable video generation, as well as in 3D\nand 4D scene generation, compared with previous methods.\n", "date": "2024-11-07", "categories": []} {"id": "2411.05006", "title": "ProEdit: Simple Progression is All You Need for High-Quality 3D Scene\n Editing", "abstract": " This paper proposes ProEdit - a simple yet effective framework for\nhigh-quality 3D scene editing guided by diffusion distillation in a novel\nprogressive manner. Inspired by the crucial observation that multi-view\ninconsistency in scene editing is rooted in the diffusion model's large\nfeasible output space (FOS), our framework controls the size of FOS and reduces\ninconsistency by decomposing the overall editing task into several subtasks,\nwhich are then executed progressively on the scene. Within this framework, we\ndesign a difficulty-aware subtask decomposition scheduler and an adaptive 3D\nGaussian splatting (3DGS) training strategy, ensuring high quality and\nefficiency in performing each subtask. Extensive evaluation shows that our\nProEdit achieves state-of-the-art results in various scenes and challenging\nediting tasks, all through a simple framework without any expensive or\nsophisticated add-ons like distillation losses, components, or training\nprocedures. Notably, ProEdit also provides a new way to control, preview, and\nselect the \"aggressivity\" of editing operation during the editing process.\n", "date": "2024-11-07", "categories": []} {"id": "2409.18336", "title": "DeBaRA: Denoising-Based 3D Room Arrangement Generation", "abstract": " Generating realistic and diverse layouts of furnished indoor 3D scenes\nunlocks multiple interactive applications impacting a wide range of industries.\nThe inherent complexity of object interactions, the limited amount of available\ndata and the requirement to fulfill spatial constraints all make generative\nmodeling for 3D scene synthesis and arrangement challenging. Current methods\naddress these challenges autoregressively or by using off-the-shelf diffusion\nobjectives by simultaneously predicting all attributes without 3D reasoning\nconsiderations. In this paper, we introduce DeBaRA, a score-based model\nspecifically tailored for precise, controllable and flexible arrangement\ngeneration in a bounded environment. We argue that the most critical component\nof a scene synthesis system is to accurately establish the size and position of\nvarious objects within a restricted area. Based on this insight, we propose a\nlightweight conditional score-based model designed with 3D spatial awareness at\nits core. We demonstrate that by focusing on spatial attributes of objects, a\nsingle trained DeBaRA model can be leveraged at test time to perform several\ndownstream applications such as scene synthesis, completion and re-arrangement.\nFurther, we introduce a novel Self Score Evaluation procedure so it can be\noptimally employed alongside external LLM models. We evaluate our approach\nthrough extensive experiments and demonstrate significant improvement upon\nstate-of-the-art approaches in a range of scenarios.\n", "date": "2024-09-26", "categories": []} {"id": "2411.02319", "title": "GenXD: Generating Any 3D and 4D Scenes", "abstract": " Recent developments in 2D visual generation have been remarkably successful.\nHowever, 3D and 4D generation remain challenging in real-world applications due\nto the lack of large-scale 4D data and effective model design. In this paper,\nwe propose to jointly investigate general 3D and 4D generation by leveraging\ncamera and object movements commonly observed in daily life. Due to the lack of\nreal-world 4D data in the community, we first propose a data curation pipeline\nto obtain camera poses and object motion strength from videos. Based on this\npipeline, we introduce a large-scale real-world 4D scene dataset: CamVid-30K.\nBy leveraging all the 3D and 4D data, we develop our framework, GenXD, which\nallows us to produce any 3D or 4D scene. We propose multiview-temporal modules,\nwhich disentangle camera and object movements, to seamlessly learn from both 3D\nand 4D data. Additionally, GenXD employs masked latent conditions to support a\nvariety of conditioning views. GenXD can generate videos that follow the camera\ntrajectory as well as consistent 3D views that can be lifted into 3D\nrepresentations. We perform extensive evaluations across various real-world and\nsynthetic datasets, demonstrating GenXD's effectiveness and versatility\ncompared to previous methods in 3D and 4D generation.\n", "date": "2024-11-04", "categories": []} {"id": "2401.15563", "title": "BrepGen: A B-rep Generative Diffusion Model with Structured Latent\n Geometry", "abstract": " This paper presents BrepGen, a diffusion-based generative approach that\ndirectly outputs a Boundary representation (B-rep) Computer-Aided Design (CAD)\nmodel. BrepGen represents a B-rep model as a novel structured latent geometry\nin a hierarchical tree. With the root node representing a whole CAD solid, each\nelement of a B-rep model (i.e., a face, an edge, or a vertex) progressively\nturns into a child-node from top to bottom. B-rep geometry information goes\ninto the nodes as the global bounding box of each primitive along with a latent\ncode describing the local geometric shape. The B-rep topology information is\nimplicitly represented by node duplication. When two faces share an edge, the\nedge curve will appear twice in the tree, and a T-junction vertex with three\nincident edges appears six times in the tree with identical node features.\nStarting from the root and progressing to the leaf, BrepGen employs\nTransformer-based diffusion models to sequentially denoise node features while\nduplicated nodes are detected and merged, recovering the B-Rep topology\ninformation. Extensive experiments show that BrepGen advances the task of CAD\nB-rep generation, surpassing existing methods on various benchmarks. Results on\nour newly collected furniture dataset further showcase its exceptional\ncapability in generating complicated geometry. While previous methods were\nlimited to generating simple prismatic shapes, BrepGen incorporates free-form\nand doubly-curved surfaces for the first time. Additional applications of\nBrepGen include CAD autocomplete and design interpolation. The code, pretrained\nmodels, and dataset are available at https://github.com/samxuxiang/BrepGen.\n", "date": "2024-01-28", "categories": []} {"id": "2405.11914", "title": "PT43D: A Probabilistic Transformer for Generating 3D Shapes from Single\n Highly-Ambiguous RGB Images", "abstract": " Generating 3D shapes from single RGB images is essential in various\napplications such as robotics. Current approaches typically target images\ncontaining clear and complete visual descriptions of the object, without\nconsidering common realistic cases where observations of objects that are\nlargely occluded or truncated. We thus propose a transformer-based\nautoregressive model to generate the probabilistic distribution of 3D shapes\nconditioned on an RGB image containing potentially highly ambiguous\nobservations of the object. To handle realistic scenarios such as occlusion or\nfield-of-view truncation, we create simulated image-to-shape training pairs\nthat enable improved fine-tuning for real-world scenarios. We then adopt\ncross-attention to effectively identify the most relevant region of interest\nfrom the input image for shape generation. This enables inference of sampled\nshapes with reasonable diversity and strong alignment with the input image. We\ntrain and test our model on our synthetic data then fine-tune and test it on\nreal-world data. Experiments demonstrate that our model outperforms state of\nthe art in both scenarios.\n", "date": "2024-05-20", "categories": []} {"id": "2406.09728", "title": "Neural Pose Representation Learning for Generating and Transferring\n Non-Rigid Object Poses", "abstract": " We propose a novel method for learning representations of poses for 3D\ndeformable objects, which specializes in 1) disentangling pose information from\nthe object's identity, 2) facilitating the learning of pose variations, and 3)\ntransferring pose information to other object identities. Based on these\nproperties, our method enables the generation of 3D deformable objects with\ndiversity in both identities and poses, using variations of a single object. It\ndoes not require explicit shape parameterization such as skeletons or joints,\npoint-level or shape-level correspondence supervision, or variations of the\ntarget object for pose transfer. To achieve pose disentanglement, compactness\nfor generative models, and transferability, we first design the pose extractor\nto represent the pose as a keypoint-based hybrid representation and the pose\napplier to learn an implicit deformation field. To better distill pose\ninformation from the object's geometry, we propose the implicit pose applier to\noutput an intrinsic mesh property, the face Jacobian. Once the extracted pose\ninformation is transferred to the target object, the pose applier is fine-tuned\nin a self-supervised manner to better describe the target object's shapes with\npose variations. The extracted poses are also used to train a cascaded\ndiffusion model to enable the generation of novel poses. Our experiments with\nthe DeformThings4D and Human datasets demonstrate state-of-the-art performance\nin pose transfer and the ability to generate diverse deformed shapes with\nvarious objects and poses.\n", "date": "2024-06-14", "categories": []} {"id": "2411.01602", "title": "DreamPolish: Domain Score Distillation With Progressive Geometry\n Generation", "abstract": " We introduce DreamPolish, a text-to-3D generation model that excels in\nproducing refined geometry and high-quality textures. In the geometry\nconstruction phase, our approach leverages multiple neural representations to\nenhance the stability of the synthesis process. Instead of relying solely on a\nview-conditioned diffusion prior in the novel sampled views, which often leads\nto undesired artifacts in the geometric surface, we incorporate an additional\nnormal estimator to polish the geometry details, conditioned on viewpoints with\nvarying field-of-views. We propose to add a surface polishing stage with only a\nfew training steps, which can effectively refine the artifacts attributed to\nlimited guidance from previous stages and produce 3D objects with more\ndesirable geometry. The key topic of texture generation using pretrained\ntext-to-image models is to find a suitable domain in the vast latent\ndistribution of these models that contains photorealistic and consistent\nrenderings. In the texture generation phase, we introduce a novel score\ndistillation objective, namely domain score distillation (DSD), to guide neural\nrepresentations toward such a domain. We draw inspiration from the\nclassifier-free guidance (CFG) in textconditioned image generation tasks and\nshow that CFG and variational distribution guidance represent distinct aspects\nin gradient guidance and are both imperative domains for the enhancement of\ntexture quality. Extensive experiments show our proposed model can produce 3D\nassets with polished surfaces and photorealistic textures, outperforming\nexisting state-of-the-art methods.\n", "date": "2024-11-03", "categories": []} {"id": "2411.02336", "title": "MVPaint: Synchronized Multi-View Diffusion for Painting Anything 3D", "abstract": " Texturing is a crucial step in the 3D asset production workflow, which\nenhances the visual appeal and diversity of 3D assets. Despite recent\nadvancements in Text-to-Texture (T2T) generation, existing methods often yield\nsubpar results, primarily due to local discontinuities, inconsistencies across\nmultiple views, and their heavy dependence on UV unwrapping outcomes. To tackle\nthese challenges, we propose a novel generation-refinement 3D texturing\nframework called MVPaint, which can generate high-resolution, seamless textures\nwhile emphasizing multi-view consistency. MVPaint mainly consists of three key\nmodules. 1) Synchronized Multi-view Generation (SMG). Given a 3D mesh model,\nMVPaint first simultaneously generates multi-view images by employing an SMG\nmodel, which leads to coarse texturing results with unpainted parts due to\nmissing observations. 2) Spatial-aware 3D Inpainting (S3I). To ensure complete\n3D texturing, we introduce the S3I method, specifically designed to effectively\ntexture previously unobserved areas. 3) UV Refinement (UVR). Furthermore,\nMVPaint employs a UVR module to improve the texture quality in the UV space,\nwhich first performs a UV-space Super-Resolution, followed by a Spatial-aware\nSeam-Smoothing algorithm for revising spatial texturing discontinuities caused\nby UV unwrapping. Moreover, we establish two T2T evaluation benchmarks: the\nObjaverse T2T benchmark and the GSO T2T benchmark, based on selected\nhigh-quality 3D meshes from the Objaverse dataset and the entire GSO dataset,\nrespectively. Extensive experimental results demonstrate that MVPaint surpasses\nexisting state-of-the-art methods. Notably, MVPaint could generate\nhigh-fidelity textures with minimal Janus issues and highly enhanced cross-view\nconsistency.\n", "date": "2024-11-04", "categories": []} {"id": "2411.00399", "title": "StyleTex: Style Image-Guided Texture Generation for 3D Models", "abstract": " Style-guided texture generation aims to generate a texture that is harmonious\nwith both the style of the reference image and the geometry of the input mesh,\ngiven a reference style image and a 3D mesh with its text description. Although\ndiffusion-based 3D texture generation methods, such as distillation sampling,\nhave numerous promising applications in stylized games and films, it requires\naddressing two challenges: 1) decouple style and content completely from the\nreference image for 3D models, and 2) align the generated texture with the\ncolor tone, style of the reference image, and the given text prompt. To this\nend, we introduce StyleTex, an innovative diffusion-model-based framework for\ncreating stylized textures for 3D models. Our key insight is to decouple style\ninformation from the reference image while disregarding content in\ndiffusion-based distillation sampling. Specifically, given a reference image,\nwe first decompose its style feature from the image CLIP embedding by\nsubtracting the embedding's orthogonal projection in the direction of the\ncontent feature, which is represented by a text CLIP embedding. Our novel\napproach to disentangling the reference image's style and content information\nallows us to generate distinct style and content features. We then inject the\nstyle feature into the cross-attention mechanism to incorporate it into the\ngeneration process, while utilizing the content feature as a negative prompt to\nfurther dissociate content information. Finally, we incorporate these\nstrategies into StyleTex to obtain stylized textures. The resulting textures\ngenerated by StyleTex retain the style of the reference image, while also\naligning with the text prompts and intrinsic details of the given 3D mesh.\nQuantitative and qualitative experiments show that our method outperforms\nexisting baseline methods by a significant margin.\n", "date": "2024-11-01", "categories": []} {"id": "2403.19655", "title": "GaussianCube: A Structured and Explicit Radiance Representation for 3D\n Generative Modeling", "abstract": " We introduce a radiance representation that is both structured and fully\nexplicit and thus greatly facilitates 3D generative modeling. Existing radiance\nrepresentations either require an implicit feature decoder, which significantly\ndegrades the modeling power of the representation, or are spatially\nunstructured, making them difficult to integrate with mainstream 3D diffusion\nmethods. We derive GaussianCube by first using a novel\ndensification-constrained Gaussian fitting algorithm, which yields\nhigh-accuracy fitting using a fixed number of free Gaussians, and then\nrearranging these Gaussians into a predefined voxel grid via Optimal Transport.\nSince GaussianCube is a structured grid representation, it allows us to use\nstandard 3D U-Net as our backbone in diffusion modeling without elaborate\ndesigns. More importantly, the high-accuracy fitting of the Gaussians allows us\nto achieve a high-quality representation with orders of magnitude fewer\nparameters than previous structured representations for comparable quality,\nranging from one to two orders of magnitude. The compactness of GaussianCube\ngreatly eases the difficulty of 3D generative modeling. Extensive experiments\nconducted on unconditional and class-conditioned object generation, digital\navatar creation, and text-to-3D synthesis all show that our model achieves\nstate-of-the-art generation results both qualitatively and quantitatively,\nunderscoring the potential of GaussianCube as a highly accurate and versatile\nradiance representation for 3D generative modeling. Project page:\nhttps://gaussiancube.github.io/.\n", "date": "2024-03-28", "categories": []} {"id": "2410.21299", "title": "TV-3DG: Mastering Text-to-3D Customized Generation with Visual Prompt", "abstract": " In recent years, advancements in generative models have significantly\nexpanded the capabilities of text-to-3D generation. Many approaches rely on\nScore Distillation Sampling (SDS) technology. However, SDS struggles to\naccommodate multi-condition inputs, such as text and visual prompts, in\ncustomized generation tasks. To explore the core reasons, we decompose SDS into\na difference term and a classifier-free guidance term. Our analysis identifies\nthe core issue as arising from the difference term and the random noise\naddition during the optimization process, both contributing to deviations from\nthe target mode during distillation. To address this, we propose a novel\nalgorithm, Classifier Score Matching (CSM), which removes the difference term\nin SDS and uses a deterministic noise addition process to reduce noise during\noptimization, effectively overcoming the low-quality limitations of SDS in our\ncustomized generation framework. Based on CSM, we integrate visual prompt\ninformation with an attention fusion mechanism and sampling guidance\ntechniques, forming the Visual Prompt CSM (VPCSM) algorithm. Furthermore, we\nintroduce a Semantic-Geometry Calibration (SGC) module to enhance quality\nthrough improved textual information integration. We present our approach as\nTV-3DG, with extensive experiments demonstrating its capability to achieve\nstable, high-quality, customized 3D generation. Project page:\n\\url{https://yjhboy.github.io/TV-3DG}\n", "date": "2024-10-16", "categories": []} {"id": "2410.19657", "title": "DiffGS: Functional Gaussian Splatting Diffusion", "abstract": " 3D Gaussian Splatting (3DGS) has shown convincing performance in rendering\nspeed and fidelity, yet the generation of Gaussian Splatting remains a\nchallenge due to its discreteness and unstructured nature. In this work, we\npropose DiffGS, a general Gaussian generator based on latent diffusion models.\nDiffGS is a powerful and efficient 3D generative model which is capable of\ngenerating Gaussian primitives at arbitrary numbers for high-fidelity rendering\nwith rasterization. The key insight is to represent Gaussian Splatting in a\ndisentangled manner via three novel functions to model Gaussian probabilities,\ncolors and transforms. Through the novel disentanglement of 3DGS, we represent\nthe discrete and unstructured 3DGS with continuous Gaussian Splatting\nfunctions, where we then train a latent diffusion model with the target of\ngenerating these Gaussian Splatting functions both unconditionally and\nconditionally. Meanwhile, we introduce a discretization algorithm to extract\nGaussians at arbitrary numbers from the generated functions via octree-guided\nsampling and optimization. We explore DiffGS for various tasks, including\nunconditional generation, conditional generation from text, image, and partial\n3DGS, as well as Point-to-Gaussian generation. We believe that DiffGS provides\na new direction for flexibly modeling and generating Gaussian Splatting.\n", "date": "2024-10-25", "categories": []} {"id": "2405.20343", "title": "Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single\n Image", "abstract": " In this work, we introduce Unique3D, a novel image-to-3D framework for\nefficiently generating high-quality 3D meshes from single-view images,\nfeaturing state-of-the-art generation fidelity and strong generalizability.\nPrevious methods based on Score Distillation Sampling (SDS) can produce\ndiversified 3D results by distilling 3D knowledge from large 2D diffusion\nmodels, but they usually suffer from long per-case optimization time with\ninconsistent issues. Recent works address the problem and generate better 3D\nresults either by finetuning a multi-view diffusion model or training a fast\nfeed-forward model. However, they still lack intricate textures and complex\ngeometries due to inconsistency and limited generated resolution. To\nsimultaneously achieve high fidelity, consistency, and efficiency in single\nimage-to-3D, we propose a novel framework Unique3D that includes a multi-view\ndiffusion model with a corresponding normal diffusion model to generate\nmulti-view images with their normal maps, a multi-level upscale process to\nprogressively improve the resolution of generated orthographic multi-views, as\nwell as an instant and consistent mesh reconstruction algorithm called ISOMER,\nwhich fully integrates the color and geometric priors into mesh results.\nExtensive experiments demonstrate that our Unique3D significantly outperforms\nother image-to-3D baselines in terms of geometric and textural details.\n", "date": "2024-05-30", "categories": []} {"id": "2410.20723", "title": "CompGS: Unleashing 2D Compositionality for Compositional Text-to-3D via\n Dynamically Optimizing 3D Gaussians", "abstract": " Recent breakthroughs in text-guided image generation have significantly\nadvanced the field of 3D generation. While generating a single high-quality 3D\nobject is now feasible, generating multiple objects with reasonable\ninteractions within a 3D space, a.k.a. compositional 3D generation, presents\nsubstantial challenges. This paper introduces CompGS, a novel generative\nframework that employs 3D Gaussian Splatting (GS) for efficient, compositional\ntext-to-3D content generation. To achieve this goal, two core designs are\nproposed: (1) 3D Gaussians Initialization with 2D compositionality: We transfer\nthe well-established 2D compositionality to initialize the Gaussian parameters\non an entity-by-entity basis, ensuring both consistent 3D priors for each\nentity and reasonable interactions among multiple entities; (2) Dynamic\nOptimization: We propose a dynamic strategy to optimize 3D Gaussians using\nScore Distillation Sampling (SDS) loss. CompGS first automatically decomposes\n3D Gaussians into distinct entity parts, enabling optimization at both the\nentity and composition levels. Additionally, CompGS optimizes across objects of\nvarying scales by dynamically adjusting the spatial parameters of each entity,\nenhancing the generation of fine-grained details, particularly in smaller\nentities. Qualitative comparisons and quantitative evaluations on T3Bench\ndemonstrate the effectiveness of CompGS in generating compositional 3D objects\nwith superior image quality and semantic alignment over existing methods.\nCompGS can also be easily extended to controllable 3D editing, facilitating\nscene generation. We hope CompGS will provide new insights to the compositional\n3D generation. Project page: https://chongjiange.github.io/compgs.html.\n", "date": "2024-10-28", "categories": []} {"id": "2305.06131", "title": "Generative AI meets 3D: A Survey on Text-to-3D in AIGC Era", "abstract": " Generative AI has made significant progress in recent years, with text-guided\ncontent generation being the most practical as it facilitates interaction\nbetween human instructions and AI-generated content (AIGC). Thanks to\nadvancements in text-to-image and 3D modeling technologies, like neural\nradiance field (NeRF), text-to-3D has emerged as a nascent yet highly active\nresearch field. Our work conducts a comprehensive survey on this topic and\nfollows up on subsequent research progress in the overall field, aiming to help\nreaders interested in this direction quickly catch up with its rapid\ndevelopment. First, we introduce 3D data representations, including both\nStructured and non-Structured data. Building on this pre-requisite, we\nintroduce various core technologies to achieve satisfactory text-to-3D results.\nAdditionally, we present mainstream baselines and research directions in recent\ntext-to-3D technology, including fidelity, efficiency, consistency,\ncontrollability, diversity, and applicability. Furthermore, we summarize the\nusage of text-to-3D technology in various applications, including avatar\ngeneration, texture generation, scene generation and 3D editing. Finally, we\ndiscuss the agenda for the future development of text-to-3D.\n", "date": "2023-05-10", "categories": []} {"id": "2410.18986", "title": "VehicleSDF: A 3D generative model for constrained engineering design via\n surrogate modeling", "abstract": " A main challenge in mechanical design is to efficiently explore the design\nspace while satisfying engineering constraints. This work explores the use of\n3D generative models to explore the design space in the context of vehicle\ndevelopment, while estimating and enforcing engineering constraints.\nSpecifically, we generate diverse 3D models of cars that meet a given set of\ngeometric specifications, while also obtaining quick estimates of performance\nparameters such as aerodynamic drag. For this, we employ a data-driven approach\n(using the ShapeNet dataset) to train VehicleSDF, a DeepSDF based model that\nrepresents potential designs in a latent space witch can be decoded into a 3D\nmodel. We then train surrogate models to estimate engineering parameters from\nthis latent space representation, enabling us to efficiently optimize latent\nvectors to match specifications. Our experiments show that we can generate\ndiverse 3D models while matching the specified geometric parameters. Finally,\nwe demonstrate that other performance parameters such as aerodynamic drag can\nbe estimated in a differentiable pipeline.\n", "date": "2024-10-09", "categories": []} {"id": "2410.17802", "title": "GenUDC: High Quality 3D Mesh Generation with Unsigned Dual Contouring\n Representation", "abstract": " Generating high-quality meshes with complex structures and realistic surfaces\nis the primary goal of 3D generative models. Existing methods typically employ\nsequence data or deformable tetrahedral grids for mesh generation. However,\nsequence-based methods have difficulty producing complex structures with many\nfaces due to memory limits. The deformable tetrahedral grid-based method\nMeshDiffusion fails to recover realistic surfaces due to the inherent ambiguity\nin deformable grids. We propose the GenUDC framework to address these\nchallenges by leveraging the Unsigned Dual Contouring (UDC) as the mesh\nrepresentation. UDC discretizes a mesh in a regular grid and divides it into\nthe face and vertex parts, recovering both complex structures and fine details.\nAs a result, the one-to-one mapping between UDC and mesh resolves the ambiguity\nproblem. In addition, GenUDC adopts a two-stage, coarse-to-fine generative\nprocess for 3D mesh generation. It first generates the face part as a rough\nshape and then the vertex part to craft a detailed shape. Extensive evaluations\ndemonstrate the superiority of UDC as a mesh representation and the favorable\nperformance of GenUDC in mesh generation. The code and trained models are\navailable at https://github.com/TrepangCat/GenUDC.\n", "date": "2024-10-23", "categories": []} {"id": "2310.05375", "title": "IPDreamer: Appearance-Controllable 3D Object Generation with Complex\n Image Prompts", "abstract": " Recent advances in 3D generation have been remarkable, with methods such as\nDreamFusion leveraging large-scale text-to-image diffusion-based models to\nguide 3D object generation. These methods enable the synthesis of detailed and\nphotorealistic textured objects. However, the appearance of 3D objects produced\nby such text-to-3D models is often unpredictable, and it is hard for\nsingle-image-to-3D methods to deal with images lacking a clear subject,\ncomplicating the generation of appearance-controllable 3D objects from complex\nimages. To address these challenges, we present IPDreamer, a novel method that\ncaptures intricate appearance features from complex $\\textbf{I}$mage\n$\\textbf{P}$rompts and aligns the synthesized 3D object with these extracted\nfeatures, enabling high-fidelity, appearance-controllable 3D object generation.\nOur experiments demonstrate that IPDreamer consistently generates high-quality\n3D objects that align with both the textual and complex image prompts,\nhighlighting its promising capability in appearance-controlled, complex 3D\nobject generation. Our code is available at\nhttps://github.com/zengbohan0217/IPDreamer.\n", "date": "2023-10-09", "categories": []} {"id": "2405.20674", "title": "4Diffusion: Multi-view Video Diffusion Model for 4D Generation", "abstract": " Current 4D generation methods have achieved noteworthy efficacy with the aid\nof advanced diffusion generative models. However, these methods lack multi-view\nspatial-temporal modeling and encounter challenges in integrating diverse prior\nknowledge from multiple diffusion models, resulting in inconsistent temporal\nappearance and flickers. In this paper, we propose a novel 4D generation\npipeline, namely 4Diffusion, aimed at generating spatial-temporally consistent\n4D content from a monocular video. We first design a unified diffusion model\ntailored for multi-view video generation by incorporating a learnable motion\nmodule into a frozen 3D-aware diffusion model to capture multi-view\nspatial-temporal correlations. After training on a curated dataset, our\ndiffusion model acquires reasonable temporal consistency and inherently\npreserves the generalizability and spatial consistency of the 3D-aware\ndiffusion model. Subsequently, we propose 4D-aware Score Distillation Sampling\nloss, which is based on our multi-view video diffusion model, to optimize 4D\nrepresentation parameterized by dynamic NeRF. This aims to eliminate\ndiscrepancies arising from multiple diffusion models, allowing for generating\nspatial-temporally consistent 4D content. Moreover, we devise an anchor loss to\nenhance the appearance details and facilitate the learning of dynamic NeRF.\nExtensive qualitative and quantitative experiments demonstrate that our method\nachieves superior performance compared to previous methods.\n", "date": "2024-05-31", "categories": []} {"id": "2407.11965", "title": "UrbanWorld: An Urban World Model for 3D City Generation", "abstract": " Cities, as the essential environment of human life, encompass diverse\nphysical elements such as buildings, roads and vegetation, which continuously\ninteract with dynamic entities like people and vehicles. Crafting realistic,\ninteractive 3D urban environments is essential for nurturing AGI systems and\nconstructing AI agents capable of perceiving, decision-making, and acting like\nhumans in real-world environments. However, creating high-fidelity 3D urban\nenvironments usually entails extensive manual labor from designers, involving\nintricate detailing and representation of complex urban elements. Therefore,\naccomplishing this automatically remains a longstanding challenge. Toward this\nproblem, we propose UrbanWorld, the first generative urban world model that can\nautomatically create a customized, realistic and interactive 3D urban world\nwith flexible control conditions. UrbanWorld incorporates four key stages in\nthe generation pipeline: flexible 3D layout generation from OSM data or urban\nlayout with semantic and height maps, urban scene design with Urban MLLM,\ncontrollable urban asset rendering via progressive 3D diffusion, and\nMLLM-assisted scene refinement. We conduct extensive quantitative analysis on\nfive visual metrics, demonstrating that UrbanWorld achieves SOTA generation\nrealism. Next, we provide qualitative results about the controllable generation\ncapabilities of UrbanWorld using both textual and image-based prompts. Lastly,\nwe verify the interactive nature of these environments by showcasing the agent\nperception and navigation within the created environments. We contribute\nUrbanWorld as an open-source tool available at\nhttps://github.com/Urban-World/UrbanWorld.\n", "date": "2024-07-16", "categories": []} {"id": "2410.16395", "title": "Joker: Conditional 3D Head Synthesis with Extreme Facial Expressions", "abstract": " We introduce Joker, a new method for the conditional synthesis of 3D human\nheads with extreme expressions. Given a single reference image of a person, we\nsynthesize a volumetric human head with the reference identity and a new\nexpression. We offer control over the expression via a 3D morphable model\n(3DMM) and textual inputs. This multi-modal conditioning signal is essential\nsince 3DMMs alone fail to define subtle emotional changes and extreme\nexpressions, including those involving the mouth cavity and tongue\narticulation. Our method is built upon a 2D diffusion-based prior that\ngeneralizes well to out-of-domain samples, such as sculptures, heavy makeup,\nand paintings while achieving high levels of expressiveness. To improve view\nconsistency, we propose a new 3D distillation technique that converts\npredictions of our 2D prior into a neural radiance field (NeRF). Both the 2D\nprior and our distillation technique produce state-of-the-art results, which\nare confirmed by our extensive evaluations. Also, to the best of our knowledge,\nour method is the first to achieve view-consistent extreme tongue articulation.\n", "date": "2024-10-21", "categories": []} {"id": "2410.16272", "title": "MvDrag3D: Drag-based Creative 3D Editing via Multi-view\n Generation-Reconstruction Priors", "abstract": " Drag-based editing has become popular in 2D content creation, driven by the\ncapabilities of image generative models. However, extending this technique to\n3D remains a challenge. Existing 3D drag-based editing methods, whether\nemploying explicit spatial transformations or relying on implicit latent\noptimization within limited-capacity 3D generative models, fall short in\nhandling significant topology changes or generating new textures across diverse\nobject categories. To overcome these limitations, we introduce MVDrag3D, a\nnovel framework for more flexible and creative drag-based 3D editing that\nleverages multi-view generation and reconstruction priors. At the core of our\napproach is the usage of a multi-view diffusion model as a strong generative\nprior to perform consistent drag editing over multiple rendered views, which is\nfollowed by a reconstruction model that reconstructs 3D Gaussians of the edited\nobject. While the initial 3D Gaussians may suffer from misalignment between\ndifferent views, we address this via view-specific deformation networks that\nadjust the position of Gaussians to be well aligned. In addition, we propose a\nmulti-view score function that distills generative priors from multiple views\nto further enhance the view consistency and visual quality. Extensive\nexperiments demonstrate that MVDrag3D provides a precise, generative, and\nflexible solution for 3D drag-based editing, supporting more versatile editing\neffects across various object categories and 3D representations.\n", "date": "2024-10-21", "categories": []} {"id": "2404.19525", "title": "MicroDreamer: Efficient 3D Generation in $\\sim$20 Seconds by Score-based\n Iterative Reconstruction", "abstract": " Optimization-based approaches, such as score distillation sampling (SDS),\nshow promise in zero-shot 3D generation but suffer from low efficiency,\nprimarily due to the high number of function evaluations (NFEs) required for\neach sample and the limitation of optimization confined to latent space. This\npaper introduces score-based iterative reconstruction (SIR), an efficient and\ngeneral algorithm mimicking a differentiable 3D reconstruction process to\nreduce the NFEs and enable optimization in pixel space. Given a single set of\nimages sampled from a multi-view score-based diffusion model, SIR repeatedly\noptimizes 3D parameters, unlike the single-step optimization in SDS. With other\nimprovements in training, we present an efficient approach called MicroDreamer\nthat generally applies to various 3D representations and 3D generation tasks.\nIn particular, MicroDreamer is 5-20 times faster than SDS in generating neural\nradiance field while retaining a comparable performance and takes about 20\nseconds to create meshes from 3D Gaussian splatting on a single A100 GPU,\nhalving the time of the fastest optimization-based baseline DreamGaussian with\nsignificantly superior performance compared to the measurement standard\ndeviation. Our code is available at https://github.com/ML-GSAI/MicroDreamer.\n", "date": "2024-04-30", "categories": []} {"id": "2402.13251", "title": "FlashTex: Fast Relightable Mesh Texturing with LightControlNet", "abstract": " Manually creating textures for 3D meshes is time-consuming, even for expert\nvisual content creators. We propose a fast approach for automatically texturing\nan input 3D mesh based on a user-provided text prompt. Importantly, our\napproach disentangles lighting from surface material/reflectance in the\nresulting texture so that the mesh can be properly relit and rendered in any\nlighting environment. We introduce LightControlNet, a new text-to-image model\nbased on the ControlNet architecture, which allows the specification of the\ndesired lighting as a conditioning image to the model. Our text-to-texture\npipeline then constructs the texture in two stages. The first stage produces a\nsparse set of visually consistent reference views of the mesh using\nLightControlNet. The second stage applies a texture optimization based on Score\nDistillation Sampling (SDS) that works with LightControlNet to increase the\ntexture quality while disentangling surface material from lighting. Our\nalgorithm is significantly faster than previous text-to-texture methods, while\nproducing high-quality and relightable textures.\n", "date": "2024-02-20", "categories": []} {"id": "2410.12928", "title": "DreamCraft3D++: Efficient Hierarchical 3D Generation with Multi-Plane\n Reconstruction Model", "abstract": " We introduce DreamCraft3D++, an extension of DreamCraft3D that enables\nefficient high-quality generation of complex 3D assets. DreamCraft3D++ inherits\nthe multi-stage generation process of DreamCraft3D, but replaces the\ntime-consuming geometry sculpting optimization with a feed-forward multi-plane\nbased reconstruction model, speeding up the process by 1000x. For texture\nrefinement, we propose a training-free IP-Adapter module that is conditioned on\nthe enhanced multi-view images to enhance texture and geometry consistency,\nproviding a 4x faster alternative to DreamCraft3D's DreamBooth fine-tuning.\nExperiments on diverse datasets demonstrate DreamCraft3D++'s ability to\ngenerate creative 3D assets with intricate geometry and realistic 360{\\deg}\ntextures, outperforming state-of-the-art image-to-3D methods in quality and\nspeed. The full implementation will be open-sourced to enable new possibilities\nin 3D content creation.\n", "date": "2024-10-16", "categories": []} {"id": "2410.13530", "title": "L3DG: Latent 3D Gaussian Diffusion", "abstract": " We propose L3DG, the first approach for generative 3D modeling of 3D\nGaussians through a latent 3D Gaussian diffusion formulation. This enables\neffective generative 3D modeling, scaling to generation of entire room-scale\nscenes which can be very efficiently rendered. To enable effective synthesis of\n3D Gaussians, we propose a latent diffusion formulation, operating in a\ncompressed latent space of 3D Gaussians. This compressed latent space is\nlearned by a vector-quantized variational autoencoder (VQ-VAE), for which we\nemploy a sparse convolutional architecture to efficiently operate on room-scale\nscenes. This way, the complexity of the costly generation process via diffusion\nis substantially reduced, allowing higher detail on object-level generation, as\nwell as scalability to large scenes. By leveraging the 3D Gaussian\nrepresentation, the generated scenes can be rendered from arbitrary viewpoints\nin real-time. We demonstrate that our approach significantly improves visual\nquality over prior work on unconditional object-level radiance field synthesis\nand showcase its applicability to room-scale scene generation.\n", "date": "2024-10-17", "categories": []} {"id": "2410.11795", "title": "Efficient Diffusion Models: A Comprehensive Survey from Principles to\n Practices", "abstract": " As one of the most popular and sought-after generative models in the recent\nyears, diffusion models have sparked the interests of many researchers and\nsteadily shown excellent advantage in various generative tasks such as image\nsynthesis, video generation, molecule design, 3D scene rendering and multimodal\ngeneration, relying on their dense theoretical principles and reliable\napplication practices. The remarkable success of these recent efforts on\ndiffusion models comes largely from progressive design principles and efficient\narchitecture, training, inference, and deployment methodologies. However, there\nhas not been a comprehensive and in-depth review to summarize these principles\nand practices to help the rapid understanding and application of diffusion\nmodels. In this survey, we provide a new efficiency-oriented perspective on\nthese existing efforts, which mainly focuses on the profound principles and\nefficient practices in architecture designs, model training, fast inference and\nreliable deployment, to guide further theoretical research, algorithm migration\nand model application for new scenarios in a reader-friendly way.\n\\url{https://github.com/ponyzym/Efficient-DMs-Survey}\n", "date": "2024-10-15", "categories": []} {"id": "2410.10399", "title": "Parameterize Structure with Differentiable Template for 3D Shape\n Generation", "abstract": " Structural representation is crucial for reconstructing and generating\neditable 3D shapes with part semantics. Recent 3D shape generation works employ\ncomplicated networks and structure definitions relying on hierarchical\nannotations and pay less attention to the details inside parts. In this paper,\nwe propose the method that parameterizes the shared structure in the same\ncategory using a differentiable template and corresponding fixed-length\nparameters. Specific parameters are fed into the template to calculate cuboids\nthat indicate a concrete shape. We utilize the boundaries of three-view\ndrawings of each cuboid to further describe the inside details. Shapes are\nrepresented with the parameters and three-view details inside cuboids, from\nwhich the SDF can be calculated to recover the object. Benefiting from our\nfixed-length parameters and three-view details, our networks for reconstruction\nand generation are simple and effective to learn the latent space. Our method\ncan reconstruct or generate diverse shapes with complicated details, and\ninterpolate them smoothly. Extensive evaluations demonstrate the superiority of\nour method on reconstruction from point cloud, generation, and interpolation.\n", "date": "2024-10-14", "categories": []} {"id": "2410.11628", "title": "Simultaneous Diffusion Sampling for Conditional LiDAR Generation", "abstract": " By enabling capturing of 3D point clouds that reflect the geometry of the\nimmediate environment, LiDAR has emerged as a primary sensor for autonomous\nsystems. If a LiDAR scan is too sparse, occluded by obstacles, or too small in\nrange, enhancing the point cloud scan by while respecting the geometry of the\nscene is useful for downstream tasks. Motivated by the explosive growth of\ninterest in generative methods in vision, conditional LiDAR generation is\nstarting to take off. This paper proposes a novel simultaneous diffusion\nsampling methodology to generate point clouds conditioned on the 3D structure\nof the scene as seen from multiple views. The key idea is to impose multi-view\ngeometric constraints on the generation process, exploiting mutual information\nfor enhanced results. Our method begins by recasting the input scan to multiple\nnew viewpoints around the scan, thus creating multiple synthetic LiDAR scans.\nThen, the synthetic and input LiDAR scans simultaneously undergo conditional\ngeneration according to our methodology. Results show that our method can\nproduce accurate and geometrically consistent enhancements to point cloud\nscans, allowing it to outperform existing methods by a large margin in a\nvariety of benchmarks.\n", "date": "2024-10-15", "categories": []} {"id": "2407.12291", "title": "JointDreamer: Ensuring Geometry Consistency and Text Congruence in\n Text-to-3D Generation via Joint Score Distillation", "abstract": " Score Distillation Sampling (SDS) by well-trained 2D diffusion models has\nshown great promise in text-to-3D generation. However, this paradigm distills\nview-agnostic 2D image distributions into the rendering distribution of 3D\nrepresentation for each view independently, overlooking the coherence across\nviews and yielding 3D inconsistency in generations. In this work, we propose\n\\textbf{J}oint \\textbf{S}core \\textbf{D}istillation (JSD), a new paradigm that\nensures coherent 3D generations. Specifically, we model the joint image\ndistribution, which introduces an energy function to capture the coherence\namong denoised images from the diffusion model. We then derive the joint score\ndistillation on multiple rendered views of the 3D representation, as opposed to\na single view in SDS. In addition, we instantiate three universal view-aware\nmodels as energy functions, demonstrating compatibility with JSD. Empirically,\nJSD significantly mitigates the 3D inconsistency problem in SDS, while\nmaintaining text congruence. Moreover, we introduce the Geometry Fading scheme\nand Classifier-Free Guidance (CFG) Switching strategy to enhance generative\ndetails. Our framework, JointDreamer, establishes a new benchmark in text-to-3D\ngeneration, achieving outstanding results with an 88.5\\% CLIP R-Precision and\n27.7\\% CLIP Score. These metrics demonstrate exceptional text congruence, as\nwell as remarkable geometric consistency and texture fidelity.\n", "date": "2024-07-17", "categories": []} {"id": "2408.13711", "title": "SceneDreamer360: Text-Driven 3D-Consistent Scene Generation with\n Panoramic Gaussian Splatting", "abstract": " Text-driven 3D scene generation has seen significant advancements recently.\nHowever, most existing methods generate single-view images using generative\nmodels and then stitch them together in 3D space. This independent generation\nfor each view often results in spatial inconsistency and implausibility in the\n3D scenes. To address this challenge, we proposed a novel text-driven\n3D-consistent scene generation model: SceneDreamer360. Our proposed method\nleverages a text-driven panoramic image generation model as a prior for 3D\nscene generation and employs 3D Gaussian Splatting (3DGS) to ensure consistency\nacross multi-view panoramic images. Specifically, SceneDreamer360 enhances the\nfine-tuned Panfusion generator with a three-stage panoramic enhancement,\nenabling the generation of high-resolution, detail-rich panoramic images.\nDuring the 3D scene construction, a novel point cloud fusion initialization\nmethod is used, producing higher quality and spatially consistent point clouds.\nOur extensive experiments demonstrate that compared to other methods,\nSceneDreamer360 with its panoramic image generation and 3DGS can produce higher\nquality, spatially consistent, and visually appealing 3D scenes from any text\nprompt. Our codes are available at\n\\url{https://github.com/liwrui/SceneDreamer360}.\n", "date": "2024-08-25", "categories": []} {"id": "2410.09592", "title": "ControLRM: Fast and Controllable 3D Generation via Large Reconstruction\n Model", "abstract": " Despite recent advancements in 3D generation methods, achieving\ncontrollability still remains a challenging issue. Current approaches utilizing\nscore-distillation sampling are hindered by laborious procedures that consume a\nsignificant amount of time. Furthermore, the process of first generating 2D\nrepresentations and then mapping them to 3D lacks internal alignment between\nthe two forms of representation. To address these challenges, we introduce\nControLRM, an end-to-end feed-forward model designed for rapid and controllable\n3D generation using a large reconstruction model (LRM). ControLRM comprises a\n2D condition generator, a condition encoding transformer, and a triplane\ndecoder transformer. Instead of training our model from scratch, we advocate\nfor a joint training framework. In the condition training branch, we lock the\ntriplane decoder and reuses the deep and robust encoding layers pretrained with\nmillions of 3D data in LRM. In the image training branch, we unlock the\ntriplane decoder to establish an implicit alignment between the 2D and 3D\nrepresentations. To ensure unbiased evaluation, we curate evaluation samples\nfrom three distinct datasets (G-OBJ, GSO, ABO) rather than relying on\ncherry-picking manual generation. The comprehensive experiments conducted on\nquantitative and qualitative comparisons of 3D controllability and generation\nquality demonstrate the strong generalization capacity of our proposed\napproach.\n", "date": "2024-10-12", "categories": []} {"id": "2410.10037", "title": "GALA: Geometry-Aware Local Adaptive Grids for Detailed 3D Generation", "abstract": " We propose GALA, a novel representation of 3D shapes that (i) excels at\ncapturing and reproducing complex geometry and surface details, (ii) is\ncomputationally efficient, and (iii) lends itself to 3D generative modelling\nwith modern, diffusion-based schemes. The key idea of GALA is to exploit both\nthe global sparsity of surfaces within a 3D volume and their local surface\nproperties. Sparsity is promoted by covering only the 3D object boundaries, not\nempty space, with an ensemble of tree root voxels. Each voxel contains an\noctree to further limit storage and compute to regions that contain surfaces.\nAdaptivity is achieved by fitting one local and geometry-aware coordinate frame\nin each non-empty leaf node. Adjusting the orientation of the local grid, as\nwell as the anisotropic scales of its axes, to the local surface shape greatly\nincreases the amount of detail that can be stored in a given amount of memory,\nwhich in turn allows for quantization without loss of quality. With our\noptimized C++/CUDA implementation, GALA can be fitted to an object in less than\n10 seconds. Moreover, the representation can efficiently be flattened and\nmanipulated with transformer networks. We provide a cascaded generation\npipeline capable of generating 3D shapes with great geometric detail.\n", "date": "2024-10-13", "categories": []} {"id": "2406.01579", "title": "Tetrahedron Splatting for 3D Generation", "abstract": " 3D representation is essential to the significant advance of 3D generation\nwith 2D diffusion priors. As a flexible representation, NeRF has been first\nadopted for 3D representation. With density-based volumetric rendering, it\nhowever suffers both intensive computational overhead and inaccurate mesh\nextraction. Using a signed distance field and Marching Tetrahedra, DMTet allows\nfor precise mesh extraction and real-time rendering but is limited in handling\nlarge topological changes in meshes, leading to optimization challenges.\nAlternatively, 3D Gaussian Splatting (3DGS) is favored in both training and\nrendering efficiency while falling short in mesh extraction. In this work, we\nintroduce a novel 3D representation, Tetrahedron Splatting (TeT-Splatting),\nthat supports easy convergence during optimization, precise mesh extraction,\nand real-time rendering simultaneously. This is achieved by integrating\nsurface-based volumetric rendering within a structured tetrahedral grid while\npreserving the desired ability of precise mesh extraction, and a tile-based\ndifferentiable tetrahedron rasterizer. Furthermore, we incorporate eikonal and\nnormal consistency regularization terms for the signed distance field to\nimprove generation quality and stability. Critically, our representation can be\ntrained without mesh extraction, making the optimization process easier to\nconverge. Our TeT-Splatting can be readily integrated in existing 3D generation\npipelines, along with polygonal mesh for texture optimization. Extensive\nexperiments show that our TeT-Splatting strikes a superior tradeoff among\nconvergence speed, render efficiency, and mesh quality as compared to previous\nalternatives under varying 3D generation settings.\n", "date": "2024-06-03", "categories": []} {"id": "2410.08824", "title": "One-shot Generative Domain Adaptation in 3D GANs", "abstract": " 3D-aware image generation necessitates extensive training data to ensure\nstable training and mitigate the risk of overfitting. This paper first\nconsiders a novel task known as One-shot 3D Generative Domain Adaptation (GDA),\naimed at transferring a pre-trained 3D generator from one domain to a new one,\nrelying solely on a single reference image. One-shot 3D GDA is characterized by\nthe pursuit of specific attributes, namely, high fidelity, large diversity,\ncross-domain consistency, and multi-view consistency. Within this paper, we\nintroduce 3D-Adapter, the first one-shot 3D GDA method, for diverse and\nfaithful generation. Our approach begins by judiciously selecting a restricted\nweight set for fine-tuning, and subsequently leverages four advanced loss\nfunctions to facilitate adaptation. An efficient progressive fine-tuning\nstrategy is also implemented to enhance the adaptation process. The synergy of\nthese three technological components empowers 3D-Adapter to achieve remarkable\nperformance, substantiated both quantitatively and qualitatively, across all\ndesired properties of 3D GDA. Furthermore, 3D-Adapter seamlessly extends its\ncapabilities to zero-shot scenarios, and preserves the potential for crucial\ntasks such as interpolation, reconstruction, and editing within the latent\nspace of the pre-trained generator. Code will be available at\nhttps://github.com/iceli1007/3D-Adapter.\n", "date": "2024-10-11", "categories": []} {"id": "2410.09009", "title": "Semantic Score Distillation Sampling for Compositional Text-to-3D\n Generation", "abstract": " Generating high-quality 3D assets from textual descriptions remains a pivotal\nchallenge in computer graphics and vision research. Due to the scarcity of 3D\ndata, state-of-the-art approaches utilize pre-trained 2D diffusion priors,\noptimized through Score Distillation Sampling (SDS). Despite progress, crafting\ncomplex 3D scenes featuring multiple objects or intricate interactions is still\ndifficult. To tackle this, recent methods have incorporated box or layout\nguidance. However, these layout-guided compositional methods often struggle to\nprovide fine-grained control, as they are generally coarse and lack\nexpressiveness. To overcome these challenges, we introduce a novel SDS\napproach, Semantic Score Distillation Sampling (SemanticSDS), designed to\neffectively improve the expressiveness and accuracy of compositional text-to-3D\ngeneration. Our approach integrates new semantic embeddings that maintain\nconsistency across different rendering views and clearly differentiate between\nvarious objects and parts. These embeddings are transformed into a semantic\nmap, which directs a region-specific SDS process, enabling precise optimization\nand compositional generation. By leveraging explicit semantic guidance, our\nmethod unlocks the compositional capabilities of existing pre-trained diffusion\nmodels, thereby achieving superior quality in 3D content generation,\nparticularly for complex objects and scenes. Experimental results demonstrate\nthat our SemanticSDS framework is highly effective for generating\nstate-of-the-art complex 3D content. Code:\nhttps://github.com/YangLing0818/SemanticSDS-3D\n", "date": "2024-10-11", "categories": []} {"id": "2405.15891", "title": "Score Distillation via Reparametrized DDIM", "abstract": " While 2D diffusion models generate realistic, high-detail images, 3D shape\ngeneration methods like Score Distillation Sampling (SDS) built on these 2D\ndiffusion models produce cartoon-like, over-smoothed shapes. To help explain\nthis discrepancy, we show that the image guidance used in Score Distillation\ncan be understood as the velocity field of a 2D denoising generative process,\nup to the choice of a noise term. In particular, after a change of variables,\nSDS resembles a high-variance version of Denoising Diffusion Implicit Models\n(DDIM) with a differently-sampled noise term: SDS introduces noise i.i.d.\nrandomly at each step, while DDIM infers it from the previous noise\npredictions. This excessive variance can lead to over-smoothing and unrealistic\noutputs. We show that a better noise approximation can be recovered by\ninverting DDIM in each SDS update step. This modification makes SDS's\ngenerative process for 2D images almost identical to DDIM. In 3D, it removes\nover-smoothing, preserves higher-frequency detail, and brings the generation\nquality closer to that of 2D samplers. Experimentally, our method achieves\nbetter or similar 3D generation quality compared to other state-of-the-art\nScore Distillation methods, all without training additional neural networks or\nmulti-view supervision, and providing useful insights into relationship between\n2D and 3D asset generation with diffusion models.\n", "date": "2024-05-24", "categories": []} {"id": "2408.16762", "title": "UV-free Texture Generation with Denoising and Geodesic Heat Diffusions", "abstract": " Seams, distortions, wasted UV space, vertex-duplication, and varying\nresolution over the surface are the most prominent issues of the standard\nUV-based texturing of meshes. These issues are particularly acute when\nautomatic UV-unwrapping techniques are used. For this reason, instead of\ngenerating textures in automatically generated UV-planes like most\nstate-of-the-art methods, we propose to represent textures as coloured\npoint-clouds whose colours are generated by a denoising diffusion probabilistic\nmodel constrained to operate on the surface of 3D objects. Our sampling and\nresolution agnostic generative model heavily relies on heat diffusion over the\nsurface of the meshes for spatial communication between points. To enable\nprocessing of arbitrarily sampled point-cloud textures and ensure long-distance\ntexture consistency we introduce a fast re-sampling of the mesh spectral\nproperties used during the heat diffusion and introduce a novel\nheat-diffusion-based self-attention mechanism. Our code and pre-trained models\nare available at github.com/simofoti/UV3-TeD.\n", "date": "2024-08-29", "categories": []} {"id": "2410.08181", "title": "RGM: Reconstructing High-fidelity 3D Car Assets with Relightable 3D-GS\n Generative Model from a Single Image", "abstract": " The generation of high-quality 3D car assets is essential for various\napplications, including video games, autonomous driving, and virtual reality.\nCurrent 3D generation methods utilizing NeRF or 3D-GS as representations for 3D\nobjects, generate a Lambertian object under fixed lighting and lack separated\nmodelings for material and global illumination. As a result, the generated\nassets are unsuitable for relighting under varying lighting conditions,\nlimiting their applicability in downstream tasks. To address this challenge, we\npropose a novel relightable 3D object generative framework that automates the\ncreation of 3D car assets, enabling the swift and accurate reconstruction of a\nvehicle's geometry, texture, and material properties from a single input image.\nOur approach begins with introducing a large-scale synthetic car dataset\ncomprising over 1,000 high-precision 3D vehicle models. We represent 3D objects\nusing global illumination and relightable 3D Gaussian primitives integrating\nwith BRDF parameters. Building on this representation, we introduce a\nfeed-forward model that takes images as input and outputs both relightable 3D\nGaussians and global illumination parameters. Experimental results demonstrate\nthat our method produces photorealistic 3D car assets that can be seamlessly\nintegrated into road scenes with different illuminations, which offers\nsubstantial practical benefits for industrial applications.\n", "date": "2024-10-10", "categories": []} {"id": "2405.20669", "title": "Hybrid Fourier Score Distillation for Efficient One Image to 3D Object\n Generation", "abstract": " Single image-to-3D generation is pivotal for crafting controllable 3D assets.\nGiven its under-constrained nature, we attempt to leverage 3D geometric priors\nfrom a novel view diffusion model and 2D appearance priors from an image\ngeneration model to guide the optimization process. We note that there is a\ndisparity between the generation priors of these two diffusion models, leading\nto their different appearance outputs. Specifically, image generation models\ntend to deliver more detailed visuals, whereas novel view models produce\nconsistent yet over-smooth results across different views. Directly combining\nthem leads to suboptimal effects due to their appearance conflicts. Hence, we\npropose a 2D-3D hybrid Fourier Score Distillation objective function, hy-FSD.\nIt optimizes 3D Gaussians using 3D priors in spatial domain to ensure geometric\nconsistency, while exploiting 2D priors in the frequency domain through Fourier\ntransform for better visual quality. hy-FSD can be integrated into existing 3D\ngeneration methods and produce significant performance gains. With this\ntechnique, we further develop an image-to-3D generation pipeline to create\nhigh-quality 3D objects within one minute, named Fourier123. Extensive\nexperiments demonstrate that Fourier123 excels in efficient generation with\nrapid convergence speed and visually-friendly generation results.\n", "date": "2024-05-31", "categories": []} {"id": "2406.10163", "title": "MeshAnything: Artist-Created Mesh Generation with Autoregressive\n Transformers", "abstract": " Recently, 3D assets created via reconstruction and generation have matched\nthe quality of manually crafted assets, highlighting their potential for\nreplacement. However, this potential is largely unrealized because these assets\nalways need to be converted to meshes for 3D industry applications, and the\nmeshes produced by current mesh extraction methods are significantly inferior\nto Artist-Created Meshes (AMs), i.e., meshes created by human artists.\nSpecifically, current mesh extraction methods rely on dense faces and ignore\ngeometric features, leading to inefficiencies, complicated post-processing, and\nlower representation quality. To address these issues, we introduce\nMeshAnything, a model that treats mesh extraction as a generation problem,\nproducing AMs aligned with specified shapes. By converting 3D assets in any 3D\nrepresentation into AMs, MeshAnything can be integrated with various 3D asset\nproduction methods, thereby enhancing their application across the 3D industry.\nThe architecture of MeshAnything comprises a VQ-VAE and a shape-conditioned\ndecoder-only transformer. We first learn a mesh vocabulary using the VQ-VAE,\nthen train the shape-conditioned decoder-only transformer on this vocabulary\nfor shape-conditioned autoregressive mesh generation. Our extensive experiments\nshow that our method generates AMs with hundreds of times fewer faces,\nsignificantly improving storage, rendering, and simulation efficiencies, while\nachieving precision comparable to previous methods.\n", "date": "2024-06-14", "categories": []} {"id": "2408.05008", "title": "FlowDreamer: Exploring High Fidelity Text-to-3D Generation via Rectified\n Flow", "abstract": " Recent advances in text-to-3D generation have made significant progress. In\nparticular, with the pretrained diffusion models, existing methods\npredominantly use Score Distillation Sampling (SDS) to train 3D models such as\nNeural RaRecent advances in text-to-3D generation have made significant\nprogress. In particular, with the pretrained diffusion models, existing methods\npredominantly use Score Distillation Sampling (SDS) to train 3D models such as\nNeural Radiance Fields (NeRF) and 3D Gaussian Splatting (3D GS). However, a\nhurdle is that they often encounter difficulties with over-smoothing textures\nand over-saturating colors. The rectified flow model -- which utilizes a simple\nordinary differential equation (ODE) to represent a straight trajectory --\nshows promise as an alternative prior to text-to-3D generation. It learns a\ntime-independent vector field, thereby reducing the ambiguity in 3D model\nupdate gradients that are calculated using time-dependent scores in the SDS\nframework. In light of this, we first develop a mathematical analysis to\nseamlessly integrate SDS with rectified flow model, paving the way for our\ninitial framework known as Vector Field Distillation Sampling (VFDS). However,\nempirical findings indicate that VFDS still results in over-smoothing outcomes.\nTherefore, we analyze the grounding reasons for such a failure from the\nperspective of ODE trajectories. On top, we propose a novel framework, named\nFlowDreamer, which yields high fidelity results with richer textual details and\nfaster convergence. The key insight is to leverage the coupling and reversible\nproperties of the rectified flow model to search for the corresponding noise,\nrather than using randomly sampled noise as in VFDS. Accordingly, we introduce\na novel Unique Couple Matching (UCM) loss, which guides the 3D model to\noptimize along the same trajectory.\n", "date": "2024-08-09", "categories": []} {"id": "2410.06756", "title": "DreamMesh4D: Video-to-4D Generation with Sparse-Controlled Gaussian-Mesh\n Hybrid Representation", "abstract": " Recent advancements in 2D/3D generative techniques have facilitated the\ngeneration of dynamic 3D objects from monocular videos. Previous methods mainly\nrely on the implicit neural radiance fields (NeRF) or explicit Gaussian\nSplatting as the underlying representation, and struggle to achieve\nsatisfactory spatial-temporal consistency and surface appearance. Drawing\ninspiration from modern 3D animation pipelines, we introduce DreamMesh4D, a\nnovel framework combining mesh representation with geometric skinning technique\nto generate high-quality 4D object from a monocular video. Instead of utilizing\nclassical texture map for appearance, we bind Gaussian splats to triangle face\nof mesh for differentiable optimization of both the texture and mesh vertices.\nIn particular, DreamMesh4D begins with a coarse mesh obtained through an\nimage-to-3D generation procedure. Sparse points are then uniformly sampled\nacross the mesh surface, and are used to build a deformation graph to drive the\nmotion of the 3D object for the sake of computational efficiency and providing\nadditional constraint. For each step, transformations of sparse control points\nare predicted using a deformation network, and the mesh vertices as well as the\nsurface Gaussians are deformed via a novel geometric skinning algorithm, which\nis a hybrid approach combining LBS (linear blending skinning) and DQS\n(dual-quaternion skinning), mitigating drawbacks associated with both\napproaches. The static surface Gaussians and mesh vertices as well as the\ndeformation network are learned via reference view photometric loss, score\ndistillation loss as well as other regularizers in a two-stage manner.\nExtensive experiments demonstrate superior performance of our method.\nFurthermore, our method is compatible with modern graphic pipelines, showcasing\nits potential in the 3D gaming and film industry.\n", "date": "2024-10-09", "categories": []} {"id": "2410.06985", "title": "Jointly Generating Multi-view Consistent PBR Textures using\n Collaborative Control", "abstract": " Multi-view consistency remains a challenge for image diffusion models. Even\nwithin the Text-to-Texture problem, where perfect geometric correspondences are\nknown a priori, many methods fail to yield aligned predictions across views,\nnecessitating non-trivial fusion methods to incorporate the results onto the\noriginal mesh. We explore this issue for a Collaborative Control workflow\nspecifically in PBR Text-to-Texture. Collaborative Control directly models PBR\nimage probability distributions, including normal bump maps; to our knowledge,\nthe only diffusion model to directly output full PBR stacks. We discuss the\ndesign decisions involved in making this model multi-view consistent, and\ndemonstrate the effectiveness of our approach in ablation studies, as well as\npractical applications.\n", "date": "2024-10-09", "categories": []} {"id": "2410.04965", "title": "Revealing Directions for Text-guided 3D Face Editing", "abstract": " 3D face editing is a significant task in multimedia, aimed at the\nmanipulation of 3D face models across various control signals. The success of\n3D-aware GAN provides expressive 3D models learned from 2D single-view images\nonly, encouraging researchers to discover semantic editing directions in its\nlatent space. However, previous methods face challenges in balancing quality,\nefficiency, and generalization. To solve the problem, we explore the\npossibility of introducing the strength of diffusion model into 3D-aware GANs.\nIn this paper, we present Face Clan, a fast and text-general approach for\ngenerating and manipulating 3D faces based on arbitrary attribute descriptions.\nTo achieve disentangled editing, we propose to diffuse on the latent space\nunder a pair of opposite prompts to estimate the mask indicating the region of\ninterest on latent codes. Based on the mask, we then apply denoising to the\nmasked latent codes to reveal the editing direction. Our method offers a\nprecisely controllable manipulation method, allowing users to intuitively\ncustomize regions of interest with the text description. Experiments\ndemonstrate the effectiveness and generalization of our Face Clan for various\npre-trained GANs. It offers an intuitive and wide application for text-guided\nface editing that contributes to the landscape of multimedia content creation.\n", "date": "2024-10-07", "categories": []} {"id": "2410.03417", "title": "Img2CAD: Conditioned 3D CAD Model Generation from Single Image with\n Structured Visual Geometry", "abstract": " In this paper, we propose Img2CAD, the first approach to our knowledge that\nuses 2D image inputs to generate CAD models with editable parameters. Unlike\nexisting AI methods for 3D model generation using text or image inputs often\nrely on mesh-based representations, which are incompatible with CAD tools and\nlack editability and fine control, Img2CAD enables seamless integration between\nAI-based 3D reconstruction and CAD software. We have identified an innovative\nintermediate representation called Structured Visual Geometry (SVG),\ncharacterized by vectorized wireframes extracted from objects. This\nrepresentation significantly enhances the performance of generating conditioned\nCAD models. Additionally, we introduce two new datasets to further support\nresearch in this area: ABC-mono, the largest known dataset comprising over\n200,000 3D CAD models with rendered images, and KOCAD, the first dataset\nfeaturing real-world captured objects alongside their ground truth CAD models,\nsupporting further research in conditioned CAD model generation.\n", "date": "2024-10-04", "categories": []} {"id": "2406.00093", "title": "Bootstrap3D: Improving Multi-view Diffusion Model with Synthetic Data", "abstract": " Recent years have witnessed remarkable progress in multi-view diffusion\nmodels for 3D content creation. However, there remains a significant gap in\nimage quality and prompt-following ability compared to 2D diffusion models. A\ncritical bottleneck is the scarcity of high-quality 3D objects with detailed\ncaptions. To address this challenge, we propose Bootstrap3D, a novel framework\nthat automatically generates an arbitrary quantity of multi-view images to\nassist in training multi-view diffusion models. Specifically, we introduce a\ndata generation pipeline that employs (1) 2D and video diffusion models to\ngenerate multi-view images based on constructed text prompts, and (2) our\nfine-tuned 3D-aware MV-LLaVA for filtering high-quality data and rewriting\ninaccurate captions. Leveraging this pipeline, we have generated 1 million\nhigh-quality synthetic multi-view images with dense descriptive captions to\naddress the shortage of high-quality 3D data. Furthermore, we present a\nTraining Timestep Reschedule (TTR) strategy that leverages the denoising\nprocess to learn multi-view consistency while maintaining the original 2D\ndiffusion prior. Extensive experiments demonstrate that Bootstrap3D can\ngenerate high-quality multi-view images with superior aesthetic quality,\nimage-text alignment, and maintained view consistency.\n", "date": "2024-05-31", "categories": []} {"id": "2406.13527", "title": "4K4DGen: Panoramic 4D Generation at 4K Resolution", "abstract": " The blooming of virtual reality and augmented reality (VR/AR) technologies\nhas driven an increasing demand for the creation of high-quality, immersive,\nand dynamic environments. However, existing generative techniques either focus\nsolely on dynamic objects or perform outpainting from a single perspective\nimage, failing to meet the requirements of VR/AR applications that need\nfree-viewpoint, 360$^{\\circ}$ virtual views where users can move in all\ndirections. In this work, we tackle the challenging task of elevating a single\npanorama to an immersive 4D experience. For the first time, we demonstrate the\ncapability to generate omnidirectional dynamic scenes with 360$^{\\circ}$ views\nat 4K (4096 $\\times$ 2048) resolution, thereby providing an immersive user\nexperience. Our method introduces a pipeline that facilitates natural scene\nanimations and optimizes a set of dynamic Gaussians using efficient splatting\ntechniques for real-time exploration. To overcome the lack of scene-scale\nannotated 4D data and models, especially in panoramic formats, we propose a\nnovel \\textbf{Panoramic Denoiser} that adapts generic 2D diffusion priors to\nanimate consistently in 360$^{\\circ}$ images, transforming them into panoramic\nvideos with dynamic scenes at targeted regions. Subsequently, we propose\n\\textbf{Dynamic Panoramic Lifting} to elevate the panoramic video into a 4D\nimmersive environment while preserving spatial and temporal consistency. By\ntransferring prior knowledge from 2D models in the perspective domain to the\npanoramic domain and the 4D lifting with spatial appearance and geometry\nregularization, we achieve high-quality Panorama-to-4D generation at a\nresolution of 4K for the first time.\n", "date": "2024-06-19", "categories": []} {"id": "2404.02148", "title": "Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of\n Video and Multi-view Diffusion Models", "abstract": " Recent advancements in 3D generation are predominantly propelled by\nimprovements in 3D-aware image diffusion models. These models are pretrained on\nInternet-scale image data and fine-tuned on massive 3D data, offering the\ncapability of producing highly consistent multi-view images. However, due to\nthe scarcity of synchronized multi-view video data, it remains challenging to\nadapt this paradigm to 4D generation directly. Despite that, the available\nvideo and 3D data are adequate for training video and multi-view diffusion\nmodels separately that can provide satisfactory dynamic and geometric priors\nrespectively. To take advantage of both, this paper presents Diffusion$^2$, a\nnovel framework for dynamic 3D content creation that reconciles the knowledge\nabout geometric consistency and temporal smoothness from these models to\ndirectly sample dense multi-view multi-frame images which can be employed to\noptimize continuous 4D representation. Specifically, we design a simple yet\neffective denoising strategy via score composition of pretrained video and\nmulti-view diffusion models based on the probability structure of the target\nimage array. To alleviate the potential conflicts between two heterogeneous\nscores, we further introduce variance-reducing sampling via interpolated steps,\nfacilitating smooth and stable generation. Owing to the high parallelism of the\nproposed image generation process and the efficiency of the modern 4D\nreconstruction pipeline, our framework can generate 4D content within few\nminutes. Notably, our method circumvents the reliance on expensive and\nhard-to-scale 4D data, thereby having the potential to benefit from the scaling\nof the foundation video and multi-view diffusion models. Extensive experiments\ndemonstrate the efficacy of our proposed framework in generating highly\nseamless and consistent 4D assets under various types of conditions.\n", "date": "2024-04-02", "categories": []} {"id": "2410.01226", "title": "Towards Native Generative Model for 3D Head Avatar", "abstract": " Creating 3D head avatars is a significant yet challenging task for many\napplicated scenarios. Previous studies have set out to learn 3D human head\ngenerative models using massive 2D image data. Although these models are highly\ngeneralizable for human appearance, their result models are not\n360$^\\circ$-renderable, and the predicted 3D geometry is unreliable. Therefore,\nsuch results cannot be used in VR, game modeling, and other scenarios that\nrequire 360$^\\circ$-renderable 3D head models. An intuitive idea is that 3D\nhead models with limited amount but high 3D accuracy are more reliable training\ndata for a high-quality 3D generative model. In this vein, we delve into how to\nlearn a native generative model for 360$^\\circ$ full head from a limited 3D\nhead dataset. Specifically, three major problems are studied: 1) how to\neffectively utilize various representations for generating the\n360$^\\circ$-renderable human head; 2) how to disentangle the appearance, shape,\nand motion of human faces to generate a 3D head model that can be edited by\nappearance and driven by motion; 3) and how to extend the generalization\ncapability of the generative model to support downstream tasks. Comprehensive\nexperiments are conducted to verify the effectiveness of the proposed model. We\nhope the proposed models and artist-designed dataset can inspire future\nresearch on learning native generative 3D head models from limited 3D datasets.\n", "date": "2024-10-02", "categories": []} {"id": "2410.01295", "title": "LaGeM: A Large Geometry Model for 3D Representation Learning and\n Diffusion", "abstract": " This paper introduces a novel hierarchical autoencoder that maps 3D models\ninto a highly compressed latent space. The hierarchical autoencoder is\nspecifically designed to tackle the challenges arising from large-scale\ndatasets and generative modeling using diffusion. Different from previous\napproaches that only work on a regular image or volume grid, our hierarchical\nautoencoder operates on unordered sets of vectors. Each level of the\nautoencoder controls different geometric levels of detail. We show that the\nmodel can be used to represent a wide range of 3D models while faithfully\nrepresenting high-resolution geometry details. The training of the new\narchitecture takes 0.70x time and 0.58x memory compared to the baseline. We\nalso explore how the new representation can be used for generative modeling.\nSpecifically, we propose a cascaded diffusion framework where each stage is\nconditioned on the previous stage. Our design extends existing cascaded designs\nfor image and volume grids to vector sets.\n", "date": "2024-10-02", "categories": []} {"id": "2312.02647", "title": "TPA3D: Triplane Attention for Fast Text-to-3D Generation", "abstract": " Due to the lack of large-scale text-3D correspondence data, recent text-to-3D\ngeneration works mainly rely on utilizing 2D diffusion models for synthesizing\n3D data. Since diffusion-based methods typically require significant\noptimization time for both training and inference, the use of GAN-based models\nwould still be desirable for fast 3D generation. In this work, we propose\nTriplane Attention for text-guided 3D generation (TPA3D), an end-to-end\ntrainable GAN-based deep learning model for fast text-to-3D generation. With\nonly 3D shape data and their rendered 2D images observed during training, our\nTPA3D is designed to retrieve detailed visual descriptions for synthesizing the\ncorresponding 3D mesh data. This is achieved by the proposed attention\nmechanisms on the extracted sentence and word-level text features. In our\nexperiments, we show that TPA3D generates high-quality 3D textured shapes\naligned with fine-grained descriptions, while impressive computation efficiency\ncan be observed.\n", "date": "2023-12-05", "categories": []} {"id": "2401.11078", "title": "UltrAvatar: A Realistic Animatable 3D Avatar Diffusion Model with\n Authenticity Guided Textures", "abstract": " Recent advances in 3D avatar generation have gained significant attentions.\nThese breakthroughs aim to produce more realistic animatable avatars, narrowing\nthe gap between virtual and real-world experiences. Most of existing works\nemploy Score Distillation Sampling (SDS) loss, combined with a differentiable\nrenderer and text condition, to guide a diffusion model in generating 3D\navatars. However, SDS often generates oversmoothed results with few facial\ndetails, thereby lacking the diversity compared with ancestral sampling. On the\nother hand, other works generate 3D avatar from a single image, where the\nchallenges of unwanted lighting effects, perspective views, and inferior image\nquality make them difficult to reliably reconstruct the 3D face meshes with the\naligned complete textures. In this paper, we propose a novel 3D avatar\ngeneration approach termed UltrAvatar with enhanced fidelity of geometry, and\nsuperior quality of physically based rendering (PBR) textures without unwanted\nlighting. To this end, the proposed approach presents a diffuse color\nextraction model and an authenticity guided texture diffusion model. The former\nremoves the unwanted lighting effects to reveal true diffuse colors so that the\ngenerated avatars can be rendered under various lighting conditions. The latter\nfollows two gradient-based guidances for generating PBR textures to render\ndiverse face-identity features and details better aligning with 3D mesh\ngeometry. We demonstrate the effectiveness and robustness of the proposed\nmethod, outperforming the state-of-the-art methods by a large margin in the\nexperiments.\n", "date": "2024-01-20", "categories": []} {"id": "2409.19989", "title": "RoCoTex: A Robust Method for Consistent Texture Synthesis with Diffusion\n Models", "abstract": " Text-to-texture generation has recently attracted increasing attention, but\nexisting methods often suffer from the problems of view inconsistencies,\napparent seams, and misalignment between textures and the underlying mesh. In\nthis paper, we propose a robust text-to-texture method for generating\nconsistent and seamless textures that are well aligned with the mesh. Our\nmethod leverages state-of-the-art 2D diffusion models, including SDXL and\nmultiple ControlNets, to capture structural features and intricate details in\nthe generated textures. The method also employs a symmetrical view synthesis\nstrategy combined with regional prompts for enhancing view consistency.\nAdditionally, it introduces novel texture blending and soft-inpainting\ntechniques, which significantly reduce the seam regions. Extensive experiments\ndemonstrate that our method outperforms existing state-of-the-art methods.\n", "date": "2024-09-30", "categories": []} {"id": "2409.18401", "title": "GenesisTex2: Stable, Consistent and High-Quality Text-to-Texture\n Generation", "abstract": " Large-scale text-guided image diffusion models have shown astonishing results\nin text-to-image (T2I) generation. However, applying these models to synthesize\ntextures for 3D geometries remains challenging due to the domain gap between 2D\nimages and textures on a 3D surface. Early works that used a\nprojecting-and-inpainting approach managed to preserve generation diversity but\noften resulted in noticeable artifacts and style inconsistencies. While recent\nmethods have attempted to address these inconsistencies, they often introduce\nother issues, such as blurring, over-saturation, or over-smoothing. To overcome\nthese challenges, we propose a novel text-to-texture synthesis framework that\nleverages pretrained diffusion models. We first introduce a local attention\nreweighing mechanism in the self-attention layers to guide the model in\nconcentrating on spatial-correlated patches across different views, thereby\nenhancing local details while preserving cross-view consistency. Additionally,\nwe propose a novel latent space merge pipeline, which further ensures\nconsistency across different viewpoints without sacrificing too much diversity.\nOur method significantly outperforms existing state-of-the-art techniques\nregarding texture consistency and visual quality, while delivering results much\nfaster than distillation-based methods. Importantly, our framework does not\nrequire additional training or fine-tuning, making it highly adaptable to a\nwide range of models available on public platforms.\n", "date": "2024-09-27", "categories": []} {"id": "2312.05295", "title": "Disentangled Clothed Avatar Generation from Text Descriptions", "abstract": " In this paper, we introduce a novel text-to-avatar generation method that\nseparately generates the human body and the clothes and allows high-quality\nanimation on the generated avatar. While recent advancements in text-to-avatar\ngeneration have yielded diverse human avatars from text prompts, these methods\ntypically combine all elements-clothes, hair, and body-into a single 3D\nrepresentation. Such an entangled approach poses challenges for downstream\ntasks like editing or animation. To overcome these limitations, we propose a\nnovel disentangled 3D avatar representation named Sequentially Offset-SMPL\n(SO-SMPL), building upon the SMPL model. SO-SMPL represents the human body and\nclothes with two separate meshes but associates them with offsets to ensure the\nphysical alignment between the body and the clothes. Then, we design a Score\nDistillation Sampling (SDS)-based distillation framework to generate the\nproposed SO-SMPL representation from text prompts. Our approach not only\nachieves higher texture and geometry quality and better semantic alignment with\ntext prompts, but also significantly improves the visual quality of character\nanimation, virtual try-on, and avatar editing. Project page:\nhttps://shanemankiw.github.io/SO-SMPL/.\n", "date": "2023-12-08", "categories": []} {"id": "2404.05705", "title": "Learning 3D-Aware GANs from Unposed Images with Template Feature Field", "abstract": " Collecting accurate camera poses of training images has been shown to well\nserve the learning of 3D-aware generative adversarial networks (GANs) yet can\nbe quite expensive in practice. This work targets learning 3D-aware GANs from\nunposed images, for which we propose to perform on-the-fly pose estimation of\ntraining images with a learned template feature field (TeFF). Concretely, in\naddition to a generative radiance field as in previous approaches, we ask the\ngenerator to also learn a field from 2D semantic features while sharing the\ndensity from the radiance field. Such a framework allows us to acquire a\ncanonical 3D feature template leveraging the dataset mean discovered by the\ngenerative model, and further efficiently estimate the pose parameters on real\ndata. Experimental results on various challenging datasets demonstrate the\nsuperiority of our approach over state-of-the-art alternatives from both the\nqualitative and the quantitative perspectives.\n", "date": "2024-04-08", "categories": []} {"id": "2409.18114", "title": "EdgeRunner: Auto-regressive Auto-encoder for Artistic Mesh Generation", "abstract": " Current auto-regressive mesh generation methods suffer from issues such as\nincompleteness, insufficient detail, and poor generalization. In this paper, we\npropose an Auto-regressive Auto-encoder (ArAE) model capable of generating\nhigh-quality 3D meshes with up to 4,000 faces at a spatial resolution of\n$512^3$. We introduce a novel mesh tokenization algorithm that efficiently\ncompresses triangular meshes into 1D token sequences, significantly enhancing\ntraining efficiency. Furthermore, our model compresses variable-length\ntriangular meshes into a fixed-length latent space, enabling training latent\ndiffusion models for better generalization. Extensive experiments demonstrate\nthe superior quality, diversity, and generalization capabilities of our model\nin both point cloud and image-conditioned mesh generation tasks.\n", "date": "2024-09-26", "categories": []} {"id": "2409.17106", "title": "Text2CAD: Generating Sequential CAD Models from Beginner-to-Expert Level\n Text Prompts", "abstract": " Prototyping complex computer-aided design (CAD) models in modern softwares\ncan be very time-consuming. This is due to the lack of intelligent systems that\ncan quickly generate simpler intermediate parts. We propose Text2CAD, the first\nAI framework for generating text-to-parametric CAD models using\ndesigner-friendly instructions for all skill levels. Furthermore, we introduce\na data annotation pipeline for generating text prompts based on natural\nlanguage instructions for the DeepCAD dataset using Mistral and LLaVA-NeXT. The\ndataset contains $\\sim170$K models and $\\sim660$K text annotations, from\nabstract CAD descriptions (e.g., generate two concentric cylinders) to detailed\nspecifications (e.g., draw two circles with center $(x,y)$ and radius $r_{1}$,\n$r_{2}$, and extrude along the normal by $d$...). Within the Text2CAD\nframework, we propose an end-to-end transformer-based auto-regressive network\nto generate parametric CAD models from input texts. We evaluate the performance\nof our model through a mixture of metrics, including visual quality, parametric\nprecision, and geometrical accuracy. Our proposed framework shows great\npotential in AI-aided design applications. Our source code and annotations will\nbe publicly available.\n", "date": "2024-09-25", "categories": []} {"id": "2409.17145", "title": "DreamWaltz-G: Expressive 3D Gaussian Avatars from Skeleton-Guided 2D\n Diffusion", "abstract": " Leveraging pretrained 2D diffusion models and score distillation sampling\n(SDS), recent methods have shown promising results for text-to-3D avatar\ngeneration. However, generating high-quality 3D avatars capable of expressive\nanimation remains challenging. In this work, we present DreamWaltz-G, a novel\nlearning framework for animatable 3D avatar generation from text. The core of\nthis framework lies in Skeleton-guided Score Distillation and Hybrid 3D\nGaussian Avatar representation. Specifically, the proposed skeleton-guided\nscore distillation integrates skeleton controls from 3D human templates into 2D\ndiffusion models, enhancing the consistency of SDS supervision in terms of view\nand human pose. This facilitates the generation of high-quality avatars,\nmitigating issues such as multiple faces, extra limbs, and blurring. The\nproposed hybrid 3D Gaussian avatar representation builds on the efficient 3D\nGaussians, combining neural implicit fields and parameterized 3D meshes to\nenable real-time rendering, stable SDS optimization, and expressive animation.\nExtensive experiments demonstrate that DreamWaltz-G is highly effective in\ngenerating and animating 3D avatars, outperforming existing methods in both\nvisual quality and animation expressiveness. Our framework further supports\ndiverse applications, including human video reenactment and multi-subject scene\ncomposition.\n", "date": "2024-09-25", "categories": []} {"id": "2303.13843", "title": "CompoNeRF: Text-guided Multi-object Compositional NeRF with Editable 3D\n Scene Layout", "abstract": " Text-to-3D form plays a crucial role in creating editable 3D scenes for\nAR/VR. Recent advances have shown promise in merging neural radiance fields\n(NeRFs) with pre-trained diffusion models for text-to-3D object generation.\nHowever, one enduring challenge is their inadequate capability to accurately\nparse and regenerate consistent multi-object environments. Specifically, these\nmodels encounter difficulties in accurately representing quantity and style\nprompted by multi-object texts, often resulting in a collapse of the rendering\nfidelity that fails to match the semantic intricacies. Moreover, amalgamating\nthese elements into a coherent 3D scene is a substantial challenge, stemming\nfrom generic distribution inherent in diffusion models. To tackle the issue of\n'guidance collapse' and further enhance scene consistency, we propose a novel\nframework, dubbed CompoNeRF, by integrating an editable 3D scene layout with\nobject-specific and scene-wide guidance mechanisms. It initiates by\ninterpreting a complex text into the layout populated with multiple NeRFs, each\npaired with a corresponding subtext prompt for precise object depiction. Next,\na tailored composition module seamlessly blends these NeRFs, promoting\nconsistency, while the dual-level text guidance reduces ambiguity and boosts\naccuracy. Noticeably, our composition design permits decomposition. This\nenables flexible scene editing and recomposition into new scenes based on the\nedited layout or text prompts. Utilizing the open-source Stable Diffusion\nmodel, CompoNeRF generates multi-object scenes with high fidelity. Remarkably,\nour framework achieves up to a \\textbf{54\\%} improvement by the multi-view CLIP\nscore metric. Our user study indicates that our method has significantly\nimproved semantic accuracy, multi-view consistency, and individual\nrecognizability for multi-object scene generation.\n", "date": "2023-03-24", "categories": []} {"id": "2406.09377", "title": "GGHead: Fast and Generalizable 3D Gaussian Heads", "abstract": " Learning 3D head priors from large 2D image collections is an important step\ntowards high-quality 3D-aware human modeling. A core requirement is an\nefficient architecture that scales well to large-scale datasets and large image\nresolutions. Unfortunately, existing 3D GANs struggle to scale to generate\nsamples at high resolutions due to their relatively slow train and render\nspeeds, and typically have to rely on 2D superresolution networks at the\nexpense of global 3D consistency. To address these challenges, we propose\nGenerative Gaussian Heads (GGHead), which adopts the recent 3D Gaussian\nSplatting representation within a 3D GAN framework. To generate a 3D\nrepresentation, we employ a powerful 2D CNN generator to predict Gaussian\nattributes in the UV space of a template head mesh. This way, GGHead exploits\nthe regularity of the template's UV layout, substantially facilitating the\nchallenging task of predicting an unstructured set of 3D Gaussians. We further\nimprove the geometric fidelity of the generated 3D representations with a novel\ntotal variation loss on rendered UV coordinates. Intuitively, this\nregularization encourages that neighboring rendered pixels should stem from\nneighboring Gaussians in the template's UV space. Taken together, our pipeline\ncan efficiently generate 3D heads trained only from single-view 2D image\nobservations. Our proposed framework matches the quality of existing 3D head\nGANs on FFHQ while being both substantially faster and fully 3D consistent. As\na result, we demonstrate real-time generation and rendering of high-quality\n3D-consistent heads at $1024^2$ resolution for the first time. Project Website:\nhttps://tobias-kirschstein.github.io/gghead\n", "date": "2024-06-13", "categories": []} {"id": "2401.09384", "title": "Diverse Part Synthesis for 3D Shape Creation", "abstract": " Methods that use neural networks for synthesizing 3D shapes in the form of a\npart-based representation have been introduced over the last few years. These\nmethods represent shapes as a graph or hierarchy of parts and enable a variety\nof applications such as shape sampling and reconstruction. However, current\nmethods do not allow easily regenerating individual shape parts according to\nuser preferences. In this paper, we investigate techniques that allow the user\nto generate multiple, diverse suggestions for individual parts. Specifically,\nwe experiment with multimodal deep generative models that allow sampling\ndiverse suggestions for shape parts and focus on models which have not been\nconsidered in previous work on shape synthesis. To provide a comparative study\nof these techniques, we introduce a method for synthesizing 3D shapes in a\npart-based representation and evaluate all the part suggestion techniques\nwithin this synthesis method. In our method, which is inspired by previous\nwork, shapes are represented as a set of parts in the form of implicit\nfunctions which are then positioned in space to form the final shape. Synthesis\nin this representation is enabled by a neural network architecture based on an\nimplicit decoder and a spatial transformer. We compare the various multimodal\ngenerative models by evaluating their performance in generating part\nsuggestions. Our contribution is to show with qualitative and quantitative\nevaluations which of the new techniques for multimodal part generation perform\nthe best and that a synthesis method based on the top-performing techniques\nallows the user to more finely control the parts that are generated in the 3D\nshapes while maintaining high shape fidelity when reconstructing shapes.\n", "date": "2024-01-17", "categories": []} {"id": "2409.05099", "title": "DreamMapping: High-Fidelity Text-to-3D Generation via Variational\n Distribution Mapping", "abstract": " Score Distillation Sampling (SDS) has emerged as a prevalent technique for\ntext-to-3D generation, enabling 3D content creation by distilling\nview-dependent information from text-to-2D guidance. However, they frequently\nexhibit shortcomings such as over-saturated color and excess smoothness. In\nthis paper, we conduct a thorough analysis of SDS and refine its formulation,\nfinding that the core design is to model the distribution of rendered images.\nFollowing this insight, we introduce a novel strategy called Variational\nDistribution Mapping (VDM), which expedites the distribution modeling process\nby regarding the rendered images as instances of degradation from\ndiffusion-based generation. This special design enables the efficient training\nof variational distribution by skipping the calculations of the Jacobians in\nthe diffusion U-Net. We also introduce timestep-dependent Distribution\nCoefficient Annealing (DCA) to further improve distilling precision. Leveraging\nVDM and DCA, we use Gaussian Splatting as the 3D representation and build a\ntext-to-3D generation framework. Extensive experiments and evaluations\ndemonstrate the capability of VDM and DCA to generate high-fidelity and\nrealistic assets with optimization efficiency.\n", "date": "2024-09-08", "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": "2404.10603", "title": "CorrespondentDream: Enhancing 3D Fidelity of Text-to-3D using Cross-View\n Correspondences", "abstract": " Leveraging multi-view diffusion models as priors for 3D optimization have\nalleviated the problem of 3D consistency, e.g., the Janus face problem or the\ncontent drift problem, in zero-shot text-to-3D models. However, the 3D\ngeometric fidelity of the output remains an unresolved issue; albeit the\nrendered 2D views are realistic, the underlying geometry may contain errors\nsuch as unreasonable concavities. In this work, we propose CorrespondentDream,\nan effective method to leverage annotation-free, cross-view correspondences\nyielded from the diffusion U-Net to provide additional 3D prior to the NeRF\noptimization process. We find that these correspondences are strongly\nconsistent with human perception, and by adopting it in our loss design, we are\nable to produce NeRF models with geometries that are more coherent with common\nsense, e.g., more smoothed object surface, yielding higher 3D fidelity. We\ndemonstrate the efficacy of our approach through various comparative\nqualitative results and a solid user study.\n", "date": "2024-04-16", "categories": []} {"id": "2409.11406", "title": "Phidias: A Generative Model for Creating 3D Content from Text, Image,\n and 3D Conditions with Reference-Augmented Diffusion", "abstract": " In 3D modeling, designers often use an existing 3D model as a reference to\ncreate new ones. This practice has inspired the development of Phidias, a novel\ngenerative model that uses diffusion for reference-augmented 3D generation.\nGiven an image, our method leverages a retrieved or user-provided 3D reference\nmodel to guide the generation process, thereby enhancing the generation\nquality, generalization ability, and controllability. Our model integrates\nthree key components: 1) meta-ControlNet that dynamically modulates the\nconditioning strength, 2) dynamic reference routing that mitigates misalignment\nbetween the input image and 3D reference, and 3) self-reference augmentations\nthat enable self-supervised training with a progressive curriculum.\nCollectively, these designs result in a clear improvement over existing\nmethods. Phidias establishes a unified framework for 3D generation using text,\nimage, and 3D conditions with versatile applications.\n", "date": "2024-09-17", "categories": []} {"id": "2409.08271", "title": "DreamBeast: Distilling 3D Fantastical Animals with Part-Aware Knowledge\n Transfer", "abstract": " We present DreamBeast, a novel method based on score distillation sampling\n(SDS) for generating fantastical 3D animal assets composed of distinct parts.\nExisting SDS methods often struggle with this generation task due to a limited\nunderstanding of part-level semantics in text-to-image diffusion models. While\nrecent diffusion models, such as Stable Diffusion 3, demonstrate a better\npart-level understanding, they are prohibitively slow and exhibit other common\nproblems associated with single-view diffusion models. DreamBeast overcomes\nthis limitation through a novel part-aware knowledge transfer mechanism. For\neach generated asset, we efficiently extract part-level knowledge from the\nStable Diffusion 3 model into a 3D Part-Affinity implicit representation. This\nenables us to instantly generate Part-Affinity maps from arbitrary camera\nviews, which we then use to modulate the guidance of a multi-view diffusion\nmodel during SDS to create 3D assets of fantastical animals. DreamBeast\nsignificantly enhances the quality of generated 3D creatures with\nuser-specified part compositions while reducing computational overhead, as\ndemonstrated by extensive quantitative and qualitative evaluations.\n", "date": "2024-09-12", "categories": []} {"id": "2409.08278", "title": "DreamHOI: Subject-Driven Generation of 3D Human-Object Interactions with\n Diffusion Priors", "abstract": " We present DreamHOI, a novel method for zero-shot synthesis of human-object\ninteractions (HOIs), enabling a 3D human model to realistically interact with\nany given object based on a textual description. This task is complicated by\nthe varying categories and geometries of real-world objects and the scarcity of\ndatasets encompassing diverse HOIs. To circumvent the need for extensive data,\nwe leverage text-to-image diffusion models trained on billions of image-caption\npairs. We optimize the articulation of a skinned human mesh using Score\nDistillation Sampling (SDS) gradients obtained from these models, which predict\nimage-space edits. However, directly backpropagating image-space gradients into\ncomplex articulation parameters is ineffective due to the local nature of such\ngradients. To overcome this, we introduce a dual implicit-explicit\nrepresentation of a skinned mesh, combining (implicit) neural radiance fields\n(NeRFs) with (explicit) skeleton-driven mesh articulation. During optimization,\nwe transition between implicit and explicit forms, grounding the NeRF\ngeneration while refining the mesh articulation. We validate our approach\nthrough extensive experiments, demonstrating its effectiveness in generating\nrealistic HOIs.\n", "date": "2024-09-12", "categories": []} {"id": "2406.01592", "title": "Text-guided Controllable Mesh Refinement for Interactive 3D Modeling", "abstract": " We propose a novel technique for adding geometric details to an input coarse\n3D mesh guided by a text prompt. Our method is composed of three stages. First,\nwe generate a single-view RGB image conditioned on the input coarse geometry\nand the input text prompt. This single-view image generation step allows the\nuser to pre-visualize the result and offers stronger conditioning for\nsubsequent multi-view generation. Second, we use our novel multi-view normal\ngeneration architecture to jointly generate six different views of the normal\nimages. The joint view generation reduces inconsistencies and leads to sharper\ndetails. Third, we optimize our mesh with respect to all views and generate a\nfine, detailed geometry as output. The resulting method produces an output\nwithin seconds and offers explicit user control over the coarse structure,\npose, and desired details of the resulting 3D mesh.\n", "date": "2024-06-03", "categories": []} {"id": "2409.07452", "title": "Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video\n Diffusion Models", "abstract": " Despite having tremendous progress in image-to-3D generation, existing\nmethods still struggle to produce multi-view consistent images with\nhigh-resolution textures in detail, especially in the paradigm of 2D diffusion\nthat lacks 3D awareness. In this work, we present High-resolution Image-to-3D\nmodel (Hi3D), a new video diffusion based paradigm that redefines a single\nimage to multi-view images as 3D-aware sequential image generation (i.e.,\norbital video generation). This methodology delves into the underlying temporal\nconsistency knowledge in video diffusion model that generalizes well to\ngeometry consistency across multiple views in 3D generation. Technically, Hi3D\nfirst empowers the pre-trained video diffusion model with 3D-aware prior\n(camera pose condition), yielding multi-view images with low-resolution texture\ndetails. A 3D-aware video-to-video refiner is learnt to further scale up the\nmulti-view images with high-resolution texture details. Such high-resolution\nmulti-view images are further augmented with novel views through 3D Gaussian\nSplatting, which are finally leveraged to obtain high-fidelity meshes via 3D\nreconstruction. Extensive experiments on both novel view synthesis and single\nview reconstruction demonstrate that our Hi3D manages to produce superior\nmulti-view consistency images with highly-detailed textures. Source code and\ndata are available at \\url{https://github.com/yanghb22-fdu/Hi3D-Official}.\n", "date": "2024-09-11", "categories": []} {"id": "2409.07454", "title": "DreamMesh: Jointly Manipulating and Texturing Triangle Meshes for\n Text-to-3D Generation", "abstract": " Learning radiance fields (NeRF) with powerful 2D diffusion models has\ngarnered popularity for text-to-3D generation. Nevertheless, the implicit 3D\nrepresentations of NeRF lack explicit modeling of meshes and textures over\nsurfaces, and such surface-undefined way may suffer from the issues, e.g.,\nnoisy surfaces with ambiguous texture details or cross-view inconsistency. To\nalleviate this, we present DreamMesh, a novel text-to-3D architecture that\npivots on well-defined surfaces (triangle meshes) to generate high-fidelity\nexplicit 3D model. Technically, DreamMesh capitalizes on a distinctive\ncoarse-to-fine scheme. In the coarse stage, the mesh is first deformed by\ntext-guided Jacobians and then DreamMesh textures the mesh with an interlaced\nuse of 2D diffusion models in a tuning free manner from multiple viewpoints. In\nthe fine stage, DreamMesh jointly manipulates the mesh and refines the texture\nmap, leading to high-quality triangle meshes with high-fidelity textured\nmaterials. Extensive experiments demonstrate that DreamMesh significantly\noutperforms state-of-the-art text-to-3D methods in faithfully generating 3D\ncontent with richer textual details and enhanced geometry. Our project page is\navailable at https://dreammesh.github.io.\n", "date": "2024-09-11", "categories": []} {"id": "2401.11067", "title": "Make-A-Shape: a Ten-Million-scale 3D Shape Model", "abstract": " Significant progress has been made in training large generative models for\nnatural language and images. Yet, the advancement of 3D generative models is\nhindered by their substantial resource demands for training, along with\ninefficient, non-compact, and less expressive representations. This paper\nintroduces Make-A-Shape, a new 3D generative model designed for efficient\ntraining on a vast scale, capable of utilizing 10 millions publicly-available\nshapes. Technical-wise, we first innovate a wavelet-tree representation to\ncompactly encode shapes by formulating the subband coefficient filtering scheme\nto efficiently exploit coefficient relations. We then make the representation\ngeneratable by a diffusion model by devising the subband coefficients packing\nscheme to layout the representation in a low-resolution grid. Further, we\nderive the subband adaptive training strategy to train our model to effectively\nlearn to generate coarse and detail wavelet coefficients. Last, we extend our\nframework to be controlled by additional input conditions to enable it to\ngenerate shapes from assorted modalities, e.g., single/multi-view images, point\nclouds, and low-resolution voxels. In our extensive set of experiments, we\ndemonstrate various applications, such as unconditional generation, shape\ncompletion, and conditional generation on a wide range of modalities. Our\napproach not only surpasses the state of the art in delivering high-quality\nresults but also efficiently generates shapes within a few seconds, often\nachieving this in just 2 seconds for most conditions. Our source code is\navailable at https://github.com/AutodeskAILab/Make-a-Shape.\n", "date": "2024-01-20", "categories": []} {"id": "2403.10094", "title": "RangeLDM: Fast Realistic LiDAR Point Cloud Generation", "abstract": " Autonomous driving demands high-quality LiDAR data, yet the cost of physical\nLiDAR sensors presents a significant scaling-up challenge. While recent efforts\nhave explored deep generative models to address this issue, they often consume\nsubstantial computational resources with slow generation speeds while suffering\nfrom a lack of realism. To address these limitations, we introduce RangeLDM, a\nnovel approach for rapidly generating high-quality range-view LiDAR point\nclouds via latent diffusion models. We achieve this by correcting range-view\ndata distribution for accurate projection from point clouds to range images via\nHough voting, which has a critical impact on generative learning. We then\ncompress the range images into a latent space with a variational autoencoder,\nand leverage a diffusion model to enhance expressivity. Additionally, we\ninstruct the model to preserve 3D structural fidelity by devising a\nrange-guided discriminator. Experimental results on KITTI-360 and nuScenes\ndatasets demonstrate both the robust expressiveness and fast speed of our LiDAR\npoint cloud generation.\n", "date": "2024-03-15", "categories": []} {"id": "2409.06322", "title": "G3PT: Unleash the power of Autoregressive Modeling in 3D Generation via\n Cross-scale Querying Transformer", "abstract": " Autoregressive transformers have revolutionized generative models in language\nprocessing and shown substantial promise in image and video generation.\nHowever, these models face significant challenges when extended to 3D\ngeneration tasks due to their reliance on next-token prediction to learn token\nsequences, which is incompatible with the unordered nature of 3D data. Instead\nof imposing an artificial order on 3D data, in this paper, we introduce G3PT, a\nscalable coarse-to-fine 3D generative model utilizing a cross-scale querying\ntransformer. The key is to map point-based 3D data into discrete tokens with\ndifferent levels of detail, naturally establishing a sequential relationship\nbetween different levels suitable for autoregressive modeling. Additionally,\nthe cross-scale querying transformer connects tokens globally across different\nlevels of detail without requiring an ordered sequence. Benefiting from this\napproach, G3PT features a versatile 3D generation pipeline that effortlessly\nsupports diverse conditional structures, enabling the generation of 3D shapes\nfrom various types of conditions. Extensive experiments demonstrate that G3PT\nachieves superior generation quality and generalization ability compared to\nprevious 3D generation methods. Most importantly, for the first time in 3D\ngeneration, scaling up G3PT reveals distinct power-law scaling behaviors.\n", "date": "2024-09-10", "categories": []} {"id": "2409.06620", "title": "MVGaussian: High-Fidelity text-to-3D Content Generation with Multi-View\n Guidance and Surface Densification", "abstract": " The field of text-to-3D content generation has made significant progress in\ngenerating realistic 3D objects, with existing methodologies like Score\nDistillation Sampling (SDS) offering promising guidance. However, these methods\noften encounter the \"Janus\" problem-multi-face ambiguities due to imprecise\nguidance. Additionally, while recent advancements in 3D gaussian splitting have\nshown its efficacy in representing 3D volumes, optimization of this\nrepresentation remains largely unexplored. This paper introduces a unified\nframework for text-to-3D content generation that addresses these critical gaps.\nOur approach utilizes multi-view guidance to iteratively form the structure of\nthe 3D model, progressively enhancing detail and accuracy. We also introduce a\nnovel densification algorithm that aligns gaussians close to the surface,\noptimizing the structural integrity and fidelity of the generated models.\nExtensive experiments validate our approach, demonstrating that it produces\nhigh-quality visual outputs with minimal time cost. Notably, our method\nachieves high-quality results within half an hour of training, offering a\nsubstantial efficiency gain over most existing methods, which require hours of\ntraining time to achieve comparable results.\n", "date": "2024-09-10", "categories": []} {"id": "2407.11398", "title": "Animate3D: Animating Any 3D Model with Multi-view Video Diffusion", "abstract": " Recent advances in 4D generation mainly focus on generating 4D content by\ndistilling pre-trained text or single-view image-conditioned models. It is\ninconvenient for them to take advantage of various off-the-shelf 3D assets with\nmulti-view attributes, and their results suffer from spatiotemporal\ninconsistency owing to the inherent ambiguity in the supervision signals. In\nthis work, we present Animate3D, a novel framework for animating any static 3D\nmodel. The core idea is two-fold: 1) We propose a novel multi-view video\ndiffusion model (MV-VDM) conditioned on multi-view renderings of the static 3D\nobject, which is trained on our presented large-scale multi-view video dataset\n(MV-Video). 2) Based on MV-VDM, we introduce a framework combining\nreconstruction and 4D Score Distillation Sampling (4D-SDS) to leverage the\nmulti-view video diffusion priors for animating 3D objects. Specifically, for\nMV-VDM, we design a new spatiotemporal attention module to enhance spatial and\ntemporal consistency by integrating 3D and video diffusion models.\nAdditionally, we leverage the static 3D model's multi-view renderings as\nconditions to preserve its identity. For animating 3D models, an effective\ntwo-stage pipeline is proposed: we first reconstruct motions directly from\ngenerated multi-view videos, followed by the introduced 4D-SDS to refine both\nappearance and motion. Benefiting from accurate motion learning, we could\nachieve straightforward mesh animation. Qualitative and quantitative\nexperiments demonstrate that Animate3D significantly outperforms previous\napproaches. Data, code, and models will be open-released.\n", "date": "2024-07-16", "categories": []} {"id": "2409.05380", "title": "Prim2Room: Layout-Controllable Room Mesh Generation from Primitives", "abstract": " We propose Prim2Room, a novel framework for controllable room mesh generation\nleveraging 2D layout conditions and 3D primitive retrieval to facilitate\nprecise 3D layout specification. Diverging from existing methods that lack\ncontrol and precision, our approach allows for detailed customization of\nroom-scale environments. To overcome the limitations of previous methods, we\nintroduce an adaptive viewpoint selection algorithm that allows the system to\ngenerate the furniture texture and geometry from more favorable views than\npredefined camera trajectories. Additionally, we employ non-rigid depth\nregistration to ensure alignment between generated objects and their\ncorresponding primitive while allowing for shape variations to maintain\ndiversity. Our method not only enhances the accuracy and aesthetic appeal of\ngenerated 3D scenes but also provides a user-friendly platform for detailed\nroom design.\n", "date": "2024-09-09", "categories": []} {"id": "2409.03718", "title": "Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with\n Image-Based Surface Representation", "abstract": " Generating high-quality 3D objects from textual descriptions remains a\nchallenging problem due to computational cost, the scarcity of 3D data, and\ncomplex 3D representations. We introduce Geometry Image Diffusion\n(GIMDiffusion), a novel Text-to-3D model that utilizes geometry images to\nefficiently represent 3D shapes using 2D images, thereby avoiding the need for\ncomplex 3D-aware architectures. By integrating a Collaborative Control\nmechanism, we exploit the rich 2D priors of existing Text-to-Image models such\nas Stable Diffusion. This enables strong generalization even with limited 3D\ntraining data (allowing us to use only high-quality training data) as well as\nretaining compatibility with guidance techniques such as IPAdapter. In short,\nGIMDiffusion enables the generation of 3D assets at speeds comparable to\ncurrent Text-to-Image models. The generated objects consist of semantically\nmeaningful, separate parts and include internal structures, enhancing both\nusability and versatility.\n", "date": "2024-09-05", "categories": []} {"id": "2208.12550", "title": "Training and Tuning Generative Neural Radiance Fields for\n Attribute-Conditional 3D-Aware Face Generation", "abstract": " Generative Neural Radiance Fields (GNeRF)-based 3D-aware GANs have showcased\nremarkable prowess in crafting high-fidelity images while upholding robust 3D\nconsistency, particularly face generation. However, specific existing models\nprioritize view consistency over disentanglement, leading to constrained\nsemantic or attribute control during the generation process. While many methods\nhave explored incorporating semantic masks or leveraging 3D Morphable Models\n(3DMM) priors to imbue models with semantic control, these methods often demand\ntraining from scratch, entailing significant computational overhead. In this\npaper, we propose a novel approach: a conditional GNeRF model that integrates\nspecific attribute labels as input, thus amplifying the controllability and\ndisentanglement capabilities of 3D-aware generative models. Our approach builds\nupon a pre-trained 3D-aware face model, and we introduce a Training as Init and\nOptimizing for Tuning (TRIOT) method to train a conditional normalized flow\nmodule to enable the facial attribute editing, then optimize the latent vector\nto improve attribute-editing precision further. Our extensive experiments\nsubstantiate the efficacy of our model, showcasing its ability to generate\nhigh-quality edits with enhanced view consistency while safeguarding non-target\nregions. The code for our model is publicly available at\nhttps://github.com/zhangqianhui/TT-GNeRF.\n", "date": "2022-08-26", "categories": []} {"id": "2405.11286", "title": "Motion Avatar: Generate Human and Animal Avatars with Arbitrary Motion", "abstract": " In recent years, there has been significant interest in creating 3D avatars\nand motions, driven by their diverse applications in areas like film-making,\nvideo games, AR/VR, and human-robot interaction. However, current efforts\nprimarily concentrate on either generating the 3D avatar mesh alone or\nproducing motion sequences, with integrating these two aspects proving to be a\npersistent challenge. Additionally, while avatar and motion generation\npredominantly target humans, extending these techniques to animals remains a\nsignificant challenge due to inadequate training data and methods. To bridge\nthese gaps, our paper presents three key contributions. Firstly, we proposed a\nnovel agent-based approach named Motion Avatar, which allows for the automatic\ngeneration of high-quality customizable human and animal avatars with motions\nthrough text queries. The method significantly advanced the progress in dynamic\n3D character generation. Secondly, we introduced a LLM planner that coordinates\nboth motion and avatar generation, which transforms a discriminative planning\ninto a customizable Q&A fashion. Lastly, we presented an animal motion dataset\nnamed Zoo-300K, comprising approximately 300,000 text-motion pairs across 65\nanimal categories and its building pipeline ZooGen, which serves as a valuable\nresource for the community. See project website\nhttps://steve-zeyu-zhang.github.io/MotionAvatar/\n", "date": "2024-05-18", "categories": []} {"id": "2409.00590", "title": "COMOGen: A Controllable Text-to-3D Multi-object Generation Framework", "abstract": " The controllability of 3D object generation methods is achieved through input\ntext. Existing text-to-3D object generation methods primarily focus on\ngenerating a single object based on a single object description. However, these\nmethods often face challenges in producing results that accurately correspond\nto our desired positions when the input text involves multiple objects. To\naddress the issue of controllability in generating multiple objects, this paper\nintroduces COMOGen, a COntrollable text-to-3D Multi-Object Generation\nframework. COMOGen enables the simultaneous generation of multiple 3D objects\nby the distillation of layout and multi-view prior knowledge. The framework\nconsists of three modules: the layout control module, the multi-view\nconsistency control module, and the 3D content enhancement module. Moreover, to\nintegrate these three modules as an integral framework, we propose Layout\nMulti-view Score Distillation, which unifies two prior knowledge and further\nenhances the diversity and quality of generated 3D content. Comprehensive\nexperiments demonstrate the effectiveness of our approach compared to the\nstate-of-the-art methods, which represents a significant step forward in\nenabling more controlled and versatile text-based 3D content generation.\n", "date": "2024-09-01", "categories": []} {"id": "2403.16210", "title": "Frankenstein: Generating Semantic-Compositional 3D Scenes in One\n Tri-Plane", "abstract": " We present Frankenstein, a diffusion-based framework that can generate\nsemantic-compositional 3D scenes in a single pass. Unlike existing methods that\noutput a single, unified 3D shape, Frankenstein simultaneously generates\nmultiple separated shapes, each corresponding to a semantically meaningful\npart. The 3D scene information is encoded in one single tri-plane tensor, from\nwhich multiple Singed Distance Function (SDF) fields can be decoded to\nrepresent the compositional shapes. During training, an auto-encoder compresses\ntri-planes into a latent space, and then the denoising diffusion process is\nemployed to approximate the distribution of the compositional scenes.\nFrankenstein demonstrates promising results in generating room interiors as\nwell as human avatars with automatically separated parts. The generated scenes\nfacilitate many downstream applications, such as part-wise re-texturing, object\nrearrangement in the room or avatar cloth re-targeting. Our project page is\navailable at: https://wolfball.github.io/frankenstein/.\n", "date": "2024-03-24", "categories": []} {"id": "2403.06702", "title": "Fast Text-to-3D-Aware Face Generation and Manipulation via Direct\n Cross-modal Mapping and Geometric Regularization", "abstract": " Text-to-3D-aware face (T3D Face) generation and manipulation is an emerging\nresearch hot spot in machine learning, which still suffers from low efficiency\nand poor quality. In this paper, we propose an End-to-End Efficient and\nEffective network for fast and accurate T3D face generation and manipulation,\ntermed $E^3$-FaceNet. Different from existing complex generation paradigms,\n$E^3$-FaceNet resorts to a direct mapping from text instructions to 3D-aware\nvisual space. We introduce a novel Style Code Enhancer to enhance cross-modal\nsemantic alignment, alongside an innovative Geometric Regularization objective\nto maintain consistency across multi-view generations. Extensive experiments on\nthree benchmark datasets demonstrate that $E^3$-FaceNet can not only achieve\npicture-like 3D face generation and manipulation, but also improve inference\nspeed by orders of magnitudes. For instance, compared with Latent3D,\n$E^3$-FaceNet speeds up the five-view generations by almost 470 times, while\nstill exceeding in generation quality. Our code is released at\nhttps://github.com/Aria-Zhangjl/E3-FaceNet.\n", "date": "2024-03-11", "categories": []} {"id": "2408.15995", "title": "TEDRA: Text-based Editing of Dynamic and Photoreal Actors", "abstract": " Over the past years, significant progress has been made in creating\nphotorealistic and drivable 3D avatars solely from videos of real humans.\nHowever, a core remaining challenge is the fine-grained and user-friendly\nediting of clothing styles by means of textual descriptions. To this end, we\npresent TEDRA, the first method allowing text-based edits of an avatar, which\nmaintains the avatar's high fidelity, space-time coherency, as well as\ndynamics, and enables skeletal pose and view control. We begin by training a\nmodel to create a controllable and high-fidelity digital replica of the real\nactor. Next, we personalize a pretrained generative diffusion model by\nfine-tuning it on various frames of the real character captured from different\ncamera angles, ensuring the digital representation faithfully captures the\ndynamics and movements of the real person. This two-stage process lays the\nfoundation for our approach to dynamic human avatar editing. Utilizing this\npersonalized diffusion model, we modify the dynamic avatar based on a provided\ntext prompt using our Personalized Normal Aligned Score Distillation Sampling\n(PNA-SDS) within a model-based guidance framework. Additionally, we propose a\ntime step annealing strategy to ensure high-quality edits. Our results\ndemonstrate a clear improvement over prior work in functionality and visual\nquality.\n", "date": "2024-08-28", "categories": []} {"id": "2408.14860", "title": "DiffSurf: A Transformer-based Diffusion Model for Generating and\n Reconstructing 3D Surfaces in Pose", "abstract": " This paper presents DiffSurf, a transformer-based denoising diffusion model\nfor generating and reconstructing 3D surfaces. Specifically, we design a\ndiffusion transformer architecture that predicts noise from noisy 3D surface\nvertices and normals. With this architecture, DiffSurf is able to generate 3D\nsurfaces in various poses and shapes, such as human bodies, hands, animals and\nman-made objects. Further, DiffSurf is versatile in that it can address various\n3D downstream tasks including morphing, body shape variation and 3D human mesh\nfitting to 2D keypoints. Experimental results on 3D human model benchmarks\ndemonstrate that DiffSurf can generate shapes with greater diversity and higher\nquality than previous generative models. Furthermore, when applied to the task\nof single-image 3D human mesh recovery, DiffSurf achieves accuracy comparable\nto prior techniques at a near real-time rate.\n", "date": "2024-08-27", "categories": []} {"id": "2403.12848", "title": "Planner3D: LLM-enhanced graph prior meets 3D indoor scene explicit\n regularization", "abstract": " Compositional 3D scene synthesis has diverse applications across a spectrum\nof industries such as robotics, films, and video games, as it closely mirrors\nthe complexity of real-world multi-object environments. Conventional works\ntypically employ shape retrieval based frameworks which naturally suffer from\nlimited shape diversity. Recent progresses have been made in object shape\ngeneration with generative models such as diffusion models, which increases the\nshape fidelity. However, these approaches separately treat 3D shape generation\nand layout generation. The synthesized scenes are usually hampered by layout\ncollision, which suggests that the scene-level fidelity is still\nunder-explored. In this paper, we aim at generating realistic and reasonable 3D\nindoor scenes from scene graph. To enrich the priors of the given scene graph\ninputs, large language model is utilized to aggregate the global-wise features\nwith local node-wise and edge-wise features. With a unified graph encoder,\ngraph features are extracted to guide joint layout-shape generation. Additional\nregularization is introduced to explicitly constrain the produced 3D layouts.\nBenchmarked on the SG-FRONT dataset, our method achieves better 3D scene\nsynthesis, especially in terms of scene-level fidelity. The source code will be\nreleased after publication.\n", "date": "2024-03-19", "categories": []} {"id": "2406.09913", "title": "OpenECAD: An Efficient Visual Language Model for Editable 3D-CAD Design", "abstract": " Computer-aided design (CAD) tools are utilized in the manufacturing industry\nfor modeling everything from cups to spacecraft. These programs are complex to\nuse and typically require years of training and experience to master.\nStructured and well-constrained 2D sketches and 3D constructions are crucial\ncomponents of CAD modeling. A well-executed CAD model can be seamlessly\nintegrated into the manufacturing process, thereby enhancing production\nefficiency. Deep generative models of 3D shapes and 3D object reconstruction\nmodels have garnered significant research interest. However, most of these\nmodels produce discrete forms of 3D objects that are not editable. Moreover,\nthe few models based on CAD operations often have substantial input\nrestrictions. In this work, we fine-tuned pre-trained models to create OpenECAD\nmodels (0.55B, 0.89B, 2.4B and 3.1B), leveraging the visual, logical, coding,\nand general capabilities of visual language models. OpenECAD models can process\nimages of 3D designs as input and generate highly structured 2D sketches and 3D\nconstruction commands, ensuring that the designs are editable. These outputs\ncan be directly used with existing CAD tools' APIs to generate project files.\nTo train our network, we created a series of OpenECAD datasets. These datasets\nare derived from existing public CAD datasets, adjusted and augmented to meet\nthe specific requirements of vision language model (VLM) training.\nAdditionally, we have introduced an approach that utilizes dependency\nrelationships to define and generate sketches, further enriching the content\nand functionality of the datasets.\n", "date": "2024-06-14", "categories": []} {"id": "2408.13149", "title": "Focus on Neighbors and Know the Whole: Towards Consistent Dense\n Multiview Text-to-Image Generator for 3D Creation", "abstract": " Generating dense multiview images from text prompts is crucial for creating\nhigh-fidelity 3D assets. Nevertheless, existing methods struggle with\nspace-view correspondences, resulting in sparse and low-quality outputs. In\nthis paper, we introduce CoSER, a novel consistent dense Multiview\nText-to-Image Generator for Text-to-3D, achieving both efficiency and quality\nby meticulously learning neighbor-view coherence and further alleviating\nambiguity through the swift traversal of all views. For achieving neighbor-view\nconsistency, each viewpoint densely interacts with adjacent viewpoints to\nperceive the global spatial structure, and aggregates information along motion\npaths explicitly defined by physical principles to refine details. To further\nenhance cross-view consistency and alleviate content drift, CoSER rapidly scan\nall views in spiral bidirectional manner to aware holistic information and then\nscores each point based on semantic material. Subsequently, we conduct weighted\ndown-sampling along the spatial dimension based on scores, thereby facilitating\nprominent information fusion across all views with lightweight computation.\nTechnically, the core module is built by integrating the attention mechanism\nwith a selective state space model, exploiting the robust learning capabilities\nof the former and the low overhead of the latter. Extensive evaluation shows\nthat CoSER is capable of producing dense, high-fidelity, content-consistent\nmultiview images that can be flexibly integrated into various 3D generation\nmodels.\n", "date": "2024-08-23", "categories": []} {"id": "2408.13674", "title": "GenCA: A Text-conditioned Generative Model for Realistic and Drivable\n Codec Avatars", "abstract": " Photo-realistic and controllable 3D avatars are crucial for various\napplications such as virtual and mixed reality (VR/MR), telepresence, gaming,\nand film production. Traditional methods for avatar creation often involve\ntime-consuming scanning and reconstruction processes for each avatar, which\nlimits their scalability. Furthermore, these methods do not offer the\nflexibility to sample new identities or modify existing ones. On the other\nhand, by learning a strong prior from data, generative models provide a\npromising alternative to traditional reconstruction methods, easing the time\nconstraints for both data capture and processing. Additionally, generative\nmethods enable downstream applications beyond reconstruction, such as editing\nand stylization. Nonetheless, the research on generative 3D avatars is still in\nits infancy, and therefore current methods still have limitations such as\ncreating static avatars, lacking photo-realism, having incomplete facial\ndetails, or having limited drivability. To address this, we propose a\ntext-conditioned generative model that can generate photo-realistic facial\navatars of diverse identities, with more complete details like hair, eyes and\nmouth interior, and which can be driven through a powerful non-parametric\nlatent expression space. Specifically, we integrate the generative and editing\ncapabilities of latent diffusion models with a strong prior model for avatar\nexpression driving.\n Our model can generate and control high-fidelity avatars, even those\nout-of-distribution. We also highlight its potential for downstream\napplications, including avatar editing and single-shot avatar reconstruction.\n", "date": "2024-08-24", "categories": []} {"id": "2312.01129", "title": "ControlDreamer: Blending Geometry and Style in Text-to-3D", "abstract": " Recent advancements in text-to-3D generation have significantly contributed\nto the automation and democratization of 3D content creation. Building upon\nthese developments, we aim to address the limitations of current methods in\nblending geometries and styles in text-to-3D generation. We introduce\nmulti-view ControlNet, a novel depth-aware multi-view diffusion model trained\non generated datasets from a carefully curated text corpus. Our multi-view\nControlNet is then integrated into our two-stage pipeline, ControlDreamer,\nenabling text-guided generation of stylized 3D models. Additionally, we present\na comprehensive benchmark for 3D style editing, encompassing a broad range of\nsubjects, including objects, animals, and characters, to further facilitate\nresearch on diverse 3D generation. Our comparative analysis reveals that this\nnew pipeline outperforms existing text-to-3D methods as evidenced by human\nevaluations and CLIP score metrics. Project page:\nhttps://controldreamer.github.io\n", "date": "2023-12-02", "categories": []} {"id": "2402.05919", "title": "Collaborative Control for Geometry-Conditioned PBR Image Generation", "abstract": " Graphics pipelines require physically-based rendering (PBR) materials, yet\ncurrent 3D content generation approaches are built on RGB models. We propose to\nmodel the PBR image distribution directly, avoiding photometric inaccuracies in\nRGB generation and the inherent ambiguity in extracting PBR from RGB. As\nexisting paradigms for cross-modal fine-tuning are not suited for PBR\ngeneration due to both a lack of data and the high dimensionality of the output\nmodalities, we propose to train a new PBR model that is tightly linked to a\nfrozen RGB model using a novel cross-network communication paradigm. As the\nbase RGB model is fully frozen, the proposed method retains its general\nperformance and remains compatible with e.g. IPAdapters for that base model.\n", "date": "2024-02-08", "categories": []} {"id": "2301.11631", "title": "HyperNeRFGAN: Hypernetwork approach to 3D NeRF GAN", "abstract": " The recent surge in popularity of deep generative models for 3D objects has\nhighlighted the need for more efficient training methods, particularly given\nthe difficulties associated with training with conventional 3D representations,\nsuch as voxels or point clouds. Neural Radiance Fields (NeRFs), which provide\nthe current benchmark in terms of quality for the generation of novel views of\ncomplex 3D scenes from a limited set of 2D images, represent a promising\nsolution to this challenge. However, the training of these models requires the\nknowledge of the respective camera positions from which the images were viewed.\nIn this paper, we overcome this limitation by introducing HyperNeRFGAN, a\nGenerative Adversarial Network (GAN) architecture employing a hypernetwork\nparadigm to transform a Gaussian noise into the weights of a NeRF architecture\nthat does not utilize viewing directions in its training phase. Consequently,\nas evidenced by the findings of our experimental study, the proposed model,\ndespite its notable simplicity in comparison to existing state-of-the-art\nalternatives, demonstrates superior performance on a diverse range of image\ndatasets where camera position estimation is challenging, particularly in the\ncontext of medical data.\n", "date": "2023-01-27", "categories": []} {"id": "2407.02034", "title": "TrAME: Trajectory-Anchored Multi-View Editing for Text-Guided 3D\n Gaussian Splatting Manipulation", "abstract": " Despite significant strides in the field of 3D scene editing, current methods\nencounter substantial challenge, particularly in preserving 3D consistency in\nmulti-view editing process. To tackle this challenge, we propose a progressive\n3D editing strategy that ensures multi-view consistency via a\nTrajectory-Anchored Scheme (TAS) with a dual-branch editing mechanism.\nSpecifically, TAS facilitates a tightly coupled iterative process between 2D\nview editing and 3D updating, preventing error accumulation yielded from\ntext-to-image process. Additionally, we explore the relationship between\noptimization-based methods and reconstruction-based methods, offering a unified\nperspective for selecting superior design choice, supporting the rationale\nbehind the designed TAS. We further present a tuning-free View-Consistent\nAttention Control (VCAC) module that leverages cross-view semantic and\ngeometric reference from the source branch to yield aligned views from the\ntarget branch during the editing of 2D views. To validate the effectiveness of\nour method, we analyze 2D examples to demonstrate the improved consistency with\nthe VCAC module. Further extensive quantitative and qualitative results in\ntext-guided 3D scene editing indicate that our method achieves superior editing\nquality compared to state-of-the-art methods. We will make the complete\ncodebase publicly available following the conclusion of the review process.\n", "date": "2024-07-02", "categories": []} {"id": "2408.11357", "title": "HumanCoser: Layered 3D Human Generation via Semantic-Aware Diffusion\n Model", "abstract": " This paper aims to generate physically-layered 3D humans from text prompts.\nExisting methods either generate 3D clothed humans as a whole or support only\ntight and simple clothing generation, which limits their applications to\nvirtual try-on and part-level editing. To achieve physically-layered 3D human\ngeneration with reusable and complex clothing, we propose a novel layer-wise\ndressed human representation based on a physically-decoupled diffusion model.\nSpecifically, to achieve layer-wise clothing generation, we propose a\ndual-representation decoupling framework for generating clothing decoupled from\nthe human body, in conjunction with an innovative multi-layer fusion volume\nrendering method. To match the clothing with different body shapes, we propose\nan SMPL-driven implicit field deformation network that enables the free\ntransfer and reuse of clothing. Extensive experiments demonstrate that our\napproach not only achieves state-of-the-art layered 3D human generation with\ncomplex clothing but also supports virtual try-on and layered human animation.\n", "date": "2024-08-21", "categories": []} {"id": "2408.11801", "title": "Story3D-Agent: Exploring 3D Storytelling Visualization with Large\n Language Models", "abstract": " Traditional visual storytelling is complex, requiring specialized knowledge\nand substantial resources, yet often constrained by human creativity and\ncreation precision. While Large Language Models (LLMs) enhance visual\nstorytelling, current approaches often limit themselves to 2D visuals or\noversimplify stories through motion synthesis and behavioral simulation,\nfailing to create comprehensive, multi-dimensional narratives. To this end, we\npresent Story3D-Agent, a pioneering approach that leverages the capabilities of\nLLMs to transform provided narratives into 3D-rendered visualizations. By\nintegrating procedural modeling, our approach enables precise control over\nmulti-character actions and motions, as well as diverse decorative elements,\nensuring the long-range and dynamic 3D representation. Furthermore, our method\nsupports narrative extension through logical reasoning, ensuring that generated\ncontent remains consistent with existing conditions. We have thoroughly\nevaluated our Story3D-Agent to validate its effectiveness, offering a basic\nframework to advance 3D story representation.\n", "date": "2024-08-21", "categories": []} {"id": "2405.20327", "title": "GECO: Generative Image-to-3D within a SECOnd", "abstract": " Recent years have seen significant advancements in 3D generation. While\nmethods like score distillation achieve impressive results, they often require\nextensive per-scene optimization, which limits their time efficiency. On the\nother hand, reconstruction-based approaches are more efficient but tend to\ncompromise quality due to their limited ability to handle uncertainty. We\nintroduce GECO, a novel method for high-quality 3D generative modeling that\noperates within a second. Our approach addresses the prevalent issues of\nuncertainty and inefficiency in existing methods through a two-stage approach.\nIn the first stage, we train a single-step multi-view generative model with\nscore distillation. Then, a second-stage distillation is applied to address the\nchallenge of view inconsistency in the multi-view generation. This two-stage\nprocess ensures a balanced approach to 3D generation, optimizing both quality\nand efficiency. Our comprehensive experiments demonstrate that GECO achieves\nhigh-quality image-to-3D mesh generation with an unprecedented level of\nefficiency. We will make the code and model publicly available.\n", "date": "2024-05-30", "categories": []} {"id": "2406.18159", "title": "Human-Aware 3D Scene Generation with Spatially-constrained Diffusion\n Models", "abstract": " Generating 3D scenes from human motion sequences supports numerous\napplications, including virtual reality and architectural design. However,\nprevious auto-regression-based human-aware 3D scene generation methods have\nstruggled to accurately capture the joint distribution of multiple objects and\ninput humans, often resulting in overlapping object generation in the same\nspace. To address this limitation, we explore the potential of diffusion models\nthat simultaneously consider all input humans and the floor plan to generate\nplausible 3D scenes. Our approach not only satisfies all input human\ninteractions but also adheres to spatial constraints with the floor plan.\nFurthermore, we introduce two spatial collision guidance mechanisms:\nhuman-object collision avoidance and object-room boundary constraints. These\nmechanisms help avoid generating scenes that conflict with human motions while\nrespecting layout constraints. To enhance the diversity and accuracy of\nhuman-guided scene generation, we have developed an automated pipeline that\nimproves the variety and plausibility of human-object interactions in the\nexisting 3D FRONT HUMAN dataset. Extensive experiments on both synthetic and\nreal-world datasets demonstrate that our framework can generate more natural\nand plausible 3D scenes with precise human-scene interactions, while\nsignificantly reducing human-object collisions compared to previous\nstate-of-the-art methods. Our code and data will be made publicly available\nupon publication of this work.\n", "date": "2024-06-26", "categories": []} {"id": "2408.05477", "title": "Scene123: One Prompt to 3D Scene Generation via Video-Assisted and\n Consistency-Enhanced MAE", "abstract": " As Artificial Intelligence Generated Content (AIGC) advances, a variety of\nmethods have been developed to generate text, images, videos, and 3D objects\nfrom single or multimodal inputs, contributing efforts to emulate human-like\ncognitive content creation. However, generating realistic large-scale scenes\nfrom a single input presents a challenge due to the complexities involved in\nensuring consistency across extrapolated views generated by models. Benefiting\nfrom recent video generation models and implicit neural representations, we\npropose Scene123, a 3D scene generation model, that not only ensures realism\nand diversity through the video generation framework but also uses implicit\nneural fields combined with Masked Autoencoders (MAE) to effectively ensures\nthe consistency of unseen areas across views. Specifically, we initially warp\nthe input image (or an image generated from text) to simulate adjacent views,\nfilling the invisible areas with the MAE model. However, these filled images\nusually fail to maintain view consistency, thus we utilize the produced views\nto optimize a neural radiance field, enhancing geometric consistency.\n Moreover, to further enhance the details and texture fidelity of generated\nviews, we employ a GAN-based Loss against images derived from the input image\nthrough the video generation model. Extensive experiments demonstrate that our\nmethod can generate realistic and consistent scenes from a single prompt. Both\nqualitative and quantitative results indicate that our approach surpasses\nexisting state-of-the-art methods. We show encourage video examples at\nhttps://yiyingyang12.github.io/Scene123.github.io/.\n", "date": "2024-08-10", "categories": []} {"id": "2408.10935", "title": "Large Point-to-Gaussian Model for Image-to-3D Generation", "abstract": " Recently, image-to-3D approaches have significantly advanced the generation\nquality and speed of 3D assets based on large reconstruction models,\nparticularly 3D Gaussian reconstruction models. Existing large 3D Gaussian\nmodels directly map 2D image to 3D Gaussian parameters, while regressing 2D\nimage to 3D Gaussian representations is challenging without 3D priors. In this\npaper, we propose a large Point-to-Gaussian model, that inputs the initial\npoint cloud produced from large 3D diffusion model conditional on 2D image to\ngenerate the Gaussian parameters, for image-to-3D generation. The point cloud\nprovides initial 3D geometry prior for Gaussian generation, thus significantly\nfacilitating image-to-3D Generation. Moreover, we present the\n\\textbf{A}ttention mechanism, \\textbf{P}rojection mechanism, and \\textbf{P}oint\nfeature extractor, dubbed as \\textbf{APP} block, for fusing the image features\nwith point cloud features. The qualitative and quantitative experiments\nextensively demonstrate the effectiveness of the proposed approach on GSO and\nObjaverse datasets, and show the proposed method achieves state-of-the-art\nperformance.\n", "date": "2024-08-20", "categories": []} {"id": "2311.12063", "title": "DatasetNeRF: Efficient 3D-aware Data Factory with Generative Radiance\n Fields", "abstract": " Progress in 3D computer vision tasks demands a huge amount of data, yet\nannotating multi-view images with 3D-consistent annotations, or point clouds\nwith part segmentation is both time-consuming and challenging. This paper\nintroduces DatasetNeRF, a novel approach capable of generating infinite,\nhigh-quality 3D-consistent 2D annotations alongside 3D point cloud\nsegmentations, while utilizing minimal 2D human-labeled annotations.\nSpecifically, we leverage the strong semantic prior within a 3D generative\nmodel to train a semantic decoder, requiring only a handful of fine-grained\nlabeled samples. Once trained, the decoder efficiently generalizes across the\nlatent space, enabling the generation of infinite data. The generated data is\napplicable across various computer vision tasks, including video segmentation\nand 3D point cloud segmentation. Our approach not only surpasses baseline\nmodels in segmentation quality, achieving superior 3D consistency and\nsegmentation precision on individual images, but also demonstrates versatility\nby being applicable to both articulated and non-articulated generative models.\nFurthermore, we explore applications stemming from our approach, such as\n3D-aware semantic editing and 3D inversion.\n", "date": "2023-11-18", "categories": []} {"id": "2408.08342", "title": "CT4D: Consistent Text-to-4D Generation with Animatable Meshes", "abstract": " Text-to-4D generation has recently been demonstrated viable by integrating a\n2D image diffusion model with a video diffusion model. However, existing models\ntend to produce results with inconsistent motions and geometric structures over\ntime. To this end, we present a novel framework, coined CT4D, which directly\noperates on animatable meshes for generating consistent 4D content from\narbitrary user-supplied prompts. The primary challenges of our mesh-based\nframework involve stably generating a mesh with details that align with the\ntext prompt while directly driving it and maintaining surface continuity. Our\nCT4D framework incorporates a unique Generate-Refine-Animate (GRA) algorithm to\nenhance the creation of text-aligned meshes. To improve surface continuity, we\ndivide a mesh into several smaller regions and implement a uniform driving\nfunction within each area. Additionally, we constrain the animating stage with\na rigidity regulation to ensure cross-region continuity. Our experimental\nresults, both qualitative and quantitative, demonstrate that our CT4D framework\nsurpasses existing text-to-4D techniques in maintaining interframe consistency\nand preserving global geometry. Furthermore, we showcase that this enhanced\nrepresentation inherently possesses the capability for combinational 4D\ngeneration and texture editing.\n", "date": "2024-08-15", "categories": []} {"id": "2406.17342", "title": "Masked Generative Extractor for Synergistic Representation and 3D\n Generation of Point Clouds", "abstract": " Representation and generative learning, as reconstruction-based methods, have\ndemonstrated their potential for mutual reinforcement across various domains.\nIn the field of point cloud processing, although existing studies have adopted\ntraining strategies from generative models to enhance representational\ncapabilities, these methods are limited by their inability to genuinely\ngenerate 3D shapes. To explore the benefits of deeply integrating 3D\nrepresentation learning and generative learning, we propose an innovative\nframework called \\textit{Point-MGE}. Specifically, this framework first\nutilizes a vector quantized variational autoencoder to reconstruct a neural\nfield representation of 3D shapes, thereby learning discrete semantic features\nof point patches. Subsequently, we design a sliding masking ratios to smooth\nthe transition from representation learning to generative learning. Moreover,\nour method demonstrates strong generalization capability in learning\nhigh-capacity models, achieving new state-of-the-art performance across\nmultiple downstream tasks. In shape classification, Point-MGE achieved an\naccuracy of 94.2% (+1.0%) on the ModelNet40 dataset and 92.9% (+5.5%) on the\nScanObjectNN dataset. Experimental results also confirmed that Point-MGE can\ngenerate high-quality 3D shapes in both unconditional and conditional settings.\n", "date": "2024-06-25", "categories": []} {"id": "2404.03736", "title": "SC4D: Sparse-Controlled Video-to-4D Generation and Motion Transfer", "abstract": " Recent advances in 2D/3D generative models enable the generation of dynamic\n3D objects from a single-view video. Existing approaches utilize score\ndistillation sampling to form the dynamic scene as dynamic NeRF or dense 3D\nGaussians. However, these methods struggle to strike a balance among reference\nview alignment, spatio-temporal consistency, and motion fidelity under\nsingle-view conditions due to the implicit nature of NeRF or the intricate\ndense Gaussian motion prediction. To address these issues, this paper proposes\nan efficient, sparse-controlled video-to-4D framework named SC4D, that\ndecouples motion and appearance to achieve superior video-to-4D generation.\nMoreover, we introduce Adaptive Gaussian (AG) initialization and Gaussian\nAlignment (GA) loss to mitigate shape degeneration issue, ensuring the fidelity\nof the learned motion and shape. Comprehensive experimental results demonstrate\nthat our method surpasses existing methods in both quality and efficiency. In\naddition, facilitated by the disentangled modeling of motion and appearance of\nSC4D, we devise a novel application that seamlessly transfers the learned\nmotion onto a diverse array of 4D entities according to textual descriptions.\n", "date": "2024-04-04", "categories": []} {"id": "2406.06432", "title": "SYM3D: Learning Symmetric Triplanes for Better 3D-Awareness of GANs", "abstract": " Despite the growing success of 3D-aware GANs, which can be trained on 2D\nimages to generate high-quality 3D assets, they still rely on multi-view images\nwith camera annotations to synthesize sufficient details from all viewing\ndirections. However, the scarce availability of calibrated multi-view image\ndatasets, especially in comparison to single-view images, has limited the\npotential of 3D GANs. Moreover, while bypassing camera pose annotations with a\ncamera distribution constraint reduces dependence on exact camera parameters,\nit still struggles to generate a consistent orientation of 3D assets. To this\nend, we propose SYM3D, a novel 3D-aware GAN designed to leverage the prevalent\nreflectional symmetry structure found in natural and man-made objects,\nalongside a proposed view-aware spatial attention mechanism in learning the 3D\nrepresentation. We evaluate SYM3D on both synthetic (ShapeNet Chairs, Cars, and\nAirplanes) and real-world datasets (ABO-Chair), demonstrating its superior\nperformance in capturing detailed geometry and texture, even when trained on\nonly single-view images. Finally, we demonstrate the effectiveness of\nincorporating symmetry regularization in helping reduce artifacts in the\nmodeling of 3D assets in the text-to-3D task. Project is at\n\\url{https://jingyang2017.github.io/sym3d.github.io/}\n", "date": "2024-06-10", "categories": []} {"id": "2312.11459", "title": "VolumeDiffusion: Flexible Text-to-3D Generation with Efficient\n Volumetric Encoder", "abstract": " This paper introduces a pioneering 3D volumetric encoder designed for\ntext-to-3D generation. To scale up the training data for the diffusion model, a\nlightweight network is developed to efficiently acquire feature volumes from\nmulti-view images. The 3D volumes are then trained on a diffusion model for\ntext-to-3D generation using a 3D U-Net. This research further addresses the\nchallenges of inaccurate object captions and high-dimensional feature volumes.\nThe proposed model, trained on the public Objaverse dataset, demonstrates\npromising outcomes in producing diverse and recognizable samples from text\nprompts. Notably, it empowers finer control over object part characteristics\nthrough textual cues, fostering model creativity by seamlessly combining\nmultiple concepts within a single object. This research significantly\ncontributes to the progress of 3D generation by introducing an efficient,\nflexible, and scalable representation methodology.\n", "date": "2023-12-18", "categories": []} {"id": "2311.16501", "title": "Context-Aware Indoor Point Cloud Object Generation through User\n Instructions", "abstract": " Indoor scene modification has emerged as a prominent area within computer\nvision, particularly for its applications in Augmented Reality (AR) and Virtual\nReality (VR). Traditional methods often rely on pre-existing object databases\nand predetermined object positions, limiting their flexibility and adaptability\nto new scenarios. In response to this challenge, we present a novel end-to-end\nmulti-modal deep neural network capable of generating point cloud objects\nseamlessly integrated with their surroundings, driven by textual instructions.\nOur model revolutionizes scene modification by enabling the creation of new\nenvironments with previously unseen object layouts, eliminating the need for\npre-stored CAD models. Leveraging Point-E as our generative model, we introduce\ninnovative techniques such as quantized position prediction and Top-K\nestimation to address the issue of false negatives resulting from ambiguous\nlanguage descriptions. Furthermore, we conduct comprehensive evaluations to\nshowcase the diversity of generated objects, the efficacy of textual\ninstructions, and the quantitative metrics, affirming the realism and\nversatility of our model in generating indoor objects. To provide a holistic\nassessment, we incorporate visual grounding as an additional metric, ensuring\nthe quality and coherence of the scenes produced by our model. Through these\nadvancements, our approach not only advances the state-of-the-art in indoor\nscene modification but also lays the foundation for future innovations in\nimmersive computing and digital environment creation.\n", "date": "2023-11-26", "categories": []} {"id": "2408.06145", "title": "Efficient and Scalable Point Cloud Generation with Sparse Point-Voxel\n Diffusion Models", "abstract": " We propose a novel point cloud U-Net diffusion architecture for 3D generative\nmodeling capable of generating high-quality and diverse 3D shapes while\nmaintaining fast generation times. Our network employs a dual-branch\narchitecture, combining the high-resolution representations of points with the\ncomputational efficiency of sparse voxels. Our fastest variant outperforms all\nnon-diffusion generative approaches on unconditional shape generation, the most\npopular benchmark for evaluating point cloud generative models, while our\nlargest model achieves state-of-the-art results among diffusion methods, with a\nruntime approximately 70% of the previously state-of-the-art PVD. Beyond\nunconditional generation, we perform extensive evaluations, including\nconditional generation on all categories of ShapeNet, demonstrating the\nscalability of our model to larger datasets, and implicit generation which\nallows our network to produce high quality point clouds on fewer timesteps,\nfurther decreasing the generation time. Finally, we evaluate the architecture's\nperformance in point cloud completion and super-resolution. Our model excels in\nall tasks, establishing it as a state-of-the-art diffusion U-Net for point\ncloud generative modeling. The code is publicly available at\nhttps://github.com/JohnRomanelis/SPVD.git.\n", "date": "2024-08-12", "categories": []} {"id": "2211.13220", "title": "TetraDiffusion: Tetrahedral Diffusion Models for 3D Shape Generation", "abstract": " Probabilistic denoising diffusion models (DDMs) have set a new standard for\n2D image generation. Extending DDMs for 3D content creation is an active field\nof research. Here, we propose TetraDiffusion, a diffusion model that operates\non a tetrahedral partitioning of 3D space to enable efficient, high-resolution\n3D shape generation. Our model introduces operators for convolution and\ntranspose convolution that act directly on the tetrahedral partition, and\nseamlessly includes additional attributes such as color. Remarkably,\nTetraDiffusion enables rapid sampling of detailed 3D objects in nearly\nreal-time with unprecedented resolution. It's also adaptable for generating 3D\nshapes conditioned on 2D images. Compared to existing 3D mesh diffusion\ntechniques, our method is up to 200 times faster in inference speed, works on\nstandard consumer hardware, and delivers superior results.\n", "date": "2022-11-23", "categories": []} {"id": "2312.00588", "title": "LucidDreaming: Controllable Object-Centric 3D Generation", "abstract": " With the recent development of generative models, Text-to-3D generations have\nalso seen significant growth, opening a door for creating video-game 3D assets\nfrom a more general public. Nonetheless, people without any professional 3D\nediting experience would find it hard to achieve precise control over the 3D\ngeneration, especially if there are multiple objects in the prompt, as using\ntext to control often leads to missing objects and imprecise locations. In this\npaper, we present LucidDreaming as an effective pipeline capable of spatial and\nnumerical control over 3D generation from only textual prompt commands or 3D\nbounding boxes. Specifically, our research demonstrates that Large Language\nModels (LLMs) possess 3D spatial awareness and can effectively translate\ntextual 3D information into precise 3D bounding boxes. We leverage LLMs to get\nindividual object information and their 3D bounding boxes as the initial step\nof our process. Then with the bounding boxes, We further propose clipped ray\nsampling and object-centric density blob bias to generate 3D objects aligning\nwith the bounding boxes. We show that our method exhibits remarkable\nadaptability across a spectrum of mainstream Score Distillation Sampling-based\n3D generation frameworks and our pipeline can even used to insert objects into\nan existing NeRF scene. Moreover, we also provide a dataset of prompts with 3D\nbounding boxes, benchmarking 3D spatial controllability. With extensive\nqualitative and quantitative experiments, we demonstrate that LucidDreaming\nachieves superior results in object placement precision and generation fidelity\ncompared to current approaches, while maintaining flexibility and ease of use\nfor non-expert users.\n", "date": "2023-11-30", "categories": []} {"id": "2312.04424", "title": "Cascade-Zero123: One Image to Highly Consistent 3D with Self-Prompted\n Nearby Views", "abstract": " Synthesizing multi-view 3D from one single image is a significant but\nchallenging task. Zero-1-to-3 methods have achieved great success by lifting a\n2D latent diffusion model to the 3D scope. The target view image is generated\nwith a single-view source image and the camera pose as condition information.\nHowever, due to the high sparsity of the single input image, Zero-1-to-3 tends\nto produce geometry and appearance inconsistency across views, especially for\ncomplex objects. To tackle this issue, we propose to supply more condition\ninformation for the generation model but in a self-prompt way. A cascade\nframework is constructed with two Zero-1-to-3 models, named Cascade-Zero123,\nwhich progressively extract 3D information from the source image. Specifically,\nseveral nearby views are first generated by the first model and then fed into\nthe second-stage model along with the source image as generation conditions.\nWith amplified self-prompted condition images, our Cascade-Zero123 generates\nmore consistent novel-view images than Zero-1-to-3. Experiment results\ndemonstrate remarkable promotion, especially for various complex and\nchallenging scenes, involving insects, humans, transparent objects, and stacked\nmultiple objects etc. More demos and code are available at\nhttps://cascadezero123.github.io.\n", "date": "2023-12-07", "categories": []} {"id": "2311.16499", "title": "InceptionHuman: Controllable Prompt-to-NeRF for Photorealistic 3D Human\n Generation", "abstract": " This paper presents InceptionHuman, a prompt-to-NeRF framework that allows\neasy control via a combination of prompts in different modalities (e.g., text,\nposes, edge, segmentation map, etc) as inputs to generate photorealistic 3D\nhumans. While many works have focused on generating 3D human models, they\nsuffer one or more of the following: lack of distinctive features, unnatural\nshading/shadows, unnatural poses/clothes, limited views, etc. InceptionHuman\nachieves consistent 3D human generation within a progressively refined NeRF\nspace with two novel modules, Iterative Pose-Aware Refinement (IPAR) and\nProgressive-Augmented Reconstruction (PAR). IPAR iteratively refines the\ndiffusion-generated images and synthesizes high-quality 3D-aware views\nconsidering the close-pose RGB values. PAR employs a pretrained diffusion prior\nto augment the generated synthetic views and adds regularization for\nview-independent appearance. Overall, the synthesis of photorealistic novel\nviews empowers the resulting 3D human NeRF from 360-degree perspectives.\nExtensive qualitative and quantitative experimental comparison show that our\nInceptionHuman models achieve state-of-the-art application quality.\n", "date": "2023-11-27", "categories": []} {"id": "2408.03178", "title": "An Object is Worth 64x64 Pixels: Generating 3D Object via Image\n Diffusion", "abstract": " We introduce a new approach for generating realistic 3D models with UV maps\nthrough a representation termed \"Object Images.\" This approach encapsulates\nsurface geometry, appearance, and patch structures within a 64x64 pixel image,\neffectively converting complex 3D shapes into a more manageable 2D format. By\ndoing so, we address the challenges of both geometric and semantic irregularity\ninherent in polygonal meshes. This method allows us to use image generation\nmodels, such as Diffusion Transformers, directly for 3D shape generation.\nEvaluated on the ABO dataset, our generated shapes with patch structures\nachieve point cloud FID comparable to recent 3D generative models, while\nnaturally supporting PBR material generation.\n", "date": "2024-08-06", "categories": []} {"id": "2408.01678", "title": "iControl3D: An Interactive System for Controllable 3D Scene Generation", "abstract": " 3D content creation has long been a complex and time-consuming process, often\nrequiring specialized skills and resources. While recent advancements have\nallowed for text-guided 3D object and scene generation, they still fall short\nof providing sufficient control over the generation process, leading to a gap\nbetween the user's creative vision and the generated results. In this paper, we\npresent iControl3D, a novel interactive system that empowers users to generate\nand render customizable 3D scenes with precise control. To this end, a 3D\ncreator interface has been developed to provide users with fine-grained control\nover the creation process. Technically, we leverage 3D meshes as an\nintermediary proxy to iteratively merge individual 2D diffusion-generated\nimages into a cohesive and unified 3D scene representation. To ensure seamless\nintegration of 3D meshes, we propose to perform boundary-aware depth alignment\nbefore fusing the newly generated mesh with the existing one in 3D space.\nAdditionally, to effectively manage depth discrepancies between remote content\nand foreground, we propose to model remote content separately with an\nenvironment map instead of 3D meshes. Finally, our neural rendering interface\nenables users to build a radiance field of their scene online and navigate the\nentire scene. Extensive experiments have been conducted to demonstrate the\neffectiveness of our system. The code will be made available at\nhttps://github.com/xingyi-li/iControl3D.\n", "date": "2024-08-03", "categories": []} {"id": "2408.01269", "title": "A General Framework to Boost 3D GS Initialization for Text-to-3D\n Generation by Lexical Richness", "abstract": " Text-to-3D content creation has recently received much attention, especially\nwith the prevalence of 3D Gaussians Splatting. In general, GS-based methods\ncomprise two key stages: initialization and rendering optimization. To achieve\ninitialization, existing works directly apply random sphere initialization or\n3D diffusion models, e.g., Point-E, to derive the initial shapes. However, such\nstrategies suffer from two critical yet challenging problems: 1) the final\nshapes are still similar to the initial ones even after training; 2) shapes can\nbe produced only from simple texts, e.g., \"a dog\", not for lexically richer\ntexts, e.g., \"a dog is sitting on the top of the airplane\". To address these\nproblems, this paper proposes a novel general framework to boost the 3D GS\nInitialization for text-to-3D generation upon the lexical richness. Our key\nidea is to aggregate 3D Gaussians into spatially uniform voxels to represent\ncomplex shapes while enabling the spatial interaction among the 3D Gaussians\nand semantic interaction between Gaussians and texts. Specifically, we first\nconstruct a voxelized representation, where each voxel holds a 3D Gaussian with\nits position, scale, and rotation fixed while setting opacity as the sole\nfactor to determine a position's occupancy. We then design an initialization\nnetwork mainly consisting of two novel components: 1) Global Information\nPerception (GIP) block and 2) Gaussians-Text Fusion (GTF) block. Such a design\nenables each 3D Gaussian to assimilate the spatial information from other areas\nand semantic information from texts. Extensive experiments show the superiority\nof our framework of high-quality 3D GS initialization against the existing\nmethods, e.g., Shap-E, by taking lexically simple, medium, and hard texts.\nAlso, our framework can be seamlessly plugged into SoTA training frameworks,\ne.g., LucidDreamer, for semantically consistent text-to-3D generation.\n", "date": "2024-08-02", "categories": []} {"id": "2408.01291", "title": "TexGen: Text-Guided 3D Texture Generation with Multi-view Sampling and\n Resampling", "abstract": " Given a 3D mesh, we aim to synthesize 3D textures that correspond to\narbitrary textual descriptions. Current methods for generating and assembling\ntextures from sampled views often result in prominent seams or excessive\nsmoothing. To tackle these issues, we present TexGen, a novel multi-view\nsampling and resampling framework for texture generation leveraging a\npre-trained text-to-image diffusion model. For view consistent sampling, first\nof all we maintain a texture map in RGB space that is parameterized by the\ndenoising step and updated after each sampling step of the diffusion model to\nprogressively reduce the view discrepancy. An attention-guided multi-view\nsampling strategy is exploited to broadcast the appearance information across\nviews. To preserve texture details, we develop a noise resampling technique\nthat aids in the estimation of noise, generating inputs for subsequent\ndenoising steps, as directed by the text prompt and current texture map.\nThrough an extensive amount of qualitative and quantitative evaluations, we\ndemonstrate that our proposed method produces significantly better texture\nquality for diverse 3D objects with a high degree of view consistency and rich\nappearance details, outperforming current state-of-the-art methods.\nFurthermore, our proposed texture generation technique can also be applied to\ntexture editing while preserving the original identity. More experimental\nresults are available at https://dong-huo.github.io/TexGen/\n", "date": "2024-08-02", "categories": []} {"id": "2408.00297", "title": "EmoTalk3D: High-Fidelity Free-View Synthesis of Emotional 3D Talking\n Head", "abstract": " We present a novel approach for synthesizing 3D talking heads with\ncontrollable emotion, featuring enhanced lip synchronization and rendering\nquality. Despite significant progress in the field, prior methods still suffer\nfrom multi-view consistency and a lack of emotional expressiveness. To address\nthese issues, we collect EmoTalk3D dataset with calibrated multi-view videos,\nemotional annotations, and per-frame 3D geometry. By training on the EmoTalk3D\ndataset, we propose a \\textit{`Speech-to-Geometry-to-Appearance'} mapping\nframework that first predicts faithful 3D geometry sequence from the audio\nfeatures, then the appearance of a 3D talking head represented by 4D Gaussians\nis synthesized from the predicted geometry. The appearance is further\ndisentangled into canonical and dynamic Gaussians, learned from multi-view\nvideos, and fused to render free-view talking head animation. Moreover, our\nmodel enables controllable emotion in the generated talking heads and can be\nrendered in wide-range views. Our method exhibits improved rendering quality\nand stability in lip motion generation while capturing dynamic facial details\nsuch as wrinkles and subtle expressions. Experiments demonstrate the\neffectiveness of our approach in generating high-fidelity and\nemotion-controllable 3D talking heads. The code and EmoTalk3D dataset are\nreleased at https://nju-3dv.github.io/projects/EmoTalk3D.\n", "date": "2024-08-01", "categories": []} {"id": "2312.14124", "title": "Neural Point Cloud Diffusion for Disentangled 3D Shape and Appearance\n Generation", "abstract": " Controllable generation of 3D assets is important for many practical\napplications like content creation in movies, games and engineering, as well as\nin AR/VR. Recently, diffusion models have shown remarkable results in\ngeneration quality of 3D objects. However, none of the existing models enable\ndisentangled generation to control the shape and appearance separately. For the\nfirst time, we present a suitable representation for 3D diffusion models to\nenable such disentanglement by introducing a hybrid point cloud and neural\nradiance field approach. We model a diffusion process over point positions\njointly with a high-dimensional feature space for a local density and radiance\ndecoder. While the point positions represent the coarse shape of the object,\nthe point features allow modeling the geometry and appearance details. This\ndisentanglement enables us to sample both independently and therefore to\ncontrol both separately. Our approach sets a new state of the art in generation\ncompared to previous disentanglement-capable methods by reduced FID scores of\n30-90% and is on-par with other non disentanglement-capable state-of-the art\nmethods.\n", "date": "2023-12-21", "categories": []} {"id": "2406.11196", "title": "Vid3D: Synthesis of Dynamic 3D Scenes using 2D Video Diffusion", "abstract": " A recent frontier in computer vision has been the task of 3D video\ngeneration, which consists of generating a time-varying 3D representation of a\nscene. To generate dynamic 3D scenes, current methods explicitly model 3D\ntemporal dynamics by jointly optimizing for consistency across both time and\nviews of the scene. In this paper, we instead investigate whether it is\nnecessary to explicitly enforce multiview consistency over time, as current\napproaches do, or if it is sufficient for a model to generate 3D\nrepresentations of each timestep independently. We hence propose a model,\nVid3D, that leverages 2D video diffusion to generate 3D videos by first\ngenerating a 2D \"seed\" of the video's temporal dynamics and then independently\ngenerating a 3D representation for each timestep in the seed video. We evaluate\nVid3D against two state-of-the-art 3D video generation methods and find that\nVid3D is achieves comparable results despite not explicitly modeling 3D\ntemporal dynamics. We further ablate how the quality of Vid3D depends on the\nnumber of views generated per frame. While we observe some degradation with\nfewer views, performance degradation remains minor. Our results thus suggest\nthat 3D temporal knowledge may not be necessary to generate high-quality\ndynamic 3D scenes, potentially enabling simpler generative algorithms for this\ntask.\n", "date": "2024-06-17", "categories": []} {"id": "2407.21428", "title": "Deformable 3D Shape Diffusion Model", "abstract": " The Gaussian diffusion model, initially designed for image generation, has\nrecently been adapted for 3D point cloud generation. However, these adaptations\nhave not fully considered the intrinsic geometric characteristics of 3D shapes,\nthereby constraining the diffusion model's potential for 3D shape manipulation.\nTo address this limitation, we introduce a novel deformable 3D shape diffusion\nmodel that facilitates comprehensive 3D shape manipulation, including point\ncloud generation, mesh deformation, and facial animation. Our approach\ninnovatively incorporates a differential deformation kernel, which deconstructs\nthe generation of geometric structures into successive non-rigid deformation\nstages. By leveraging a probabilistic diffusion model to simulate this\nstep-by-step process, our method provides a versatile and efficient solution\nfor a wide range of applications, spanning from graphics rendering to facial\nexpression animation. Empirical evidence highlights the effectiveness of our\napproach, demonstrating state-of-the-art performance in point cloud generation\nand competitive results in mesh deformation. Additionally, extensive visual\ndemonstrations reveal the significant potential of our approach for practical\napplications. Our method presents a unique pathway for advancing 3D shape\nmanipulation and unlocking new opportunities in the realm of virtual reality.\n", "date": "2024-07-31", "categories": []} {"id": "2312.00085", "title": "X-Dreamer: Creating High-quality 3D Content by Bridging the Domain Gap\n Between Text-to-2D and Text-to-3D Generation", "abstract": " In recent times, automatic text-to-3D content creation has made significant\nprogress, driven by the development of pretrained 2D diffusion models. Existing\ntext-to-3D methods typically optimize the 3D representation to ensure that the\nrendered image aligns well with the given text, as evaluated by the pretrained\n2D diffusion model. Nevertheless, a substantial domain gap exists between 2D\nimages and 3D assets, primarily attributed to variations in camera-related\nattributes and the exclusive presence of foreground objects. Consequently,\nemploying 2D diffusion models directly for optimizing 3D representations may\nlead to suboptimal outcomes. To address this issue, we present X-Dreamer, a\nnovel approach for high-quality text-to-3D content creation that effectively\nbridges the gap between text-to-2D and text-to-3D synthesis. The key components\nof X-Dreamer are two innovative designs: Camera-Guided Low-Rank Adaptation\n(CG-LoRA) and Attention-Mask Alignment (AMA) Loss. CG-LoRA dynamically\nincorporates camera information into the pretrained diffusion models by\nemploying camera-dependent generation for trainable parameters. This\nintegration enhances the alignment between the generated 3D assets and the\ncamera's perspective. AMA loss guides the attention map of the pretrained\ndiffusion model using the binary mask of the 3D object, prioritizing the\ncreation of the foreground object. This module ensures that the model focuses\non generating accurate and detailed foreground objects. Extensive evaluations\ndemonstrate the effectiveness of our proposed method compared to existing\ntext-to-3D approaches. Our project webpage:\nhttps://xmu-xiaoma666.github.io/Projects/X-Dreamer/ .\n", "date": "2023-11-30", "categories": []} {"id": "2407.20727", "title": "SceneTeller: Language-to-3D Scene Generation", "abstract": " Designing high-quality indoor 3D scenes is important in many practical\napplications, such as room planning or game development. Conventionally, this\nhas been a time-consuming process which requires both artistic skill and\nfamiliarity with professional software, making it hardly accessible for layman\nusers. However, recent advances in generative AI have established solid\nfoundation for democratizing 3D design. In this paper, we propose a pioneering\napproach for text-based 3D room design. Given a prompt in natural language\ndescribing the object placement in the room, our method produces a high-quality\n3D scene corresponding to it. With an additional text prompt the users can\nchange the appearance of the entire scene or of individual objects in it. Built\nusing in-context learning, CAD model retrieval and 3D-Gaussian-Splatting-based\nstylization, our turnkey pipeline produces state-of-the-art 3D scenes, while\nbeing easy to use even for novices. Our project page is available at\nhttps://sceneteller.github.io/.\n", "date": "2024-07-30", "categories": []} {"id": "2312.06644", "title": "AnyHome: Open-Vocabulary Generation of Structured and Textured 3D Homes", "abstract": " Inspired by cognitive theories, we introduce AnyHome, a framework that\ntranslates any text into well-structured and textured indoor scenes at a\nhouse-scale. By prompting Large Language Models (LLMs) with designed templates,\nour approach converts provided textual narratives into amodal structured\nrepresentations. These representations guarantee consistent and realistic\nspatial layouts by directing the synthesis of a geometry mesh within defined\nconstraints. A Score Distillation Sampling process is then employed to refine\nthe geometry, followed by an egocentric inpainting process that adds lifelike\ntextures to it. AnyHome stands out with its editability, customizability,\ndiversity, and realism. The structured representations for scenes allow for\nextensive editing at varying levels of granularity. Capable of interpreting\ntexts ranging from simple labels to detailed narratives, AnyHome generates\ndetailed geometries and textures that outperform existing methods in both\nquantitative and qualitative measures.\n", "date": "2023-12-11", "categories": []} {"id": "2403.01807", "title": "ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models", "abstract": " 3D asset generation is getting massive amounts of attention, inspired by the\nrecent success of text-guided 2D content creation. Existing text-to-3D methods\nuse pretrained text-to-image diffusion models in an optimization problem or\nfine-tune them on synthetic data, which often results in non-photorealistic 3D\nobjects without backgrounds. In this paper, we present a method that leverages\npretrained text-to-image models as a prior, and learn to generate multi-view\nimages in a single denoising process from real-world data. Concretely, we\npropose to integrate 3D volume-rendering and cross-frame-attention layers into\neach block of the existing U-Net network of the text-to-image model. Moreover,\nwe design an autoregressive generation that renders more 3D-consistent images\nat any viewpoint. We train our model on real-world datasets of objects and\nshowcase its capabilities to generate instances with a variety of high-quality\nshapes and textures in authentic surroundings. Compared to the existing\nmethods, the results generated by our method are consistent, and have favorable\nvisual quality (-30% FID, -37% KID).\n", "date": "2024-03-04", "categories": []} {"id": "2405.10988", "title": "Flow Score Distillation for Diverse Text-to-3D Generation", "abstract": " Recent advancements in Text-to-3D generation have yielded remarkable\nprogress, particularly through methods that rely on Score Distillation Sampling\n(SDS). While SDS exhibits the capability to create impressive 3D assets, it is\nhindered by its inherent maximum-likelihood-seeking essence, resulting in\nlimited diversity in generation outcomes. In this paper, we discover that the\nDenoise Diffusion Implicit Models (DDIM) generation process (\\ie PF-ODE) can be\nsuccinctly expressed using an analogue of SDS loss. One step further, one can\nsee SDS as a generalized DDIM generation process. Following this insight, we\nshow that the noise sampling strategy in the noise addition stage significantly\nrestricts the diversity of generation results. To address this limitation, we\npresent an innovative noise sampling approach and introduce a novel text-to-3D\nmethod called Flow Score Distillation (FSD). Our validation experiments across\nvarious text-to-image Diffusion Models demonstrate that FSD substantially\nenhances generation diversity without compromising quality.\n", "date": "2024-05-16", "categories": []} {"id": "2407.19035", "title": "ScalingGaussian: Enhancing 3D Content Creation with Generative Gaussian\n Splatting", "abstract": " The creation of high-quality 3D assets is paramount for applications in\ndigital heritage preservation, entertainment, and robotics. Traditionally, this\nprocess necessitates skilled professionals and specialized software for the\nmodeling, texturing, and rendering of 3D objects. However, the rising demand\nfor 3D assets in gaming and virtual reality (VR) has led to the creation of\naccessible image-to-3D technologies, allowing non-professionals to produce 3D\ncontent and decreasing dependence on expert input. Existing methods for 3D\ncontent generation struggle to simultaneously achieve detailed textures and\nstrong geometric consistency. We introduce a novel 3D content creation\nframework, ScalingGaussian, which combines 3D and 2D diffusion models to\nachieve detailed textures and geometric consistency in generated 3D assets.\nInitially, a 3D diffusion model generates point clouds, which are then\ndensified through a process of selecting local regions, introducing Gaussian\nnoise, followed by using local density-weighted selection. To refine the 3D\ngaussians, we utilize a 2D diffusion model with Score Distillation Sampling\n(SDS) loss, guiding the 3D Gaussians to clone and split. Finally, the 3D\nGaussians are converted into meshes, and the surface textures are optimized\nusing Mean Square Error(MSE) and Gradient Profile Prior(GPP) losses. Our method\naddresses the common issue of sparse point clouds in 3D diffusion, resulting in\nimproved geometric structure and detailed textures. Experiments on image-to-3D\ntasks demonstrate that our approach efficiently generates high-quality 3D\nassets.\n", "date": "2024-07-26", "categories": []} {"id": "2407.19225", "title": "Magic3DSketch: Create Colorful 3D Models From Sketch-Based 3D Modeling\n Guided by Text and Language-Image Pre-Training", "abstract": " The requirement for 3D content is growing as AR/VR application emerges. At\nthe same time, 3D modelling is only available for skillful experts, because\ntraditional methods like Computer-Aided Design (CAD) are often too\nlabor-intensive and skill-demanding, making it challenging for novice users.\nOur proposed method, Magic3DSketch, employs a novel technique that encodes\nsketches to predict a 3D mesh, guided by text descriptions and leveraging\nexternal prior knowledge obtained through text and language-image pre-training.\nThe integration of language-image pre-trained neural networks complements the\nsparse and ambiguous nature of single-view sketch inputs. Our method is also\nmore useful and offers higher degree of controllability compared to existing\ntext-to-3D approaches, according to our user study. Moreover, Magic3DSketch\nachieves state-of-the-art performance in both synthetic and real dataset with\nthe capability of producing more detailed structures and realistic shapes with\nthe help of text input. Users are also more satisfied with models obtained by\nMagic3DSketch according to our user study. Additionally, we are also the first,\nto our knowledge, add color based on text description to the sketch-derived\nshapes. By combining sketches and text guidance with the help of language-image\npretrained models, our Magic3DSketch can allow novice users to create custom 3D\nmodels with minimal effort and maximum creative freedom, with the potential to\nrevolutionize future 3D modeling pipelines.\n", "date": "2024-07-27", "categories": []} {"id": "2407.19548", "title": "Cycle3D: High-quality and Consistent Image-to-3D Generation via\n Generation-Reconstruction Cycle", "abstract": " Recent 3D large reconstruction models typically employ a two-stage process,\nincluding first generate multi-view images by a multi-view diffusion model, and\nthen utilize a feed-forward model to reconstruct images to 3D content.However,\nmulti-view diffusion models often produce low-quality and inconsistent images,\nadversely affecting the quality of the final 3D reconstruction. To address this\nissue, we propose a unified 3D generation framework called Cycle3D, which\ncyclically utilizes a 2D diffusion-based generation module and a feed-forward\n3D reconstruction module during the multi-step diffusion process. Concretely,\n2D diffusion model is applied for generating high-quality texture, and the\nreconstruction model guarantees multi-view consistency.Moreover, 2D diffusion\nmodel can further control the generated content and inject reference-view\ninformation for unseen views, thereby enhancing the diversity and texture\nconsistency of 3D generation during the denoising process. Extensive\nexperiments demonstrate the superior ability of our method to create 3D content\nwith high-quality and consistency compared with state-of-the-art baselines.\n", "date": "2024-07-28", "categories": []} {"id": "2404.06903", "title": "DreamScene360: Unconstrained Text-to-3D Scene Generation with Panoramic\n Gaussian Splatting", "abstract": " The increasing demand for virtual reality applications has highlighted the\nsignificance of crafting immersive 3D assets. We present a text-to-3D\n360$^{\\circ}$ scene generation pipeline that facilitates the creation of\ncomprehensive 360$^{\\circ}$ scenes for in-the-wild environments in a matter of\nminutes. Our approach utilizes the generative power of a 2D diffusion model and\nprompt self-refinement to create a high-quality and globally coherent panoramic\nimage. This image acts as a preliminary \"flat\" (2D) scene representation.\nSubsequently, it is lifted into 3D Gaussians, employing splatting techniques to\nenable real-time exploration. To produce consistent 3D geometry, our pipeline\nconstructs a spatially coherent structure by aligning the 2D monocular depth\ninto a globally optimized point cloud. This point cloud serves as the initial\nstate for the centroids of 3D Gaussians. In order to address invisible issues\ninherent in single-view inputs, we impose semantic and geometric constraints on\nboth synthesized and input camera views as regularizations. These guide the\noptimization of Gaussians, aiding in the reconstruction of unseen regions. In\nsummary, our method offers a globally consistent 3D scene within a\n360$^{\\circ}$ perspective, providing an enhanced immersive experience over\nexisting techniques. Project website at: http://dreamscene360.github.io/\n", "date": "2024-04-10", "categories": []} {"id": "2311.17050", "title": "Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using\n Diffusion Models", "abstract": " We present Surf-D, a novel method for generating high-quality 3D shapes as\nSurfaces with arbitrary topologies using Diffusion models. Previous methods\nexplored shape generation with different representations and they suffer from\nlimited topologies and poor geometry details. To generate high-quality surfaces\nof arbitrary topologies, we use the Unsigned Distance Field (UDF) as our\nsurface representation to accommodate arbitrary topologies. Furthermore, we\npropose a new pipeline that employs a point-based AutoEncoder to learn a\ncompact and continuous latent space for accurately encoding UDF and support\nhigh-resolution mesh extraction. We further show that our new pipeline\nsignificantly outperforms the prior approaches to learning the distance fields,\nsuch as the grid-based AutoEncoder, which is not scalable and incapable of\nlearning accurate UDF. In addition, we adopt a curriculum learning strategy to\nefficiently embed various surfaces. With the pretrained shape latent space, we\nemploy a latent diffusion model to acquire the distribution of various shapes.\nExtensive experiments are presented on using Surf-D for unconditional\ngeneration, category conditional generation, image conditional generation, and\ntext-to-shape tasks. The experiments demonstrate the superior performance of\nSurf-D in shape generation across multiple modalities as conditions. Visit our\nproject page at https://yzmblog.github.io/projects/SurfD/.\n", "date": "2023-11-28", "categories": []} {"id": "2407.16260", "title": "DreamDissector: Learning Disentangled Text-to-3D Generation from 2D\n Diffusion Priors", "abstract": " Text-to-3D generation has recently seen significant progress. To enhance its\npracticality in real-world applications, it is crucial to generate multiple\nindependent objects with interactions, similar to layer-compositing in 2D image\nediting. However, existing text-to-3D methods struggle with this task, as they\nare designed to generate either non-independent objects or independent objects\nlacking spatially plausible interactions. Addressing this, we propose\nDreamDissector, a text-to-3D method capable of generating multiple independent\nobjects with interactions. DreamDissector accepts a multi-object text-to-3D\nNeRF as input and produces independent textured meshes. To achieve this, we\nintroduce the Neural Category Field (NeCF) for disentangling the input NeRF.\nAdditionally, we present the Category Score Distillation Sampling (CSDS),\nfacilitated by a Deep Concept Mining (DCM) module, to tackle the concept gap\nissue in diffusion models. By leveraging NeCF and CSDS, we can effectively\nderive sub-NeRFs from the original scene. Further refinement enhances geometry\nand texture. Our experimental results validate the effectiveness of\nDreamDissector, providing users with novel means to control 3D synthesis at the\nobject level and potentially opening avenues for various creative applications\nin the future.\n", "date": "2024-07-23", "categories": []} {"id": "2407.16511", "title": "DreamVTON: Customizing 3D Virtual Try-on with Personalized Diffusion\n Models", "abstract": " Image-based 3D Virtual Try-ON (VTON) aims to sculpt the 3D human according to\nperson and clothes images, which is data-efficient (i.e., getting rid of\nexpensive 3D data) but challenging. Recent text-to-3D methods achieve\nremarkable improvement in high-fidelity 3D human generation, demonstrating its\npotential for 3D virtual try-on. Inspired by the impressive success of\npersonalized diffusion models (e.g., Dreambooth and LoRA) for 2D VTON, it is\nstraightforward to achieve 3D VTON by integrating the personalization technique\ninto the diffusion-based text-to-3D framework. However, employing the\npersonalized module in a pre-trained diffusion model (e.g., StableDiffusion\n(SD)) would degrade the model's capability for multi-view or multi-domain\nsynthesis, which is detrimental to the geometry and texture optimization guided\nby Score Distillation Sampling (SDS) loss. In this work, we propose a novel\ncustomizing 3D human try-on model, named \\textbf{DreamVTON}, to separately\noptimize the geometry and texture of the 3D human. Specifically, a personalized\nSD with multi-concept LoRA is proposed to provide the generative prior about\nthe specific person and clothes, while a Densepose-guided ControlNet is\nexploited to guarantee consistent prior about body pose across various camera\nviews. Besides, to avoid the inconsistent multi-view priors from the\npersonalized SD dominating the optimization, DreamVTON introduces a\ntemplate-based optimization mechanism, which employs mask templates for\ngeometry shape learning and normal/RGB templates for geometry/texture details\nlearning. Furthermore, for the geometry optimization phase, DreamVTON\nintegrates a normal-style LoRA into personalized SD to enhance normal map\ngenerative prior, facilitating smooth geometry modeling.\n", "date": "2024-07-23", "categories": []} {"id": "2312.05804", "title": "Layered 3D Human Generation via Semantic-Aware Diffusion Model", "abstract": " The generation of 3D clothed humans has attracted increasing attention in\nrecent years. However, existing work cannot generate layered high-quality 3D\nhumans with consistent body structures. As a result, these methods are unable\nto arbitrarily and separately change and edit the body and clothing of the\nhuman. In this paper, we propose a text-driven layered 3D human generation\nframework based on a novel physically-decoupled semantic-aware diffusion model.\nTo keep the generated clothing consistent with the target text, we propose a\nsemantic-confidence strategy for clothing that can eliminate the non-clothing\ncontent generated by the model. To match the clothing with different body\nshapes, we propose a SMPL-driven implicit field deformation network that\nenables the free transfer and reuse of clothing. Besides, we introduce uniform\nshape priors based on the SMPL model for body and clothing, respectively, which\ngenerates more diverse 3D content without being constrained by specific\ntemplates. The experimental results demonstrate that the proposed method not\nonly generates 3D humans with consistent body structures but also allows free\nediting in a layered manner. The source code will be made public.\n", "date": "2023-12-10", "categories": []} {"id": "2407.13584", "title": "Connecting Consistency Distillation to Score Distillation for Text-to-3D\n Generation", "abstract": " Although recent advancements in text-to-3D generation have significantly\nimproved generation quality, issues like limited level of detail and low\nfidelity still persist, which requires further improvement. To understand the\nessence of those issues, we thoroughly analyze current score distillation\nmethods by connecting theories of consistency distillation to score\ndistillation. Based on the insights acquired through analysis, we propose an\noptimization framework, Guided Consistency Sampling (GCS), integrated with 3D\nGaussian Splatting (3DGS) to alleviate those issues. Additionally, we have\nobserved the persistent oversaturation in the rendered views of generated 3D\nassets. From experiments, we find that it is caused by unwanted accumulated\nbrightness in 3DGS during optimization. To mitigate this issue, we introduce a\nBrightness-Equalized Generation (BEG) scheme in 3DGS rendering. Experimental\nresults demonstrate that our approach generates 3D assets with more details and\nhigher fidelity than state-of-the-art methods. The codes are released at\nhttps://github.com/LMozart/ECCV2024-GCS-BEG.\n", "date": "2024-07-18", "categories": []} {"id": "2407.15187", "title": "HoloDreamer: Holistic 3D Panoramic World Generation from Text\n Descriptions", "abstract": " 3D scene generation is in high demand across various domains, including\nvirtual reality, gaming, and the film industry. Owing to the powerful\ngenerative capabilities of text-to-image diffusion models that provide reliable\npriors, the creation of 3D scenes using only text prompts has become viable,\nthereby significantly advancing researches in text-driven 3D scene generation.\nIn order to obtain multiple-view supervision from 2D diffusion models,\nprevailing methods typically employ the diffusion model to generate an initial\nlocal image, followed by iteratively outpainting the local image using\ndiffusion models to gradually generate scenes. Nevertheless, these\noutpainting-based approaches prone to produce global inconsistent scene\ngeneration results without high degree of completeness, restricting their\nbroader applications. To tackle these problems, we introduce HoloDreamer, a\nframework that first generates high-definition panorama as a holistic\ninitialization of the full 3D scene, then leverage 3D Gaussian Splatting\n(3D-GS) to quickly reconstruct the 3D scene, thereby facilitating the creation\nof view-consistent and fully enclosed 3D scenes. Specifically, we propose\nStylized Equirectangular Panorama Generation, a pipeline that combines multiple\ndiffusion models to enable stylized and detailed equirectangular panorama\ngeneration from complex text prompts. Subsequently, Enhanced Two-Stage Panorama\nReconstruction is introduced, conducting a two-stage optimization of 3D-GS to\ninpaint the missing region and enhance the integrity of the scene.\nComprehensive experiments demonstrated that our method outperforms prior works\nin terms of overall visual consistency and harmony as well as reconstruction\nquality and rendering robustness when generating fully enclosed scenes.\n", "date": "2024-07-21", "categories": []} {"id": "2403.12034", "title": "VFusion3D: Learning Scalable 3D Generative Models from Video Diffusion\n Models", "abstract": " This paper presents a novel method for building scalable 3D generative models\nutilizing pre-trained video diffusion models. The primary obstacle in\ndeveloping foundation 3D generative models is the limited availability of 3D\ndata. Unlike images, texts, or videos, 3D data are not readily accessible and\nare difficult to acquire. This results in a significant disparity in scale\ncompared to the vast quantities of other types of data. To address this issue,\nwe propose using a video diffusion model, trained with extensive volumes of\ntext, images, and videos, as a knowledge source for 3D data. By unlocking its\nmulti-view generative capabilities through fine-tuning, we generate a\nlarge-scale synthetic multi-view dataset to train a feed-forward 3D generative\nmodel. The proposed model, VFusion3D, trained on nearly 3M synthetic multi-view\ndata, can generate a 3D asset from a single image in seconds and achieves\nsuperior performance when compared to current SOTA feed-forward 3D generative\nmodels, with users preferring our results over 90% of the time.\n", "date": "2024-03-18", "categories": []} {"id": "2404.03575", "title": "DreamScene: 3D Gaussian-based Text-to-3D Scene Generation via Formation\n Pattern Sampling", "abstract": " Text-to-3D scene generation holds immense potential for the gaming, film, and\narchitecture sectors. Despite significant progress, existing methods struggle\nwith maintaining high quality, consistency, and editing flexibility. In this\npaper, we propose DreamScene, a 3D Gaussian-based novel text-to-3D scene\ngeneration framework, to tackle the aforementioned three challenges mainly via\ntwo strategies. First, DreamScene employs Formation Pattern Sampling (FPS), a\nmulti-timestep sampling strategy guided by the formation patterns of 3D\nobjects, to form fast, semantically rich, and high-quality representations. FPS\nuses 3D Gaussian filtering for optimization stability, and leverages\nreconstruction techniques to generate plausible textures. Second, DreamScene\nemploys a progressive three-stage camera sampling strategy, specifically\ndesigned for both indoor and outdoor settings, to effectively ensure\nobject-environment integration and scene-wide 3D consistency. Last, DreamScene\nenhances scene editing flexibility by integrating objects and environments,\nenabling targeted adjustments. Extensive experiments validate DreamScene's\nsuperiority over current state-of-the-art techniques, heralding its\nwide-ranging potential for diverse applications. Code and demos will be\nreleased at https://dreamscene-project.github.io .\n", "date": "2024-04-04", "categories": []} {"id": "2407.13976", "title": "PlacidDreamer: Advancing Harmony in Text-to-3D Generation", "abstract": " Recently, text-to-3D generation has attracted significant attention,\nresulting in notable performance enhancements. Previous methods utilize\nend-to-end 3D generation models to initialize 3D Gaussians, multi-view\ndiffusion models to enforce multi-view consistency, and text-to-image diffusion\nmodels to refine details with score distillation algorithms. However, these\nmethods exhibit two limitations. Firstly, they encounter conflicts in\ngeneration directions since different models aim to produce diverse 3D assets.\nSecondly, the issue of over-saturation in score distillation has not been\nthoroughly investigated and solved. To address these limitations, we propose\nPlacidDreamer, a text-to-3D framework that harmonizes initialization,\nmulti-view generation, and text-conditioned generation with a single multi-view\ndiffusion model, while simultaneously employing a novel score distillation\nalgorithm to achieve balanced saturation. To unify the generation direction, we\nintroduce the Latent-Plane module, a training-friendly plug-in extension that\nenables multi-view diffusion models to provide fast geometry reconstruction for\ninitialization and enhanced multi-view images to personalize the text-to-image\ndiffusion model. To address the over-saturation problem, we propose to view\nscore distillation as a multi-objective optimization problem and introduce the\nBalanced Score Distillation algorithm, which offers a Pareto Optimal solution\nthat achieves both rich details and balanced saturation. Extensive experiments\nvalidate the outstanding capabilities of our PlacidDreamer. The code is\navailable at \\url{https://github.com/HansenHuang0823/PlacidDreamer}.\n", "date": "2024-07-19", "categories": []} {"id": "2311.12085", "title": "Pyramid Diffusion for Fine 3D Large Scene Generation", "abstract": " Diffusion models have shown remarkable results in generating 2D images and\nsmall-scale 3D objects. However, their application to the synthesis of\nlarge-scale 3D scenes has been rarely explored. This is mainly due to the\ninherent complexity and bulky size of 3D scenery data, particularly outdoor\nscenes, and the limited availability of comprehensive real-world datasets,\nwhich makes training a stable scene diffusion model challenging. In this work,\nwe explore how to effectively generate large-scale 3D scenes using the\ncoarse-to-fine paradigm. We introduce a framework, the Pyramid Discrete\nDiffusion model (PDD), which employs scale-varied diffusion models to\nprogressively generate high-quality outdoor scenes. Experimental results of PDD\ndemonstrate our successful exploration in generating 3D scenes both\nunconditionally and conditionally. We further showcase the data compatibility\nof the PDD model, due to its multi-scale architecture: a PDD model trained on\none dataset can be easily fine-tuned with another dataset. Code is available at\nhttps://github.com/yuhengliu02/pyramid-discrete-diffusion.\n", "date": "2023-11-20", "categories": []} {"id": "2407.13677", "title": "PASTA: Controllable Part-Aware Shape Generation with Autoregressive\n Transformers", "abstract": " The increased demand for tools that automate the 3D content creation process\nled to tremendous progress in deep generative models that can generate diverse\n3D objects of high fidelity. In this paper, we present PASTA, an autoregressive\ntransformer architecture for generating high quality 3D shapes. PASTA comprises\ntwo main components: An autoregressive transformer that generates objects as a\nsequence of cuboidal primitives and a blending network, implemented with a\ntransformer decoder that composes the sequences of cuboids and synthesizes high\nquality meshes for each object. Our model is trained in two stages: First we\ntrain our autoregressive generative model using only annotated cuboidal parts\nas supervision and next, we train our blending network using explicit 3D\nsupervision, in the form of watertight meshes. Evaluations on various ShapeNet\nobjects showcase the ability of our model to perform shape generation from\ndiverse inputs \\eg from scratch, from a partial object, from text and images,\nas well size-guided generation, by explicitly conditioning on a bounding box\nthat defines the object's boundaries. Moreover, as our model considers the\nunderlying part-based structure of a 3D object, we are able to select a\nspecific part and produce shapes with meaningful variations of this part. As\nevidenced by our experiments, our model generates 3D shapes that are both more\nrealistic and diverse than existing part-based and non part-based methods,\nwhile at the same time is simpler to implement and train.\n", "date": "2024-07-18", "categories": []} {"id": "2312.02192", "title": "DiverseDream: Diverse Text-to-3D Synthesis with Augmented Text Embedding", "abstract": " Text-to-3D synthesis has recently emerged as a new approach to sampling 3D\nmodels by adopting pretrained text-to-image models as guiding visual priors. An\nintriguing but underexplored problem with existing text-to-3D methods is that\n3D models obtained from the sampling-by-optimization procedure tend to have\nmode collapses, and hence poor diversity in their results. In this paper, we\nprovide an analysis and identify potential causes of such a limited diversity,\nwhich motivates us to devise a new method that considers the joint generation\nof different 3D models from the same text prompt. We propose to use augmented\ntext prompts via textual inversion of reference images to diversify the joint\ngeneration. We show that our method leads to improved diversity in text-to-3D\nsynthesis qualitatively and quantitatively. Project page:\nhttps://diversedream.github.io\n", "date": "2023-12-02", "categories": []} {"id": "2407.09822", "title": "VividDreamer: Invariant Score Distillation For Hyper-Realistic\n Text-to-3D Generation", "abstract": " This paper presents Invariant Score Distillation (ISD), a novel method for\nhigh-fidelity text-to-3D generation. ISD aims to tackle the over-saturation and\nover-smoothing problems in Score Distillation Sampling (SDS). In this paper,\nSDS is decoupled into a weighted sum of two components: the reconstruction term\nand the classifier-free guidance term. We experimentally found that\nover-saturation stems from the large classifier-free guidance scale and\nover-smoothing comes from the reconstruction term. To overcome these problems,\nISD utilizes an invariant score term derived from DDIM sampling to replace the\nreconstruction term in SDS. This operation allows the utilization of a medium\nclassifier-free guidance scale and mitigates the reconstruction-related errors,\nthus preventing the over-smoothing and over-saturation of results. Extensive\nexperiments demonstrate that our method greatly enhances SDS and produces\nrealistic 3D objects through single-stage optimization.\n", "date": "2024-07-13", "categories": []} {"id": "2407.12267", "title": "Generating 3D House Wireframes with Semantics", "abstract": " We present a new approach for generating 3D house wireframes with semantic\nenrichment using an autoregressive model. Unlike conventional generative models\nthat independently process vertices, edges, and faces, our approach employs a\nunified wire-based representation for improved coherence in learning 3D\nwireframe structures. By re-ordering wire sequences based on semantic meanings,\nwe facilitate seamless semantic integration during sequence generation. Our\ntwo-phase technique merges a graph-based autoencoder with a transformer-based\ndecoder to learn latent geometric tokens and generate semantic-aware\nwireframes. Through iterative prediction and decoding during inference, our\nmodel produces detailed wireframes that can be easily segmented into distinct\ncomponents, such as walls, roofs, and rooms, reflecting the semantic essence of\nthe shape. Empirical results on a comprehensive house dataset validate the\nsuperior accuracy, novelty, and semantic fidelity of our model compared to\nexisting generative models. More results and details can be found on\nhttps://vcc.tech/research/2024/3DWire.\n", "date": "2024-07-17", "categories": []} {"id": "2407.12684", "title": "4Dynamic: Text-to-4D Generation with Hybrid Priors", "abstract": " Due to the fascinating generative performance of text-to-image diffusion\nmodels, growing text-to-3D generation works explore distilling the 2D\ngenerative priors into 3D, using the score distillation sampling (SDS) loss, to\nbypass the data scarcity problem. The existing text-to-3D methods have achieved\npromising results in realism and 3D consistency, but text-to-4D generation\nstill faces challenges, including lack of realism and insufficient dynamic\nmotions. In this paper, we propose a novel method for text-to-4D generation,\nwhich ensures the dynamic amplitude and authenticity through direct supervision\nprovided by a video prior. Specifically, we adopt a text-to-video diffusion\nmodel to generate a reference video and divide 4D generation into two stages:\nstatic generation and dynamic generation. The static 3D generation is achieved\nunder the guidance of the input text and the first frame of the reference\nvideo, while in the dynamic generation stage, we introduce a customized SDS\nloss to ensure multi-view consistency, a video-based SDS loss to improve\ntemporal consistency, and most importantly, direct priors from the reference\nvideo to ensure the quality of geometry and texture. Moreover, we design a\nprior-switching training strategy to avoid conflicts between different priors\nand fully leverage the benefits of each prior. In addition, to enrich the\ngenerated motion, we further introduce a dynamic modeling representation\ncomposed of a deformation network and a topology network, which ensures dynamic\ncontinuity while modeling topological changes. Our method not only supports\ntext-to-4D generation but also enables 4D generation from monocular videos. The\ncomparison experiments demonstrate the superiority of our method compared to\nexisting methods.\n", "date": "2024-07-17", "categories": []} {"id": "2407.12739", "title": "GroundUp: Rapid Sketch-Based 3D City Massing", "abstract": " We propose GroundUp, the first sketch-based ideation tool for 3D city massing\nof urban areas. We focus on early-stage urban design, where sketching is a\ncommon tool and the design starts from balancing building volumes (masses) and\nopen spaces. With Human-Centered AI in mind, we aim to help architects quickly\nrevise their ideas by easily switching between 2D sketches and 3D models,\nallowing for smoother iteration and sharing of ideas. Inspired by feedback from\narchitects and existing workflows, our system takes as a first input a user\nsketch of multiple buildings in a top-down view. The user then draws a\nperspective sketch of the envisioned site. Our method is designed to exploit\nthe complementarity of information in the two sketches and allows users to\nquickly preview and adjust the inferred 3D shapes. Our model has two main\ncomponents. First, we propose a novel sketch-to-depth prediction network for\nperspective sketches that exploits top-down sketch shapes. Second, we use depth\ncues derived from the perspective sketch as a condition to our diffusion model,\nwhich ultimately completes the geometry in a top-down view. Thus, our final 3D\ngeometry is represented as a heightfield, allowing users to construct the city\n`from the ground up'.\n", "date": "2024-07-17", "categories": []} {"id": "2403.12957", "title": "GVGEN: Text-to-3D Generation with Volumetric Representation", "abstract": " In recent years, 3D Gaussian splatting has emerged as a powerful technique\nfor 3D reconstruction and generation, known for its fast and high-quality\nrendering capabilities. To address these shortcomings, this paper introduces a\nnovel diffusion-based framework, GVGEN, designed to efficiently generate 3D\nGaussian representations from text input. We propose two innovative\ntechniques:(1) Structured Volumetric Representation. We first arrange\ndisorganized 3D Gaussian points as a structured form GaussianVolume. This\ntransformation allows the capture of intricate texture details within a volume\ncomposed of a fixed number of Gaussians. To better optimize the representation\nof these details, we propose a unique pruning and densifying method named the\nCandidate Pool Strategy, enhancing detail fidelity through selective\noptimization. (2) Coarse-to-fine Generation Pipeline. To simplify the\ngeneration of GaussianVolume and empower the model to generate instances with\ndetailed 3D geometry, we propose a coarse-to-fine pipeline. It initially\nconstructs a basic geometric structure, followed by the prediction of complete\nGaussian attributes. Our framework, GVGEN, demonstrates superior performance in\nqualitative and quantitative assessments compared to existing 3D generation\nmethods. Simultaneously, it maintains a fast generation speed ($\\sim$7\nseconds), effectively striking a balance between quality and efficiency. Our\nproject page is: https://gvgen.github.io/\n", "date": "2024-03-19", "categories": []} {"id": "2403.15612", "title": "InterFusion: Text-Driven Generation of 3D Human-Object Interaction", "abstract": " In this study, we tackle the complex task of generating 3D human-object\ninteractions (HOI) from textual descriptions in a zero-shot text-to-3D manner.\nWe identify and address two key challenges: the unsatisfactory outcomes of\ndirect text-to-3D methods in HOI, largely due to the lack of paired\ntext-interaction data, and the inherent difficulties in simultaneously\ngenerating multiple concepts with complex spatial relationships. To effectively\naddress these issues, we present InterFusion, a two-stage framework\nspecifically designed for HOI generation. InterFusion involves human pose\nestimations derived from text as geometric priors, which simplifies the\ntext-to-3D conversion process and introduces additional constraints for\naccurate object generation. At the first stage, InterFusion extracts 3D human\nposes from a synthesized image dataset depicting a wide range of interactions,\nsubsequently mapping these poses to interaction descriptions. The second stage\nof InterFusion capitalizes on the latest developments in text-to-3D generation,\nenabling the production of realistic and high-quality 3D HOI scenes. This is\nachieved through a local-global optimization process, where the generation of\nhuman body and object is optimized separately, and jointly refined with a\nglobal optimization of the entire scene, ensuring a seamless and contextually\ncoherent integration. Our experimental results affirm that InterFusion\nsignificantly outperforms existing state-of-the-art methods in 3D HOI\ngeneration.\n", "date": "2024-03-22", "categories": []} {"id": "2404.05680", "title": "SphereHead: Stable 3D Full-head Synthesis with Spherical Tri-plane\n Representation", "abstract": " While recent advances in 3D-aware Generative Adversarial Networks (GANs) have\naided the development of near-frontal view human face synthesis, the challenge\nof comprehensively synthesizing a full 3D head viewable from all angles still\npersists. Although PanoHead proves the possibilities of using a large-scale\ndataset with images of both frontal and back views for full-head synthesis, it\noften causes artifacts for back views. Based on our in-depth analysis, we found\nthe reasons are mainly twofold. First, from network architecture perspective,\nwe found each plane in the utilized tri-plane/tri-grid representation space\ntends to confuse the features from both sides, causing \"mirroring\" artifacts\n(e.g., the glasses appear in the back). Second, from data supervision aspect,\nwe found that existing discriminator training in 3D GANs mainly focuses on the\nquality of the rendered image itself, and does not care much about its\nplausibility with the perspective from which it was rendered. This makes it\npossible to generate \"face\" in non-frontal views, due to its easiness to fool\nthe discriminator. In response, we propose SphereHead, a novel tri-plane\nrepresentation in the spherical coordinate system that fits the human head's\ngeometric characteristics and efficiently mitigates many of the generated\nartifacts. We further introduce a view-image consistency loss for the\ndiscriminator to emphasize the correspondence of the camera parameters and the\nimages. The combination of these efforts results in visually superior outcomes\nwith significantly fewer artifacts. Our code and dataset are publicly available\nat https://lhyfst.github.io/spherehead.\n", "date": "2024-04-08", "categories": []} {"id": "2312.08754", "title": "UniDream: Unifying Diffusion Priors for Relightable Text-to-3D\n Generation", "abstract": " Recent advancements in text-to-3D generation technology have significantly\nadvanced the conversion of textual descriptions into imaginative\nwell-geometrical and finely textured 3D objects. Despite these developments, a\nprevalent limitation arises from the use of RGB data in diffusion or\nreconstruction models, which often results in models with inherent lighting and\nshadows effects that detract from their realism, thereby limiting their\nusability in applications that demand accurate relighting capabilities. To\nbridge this gap, we present UniDream, a text-to-3D generation framework by\nincorporating unified diffusion priors. Our approach consists of three main\ncomponents: (1) a dual-phase training process to get albedo-normal aligned\nmulti-view diffusion and reconstruction models, (2) a progressive generation\nprocedure for geometry and albedo-textures based on Score Distillation Sample\n(SDS) using the trained reconstruction and diffusion models, and (3) an\ninnovative application of SDS for finalizing PBR generation while keeping a\nfixed albedo based on Stable Diffusion model. Extensive evaluations demonstrate\nthat UniDream surpasses existing methods in generating 3D objects with clearer\nalbedo textures, smoother surfaces, enhanced realism, and superior relighting\ncapabilities.\n", "date": "2023-12-14", "categories": []} {"id": "2312.13663", "title": "Free-Editor: Zero-shot Text-driven 3D Scene Editing", "abstract": " Text-to-Image (T2I) diffusion models have recently gained traction for their\nversatility and user-friendliness in 2D content generation and editing.\nHowever, training a diffusion model specifically for 3D scene editing is\nchallenging due to the scarcity of large-scale datasets. Currently, editing 3D\nscenes necessitates either retraining the model to accommodate various 3D edits\nor developing specific methods tailored to each unique editing type. Moreover,\nstate-of-the-art (SOTA) techniques require multiple synchronized edited images\nfrom the same scene to enable effective scene editing. Given the current\nlimitations of T2I models, achieving consistent editing effects across multiple\nimages remains difficult, leading to multi-view inconsistency in editing. This\ninconsistency undermines the performance of 3D scene editing when these images\nare utilized. In this study, we introduce a novel, training-free 3D scene\nediting technique called \\textsc{Free-Editor}, which enables users to edit 3D\nscenes without the need for model retraining during the testing phase. Our\nmethod effectively addresses the issue of multi-view style inconsistency found\nin state-of-the-art (SOTA) methods through the implementation of a single-view\nediting scheme. Specifically, we demonstrate that editing a particular 3D scene\ncan be achieved by modifying only a single view. To facilitate this, we present\nan Edit Transformer that ensures intra-view consistency and inter-view style\ntransfer using self-view and cross-view attention mechanisms, respectively. By\neliminating the need for model retraining and multi-view editing, our approach\nsignificantly reduces editing time and memory resource requirements, achieving\nruntimes approximately 20 times faster than SOTA methods. We have performed\nextensive experiments on various benchmark datasets, showcasing the diverse\nediting capabilities of our proposed technique.\n", "date": "2023-12-21", "categories": []} {"id": "2403.08733", "title": "GaussCtrl: Multi-View Consistent Text-Driven 3D Gaussian Splatting\n Editing", "abstract": " We propose GaussCtrl, a text-driven method to edit a 3D scene reconstructed\nby the 3D Gaussian Splatting (3DGS).\n Our method first renders a collection of images by using the 3DGS and edits\nthem by using a pre-trained 2D diffusion model (ControlNet) based on the input\nprompt, which is then used to optimise the 3D model.\n Our key contribution is multi-view consistent editing, which enables editing\nall images together instead of iteratively editing one image while updating the\n3D model as in previous works.\n It leads to faster editing as well as higher visual quality.\n This is achieved by the two terms:\n (a) depth-conditioned editing that enforces geometric consistency across\nmulti-view images by leveraging naturally consistent depth maps.\n (b) attention-based latent code alignment that unifies the appearance of\nedited images by conditioning their editing to several reference views through\nself and cross-view attention between images' latent representations.\n Experiments demonstrate that our method achieves faster editing and better\nvisual results than previous state-of-the-art methods.\n", "date": "2024-03-13", "categories": []} {"id": "2403.17541", "title": "WordRobe: Text-Guided Generation of Textured 3D Garments", "abstract": " In this paper, we tackle a new and challenging problem of text-driven\ngeneration of 3D garments with high-quality textures. We propose \"WordRobe\", a\nnovel framework for the generation of unposed & textured 3D garment meshes from\nuser-friendly text prompts. We achieve this by first learning a latent\nrepresentation of 3D garments using a novel coarse-to-fine training strategy\nand a loss for latent disentanglement, promoting better latent interpolation.\nSubsequently, we align the garment latent space to the CLIP embedding space in\na weakly supervised manner, enabling text-driven 3D garment generation and\nediting. For appearance modeling, we leverage the zero-shot generation\ncapability of ControlNet to synthesize view-consistent texture maps in a single\nfeed-forward inference step, thereby drastically decreasing the generation time\nas compared to existing methods. We demonstrate superior performance over\ncurrent SOTAs for learning 3D garment latent space, garment interpolation, and\ntext-driven texture synthesis, supported by quantitative evaluation and\nqualitative user study. The unposed 3D garment meshes generated using WordRobe\ncan be directly fed to standard cloth simulation & animation pipelines without\nany post-processing.\n", "date": "2024-03-26", "categories": []} {"id": "2404.06119", "title": "DreamView: Injecting View-specific Text Guidance into Text-to-3D\n Generation", "abstract": " Text-to-3D generation, which synthesizes 3D assets according to an overall\ntext description, has significantly progressed. However, a challenge arises\nwhen the specific appearances need customizing at designated viewpoints but\nreferring solely to the overall description for generating 3D objects. For\ninstance, ambiguity easily occurs when producing a T-shirt with distinct\npatterns on its front and back using a single overall text guidance. In this\nwork, we propose DreamView, a text-to-image approach enabling multi-view\ncustomization while maintaining overall consistency by adaptively injecting the\nview-specific and overall text guidance through a collaborative text guidance\ninjection module, which can also be lifted to 3D generation via score\ndistillation sampling. DreamView is trained with large-scale rendered\nmulti-view images and their corresponding view-specific texts to learn to\nbalance the separate content manipulation in each view and the global\nconsistency of the overall object, resulting in a dual achievement of\ncustomization and consistency. Consequently, DreamView empowers artists to\ndesign 3D objects creatively, fostering the creation of more innovative and\ndiverse 3D assets. Code and model will be released at\nhttps://github.com/iSEE-Laboratory/DreamView.\n", "date": "2024-04-09", "categories": []} {"id": "2407.10102", "title": "3DEgo: 3D Editing on the Go!", "abstract": " We introduce 3DEgo to address a novel problem of directly synthesizing\nphotorealistic 3D scenes from monocular videos guided by textual prompts.\nConventional methods construct a text-conditioned 3D scene through a\nthree-stage process, involving pose estimation using Structure-from-Motion\n(SfM) libraries like COLMAP, initializing the 3D model with unedited images,\nand iteratively updating the dataset with edited images to achieve a 3D scene\nwith text fidelity. Our framework streamlines the conventional multi-stage 3D\nediting process into a single-stage workflow by overcoming the reliance on\nCOLMAP and eliminating the cost of model initialization. We apply a diffusion\nmodel to edit video frames prior to 3D scene creation by incorporating our\ndesigned noise blender module for enhancing multi-view editing consistency, a\nstep that does not require additional training or fine-tuning of T2I diffusion\nmodels. 3DEgo utilizes 3D Gaussian Splatting to create 3D scenes from the\nmulti-view consistent edited frames, capitalizing on the inherent temporal\ncontinuity and explicit point cloud data. 3DEgo demonstrates remarkable editing\nprecision, speed, and adaptability across a variety of video sources, as\nvalidated by extensive evaluations on six datasets, including our own prepared\nGS25 dataset. Project Page: https://3dego.github.io/\n", "date": "2024-07-14", "categories": []} {"id": "2407.10330", "title": "Tree-D Fusion: Simulation-Ready Tree Dataset from Single Images with\n Diffusion Priors", "abstract": " We introduce Tree D-fusion, featuring the first collection of 600,000\nenvironmentally aware, 3D simulation-ready tree models generated through\nDiffusion priors. Each reconstructed 3D tree model corresponds to an image from\nGoogle's Auto Arborist Dataset, comprising street view images and associated\ngenus labels of trees across North America. Our method distills the scores of\ntwo tree-adapted diffusion models by utilizing text prompts to specify a tree\ngenus, thus facilitating shape reconstruction. This process involves\nreconstructing a 3D tree envelope filled with point markers, which are\nsubsequently utilized to estimate the tree's branching structure using the\nspace colonization algorithm conditioned on a specified genus.\n", "date": "2024-07-14", "categories": []} {"id": "2407.10558", "title": "ConTEXTure: Consistent Multiview Images to Texture", "abstract": " We introduce ConTEXTure, a generative network designed to create a texture\nmap/atlas for a given 3D mesh using images from multiple viewpoints. The\nprocess begins with generating a front-view image from a text prompt, such as\n'Napoleon, front view', describing the 3D mesh. Additional images from\ndifferent viewpoints are derived from this front-view image and camera poses\nrelative to it. ConTEXTure builds upon the TEXTure network, which uses text\nprompts for six viewpoints (e.g., 'Napoleon, front view', 'Napoleon, left\nview', etc.). However, TEXTure often generates images for non-front viewpoints\nthat do not accurately represent those viewpoints.To address this issue, we\nemploy Zero123++, which generates multiple view-consistent images for the six\nspecified viewpoints simultaneously, conditioned on the initial front-view\nimage and the depth maps of the mesh for the six viewpoints. By utilizing these\nview-consistent images, ConTEXTure learns the texture atlas from all viewpoint\nimages concurrently, unlike previous methods that do so sequentially. This\napproach ensures that the rendered images from various viewpoints, including\nback, side, bottom, and top, are free from viewpoint irregularities.\n", "date": "2024-07-15", "categories": []} {"id": "2310.06744", "title": "HiFi-123: Towards High-fidelity One Image to 3D Content Generation", "abstract": " Recent advances in diffusion models have enabled 3D generation from a single\nimage. However, current methods often produce suboptimal results for novel\nviews, with blurred textures and deviations from the reference image, limiting\ntheir practical applications. In this paper, we introduce HiFi-123, a method\ndesigned for high-fidelity and multi-view consistent 3D generation. Our\ncontributions are twofold: First, we propose a Reference-Guided Novel View\nEnhancement (RGNV) technique that significantly improves the fidelity of\ndiffusion-based zero-shot novel view synthesis methods. Second, capitalizing on\nthe RGNV, we present a novel Reference-Guided State Distillation (RGSD) loss.\nWhen incorporated into the optimization-based image-to-3D pipeline, our method\nsignificantly improves 3D generation quality, achieving state-of-the-art\nperformance. Comprehensive evaluations demonstrate the effectiveness of our\napproach over existing methods, both qualitatively and quantitatively. Video\nresults are available on the project page.\n", "date": "2023-10-10", "categories": []} {"id": "2403.18241", "title": "NeuSDFusion: A Spatial-Aware Generative Model for 3D Shape Completion,\n Reconstruction, and Generation", "abstract": " 3D shape generation aims to produce innovative 3D content adhering to\nspecific conditions and constraints. Existing methods often decompose 3D shapes\ninto a sequence of localized components, treating each element in isolation\nwithout considering spatial consistency. As a result, these approaches exhibit\nlimited versatility in 3D data representation and shape generation, hindering\ntheir ability to generate highly diverse 3D shapes that comply with the\nspecified constraints. In this paper, we introduce a novel spatial-aware 3D\nshape generation framework that leverages 2D plane representations for enhanced\n3D shape modeling. To ensure spatial coherence and reduce memory usage, we\nincorporate a hybrid shape representation technique that directly learns a\ncontinuous signed distance field representation of the 3D shape using\northogonal 2D planes. Additionally, we meticulously enforce spatial\ncorrespondences across distinct planes using a transformer-based autoencoder\nstructure, promoting the preservation of spatial relationships in the generated\n3D shapes. This yields an algorithm that consistently outperforms\nstate-of-the-art 3D shape generation methods on various tasks, including\nunconditional shape generation, multi-modal shape completion, single-view\nreconstruction, and text-to-shape synthesis. Our project page is available at\nhttps://weizheliu.github.io/NeuSDFusion/ .\n", "date": "2024-03-27", "categories": []} {"id": "2406.07516", "title": "Instant 3D Human Avatar Generation using Image Diffusion Models", "abstract": " We present AvatarPopUp, a method for fast, high quality 3D human avatar\ngeneration from different input modalities, such as images and text prompts and\nwith control over the generated pose and shape. The common theme is the use of\ndiffusion-based image generation networks that are specialized for each\nparticular task, followed by a 3D lifting network. We purposefully decouple the\ngeneration from the 3D modeling which allow us to leverage powerful image\nsynthesis priors, trained on billions of text-image pairs. We fine-tune latent\ndiffusion networks with additional image conditioning for image generation and\nback-view prediction, and to support qualitatively different multiple 3D\nhypotheses. Our partial fine-tuning approach allows to adapt the networks for\neach task without inducing catastrophic forgetting. In our experiments, we\ndemonstrate that our method produces accurate, high-quality 3D avatars with\ndiverse appearance that respect the multimodal text, image, and body control\nsignals. Our approach can produce a 3D model in as few as 2 seconds, a four\norders of magnitude speedup wrt the vast majority of existing methods, most of\nwhich solve only a subset of our tasks, and with fewer controls. AvatarPopUp\nenables applications that require the controlled 3D generation of human avatars\nat scale. The project website can be found at\nhttps://www.nikoskolot.com/avatarpopup/.\n", "date": "2024-06-11", "categories": []} {"id": "2407.06516", "title": "VQA-Diff: Exploiting VQA and Diffusion for Zero-Shot Image-to-3D Vehicle\n Asset Generation in Autonomous Driving", "abstract": " Generating 3D vehicle assets from in-the-wild observations is crucial to\nautonomous driving. Existing image-to-3D methods cannot well address this\nproblem because they learn generation merely from image RGB information without\na deeper understanding of in-the-wild vehicles (such as car models,\nmanufacturers, etc.). This leads to their poor zero-shot prediction capability\nto handle real-world observations with occlusion or tricky viewing angles. To\nsolve this problem, in this work, we propose VQA-Diff, a novel framework that\nleverages in-the-wild vehicle images to create photorealistic 3D vehicle assets\nfor autonomous driving. VQA-Diff exploits the real-world knowledge inherited\nfrom the Large Language Model in the Visual Question Answering (VQA) model for\nrobust zero-shot prediction and the rich image prior knowledge in the Diffusion\nmodel for structure and appearance generation. In particular, we utilize a\nmulti-expert Diffusion Models strategy to generate the structure information\nand employ a subject-driven structure-controlled generation mechanism to model\nappearance information. As a result, without the necessity to learn from a\nlarge-scale image-to-3D vehicle dataset collected from the real world, VQA-Diff\nstill has a robust zero-shot image-to-novel-view generation ability. We conduct\nexperiments on various datasets, including Pascal 3D+, Waymo, and Objaverse, to\ndemonstrate that VQA-Diff outperforms existing state-of-the-art methods both\nqualitatively and quantitatively.\n", "date": "2024-07-09", "categories": []} {"id": "2407.06938", "title": "RodinHD: High-Fidelity 3D Avatar Generation with Diffusion Models", "abstract": " We present RodinHD, which can generate high-fidelity 3D avatars from a\nportrait image. Existing methods fail to capture intricate details such as\nhairstyles which we tackle in this paper. We first identify an overlooked\nproblem of catastrophic forgetting that arises when fitting triplanes\nsequentially on many avatars, caused by the MLP decoder sharing scheme. To\novercome this issue, we raise a novel data scheduling strategy and a weight\nconsolidation regularization term, which improves the decoder's capability of\nrendering sharper details. Additionally, we optimize the guiding effect of the\nportrait image by computing a finer-grained hierarchical representation that\ncaptures rich 2D texture cues, and injecting them to the 3D diffusion model at\nmultiple layers via cross-attention. When trained on 46K avatars with a noise\nschedule optimized for triplanes, the resulting model can generate 3D avatars\nwith notably better details than previous methods and can generalize to\nin-the-wild portrait input.\n", "date": "2024-07-09", "categories": []} {"id": "2402.17214", "title": "CharacterGen: Efficient 3D Character Generation from Single Images with\n Multi-View Pose Canonicalization", "abstract": " In the field of digital content creation, generating high-quality 3D\ncharacters from single images is challenging, especially given the complexities\nof various body poses and the issues of self-occlusion and pose ambiguity. In\nthis paper, we present CharacterGen, a framework developed to efficiently\ngenerate 3D characters. CharacterGen introduces a streamlined generation\npipeline along with an image-conditioned multi-view diffusion model. This model\neffectively calibrates input poses to a canonical form while retaining key\nattributes of the input image, thereby addressing the challenges posed by\ndiverse poses. A transformer-based, generalizable sparse-view reconstruction\nmodel is the other core component of our approach, facilitating the creation of\ndetailed 3D models from multi-view images. We also adopt a\ntexture-back-projection strategy to produce high-quality texture maps.\nAdditionally, we have curated a dataset of anime characters, rendered in\nmultiple poses and views, to train and evaluate our model. Our approach has\nbeen thoroughly evaluated through quantitative and qualitative experiments,\nshowing its proficiency in generating 3D characters with high-quality shapes\nand textures, ready for downstream applications such as rigging and animation.\n", "date": "2024-02-27", "categories": []} {"id": "2403.14121", "title": "External Knowledge Enhanced 3D Scene Generation from Sketch", "abstract": " Generating realistic 3D scenes is challenging due to the complexity of room\nlayouts and object geometries.We propose a sketch based knowledge enhanced\ndiffusion architecture (SEK) for generating customized, diverse, and plausible\n3D scenes. SEK conditions the denoising process with a hand-drawn sketch of the\ntarget scene and cues from an object relationship knowledge base. We first\nconstruct an external knowledge base containing object relationships and then\nleverage knowledge enhanced graph reasoning to assist our model in\nunderstanding hand-drawn sketches. A scene is represented as a combination of\n3D objects and their relationships, and then incrementally diffused to reach a\nGaussian distribution.We propose a 3D denoising scene transformer that learns\nto reverse the diffusion process, conditioned by a hand-drawn sketch along with\nknowledge cues, to regressively generate the scene including the 3D object\ninstances as well as their layout. Experiments on the 3D-FRONT dataset show\nthat our model improves FID, CKL by 17.41%, 37.18% in 3D scene generation and\nFID, KID by 19.12%, 20.06% in 3D scene completion compared to the nearest\ncompetitor DiffuScene.\n", "date": "2024-03-21", "categories": []} {"id": "2404.09465", "title": "PhyScene: Physically Interactable 3D Scene Synthesis for Embodied AI", "abstract": " With recent developments in Embodied Artificial Intelligence (EAI) research,\nthere has been a growing demand for high-quality, large-scale interactive scene\ngeneration. While prior methods in scene synthesis have prioritized the\nnaturalness and realism of the generated scenes, the physical plausibility and\ninteractivity of scenes have been largely left unexplored. To address this\ndisparity, we introduce PhyScene, a novel method dedicated to generating\ninteractive 3D scenes characterized by realistic layouts, articulated objects,\nand rich physical interactivity tailored for embodied agents. Based on a\nconditional diffusion model for capturing scene layouts, we devise novel\nphysics- and interactivity-based guidance mechanisms that integrate constraints\nfrom object collision, room layout, and object reachability. Through extensive\nexperiments, we demonstrate that PhyScene effectively leverages these guidance\nfunctions for physically interactable scene synthesis, outperforming existing\nstate-of-the-art scene synthesis methods by a large margin. Our findings\nsuggest that the scenes generated by PhyScene hold considerable potential for\nfacilitating diverse skill acquisition among agents within interactive\nenvironments, thereby catalyzing further advancements in embodied AI research.\nProject website: http://physcene.github.io.\n", "date": "2024-04-15", "categories": []} {"id": "2312.10111", "title": "Plasticine3D: 3D Non-Rigid Editing with Text Guidance by Multi-View\n Embedding Optimization", "abstract": " With the help of Score Distillation Sampling (SDS) and the rapid development\nof neural 3D representations, some methods have been proposed to perform 3D\nediting such as adding additional geometries, or overwriting textures. However,\ngeneralized 3D non-rigid editing task, which requires changing both the\nstructure (posture or composition) and appearance (texture) of the original\nobject, remains to be challenging in 3D editing field. In this paper, we\npropose Plasticine3D, a novel text-guided fine-grained controlled 3D editing\npipeline that can perform 3D non-rigid editing with large structure\ndeformations. Our work divides the editing process into a geometry editing\nstage and a texture editing stage to achieve separate control of structure and\nappearance. In order to maintain the details of the original object from\ndifferent viewpoints, we propose a Multi-View-Embedding (MVE) Optimization\nstrategy to ensure that the guidance model learns the features of the original\nobject from various viewpoints. For the purpose of fine-grained control, we\npropose Embedding-Fusion (EF) to blend the original characteristics with the\nediting objectives in the embedding space, and control the extent of editing by\nadjusting the fusion rate. Furthermore, in order to address the issue of\ngradual loss of details during the generation process under high editing\nintensity, as well as the problem of insignificant editing effects in some\nscenarios, we propose Score Projection Sampling (SPS) as a replacement of score\ndistillation sampling, which introduces additional optimization phases for\nediting target enhancement and original detail maintenance, leading to better\nediting quality. Extensive experiments demonstrate the effectiveness of our\nmethod on 3D non-rigid editing tasks\n", "date": "2023-12-15", "categories": []} {"id": "2407.05388", "title": "Forest2Seq: Revitalizing Order Prior for Sequential Indoor Scene\n Synthesis", "abstract": " Synthesizing realistic 3D indoor scenes is a challenging task that\ntraditionally relies on manual arrangement and annotation by expert designers.\nRecent advances in autoregressive models have automated this process, but they\noften lack semantic understanding of the relationships and hierarchies present\nin real-world scenes, yielding limited performance. In this paper, we propose\nForest2Seq, a framework that formulates indoor scene synthesis as an\norder-aware sequential learning problem. Forest2Seq organizes the inherently\nunordered collection of scene objects into structured, ordered hierarchical\nscene trees and forests. By employing a clustering-based algorithm and a\nbreadth-first traversal, Forest2Seq derives meaningful orderings and utilizes a\ntransformer to generate realistic 3D scenes autoregressively. Experimental\nresults on standard benchmarks demonstrate Forest2Seq's superiority in\nsynthesizing more realistic scenes compared to top-performing baselines, with\nsignificant improvements in FID and KL scores. Our additional experiments for\ndownstream tasks and ablation studies also confirm the importance of\nincorporating order as a prior in 3D scene generation.\n", "date": "2024-07-07", "categories": []} {"id": "2407.06191", "title": "Tailor3D: Customized 3D Assets Editing and Generation with Dual-Side\n Images", "abstract": " Recent advances in 3D AIGC have shown promise in directly creating 3D objects\nfrom text and images, offering significant cost savings in animation and\nproduct design. However, detailed edit and customization of 3D assets remains a\nlong-standing challenge. Specifically, 3D Generation methods lack the ability\nto follow finely detailed instructions as precisely as their 2D image creation\ncounterparts. Imagine you can get a toy through 3D AIGC but with undesired\naccessories and dressing. To tackle this challenge, we propose a novel pipeline\ncalled Tailor3D, which swiftly creates customized 3D assets from editable\ndual-side images. We aim to emulate a tailor's ability to locally change\nobjects or perform overall style transfer. Unlike creating 3D assets from\nmultiple views, using dual-side images eliminates conflicts on overlapping\nareas that occur when editing individual views. Specifically, it begins by\nediting the front view, then generates the back view of the object through\nmulti-view diffusion. Afterward, it proceeds to edit the back views. Finally, a\nDual-sided LRM is proposed to seamlessly stitch together the front and back 3D\nfeatures, akin to a tailor sewing together the front and back of a garment. The\nDual-sided LRM rectifies imperfect consistencies between the front and back\nviews, enhancing editing capabilities and reducing memory burdens while\nseamlessly integrating them into a unified 3D representation with the LoRA\nTriplane Transformer. Experimental results demonstrate Tailor3D's effectiveness\nacross various 3D generation and editing tasks, including 3D generative fill\nand style transfer. It provides a user-friendly, efficient solution for editing\n3D assets, with each editing step taking only seconds to complete.\n", "date": "2024-07-08", "categories": []} {"id": "2401.05293", "title": "Score Distillation Sampling with Learned Manifold Corrective", "abstract": " Score Distillation Sampling (SDS) is a recent but already widely popular\nmethod that relies on an image diffusion model to control optimization problems\nusing text prompts. In this paper, we conduct an in-depth analysis of the SDS\nloss function, identify an inherent problem with its formulation, and propose a\nsurprisingly easy but effective fix. Specifically, we decompose the loss into\ndifferent factors and isolate the component responsible for noisy gradients. In\nthe original formulation, high text guidance is used to account for the noise,\nleading to unwanted side effects such as oversaturation or repeated detail.\nInstead, we train a shallow network mimicking the timestep-dependent frequency\nbias of the image diffusion model in order to effectively factor it out. We\ndemonstrate the versatility and the effectiveness of our novel loss formulation\nthrough qualitative and quantitative experiments, including optimization-based\nimage synthesis and editing, zero-shot image translation network training, and\ntext-to-3D synthesis.\n", "date": "2024-01-10", "categories": []} {"id": "2407.02599", "title": "Meta 3D Gen", "abstract": " We introduce Meta 3D Gen (3DGen), a new state-of-the-art, fast pipeline for\ntext-to-3D asset generation. 3DGen offers 3D asset creation with high prompt\nfidelity and high-quality 3D shapes and textures in under a minute. It supports\nphysically-based rendering (PBR), necessary for 3D asset relighting in\nreal-world applications. Additionally, 3DGen supports generative retexturing of\npreviously generated (or artist-created) 3D shapes using additional textual\ninputs provided by the user. 3DGen integrates key technical components, Meta 3D\nAssetGen and Meta 3D TextureGen, that we developed for text-to-3D and\ntext-to-texture generation, respectively. By combining their strengths, 3DGen\nrepresents 3D objects simultaneously in three ways: in view space, in\nvolumetric space, and in UV (or texture) space. The integration of these two\ntechniques achieves a win rate of 68% with respect to the single-stage model.\nWe compare 3DGen to numerous industry baselines, and show that it outperforms\nthem in terms of prompt fidelity and visual quality for complex textual\nprompts, while being significantly faster.\n", "date": "2024-07-02", "categories": []} {"id": "2404.01241", "title": "StructLDM: Structured Latent Diffusion for 3D Human Generation", "abstract": " Recent 3D human generative models have achieved remarkable progress by\nlearning 3D-aware GANs from 2D images. However, existing 3D human generative\nmethods model humans in a compact 1D latent space, ignoring the articulated\nstructure and semantics of human body topology. In this paper, we explore more\nexpressive and higher-dimensional latent space for 3D human modeling and\npropose StructLDM, a diffusion-based unconditional 3D human generative model,\nwhich is learned from 2D images. StructLDM solves the challenges imposed due to\nthe high-dimensional growth of latent space with three key designs: 1) A\nsemantic structured latent space defined on the dense surface manifold of a\nstatistical human body template. 2) A structured 3D-aware auto-decoder that\nfactorizes the global latent space into several semantic body parts\nparameterized by a set of conditional structured local NeRFs anchored to the\nbody template, which embeds the properties learned from the 2D training data\nand can be decoded to render view-consistent humans under different poses and\nclothing styles. 3) A structured latent diffusion model for generative human\nappearance sampling. Extensive experiments validate StructLDM's\nstate-of-the-art generation performance and illustrate the expressiveness of\nthe structured latent space over the well-adopted 1D latent space. Notably,\nStructLDM enables different levels of controllable 3D human generation and\nediting, including pose/view/shape control, and high-level tasks including\ncompositional generations, part-aware clothing editing, 3D virtual try-on, etc.\nOur project page is at: https://taohuumd.github.io/projects/StructLDM/.\n", "date": "2024-04-01", "categories": []} {"id": "2407.02040", "title": "ScaleDreamer: Scalable Text-to-3D Synthesis with Asynchronous Score\n Distillation", "abstract": " By leveraging the text-to-image diffusion priors, score distillation can\nsynthesize 3D contents without paired text-3D training data. Instead of\nspending hours of online optimization per text prompt, recent studies have been\nfocused on learning a text-to-3D generative network for amortizing multiple\ntext-3D relations, which can synthesize 3D contents in seconds. However,\nexisting score distillation methods are hard to scale up to a large amount of\ntext prompts due to the difficulties in aligning pretrained diffusion prior\nwith the distribution of rendered images from various text prompts. Current\nstate-of-the-arts such as Variational Score Distillation finetune the\npretrained diffusion model to minimize the noise prediction error so as to\nalign the distributions, which are however unstable to train and will impair\nthe model's comprehension capability to numerous text prompts. Based on the\nobservation that the diffusion models tend to have lower noise prediction\nerrors at earlier timesteps, we propose Asynchronous Score Distillation (ASD),\nwhich minimizes the noise prediction error by shifting the diffusion timestep\nto earlier ones. ASD is stable to train and can scale up to 100k prompts. It\nreduces the noise prediction error without changing the weights of pre-trained\ndiffusion model, thus keeping its strong comprehension capability to prompts.\nWe conduct extensive experiments across different 2D diffusion models,\nincluding Stable Diffusion and MVDream, and text-to-3D generators, including\nHyper-iNGP, 3DConv-Net and Triplane-Transformer. The results demonstrate ASD's\neffectiveness in stable 3D generator training, high-quality 3D content\nsynthesis, and its superior prompt-consistency, especially under large prompt\ncorpus.\n", "date": "2024-07-02", "categories": []} {"id": "2407.02430", "title": "Meta 3D TextureGen: Fast and Consistent Texture Generation for 3D\n Objects", "abstract": " The recent availability and adaptability of text-to-image models has sparked\na new era in many related domains that benefit from the learned text priors as\nwell as high-quality and fast generation capabilities, one of which is texture\ngeneration for 3D objects. Although recent texture generation methods achieve\nimpressive results by using text-to-image networks, the combination of global\nconsistency, quality, and speed, which is crucial for advancing texture\ngeneration to real-world applications, remains elusive. To that end, we\nintroduce Meta 3D TextureGen: a new feedforward method comprised of two\nsequential networks aimed at generating high-quality and globally consistent\ntextures for arbitrary geometries of any complexity degree in less than 20\nseconds. Our method achieves state-of-the-art results in quality and speed by\nconditioning a text-to-image model on 3D semantics in 2D space and fusing them\ninto a complete and high-resolution UV texture map, as demonstrated by\nextensive qualitative and quantitative evaluations. In addition, we introduce a\ntexture enhancement network that is capable of up-scaling any texture by an\narbitrary ratio, producing 4k pixel resolution textures.\n", "date": "2024-07-02", "categories": []} {"id": "2407.02445", "title": "Meta 3D AssetGen: Text-to-Mesh Generation with High-Quality Geometry,\n Texture, and PBR Materials", "abstract": " We present Meta 3D AssetGen (AssetGen), a significant advancement in\ntext-to-3D generation which produces faithful, high-quality meshes with texture\nand material control. Compared to works that bake shading in the 3D object's\nappearance, AssetGen outputs physically-based rendering (PBR) materials,\nsupporting realistic relighting. AssetGen generates first several views of the\nobject with factored shaded and albedo appearance channels, and then\nreconstructs colours, metalness and roughness in 3D, using a deferred shading\nloss for efficient supervision. It also uses a sign-distance function to\nrepresent 3D shape more reliably and introduces a corresponding loss for direct\nshape supervision. This is implemented using fused kernels for high memory\nefficiency. After mesh extraction, a texture refinement transformer operating\nin UV space significantly improves sharpness and details. AssetGen achieves 17%\nimprovement in Chamfer Distance and 40% in LPIPS over the best concurrent work\nfor few-view reconstruction, and a human preference of 72% over the best\nindustry competitors of comparable speed, including those that support PBR.\nProject page with generated assets: https://assetgen.github.io\n", "date": "2024-07-02", "categories": []} {"id": "2406.16695", "title": "Geometry-Aware Score Distillation via 3D Consistent Noising and Gradient\n Consistency Modeling", "abstract": " Score distillation sampling (SDS), the methodology in which the score from\npretrained 2D diffusion models is distilled into 3D representation, has\nrecently brought significant advancements in text-to-3D generation task.\nHowever, this approach is still confronted with critical geometric\ninconsistency problems such as the Janus problem. Starting from a hypothesis\nthat such inconsistency problems may be induced by multiview inconsistencies\nbetween 2D scores predicted from various viewpoints, we introduce GSD, a simple\nand general plug-and-play framework for incorporating 3D consistency and\ntherefore geometry awareness into the SDS process. Our methodology is composed\nof three components: 3D consistent noising, designed to produce 3D consistent\nnoise maps that perfectly follow the standard Gaussian distribution,\ngeometry-based gradient warping for identifying correspondences between\npredicted gradients of different viewpoints, and novel gradient consistency\nloss to optimize the scene geometry toward producing more consistent gradients.\nWe demonstrate that our method significantly improves performance, successfully\naddressing the geometric inconsistency problems in text-to-3D generation task\nwith minimal computation cost and being compatible with existing score\ndistillation-based models. Our project page is available at\nhttps://ku-cvlab.github.io/GSD/.\n", "date": "2024-06-24", "categories": []} {"id": "2406.18539", "title": "TexPainter: Generative Mesh Texturing with Multi-view Consistency", "abstract": " The recent success of pre-trained diffusion models unlocks the possibility of\nthe automatic generation of textures for arbitrary 3D meshes in the wild.\nHowever, these models are trained in the screen space, while converting them to\na multi-view consistent texture image poses a major obstacle to the output\nquality. In this paper, we propose a novel method to enforce multi-view\nconsistency. Our method is based on the observation that latent space in a\npre-trained diffusion model is noised separately for each camera view, making\nit difficult to achieve multi-view consistency by directly manipulating the\nlatent codes. Based on the celebrated Denoising Diffusion Implicit Models\n(DDIM) scheme, we propose to use an optimization-based color-fusion to enforce\nconsistency and indirectly modify the latent codes by gradient\nback-propagation. Our method further relaxes the sequential dependency\nassumption among the camera views. By evaluating on a series of general 3D\nmodels, we find our simple approach improves consistency and overall quality of\nthe generated textures as compared to competing state-of-the-arts. Our\nimplementation is available at: https://github.com/Quantuman134/TexPainter\n", "date": "2024-05-17", "categories": []} {"id": "2406.18462", "title": "GaussianDreamerPro: Text to Manipulable 3D Gaussians with Highly\n Enhanced Quality", "abstract": " Recently, 3D Gaussian splatting (3D-GS) has achieved great success in\nreconstructing and rendering real-world scenes. To transfer the high rendering\nquality to generation tasks, a series of research works attempt to generate\n3D-Gaussian assets from text. However, the generated assets have not achieved\nthe same quality as those in reconstruction tasks. We observe that Gaussians\ntend to grow without control as the generation process may cause indeterminacy.\nAiming at highly enhancing the generation quality, we propose a novel framework\nnamed GaussianDreamerPro. The main idea is to bind Gaussians to reasonable\ngeometry, which evolves over the whole generation process. Along different\nstages of our framework, both the geometry and appearance can be enriched\nprogressively. The final output asset is constructed with 3D Gaussians bound to\nmesh, which shows significantly enhanced details and quality compared with\nprevious methods. Notably, the generated asset can also be seamlessly\nintegrated into downstream manipulation pipelines, e.g. animation, composition,\nand simulation etc., greatly promoting its potential in wide applications.\nDemos are available at https://taoranyi.com/gaussiandreamerpro/.\n", "date": "2024-06-26", "categories": []} {"id": "2312.03806", "title": "XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies", "abstract": " We present XCube (abbreviated as $\\mathcal{X}^3$), a novel generative model\nfor high-resolution sparse 3D voxel grids with arbitrary attributes. Our model\ncan generate millions of voxels with a finest effective resolution of up to\n$1024^3$ in a feed-forward fashion without time-consuming test-time\noptimization. To achieve this, we employ a hierarchical voxel latent diffusion\nmodel which generates progressively higher resolution grids in a coarse-to-fine\nmanner using a custom framework built on the highly efficient VDB data\nstructure. Apart from generating high-resolution objects, we demonstrate the\neffectiveness of XCube on large outdoor scenes at scales of 100m$\\times$100m\nwith a voxel size as small as 10cm. We observe clear qualitative and\nquantitative improvements over past approaches. In addition to unconditional\ngeneration, we show that our model can be used to solve a variety of tasks such\nas user-guided editing, scene completion from a single scan, and text-to-3D.\nThe source code and more results can be found at\nhttps://research.nvidia.com/labs/toronto-ai/xcube/.\n", "date": "2023-12-06", "categories": []} {"id": "2406.17396", "title": "SyncNoise: Geometrically Consistent Noise Prediction for Text-based 3D\n Scene Editing", "abstract": " Text-based 2D diffusion models have demonstrated impressive capabilities in\nimage generation and editing. Meanwhile, the 2D diffusion models also exhibit\nsubstantial potentials for 3D editing tasks. However, how to achieve consistent\nedits across multiple viewpoints remains a challenge. While the iterative\ndataset update method is capable of achieving global consistency, it suffers\nfrom slow convergence and over-smoothed textures. We propose SyncNoise, a novel\ngeometry-guided multi-view consistent noise editing approach for high-fidelity\n3D scene editing. SyncNoise synchronously edits multiple views with 2D\ndiffusion models while enforcing multi-view noise predictions to be\ngeometrically consistent, which ensures global consistency in both semantic\nstructure and low-frequency appearance. To further enhance local consistency in\nhigh-frequency details, we set a group of anchor views and propagate them to\ntheir neighboring frames through cross-view reprojection. To improve the\nreliability of multi-view correspondences, we introduce depth supervision\nduring training to enhance the reconstruction of precise geometries. Our method\nachieves high-quality 3D editing results respecting the textual instructions,\nespecially in scenes with complex textures, by enhancing geometric consistency\nat the noise and pixel levels.\n", "date": "2024-06-25", "categories": []} {"id": "2406.17601", "title": "Director3D: Real-world Camera Trajectory and 3D Scene Generation from\n Text", "abstract": " Recent advancements in 3D generation have leveraged synthetic datasets with\nground truth 3D assets and predefined cameras. However, the potential of\nadopting real-world datasets, which can produce significantly more realistic 3D\nscenes, remains largely unexplored. In this work, we delve into the key\nchallenge of the complex and scene-specific camera trajectories found in\nreal-world captures. We introduce Director3D, a robust open-world text-to-3D\ngeneration framework, designed to generate both real-world 3D scenes and\nadaptive camera trajectories. To achieve this, (1) we first utilize a\nTrajectory Diffusion Transformer, acting as the Cinematographer, to model the\ndistribution of camera trajectories based on textual descriptions. (2) Next, a\nGaussian-driven Multi-view Latent Diffusion Model serves as the Decorator,\nmodeling the image sequence distribution given the camera trajectories and\ntexts. This model, fine-tuned from a 2D diffusion model, directly generates\npixel-aligned 3D Gaussians as an immediate 3D scene representation for\nconsistent denoising. (3) Lastly, the 3D Gaussians are refined by a novel SDS++\nloss as the Detailer, which incorporates the prior of the 2D diffusion model.\nExtensive experiments demonstrate that Director3D outperforms existing methods,\noffering superior performance in real-world 3D generation.\n", "date": "2024-06-25", "categories": []} {"id": "2406.16273", "title": "YouDream: Generating Anatomically Controllable Consistent Text-to-3D\n Animals", "abstract": " 3D generation guided by text-to-image diffusion models enables the creation\nof visually compelling assets. However previous methods explore generation\nbased on image or text. The boundaries of creativity are limited by what can be\nexpressed through words or the images that can be sourced. We present YouDream,\na method to generate high-quality anatomically controllable animals. YouDream\nis guided using a text-to-image diffusion model controlled by 2D views of a 3D\npose prior. Our method generates 3D animals that are not possible to create\nusing previous text-to-3D generative methods. Additionally, our method is\ncapable of preserving anatomic consistency in the generated animals, an area\nwhere prior text-to-3D approaches often struggle. Moreover, we design a fully\nautomated pipeline for generating commonly found animals. To circumvent the\nneed for human intervention to create a 3D pose, we propose a multi-agent LLM\nthat adapts poses from a limited library of animal 3D poses to represent the\ndesired animal. A user study conducted on the outcomes of YouDream demonstrates\nthe preference of the animal models generated by our method over others.\nTurntable results and code are released at https://youdream3d.github.io/\n", "date": "2024-06-24", "categories": []} {"id": "2406.16815", "title": "ClotheDreamer: Text-Guided Garment Generation with 3D Gaussians", "abstract": " High-fidelity 3D garment synthesis from text is desirable yet challenging for\ndigital avatar creation. Recent diffusion-based approaches via Score\nDistillation Sampling (SDS) have enabled new possibilities but either\nintricately couple with human body or struggle to reuse. We introduce\nClotheDreamer, a 3D Gaussian-based method for generating wearable,\nproduction-ready 3D garment assets from text prompts. We propose a novel\nrepresentation Disentangled Clothe Gaussian Splatting (DCGS) to enable separate\noptimization. DCGS represents clothed avatar as one Gaussian model but freezes\nbody Gaussian splats. To enhance quality and completeness, we incorporate\nbidirectional SDS to supervise clothed avatar and garment RGBD renderings\nrespectively with pose conditions and propose a new pruning strategy for loose\nclothing. Our approach can also support custom clothing templates as input.\nBenefiting from our design, the synthetic 3D garment can be easily applied to\nvirtual try-on and support physically accurate animation. Extensive experiments\nshowcase our method's superior and competitive performance. Our project page is\nat https://ggxxii.github.io/clothedreamer.\n", "date": "2024-06-24", "categories": []} {"id": "2406.14964", "title": "VividDreamer: Towards High-Fidelity and Efficient Text-to-3D Generation", "abstract": " Text-to-3D generation aims to create 3D assets from text-to-image diffusion\nmodels. However, existing methods face an inherent bottleneck in generation\nquality because the widely-used objectives such as Score Distillation Sampling\n(SDS) inappropriately omit U-Net jacobians for swift generation, leading to\nsignificant bias compared to the \"true\" gradient obtained by full denoising\nsampling. This bias brings inconsistent updating direction, resulting in\nimplausible 3D generation e.g., color deviation, Janus problem, and\nsemantically inconsistent details). In this work, we propose Pose-dependent\nConsistency Distillation Sampling (PCDS), a novel yet efficient objective for\ndiffusion-based 3D generation tasks. Specifically, PCDS builds the\npose-dependent consistency function within diffusion trajectories, allowing to\napproximate true gradients through minimal sampling steps (1-3). Compared to\nSDS, PCDS can acquire a more accurate updating direction with the same sampling\ntime (1 sampling step), while enabling few-step (2-3) sampling to trade compute\nfor higher generation quality. For efficient generation, we propose a\ncoarse-to-fine optimization strategy, which first utilizes 1-step PCDS to\ncreate the basic structure of 3D objects, and then gradually increases PCDS\nsteps to generate fine-grained details. Extensive experiments demonstrate that\nour approach outperforms the state-of-the-art in generation quality and\ntraining efficiency, conspicuously alleviating the implausible 3D generation\nissues caused by the deviated updating direction. Moreover, it can be simply\napplied to many 3D generative applications to yield impressive 3D assets,\nplease see our project page: https://narcissusex.github.io/VividDreamer.\n", "date": "2024-06-21", "categories": []} {"id": "2406.13099", "title": "Sampling 3D Gaussian Scenes in Seconds with Latent Diffusion Models", "abstract": " We present a latent diffusion model over 3D scenes, that can be trained using\nonly 2D image data. To achieve this, we first design an autoencoder that maps\nmulti-view images to 3D Gaussian splats, and simultaneously builds a compressed\nlatent representation of these splats. Then, we train a multi-view diffusion\nmodel over the latent space to learn an efficient generative model. This\npipeline does not require object masks nor depths, and is suitable for complex\nscenes with arbitrary camera positions. We conduct careful experiments on two\nlarge-scale datasets of complex real-world scenes -- MVImgNet and\nRealEstate10K. We show that our approach enables generating 3D scenes in as\nlittle as 0.2 seconds, either from scratch, from a single input view, or from\nsparse input views. It produces diverse and high-quality results while running\nan order of magnitude faster than non-latent diffusion models and earlier\nNeRF-based generative models\n", "date": "2024-06-18", "categories": []} {"id": "2406.13897", "title": "CLAY: A Controllable Large-scale Generative Model for Creating\n High-quality 3D Assets", "abstract": " In the realm of digital creativity, our potential to craft intricate 3D\nworlds from imagination is often hampered by the limitations of existing\ndigital tools, which demand extensive expertise and efforts. To narrow this\ndisparity, we introduce CLAY, a 3D geometry and material generator designed to\neffortlessly transform human imagination into intricate 3D digital structures.\nCLAY supports classic text or image inputs as well as 3D-aware controls from\ndiverse primitives (multi-view images, voxels, bounding boxes, point clouds,\nimplicit representations, etc). At its core is a large-scale generative model\ncomposed of a multi-resolution Variational Autoencoder (VAE) and a minimalistic\nlatent Diffusion Transformer (DiT), to extract rich 3D priors directly from a\ndiverse range of 3D geometries. Specifically, it adopts neural fields to\nrepresent continuous and complete surfaces and uses a geometry generative\nmodule with pure transformer blocks in latent space. We present a progressive\ntraining scheme to train CLAY on an ultra large 3D model dataset obtained\nthrough a carefully designed processing pipeline, resulting in a 3D native\ngeometry generator with 1.5 billion parameters. For appearance generation, CLAY\nsets out to produce physically-based rendering (PBR) textures by employing a\nmulti-view material diffusion model that can generate 2K resolution textures\nwith diffuse, roughness, and metallic modalities. We demonstrate using CLAY for\na range of controllable 3D asset creations, from sketchy conceptual designs to\nproduction ready assets with intricate details. Even first time users can\neasily use CLAY to bring their vivid 3D imaginations to life, unleashing\nunlimited creativity.\n", "date": "2024-05-30", "categories": []} {"id": "2405.20853", "title": "MeshXL: Neural Coordinate Field for Generative 3D Foundation Models", "abstract": " The polygon mesh representation of 3D data exhibits great flexibility, fast\nrendering speed, and storage efficiency, which is widely preferred in various\napplications. However, given its unstructured graph representation, the direct\ngeneration of high-fidelity 3D meshes is challenging. Fortunately, with a\npre-defined ordering strategy, 3D meshes can be represented as sequences, and\nthe generation process can be seamlessly treated as an auto-regressive problem.\nIn this paper, we validate the Neural Coordinate Field (NeurCF), an explicit\ncoordinate representation with implicit neural embeddings, is a\nsimple-yet-effective representation for large-scale sequential mesh modeling.\nAfter that, we present MeshXL, a family of generative pre-trained\nauto-regressive models, which addresses the process of 3D mesh generation with\nmodern large language model approaches. Extensive experiments show that MeshXL\nis able to generate high-quality 3D meshes, and can also serve as foundation\nmodels for various down-stream applications.\n", "date": "2024-05-31", "categories": []} {"id": "2401.16465", "title": "DressCode: Autoregressively Sewing and Generating Garments from Text\n Guidance", "abstract": " Apparel's significant role in human appearance underscores the importance of\ngarment digitalization for digital human creation. Recent advances in 3D\ncontent creation are pivotal for digital human creation. Nonetheless, garment\ngeneration from text guidance is still nascent. We introduce a text-driven 3D\ngarment generation framework, DressCode, which aims to democratize design for\nnovices and offer immense potential in fashion design, virtual try-on, and\ndigital human creation. We first introduce SewingGPT, a GPT-based architecture\nintegrating cross-attention with text-conditioned embedding to generate sewing\npatterns with text guidance. We then tailor a pre-trained Stable Diffusion to\ngenerate tile-based Physically-based Rendering (PBR) textures for the garments.\nBy leveraging a large language model, our framework generates CG-friendly\ngarments through natural language interaction. It also facilitates pattern\ncompletion and texture editing, streamlining the design process through\nuser-friendly interaction. This framework fosters innovation by allowing\ncreators to freely experiment with designs and incorporate unique elements into\ntheir work. With comprehensive evaluations and comparisons with other\nstate-of-the-art methods, our method showcases superior quality and alignment\nwith input prompts. User studies further validate our high-quality rendering\nresults, highlighting its practical utility and potential in production\nsettings. Our project page is https://IHe-KaiI.github.io/DressCode/.\n", "date": "2024-01-29", "categories": []} {"id": "2406.11202", "title": "Consistency^2: Consistent and Fast 3D Painting with Latent Consistency\n Models", "abstract": " Generative 3D Painting is among the top productivity boosters in\nhigh-resolution 3D asset management and recycling. Ever since text-to-image\nmodels became accessible for inference on consumer hardware, the performance of\n3D Painting methods has consistently improved and is currently close to\nplateauing. At the core of most such models lies denoising diffusion in the\nlatent space, an inherently time-consuming iterative process. Multiple\ntechniques have been developed recently to accelerate generation and reduce\nsampling iterations by orders of magnitude. Designed for 2D generative imaging,\nthese techniques do not come with recipes for lifting them into 3D. In this\npaper, we address this shortcoming by proposing a Latent Consistency Model\n(LCM) adaptation for the task at hand. We analyze the strengths and weaknesses\nof the proposed model and evaluate it quantitatively and qualitatively. Based\non the Objaverse dataset samples study, our 3D painting method attains strong\npreference in all evaluations. Source code is available at\nhttps://github.com/kongdai123/consistency2.\n", "date": "2024-06-17", "categories": []} {"id": "2406.11283", "title": "Enhancing Generalizability of Representation Learning for Data-Efficient\n 3D Scene Understanding", "abstract": " The field of self-supervised 3D representation learning has emerged as a\npromising solution to alleviate the challenge presented by the scarcity of\nextensive, well-annotated datasets. However, it continues to be hindered by the\nlack of diverse, large-scale, real-world 3D scene datasets for source data. To\naddress this shortfall, we propose Generalizable Representation Learning (GRL),\nwhere we devise a generative Bayesian network to produce diverse synthetic\nscenes with real-world patterns, and conduct pre-training with a joint\nobjective. By jointly learning a coarse-to-fine contrastive learning task and\nan occlusion-aware reconstruction task, the model is primed with transferable,\ngeometry-informed representations. Post pre-training on synthetic data, the\nacquired knowledge of the model can be seamlessly transferred to two principal\ndownstream tasks associated with 3D scene understanding, namely 3D object\ndetection and 3D semantic segmentation, using real-world benchmark datasets. A\nthorough series of experiments robustly display our method's consistent\nsuperiority over existing state-of-the-art pre-training approaches.\n", "date": "2024-06-17", "categories": []} {"id": "2311.15421", "title": "Wired Perspectives: Multi-View Wire Art Embraces Generative AI", "abstract": " Creating multi-view wire art (MVWA), a static 3D sculpture with diverse\ninterpretations from different viewpoints, is a complex task even for skilled\nartists. In response, we present DreamWire, an AI system enabling everyone to\ncraft MVWA easily. Users express their vision through text prompts or\nscribbles, freeing them from intricate 3D wire organisation. Our approach\nsynergises 3D B\\'ezier curves, Prim's algorithm, and knowledge distillation\nfrom diffusion models or their variants (e.g., ControlNet). This blend enables\nthe system to represent 3D wire art, ensuring spatial continuity and overcoming\ndata scarcity. Extensive evaluation and analysis are conducted to shed insight\non the inner workings of the proposed system, including the trade-off between\nconnectivity and visual aesthetics.\n", "date": "2023-11-26", "categories": []} {"id": "2406.04254", "title": "GeoGen: Geometry-Aware Generative Modeling via Signed Distance Functions", "abstract": " We introduce a new generative approach for synthesizing 3D geometry and\nimages from single-view collections. Most existing approaches predict\nvolumetric density to render multi-view consistent images. By employing\nvolumetric rendering using neural radiance fields, they inherit a key\nlimitation: the generated geometry is noisy and unconstrained, limiting the\nquality and utility of the output meshes. To address this issue, we propose\nGeoGen, a new SDF-based 3D generative model trained in an end-to-end manner.\nInitially, we reinterpret the volumetric density as a Signed Distance Function\n(SDF). This allows us to introduce useful priors to generate valid meshes.\nHowever, those priors prevent the generative model from learning details,\nlimiting the applicability of the method to real-world scenarios. To alleviate\nthat problem, we make the transformation learnable and constrain the rendered\ndepth map to be consistent with the zero-level set of the SDF. Through the lens\nof adversarial training, we encourage the network to produce higher fidelity\ndetails on the output meshes. For evaluation, we introduce a synthetic dataset\nof human avatars captured from 360-degree camera angles, to overcome the\nchallenges presented by real-world datasets, which often lack 3D consistency\nand do not cover all camera angles. Our experiments on multiple datasets show\nthat GeoGen produces visually and quantitatively better geometry than the\nprevious generative models based on neural radiance fields.\n", "date": "2024-06-06", "categories": []} {"id": "2406.09850", "title": "GradeADreamer: Enhanced Text-to-3D Generation Using Gaussian Splatting\n and Multi-View Diffusion", "abstract": " Text-to-3D generation has shown promising results, yet common challenges such\nas the Multi-face Janus problem and extended generation time for high-quality\nassets. In this paper, we address these issues by introducing a novel\nthree-stage training pipeline called GradeADreamer. This pipeline is capable of\nproducing high-quality assets with a total generation time of under 30 minutes\nusing only a single RTX 3090 GPU. Our proposed method employs a Multi-view\nDiffusion Model, MVDream, to generate Gaussian Splats as a prior, followed by\nrefining geometry and texture using StableDiffusion. Experimental results\ndemonstrate that our approach significantly mitigates the Multi-face Janus\nproblem and achieves the highest average user preference ranking compared to\nprevious state-of-the-art methods. The project code is available at\nhttps://github.com/trapoom555/GradeADreamer.\n", "date": "2024-06-14", "categories": []} {"id": "2406.10000", "title": "OrientDream: Streamlining Text-to-3D Generation with Explicit\n Orientation Control", "abstract": " In the evolving landscape of text-to-3D technology, Dreamfusion has showcased\nits proficiency by utilizing Score Distillation Sampling (SDS) to optimize\nimplicit representations such as NeRF. This process is achieved through the\ndistillation of pretrained large-scale text-to-image diffusion models. However,\nDreamfusion encounters fidelity and efficiency constraints: it faces the\nmulti-head Janus issue and exhibits a relatively slow optimization process. To\ncircumvent these challenges, we introduce OrientDream, a camera orientation\nconditioned framework designed for efficient and multi-view consistent 3D\ngeneration from textual prompts. Our strategy emphasizes the implementation of\nan explicit camera orientation conditioned feature in the pre-training of a 2D\ntext-to-image diffusion module. This feature effectively utilizes data from\nMVImgNet, an extensive external multi-view dataset, to refine and bolster its\nfunctionality. Subsequently, we utilize the pre-conditioned 2D images as a\nbasis for optimizing a randomly initialized implicit representation (NeRF).\nThis process is significantly expedited by a decoupled back-propagation\ntechnique, allowing for multiple updates of implicit parameters per\noptimization cycle. Our experiments reveal that our method not only produces\nhigh-quality NeRF models with consistent multi-view properties but also\nachieves an optimization speed significantly greater than existing methods, as\nquantified by comparative metrics.\n", "date": "2024-06-14", "categories": []} {"id": "2401.09050", "title": "Consistent3D: Towards Consistent High-Fidelity Text-to-3D Generation\n with Deterministic Sampling Prior", "abstract": " Score distillation sampling (SDS) and its variants have greatly boosted the\ndevelopment of text-to-3D generation, but are vulnerable to geometry collapse\nand poor textures yet. To solve this issue, we first deeply analyze the SDS and\nfind that its distillation sampling process indeed corresponds to the\ntrajectory sampling of a stochastic differential equation (SDE): SDS samples\nalong an SDE trajectory to yield a less noisy sample which then serves as a\nguidance to optimize a 3D model. However, the randomness in SDE sampling often\nleads to a diverse and unpredictable sample which is not always less noisy, and\nthus is not a consistently correct guidance, explaining the vulnerability of\nSDS. Since for any SDE, there always exists an ordinary differential equation\n(ODE) whose trajectory sampling can deterministically and consistently converge\nto the desired target point as the SDE, we propose a novel and effective\n\"Consistent3D\" method that explores the ODE deterministic sampling prior for\ntext-to-3D generation. Specifically, at each training iteration, given a\nrendered image by a 3D model, we first estimate its desired 3D score function\nby a pre-trained 2D diffusion model, and build an ODE for trajectory sampling.\nNext, we design a consistency distillation sampling loss which samples along\nthe ODE trajectory to generate two adjacent samples and uses the less noisy\nsample to guide another more noisy one for distilling the deterministic prior\ninto the 3D model. Experimental results show the efficacy of our Consistent3D\nin generating high-fidelity and diverse 3D objects and large-scale scenes, as\nshown in Fig. 1. The codes are available at\nhttps://github.com/sail-sg/Consistent3D.\n", "date": "2024-01-17", "categories": []} {"id": "2405.03894", "title": "MVDiff: Scalable and Flexible Multi-View Diffusion for 3D Object\n Reconstruction from Single-View", "abstract": " Generating consistent multiple views for 3D reconstruction tasks is still a\nchallenge to existing image-to-3D diffusion models. Generally, incorporating 3D\nrepresentations into diffusion model decrease the model's speed as well as\ngeneralizability and quality. This paper proposes a general framework to\ngenerate consistent multi-view images from single image or leveraging scene\nrepresentation transformer and view-conditioned diffusion model. In the model,\nwe introduce epipolar geometry constraints and multi-view attention to enforce\n3D consistency. From as few as one image input, our model is able to generate\n3D meshes surpassing baselines methods in evaluation metrics, including PSNR,\nSSIM and LPIPS.\n", "date": "2024-05-06", "categories": []} {"id": "2406.08953", "title": "Preserving Identity with Variational Score for General-purpose 3D\n Editing", "abstract": " We present Piva (Preserving Identity with Variational Score Distillation), a\nnovel optimization-based method for editing images and 3D models based on\ndiffusion models. Specifically, our approach is inspired by the recently\nproposed method for 2D image editing - Delta Denoising Score (DDS). We pinpoint\nthe limitations in DDS for 2D and 3D editing, which causes detail loss and\nover-saturation. To address this, we propose an additional score distillation\nterm that enforces identity preservation. This results in a more stable editing\nprocess, gradually optimizing NeRF models to match target prompts while\nretaining crucial input characteristics. We demonstrate the effectiveness of\nour approach in zero-shot image and neural field editing. Our method\nsuccessfully alters visual attributes, adds both subtle and substantial\nstructural elements, translates shapes, and achieves competitive results on\nstandard 2D and 3D editing benchmarks. Additionally, our method imposes no\nconstraints like masking or pre-training, making it compatible with a wide\nrange of pre-trained diffusion models. This allows for versatile editing\nwithout needing neural field-to-mesh conversion, offering a more user-friendly\nexperience.\n", "date": "2024-06-13", "categories": []} {"id": "2406.07742", "title": "C3DAG: Controlled 3D Animal Generation using 3D pose guidance", "abstract": " Recent advancements in text-to-3D generation have demonstrated the ability to\ngenerate high quality 3D assets. However while generating animals these methods\nunderperform, often portraying inaccurate anatomy and geometry. Towards\nameliorating this defect, we present C3DAG, a novel pose-Controlled text-to-3D\nAnimal Generation framework which generates a high quality 3D animal consistent\nwith a given pose. We also introduce an automatic 3D shape creator tool, that\nallows dynamic pose generation and modification via a web-based tool, and that\ngenerates a 3D balloon animal using simple geometries. A NeRF is then\ninitialized using this 3D shape using depth-controlled SDS. In the next stage,\nthe pre-trained NeRF is fine-tuned using quadruped-pose-controlled SDS. The\npipeline that we have developed not only produces geometrically and\nanatomically consistent results, but also renders highly controlled 3D animals,\nunlike prior methods which do not allow fine-grained pose control.\n", "date": "2024-06-11", "categories": []} {"id": "2406.08292", "title": "Outdoor Scene Extrapolation with Hierarchical Generative Cellular\n Automata", "abstract": " We aim to generate fine-grained 3D geometry from large-scale sparse LiDAR\nscans, abundantly captured by autonomous vehicles (AV). Contrary to prior work\non AV scene completion, we aim to extrapolate fine geometry from unlabeled and\nbeyond spatial limits of LiDAR scans, taking a step towards generating\nrealistic, high-resolution simulation-ready 3D street environments. We propose\nhierarchical Generative Cellular Automata (hGCA), a spatially scalable\nconditional 3D generative model, which grows geometry recursively with local\nkernels following, in a coarse-to-fine manner, equipped with a light-weight\nplanner to induce global consistency. Experiments on synthetic scenes show that\nhGCA generates plausible scene geometry with higher fidelity and completeness\ncompared to state-of-the-art baselines. Our model generalizes strongly from\nsim-to-real, qualitatively outperforming baselines on the Waymo-open dataset.\nWe also show anecdotal evidence of the ability to create novel objects from\nreal-world geometric cues even when trained on limited synthetic content. More\nresults and details can be found on\nhttps://research.nvidia.com/labs/toronto-ai/hGCA/.\n", "date": "2024-06-12", "categories": []} {"id": "2312.00093", "title": "GraphDreamer: Compositional 3D Scene Synthesis from Scene Graphs", "abstract": " As pretrained text-to-image diffusion models become increasingly powerful,\nrecent efforts have been made to distill knowledge from these text-to-image\npretrained models for optimizing a text-guided 3D model. Most of the existing\nmethods generate a holistic 3D model from a plain text input. This can be\nproblematic when the text describes a complex scene with multiple objects,\nbecause the vectorized text embeddings are inherently unable to capture a\ncomplex description with multiple entities and relationships. Holistic 3D\nmodeling of the entire scene further prevents accurate grounding of text\nentities and concepts. To address this limitation, we propose GraphDreamer, a\nnovel framework to generate compositional 3D scenes from scene graphs, where\nobjects are represented as nodes and their interactions as edges. By exploiting\nnode and edge information in scene graphs, our method makes better use of the\npretrained text-to-image diffusion model and is able to fully disentangle\ndifferent objects without image-level supervision. To facilitate modeling of\nobject-wise relationships, we use signed distance fields as representation and\nimpose a constraint to avoid inter-penetration of objects. To avoid manual\nscene graph creation, we design a text prompt for ChatGPT to generate scene\ngraphs based on text inputs. We conduct both qualitative and quantitative\nexperiments to validate the effectiveness of GraphDreamer in generating\nhigh-fidelity compositional 3D scenes with disentangled object entities.\n", "date": "2023-11-30", "categories": []} {"id": "2402.07207", "title": "GALA3D: Towards Text-to-3D Complex Scene Generation via Layout-guided\n Generative Gaussian Splatting", "abstract": " We present GALA3D, generative 3D GAussians with LAyout-guided control, for\neffective compositional text-to-3D generation. We first utilize large language\nmodels (LLMs) to generate the initial layout and introduce a layout-guided 3D\nGaussian representation for 3D content generation with adaptive geometric\nconstraints. We then propose an instance-scene compositional optimization\nmechanism with conditioned diffusion to collaboratively generate realistic 3D\nscenes with consistent geometry, texture, scale, and accurate interactions\namong multiple objects while simultaneously adjusting the coarse layout priors\nextracted from the LLMs to align with the generated scene. Experiments show\nthat GALA3D is a user-friendly, end-to-end framework for state-of-the-art\nscene-level 3D content generation and controllable editing while ensuring the\nhigh fidelity of object-level entities within the scene. The source codes and\nmodels will be available at gala3d.github.io.\n", "date": "2024-02-11", "categories": []} {"id": "2312.17142", "title": "DreamGaussian4D: Generative 4D Gaussian Splatting", "abstract": " 4D content generation has achieved remarkable progress recently. However,\nexisting methods suffer from long optimization times, a lack of motion\ncontrollability, and a low quality of details. In this paper, we introduce\nDreamGaussian4D (DG4D), an efficient 4D generation framework that builds on\nGaussian Splatting (GS). Our key insight is that combining explicit modeling of\nspatial transformations with static GS makes an efficient and powerful\nrepresentation for 4D generation. Moreover, video generation methods have the\npotential to offer valuable spatial-temporal priors, enhancing the high-quality\n4D generation. Specifically, we propose an integral framework with two major\nmodules: 1) Image-to-4D GS - we initially generate static GS with\nDreamGaussianHD, followed by HexPlane-based dynamic generation with Gaussian\ndeformation; and 2) Video-to-Video Texture Refinement - we refine the generated\nUV-space texture maps and meanwhile enhance their temporal consistency by\nutilizing a pre-trained image-to-video diffusion model. Notably, DG4D reduces\nthe optimization time from several hours to just a few minutes, allows the\ngenerated 3D motion to be visually controlled, and produces animated meshes\nthat can be realistically rendered in 3D engines.\n", "date": "2023-12-28", "categories": []} {"id": "2406.05543", "title": "VP-LLM: Text-Driven 3D Volume Completion with Large Language Models\n through Patchification", "abstract": " Recent conditional 3D completion works have mainly relied on CLIP or BERT to\nencode textual information, which cannot support complex instruction.\nMeanwhile, large language models (LLMs) have shown great potential in\nmulti-modal understanding and generation tasks. Inspired by the recent\nadvancements of LLM, we present Volume Patch LLM (VP-LLM), which leverages LLMs\nto perform conditional 3D completion in a single-forward pass. To integrate a\n3D model into the LLM tokenization configuration, the incomplete 3D object is\nfirst divided into small patches that can be encoded independently. These\nencoded patches are then fed into an LLM along with the text prompt,\ninstructing the LLM to capture the relations between these patches as well as\ninjecting semantic meanings into the 3D object. Our results demonstrate a\nstrong ability of LLMs to interpret complex text instructions and understand 3D\nobjects, surpassing state-of-the-art diffusion-based 3D completion models in\ngeneration quality.\n", "date": "2024-06-08", "categories": []} {"id": "2312.04043", "title": "Doodle Your 3D: From Abstract Freehand Sketches to Precise 3D Shapes", "abstract": " In this paper, we democratise 3D content creation, enabling precise\ngeneration of 3D shapes from abstract sketches while overcoming limitations\ntied to drawing skills. We introduce a novel part-level modelling and alignment\nframework that facilitates abstraction modelling and cross-modal\ncorrespondence. Leveraging the same part-level decoder, our approach seamlessly\nextends to sketch modelling by establishing correspondence between CLIPasso\nedgemaps and projected 3D part regions, eliminating the need for a dataset\npairing human sketches and 3D shapes. Additionally, our method introduces a\nseamless in-position editing process as a byproduct of cross-modal part-aligned\nmodelling. Operating in a low-dimensional implicit space, our approach\nsignificantly reduces computational demands and processing time.\n", "date": "2023-12-07", "categories": []} {"id": "2406.04322", "title": "DIRECT-3D: Learning Direct Text-to-3D Generation on Massive Noisy 3D\n Data", "abstract": " We present DIRECT-3D, a diffusion-based 3D generative model for creating\nhigh-quality 3D assets (represented by Neural Radiance Fields) from text\nprompts. Unlike recent 3D generative models that rely on clean and well-aligned\n3D data, limiting them to single or few-class generation, our model is directly\ntrained on extensive noisy and unaligned `in-the-wild' 3D assets, mitigating\nthe key challenge (i.e., data scarcity) in large-scale 3D generation. In\nparticular, DIRECT-3D is a tri-plane diffusion model that integrates two\ninnovations: 1) A novel learning framework where noisy data are filtered and\naligned automatically during the training process. Specifically, after an\ninitial warm-up phase using a small set of clean data, an iterative\noptimization is introduced in the diffusion process to explicitly estimate the\n3D pose of objects and select beneficial data based on conditional density. 2)\nAn efficient 3D representation that is achieved by disentangling object\ngeometry and color features with two separate conditional diffusion models that\nare optimized hierarchically. Given a prompt input, our model generates\nhigh-quality, high-resolution, realistic, and complex 3D objects with accurate\ngeometric details in seconds. We achieve state-of-the-art performance in both\nsingle-class generation and text-to-3D generation. We also demonstrate that\nDIRECT-3D can serve as a useful 3D geometric prior of objects, for example to\nalleviate the well-known Janus problem in 2D-lifting methods such as\nDreamFusion. The code and models are available for research purposes at:\nhttps://github.com/qihao067/direct3d.\n", "date": "2024-06-06", "categories": []} {"id": "2406.04629", "title": "STAR: Skeleton-aware Text-based 4D Avatar Generation with In-Network\n Motion Retargeting", "abstract": " The creation of 4D avatars (i.e., animated 3D avatars) from text description\ntypically uses text-to-image (T2I) diffusion models to synthesize 3D avatars in\nthe canonical space and subsequently applies animation with target motions.\nHowever, such an optimization-by-animation paradigm has several drawbacks. (1)\nFor pose-agnostic optimization, the rendered images in canonical pose for naive\nScore Distillation Sampling (SDS) exhibit domain gap and cannot preserve\nview-consistency using only T2I priors, and (2) For post hoc animation, simply\napplying the source motions to target 3D avatars yields translation artifacts\nand misalignment. To address these issues, we propose Skeleton-aware Text-based\n4D Avatar generation with in-network motion Retargeting (STAR). STAR considers\nthe geometry and skeleton differences between the template mesh and target\navatar, and corrects the mismatched source motion by resorting to the\npretrained motion retargeting techniques. With the informatively retargeted and\nocclusion-aware skeleton, we embrace the skeleton-conditioned T2I and\ntext-to-video (T2V) priors, and propose a hybrid SDS module to coherently\nprovide multi-view and frame-consistent supervision signals. Hence, STAR can\nprogressively optimize the geometry, texture, and motion in an end-to-end\nmanner. The quantitative and qualitative experiments demonstrate our proposed\nSTAR can synthesize high-quality 4D avatars with vivid animations that align\nwell with the text description. Additional ablation studies shows the\ncontributions of each component in STAR. The source code and demos are\navailable at:\n\\href{https://star-avatar.github.io}{https://star-avatar.github.io}.\n", "date": "2024-06-07", "categories": []} {"id": "2406.04983", "title": "CityCraft: A Real Crafter for 3D City Generation", "abstract": " City scene generation has gained significant attention in autonomous driving,\nsmart city development, and traffic simulation. It helps enhance infrastructure\nplanning and monitoring solutions. Existing methods have employed a two-stage\nprocess involving city layout generation, typically using Variational\nAutoencoders (VAEs), Generative Adversarial Networks (GANs), or Transformers,\nfollowed by neural rendering. These techniques often exhibit limited diversity\nand noticeable artifacts in the rendered city scenes. The rendered scenes lack\nvariety, resembling the training images, resulting in monotonous styles.\nAdditionally, these methods lack planning capabilities, leading to less\nrealistic generated scenes. In this paper, we introduce CityCraft, an\ninnovative framework designed to enhance both the diversity and quality of\nurban scene generation. Our approach integrates three key stages: initially, a\ndiffusion transformer (DiT) model is deployed to generate diverse and\ncontrollable 2D city layouts. Subsequently, a Large Language Model(LLM) is\nutilized to strategically make land-use plans within these layouts based on\nuser prompts and language guidelines. Based on the generated layout and city\nplan, we utilize the asset retrieval module and Blender for precise asset\nplacement and scene construction. Furthermore, we contribute two new datasets\nto the field: 1)CityCraft-OSM dataset including 2D semantic layouts of urban\nareas, corresponding satellite images, and detailed annotations. 2)\nCityCraft-Buildings dataset, featuring thousands of diverse, high-quality 3D\nbuilding assets. CityCraft achieves state-of-the-art performance in generating\nrealistic 3D cities.\n", "date": "2024-06-07", "categories": []} {"id": "2406.05038", "title": "Efficient 3D Shape Generation via Diffusion Mamba with Bidirectional\n SSMs", "abstract": " Recent advancements in sequence modeling have led to the development of the\nMamba architecture, noted for its selective state space approach, offering a\npromising avenue for efficient long sequence handling. However, its application\nin 3D shape generation, particularly at high resolutions, remains\nunderexplored. Traditional diffusion transformers (DiT) with self-attention\nmechanisms, despite their potential, face scalability challenges due to the\ncubic complexity of attention operations as input length increases. This\ncomplexity becomes a significant hurdle when dealing with high-resolution voxel\nsizes. To address this challenge, we introduce a novel diffusion architecture\ntailored for 3D point clouds generation-Diffusion Mamba (DiM-3D). This\narchitecture forgoes traditional attention mechanisms, instead utilizing the\ninherent efficiency of the Mamba architecture to maintain linear complexity\nwith respect to sequence length. DiM-3D is characterized by fast inference\ntimes and substantially lower computational demands, quantified in reduced\nGflops, thereby addressing the key scalability issues of prior models. Our\nempirical results on the ShapeNet benchmark demonstrate that DiM-3D achieves\nstate-of-the-art performance in generating high-fidelity and diverse 3D shapes.\nAdditionally, DiM-3D shows superior capabilities in tasks like 3D point cloud\ncompletion. This not only proves the model's scalability but also underscores\nits efficiency in generating detailed, high-resolution voxels necessary for\nadvanced 3D shape modeling, particularly excelling in environments requiring\nhigh-resolution voxel sizes. Through these findings, we illustrate the\nexceptional scalability and efficiency of the Diffusion Mamba framework in 3D\nshape generation, setting a new standard for the field and paving the way for\nfuture explorations in high-resolution 3D modeling technologies.\n", "date": "2024-06-07", "categories": []} {"id": "2309.00610", "title": "CityDreamer: Compositional Generative Model of Unbounded 3D Cities", "abstract": " 3D city generation is a desirable yet challenging task, since humans are more\nsensitive to structural distortions in urban environments. Additionally,\ngenerating 3D cities is more complex than 3D natural scenes since buildings, as\nobjects of the same class, exhibit a wider range of appearances compared to the\nrelatively consistent appearance of objects like trees in natural scenes. To\naddress these challenges, we propose \\textbf{CityDreamer}, a compositional\ngenerative model designed specifically for unbounded 3D cities. Our key insight\nis that 3D city generation should be a composition of different types of neural\nfields: 1) various building instances, and 2) background stuff, such as roads\nand green lands. Specifically, we adopt the bird's eye view scene\nrepresentation and employ a volumetric render for both instance-oriented and\nstuff-oriented neural fields. The generative hash grid and periodic positional\nembedding are tailored as scene parameterization to suit the distinct\ncharacteristics of building instances and background stuff. Furthermore, we\ncontribute a suite of CityGen Datasets, including OSM and GoogleEarth, which\ncomprises a vast amount of real-world city imagery to enhance the realism of\nthe generated 3D cities both in their layouts and appearances. CityDreamer\nachieves state-of-the-art performance not only in generating realistic 3D\ncities but also in localized editing within the generated cities.\n", "date": "2023-09-01", "categories": []} {"id": "2406.04253", "title": "A Survey on 3D Human Avatar Modeling -- From Reconstruction to\n Generation", "abstract": " 3D modeling has long been an important area in computer vision and computer\ngraphics. Recently, thanks to the breakthroughs in neural representations and\ngenerative models, we witnessed a rapid development of 3D modeling. 3D human\nmodeling, lying at the core of many real-world applications, such as gaming and\nanimation, has attracted significant attention. Over the past few years, a\nlarge body of work on creating 3D human avatars has been introduced, forming a\nnew and abundant knowledge base for 3D human modeling. The scale of the\nliterature makes it difficult for individuals to keep track of all the works.\nThis survey aims to provide a comprehensive overview of these emerging\ntechniques for 3D human avatar modeling, from both reconstruction and\ngeneration perspectives. Firstly, we review representative methods for 3D human\nreconstruction, including methods based on pixel-aligned implicit function,\nneural radiance field, and 3D Gaussian Splatting, etc. We then summarize\nrepresentative methods for 3D human generation, especially those using large\nlanguage models like CLIP, diffusion models, and various 3D representations,\nwhich demonstrate state-of-the-art performance. Finally, we discuss our\nreflection on existing methods and open challenges for 3D human avatar\nmodeling, shedding light on future research.\n", "date": "2024-06-06", "categories": []} {"id": "2406.00687", "title": "Lay-A-Scene: Personalized 3D Object Arrangement Using Text-to-Image\n Priors", "abstract": " Generating 3D visual scenes is at the forefront of visual generative AI, but\ncurrent 3D generation techniques struggle with generating scenes with multiple\nhigh-resolution objects. Here we introduce Lay-A-Scene, which solves the task\nof Open-set 3D Object Arrangement, effectively arranging unseen objects. Given\na set of 3D objects, the task is to find a plausible arrangement of these\nobjects in a scene. We address this task by leveraging pre-trained\ntext-to-image models. We personalize the model and explain how to generate\nimages of a scene that contains multiple predefined objects without neglecting\nany of them. Then, we describe how to infer the 3D poses and arrangement of\nobjects from a 2D generated image by finding a consistent projection of objects\nonto the 2D scene. We evaluate the quality of Lay-A-Scene using 3D objects from\nObjaverse and human raters and find that it often generates coherent and\nfeasible 3D object arrangements.\n", "date": "2024-06-02", "categories": []} {"id": "2404.14329", "title": "X-Ray: A Sequential 3D Representation For Generation", "abstract": " We introduce X-Ray, a novel 3D sequential representation inspired by the\npenetrability of x-ray scans. X-Ray transforms a 3D object into a series of\nsurface frames at different layers, making it suitable for generating 3D models\nfrom images. Our method utilizes ray casting from the camera center to capture\ngeometric and textured details, including depth, normal, and color, across all\nintersected surfaces. This process efficiently condenses the whole 3D object\ninto a multi-frame video format, motivating the utilize of a network\narchitecture similar to those in video diffusion models. This design ensures an\nefficient 3D representation by focusing solely on surface information. Also, we\npropose a two-stage pipeline to generate 3D objects from X-Ray Diffusion Model\nand Upsampler. We demonstrate the practicality and adaptability of our X-Ray\nrepresentation by synthesizing the complete visible and hidden surfaces of a 3D\nobject from a single input image. Experimental results reveal the\nstate-of-the-art superiority of our representation in enhancing the accuracy of\n3D generation, paving the way for new 3D representation research and practical\napplications.\n", "date": "2024-04-22", "categories": []} {"id": "2405.14832", "title": "Direct3D: Scalable Image-to-3D Generation via 3D Latent Diffusion\n Transformer", "abstract": " Generating high-quality 3D assets from text and images has long been\nchallenging, primarily due to the absence of scalable 3D representations\ncapable of capturing intricate geometry distributions. In this work, we\nintroduce Direct3D, a native 3D generative model scalable to in-the-wild input\nimages, without requiring a multiview diffusion model or SDS optimization. Our\napproach comprises two primary components: a Direct 3D Variational Auto-Encoder\n(D3D-VAE) and a Direct 3D Diffusion Transformer (D3D-DiT). D3D-VAE efficiently\nencodes high-resolution 3D shapes into a compact and continuous latent triplane\nspace. Notably, our method directly supervises the decoded geometry using a\nsemi-continuous surface sampling strategy, diverging from previous methods\nrelying on rendered images as supervision signals. D3D-DiT models the\ndistribution of encoded 3D latents and is specifically designed to fuse\npositional information from the three feature maps of the triplane latent,\nenabling a native 3D generative model scalable to large-scale 3D datasets.\nAdditionally, we introduce an innovative image-to-3D generation pipeline\nincorporating semantic and pixel-level image conditions, allowing the model to\nproduce 3D shapes consistent with the provided conditional image input.\nExtensive experiments demonstrate the superiority of our large-scale\npre-trained Direct3D over previous image-to-3D approaches, achieving\nsignificantly better generation quality and generalization ability, thus\nestablishing a new state-of-the-art for 3D content creation. Project page:\nhttps://nju-3dv.github.io/projects/Direct3D/.\n", "date": "2024-05-23", "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": "2212.04981", "title": "LoopDraw: a Loop-Based Autoregressive Model for Shape Synthesis and\n Editing", "abstract": " There is no settled universal 3D representation for geometry with many\nalternatives such as point clouds, meshes, implicit functions, and voxels to\nname a few. In this work, we present a new, compelling alternative for\nrepresenting shapes using a sequence of cross-sectional closed loops. The loops\nacross all planes form an organizational hierarchy which we leverage for\nautoregressive shape synthesis and editing. Loops are a non-local description\nof the underlying shape, as simple loop manipulations (such as shifts) result\nin significant structural changes to the geometry. This is in contrast to\nmanipulating local primitives such as points in a point cloud or a triangle in\na triangle mesh. We further demonstrate that loops are intuitive and natural\nprimitive for analyzing and editing shapes, both computationally and for users.\n", "date": "2022-12-09", "categories": []} {"id": "2405.19203", "title": "$E^{3}$Gen: Efficient, Expressive and Editable Avatars Generation", "abstract": " This paper aims to introduce 3D Gaussian for efficient, expressive, and\neditable digital avatar generation. This task faces two major challenges: (1)\nThe unstructured nature of 3D Gaussian makes it incompatible with current\ngeneration pipelines; (2) the expressive animation of 3D Gaussian in a\ngenerative setting that involves training with multiple subjects remains\nunexplored. In this paper, we propose a novel avatar generation method named\n$E^3$Gen, to effectively address these challenges. First, we propose a novel\ngenerative UV features plane representation that encodes unstructured 3D\nGaussian onto a structured 2D UV space defined by the SMPL-X parametric model.\nThis novel representation not only preserves the representation ability of the\noriginal 3D Gaussian but also introduces a shared structure among subjects to\nenable generative learning of the diffusion model. To tackle the second\nchallenge, we propose a part-aware deformation module to achieve robust and\naccurate full-body expressive pose control. Extensive experiments demonstrate\nthat our method achieves superior performance in avatar generation and enables\nexpressive full-body pose control and editing. Our project page is\nhttps://olivia23333.github.io/E3Gen.\n", "date": "2024-05-29", "categories": []} {"id": "2405.20334", "title": "VividDream: Generating 3D Scene with Ambient Dynamics", "abstract": " We introduce VividDream, a method for generating explorable 4D scenes with\nambient dynamics from a single input image or text prompt. VividDream first\nexpands an input image into a static 3D point cloud through iterative\ninpainting and geometry merging. An ensemble of animated videos is then\ngenerated using video diffusion models with quality refinement techniques and\nconditioned on renderings of the static 3D scene from the sampled camera\ntrajectories. We then optimize a canonical 4D scene representation using an\nanimated video ensemble, with per-video motion embeddings and visibility masks\nto mitigate inconsistencies. The resulting 4D scene enables free-view\nexploration of a 3D scene with plausible ambient scene dynamics. Experiments\ndemonstrate that VividDream can provide human viewers with compelling 4D\nexperiences generated based on diverse real images and text prompts.\n", "date": "2024-05-30", "categories": []} {"id": "2310.12945", "title": "3D-GPT: Procedural 3D Modeling with Large Language Models", "abstract": " In the pursuit of efficient automated content creation, procedural\ngeneration, leveraging modifiable parameters and rule-based systems, emerges as\na promising approach. Nonetheless, it could be a demanding endeavor, given its\nintricate nature necessitating a deep understanding of rules, algorithms, and\nparameters. To reduce workload, we introduce 3D-GPT, a framework utilizing\nlarge language models~(LLMs) for instruction-driven 3D modeling. 3D-GPT\npositions LLMs as proficient problem solvers, dissecting the procedural 3D\nmodeling tasks into accessible segments and appointing the apt agent for each\ntask. 3D-GPT integrates three core agents: the task dispatch agent, the\nconceptualization agent, and the modeling agent. They collaboratively achieve\ntwo objectives. First, it enhances concise initial scene descriptions, evolving\nthem into detailed forms while dynamically adapting the text based on\nsubsequent instructions. Second, it integrates procedural generation,\nextracting parameter values from enriched text to effortlessly interface with\n3D software for asset creation. Our empirical investigations confirm that\n3D-GPT not only interprets and executes instructions, delivering reliable\nresults but also collaborates effectively with human designers. Furthermore, it\nseamlessly integrates with Blender, unlocking expanded manipulation\npossibilities. Our work highlights the potential of LLMs in 3D modeling,\noffering a basic framework for future advancements in scene generation and\nanimation.\n", "date": "2023-10-19", "categories": []} {"id": "2312.09250", "title": "Single Mesh Diffusion Models with Field Latents for Texture Generation", "abstract": " We introduce a framework for intrinsic latent diffusion models operating\ndirectly on the surfaces of 3D shapes, with the goal of synthesizing\nhigh-quality textures. Our approach is underpinned by two contributions: field\nlatents, a latent representation encoding textures as discrete vector fields on\nthe mesh vertices, and field latent diffusion models, which learn to denoise a\ndiffusion process in the learned latent space on the surface. We consider a\nsingle-textured-mesh paradigm, where our models are trained to generate\nvariations of a given texture on a mesh. We show the synthesized textures are\nof superior fidelity compared those from existing single-textured-mesh\ngenerative models. Our models can also be adapted for user-controlled editing\ntasks such as inpainting and label-guided generation. The efficacy of our\napproach is due in part to the equivariance of our proposed framework under\nisometries, allowing our models to seamlessly reproduce details across locally\nsimilar regions and opening the door to a notion of generative texture\ntransfer.\n", "date": "2023-12-14", "categories": []} {"id": "2405.16570", "title": "ID-to-3D: Expressive ID-guided 3D Heads via Score Distillation Sampling", "abstract": " We propose ID-to-3D, a method to generate identity- and text-guided 3D human\nheads with disentangled expressions, starting from even a single casually\ncaptured in-the-wild image of a subject. The foundation of our approach is\nanchored in compositionality, alongside the use of task-specific 2D diffusion\nmodels as priors for optimization. First, we extend a foundational model with a\nlightweight expression-aware and ID-aware architecture, and create 2D priors\nfor geometry and texture generation, via fine-tuning only 0.2% of its available\ntraining parameters. Then, we jointly leverage a neural parametric\nrepresentation for the expressions of each subject and a multi-stage generation\nof highly detailed geometry and albedo texture. This combination of strong face\nidentity embeddings and our neural representation enables accurate\nreconstruction of not only facial features but also accessories and hair and\ncan be meshed to provide render-ready assets for gaming and telepresence. Our\nresults achieve an unprecedented level of identity-consistent and high-quality\ntexture and geometry generation, generalizing to a ``world'' of unseen 3D\nidentities, without relying on large 3D captured datasets of human assets.\n", "date": "2024-05-26", "categories": []} {"id": "2405.18132", "title": "EG4D: Explicit Generation of 4D Object without Score Distillation", "abstract": " In recent years, the increasing demand for dynamic 3D assets in design and\ngaming applications has given rise to powerful generative pipelines capable of\nsynthesizing high-quality 4D objects. Previous methods generally rely on score\ndistillation sampling (SDS) algorithm to infer the unseen views and motion of\n4D objects, thus leading to unsatisfactory results with defects like\nover-saturation and Janus problem. Therefore, inspired by recent progress of\nvideo diffusion models, we propose to optimize a 4D representation by\nexplicitly generating multi-view videos from one input image. However, it is\nfar from trivial to handle practical challenges faced by such a pipeline,\nincluding dramatic temporal inconsistency, inter-frame geometry and texture\ndiversity, and semantic defects brought by video generation results. To address\nthese issues, we propose DG4D, a novel multi-stage framework that generates\nhigh-quality and consistent 4D assets without score distillation. Specifically,\ncollaborative techniques and solutions are developed, including an attention\ninjection strategy to synthesize temporal-consistent multi-view videos, a\nrobust and efficient dynamic reconstruction method based on Gaussian Splatting,\nand a refinement stage with diffusion prior for semantic restoration. The\nqualitative results and user preference study demonstrate that our framework\noutperforms the baselines in generation quality by a considerable margin. Code\nwill be released at \\url{https://github.com/jasongzy/EG4D}.\n", "date": "2024-05-28", "categories": []} {"id": "2311.17984", "title": "4D-fy: Text-to-4D Generation Using Hybrid Score Distillation Sampling", "abstract": " Recent breakthroughs in text-to-4D generation rely on pre-trained\ntext-to-image and text-to-video models to generate dynamic 3D scenes. However,\ncurrent text-to-4D methods face a three-way tradeoff between the quality of\nscene appearance, 3D structure, and motion. For example, text-to-image models\nand their 3D-aware variants are trained on internet-scale image datasets and\ncan be used to produce scenes with realistic appearance and 3D structure -- but\nno motion. Text-to-video models are trained on relatively smaller video\ndatasets and can produce scenes with motion, but poorer appearance and 3D\nstructure. While these models have complementary strengths, they also have\nopposing weaknesses, making it difficult to combine them in a way that\nalleviates this three-way tradeoff. Here, we introduce hybrid score\ndistillation sampling, an alternating optimization procedure that blends\nsupervision signals from multiple pre-trained diffusion models and incorporates\nbenefits of each for high-fidelity text-to-4D generation. Using hybrid SDS, we\ndemonstrate synthesis of 4D scenes with compelling appearance, 3D structure,\nand motion.\n", "date": "2023-11-29", "categories": []} {"id": "2404.00987", "title": "FlexiDreamer: Single Image-to-3D Generation with FlexiCubes", "abstract": " 3D content generation has wide applications in various fields. One of its\ndominant paradigms is by sparse-view reconstruction using multi-view images\ngenerated by diffusion models. However, since directly reconstructing triangle\nmeshes from multi-view images is challenging, most methodologies opt to an\nimplicit representation (such as NeRF) during the sparse-view reconstruction\nand acquire the target mesh by a post-processing extraction. However, the\nimplicit representation takes extensive time to train and the post-extraction\nalso leads to undesirable visual artifacts. In this paper, we propose\nFlexiDreamer, a novel framework that directly reconstructs high-quality meshes\nfrom multi-view generated images. We utilize an advanced gradient-based mesh\noptimization, namely FlexiCubes, for multi-view mesh reconstruction, which\nenables us to generate 3D meshes in an end-to-end manner. To address the\nreconstruction artifacts owing to the inconsistencies from generated images, we\ndesign a hybrid positional encoding scheme to improve the reconstruction\ngeometry and an orientation-aware texture mapping to mitigate surface ghosting.\nTo further enhance the results, we respectively incorporate eikonal and smooth\nregularizations to reduce geometric holes and surface noise. Our approach can\ngenerate high-fidelity 3D meshes in the single image-to-3D downstream task with\napproximately 1 minute, significantly outperforming previous methods.\n", "date": "2024-04-01", "categories": []} {"id": "2405.15914", "title": "ExactDreamer: High-Fidelity Text-to-3D Content Creation via Exact Score\n Matching", "abstract": " Text-to-3D content creation is a rapidly evolving research area. Given the\nscarcity of 3D data, current approaches often adapt pre-trained 2D diffusion\nmodels for 3D synthesis. Among these approaches, Score Distillation Sampling\n(SDS) has been widely adopted. However, the issue of over-smoothing poses a\nsignificant limitation on the high-fidelity generation of 3D models. To address\nthis challenge, LucidDreamer replaces the Denoising Diffusion Probabilistic\nModel (DDPM) in SDS with the Denoising Diffusion Implicit Model (DDIM) to\nconstruct Interval Score Matching (ISM). However, ISM inevitably inherits\ninconsistencies from DDIM, causing reconstruction errors during the DDIM\ninversion process. This results in poor performance in the detailed generation\nof 3D objects and loss of content. To alleviate these problems, we propose a\nnovel method named Exact Score Matching (ESM). Specifically, ESM leverages\nauxiliary variables to mathematically guarantee exact recovery in the DDIM\nreverse process. Furthermore, to effectively capture the dynamic changes of the\noriginal and auxiliary variables, the LoRA of a pre-trained diffusion model\nimplements these exact paths. Extensive experiments demonstrate the\neffectiveness of ESM in text-to-3D generation, particularly highlighting its\nsuperiority in detailed generation.\n", "date": "2024-05-24", "categories": []} {"id": "2405.16645", "title": "Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video\n Diffusion Models", "abstract": " The availability of large-scale multimodal datasets and advancements in\ndiffusion models have significantly accelerated progress in 4D content\ngeneration. Most prior approaches rely on multiple image or video diffusion\nmodels, utilizing score distillation sampling for optimization or generating\npseudo novel views for direct supervision. However, these methods are hindered\nby slow optimization speeds and multi-view inconsistency issues. Spatial and\ntemporal consistency in 4D geometry has been extensively explored respectively\nin 3D-aware diffusion models and traditional monocular video diffusion models.\nBuilding on this foundation, we propose a strategy to migrate the temporal\nconsistency in video diffusion models to the spatial-temporal consistency\nrequired for 4D generation. Specifically, we present a novel framework,\n\\textbf{Diffusion4D}, for efficient and scalable 4D content generation.\nLeveraging a meticulously curated dynamic 3D dataset, we develop a 4D-aware\nvideo diffusion model capable of synthesizing orbital views of dynamic 3D\nassets. To control the dynamic strength of these assets, we introduce a\n3D-to-4D motion magnitude metric as guidance. Additionally, we propose a novel\nmotion magnitude reconstruction loss and 3D-aware classifier-free guidance to\nrefine the learning and generation of motion dynamics. After obtaining orbital\nviews of the 4D asset, we perform explicit 4D construction with Gaussian\nsplatting in a coarse-to-fine manner. The synthesized multi-view consistent 4D\nimage set enables us to swiftly generate high-fidelity and diverse 4D assets\nwithin just several minutes. Extensive experiments demonstrate that our method\nsurpasses prior state-of-the-art techniques in terms of generation efficiency\nand 4D geometry consistency across various prompt modalities.\n", "date": "2024-05-26", "categories": []} {"id": "2405.16890", "title": "PivotMesh: Generic 3D Mesh Generation via Pivot Vertices Guidance", "abstract": " Generating compact and sharply detailed 3D meshes poses a significant\nchallenge for current 3D generative models. Different from extracting dense\nmeshes from neural representation, some recent works try to model the native\nmesh distribution (i.e., a set of triangles), which generates more compact\nresults as humans crafted. However, due to the complexity and variety of mesh\ntopology, these methods are typically limited to small datasets with specific\ncategories and are hard to extend. In this paper, we introduce a generic and\nscalable mesh generation framework PivotMesh, which makes an initial attempt to\nextend the native mesh generation to large-scale datasets. We employ a\ntransformer-based auto-encoder to encode meshes into discrete tokens and decode\nthem from face level to vertex level hierarchically. Subsequently, to model the\ncomplex typology, we first learn to generate pivot vertices as coarse mesh\nrepresentation and then generate the complete mesh tokens with the same\nauto-regressive Transformer. This reduces the difficulty compared with directly\nmodeling the mesh distribution and further improves the model controllability.\nPivotMesh demonstrates its versatility by effectively learning from both small\ndatasets like Shapenet, and large-scale datasets like Objaverse and\nObjaverse-xl. Extensive experiments indicate that PivotMesh can generate\ncompact and sharp 3D meshes across various categories, highlighting its great\npotential for native mesh modeling.\n", "date": "2024-05-27", "categories": []} {"id": "2405.17176", "title": "DreamMat: High-quality PBR Material Generation with Geometry- and\n Light-aware Diffusion Models", "abstract": " 2D diffusion model, which often contains unwanted baked-in shading effects\nand results in unrealistic rendering effects in the downstream applications.\nGenerating Physically Based Rendering (PBR) materials instead of just RGB\ntextures would be a promising solution. However, directly distilling the PBR\nmaterial parameters from 2D diffusion models still suffers from incorrect\nmaterial decomposition, such as baked-in shading effects in albedo. We\nintroduce DreamMat, an innovative approach to resolve the aforementioned\nproblem, to generate high-quality PBR materials from text descriptions. We find\nout that the main reason for the incorrect material distillation is that\nlarge-scale 2D diffusion models are only trained to generate final shading\ncolors, resulting in insufficient constraints on material decomposition during\ndistillation. To tackle this problem, we first finetune a new light-aware 2D\ndiffusion model to condition on a given lighting environment and generate the\nshading results on this specific lighting condition. Then, by applying the same\nenvironment lights in the material distillation, DreamMat can generate\nhigh-quality PBR materials that are not only consistent with the given geometry\nbut also free from any baked-in shading effects in albedo. Extensive\nexperiments demonstrate that the materials produced through our methods exhibit\ngreater visual appeal to users and achieve significantly superior rendering\nquality compared to baseline methods, which are preferable for downstream tasks\nsuch as game and film production.\n", "date": "2024-05-27", "categories": []} {"id": "2401.17053", "title": "BlockFusion: Expandable 3D Scene Generation using Latent Tri-plane\n Extrapolation", "abstract": " We present BlockFusion, a diffusion-based model that generates 3D scenes as\nunit blocks and seamlessly incorporates new blocks to extend the scene.\nBlockFusion is trained using datasets of 3D blocks that are randomly cropped\nfrom complete 3D scene meshes. Through per-block fitting, all training blocks\nare converted into the hybrid neural fields: with a tri-plane containing the\ngeometry features, followed by a Multi-layer Perceptron (MLP) for decoding the\nsigned distance values. A variational auto-encoder is employed to compress the\ntri-planes into the latent tri-plane space, on which the denoising diffusion\nprocess is performed. Diffusion applied to the latent representations allows\nfor high-quality and diverse 3D scene generation. To expand a scene during\ngeneration, one needs only to append empty blocks to overlap with the current\nscene and extrapolate existing latent tri-planes to populate new blocks. The\nextrapolation is done by conditioning the generation process with the feature\nsamples from the overlapping tri-planes during the denoising iterations. Latent\ntri-plane extrapolation produces semantically and geometrically meaningful\ntransitions that harmoniously blend with the existing scene. A 2D layout\nconditioning mechanism is used to control the placement and arrangement of\nscene elements. Experimental results indicate that BlockFusion is capable of\ngenerating diverse, geometrically consistent and unbounded large 3D scenes with\nunprecedented high-quality shapes in both indoor and outdoor scenarios.\n", "date": "2024-01-30", "categories": []} {"id": "2212.04965", "title": "Seeing a Rose in Five Thousand Ways", "abstract": " What is a rose, visually? A rose comprises its intrinsics, including the\ndistribution of geometry, texture, and material specific to its object\ncategory. With knowledge of these intrinsic properties, we may render roses of\ndifferent sizes and shapes, in different poses, and under different lighting\nconditions. In this work, we build a generative model that learns to capture\nsuch object intrinsics from a single image, such as a photo of a bouquet. Such\nan image includes multiple instances of an object type. These instances all\nshare the same intrinsics, but appear different due to a combination of\nvariance within these intrinsics and differences in extrinsic factors, such as\npose and illumination. Experiments show that our model successfully learns\nobject intrinsics (distribution of geometry, texture, and material) for a wide\nrange of objects, each from a single Internet image. Our method achieves\nsuperior results on multiple downstream tasks, including intrinsic image\ndecomposition, shape and image generation, view synthesis, and relighting.\n", "date": "2022-12-09", "categories": []} {"id": "2403.05154", "title": "GSEdit: Efficient Text-Guided Editing of 3D Objects via Gaussian\n Splatting", "abstract": " We present GSEdit, a pipeline for text-guided 3D object editing based on\nGaussian Splatting models. Our method enables the editing of the style and\nappearance of 3D objects without altering their main details, all in a matter\nof minutes on consumer hardware. We tackle the problem by leveraging Gaussian\nsplatting to represent 3D scenes, and we optimize the model while progressively\nvarying the image supervision by means of a pretrained image-based diffusion\nmodel. The input object may be given as a 3D triangular mesh, or directly\nprovided as Gaussians from a generative model such as DreamGaussian. GSEdit\nensures consistency across different viewpoints, maintaining the integrity of\nthe original object's information. Compared to previously proposed methods\nrelying on NeRF-like MLP models, GSEdit stands out for its efficiency, making\n3D editing tasks much faster. Our editing process is refined via the\napplication of the SDS loss, ensuring that our edits are both precise and\naccurate. Our comprehensive evaluation demonstrates that GSEdit effectively\nalters object shape and appearance following the given textual instructions\nwhile preserving their coherence and detail.\n", "date": "2024-03-08", "categories": []} {"id": "2405.12420", "title": "GarmentDreamer: 3DGS Guided Garment Synthesis with Diverse Geometry and\n Texture Details", "abstract": " Traditional 3D garment creation is labor-intensive, involving sketching,\nmodeling, UV mapping, and texturing, which are time-consuming and costly.\nRecent advances in diffusion-based generative models have enabled new\npossibilities for 3D garment generation from text prompts, images, and videos.\nHowever, existing methods either suffer from inconsistencies among multi-view\nimages or require additional processes to separate cloth from the underlying\nhuman model. In this paper, we propose GarmentDreamer, a novel method that\nleverages 3D Gaussian Splatting (GS) as guidance to generate wearable,\nsimulation-ready 3D garment meshes from text prompts. In contrast to using\nmulti-view images directly predicted by generative models as guidance, our 3DGS\nguidance ensures consistent optimization in both garment deformation and\ntexture synthesis. Our method introduces a novel garment augmentation module,\nguided by normal and RGBA information, and employs implicit Neural Texture\nFields (NeTF) combined with Score Distillation Sampling (SDS) to generate\ndiverse geometric and texture details. We validate the effectiveness of our\napproach through comprehensive qualitative and quantitative experiments,\nshowcasing the superior performance of GarmentDreamer over state-of-the-art\nalternatives. Our project page is available at:\nhttps://xuan-li.github.io/GarmentDreamerDemo/.\n", "date": "2024-05-20", "categories": []} {"id": "2307.05468", "title": "My3DGen: A Scalable Personalized 3D Generative Model", "abstract": " In recent years, generative 3D face models (e.g., EG3D) have been developed\nto tackle the problem of synthesizing photo-realistic faces. However, these\nmodels are often unable to capture facial features unique to each individual,\nhighlighting the importance of personalization. Some prior works have shown\npromise in personalizing generative face models, but these studies primarily\nfocus on 2D settings. Also, these methods require both fine-tuning and storing\na large number of parameters for each user, posing a hindrance to achieving\nscalable personalization. Another challenge of personalization is the limited\nnumber of training images available for each individual, which often leads to\noverfitting when using full fine-tuning methods. Our proposed approach,\nMy3DGen, generates a personalized 3D prior of an individual using as few as 50\ntraining images. My3DGen allows for novel view synthesis, semantic editing of a\ngiven face (e.g. adding a smile), and synthesizing novel appearances, all while\npreserving the original person's identity. We decouple the 3D facial features\ninto global features and personalized features by freezing the pre-trained EG3D\nand training additional personalized weights through low-rank decomposition. As\na result, My3DGen introduces only $\\textbf{240K}$ personalized parameters per\nindividual, leading to a $\\textbf{127}\\times$ reduction in trainable parameters\ncompared to the $\\textbf{30.6M}$ required for fine-tuning the entire parameter\nspace. Despite this significant reduction in storage, our model preserves\nidentity features without compromising the quality of downstream applications.\n", "date": "2023-07-11", "categories": []} {"id": "2311.17082", "title": "DreamPropeller: Supercharge Text-to-3D Generation with Parallel Sampling", "abstract": " Recent methods such as Score Distillation Sampling (SDS) and Variational\nScore Distillation (VSD) using 2D diffusion models for text-to-3D generation\nhave demonstrated impressive generation quality. However, the long generation\ntime of such algorithms significantly degrades the user experience. To tackle\nthis problem, we propose DreamPropeller, a drop-in acceleration algorithm that\ncan be wrapped around any existing text-to-3D generation pipeline based on\nscore distillation. Our framework generalizes Picard iterations, a classical\nalgorithm for parallel sampling an ODE path, and can account for non-ODE paths\nsuch as momentum-based gradient updates and changes in dimensions during the\noptimization process as in many cases of 3D generation. We show that our\nalgorithm trades parallel compute for wallclock time and empirically achieves\nup to 4.7x speedup with a negligible drop in generation quality for all tested\nframeworks.\n", "date": "2023-11-28", "categories": []} {"id": "2404.01655", "title": "FashionEngine: Interactive 3D Human Generation and Editing via\n Multimodal Controls", "abstract": " We present FashionEngine, an interactive 3D human generation and editing\nsystem that creates 3D digital humans via user-friendly multimodal controls\nsuch as natural languages, visual perceptions, and hand-drawing sketches.\nFashionEngine automates the 3D human production with three key components: 1) A\npre-trained 3D human diffusion model that learns to model 3D humans in a\nsemantic UV latent space from 2D image training data, which provides strong\npriors for diverse generation and editing tasks. 2) Multimodality-UV Space\nencoding the texture appearance, shape topology, and textual semantics of human\nclothing in a canonical UV-aligned space, which faithfully aligns the user\nmultimodal inputs with the implicit UV latent space for controllable 3D human\nediting. The multimodality-UV space is shared across different user inputs,\nsuch as texts, images, and sketches, which enables various joint multimodal\nediting tasks. 3) Multimodality-UV Aligned Sampler learns to sample\nhigh-quality and diverse 3D humans from the diffusion prior. Extensive\nexperiments validate FashionEngine's state-of-the-art performance for\nconditional generation/editing tasks. In addition, we present an interactive\nuser interface for our FashionEngine that enables both conditional and\nunconditional generation tasks, and editing tasks including pose/view/shape\ncontrol, text-, image-, and sketch-driven 3D human editing and 3D virtual\ntry-on, in a unified framework. Our project page is at:\nhttps://taohuumd.github.io/projects/FashionEngine.\n", "date": "2024-04-02", "categories": []} {"id": "2405.11252", "title": "Dreamer XL: Towards High-Resolution Text-to-3D Generation via Trajectory\n Score Matching", "abstract": " In this work, we propose a novel Trajectory Score Matching (TSM) method that\naims to solve the pseudo ground truth inconsistency problem caused by the\naccumulated error in Interval Score Matching (ISM) when using the Denoising\nDiffusion Implicit Models (DDIM) inversion process. Unlike ISM which adopts the\ninversion process of DDIM to calculate on a single path, our TSM method\nleverages the inversion process of DDIM to generate two paths from the same\nstarting point for calculation. Since both paths start from the same starting\npoint, TSM can reduce the accumulated error compared to ISM, thus alleviating\nthe problem of pseudo ground truth inconsistency. TSM enhances the stability\nand consistency of the model's generated paths during the distillation process.\nWe demonstrate this experimentally and further show that ISM is a special case\nof TSM. Furthermore, to optimize the current multi-stage optimization process\nfrom high-resolution text to 3D generation, we adopt Stable Diffusion XL for\nguidance. In response to the issues of abnormal replication and splitting\ncaused by unstable gradients during the 3D Gaussian splatting process when\nusing Stable Diffusion XL, we propose a pixel-by-pixel gradient clipping\nmethod. Extensive experiments show that our model significantly surpasses the\nstate-of-the-art models in terms of visual quality and performance. Code:\n\\url{https://github.com/xingy038/Dreamer-XL}.\n", "date": "2024-05-18", "categories": []} {"id": "2405.10508", "title": "ART3D: 3D Gaussian Splatting for Text-Guided Artistic Scenes Generation", "abstract": " In this paper, we explore the existing challenges in 3D artistic scene\ngeneration by introducing ART3D, a novel framework that combines diffusion\nmodels and 3D Gaussian splatting techniques. Our method effectively bridges the\ngap between artistic and realistic images through an innovative image semantic\ntransfer algorithm. By leveraging depth information and an initial artistic\nimage, we generate a point cloud map, addressing domain differences.\nAdditionally, we propose a depth consistency module to enhance 3D scene\nconsistency. Finally, the 3D scene serves as initial points for optimizing\nGaussian splats. Experimental results demonstrate ART3D's superior performance\nin both content and structural consistency metrics when compared to existing\nmethods. ART3D significantly advances the field of AI in art creation by\nproviding an innovative solution for generating high-quality 3D artistic\nscenes.\n", "date": "2024-05-17", "categories": []} {"id": "2405.09874", "title": "Dual3D: Efficient and Consistent Text-to-3D Generation with Dual-mode\n Multi-view Latent Diffusion", "abstract": " We present Dual3D, a novel text-to-3D generation framework that generates\nhigh-quality 3D assets from texts in only $1$ minute.The key component is a\ndual-mode multi-view latent diffusion model. Given the noisy multi-view\nlatents, the 2D mode can efficiently denoise them with a single latent\ndenoising network, while the 3D mode can generate a tri-plane neural surface\nfor consistent rendering-based denoising. Most modules for both modes are tuned\nfrom a pre-trained text-to-image latent diffusion model to circumvent the\nexpensive cost of training from scratch. To overcome the high rendering cost\nduring inference, we propose the dual-mode toggling inference strategy to use\nonly $1/10$ denoising steps with 3D mode, successfully generating a 3D asset in\njust $10$ seconds without sacrificing quality. The texture of the 3D asset can\nbe further enhanced by our efficient texture refinement process in a short\ntime. Extensive experiments demonstrate that our method delivers\nstate-of-the-art performance while significantly reducing generation time. Our\nproject page is available at https://dual3d.github.io\n", "date": "2024-05-16", "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": "2403.15383", "title": "ThemeStation: Generating Theme-Aware 3D Assets from Few Exemplars", "abstract": " Real-world applications often require a large gallery of 3D assets that share\na consistent theme. While remarkable advances have been made in general 3D\ncontent creation from text or image, synthesizing customized 3D assets\nfollowing the shared theme of input 3D exemplars remains an open and\nchallenging problem. In this work, we present ThemeStation, a novel approach\nfor theme-aware 3D-to-3D generation. ThemeStation synthesizes customized 3D\nassets based on given few exemplars with two goals: 1) unity for generating 3D\nassets that thematically align with the given exemplars and 2) diversity for\ngenerating 3D assets with a high degree of variations. To this end, we design a\ntwo-stage framework that draws a concept image first, followed by a\nreference-informed 3D modeling stage. We propose a novel dual score\ndistillation (DSD) loss to jointly leverage priors from both the input\nexemplars and the synthesized concept image. Extensive experiments and user\nstudies confirm that ThemeStation surpasses prior works in producing diverse\ntheme-aware 3D models with impressive quality. ThemeStation also enables\nvarious applications such as controllable 3D-to-3D generation.\n", "date": "2024-03-22", "categories": []} {"id": "2405.09431", "title": "A Survey On Text-to-3D Contents Generation In The Wild", "abstract": " 3D content creation plays a vital role in various applications, such as\ngaming, robotics simulation, and virtual reality. However, the process is\nlabor-intensive and time-consuming, requiring skilled designers to invest\nconsiderable effort in creating a single 3D asset. To address this challenge,\ntext-to-3D generation technologies have emerged as a promising solution for\nautomating 3D creation. Leveraging the success of large vision language models,\nthese techniques aim to generate 3D content based on textual descriptions.\nDespite recent advancements in this area, existing solutions still face\nsignificant limitations in terms of generation quality and efficiency. In this\nsurvey, we conduct an in-depth investigation of the latest text-to-3D creation\nmethods. We provide a comprehensive background on text-to-3D creation,\nincluding discussions on datasets employed in training and evaluation metrics\nused to assess the quality of generated 3D models. Then, we delve into the\nvarious 3D representations that serve as the foundation for the 3D generation\nprocess. Furthermore, we present a thorough comparison of the rapidly growing\nliterature on generative pipelines, categorizing them into feedforward\ngenerators, optimization-based generation, and view reconstruction approaches.\nBy examining the strengths and weaknesses of these methods, we aim to shed\nlight on their respective capabilities and limitations. Lastly, we point out\nseveral promising avenues for future research. With this survey, we hope to\ninspire researchers further to explore the potential of open-vocabulary\ntext-conditioned 3D content creation.\n", "date": "2024-05-15", "categories": []} {"id": "2405.08054", "title": "Coin3D: Controllable and Interactive 3D Assets Generation with\n Proxy-Guided Conditioning", "abstract": " As humans, we aspire to create media content that is both freely willed and\nreadily controlled. Thanks to the prominent development of generative\ntechniques, we now can easily utilize 2D diffusion methods to synthesize images\ncontrolled by raw sketch or designated human poses, and even progressively\nedit/regenerate local regions with masked inpainting. However, similar\nworkflows in 3D modeling tasks are still unavailable due to the lack of\ncontrollability and efficiency in 3D generation. In this paper, we present a\nnovel controllable and interactive 3D assets modeling framework, named Coin3D.\nCoin3D allows users to control the 3D generation using a coarse geometry proxy\nassembled from basic shapes, and introduces an interactive generation workflow\nto support seamless local part editing while delivering responsive 3D object\npreviewing within a few seconds. To this end, we develop several techniques,\nincluding the 3D adapter that applies volumetric coarse shape control to the\ndiffusion model, proxy-bounded editing strategy for precise part editing,\nprogressive volume cache to support responsive preview, and volume-SDS to\nensure consistent mesh reconstruction. Extensive experiments of interactive\ngeneration and editing on diverse shape proxies demonstrate that our method\nachieves superior controllability and flexibility in the 3D assets generation\ntask.\n", "date": "2024-05-13", "categories": []} {"id": "2405.08055", "title": "DiffTF++: 3D-aware Diffusion Transformer for Large-Vocabulary 3D\n Generation", "abstract": " Generating diverse and high-quality 3D assets automatically poses a\nfundamental yet challenging task in 3D computer vision. Despite extensive\nefforts in 3D generation, existing optimization-based approaches struggle to\nproduce large-scale 3D assets efficiently. Meanwhile, feed-forward methods\noften focus on generating only a single category or a few categories, limiting\ntheir generalizability. Therefore, we introduce a diffusion-based feed-forward\nframework to address these challenges with a single model. To handle the large\ndiversity and complexity in geometry and texture across categories efficiently,\nwe 1) adopt improved triplane to guarantee efficiency; 2) introduce the\n3D-aware transformer to aggregate the generalized 3D knowledge with specialized\n3D features; and 3) devise the 3D-aware encoder/decoder to enhance the\ngeneralized 3D knowledge. Building upon our 3D-aware Diffusion model with\nTransFormer, DiffTF, we propose a stronger version for 3D generation, i.e.,\nDiffTF++. It boils down to two parts: multi-view reconstruction loss and\ntriplane refinement. Specifically, we utilize multi-view reconstruction loss to\nfine-tune the diffusion model and triplane decoder, thereby avoiding the\nnegative influence caused by reconstruction errors and improving texture\nsynthesis. By eliminating the mismatch between the two stages, the generative\nperformance is enhanced, especially in texture. Additionally, a 3D-aware\nrefinement process is introduced to filter out artifacts and refine triplanes,\nresulting in the generation of more intricate and reasonable details. Extensive\nexperiments on ShapeNet and OmniObject3D convincingly demonstrate the\neffectiveness of our proposed modules and the state-of-the-art 3D object\ngeneration performance with large diversity, rich semantics, and high quality.\n", "date": "2024-05-13", "categories": []} {"id": "2310.08529", "title": "GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging\n 2D and 3D Diffusion Models", "abstract": " In recent times, the generation of 3D assets from text prompts has shown\nimpressive results. Both 2D and 3D diffusion models can help generate decent 3D\nobjects based on prompts. 3D diffusion models have good 3D consistency, but\ntheir quality and generalization are limited as trainable 3D data is expensive\nand hard to obtain. 2D diffusion models enjoy strong abilities of\ngeneralization and fine generation, but 3D consistency is hard to guarantee.\nThis paper attempts to bridge the power from the two types of diffusion models\nvia the recent explicit and efficient 3D Gaussian splatting representation. A\nfast 3D object generation framework, named as GaussianDreamer, is proposed,\nwhere the 3D diffusion model provides priors for initialization and the 2D\ndiffusion model enriches the geometry and appearance. Operations of noisy point\ngrowing and color perturbation are introduced to enhance the initialized\nGaussians. Our GaussianDreamer can generate a high-quality 3D instance or 3D\navatar within 15 minutes on one GPU, much faster than previous methods, while\nthe generated instances can be directly rendered in real time. Demos and code\nare available at https://taoranyi.com/gaussiandreamer/.\n", "date": "2023-10-12", "categories": []} {"id": "2405.06547", "title": "OneTo3D: One Image to Re-editable Dynamic 3D Model and Video Generation", "abstract": " One image to editable dynamic 3D model and video generation is novel\ndirection and change in the research area of single image to 3D representation\nor 3D reconstruction of image. Gaussian Splatting has demonstrated its\nadvantages in implicit 3D reconstruction, compared with the original Neural\nRadiance Fields. As the rapid development of technologies and principles,\npeople tried to used the Stable Diffusion models to generate targeted models\nwith text instructions. However, using the normal implicit machine learning\nmethods is hard to gain the precise motions and actions control, further more,\nit is difficult to generate a long content and semantic continuous 3D video. To\naddress this issue, we propose the OneTo3D, a method and theory to used one\nsingle image to generate the editable 3D model and generate the targeted\nsemantic continuous time-unlimited 3D video. We used a normal basic Gaussian\nSplatting model to generate the 3D model from a single image, which requires\nless volume of video memory and computer calculation ability. Subsequently, we\ndesigned an automatic generation and self-adaptive binding mechanism for the\nobject armature. Combined with the re-editable motions and actions analyzing\nand controlling algorithm we proposed, we can achieve a better performance than\nthe SOTA projects in the area of building the 3D model precise motions and\nactions control, and generating a stable semantic continuous time-unlimited 3D\nvideo with the input text instructions. Here we will analyze the detailed\nimplementation methods and theories analyses. Relative comparisons and\nconclusions will be presented. The project code is open source.\n", "date": "2024-05-10", "categories": []} {"id": "2405.05768", "title": "FastScene: Text-Driven Fast 3D Indoor Scene Generation via Panoramic\n Gaussian Splatting", "abstract": " Text-driven 3D indoor scene generation holds broad applications, ranging from\ngaming and smart homes to AR/VR applications. Fast and high-fidelity scene\ngeneration is paramount for ensuring user-friendly experiences. However,\nexisting methods are characterized by lengthy generation processes or\nnecessitate the intricate manual specification of motion parameters, which\nintroduces inconvenience for users. Furthermore, these methods often rely on\nnarrow-field viewpoint iterative generations, compromising global consistency\nand overall scene quality. To address these issues, we propose FastScene, a\nframework for fast and higher-quality 3D scene generation, while maintaining\nthe scene consistency. Specifically, given a text prompt, we generate a\npanorama and estimate its depth, since the panorama encompasses information\nabout the entire scene and exhibits explicit geometric constraints. To obtain\nhigh-quality novel views, we introduce the Coarse View Synthesis (CVS) and\nProgressive Novel View Inpainting (PNVI) strategies, ensuring both scene\nconsistency and view quality. Subsequently, we utilize Multi-View Projection\n(MVP) to form perspective views, and apply 3D Gaussian Splatting (3DGS) for\nscene reconstruction. Comprehensive experiments demonstrate FastScene surpasses\nother methods in both generation speed and quality with better scene\nconsistency. Notably, guided only by a text prompt, FastScene can generate a 3D\nscene within a mere 15 minutes, which is at least one hour faster than\nstate-of-the-art methods, making it a paradigm for user-friendly scene\ngeneration.\n", "date": "2024-05-09", "categories": []} {"id": "2311.17834", "title": "Spice-E : Structural Priors in 3D Diffusion using Cross-Entity Attention", "abstract": " We are witnessing rapid progress in automatically generating and manipulating\n3D assets due to the availability of pretrained text-image diffusion models.\nHowever, time-consuming optimization procedures are required for synthesizing\neach sample, hindering their potential for democratizing 3D content creation.\nConversely, 3D diffusion models now train on million-scale 3D datasets,\nyielding high-quality text-conditional 3D samples within seconds. In this work,\nwe present Spice-E - a neural network that adds structural guidance to 3D\ndiffusion models, extending their usage beyond text-conditional generation. At\nits core, our framework introduces a cross-entity attention mechanism that\nallows for multiple entities (in particular, paired input and guidance 3D\nshapes) to interact via their internal representations within the denoising\nnetwork. We utilize this mechanism for learning task-specific structural priors\nin 3D diffusion models from auxiliary guidance shapes. We show that our\napproach supports a variety of applications, including 3D stylization, semantic\nshape editing and text-conditional abstraction-to-3D, which transforms\nprimitive-based abstractions into highly-expressive shapes. Extensive\nexperiments demonstrate that Spice-E achieves SOTA performance over these tasks\nwhile often being considerably faster than alternative methods. Importantly,\nthis is accomplished without tailoring our approach for any specific task.\n", "date": "2023-11-29", "categories": []} {"id": "2312.07539", "title": "HeadArtist: Text-conditioned 3D Head Generation with Self Score\n Distillation", "abstract": " This work presents HeadArtist for 3D head generation from text descriptions.\nWith a landmark-guided ControlNet serving as the generative prior, we come up\nwith an efficient pipeline that optimizes a parameterized 3D head model under\nthe supervision of the prior distillation itself. We call such a process self\nscore distillation (SSD). In detail, given a sampled camera pose, we first\nrender an image and its corresponding landmarks from the head model, and add\nsome particular level of noise onto the image. The noisy image, landmarks, and\ntext condition are then fed into the frozen ControlNet twice for noise\nprediction. Two different classifier-free guidance (CFG) weights are applied\nduring these two predictions, and the prediction difference offers a direction\non how the rendered image can better match the text of interest. Experimental\nresults suggest that our approach delivers high-quality 3D head sculptures with\nadequate geometry and photorealistic appearance, significantly outperforming\nstate-ofthe-art methods. We also show that the same pipeline well supports\nediting the generated heads, including both geometry deformation and appearance\nchange.\n", "date": "2023-12-12", "categories": []} {"id": "2311.16854", "title": "A Unified Approach for Text- and Image-guided 4D Scene Generation", "abstract": " Large-scale diffusion generative models are greatly simplifying image, video\nand 3D asset creation from user-provided text prompts and images. However, the\nchallenging problem of text-to-4D dynamic 3D scene generation with diffusion\nguidance remains largely unexplored. We propose Dream-in-4D, which features a\nnovel two-stage approach for text-to-4D synthesis, leveraging (1) 3D and 2D\ndiffusion guidance to effectively learn a high-quality static 3D asset in the\nfirst stage; (2) a deformable neural radiance field that explicitly\ndisentangles the learned static asset from its deformation, preserving quality\nduring motion learning; and (3) a multi-resolution feature grid for the\ndeformation field with a displacement total variation loss to effectively learn\nmotion with video diffusion guidance in the second stage. Through a user\npreference study, we demonstrate that our approach significantly advances image\nand motion quality, 3D consistency and text fidelity for text-to-4D generation\ncompared to baseline approaches. Thanks to its motion-disentangled\nrepresentation, Dream-in-4D can also be easily adapted for controllable\ngeneration where appearance is defined by one or multiple images, without the\nneed to modify the motion learning stage. Thus, our method offers, for the\nfirst time, a unified approach for text-to-4D, image-to-4D and personalized 4D\ngeneration tasks.\n", "date": "2023-11-28", "categories": []} {"id": "2403.02234", "title": "3DTopia: Large Text-to-3D Generation Model with Hybrid Diffusion Priors", "abstract": " We present a two-stage text-to-3D generation system, namely 3DTopia, which\ngenerates high-quality general 3D assets within 5 minutes using hybrid\ndiffusion priors. The first stage samples from a 3D diffusion prior directly\nlearned from 3D data. Specifically, it is powered by a text-conditioned\ntri-plane latent diffusion model, which quickly generates coarse 3D samples for\nfast prototyping. The second stage utilizes 2D diffusion priors to further\nrefine the texture of coarse 3D models from the first stage. The refinement\nconsists of both latent and pixel space optimization for high-quality texture\ngeneration. To facilitate the training of the proposed system, we clean and\ncaption the largest open-source 3D dataset, Objaverse, by combining the power\nof vision language models and large language models. Experiment results are\nreported qualitatively and quantitatively to show the performance of the\nproposed system. Our codes and models are available at\nhttps://github.com/3DTopia/3DTopia\n", "date": "2024-03-04", "categories": []} {"id": "2306.12422", "title": "DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D\n Generation", "abstract": " Text-to-image diffusion models pre-trained on billions of image-text pairs\nhave recently enabled 3D content creation by optimizing a randomly initialized\ndifferentiable 3D representation with score distillation. However, the\noptimization process suffers slow convergence and the resultant 3D models often\nexhibit two limitations: (a) quality concerns such as missing attributes and\ndistorted shape and texture; (b) extremely low diversity comparing to\ntext-guided image synthesis. In this paper, we show that the conflict between\nthe 3D optimization process and uniform timestep sampling in score distillation\nis the main reason for these limitations. To resolve this conflict, we propose\nto prioritize timestep sampling with monotonically non-increasing functions,\nwhich aligns the 3D optimization process with the sampling process of diffusion\nmodel. Extensive experiments show that our simple redesign significantly\nimproves 3D content creation with faster convergence, better quality and\ndiversity.\n", "date": "2023-06-21", "categories": []} {"id": "2401.03470", "title": "FurniScene: A Large-scale 3D Room Dataset with Intricate Furnishing\n Scenes", "abstract": " Indoor scene generation has attracted significant attention recently as it is\ncrucial for applications of gaming, virtual reality, and interior design.\nCurrent indoor scene generation methods can produce reasonable room layouts but\noften lack diversity and realism. This is primarily due to the limited coverage\nof existing datasets, including only large furniture without tiny furnishings\nin daily life. To address these challenges, we propose FurniScene, a\nlarge-scale 3D room dataset with intricate furnishing scenes from interior\ndesign professionals. Specifically, the FurniScene consists of 11,698 rooms and\n39,691 unique furniture CAD models with 89 different types, covering things\nfrom large beds to small teacups on the coffee table. To better suit\nfine-grained indoor scene layout generation, we introduce a novel Two-Stage\nDiffusion Scene Model (TSDSM) and conduct an evaluation benchmark for various\nindoor scene generation based on FurniScene. Quantitative and qualitative\nevaluations demonstrate the capability of our method to generate highly\nrealistic indoor scenes. Our dataset and code will be publicly available soon.\n", "date": "2024-01-07", "categories": []} {"id": "2306.08068", "title": "DORSal: Diffusion for Object-centric Representations of Scenes et al", "abstract": " Recent progress in 3D scene understanding enables scalable learning of\nrepresentations across large datasets of diverse scenes. As a consequence,\ngeneralization to unseen scenes and objects, rendering novel views from just a\nsingle or a handful of input images, and controllable scene generation that\nsupports editing, is now possible. However, training jointly on a large number\nof scenes typically compromises rendering quality when compared to single-scene\noptimized models such as NeRFs. In this paper, we leverage recent progress in\ndiffusion models to equip 3D scene representation learning models with the\nability to render high-fidelity novel views, while retaining benefits such as\nobject-level scene editing to a large degree. In particular, we propose DORSal,\nwhich adapts a video diffusion architecture for 3D scene generation conditioned\non frozen object-centric slot-based representations of scenes. On both complex\nsynthetic multi-object scenes and on the real-world large-scale Street View\ndataset, we show that DORSal enables scalable neural rendering of 3D scenes\nwith object-level editing and improves upon existing approaches.\n", "date": "2023-06-13", "categories": []} {"id": "2404.04562", "title": "Diffusion Time-step Curriculum for One Image to 3D Generation", "abstract": " Score distillation sampling~(SDS) has been widely adopted to overcome the\nabsence of unseen views in reconstructing 3D objects from a \\textbf{single}\nimage. It leverages pre-trained 2D diffusion models as teacher to guide the\nreconstruction of student 3D models. Despite their remarkable success,\nSDS-based methods often encounter geometric artifacts and texture saturation.\nWe find out the crux is the overlooked indiscriminate treatment of diffusion\ntime-steps during optimization: it unreasonably treats the student-teacher\nknowledge distillation to be equal at all time-steps and thus entangles\ncoarse-grained and fine-grained modeling. Therefore, we propose the Diffusion\nTime-step Curriculum one-image-to-3D pipeline (DTC123), which involves both the\nteacher and student models collaborating with the time-step curriculum in a\ncoarse-to-fine manner. Extensive experiments on NeRF4, RealFusion15, GSO and\nLevel50 benchmark demonstrate that DTC123 can produce multi-view consistent,\nhigh-quality, and diverse 3D assets. Codes and more generation demos will be\nreleased in https://github.com/yxymessi/DTC123.\n", "date": "2024-04-06", "categories": []} {"id": "2402.02972", "title": "Retrieval-Augmented Score Distillation for Text-to-3D Generation", "abstract": " Text-to-3D generation has achieved significant success by incorporating\npowerful 2D diffusion models, but insufficient 3D prior knowledge also leads to\nthe inconsistency of 3D geometry. Recently, since large-scale multi-view\ndatasets have been released, fine-tuning the diffusion model on the multi-view\ndatasets becomes a mainstream to solve the 3D inconsistency problem. However,\nit has confronted with fundamental difficulties regarding the limited quality\nand diversity of 3D data, compared with 2D data. To sidestep these trade-offs,\nwe explore a retrieval-augmented approach tailored for score distillation,\ndubbed ReDream. We postulate that both expressiveness of 2D diffusion models\nand geometric consistency of 3D assets can be fully leveraged by employing the\nsemantically relevant assets directly within the optimization process. To this\nend, we introduce novel framework for retrieval-based quality enhancement in\ntext-to-3D generation. We leverage the retrieved asset to incorporate its\ngeometric prior in the variational objective and adapt the diffusion model's 2D\nprior toward view consistency, achieving drastic improvements in both geometry\nand fidelity of generated scenes. We conduct extensive experiments to\ndemonstrate that ReDream exhibits superior quality with increased geometric\nconsistency. Project page is available at https://ku-cvlab.github.io/ReDream/.\n", "date": "2024-02-05", "categories": []} {"id": "2405.00954", "title": "X-Oscar: A Progressive Framework for High-quality Text-guided 3D\n Animatable Avatar Generation", "abstract": " Recent advancements in automatic 3D avatar generation guided by text have\nmade significant progress. However, existing methods have limitations such as\noversaturation and low-quality output. To address these challenges, we propose\nX-Oscar, a progressive framework for generating high-quality animatable avatars\nfrom text prompts. It follows a sequential Geometry->Texture->Animation\nparadigm, simplifying optimization through step-by-step generation. To tackle\noversaturation, we introduce Adaptive Variational Parameter (AVP), representing\navatars as an adaptive distribution during training. Additionally, we present\nAvatar-aware Score Distillation Sampling (ASDS), a novel technique that\nincorporates avatar-aware noise into rendered images for improved generation\nquality during optimization. Extensive evaluations confirm the superiority of\nX-Oscar over existing text-to-3D and text-to-avatar approaches. Our anonymous\nproject page: https://xmu-xiaoma666.github.io/Projects/X-Oscar/.\n", "date": "2024-05-02", "categories": []} {"id": "2403.00372", "title": "HyperSDFusion: Bridging Hierarchical Structures in Language and Geometry\n for Enhanced 3D Text2Shape Generation", "abstract": " 3D shape generation from text is a fundamental task in 3D representation\nlearning. The text-shape pairs exhibit a hierarchical structure, where a\ngeneral text like ``chair\" covers all 3D shapes of the chair, while more\ndetailed prompts refer to more specific shapes. Furthermore, both text and 3D\nshapes are inherently hierarchical structures. However, existing Text2Shape\nmethods, such as SDFusion, do not exploit that. In this work, we propose\nHyperSDFusion, a dual-branch diffusion model that generates 3D shapes from a\ngiven text. Since hyperbolic space is suitable for handling hierarchical data,\nwe propose to learn the hierarchical representations of text and 3D shapes in\nhyperbolic space. First, we introduce a hyperbolic text-image encoder to learn\nthe sequential and multi-modal hierarchical features of text in hyperbolic\nspace. In addition, we design a hyperbolic text-graph convolution module to\nlearn the hierarchical features of text in hyperbolic space. In order to fully\nutilize these text features, we introduce a dual-branch structure to embed text\nfeatures in 3D feature space. At last, to endow the generated 3D shapes with a\nhierarchical structure, we devise a hyperbolic hierarchical loss. Our method is\nthe first to explore the hyperbolic hierarchical representation for\ntext-to-shape generation. Experimental results on the existing text-to-shape\npaired dataset, Text2Shape, achieved state-of-the-art results. We release our\nimplementation under HyperSDFusion.github.io.\n", "date": "2024-03-01", "categories": []} {"id": "2404.19758", "title": "Invisible Stitch: Generating Smooth 3D Scenes with Depth Inpainting", "abstract": " 3D scene generation has quickly become a challenging new research direction,\nfueled by consistent improvements of 2D generative diffusion models. Most prior\nwork in this area generates scenes by iteratively stitching newly generated\nframes with existing geometry. These works often depend on pre-trained\nmonocular depth estimators to lift the generated images into 3D, fusing them\nwith the existing scene representation. These approaches are then often\nevaluated via a text metric, measuring the similarity between the generated\nimages and a given text prompt. In this work, we make two fundamental\ncontributions to the field of 3D scene generation. First, we note that lifting\nimages to 3D with a monocular depth estimation model is suboptimal as it\nignores the geometry of the existing scene. We thus introduce a novel depth\ncompletion model, trained via teacher distillation and self-training to learn\nthe 3D fusion process, resulting in improved geometric coherence of the scene.\nSecond, we introduce a new benchmarking scheme for scene generation methods\nthat is based on ground truth geometry, and thus measures the quality of the\nstructure of the scene.\n", "date": "2024-04-30", "categories": []} {"id": "2311.08403", "title": "Instant3D: Instant Text-to-3D Generation", "abstract": " Text-to-3D generation has attracted much attention from the computer vision\ncommunity. Existing methods mainly optimize a neural field from scratch for\neach text prompt, relying on heavy and repetitive training cost which impedes\ntheir practical deployment. In this paper, we propose a novel framework for\nfast text-to-3D generation, dubbed Instant3D. Once trained, Instant3D is able\nto create a 3D object for an unseen text prompt in less than one second with a\nsingle run of a feedforward network. We achieve this remarkable speed by\ndevising a new network that directly constructs a 3D triplane from a text\nprompt. The core innovation of our Instant3D lies in our exploration of\nstrategies to effectively inject text conditions into the network. In\nparticular, we propose to combine three key mechanisms: cross-attention, style\ninjection, and token-to-plane transformation, which collectively ensure precise\nalignment of the output with the input text. Furthermore, we propose a simple\nyet effective activation function, the scaled-sigmoid, to replace the original\nsigmoid function, which speeds up the training convergence by more than ten\ntimes. Finally, to address the Janus (multi-head) problem in 3D generation, we\npropose an adaptive Perp-Neg algorithm that can dynamically adjust its concept\nnegation scales according to the severity of the Janus problem during training,\neffectively reducing the multi-head effect. Extensive experiments on a wide\nvariety of benchmark datasets demonstrate that the proposed algorithm performs\nfavorably against the state-of-the-art methods both qualitatively and\nquantitatively, while achieving significantly better efficiency. The code,\ndata, and models are available at https://github.com/ming1993li/Instant3DCodes.\n", "date": "2023-11-14", "categories": []} {"id": "2404.18065", "title": "Grounded Compositional and Diverse Text-to-3D with Pretrained Multi-View\n Diffusion Model", "abstract": " In this paper, we propose an effective two-stage approach named\nGrounded-Dreamer to generate 3D assets that can accurately follow complex,\ncompositional text prompts while achieving high fidelity by using a pre-trained\nmulti-view diffusion model. Multi-view diffusion models, such as MVDream, have\nshown to generate high-fidelity 3D assets using score distillation sampling\n(SDS). However, applied naively, these methods often fail to comprehend\ncompositional text prompts, and may often entirely omit certain subjects or\nparts. To address this issue, we first advocate leveraging text-guided 4-view\nimages as the bottleneck in the text-to-3D pipeline. We then introduce an\nattention refocusing mechanism to encourage text-aligned 4-view image\ngeneration, without the necessity to re-train the multi-view diffusion model or\ncraft a high-quality compositional 3D dataset. We further propose a hybrid\noptimization strategy to encourage synergy between the SDS loss and the sparse\nRGB reference images. Our method consistently outperforms previous\nstate-of-the-art (SOTA) methods in generating compositional 3D assets,\nexcelling in both quality and accuracy, and enabling diverse 3D from the same\ntext prompt.\n", "date": "2024-04-28", "categories": []} {"id": "2404.17419", "title": "Multi-view Image Prompted Multi-view Diffusion for Improved 3D\n Generation", "abstract": " Using image as prompts for 3D generation demonstrate particularly strong\nperformances compared to using text prompts alone, for images provide a more\nintuitive guidance for the 3D generation process. In this work, we delve into\nthe potential of using multiple image prompts, instead of a single image\nprompt, for 3D generation. Specifically, we build on ImageDream, a novel\nimage-prompt multi-view diffusion model, to support multi-view images as the\ninput prompt. Our method, dubbed MultiImageDream, reveals that transitioning\nfrom a single-image prompt to multiple-image prompts enhances the performance\nof multi-view and 3D object generation according to various quantitative\nevaluation metrics and qualitative assessments. This advancement is achieved\nwithout the necessity of fine-tuning the pre-trained ImageDream multi-view\ndiffusion model.\n", "date": "2024-04-26", "categories": []} {"id": "2401.14828", "title": "TIP-Editor: An Accurate 3D Editor Following Both Text-Prompts And\n Image-Prompts", "abstract": " Text-driven 3D scene editing has gained significant attention owing to its\nconvenience and user-friendliness. However, existing methods still lack\naccurate control of the specified appearance and location of the editing result\ndue to the inherent limitations of the text description. To this end, we\npropose a 3D scene editing framework, TIPEditor, that accepts both text and\nimage prompts and a 3D bounding box to specify the editing region. With the\nimage prompt, users can conveniently specify the detailed appearance/style of\nthe target content in complement to the text description, enabling accurate\ncontrol of the appearance. Specifically, TIP-Editor employs a stepwise 2D\npersonalization strategy to better learn the representation of the existing\nscene and the reference image, in which a localization loss is proposed to\nencourage correct object placement as specified by the bounding box.\nAdditionally, TIPEditor utilizes explicit and flexible 3D Gaussian splatting as\nthe 3D representation to facilitate local editing while keeping the background\nunchanged. Extensive experiments have demonstrated that TIP-Editor conducts\naccurate editing following the text and image prompts in the specified bounding\nbox region, consistently outperforming the baselines in editing quality, and\nthe alignment to the prompts, qualitatively and quantitatively.\n", "date": "2024-01-26", "categories": []} {"id": "2404.16510", "title": "Interactive3D: Create What You Want by Interactive 3D Generation", "abstract": " 3D object generation has undergone significant advancements, yielding\nhigh-quality results. However, fall short of achieving precise user control,\noften yielding results that do not align with user expectations, thus limiting\ntheir applicability. User-envisioning 3D object generation faces significant\nchallenges in realizing its concepts using current generative models due to\nlimited interaction capabilities. Existing methods mainly offer two approaches:\n(i) interpreting textual instructions with constrained controllability, or (ii)\nreconstructing 3D objects from 2D images. Both of them limit customization to\nthe confines of the 2D reference and potentially introduce undesirable\nartifacts during the 3D lifting process, restricting the scope for direct and\nversatile 3D modifications. In this work, we introduce Interactive3D, an\ninnovative framework for interactive 3D generation that grants users precise\ncontrol over the generative process through extensive 3D interaction\ncapabilities. Interactive3D is constructed in two cascading stages, utilizing\ndistinct 3D representations. The first stage employs Gaussian Splatting for\ndirect user interaction, allowing modifications and guidance of the generative\ndirection at any intermediate step through (i) Adding and Removing components,\n(ii) Deformable and Rigid Dragging, (iii) Geometric Transformations, and (iv)\nSemantic Editing. Subsequently, the Gaussian splats are transformed into\nInstantNGP. We introduce a novel (v) Interactive Hash Refinement module to\nfurther add details and extract the geometry in the second stage. Our\nexperiments demonstrate that Interactive3D markedly improves the\ncontrollability and quality of 3D generation. Our project webpage is available\nat \\url{https://interactive-3d.github.io/}.\n", "date": "2024-04-25", "categories": []} {"id": "2404.16748", "title": "TELA: Text to Layer-wise 3D Clothed Human Generation", "abstract": " This paper addresses the task of 3D clothed human generation from textural\ndescriptions. Previous works usually encode the human body and clothes as a\nholistic model and generate the whole model in a single-stage optimization,\nwhich makes them struggle for clothing editing and meanwhile lose fine-grained\ncontrol over the whole generation process. To solve this, we propose a\nlayer-wise clothed human representation combined with a progressive\noptimization strategy, which produces clothing-disentangled 3D human models\nwhile providing control capacity for the generation process. The basic idea is\nprogressively generating a minimal-clothed human body and layer-wise clothes.\nDuring clothing generation, a novel stratified compositional rendering method\nis proposed to fuse multi-layer human models, and a new loss function is\nutilized to help decouple the clothing model from the human body. The proposed\nmethod achieves high-quality disentanglement, which thereby provides an\neffective way for 3D garment generation. Extensive experiments demonstrate that\nour approach achieves state-of-the-art 3D clothed human generation while also\nsupporting cloth editing applications such as virtual try-on. Project page:\nhttp://jtdong.com/tela_layer/\n", "date": "2024-04-25", "categories": []} {"id": "2312.09222", "title": "Mosaic-SDF for 3D Generative Models", "abstract": " Current diffusion or flow-based generative models for 3D shapes divide to\ntwo: distilling pre-trained 2D image diffusion models, and training directly on\n3D shapes. When training a diffusion or flow models on 3D shapes a crucial\ndesign choice is the shape representation. An effective shape representation\nneeds to adhere three design principles: it should allow an efficient\nconversion of large 3D datasets to the representation form; it should provide a\ngood tradeoff of approximation power versus number of parameters; and it should\nhave a simple tensorial form that is compatible with existing powerful neural\narchitectures. While standard 3D shape representations such as volumetric grids\nand point clouds do not adhere to all these principles simultaneously, we\nadvocate in this paper a new representation that does. We introduce Mosaic-SDF\n(M-SDF): a simple 3D shape representation that approximates the Signed Distance\nFunction (SDF) of a given shape by using a set of local grids spread near the\nshape's boundary. The M-SDF representation is fast to compute for each shape\nindividually making it readily parallelizable; it is parameter efficient as it\nonly covers the space around the shape's boundary; and it has a simple matrix\nform, compatible with Transformer-based architectures. We demonstrate the\nefficacy of the M-SDF representation by using it to train a 3D generative flow\nmodel including class-conditioned generation with the 3D Warehouse dataset, and\ntext-to-3D generation using a dataset of about 600k caption-shape pairs.\n", "date": "2023-12-14", "categories": []} {"id": "2404.01887", "title": "3D scene generation from scene graphs and self-attention", "abstract": " Synthesizing realistic and diverse indoor 3D scene layouts in a controllable\nfashion opens up applications in simulated navigation and virtual reality. As\nconcise and robust representations of a scene, scene graphs have proven to be\nwell-suited as the semantic control on the generated layout. We present a\nvariant of the conditional variational autoencoder (cVAE) model to synthesize\n3D scenes from scene graphs and floor plans. We exploit the properties of\nself-attention layers to capture high-level relationships between objects in a\nscene, and use these as the building blocks of our model. Our model, leverages\ngraph transformers to estimate the size, dimension and orientation of the\nobjects in a room while satisfying relationships in the given scene graph. Our\nexperiments shows self-attention layers leads to sparser (7.9x compared to\nGraphto3D) and more diverse scenes (16%).\n", "date": "2024-04-02", "categories": []} {"id": "2404.15538", "title": "DreamCraft: Text-Guided Generation of Functional 3D Environments in\n Minecraft", "abstract": " Procedural Content Generation (PCG) algorithms enable the automatic\ngeneration of complex and diverse artifacts. However, they don't provide\nhigh-level control over the generated content and typically require domain\nexpertise. In contrast, text-to-3D methods allow users to specify desired\ncharacteristics in natural language, offering a high amount of flexibility and\nexpressivity. But unlike PCG, such approaches cannot guarantee functionality,\nwhich is crucial for certain applications like game design. In this paper, we\npresent a method for generating functional 3D artifacts from free-form text\nprompts in the open-world game Minecraft. Our method, DreamCraft, trains\nquantized Neural Radiance Fields (NeRFs) to represent artifacts that, when\nviewed in-game, match given text descriptions. We find that DreamCraft produces\nmore aligned in-game artifacts than a baseline that post-processes the output\nof an unconstrained NeRF. Thanks to the quantized representation of the\nenvironment, functional constraints can be integrated using specialized loss\nterms. We show how this can be leveraged to generate 3D structures that match a\ntarget distribution or obey certain adjacency rules over the block types.\nDreamCraft inherits a high degree of expressivity and controllability from the\nNeRF, while still being able to incorporate functional constraints through\ndomain-specific objectives.\n", "date": "2024-04-23", "categories": []} {"id": "2312.09067", "title": "Holodeck: Language Guided Generation of 3D Embodied AI Environments", "abstract": " 3D simulated environments play a critical role in Embodied AI, but their\ncreation requires expertise and extensive manual effort, restricting their\ndiversity and scope. To mitigate this limitation, we present Holodeck, a system\nthat generates 3D environments to match a user-supplied prompt fully\nautomatedly. Holodeck can generate diverse scenes, e.g., arcades, spas, and\nmuseums, adjust the designs for styles, and can capture the semantics of\ncomplex queries such as \"apartment for a researcher with a cat\" and \"office of\na professor who is a fan of Star Wars\". Holodeck leverages a large language\nmodel (i.e., GPT-4) for common sense knowledge about what the scene might look\nlike and uses a large collection of 3D assets from Objaverse to populate the\nscene with diverse objects. To address the challenge of positioning objects\ncorrectly, we prompt GPT-4 to generate spatial relational constraints between\nobjects and then optimize the layout to satisfy those constraints. Our\nlarge-scale human evaluation shows that annotators prefer Holodeck over\nmanually designed procedural baselines in residential scenes and that Holodeck\ncan produce high-quality outputs for diverse scene types. We also demonstrate\nan exciting application of Holodeck in Embodied AI, training agents to navigate\nin novel scenes like music rooms and daycares without human-constructed data,\nwhich is a significant step forward in developing general-purpose embodied\nagents.\n", "date": "2023-12-14", "categories": []} {"id": "2312.09069", "title": "PI3D: Efficient Text-to-3D Generation with Pseudo-Image Diffusion", "abstract": " Diffusion models trained on large-scale text-image datasets have demonstrated\na strong capability of controllable high-quality image generation from\narbitrary text prompts. However, the generation quality and generalization\nability of 3D diffusion models is hindered by the scarcity of high-quality and\nlarge-scale 3D datasets. In this paper, we present PI3D, a framework that fully\nleverages the pre-trained text-to-image diffusion models' ability to generate\nhigh-quality 3D shapes from text prompts in minutes. The core idea is to\nconnect the 2D and 3D domains by representing a 3D shape as a set of Pseudo RGB\nImages. We fine-tune an existing text-to-image diffusion model to produce such\npseudo-images using a small number of text-3D pairs. Surprisingly, we find that\nit can already generate meaningful and consistent 3D shapes given complex text\ndescriptions. We further take the generated shapes as the starting point for a\nlightweight iterative refinement using score distillation sampling to achieve\nhigh-quality generation under a low budget. PI3D generates a single 3D shape\nfrom text in only 3 minutes and the quality is validated to outperform existing\n3D generative models by a large margin.\n", "date": "2023-12-14", "categories": []} {"id": "2401.02683", "title": "Geometric-Facilitated Denoising Diffusion Model for 3D Molecule\n Generation", "abstract": " Denoising diffusion models have shown great potential in multiple research\nareas. Existing diffusion-based generative methods on de novo 3D molecule\ngeneration face two major challenges. Since majority heavy atoms in molecules\nallow connections to multiple atoms through single bonds, solely using\npair-wise distance to model molecule geometries is insufficient. Therefore, the\nfirst one involves proposing an effective neural network as the denoising\nkernel that is capable to capture complex multi-body interatomic relationships\nand learn high-quality features. Due to the discrete nature of graphs,\nmainstream diffusion-based methods for molecules heavily rely on predefined\nrules and generate edges in an indirect manner. The second challenge involves\naccommodating molecule generation to diffusion and accurately predicting the\nexistence of bonds. In our research, we view the iterative way of updating\nmolecule conformations in diffusion process is consistent with molecular\ndynamics and introduce a novel molecule generation method named\nGeometric-Facilitated Molecular Diffusion (GFMDiff). For the first challenge,\nwe introduce a Dual-Track Transformer Network (DTN) to fully excevate global\nspatial relationships and learn high quality representations which contribute\nto accurate predictions of features and geometries. As for the second\nchallenge, we design Geometric-Facilitated Loss (GFLoss) which intervenes the\nformation of bonds during the training period, instead of directly embedding\nedges into the latent space. Comprehensive experiments on current benchmarks\ndemonstrate the superiority of GFMDiff.\n", "date": "2024-01-05", "categories": []} {"id": "2404.00815", "title": "Towards Realistic Scene Generation with LiDAR Diffusion Models", "abstract": " Diffusion models (DMs) excel in photo-realistic image synthesis, but their\nadaptation to LiDAR scene generation poses a substantial hurdle. This is\nprimarily because DMs operating in the point space struggle to preserve the\ncurve-like patterns and 3D geometry of LiDAR scenes, which consumes much of\ntheir representation power. In this paper, we propose LiDAR Diffusion Models\n(LiDMs) to generate LiDAR-realistic scenes from a latent space tailored to\ncapture the realism of LiDAR scenes by incorporating geometric priors into the\nlearning pipeline. Our method targets three major desiderata: pattern realism,\ngeometry realism, and object realism. Specifically, we introduce curve-wise\ncompression to simulate real-world LiDAR patterns, point-wise coordinate\nsupervision to learn scene geometry, and patch-wise encoding for a full 3D\nobject context. With these three core designs, our method achieves competitive\nperformance on unconditional LiDAR generation in 64-beam scenario and state of\nthe art on conditional LiDAR generation, while maintaining high efficiency\ncompared to point-based DMs (up to 107$\\times$ faster). Furthermore, by\ncompressing LiDAR scenes into a latent space, we enable the controllability of\nDMs with various conditions such as semantic maps, camera views, and text\nprompts.\n", "date": "2024-03-31", "categories": []} {"id": "2308.16512", "title": "MVDream: Multi-view Diffusion for 3D Generation", "abstract": " We introduce MVDream, a diffusion model that is able to generate consistent\nmulti-view images from a given text prompt. Learning from both 2D and 3D data,\na multi-view diffusion model can achieve the generalizability of 2D diffusion\nmodels and the consistency of 3D renderings. We demonstrate that such a\nmulti-view diffusion model is implicitly a generalizable 3D prior agnostic to\n3D representations. It can be applied to 3D generation via Score Distillation\nSampling, significantly enhancing the consistency and stability of existing\n2D-lifting methods. It can also learn new concepts from a few 2D examples, akin\nto DreamBooth, but for 3D generation.\n", "date": "2023-08-31", "categories": []} {"id": "2310.02977", "title": "T$^3$Bench: Benchmarking Current Progress in Text-to-3D Generation", "abstract": " Recent methods in text-to-3D leverage powerful pretrained diffusion models to\noptimize NeRF. Notably, these methods are able to produce high-quality 3D\nscenes without training on 3D data. Due to the open-ended nature of the task,\nmost studies evaluate their results with subjective case studies and user\nexperiments, thereby presenting a challenge in quantitatively addressing the\nquestion: How has current progress in Text-to-3D gone so far? In this paper, we\nintroduce T$^3$Bench, the first comprehensive text-to-3D benchmark containing\ndiverse text prompts of three increasing complexity levels that are specially\ndesigned for 3D generation. To assess both the subjective quality and the text\nalignment, we propose two automatic metrics based on multi-view images produced\nby the 3D contents. The quality metric combines multi-view text-image scores\nand regional convolution to detect quality and view inconsistency. The\nalignment metric uses multi-view captioning and GPT-4 evaluation to measure\ntext-3D consistency. Both metrics closely correlate with different dimensions\nof human judgments, providing a paradigm for efficiently evaluating text-to-3D\nmodels. The benchmarking results, shown in Fig. 1, reveal performance\ndifferences among an extensive 10 prevalent text-to-3D methods. Our analysis\nfurther highlights the common struggles for current methods on generating\nsurroundings and multi-object scenes, as well as the bottleneck of leveraging\n2D guidance for 3D generation. Our project page is available at:\nhttps://t3bench.com.\n", "date": "2023-10-04", "categories": []} {"id": "2404.10142", "title": "Shaping Realities: Enhancing 3D Generative AI with Fabrication\n Constraints", "abstract": " Generative AI tools are becoming more prevalent in 3D modeling, enabling\nusers to manipulate or create new models with text or images as inputs. This\nmakes it easier for users to rapidly customize and iterate on their 3D designs\nand explore new creative ideas. These methods focus on the aesthetic quality of\nthe 3D models, refining them to look similar to the prompts provided by the\nuser. However, when creating 3D models intended for fabrication, designers need\nto trade-off the aesthetic qualities of a 3D model with their intended physical\nproperties. To be functional post-fabrication, 3D models have to satisfy\nstructural constraints informed by physical principles. Currently, such\nrequirements are not enforced by generative AI tools. This leads to the\ndevelopment of aesthetically appealing, but potentially non-functional 3D\ngeometry, that would be hard to fabricate and use in the real world. This\nworkshop paper highlights the limitations of generative AI tools in translating\ndigital creations into the physical world and proposes new augmentations to\ngenerative AI tools for creating physically viable 3D models. We advocate for\nthe development of tools that manipulate or generate 3D models by considering\nnot only the aesthetic appearance but also using physical properties as\nconstraints. This exploration seeks to bridge the gap between digital\ncreativity and real-world applicability, extending the creative potential of\ngenerative AI into the tangible domain.\n", "date": "2024-04-15", "categories": []} {"id": "2404.10279", "title": "EucliDreamer: Fast and High-Quality Texturing for 3D Models with\n Depth-Conditioned Stable Diffusion", "abstract": " We present EucliDreamer, a simple and effective method to generate textures\nfor 3D models given text prompts and meshes. The texture is parametrized as an\nimplicit function on the 3D surface, which is optimized with the Score\nDistillation Sampling (SDS) process and differentiable rendering. To generate\nhigh-quality textures, we leverage a depth-conditioned Stable Diffusion model\nguided by the depth image rendered from the mesh. We test our approach on 3D\nmodels in Objaverse and conducted a user study, which shows its superior\nquality compared to existing texturing methods like Text2Tex. In addition, our\nmethod converges 2 times faster than DreamFusion. Through text prompting,\ntextures of diverse art styles can be produced. We hope Euclidreamer proides a\nviable solution to automate a labor-intensive stage in 3D content creation.\n", "date": "2024-04-16", "categories": []} {"id": "2404.10394", "title": "Portrait3D: Text-Guided High-Quality 3D Portrait Generation Using\n Pyramid Representation and GANs Prior", "abstract": " Existing neural rendering-based text-to-3D-portrait generation methods\ntypically make use of human geometry prior and diffusion models to obtain\nguidance. However, relying solely on geometry information introduces issues\nsuch as the Janus problem, over-saturation, and over-smoothing. We present\nPortrait3D, a novel neural rendering-based framework with a novel joint\ngeometry-appearance prior to achieve text-to-3D-portrait generation that\novercomes the aforementioned issues. To accomplish this, we train a 3D portrait\ngenerator, 3DPortraitGAN-Pyramid, as a robust prior. This generator is capable\nof producing 360{\\deg} canonical 3D portraits, serving as a starting point for\nthe subsequent diffusion-based generation process. To mitigate the \"grid-like\"\nartifact caused by the high-frequency information in the feature-map-based 3D\nrepresentation commonly used by most 3D-aware GANs, we integrate a novel\npyramid tri-grid 3D representation into 3DPortraitGAN-Pyramid. To generate 3D\nportraits from text, we first project a randomly generated image aligned with\nthe given prompt into the pre-trained 3DPortraitGAN-Pyramid's latent space. The\nresulting latent code is then used to synthesize a pyramid tri-grid. Beginning\nwith the obtained pyramid tri-grid, we use score distillation sampling to\ndistill the diffusion model's knowledge into the pyramid tri-grid. Following\nthat, we utilize the diffusion model to refine the rendered images of the 3D\nportrait and then use these refined images as training data to further optimize\nthe pyramid tri-grid, effectively eliminating issues with unrealistic color and\nunnatural artifacts. Our experimental results show that Portrait3D can produce\nrealistic, high-quality, and canonical 3D portraits that align with the prompt.\n", "date": "2024-04-16", "categories": []} {"id": "2303.16611", "title": "4D Facial Expression Diffusion Model", "abstract": " Facial expression generation is one of the most challenging and long-sought\naspects of character animation, with many interesting applications. The\nchallenging task, traditionally having relied heavily on digital craftspersons,\nremains yet to be explored. In this paper, we introduce a generative framework\nfor generating 3D facial expression sequences (i.e. 4D faces) that can be\nconditioned on different inputs to animate an arbitrary 3D face mesh. It is\ncomposed of two tasks: (1) Learning the generative model that is trained over a\nset of 3D landmark sequences, and (2) Generating 3D mesh sequences of an input\nfacial mesh driven by the generated landmark sequences. The generative model is\nbased on a Denoising Diffusion Probabilistic Model (DDPM), which has achieved\nremarkable success in generative tasks of other domains. While it can be\ntrained unconditionally, its reverse process can still be conditioned by\nvarious condition signals. This allows us to efficiently develop several\ndownstream tasks involving various conditional generation, by using expression\nlabels, text, partial sequences, or simply a facial geometry. To obtain the\nfull mesh deformation, we then develop a landmark-guided encoder-decoder to\napply the geometrical deformation embedded in landmarks on a given facial mesh.\nExperiments show that our model has learned to generate realistic, quality\nexpressions solely from the dataset of relatively small size, improving over\nthe state-of-the-art methods. Videos and qualitative comparisons with other\nmethods can be found at \\url{https://github.com/ZOUKaifeng/4DFM}.\n", "date": "2023-03-29", "categories": []} {"id": "2309.03453", "title": "SyncDreamer: Generating Multiview-consistent Images from a Single-view\n Image", "abstract": " In this paper, we present a novel diffusion model called that generates\nmultiview-consistent images from a single-view image. Using pretrained\nlarge-scale 2D diffusion models, recent work Zero123 demonstrates the ability\nto generate plausible novel views from a single-view image of an object.\nHowever, maintaining consistency in geometry and colors for the generated\nimages remains a challenge. To address this issue, we propose a synchronized\nmultiview diffusion model that models the joint probability distribution of\nmultiview images, enabling the generation of multiview-consistent images in a\nsingle reverse process. SyncDreamer synchronizes the intermediate states of all\nthe generated images at every step of the reverse process through a 3D-aware\nfeature attention mechanism that correlates the corresponding features across\ndifferent views. Experiments show that SyncDreamer generates images with high\nconsistency across different views, thus making it well-suited for various 3D\ngeneration tasks such as novel-view-synthesis, text-to-3D, and image-to-3D.\n", "date": "2023-09-07", "categories": []} {"id": "2404.07191", "title": "InstantMesh: Efficient 3D Mesh Generation from a Single Image with\n Sparse-view Large Reconstruction Models", "abstract": " We present InstantMesh, a feed-forward framework for instant 3D mesh\ngeneration from a single image, featuring state-of-the-art generation quality\nand significant training scalability. By synergizing the strengths of an\noff-the-shelf multiview diffusion model and a sparse-view reconstruction model\nbased on the LRM architecture, InstantMesh is able to create diverse 3D assets\nwithin 10 seconds. To enhance the training efficiency and exploit more\ngeometric supervisions, e.g, depths and normals, we integrate a differentiable\niso-surface extraction module into our framework and directly optimize on the\nmesh representation. Experimental results on public datasets demonstrate that\nInstantMesh significantly outperforms other latest image-to-3D baselines, both\nqualitatively and quantitatively. We release all the code, weights, and demo of\nInstantMesh, with the intention that it can make substantial contributions to\nthe community of 3D generative AI and empower both researchers and content\ncreators.\n", "date": "2024-04-10", "categories": []} {"id": "2404.09540", "title": "Text-Driven Diverse Facial Texture Generation via Progressive\n Latent-Space Refinement", "abstract": " Automatic 3D facial texture generation has gained significant interest\nrecently. Existing approaches may not support the traditional physically based\nrendering pipeline or rely on 3D data captured by Light Stage. Our key\ncontribution is a progressive latent space refinement approach that can\nbootstrap from 3D Morphable Models (3DMMs)-based texture maps generated from\nfacial images to generate high-quality and diverse PBR textures, including\nalbedo, normal, and roughness. It starts with enhancing Generative Adversarial\nNetworks (GANs) for text-guided and diverse texture generation. To this end, we\ndesign a self-supervised paradigm to overcome the reliance on ground truth 3D\ntextures and train the generative model with only entangled texture maps.\nBesides, we foster mutual enhancement between GANs and Score Distillation\nSampling (SDS). SDS boosts GANs with more generative modes, while GANs promote\nmore efficient optimization of SDS. Furthermore, we introduce an edge-aware SDS\nfor multi-view consistent facial structure. Experiments demonstrate that our\nmethod outperforms existing 3D texture generation methods regarding\nphoto-realistic quality, diversity, and efficiency.\n", "date": "2024-04-15", "categories": []} {"id": "2311.13570", "title": "WildFusion: Learning 3D-Aware Latent Diffusion Models in View Space", "abstract": " Modern learning-based approaches to 3D-aware image synthesis achieve high\nphotorealism and 3D-consistent viewpoint changes for the generated images.\nExisting approaches represent instances in a shared canonical space. However,\nfor in-the-wild datasets a shared canonical system can be difficult to define\nor might not even exist. In this work, we instead model instances in view\nspace, alleviating the need for posed images and learned camera distributions.\nWe find that in this setting, existing GAN-based methods are prone to\ngenerating flat geometry and struggle with distribution coverage. We hence\npropose WildFusion, a new approach to 3D-aware image synthesis based on latent\ndiffusion models (LDMs). We first train an autoencoder that infers a compressed\nlatent representation, which additionally captures the images' underlying 3D\nstructure and enables not only reconstruction but also novel view synthesis. To\nlearn a faithful 3D representation, we leverage cues from monocular depth\nprediction. Then, we train a diffusion model in the 3D-aware latent space,\nthereby enabling synthesis of high-quality 3D-consistent image samples,\noutperforming recent state-of-the-art GAN-based methods. Importantly, our\n3D-aware LDM is trained without any direct supervision from multiview images or\n3D geometry and does not require posed images or learned pose or camera\ndistributions. It directly learns a 3D representation without relying on\ncanonical camera coordinates. This opens up promising research avenues for\nscalable 3D-aware image synthesis and 3D content creation from in-the-wild\nimage data. See https://katjaschwarz.github.io/wildfusion for videos of our 3D\nresults.\n", "date": "2023-11-22", "categories": []} {"id": "2312.03884", "title": "WonderJourney: Going from Anywhere to Everywhere", "abstract": " We introduce WonderJourney, a modularized framework for perpetual 3D scene\ngeneration. Unlike prior work on view generation that focuses on a single type\nof scenes, we start at any user-provided location (by a text description or an\nimage) and generate a journey through a long sequence of diverse yet coherently\nconnected 3D scenes. We leverage an LLM to generate textual descriptions of the\nscenes in this journey, a text-driven point cloud generation pipeline to make a\ncompelling and coherent sequence of 3D scenes, and a large VLM to verify the\ngenerated scenes. We show compelling, diverse visual results across various\nscene types and styles, forming imaginary \"wonderjourneys\". Project website:\nhttps://kovenyu.com/WonderJourney/\n", "date": "2023-12-06", "categories": []} {"id": "2312.16837", "title": "DiffusionGAN3D: Boosting Text-guided 3D Generation and Domain Adaptation\n by Combining 3D GANs and Diffusion Priors", "abstract": " Text-guided domain adaptation and generation of 3D-aware portraits find many\napplications in various fields. However, due to the lack of training data and\nthe challenges in handling the high variety of geometry and appearance, the\nexisting methods for these tasks suffer from issues like inflexibility,\ninstability, and low fidelity. In this paper, we propose a novel framework\nDiffusionGAN3D, which boosts text-guided 3D domain adaptation and generation by\ncombining 3D GANs and diffusion priors. Specifically, we integrate the\npre-trained 3D generative models (e.g., EG3D) and text-to-image diffusion\nmodels. The former provides a strong foundation for stable and high-quality\navatar generation from text. And the diffusion models in turn offer powerful\npriors and guide the 3D generator finetuning with informative direction to\nachieve flexible and efficient text-guided domain adaptation. To enhance the\ndiversity in domain adaptation and the generation capability in text-to-avatar,\nwe introduce the relative distance loss and case-specific learnable triplane\nrespectively. Besides, we design a progressive texture refinement module to\nimprove the texture quality for both tasks above. Extensive experiments\ndemonstrate that the proposed framework achieves excellent results in both\ndomain adaptation and text-to-avatar tasks, outperforming existing methods in\nterms of generation quality and efficiency. The project homepage is at\nhttps://younglbw.github.io/DiffusionGAN3D-homepage/.\n", "date": "2023-12-28", "categories": []} {"id": "2402.16994", "title": "GEM3D: GEnerative Medial Abstractions for 3D Shape Synthesis", "abstract": " We introduce GEM3D -- a new deep, topology-aware generative model of 3D\nshapes. The key ingredient of our method is a neural skeleton-based\nrepresentation encoding information on both shape topology and geometry.\nThrough a denoising diffusion probabilistic model, our method first generates\nskeleton-based representations following the Medial Axis Transform (MAT), then\ngenerates surfaces through a skeleton-driven neural implicit formulation. The\nneural implicit takes into account the topological and geometric information\nstored in the generated skeleton representations to yield surfaces that are\nmore topologically and geometrically accurate compared to previous neural field\nformulations. We discuss applications of our method in shape synthesis and\npoint cloud reconstruction tasks, and evaluate our method both qualitatively\nand quantitatively. We demonstrate significantly more faithful surface\nreconstruction and diverse shape generation results compared to the\nstate-of-the-art, also involving challenging scenarios of reconstructing and\nsynthesizing structurally complex, high-genus shape surfaces from Thingi10K and\nShapeNet.\n", "date": "2024-02-26", "categories": []} {"id": "2312.13980", "title": "Carve3D: Improving Multi-view Reconstruction Consistency for Diffusion\n Models with RL Finetuning", "abstract": " Multi-view diffusion models, obtained by applying Supervised Finetuning (SFT)\nto text-to-image diffusion models, have driven recent breakthroughs in\ntext-to-3D research. However, due to the limited size and quality of existing\n3D datasets, they still suffer from multi-view inconsistencies and Neural\nRadiance Field (NeRF) reconstruction artifacts. We argue that multi-view\ndiffusion models can benefit from further Reinforcement Learning Finetuning\n(RLFT), which allows models to learn from the data generated by themselves and\nimprove beyond their dataset limitations during SFT. To this end, we introduce\nCarve3D, an improved RLFT algorithm coupled with a novel Multi-view\nReconstruction Consistency (MRC) metric, to enhance the consistency of\nmulti-view diffusion models. To measure the MRC metric on a set of multi-view\nimages, we compare them with their corresponding NeRF renderings at the same\ncamera viewpoints. The resulting model, which we denote as Carve3DM,\ndemonstrates superior multi-view consistency and NeRF reconstruction quality\nthan existing models. Our results suggest that pairing SFT with Carve3D's RLFT\nis essential for developing multi-view-consistent diffusion models, mirroring\nthe standard Large Language Model (LLM) alignment pipeline. Our code, training\nand testing data, and video results are available at:\nhttps://desaixie.github.io/carve-3d.\n", "date": "2023-12-21", "categories": []} {"id": "2404.06780", "title": "Urban Architect: Steerable 3D Urban Scene Generation with Layout Prior", "abstract": " Text-to-3D generation has achieved remarkable success via large-scale\ntext-to-image diffusion models. Nevertheless, there is no paradigm for scaling\nup the methodology to urban scale. Urban scenes, characterized by numerous\nelements, intricate arrangement relationships, and vast scale, present a\nformidable barrier to the interpretability of ambiguous textual descriptions\nfor effective model optimization. In this work, we surmount the limitations by\nintroducing a compositional 3D layout representation into text-to-3D paradigm,\nserving as an additional prior. It comprises a set of semantic primitives with\nsimple geometric structures and explicit arrangement relationships,\ncomplementing textual descriptions and enabling steerable generation. Upon\nthis, we propose two modifications -- (1) We introduce Layout-Guided\nVariational Score Distillation to address model optimization inadequacies. It\nconditions the score distillation sampling process with geometric and semantic\nconstraints of 3D layouts. (2) To handle the unbounded nature of urban scenes,\nwe represent 3D scene with a Scalable Hash Grid structure, incrementally\nadapting to the growing scale of urban scenes. Extensive experiments\nsubstantiate the capability of our framework to scale text-to-3D generation to\nlarge-scale urban scenes that cover over 1000m driving distance for the first\ntime. We also present various scene editing demonstrations, showing the powers\nof steerable urban scene generation. Website: https://urbanarchitect.github.io.\n", "date": "2024-04-10", "categories": []} {"id": "2404.06851", "title": "UDiFF: Generating Conditional Unsigned Distance Fields with Optimal\n Wavelet Diffusion", "abstract": " Diffusion models have shown remarkable results for image generation, editing\nand inpainting. Recent works explore diffusion models for 3D shape generation\nwith neural implicit functions, i.e., signed distance function and occupancy\nfunction. However, they are limited to shapes with closed surfaces, which\nprevents them from generating diverse 3D real-world contents containing open\nsurfaces. In this work, we present UDiFF, a 3D diffusion model for unsigned\ndistance fields (UDFs) which is capable to generate textured 3D shapes with\nopen surfaces from text conditions or unconditionally. Our key idea is to\ngenerate UDFs in spatial-frequency domain with an optimal wavelet\ntransformation, which produces a compact representation space for UDF\ngeneration. Specifically, instead of selecting an appropriate wavelet\ntransformation which requires expensive manual efforts and still leads to large\ninformation loss, we propose a data-driven approach to learn the optimal\nwavelet transformation for UDFs. We evaluate UDiFF to show our advantages by\nnumerical and visual comparisons with the latest methods on widely used\nbenchmarks. Page: https://weiqi-zhang.github.io/UDiFF.\n", "date": "2024-04-10", "categories": []} {"id": "2404.06091", "title": "Hash3D: Training-free Acceleration for 3D Generation", "abstract": " The evolution of 3D generative modeling has been notably propelled by the\nadoption of 2D diffusion models. Despite this progress, the cumbersome\noptimization process per se presents a critical hurdle to efficiency. In this\npaper, we introduce Hash3D, a universal acceleration for 3D generation without\nmodel training. Central to Hash3D is the insight that feature-map redundancy is\nprevalent in images rendered from camera positions and diffusion time-steps in\nclose proximity. By effectively hashing and reusing these feature maps across\nneighboring timesteps and camera angles, Hash3D substantially prevents\nredundant calculations, thus accelerating the diffusion model's inference in 3D\ngeneration tasks. We achieve this through an adaptive grid-based hashing.\nSurprisingly, this feature-sharing mechanism not only speed up the generation\nbut also enhances the smoothness and view consistency of the synthesized 3D\nobjects. Our experiments covering 5 text-to-3D and 3 image-to-3D models,\ndemonstrate Hash3D's versatility to speed up optimization, enhancing efficiency\nby 1.3 to 4 times. Additionally, Hash3D's integration with 3D Gaussian\nsplatting largely speeds up 3D model creation, reducing text-to-3D processing\nto about 10 minutes and image-to-3D conversion to roughly 30 seconds. The\nproject page is at https://adamdad.github.io/hash3D/.\n", "date": "2024-04-09", "categories": []} {"id": "2404.01843", "title": "Sketch3D: Style-Consistent Guidance for Sketch-to-3D Generation", "abstract": " Recently, image-to-3D approaches have achieved significant results with a\nnatural image as input. However, it is not always possible to access these\nenriched color input samples in practical applications, where only sketches are\navailable. Existing sketch-to-3D researches suffer from limitations in broad\napplications due to the challenges of lacking color information and multi-view\ncontent. To overcome them, this paper proposes a novel generation paradigm\nSketch3D to generate realistic 3D assets with shape aligned with the input\nsketch and color matching the textual description. Concretely, Sketch3D first\ninstantiates the given sketch in the reference image through the\nshape-preserving generation process. Second, the reference image is leveraged\nto deduce a coarse 3D Gaussian prior, and multi-view style-consistent guidance\nimages are generated based on the renderings of the 3D Gaussians. Finally,\nthree strategies are designed to optimize 3D Gaussians, i.e., structural\noptimization via a distribution transfer mechanism, color optimization with a\nstraightforward MSE loss and sketch similarity optimization with a CLIP-based\ngeometric similarity loss. Extensive visual comparisons and quantitative\nanalysis illustrate the advantage of our Sketch3D in generating realistic 3D\nassets while preserving consistency with the input.\n", "date": "2024-04-02", "categories": []} {"id": "2306.08103", "title": "Generating Images with 3D Annotations Using Diffusion Models", "abstract": " Diffusion models have emerged as a powerful generative method, capable of\nproducing stunning photo-realistic images from natural language descriptions.\nHowever, these models lack explicit control over the 3D structure in the\ngenerated images. Consequently, this hinders our ability to obtain detailed 3D\nannotations for the generated images or to craft instances with specific poses\nand distances. In this paper, we propose 3D Diffusion Style Transfer (3D-DST),\nwhich incorporates 3D geometry control into diffusion models. Our method\nexploits ControlNet, which extends diffusion models by using visual prompts in\naddition to text prompts. We generate images of the 3D objects taken from 3D\nshape repositories (e.g., ShapeNet and Objaverse), render them from a variety\nof poses and viewing directions, compute the edge maps of the rendered images,\nand use these edge maps as visual prompts to generate realistic images. With\nexplicit 3D geometry control, we can easily change the 3D structures of the\nobjects in the generated images and obtain ground-truth 3D annotations\nautomatically. This allows us to improve a wide range of vision tasks, e.g.,\nclassification and 3D pose estimation, in both in-distribution (ID) and\nout-of-distribution (OOD) settings. We demonstrate the effectiveness of our\nmethod through extensive experiments on ImageNet-100/200, ImageNet-R,\nPASCAL3D+, ObjectNet3D, and OOD-CV. The results show that our method\nsignificantly outperforms existing methods, e.g., 3.8 percentage points on\nImageNet-100 using DeiT-B.\n", "date": "2023-06-13", "categories": []} {"id": "2403.19773", "title": "ShapeFusion: A 3D diffusion model for localized shape editing", "abstract": " In the realm of 3D computer vision, parametric models have emerged as a\nground-breaking methodology for the creation of realistic and expressive 3D\navatars. Traditionally, they rely on Principal Component Analysis (PCA), given\nits ability to decompose data to an orthonormal space that maximally captures\nshape variations. However, due to the orthogonality constraints and the global\nnature of PCA's decomposition, these models struggle to perform localized and\ndisentangled editing of 3D shapes, which severely affects their use in\napplications requiring fine control such as face sculpting. In this paper, we\nleverage diffusion models to enable diverse and fully localized edits on 3D\nmeshes, while completely preserving the un-edited regions. We propose an\neffective diffusion masking training strategy that, by design, facilitates\nlocalized manipulation of any shape region, without being limited to predefined\nregions or to sparse sets of predefined control vertices. Following our\nframework, a user can explicitly set their manipulation region of choice and\ndefine an arbitrary set of vertices as handles to edit a 3D mesh. Compared to\nthe current state-of-the-art our method leads to more interpretable shape\nmanipulations than methods relying on latent code state, greater localization\nand generation diversity while offering faster inference than optimization\nbased approaches. Project page: https://rolpotamias.github.io/Shapefusion/\n", "date": "2024-03-28", "categories": []} {"id": "2404.03042", "title": "AWOL: Analysis WithOut synthesis using Language", "abstract": " Many classical parametric 3D shape models exist, but creating novel shapes\nwith such models requires expert knowledge of their parameters. For example,\nimagine creating a specific type of tree using procedural graphics or a new\nkind of animal from a statistical shape model. Our key idea is to leverage\nlanguage to control such existing models to produce novel shapes. This involves\nlearning a mapping between the latent space of a vision-language model and the\nparameter space of the 3D model, which we do using a small set of shape and\ntext pairs. Our hypothesis is that mapping from language to parameters allows\nus to generate parameters for objects that were never seen during training. If\nthe mapping between language and parameters is sufficiently smooth, then\ninterpolation or generalization in language should translate appropriately into\nnovel 3D shapes. We test our approach with two very different types of\nparametric shape models (quadrupeds and arboreal trees). We use a learned\nstatistical shape model of quadrupeds and show that we can use text to generate\nnew animals not present during training. In particular, we demonstrate\nstate-of-the-art shape estimation of 3D dogs. This work also constitutes the\nfirst language-driven method for generating 3D trees. Finally, embedding images\nin the CLIP latent space enables us to generate animals and trees directly from\nimages.\n", "date": "2024-04-03", "categories": []} {"id": "2404.03656", "title": "MVD-Fusion: Single-view 3D via Depth-consistent Multi-view Generation", "abstract": " We present MVD-Fusion: a method for single-view 3D inference via generative\nmodeling of multi-view-consistent RGB-D images. While recent methods pursuing\n3D inference advocate learning novel-view generative models, these generations\nare not 3D-consistent and require a distillation process to generate a 3D\noutput. We instead cast the task of 3D inference as directly generating\nmutually-consistent multiple views and build on the insight that additionally\ninferring depth can provide a mechanism for enforcing this consistency.\nSpecifically, we train a denoising diffusion model to generate multi-view RGB-D\nimages given a single RGB input image and leverage the (intermediate noisy)\ndepth estimates to obtain reprojection-based conditioning to maintain\nmulti-view consistency. We train our model using large-scale synthetic dataset\nObajverse as well as the real-world CO3D dataset comprising of generic camera\nviewpoints. We demonstrate that our approach can yield more accurate synthesis\ncompared to recent state-of-the-art, including distillation-based 3D inference\nand prior multi-view generation methods. We also evaluate the geometry induced\nby our multi-view depth prediction and find that it yields a more accurate\nrepresentation than other direct 3D inference approaches.\n", "date": "2024-04-04", "categories": []} {"id": "2403.00939", "title": "G3DR: Generative 3D Reconstruction in ImageNet", "abstract": " We introduce a novel 3D generative method, Generative 3D Reconstruction\n(G3DR) in ImageNet, capable of generating diverse and high-quality 3D objects\nfrom single images, addressing the limitations of existing methods. At the\nheart of our framework is a novel depth regularization technique that enables\nthe generation of scenes with high-geometric fidelity. G3DR also leverages a\npretrained language-vision model, such as CLIP, to enable reconstruction in\nnovel views and improve the visual realism of generations. Additionally, G3DR\ndesigns a simple but effective sampling procedure to further improve the\nquality of generations. G3DR offers diverse and efficient 3D asset generation\nbased on class or text conditioning. Despite its simplicity, G3DR is able to\nbeat state-of-theart methods, improving over them by up to 22% in perceptual\nmetrics and 90% in geometry scores, while needing only half of the training\ntime. Code is available at https://github.com/preddy5/G3DR\n", "date": "2024-03-01", "categories": []} {"id": "2404.02686", "title": "Design2Cloth: 3D Cloth Generation from 2D Masks", "abstract": " In recent years, there has been a significant shift in the field of digital\navatar research, towards modeling, animating and reconstructing clothed human\nrepresentations, as a key step towards creating realistic avatars. However,\ncurrent 3D cloth generation methods are garment specific or trained completely\non synthetic data, hence lacking fine details and realism. In this work, we\nmake a step towards automatic realistic garment design and propose\nDesign2Cloth, a high fidelity 3D generative model trained on a real world\ndataset from more than 2000 subject scans. To provide vital contribution to the\nfashion industry, we developed a user-friendly adversarial model capable of\ngenerating diverse and detailed clothes simply by drawing a 2D cloth mask.\nUnder a series of both qualitative and quantitative experiments, we showcase\nthat Design2Cloth outperforms current state-of-the-art cloth generative models\nby a large margin. In addition to the generative properties of our network, we\nshowcase that the proposed method can be used to achieve high quality\nreconstructions from single in-the-wild images and 3D scans. Dataset, code and\npre-trained model will become publicly available.\n", "date": "2024-04-03", "categories": []} {"id": "2309.16585", "title": "Text-to-3D using Gaussian Splatting", "abstract": " Automatic text-to-3D generation that combines Score Distillation Sampling\n(SDS) with the optimization of volume rendering has achieved remarkable\nprogress in synthesizing realistic 3D objects. Yet most existing text-to-3D\nmethods by SDS and volume rendering suffer from inaccurate geometry, e.g., the\nJanus issue, since it is hard to explicitly integrate 3D priors into implicit\n3D representations. Besides, it is usually time-consuming for them to generate\nelaborate 3D models with rich colors. In response, this paper proposes GSGEN, a\nnovel method that adopts Gaussian Splatting, a recent state-of-the-art\nrepresentation, to text-to-3D generation. GSGEN aims at generating high-quality\n3D objects and addressing existing shortcomings by exploiting the explicit\nnature of Gaussian Splatting that enables the incorporation of 3D prior.\nSpecifically, our method adopts a progressive optimization strategy, which\nincludes a geometry optimization stage and an appearance refinement stage. In\ngeometry optimization, a coarse representation is established under 3D point\ncloud diffusion prior along with the ordinary 2D SDS optimization, ensuring a\nsensible and 3D-consistent rough shape. Subsequently, the obtained Gaussians\nundergo an iterative appearance refinement to enrich texture details. In this\nstage, we increase the number of Gaussians by compactness-based densification\nto enhance continuity and improve fidelity. With these designs, our approach\ncan generate 3D assets with delicate details and accurate geometry. Extensive\nevaluations demonstrate the effectiveness of our method, especially for\ncapturing high-frequency components. Our code is available at\nhttps://github.com/gsgen3d/gsgen\n", "date": "2023-09-28", "categories": []} {"id": "2311.13831", "title": "Posterior Distillation Sampling", "abstract": " We introduce Posterior Distillation Sampling (PDS), a novel optimization\nmethod for parametric image editing based on diffusion models. Existing\noptimization-based methods, which leverage the powerful 2D prior of diffusion\nmodels to handle various parametric images, have mainly focused on generation.\nUnlike generation, editing requires a balance between conforming to the target\nattribute and preserving the identity of the source content. Recent 2D image\nediting methods have achieved this balance by leveraging the stochastic latent\nencoded in the generative process of diffusion models. To extend the editing\ncapabilities of diffusion models shown in pixel space to parameter space, we\nreformulate the 2D image editing method into an optimization form named PDS.\nPDS matches the stochastic latents of the source and the target, enabling the\nsampling of targets in diverse parameter spaces that align with a desired\nattribute while maintaining the source's identity. We demonstrate that this\noptimization resembles running a generative process with the target attribute,\nbut aligning this process with the trajectory of the source's generative\nprocess. Extensive editing results in Neural Radiance Fields and Scalable\nVector Graphics representations demonstrate that PDS is capable of sampling\ntargets to fulfill the aforementioned balance across various parameter spaces.\n", "date": "2023-11-23", "categories": []} {"id": "2312.06725", "title": "EpiDiff: Enhancing Multi-View Synthesis via Localized\n Epipolar-Constrained Diffusion", "abstract": " Generating multiview images from a single view facilitates the rapid\ngeneration of a 3D mesh conditioned on a single image. Recent methods that\nintroduce 3D global representation into diffusion models have shown the\npotential to generate consistent multiviews, but they have reduced generation\nspeed and face challenges in maintaining generalizability and quality. To\naddress this issue, we propose EpiDiff, a localized interactive multiview\ndiffusion model. At the core of the proposed approach is to insert a\nlightweight epipolar attention block into the frozen diffusion model,\nleveraging epipolar constraints to enable cross-view interaction among feature\nmaps of neighboring views. The newly initialized 3D modeling module preserves\nthe original feature distribution of the diffusion model, exhibiting\ncompatibility with a variety of base diffusion models. Experiments show that\nEpiDiff generates 16 multiview images in just 12 seconds, and it surpasses\nprevious methods in quality evaluation metrics, including PSNR, SSIM and LPIPS.\nAdditionally, EpiDiff can generate a more diverse distribution of views,\nimproving the reconstruction quality from generated multiviews. Please see our\nproject page at https://huanngzh.github.io/EpiDiff/.\n", "date": "2023-12-11", "categories": []} {"id": "2401.04728", "title": "Morphable Diffusion: 3D-Consistent Diffusion for Single-image Avatar\n Creation", "abstract": " Recent advances in generative diffusion models have enabled the previously\nunfeasible capability of generating 3D assets from a single input image or a\ntext prompt. In this work, we aim to enhance the quality and functionality of\nthese models for the task of creating controllable, photorealistic human\navatars. We achieve this by integrating a 3D morphable model into the\nstate-of-the-art multi-view-consistent diffusion approach. We demonstrate that\naccurate conditioning of a generative pipeline on the articulated 3D model\nenhances the baseline model performance on the task of novel view synthesis\nfrom a single image. More importantly, this integration facilitates a seamless\nand accurate incorporation of facial expression and body pose control into the\ngeneration process. To the best of our knowledge, our proposed framework is the\nfirst diffusion model to enable the creation of fully 3D-consistent,\nanimatable, and photorealistic human avatars from a single image of an unseen\nsubject; extensive quantitative and qualitative evaluations demonstrate the\nadvantages of our approach over existing state-of-the-art avatar creation\nmodels on both novel view and novel expression synthesis tasks. The code for\nour project is publicly available.\n", "date": "2024-01-09", "categories": []} {"id": "2402.10636", "title": "PEGASUS: Personalized Generative 3D Avatars with Composable Attributes", "abstract": " We present PEGASUS, a method for constructing a personalized generative 3D\nface avatar from monocular video sources. Our generative 3D avatar enables\ndisentangled controls to selectively alter the facial attributes (e.g., hair or\nnose) while preserving the identity. Our approach consists of two stages:\nsynthetic database generation and constructing a personalized generative\navatar. We generate a synthetic video collection of the target identity with\nvarying facial attributes, where the videos are synthesized by borrowing the\nattributes from monocular videos of diverse identities. Then, we build a\nperson-specific generative 3D avatar that can modify its attributes\ncontinuously while preserving its identity. Through extensive experiments, we\ndemonstrate that our method of generating a synthetic database and creating a\n3D generative avatar is the most effective in preserving identity while\nachieving high realism. Subsequently, we introduce a zero-shot approach to\nachieve the same goal of generative modeling more efficiently by leveraging a\npreviously constructed personalized generative model.\n", "date": "2024-02-16", "categories": []} {"id": "2401.10786", "title": "Sat2Scene: 3D Urban Scene Generation from Satellite Images with\n Diffusion", "abstract": " Directly generating scenes from satellite imagery offers exciting\npossibilities for integration into applications like games and map services.\nHowever, challenges arise from significant view changes and scene scale.\nPrevious efforts mainly focused on image or video generation, lacking\nexploration into the adaptability of scene generation for arbitrary views.\nExisting 3D generation works either operate at the object level or are\ndifficult to utilize the geometry obtained from satellite imagery. To overcome\nthese limitations, we propose a novel architecture for direct 3D scene\ngeneration by introducing diffusion models into 3D sparse representations and\ncombining them with neural rendering techniques. Specifically, our approach\ngenerates texture colors at the point level for a given geometry using a 3D\ndiffusion model first, which is then transformed into a scene representation in\na feed-forward manner. The representation can be utilized to render arbitrary\nviews which would excel in both single-frame quality and inter-frame\nconsistency. Experiments in two city-scale datasets show that our model\ndemonstrates proficiency in generating photo-realistic street-view image\nsequences and cross-view urban scenes from satellite imagery.\n", "date": "2024-01-19", "categories": []} {"id": "2404.01296", "title": "MagicMirror: Fast and High-Quality Avatar Generation with a Constrained\n Search Space", "abstract": " We introduce a novel framework for 3D human avatar generation and\npersonalization, leveraging text prompts to enhance user engagement and\ncustomization. Central to our approach are key innovations aimed at overcoming\nthe challenges in photo-realistic avatar synthesis. Firstly, we utilize a\nconditional Neural Radiance Fields (NeRF) model, trained on a large-scale\nunannotated multi-view dataset, to create a versatile initial solution space\nthat accelerates and diversifies avatar generation. Secondly, we develop a\ngeometric prior, leveraging the capabilities of Text-to-Image Diffusion Models,\nto ensure superior view invariance and enable direct optimization of avatar\ngeometry. These foundational ideas are complemented by our optimization\npipeline built on Variational Score Distillation (VSD), which mitigates texture\nloss and over-saturation issues. As supported by our extensive experiments,\nthese strategies collectively enable the creation of custom avatars with\nunparalleled visual quality and better adherence to input text prompts. You can\nfind more results and videos in our website:\nhttps://syntec-research.github.io/MagicMirror\n", "date": "2024-04-01", "categories": []} {"id": "2309.16653", "title": "DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content\n Creation", "abstract": " Recent advances in 3D content creation mostly leverage optimization-based 3D\ngeneration via score distillation sampling (SDS). Though promising results have\nbeen exhibited, these methods often suffer from slow per-sample optimization,\nlimiting their practical usage. In this paper, we propose DreamGaussian, a\nnovel 3D content generation framework that achieves both efficiency and quality\nsimultaneously. Our key insight is to design a generative 3D Gaussian Splatting\nmodel with companioned mesh extraction and texture refinement in UV space. In\ncontrast to the occupancy pruning used in Neural Radiance Fields, we\ndemonstrate that the progressive densification of 3D Gaussians converges\nsignificantly faster for 3D generative tasks. To further enhance the texture\nquality and facilitate downstream applications, we introduce an efficient\nalgorithm to convert 3D Gaussians into textured meshes and apply a fine-tuning\nstage to refine the details. Extensive experiments demonstrate the superior\nefficiency and competitive generation quality of our proposed approach.\nNotably, DreamGaussian produces high-quality textured meshes in just 2 minutes\nfrom a single-view image, achieving approximately 10 times acceleration\ncompared to existing methods.\n", "date": "2023-09-28", "categories": []} {"id": "2312.11461", "title": "GAvatar: Animatable 3D Gaussian Avatars with Implicit Mesh Learning", "abstract": " Gaussian splatting has emerged as a powerful 3D representation that harnesses\nthe advantages of both explicit (mesh) and implicit (NeRF) 3D representations.\nIn this paper, we seek to leverage Gaussian splatting to generate realistic\nanimatable avatars from textual descriptions, addressing the limitations (e.g.,\nflexibility and efficiency) imposed by mesh or NeRF-based representations.\nHowever, a naive application of Gaussian splatting cannot generate high-quality\nanimatable avatars and suffers from learning instability; it also cannot\ncapture fine avatar geometries and often leads to degenerate body parts. To\ntackle these problems, we first propose a primitive-based 3D Gaussian\nrepresentation where Gaussians are defined inside pose-driven primitives to\nfacilitate animation. Second, to stabilize and amortize the learning of\nmillions of Gaussians, we propose to use neural implicit fields to predict the\nGaussian attributes (e.g., colors). Finally, to capture fine avatar geometries\nand extract detailed meshes, we propose a novel SDF-based implicit mesh\nlearning approach for 3D Gaussians that regularizes the underlying geometries\nand extracts highly detailed textured meshes. Our proposed method, GAvatar,\nenables the large-scale generation of diverse animatable avatars using only\ntext prompts. GAvatar significantly surpasses existing methods in terms of both\nappearance and geometry quality, and achieves extremely fast rendering (100\nfps) at 1K resolution.\n", "date": "2023-12-18", "categories": []} {"id": "2312.03795", "title": "AnimatableDreamer: Text-Guided Non-rigid 3D Model Generation and\n Reconstruction with Canonical Score Distillation", "abstract": " Advances in 3D generation have facilitated sequential 3D model generation\n(a.k.a 4D generation), yet its application for animatable objects with large\nmotion remains scarce. Our work proposes AnimatableDreamer, a text-to-4D\ngeneration framework capable of generating diverse categories of non-rigid\nobjects on skeletons extracted from a monocular video. At its core,\nAnimatableDreamer is equipped with our novel optimization design dubbed\nCanonical Score Distillation (CSD), which lifts 2D diffusion for temporal\nconsistent 4D generation. CSD, designed from a score gradient perspective,\ngenerates a canonical model with warp-robustness across different\narticulations. Notably, it also enhances the authenticity of bones and skinning\nby integrating inductive priors from a diffusion model. Furthermore, with\nmulti-view distillation, CSD infers invisible regions, thereby improving the\nfidelity of monocular non-rigid reconstruction. Extensive experiments\ndemonstrate the capability of our method in generating high-flexibility\ntext-guided 3D models from the monocular video, while also showing improved\nreconstruction performance over existing non-rigid reconstruction methods.\n", "date": "2023-12-06", "categories": []} {"id": "2401.01647", "title": "SIGNeRF: Scene Integrated Generation for Neural Radiance Fields", "abstract": " Advances in image diffusion models have recently led to notable improvements\nin the generation of high-quality images. In combination with Neural Radiance\nFields (NeRFs), they enabled new opportunities in 3D generation. However, most\ngenerative 3D approaches are object-centric and applying them to editing\nexisting photorealistic scenes is not trivial. We propose SIGNeRF, a novel\napproach for fast and controllable NeRF scene editing and scene-integrated\nobject generation. A new generative update strategy ensures 3D consistency\nacross the edited images, without requiring iterative optimization. We find\nthat depth-conditioned diffusion models inherently possess the capability to\ngenerate 3D consistent views by requesting a grid of images instead of single\nviews. Based on these insights, we introduce a multi-view reference sheet of\nmodified images. Our method updates an image collection consistently based on\nthe reference sheet and refines the original NeRF with the newly generated\nimage set in one go. By exploiting the depth conditioning mechanism of the\nimage diffusion model, we gain fine control over the spatial location of the\nedit and enforce shape guidance by a selected region or an external mesh.\n", "date": "2024-01-03", "categories": []} {"id": "2312.03611", "title": "DreamComposer: Controllable 3D Object Generation via Multi-View\n Conditions", "abstract": " Utilizing pre-trained 2D large-scale generative models, recent works are\ncapable of generating high-quality novel views from a single in-the-wild image.\nHowever, due to the lack of information from multiple views, these works\nencounter difficulties in generating controllable novel views. In this paper,\nwe present DreamComposer, a flexible and scalable framework that can enhance\nexisting view-aware diffusion models by injecting multi-view conditions.\nSpecifically, DreamComposer first uses a view-aware 3D lifting module to obtain\n3D representations of an object from multiple views. Then, it renders the\nlatent features of the target view from 3D representations with the multi-view\nfeature fusion module. Finally the target view features extracted from\nmulti-view inputs are injected into a pre-trained diffusion model. Experiments\nshow that DreamComposer is compatible with state-of-the-art diffusion models\nfor zero-shot novel view synthesis, further enhancing them to generate\nhigh-fidelity novel view images with multi-view conditions, ready for\ncontrollable 3D object reconstruction and various other applications.\n", "date": "2023-12-06", "categories": []} {"id": "2402.12225", "title": "Pushing Auto-regressive Models for 3D Shape Generation at Capacity and\n Scalability", "abstract": " Auto-regressive models have achieved impressive results in 2D image\ngeneration by modeling joint distributions in grid space. In this paper, we\nextend auto-regressive models to 3D domains, and seek a stronger ability of 3D\nshape generation by improving auto-regressive models at capacity and\nscalability simultaneously. Firstly, we leverage an ensemble of publicly\navailable 3D datasets to facilitate the training of large-scale models. It\nconsists of a comprehensive collection of approximately 900,000 objects, with\nmultiple properties of meshes, points, voxels, rendered images, and text\ncaptions. This diverse labeled dataset, termed Objaverse-Mix, empowers our\nmodel to learn from a wide range of object variations. However, directly\napplying 3D auto-regression encounters critical challenges of high\ncomputational demands on volumetric grids and ambiguous auto-regressive order\nalong grid dimensions, resulting in inferior quality of 3D shapes. To this end,\nwe then present a novel framework Argus3D in terms of capacity. Concretely, our\napproach introduces discrete representation learning based on a latent vector\ninstead of volumetric grids, which not only reduces computational costs but\nalso preserves essential geometric details by learning the joint distributions\nin a more tractable order. The capacity of conditional generation can thus be\nrealized by simply concatenating various conditioning inputs to the latent\nvector, such as point clouds, categories, images, and texts. In addition,\nthanks to the simplicity of our model architecture, we naturally scale up our\napproach to a larger model with an impressive 3.6 billion parameters, further\nenhancing the quality of versatile 3D generation. Extensive experiments on four\ngeneration tasks demonstrate that Argus3D can synthesize diverse and faithful\nshapes across multiple categories, achieving remarkable performance.\n", "date": "2024-02-19", "categories": []} {"id": "2403.17213", "title": "AnimateMe: 4D Facial Expressions via Diffusion Models", "abstract": " The field of photorealistic 3D avatar reconstruction and generation has\ngarnered significant attention in recent years; however, animating such avatars\nremains challenging. Recent advances in diffusion models have notably enhanced\nthe capabilities of generative models in 2D animation. In this work, we\ndirectly utilize these models within the 3D domain to achieve controllable and\nhigh-fidelity 4D facial animation. By integrating the strengths of diffusion\nprocesses and geometric deep learning, we employ Graph Neural Networks (GNNs)\nas denoising diffusion models in a novel approach, formulating the diffusion\nprocess directly on the mesh space and enabling the generation of 3D facial\nexpressions. This facilitates the generation of facial deformations through a\nmesh-diffusion-based model. Additionally, to ensure temporal coherence in our\nanimations, we propose a consistent noise sampling method. Under a series of\nboth quantitative and qualitative experiments, we showcase that the proposed\nmethod outperforms prior work in 4D expression synthesis by generating\nhigh-fidelity extreme expressions. Furthermore, we applied our method to\ntextured 4D facial expression generation, implementing a straightforward\nextension that involves training on a large-scale textured 4D facial expression\ndatabase.\n", "date": "2024-03-25", "categories": []} {"id": "2403.17237", "title": "DreamPolisher: Towards High-Quality Text-to-3D Generation via Geometric\n Diffusion", "abstract": " We present DreamPolisher, a novel Gaussian Splatting based method with\ngeometric guidance, tailored to learn cross-view consistency and intricate\ndetail from textual descriptions. While recent progress on text-to-3D\ngeneration methods have been promising, prevailing methods often fail to ensure\nview-consistency and textural richness. This problem becomes particularly\nnoticeable for methods that work with text input alone. To address this, we\npropose a two-stage Gaussian Splatting based approach that enforces geometric\nconsistency among views. Initially, a coarse 3D generation undergoes refinement\nvia geometric optimization. Subsequently, we use a ControlNet driven refiner\ncoupled with the geometric consistency term to improve both texture fidelity\nand overall consistency of the generated 3D asset. Empirical evaluations across\ndiverse textual prompts spanning various object categories demonstrate the\nefficacy of DreamPolisher in generating consistent and realistic 3D objects,\naligning closely with the semantics of the textual instructions.\n", "date": "2024-03-25", "categories": []} {"id": "2403.17782", "title": "GenesisTex: Adapting Image Denoising Diffusion to Texture Space", "abstract": " We present GenesisTex, a novel method for synthesizing textures for 3D\ngeometries from text descriptions. GenesisTex adapts the pretrained image\ndiffusion model to texture space by texture space sampling. Specifically, we\nmaintain a latent texture map for each viewpoint, which is updated with\npredicted noise on the rendering of the corresponding viewpoint. The sampled\nlatent texture maps are then decoded into a final texture map. During the\nsampling process, we focus on both global and local consistency across multiple\nviewpoints: global consistency is achieved through the integration of style\nconsistency mechanisms within the noise prediction network, and low-level\nconsistency is achieved by dynamically aligning latent textures. Finally, we\napply reference-based inpainting and img2img on denser views for texture\nrefinement. Our approach overcomes the limitations of slow optimization in\ndistillation-based methods and instability in inpainting-based methods.\nExperiments on meshes from various sources demonstrate that our method\nsurpasses the baseline methods quantitatively and qualitatively.\n", "date": "2024-03-26", "categories": []} {"id": "2310.14729", "title": "MAS: Multi-view Ancestral Sampling for 3D motion generation using 2D\n diffusion", "abstract": " We introduce Multi-view Ancestral Sampling (MAS), a method for 3D motion\ngeneration, using 2D diffusion models that were trained on motions obtained\nfrom in-the-wild videos. As such, MAS opens opportunities to exciting and\ndiverse fields of motion previously under-explored as 3D data is scarce and\nhard to collect. MAS works by simultaneously denoising multiple 2D motion\nsequences representing different views of the same 3D motion. It ensures\nconsistency across all views at each diffusion step by combining the individual\ngenerations into a unified 3D sequence, and projecting it back to the original\nviews. We demonstrate MAS on 2D pose data acquired from videos depicting\nprofessional basketball maneuvers, rhythmic gymnastic performances featuring a\nball apparatus, and horse races. In each of these domains, 3D motion capture is\narduous, and yet, MAS generates diverse and realistic 3D sequences. Unlike the\nScore Distillation approach, which optimizes each sample by repeatedly applying\nsmall fixes, our method uses a sampling process that was constructed for the\ndiffusion framework. As we demonstrate, MAS avoids common issues such as\nout-of-domain sampling and mode-collapse. https://guytevet.github.io/mas-page/\n", "date": "2023-10-23", "categories": []} {"id": "2312.04784", "title": "Reality's Canvas, Language's Brush: Crafting 3D Avatars from Monocular\n Video", "abstract": " Recent advancements in 3D avatar generation excel with multi-view supervision\nfor photorealistic models. However, monocular counterparts lag in quality\ndespite broader applicability. We propose ReCaLaB to close this gap. ReCaLaB is\na fully-differentiable pipeline that learns high-fidelity 3D human avatars from\njust a single RGB video. A pose-conditioned deformable NeRF is optimized to\nvolumetrically represent a human subject in canonical T-pose. The canonical\nrepresentation is then leveraged to efficiently associate neural textures using\n2D-3D correspondences. This enables the separation of diffused color generation\nand lighting correction branches that jointly compose an RGB prediction. The\ndesign allows to control intermediate results for human pose, body shape,\ntexture, and lighting with text prompts. An image-conditioned diffusion model\nthereby helps to animate appearance and pose of the 3D avatar to create video\nsequences with previously unseen human motion. Extensive experiments show that\nReCaLaB outperforms previous monocular approaches in terms of image quality for\nimage synthesis tasks. Moreover, natural language offers an intuitive user\ninterface for creative manipulation of 3D human avatars.\n", "date": "2023-12-08", "categories": []} {"id": "2403.15441", "title": "Unified Generative Modeling of 3D Molecules via Bayesian Flow Networks", "abstract": " Advanced generative model (e.g., diffusion model) derived from simplified\ncontinuity assumptions of data distribution, though showing promising progress,\nhas been difficult to apply directly to geometry generation applications due to\nthe multi-modality and noise-sensitive nature of molecule geometry. This work\nintroduces Geometric Bayesian Flow Networks (GeoBFN), which naturally fits\nmolecule geometry by modeling diverse modalities in the differentiable\nparameter space of distributions. GeoBFN maintains the SE-(3) invariant density\nmodeling property by incorporating equivariant inter-dependency modeling on\nparameters of distributions and unifying the probabilistic modeling of\ndifferent modalities. Through optimized training and sampling techniques, we\ndemonstrate that GeoBFN achieves state-of-the-art performance on multiple 3D\nmolecule generation benchmarks in terms of generation quality (90.87% molecule\nstability in QM9 and 85.6% atom stability in GEOM-DRUG. GeoBFN can also conduct\nsampling with any number of steps to reach an optimal trade-off between\nefficiency and quality (e.g., 20-times speedup without sacrificing\nperformance).\n", "date": "2024-03-17", "categories": []} {"id": "2403.17001", "title": "VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation", "abstract": " Recent innovations on text-to-3D generation have featured Score Distillation\nSampling (SDS), which enables the zero-shot learning of implicit 3D models\n(NeRF) by directly distilling prior knowledge from 2D diffusion models.\nHowever, current SDS-based models still struggle with intricate text prompts\nand commonly result in distorted 3D models with unrealistic textures or\ncross-view inconsistency issues. In this work, we introduce a novel Visual\nPrompt-guided text-to-3D diffusion model (VP3D) that explicitly unleashes the\nvisual appearance knowledge in 2D visual prompt to boost text-to-3D generation.\nInstead of solely supervising SDS with text prompt, VP3D first capitalizes on\n2D diffusion model to generate a high-quality image from input text, which\nsubsequently acts as visual prompt to strengthen SDS optimization with explicit\nvisual appearance. Meanwhile, we couple the SDS optimization with additional\ndifferentiable reward function that encourages rendering images of 3D models to\nbetter visually align with 2D visual prompt and semantically match with text\nprompt. Through extensive experiments, we show that the 2D Visual Prompt in our\nVP3D significantly eases the learning of visual appearance of 3D models and\nthus leads to higher visual fidelity with more detailed textures. It is also\nappealing in view that when replacing the self-generating visual prompt with a\ngiven reference image, VP3D is able to trigger a new task of stylized\ntext-to-3D generation. Our project page is available at\nhttps://vp3d-cvpr24.github.io.\n", "date": "2024-03-25", "categories": []} {"id": "2403.14966", "title": "DreamFlow: High-Quality Text-to-3D Generation by Approximating\n Probability Flow", "abstract": " Recent progress in text-to-3D generation has been achieved through the\nutilization of score distillation methods: they make use of the pre-trained\ntext-to-image (T2I) diffusion models by distilling via the diffusion model\ntraining objective. However, such an approach inevitably results in the use of\nrandom timesteps at each update, which increases the variance of the gradient\nand ultimately prolongs the optimization process. In this paper, we propose to\nenhance the text-to-3D optimization by leveraging the T2I diffusion prior in\nthe generative sampling process with a predetermined timestep schedule. To this\nend, we interpret text-to3D optimization as a multi-view image-to-image\ntranslation problem, and propose a solution by approximating the probability\nflow. By leveraging the proposed novel optimization algorithm, we design\nDreamFlow, a practical three-stage coarseto-fine text-to-3D optimization\nframework that enables fast generation of highquality and high-resolution\n(i.e., 1024x1024) 3D contents. For example, we demonstrate that DreamFlow is 5\ntimes faster than the existing state-of-the-art text-to-3D method, while\nproducing more photorealistic 3D contents. Visit our project page\n(https://kyungmnlee.github.io/dreamflow.github.io/) for visualizations.\n", "date": "2024-03-22", "categories": []} {"id": "2403.15385", "title": "LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesis", "abstract": " Recent text-to-3D generation approaches produce impressive 3D results but\nrequire time-consuming optimization that can take up to an hour per prompt.\nAmortized methods like ATT3D optimize multiple prompts simultaneously to\nimprove efficiency, enabling fast text-to-3D synthesis. However, they cannot\ncapture high-frequency geometry and texture details and struggle to scale to\nlarge prompt sets, so they generalize poorly. We introduce LATTE3D, addressing\nthese limitations to achieve fast, high-quality generation on a significantly\nlarger prompt set. Key to our method is 1) building a scalable architecture and\n2) leveraging 3D data during optimization through 3D-aware diffusion priors,\nshape regularization, and model initialization to achieve robustness to diverse\nand complex training prompts. LATTE3D amortizes both neural field and textured\nsurface generation to produce highly detailed textured meshes in a single\nforward pass. LATTE3D generates 3D objects in 400ms, and can be further\nenhanced with fast test-time optimization.\n", "date": "2024-03-22", "categories": []} {"id": "2308.13223", "title": "EfficientDreamer: High-Fidelity and Robust 3D Creation via\n Orthogonal-view Diffusion Prior", "abstract": " While image diffusion models have made significant progress in text-driven 3D\ncontent creation, they often fail to accurately capture the intended meaning of\ntext prompts, especially for view information. This limitation leads to the\nJanus problem, where multi-faced 3D models are generated under the guidance of\nsuch diffusion models. In this paper, we propose a robust high-quality 3D\ncontent generation pipeline by exploiting orthogonal-view image guidance.\nFirst, we introduce a novel 2D diffusion model that generates an image\nconsisting of four orthogonal-view sub-images based on the given text prompt.\nThen, the 3D content is created using this diffusion model. Notably, the\ngenerated orthogonal-view image provides strong geometric structure priors and\nthus improves 3D consistency. As a result, it effectively resolves the Janus\nproblem and significantly enhances the quality of 3D content creation.\nAdditionally, we present a 3D synthesis fusion network that can further improve\nthe details of the generated 3D contents. Both quantitative and qualitative\nevaluations demonstrate that our method surpasses previous text-to-3D\ntechniques. Project page: https://efficientdreamer.github.io.\n", "date": "2023-08-25", "categories": []} {"id": "2310.02712", "title": "ED-NeRF: Efficient Text-Guided Editing of 3D Scene with Latent Space\n NeRF", "abstract": " Recently, there has been a significant advancement in text-to-image diffusion\nmodels, leading to groundbreaking performance in 2D image generation. These\nadvancements have been extended to 3D models, enabling the generation of novel\n3D objects from textual descriptions. This has evolved into NeRF editing\nmethods, which allow the manipulation of existing 3D objects through textual\nconditioning. However, existing NeRF editing techniques have faced limitations\nin their performance due to slow training speeds and the use of loss functions\nthat do not adequately consider editing. To address this, here we present a\nnovel 3D NeRF editing approach dubbed ED-NeRF by successfully embedding\nreal-world scenes into the latent space of the latent diffusion model (LDM)\nthrough a unique refinement layer. This approach enables us to obtain a NeRF\nbackbone that is not only faster but also more amenable to editing compared to\ntraditional image space NeRF editing. Furthermore, we propose an improved loss\nfunction tailored for editing by migrating the delta denoising score (DDS)\ndistillation loss, originally used in 2D image editing to the three-dimensional\ndomain. This novel loss function surpasses the well-known score distillation\nsampling (SDS) loss in terms of suitability for editing purposes. Our\nexperimental results demonstrate that ED-NeRF achieves faster editing speed\nwhile producing improved output quality compared to state-of-the-art 3D editing\nmodels.\n", "date": "2023-10-04", "categories": []} {"id": "2311.15980", "title": "Direct2.5: Diverse Text-to-3D Generation via Multi-view 2.5D Diffusion", "abstract": " Recent advances in generative AI have unveiled significant potential for the\ncreation of 3D content. However, current methods either apply a pre-trained 2D\ndiffusion model with the time-consuming score distillation sampling (SDS), or a\ndirect 3D diffusion model trained on limited 3D data losing generation\ndiversity. In this work, we approach the problem by employing a multi-view 2.5D\ndiffusion fine-tuned from a pre-trained 2D diffusion model. The multi-view 2.5D\ndiffusion directly models the structural distribution of 3D data, while still\nmaintaining the strong generalization ability of the original 2D diffusion\nmodel, filling the gap between 2D diffusion-based and direct 3D diffusion-based\nmethods for 3D content generation. During inference, multi-view normal maps are\ngenerated using the 2.5D diffusion, and a novel differentiable rasterization\nscheme is introduced to fuse the almost consistent multi-view normal maps into\na consistent 3D model. We further design a normal-conditioned multi-view image\ngeneration module for fast appearance generation given the 3D geometry. Our\nmethod is a one-pass diffusion process and does not require any SDS\noptimization as post-processing. We demonstrate through extensive experiments\nthat, our direct 2.5D generation with the specially-designed fusion scheme can\nachieve diverse, mode-seeking-free, and high-fidelity 3D content generation in\nonly 10 seconds. Project page: https://nju-3dv.github.io/projects/direct25.\n", "date": "2023-11-27", "categories": []} {"id": "2403.14613", "title": "DreamReward: Text-to-3D Generation with Human Preference", "abstract": " 3D content creation from text prompts has shown remarkable success recently.\nHowever, current text-to-3D methods often generate 3D results that do not align\nwell with human preferences. In this paper, we present a comprehensive\nframework, coined DreamReward, to learn and improve text-to-3D models from\nhuman preference feedback. To begin with, we collect 25k expert comparisons\nbased on a systematic annotation pipeline including rating and ranking. Then,\nwe build Reward3D -- the first general-purpose text-to-3D human preference\nreward model to effectively encode human preferences. Building upon the 3D\nreward model, we finally perform theoretical analysis and present the Reward3D\nFeedback Learning (DreamFL), a direct tuning algorithm to optimize the\nmulti-view diffusion models with a redefined scorer. Grounded by theoretical\nproof and extensive experiment comparisons, our DreamReward successfully\ngenerates high-fidelity and 3D consistent results with significant boosts in\nprompt alignment with human intention. Our results demonstrate the great\npotential for learning from human feedback to improve text-to-3D models.\n", "date": "2024-03-21", "categories": []} {"id": "2303.12236", "title": "SALAD: Part-Level Latent Diffusion for 3D Shape Generation and\n Manipulation", "abstract": " We present a cascaded diffusion model based on a part-level implicit 3D\nrepresentation. Our model achieves state-of-the-art generation quality and also\nenables part-level shape editing and manipulation without any additional\ntraining in conditional setup. Diffusion models have demonstrated impressive\ncapabilities in data generation as well as zero-shot completion and editing via\na guided reverse process. Recent research on 3D diffusion models has focused on\nimproving their generation capabilities with various data representations,\nwhile the absence of structural information has limited their capability in\ncompletion and editing tasks. We thus propose our novel diffusion model using a\npart-level implicit representation. To effectively learn diffusion with\nhigh-dimensional embedding vectors of parts, we propose a cascaded framework,\nlearning diffusion first on a low-dimensional subspace encoding extrinsic\nparameters of parts and then on the other high-dimensional subspace encoding\nintrinsic attributes. In the experiments, we demonstrate the outperformance of\nour method compared with the previous ones both in generation and part-level\ncompletion and manipulation tasks.\n", "date": "2023-03-21", "categories": []} {"id": "2312.09570", "title": "CAGE: Controllable Articulation GEneration", "abstract": " We address the challenge of generating 3D articulated objects in a\ncontrollable fashion. Currently, modeling articulated 3D objects is either\nachieved through laborious manual authoring, or using methods from prior work\nthat are hard to scale and control directly. We leverage the interplay between\npart shape, connectivity, and motion using a denoising diffusion-based method\nwith attention modules designed to extract correlations between part\nattributes. Our method takes an object category label and a part connectivity\ngraph as input and generates an object's geometry and motion parameters. The\ngenerated objects conform to user-specified constraints on the object category,\npart shape, and part articulation. Our experiments show that our method\noutperforms the state-of-the-art in articulated object generation, producing\nmore realistic objects while conforming better to user constraints.\n Video Summary at: http://youtu.be/cH_rbKbyTpE\n", "date": "2023-12-15", "categories": []} {"id": "2401.00711", "title": "Text2Avatar: Text to 3D Human Avatar Generation with Codebook-Driven\n Body Controllable Attribute", "abstract": " Generating 3D human models directly from text helps reduce the cost and time\nof character modeling. However, achieving multi-attribute controllable and\nrealistic 3D human avatar generation is still challenging due to feature\ncoupling and the scarcity of realistic 3D human avatar datasets. To address\nthese issues, we propose Text2Avatar, which can generate realistic-style 3D\navatars based on the coupled text prompts. Text2Avatar leverages a discrete\ncodebook as an intermediate feature to establish a connection between text and\navatars, enabling the disentanglement of features. Furthermore, to alleviate\nthe scarcity of realistic style 3D human avatar data, we utilize a pre-trained\nunconditional 3D human avatar generation model to obtain a large amount of 3D\navatar pseudo data, which allows Text2Avatar to achieve realistic style\ngeneration. Experimental results demonstrate that our method can generate\nrealistic 3D avatars from coupled textual data, which is challenging for other\nexisting methods in this field.\n", "date": "2024-01-01", "categories": []} {"id": "2403.13289", "title": "Text-to-3D Shape Generation", "abstract": " Recent years have seen an explosion of work and interest in text-to-3D shape\ngeneration. Much of the progress is driven by advances in 3D representations,\nlarge-scale pretraining and representation learning for text and image data\nenabling generative AI models, and differentiable rendering. Computational\nsystems that can perform text-to-3D shape generation have captivated the\npopular imagination as they enable non-expert users to easily create 3D content\ndirectly from text. However, there are still many limitations and challenges\nremaining in this problem space. In this state-of-the-art report, we provide a\nsurvey of the underlying technology and methods enabling text-to-3D shape\ngeneration to summarize the background literature. We then derive a systematic\ncategorization of recent work on text-to-3D shape generation based on the type\nof supervision data required. Finally, we discuss limitations of the existing\ncategories of methods, and delineate promising directions for future work.\n", "date": "2024-03-20", "categories": []} {"id": "2403.13524", "title": "Compress3D: a Compressed Latent Space for 3D Generation from a Single\n Image", "abstract": " 3D generation has witnessed significant advancements, yet efficiently\nproducing high-quality 3D assets from a single image remains challenging. In\nthis paper, we present a triplane autoencoder, which encodes 3D models into a\ncompact triplane latent space to effectively compress both the 3D geometry and\ntexture information. Within the autoencoder framework, we introduce a 3D-aware\ncross-attention mechanism, which utilizes low-resolution latent representations\nto query features from a high-resolution 3D feature volume, thereby enhancing\nthe representation capacity of the latent space. Subsequently, we train a\ndiffusion model on this refined latent space. In contrast to solely relying on\nimage embedding for 3D generation, our proposed method advocates for the\nsimultaneous utilization of both image embedding and shape embedding as\nconditions. Specifically, the shape embedding is estimated via a diffusion\nprior model conditioned on the image embedding. Through comprehensive\nexperiments, we demonstrate that our method outperforms state-of-the-art\nalgorithms, achieving superior performance while requiring less training data\nand time. Our approach enables the generation of high-quality 3D assets in\nmerely 7 seconds on a single A100 GPU.\n", "date": "2024-03-20", "categories": []} {"id": "2402.01166", "title": "A Comprehensive Survey on 3D Content Generation", "abstract": " Recent years have witnessed remarkable advances in artificial intelligence\ngenerated content(AIGC), with diverse input modalities, e.g., text, image,\nvideo, audio and 3D. The 3D is the most close visual modality to real-world 3D\nenvironment and carries enormous knowledge. The 3D content generation shows\nboth academic and practical values while also presenting formidable technical\nchallenges. This review aims to consolidate developments within the burgeoning\ndomain of 3D content generation. Specifically, a new taxonomy is proposed that\ncategorizes existing approaches into three types: 3D native generative methods,\n2D prior-based 3D generative methods, and hybrid 3D generative methods. The\nsurvey covers approximately 60 papers spanning the major techniques. Besides,\nwe discuss limitations of current 3D content generation techniques, and point\nout open challenges as well as promising directions for future work.\nAccompanied with this survey, we have established a project website where the\nresources on 3D content generation research are provided. The project page is\navailable at https://github.com/hitcslj/Awesome-AIGC-3D.\n", "date": "2024-02-02", "categories": []} {"id": "2402.03908", "title": "EscherNet: A Generative Model for Scalable View Synthesis", "abstract": " We introduce EscherNet, a multi-view conditioned diffusion model for view\nsynthesis. EscherNet learns implicit and generative 3D representations coupled\nwith a specialised camera positional encoding, allowing precise and continuous\nrelative control of the camera transformation between an arbitrary number of\nreference and target views. EscherNet offers exceptional generality,\nflexibility, and scalability in view synthesis -- it can generate more than 100\nconsistent target views simultaneously on a single consumer-grade GPU, despite\nbeing trained with a fixed number of 3 reference views to 3 target views. As a\nresult, EscherNet not only addresses zero-shot novel view synthesis, but also\nnaturally unifies single- and multi-image 3D reconstruction, combining these\ndiverse tasks into a single, cohesive framework. Our extensive experiments\ndemonstrate that EscherNet achieves state-of-the-art performance in multiple\nbenchmarks, even when compared to methods specifically tailored for each\nindividual problem. This remarkable versatility opens up new directions for\ndesigning scalable neural architectures for 3D vision. Project page:\nhttps://kxhit.github.io/EscherNet.\n", "date": "2024-02-06", "categories": []} {"id": "2403.11697", "title": "Urban Scene Diffusion through Semantic Occupancy Map", "abstract": " Generating unbounded 3D scenes is crucial for large-scale scene understanding\nand simulation. Urban scenes, unlike natural landscapes, consist of various\ncomplex man-made objects and structures such as roads, traffic signs, vehicles,\nand buildings. To create a realistic and detailed urban scene, it is crucial to\naccurately represent the geometry and semantics of the underlying objects,\ngoing beyond their visual appearance. In this work, we propose UrbanDiffusion,\na 3D diffusion model that is conditioned on a Bird's-Eye View (BEV) map and\ngenerates an urban scene with geometry and semantics in the form of semantic\noccupancy map. Our model introduces a novel paradigm that learns the data\ndistribution of scene-level structures within a latent space and further\nenables the expansion of the synthesized scene into an arbitrary scale. After\ntraining on real-world driving datasets, our model can generate a wide range of\ndiverse urban scenes given the BEV maps from the held-out set and also\ngeneralize to the synthesized maps from a driving simulator. We further\ndemonstrate its application to scene image synthesis with a pretrained image\ngenerator as a prior.\n", "date": "2024-03-18", "categories": []} {"id": "2403.12032", "title": "Generic 3D Diffusion Adapter Using Controlled Multi-View Editing", "abstract": " Open-domain 3D object synthesis has been lagging behind image synthesis due\nto limited data and higher computational complexity. To bridge this gap, recent\nworks have investigated multi-view diffusion but often fall short in either 3D\nconsistency, visual quality, or efficiency. This paper proposes MVEdit, which\nfunctions as a 3D counterpart of SDEdit, employing ancestral sampling to\njointly denoise multi-view images and output high-quality textured meshes.\nBuilt on off-the-shelf 2D diffusion models, MVEdit achieves 3D consistency\nthrough a training-free 3D Adapter, which lifts the 2D views of the last\ntimestep into a coherent 3D representation, then conditions the 2D views of the\nnext timestep using rendered views, without uncompromising visual quality. With\nan inference time of only 2-5 minutes, this framework achieves better trade-off\nbetween quality and speed than score distillation. MVEdit is highly versatile\nand extendable, with a wide range of applications including text/image-to-3D\ngeneration, 3D-to-3D editing, and high-quality texture synthesis. In\nparticular, evaluations demonstrate state-of-the-art performance in both\nimage-to-3D and text-guided texture generation tasks. Additionally, we\nintroduce a method for fine-tuning 2D latent diffusion models on small 3D\ndatasets with limited resources, enabling fast low-resolution text-to-3D\ninitialization.\n", "date": "2024-03-18", "categories": []} {"id": "2403.12409", "title": "ComboVerse: Compositional 3D Assets Creation Using Spatially-Aware\n Diffusion Guidance", "abstract": " Generating high-quality 3D assets from a given image is highly desirable in\nvarious applications such as AR/VR. Recent advances in single-image 3D\ngeneration explore feed-forward models that learn to infer the 3D model of an\nobject without optimization. Though promising results have been achieved in\nsingle object generation, these methods often struggle to model complex 3D\nassets that inherently contain multiple objects. In this work, we present\nComboVerse, a 3D generation framework that produces high-quality 3D assets with\ncomplex compositions by learning to combine multiple models. 1) We first\nperform an in-depth analysis of this ``multi-object gap'' from both model and\ndata perspectives. 2) Next, with reconstructed 3D models of different objects,\nwe seek to adjust their sizes, rotation angles, and locations to create a 3D\nasset that matches the given image. 3) To automate this process, we apply\nspatially-aware score distillation sampling (SSDS) from pretrained diffusion\nmodels to guide the positioning of objects. Our proposed framework emphasizes\nspatial alignment of objects, compared with standard score distillation\nsampling, and thus achieves more accurate results. Extensive experiments\nvalidate ComboVerse achieves clear improvements over existing methods in\ngenerating compositional 3D assets.\n", "date": "2024-03-19", "categories": []} {"id": "2403.12438", "title": "Precise-Physics Driven Text-to-3D Generation", "abstract": " Text-to-3D generation has shown great promise in generating novel 3D content\nbased on given text prompts. However, existing generative methods mostly focus\non geometric or visual plausibility while ignoring precise physics perception\nfor the generated 3D shapes. This greatly hinders the practicality of generated\n3D shapes in real-world applications. In this work, we propose Phy3DGen, a\nprecise-physics-driven text-to-3D generation method. By analyzing the solid\nmechanics of generated 3D shapes, we reveal that the 3D shapes generated by\nexisting text-to-3D generation methods are impractical for real-world\napplications as the generated 3D shapes do not conform to the laws of physics.\nTo this end, we leverage 3D diffusion models to provide 3D shape priors and\ndesign a data-driven differentiable physics layer to optimize 3D shape priors\nwith solid mechanics. This allows us to optimize geometry efficiently and learn\nprecise physics information about 3D shapes at the same time. Experimental\nresults demonstrate that our method can consider both geometric plausibility\nand precise physics perception, further bridging 3D virtual modeling and\nprecise physical worlds.\n", "date": "2024-03-19", "categories": []} {"id": "2403.12906", "title": "TexDreamer: Towards Zero-Shot High-Fidelity 3D Human Texture Generation", "abstract": " Texturing 3D humans with semantic UV maps remains a challenge due to the\ndifficulty of acquiring reasonably unfolded UV. Despite recent text-to-3D\nadvancements in supervising multi-view renderings using large text-to-image\n(T2I) models, issues persist with generation speed, text consistency, and\ntexture quality, resulting in data scarcity among existing datasets. We present\nTexDreamer, the first zero-shot multimodal high-fidelity 3D human texture\ngeneration model. Utilizing an efficient texture adaptation finetuning\nstrategy, we adapt large T2I model to a semantic UV structure while preserving\nits original generalization capability. Leveraging a novel feature translator\nmodule, the trained model is capable of generating high-fidelity 3D human\ntextures from either text or image within seconds. Furthermore, we introduce\nArTicuLated humAn textureS (ATLAS), the largest high-resolution (1024 X 1024)\n3D human texture dataset which contains 50k high-fidelity textures with text\ndescriptions.\n", "date": "2024-03-19", "categories": []} {"id": "2310.11784", "title": "Progressive3D: Progressively Local Editing for Text-to-3D Content\n Creation with Complex Semantic Prompts", "abstract": " Recent text-to-3D generation methods achieve impressive 3D content creation\ncapacity thanks to the advances in image diffusion models and optimizing\nstrategies. However, current methods struggle to generate correct 3D content\nfor a complex prompt in semantics, i.e., a prompt describing multiple\ninteracted objects binding with different attributes. In this work, we propose\na general framework named Progressive3D, which decomposes the entire generation\ninto a series of locally progressive editing steps to create precise 3D content\nfor complex prompts, and we constrain the content change to only occur in\nregions determined by user-defined region prompts in each editing step.\nFurthermore, we propose an overlapped semantic component suppression technique\nto encourage the optimization process to focus more on the semantic differences\nbetween prompts. Extensive experiments demonstrate that the proposed\nProgressive3D framework generates precise 3D content for prompts with complex\nsemantics and is general for various text-to-3D methods driven by different 3D\nrepresentations.\n", "date": "2023-10-18", "categories": []} {"id": "2403.07773", "title": "SemCity: Semantic Scene Generation with Triplane Diffusion", "abstract": " We present \"SemCity,\" a 3D diffusion model for semantic scene generation in\nreal-world outdoor environments. Most 3D diffusion models focus on generating a\nsingle object, synthetic indoor scenes, or synthetic outdoor scenes, while the\ngeneration of real-world outdoor scenes is rarely addressed. In this paper, we\nconcentrate on generating a real-outdoor scene through learning a diffusion\nmodel on a real-world outdoor dataset. In contrast to synthetic data,\nreal-outdoor datasets often contain more empty spaces due to sensor\nlimitations, causing challenges in learning real-outdoor distributions. To\naddress this issue, we exploit a triplane representation as a proxy form of\nscene distributions to be learned by our diffusion model. Furthermore, we\npropose a triplane manipulation that integrates seamlessly with our triplane\ndiffusion model. The manipulation improves our diffusion model's applicability\nin a variety of downstream tasks related to outdoor scene generation such as\nscene inpainting, scene outpainting, and semantic scene completion refinements.\nIn experimental results, we demonstrate that our triplane diffusion model shows\nmeaningful generation results compared with existing work in a real-outdoor\ndataset, SemanticKITTI. We also show our triplane manipulation facilitates\nseamlessly adding, removing, or modifying objects within a scene. Further, it\nalso enables the expansion of scenes toward a city-level scale. Finally, we\nevaluate our method on semantic scene completion refinements where our\ndiffusion model enhances predictions of semantic scene completion networks by\nlearning scene distribution. Our code is available at\nhttps://github.com/zoomin-lee/SemCity.\n", "date": "2024-03-12", "categories": []} {"id": "2403.11878", "title": "InTeX: Interactive Text-to-texture Synthesis via Unified Depth-aware\n Inpainting", "abstract": " Text-to-texture synthesis has become a new frontier in 3D content creation\nthanks to the recent advances in text-to-image models. Existing methods\nprimarily adopt a combination of pretrained depth-aware diffusion and\ninpainting models, yet they exhibit shortcomings such as 3D inconsistency and\nlimited controllability. To address these challenges, we introduce InteX, a\nnovel framework for interactive text-to-texture synthesis. 1) InteX includes a\nuser-friendly interface that facilitates interaction and control throughout the\nsynthesis process, enabling region-specific repainting and precise texture\nediting. 2) Additionally, we develop a unified depth-aware inpainting model\nthat integrates depth information with inpainting cues, effectively mitigating\n3D inconsistencies and improving generation speed. Through extensive\nexperiments, our framework has proven to be both practical and effective in\ntext-to-texture synthesis, paving the way for high-quality 3D content creation.\n", "date": "2024-03-18", "categories": []} {"id": "2403.11990", "title": "GetMesh: A Controllable Model for High-quality Mesh Generation and\n Manipulation", "abstract": " Mesh is a fundamental representation of 3D assets in various industrial\napplications, and is widely supported by professional softwares. However, due\nto its irregular structure, mesh creation and manipulation is often\ntime-consuming and labor-intensive. In this paper, we propose a highly\ncontrollable generative model, GetMesh, for mesh generation and manipulation\nacross different categories. By taking a varying number of points as the latent\nrepresentation, and re-organizing them as triplane representation, GetMesh\ngenerates meshes with rich and sharp details, outperforming both\nsingle-category and multi-category counterparts. Moreover, it also enables\nfine-grained control over the generation process that previous mesh generative\nmodels cannot achieve, where changing global/local mesh topologies,\nadding/removing mesh parts, and combining mesh parts across categories can be\nintuitively, efficiently, and robustly accomplished by adjusting the number,\npositions or features of latent points. Project page is\nhttps://getmesh.github.io.\n", "date": "2024-03-18", "categories": []} {"id": "2403.12008", "title": "SV3D: Novel Multi-view Synthesis and 3D Generation from a Single Image\n using Latent Video Diffusion", "abstract": " We present Stable Video 3D (SV3D) -- a latent video diffusion model for\nhigh-resolution, image-to-multi-view generation of orbital videos around a 3D\nobject. Recent work on 3D generation propose techniques to adapt 2D generative\nmodels for novel view synthesis (NVS) and 3D optimization. However, these\nmethods have several disadvantages due to either limited views or inconsistent\nNVS, thereby affecting the performance of 3D object generation. In this work,\nwe propose SV3D that adapts image-to-video diffusion model for novel multi-view\nsynthesis and 3D generation, thereby leveraging the generalization and\nmulti-view consistency of the video models, while further adding explicit\ncamera control for NVS. We also propose improved 3D optimization techniques to\nuse SV3D and its NVS outputs for image-to-3D generation. Extensive experimental\nresults on multiple datasets with 2D and 3D metrics as well as user study\ndemonstrate SV3D's state-of-the-art performance on NVS as well as 3D\nreconstruction compared to prior works.\n", "date": "2024-03-18", "categories": []} {"id": "2403.12010", "title": "VideoMV: Consistent Multi-View Generation Based on Large Video\n Generative Model", "abstract": " Generating multi-view images based on text or single-image prompts is a\ncritical capability for the creation of 3D content. Two fundamental questions\non this topic are what data we use for training and how to ensure multi-view\nconsistency. This paper introduces a novel framework that makes fundamental\ncontributions to both questions. Unlike leveraging images from 2D diffusion\nmodels for training, we propose a dense consistent multi-view generation model\nthat is fine-tuned from off-the-shelf video generative models. Images from\nvideo generative models are more suitable for multi-view generation because the\nunderlying network architecture that generates them employs a temporal module\nto enforce frame consistency. Moreover, the video data sets used to train these\nmodels are abundant and diverse, leading to a reduced train-finetuning domain\ngap. To enhance multi-view consistency, we introduce a 3D-Aware Denoising\nSampling, which first employs a feed-forward reconstruction module to get an\nexplicit global 3D model, and then adopts a sampling strategy that effectively\ninvolves images rendered from the global 3D model into the denoising sampling\nloop to improve the multi-view consistency of the final images. As a\nby-product, this module also provides a fast way to create 3D assets\nrepresented by 3D Gaussians within a few seconds. Our approach can generate 24\ndense views and converges much faster in training than state-of-the-art\napproaches (4 GPU hours versus many thousand GPU hours) with comparable visual\nquality and consistency. By further fine-tuning, our approach outperforms\nexisting state-of-the-art methods in both quantitative metrics and visual\neffects. Our project page is aigc3d.github.io/VideoMV.\n", "date": "2024-03-18", "categories": []} {"id": "2310.03020", "title": "Consistent-1-to-3: Consistent Image to 3D View Synthesis via\n Geometry-aware Diffusion Models", "abstract": " Zero-shot novel view synthesis (NVS) from a single image is an essential\nproblem in 3D object understanding. While recent approaches that leverage\npre-trained generative models can synthesize high-quality novel views from\nin-the-wild inputs, they still struggle to maintain 3D consistency across\ndifferent views. In this paper, we present Consistent-1-to-3, which is a\ngenerative framework that significantly mitigates this issue. Specifically, we\ndecompose the NVS task into two stages: (i) transforming observed regions to a\nnovel view, and (ii) hallucinating unseen regions. We design a scene\nrepresentation transformer and view-conditioned diffusion model for performing\nthese two stages respectively. Inside the models, to enforce 3D consistency, we\npropose to employ epipolor-guided attention to incorporate geometry\nconstraints, and multi-view attention to better aggregate multi-view\ninformation. Finally, we design a hierarchy generation paradigm to generate\nlong sequences of consistent views, allowing a full 360-degree observation of\nthe provided object image. Qualitative and quantitative evaluation over\nmultiple datasets demonstrates the effectiveness of the proposed mechanisms\nagainst state-of-the-art approaches. Our project page is at\nhttps://jianglongye.com/consistent123/\n", "date": "2023-10-04", "categories": []} {"id": "2403.09675", "title": "Open-Universe Indoor Scene Generation using LLM Program Synthesis and\n Uncurated Object Databases", "abstract": " We present a system for generating indoor scenes in response to text prompts.\nThe prompts are not limited to a fixed vocabulary of scene descriptions, and\nthe objects in generated scenes are not restricted to a fixed set of object\ncategories -- we call this setting indoor scene generation. Unlike most prior\nwork on indoor scene generation, our system does not require a large training\ndataset of existing 3D scenes. Instead, it leverages the world knowledge\nencoded in pre-trained large language models (LLMs) to synthesize programs in a\ndomain-specific layout language that describe objects and spatial relations\nbetween them. Executing such a program produces a specification of a constraint\nsatisfaction problem, which the system solves using a gradient-based\noptimization scheme to produce object positions and orientations. To produce\nobject geometry, the system retrieves 3D meshes from a database. Unlike prior\nwork which uses databases of category-annotated, mutually-aligned meshes, we\ndevelop a pipeline using vision-language models (VLMs) to retrieve meshes from\nmassive databases of un-annotated, inconsistently-aligned meshes. Experimental\nevaluations show that our system outperforms generative models trained on 3D\ndata for traditional, closed-universe scene generation tasks; it also\noutperforms a recent LLM-based layout generation method on open-universe scene\ngeneration.\n", "date": "2024-02-05", "categories": []} {"id": "2403.10166", "title": "SemanticHuman-HD: High-Resolution Semantic Disentangled 3D Human\n Generation", "abstract": " With the development of neural radiance fields and generative models,\nnumerous methods have been proposed for learning 3D human generation from 2D\nimages. These methods allow control over the pose of the generated 3D human and\nenable rendering from different viewpoints. However, none of these methods\nexplore semantic disentanglement in human image synthesis, i.e., they can not\ndisentangle the generation of different semantic parts, such as the body, tops,\nand bottoms. Furthermore, existing methods are limited to synthesize images at\n$512^2$ resolution due to the high computational cost of neural radiance\nfields. To address these limitations, we introduce SemanticHuman-HD, the first\nmethod to achieve semantic disentangled human image synthesis. Notably,\nSemanticHuman-HD is also the first method to achieve 3D-aware image synthesis\nat $1024^2$ resolution, benefiting from our proposed 3D-aware super-resolution\nmodule. By leveraging the depth maps and semantic masks as guidance for the\n3D-aware super-resolution, we significantly reduce the number of sampling\npoints during volume rendering, thereby reducing the computational cost. Our\ncomparative experiments demonstrate the superiority of our method. The\neffectiveness of each proposed component is also verified through ablation\nstudies. Moreover, our method opens up exciting possibilities for various\napplications, including 3D garment generation, semantic-aware image synthesis,\ncontrollable image synthesis, and out-of-domain image synthesis.\n", "date": "2024-03-15", "categories": []} {"id": "2403.10395", "title": "Isotropic3D: Image-to-3D Generation Based on a Single CLIP Embedding", "abstract": " Encouraged by the growing availability of pre-trained 2D diffusion models,\nimage-to-3D generation by leveraging Score Distillation Sampling (SDS) is\nmaking remarkable progress. Most existing methods combine novel-view lifting\nfrom 2D diffusion models which usually take the reference image as a condition\nwhile applying hard L2 image supervision at the reference view. Yet heavily\nadhering to the image is prone to corrupting the inductive knowledge of the 2D\ndiffusion model leading to flat or distorted 3D generation frequently. In this\nwork, we reexamine image-to-3D in a novel perspective and present Isotropic3D,\nan image-to-3D generation pipeline that takes only an image CLIP embedding as\ninput. Isotropic3D allows the optimization to be isotropic w.r.t. the azimuth\nangle by solely resting on the SDS loss. The core of our framework lies in a\ntwo-stage diffusion model fine-tuning. Firstly, we fine-tune a text-to-3D\ndiffusion model by substituting its text encoder with an image encoder, by\nwhich the model preliminarily acquires image-to-image capabilities. Secondly,\nwe perform fine-tuning using our Explicit Multi-view Attention (EMA) which\ncombines noisy multi-view images with the noise-free reference image as an\nexplicit condition. CLIP embedding is sent to the diffusion model throughout\nthe whole process while reference images are discarded once after fine-tuning.\nAs a result, with a single image CLIP embedding, Isotropic3D is capable of\ngenerating multi-view mutually consistent images and also a 3D model with more\nsymmetrical and neat content, well-proportioned geometry, rich colored texture,\nand less distortion compared with existing image-to-3D methods while still\npreserving the similarity to the reference image to a large extent. The project\npage is available at https://isotropic3d.github.io/. The code and models are\navailable at https://github.com/pkunliu/Isotropic3D.\n", "date": "2024-03-15", "categories": []} {"id": "2309.17175", "title": "TextField3D: Towards Enhancing Open-Vocabulary 3D Generation with Noisy\n Text Fields", "abstract": " Recent works learn 3D representation explicitly under text-3D guidance.\nHowever, limited text-3D data restricts the vocabulary scale and text control\nof generations. Generators may easily fall into a stereotype concept for\ncertain text prompts, thus losing open-vocabulary generation ability. To tackle\nthis issue, we introduce a conditional 3D generative model, namely TextField3D.\nSpecifically, rather than using the text prompts as input directly, we suggest\nto inject dynamic noise into the latent space of given text prompts, i.e.,\nNoisy Text Fields (NTFs). In this way, limited 3D data can be mapped to the\nappropriate range of textual latent space that is expanded by NTFs. To this\nend, an NTFGen module is proposed to model general text latent code in noisy\nfields. Meanwhile, an NTFBind module is proposed to align view-invariant image\nlatent code to noisy fields, further supporting image-conditional 3D\ngeneration. To guide the conditional generation in both geometry and texture,\nmulti-modal discrimination is constructed with a text-3D discriminator and a\ntext-2.5D discriminator. Compared to previous methods, TextField3D includes\nthree merits: 1) large vocabulary, 2) text consistency, and 3) low latency.\nExtensive experiments demonstrate that our method achieves a potential\nopen-vocabulary 3D generation capability.\n", "date": "2023-09-29", "categories": []} {"id": "2311.15573", "title": "EucliDreamer: Fast and High-Quality Texturing for 3D Models with Stable\n Diffusion Depth", "abstract": " This paper presents a novel method to generate textures for 3D models given\ntext prompts and 3D meshes. Additional depth information is taken into account\nto perform the Score Distillation Sampling (SDS) process with depth conditional\nStable Diffusion. We ran our model over the open-source dataset Objaverse and\nconducted a user study to compare the results with those of various 3D\ntexturing methods. We have shown that our model can generate more satisfactory\nresults and produce various art styles for the same object. In addition, we\nachieved faster time when generating textures of comparable quality. We also\nconduct thorough ablation studies of how different factors may affect\ngeneration quality, including sampling steps, guidance scale, negative prompts,\ndata augmentation, elevation range, and alternatives to SDS.\n", "date": "2023-11-27", "categories": []} {"id": "2311.17061", "title": "HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting", "abstract": " Realistic 3D human generation from text prompts is a desirable yet\nchallenging task. Existing methods optimize 3D representations like mesh or\nneural fields via score distillation sampling (SDS), which suffers from\ninadequate fine details or excessive training time. In this paper, we propose\nan efficient yet effective framework, HumanGaussian, that generates\nhigh-quality 3D humans with fine-grained geometry and realistic appearance. Our\nkey insight is that 3D Gaussian Splatting is an efficient renderer with\nperiodic Gaussian shrinkage or growing, where such adaptive density control can\nbe naturally guided by intrinsic human structures. Specifically, 1) we first\npropose a Structure-Aware SDS that simultaneously optimizes human appearance\nand geometry. The multi-modal score function from both RGB and depth space is\nleveraged to distill the Gaussian densification and pruning process. 2)\nMoreover, we devise an Annealed Negative Prompt Guidance by decomposing SDS\ninto a noisier generative score and a cleaner classifier score, which well\naddresses the over-saturation issue. The floating artifacts are further\neliminated based on Gaussian size in a prune-only phase to enhance generation\nsmoothness. Extensive experiments demonstrate the superior efficiency and\ncompetitive quality of our framework, rendering vivid 3D humans under diverse\nscenarios. Project Page: https://alvinliu0.github.io/projects/HumanGaussian\n", "date": "2023-11-28", "categories": []} {"id": "2403.08902", "title": "Envision3D: One Image to 3D with Anchor Views Interpolation", "abstract": " We present Envision3D, a novel method for efficiently generating high-quality\n3D content from a single image. Recent methods that extract 3D content from\nmulti-view images generated by diffusion models show great potential. However,\nit is still challenging for diffusion models to generate dense multi-view\nconsistent images, which is crucial for the quality of 3D content extraction.\nTo address this issue, we propose a novel cascade diffusion framework, which\ndecomposes the challenging dense views generation task into two tractable\nstages, namely anchor views generation and anchor views interpolation. In the\nfirst stage, we train the image diffusion model to generate global consistent\nanchor views conditioning on image-normal pairs. Subsequently, leveraging our\nvideo diffusion model fine-tuned on consecutive multi-view images, we conduct\ninterpolation on the previous anchor views to generate extra dense views. This\nframework yields dense, multi-view consistent images, providing comprehensive\n3D information. To further enhance the overall generation quality, we introduce\na coarse-to-fine sampling strategy for the reconstruction algorithm to robustly\nextract textured meshes from the generated dense images. Extensive experiments\ndemonstrate that our method is capable of generating high-quality 3D content in\nterms of texture and geometry, surpassing previous image-to-3D baseline\nmethods.\n", "date": "2024-03-13", "categories": []} {"id": "2403.09140", "title": "Sculpt3D: Multi-View Consistent Text-to-3D Generation with Sparse 3D\n Prior", "abstract": " Recent works on text-to-3d generation show that using only 2D diffusion\nsupervision for 3D generation tends to produce results with inconsistent\nappearances (e.g., faces on the back view) and inaccurate shapes (e.g., animals\nwith extra legs). Existing methods mainly address this issue by retraining\ndiffusion models with images rendered from 3D data to ensure multi-view\nconsistency while struggling to balance 2D generation quality with 3D\nconsistency. In this paper, we present a new framework Sculpt3D that equips the\ncurrent pipeline with explicit injection of 3D priors from retrieved reference\nobjects without re-training the 2D diffusion model. Specifically, we\ndemonstrate that high-quality and diverse 3D geometry can be guaranteed by\nkeypoints supervision through a sparse ray sampling approach. Moreover, to\nensure accurate appearances of different views, we further modulate the output\nof the 2D diffusion model to the correct patterns of the template views without\naltering the generated object's style. These two decoupled designs effectively\nharness 3D information from reference objects to generate 3D objects while\npreserving the generation quality of the 2D diffusion model. Extensive\nexperiments show our method can largely improve the multi-view consistency\nwhile retaining fidelity and diversity. Our project page is available at:\nhttps://stellarcheng.github.io/Sculpt3D/.\n", "date": "2024-03-14", "categories": []} {"id": "2403.09439", "title": "3D-SceneDreamer: Text-Driven 3D-Consistent Scene Generation", "abstract": " Text-driven 3D scene generation techniques have made rapid progress in recent\nyears. Their success is mainly attributed to using existing generative models\nto iteratively perform image warping and inpainting to generate 3D scenes.\nHowever, these methods heavily rely on the outputs of existing models, leading\nto error accumulation in geometry and appearance that prevent the models from\nbeing used in various scenarios (e.g., outdoor and unreal scenarios). To\naddress this limitation, we generatively refine the newly generated local views\nby querying and aggregating global 3D information, and then progressively\ngenerate the 3D scene. Specifically, we employ a tri-plane features-based NeRF\nas a unified representation of the 3D scene to constrain global 3D consistency,\nand propose a generative refinement network to synthesize new contents with\nhigher quality by exploiting the natural image prior from 2D diffusion model as\nwell as the global 3D information of the current scene. Our extensive\nexperiments demonstrate that, in comparison to previous methods, our approach\nsupports wide variety of scene generation and arbitrary camera trajectories\nwith improved visual quality and 3D consistency.\n", "date": "2024-03-14", "categories": []} {"id": "2403.09625", "title": "Make-Your-3D: Fast and Consistent Subject-Driven 3D Content Generation", "abstract": " Recent years have witnessed the strong power of 3D generation models, which\noffer a new level of creative flexibility by allowing users to guide the 3D\ncontent generation process through a single image or natural language. However,\nit remains challenging for existing 3D generation methods to create\nsubject-driven 3D content across diverse prompts. In this paper, we introduce a\nnovel 3D customization method, dubbed Make-Your-3D that can personalize\nhigh-fidelity and consistent 3D content from only a single image of a subject\nwith text description within 5 minutes. Our key insight is to harmonize the\ndistributions of a multi-view diffusion model and an identity-specific 2D\ngenerative model, aligning them with the distribution of the desired 3D\nsubject. Specifically, we design a co-evolution framework to reduce the\nvariance of distributions, where each model undergoes a process of learning\nfrom the other through identity-aware optimization and subject-prior\noptimization, respectively. Extensive experiments demonstrate that our method\ncan produce high-quality, consistent, and subject-specific 3D content with\ntext-driven modifications that are unseen in subject image.\n", "date": "2024-03-14", "categories": []} {"id": "2303.14207", "title": "DiffuScene: Denoising Diffusion Models for Generative Indoor Scene\n Synthesis", "abstract": " We present DiffuScene for indoor 3D scene synthesis based on a novel scene\nconfiguration denoising diffusion model. It generates 3D instance properties\nstored in an unordered object set and retrieves the most similar geometry for\neach object configuration, which is characterized as a concatenation of\ndifferent attributes, including location, size, orientation, semantics, and\ngeometry features. We introduce a diffusion network to synthesize a collection\nof 3D indoor objects by denoising a set of unordered object attributes.\nUnordered parametrization simplifies and eases the joint distribution\napproximation. The shape feature diffusion facilitates natural object\nplacements, including symmetries. Our method enables many downstream\napplications, including scene completion, scene arrangement, and\ntext-conditioned scene synthesis. Experiments on the 3D-FRONT dataset show that\nour method can synthesize more physically plausible and diverse indoor scenes\nthan state-of-the-art methods. Extensive ablation studies verify the\neffectiveness of our design choice in scene diffusion models.\n", "date": "2023-03-24", "categories": []} {"id": "2312.06439", "title": "DreamControl: Control-Based Text-to-3D Generation with 3D Self-Prior", "abstract": " 3D generation has raised great attention in recent years. With the success of\ntext-to-image diffusion models, the 2D-lifting technique becomes a promising\nroute to controllable 3D generation. However, these methods tend to present\ninconsistent geometry, which is also known as the Janus problem. We observe\nthat the problem is caused mainly by two aspects, i.e., viewpoint bias in 2D\ndiffusion models and overfitting of the optimization objective. To address it,\nwe propose a two-stage 2D-lifting framework, namely DreamControl, which\noptimizes coarse NeRF scenes as 3D self-prior and then generates fine-grained\nobjects with control-based score distillation. Specifically, adaptive viewpoint\nsampling and boundary integrity metric are proposed to ensure the consistency\nof generated priors. The priors are then regarded as input conditions to\nmaintain reasonable geometries, in which conditional LoRA and weighted score\nare further proposed to optimize detailed textures. DreamControl can generate\nhigh-quality 3D content in terms of both geometry consistency and texture\nfidelity. Moreover, our control-based optimization guidance is applicable to\nmore downstream tasks, including user-guided generation and 3D animation. The\nproject page is available at https://github.com/tyhuang0428/DreamControl.\n", "date": "2023-12-11", "categories": []} {"id": "2305.18766", "title": "HiFA: High-fidelity Text-to-3D Generation with Advanced Diffusion\n Guidance", "abstract": " The advancements in automatic text-to-3D generation have been remarkable.\nMost existing methods use pre-trained text-to-image diffusion models to\noptimize 3D representations like Neural Radiance Fields (NeRFs) via\nlatent-space denoising score matching. Yet, these methods often result in\nartifacts and inconsistencies across different views due to their suboptimal\noptimization approaches and limited understanding of 3D geometry. Moreover, the\ninherent constraints of NeRFs in rendering crisp geometry and stable textures\nusually lead to a two-stage optimization to attain high-resolution details.\nThis work proposes holistic sampling and smoothing approaches to achieve\nhigh-quality text-to-3D generation, all in a single-stage optimization. We\ncompute denoising scores in the text-to-image diffusion model's latent and\nimage spaces. Instead of randomly sampling timesteps (also referred to as noise\nlevels in denoising score matching), we introduce a novel timestep annealing\napproach that progressively reduces the sampled timestep throughout\noptimization. To generate high-quality renderings in a single-stage\noptimization, we propose regularization for the variance of z-coordinates along\nNeRF rays. To address texture flickering issues in NeRFs, we introduce a kernel\nsmoothing technique that refines importance sampling weights coarse-to-fine,\nensuring accurate and thorough sampling in high-density regions. Extensive\nexperiments demonstrate the superiority of our method over previous approaches,\nenabling the generation of highly detailed and view-consistent 3D assets\nthrough a single-stage training process.\n", "date": "2023-05-30", "categories": []} {"id": "2403.06317", "title": "An End-to-End Deep Learning Generative Framework for Refinable Shape\n Matching and Generation", "abstract": " Generative modelling for shapes is a prerequisite for In-Silico Clinical\nTrials (ISCTs), which aim to cost-effectively validate medical device\ninterventions using synthetic anatomical shapes, often represented as 3D\nsurface meshes. However, constructing AI models to generate shapes closely\nresembling the real mesh samples is challenging due to variable vertex counts,\nconnectivities, and the lack of dense vertex-wise correspondences across the\ntraining data. Employing graph representations for meshes, we develop a novel\nunsupervised geometric deep-learning model to establish refinable shape\ncorrespondences in a latent space, construct a population-derived atlas and\ngenerate realistic synthetic shapes. We additionally extend our proposed base\nmodel to a joint shape generative-clustering multi-atlas framework to\nincorporate further variability and preserve more details in the generated\nshapes. Experimental results using liver and left-ventricular models\ndemonstrate the approach's applicability to computational medicine,\nhighlighting its suitability for ISCTs through a comparative analysis.\n", "date": "2024-03-10", "categories": []} {"id": "2403.06470", "title": "3D-aware Image Generation and Editing with Multi-modal Conditions", "abstract": " 3D-consistent image generation from a single 2D semantic label is an\nimportant and challenging research topic in computer graphics and computer\nvision. Although some related works have made great progress in this field,\nmost of the existing methods suffer from poor disentanglement performance of\nshape and appearance, and lack multi-modal control. In this paper, we propose a\nnovel end-to-end 3D-aware image generation and editing model incorporating\nmultiple types of conditional inputs, including pure noise, text and reference\nimage. On the one hand, we dive into the latent space of 3D Generative\nAdversarial Networks (GANs) and propose a novel disentanglement strategy to\nseparate appearance features from shape features during the generation process.\nOn the other hand, we propose a unified framework for flexible image generation\nand editing tasks with multi-modal conditions. Our method can generate diverse\nimages with distinct noises, edit the attribute through a text description and\nconduct style transfer by giving a reference RGB image. Extensive experiments\ndemonstrate that the proposed method outperforms alternative approaches both\nqualitatively and quantitatively on image generation and editing.\n", "date": "2024-03-11", "categories": []} {"id": "2403.06738", "title": "V3D: Video Diffusion Models are Effective 3D Generators", "abstract": " Automatic 3D generation has recently attracted widespread attention. Recent\nmethods have greatly accelerated the generation speed, but usually produce\nless-detailed objects due to limited model capacity or 3D data. Motivated by\nrecent advancements in video diffusion models, we introduce V3D, which\nleverages the world simulation capacity of pre-trained video diffusion models\nto facilitate 3D generation. To fully unleash the potential of video diffusion\nto perceive the 3D world, we further introduce geometrical consistency prior\nand extend the video diffusion model to a multi-view consistent 3D generator.\nBenefiting from this, the state-of-the-art video diffusion model could be\nfine-tuned to generate 360degree orbit frames surrounding an object given a\nsingle image. With our tailored reconstruction pipelines, we can generate\nhigh-quality meshes or 3D Gaussians within 3 minutes. Furthermore, our method\ncan be extended to scene-level novel view synthesis, achieving precise control\nover the camera path with sparse input views. Extensive experiments demonstrate\nthe superior performance of the proposed approach, especially in terms of\ngeneration quality and multi-view consistency. Our code is available at\nhttps://github.com/heheyas/V3D\n", "date": "2024-03-11", "categories": []} {"id": "2403.05034", "title": "CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction\n Model", "abstract": " Feed-forward 3D generative models like the Large Reconstruction Model (LRM)\nhave demonstrated exceptional generation speed. However, the transformer-based\nmethods do not leverage the geometric priors of the triplane component in their\narchitecture, often leading to sub-optimal quality given the limited size of 3D\ndata and slow training. In this work, we present the Convolutional\nReconstruction Model (CRM), a high-fidelity feed-forward single image-to-3D\ngenerative model. Recognizing the limitations posed by sparse 3D data, we\nhighlight the necessity of integrating geometric priors into network design.\nCRM builds on the key observation that the visualization of triplane exhibits\nspatial correspondence of six orthographic images. First, it generates six\northographic view images from a single input image, then feeds these images\ninto a convolutional U-Net, leveraging its strong pixel-level alignment\ncapabilities and significant bandwidth to create a high-resolution triplane.\nCRM further employs Flexicubes as geometric representation, facilitating direct\nend-to-end optimization on textured meshes. Overall, our model delivers a\nhigh-fidelity textured mesh from an image in just 10 seconds, without any\ntest-time optimization.\n", "date": "2024-03-08", "categories": []} {"id": "2403.04225", "title": "3DTextureTransformer: Geometry Aware Texture Generation for Arbitrary\n Mesh Topology", "abstract": " Learning to generate textures for a novel 3D mesh given a collection of 3D\nmeshes and real-world 2D images is an important problem with applications in\nvarious domains such as 3D simulation, augmented and virtual reality, gaming,\narchitecture, and design. Existing solutions either do not produce high-quality\ntextures or deform the original high-resolution input mesh topology into a\nregular grid to make this generation easier but also lose the original mesh\ntopology. In this paper, we present a novel framework called the\n3DTextureTransformer that enables us to generate high-quality textures without\ndeforming the original, high-resolution input mesh. Our solution, a hybrid of\ngeometric deep learning and StyleGAN-like architecture, is flexible enough to\nwork on arbitrary mesh topologies and also easily extensible to texture\ngeneration for point cloud representations. Our solution employs a\nmessage-passing framework in 3D in conjunction with a StyleGAN-like\narchitecture for 3D texture generation. The architecture achieves\nstate-of-the-art performance among a class of solutions that can learn from a\ncollection of 3D geometry and real-world 2D images while working with any\narbitrary mesh topology.\n", "date": "2024-03-07", "categories": []} {"id": "2403.03206", "title": "Scaling Rectified Flow Transformers for High-Resolution Image Synthesis", "abstract": " Diffusion models create data from noise by inverting the forward paths of\ndata towards noise and have emerged as a powerful generative modeling technique\nfor high-dimensional, perceptual data such as images and videos. Rectified flow\nis a recent generative model formulation that connects data and noise in a\nstraight line. Despite its better theoretical properties and conceptual\nsimplicity, it is not yet decisively established as standard practice. In this\nwork, we improve existing noise sampling techniques for training rectified flow\nmodels by biasing them towards perceptually relevant scales. Through a\nlarge-scale study, we demonstrate the superior performance of this approach\ncompared to established diffusion formulations for high-resolution\ntext-to-image synthesis. Additionally, we present a novel transformer-based\narchitecture for text-to-image generation that uses separate weights for the\ntwo modalities and enables a bidirectional flow of information between image\nand text tokens, improving text comprehension, typography, and human preference\nratings. We demonstrate that this architecture follows predictable scaling\ntrends and correlates lower validation loss to improved text-to-image synthesis\nas measured by various metrics and human evaluations. Our largest models\noutperform state-of-the-art models, and we will make our experimental data,\ncode, and model weights publicly available.\n", "date": "2024-03-05", "categories": []} {"id": "2309.09256", "title": "LiDAR Data Synthesis with Denoising Diffusion Probabilistic Models", "abstract": " Generative modeling of 3D LiDAR data is an emerging task with promising\napplications for autonomous mobile robots, such as scalable simulation, scene\nmanipulation, and sparse-to-dense completion of LiDAR point clouds. While\nexisting approaches have demonstrated the feasibility of image-based LiDAR data\ngeneration using deep generative models, they still struggle with fidelity and\ntraining stability. In this work, we present R2DM, a novel generative model for\nLiDAR data that can generate diverse and high-fidelity 3D scene point clouds\nbased on the image representation of range and reflectance intensity. Our\nmethod is built upon denoising diffusion probabilistic models (DDPMs), which\nhave shown impressive results among generative model frameworks in recent\nyears. To effectively train DDPMs in the LiDAR domain, we first conduct an\nin-depth analysis of data representation, loss functions, and spatial inductive\nbiases. Leveraging our R2DM model, we also introduce a flexible LiDAR\ncompletion pipeline based on the powerful capabilities of DDPMs. We demonstrate\nthat our method surpasses existing methods in generating tasks on the KITTI-360\nand KITTI-Raw datasets, as well as in the completion task on the KITTI-360\ndataset. Our project page can be found at https://kazuto1011.github.io/r2dm.\n", "date": "2023-09-17", "categories": []} {"id": "2403.01430", "title": "On Diffusion Process in SE(3)-invariant Space", "abstract": " Sampling viable 3D structures (e.g., molecules and point clouds) with\nSE(3)-invariance using diffusion-based models proved promising in a variety of\nreal-world applications, wherein SE(3)-invariant properties can be naturally\ncharacterized by the inter-point distance manifold. However, due to the\nnon-trivial geometry, we still lack a comprehensive understanding of the\ndiffusion mechanism within such SE(3)-invariant space. This study addresses\nthis gap by mathematically delineating the diffusion mechanism under\nSE(3)-invariance, via zooming into the interaction behavior between coordinates\nand the inter-point distance manifold through the lens of differential\ngeometry. Upon this analysis, we propose accurate and projection-free diffusion\nSDE and ODE accordingly. Such formulations enable enhancing the performance and\nthe speed of generation pathways; meanwhile offering valuable insights into\nother systems incorporating SE(3)-invariance.\n", "date": "2024-03-03", "categories": []} {"id": "2403.02151", "title": "TripoSR: Fast 3D Object Reconstruction from a Single Image", "abstract": " This technical report introduces TripoSR, a 3D reconstruction model\nleveraging transformer architecture for fast feed-forward 3D generation,\nproducing 3D mesh from a single image in under 0.5 seconds. Building upon the\nLRM network architecture, TripoSR integrates substantial improvements in data\nprocessing, model design, and training techniques. Evaluations on public\ndatasets show that TripoSR exhibits superior performance, both quantitatively\nand qualitatively, compared to other open-source alternatives. Released under\nthe MIT license, TripoSR is intended to empower researchers, developers, and\ncreatives with the latest advancements in 3D generative AI.\n", "date": "2024-03-04", "categories": []} {"id": "2403.02217", "title": "DragTex: Generative Point-Based Texture Editing on 3D Mesh", "abstract": " Creating 3D textured meshes using generative artificial intelligence has\ngarnered significant attention recently. While existing methods support\ntext-based generative texture generation or editing on 3D meshes, they often\nstruggle to precisely control pixels of texture images through more intuitive\ninteraction. While 2D images can be edited generatively using drag interaction,\napplying this type of methods directly to 3D mesh textures still leads to\nissues such as the lack of local consistency among multiple views, error\naccumulation and long training times. To address these challenges, we propose a\ngenerative point-based 3D mesh texture editing method called DragTex. This\nmethod utilizes a diffusion model to blend locally inconsistent textures in the\nregion near the deformed silhouette between different views, enabling locally\nconsistent texture editing. Besides, we fine-tune a decoder to reduce\nreconstruction errors in the non-drag region, thereby mitigating overall error\naccumulation. Moreover, we train LoRA using multi-view images instead of\ntraining each view individually, which significantly shortens the training\ntime. The experimental results show that our method effectively achieves\ndragging textures on 3D meshes and generates plausible textures that align with\nthe desired intent of drag interaction.\n", "date": "2024-03-04", "categories": []} {"id": "2402.18780", "title": "A Quantitative Evaluation of Score Distillation Sampling Based\n Text-to-3D", "abstract": " The development of generative models that create 3D content from a text\nprompt has made considerable strides thanks to the use of the score\ndistillation sampling (SDS) method on pre-trained diffusion models for image\ngeneration. However, the SDS method is also the source of several artifacts,\nsuch as the Janus problem, the misalignment between the text prompt and the\ngenerated 3D model, and 3D model inaccuracies. While existing methods heavily\nrely on the qualitative assessment of these artifacts through visual inspection\nof a limited set of samples, in this work we propose more objective\nquantitative evaluation metrics, which we cross-validate via human ratings, and\nshow analysis of the failure cases of the SDS technique. We demonstrate the\neffectiveness of this analysis by designing a novel computationally efficient\nbaseline model that achieves state-of-the-art performance on the proposed\nmetrics while addressing all the above-mentioned artifacts.\n", "date": "2024-02-29", "categories": []} {"id": "2312.02209", "title": "AttriHuman-3D: Editable 3D Human Avatar Generation with Attribute\n Decomposition and Indexing", "abstract": " Editable 3D-aware generation, which supports user-interacted editing, has\nwitnessed rapid development recently. However, existing editable 3D GANs either\nfail to achieve high-accuracy local editing or suffer from huge computational\ncosts. We propose AttriHuman-3D, an editable 3D human generation model, which\naddress the aforementioned problems with attribute decomposition and indexing.\nThe core idea of the proposed model is to generate all attributes (e.g. human\nbody, hair, clothes and so on) in an overall attribute space with six feature\nplanes, which are then decomposed and manipulated with different attribute\nindexes. To precisely extract features of different attributes from the\ngenerated feature planes, we propose a novel attribute indexing method as well\nas an orthogonal projection regularization to enhance the disentanglement. We\nalso introduce a hyper-latent training strategy and an attribute-specific\nsampling strategy to avoid style entanglement and misleading punishment from\nthe discriminator. Our method allows users to interactively edit selected\nattributes in the generated 3D human avatars while keeping others fixed. Both\nqualitative and quantitative experiments demonstrate that our model provides a\nstrong disentanglement between different attributes, allows fine-grained image\nediting and generates high-quality 3D human avatars.\n", "date": "2023-12-03", "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": "2402.17115", "title": "CharNeRF: 3D Character Generation from Concept Art", "abstract": " 3D modeling holds significant importance in the realms of AR/VR and gaming,\nallowing for both artistic creativity and practical applications. However, the\nprocess is often time-consuming and demands a high level of skill. In this\npaper, we present a novel approach to create volumetric representations of 3D\ncharacters from consistent turnaround concept art, which serves as the standard\ninput in the 3D modeling industry. While Neural Radiance Field (NeRF) has been\na game-changer in image-based 3D reconstruction, to the best of our knowledge,\nthere is no known research that optimizes the pipeline for concept art. To\nharness the potential of concept art, with its defined body poses and specific\nview angles, we propose encoding it as priors for our model. We train the\nnetwork to make use of these priors for various 3D points through a learnable\nview-direction-attended multi-head self-attention layer. Additionally, we\ndemonstrate that a combination of ray sampling and surface sampling enhances\nthe inference capabilities of our network. Our model is able to generate\nhigh-quality 360-degree views of characters. Subsequently, we provide a simple\nguideline to better leverage our model to extract the 3D mesh. It is important\nto note that our model's inferencing capabilities are influenced by the\ntraining data's characteristics, primarily focusing on characters with a single\nhead, two arms, and two legs. Nevertheless, our methodology remains versatile\nand adaptable to concept art from diverse subject matters, without imposing any\nspecific assumptions on the data.\n", "date": "2024-02-27", "categories": []} {"id": "2402.17292", "title": "DivAvatar: Diverse 3D Avatar Generation with a Single Prompt", "abstract": " Text-to-Avatar generation has recently made significant strides due to\nadvancements in diffusion models. However, most existing work remains\nconstrained by limited diversity, producing avatars with subtle differences in\nappearance for a given text prompt. We design DivAvatar, a novel framework that\ngenerates diverse avatars, empowering 3D creatives with a multitude of distinct\nand richly varied 3D avatars from a single text prompt. Different from most\nexisting work that exploits scene-specific 3D representations such as NeRF,\nDivAvatar finetunes a 3D generative model (i.e., EVA3D), allowing diverse\navatar generation from simply noise sampling in inference time. DivAvatar has\ntwo key designs that help achieve generation diversity and visual quality. The\nfirst is a noise sampling technique during training phase which is critical in\ngenerating diverse appearances. The second is a semantic-aware zoom mechanism\nand a novel depth loss, the former producing appearances of high textual\nfidelity by separate fine-tuning of specific body parts and the latter\nimproving geometry quality greatly by smoothing the generated mesh in the\nfeatures space. Extensive experiments show that DivAvatar is highly versatile\nin generating avatars of diverse appearances.\n", "date": "2024-02-27", "categories": []} {"id": "2402.14253", "title": "MVD$^2$: Efficient Multiview 3D Reconstruction for Multiview Diffusion", "abstract": " As a promising 3D generation technique, multiview diffusion (MVD) has\nreceived a lot of attention due to its advantages in terms of generalizability,\nquality, and efficiency. By finetuning pretrained large image diffusion models\nwith 3D data, the MVD methods first generate multiple views of a 3D object\nbased on an image or text prompt and then reconstruct 3D shapes with multiview\n3D reconstruction. However, the sparse views and inconsistent details in the\ngenerated images make 3D reconstruction challenging. We present MVD$^2$, an\nefficient 3D reconstruction method for multiview diffusion (MVD) images.\nMVD$^2$ aggregates image features into a 3D feature volume by projection and\nconvolution and then decodes volumetric features into a 3D mesh. We train\nMVD$^2$ with 3D shape collections and MVD images prompted by rendered views of\n3D shapes. To address the discrepancy between the generated multiview images\nand ground-truth views of the 3D shapes, we design a simple-yet-efficient\nview-dependent training scheme. MVD$^2$ improves the 3D generation quality of\nMVD and is fast and robust to various MVD methods. After training, it can\nefficiently decode 3D meshes from multiview images within one second. We train\nMVD$^2$ with Zero-123++ and ObjectVerse-LVIS 3D dataset and demonstrate its\nsuperior performance in generating 3D models from multiview images generated by\ndifferent MVD methods, using both synthetic and real images as prompts.\n", "date": "2024-02-22", "categories": []} {"id": "2211.09869", "title": "RenderDiffusion: Image Diffusion for 3D Reconstruction, Inpainting and\n Generation", "abstract": " Diffusion models currently achieve state-of-the-art performance for both\nconditional and unconditional image generation. However, so far, image\ndiffusion models do not support tasks required for 3D understanding, such as\nview-consistent 3D generation or single-view object reconstruction. In this\npaper, we present RenderDiffusion, the first diffusion model for 3D generation\nand inference, trained using only monocular 2D supervision. Central to our\nmethod is a novel image denoising architecture that generates and renders an\nintermediate three-dimensional representation of a scene in each denoising\nstep. This enforces a strong inductive structure within the diffusion process,\nproviding a 3D consistent representation while only requiring 2D supervision.\nThe resulting 3D representation can be rendered from any view. We evaluate\nRenderDiffusion on FFHQ, AFHQ, ShapeNet and CLEVR datasets, showing competitive\nperformance for generation of 3D scenes and inference of 3D scenes from 2D\nimages. Additionally, our diffusion-based approach allows us to use 2D\ninpainting to edit 3D scenes.\n", "date": "2022-11-17", "categories": []} {"id": "2305.15399", "title": "Sin3DM: Learning a Diffusion Model from a Single 3D Textured Shape", "abstract": " Synthesizing novel 3D models that resemble the input example has long been\npursued by graphics artists and machine learning researchers. In this paper, we\npresent Sin3DM, a diffusion model that learns the internal patch distribution\nfrom a single 3D textured shape and generates high-quality variations with fine\ngeometry and texture details. Training a diffusion model directly in 3D would\ninduce large memory and computational cost. Therefore, we first compress the\ninput into a lower-dimensional latent space and then train a diffusion model on\nit. Specifically, we encode the input 3D textured shape into triplane feature\nmaps that represent the signed distance and texture fields of the input. The\ndenoising network of our diffusion model has a limited receptive field to avoid\noverfitting, and uses triplane-aware 2D convolution blocks to improve the\nresult quality. Aside from randomly generating new samples, our model also\nfacilitates applications such as retargeting, outpainting and local editing.\nThrough extensive qualitative and quantitative evaluation, we show that our\nmethod outperforms prior methods in generation quality of 3D shapes.\n", "date": "2023-05-24", "categories": []} {"id": "2306.06088", "title": "SENS: Part-Aware Sketch-based Implicit Neural Shape Modeling", "abstract": " We present SENS, a novel method for generating and editing 3D models from\nhand-drawn sketches, including those of abstract nature. Our method allows\nusers to quickly and easily sketch a shape, and then maps the sketch into the\nlatent space of a part-aware neural implicit shape architecture. SENS analyzes\nthe sketch and encodes its parts into ViT patch encoding, subsequently feeding\nthem into a transformer decoder that converts them to shape embeddings suitable\nfor editing 3D neural implicit shapes. SENS provides intuitive sketch-based\ngeneration and editing, and also succeeds in capturing the intent of the user's\nsketch to generate a variety of novel and expressive 3D shapes, even from\nabstract and imprecise sketches. Additionally, SENS supports refinement via\npart reconstruction, allowing for nuanced adjustments and artifact removal. It\nalso offers part-based modeling capabilities, enabling the combination of\nfeatures from multiple sketches to create more complex and customized 3D\nshapes. We demonstrate the effectiveness of our model compared to the\nstate-of-the-art using objective metric evaluation criteria and a user study,\nboth indicating strong performance on sketches with a medium level of\nabstraction. Furthermore, we showcase our method's intuitive sketch-based shape\nediting capabilities, and validate it through a usability study.\n", "date": "2023-06-09", "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": "2309.17261", "title": "Consistent123: One Image to Highly Consistent 3D Asset Using Case-Aware\n Diffusion Priors", "abstract": " Reconstructing 3D objects from a single image guided by pretrained diffusion\nmodels has demonstrated promising outcomes. However, due to utilizing the\ncase-agnostic rigid strategy, their generalization ability to arbitrary cases\nand the 3D consistency of reconstruction are still poor. In this work, we\npropose Consistent123, a case-aware two-stage method for highly consistent 3D\nasset reconstruction from one image with both 2D and 3D diffusion priors. In\nthe first stage, Consistent123 utilizes only 3D structural priors for\nsufficient geometry exploitation, with a CLIP-based case-aware adaptive\ndetection mechanism embedded within this process. In the second stage, 2D\ntexture priors are introduced and progressively take on a dominant guiding\nrole, delicately sculpting the details of the 3D model. Consistent123 aligns\nmore closely with the evolving trends in guidance requirements, adaptively\nproviding adequate 3D geometric initialization and suitable 2D texture\nrefinement for different objects. Consistent123 can obtain highly 3D-consistent\nreconstruction and exhibits strong generalization ability across various\nobjects. Qualitative and quantitative experiments show that our method\nsignificantly outperforms state-of-the-art image-to-3D methods. See\nhttps://Consistent123.github.io for a more comprehensive exploration of our\ngenerated 3D assets.\n", "date": "2023-09-29", "categories": []} {"id": "2402.11241", "title": "DiffPoint: Single and Multi-view Point Cloud Reconstruction with ViT\n Based Diffusion Model", "abstract": " As the task of 2D-to-3D reconstruction has gained significant attention in\nvarious real-world scenarios, it becomes crucial to be able to generate\nhigh-quality point clouds. Despite the recent success of deep learning models\nin generating point clouds, there are still challenges in producing\nhigh-fidelity results due to the disparities between images and point clouds.\nWhile vision transformers (ViT) and diffusion models have shown promise in\nvarious vision tasks, their benefits for reconstructing point clouds from\nimages have not been demonstrated yet. In this paper, we first propose a neat\nand powerful architecture called DiffPoint that combines ViT and diffusion\nmodels for the task of point cloud reconstruction. At each diffusion step, we\ndivide the noisy point clouds into irregular patches. Then, using a standard\nViT backbone that treats all inputs as tokens (including time information,\nimage embeddings, and noisy patches), we train our model to predict target\npoints based on input images. We evaluate DiffPoint on both single-view and\nmulti-view reconstruction tasks and achieve state-of-the-art results.\nAdditionally, we introduce a unified and flexible feature fusion module for\naggregating image features from single or multiple input images. Furthermore,\nour work demonstrates the feasibility of applying unified architectures across\nlanguages and images to improve 3D reconstruction tasks.\n", "date": "2024-02-17", "categories": []} {"id": "2402.09052", "title": "L3GO: Language Agents with Chain-of-3D-Thoughts for Generating\n Unconventional Objects", "abstract": " Diffusion-based image generation models such as DALL-E 3 and Stable\nDiffusion-XL demonstrate remarkable capabilities in generating images with\nrealistic and unique compositions. Yet, these models are not robust in\nprecisely reasoning about physical and spatial configurations of objects,\nespecially when instructed with unconventional, thereby out-of-distribution\ndescriptions, such as \"a chair with five legs\". In this paper, we propose a\nlanguage agent with chain-of-3D-thoughts (L3GO), an inference-time approach\nthat can reason about part-based 3D mesh generation of unconventional objects\nthat current data-driven diffusion models struggle with. More concretely, we\nuse large language models as agents to compose a desired object via\ntrial-and-error within the 3D simulation environment. To facilitate our\ninvestigation, we develop a new benchmark, Unconventionally Feasible Objects\n(UFO), as well as SimpleBlenv, a wrapper environment built on top of Blender\nwhere language agents can build and compose atomic building blocks via API\ncalls. Human and automatic GPT-4V evaluations show that our approach surpasses\nthe standard GPT-4 and other language agents (e.g., ReAct and Reflexion) for 3D\nmesh generation on ShapeNet. Moreover, when tested on our UFO benchmark, our\napproach outperforms other state-of-the-art text-to-2D image and text-to-3D\nmodels based on human evaluation.\n", "date": "2024-02-14", "categories": []} {"id": "2402.08682", "title": "IM-3D: Iterative Multiview Diffusion and Reconstruction for High-Quality\n 3D Generation", "abstract": " Most text-to-3D generators build upon off-the-shelf text-to-image models\ntrained on billions of images. They use variants of Score Distillation Sampling\n(SDS), which is slow, somewhat unstable, and prone to artifacts. A mitigation\nis to fine-tune the 2D generator to be multi-view aware, which can help\ndistillation or can be combined with reconstruction networks to output 3D\nobjects directly. In this paper, we further explore the design space of\ntext-to-3D models. We significantly improve multi-view generation by\nconsidering video instead of image generators. Combined with a 3D\nreconstruction algorithm which, by using Gaussian splatting, can optimize a\nrobust image-based loss, we directly produce high-quality 3D outputs from the\ngenerated views. Our new method, IM-3D, reduces the number of evaluations of\nthe 2D generator network 10-100x, resulting in a much more efficient pipeline,\nbetter quality, fewer geometric inconsistencies, and higher yield of usable 3D\nassets.\n", "date": "2024-02-13", "categories": []} {"id": "2402.05235", "title": "SPAD : Spatially Aware Multiview Diffusers", "abstract": " We present SPAD, a novel approach for creating consistent multi-view images\nfrom text prompts or single images. To enable multi-view generation, we\nrepurpose a pretrained 2D diffusion model by extending its self-attention\nlayers with cross-view interactions, and fine-tune it on a high quality subset\nof Objaverse. We find that a naive extension of the self-attention proposed in\nprior work (e.g. MVDream) leads to content copying between views. Therefore, we\nexplicitly constrain the cross-view attention based on epipolar geometry. To\nfurther enhance 3D consistency, we utilize Plucker coordinates derived from\ncamera rays and inject them as positional encoding. This enables SPAD to reason\nover spatial proximity in 3D well. In contrast to recent works that can only\ngenerate views at fixed azimuth and elevation, SPAD offers full camera control\nand achieves state-of-the-art results in novel view synthesis on unseen objects\nfrom the Objaverse and Google Scanned Objects datasets. Finally, we demonstrate\nthat text-to-3D generation using SPAD prevents the multi-face Janus issue. See\nmore details at our webpage: https://yashkant.github.io/spad\n", "date": "2024-02-07", "categories": []} {"id": "2402.05728", "title": "CTGAN: Semantic-guided Conditional Texture Generator for 3D Shapes", "abstract": " The entertainment industry relies on 3D visual content to create immersive\nexperiences, but traditional methods for creating textured 3D models can be\ntime-consuming and subjective. Generative networks such as StyleGAN have\nadvanced image synthesis, but generating 3D objects with high-fidelity textures\nis still not well explored, and existing methods have limitations. We propose\nthe Semantic-guided Conditional Texture Generator (CTGAN), producing\nhigh-quality textures for 3D shapes that are consistent with the viewing angle\nwhile respecting shape semantics. CTGAN utilizes the disentangled nature of\nStyleGAN to finely manipulate the input latent codes, enabling explicit control\nover both the style and structure of the generated textures. A coarse-to-fine\nencoder architecture is introduced to enhance control over the structure of the\nresulting textures via input segmentation. Experimental results show that CTGAN\noutperforms existing methods on multiple quality metrics and achieves\nstate-of-the-art performance on texture generation in both conditional and\nunconditional settings.\n", "date": "2024-02-08", "categories": []} {"id": "2402.05803", "title": "AvatarMMC: 3D Head Avatar Generation and Editing with Multi-Modal\n Conditioning", "abstract": " We introduce an approach for 3D head avatar generation and editing with\nmulti-modal conditioning based on a 3D Generative Adversarial Network (GAN) and\na Latent Diffusion Model (LDM). 3D GANs can generate high-quality head avatars\ngiven a single or no condition. However, it is challenging to generate samples\nthat adhere to multiple conditions of different modalities. On the other hand,\nLDMs excel at learning complex conditional distributions. To this end, we\npropose to exploit the conditioning capabilities of LDMs to enable multi-modal\ncontrol over the latent space of a pre-trained 3D GAN. Our method can generate\nand edit 3D head avatars given a mixture of control signals such as RGB input,\nsegmentation masks, and global attributes. This provides better control over\nthe generation and editing of synthetic avatars both globally and locally.\nExperiments show that our proposed approach outperforms a solely GAN-based\napproach both qualitatively and quantitatively on generation and editing tasks.\nTo the best of our knowledge, our approach is the first to introduce\nmulti-modal conditioning to 3D avatar generation and editing.\n\\\\href{avatarmmc-sig24.github.io}{Project Page}\n", "date": "2024-02-08", "categories": []} {"id": "2312.09305", "title": "Stable Score Distillation for High-Quality 3D Generation", "abstract": " Although Score Distillation Sampling (SDS) has exhibited remarkable\nperformance in conditional 3D content generation, a comprehensive understanding\nof its formulation is still lacking, hindering the development of 3D\ngeneration. In this work, we decompose SDS as a combination of three functional\ncomponents, namely mode-seeking, mode-disengaging and variance-reducing terms,\nanalyzing the properties of each. We show that problems such as over-smoothness\nand implausibility result from the intrinsic deficiency of the first two terms\nand propose a more advanced variance-reducing term than that introduced by SDS.\nBased on the analysis, we propose a simple yet effective approach named Stable\nScore Distillation (SSD) which strategically orchestrates each term for\nhigh-quality 3D generation and can be readily incorporated to various 3D\ngeneration frameworks and 3D representations. Extensive experiments validate\nthe efficacy of our approach, demonstrating its ability to generate\nhigh-fidelity 3D content without succumbing to issues such as over-smoothness.\n", "date": "2023-12-14", "categories": []} {"id": "2402.04717", "title": "InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with\n Semantic Graph Prior", "abstract": " Comprehending natural language instructions is a charming property for 3D\nindoor scene synthesis systems. Existing methods directly model object joint\ndistributions and express object relations implicitly within a scene, thereby\nhindering the controllability of generation. We introduce InstructScene, a\nnovel generative framework that integrates a semantic graph prior and a layout\ndecoder to improve controllability and fidelity for 3D scene synthesis. The\nproposed semantic graph prior jointly learns scene appearances and layout\ndistributions, exhibiting versatility across various downstream tasks in a\nzero-shot manner. To facilitate the benchmarking for text-driven 3D scene\nsynthesis, we curate a high-quality dataset of scene-instruction pairs with\nlarge language and multimodal models. Extensive experimental results reveal\nthat the proposed method surpasses existing state-of-the-art approaches by a\nlarge margin. Thorough ablation studies confirm the efficacy of crucial design\ncomponents. Project page: https://chenguolin.github.io/projects/InstructScene.\n", "date": "2024-02-07", "categories": []} {"id": "2402.05054", "title": "LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content\n Creation", "abstract": " 3D content creation has achieved significant progress in terms of both\nquality and speed. Although current feed-forward models can produce 3D objects\nin seconds, their resolution is constrained by the intensive computation\nrequired during training. In this paper, we introduce Large Multi-View Gaussian\nModel (LGM), a novel framework designed to generate high-resolution 3D models\nfrom text prompts or single-view images. Our key insights are two-fold: 1) 3D\nRepresentation: We propose multi-view Gaussian features as an efficient yet\npowerful representation, which can then be fused together for differentiable\nrendering. 2) 3D Backbone: We present an asymmetric U-Net as a high-throughput\nbackbone operating on multi-view images, which can be produced from text or\nsingle-view image input by leveraging multi-view diffusion models. Extensive\nexperiments demonstrate the high fidelity and efficiency of our approach.\nNotably, we maintain the fast speed to generate 3D objects within 5 seconds\nwhile boosting the training resolution to 512, thereby achieving\nhigh-resolution 3D content generation.\n", "date": "2024-02-07", "categories": []} {"id": "2303.07937", "title": "Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D\n Generation", "abstract": " Text-to-3D generation has shown rapid progress in recent days with the advent\nof score distillation, a methodology of using pretrained text-to-2D diffusion\nmodels to optimize neural radiance field (NeRF) in the zero-shot setting.\nHowever, the lack of 3D awareness in the 2D diffusion models destabilizes score\ndistillation-based methods from reconstructing a plausible 3D scene. To address\nthis issue, we propose 3DFuse, a novel framework that incorporates 3D awareness\ninto pretrained 2D diffusion models, enhancing the robustness and 3D\nconsistency of score distillation-based methods. We realize this by first\nconstructing a coarse 3D structure of a given text prompt and then utilizing\nprojected, view-specific depth map as a condition for the diffusion model.\nAdditionally, we introduce a training strategy that enables the 2D diffusion\nmodel learns to handle the errors and sparsity within the coarse 3D structure\nfor robust generation, as well as a method for ensuring semantic consistency\nthroughout all viewpoints of the scene. Our framework surpasses the limitations\nof prior arts, and has significant implications for 3D consistent generation of\n2D diffusion models.\n", "date": "2023-03-14", "categories": []} {"id": "2401.17212", "title": "ContactGen: Contact-Guided Interactive 3D Human Generation for Partners", "abstract": " Among various interactions between humans, such as eye contact and gestures,\nphysical interactions by contact can act as an essential moment in\nunderstanding human behaviors. Inspired by this fact, given a 3D partner human\nwith the desired interaction label, we introduce a new task of 3D human\ngeneration in terms of physical contact. Unlike previous works of interacting\nwith static objects or scenes, a given partner human can have diverse poses and\ndifferent contact regions according to the type of interaction. To handle this\nchallenge, we propose a novel method of generating interactive 3D humans for a\ngiven partner human based on a guided diffusion framework. Specifically, we\nnewly present a contact prediction module that adaptively estimates potential\ncontact regions between two input humans according to the interaction label.\nUsing the estimated potential contact regions as complementary guidances, we\ndynamically enforce ContactGen to generate interactive 3D humans for a given\npartner human within a guided diffusion model. We demonstrate ContactGen on the\nCHI3D dataset, where our method generates physically plausible and diverse\nposes compared to comparison methods.\n", "date": "2024-01-30", "categories": []} {"id": "2311.02826", "title": "InstructPix2NeRF: Instructed 3D Portrait Editing from a Single Image", "abstract": " With the success of Neural Radiance Field (NeRF) in 3D-aware portrait\nediting, a variety of works have achieved promising results regarding both\nquality and 3D consistency. However, these methods heavily rely on per-prompt\noptimization when handling natural language as editing instructions. Due to the\nlack of labeled human face 3D datasets and effective architectures, the area of\nhuman-instructed 3D-aware editing for open-world portraits in an end-to-end\nmanner remains under-explored. To solve this problem, we propose an end-to-end\ndiffusion-based framework termed InstructPix2NeRF, which enables instructed\n3D-aware portrait editing from a single open-world image with human\ninstructions. At its core lies a conditional latent 3D diffusion process that\nlifts 2D editing to 3D space by learning the correlation between the paired\nimages' difference and the instructions via triplet data. With the help of our\nproposed token position randomization strategy, we could even achieve\nmulti-semantic editing through one single pass with the portrait identity\nwell-preserved. Besides, we further propose an identity consistency module that\ndirectly modulates the extracted identity signals into our diffusion process,\nwhich increases the multi-view 3D identity consistency. Extensive experiments\nverify the effectiveness of our method and show its superiority against strong\nbaselines quantitatively and qualitatively. Source code and pre-trained models\ncan be found on our project page:\n\\url{https://mybabyyh.github.io/InstructPix2NeRF}.\n", "date": "2023-11-06", "categories": []} {"id": "2402.00225", "title": "Geometry aware 3D generation from in-the-wild images in ImageNet", "abstract": " Generating accurate 3D models is a challenging problem that traditionally\nrequires explicit learning from 3D datasets using supervised learning. Although\nrecent advances have shown promise in learning 3D models from 2D images, these\nmethods often rely on well-structured datasets with multi-view images of each\ninstance or camera pose information. Furthermore, these datasets usually\ncontain clean backgrounds with simple shapes, making them expensive to acquire\nand hard to generalize, which limits the applicability of these methods. To\novercome these limitations, we propose a method for reconstructing 3D geometry\nfrom the diverse and unstructured Imagenet dataset without camera pose\ninformation. We use an efficient triplane representation to learn 3D models\nfrom 2D images and modify the architecture of the generator backbone based on\nStyleGAN2 to adapt to the highly diverse dataset. To prevent mode collapse and\nimprove the training stability on diverse data, we propose to use multi-view\ndiscrimination. The trained generator can produce class-conditional 3D models\nas well as renderings from arbitrary viewpoints. The class-conditional\ngeneration results demonstrate significant improvement over the current\nstate-of-the-art method. Additionally, using PTI, we can efficiently\nreconstruct the whole 3D geometry from single-view images.\n", "date": "2024-01-31", "categories": []} {"id": "2402.01241", "title": "Can Shape-Infused Joint Embeddings Improve Image-Conditioned 3D\n Diffusion?", "abstract": " Recent advancements in deep generative models, particularly with the\napplication of CLIP (Contrastive Language Image Pretraining) to Denoising\nDiffusion Probabilistic Models (DDPMs), have demonstrated remarkable\neffectiveness in text to image generation. The well structured embedding space\nof CLIP has also been extended to image to shape generation with DDPMs,\nyielding notable results. Despite these successes, some fundamental questions\narise: Does CLIP ensure the best results in shape generation from images? Can\nwe leverage conditioning to bring explicit 3D knowledge into the generative\nprocess and obtain better quality? This study introduces CISP (Contrastive\nImage Shape Pre training), designed to enhance 3D shape synthesis guided by 2D\nimages. CISP aims to enrich the CLIP framework by aligning 2D images with 3D\nshapes in a shared embedding space, specifically capturing 3D characteristics\npotentially overlooked by CLIP's text image focus. Our comprehensive analysis\nassesses CISP's guidance performance against CLIP guided models, focusing on\ngeneration quality, diversity, and coherence of the produced shapes with the\nconditioning image. We find that, while matching CLIP in generation quality and\ndiversity, CISP substantially improves coherence with input images,\nunderscoring the value of incorporating 3D knowledge into generative models.\nThese findings suggest a promising direction for advancing the synthesis of 3D\nvisual content by integrating multimodal systems with 3D representations.\n", "date": "2024-02-02", "categories": []} {"id": "2402.00864", "title": "ViCA-NeRF: View-Consistency-Aware 3D Editing of Neural Radiance Fields", "abstract": " We introduce ViCA-NeRF, the first view-consistency-aware method for 3D\nediting with text instructions. In addition to the implicit neural radiance\nfield (NeRF) modeling, our key insight is to exploit two sources of\nregularization that explicitly propagate the editing information across\ndifferent views, thus ensuring multi-view consistency. For geometric\nregularization, we leverage the depth information derived from NeRF to\nestablish image correspondences between different views. For learned\nregularization, we align the latent codes in the 2D diffusion model between\nedited and unedited images, enabling us to edit key views and propagate the\nupdate throughout the entire scene. Incorporating these two strategies, our\nViCA-NeRF operates in two stages. In the initial stage, we blend edits from\ndifferent views to create a preliminary 3D edit. This is followed by a second\nstage of NeRF training, dedicated to further refining the scene's appearance.\nExperimental results demonstrate that ViCA-NeRF provides more flexible,\nefficient (3 times faster) editing with higher levels of consistency and\ndetails, compared with the state of the art. Our code is publicly available.\n", "date": "2024-02-01", "categories": []} {"id": "2402.00867", "title": "AToM: Amortized Text-to-Mesh using 2D Diffusion", "abstract": " We introduce Amortized Text-to-Mesh (AToM), a feed-forward text-to-mesh\nframework optimized across multiple text prompts simultaneously. In contrast to\nexisting text-to-3D methods that often entail time-consuming per-prompt\noptimization and commonly output representations other than polygonal meshes,\nAToM directly generates high-quality textured meshes in less than 1 second with\naround 10 times reduction in the training cost, and generalizes to unseen\nprompts. Our key idea is a novel triplane-based text-to-mesh architecture with\na two-stage amortized optimization strategy that ensures stable training and\nenables scalability. Through extensive experiments on various prompt\nbenchmarks, AToM significantly outperforms state-of-the-art amortized\napproaches with over 4 times higher accuracy (in DF415 dataset) and produces\nmore distinguishable and higher-quality 3D outputs. AToM demonstrates strong\ngeneralizability, offering finegrained 3D assets for unseen interpolated\nprompts without further optimization during inference, unlike per-prompt\nsolutions.\n", "date": "2024-02-01", "categories": []} {"id": "2305.11588", "title": "Text2NeRF: Text-Driven 3D Scene Generation with Neural Radiance Fields", "abstract": " Text-driven 3D scene generation is widely applicable to video gaming, film\nindustry, and metaverse applications that have a large demand for 3D scenes.\nHowever, existing text-to-3D generation methods are limited to producing 3D\nobjects with simple geometries and dreamlike styles that lack realism. In this\nwork, we present Text2NeRF, which is able to generate a wide range of 3D scenes\nwith complicated geometric structures and high-fidelity textures purely from a\ntext prompt. To this end, we adopt NeRF as the 3D representation and leverage a\npre-trained text-to-image diffusion model to constrain the 3D reconstruction of\nthe NeRF to reflect the scene description. Specifically, we employ the\ndiffusion model to infer the text-related image as the content prior and use a\nmonocular depth estimation method to offer the geometric prior. Both content\nand geometric priors are utilized to update the NeRF model. To guarantee\ntextured and geometric consistency between different views, we introduce a\nprogressive scene inpainting and updating strategy for novel view synthesis of\nthe scene. Our method requires no additional training data but only a natural\nlanguage description of the scene as the input. Extensive experiments\ndemonstrate that our Text2NeRF outperforms existing methods in producing\nphoto-realistic, multi-view consistent, and diverse 3D scenes from a variety of\nnatural language prompts. Our code is available at\nhttps://github.com/eckertzhang/Text2NeRF.\n", "date": "2023-05-19", "categories": []} {"id": "2401.17603", "title": "Topology-Aware Latent Diffusion for 3D Shape Generation", "abstract": " We introduce a new generative model that combines latent diffusion with\npersistent homology to create 3D shapes with high diversity, with a special\nemphasis on their topological characteristics. Our method involves representing\n3D shapes as implicit fields, then employing persistent homology to extract\ntopological features, including Betti numbers and persistence diagrams. The\nshape generation process consists of two steps. Initially, we employ a\ntransformer-based autoencoding module to embed the implicit representation of\neach 3D shape into a set of latent vectors. Subsequently, we navigate through\nthe learned latent space via a diffusion model. By strategically incorporating\ntopological features into the diffusion process, our generative module is able\nto produce a richer variety of 3D shapes with different topological structures.\nFurthermore, our framework is flexible, supporting generation tasks constrained\nby a variety of inputs, including sparse and partial point clouds, as well as\nsketches. By modifying the persistence diagrams, we can alter the topology of\nthe shapes generated from these input modalities.\n", "date": "2024-01-31", "categories": []} {"id": "2401.17807", "title": "Advances in 3D Generation: A Survey", "abstract": " Generating 3D models lies at the core of computer graphics and has been the\nfocus of decades of research. With the emergence of advanced neural\nrepresentations and generative models, the field of 3D content generation is\ndeveloping rapidly, enabling the creation of increasingly high-quality and\ndiverse 3D models. The rapid growth of this field makes it difficult to stay\nabreast of all recent developments. In this survey, we aim to introduce the\nfundamental methodologies of 3D generation methods and establish a structured\nroadmap, encompassing 3D representation, generation methods, datasets, and\ncorresponding applications. Specifically, we introduce the 3D representations\nthat serve as the backbone for 3D generation. Furthermore, we provide a\ncomprehensive overview of the rapidly growing literature on generation methods,\ncategorized by the type of algorithmic paradigms, including feedforward\ngeneration, optimization-based generation, procedural generation, and\ngenerative novel view synthesis. Lastly, we discuss available datasets,\napplications, and open challenges. We hope this survey will help readers\nexplore this exciting topic and foster further advancements in the field of 3D\ncontent generation.\n", "date": "2024-01-31", "categories": []} {"id": "2401.17895", "title": "ReplaceAnything3D:Text-Guided 3D Scene Editing with Compositional Neural\n Radiance Fields", "abstract": " We introduce ReplaceAnything3D model (RAM3D), a novel text-guided 3D scene\nediting method that enables the replacement of specific objects within a scene.\nGiven multi-view images of a scene, a text prompt describing the object to\nreplace, and a text prompt describing the new object, our Erase-and-Replace\napproach can effectively swap objects in the scene with newly generated content\nwhile maintaining 3D consistency across multiple viewpoints. We demonstrate the\nversatility of ReplaceAnything3D by applying it to various realistic 3D scenes,\nshowcasing results of modified foreground objects that are well-integrated with\nthe rest of the scene without affecting its overall integrity.\n", "date": "2024-01-31", "categories": []} {"id": "2401.14257", "title": "Sketch2NeRF: Multi-view Sketch-guided Text-to-3D Generation", "abstract": " Recently, text-to-3D approaches have achieved high-fidelity 3D content\ngeneration using text description. However, the generated objects are\nstochastic and lack fine-grained control. Sketches provide a cheap approach to\nintroduce such fine-grained control. Nevertheless, it is challenging to achieve\nflexible control from these sketches due to their abstraction and ambiguity. In\nthis paper, we present a multi-view sketch-guided text-to-3D generation\nframework (namely, Sketch2NeRF) to add sketch control to 3D generation.\nSpecifically, our method leverages pretrained 2D diffusion models (e.g., Stable\nDiffusion and ControlNet) to supervise the optimization of a 3D scene\nrepresented by a neural radiance field (NeRF). We propose a novel synchronized\ngeneration and reconstruction method to effectively optimize the NeRF. In the\nexperiments, we collected two kinds of multi-view sketch datasets to evaluate\nthe proposed method. We demonstrate that our method can synthesize 3D\nconsistent contents with fine-grained sketch control while being high-fidelity\nto text prompts. Extensive results show that our method achieves\nstate-of-the-art performance in terms of sketch similarity and text alignment.\n", "date": "2024-01-25", "categories": []} {"id": "2401.13203", "title": "Style-Consistent 3D Indoor Scene Synthesis with Decoupled Objects", "abstract": " Controllable 3D indoor scene synthesis stands at the forefront of\ntechnological progress, offering various applications like gaming, film, and\naugmented/virtual reality. The capability to stylize and de-couple objects\nwithin these scenarios is a crucial factor, providing an advanced level of\ncontrol throughout the editing process. This control extends not just to\nmanipulating geometric attributes like translation and scaling but also\nincludes managing appearances, such as stylization. Current methods for scene\nstylization are limited to applying styles to the entire scene, without the\nability to separate and customize individual objects. Addressing the\nintricacies of this challenge, we introduce a unique pipeline designed for\nsynthesis 3D indoor scenes. Our approach involves strategically placing objects\nwithin the scene, utilizing information from professionally designed bounding\nboxes. Significantly, our pipeline prioritizes maintaining style consistency\nacross multiple objects within the scene, ensuring a cohesive and visually\nappealing result aligned with the desired aesthetic. The core strength of our\npipeline lies in its ability to generate 3D scenes that are not only visually\nimpressive but also exhibit features like photorealism, multi-view consistency,\nand diversity. These scenes are crafted in response to various natural language\nprompts, demonstrating the versatility and adaptability of our model.\n", "date": "2024-01-24", "categories": []} {"id": "2310.12474", "title": "Enhancing High-Resolution 3D Generation through Pixel-wise Gradient\n Clipping", "abstract": " High-resolution 3D object generation remains a challenging task primarily due\nto the limited availability of comprehensive annotated training data. Recent\nadvancements have aimed to overcome this constraint by harnessing image\ngenerative models, pretrained on extensive curated web datasets, using\nknowledge transfer techniques like Score Distillation Sampling (SDS).\nEfficiently addressing the requirements of high-resolution rendering often\nnecessitates the adoption of latent representation-based models, such as the\nLatent Diffusion Model (LDM). In this framework, a significant challenge\narises: To compute gradients for individual image pixels, it is necessary to\nbackpropagate gradients from the designated latent space through the frozen\ncomponents of the image model, such as the VAE encoder used within LDM.\nHowever, this gradient propagation pathway has never been optimized, remaining\nuncontrolled during training. We find that the unregulated gradients adversely\naffect the 3D model's capacity in acquiring texture-related information from\nthe image generative model, leading to poor quality appearance synthesis. To\naddress this overarching challenge, we propose an innovative operation termed\nPixel-wise Gradient Clipping (PGC) designed for seamless integration into\nexisting 3D generative models, thereby enhancing their synthesis quality.\nSpecifically, we control the magnitude of stochastic gradients by clipping the\npixel-wise gradients efficiently, while preserving crucial texture-related\ngradient directions. Despite this simplicity and minimal extra cost, extensive\nexperiments demonstrate the efficacy of our PGC in enhancing the performance of\nexisting 3D generative models for high-resolution object rendering.\n", "date": "2023-10-19", "categories": []} {"id": "2401.09416", "title": "TextureDreamer: Image-guided Texture Synthesis through Geometry-aware\n Diffusion", "abstract": " We present TextureDreamer, a novel image-guided texture synthesis method to\ntransfer relightable textures from a small number of input images (3 to 5) to\ntarget 3D shapes across arbitrary categories. Texture creation is a pivotal\nchallenge in vision and graphics. Industrial companies hire experienced artists\nto manually craft textures for 3D assets. Classical methods require densely\nsampled views and accurately aligned geometry, while learning-based methods are\nconfined to category-specific shapes within the dataset. In contrast,\nTextureDreamer can transfer highly detailed, intricate textures from real-world\nenvironments to arbitrary objects with only a few casually captured images,\npotentially significantly democratizing texture creation. Our core idea,\npersonalized geometry-aware score distillation (PGSD), draws inspiration from\nrecent advancements in diffuse models, including personalized modeling for\ntexture information extraction, variational score distillation for detailed\nappearance synthesis, and explicit geometry guidance with ControlNet. Our\nintegration and several essential modifications substantially improve the\ntexture quality. Experiments on real images spanning different categories show\nthat TextureDreamer can successfully transfer highly realistic, semantic\nmeaningful texture to arbitrary objects, surpassing the visual quality of\nprevious state-of-the-art.\n", "date": "2024-01-17", "categories": []} {"id": "2311.02700", "title": "A Generative Multi-Resolution Pyramid and Normal-Conditioning 3D Cloth\n Draping", "abstract": " RGB cloth generation has been deeply studied in the related literature,\nhowever, 3D garment generation remains an open problem. In this paper, we build\na conditional variational autoencoder for 3D garment generation and draping. We\npropose a pyramid network to add garment details progressively in a canonical\nspace, i.e. unposing and unshaping the garments w.r.t. the body. We study\nconditioning the network on surface normal UV maps, as an intermediate\nrepresentation, which is an easier problem to optimize than 3D coordinates. Our\nresults on two public datasets, CLOTH3D and CAPE, show that our model is\nrobust, controllable in terms of detail generation by the use of\nmulti-resolution pyramids, and achieves state-of-the-art results that can\nhighly generalize to unseen garments, poses, and shapes even when training with\nsmall amounts of data.\n", "date": "2023-11-05", "categories": []} {"id": "2401.06437", "title": "3D-PreMise: Can Large Language Models Generate 3D Shapes with Sharp\n Features and Parametric Control?", "abstract": " Recent advancements in implicit 3D representations and generative models have\nmarkedly propelled the field of 3D object generation forward. However, it\nremains a significant challenge to accurately model geometries with defined\nsharp features under parametric controls, which is crucial in fields like\nindustrial design and manufacturing. To bridge this gap, we introduce a\nframework that employs Large Language Models (LLMs) to generate text-driven 3D\nshapes, manipulating 3D software via program synthesis. We present 3D-PreMise,\na dataset specifically tailored for 3D parametric modeling of industrial\nshapes, designed to explore state-of-the-art LLMs within our proposed pipeline.\nOur work reveals effective generation strategies and delves into the\nself-correction capabilities of LLMs using a visual interface. Our work\nhighlights both the potential and limitations of LLMs in 3D parametric modeling\nfor industrial applications.\n", "date": "2024-01-12", "categories": []} {"id": "2401.07727", "title": "HexaGen3D: StableDiffusion is just one step away from Fast and Diverse\n Text-to-3D Generation", "abstract": " Despite the latest remarkable advances in generative modeling, efficient\ngeneration of high-quality 3D assets from textual prompts remains a difficult\ntask. A key challenge lies in data scarcity: the most extensive 3D datasets\nencompass merely millions of assets, while their 2D counterparts contain\nbillions of text-image pairs. To address this, we propose a novel approach\nwhich harnesses the power of large, pretrained 2D diffusion models. More\nspecifically, our approach, HexaGen3D, fine-tunes a pretrained text-to-image\nmodel to jointly predict 6 orthographic projections and the corresponding\nlatent triplane. We then decode these latents to generate a textured mesh.\nHexaGen3D does not require per-sample optimization, and can infer high-quality\nand diverse objects from textual prompts in 7 seconds, offering significantly\nbetter quality-to-latency trade-offs when comparing to existing approaches.\nFurthermore, HexaGen3D demonstrates strong generalization to new objects or\ncompositions.\n", "date": "2024-01-15", "categories": []} {"id": "2312.13941", "title": "Controllable 3D Face Generation with Conditional Style Code Diffusion", "abstract": " Generating photorealistic 3D faces from given conditions is a challenging\ntask. Existing methods often rely on time-consuming one-by-one optimization\napproaches, which are not efficient for modeling the same distribution content,\ne.g., faces. Additionally, an ideal controllable 3D face generation model\nshould consider both facial attributes and expressions. Thus we propose a novel\napproach called TEx-Face(TExt & Expression-to-Face) that addresses these\nchallenges by dividing the task into three components, i.e., 3D GAN Inversion,\nConditional Style Code Diffusion, and 3D Face Decoding. For 3D GAN inversion,\nwe introduce two methods which aim to enhance the representation of style codes\nand alleviate 3D inconsistencies. Furthermore, we design a style code denoiser\nto incorporate multiple conditions into the style code and propose a data\naugmentation strategy to address the issue of insufficient paired\nvisual-language data. Extensive experiments conducted on FFHQ, CelebA-HQ, and\nCelebA-Dialog demonstrate the promising performance of our TEx-Face in\nachieving the efficient and controllable generation of photorealistic 3D faces.\nThe code will be available at https://github.com/sxl142/TEx-Face.\n", "date": "2023-12-21", "categories": []} {"id": "2306.10830", "title": "3D VR Sketch Guided 3D Shape Prototyping and Exploration", "abstract": " 3D shape modeling is labor-intensive, time-consuming, and requires years of\nexpertise. To facilitate 3D shape modeling, we propose a 3D shape generation\nnetwork that takes a 3D VR sketch as a condition. We assume that sketches are\ncreated by novices without art training and aim to reconstruct geometrically\nrealistic 3D shapes of a given category. To handle potential sketch ambiguity,\nour method creates multiple 3D shapes that align with the original sketch's\nstructure. We carefully design our method, training the model step-by-step and\nleveraging multi-modal 3D shape representation to support training with limited\ntraining data. To guarantee the realism of generated 3D shapes we leverage the\nnormalizing flow that models the distribution of the latent space of 3D shapes.\nTo encourage the fidelity of the generated 3D shapes to an input sketch, we\npropose a dedicated loss that we deploy at different stages of the training\nprocess. The code is available at https://github.com/Rowl1ng/3Dsketch2shape.\n", "date": "2023-06-19", "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": "2301.00950", "title": "Class-Continuous Conditional Generative Neural Radiance Field", "abstract": " The 3D-aware image synthesis focuses on conserving spatial consistency\nbesides generating high-resolution images with fine details. Recently, Neural\nRadiance Field (NeRF) has been introduced for synthesizing novel views with low\ncomputational cost and superior performance. While several works investigate a\ngenerative NeRF and show remarkable achievement, they cannot handle conditional\nand continuous feature manipulation in the generation procedure. In this work,\nwe introduce a novel model, called Class-Continuous Conditional Generative NeRF\n($\\text{C}^{3}$G-NeRF), which can synthesize conditionally manipulated\nphotorealistic 3D-consistent images by projecting conditional features to the\ngenerator and the discriminator. The proposed $\\text{C}^{3}$G-NeRF is evaluated\nwith three image datasets, AFHQ, CelebA, and Cars. As a result, our model shows\nstrong 3D-consistency with fine details and smooth interpolation in conditional\nfeature manipulation. For instance, $\\text{C}^{3}$G-NeRF exhibits a Fr\\'echet\nInception Distance (FID) of 7.64 in 3D-aware face image synthesis with a\n$\\text{128}^{2}$ resolution. Additionally, we provide FIDs of generated\n3D-aware images of each class of the datasets as it is possible to synthesize\nclass-conditional images with $\\text{C}^{3}$G-NeRF.\n", "date": "2023-01-03", "categories": []} {"id": "2307.03538", "title": "Language-free Compositional Action Generation via Decoupling Refinement", "abstract": " Composing simple elements into complex concepts is crucial yet challenging,\nespecially for 3D action generation. Existing methods largely rely on extensive\nneural language annotations to discern composable latent semantics, a process\nthat is often costly and labor-intensive. In this study, we introduce a novel\nframework to generate compositional actions without reliance on language\nauxiliaries. Our approach consists of three main components: Action Coupling,\nConditional Action Generation, and Decoupling Refinement. Action Coupling\nutilizes an energy model to extract the attention masks of each sub-action,\nsubsequently integrating two actions using these attentions to generate\npseudo-training examples. Then, we employ a conditional generative model, CVAE,\nto learn a latent space, facilitating the diverse generation. Finally, we\npropose Decoupling Refinement, which leverages a self-supervised pre-trained\nmodel MAE to ensure semantic consistency between the sub-actions and\ncompositional actions. This refinement process involves rendering generated 3D\nactions into 2D space, decoupling these images into two sub-segments, using the\nMAE model to restore the complete image from sub-segments, and constraining the\nrecovered images to match images rendered from raw sub-actions. Due to the lack\nof existing datasets containing both sub-actions and compositional actions, we\ncreated two new datasets, named HumanAct-C and UESTC-C, and present a\ncorresponding evaluation metric. Both qualitative and quantitative assessments\nare conducted to show our efficacy.\n", "date": "2023-07-07", "categories": []} {"id": "2401.04099", "title": "AGG: Amortized Generative 3D Gaussians for Single Image to 3D", "abstract": " Given the growing need for automatic 3D content creation pipelines, various\n3D representations have been studied to generate 3D objects from a single\nimage. Due to its superior rendering efficiency, 3D Gaussian splatting-based\nmodels have recently excelled in both 3D reconstruction and generation. 3D\nGaussian splatting approaches for image to 3D generation are often\noptimization-based, requiring many computationally expensive score-distillation\nsteps. To overcome these challenges, we introduce an Amortized Generative 3D\nGaussian framework (AGG) that instantly produces 3D Gaussians from a single\nimage, eliminating the need for per-instance optimization. Utilizing an\nintermediate hybrid representation, AGG decomposes the generation of 3D\nGaussian locations and other appearance attributes for joint optimization.\nMoreover, we propose a cascaded pipeline that first generates a coarse\nrepresentation of the 3D data and later upsamples it with a 3D Gaussian\nsuper-resolution module. Our method is evaluated against existing\noptimization-based 3D Gaussian frameworks and sampling-based pipelines\nutilizing other 3D representations, where AGG showcases competitive generation\nabilities both qualitatively and quantitatively while being several orders of\nmagnitude faster. Project page: https://ir1d.github.io/AGG/\n", "date": "2024-01-08", "categories": []} {"id": "2401.02620", "title": "Progress and Prospects in 3D Generative AI: A Technical Overview\n including 3D human", "abstract": " While AI-generated text and 2D images continue to expand its territory, 3D\ngeneration has gradually emerged as a trend that cannot be ignored. Since the\nyear 2023 an abundant amount of research papers has emerged in the domain of 3D\ngeneration. This growth encompasses not just the creation of 3D objects, but\nalso the rapid development of 3D character and motion generation. Several key\nfactors contribute to this progress. The enhanced fidelity in stable diffusion,\ncoupled with control methods that ensure multi-view consistency, and realistic\nhuman models like SMPL-X, contribute synergistically to the production of 3D\nmodels with remarkable consistency and near-realistic appearances. The\nadvancements in neural network-based 3D storing and rendering models, such as\nNeural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have\naccelerated the efficiency and realism of neural rendered models. Furthermore,\nthe multimodality capabilities of large language models have enabled language\ninputs to transcend into human motion outputs. This paper aims to provide a\ncomprehensive overview and summary of the relevant papers published mostly\nduring the latter half year of 2023. It will begin by discussing the AI\ngenerated object models in 3D, followed by the generated 3D human models, and\nfinally, the generated 3D human motions, culminating in a conclusive summary\nand a vision for the future.\n", "date": "2024-01-05", "categories": []} {"id": "2401.02937", "title": "Locally Adaptive Neural 3D Morphable Models", "abstract": " We present the Locally Adaptive Morphable Model (LAMM), a highly flexible\nAuto-Encoder (AE) framework for learning to generate and manipulate 3D meshes.\nWe train our architecture following a simple self-supervised training scheme in\nwhich input displacements over a set of sparse control vertices are used to\noverwrite the encoded geometry in order to transform one training sample into\nanother. During inference, our model produces a dense output that adheres\nlocally to the specified sparse geometry while maintaining the overall\nappearance of the encoded object. This approach results in state-of-the-art\nperformance in both disentangling manipulated geometry and 3D mesh\nreconstruction. To the best of our knowledge LAMM is the first end-to-end\nframework that enables direct local control of 3D vertex geometry in a single\nforward pass. A very efficient computational graph allows our network to train\nwith only a fraction of the memory required by previous methods and run faster\nduring inference, generating 12k vertex meshes at $>$60fps on a single CPU\nthread. We further leverage local geometry control as a primitive for higher\nlevel editing operations and present a set of derivative capabilities such as\nswapping and sampling object parts. Code and pretrained models can be found at\nhttps://github.com/michaeltrs/LAMM.\n", "date": "2024-01-05", "categories": []} {"id": "2401.02411", "title": "What You See is What You GAN: Rendering Every Pixel for High-Fidelity\n Geometry in 3D GANs", "abstract": " 3D-aware Generative Adversarial Networks (GANs) have shown remarkable\nprogress in learning to generate multi-view-consistent images and 3D geometries\nof scenes from collections of 2D images via neural volume rendering. Yet, the\nsignificant memory and computational costs of dense sampling in volume\nrendering have forced 3D GANs to adopt patch-based training or employ\nlow-resolution rendering with post-processing 2D super resolution, which\nsacrifices multiview consistency and the quality of resolved geometry.\nConsequently, 3D GANs have not yet been able to fully resolve the rich 3D\ngeometry present in 2D images. In this work, we propose techniques to scale\nneural volume rendering to the much higher resolution of native 2D images,\nthereby resolving fine-grained 3D geometry with unprecedented detail. Our\napproach employs learning-based samplers for accelerating neural rendering for\n3D GAN training using up to 5 times fewer depth samples. This enables us to\nexplicitly \"render every pixel\" of the full-resolution image during training\nand inference without post-processing superresolution in 2D. Together with our\nstrategy to learn high-quality surface geometry, our method synthesizes\nhigh-resolution 3D geometry and strictly view-consistent images while\nmaintaining image quality on par with baselines relying on post-processing\nsuper resolution. We demonstrate state-of-the-art 3D gemetric quality on FFHQ\nand AFHQ, setting a new standard for unsupervised learning of 3D shapes in 3D\nGANs.\n", "date": "2024-01-04", "categories": []} {"id": "2312.13763", "title": "Align Your Gaussians: Text-to-4D with Dynamic 3D Gaussians and Composed\n Diffusion Models", "abstract": " Text-guided diffusion models have revolutionized image and video generation\nand have also been successfully used for optimization-based 3D object\nsynthesis. Here, we instead focus on the underexplored text-to-4D setting and\nsynthesize dynamic, animated 3D objects using score distillation methods with\nan additional temporal dimension. Compared to previous work, we pursue a novel\ncompositional generation-based approach, and combine text-to-image,\ntext-to-video, and 3D-aware multiview diffusion models to provide feedback\nduring 4D object optimization, thereby simultaneously enforcing temporal\nconsistency, high-quality visual appearance and realistic geometry. Our method,\ncalled Align Your Gaussians (AYG), leverages dynamic 3D Gaussian Splatting with\ndeformation fields as 4D representation. Crucial to AYG is a novel method to\nregularize the distribution of the moving 3D Gaussians and thereby stabilize\nthe optimization and induce motion. We also propose a motion amplification\nmechanism as well as a new autoregressive synthesis scheme to generate and\ncombine multiple 4D sequences for longer generation. These techniques allow us\nto synthesize vivid dynamic scenes, outperform previous work qualitatively and\nquantitatively and achieve state-of-the-art text-to-4D performance. Due to the\nGaussian 4D representation, different 4D animations can be seamlessly combined,\nas we demonstrate. AYG opens up promising avenues for animation, simulation and\ndigital content creation as well as synthetic data generation.\n", "date": "2023-12-21", "categories": []} {"id": "2401.01173", "title": "En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D\n Synthetic Data", "abstract": " We present En3D, an enhanced generative scheme for sculpting high-quality 3D\nhuman avatars. Unlike previous works that rely on scarce 3D datasets or limited\n2D collections with imbalanced viewing angles and imprecise pose priors, our\napproach aims to develop a zero-shot 3D generative scheme capable of producing\nvisually realistic, geometrically accurate and content-wise diverse 3D humans\nwithout relying on pre-existing 3D or 2D assets. To address this challenge, we\nintroduce a meticulously crafted workflow that implements accurate physical\nmodeling to learn the enhanced 3D generative model from synthetic 2D data.\nDuring inference, we integrate optimization modules to bridge the gap between\nrealistic appearances and coarse 3D shapes. Specifically, En3D comprises three\nmodules: a 3D generator that accurately models generalizable 3D humans with\nrealistic appearance from synthesized balanced, diverse, and structured human\nimages; a geometry sculptor that enhances shape quality using multi-view normal\nconstraints for intricate human anatomy; and a texturing module that\ndisentangles explicit texture maps with fidelity and editability, leveraging\nsemantical UV partitioning and a differentiable rasterizer. Experimental\nresults show that our approach significantly outperforms prior works in terms\nof image quality, geometry accuracy and content diversity. We also showcase the\napplicability of our generated avatars for animation and editing, as well as\nthe scalability of our approach for content-style free adaptation.\n", "date": "2024-01-02", "categories": []} {"id": "2305.16283", "title": "CommonScenes: Generating Commonsense 3D Indoor Scenes with Scene Graph\n Diffusion", "abstract": " Controllable scene synthesis aims to create interactive environments for\nvarious industrial use cases. Scene graphs provide a highly suitable interface\nto facilitate these applications by abstracting the scene context in a compact\nmanner. Existing methods, reliant on retrieval from extensive databases or\npre-trained shape embeddings, often overlook scene-object and object-object\nrelationships, leading to inconsistent results due to their limited generation\ncapacity. To address this issue, we present CommonScenes, a fully generative\nmodel that converts scene graphs into corresponding controllable 3D scenes,\nwhich are semantically realistic and conform to commonsense. Our pipeline\nconsists of two branches, one predicting the overall scene layout via a\nvariational auto-encoder and the other generating compatible shapes via latent\ndiffusion, capturing global scene-object and local inter-object relationships\nin the scene graph while preserving shape diversity. The generated scenes can\nbe manipulated by editing the input scene graph and sampling the noise in the\ndiffusion model. Due to lacking a scene graph dataset offering high-quality\nobject-level meshes with relations, we also construct SG-FRONT, enriching the\noff-the-shelf indoor dataset 3D-FRONT with additional scene graph labels.\nExtensive experiments are conducted on SG-FRONT where CommonScenes shows clear\nadvantages over other methods regarding generation consistency, quality, and\ndiversity. Codes and the dataset will be released upon acceptance.\n", "date": "2023-05-25", "categories": []} {"id": "2312.08563", "title": "Efficient-NeRF2NeRF: Streamlining Text-Driven 3D Editing with Multiview\n Correspondence-Enhanced Diffusion Models", "abstract": " The advancement of text-driven 3D content editing has been blessed by the\nprogress from 2D generative diffusion models. However, a major obstacle\nhindering the widespread adoption of 3D content editing is its time-intensive\nprocessing. This challenge arises from the iterative and refining steps\nrequired to achieve consistent 3D outputs from 2D image-based generative\nmodels. Recent state-of-the-art methods typically require optimization time\nranging from tens of minutes to several hours to edit a 3D scene using a single\nGPU. In this work, we propose that by incorporating correspondence\nregularization into diffusion models, the process of 3D editing can be\nsignificantly accelerated. This approach is inspired by the notion that the\nestimated samples during diffusion should be multiview-consistent during the\ndiffusion generation process. By leveraging this multiview consistency, we can\nedit 3D content at a much faster speed. In most scenarios, our proposed\ntechnique brings a 10$\\times$ speed-up compared to the baseline method and\ncompletes the editing of a 3D scene in 2 minutes with comparable quality.\n", "date": "2023-12-13", "categories": []} {"id": "2312.13271", "title": "Repaint123: Fast and High-quality One Image to 3D Generation with\n Progressive Controllable 2D Repainting", "abstract": " Recent one image to 3D generation methods commonly adopt Score Distillation\nSampling (SDS). Despite the impressive results, there are multiple deficiencies\nincluding multi-view inconsistency, over-saturated and over-smoothed textures,\nas well as the slow generation speed. To address these deficiencies, we present\nRepaint123 to alleviate multi-view bias as well as texture degradation and\nspeed up the generation process. The core idea is to combine the powerful image\ngeneration capability of the 2D diffusion model and the texture alignment\nability of the repainting strategy for generating high-quality multi-view\nimages with consistency. We further propose visibility-aware adaptive\nrepainting strength for overlap regions to enhance the generated image quality\nin the repainting process. The generated high-quality and multi-view consistent\nimages enable the use of simple Mean Square Error (MSE) loss for fast 3D\ncontent generation. We conduct extensive experiments and show that our method\nhas a superior ability to generate high-quality 3D content with multi-view\nconsistency and fine textures in 2 minutes from scratch. Our project page is\navailable at https://pku-yuangroup.github.io/repaint123/.\n", "date": "2023-12-20", "categories": []} {"id": "2306.15419", "title": "Freestyle 3D-Aware Portrait Synthesis Based on Compositional Generative\n Priors", "abstract": " Efficiently generating a freestyle 3D portrait with high quality and\n3D-consistency is a promising yet challenging task. The portrait styles\ngenerated by most existing methods are usually restricted by their 3D\ngenerators, which are learned in specific facial datasets, such as FFHQ. To get\nthe diverse 3D portraits, one can build a large-scale multi-style database to\nretrain a 3D-aware generator, or use a off-the-shelf tool to do the style\ntranslation. However, the former is time-consuming due to data collection and\ntraining process, the latter may destroy the multi-view consistency. To tackle\nthis problem, we propose a novel text-driven 3D-aware portrait synthesis\nframework that can generate out-of-distribution portrait styles. Specifically,\nfor a given portrait style prompt, we first composite two generative priors, a\n3D-aware GAN generator and a text-guided image editor, to quickly construct a\nfew-shot stylized portrait set. Then we map the special style domain of this\nset to our proposed 3D latent feature generator and obtain a 3D representation\ncontaining the given style information. Finally we use a pre-trained 3D\nrenderer to generate view-consistent stylized portraits from the 3D\nrepresentation. Extensive experimental results show that our method is capable\nof synthesizing high-quality 3D portraits with specified styles in a few\nminutes, outperforming the state-of-the-art.\n", "date": "2023-06-27", "categories": []} {"id": "2311.16918", "title": "RichDreamer: A Generalizable Normal-Depth Diffusion Model for Detail\n Richness in Text-to-3D", "abstract": " Lifting 2D diffusion for 3D generation is a challenging problem due to the\nlack of geometric prior and the complex entanglement of materials and lighting\nin natural images. Existing methods have shown promise by first creating the\ngeometry through score-distillation sampling (SDS) applied to rendered surface\nnormals, followed by appearance modeling. However, relying on a 2D RGB\ndiffusion model to optimize surface normals is suboptimal due to the\ndistribution discrepancy between natural images and normals maps, leading to\ninstability in optimization. In this paper, recognizing that the normal and\ndepth information effectively describe scene geometry and be automatically\nestimated from images, we propose to learn a generalizable Normal-Depth\ndiffusion model for 3D generation. We achieve this by training on the\nlarge-scale LAION dataset together with the generalizable image-to-depth and\nnormal prior models. In an attempt to alleviate the mixed illumination effects\nin the generated materials, we introduce an albedo diffusion model to impose\ndata-driven constraints on the albedo component. Our experiments show that when\nintegrated into existing text-to-3D pipelines, our models significantly enhance\nthe detail richness, achieving state-of-the-art results. Our project page is\nhttps://aigc3d.github.io/richdreamer/.\n", "date": "2023-11-28", "categories": []} {"id": "2312.08889", "title": "SEEAvatar: Photorealistic Text-to-3D Avatar Generation with Constrained\n Geometry and Appearance", "abstract": " Powered by large-scale text-to-image generation models, text-to-3D avatar\ngeneration has made promising progress. However, most methods fail to produce\nphotorealistic results, limited by imprecise geometry and low-quality\nappearance. Towards more practical avatar generation, we present SEEAvatar, a\nmethod for generating photorealistic 3D avatars from text with SElf-Evolving\nconstraints for decoupled geometry and appearance. For geometry, we propose to\nconstrain the optimized avatar in a decent global shape with a template avatar.\nThe template avatar is initialized with human prior and can be updated by the\noptimized avatar periodically as an evolving template, which enables more\nflexible shape generation. Besides, the geometry is also constrained by the\nstatic human prior in local parts like face and hands to maintain the delicate\nstructures. For appearance generation, we use diffusion model enhanced by\nprompt engineering to guide a physically based rendering pipeline to generate\nrealistic textures. The lightness constraint is applied on the albedo texture\nto suppress incorrect lighting effect. Experiments show that our method\noutperforms previous methods on both global and local geometry and appearance\nquality by a large margin. Since our method can produce high-quality meshes and\ntextures, such assets can be directly applied in classic graphics pipeline for\nrealistic rendering under any lighting condition. Project page at:\nhttps://yoxu515.github.io/SEEAvatar/.\n", "date": "2023-12-13", "categories": []} {"id": "2312.15430", "title": "Make-A-Character: High Quality Text-to-3D Character Generation within\n Minutes", "abstract": " There is a growing demand for customized and expressive 3D characters with\nthe emergence of AI agents and Metaverse, but creating 3D characters using\ntraditional computer graphics tools is a complex and time-consuming task. To\naddress these challenges, we propose a user-friendly framework named\nMake-A-Character (Mach) to create lifelike 3D avatars from text descriptions.\nThe framework leverages the power of large language and vision models for\ntextual intention understanding and intermediate image generation, followed by\na series of human-oriented visual perception and 3D generation modules. Our\nsystem offers an intuitive approach for users to craft controllable, realistic,\nfully-realized 3D characters that meet their expectations within 2 minutes,\nwhile also enabling easy integration with existing CG pipeline for dynamic\nexpressiveness. For more information, please visit the project page at\nhttps://human3daigc.github.io/MACH/.\n", "date": "2023-12-24", "categories": []} {"id": "2312.15980", "title": "HarmonyView: Harmonizing Consistency and Diversity in One-Image-to-3D", "abstract": " Recent progress in single-image 3D generation highlights the importance of\nmulti-view coherency, leveraging 3D priors from large-scale diffusion models\npretrained on Internet-scale images. However, the aspect of novel-view\ndiversity remains underexplored within the research landscape due to the\nambiguity in converting a 2D image into 3D content, where numerous potential\nshapes can emerge. Here, we aim to address this research gap by simultaneously\naddressing both consistency and diversity. Yet, striking a balance between\nthese two aspects poses a considerable challenge due to their inherent\ntrade-offs. This work introduces HarmonyView, a simple yet effective diffusion\nsampling technique adept at decomposing two intricate aspects in single-image\n3D generation: consistency and diversity. This approach paves the way for a\nmore nuanced exploration of the two critical dimensions within the sampling\nprocess. Moreover, we propose a new evaluation metric based on CLIP image and\ntext encoders to comprehensively assess the diversity of the generated views,\nwhich closely aligns with human evaluators' judgments. In experiments,\nHarmonyView achieves a harmonious balance, demonstrating a win-win scenario in\nboth consistency and diversity.\n", "date": "2023-12-26", "categories": []} {"id": "2312.13913", "title": "Paint3D: Paint Anything 3D with Lighting-Less Texture Diffusion Models", "abstract": " This paper presents Paint3D, a novel coarse-to-fine generative framework that\nis capable of producing high-resolution, lighting-less, and diverse 2K UV\ntexture maps for untextured 3D meshes conditioned on text or image inputs. The\nkey challenge addressed is generating high-quality textures without embedded\nillumination information, which allows the textures to be re-lighted or\nre-edited within modern graphics pipelines. To achieve this, our method first\nleverages a pre-trained depth-aware 2D diffusion model to generate\nview-conditional images and perform multi-view texture fusion, producing an\ninitial coarse texture map. However, as 2D models cannot fully represent 3D\nshapes and disable lighting effects, the coarse texture map exhibits incomplete\nareas and illumination artifacts. To resolve this, we train separate UV\nInpainting and UVHD diffusion models specialized for the shape-aware refinement\nof incomplete areas and the removal of illumination artifacts. Through this\ncoarse-to-fine process, Paint3D can produce high-quality 2K UV textures that\nmaintain semantic consistency while being lighting-less, significantly\nadvancing the state-of-the-art in texturing 3D objects.\n", "date": "2023-12-21", "categories": []} {"id": "2312.03763", "title": "Gaussian3Diff: 3D Gaussian Diffusion for 3D Full Head Synthesis and\n Editing", "abstract": " We present a novel framework for generating photorealistic 3D human head and\nsubsequently manipulating and reposing them with remarkable flexibility. The\nproposed approach leverages an implicit function representation of 3D human\nheads, employing 3D Gaussians anchored on a parametric face model. To enhance\nrepresentational capabilities and encode spatial information, we embed a\nlightweight tri-plane payload within each Gaussian rather than directly storing\ncolor and opacity. Additionally, we parameterize the Gaussians in a 2D UV space\nvia a 3DMM, enabling effective utilization of the diffusion model for 3D head\navatar generation. Our method facilitates the creation of diverse and realistic\n3D human heads with fine-grained editing over facial features and expressions.\nExtensive experiments demonstrate the effectiveness of our method.\n", "date": "2023-12-05", "categories": []} {"id": "2312.13324", "title": "ShowRoom3D: Text to High-Quality 3D Room Generation Using 3D Priors", "abstract": " We introduce ShowRoom3D, a three-stage approach for generating high-quality\n3D room-scale scenes from texts. Previous methods using 2D diffusion priors to\noptimize neural radiance fields for generating room-scale scenes have shown\nunsatisfactory quality. This is primarily attributed to the limitations of 2D\npriors lacking 3D awareness and constraints in the training methodology. In\nthis paper, we utilize a 3D diffusion prior, MVDiffusion, to optimize the 3D\nroom-scale scene. Our contributions are in two aspects. Firstly, we propose a\nprogressive view selection process to optimize NeRF. This involves dividing the\ntraining process into three stages, gradually expanding the camera sampling\nscope. Secondly, we propose the pose transformation method in the second stage.\nIt will ensure MVDiffusion provide the accurate view guidance. As a result,\nShowRoom3D enables the generation of rooms with improved structural integrity,\nenhanced clarity from any view, reduced content repetition, and higher\nconsistency across different perspectives. Extensive experiments demonstrate\nthat our method, significantly outperforms state-of-the-art approaches by a\nlarge margin in terms of user study.\n", "date": "2023-12-20", "categories": []} {"id": "2303.11938", "title": "3D-CLFusion: Fast Text-to-3D Rendering with Contrastive Latent Diffusion", "abstract": " We tackle the task of text-to-3D creation with pre-trained latent-based NeRFs\n(NeRFs that generate 3D objects given input latent code). Recent works such as\nDreamFusion and Magic3D have shown great success in generating 3D content using\nNeRFs and text prompts, but the current approach of optimizing a NeRF for every\ntext prompt is 1) extremely time-consuming and 2) often leads to low-resolution\noutputs. To address these challenges, we propose a novel method named\n3D-CLFusion which leverages the pre-trained latent-based NeRFs and performs\nfast 3D content creation in less than a minute. In particular, we introduce a\nlatent diffusion prior network for learning the w latent from the input CLIP\ntext/image embeddings. This pipeline allows us to produce the w latent without\nfurther optimization during inference and the pre-trained NeRF is able to\nperform multi-view high-resolution 3D synthesis based on the latent. We note\nthat the novelty of our model lies in that we introduce contrastive learning\nduring training the diffusion prior which enables the generation of the valid\nview-invariant latent code. We demonstrate through experiments the\neffectiveness of our proposed view-invariant diffusion process for fast\ntext-to-3D creation, e.g., 100 times faster than DreamFusion. We note that our\nmodel is able to serve as the role of a plug-and-play tool for text-to-3D with\npre-trained NeRFs.\n", "date": "2023-03-21", "categories": []} {"id": "2303.15413", "title": "Debiasing Scores and Prompts of 2D Diffusion for View-consistent\n Text-to-3D Generation", "abstract": " Existing score-distilling text-to-3D generation techniques, despite their\nconsiderable promise, often encounter the view inconsistency problem. One of\nthe most notable issues is the Janus problem, where the most canonical view of\nan object (\\textit{e.g}., face or head) appears in other views. In this work,\nwe explore existing frameworks for score-distilling text-to-3D generation and\nidentify the main causes of the view inconsistency problem -- the embedded bias\nof 2D diffusion models. Based on these findings, we propose two approaches to\ndebias the score-distillation frameworks for view-consistent text-to-3D\ngeneration. Our first approach, called score debiasing, involves cutting off\nthe score estimated by 2D diffusion models and gradually increasing the\ntruncation value throughout the optimization process. Our second approach,\ncalled prompt debiasing, identifies conflicting words between user prompts and\nview prompts using a language model, and adjusts the discrepancy between view\nprompts and the viewing direction of an object. Our experimental results show\nthat our methods improve the realism of the generated 3D objects by\nsignificantly reducing artifacts and achieve a good trade-off between\nfaithfulness to the 2D diffusion models and 3D consistency with little\noverhead. Our project page is available\nat~\\url{https://susunghong.github.io/Debiased-Score-Distillation-Sampling/}.\n", "date": "2023-03-27", "categories": []} {"id": "2308.14078", "title": "Sparse3D: Distilling Multiview-Consistent Diffusion for Object\n Reconstruction from Sparse Views", "abstract": " Reconstructing 3D objects from extremely sparse views is a long-standing and\nchallenging problem. While recent techniques employ image diffusion models for\ngenerating plausible images at novel viewpoints or for distilling pre-trained\ndiffusion priors into 3D representations using score distillation sampling\n(SDS), these methods often struggle to simultaneously achieve high-quality,\nconsistent, and detailed results for both novel-view synthesis (NVS) and\ngeometry. In this work, we present Sparse3D, a novel 3D reconstruction method\ntailored for sparse view inputs. Our approach distills robust priors from a\nmultiview-consistent diffusion model to refine a neural radiance field.\nSpecifically, we employ a controller that harnesses epipolar features from\ninput views, guiding a pre-trained diffusion model, such as Stable Diffusion,\nto produce novel-view images that maintain 3D consistency with the input. By\ntapping into 2D priors from powerful image diffusion models, our integrated\nmodel consistently delivers high-quality results, even when faced with\nopen-world objects. To address the blurriness introduced by conventional SDS,\nwe introduce the category-score distillation sampling (C-SDS) to enhance\ndetail. We conduct experiments on CO3DV2 which is a multi-view dataset of\nreal-world objects. Both quantitative and qualitative evaluations demonstrate\nthat our approach outperforms previous state-of-the-art works on the metrics\nregarding NVS and geometry reconstruction.\n", "date": "2023-08-27", "categories": []} {"id": "2312.04875", "title": "MVDD: Multi-View Depth Diffusion Models", "abstract": " Denoising diffusion models have demonstrated outstanding results in 2D image\ngeneration, yet it remains a challenge to replicate its success in 3D shape\ngeneration. In this paper, we propose leveraging multi-view depth, which\nrepresents complex 3D shapes in a 2D data format that is easy to denoise. We\npair this representation with a diffusion model, MVDD, that is capable of\ngenerating high-quality dense point clouds with 20K+ points with fine-grained\ndetails. To enforce 3D consistency in multi-view depth, we introduce an\nepipolar line segment attention that conditions the denoising step for a view\non its neighboring views. Additionally, a depth fusion module is incorporated\ninto diffusion steps to further ensure the alignment of depth maps. When\naugmented with surface reconstruction, MVDD can also produce high-quality 3D\nmeshes. Furthermore, MVDD stands out in other tasks such as depth completion,\nand can serve as a 3D prior, significantly boosting many downstream tasks, such\nas GAN inversion. State-of-the-art results from extensive experiments\ndemonstrate MVDD's excellent ability in 3D shape generation, depth completion,\nand its potential as a 3D prior for downstream tasks.\n", "date": "2023-12-08", "categories": []} {"id": "2312.11666", "title": "HAAR: Text-Conditioned Generative Model of 3D Strand-based Human\n Hairstyles", "abstract": " We present HAAR, a new strand-based generative model for 3D human hairstyles.\nSpecifically, based on textual inputs, HAAR produces 3D hairstyles that could\nbe used as production-level assets in modern computer graphics engines. Current\nAI-based generative models take advantage of powerful 2D priors to reconstruct\n3D content in the form of point clouds, meshes, or volumetric functions.\nHowever, by using the 2D priors, they are intrinsically limited to only\nrecovering the visual parts. Highly occluded hair structures can not be\nreconstructed with those methods, and they only model the ''outer shell'',\nwhich is not ready to be used in physics-based rendering or simulation\npipelines. In contrast, we propose a first text-guided generative method that\nuses 3D hair strands as an underlying representation. Leveraging 2D visual\nquestion-answering (VQA) systems, we automatically annotate synthetic hair\nmodels that are generated from a small set of artist-created hairstyles. This\nallows us to train a latent diffusion model that operates in a common hairstyle\nUV space. In qualitative and quantitative studies, we demonstrate the\ncapabilities of the proposed model and compare it to existing hairstyle\ngeneration approaches.\n", "date": "2023-12-18", "categories": []} {"id": "2312.11774", "title": "Text-Image Conditioned Diffusion for Consistent Text-to-3D Generation", "abstract": " By lifting the pre-trained 2D diffusion models into Neural Radiance Fields\n(NeRFs), text-to-3D generation methods have made great progress. Many\nstate-of-the-art approaches usually apply score distillation sampling (SDS) to\noptimize the NeRF representations, which supervises the NeRF optimization with\npre-trained text-conditioned 2D diffusion models such as Imagen. However, the\nsupervision signal provided by such pre-trained diffusion models only depends\non text prompts and does not constrain the multi-view consistency. To inject\nthe cross-view consistency into diffusion priors, some recent works finetune\nthe 2D diffusion model with multi-view data, but still lack fine-grained view\ncoherence. To tackle this challenge, we incorporate multi-view image conditions\ninto the supervision signal of NeRF optimization, which explicitly enforces\nfine-grained view consistency. With such stronger supervision, our proposed\ntext-to-3D method effectively mitigates the generation of floaters (due to\nexcessive densities) and completely empty spaces (due to insufficient\ndensities). Our quantitative evaluations on the T$^3$Bench dataset demonstrate\nthat our method achieves state-of-the-art performance over existing text-to-3D\nmethods. We will make the code publicly available.\n", "date": "2023-12-19", "categories": []} {"id": "2312.11856", "title": "Self-supervised Learning for Enhancing Geometrical Modeling in 3D-Aware\n Generative Adversarial Network", "abstract": " 3D-aware Generative Adversarial Networks (3D-GANs) currently exhibit\nartifacts in their 3D geometrical modeling, such as mesh imperfections and\nholes. These shortcomings are primarily attributed to the limited availability\nof annotated 3D data, leading to a constrained \"valid latent area\" for\nsatisfactory modeling. To address this, we present a Self-Supervised Learning\n(SSL) technique tailored as an auxiliary loss for any 3D-GAN, designed to\nimprove its 3D geometrical modeling capabilities. Our approach pioneers an\ninversion technique for 3D-GANs, integrating an encoder that performs adaptive\nspatially-varying range operations. Utilizing this inversion, we introduce the\nCyclic Generative Constraint (CGC), aiming to densify the valid latent space.\nThe CGC operates via augmented local latent vectors that maintain the same\ngeometric form, and it imposes constraints on the cycle path outputs,\nspecifically the generator-encoder-generator sequence. This SSL methodology\nseamlessly integrates with the inherent GAN loss, ensuring the integrity of\npre-existing 3D-GAN architectures without necessitating alterations. We\nvalidate our approach with comprehensive experiments across various datasets\nand architectures, underscoring its efficacy. Our project website:\nhttps://3dgan-ssl.github.io\n", "date": "2023-12-19", "categories": []} {"id": "2312.10120", "title": "MVHuman: Tailoring 2D Diffusion with Multi-view Sampling For Realistic\n 3D Human Generation", "abstract": " Recent months have witnessed rapid progress in 3D generation based on\ndiffusion models. Most advances require fine-tuning existing 2D Stable\nDiffsuions into multi-view settings or tedious distilling operations and hence\nfall short of 3D human generation due to the lack of diverse 3D human datasets.\nWe present an alternative scheme named MVHuman to generate human radiance\nfields from text guidance, with consistent multi-view images directly sampled\nfrom pre-trained Stable Diffsuions without any fine-tuning or distilling. Our\ncore is a multi-view sampling strategy to tailor the denoising processes of the\npre-trained network for generating consistent multi-view images. It encompasses\nview-consistent conditioning, replacing the original noises with\n``consistency-guided noises'', optimizing latent codes, as well as utilizing\ncross-view attention layers. With the multi-view images through the sampling\nprocess, we adopt geometry refinement and 3D radiance field generation followed\nby a subsequent neural blending scheme for free-view rendering. Extensive\nexperiments demonstrate the efficacy of our method, as well as its superiority\nto state-of-the-art 3D human generation methods.\n", "date": "2023-12-15", "categories": []} {"id": "2312.11417", "title": "PolyDiff: Generating 3D Polygonal Meshes with Diffusion Models", "abstract": " We introduce PolyDiff, the first diffusion-based approach capable of directly\ngenerating realistic and diverse 3D polygonal meshes. In contrast to methods\nthat use alternate 3D shape representations (e.g. implicit representations),\nour approach is a discrete denoising diffusion probabilistic model that\noperates natively on the polygonal mesh data structure. This enables learning\nof both the geometric properties of vertices and the topological\ncharacteristics of faces. Specifically, we treat meshes as quantized triangle\nsoups, progressively corrupted with categorical noise in the forward diffusion\nphase. In the reverse diffusion phase, a transformer-based denoising network is\ntrained to revert the noising process, restoring the original mesh structure.\nAt inference, new meshes can be generated by applying this denoising network\niteratively, starting with a completely noisy triangle soup. Consequently, our\nmodel is capable of producing high-quality 3D polygonal meshes, ready for\nintegration into downstream 3D workflows. Our extensive experimental analysis\nshows that PolyDiff achieves a significant advantage (avg. FID and JSD\nimprovement of 18.2 and 5.8 respectively) over current state-of-the-art\nmethods.\n", "date": "2023-12-18", "categories": []} {"id": "2312.09682", "title": "Exploring the Feasibility of Generating Realistic 3D Models of\n Endangered Species Using DreamGaussian: An Analysis of Elevation Angle's\n Impact on Model Generation", "abstract": " Many species face the threat of extinction. It's important to study these\nspecies and gather information about them as much as possible to preserve\nbiodiversity. Due to the rarity of endangered species, there is a limited\namount of data available, making it difficult to apply data requiring\ngenerative AI methods to this domain. We aim to study the feasibility of\ngenerating consistent and real-like 3D models of endangered animals using\nlimited data. Such a phenomenon leads us to utilize zero-shot stable diffusion\nmodels that can generate a 3D model out of a single image of the target\nspecies. This paper investigates the intricate relationship between elevation\nangle and the output quality of 3D model generation, focusing on the innovative\napproach presented in DreamGaussian. DreamGaussian, a novel framework utilizing\nGenerative Gaussian Splatting along with novel mesh extraction and refinement\nalgorithms, serves as the focal point of our study. We conduct a comprehensive\nanalysis, analyzing the effect of varying elevation angles on DreamGaussian's\nability to reconstruct 3D scenes accurately. Through an empirical evaluation,\nwe demonstrate how changes in elevation angle impact the generated images'\nspatial coherence, structural integrity, and perceptual realism. We observed\nthat giving a correct elevation angle with the input image significantly\naffects the result of the generated 3D model. We hope this study to be\ninfluential for the usability of AI to preserve endangered animals; while the\npenultimate aim is to obtain a model that can output biologically consistent 3D\nmodels via small samples, the qualitative interpretation of an existing\nstate-of-the-art model such as DreamGaussian will be a step forward in our\ngoal.\n", "date": "2023-12-15", "categories": []} {"id": "2312.08885", "title": "SceneWiz3D: Towards Text-guided 3D Scene Composition", "abstract": " We are witnessing significant breakthroughs in the technology for generating\n3D objects from text. Existing approaches either leverage large text-to-image\nmodels to optimize a 3D representation or train 3D generators on object-centric\ndatasets. Generating entire scenes, however, remains very challenging as a\nscene contains multiple 3D objects, diverse and scattered. In this work, we\nintroduce SceneWiz3D, a novel approach to synthesize high-fidelity 3D scenes\nfrom text. We marry the locality of objects with globality of scenes by\nintroducing a hybrid 3D representation: explicit for objects and implicit for\nscenes. Remarkably, an object, being represented explicitly, can be either\ngenerated from text using conventional text-to-3D approaches, or provided by\nusers. To configure the layout of the scene and automatically place objects, we\napply the Particle Swarm Optimization technique during the optimization\nprocess. Furthermore, it is difficult for certain parts of the scene (e.g.,\ncorners, occlusion) to receive multi-view supervision, leading to inferior\ngeometry. We incorporate an RGBD panorama diffusion model to mitigate it,\nresulting in high-quality geometry. Extensive evaluation supports that our\napproach achieves superior quality over previous approaches, enabling the\ngeneration of detailed and view-consistent 3D scenes.\n", "date": "2023-12-13", "categories": []} {"id": "2312.09242", "title": "Text2Immersion: Generative Immersive Scene with 3D Gaussians", "abstract": " We introduce Text2Immersion, an elegant method for producing high-quality 3D\nimmersive scenes from text prompts. Our proposed pipeline initiates by\nprogressively generating a Gaussian cloud using pre-trained 2D diffusion and\ndepth estimation models. This is followed by a refining stage on the Gaussian\ncloud, interpolating and refining it to enhance the details of the generated\nscene. Distinct from prevalent methods that focus on single object or indoor\nscenes, or employ zoom-out trajectories, our approach generates diverse scenes\nwith various objects, even extending to the creation of imaginary scenes.\nConsequently, Text2Immersion can have wide-ranging implications for various\napplications such as virtual reality, game development, and automated content\ncreation. Extensive evaluations demonstrate that our system surpasses other\nmethods in rendering quality and diversity, further progressing towards\ntext-driven 3D scene generation. We will make the source code publicly\naccessible at the project page.\n", "date": "2023-12-14", "categories": []} {"id": "2312.09246", "title": "SHAP-EDITOR: Instruction-guided Latent 3D Editing in Seconds", "abstract": " We propose a novel feed-forward 3D editing framework called Shap-Editor.\nPrior research on editing 3D objects primarily concentrated on editing\nindividual objects by leveraging off-the-shelf 2D image editing networks. This\nis achieved via a process called distillation, which transfers knowledge from\nthe 2D network to 3D assets. Distillation necessitates at least tens of minutes\nper asset to attain satisfactory editing results, and is thus not very\npractical. In contrast, we ask whether 3D editing can be carried out directly\nby a feed-forward network, eschewing test-time optimisation. In particular, we\nhypothesise that editing can be greatly simplified by first encoding 3D objects\nin a suitable latent space. We validate this hypothesis by building upon the\nlatent space of Shap-E. We demonstrate that direct 3D editing in this space is\npossible and efficient by building a feed-forward editor network that only\nrequires approximately one second per edit. Our experiments show that\nShap-Editor generalises well to both in-distribution and out-of-distribution 3D\nassets with different prompts, exhibiting comparable performance with methods\nthat carry out test-time optimisation for each edited instance.\n", "date": "2023-12-14", "categories": []} {"id": "2312.08094", "title": "3DGEN: A GAN-based approach for generating novel 3D models from image\n data", "abstract": " The recent advances in text and image synthesis show a great promise for the\nfuture of generative models in creative fields. However, a less explored area\nis the one of 3D model generation, with a lot of potential applications to game\ndesign, video production, and physical product design. In our paper, we present\n3DGEN, a model that leverages the recent work on both Neural Radiance Fields\nfor object reconstruction and GAN-based image generation. We show that the\nproposed architecture can generate plausible meshes for objects of the same\ncategory as the training images and compare the resulting meshes with the\nstate-of-the-art baselines, leading to visible uplifts in generation quality.\n", "date": "2023-12-13", "categories": []} {"id": "2312.07231", "title": "Fast Training of Diffusion Transformer with Extreme Masking for 3D Point\n Clouds Generation", "abstract": " Diffusion Transformers have recently shown remarkable effectiveness in\ngenerating high-quality 3D point clouds. However, training voxel-based\ndiffusion models for high-resolution 3D voxels remains prohibitively expensive\ndue to the cubic complexity of attention operators, which arises from the\nadditional dimension of voxels. Motivated by the inherent redundancy of 3D\ncompared to 2D, we propose FastDiT-3D, a novel masked diffusion transformer\ntailored for efficient 3D point cloud generation, which greatly reduces\ntraining costs. Specifically, we draw inspiration from masked autoencoders to\ndynamically operate the denoising process on masked voxelized point clouds. We\nalso propose a novel voxel-aware masking strategy to adaptively aggregate\nbackground/foreground information from voxelized point clouds. Our method\nachieves state-of-the-art performance with an extreme masking ratio of nearly\n99%. Moreover, to improve multi-category 3D generation, we introduce\nMixture-of-Expert (MoE) in 3D diffusion model. Each category can learn a\ndistinct diffusion path with different experts, relieving gradient conflict.\nExperimental results on the ShapeNet dataset demonstrate that our method\nachieves state-of-the-art high-fidelity and diverse 3D point cloud generation\nperformance. Our FastDiT-3D improves 1-Nearest Neighbor Accuracy and Coverage\nmetrics when generating 128-resolution voxel point clouds, using only 6.5% of\nthe original training cost.\n", "date": "2023-12-12", "categories": []} {"id": "2312.06655", "title": "Sherpa3D: Boosting High-Fidelity Text-to-3D Generation via Coarse 3D\n Prior", "abstract": " Recently, 3D content creation from text prompts has demonstrated remarkable\nprogress by utilizing 2D and 3D diffusion models. While 3D diffusion models\nensure great multi-view consistency, their ability to generate high-quality and\ndiverse 3D assets is hindered by the limited 3D data. In contrast, 2D diffusion\nmodels find a distillation approach that achieves excellent generalization and\nrich details without any 3D data. However, 2D lifting methods suffer from\ninherent view-agnostic ambiguity thereby leading to serious multi-face Janus\nissues, where text prompts fail to provide sufficient guidance to learn\ncoherent 3D results. Instead of retraining a costly viewpoint-aware model, we\nstudy how to fully exploit easily accessible coarse 3D knowledge to enhance the\nprompts and guide 2D lifting optimization for refinement. In this paper, we\npropose Sherpa3D, a new text-to-3D framework that achieves high-fidelity,\ngeneralizability, and geometric consistency simultaneously. Specifically, we\ndesign a pair of guiding strategies derived from the coarse 3D prior generated\nby the 3D diffusion model: a structural guidance for geometric fidelity and a\nsemantic guidance for 3D coherence. Employing the two types of guidance, the 2D\ndiffusion model enriches the 3D content with diversified and high-quality\nresults. Extensive experiments show the superiority of our Sherpa3D over the\nstate-of-the-art text-to-3D methods in terms of quality and 3D consistency.\n", "date": "2023-12-11", "categories": []} {"id": "2312.06663", "title": "CAD: Photorealistic 3D Generation via Adversarial Distillation", "abstract": " The increased demand for 3D data in AR/VR, robotics and gaming applications,\ngave rise to powerful generative pipelines capable of synthesizing high-quality\n3D objects. Most of these models rely on the Score Distillation Sampling (SDS)\nalgorithm to optimize a 3D representation such that the rendered image\nmaintains a high likelihood as evaluated by a pre-trained diffusion model.\nHowever, finding a correct mode in the high-dimensional distribution produced\nby the diffusion model is challenging and often leads to issues such as\nover-saturation, over-smoothing, and Janus-like artifacts. In this paper, we\npropose a novel learning paradigm for 3D synthesis that utilizes pre-trained\ndiffusion models. Instead of focusing on mode-seeking, our method directly\nmodels the distribution discrepancy between multi-view renderings and diffusion\npriors in an adversarial manner, which unlocks the generation of high-fidelity\nand photorealistic 3D content, conditioned on a single image and prompt.\nMoreover, by harnessing the latent space of GANs and expressive diffusion model\npriors, our method facilitates a wide variety of 3D applications including\nsingle-view reconstruction, high diversity generation and continuous 3D\ninterpolation in the open domain. The experiments demonstrate the superiority\nof our pipeline compared to previous works in terms of generation quality and\ndiversity.\n", "date": "2023-12-11", "categories": []} {"id": "2312.04806", "title": "RL Dreams: Policy Gradient Optimization for Score Distillation based 3D\n Generation", "abstract": " 3D generation has rapidly accelerated in the past decade owing to the\nprogress in the field of generative modeling. Score Distillation Sampling (SDS)\nbased rendering has improved 3D asset generation to a great extent. Further,\nthe recent work of Denoising Diffusion Policy Optimization (DDPO) demonstrates\nthat the diffusion process is compatible with policy gradient methods and has\nbeen demonstrated to improve the 2D diffusion models using an aesthetic scoring\nfunction. We first show that this aesthetic scorer acts as a strong guide for a\nvariety of SDS-based methods and demonstrates its effectiveness in text-to-3D\nsynthesis. Further, we leverage the DDPO approach to improve the quality of the\n3D rendering obtained from 2D diffusion models. Our approach, DDPO3D, employs\nthe policy gradient method in tandem with aesthetic scoring. To the best of our\nknowledge, this is the first method that extends policy gradient methods to 3D\nscore-based rendering and shows improvement across SDS-based methods such as\nDreamGaussian, which are currently driving research in text-to-3D synthesis.\nOur approach is compatible with score distillation-based methods, which would\nfacilitate the integration of diverse reward functions into the generative\nprocess. Our project page can be accessed via https://ddpo3d.github.io.\n", "date": "2023-12-08", "categories": []} {"id": "2312.04820", "title": "Learn to Optimize Denoising Scores for 3D Generation: A Unified and\n Improved Diffusion Prior on NeRF and 3D Gaussian Splatting", "abstract": " We propose a unified framework aimed at enhancing the diffusion priors for 3D\ngeneration tasks. Despite the critical importance of these tasks, existing\nmethodologies often struggle to generate high-caliber results. We begin by\nexamining the inherent limitations in previous diffusion priors. We identify a\ndivergence between the diffusion priors and the training procedures of\ndiffusion models that substantially impairs the quality of 3D generation. To\naddress this issue, we propose a novel, unified framework that iteratively\noptimizes both the 3D model and the diffusion prior. Leveraging the different\nlearnable parameters of the diffusion prior, our approach offers multiple\nconfigurations, affording various trade-offs between performance and\nimplementation complexity. Notably, our experimental results demonstrate that\nour method markedly surpasses existing techniques, establishing new\nstate-of-the-art in the realm of text-to-3D generation. Furthermore, our\napproach exhibits impressive performance on both NeRF and the newly introduced\n3D Gaussian Splatting backbones. Additionally, our framework yields insightful\ncontributions to the understanding of recent score distillation methods, such\nas the VSD and DDS loss.\n", "date": "2023-12-08", "categories": []} {"id": "2312.04963", "title": "Text-to-3D Generation with Bidirectional Diffusion using both 2D and 3D\n priors", "abstract": " Most 3D generation research focuses on up-projecting 2D foundation models\ninto the 3D space, either by minimizing 2D Score Distillation Sampling (SDS)\nloss or fine-tuning on multi-view datasets. Without explicit 3D priors, these\nmethods often lead to geometric anomalies and multi-view inconsistency.\nRecently, researchers have attempted to improve the genuineness of 3D objects\nby directly training on 3D datasets, albeit at the cost of low-quality texture\ngeneration due to the limited texture diversity in 3D datasets. To harness the\nadvantages of both approaches, we propose Bidirectional Diffusion(BiDiff), a\nunified framework that incorporates both a 3D and a 2D diffusion process, to\npreserve both 3D fidelity and 2D texture richness, respectively. Moreover, as a\nsimple combination may yield inconsistent generation results, we further bridge\nthem with novel bidirectional guidance. In addition, our method can be used as\nan initialization of optimization-based models to further improve the quality\nof 3D model and efficiency of optimization, reducing the generation process\nfrom 3.4 hours to 20 minutes. Experimental results have shown that our model\nachieves high-quality, diverse, and scalable 3D generation. Project website:\nhttps://bidiff.github.io/.\n", "date": "2023-12-07", "categories": []} {"id": "2312.05208", "title": "ControlRoom3D: Room Generation using Semantic Proxy Rooms", "abstract": " Manually creating 3D environments for AR/VR applications is a complex process\nrequiring expert knowledge in 3D modeling software. Pioneering works facilitate\nthis process by generating room meshes conditioned on textual style\ndescriptions. Yet, many of these automatically generated 3D meshes do not\nadhere to typical room layouts, compromising their plausibility, e.g., by\nplacing several beds in one bedroom. To address these challenges, we present\nControlRoom3D, a novel method to generate high-quality room meshes. Central to\nour approach is a user-defined 3D semantic proxy room that outlines a rough\nroom layout based on semantic bounding boxes and a textual description of the\noverall room style. Our key insight is that when rendered to 2D, this 3D\nrepresentation provides valuable geometric and semantic information to control\npowerful 2D models to generate 3D consistent textures and geometry that aligns\nwell with the proxy room. Backed up by an extensive study including\nquantitative metrics and qualitative user evaluations, our method generates\ndiverse and globally plausible 3D room meshes, thus empowering users to design\n3D rooms effortlessly without specialized knowledge.\n", "date": "2023-12-08", "categories": []} {"id": "2302.01330", "title": "SceneDreamer: Unbounded 3D Scene Generation from 2D Image Collections", "abstract": " In this work, we present SceneDreamer, an unconditional generative model for\nunbounded 3D scenes, which synthesizes large-scale 3D landscapes from random\nnoise. Our framework is learned from in-the-wild 2D image collections only,\nwithout any 3D annotations. At the core of SceneDreamer is a principled\nlearning paradigm comprising 1) an efficient yet expressive 3D scene\nrepresentation, 2) a generative scene parameterization, and 3) an effective\nrenderer that can leverage the knowledge from 2D images. Our approach begins\nwith an efficient bird's-eye-view (BEV) representation generated from simplex\nnoise, which includes a height field for surface elevation and a semantic field\nfor detailed scene semantics. This BEV scene representation enables 1)\nrepresenting a 3D scene with quadratic complexity, 2) disentangled geometry and\nsemantics, and 3) efficient training. Moreover, we propose a novel generative\nneural hash grid to parameterize the latent space based on 3D positions and\nscene semantics, aiming to encode generalizable features across various scenes.\nLastly, a neural volumetric renderer, learned from 2D image collections through\nadversarial training, is employed to produce photorealistic images. Extensive\nexperiments demonstrate the effectiveness of SceneDreamer and superiority over\nstate-of-the-art methods in generating vivid yet diverse unbounded 3D worlds.\n", "date": "2023-02-02", "categories": []} {"id": "2312.03869", "title": "Inpaint3D: 3D Scene Content Generation using 2D Inpainting Diffusion", "abstract": " This paper presents a novel approach to inpainting 3D regions of a scene,\ngiven masked multi-view images, by distilling a 2D diffusion model into a\nlearned 3D scene representation (e.g. a NeRF). Unlike 3D generative methods\nthat explicitly condition the diffusion model on camera pose or multi-view\ninformation, our diffusion model is conditioned only on a single masked 2D\nimage. Nevertheless, we show that this 2D diffusion model can still serve as a\ngenerative prior in a 3D multi-view reconstruction problem where we optimize a\nNeRF using a combination of score distillation sampling and NeRF reconstruction\nlosses. Predicted depth is used as additional supervision to encourage accurate\ngeometry. We compare our approach to 3D inpainting methods that focus on object\nremoval. Because our method can generate content to fill any 3D masked region,\nwe additionally demonstrate 3D object completion, 3D object replacement, and 3D\nscene completion.\n", "date": "2023-12-06", "categories": []} {"id": "2312.04543", "title": "HyperDreamer: Hyper-Realistic 3D Content Generation and Editing from a\n Single Image", "abstract": " 3D content creation from a single image is a long-standing yet highly\ndesirable task. Recent advances introduce 2D diffusion priors, yielding\nreasonable results. However, existing methods are not hyper-realistic enough\nfor post-generation usage, as users cannot view, render and edit the resulting\n3D content from a full range. To address these challenges, we introduce\nHyperDreamer with several key designs and appealing properties: 1) Viewable:\n360 degree mesh modeling with high-resolution textures enables the creation of\nvisually compelling 3D models from a full range of observation points. 2)\nRenderable: Fine-grained semantic segmentation and data-driven priors are\nincorporated as guidance to learn reasonable albedo, roughness, and specular\nproperties of the materials, enabling semantic-aware arbitrary material\nestimation. 3) Editable: For a generated model or their own data, users can\ninteractively select any region via a few clicks and efficiently edit the\ntexture with text-based guidance. Extensive experiments demonstrate the\neffectiveness of HyperDreamer in modeling region-aware materials with\nhigh-resolution textures and enabling user-friendly editing. We believe that\nHyperDreamer holds promise for advancing 3D content creation and finding\napplications in various domains.\n", "date": "2023-12-07", "categories": []} {"id": "2312.04559", "title": "PrimDiffusion: Volumetric Primitives Diffusion for 3D Human Generation", "abstract": " We present PrimDiffusion, the first diffusion-based framework for 3D human\ngeneration. Devising diffusion models for 3D human generation is difficult due\nto the intensive computational cost of 3D representations and the articulated\ntopology of 3D humans. To tackle these challenges, our key insight is operating\nthe denoising diffusion process directly on a set of volumetric primitives,\nwhich models the human body as a number of small volumes with radiance and\nkinematic information. This volumetric primitives representation marries the\ncapacity of volumetric representations with the efficiency of primitive-based\nrendering. Our PrimDiffusion framework has three appealing properties: 1)\ncompact and expressive parameter space for the diffusion model, 2) flexible 3D\nrepresentation that incorporates human prior, and 3) decoder-free rendering for\nefficient novel-view and novel-pose synthesis. Extensive experiments validate\nthat PrimDiffusion outperforms state-of-the-art methods in 3D human generation.\nNotably, compared to GAN-based methods, our PrimDiffusion supports real-time\nrendering of high-quality 3D humans at a resolution of $512\\times512$ once the\ndenoising process is done. We also demonstrate the flexibility of our framework\non training-free conditional generation such as texture transfer and 3D\ninpainting.\n", "date": "2023-12-07", "categories": []} {"id": "2312.02189", "title": "StableDreamer: Taming Noisy Score Distillation Sampling for Text-to-3D", "abstract": " In the realm of text-to-3D generation, utilizing 2D diffusion models through\nscore distillation sampling (SDS) frequently leads to issues such as blurred\nappearances and multi-faced geometry, primarily due to the intrinsically noisy\nnature of the SDS loss. Our analysis identifies the core of these challenges as\nthe interaction among noise levels in the 2D diffusion process, the\narchitecture of the diffusion network, and the 3D model representation. To\novercome these limitations, we present StableDreamer, a methodology\nincorporating three advances. First, inspired by InstructNeRF2NeRF, we\nformalize the equivalence of the SDS generative prior and a simple supervised\nL2 reconstruction loss. This finding provides a novel tool to debug SDS, which\nwe use to show the impact of time-annealing noise levels on reducing\nmulti-faced geometries. Second, our analysis shows that while image-space\ndiffusion contributes to geometric precision, latent-space diffusion is crucial\nfor vivid color rendition. Based on this observation, StableDreamer introduces\na two-stage training strategy that effectively combines these aspects,\nresulting in high-fidelity 3D models. Third, we adopt an anisotropic 3D\nGaussians representation, replacing Neural Radiance Fields (NeRFs), to enhance\nthe overall quality, reduce memory usage during training, and accelerate\nrendering speeds, and better capture semi-transparent objects. StableDreamer\nreduces multi-face geometries, generates fine details, and converges stably.\n", "date": "2023-12-02", "categories": []} {"id": "2312.02201", "title": "ImageDream: Image-Prompt Multi-view Diffusion for 3D Generation", "abstract": " We introduce \"ImageDream,\" an innovative image-prompt, multi-view diffusion\nmodel for 3D object generation. ImageDream stands out for its ability to\nproduce 3D models of higher quality compared to existing state-of-the-art,\nimage-conditioned methods. Our approach utilizes a canonical camera\ncoordination for the objects in images, improving visual geometry accuracy. The\nmodel is designed with various levels of control at each block inside the\ndiffusion model based on the input image, where global control shapes the\noverall object layout and local control fine-tunes the image details. The\neffectiveness of ImageDream is demonstrated through extensive evaluations using\na standard prompt list. For more information, visit our project page at\nhttps://Image-Dream.github.io.\n", "date": "2023-12-02", "categories": []} {"id": "2306.00783", "title": "FaceDNeRF: Semantics-Driven Face Reconstruction, Prompt Editing and\n Relighting with Diffusion Models", "abstract": " The ability to create high-quality 3D faces from a single image has become\nincreasingly important with wide applications in video conferencing, AR/VR, and\nadvanced video editing in movie industries. In this paper, we propose Face\nDiffusion NeRF (FaceDNeRF), a new generative method to reconstruct high-quality\nFace NeRFs from single images, complete with semantic editing and relighting\ncapabilities. FaceDNeRF utilizes high-resolution 3D GAN inversion and expertly\ntrained 2D latent-diffusion model, allowing users to manipulate and construct\nFace NeRFs in zero-shot learning without the need for explicit 3D data. With\ncarefully designed illumination and identity preserving loss, as well as\nmulti-modal pre-training, FaceDNeRF offers users unparalleled control over the\nediting process enabling them to create and edit face NeRFs using just\nsingle-view images, text prompts, and explicit target lighting. The advanced\nfeatures of FaceDNeRF have been designed to produce more impressive results\nthan existing 2D editing approaches that rely on 2D segmentation maps for\neditable attributes. Experiments show that our FaceDNeRF achieves exceptionally\nrealistic results and unprecedented flexibility in editing compared with\nstate-of-the-art 3D face reconstruction and editing methods. Our code will be\navailable at https://github.com/BillyXYB/FaceDNeRF.\n", "date": "2023-06-01", "categories": []} {"id": "2311.11284", "title": "LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval\n Score Matching", "abstract": " The recent advancements in text-to-3D generation mark a significant milestone\nin generative models, unlocking new possibilities for creating imaginative 3D\nassets across various real-world scenarios. While recent advancements in\ntext-to-3D generation have shown promise, they often fall short in rendering\ndetailed and high-quality 3D models. This problem is especially prevalent as\nmany methods base themselves on Score Distillation Sampling (SDS). This paper\nidentifies a notable deficiency in SDS, that it brings inconsistent and\nlow-quality updating direction for the 3D model, causing the over-smoothing\neffect. To address this, we propose a novel approach called Interval Score\nMatching (ISM). ISM employs deterministic diffusing trajectories and utilizes\ninterval-based score matching to counteract over-smoothing. Furthermore, we\nincorporate 3D Gaussian Splatting into our text-to-3D generation pipeline.\nExtensive experiments show that our model largely outperforms the\nstate-of-the-art in quality and training efficiency.\n", "date": "2023-11-19", "categories": []} {"id": "2311.17971", "title": "GeoDream: Disentangling 2D and Geometric Priors for High-Fidelity and\n Consistent 3D Generation", "abstract": " Text-to-3D generation by distilling pretrained large-scale text-to-image\ndiffusion models has shown great promise but still suffers from inconsistent 3D\ngeometric structures (Janus problems) and severe artifacts. The aforementioned\nproblems mainly stem from 2D diffusion models lacking 3D awareness during the\nlifting. In this work, we present GeoDream, a novel method that incorporates\nexplicit generalized 3D priors with 2D diffusion priors to enhance the\ncapability of obtaining unambiguous 3D consistent geometric structures without\nsacrificing diversity or fidelity. Specifically, we first utilize a multi-view\ndiffusion model to generate posed images and then construct cost volume from\nthe predicted image, which serves as native 3D geometric priors, ensuring\nspatial consistency in 3D space. Subsequently, we further propose to harness 3D\ngeometric priors to unlock the great potential of 3D awareness in 2D diffusion\npriors via a disentangled design. Notably, disentangling 2D and 3D priors\nallows us to refine 3D geometric priors further. We justify that the refined 3D\ngeometric priors aid in the 3D-aware capability of 2D diffusion priors, which\nin turn provides superior guidance for the refinement of 3D geometric priors.\nOur numerical and visual comparisons demonstrate that GeoDream generates more\n3D consistent textured meshes with high-resolution realistic renderings (i.e.,\n1024 $\\times$ 1024) and adheres more closely to semantic coherence.\n", "date": "2023-11-29", "categories": []} {"id": "2312.02136", "title": "BerfScene: Bev-conditioned Equivariant Radiance Fields for Infinite 3D\n Scene Generation", "abstract": " Generating large-scale 3D scenes cannot simply apply existing 3D object\nsynthesis technique since 3D scenes usually hold complex spatial configurations\nand consist of a number of objects at varying scales. We thus propose a\npractical and efficient 3D representation that incorporates an equivariant\nradiance field with the guidance of a bird's-eye view (BEV) map. Concretely,\nobjects of synthesized 3D scenes could be easily manipulated through steering\nthe corresponding BEV maps. Moreover, by adequately incorporating positional\nencoding and low-pass filters into the generator, the representation becomes\nequivariant to the given BEV map. Such equivariance allows us to produce\nlarge-scale, even infinite-scale, 3D scenes via synthesizing local scenes and\nthen stitching them with smooth consistency. Extensive experiments on 3D scene\ndatasets demonstrate the effectiveness of our approach. Our project website is\nat https://zqh0253.github.io/BerfScene/.\n", "date": "2023-12-04", "categories": []} {"id": "2311.17618", "title": "ShapeGPT: 3D Shape Generation with A Unified Multi-modal Language Model", "abstract": " The advent of large language models, enabling flexibility through\ninstruction-driven approaches, has revolutionized many traditional generative\ntasks, but large models for 3D data, particularly in comprehensively handling\n3D shapes with other modalities, are still under-explored. By achieving\ninstruction-based shape generations, versatile multimodal generative shape\nmodels can significantly benefit various fields like 3D virtual construction\nand network-aided design. In this work, we present ShapeGPT, a shape-included\nmulti-modal framework to leverage strong pre-trained language models to address\nmultiple shape-relevant tasks. Specifically, ShapeGPT employs a\nword-sentence-paragraph framework to discretize continuous shapes into shape\nwords, further assembles these words for shape sentences, as well as integrates\nshape with instructional text for multi-modal paragraphs. To learn this\nshape-language model, we use a three-stage training scheme, including shape\nrepresentation, multimodal alignment, and instruction-based generation, to\nalign shape-language codebooks and learn the intricate correlations among these\nmodalities. Extensive experiments demonstrate that ShapeGPT achieves comparable\nperformance across shape-relevant tasks, including text-to-shape,\nshape-to-text, shape completion, and shape editing.\n", "date": "2023-11-29", "categories": []} {"id": "2312.00063", "title": "MoMask: Generative Masked Modeling of 3D Human Motions", "abstract": " We introduce MoMask, a novel masked modeling framework for text-driven 3D\nhuman motion generation. In MoMask, a hierarchical quantization scheme is\nemployed to represent human motion as multi-layer discrete motion tokens with\nhigh-fidelity details. Starting at the base layer, with a sequence of motion\ntokens obtained by vector quantization, the residual tokens of increasing\norders are derived and stored at the subsequent layers of the hierarchy. This\nis consequently followed by two distinct bidirectional transformers. For the\nbase-layer motion tokens, a Masked Transformer is designated to predict\nrandomly masked motion tokens conditioned on text input at training stage.\nDuring generation (i.e. inference) stage, starting from an empty sequence, our\nMasked Transformer iteratively fills up the missing tokens; Subsequently, a\nResidual Transformer learns to progressively predict the next-layer tokens\nbased on the results from current layer. Extensive experiments demonstrate that\nMoMask outperforms the state-of-art methods on the text-to-motion generation\ntask, with an FID of 0.045 (vs e.g. 0.141 of T2M-GPT) on the HumanML3D dataset,\nand 0.228 (vs 0.514) on KIT-ML, respectively. MoMask can also be seamlessly\napplied in related tasks without further model fine-tuning, such as text-guided\ntemporal inpainting.\n", "date": "2023-11-29", "categories": []} {"id": "2312.00375", "title": "Text-Guided 3D Face Synthesis -- From Generation to Editing", "abstract": " Text-guided 3D face synthesis has achieved remarkable results by leveraging\ntext-to-image (T2I) diffusion models. However, most existing works focus solely\non the direct generation, ignoring the editing, restricting them from\nsynthesizing customized 3D faces through iterative adjustments. In this paper,\nwe propose a unified text-guided framework from face generation to editing. In\nthe generation stage, we propose a geometry-texture decoupled generation to\nmitigate the loss of geometric details caused by coupling. Besides, decoupling\nenables us to utilize the generated geometry as a condition for texture\ngeneration, yielding highly geometry-texture aligned results. We further employ\na fine-tuned texture diffusion model to enhance texture quality in both RGB and\nYUV space. In the editing stage, we first employ a pre-trained diffusion model\nto update facial geometry or texture based on the texts. To enable sequential\nediting, we introduce a UV domain consistency preservation regularization,\npreventing unintentional changes to irrelevant facial attributes. Besides, we\npropose a self-guided consistency weight strategy to improve editing efficacy\nwhile preserving consistency. Through comprehensive experiments, we showcase\nour method's superiority in face synthesis. Project page:\nhttps://faceg2e.github.io/.\n", "date": "2023-12-01", "categories": []} {"id": "2304.00916", "title": "DreamAvatar: Text-and-Shape Guided 3D Human Avatar Generation via\n Diffusion Models", "abstract": " We present DreamAvatar, a text-and-shape guided framework for generating\nhigh-quality 3D human avatars with controllable poses. While encouraging\nresults have been reported by recent methods on text-guided 3D common object\ngeneration, generating high-quality human avatars remains an open challenge due\nto the complexity of the human body's shape, pose, and appearance. We propose\nDreamAvatar to tackle this challenge, which utilizes a trainable NeRF for\npredicting density and color for 3D points and pretrained text-to-image\ndiffusion models for providing 2D self-supervision. Specifically, we leverage\nthe SMPL model to provide shape and pose guidance for the generation. We\nintroduce a dual-observation-space design that involves the joint optimization\nof a canonical space and a posed space that are related by a learnable\ndeformation field. This facilitates the generation of more complete textures\nand geometry faithful to the target pose. We also jointly optimize the losses\ncomputed from the full body and from the zoomed-in 3D head to alleviate the\ncommon multi-face ''Janus'' problem and improve facial details in the generated\navatars. Extensive evaluations demonstrate that DreamAvatar significantly\noutperforms existing methods, establishing a new state-of-the-art for\ntext-and-shape guided 3D human avatar generation.\n", "date": "2023-04-03", "categories": []} {"id": "2306.00519", "title": "DiffInDScene: Diffusion-based High-Quality 3D Indoor Scene Generation", "abstract": " We present DiffInDScene, a novel framework for tackling the problem of\nhigh-quality 3D indoor scene generation, which is challenging due to the\ncomplexity and diversity of the indoor scene geometry. Although diffusion-based\ngenerative models have previously demonstrated impressive performance in image\ngeneration and object-level 3D generation, they have not yet been applied to\nroom-level 3D generation due to their computationally intensive costs. In\nDiffInDScene, we propose a cascaded 3D diffusion pipeline that is efficient and\npossesses strong generative performance for Truncated Signed Distance Function\n(TSDF). The whole pipeline is designed to run on a sparse occupancy space in a\ncoarse-to-fine fashion. Inspired by KinectFusion's incremental alignment and\nfusion of local TSDF volumes, we propose a diffusion-based SDF fusion approach\nthat iteratively diffuses and fuses local TSDF volumes, facilitating the\ngeneration of an entire room environment. The generated results demonstrate\nthat our work is capable to achieve high-quality room generation directly in\nthree-dimensional space, starting from scratch. In addition to the scene\ngeneration, the final part of DiffInDScene can be used as a post-processing\nmodule to refine the 3D reconstruction results from multi-view stereo.\nAccording to the user study, the mesh quality generated by our DiffInDScene can\neven outperform the ground truth mesh provided by ScanNet. Please visit our\nproject page for the latest progress and demonstrations:\nhttps://github.com/AkiraHero/diffindscene.\n", "date": "2023-06-01", "categories": []} {"id": "2309.04917", "title": "Editing 3D Scenes via Text Prompts without Retraining", "abstract": " Numerous diffusion models have recently been applied to image synthesis and\nediting. However, editing 3D scenes is still in its early stages. It poses\nvarious challenges, such as the requirement to design specific methods for\ndifferent editing types, retraining new models for various 3D scenes, and the\nabsence of convenient human interaction during editing. To tackle these issues,\nwe introduce a text-driven editing method, termed DN2N, which allows for the\ndirect acquisition of a NeRF model with universal editing capabilities,\neliminating the requirement for retraining. Our method employs off-the-shelf\ntext-based editing models of 2D images to modify the 3D scene images, followed\nby a filtering process to discard poorly edited images that disrupt 3D\nconsistency. We then consider the remaining inconsistency as a problem of\nremoving noise perturbation, which can be solved by generating training data\nwith similar perturbation characteristics for training. We further propose\ncross-view regularization terms to help the generalized NeRF model mitigate\nthese perturbations. Our text-driven method allows users to edit a 3D scene\nwith their desired description, which is more friendly, intuitive, and\npractical than prior works. Empirical results show that our method achieves\nmultiple editing types, including but not limited to appearance editing,\nweather transition, material changing, and style transfer. Most importantly,\nour method generalizes well with editing abilities shared among a set of model\nparameters without requiring a customized editing model for some specific\nscenes, thus inferring novel views with editing effects directly from user\ninput. The project website is available at https://sk-fun.fun/DN2N\n", "date": "2023-09-10", "categories": []} {"id": "2311.01714", "title": "EXIM: A Hybrid Explicit-Implicit Representation for Text-Guided 3D Shape\n Generation", "abstract": " This paper presents a new text-guided technique for generating 3D shapes. The\ntechnique leverages a hybrid 3D shape representation, namely EXIM, combining\nthe strengths of explicit and implicit representations. Specifically, the\nexplicit stage controls the topology of the generated 3D shapes and enables\nlocal modifications, whereas the implicit stage refines the shape and paints it\nwith plausible colors. Also, the hybrid approach separates the shape and color\nand generates color conditioned on shape to ensure shape-color consistency.\nUnlike the existing state-of-the-art methods, we achieve high-fidelity shape\ngeneration from natural-language descriptions without the need for\ntime-consuming per-shape optimization or reliance on human-annotated texts\nduring training or test-time optimization. Further, we demonstrate the\napplicability of our approach to generate indoor scenes with consistent styles\nusing text-induced 3D shapes. Through extensive experiments, we demonstrate the\ncompelling quality of our results and the high coherency of our generated\nshapes with the input texts, surpassing the performance of existing methods by\na significant margin. Codes and models are released at\nhttps://github.com/liuzhengzhe/EXIM.\n", "date": "2023-11-03", "categories": []} {"id": "2310.01406", "title": "HumanNorm: Learning Normal Diffusion Model for High-quality and\n Realistic 3D Human Generation", "abstract": " Recent text-to-3D methods employing diffusion models have made significant\nadvancements in 3D human generation. However, these approaches face challenges\ndue to the limitations of text-to-image diffusion models, which lack an\nunderstanding of 3D structures. Consequently, these methods struggle to achieve\nhigh-quality human generation, resulting in smooth geometry and cartoon-like\nappearances. In this paper, we propose HumanNorm, a novel approach for\nhigh-quality and realistic 3D human generation. The main idea is to enhance the\nmodel's 2D perception of 3D geometry by learning a normal-adapted diffusion\nmodel and a normal-aligned diffusion model. The normal-adapted diffusion model\ncan generate high-fidelity normal maps corresponding to user prompts with\nview-dependent and body-aware text. The normal-aligned diffusion model learns\nto generate color images aligned with the normal maps, thereby transforming\nphysical geometry details into realistic appearance. Leveraging the proposed\nnormal diffusion model, we devise a progressive geometry generation strategy\nand a multi-step Score Distillation Sampling (SDS) loss to enhance the\nperformance of 3D human generation. Comprehensive experiments substantiate\nHumanNorm's ability to generate 3D humans with intricate geometry and realistic\nappearances. HumanNorm outperforms existing text-to-3D methods in both geometry\nand texture quality. The project page of HumanNorm is\nhttps://humannorm.github.io/.\n", "date": "2023-10-02", "categories": []} {"id": "2311.17261", "title": "SceneTex: High-Quality Texture Synthesis for Indoor Scenes via Diffusion\n Priors", "abstract": " We propose SceneTex, a novel method for effectively generating high-quality\nand style-consistent textures for indoor scenes using depth-to-image diffusion\npriors. Unlike previous methods that either iteratively warp 2D views onto a\nmesh surface or distillate diffusion latent features without accurate geometric\nand style cues, SceneTex formulates the texture synthesis task as an\noptimization problem in the RGB space where style and geometry consistency are\nproperly reflected. At its core, SceneTex proposes a multiresolution texture\nfield to implicitly encode the mesh appearance. We optimize the target texture\nvia a score-distillation-based objective function in respective RGB renderings.\nTo further secure the style consistency across views, we introduce a\ncross-attention decoder to predict the RGB values by cross-attending to the\npre-sampled reference locations in each instance. SceneTex enables various and\naccurate texture synthesis for 3D-FRONT scenes, demonstrating significant\nimprovements in visual quality and prompt fidelity over the prior texture\ngeneration methods.\n", "date": "2023-11-28", "categories": []} {"id": "2311.17737", "title": "GenZI: Zero-Shot 3D Human-Scene Interaction Generation", "abstract": " Can we synthesize 3D humans interacting with scenes without learning from any\n3D human-scene interaction data? We propose GenZI, the first zero-shot approach\nto generating 3D human-scene interactions. Key to GenZI is our distillation of\ninteraction priors from large vision-language models (VLMs), which have learned\na rich semantic space of 2D human-scene compositions. Given a natural language\ndescription and a coarse point location of the desired interaction in a 3D\nscene, we first leverage VLMs to imagine plausible 2D human interactions\ninpainted into multiple rendered views of the scene. We then formulate a robust\niterative optimization to synthesize the pose and shape of a 3D human model in\nthe scene, guided by consistency with the 2D interaction hypotheses. In\ncontrast to existing learning-based approaches, GenZI circumvents the\nconventional need for captured 3D interaction data, and allows for flexible\ncontrol of the 3D interaction synthesis with easy-to-use text prompts.\nExtensive experiments show that our zero-shot approach has high flexibility and\ngenerality, making it applicable to diverse scene types, including both indoor\nand outdoor environments.\n", "date": "2023-11-29", "categories": []} {"id": "2311.17857", "title": "Gaussian Shell Maps for Efficient 3D Human Generation", "abstract": " Efficient generation of 3D digital humans is important in several industries,\nincluding virtual reality, social media, and cinematic production. 3D\ngenerative adversarial networks (GANs) have demonstrated state-of-the-art\n(SOTA) quality and diversity for generated assets. Current 3D GAN\narchitectures, however, typically rely on volume representations, which are\nslow to render, thereby hampering the GAN training and requiring\nmulti-view-inconsistent 2D upsamplers. Here, we introduce Gaussian Shell Maps\n(GSMs) as a framework that connects SOTA generator network architectures with\nemerging 3D Gaussian rendering primitives using an articulable multi\nshell--based scaffold. In this setting, a CNN generates a 3D texture stack with\nfeatures that are mapped to the shells. The latter represent inflated and\ndeflated versions of a template surface of a digital human in a canonical body\npose. Instead of rasterizing the shells directly, we sample 3D Gaussians on the\nshells whose attributes are encoded in the texture features. These Gaussians\nare efficiently and differentiably rendered. The ability to articulate the\nshells is important during GAN training and, at inference time, to deform a\nbody into arbitrary user-defined poses. Our efficient rendering scheme bypasses\nthe need for view-inconsistent upsamplers and achieves high-quality multi-view\nconsistent renderings at a native resolution of $512 \\times 512$ pixels. We\ndemonstrate that GSMs successfully generate 3D humans when trained on\nsingle-view datasets, including SHHQ and DeepFashion.\n", "date": "2023-11-29", "categories": []} {"id": "2311.17907", "title": "CG3D: Compositional Generation for Text-to-3D via Gaussian Splatting", "abstract": " With the onset of diffusion-based generative models and their ability to\ngenerate text-conditioned images, content generation has received a massive\ninvigoration. Recently, these models have been shown to provide useful guidance\nfor the generation of 3D graphics assets. However, existing work in\ntext-conditioned 3D generation faces fundamental constraints: (i) inability to\ngenerate detailed, multi-object scenes, (ii) inability to textually control\nmulti-object configurations, and (iii) physically realistic scene composition.\nIn this work, we propose CG3D, a method for compositionally generating scalable\n3D assets that resolves these constraints. We find that explicit Gaussian\nradiance fields, parameterized to allow for compositions of objects, possess\nthe capability to enable semantically and physically consistent scenes. By\nutilizing a guidance framework built around this explicit representation, we\nshow state of the art results, capable of even exceeding the guiding diffusion\nmodel in terms of object combinations and physics accuracy.\n", "date": "2023-11-29", "categories": []} {"id": "2311.17917", "title": "AvatarStudio: High-fidelity and Animatable 3D Avatar Creation from Text", "abstract": " We study the problem of creating high-fidelity and animatable 3D avatars from\nonly textual descriptions. Existing text-to-avatar methods are either limited\nto static avatars which cannot be animated or struggle to generate animatable\navatars with promising quality and precise pose control. To address these\nlimitations, we propose AvatarStudio, a coarse-to-fine generative model that\ngenerates explicit textured 3D meshes for animatable human avatars.\nSpecifically, AvatarStudio begins with a low-resolution NeRF-based\nrepresentation for coarse generation, followed by incorporating SMPL-guided\narticulation into the explicit mesh representation to support avatar animation\nand high resolution rendering. To ensure view consistency and pose\ncontrollability of the resulting avatars, we introduce a 2D diffusion model\nconditioned on DensePose for Score Distillation Sampling supervision. By\neffectively leveraging the synergy between the articulated mesh representation\nand the DensePose-conditional diffusion model, AvatarStudio can create\nhigh-quality avatars from text that are ready for animation, significantly\noutperforming previous methods. Moreover, it is competent for many\napplications, e.g., multimodal avatar animations and style-guided avatar\ncreation. For more results, please refer to our project page:\nhttp://jeff95.me/projects/avatarstudio.html\n", "date": "2023-11-29", "categories": []} {"id": "2311.14494", "title": "MVControl: Adding Conditional Control to Multi-view Diffusion for\n Controllable Text-to-3D Generation", "abstract": " We introduce MVControl, a novel neural network architecture that enhances\nexisting pre-trained multi-view 2D diffusion models by incorporating additional\ninput conditions, e.g. edge maps. Our approach enables the generation of\ncontrollable multi-view images and view-consistent 3D content. To achieve\ncontrollable multi-view image generation, we leverage MVDream as our base\nmodel, and train a new neural network module as additional plugin for\nend-to-end task-specific condition learning. To precisely control the shapes\nand views of generated images, we innovatively propose a new conditioning\nmechanism that predicts an embedding encapsulating the input spatial and view\nconditions, which is then injected to the network globally. Once MVControl is\ntrained, score-distillation (SDS) loss based optimization can be performed to\ngenerate 3D content, in which process we propose to use a hybrid diffusion\nprior. The hybrid prior relies on a pre-trained Stable-Diffusion network and\nour trained MVControl for additional guidance. Extensive experiments\ndemonstrate that our method achieves robust generalization and enables the\ncontrollable generation of high-quality 3D content. Code available at\nhttps://github.com/WU-CVGL/MVControl/.\n", "date": "2023-11-24", "categories": []} {"id": "2311.16961", "title": "HumanRef: Single Image to 3D Human Generation via Reference-Guided\n Diffusion", "abstract": " Generating a 3D human model from a single reference image is challenging\nbecause it requires inferring textures and geometries in invisible views while\nmaintaining consistency with the reference image. Previous methods utilizing 3D\ngenerative models are limited by the availability of 3D training data.\nOptimization-based methods that lift text-to-image diffusion models to 3D\ngeneration often fail to preserve the texture details of the reference image,\nresulting in inconsistent appearances in different views. In this paper, we\npropose HumanRef, a 3D human generation framework from a single-view input. To\nensure the generated 3D model is photorealistic and consistent with the input\nimage, HumanRef introduces a novel method called reference-guided score\ndistillation sampling (Ref-SDS), which effectively incorporates image guidance\ninto the generation process. Furthermore, we introduce region-aware attention\nto Ref-SDS, ensuring accurate correspondence between different body regions.\nExperimental results demonstrate that HumanRef outperforms state-of-the-art\nmethods in generating 3D clothed humans with fine geometry, photorealistic\ntextures, and view-consistent appearances.\n", "date": "2023-11-28", "categories": []} {"id": "2311.15435", "title": "Functional Diffusion", "abstract": " We propose a new class of generative diffusion models, called functional\ndiffusion. In contrast to previous work, functional diffusion works on samples\nthat are represented by functions with a continuous domain. Functional\ndiffusion can be seen as an extension of classical diffusion models to an\ninfinite-dimensional domain. Functional diffusion is very versatile as images,\nvideos, audio, 3D shapes, deformations, \\etc, can be handled by the same\nframework with minimal changes. In addition, functional diffusion is especially\nsuited for irregular data or data defined in non-standard domains. In our work,\nwe derive the necessary foundations for functional diffusion and propose a\nfirst implementation based on the transformer architecture. We show generative\nresults on complicated signed distance functions and deformation functions\ndefined on 3D surfaces.\n", "date": "2023-11-26", "categories": []} {"id": "2311.15475", "title": "MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers", "abstract": " We introduce MeshGPT, a new approach for generating triangle meshes that\nreflects the compactness typical of artist-created meshes, in contrast to dense\ntriangle meshes extracted by iso-surfacing methods from neural fields. Inspired\nby recent advances in powerful large language models, we adopt a sequence-based\napproach to autoregressively generate triangle meshes as sequences of\ntriangles. We first learn a vocabulary of latent quantized embeddings, using\ngraph convolutions, which inform these embeddings of the local mesh geometry\nand topology. These embeddings are sequenced and decoded into triangles by a\ndecoder, ensuring that they can effectively reconstruct the mesh. A transformer\nis then trained on this learned vocabulary to predict the index of the next\nembedding given previous embeddings. Once trained, our model can be\nautoregressively sampled to generate new triangle meshes, directly generating\ncompact meshes with sharp edges, more closely imitating the efficient\ntriangulation patterns of human-crafted meshes. MeshGPT demonstrates a notable\nimprovement over state of the art mesh generation methods, with a 9% increase\nin shape coverage and a 30-point enhancement in FID scores across various\ncategories.\n", "date": "2023-11-27", "categories": []} {"id": "2311.15561", "title": "ET3D: Efficient Text-to-3D Generation via Multi-View Distillation", "abstract": " Recent breakthroughs in text-to-image generation has shown encouraging\nresults via large generative models. Due to the scarcity of 3D assets, it is\nhardly to transfer the success of text-to-image generation to that of\ntext-to-3D generation. Existing text-to-3D generation methods usually adopt the\nparadigm of DreamFusion, which conducts per-asset optimization by distilling a\npretrained text-to-image diffusion model. The generation speed usually ranges\nfrom several minutes to tens of minutes per 3D asset, which degrades the user\nexperience and also imposes a burden to the service providers due to the high\ncomputational budget.\n In this work, we present an efficient text-to-3D generation method, which\nrequires only around 8 $ms$ to generate a 3D asset given the text prompt on a\nconsumer graphic card. The main insight is that we exploit the images generated\nby a large pre-trained text-to-image diffusion model, to supervise the training\nof a text conditioned 3D generative adversarial network. Once the network is\ntrained, we are able to efficiently generate a 3D asset via a single forward\npass. Our method requires no 3D training data and provides an alternative\napproach for efficient text-to-3D generation by distilling pre-trained image\ndiffusion models.\n", "date": "2023-11-27", "categories": []} {"id": "2311.06214", "title": "Instant3D: Fast Text-to-3D with Sparse-View Generation and Large\n Reconstruction Model", "abstract": " Text-to-3D with diffusion models has achieved remarkable progress in recent\nyears. However, existing methods either rely on score distillation-based\noptimization which suffer from slow inference, low diversity and Janus\nproblems, or are feed-forward methods that generate low-quality results due to\nthe scarcity of 3D training data. In this paper, we propose Instant3D, a novel\nmethod that generates high-quality and diverse 3D assets from text prompts in a\nfeed-forward manner. We adopt a two-stage paradigm, which first generates a\nsparse set of four structured and consistent views from text in one shot with a\nfine-tuned 2D text-to-image diffusion model, and then directly regresses the\nNeRF from the generated images with a novel transformer-based sparse-view\nreconstructor. Through extensive experiments, we demonstrate that our method\ncan generate diverse 3D assets of high visual quality within 20 seconds, which\nis two orders of magnitude faster than previous optimization-based methods that\ncan take 1 to 10 hours. Our project webpage: https://jiahao.ai/instant3d/.\n", "date": "2023-11-10", "categories": []} {"id": "2311.13384", "title": "LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes", "abstract": " With the widespread usage of VR devices and contents, demands for 3D scene\ngeneration techniques become more popular. Existing 3D scene generation models,\nhowever, limit the target scene to specific domain, primarily due to their\ntraining strategies using 3D scan dataset that is far from the real-world. To\naddress such limitation, we propose LucidDreamer, a domain-free scene\ngeneration pipeline by fully leveraging the power of existing large-scale\ndiffusion-based generative model. Our LucidDreamer has two alternate steps:\nDreaming and Alignment. First, to generate multi-view consistent images from\ninputs, we set the point cloud as a geometrical guideline for each image\ngeneration. Specifically, we project a portion of point cloud to the desired\nview and provide the projection as a guidance for inpainting using the\ngenerative model. The inpainted images are lifted to 3D space with estimated\ndepth maps, composing a new points. Second, to aggregate the new points into\nthe 3D scene, we propose an aligning algorithm which harmoniously integrates\nthe portions of newly generated 3D scenes. The finally obtained 3D scene serves\nas initial points for optimizing Gaussian splats. LucidDreamer produces\nGaussian splats that are highly-detailed compared to the previous 3D scene\ngeneration methods, with no constraint on domain of the target scene. Project\npage: https://luciddreamer-cvlab.github.io/\n", "date": "2023-11-22", "categories": []} {"id": "2311.13617", "title": "Boosting3D: High-Fidelity Image-to-3D by Boosting 2D Diffusion Prior to\n 3D Prior with Progressive Learning", "abstract": " We present Boosting3D, a multi-stage single image-to-3D generation method\nthat can robustly generate reasonable 3D objects in different data domains. The\npoint of this work is to solve the view consistency problem in single\nimage-guided 3D generation by modeling a reasonable geometric structure. For\nthis purpose, we propose to utilize better 3D prior to training the NeRF. More\nspecifically, we train an object-level LoRA for the target object using\noriginal image and the rendering output of NeRF. And then we train the LoRA and\nNeRF using a progressive training strategy. The LoRA and NeRF will boost each\nother while training. After the progressive training, the LoRA learns the 3D\ninformation of the generated object and eventually turns to an object-level 3D\nprior. In the final stage, we extract the mesh from the trained NeRF and use\nthe trained LoRA to optimize the structure and appearance of the mesh. The\nexperiments demonstrate the effectiveness of the proposed method. Boosting3D\nlearns object-specific 3D prior which is beyond the ability of pre-trained\ndiffusion priors and achieves state-of-the-art performance in the single\nimage-to-3d generation task.\n", "date": "2023-11-22", "categories": []} {"id": "2305.16213", "title": "ProlificDreamer: High-Fidelity and Diverse Text-to-3D Generation with\n Variational Score Distillation", "abstract": " Score distillation sampling (SDS) has shown great promise in text-to-3D\ngeneration by distilling pretrained large-scale text-to-image diffusion models,\nbut suffers from over-saturation, over-smoothing, and low-diversity problems.\nIn this work, we propose to model the 3D parameter as a random variable instead\nof a constant as in SDS and present variational score distillation (VSD), a\nprincipled particle-based variational framework to explain and address the\naforementioned issues in text-to-3D generation. We show that SDS is a special\ncase of VSD and leads to poor samples with both small and large CFG weights. In\ncomparison, VSD works well with various CFG weights as ancestral sampling from\ndiffusion models and simultaneously improves the diversity and sample quality\nwith a common CFG weight (i.e., $7.5$). We further present various improvements\nin the design space for text-to-3D such as distillation time schedule and\ndensity initialization, which are orthogonal to the distillation algorithm yet\nnot well explored. Our overall approach, dubbed ProlificDreamer, can generate\nhigh rendering resolution (i.e., $512\\times512$) and high-fidelity NeRF with\nrich structure and complex effects (e.g., smoke and drops). Further,\ninitialized from NeRF, meshes fine-tuned by VSD are meticulously detailed and\nphoto-realistic. Project page and codes:\nhttps://ml.cs.tsinghua.edu.cn/prolificdreamer/\n", "date": "2023-05-25", "categories": []} {"id": "2306.04889", "title": "ShaDDR: Interactive Example-Based Geometry and Texture Generation via 3D\n Shape Detailization and Differentiable Rendering", "abstract": " We present ShaDDR, an example-based deep generative neural network which\nproduces a high-resolution textured 3D shape through geometry detailization and\nconditional texture generation applied to an input coarse voxel shape. Trained\non a small set of detailed and textured exemplar shapes, our method learns to\ndetailize the geometry via multi-resolution voxel upsampling and generate\ntextures on voxel surfaces via differentiable rendering against exemplar\ntexture images from a few views. The generation is interactive, taking less\nthan 1 second to produce a 3D model with voxel resolutions up to 512^3. The\ngenerated shape preserves the overall structure of the input coarse voxel\nmodel, while the style of the generated geometric details and textures can be\nmanipulated through learned latent codes. In the experiments, we show that our\nmethod can generate higher-resolution shapes with plausible and improved\ngeometric details and clean textures compared to prior works. Furthermore, we\nshowcase the ability of our method to learn geometric details and textures from\nshapes reconstructed from real-world photos. In addition, we have developed an\ninteractive modeling application to demonstrate the generalizability of our\nmethod to various user inputs and the controllability it offers, allowing users\nto interactively sculpt a coarse voxel shape to define the overall structure of\nthe detailized 3D shape. Code and data are available at\nhttps://github.com/qiminchen/ShaDDR.\n", "date": "2023-06-08", "categories": []} {"id": "2311.13574", "title": "XAGen: 3D Expressive Human Avatars Generation", "abstract": " Recent advances in 3D-aware GAN models have enabled the generation of\nrealistic and controllable human body images. However, existing methods focus\non the control of major body joints, neglecting the manipulation of expressive\nattributes, such as facial expressions, jaw poses, hand poses, and so on. In\nthis work, we present XAGen, the first 3D generative model for human avatars\ncapable of expressive control over body, face, and hands. To enhance the\nfidelity of small-scale regions like face and hands, we devise a multi-scale\nand multi-part 3D representation that models fine details. Based on this\nrepresentation, we propose a multi-part rendering technique that disentangles\nthe synthesis of body, face, and hands to ease model training and enhance\ngeometric quality. Furthermore, we design multi-part discriminators that\nevaluate the quality of the generated avatars with respect to their appearance\nand fine-grained control capabilities. Experiments show that XAGen surpasses\nstate-of-the-art methods in terms of realism, diversity, and expressive control\nabilities. Code and data will be made available at\nhttps://showlab.github.io/xagen.\n", "date": "2023-11-22", "categories": []} {"id": "2311.10123", "title": "MetaDreamer: Efficient Text-to-3D Creation With Disentangling Geometry\n and Texture", "abstract": " Generative models for 3D object synthesis have seen significant advancements\nwith the incorporation of prior knowledge distilled from 2D diffusion models.\nNevertheless, challenges persist in the form of multi-view geometric\ninconsistencies and slow generation speeds within the existing 3D synthesis\nframeworks. This can be attributed to two factors: firstly, the deficiency of\nabundant geometric a priori knowledge in optimization, and secondly, the\nentanglement issue between geometry and texture in conventional 3D generation\nmethods.In response, we introduce MetaDreammer, a two-stage optimization\napproach that leverages rich 2D and 3D prior knowledge. In the first stage, our\nemphasis is on optimizing the geometric representation to ensure multi-view\nconsistency and accuracy of 3D objects. In the second stage, we concentrate on\nfine-tuning the geometry and optimizing the texture, thereby achieving a more\nrefined 3D object. Through leveraging 2D and 3D prior knowledge in two stages,\nrespectively, we effectively mitigate the interdependence between geometry and\ntexture. MetaDreamer establishes clear optimization objectives for each stage,\nresulting in significant time savings in the 3D generation process. Ultimately,\nMetaDreamer can generate high-quality 3D objects based on textual prompts\nwithin 20 minutes, and to the best of our knowledge, it is the most efficient\ntext-to-3D generation method. Furthermore, we introduce image control into the\nprocess, enhancing the controllability of 3D generation. Extensive empirical\nevidence confirms that our method is not only highly efficient but also\nachieves a quality level that is at the forefront of current state-of-the-art\n3D generation techniques.\n", "date": "2023-11-16", "categories": []} {"id": "2311.09571", "title": "3D Paintbrush: Local Stylization of 3D Shapes with Cascaded Score\n Distillation", "abstract": " In this work we develop 3D Paintbrush, a technique for automatically\ntexturing local semantic regions on meshes via text descriptions. Our method is\ndesigned to operate directly on meshes, producing texture maps which seamlessly\nintegrate into standard graphics pipelines. We opt to simultaneously produce a\nlocalization map (to specify the edit region) and a texture map which conforms\nto it. This synergistic approach improves the quality of both the localization\nand the stylization. To enhance the details and resolution of the textured\narea, we leverage multiple stages of a cascaded diffusion model to supervise\nour local editing technique with generative priors learned from images at\ndifferent resolutions. Our technique, referred to as Cascaded Score\nDistillation (CSD), simultaneously distills scores at multiple resolutions in a\ncascaded fashion, enabling control over both the granularity and global\nunderstanding of the supervision. We demonstrate the effectiveness of 3D\nPaintbrush to locally texture a variety of shapes within different semantic\nregions. Project page: https://threedle.github.io/3d-paintbrush\n", "date": "2023-11-16", "categories": []} {"id": "2311.09217", "title": "DMV3D: Denoising Multi-View Diffusion using 3D Large Reconstruction\n Model", "abstract": " We propose \\textbf{DMV3D}, a novel 3D generation approach that uses a\ntransformer-based 3D large reconstruction model to denoise multi-view\ndiffusion. Our reconstruction model incorporates a triplane NeRF representation\nand can denoise noisy multi-view images via NeRF reconstruction and rendering,\nachieving single-stage 3D generation in $\\sim$30s on single A100 GPU. We train\n\\textbf{DMV3D} on large-scale multi-view image datasets of highly diverse\nobjects using only image reconstruction losses, without accessing 3D assets. We\ndemonstrate state-of-the-art results for the single-image reconstruction\nproblem where probabilistic modeling of unseen object parts is required for\ngenerating diverse reconstructions with sharp textures. We also show\nhigh-quality text-to-3D generation results outperforming previous 3D diffusion\nmodels. Our project website is at: https://justimyhxu.github.io/projects/dmv3d/ .\n", "date": "2023-11-15", "categories": []} {"id": "2311.09221", "title": "Single-Image 3D Human Digitization with Shape-Guided Diffusion", "abstract": " We present an approach to generate a 360-degree view of a person with a\nconsistent, high-resolution appearance from a single input image. NeRF and its\nvariants typically require videos or images from different viewpoints. Most\nexisting approaches taking monocular input either rely on ground-truth 3D scans\nfor supervision or lack 3D consistency. While recent 3D generative models show\npromise of 3D consistent human digitization, these approaches do not generalize\nwell to diverse clothing appearances, and the results lack photorealism. Unlike\nexisting work, we utilize high-capacity 2D diffusion models pretrained for\ngeneral image synthesis tasks as an appearance prior of clothed humans. To\nachieve better 3D consistency while retaining the input identity, we\nprogressively synthesize multiple views of the human in the input image by\ninpainting missing regions with shape-guided diffusion conditioned on\nsilhouette and surface normal. We then fuse these synthesized multi-view images\nvia inverse rendering to obtain a fully textured high-resolution 3D mesh of the\ngiven person. Experiments show that our approach outperforms prior methods and\nachieves photorealistic 360-degree synthesis of a wide range of clothed humans\nwith complex textures from a single image.\n", "date": "2023-11-15", "categories": []} {"id": "2301.13591", "title": "Zero3D: Semantic-Driven Multi-Category 3D Shape Generation", "abstract": " Semantic-driven 3D shape generation aims to generate 3D objects conditioned\non text. Previous works face problems with single-category generation,\nlow-frequency 3D details, and requiring a large number of paired datasets for\ntraining. To tackle these challenges, we propose a multi-category conditional\ndiffusion model. Specifically, 1) to alleviate the problem of lack of\nlarge-scale paired data, we bridge the text, 2D image and 3D shape based on the\npre-trained CLIP model, and 2) to obtain the multi-category 3D shape feature,\nwe apply the conditional flow model to generate 3D shape vector conditioned on\nCLIP embedding. 3) to generate multi-category 3D shape, we employ the\nhidden-layer diffusion model conditioned on the multi-category shape vector,\nwhich greatly reduces the training time and memory consumption.\n", "date": "2023-01-31", "categories": []} {"id": "2311.07885", "title": "One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View\n Generation and 3D Diffusion", "abstract": " Recent advancements in open-world 3D object generation have been remarkable,\nwith image-to-3D methods offering superior fine-grained control over their\ntext-to-3D counterparts. However, most existing models fall short in\nsimultaneously providing rapid generation speeds and high fidelity to input\nimages - two features essential for practical applications. In this paper, we\npresent One-2-3-45++, an innovative method that transforms a single image into\na detailed 3D textured mesh in approximately one minute. Our approach aims to\nfully harness the extensive knowledge embedded in 2D diffusion models and\npriors from valuable yet limited 3D data. This is achieved by initially\nfinetuning a 2D diffusion model for consistent multi-view image generation,\nfollowed by elevating these images to 3D with the aid of multi-view conditioned\n3D native diffusion models. Extensive experimental evaluations demonstrate that\nour method can produce high-quality, diverse 3D assets that closely mirror the\noriginal input image. Our project webpage:\nhttps://sudo-ai-3d.github.io/One2345plus_page.\n", "date": "2023-11-14", "categories": []} {"id": "2301.09091", "title": "BallGAN: 3D-aware Image Synthesis with a Spherical Background", "abstract": " 3D-aware GANs aim to synthesize realistic 3D scenes such that they can be\nrendered in arbitrary perspectives to produce images. Although previous methods\nproduce realistic images, they suffer from unstable training or degenerate\nsolutions where the 3D geometry is unnatural. We hypothesize that the 3D\ngeometry is underdetermined due to the insufficient constraint, i.e., being\nclassified as real image to the discriminator is not enough. To solve this\nproblem, we propose to approximate the background as a spherical surface and\nrepresent a scene as a union of the foreground placed in the sphere and the\nthin spherical background. It reduces the degree of freedom in the background\nfield. Accordingly, we modify the volume rendering equation and incorporate\ndedicated constraints to design a novel 3D-aware GAN framework named BallGAN.\nBallGAN has multiple advantages as follows. 1) It produces more reasonable 3D\ngeometry; the images of a scene across different viewpoints have better\nphotometric consistency and fidelity than the state-of-the-art methods. 2) The\ntraining becomes much more stable. 3) The foreground can be separately rendered\non top of different arbitrary backgrounds.\n", "date": "2023-01-22", "categories": []} {"id": "2310.15008", "title": "Wonder3D: Single Image to 3D using Cross-Domain Diffusion", "abstract": " In this work, we introduce Wonder3D, a novel method for efficiently\ngenerating high-fidelity textured meshes from single-view images.Recent methods\nbased on Score Distillation Sampling (SDS) have shown the potential to recover\n3D geometry from 2D diffusion priors, but they typically suffer from\ntime-consuming per-shape optimization and inconsistent geometry. In contrast,\ncertain works directly produce 3D information via fast network inferences, but\ntheir results are often of low quality and lack geometric details. To\nholistically improve the quality, consistency, and efficiency of image-to-3D\ntasks, we propose a cross-domain diffusion model that generates multi-view\nnormal maps and the corresponding color images. To ensure consistency, we\nemploy a multi-view cross-domain attention mechanism that facilitates\ninformation exchange across views and modalities. Lastly, we introduce a\ngeometry-aware normal fusion algorithm that extracts high-quality surfaces from\nthe multi-view 2D representations. Our extensive evaluations demonstrate that\nour method achieves high-quality reconstruction results, robust generalization,\nand reasonably good efficiency compared to prior works.\n", "date": "2023-10-23", "categories": []} {"id": "2311.05230", "title": "ConRad: Image Constrained Radiance Fields for 3D Generation from a\n Single Image", "abstract": " We present a novel method for reconstructing 3D objects from a single RGB\nimage. Our method leverages the latest image generation models to infer the\nhidden 3D structure while remaining faithful to the input image. While existing\nmethods obtain impressive results in generating 3D models from text prompts,\nthey do not provide an easy approach for conditioning on input RGB data.\nNa\\\"ive extensions of these methods often lead to improper alignment in\nappearance between the input image and the 3D reconstructions. We address these\nchallenges by introducing Image Constrained Radiance Fields (ConRad), a novel\nvariant of neural radiance fields. ConRad is an efficient 3D representation\nthat explicitly captures the appearance of an input image in one viewpoint. We\npropose a training algorithm that leverages the single RGB image in conjunction\nwith pretrained Diffusion Models to optimize the parameters of a ConRad\nrepresentation. Extensive experiments show that ConRad representations can\nsimplify preservation of image details while producing a realistic 3D\nreconstruction. Compared to existing state-of-the-art baselines, we show that\nour 3D reconstructions remain more faithful to the input and produce more\nconsistent 3D models while demonstrating significantly improved quantitative\nperformance on a ShapeNet object benchmark.\n", "date": "2023-11-09", "categories": []} {"id": "2311.05461", "title": "Control3D: Towards Controllable Text-to-3D Generation", "abstract": " Recent remarkable advances in large-scale text-to-image diffusion models have\ninspired a significant breakthrough in text-to-3D generation, pursuing 3D\ncontent creation solely from a given text prompt. However, existing text-to-3D\ntechniques lack a crucial ability in the creative process: interactively\ncontrol and shape the synthetic 3D contents according to users' desired\nspecifications (e.g., sketch). To alleviate this issue, we present the first\nattempt for text-to-3D generation conditioning on the additional hand-drawn\nsketch, namely Control3D, which enhances controllability for users. In\nparticular, a 2D conditioned diffusion model (ControlNet) is remoulded to guide\nthe learning of 3D scene parameterized as NeRF, encouraging each view of 3D\nscene aligned with the given text prompt and hand-drawn sketch. Moreover, we\nexploit a pre-trained differentiable photo-to-sketch model to directly estimate\nthe sketch of the rendered image over synthetic 3D scene. Such estimated sketch\nalong with each sampled view is further enforced to be geometrically consistent\nwith the given sketch, pursuing better controllable text-to-3D generation.\nThrough extensive experiments, we demonstrate that our proposal can generate\naccurate and faithful 3D scenes that align closely with the input text prompts\nand sketches.\n", "date": "2023-11-09", "categories": []} {"id": "2311.05464", "title": "3DStyle-Diffusion: Pursuing Fine-grained Text-driven 3D Stylization with\n 2D Diffusion Models", "abstract": " 3D content creation via text-driven stylization has played a fundamental\nchallenge to multimedia and graphics community. Recent advances of cross-modal\nfoundation models (e.g., CLIP) have made this problem feasible. Those\napproaches commonly leverage CLIP to align the holistic semantics of stylized\nmesh with the given text prompt. Nevertheless, it is not trivial to enable more\ncontrollable stylization of fine-grained details in 3D meshes solely based on\nsuch semantic-level cross-modal supervision. In this work, we propose a new\n3DStyle-Diffusion model that triggers fine-grained stylization of 3D meshes\nwith additional controllable appearance and geometric guidance from 2D\nDiffusion models. Technically, 3DStyle-Diffusion first parameterizes the\ntexture of 3D mesh into reflectance properties and scene lighting using\nimplicit MLP networks. Meanwhile, an accurate depth map of each sampled view is\nachieved conditioned on 3D mesh. Then, 3DStyle-Diffusion leverages a\npre-trained controllable 2D Diffusion model to guide the learning of rendered\nimages, encouraging the synthesized image of each view semantically aligned\nwith text prompt and geometrically consistent with depth map. This way\nelegantly integrates both image rendering via implicit MLP networks and\ndiffusion process of image synthesis in an end-to-end fashion, enabling a\nhigh-quality fine-grained stylization of 3D meshes. We also build a new dataset\nderived from Objaverse and the evaluation protocol for this task. Through both\nqualitative and quantitative experiments, we validate the capability of our\n3DStyle-Diffusion. Source code and data are available at\n\\url{https://github.com/yanghb22-fdu/3DStyle-Diffusion-Official}.\n", "date": "2023-11-09", "categories": []} {"id": "2305.12529", "title": "DreamWaltz: Make a Scene with Complex 3D Animatable Avatars", "abstract": " We present DreamWaltz, a novel framework for generating and animating complex\n3D avatars given text guidance and parametric human body prior. While recent\nmethods have shown encouraging results for text-to-3D generation of common\nobjects, creating high-quality and animatable 3D avatars remains challenging.\nTo create high-quality 3D avatars, DreamWaltz proposes 3D-consistent\nocclusion-aware Score Distillation Sampling (SDS) to optimize implicit neural\nrepresentations with canonical poses. It provides view-aligned supervision via\n3D-aware skeleton conditioning which enables complex avatar generation without\nartifacts and multiple faces. For animation, our method learns an animatable 3D\navatar representation from abundant image priors of diffusion model conditioned\non various poses, which could animate complex non-rigged avatars given\narbitrary poses without retraining. Extensive evaluations demonstrate that\nDreamWaltz is an effective and robust approach for creating 3D avatars that can\ntake on complex shapes and appearances as well as novel poses for animation.\nThe proposed framework further enables the creation of complex scenes with\ndiverse compositions, including avatar-avatar, avatar-object and avatar-scene\ninteractions. See https://dreamwaltz3d.github.io/ for more vivid 3D avatar and\nanimation results.\n", "date": "2023-05-21", "categories": []} {"id": "2311.02848", "title": "Consistent4D: Consistent 360{\\deg} Dynamic Object Generation from\n Monocular Video", "abstract": " In this paper, we present Consistent4D, a novel approach for generating 4D\ndynamic objects from uncalibrated monocular videos. Uniquely, we cast the\n360-degree dynamic object reconstruction as a 4D generation problem,\neliminating the need for tedious multi-view data collection and camera\ncalibration. This is achieved by leveraging the object-level 3D-aware image\ndiffusion model as the primary supervision signal for training Dynamic Neural\nRadiance Fields (DyNeRF). Specifically, we propose a Cascade DyNeRF to\nfacilitate stable convergence and temporal continuity under the supervision\nsignal which is discrete along the time axis. To achieve spatial and temporal\nconsistency, we further introduce an Interpolation-driven Consistency Loss. It\nis optimized by minimizing the discrepancy between rendered frames from DyNeRF\nand interpolated frames from a pre-trained video interpolation model. Extensive\nexperiments show that our Consistent4D can perform competitively to prior art\nalternatives, opening up new possibilities for 4D dynamic object generation\nfrom monocular videos, whilst also demonstrating advantage for conventional\ntext-to-3D generation tasks. Our project page is\nhttps://consistent4d.github.io/.\n", "date": "2023-11-06", "categories": []} {"id": "2311.03226", "title": "LDM3D-VR: Latent Diffusion Model for 3D VR", "abstract": " Latent diffusion models have proven to be state-of-the-art in the creation\nand manipulation of visual outputs. However, as far as we know, the generation\nof depth maps jointly with RGB is still limited. We introduce LDM3D-VR, a suite\nof diffusion models targeting virtual reality development that includes\nLDM3D-pano and LDM3D-SR. These models enable the generation of panoramic RGBD\nbased on textual prompts and the upscaling of low-resolution inputs to\nhigh-resolution RGBD, respectively. Our models are fine-tuned from existing\npretrained models on datasets containing panoramic/high-resolution RGB images,\ndepth maps and captions. Both models are evaluated in comparison to existing\nrelated methods.\n", "date": "2023-11-06", "categories": []} {"id": "2310.19415", "title": "Text-to-3D with Classifier Score Distillation", "abstract": " Text-to-3D generation has made remarkable progress recently, particularly\nwith methods based on Score Distillation Sampling (SDS) that leverages\npre-trained 2D diffusion models. While the usage of classifier-free guidance is\nwell acknowledged to be crucial for successful optimization, it is considered\nan auxiliary trick rather than the most essential component. In this paper, we\nre-evaluate the role of classifier-free guidance in score distillation and\ndiscover a surprising finding: the guidance alone is enough for effective\ntext-to-3D generation tasks. We name this method Classifier Score Distillation\n(CSD), which can be interpreted as using an implicit classification model for\ngeneration. This new perspective reveals new insights for understanding\nexisting techniques. We validate the effectiveness of CSD across a variety of\ntext-to-3D tasks including shape generation, texture synthesis, and shape\nediting, achieving results superior to those of state-of-the-art methods. Our\nproject page is https://xinyu-andy.github.io/Classifier-Score-Distillation\n", "date": "2023-10-30", "categories": []} {"id": "2211.13251", "title": "CGOF++: Controllable 3D Face Synthesis with Conditional Generative\n Occupancy Fields", "abstract": " Capitalizing on the recent advances in image generation models, existing\ncontrollable face image synthesis methods are able to generate high-fidelity\nimages with some levels of controllability, e.g., controlling the shapes,\nexpressions, textures, and poses of the generated face images. However,\nprevious methods focus on controllable 2D image generative models, which are\nprone to producing inconsistent face images under large expression and pose\nchanges. In this paper, we propose a new NeRF-based conditional 3D face\nsynthesis framework, which enables 3D controllability over the generated face\nimages by imposing explicit 3D conditions from 3D face priors. At its core is a\nconditional Generative Occupancy Field (cGOF++) that effectively enforces the\nshape of the generated face to conform to a given 3D Morphable Model (3DMM)\nmesh, built on top of EG3D [1], a recent tri-plane-based generative model. To\nachieve accurate control over fine-grained 3D face shapes of the synthesized\nimages, we additionally incorporate a 3D landmark loss as well as a volume\nwarping loss into our synthesis framework. Experiments validate the\neffectiveness of the proposed method, which is able to generate high-fidelity\nface images and shows more precise 3D controllability than state-of-the-art\n2D-based controllable face synthesis methods.\n", "date": "2022-11-23", "categories": []} {"id": "2304.12483", "title": "Towards Realistic Generative 3D Face Models", "abstract": " In recent years, there has been significant progress in 2D generative face\nmodels fueled by applications such as animation, synthetic data generation, and\ndigital avatars. However, due to the absence of 3D information, these 2D models\noften struggle to accurately disentangle facial attributes like pose,\nexpression, and illumination, limiting their editing capabilities. To address\nthis limitation, this paper proposes a 3D controllable generative face model to\nproduce high-quality albedo and precise 3D shape leveraging existing 2D\ngenerative models. By combining 2D face generative models with semantic face\nmanipulation, this method enables editing of detailed 3D rendered faces. The\nproposed framework utilizes an alternating descent optimization approach over\nshape and albedo. Differentiable rendering is used to train high-quality shapes\nand albedo without 3D supervision. Moreover, this approach outperforms the\nstate-of-the-art (SOTA) methods in the well-known NoW benchmark for shape\nreconstruction. It also outperforms the SOTA reconstruction models in\nrecovering rendered faces' identities across novel poses by an average of 10%.\nAdditionally, the paper demonstrates direct control of expressions in 3D faces\nby exploiting latent space leading to text-based editing of 3D faces.\n", "date": "2023-04-24", "categories": []} {"id": "2304.06700", "title": "Control3Diff: Learning Controllable 3D Diffusion Models from Single-view\n Images", "abstract": " Diffusion models have recently become the de-facto approach for generative\nmodeling in the 2D domain. However, extending diffusion models to 3D is\nchallenging due to the difficulties in acquiring 3D ground truth data for\ntraining. On the other hand, 3D GANs that integrate implicit 3D representations\ninto GANs have shown remarkable 3D-aware generation when trained only on\nsingle-view image datasets. However, 3D GANs do not provide straightforward\nways to precisely control image synthesis. To address these challenges, We\npresent Control3Diff, a 3D diffusion model that combines the strengths of\ndiffusion models and 3D GANs for versatile, controllable 3D-aware image\nsynthesis for single-view datasets. Control3Diff explicitly models the\nunderlying latent distribution (optionally conditioned on external inputs),\nthus enabling direct control during the diffusion process. Moreover, our\napproach is general and applicable to any type of controlling input, allowing\nus to train it with the same diffusion objective without any auxiliary\nsupervision. We validate the efficacy of Control3Diff on standard image\ngeneration benchmarks, including FFHQ, AFHQ, and ShapeNet, using various\nconditioning inputs such as images, sketches, and text prompts. Please see the\nproject website (\\url{https://jiataogu.me/control3diff}) for video comparisons.\n", "date": "2023-04-13", "categories": []} {"id": "2310.16818", "title": "DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion\n Prior", "abstract": " We present DreamCraft3D, a hierarchical 3D content generation method that\nproduces high-fidelity and coherent 3D objects. We tackle the problem by\nleveraging a 2D reference image to guide the stages of geometry sculpting and\ntexture boosting. A central focus of this work is to address the consistency\nissue that existing works encounter. To sculpt geometries that render\ncoherently, we perform score distillation sampling via a view-dependent\ndiffusion model. This 3D prior, alongside several training strategies,\nprioritizes the geometry consistency but compromises the texture fidelity. We\nfurther propose Bootstrapped Score Distillation to specifically boost the\ntexture. We train a personalized diffusion model, Dreambooth, on the augmented\nrenderings of the scene, imbuing it with 3D knowledge of the scene being\noptimized. The score distillation from this 3D-aware diffusion prior provides\nview-consistent guidance for the scene. Notably, through an alternating\noptimization of the diffusion prior and 3D scene representation, we achieve\nmutually reinforcing improvements: the optimized 3D scene aids in training the\nscene-specific diffusion model, which offers increasingly view-consistent\nguidance for 3D optimization. The optimization is thus bootstrapped and leads\nto substantial texture boosting. With tailored 3D priors throughout the\nhierarchical generation, DreamCraft3D generates coherent 3D objects with\nphotorealistic renderings, advancing the state-of-the-art in 3D content\ngeneration. Code available at https://github.com/deepseek-ai/DreamCraft3D.\n", "date": "2023-10-25", "categories": []} {"id": "2212.07275", "title": "PhoMoH: Implicit Photorealistic 3D Models of Human Heads", "abstract": " We present PhoMoH, a neural network methodology to construct generative\nmodels of photo-realistic 3D geometry and appearance of human heads including\nhair, beards, an oral cavity, and clothing. In contrast to prior work, PhoMoH\nmodels the human head using neural fields, thus supporting complex topology.\nInstead of learning a head model from scratch, we propose to augment an\nexisting expressive head model with new features. Concretely, we learn a highly\ndetailed geometry network layered on top of a mid-resolution head model\ntogether with a detailed, local geometry-aware, and disentangled color field.\nOur proposed architecture allows us to learn photo-realistic human head models\nfrom relatively little data. The learned generative geometry and appearance\nnetworks can be sampled individually and enable the creation of diverse and\nrealistic human heads. Extensive experiments validate our method qualitatively\nand across different metrics.\n", "date": "2022-12-14", "categories": []} {"id": "2310.15948", "title": "Language-driven Scene Synthesis using Multi-conditional Diffusion Model", "abstract": " Scene synthesis is a challenging problem with several industrial\napplications. Recently, substantial efforts have been directed to synthesize\nthe scene using human motions, room layouts, or spatial graphs as the input.\nHowever, few studies have addressed this problem from multiple modalities,\nespecially combining text prompts. In this paper, we propose a language-driven\nscene synthesis task, which is a new task that integrates text prompts, human\nmotion, and existing objects for scene synthesis. Unlike other single-condition\nsynthesis tasks, our problem involves multiple conditions and requires a\nstrategy for processing and encoding them into a unified space. To address the\nchallenge, we present a multi-conditional diffusion model, which differs from\nthe implicit unification approach of other diffusion literature by explicitly\npredicting the guiding points for the original data distribution. We\ndemonstrate that our approach is theoretically supportive. The intensive\nexperiment results illustrate that our method outperforms state-of-the-art\nbenchmarks and enables natural scene editing applications. The source code and\ndataset can be accessed at https://lang-scene-synth.github.io/.\n", "date": "2023-10-24", "categories": []} {"id": "2310.13772", "title": "TexFusion: Synthesizing 3D Textures with Text-Guided Image Diffusion\n Models", "abstract": " We present TexFusion (Texture Diffusion), a new method to synthesize textures\nfor given 3D geometries, using large-scale text-guided image diffusion models.\nIn contrast to recent works that leverage 2D text-to-image diffusion models to\ndistill 3D objects using a slow and fragile optimization process, TexFusion\nintroduces a new 3D-consistent generation technique specifically designed for\ntexture synthesis that employs regular diffusion model sampling on different 2D\nrendered views. Specifically, we leverage latent diffusion models, apply the\ndiffusion model's denoiser on a set of 2D renders of the 3D object, and\naggregate the different denoising predictions on a shared latent texture map.\nFinal output RGB textures are produced by optimizing an intermediate neural\ncolor field on the decodings of 2D renders of the latent texture. We thoroughly\nvalidate TexFusion and show that we can efficiently generate diverse, high\nquality and globally coherent textures. We achieve state-of-the-art text-guided\ntexture synthesis performance using only image diffusion models, while avoiding\nthe pitfalls of previous distillation-based methods. The text-conditioning\noffers detailed control and we also do not rely on any ground truth 3D textures\nfor training. This makes our method versatile and applicable to a broad range\nof geometry and texture types. We hope that TexFusion will advance AI-based\ntexturing of 3D assets for applications in virtual reality, game design,\nsimulation, and more.\n", "date": "2023-10-20", "categories": []} {"id": "2310.15110", "title": "Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model", "abstract": " We report Zero123++, an image-conditioned diffusion model for generating\n3D-consistent multi-view images from a single input view. To take full\nadvantage of pretrained 2D generative priors, we develop various conditioning\nand training schemes to minimize the effort of finetuning from off-the-shelf\nimage diffusion models such as Stable Diffusion. Zero123++ excels in producing\nhigh-quality, consistent multi-view images from a single image, overcoming\ncommon issues like texture degradation and geometric misalignment. Furthermore,\nwe showcase the feasibility of training a ControlNet on Zero123++ for enhanced\ncontrol over the generation process. The code is available at\nhttps://github.com/SUDO-AI-3D/zero123plus.\n", "date": "2023-10-23", "categories": []} {"id": "2305.14312", "title": "Text-guided 3D Human Generation from 2D Collections", "abstract": " 3D human modeling has been widely used for engaging interaction in gaming,\nfilm, and animation. The customization of these characters is crucial for\ncreativity and scalability, which highlights the importance of controllability.\nIn this work, we introduce Text-guided 3D Human Generation (\\texttt{T3H}),\nwhere a model is to generate a 3D human, guided by the fashion description.\nThere are two goals: 1) the 3D human should render articulately, and 2) its\noutfit is controlled by the given text. To address this \\texttt{T3H} task, we\npropose Compositional Cross-modal Human (CCH). CCH adopts cross-modal attention\nto fuse compositional human rendering with the extracted fashion semantics.\nEach human body part perceives relevant textual guidance as its visual\npatterns. We incorporate the human prior and semantic discrimination to enhance\n3D geometry transformation and fine-grained consistency, enabling it to learn\nfrom 2D collections for data efficiency. We conduct evaluations on DeepFashion\nand SHHQ with diverse fashion attributes covering the shape, fabric, and color\nof upper and lower clothing. Extensive experiments demonstrate that CCH\nachieves superior results for \\texttt{T3H} with high efficiency.\n", "date": "2023-05-23", "categories": []} {"id": "2307.16605", "title": "VPP: Efficient Conditional 3D Generation via Voxel-Point Progressive\n Representation", "abstract": " Conditional 3D generation is undergoing a significant advancement, enabling\nthe free creation of 3D content from inputs such as text or 2D images. However,\nprevious approaches have suffered from low inference efficiency, limited\ngeneration categories, and restricted downstream applications. In this work, we\nrevisit the impact of different 3D representations on generation quality and\nefficiency. We propose a progressive generation method through Voxel-Point\nProgressive Representation (VPP). VPP leverages structured voxel representation\nin the proposed Voxel Semantic Generator and the sparsity of unstructured point\nrepresentation in the Point Upsampler, enabling efficient generation of\nmulti-category objects. VPP can generate high-quality 8K point clouds within\n0.2 seconds. Additionally, the masked generation Transformer allows for various\n3D downstream tasks, such as generation, editing, completion, and pre-training.\nExtensive experiments demonstrate that VPP efficiently generates high-fidelity\nand diverse 3D shapes across different categories, while also exhibiting\nexcellent representation transfer performance. Codes will be released at\n\\url{https://github.com/qizekun/VPP}.\n", "date": "2023-07-28", "categories": []} {"id": "2310.02596", "title": "SweetDreamer: Aligning Geometric Priors in 2D Diffusion for Consistent\n Text-to-3D", "abstract": " It is inherently ambiguous to lift 2D results from pre-trained diffusion\nmodels to a 3D world for text-to-3D generation. 2D diffusion models solely\nlearn view-agnostic priors and thus lack 3D knowledge during the lifting,\nleading to the multi-view inconsistency problem. We find that this problem\nprimarily stems from geometric inconsistency, and avoiding misplaced geometric\nstructures substantially mitigates the problem in the final outputs. Therefore,\nwe improve the consistency by aligning the 2D geometric priors in diffusion\nmodels with well-defined 3D shapes during the lifting, addressing the vast\nmajority of the problem. This is achieved by fine-tuning the 2D diffusion model\nto be viewpoint-aware and to produce view-specific coordinate maps of\ncanonically oriented 3D objects. In our process, only coarse 3D information is\nused for aligning. This \"coarse\" alignment not only resolves the multi-view\ninconsistency in geometries but also retains the ability in 2D diffusion models\nto generate detailed and diversified high-quality objects unseen in the 3D\ndatasets. Furthermore, our aligned geometric priors (AGP) are generic and can\nbe seamlessly integrated into various state-of-the-art pipelines, obtaining\nhigh generalizability in terms of unseen shapes and visual appearance while\ngreatly alleviating the multi-view inconsistency problem. Our method represents\na new state-of-the-art performance with an 85+% consistency rate by human\nevaluation, while many previous methods are around 30%. Our project page is\nhttps://sweetdreamer3d.github.io/\n", "date": "2023-10-04", "categories": []} {"id": "2309.08523", "title": "Breathing New Life into 3D Assets with Generative Repainting", "abstract": " Diffusion-based text-to-image models ignited immense attention from the\nvision community, artists, and content creators. Broad adoption of these models\nis due to significant improvement in the quality of generations and efficient\nconditioning on various modalities, not just text. However, lifting the rich\ngenerative priors of these 2D models into 3D is challenging. Recent works have\nproposed various pipelines powered by the entanglement of diffusion models and\nneural fields. We explore the power of pretrained 2D diffusion models and\nstandard 3D neural radiance fields as independent, standalone tools and\ndemonstrate their ability to work together in a non-learned fashion. Such\nmodularity has the intrinsic advantage of eased partial upgrades, which became\nan important property in such a fast-paced domain. Our pipeline accepts any\nlegacy renderable geometry, such as textured or untextured meshes, orchestrates\nthe interaction between 2D generative refinement and 3D consistency enforcement\ntools, and outputs a painted input geometry in several formats. We conduct a\nlarge-scale study on a wide range of objects and categories from the\nShapeNetSem dataset and demonstrate the advantages of our approach, both\nqualitatively and quantitatively. Project page:\nhttps://www.obukhov.ai/repainting_3d_assets\n", "date": "2023-09-15", "categories": []} {"id": "2310.09458", "title": "PaintHuman: Towards High-fidelity Text-to-3D Human Texturing via\n Denoised Score Distillation", "abstract": " Recent advances in zero-shot text-to-3D human generation, which employ the\nhuman model prior (eg, SMPL) or Score Distillation Sampling (SDS) with\npre-trained text-to-image diffusion models, have been groundbreaking. However,\nSDS may provide inaccurate gradient directions under the weak diffusion\nguidance, as it tends to produce over-smoothed results and generate body\ntextures that are inconsistent with the detailed mesh geometry. Therefore,\ndirectly leverage existing strategies for high-fidelity text-to-3D human\ntexturing is challenging. In this work, we propose a model called PaintHuman to\naddresses the challenges from two aspects. We first propose a novel score\nfunction, Denoised Score Distillation (DSD), which directly modifies the SDS by\nintroducing negative gradient components to iteratively correct the gradient\ndirection and generate high-quality textures. In addition, we use the depth map\nas a geometric guidance to ensure the texture is semantically aligned to human\nmesh surfaces. To guarantee the quality of rendered results, we employ\ngeometry-aware networks to predict surface materials and render realistic human\ntextures. Extensive experiments, benchmarked against state-of-the-art methods,\nvalidate the efficacy of our approach.\n", "date": "2023-10-14", "categories": []} {"id": "2310.10343", "title": "ConsistNet: Enforcing 3D Consistency for Multi-view Images Diffusion", "abstract": " Given a single image of a 3D object, this paper proposes a novel method\n(named ConsistNet) that is able to generate multiple images of the same object,\nas if seen they are captured from different viewpoints, while the 3D\n(multi-view) consistencies among those multiple generated images are\neffectively exploited. Central to our method is a multi-view consistency block\nwhich enables information exchange across multiple single-view diffusion\nprocesses based on the underlying multi-view geometry principles. ConsistNet is\nan extension to the standard latent diffusion model, and consists of two\nsub-modules: (a) a view aggregation module that unprojects multi-view features\ninto global 3D volumes and infer consistency, and (b) a ray aggregation module\nthat samples and aggregate 3D consistent features back to each view to enforce\nconsistency. Our approach departs from previous methods in multi-view image\ngeneration, in that it can be easily dropped-in pre-trained LDMs without\nrequiring explicit pixel correspondences or depth prediction. Experiments show\nthat our method effectively learns 3D consistency over a frozen Zero123\nbackbone and can generate 16 surrounding views of the object within 40 seconds\non a single A100 GPU. Our code will be made available on\nhttps://github.com/JiayuYANG/ConsistNet\n", "date": "2023-10-16", "categories": []} {"id": "2211.14108", "title": "3DDesigner: Towards Photorealistic 3D Object Generation and Editing with\n Text-guided Diffusion Models", "abstract": " Text-guided diffusion models have shown superior performance in image/video\ngeneration and editing. While few explorations have been performed in 3D\nscenarios. In this paper, we discuss three fundamental and interesting problems\non this topic. First, we equip text-guided diffusion models to achieve\n3D-consistent generation. Specifically, we integrate a NeRF-like neural field\nto generate low-resolution coarse results for a given camera view. Such results\ncan provide 3D priors as condition information for the following diffusion\nprocess. During denoising diffusion, we further enhance the 3D consistency by\nmodeling cross-view correspondences with a novel two-stream (corresponding to\ntwo different views) asynchronous diffusion process. Second, we study 3D local\nediting and propose a two-step solution that can generate 360-degree\nmanipulated results by editing an object from a single view. Step 1, we propose\nto perform 2D local editing by blending the predicted noises. Step 2, we\nconduct a noise-to-text inversion process that maps 2D blended noises into the\nview-independent text embedding space. Once the corresponding text embedding is\nobtained, 360-degree images can be generated. Last but not least, we extend our\nmodel to perform one-shot novel view synthesis by fine-tuning on a single\nimage, firstly showing the potential of leveraging text guidance for novel view\nsynthesis. Extensive experiments and various applications show the prowess of\nour 3DDesigner. The project page is available at\nhttps://3ddesigner-diffusion.github.io/.\n", "date": "2022-11-25", "categories": []} {"id": "2310.08092", "title": "Consistent123: Improve Consistency for One Image to 3D Object Synthesis", "abstract": " Large image diffusion models enable novel view synthesis with high quality\nand excellent zero-shot capability. However, such models based on\nimage-to-image translation have no guarantee of view consistency, limiting the\nperformance for downstream tasks like 3D reconstruction and image-to-3D\ngeneration. To empower consistency, we propose Consistent123 to synthesize\nnovel views simultaneously by incorporating additional cross-view attention\nlayers and the shared self-attention mechanism. The proposed attention\nmechanism improves the interaction across all synthesized views, as well as the\nalignment between the condition view and novel views. In the sampling stage,\nsuch architecture supports simultaneously generating an arbitrary number of\nviews while training at a fixed length. We also introduce a progressive\nclassifier-free guidance strategy to achieve the trade-off between texture and\ngeometry for synthesized object views. Qualitative and quantitative experiments\nshow that Consistent123 outperforms baselines in view consistency by a large\nmargin. Furthermore, we demonstrate a significant improvement of Consistent123\non varying downstream tasks, showing its great potential in the 3D generation\nfield. The project page is available at consistent-123.github.io.\n", "date": "2023-10-12", "categories": []} {"id": "2206.07255", "title": "GRAM-HD: 3D-Consistent Image Generation at High Resolution with\n Generative Radiance Manifolds", "abstract": " Recent works have shown that 3D-aware GANs trained on unstructured single\nimage collections can generate multiview images of novel instances. The key\nunderpinnings to achieve this are a 3D radiance field generator and a volume\nrendering process. However, existing methods either cannot generate\nhigh-resolution images (e.g., up to 256X256) due to the high computation cost\nof neural volume rendering, or rely on 2D CNNs for image-space upsampling which\njeopardizes the 3D consistency across different views. This paper proposes a\nnovel 3D-aware GAN that can generate high resolution images (up to 1024X1024)\nwhile keeping strict 3D consistency as in volume rendering. Our motivation is\nto achieve super-resolution directly in the 3D space to preserve 3D\nconsistency. We avoid the otherwise prohibitively-expensive computation cost by\napplying 2D convolutions on a set of 2D radiance manifolds defined in the\nrecent generative radiance manifold (GRAM) approach, and apply dedicated loss\nfunctions for effective GAN training at high resolution. Experiments on FFHQ\nand AFHQv2 datasets show that our method can produce high-quality 3D-consistent\nresults that significantly outperform existing methods. It makes a significant\nstep towards closing the gap between traditional 2D image generation and\n3D-consistent free-view generation.\n", "date": "2022-06-15", "categories": []} {"id": "2305.03040", "title": "TUVF: Learning Generalizable Texture UV Radiance Fields", "abstract": " Textures are a vital aspect of creating visually appealing and realistic 3D\nmodels. In this paper, we study the problem of generating high-fidelity texture\ngiven shapes of 3D assets, which has been relatively less explored compared\nwith generic 3D shape modeling. Our goal is to facilitate a controllable\ntexture generation process, such that one texture code can correspond to a\nparticular appearance style independent of any input shapes from a category. We\nintroduce Texture UV Radiance Fields (TUVF) that generate textures in a\nlearnable UV sphere space rather than directly on the 3D shape. This allows the\ntexture to be disentangled from the underlying shape and transferable to other\nshapes that share the same UV space, i.e., from the same category. We integrate\nthe UV sphere space with the radiance field, which provides a more efficient\nand accurate representation of textures than traditional texture maps. We\nperform our experiments on synthetic and real-world object datasets where we\nachieve not only realistic synthesis but also substantial improvements over\nstate-of-the-arts on texture controlling and editing. Project Page:\nhttps://www.anjiecheng.me/TUVF\n", "date": "2023-05-04", "categories": []} {"id": "2310.02714", "title": "GETAvatar: Generative Textured Meshes for Animatable Human Avatars", "abstract": " We study the problem of 3D-aware full-body human generation, aiming at\ncreating animatable human avatars with high-quality textures and geometries.\nGenerally, two challenges remain in this field: i) existing methods struggle to\ngenerate geometries with rich realistic details such as the wrinkles of\ngarments; ii) they typically utilize volumetric radiance fields and neural\nrenderers in the synthesis process, making high-resolution rendering\nnon-trivial. To overcome these problems, we propose GETAvatar, a Generative\nmodel that directly generates Explicit Textured 3D meshes for animatable human\nAvatar, with photo-realistic appearance and fine geometric details.\nSpecifically, we first design an articulated 3D human representation with\nexplicit surface modeling, and enrich the generated humans with realistic\nsurface details by learning from the 2D normal maps of 3D scan data. Second,\nwith the explicit mesh representation, we can use a rasterization-based\nrenderer to perform surface rendering, allowing us to achieve high-resolution\nimage generation efficiently. Extensive experiments demonstrate that GETAvatar\nachieves state-of-the-art performance on 3D-aware human generation both in\nappearance and geometry quality. Notably, GETAvatar can generate images at\n512x512 resolution with 17FPS and 1024x1024 resolution with 14FPS, improving\nupon previous methods by 2x. Our code and models will be available.\n", "date": "2023-10-04", "categories": []} {"id": "2310.02753", "title": "MUNCH: Modelling Unique 'N Controllable Heads", "abstract": " The automated generation of 3D human heads has been an intriguing and\nchallenging task for computer vision researchers. Prevailing methods synthesize\nrealistic avatars but with limited control over the diversity and quality of\nrendered outputs and suffer from limited correlation between shape and texture\nof the character. We propose a method that offers quality, diversity, control,\nand realism along with explainable network design, all desirable features to\ngame-design artists in the domain. First, our proposed Geometry Generator\nidentifies disentangled latent directions and generate novel and diverse\nsamples. A Render Map Generator then learns to synthesize multiply high-fidelty\nphysically-based render maps including Albedo, Glossiness, Specular, and\nNormals. For artists preferring fine-grained control over the output, we\nintroduce a novel Color Transformer Model that allows semantic color control\nover generated maps. We also introduce quantifiable metrics called Uniqueness\nand Novelty and a combined metric to test the overall performance of our model.\nDemo for both shapes and textures can be found:\nhttps://munch-seven.vercel.app/. We will release our model along with the\nsynthetic dataset.\n", "date": "2023-10-04", "categories": []} {"id": "2210.14267", "title": "A Survey on Deep Generative 3D-aware Image Synthesis", "abstract": " Recent years have seen remarkable progress in deep learning powered visual\ncontent creation. This includes deep generative 3D-aware image synthesis, which\nproduces high-idelity images in a 3D-consistent manner while simultaneously\ncapturing compact surfaces of objects from pure image collections without the\nneed for any 3D supervision, thus bridging the gap between 2D imagery and 3D\nreality. The ield of computer vision has been recently captivated by the task\nof deep generative 3D-aware image synthesis, with hundreds of papers appearing\nin top-tier journals and conferences over the past few years (mainly the past\ntwo years), but there lacks a comprehensive survey of this remarkable and swift\nprogress. Our survey aims to introduce new researchers to this topic, provide a\nuseful reference for related works, and stimulate future research directions\nthrough our discussion section. Apart from the presented papers, we aim to\nconstantly update the latest relevant papers along with corresponding\nimplementations at https://weihaox.github.io/3D-aware-Gen.\n", "date": "2022-10-25", "categories": []} {"id": "2302.06857", "title": "Make Your Brief Stroke Real and Stereoscopic: 3D-Aware Simplified Sketch\n to Portrait Generation", "abstract": " Creating the photo-realistic version of people sketched portraits is useful\nto various entertainment purposes. Existing studies only generate portraits in\nthe 2D plane with fixed views, making the results less vivid. In this paper, we\npresent Stereoscopic Simplified Sketch-to-Portrait (SSSP), which explores the\npossibility of creating Stereoscopic 3D-aware portraits from simple contour\nsketches by involving 3D generative models. Our key insight is to design\nsketch-aware constraints that can fully exploit the prior knowledge of a\ntri-plane-based 3D-aware generative model. Specifically, our designed\nregion-aware volume rendering strategy and global consistency constraint\nfurther enhance detail correspondences during sketch encoding. Moreover, in\norder to facilitate the usage of layman users, we propose a Contour-to-Sketch\nmodule with vector quantized representations, so that easily drawn contours can\ndirectly guide the generation of 3D portraits. Extensive comparisons show that\nour method generates high-quality results that match the sketch. Our usability\nstudy verifies that our system is greatly preferred by user.\n", "date": "2023-02-14", "categories": []} {"id": "2309.15941", "title": "AutoEncoding Tree for City Generation and Applications", "abstract": " City modeling and generation have attracted an increased interest in various\napplications, including gaming, urban planning, and autonomous driving. Unlike\nprevious works focused on the generation of single objects or indoor scenes,\nthe huge volumes of spatial data in cities pose a challenge to the generative\nmodels. Furthermore, few publicly available 3D real-world city datasets also\nhinder the development of methods for city generation. In this paper, we first\ncollect over 3,000,000 geo-referenced objects for the city of New York, Zurich,\nTokyo, Berlin, Boston and several other large cities. Based on this dataset, we\npropose AETree, a tree-structured auto-encoder neural network, for city\ngeneration. Specifically, we first propose a novel Spatial-Geometric Distance\n(SGD) metric to measure the similarity between building layouts and then\nconstruct a binary tree over the raw geometric data of building based on the\nSGD metric. Next, we present a tree-structured network whose encoder learns to\nextract and merge spatial information from bottom-up iteratively. The resulting\nglobal representation is reversely decoded for reconstruction or generation. To\naddress the issue of long-dependency as the level of the tree increases, a Long\nShort-Term Memory (LSTM) Cell is employed as a basic network element of the\nproposed AETree. Moreover, we introduce a novel metric, Overlapping Area Ratio\n(OAR), to quantitatively evaluate the generation results. Experiments on the\ncollected dataset demonstrate the effectiveness of the proposed model on 2D and\n3D city generation. Furthermore, the latent features learned by AETree can\nserve downstream urban planning applications.\n", "date": "2023-09-27", "categories": []} {"id": "2303.13873", "title": "Fantasia3D: Disentangling Geometry and Appearance for High-quality\n Text-to-3D Content Creation", "abstract": " Automatic 3D content creation has achieved rapid progress recently due to the\navailability of pre-trained, large language models and image diffusion models,\nforming the emerging topic of text-to-3D content creation. Existing text-to-3D\nmethods commonly use implicit scene representations, which couple the geometry\nand appearance via volume rendering and are suboptimal in terms of recovering\nfiner geometries and achieving photorealistic rendering; consequently, they are\nless effective for generating high-quality 3D assets. In this work, we propose\na new method of Fantasia3D for high-quality text-to-3D content creation. Key to\nFantasia3D is the disentangled modeling and learning of geometry and\nappearance. For geometry learning, we rely on a hybrid scene representation,\nand propose to encode surface normal extracted from the representation as the\ninput of the image diffusion model. For appearance modeling, we introduce the\nspatially varying bidirectional reflectance distribution function (BRDF) into\nthe text-to-3D task, and learn the surface material for photorealistic\nrendering of the generated surface. Our disentangled framework is more\ncompatible with popular graphics engines, supporting relighting, editing, and\nphysical simulation of the generated 3D assets. We conduct thorough experiments\nthat show the advantages of our method over existing ones under different\ntext-to-3D task settings. Project page and source codes:\nhttps://fantasia3d.github.io/.\n", "date": "2023-03-24", "categories": []} {"id": "2309.15830", "title": "OrthoPlanes: A Novel Representation for Better 3D-Awareness of GANs", "abstract": " We present a new method for generating realistic and view-consistent images\nwith fine geometry from 2D image collections. Our method proposes a hybrid\nexplicit-implicit representation called \\textbf{OrthoPlanes}, which encodes\nfine-grained 3D information in feature maps that can be efficiently generated\nby modifying 2D StyleGANs. Compared to previous representations, our method has\nbetter scalability and expressiveness with clear and explicit information. As a\nresult, our method can handle more challenging view-angles and synthesize\narticulated objects with high spatial degree of freedom. Experiments\ndemonstrate that our method achieves state-of-the-art results on FFHQ and SHHQ\ndatasets, both quantitatively and qualitatively. Project page:\n\\url{https://orthoplanes.github.io/}.\n", "date": "2023-09-27", "categories": []} {"id": "2303.15181", "title": "DreamStone: Image as Stepping Stone for Text-Guided 3D Shape Generation", "abstract": " In this paper, we present a new text-guided 3D shape generation approach\nDreamStone that uses images as a stepping stone to bridge the gap between text\nand shape modalities for generating 3D shapes without requiring paired text and\n3D data. The core of our approach is a two-stage feature-space alignment\nstrategy that leverages a pre-trained single-view reconstruction (SVR) model to\nmap CLIP features to shapes: to begin with, map the CLIP image feature to the\ndetail-rich 3D shape space of the SVR model, then map the CLIP text feature to\nthe 3D shape space through encouraging the CLIP-consistency between rendered\nimages and the input text. Besides, to extend beyond the generative capability\nof the SVR model, we design a text-guided 3D shape stylization module that can\nenhance the output shapes with novel structures and textures. Further, we\nexploit pre-trained text-to-image diffusion models to enhance the generative\ndiversity, fidelity, and stylization capability. Our approach is generic,\nflexible, and scalable, and it can be easily integrated with various SVR models\nto expand the generative space and improve the generative fidelity. Extensive\nexperimental results demonstrate that our approach outperforms the\nstate-of-the-art methods in terms of generative quality and consistency with\nthe input text. Codes and models are released at\nhttps://github.com/liuzhengzhe/DreamStone-ISS.\n", "date": "2023-03-24", "categories": []} {"id": "2309.14600", "title": "Progressive Text-to-3D Generation for Automatic 3D Prototyping", "abstract": " Text-to-3D generation is to craft a 3D object according to a natural language\ndescription. This can significantly reduce the workload for manually designing\n3D models and provide a more natural way of interaction for users. However,\nthis problem remains challenging in recovering the fine-grained details\neffectively and optimizing a large-size 3D output efficiently. Inspired by the\nsuccess of progressive learning, we propose a Multi-Scale Triplane Network\n(MTN) and a new progressive learning strategy. As the name implies, the\nMulti-Scale Triplane Network consists of four triplanes transitioning from low\nto high resolution. The low-resolution triplane could serve as an initial shape\nfor the high-resolution ones, easing the optimization difficulty. To further\nenable the fine-grained details, we also introduce the progressive learning\nstrategy, which explicitly demands the network to shift its focus of attention\nfrom simple coarse-grained patterns to difficult fine-grained patterns. Our\nexperiment verifies that the proposed method performs favorably against\nexisting methods. For even the most challenging descriptions, where most\nexisting methods struggle to produce a viable shape, our proposed method\nconsistently delivers. We aspire for our work to pave the way for automatic 3D\nprototyping via natural language descriptions.\n", "date": "2023-09-26", "categories": []} {"id": "2309.13006", "title": "Deep3DSketch+: Rapid 3D Modeling from Single Free-hand Sketches", "abstract": " The rapid development of AR/VR brings tremendous demands for 3D content.\nWhile the widely-used Computer-Aided Design (CAD) method requires a\ntime-consuming and labor-intensive modeling process, sketch-based 3D modeling\noffers a potential solution as a natural form of computer-human interaction.\nHowever, the sparsity and ambiguity of sketches make it challenging to generate\nhigh-fidelity content reflecting creators' ideas. Precise drawing from multiple\nviews or strategic step-by-step drawings is often required to tackle the\nchallenge but is not friendly to novice users. In this work, we introduce a\nnovel end-to-end approach, Deep3DSketch+, which performs 3D modeling using only\na single free-hand sketch without inputting multiple sketches or view\ninformation. Specifically, we introduce a lightweight generation network for\nefficient inference in real-time and a structural-aware adversarial training\napproach with a Stroke Enhancement Module (SEM) to capture the structural\ninformation to facilitate learning of the realistic and fine-detailed shape\nstructures for high-fidelity performance. Extensive experiments demonstrated\nthe effectiveness of our approach with the state-of-the-art (SOTA) performance\non both synthetic and real datasets.\n", "date": "2023-09-22", "categories": []} {"id": "2303.12048", "title": "Vox-E: Text-guided Voxel Editing of 3D Objects", "abstract": " Large scale text-guided diffusion models have garnered significant attention\ndue to their ability to synthesize diverse images that convey complex visual\nconcepts. This generative power has more recently been leveraged to perform\ntext-to-3D synthesis. In this work, we present a technique that harnesses the\npower of latent diffusion models for editing existing 3D objects. Our method\ntakes oriented 2D images of a 3D object as input and learns a grid-based\nvolumetric representation of it. To guide the volumetric representation to\nconform to a target text prompt, we follow unconditional text-to-3D methods and\noptimize a Score Distillation Sampling (SDS) loss. However, we observe that\ncombining this diffusion-guided loss with an image-based regularization loss\nthat encourages the representation not to deviate too strongly from the input\nobject is challenging, as it requires achieving two conflicting goals while\nviewing only structure-and-appearance coupled 2D projections. Thus, we\nintroduce a novel volumetric regularization loss that operates directly in 3D\nspace, utilizing the explicit nature of our 3D representation to enforce\ncorrelation between the global structure of the original and edited object.\nFurthermore, we present a technique that optimizes cross-attention volumetric\ngrids to refine the spatial extent of the edits. Extensive experiments and\ncomparisons demonstrate the effectiveness of our approach in creating a myriad\nof edits which cannot be achieved by prior works.\n", "date": "2023-03-21", "categories": []} {"id": "2309.10388", "title": "SideGAN: 3D-Aware Generative Model for Improved Side-View Image\n Synthesis", "abstract": " While recent 3D-aware generative models have shown photo-realistic image\nsynthesis with multi-view consistency, the synthesized image quality degrades\ndepending on the camera pose (e.g., a face with a blurry and noisy boundary at\na side viewpoint). Such degradation is mainly caused by the difficulty of\nlearning both pose consistency and photo-realism simultaneously from a dataset\nwith heavily imbalanced poses. In this paper, we propose SideGAN, a novel 3D\nGAN training method to generate photo-realistic images irrespective of the\ncamera pose, especially for faces of side-view angles. To ease the challenging\nproblem of learning photo-realistic and pose-consistent image synthesis, we\nsplit the problem into two subproblems, each of which can be solved more\neasily. Specifically, we formulate the problem as a combination of two simple\ndiscrimination problems, one of which learns to discriminate whether a\nsynthesized image looks real or not, and the other learns to discriminate\nwhether a synthesized image agrees with the camera pose. Based on this, we\npropose a dual-branched discriminator with two discrimination branches. We also\npropose a pose-matching loss to learn the pose consistency of 3D GANs. In\naddition, we present a pose sampling strategy to increase learning\nopportunities for steep angles in a pose-imbalanced dataset. With extensive\nvalidation, we demonstrate that our approach enables 3D GANs to generate\nhigh-quality geometries and photo-realistic images irrespective of the camera\npose.\n", "date": "2023-09-19", "categories": []} {"id": "2305.11870", "title": "Chupa: Carving 3D Clothed Humans from Skinned Shape Priors using 2D\n Diffusion Probabilistic Models", "abstract": " We propose a 3D generation pipeline that uses diffusion models to generate\nrealistic human digital avatars. Due to the wide variety of human identities,\nposes, and stochastic details, the generation of 3D human meshes has been a\nchallenging problem. To address this, we decompose the problem into 2D normal\nmap generation and normal map-based 3D reconstruction. Specifically, we first\nsimultaneously generate realistic normal maps for the front and backside of a\nclothed human, dubbed dual normal maps, using a pose-conditional diffusion\nmodel. For 3D reconstruction, we \"carve\" the prior SMPL-X mesh to a detailed 3D\nmesh according to the normal maps through mesh optimization. To further enhance\nthe high-frequency details, we present a diffusion resampling scheme on both\nbody and facial regions, thus encouraging the generation of realistic digital\navatars. We also seamlessly incorporate a recent text-to-image diffusion model\nto support text-based human identity control. Our method, namely, Chupa, is\ncapable of generating realistic 3D clothed humans with better perceptual\nquality and identity variety.\n", "date": "2023-05-19", "categories": []} {"id": "2309.07920", "title": "Large-Vocabulary 3D Diffusion Model with Transformer", "abstract": " Creating diverse and high-quality 3D assets with an automatic generative\nmodel is highly desirable. Despite extensive efforts on 3D generation, most\nexisting works focus on the generation of a single category or a few\ncategories. In this paper, we introduce a diffusion-based feed-forward\nframework for synthesizing massive categories of real-world 3D objects with a\nsingle generative model. Notably, there are three major challenges for this\nlarge-vocabulary 3D generation: a) the need for expressive yet efficient 3D\nrepresentation; b) large diversity in geometry and texture across categories;\nc) complexity in the appearances of real-world objects. To this end, we propose\na novel triplane-based 3D-aware Diffusion model with TransFormer, DiffTF, for\nhandling challenges via three aspects. 1) Considering efficiency and\nrobustness, we adopt a revised triplane representation and improve the fitting\nspeed and accuracy. 2) To handle the drastic variations in geometry and\ntexture, we regard the features of all 3D objects as a combination of\ngeneralized 3D knowledge and specialized 3D features. To extract generalized 3D\nknowledge from diverse categories, we propose a novel 3D-aware transformer with\nshared cross-plane attention. It learns the cross-plane relations across\ndifferent planes and aggregates the generalized 3D knowledge with specialized\n3D features. 3) In addition, we devise the 3D-aware encoder/decoder to enhance\nthe generalized 3D knowledge in the encoded triplanes for handling categories\nwith complex appearances. Extensive experiments on ShapeNet and OmniObject3D\n(over 200 diverse real-world categories) convincingly demonstrate that a single\nDiffTF model achieves state-of-the-art large-vocabulary 3D object generation\nperformance with large diversity, rich semantics, and high quality.\n", "date": "2023-09-14", "categories": []} {"id": "2307.06526", "title": "AvatarFusion: Zero-shot Generation of Clothing-Decoupled 3D Avatars\n Using 2D Diffusion", "abstract": " Large-scale pre-trained vision-language models allow for the zero-shot\ntext-based generation of 3D avatars. The previous state-of-the-art method\nutilized CLIP to supervise neural implicit models that reconstructed a human\nbody mesh. However, this approach has two limitations. Firstly, the lack of\navatar-specific models can cause facial distortion and unrealistic clothing in\nthe generated avatars. Secondly, CLIP only provides optimization direction for\nthe overall appearance, resulting in less impressive results. To address these\nlimitations, we propose AvatarFusion, the first framework to use a latent\ndiffusion model to provide pixel-level guidance for generating human-realistic\navatars while simultaneously segmenting clothing from the avatar's body.\nAvatarFusion includes the first clothing-decoupled neural implicit avatar model\nthat employs a novel Dual Volume Rendering strategy to render the decoupled\nskin and clothing sub-models in one space. We also introduce a novel\noptimization method, called Pixel-Semantics Difference-Sampling (PS-DS), which\nsemantically separates the generation of body and clothes, and generates a\nvariety of clothing styles. Moreover, we establish the first benchmark for\nzero-shot text-to-avatar generation. Our experimental results demonstrate that\nour framework outperforms previous approaches, with significant improvements\nobserved in all metrics. Additionally, since our model is clothing-decoupled,\nwe can exchange the clothes of avatars. Code are available on our project page\nhttps://hansenhuang0823.github.io/AvatarFusion.\n", "date": "2023-07-13", "categories": []} {"id": "2211.10865", "title": "IC3D: Image-Conditioned 3D Diffusion for Shape Generation", "abstract": " In recent years, Denoising Diffusion Probabilistic Models (DDPMs) have\ndemonstrated exceptional performance in various 2D generative tasks. Following\nthis success, DDPMs have been extended to 3D shape generation, surpassing\nprevious methodologies in this domain. While many of these models are\nunconditional, some have explored the potential of using guidance from\ndifferent modalities. In particular, image guidance for 3D generation has been\nexplored through the utilization of CLIP embeddings. However, these embeddings\nare designed to align images and text, and do not necessarily capture the\nspecific details needed for shape generation. To address this limitation and\nenhance image-guided 3D DDPMs with augmented 3D understanding, we introduce\nCISP (Contrastive Image-Shape Pre-training), obtaining a well-structured\nimage-shape joint embedding space. Building upon CISP, we then introduce IC3D,\na DDPM that harnesses CISP's guidance for 3D shape generation from single-view\nimages. This generative diffusion model outperforms existing benchmarks in both\nquality and diversity of generated 3D shapes. Moreover, despite IC3D's\ngenerative nature, its generated shapes are preferred by human evaluators over\na competitive single-view 3D reconstruction model. These properties contribute\nto a coherent embedding space, enabling latent interpolation and conditioned\ngeneration also from out-of-distribution images. We find IC3D able to generate\ncoherent and diverse completions also when presented with occluded views,\nrendering it applicable in controlled real-world scenarios.\n", "date": "2022-11-20", "categories": []} {"id": "2309.07125", "title": "Text-Guided Generation and Editing of Compositional 3D Avatars", "abstract": " Our goal is to create a realistic 3D facial avatar with hair and accessories\nusing only a text description. While this challenge has attracted significant\nrecent interest, existing methods either lack realism, produce unrealistic\nshapes, or do not support editing, such as modifications to the hairstyle. We\nargue that existing methods are limited because they employ a monolithic\nmodeling approach, using a single representation for the head, face, hair, and\naccessories. Our observation is that the hair and face, for example, have very\ndifferent structural qualities that benefit from different representations.\nBuilding on this insight, we generate avatars with a compositional model, in\nwhich the head, face, and upper body are represented with traditional 3D\nmeshes, and the hair, clothing, and accessories with neural radiance fields\n(NeRF). The model-based mesh representation provides a strong geometric prior\nfor the face region, improving realism while enabling editing of the person's\nappearance. By using NeRFs to represent the remaining components, our method is\nable to model and synthesize parts with complex geometry and appearance, such\nas curly hair and fluffy scarves. Our novel system synthesizes these\nhigh-quality compositional avatars from text descriptions. The experimental\nresults demonstrate that our method, Text-guided generation and Editing of\nCompositional Avatars (TECA), produces avatars that are more realistic than\nthose of recent methods while being editable because of their compositional\nnature. For example, our TECA enables the seamless transfer of compositional\nfeatures like hairstyles, scarves, and other accessories between avatars. This\ncapability supports applications such as virtual try-on.\n", "date": "2023-09-13", "categories": []} {"id": "2303.11989", "title": "Text2Room: Extracting Textured 3D Meshes from 2D Text-to-Image Models", "abstract": " We present Text2Room, a method for generating room-scale textured 3D meshes\nfrom a given text prompt as input. To this end, we leverage pre-trained 2D\ntext-to-image models to synthesize a sequence of images from different poses.\nIn order to lift these outputs into a consistent 3D scene representation, we\ncombine monocular depth estimation with a text-conditioned inpainting model.\nThe core idea of our approach is a tailored viewpoint selection such that the\ncontent of each image can be fused into a seamless, textured 3D mesh. More\nspecifically, we propose a continuous alignment strategy that iteratively fuses\nscene frames with the existing geometry to create a seamless mesh. Unlike\nexisting works that focus on generating single objects or zoom-out trajectories\nfrom text, our method generates complete 3D scenes with multiple objects and\nexplicit 3D geometry. We evaluate our approach using qualitative and\nquantitative metrics, demonstrating it as the first method to generate\nroom-scale 3D geometry with compelling textures from only text as input.\n", "date": "2023-03-21", "categories": []} {"id": "2303.12074", "title": "CC3D: Layout-Conditioned Generation of Compositional 3D Scenes", "abstract": " In this work, we introduce CC3D, a conditional generative model that\nsynthesizes complex 3D scenes conditioned on 2D semantic scene layouts, trained\nusing single-view images. Different from most existing 3D GANs that limit their\napplicability to aligned single objects, we focus on generating complex scenes\nwith multiple objects, by modeling the compositional nature of 3D scenes. By\ndevising a 2D layout-based approach for 3D synthesis and implementing a new 3D\nfield representation with a stronger geometric inductive bias, we have created\na 3D GAN that is both efficient and of high quality, while allowing for a more\ncontrollable generation process. Our evaluations on synthetic 3D-FRONT and\nreal-world KITTI-360 datasets demonstrate that our model generates scenes of\nimproved visual and geometric quality in comparison to previous works.\n", "date": "2023-03-21", "categories": []} {"id": "2309.04800", "title": "VeRi3D: Generative Vertex-based Radiance Fields for 3D Controllable\n Human Image Synthesis", "abstract": " Unsupervised learning of 3D-aware generative adversarial networks has lately\nmade much progress. Some recent work demonstrates promising results of learning\nhuman generative models using neural articulated radiance fields, yet their\ngeneralization ability and controllability lag behind parametric human models,\ni.e., they do not perform well when generalizing to novel pose/shape and are\nnot part controllable. To solve these problems, we propose VeRi3D, a generative\nhuman vertex-based radiance field parameterized by vertices of the parametric\nhuman template, SMPL. We map each 3D point to the local coordinate system\ndefined on its neighboring vertices, and use the corresponding vertex feature\nand local coordinates for mapping it to color and density values. We\ndemonstrate that our simple approach allows for generating photorealistic human\nimages with free control over camera pose, human pose, shape, as well as\nenabling part-level editing.\n", "date": "2023-09-09", "categories": []} {"id": "2309.01855", "title": "SMPLitex: A Generative Model and Dataset for 3D Human Texture Estimation\n from Single Image", "abstract": " We propose SMPLitex, a method for estimating and manipulating the complete 3D\nappearance of humans captured from a single image. SMPLitex builds upon the\nrecently proposed generative models for 2D images, and extends their use to the\n3D domain through pixel-to-surface correspondences computed on the input image.\nTo this end, we first train a generative model for complete 3D human\nappearance, and then fit it into the input image by conditioning the generative\nmodel to the visible parts of the subject. Furthermore, we propose a new\ndataset of high-quality human textures built by sampling SMPLitex conditioned\non subject descriptions and images. We quantitatively and qualitatively\nevaluate our method in 3 publicly available datasets, demonstrating that\nSMPLitex significantly outperforms existing methods for human texture\nestimation while allowing for a wider variety of tasks such as editing,\nsynthesis, and manipulation\n", "date": "2023-09-04", "categories": []} {"id": "2306.13455", "title": "DreamEditor: Text-Driven 3D Scene Editing with Neural Fields", "abstract": " Neural fields have achieved impressive advancements in view synthesis and\nscene reconstruction. However, editing these neural fields remains challenging\ndue to the implicit encoding of geometry and texture information. In this\npaper, we propose DreamEditor, a novel framework that enables users to perform\ncontrolled editing of neural fields using text prompts. By representing scenes\nas mesh-based neural fields, DreamEditor allows localized editing within\nspecific regions. DreamEditor utilizes the text encoder of a pretrained\ntext-to-Image diffusion model to automatically identify the regions to be\nedited based on the semantics of the text prompts. Subsequently, DreamEditor\noptimizes the editing region and aligns its geometry and texture with the text\nprompts through score distillation sampling [29]. Extensive experiments have\ndemonstrated that DreamEditor can accurately edit neural fields of real-world\nscenes according to the given text prompts while ensuring consistency in\nirrelevant areas. DreamEditor generates highly realistic textures and geometry,\nsignificantly surpassing previous works in both quantitative and qualitative\nevaluations.\n", "date": "2023-06-23", "categories": []} {"id": "2309.03550", "title": "Text2Control3D: Controllable 3D Avatar Generation in Neural Radiance\n Fields using Geometry-Guided Text-to-Image Diffusion Model", "abstract": " Recent advances in diffusion models such as ControlNet have enabled\ngeometrically controllable, high-fidelity text-to-image generation. However,\nnone of them addresses the question of adding such controllability to\ntext-to-3D generation. In response, we propose Text2Control3D, a controllable\ntext-to-3D avatar generation method whose facial expression is controllable\ngiven a monocular video casually captured with hand-held camera. Our main\nstrategy is to construct the 3D avatar in Neural Radiance Fields (NeRF)\noptimized with a set of controlled viewpoint-aware images that we generate from\nControlNet, whose condition input is the depth map extracted from the input\nvideo. When generating the viewpoint-aware images, we utilize cross-reference\nattention to inject well-controlled, referential facial expression and\nappearance via cross attention. We also conduct low-pass filtering of Gaussian\nlatent of the diffusion model in order to ameliorate the viewpoint-agnostic\ntexture problem we observed from our empirical analysis, where the\nviewpoint-aware images contain identical textures on identical pixel positions\nthat are incomprehensible in 3D. Finally, to train NeRF with the images that\nare viewpoint-aware yet are not strictly consistent in geometry, our approach\nconsiders per-image geometric variation as a view of deformation from a shared\n3D canonical space. Consequently, we construct the 3D avatar in a canonical\nspace of deformable NeRF by learning a set of per-image deformation via\ndeformation field table. We demonstrate the empirical results and discuss the\neffectiveness of our method.\n", "date": "2023-09-07", "categories": []} {"id": "2309.03599", "title": "Chasing Consistency in Text-to-3D Generation from a Single Image", "abstract": " Text-to-3D generation from a single-view image is a popular but challenging\ntask in 3D vision. Although numerous methods have been proposed, existing works\nstill suffer from the inconsistency issues, including 1) semantic\ninconsistency, 2) geometric inconsistency, and 3) saturation inconsistency,\nresulting in distorted, overfitted, and over-saturated generations. In light of\nthe above issues, we present Consist3D, a three-stage framework Chasing for\nsemantic-, geometric-, and saturation-Consistent Text-to-3D generation from a\nsingle image, in which the first two stages aim to learn parameterized\nconsistency tokens, and the last stage is for optimization. Specifically, the\nsemantic encoding stage learns a token independent of views and estimations,\npromoting semantic consistency and robustness. Meanwhile, the geometric\nencoding stage learns another token with comprehensive geometry and\nreconstruction constraints under novel-view estimations, reducing overfitting\nand encouraging geometric consistency. Finally, the optimization stage benefits\nfrom the semantic and geometric tokens, allowing a low classifier-free guidance\nscale and therefore preventing oversaturation. Experimental results demonstrate\nthat Consist3D produces more consistent, faithful, and photo-realistic 3D\nassets compared to previous state-of-the-art methods. Furthermore, Consist3D\nalso allows background and object editing through text prompts.\n", "date": "2023-09-07", "categories": []} {"id": "2212.01381", "title": "LatentSwap3D: Semantic Edits on 3D Image GANs", "abstract": " 3D GANs have the ability to generate latent codes for entire 3D volumes\nrather than only 2D images. These models offer desirable features like\nhigh-quality geometry and multi-view consistency, but, unlike their 2D\ncounterparts, complex semantic image editing tasks for 3D GANs have only been\npartially explored. To address this problem, we propose LatentSwap3D, a\nsemantic edit approach based on latent space discovery that can be used with\nany off-the-shelf 3D or 2D GAN model and on any dataset. LatentSwap3D relies on\nidentifying the latent code dimensions corresponding to specific attributes by\nfeature ranking using a random forest classifier. It then performs the edit by\nswapping the selected dimensions of the image being edited with the ones from\nan automatically selected reference image. Compared to other latent space\ncontrol-based edit methods, which were mainly designed for 2D GANs, our method\non 3D GANs provides remarkably consistent semantic edits in a disentangled\nmanner and outperforms others both qualitatively and quantitatively. We show\nresults on seven 3D GANs (pi-GAN, GIRAFFE, StyleSDF, MVCGAN, EG3D, StyleNeRF,\nand VolumeGAN) and on five datasets (FFHQ, AFHQ, Cats, MetFaces, and CompCars).\n", "date": "2022-12-02", "categories": []} {"id": "2309.02186", "title": "AniPortraitGAN: Animatable 3D Portrait Generation from 2D Image\n Collections", "abstract": " Previous animatable 3D-aware GANs for human generation have primarily focused\non either the human head or full body. However, head-only videos are relatively\nuncommon in real life, and full body generation typically does not deal with\nfacial expression control and still has challenges in generating high-quality\nresults. Towards applicable video avatars, we present an animatable 3D-aware\nGAN that generates portrait images with controllable facial expression, head\npose, and shoulder movements. It is a generative model trained on unstructured\n2D image collections without using 3D or video data. For the new task, we base\nour method on the generative radiance manifold representation and equip it with\nlearnable facial and head-shoulder deformations. A dual-camera rendering and\nadversarial learning scheme is proposed to improve the quality of the generated\nfaces, which is critical for portrait images. A pose deformation processing\nnetwork is developed to generate plausible deformations for challenging regions\nsuch as long hair. Experiments show that our method, trained on unstructured 2D\nimages, can generate diverse and high-quality 3D portraits with desired control\nover different properties.\n", "date": "2023-09-05", "categories": []} {"id": "2306.07881", "title": "Viewset Diffusion: (0-)Image-Conditioned 3D Generative Models from 2D\n Data", "abstract": " We present Viewset Diffusion, a diffusion-based generator that outputs 3D\nobjects while only using multi-view 2D data for supervision. We note that there\nexists a one-to-one mapping between viewsets, i.e., collections of several 2D\nviews of an object, and 3D models. Hence, we train a diffusion model to\ngenerate viewsets, but design the neural network generator to reconstruct\ninternally corresponding 3D models, thus generating those too. We fit a\ndiffusion model to a large number of viewsets for a given category of objects.\nThe resulting generator can be conditioned on zero, one or more input views.\nConditioned on a single view, it performs 3D reconstruction accounting for the\nambiguity of the task and allowing to sample multiple solutions compatible with\nthe input. The model performs reconstruction efficiently, in a feed-forward\nmanner, and is trained using only rendering losses using as few as three views\nper viewset. Project page: szymanowiczs.github.io/viewset-diffusion.\n", "date": "2023-06-13", "categories": []} {"id": "2309.00158", "title": "BuilDiff: 3D Building Shape Generation using Single-Image Conditional\n Point Cloud Diffusion Models", "abstract": " 3D building generation with low data acquisition costs, such as single\nimage-to-3D, becomes increasingly important. However, most of the existing\nsingle image-to-3D building creation works are restricted to those images with\nspecific viewing angles, hence they are difficult to scale to general-view\nimages that commonly appear in practical cases. To fill this gap, we propose a\nnovel 3D building shape generation method exploiting point cloud diffusion\nmodels with image conditioning schemes, which demonstrates flexibility to the\ninput images. By cooperating two conditional diffusion models and introducing a\nregularization strategy during denoising process, our method is able to\nsynthesize building roofs while maintaining the overall structures. We validate\nour framework on two newly built datasets and extensive experiments show that\nour method outperforms previous works in terms of building generation quality.\n", "date": "2023-08-31", "categories": []} {"id": "2308.16758", "title": "Towards High-Fidelity Text-Guided 3D Face Generation and Manipulation\n Using only Images", "abstract": " Generating 3D faces from textual descriptions has a multitude of\napplications, such as gaming, movie, and robotics. Recent progresses have\ndemonstrated the success of unconditional 3D face generation and text-to-3D\nshape generation. However, due to the limited text-3D face data pairs,\ntext-driven 3D face generation remains an open problem. In this paper, we\npropose a text-guided 3D faces generation method, refer as TG-3DFace, for\ngenerating realistic 3D faces using text guidance. Specifically, we adopt an\nunconditional 3D face generation framework and equip it with text conditions,\nwhich learns the text-guided 3D face generation with only text-2D face data. On\ntop of that, we propose two text-to-face cross-modal alignment techniques,\nincluding the global contrastive learning and the fine-grained alignment\nmodule, to facilitate high semantic consistency between generated 3D faces and\ninput texts. Besides, we present directional classifier guidance during the\ninference process, which encourages creativity for out-of-domain generations.\nCompared to the existing methods, TG-3DFace creates more realistic and\naesthetically pleasing 3D faces, boosting 9% multi-view consistency (MVIC) over\nLatent3D. The rendered face images generated by TG-3DFace achieve higher FID\nand CLIP score than text-to-2D face/image generation models, demonstrating our\nsuperiority in generating realistic and semantic-consistent textures.\n", "date": "2023-08-31", "categories": []} {"id": "2306.03038", "title": "HeadSculpt: Crafting 3D Head Avatars with Text", "abstract": " Recently, text-guided 3D generative methods have made remarkable advancements\nin producing high-quality textures and geometry, capitalizing on the\nproliferation of large vision-language and image diffusion models. However,\nexisting methods still struggle to create high-fidelity 3D head avatars in two\naspects: (1) They rely mostly on a pre-trained text-to-image diffusion model\nwhilst missing the necessary 3D awareness and head priors. This makes them\nprone to inconsistency and geometric distortions in the generated avatars. (2)\nThey fall short in fine-grained editing. This is primarily due to the inherited\nlimitations from the pre-trained 2D image diffusion models, which become more\npronounced when it comes to 3D head avatars. In this work, we address these\nchallenges by introducing a versatile coarse-to-fine pipeline dubbed HeadSculpt\nfor crafting (i.e., generating and editing) 3D head avatars from textual\nprompts. Specifically, we first equip the diffusion model with 3D awareness by\nleveraging landmark-based control and a learned textual embedding representing\nthe back view appearance of heads, enabling 3D-consistent head avatar\ngenerations. We further propose a novel identity-aware editing score\ndistillation strategy to optimize a textured mesh with a high-resolution\ndifferentiable rendering technique. This enables identity preservation while\nfollowing the editing instruction. We showcase HeadSculpt's superior fidelity\nand editing capabilities through comprehensive experiments and comparisons with\nexisting methods.\n", "date": "2023-06-05", "categories": []} {"id": "2210.15663", "title": "Deep Generative Models on 3D Representations: A Survey", "abstract": " Generative models aim to learn the distribution of observed data by\ngenerating new instances. With the advent of neural networks, deep generative\nmodels, including variational autoencoders (VAEs), generative adversarial\nnetworks (GANs), and diffusion models (DMs), have progressed remarkably in\nsynthesizing 2D images. Recently, researchers started to shift focus from 2D to\n3D space, considering that 3D data is more closely aligned with our physical\nworld and holds immense practical potential. However, unlike 2D images, which\npossess an inherent and efficient representation (\\textit{i.e.}, a pixel grid),\nrepresenting 3D data poses significantly greater challenges. Ideally, a robust\n3D representation should be capable of accurately modeling complex shapes and\nappearances while being highly efficient in handling high-resolution data with\nhigh processing speeds and low memory requirements. Regrettably, existing 3D\nrepresentations, such as point clouds, meshes, and neural fields, often fail to\nsatisfy all of these requirements simultaneously. In this survey, we thoroughly\nreview the ongoing developments of 3D generative models, including methods that\nemploy 2D and 3D supervision. Our analysis centers on generative models, with a\nparticular focus on the representations utilized in this context. We believe\nour survey will help the community to track the field's evolution and to spark\ninnovative ideas to propel progress towards solving this challenging task.\n", "date": "2022-10-27", "categories": []} {"id": "2304.02163", "title": "GINA-3D: Learning to Generate Implicit Neural Assets in the Wild", "abstract": " Modeling the 3D world from sensor data for simulation is a scalable way of\ndeveloping testing and validation environments for robotic learning problems\nsuch as autonomous driving. However, manually creating or re-creating\nreal-world-like environments is difficult, expensive, and not scalable. Recent\ngenerative model techniques have shown promising progress to address such\nchallenges by learning 3D assets using only plentiful 2D images -- but still\nsuffer limitations as they leverage either human-curated image datasets or\nrenderings from manually-created synthetic 3D environments. In this paper, we\nintroduce GINA-3D, a generative model that uses real-world driving data from\ncamera and LiDAR sensors to create realistic 3D implicit neural assets of\ndiverse vehicles and pedestrians. Compared to the existing image datasets, the\nreal-world driving setting poses new challenges due to occlusions,\nlighting-variations and long-tail distributions. GINA-3D tackles these\nchallenges by decoupling representation learning and generative modeling into\ntwo stages with a learned tri-plane latent structure, inspired by recent\nadvances in generative modeling of images. To evaluate our approach, we\nconstruct a large-scale object-centric dataset containing over 1.2M images of\nvehicles and pedestrians from the Waymo Open Dataset, and a new set of 80K\nimages of long-tail instances such as construction equipment, garbage trucks,\nand cable cars. We compare our model with existing approaches and demonstrate\nthat it achieves state-of-the-art performance in quality and diversity for both\ngenerated images and geometries.\n", "date": "2023-04-04", "categories": []} {"id": "2308.14152", "title": "Unaligned 2D to 3D Translation with Conditional Vector-Quantized Code\n Diffusion using Transformers", "abstract": " Generating 3D images of complex objects conditionally from a few 2D views is\na difficult synthesis problem, compounded by issues such as domain gap and\ngeometric misalignment. For instance, a unified framework such as Generative\nAdversarial Networks cannot achieve this unless they explicitly define both a\ndomain-invariant and geometric-invariant joint latent distribution, whereas\nNeural Radiance Fields are generally unable to handle both issues as they\noptimize at the pixel level. By contrast, we propose a simple and novel 2D to\n3D synthesis approach based on conditional diffusion with vector-quantized\ncodes. Operating in an information-rich code space enables high-resolution 3D\nsynthesis via full-coverage attention across the views. Specifically, we\ngenerate the 3D codes (e.g. for CT images) conditional on previously generated\n3D codes and the entire codebook of two 2D views (e.g. 2D X-rays). Qualitative\nand quantitative results demonstrate state-of-the-art performance over\nspecialized methods across varied evaluation criteria, including fidelity\nmetrics such as density, coverage, and distortion metrics for two complex\nvolumetric imagery datasets from in real-world scenarios.\n", "date": "2023-08-27", "categories": []} {"id": "2308.14244", "title": "HoloFusion: Towards Photo-realistic 3D Generative Modeling", "abstract": " Diffusion-based image generators can now produce high-quality and diverse\nsamples, but their success has yet to fully translate to 3D generation:\nexisting diffusion methods can either generate low-resolution but 3D consistent\noutputs, or detailed 2D views of 3D objects but with potential structural\ndefects and lacking view consistency or realism. We present HoloFusion, a\nmethod that combines the best of these approaches to produce high-fidelity,\nplausible, and diverse 3D samples while learning from a collection of\nmulti-view 2D images only. The method first generates coarse 3D samples using a\nvariant of the recently proposed HoloDiffusion generator. Then, it\nindependently renders and upsamples a large number of views of the coarse 3D\nmodel, super-resolves them to add detail, and distills those into a single,\nhigh-fidelity implicit 3D representation, which also ensures view consistency\nof the final renders. The super-resolution network is trained as an integral\npart of HoloFusion, end-to-end, and the final distillation uses a new sampling\nscheme to capture the space of super-resolved signals. We compare our method\nagainst existing baselines, including DreamFusion, Get3D, EG3D, and\nHoloDiffusion, and achieve, to the best of our knowledge, the most realistic\nresults on the challenging CO3Dv2 dataset.\n", "date": "2023-08-28", "categories": []} {"id": "2304.06714", "title": "Single-Stage Diffusion NeRF: A Unified Approach to 3D Generation and\n Reconstruction", "abstract": " 3D-aware image synthesis encompasses a variety of tasks, such as scene\ngeneration and novel view synthesis from images. Despite numerous task-specific\nmethods, developing a comprehensive model remains challenging. In this paper,\nwe present SSDNeRF, a unified approach that employs an expressive diffusion\nmodel to learn a generalizable prior of neural radiance fields (NeRF) from\nmulti-view images of diverse objects. Previous studies have used two-stage\napproaches that rely on pretrained NeRFs as real data to train diffusion\nmodels. In contrast, we propose a new single-stage training paradigm with an\nend-to-end objective that jointly optimizes a NeRF auto-decoder and a latent\ndiffusion model, enabling simultaneous 3D reconstruction and prior learning,\neven from sparsely available views. At test time, we can directly sample the\ndiffusion prior for unconditional generation, or combine it with arbitrary\nobservations of unseen objects for NeRF reconstruction. SSDNeRF demonstrates\nrobust results comparable to or better than leading task-specific methods in\nunconditional generation and single/sparse-view 3D reconstruction.\n", "date": "2023-04-13", "categories": []} {"id": "2308.10608", "title": "FocalDreamer: Text-driven 3D Editing via Focal-fusion Assembly", "abstract": " While text-3D editing has made significant strides in leveraging score\ndistillation sampling, emerging approaches still fall short in delivering\nseparable, precise and consistent outcomes that are vital to content creation.\nIn response, we introduce FocalDreamer, a framework that merges base shape with\neditable parts according to text prompts for fine-grained editing within\ndesired regions. Specifically, equipped with geometry union and dual-path\nrendering, FocalDreamer assembles independent 3D parts into a complete object,\ntailored for convenient instance reuse and part-wise control. We propose\ngeometric focal loss and style consistency regularization, which encourage\nfocal fusion and congruent overall appearance. Furthermore, FocalDreamer\ngenerates high-fidelity geometry and PBR textures which are compatible with\nwidely-used graphics engines. Extensive experiments have highlighted the\nsuperior editing capabilities of FocalDreamer in both quantitative and\nqualitative evaluations.\n", "date": "2023-08-21", "categories": []} {"id": "2308.11473", "title": "IT3D: Improved Text-to-3D Generation with Explicit View Synthesis", "abstract": " Recent strides in Text-to-3D techniques have been propelled by distilling\nknowledge from powerful large text-to-image diffusion models (LDMs).\nNonetheless, existing Text-to-3D approaches often grapple with challenges such\nas over-saturation, inadequate detailing, and unrealistic outputs. This study\npresents a novel strategy that leverages explicitly synthesized multi-view\nimages to address these issues. Our approach involves the utilization of\nimage-to-image pipelines, empowered by LDMs, to generate posed high-quality\nimages based on the renderings of coarse 3D models. Although the generated\nimages mostly alleviate the aforementioned issues, challenges such as view\ninconsistency and significant content variance persist due to the inherent\ngenerative nature of large diffusion models, posing extensive difficulties in\nleveraging these images effectively. To overcome this hurdle, we advocate\nintegrating a discriminator alongside a novel Diffusion-GAN dual training\nstrategy to guide the training of 3D models. For the incorporated\ndiscriminator, the synthesized multi-view images are considered real data,\nwhile the renderings of the optimized 3D models function as fake data. We\nconduct a comprehensive set of experiments that demonstrate the effectiveness\nof our method over baseline approaches.\n", "date": "2023-08-22", "categories": []} {"id": "2303.10735", "title": "SKED: Sketch-guided Text-based 3D Editing", "abstract": " Text-to-image diffusion models are gradually introduced into computer\ngraphics, recently enabling the development of Text-to-3D pipelines in an open\ndomain. However, for interactive editing purposes, local manipulations of\ncontent through a simplistic textual interface can be arduous. Incorporating\nuser guided sketches with Text-to-image pipelines offers users more intuitive\ncontrol. Still, as state-of-the-art Text-to-3D pipelines rely on optimizing\nNeural Radiance Fields (NeRF) through gradients from arbitrary rendering views,\nconditioning on sketches is not straightforward. In this paper, we present\nSKED, a technique for editing 3D shapes represented by NeRFs. Our technique\nutilizes as few as two guiding sketches from different views to alter an\nexisting neural field. The edited region respects the prompt semantics through\na pre-trained diffusion model. To ensure the generated output adheres to the\nprovided sketches, we propose novel loss functions to generate the desired\nedits while preserving the density and radiance of the base instance. We\ndemonstrate the effectiveness of our proposed method through several\nqualitative and quantitative experiments. https://sked-paper.github.io/\n", "date": "2023-03-19", "categories": []} {"id": "2303.17606", "title": "AvatarCraft: Transforming Text into Neural Human Avatars with\n Parameterized Shape and Pose Control", "abstract": " Neural implicit fields are powerful for representing 3D scenes and generating\nhigh-quality novel views, but it remains challenging to use such implicit\nrepresentations for creating a 3D human avatar with a specific identity and\nartistic style that can be easily animated. Our proposed method, AvatarCraft,\naddresses this challenge by using diffusion models to guide the learning of\ngeometry and texture for a neural avatar based on a single text prompt. We\ncarefully design the optimization framework of neural implicit fields,\nincluding a coarse-to-fine multi-bounding box training strategy, shape\nregularization, and diffusion-based constraints, to produce high-quality\ngeometry and texture. Additionally, we make the human avatar animatable by\ndeforming the neural implicit field with an explicit warping field that maps\nthe target human mesh to a template human mesh, both represented using\nparametric human models. This simplifies animation and reshaping of the\ngenerated avatar by controlling pose and shape parameters. Extensive\nexperiments on various text descriptions show that AvatarCraft is effective and\nrobust in creating human avatars and rendering novel views, poses, and shapes.\nOur project page is: https://avatar-craft.github.io/.\n", "date": "2023-03-30", "categories": []} {"id": "2305.01921", "title": "DiffFacto: Controllable Part-Based 3D Point Cloud Generation with Cross\n Diffusion", "abstract": " While the community of 3D point cloud generation has witnessed a big growth\nin recent years, there still lacks an effective way to enable intuitive user\ncontrol in the generation process, hence limiting the general utility of such\nmethods. Since an intuitive way of decomposing a shape is through its parts, we\npropose to tackle the task of controllable part-based point cloud generation.\nWe introduce DiffFacto, a novel probabilistic generative model that learns the\ndistribution of shapes with part-level control. We propose a factorization that\nmodels independent part style and part configuration distributions and presents\na novel cross-diffusion network that enables us to generate coherent and\nplausible shapes under our proposed factorization. Experiments show that our\nmethod is able to generate novel shapes with multiple axes of control. It\nachieves state-of-the-art part-level generation quality and generates plausible\nand coherent shapes while enabling various downstream editing applications such\nas shape interpolation, mixing, and transformation editing. Project website:\nhttps://difffacto.github.io/\n", "date": "2023-05-03", "categories": []} {"id": "2306.11510", "title": "Pushing the Limits of 3D Shape Generation at Scale", "abstract": " We present a significant breakthrough in 3D shape generation by scaling it to\nunprecedented dimensions. Through the adaptation of the Auto-Regressive model\nand the utilization of large language models, we have developed a remarkable\nmodel with an astounding 3.6 billion trainable parameters, establishing it as\nthe largest 3D shape generation model to date, named Argus-3D. Our approach\naddresses the limitations of existing methods by enhancing the quality and\ndiversity of generated 3D shapes. To tackle the challenges of high-resolution\n3D shape generation, our model incorporates tri-plane features as latent\nrepresentations, effectively reducing computational complexity. Additionally,\nwe introduce a discrete codebook for efficient quantization of these\nrepresentations. Leveraging the power of transformers, we enable multi-modal\nconditional generation, facilitating the production of diverse and visually\nimpressive 3D shapes. To train our expansive model, we leverage an ensemble of\npublicly-available 3D datasets, consisting of a comprehensive collection of\napproximately 900,000 objects from renowned repositories such as ModelNet40,\nShapeNet, Pix3D, 3D-Future, and Objaverse. This diverse dataset empowers our\nmodel to learn from a wide range of object variations, bolstering its ability\nto generate high-quality and diverse 3D shapes. Extensive experimentation\ndemonstrate the remarkable efficacy of our approach in significantly improving\nthe visual quality of generated 3D shapes. By pushing the boundaries of 3D\ngeneration, introducing novel methods for latent representation learning, and\nharnessing the power of transformers for multi-modal conditional generation,\nour contributions pave the way for substantial advancements in the field. Our\nwork unlocks new possibilities for applications in gaming, virtual reality,\nproduct design, and other domains that demand high-quality and diverse 3D\nobjects.\n", "date": "2023-06-20", "categories": []} {"id": "2308.10490", "title": "Texture Generation on 3D Meshes with Point-UV Diffusion", "abstract": " In this work, we focus on synthesizing high-quality textures on 3D meshes. We\npresent Point-UV diffusion, a coarse-to-fine pipeline that marries the\ndenoising diffusion model with UV mapping to generate 3D consistent and\nhigh-quality texture images in UV space. We start with introducing a point\ndiffusion model to synthesize low-frequency texture components with our\ntailored style guidance to tackle the biased color distribution. The derived\ncoarse texture offers global consistency and serves as a condition for the\nsubsequent UV diffusion stage, aiding in regularizing the model to generate a\n3D consistent UV texture image. Then, a UV diffusion model with hybrid\nconditions is developed to enhance the texture fidelity in the 2D UV space. Our\nmethod can process meshes of any genus, generating diversified,\ngeometry-compatible, and high-fidelity textures. Code is available at\nhttps://cvmi-lab.github.io/Point-UV-Diffusion\n", "date": "2023-08-21", "categories": []} {"id": "2308.10898", "title": "Few-Shot Physically-Aware Articulated Mesh Generation via Hierarchical\n Deformation", "abstract": " We study the problem of few-shot physically-aware articulated mesh\ngeneration. By observing an articulated object dataset containing only a few\nexamples, we wish to learn a model that can generate diverse meshes with high\nvisual fidelity and physical validity. Previous mesh generative models either\nhave difficulties in depicting a diverse data space from only a few examples or\nfail to ensure physical validity of their samples. Regarding the above\nchallenges, we propose two key innovations, including 1) a hierarchical mesh\ndeformation-based generative model based upon the divide-and-conquer philosophy\nto alleviate the few-shot challenge by borrowing transferrable deformation\npatterns from large scale rigid meshes and 2) a physics-aware deformation\ncorrection scheme to encourage physically plausible generations. We conduct\nextensive experiments on 6 articulated categories to demonstrate the\nsuperiority of our method in generating articulated meshes with better\ndiversity, higher visual fidelity, and better physical validity over previous\nmethods in the few-shot setting. Further, we validate solid contributions of\nour two innovations in the ablation study. Project page with code is available\nat https://meowuu7.github.io/few-arti-obj-gen.\n", "date": "2023-08-21", "categories": []} {"id": "2308.10899", "title": "TADA! Text to Animatable Digital Avatars", "abstract": " We introduce TADA, a simple-yet-effective approach that takes textual\ndescriptions and produces expressive 3D avatars with high-quality geometry and\nlifelike textures, that can be animated and rendered with traditional graphics\npipelines. Existing text-based character generation methods are limited in\nterms of geometry and texture quality, and cannot be realistically animated due\nto inconsistent alignment between the geometry and the texture, particularly in\nthe face region. To overcome these limitations, TADA leverages the synergy of a\n2D diffusion model and an animatable parametric body model. Specifically, we\nderive an optimizable high-resolution body model from SMPL-X with 3D\ndisplacements and a texture map, and use hierarchical rendering with score\ndistillation sampling (SDS) to create high-quality, detailed, holistic 3D\navatars from text. To ensure alignment between the geometry and texture, we\nrender normals and RGB images of the generated character and exploit their\nlatent embeddings in the SDS training process. We further introduce various\nexpression parameters to deform the generated character during training,\nensuring that the semantics of our generated character remain consistent with\nthe original SMPL-X model, resulting in an animatable character. Comprehensive\nevaluations demonstrate that TADA significantly surpasses existing approaches\non both qualitative and quantitative measures. TADA enables creation of\nlarge-scale digital character assets that are ready for animation and\nrendering, while also being easily editable through natural language. The code\nwill be public for research purposes.\n", "date": "2023-08-21", "categories": []} {"id": "2308.09278", "title": "MATLABER: Material-Aware Text-to-3D via LAtent BRDF auto-EncodeR", "abstract": " Based on powerful text-to-image diffusion models, text-to-3D generation has\nmade significant progress in generating compelling geometry and appearance.\nHowever, existing methods still struggle to recover high-fidelity object\nmaterials, either only considering Lambertian reflectance, or failing to\ndisentangle BRDF materials from the environment lights. In this work, we\npropose Material-Aware Text-to-3D via LAtent BRDF auto-EncodeR\n(\\textbf{MATLABER}) that leverages a novel latent BRDF auto-encoder for\nmaterial generation. We train this auto-encoder with large-scale real-world\nBRDF collections and ensure the smoothness of its latent space, which\nimplicitly acts as a natural distribution of materials. During appearance\nmodeling in text-to-3D generation, the latent BRDF embeddings, rather than BRDF\nparameters, are predicted via a material network. Through exhaustive\nexperiments, our approach demonstrates the superiority over existing ones in\ngenerating realistic and coherent object materials. Moreover, high-quality\nmaterials naturally enable multiple downstream tasks such as relighting and\nmaterial editing. Code and model will be publicly available at\n\\url{https://sheldontsui.github.io/projects/Matlaber}.\n", "date": "2023-08-18", "categories": []} {"id": "2308.09705", "title": "Guide3D: Create 3D Avatars from Text and Image Guidance", "abstract": " Recently, text-to-image generation has exhibited remarkable advancements,\nwith the ability to produce visually impressive results. In contrast,\ntext-to-3D generation has not yet reached a comparable level of quality.\nExisting methods primarily rely on text-guided score distillation sampling\n(SDS), and they encounter difficulties in transferring 2D attributes of the\ngenerated images to 3D content. In this work, we aim to develop an effective 3D\ngenerative model capable of synthesizing high-resolution textured meshes by\nleveraging both textual and image information. To this end, we introduce\nGuide3D, a zero-shot text-and-image-guided generative model for 3D avatar\ngeneration based on diffusion models. Our model involves (1) generating\nsparse-view images of a text-consistent character using diffusion models, and\n(2) jointly optimizing multi-resolution differentiable marching tetrahedral\ngrids with pixel-aligned image features. We further propose a similarity-aware\nfeature fusion strategy for efficiently integrating features from different\nviews. Moreover, we introduce two novel training objectives as an alternative\nto calculating SDS, significantly enhancing the optimization process. We\nthoroughly evaluate the performance and components of our framework, which\noutperforms the current state-of-the-art in producing topologically and\nstructurally correct geometry and high-resolution textures. Guide3D enables the\ndirect transfer of 2D-generated images to the 3D space. Our code will be made\npublicly available.\n", "date": "2023-08-18", "categories": []} {"id": "2308.09712", "title": "HumanLiff: Layer-wise 3D Human Generation with Diffusion Model", "abstract": " 3D human generation from 2D images has achieved remarkable progress through\nthe synergistic utilization of neural rendering and generative models. Existing\n3D human generative models mainly generate a clothed 3D human as an\nundetectable 3D model in a single pass, while rarely considering the layer-wise\nnature of a clothed human body, which often consists of the human body and\nvarious clothes such as underwear, outerwear, trousers, shoes, etc. In this\nwork, we propose HumanLiff, the first layer-wise 3D human generative model with\na unified diffusion process. Specifically, HumanLiff firstly generates\nminimal-clothed humans, represented by tri-plane features, in a canonical\nspace, and then progressively generates clothes in a layer-wise manner. In this\nway, the 3D human generation is thus formulated as a sequence of\ndiffusion-based 3D conditional generation. To reconstruct more fine-grained 3D\nhumans with tri-plane representation, we propose a tri-plane shift operation\nthat splits each tri-plane into three sub-planes and shifts these sub-planes to\nenable feature grid subdivision. To further enhance the controllability of 3D\ngeneration with 3D layered conditions, HumanLiff hierarchically fuses tri-plane\nfeatures and 3D layered conditions to facilitate the 3D diffusion model\nlearning. Extensive experiments on two layer-wise 3D human datasets, SynBody\n(synthetic) and TightCap (real-world), validate that HumanLiff significantly\noutperforms state-of-the-art methods in layer-wise 3D human generation. Our\ncode will be available at https://skhu101.github.io/HumanLiff.\n", "date": "2023-08-18", "categories": []} {"id": "2212.01103", "title": "3D-TOGO: Towards Text-Guided Cross-Category 3D Object Generation", "abstract": " Text-guided 3D object generation aims to generate 3D objects described by\nuser-defined captions, which paves a flexible way to visualize what we\nimagined. Although some works have been devoted to solving this challenging\ntask, these works either utilize some explicit 3D representations (e.g., mesh),\nwhich lack texture and require post-processing for rendering photo-realistic\nviews; or require individual time-consuming optimization for every single case.\nHere, we make the first attempt to achieve generic text-guided cross-category\n3D object generation via a new 3D-TOGO model, which integrates a text-to-views\ngeneration module and a views-to-3D generation module. The text-to-views\ngeneration module is designed to generate different views of the target 3D\nobject given an input caption. prior-guidance, caption-guidance and view\ncontrastive learning are proposed for achieving better view-consistency and\ncaption similarity. Meanwhile, a pixelNeRF model is adopted for the views-to-3D\ngeneration module to obtain the implicit 3D neural representation from the\npreviously-generated views. Our 3D-TOGO model generates 3D objects in the form\nof the neural radiance field with good texture and requires no time-cost\noptimization for every single caption. Besides, 3D-TOGO can control the\ncategory, color and shape of generated 3D objects with the input caption.\nExtensive experiments on the largest 3D object dataset (i.e., ABO) are\nconducted to verify that 3D-TOGO can better generate high-quality 3D objects\naccording to the input captions across 98 different categories, in terms of\nPSNR, SSIM, LPIPS and CLIP-score, compared with text-NeRF and Dreamfields.\n", "date": "2022-12-02", "categories": []} {"id": "2307.14051", "title": "3D Semantic Subspace Traverser: Empowering 3D Generative Model with\n Shape Editing Capability", "abstract": " Shape generation is the practice of producing 3D shapes as various\nrepresentations for 3D content creation. Previous studies on 3D shape\ngeneration have focused on shape quality and structure, without or less\nconsidering the importance of semantic information. Consequently, such\ngenerative models often fail to preserve the semantic consistency of shape\nstructure or enable manipulation of the semantic attributes of shapes during\ngeneration. In this paper, we proposed a novel semantic generative model named\n3D Semantic Subspace Traverser that utilizes semantic attributes for\ncategory-specific 3D shape generation and editing. Our method utilizes implicit\nfunctions as the 3D shape representation and combines a novel latent-space GAN\nwith a linear subspace model to discover semantic dimensions in the local\nlatent space of 3D shapes. Each dimension of the subspace corresponds to a\nparticular semantic attribute, and we can edit the attributes of generated\nshapes by traversing the coefficients of those dimensions. Experimental results\ndemonstrate that our method can produce plausible shapes with complex\nstructures and enable the editing of semantic attributes. The code and trained\nmodels are available at\nhttps://github.com/TrepangCat/3D_Semantic_Subspace_Traverser\n", "date": "2023-07-26", "categories": []} {"id": "2301.09637", "title": "InfiniCity: Infinite-Scale City Synthesis", "abstract": " Toward infinite-scale 3D city synthesis, we propose a novel framework,\nInfiniCity, which constructs and renders an unconstrainedly large and\n3D-grounded environment from random noises. InfiniCity decomposes the seemingly\nimpractical task into three feasible modules, taking advantage of both 2D and\n3D data. First, an infinite-pixel image synthesis module generates\narbitrary-scale 2D maps from the bird's-eye view. Next, an octree-based voxel\ncompletion module lifts the generated 2D map to 3D octrees. Finally, a\nvoxel-based neural rendering module texturizes the voxels and renders 2D\nimages. InfiniCity can thus synthesize arbitrary-scale and traversable 3D city\nenvironments, and allow flexible and interactive editing from users. We\nquantitatively and qualitatively demonstrate the efficacy of the proposed\nframework. Project page: https://hubert0527.github.io/infinicity/\n", "date": "2023-01-23", "categories": []} {"id": "2206.14797", "title": "3D-Aware Video Generation", "abstract": " Generative models have emerged as an essential building block for many image\nsynthesis and editing tasks. Recent advances in this field have also enabled\nhigh-quality 3D or video content to be generated that exhibits either\nmulti-view or temporal consistency. With our work, we explore 4D generative\nadversarial networks (GANs) that learn unconditional generation of 3D-aware\nvideos. By combining neural implicit representations with time-aware\ndiscriminator, we develop a GAN framework that synthesizes 3D video supervised\nonly with monocular videos. We show that our method learns a rich embedding of\ndecomposable 3D structures and motions that enables new visual effects of\nspatio-temporal renderings while producing imagery with quality comparable to\nthat of existing 3D or video GANs.\n", "date": "2022-06-29", "categories": []} {"id": "2308.04468", "title": "3D Scene Diffusion Guidance using Scene Graphs", "abstract": " Guided synthesis of high-quality 3D scenes is a challenging task. Diffusion\nmodels have shown promise in generating diverse data, including 3D scenes.\nHowever, current methods rely directly on text embeddings for controlling the\ngeneration, limiting the incorporation of complex spatial relationships between\nobjects. We propose a novel approach for 3D scene diffusion guidance using\nscene graphs. To leverage the relative spatial information the scene graphs\nprovide, we make use of relational graph convolutional blocks within our\ndenoising network. We show that our approach significantly improves the\nalignment between scene description and generated scene.\n", "date": "2023-08-08", "categories": []} {"id": "2303.09036", "title": "Mimic3D: Thriving 3D-Aware GANs via 3D-to-2D Imitation", "abstract": " Generating images with both photorealism and multiview 3D consistency is\ncrucial for 3D-aware GANs, yet existing methods struggle to achieve them\nsimultaneously. Improving the photorealism via CNN-based 2D super-resolution\ncan break the strict 3D consistency, while keeping the 3D consistency by\nlearning high-resolution 3D representations for direct rendering often\ncompromises image quality. In this paper, we propose a novel learning strategy,\nnamely 3D-to-2D imitation, which enables a 3D-aware GAN to generate\nhigh-quality images while maintaining their strict 3D consistency, by letting\nthe images synthesized by the generator's 3D rendering branch to mimic those\ngenerated by its 2D super-resolution branch. We also introduce 3D-aware\nconvolutions into the generator for better 3D representation learning, which\nfurther improves the image generation quality. With the above strategies, our\nmethod reaches FID scores of 5.4 and 4.3 on FFHQ and AFHQ-v2 Cats,\nrespectively, at 512x512 resolution, largely outperforming existing 3D-aware\nGANs using direct 3D rendering and coming very close to the previous\nstate-of-the-art method that leverages 2D super-resolution. Project website:\nhttps://seanchenxy.github.io/Mimic3DWeb.\n", "date": "2023-03-16", "categories": []} {"id": "2308.02874", "title": "Sketch and Text Guided Diffusion Model for Colored Point Cloud\n Generation", "abstract": " Diffusion probabilistic models have achieved remarkable success in text\nguided image generation. However, generating 3D shapes is still challenging due\nto the lack of sufficient data containing 3D models along with their\ndescriptions. Moreover, text based descriptions of 3D shapes are inherently\nambiguous and lack details. In this paper, we propose a sketch and text guided\nprobabilistic diffusion model for colored point cloud generation that\nconditions the denoising process jointly with a hand drawn sketch of the object\nand its textual description. We incrementally diffuse the point coordinates and\ncolor values in a joint diffusion process to reach a Gaussian distribution.\nColored point cloud generation thus amounts to learning the reverse diffusion\nprocess, conditioned by the sketch and text, to iteratively recover the desired\nshape and color. Specifically, to learn effective sketch-text embedding, our\nmodel adaptively aggregates the joint embedding of text prompt and the sketch\nbased on a capsule attention network. Our model uses staged diffusion to\ngenerate the shape and then assign colors to different parts conditioned on the\nappearance prompt while preserving precise shapes from the first stage. This\ngives our model the flexibility to extend to multiple tasks, such as appearance\nre-editing and part segmentation. Experimental results demonstrate that our\nmodel outperforms recent state-of-the-art in point cloud generation.\n", "date": "2023-08-05", "categories": []} {"id": "2308.03610", "title": "AvatarVerse: High-quality & Stable 3D Avatar Creation from Text and Pose", "abstract": " Creating expressive, diverse and high-quality 3D avatars from highly\ncustomized text descriptions and pose guidance is a challenging task, due to\nthe intricacy of modeling and texturing in 3D that ensure details and various\nstyles (realistic, fictional, etc). We present AvatarVerse, a stable pipeline\nfor generating expressive high-quality 3D avatars from nothing but text\ndescriptions and pose guidance. In specific, we introduce a 2D diffusion model\nconditioned on DensePose signal to establish 3D pose control of avatars through\n2D images, which enhances view consistency from partially observed scenarios.\nIt addresses the infamous Janus Problem and significantly stablizes the\ngeneration process. Moreover, we propose a progressive high-resolution 3D\nsynthesis strategy, which obtains substantial improvement over the quality of\nthe created 3D avatars. To this end, the proposed AvatarVerse pipeline achieves\nzero-shot 3D modeling of 3D avatars that are not only more expressive, but also\nin higher quality and fidelity than previous works. Rigorous qualitative\nevaluations and user studies showcase AvatarVerse's superiority in synthesizing\nhigh-fidelity 3D avatars, leading to a new standard in high-quality and stable\n3D avatar creation. Our project page is: https://avatarverse3d.github.io\n", "date": "2023-08-07", "categories": []} {"id": "2307.16183", "title": "HD-Fusion: Detailed Text-to-3D Generation Leveraging Multiple Noise\n Estimation", "abstract": " In this paper, we study Text-to-3D content generation leveraging 2D diffusion\npriors to enhance the quality and detail of the generated 3D models. Recent\nprogress (Magic3D) in text-to-3D has shown that employing high-resolution\n(e.g., 512 x 512) renderings can lead to the production of high-quality 3D\nmodels using latent diffusion priors. To enable rendering at even higher\nresolutions, which has the potential to further augment the quality and detail\nof the models, we propose a novel approach that combines multiple noise\nestimation processes with a pretrained 2D diffusion prior. Distinct from the\nBar-Tal et al.s' study which binds multiple denoised results to generate images\nfrom texts, our approach integrates the computation of scoring distillation\nlosses such as SDS loss and VSD loss which are essential techniques for the 3D\ncontent generation with 2D diffusion priors. We experimentally evaluated the\nproposed approach. The results show that the proposed approach can generate\nhigh-quality details compared to the baselines.\n", "date": "2023-07-30", "categories": []} {"id": "2307.14918", "title": "GET3D--: Learning GET3D from Unconstrained Image Collections", "abstract": " The demand for efficient 3D model generation techniques has grown\nexponentially, as manual creation of 3D models is time-consuming and requires\nspecialized expertise. While generative models have shown potential in creating\n3D textured shapes from 2D images, their applicability in 3D industries is\nlimited due to the lack of a well-defined camera distribution in real-world\nscenarios, resulting in low-quality shapes. To overcome this limitation, we\npropose GET3D--, the first method that directly generates textured 3D shapes\nfrom 2D images with unknown pose and scale. GET3D-- comprises a 3D shape\ngenerator and a learnable camera sampler that captures the 6D external changes\non the camera. In addition, We propose a novel training schedule to stably\noptimize both the shape generator and camera sampler in a unified framework. By\ncontrolling external variations using the learnable camera sampler, our method\ncan generate aligned shapes with clear textures. Extensive experiments\ndemonstrate the efficacy of GET3D--, which precisely fits the 6D camera pose\ndistribution and generates high-quality shapes on both synthetic and realistic\nunconstrained datasets.\n", "date": "2023-07-27", "categories": []} {"id": "2307.13908", "title": "Points-to-3D: Bridging the Gap between Sparse Points and\n Shape-Controllable Text-to-3D Generation", "abstract": " Text-to-3D generation has recently garnered significant attention, fueled by\n2D diffusion models trained on billions of image-text pairs. Existing methods\nprimarily rely on score distillation to leverage the 2D diffusion priors to\nsupervise the generation of 3D models, e.g., NeRF. However, score distillation\nis prone to suffer the view inconsistency problem, and implicit NeRF modeling\ncan also lead to an arbitrary shape, thus leading to less realistic and\nuncontrollable 3D generation. In this work, we propose a flexible framework of\nPoints-to-3D to bridge the gap between sparse yet freely available 3D points\nand realistic shape-controllable 3D generation by distilling the knowledge from\nboth 2D and 3D diffusion models. The core idea of Points-to-3D is to introduce\ncontrollable sparse 3D points to guide the text-to-3D generation. Specifically,\nwe use the sparse point cloud generated from the 3D diffusion model, Point-E,\nas the geometric prior, conditioned on a single reference image. To better\nutilize the sparse 3D points, we propose an efficient point cloud guidance loss\nto adaptively drive the NeRF's geometry to align with the shape of the sparse\n3D points. In addition to controlling the geometry, we propose to optimize the\nNeRF for a more view-consistent appearance. To be specific, we perform score\ndistillation to the publicly available 2D image diffusion model ControlNet,\nconditioned on text as well as depth map of the learned compact geometry.\nQualitative and quantitative comparisons demonstrate that Points-to-3D improves\nview consistency and achieves good shape controllability for text-to-3D\ngeneration. Points-to-3D provides users with a new way to improve and control\ntext-to-3D generation.\n", "date": "2023-07-26", "categories": []} {"id": "2302.01162", "title": "Get3DHuman: Lifting StyleGAN-Human into a 3D Generative Model using\n Pixel-aligned Reconstruction Priors", "abstract": " Fast generation of high-quality 3D digital humans is important to a vast\nnumber of applications ranging from entertainment to professional concerns.\nRecent advances in differentiable rendering have enabled the training of 3D\ngenerative models without requiring 3D ground truths. However, the quality of\nthe generated 3D humans still has much room to improve in terms of both\nfidelity and diversity. In this paper, we present Get3DHuman, a novel 3D human\nframework that can significantly boost the realism and diversity of the\ngenerated outcomes by only using a limited budget of 3D ground-truth data. Our\nkey observation is that the 3D generator can profit from human-related priors\nlearned through 2D human generators and 3D reconstructors. Specifically, we\nbridge the latent space of Get3DHuman with that of StyleGAN-Human via a\nspecially-designed prior network, where the input latent code is mapped to the\nshape and texture feature volumes spanned by the pixel-aligned 3D\nreconstructor. The outcomes of the prior network are then leveraged as the\nsupervisory signals for the main generator network. To ensure effective\ntraining, we further propose three tailored losses applied to the generated\nfeature volumes and the intermediate feature maps. Extensive experiments\ndemonstrate that Get3DHuman greatly outperforms the other state-of-the-art\napproaches and can support a wide range of applications including shape\ninterpolation, shape re-texturing, and single-view reconstruction through\nlatent inversion.\n", "date": "2023-02-02", "categories": []} {"id": "2303.12865", "title": "NeRF-GAN Distillation for Efficient 3D-Aware Generation with\n Convolutions", "abstract": " Pose-conditioned convolutional generative models struggle with high-quality\n3D-consistent image generation from single-view datasets, due to their lack of\nsufficient 3D priors. Recently, the integration of Neural Radiance Fields\n(NeRFs) and generative models, such as Generative Adversarial Networks (GANs),\nhas transformed 3D-aware generation from single-view images. NeRF-GANs exploit\nthe strong inductive bias of neural 3D representations and volumetric rendering\nat the cost of higher computational complexity. This study aims at revisiting\npose-conditioned 2D GANs for efficient 3D-aware generation at inference time by\ndistilling 3D knowledge from pretrained NeRF-GANs. We propose a simple and\neffective method, based on re-using the well-disentangled latent space of a\npre-trained NeRF-GAN in a pose-conditioned convolutional network to directly\ngenerate 3D-consistent images corresponding to the underlying 3D\nrepresentations. Experiments on several datasets demonstrate that the proposed\nmethod obtains results comparable with volumetric rendering in terms of quality\nand 3D consistency while benefiting from the computational advantage of\nconvolutional networks. The code will be available at:\nhttps://github.com/mshahbazi72/NeRF-GAN-Distillation\n", "date": "2023-03-22", "categories": []} {"id": "2306.17843", "title": "Magic123: One Image to High-Quality 3D Object Generation Using Both 2D\n and 3D Diffusion Priors", "abstract": " We present Magic123, a two-stage coarse-to-fine approach for high-quality,\ntextured 3D meshes generation from a single unposed image in the wild using\nboth2D and 3D priors. In the first stage, we optimize a neural radiance field\nto produce a coarse geometry. In the second stage, we adopt a memory-efficient\ndifferentiable mesh representation to yield a high-resolution mesh with a\nvisually appealing texture. In both stages, the 3D content is learned through\nreference view supervision and novel views guided by a combination of 2D and 3D\ndiffusion priors. We introduce a single trade-off parameter between the 2D and\n3D priors to control exploration (more imaginative) and exploitation (more\nprecise) of the generated geometry. Additionally, we employ textual inversion\nand monocular depth regularization to encourage consistent appearances across\nviews and to prevent degenerate solutions, respectively. Magic123 demonstrates\na significant improvement over previous image-to-3D techniques, as validated\nthrough extensive experiments on synthetic benchmarks and diverse real-world\nimages. Our code, models, and generated 3D assets are available at\nhttps://github.com/guochengqian/Magic123.\n", "date": "2023-06-30", "categories": []} {"id": "2307.12049", "title": "Patch-Wise Point Cloud Generation: A Divide-and-Conquer Approach", "abstract": " A generative model for high-fidelity point clouds is of great importance in\nsynthesizing 3d environments for applications such as autonomous driving and\nrobotics. Despite the recent success of deep generative models for 2d images,\nit is non-trivial to generate 3d point clouds without a comprehensive\nunderstanding of both local and global geometric structures. In this paper, we\ndevise a new 3d point cloud generation framework using a divide-and-conquer\napproach, where the whole generation process can be divided into a set of\npatch-wise generation tasks. Specifically, all patch generators are based on\nlearnable priors, which aim to capture the information of geometry primitives.\nWe introduce point- and patch-wise transformers to enable the interactions\nbetween points and patches. Therefore, the proposed divide-and-conquer approach\ncontributes to a new understanding of point cloud generation from the geometry\nconstitution of 3d shapes. Experimental results on a variety of object\ncategories from the most popular point cloud dataset, ShapeNet, show the\neffectiveness of the proposed patch-wise point cloud generation, where it\nclearly outperforms recent state-of-the-art methods for high-fidelity point\ncloud generation.\n", "date": "2023-07-22", "categories": []} {"id": "2307.03847", "title": "Blocks2World: Controlling Realistic Scenes with Editable Primitives", "abstract": " We present Blocks2World, a novel method for 3D scene rendering and editing\nthat leverages a two-step process: convex decomposition of images and\nconditioned synthesis. Our technique begins by extracting 3D parallelepipeds\nfrom various objects in a given scene using convex decomposition, thus\nobtaining a primitive representation of the scene. These primitives are then\nutilized to generate paired data through simple ray-traced depth maps. The next\nstage involves training a conditioned model that learns to generate images from\nthe 2D-rendered convex primitives. This step establishes a direct mapping\nbetween the 3D model and its 2D representation, effectively learning the\ntransition from a 3D model to an image. Once the model is fully trained, it\noffers remarkable control over the synthesis of novel and edited scenes. This\nis achieved by manipulating the primitives at test time, including translating\nor adding them, thereby enabling a highly customizable scene rendering process.\nOur method provides a fresh perspective on 3D scene rendering and editing,\noffering control and flexibility. It opens up new avenues for research and\napplications in the field, including authoring and data augmentation.\n", "date": "2023-07-07", "categories": []} {"id": "2307.04787", "title": "Collaborative Score Distillation for Consistent Visual Synthesis", "abstract": " Generative priors of large-scale text-to-image diffusion models enable a wide\nrange of new generation and editing applications on diverse visual modalities.\nHowever, when adapting these priors to complex visual modalities, often\nrepresented as multiple images (e.g., video), achieving consistency across a\nset of images is challenging. In this paper, we address this challenge with a\nnovel method, Collaborative Score Distillation (CSD). CSD is based on the Stein\nVariational Gradient Descent (SVGD). Specifically, we propose to consider\nmultiple samples as \"particles\" in the SVGD update and combine their score\nfunctions to distill generative priors over a set of images synchronously.\nThus, CSD facilitates seamless integration of information across 2D images,\nleading to a consistent visual synthesis across multiple samples. We show the\neffectiveness of CSD in a variety of tasks, encompassing the visual editing of\npanorama images, videos, and 3D scenes. Our results underline the competency of\nCSD as a versatile method for enhancing inter-sample consistency, thereby\nbroadening the applicability of text-to-image diffusion models.\n", "date": "2023-07-04", "categories": []} {"id": "2307.04859", "title": "Articulated 3D Head Avatar Generation using Text-to-Image Diffusion\n Models", "abstract": " The ability to generate diverse 3D articulated head avatars is vital to a\nplethora of applications, including augmented reality, cinematography, and\neducation. Recent work on text-guided 3D object generation has shown great\npromise in addressing these needs. These methods directly leverage pre-trained\n2D text-to-image diffusion models to generate 3D-multi-view-consistent radiance\nfields of generic objects. However, due to the lack of geometry and texture\npriors, these methods have limited control over the generated 3D objects,\nmaking it difficult to operate inside a specific domain, e.g., human heads. In\nthis work, we develop a new approach to text-guided 3D head avatar generation\nto address this limitation. Our framework directly operates on the geometry and\ntexture of an articulable 3D morphable model (3DMM) of a head, and introduces\nnovel optimization procedures to update the geometry and texture while keeping\nthe 2D and 3D facial features aligned. The result is a 3D head avatar that is\nconsistent with the text description and can be readily articulated using the\ndeformation model of the 3DMM. We show that our diffusion-based articulated\nhead avatars outperform state-of-the-art approaches for this task. The latter\nare typically based on CLIP, which is known to provide limited diversity of\ngeneration and accuracy for 3D object generation.\n", "date": "2023-07-10", "categories": []} {"id": "2307.05445", "title": "AutoDecoding Latent 3D Diffusion Models", "abstract": " We present a novel approach to the generation of static and articulated 3D\nassets that has a 3D autodecoder at its core. The 3D autodecoder framework\nembeds properties learned from the target dataset in the latent space, which\ncan then be decoded into a volumetric representation for rendering\nview-consistent appearance and geometry. We then identify the appropriate\nintermediate volumetric latent space, and introduce robust normalization and\nde-normalization operations to learn a 3D diffusion from 2D images or monocular\nvideos of rigid or articulated objects. Our approach is flexible enough to use\neither existing camera supervision or no camera information at all -- instead\nefficiently learning it during training. Our evaluations demonstrate that our\ngeneration results outperform state-of-the-art alternatives on various\nbenchmark datasets and metrics, including multi-view image datasets of\nsynthetic objects, real in-the-wild videos of moving people, and a large-scale,\nreal video dataset of static objects.\n", "date": "2023-07-07", "categories": []} {"id": "2307.05462", "title": "Efficient 3D Articulated Human Generation with Layered Surface Volumes", "abstract": " Access to high-quality and diverse 3D articulated digital human assets is\ncrucial in various applications, ranging from virtual reality to social\nplatforms. Generative approaches, such as 3D generative adversarial networks\n(GANs), are rapidly replacing laborious manual content creation tools. However,\nexisting 3D GAN frameworks typically rely on scene representations that\nleverage either template meshes, which are fast but offer limited quality, or\nvolumes, which offer high capacity but are slow to render, thereby limiting the\n3D fidelity in GAN settings. In this work, we introduce layered surface volumes\n(LSVs) as a new 3D object representation for articulated digital humans. LSVs\nrepresent a human body using multiple textured mesh layers around a\nconventional template. These layers are rendered using alpha compositing with\nfast differentiable rasterization, and they can be interpreted as a volumetric\nrepresentation that allocates its capacity to a manifold of finite thickness\naround the template. Unlike conventional single-layer templates that struggle\nwith representing fine off-surface details like hair or accessories, our\nsurface volumes naturally capture such details. LSVs can be articulated, and\nthey exhibit exceptional efficiency in GAN settings, where a 2D generator\nlearns to synthesize the RGBA textures for the individual layers. Trained on\nunstructured, single-view 2D image datasets, our LSV-GAN generates high-quality\nand view-consistent 3D articulated digital humans without the need for\nview-inconsistent 2D upsampling networks.\n", "date": "2023-07-11", "categories": []} {"id": "2307.03869", "title": "Sketch-A-Shape: Zero-Shot Sketch-to-3D Shape Generation", "abstract": " Significant progress has recently been made in creative applications of large\npre-trained models for downstream tasks in 3D vision, such as text-to-shape\ngeneration. This motivates our investigation of how these pre-trained models\ncan be used effectively to generate 3D shapes from sketches, which has largely\nremained an open challenge due to the limited sketch-shape paired datasets and\nthe varying level of abstraction in the sketches. We discover that conditioning\na 3D generative model on the features (obtained from a frozen large pre-trained\nvision model) of synthetic renderings during training enables us to effectively\ngenerate 3D shapes from sketches at inference time. This suggests that the\nlarge pre-trained vision model features carry semantic signals that are\nresilient to domain shifts, i.e., allowing us to use only RGB renderings, but\ngeneralizing to sketches at inference time. We conduct a comprehensive set of\nexperiments investigating different design factors and demonstrate the\neffectiveness of our straightforward approach for generation of multiple 3D\nshapes per each input sketch regardless of their level of abstraction without\nrequiring any paired datasets during training.\n", "date": "2023-07-08", "categories": []} {"id": "2307.01831", "title": "DiT-3D: Exploring Plain Diffusion Transformers for 3D Shape Generation", "abstract": " Recent Diffusion Transformers (e.g., DiT) have demonstrated their powerful\neffectiveness in generating high-quality 2D images. However, it is still being\ndetermined whether the Transformer architecture performs equally well in 3D\nshape generation, as previous 3D diffusion methods mostly adopted the U-Net\narchitecture. To bridge this gap, we propose a novel Diffusion Transformer for\n3D shape generation, namely DiT-3D, which can directly operate the denoising\nprocess on voxelized point clouds using plain Transformers. Compared to\nexisting U-Net approaches, our DiT-3D is more scalable in model size and\nproduces much higher quality generations. Specifically, the DiT-3D adopts the\ndesign philosophy of DiT but modifies it by incorporating 3D positional and\npatch embeddings to adaptively aggregate input from voxelized point clouds. To\nreduce the computational cost of self-attention in 3D shape generation, we\nincorporate 3D window attention into Transformer blocks, as the increased 3D\ntoken length resulting from the additional dimension of voxels can lead to high\ncomputation. Finally, linear and devoxelization layers are used to predict the\ndenoised point clouds. In addition, our transformer architecture supports\nefficient fine-tuning from 2D to 3D, where the pre-trained DiT-2D checkpoint on\nImageNet can significantly improve DiT-3D on ShapeNet. Experimental results on\nthe ShapeNet dataset demonstrate that the proposed DiT-3D achieves\nstate-of-the-art performance in high-fidelity and diverse 3D point cloud\ngeneration. In particular, our DiT-3D decreases the 1-Nearest Neighbor Accuracy\nof the state-of-the-art method by 4.59 and increases the Coverage metric by\n3.51 when evaluated on Chamfer Distance.\n", "date": "2023-07-04", "categories": []} {"id": "2307.01957", "title": "Hybrid Neural Diffeomorphic Flow for Shape Representation and Generation\n via Triplane", "abstract": " Deep Implicit Functions (DIFs) have gained popularity in 3D computer vision\ndue to their compactness and continuous representation capabilities. However,\naddressing dense correspondences and semantic relationships across DIF-encoded\nshapes remains a critical challenge, limiting their applications in texture\ntransfer and shape analysis. Moreover, recent endeavors in 3D shape generation\nusing DIFs often neglect correspondence and topology preservation. This paper\npresents HNDF (Hybrid Neural Diffeomorphic Flow), a method that implicitly\nlearns the underlying representation and decomposes intricate dense\ncorrespondences into explicitly axis-aligned triplane features. To avoid\nsuboptimal representations trapped in local minima, we propose hybrid\nsupervision that captures both local and global correspondences. Unlike\nconventional approaches that directly generate new 3D shapes, we further\nexplore the idea of shape generation with deformed template shape via\ndiffeomorphic flows, where the deformation is encoded by the generated triplane\nfeatures. Leveraging a pre-existing 2D diffusion model, we produce high-quality\nand diverse 3D diffeomorphic flows through generated triplanes features,\nensuring topological consistency with the template shape. Extensive experiments\non medical image organ segmentation datasets evaluate the effectiveness of HNDF\nin 3D shape representation and generation.\n", "date": "2023-07-04", "categories": []} {"id": "2304.14005", "title": "ContraNeRF: 3D-Aware Generative Model via Contrastive Learning with\n Unsupervised Implicit Pose Embedding", "abstract": " Although 3D-aware GANs based on neural radiance fields have achieved\ncompetitive performance, their applicability is still limited to objects or\nscenes with the ground-truths or prediction models for clearly defined\ncanonical camera poses. To extend the scope of applicable datasets, we propose\na novel 3D-aware GAN optimization technique through contrastive learning with\nimplicit pose embeddings. To this end, we first revise the discriminator design\nand remove dependency on ground-truth camera poses. Then, to capture complex\nand challenging 3D scene structures more effectively, we make the discriminator\nestimate a high-dimensional implicit pose embedding from a given image and\nperform contrastive learning on the pose embedding. The proposed approach can\nbe employed for the dataset, where the canonical camera pose is ill-defined\nbecause it does not look up or estimate camera poses. Experimental results show\nthat our algorithm outperforms existing methods by large margins on the\ndatasets with multiple object categories and inconsistent canonical camera\nposes.\n", "date": "2023-04-27", "categories": []} {"id": "2306.17115", "title": "Michelangelo: Conditional 3D Shape Generation based on Shape-Image-Text\n Aligned Latent Representation", "abstract": " We present a novel alignment-before-generation approach to tackle the\nchallenging task of generating general 3D shapes based on 2D images or texts.\nDirectly learning a conditional generative model from images or texts to 3D\nshapes is prone to producing inconsistent results with the conditions because\n3D shapes have an additional dimension whose distribution significantly differs\nfrom that of 2D images and texts. To bridge the domain gap among the three\nmodalities and facilitate multi-modal-conditioned 3D shape generation, we\nexplore representing 3D shapes in a shape-image-text-aligned space. Our\nframework comprises two models: a Shape-Image-Text-Aligned Variational\nAuto-Encoder (SITA-VAE) and a conditional Aligned Shape Latent Diffusion Model\n(ASLDM). The former model encodes the 3D shapes into the shape latent space\naligned to the image and text and reconstructs the fine-grained 3D neural\nfields corresponding to given shape embeddings via the transformer-based\ndecoder. The latter model learns a probabilistic mapping function from the\nimage or text space to the latent shape space. Our extensive experiments\ndemonstrate that our proposed approach can generate higher-quality and more\ndiverse 3D shapes that better semantically conform to the visual or textural\nconditional inputs, validating the effectiveness of the\nshape-image-text-aligned space for cross-modality 3D shape generation.\n", "date": "2023-06-29", "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": "2306.15762", "title": "Toward Mesh-Invariant 3D Generative Deep Learning with Geometric\n Measures", "abstract": " 3D generative modeling is accelerating as the technology allowing the capture\nof geometric data is developing. However, the acquired data is often\ninconsistent, resulting in unregistered meshes or point clouds. Many generative\nlearning algorithms require correspondence between each point when comparing\nthe predicted shape and the target shape. We propose an architecture able to\ncope with different parameterizations, even during the training phase. In\nparticular, our loss function is built upon a kernel-based metric over a\nrepresentation of meshes using geometric measures such as currents and\nvarifolds. The latter allows to implement an efficient dissimilarity measure\nwith many desirable properties such as robustness to resampling of the mesh or\npoint cloud. We demonstrate the efficiency and resilience of our model with a\ngenerative learning task of human faces.\n", "date": "2023-06-27", "categories": []} {"id": "2306.16329", "title": "DiffComplete: Diffusion-based Generative 3D Shape Completion", "abstract": " We introduce a new diffusion-based approach for shape completion on 3D range\nscans. Compared with prior deterministic and probabilistic methods, we strike a\nbalance between realism, multi-modality, and high fidelity. We propose\nDiffComplete by casting shape completion as a generative task conditioned on\nthe incomplete shape. Our key designs are two-fold. First, we devise a\nhierarchical feature aggregation mechanism to inject conditional features in a\nspatially-consistent manner. So, we can capture both local details and broader\ncontexts of the conditional inputs to control the shape completion. Second, we\npropose an occupancy-aware fusion strategy in our model to enable the\ncompletion of multiple partial shapes and introduce higher flexibility on the\ninput conditions. DiffComplete sets a new SOTA performance (e.g., 40% decrease\non l_1 error) on two large-scale 3D shape completion benchmarks. Our completed\nshapes not only have a realistic outlook compared with the deterministic\nmethods but also exhibit high similarity to the ground truths compared with the\nprobabilistic alternatives. Further, DiffComplete has strong generalizability\non objects of entirely unseen classes for both synthetic and real data,\neliminating the need for model re-training in various applications.\n", "date": "2023-06-28", "categories": []} {"id": "2306.12570", "title": "Local 3D Editing via 3D Distillation of CLIP Knowledge", "abstract": " 3D content manipulation is an important computer vision task with many\nreal-world applications (e.g., product design, cartoon generation, and 3D\nAvatar editing). Recently proposed 3D GANs can generate diverse photorealistic\n3D-aware contents using Neural Radiance fields (NeRF). However, manipulation of\nNeRF still remains a challenging problem since the visual quality tends to\ndegrade after manipulation and suboptimal control handles such as 2D semantic\nmaps are used for manipulations. While text-guided manipulations have shown\npotential in 3D editing, such approaches often lack locality. To overcome these\nproblems, we propose Local Editing NeRF (LENeRF), which only requires text\ninputs for fine-grained and localized manipulation. Specifically, we present\nthree add-on modules of LENeRF, the Latent Residual Mapper, the Attention Field\nNetwork, and the Deformation Network, which are jointly used for local\nmanipulations of 3D features by estimating a 3D attention field. The 3D\nattention field is learned in an unsupervised way, by distilling the zero-shot\nmask generation capability of CLIP to the 3D space with multi-view guidance. We\nconduct diverse experiments and thorough evaluations both quantitatively and\nqualitatively.\n", "date": "2023-06-21", "categories": []} {"id": "2212.06384", "title": "PV3D: A 3D Generative Model for Portrait Video Generation", "abstract": " Recent advances in generative adversarial networks (GANs) have demonstrated\nthe capabilities of generating stunning photo-realistic portrait images. While\nsome prior works have applied such image GANs to unconditional 2D portrait\nvideo generation and static 3D portrait synthesis, there are few works\nsuccessfully extending GANs for generating 3D-aware portrait videos. In this\nwork, we propose PV3D, the first generative framework that can synthesize\nmulti-view consistent portrait videos. Specifically, our method extends the\nrecent static 3D-aware image GAN to the video domain by generalizing the 3D\nimplicit neural representation to model the spatio-temporal space. To introduce\nmotion dynamics to the generation process, we develop a motion generator by\nstacking multiple motion layers to generate motion features via modulated\nconvolution. To alleviate motion ambiguities caused by camera/human motions, we\npropose a simple yet effective camera condition strategy for PV3D, enabling\nboth temporal and multi-view consistent video generation. Moreover, PV3D\nintroduces two discriminators for regularizing the spatial and temporal domains\nto ensure the plausibility of the generated portrait videos. These elaborated\ndesigns enable PV3D to generate 3D-aware motion-plausible portrait videos with\nhigh-quality appearance and geometry, significantly outperforming prior works.\nAs a result, PV3D is able to support many downstream applications such as\nanimating static portraits and view-consistent video motion editing. Code and\nmodels are released at https://showlab.github.io/pv3d.\n", "date": "2022-12-13", "categories": []} {"id": "2306.12423", "title": "Benchmarking and Analyzing 3D-aware Image Synthesis with a Modularized\n Codebase", "abstract": " Despite the rapid advance of 3D-aware image synthesis, existing studies\nusually adopt a mixture of techniques and tricks, leaving it unclear how each\npart contributes to the final performance in terms of generality. Following the\nmost popular and effective paradigm in this field, which incorporates a neural\nradiance field (NeRF) into the generator of a generative adversarial network\n(GAN), we build a well-structured codebase, dubbed Carver, through modularizing\nthe generation process. Such a design allows researchers to develop and replace\neach module independently, and hence offers an opportunity to fairly compare\nvarious approaches and recognize their contributions from the module\nperspective. The reproduction of a range of cutting-edge algorithms\ndemonstrates the availability of our modularized codebase. We also perform a\nvariety of in-depth analyses, such as the comparison across different types of\npoint feature, the necessity of the tailing upsampler in the generator, the\nreliance on the camera pose prior, etc., which deepen our understanding of\nexisting methods and point out some further directions of the research work. We\nrelease code and models at https://github.com/qiuyu96/Carver to facilitate the\ndevelopment and evaluation of this field.\n", "date": "2023-06-21", "categories": []} {"id": "2306.09551", "title": "Edit-DiffNeRF: Editing 3D Neural Radiance Fields using 2D Diffusion\n Model", "abstract": " Recent research has demonstrated that the combination of pretrained diffusion\nmodels with neural radiance fields (NeRFs) has emerged as a promising approach\nfor text-to-3D generation. Simply coupling NeRF with diffusion models will\nresult in cross-view inconsistency and degradation of stylized view syntheses.\nTo address this challenge, we propose the Edit-DiffNeRF framework, which is\ncomposed of a frozen diffusion model, a proposed delta module to edit the\nlatent semantic space of the diffusion model, and a NeRF. Instead of training\nthe entire diffusion for each scene, our method focuses on editing the latent\nsemantic space in frozen pretrained diffusion models by the delta module. This\nfundamental change to the standard diffusion framework enables us to make\nfine-grained modifications to the rendered views and effectively consolidate\nthese instructions in a 3D scene via NeRF training. As a result, we are able to\nproduce an edited 3D scene that faithfully aligns to input text instructions.\nFurthermore, to ensure semantic consistency across different viewpoints, we\npropose a novel multi-view semantic consistency loss that extracts a latent\nsemantic embedding from the input view as a prior, and aim to reconstruct it in\ndifferent views. Our proposed method has been shown to effectively edit\nreal-world 3D scenes, resulting in 25% improvement in the alignment of the\nperformed 3D edits with text instructions compared to prior work.\n", "date": "2023-06-15", "categories": []} {"id": "2306.09864", "title": "AvatarBooth: High-Quality and Customizable 3D Human Avatar Generation", "abstract": " We introduce AvatarBooth, a novel method for generating high-quality 3D\navatars using text prompts or specific images. Unlike previous approaches that\ncan only synthesize avatars based on simple text descriptions, our method\nenables the creation of personalized avatars from casually captured face or\nbody images, while still supporting text-based model generation and editing.\nOur key contribution is the precise avatar generation control by using dual\nfine-tuned diffusion models separately for the human face and body. This\nenables us to capture intricate details of facial appearance, clothing, and\naccessories, resulting in highly realistic avatar generations. Furthermore, we\nintroduce pose-consistent constraint to the optimization process to enhance the\nmulti-view consistency of synthesized head images from the diffusion model and\nthus eliminate interference from uncontrolled human poses. In addition, we\npresent a multi-resolution rendering strategy that facilitates coarse-to-fine\nsupervision of 3D avatar generation, thereby enhancing the performance of the\nproposed system. The resulting avatar model can be further edited using\nadditional text descriptions and driven by motion sequences. Experiments show\nthat AvatarBooth outperforms previous text-to-3D methods in terms of rendering\nand geometric quality from either text prompts or specific images. Please check\nour project website at https://zeng-yifei.github.io/avatarbooth_page/.\n", "date": "2023-06-16", "categories": []} {"id": "2306.09329", "title": "DreamHuman: Animatable 3D Avatars from Text", "abstract": " We present DreamHuman, a method to generate realistic animatable 3D human\navatar models solely from textual descriptions. Recent text-to-3D methods have\nmade considerable strides in generation, but are still lacking in important\naspects. Control and often spatial resolution remain limited, existing methods\nproduce fixed rather than animated 3D human models, and anthropometric\nconsistency for complex structures like people remains a challenge. DreamHuman\nconnects large text-to-image synthesis models, neural radiance fields, and\nstatistical human body models in a novel modeling and optimization framework.\nThis makes it possible to generate dynamic 3D human avatars with high-quality\ntextures and learned, instance-specific, surface deformations. We demonstrate\nthat our method is capable to generate a wide variety of animatable, realistic\n3D human models from text. Our 3D models have diverse appearance, clothing,\nskin tones and body shapes, and significantly outperform both generic\ntext-to-3D approaches and previous text-based 3D avatar generators in visual\nfidelity. For more results and animations please check our website at\nhttps://dream-human.github.io.\n", "date": "2023-06-15", "categories": []} {"id": "2306.07349", "title": "ATT3D: Amortized Text-to-3D Object Synthesis", "abstract": " Text-to-3D modelling has seen exciting progress by combining generative\ntext-to-image models with image-to-3D methods like Neural Radiance Fields.\nDreamFusion recently achieved high-quality results but requires a lengthy,\nper-prompt optimization to create 3D objects. To address this, we amortize\noptimization over text prompts by training on many prompts simultaneously with\na unified model, instead of separately. With this, we share computation across\na prompt set, training in less time than per-prompt optimization. Our framework\n- Amortized text-to-3D (ATT3D) - enables knowledge-sharing between prompts to\ngeneralize to unseen setups and smooth interpolations between text for novel\nassets and simple animations.\n", "date": "2023-06-06", "categories": []} {"id": "2306.06212", "title": "Aladdin: Zero-Shot Hallucination of Stylized 3D Assets from Abstract\n Scene Descriptions", "abstract": " What constitutes the \"vibe\" of a particular scene? What should one find in \"a\nbusy, dirty city street\", \"an idyllic countryside\", or \"a crime scene in an\nabandoned living room\"? The translation from abstract scene descriptions to\nstylized scene elements cannot be done with any generality by extant systems\ntrained on rigid and limited indoor datasets. In this paper, we propose to\nleverage the knowledge captured by foundation models to accomplish this\ntranslation. We present a system that can serve as a tool to generate stylized\nassets for 3D scenes described by a short phrase, without the need to enumerate\nthe objects to be found within the scene or give instructions on their\nappearance. Additionally, it is robust to open-world concepts in a way that\ntraditional methods trained on limited data are not, affording more creative\nfreedom to the 3D artist. Our system demonstrates this using a foundation model\n\"team\" composed of a large language model, a vision-language model and several\nimage diffusion models, which communicate using an interpretable and\nuser-editable intermediate representation, thus allowing for more versatile and\ncontrollable stylized asset generation for 3D artists. We introduce novel\nmetrics for this task, and show through human evaluations that in 91% of the\ncases, our system outputs are judged more faithful to the semantics of the\ninput scene description than the baseline, thus highlighting the potential of\nthis approach to radically accelerate the 3D content creation process for 3D\nartists.\n", "date": "2023-06-09", "categories": []} {"id": "2306.07154", "title": "InstructP2P: Learning to Edit 3D Point Clouds with Text Instructions", "abstract": " Enhancing AI systems to perform tasks following human instructions can\nsignificantly boost productivity. In this paper, we present InstructP2P, an\nend-to-end framework for 3D shape editing on point clouds, guided by high-level\ntextual instructions. InstructP2P extends the capabilities of existing methods\nby synergizing the strengths of a text-conditioned point cloud diffusion model,\nPoint-E, and powerful language models, enabling color and geometry editing\nusing language instructions. To train InstructP2P, we introduce a new shape\nediting dataset, constructed by integrating a shape segmentation dataset,\noff-the-shelf shape programs, and diverse edit instructions generated by a\nlarge language model, ChatGPT. Our proposed method allows for editing both\ncolor and geometry of specific regions in a single forward pass, while leaving\nother regions unaffected. In our experiments, InstructP2P shows generalization\ncapabilities, adapting to novel shape categories and instructions, despite\nbeing trained on a limited amount of data.\n", "date": "2023-06-12", "categories": []} {"id": "2306.02083", "title": "Efficient Text-Guided 3D-Aware Portrait Generation with Score\n Distillation Sampling on Distribution", "abstract": " Text-to-3D is an emerging task that allows users to create 3D content with\ninfinite possibilities. Existing works tackle the problem by optimizing a 3D\nrepresentation with guidance from pre-trained diffusion models. An apparent\ndrawback is that they need to optimize from scratch for each prompt, which is\ncomputationally expensive and often yields poor visual fidelity. In this paper,\nwe propose DreamPortrait, which aims to generate text-guided 3D-aware portraits\nin a single-forward pass for efficiency. To achieve this, we extend Score\nDistillation Sampling from datapoint to distribution formulation, which injects\nsemantic prior into a 3D distribution. However, the direct extension will lead\nto the mode collapse problem since the objective only pursues semantic\nalignment. Hence, we propose to optimize a distribution with hierarchical\ncondition adapters and GAN loss regularization. For better 3D modeling, we\nfurther design a 3D-aware gated cross-attention mechanism to explicitly let the\nmodel perceive the correspondence between the text and the 3D-aware space.\nThese elaborated designs enable our model to generate portraits with robust\nmulti-view semantic consistency, eliminating the need for optimization-based\nmethods. Extensive experiments demonstrate our model's highly competitive\nperformance and significant speed boost against existing methods.\n", "date": "2023-06-03", "categories": []} {"id": "2306.02741", "title": "ZIGNeRF: Zero-shot 3D Scene Representation with Invertible Generative\n Neural Radiance Fields", "abstract": " Generative Neural Radiance Fields (NeRFs) have demonstrated remarkable\nproficiency in synthesizing multi-view images by learning the distribution of a\nset of unposed images. Despite the aptitude of existing generative NeRFs in\ngenerating 3D-consistent high-quality random samples within data distribution,\nthe creation of a 3D representation of a singular input image remains a\nformidable challenge. In this manuscript, we introduce ZIGNeRF, an innovative\nmodel that executes zero-shot Generative Adversarial Network (GAN) inversion\nfor the generation of multi-view images from a single out-of-domain image. The\nmodel is underpinned by a novel inverter that maps out-of-domain images into\nthe latent code of the generator manifold. Notably, ZIGNeRF is capable of\ndisentangling the object from the background and executing 3D operations such\nas 360-degree rotation or depth and horizontal translation. The efficacy of our\nmodel is validated using multiple real-image datasets: Cats, AFHQ, CelebA,\nCelebA-HQ, and CompCars.\n", "date": "2023-06-05", "categories": []} {"id": "2306.02903", "title": "Instruct-Video2Avatar: Video-to-Avatar Generation with Instructions", "abstract": " We propose a method for synthesizing edited photo-realistic digital avatars\nwith text instructions. Given a short monocular RGB video and text\ninstructions, our method uses an image-conditioned diffusion model to edit one\nhead image and uses the video stylization method to accomplish the editing of\nother head images. Through iterative training and update (three times or more),\nour method synthesizes edited photo-realistic animatable 3D neural head avatars\nwith a deformable neural radiance field head synthesis method. In quantitative\nand qualitative studies on various subjects, our method outperforms\nstate-of-the-art methods.\n", "date": "2023-06-05", "categories": []} {"id": "2303.12789", "title": "Instruct-NeRF2NeRF: Editing 3D Scenes with Instructions", "abstract": " We propose a method for editing NeRF scenes with text-instructions. Given a\nNeRF of a scene and the collection of images used to reconstruct it, our method\nuses an image-conditioned diffusion model (InstructPix2Pix) to iteratively edit\nthe input images while optimizing the underlying scene, resulting in an\noptimized 3D scene that respects the edit instruction. We demonstrate that our\nproposed method is able to edit large-scale, real-world scenes, and is able to\naccomplish more realistic, targeted edits than prior work.\n", "date": "2023-03-22", "categories": []} {"id": "2305.19012", "title": "StyleAvatar3D: Leveraging Image-Text Diffusion Models for High-Fidelity\n 3D Avatar Generation", "abstract": " The recent advancements in image-text diffusion models have stimulated\nresearch interest in large-scale 3D generative models. Nevertheless, the\nlimited availability of diverse 3D resources presents significant challenges to\nlearning. In this paper, we present a novel method for generating high-quality,\nstylized 3D avatars that utilizes pre-trained image-text diffusion models for\ndata generation and a Generative Adversarial Network (GAN)-based 3D generation\nnetwork for training. Our method leverages the comprehensive priors of\nappearance and geometry offered by image-text diffusion models to generate\nmulti-view images of avatars in various styles. During data generation, we\nemploy poses extracted from existing 3D models to guide the generation of\nmulti-view images. To address the misalignment between poses and images in\ndata, we investigate view-specific prompts and develop a coarse-to-fine\ndiscriminator for GAN training. We also delve into attribute-related prompts to\nincrease the diversity of the generated avatars. Additionally, we develop a\nlatent diffusion model within the style space of StyleGAN to enable the\ngeneration of avatars based on image inputs. Our approach demonstrates superior\nperformance over current state-of-the-art methods in terms of visual quality\nand diversity of the produced avatars.\n", "date": "2023-05-30", "categories": []} {"id": "2305.13460", "title": "'Tax-free' 3DMM Conditional Face Generation", "abstract": " 3DMM conditioned face generation has gained traction due to its well-defined\ncontrollability; however, the trade-off is lower sample quality: Previous works\nsuch as DiscoFaceGAN and 3D-FM GAN show a significant FID gap compared to the\nunconditional StyleGAN, suggesting that there is a quality tax to pay for\ncontrollability. In this paper, we challenge the assumption that quality and\ncontrollability cannot coexist. To pinpoint the previous issues, we\nmathematically formalize the problem of 3DMM conditioned face generation. Then,\nwe devise simple solutions to the problem under our proposed framework. This\nresults in a new model that effectively removes the quality tax between 3DMM\nconditioned face GANs and the unconditional StyleGAN.\n", "date": "2023-05-22", "categories": []} {"id": "2305.14345", "title": "NCHO: Unsupervised Learning for Neural 3D Composition of Humans and\n Objects", "abstract": " Deep generative models have been recently extended to synthesizing 3D digital\nhumans. However, previous approaches treat clothed humans as a single chunk of\ngeometry without considering the compositionality of clothing and accessories.\nAs a result, individual items cannot be naturally composed into novel\nidentities, leading to limited expressiveness and controllability of generative\n3D avatars. While several methods attempt to address this by leveraging\nsynthetic data, the interaction between humans and objects is not authentic due\nto the domain gap, and manual asset creation is difficult to scale for a wide\nvariety of objects. In this work, we present a novel framework for learning a\ncompositional generative model of humans and objects (backpacks, coats,\nscarves, and more) from real-world 3D scans. Our compositional model is\ninteraction-aware, meaning the spatial relationship between humans and objects,\nand the mutual shape change by physical contact is fully incorporated. The key\nchallenge is that, since humans and objects are in contact, their 3D scans are\nmerged into a single piece. To decompose them without manual annotations, we\npropose to leverage two sets of 3D scans of a single person with and without\nobjects. Our approach learns to decompose objects and naturally compose them\nback into a generative human model in an unsupervised manner. Despite our\nsimple setup requiring only the capture of a single subject with objects, our\nexperiments demonstrate the strong generalization of our model by enabling the\nnatural composition of objects to diverse identities in various poses and the\ncomposition of multiple objects, which is unseen in training data.\nhttps://taeksuu.github.io/ncho/\n", "date": "2023-05-23", "categories": []} {"id": "2305.15753", "title": "T2TD: Text-3D Generation Model based on Prior Knowledge Guidance", "abstract": " In recent years, 3D models have been utilized in many applications, such as\nauto-driver, 3D reconstruction, VR, and AR. However, the scarcity of 3D model\ndata does not meet its practical demands. Thus, generating high-quality 3D\nmodels efficiently from textual descriptions is a promising but challenging way\nto solve this problem. In this paper, inspired by the ability of human beings\nto complement visual information details from ambiguous descriptions based on\ntheir own experience, we propose a novel text-3D generation model (T2TD), which\nintroduces the related shapes or textual information as the prior knowledge to\nimprove the performance of the 3D generation model. In this process, we first\nintroduce the text-3D knowledge graph to save the relationship between 3D\nmodels and textual semantic information, which can provide the related shapes\nto guide the target 3D model generation. Second, we integrate an effective\ncausal inference model to select useful feature information from these related\nshapes, which removes the unrelated shape information and only maintains\nfeature information that is strongly relevant to the textual description.\nMeanwhile, to effectively integrate multi-modal prior knowledge into textual\ninformation, we adopt a novel multi-layer transformer structure to\nprogressively fuse related shape and textual information, which can effectively\ncompensate for the lack of structural information in the text and enhance the\nfinal performance of the 3D generation model. The final experimental results\ndemonstrate that our approach significantly improves 3D model generation\nquality and outperforms the SOTA methods on the text2shape datasets.\n", "date": "2023-05-25", "categories": []} {"id": "2305.15808", "title": "Towards Language-guided Interactive 3D Generation: LLMs as Layout\n Interpreter with Generative Feedback", "abstract": " Generating and editing a 3D scene guided by natural language poses a\nchallenge, primarily due to the complexity of specifying the positional\nrelations and volumetric changes within the 3D space. Recent advancements in\nLarge Language Models (LLMs) have demonstrated impressive reasoning,\nconversational, and zero-shot generation abilities across various domains.\nSurprisingly, these models also show great potential in realizing and\ninterpreting the 3D space. In light of this, we propose a novel language-guided\ninteractive 3D generation system, dubbed LI3D, that integrates LLMs as a 3D\nlayout interpreter into the off-the-shelf layout-to-3D generative models,\nallowing users to flexibly and interactively generate visual content.\nSpecifically, we design a versatile layout structure base on the bounding boxes\nand semantics to prompt the LLMs to model the spatial generation and reasoning\nfrom language. Our system also incorporates LLaVA, a large language and vision\nassistant, to provide generative feedback from the visual aspect for improving\nthe visual quality of generated content. We validate the effectiveness of LI3D,\nprimarily in 3D generation and editing through multi-round interactions, which\ncan be flexibly extended to 2D generation and editing. Various experiments\ndemonstrate the potential benefits of incorporating LLMs in generative AI for\napplications, e.g., metaverse. Moreover, we benchmark the layout reasoning\nperformance of LLMs with neural visual artist tasks, revealing their emergent\nability in the spatial layout domain.\n", "date": "2023-05-25", "categories": []} {"id": "2305.16315", "title": "NAP: Neural 3D Articulation Prior", "abstract": " We propose Neural 3D Articulation Prior (NAP), the first 3D deep generative\nmodel to synthesize 3D articulated object models. Despite the extensive\nresearch on generating 3D objects, compositions, or scenes, there remains a\nlack of focus on capturing the distribution of articulated objects, a common\nobject category for human and robot interaction. To generate articulated\nobjects, we first design a novel articulation tree/graph parameterization and\nthen apply a diffusion-denoising probabilistic model over this representation\nwhere articulated objects can be generated via denoising from random complete\ngraphs. In order to capture both the geometry and the motion structure whose\ndistribution will affect each other, we design a graph-attention denoising\nnetwork for learning the reverse diffusion process. We propose a novel distance\nthat adapts widely used 3D generation metrics to our novel task to evaluate\ngeneration quality, and experiments demonstrate our high performance in\narticulated object generation. We also demonstrate several conditioned\ngeneration applications, including Part2Motion, PartNet-Imagination,\nMotion2Part, and GAPart2Object.\n", "date": "2023-05-25", "categories": []} {"id": "2211.01427", "title": "CLIP-Sculptor: Zero-Shot Generation of High-Fidelity and Diverse Shapes\n from Natural Language", "abstract": " Recent works have demonstrated that natural language can be used to generate\nand edit 3D shapes. However, these methods generate shapes with limited\nfidelity and diversity. We introduce CLIP-Sculptor, a method to address these\nconstraints by producing high-fidelity and diverse 3D shapes without the need\nfor (text, shape) pairs during training. CLIP-Sculptor achieves this in a\nmulti-resolution approach that first generates in a low-dimensional latent\nspace and then upscales to a higher resolution for improved shape fidelity. For\nimproved shape diversity, we use a discrete latent space which is modeled using\na transformer conditioned on CLIP's image-text embedding space. We also present\na novel variant of classifier-free guidance, which improves the\naccuracy-diversity trade-off. Finally, we perform extensive experiments\ndemonstrating that CLIP-Sculptor outperforms state-of-the-art baselines. The\ncode is available at https://ivl.cs.brown.edu/#/projects/clip-sculptor.\n", "date": "2022-11-02", "categories": []} {"id": "2211.06009", "title": "What's the Situation with Intelligent Mesh Generation: A Survey and\n Perspectives", "abstract": " Intelligent Mesh Generation (IMG) represents a novel and promising field of\nresearch, utilizing machine learning techniques to generate meshes. Despite its\nrelative infancy, IMG has significantly broadened the adaptability and\npracticality of mesh generation techniques, delivering numerous breakthroughs\nand unveiling potential future pathways. However, a noticeable void exists in\nthe contemporary literature concerning comprehensive surveys of IMG methods.\nThis paper endeavors to fill this gap by providing a systematic and thorough\nsurvey of the current IMG landscape. With a focus on 113 preliminary IMG\nmethods, we undertake a meticulous analysis from various angles, encompassing\ncore algorithm techniques and their application scope, agent learning\nobjectives, data types, targeted challenges, as well as advantages and\nlimitations. We have curated and categorized the literature, proposing three\nunique taxonomies based on key techniques, output mesh unit elements, and\nrelevant input data types. This paper also underscores several promising future\nresearch directions and challenges in IMG. To augment reader accessibility, a\ndedicated IMG project page is available at\n\\url{https://github.com/xzb030/IMG_Survey}.\n", "date": "2022-11-11", "categories": []} {"id": "2303.16509", "title": "HoloDiffusion: Training a 3D Diffusion Model using 2D Images", "abstract": " Diffusion models have emerged as the best approach for generative modeling of\n2D images. Part of their success is due to the possibility of training them on\nmillions if not billions of images with a stable learning objective. However,\nextending these models to 3D remains difficult for two reasons. First, finding\na large quantity of 3D training data is much more complex than for 2D images.\nSecond, while it is conceptually trivial to extend the models to operate on 3D\nrather than 2D grids, the associated cubic growth in memory and compute\ncomplexity makes this infeasible. We address the first challenge by introducing\na new diffusion setup that can be trained, end-to-end, with only posed 2D\nimages for supervision; and the second challenge by proposing an image\nformation model that decouples model memory from spatial memory. We evaluate\nour method on real-world data, using the CO3D dataset which has not been used\nto train 3D generative models before. We show that our diffusion models are\nscalable, train robustly, and are competitive in terms of sample quality and\nfidelity to existing approaches for 3D generative modeling.\n", "date": "2023-03-29", "categories": []} {"id": "2305.10853", "title": "LDM3D: Latent Diffusion Model for 3D", "abstract": " This research paper proposes a Latent Diffusion Model for 3D (LDM3D) that\ngenerates both image and depth map data from a given text prompt, allowing\nusers to generate RGBD images from text prompts. The LDM3D model is fine-tuned\non a dataset of tuples containing an RGB image, depth map and caption, and\nvalidated through extensive experiments. We also develop an application called\nDepthFusion, which uses the generated RGB images and depth maps to create\nimmersive and interactive 360-degree-view experiences using TouchDesigner. This\ntechnology has the potential to transform a wide range of industries, from\nentertainment and gaming to architecture and design. Overall, this paper\npresents a significant contribution to the field of generative AI and computer\nvision, and showcases the potential of LDM3D and DepthFusion to revolutionize\ncontent creation and digital experiences. A short video summarizing the\napproach can be found at https://t.ly/tdi2.\n", "date": "2023-05-18", "categories": []} {"id": "2305.11337", "title": "RoomDreamer: Text-Driven 3D Indoor Scene Synthesis with Coherent\n Geometry and Texture", "abstract": " The techniques for 3D indoor scene capturing are widely used, but the meshes\nproduced leave much to be desired. In this paper, we propose \"RoomDreamer\",\nwhich leverages powerful natural language to synthesize a new room with a\ndifferent style. Unlike existing image synthesis methods, our work addresses\nthe challenge of synthesizing both geometry and texture aligned to the input\nscene structure and prompt simultaneously. The key insight is that a scene\nshould be treated as a whole, taking into account both scene texture and\ngeometry. The proposed framework consists of two significant components:\nGeometry Guided Diffusion and Mesh Optimization. Geometry Guided Diffusion for\n3D Scene guarantees the consistency of the scene style by applying the 2D prior\nto the entire scene simultaneously. Mesh Optimization improves the geometry and\ntexture jointly and eliminates the artifacts in the scanned scene. To validate\nthe proposed method, real indoor scenes scanned with smartphones are used for\nextensive experiments, through which the effectiveness of our method is\ndemonstrated.\n", "date": "2023-05-18", "categories": []} {"id": "2305.11664", "title": "Few-shot 3D Shape Generation", "abstract": " Realistic and diverse 3D shape generation is helpful for a wide variety of\napplications such as virtual reality, gaming, and animation. Modern generative\nmodels, such as GANs and diffusion models, learn from large-scale datasets and\ngenerate new samples following similar data distributions. However, when\ntraining data is limited, deep neural generative networks overfit and tend to\nreplicate training samples. Prior works focus on few-shot image generation to\nproduce high-quality and diverse results using a few target images.\nUnfortunately, abundant 3D shape data is typically hard to obtain as well. In\nthis work, we make the first attempt to realize few-shot 3D shape generation by\nadapting generative models pre-trained on large source domains to target\ndomains using limited data. To relieve overfitting and keep considerable\ndiversity, we propose to maintain the probability distributions of the pairwise\nrelative distances between adapted samples at feature-level and shape-level\nduring domain adaptation. Our approach only needs the silhouettes of few-shot\ntarget samples as training data to learn target geometry distributions and\nachieve generated shapes with diverse topology and textures. Moreover, we\nintroduce several metrics to evaluate the quality and diversity of few-shot 3D\nshape generation. The effectiveness of our approach is demonstrated\nqualitatively and quantitatively under a series of few-shot 3D shape adaptation\nsetups.\n", "date": "2023-05-19", "categories": []} {"id": "2305.05901", "title": "Text-guided High-definition Consistency Texture Model", "abstract": " With the advent of depth-to-image diffusion models, text-guided generation,\nediting, and transfer of realistic textures are no longer difficult. However,\ndue to the limitations of pre-trained diffusion models, they can only create\nlow-resolution, inconsistent textures. To address this issue, we present the\nHigh-definition Consistency Texture Model (HCTM), a novel method that can\ngenerate high-definition and consistent textures for 3D meshes according to the\ntext prompts. We achieve this by leveraging a pre-trained depth-to-image\ndiffusion model to generate single viewpoint results based on the text prompt\nand a depth map. We fine-tune the diffusion model with Parameter-Efficient\nFine-Tuning to quickly learn the style of the generated result, and apply the\nmulti-diffusion strategy to produce high-resolution and consistent results from\ndifferent viewpoints. Furthermore, we propose a strategy that prevents the\nappearance of noise on the textures caused by backpropagation. Our proposed\napproach has demonstrated promising results in generating high-definition and\nconsistent textures for 3D meshes, as demonstrated through a series of\nexperiments.\n", "date": "2023-05-10", "categories": []} {"id": "2305.04461", "title": "Locally Attentional SDF Diffusion for Controllable 3D Shape Generation", "abstract": " Although the recent rapid evolution of 3D generative neural networks greatly\nimproves 3D shape generation, it is still not convenient for ordinary users to\ncreate 3D shapes and control the local geometry of generated shapes. To address\nthese challenges, we propose a diffusion-based 3D generation framework --\nlocally attentional SDF diffusion, to model plausible 3D shapes, via 2D sketch\nimage input. Our method is built on a two-stage diffusion model. The first\nstage, named occupancy-diffusion, aims to generate a low-resolution occupancy\nfield to approximate the shape shell. The second stage, named SDF-diffusion,\nsynthesizes a high-resolution signed distance field within the occupied voxels\ndetermined by the first stage to extract fine geometry. Our model is empowered\nby a novel view-aware local attention mechanism for image-conditioned shape\ngeneration, which takes advantage of 2D image patch features to guide 3D voxel\nfeature learning, greatly improving local controllability and model\ngeneralizability. Through extensive experiments in sketch-conditioned and\ncategory-conditioned 3D shape generation tasks, we validate and demonstrate the\nability of our method to provide plausible and diverse 3D shapes, as well as\nits superior controllability and generalizability over existing work. Our code\nand trained models are available at\nhttps://zhengxinyang.github.io/projects/LAS-Diffusion.html\n", "date": "2023-05-08", "categories": []} {"id": "2212.03293", "title": "Diffusion-SDF: Text-to-Shape via Voxelized Diffusion", "abstract": " With the rising industrial attention to 3D virtual modeling technology,\ngenerating novel 3D content based on specified conditions (e.g. text) has\nbecome a hot issue. In this paper, we propose a new generative 3D modeling\nframework called Diffusion-SDF for the challenging task of text-to-shape\nsynthesis. Previous approaches lack flexibility in both 3D data representation\nand shape generation, thereby failing to generate highly diversified 3D shapes\nconforming to the given text descriptions. To address this, we propose a SDF\nautoencoder together with the Voxelized Diffusion model to learn and generate\nrepresentations for voxelized signed distance fields (SDFs) of 3D shapes.\nSpecifically, we design a novel UinU-Net architecture that implants a\nlocal-focused inner network inside the standard U-Net architecture, which\nenables better reconstruction of patch-independent SDF representations. We\nextend our approach to further text-to-shape tasks including text-conditioned\nshape completion and manipulation. Experimental results show that Diffusion-SDF\ngenerates both higher quality and more diversified 3D shapes that conform well\nto given text descriptions when compared to previous approaches. Code is\navailable at: https://github.com/ttlmh/Diffusion-SDF\n", "date": "2022-12-06", "categories": []} {"id": "2305.03302", "title": "High-Fidelity 3D Face Generation from Natural Language Descriptions", "abstract": " Synthesizing high-quality 3D face models from natural language descriptions\nis very valuable for many applications, including avatar creation, virtual\nreality, and telepresence. However, little research ever tapped into this task.\nWe argue the major obstacle lies in 1) the lack of high-quality 3D face data\nwith descriptive text annotation, and 2) the complex mapping relationship\nbetween descriptive language space and shape/appearance space. To solve these\nproblems, we build Describe3D dataset, the first large-scale dataset with\nfine-grained text descriptions for text-to-3D face generation task. Then we\npropose a two-stage framework to first generate a 3D face that matches the\nconcrete descriptions, then optimize the parameters in the 3D shape and texture\nspace with abstract description to refine the 3D face model. Extensive\nexperimental results show that our method can produce a faithful 3D face that\nconforms to the input descriptions with higher accuracy and quality than\nprevious methods. The code and Describe3D dataset are released at\nhttps://github.com/zhuhao-nju/describe3d .\n", "date": "2023-05-05", "categories": []} {"id": "2305.02463", "title": "Shap-E: Generating Conditional 3D Implicit Functions", "abstract": " We present Shap-E, a conditional generative model for 3D assets. Unlike\nrecent work on 3D generative models which produce a single output\nrepresentation, Shap-E directly generates the parameters of implicit functions\nthat can be rendered as both textured meshes and neural radiance fields. We\ntrain Shap-E in two stages: first, we train an encoder that deterministically\nmaps 3D assets into the parameters of an implicit function; second, we train a\nconditional diffusion model on outputs of the encoder. When trained on a large\ndataset of paired 3D and text data, our resulting models are capable of\ngenerating complex and diverse 3D assets in a matter of seconds. When compared\nto Point-E, an explicit generative model over point clouds, Shap-E converges\nfaster and reaches comparable or better sample quality despite modeling a\nhigher-dimensional, multi-representation output space. We release model\nweights, inference code, and samples at https://github.com/openai/shap-e.\n", "date": "2023-05-03", "categories": []} {"id": "2305.02312", "title": "AG3D: Learning to Generate 3D Avatars from 2D Image Collections", "abstract": " While progress in 2D generative models of human appearance has been rapid,\nmany applications require 3D avatars that can be animated and rendered.\nUnfortunately, most existing methods for learning generative models of 3D\nhumans with diverse shape and appearance require 3D training data, which is\nlimited and expensive to acquire. The key to progress is hence to learn\ngenerative models of 3D avatars from abundant unstructured 2D image\ncollections. However, learning realistic and complete 3D appearance and\ngeometry in this under-constrained setting remains challenging, especially in\nthe presence of loose clothing such as dresses. In this paper, we propose a new\nadversarial generative model of realistic 3D people from 2D images. Our method\ncaptures shape and deformation of the body and loose clothing by adopting a\nholistic 3D generator and integrating an efficient and flexible articulation\nmodule. To improve realism, we train our model using multiple discriminators\nwhile also integrating geometric cues in the form of predicted 2D normal maps.\nWe experimentally find that our method outperforms previous 3D- and\narticulation-aware methods in terms of geometry and appearance. We validate the\neffectiveness of our model and the importance of each component via systematic\nablation studies.\n", "date": "2023-05-03", "categories": []} {"id": "2301.11445", "title": "3DShape2VecSet: A 3D Shape Representation for Neural Fields and\n Generative Diffusion Models", "abstract": " We introduce 3DShape2VecSet, a novel shape representation for neural fields\ndesigned for generative diffusion models. Our shape representation can encode\n3D shapes given as surface models or point clouds, and represents them as\nneural fields. The concept of neural fields has previously been combined with a\nglobal latent vector, a regular grid of latent vectors, or an irregular grid of\nlatent vectors. Our new representation encodes neural fields on top of a set of\nvectors. We draw from multiple concepts, such as the radial basis function\nrepresentation and the cross attention and self-attention function, to design a\nlearnable representation that is especially suitable for processing with\ntransformers. Our results show improved performance in 3D shape encoding and 3D\nshape generative modeling tasks. We demonstrate a wide variety of generative\napplications: unconditioned generation, category-conditioned generation,\ntext-conditioned generation, point-cloud completion, and image-conditioned\ngeneration.\n", "date": "2023-01-26", "categories": []} {"id": "2304.14473", "title": "Learning a Diffusion Prior for NeRFs", "abstract": " Neural Radiance Fields (NeRFs) have emerged as a powerful neural 3D\nrepresentation for objects and scenes derived from 2D data. Generating NeRFs,\nhowever, remains difficult in many scenarios. For instance, training a NeRF\nwith only a small number of views as supervision remains challenging since it\nis an under-constrained problem. In such settings, it calls for some inductive\nprior to filter out bad local minima. One way to introduce such inductive\npriors is to learn a generative model for NeRFs modeling a certain class of\nscenes. In this paper, we propose to use a diffusion model to generate NeRFs\nencoded on a regularized grid. We show that our model can sample realistic\nNeRFs, while at the same time allowing conditional generations, given a certain\nobservation as guidance.\n", "date": "2023-04-27", "categories": []} {"id": "2304.04968", "title": "Re-imagine the Negative Prompt Algorithm: Transform 2D Diffusion into\n 3D, alleviate Janus problem and Beyond", "abstract": " Although text-to-image diffusion models have made significant strides in\ngenerating images from text, they are sometimes more inclined to generate\nimages like the data on which the model was trained rather than the provided\ntext. This limitation has hindered their usage in both 2D and 3D applications.\nTo address this problem, we explored the use of negative prompts but found that\nthe current implementation fails to produce desired results, particularly when\nthere is an overlap between the main and negative prompts. To overcome this\nissue, we propose Perp-Neg, a new algorithm that leverages the geometrical\nproperties of the score space to address the shortcomings of the current\nnegative prompts algorithm. Perp-Neg does not require any training or\nfine-tuning of the model. Moreover, we experimentally demonstrate that Perp-Neg\nprovides greater flexibility in generating images by enabling users to edit out\nunwanted concepts from the initially generated images in 2D cases. Furthermore,\nto extend the application of Perp-Neg to 3D, we conducted a thorough\nexploration of how Perp-Neg can be used in 2D to condition the diffusion model\nto generate desired views, rather than being biased toward the canonical views.\nFinally, we applied our 2D intuition to integrate Perp-Neg with the\nstate-of-the-art text-to-3D (DreamFusion) method, effectively addressing its\nJanus (multi-head) problem. Our project page is available at\nhttps://Perp-Neg.github.io/\n", "date": "2023-04-11", "categories": []} {"id": "2304.12670", "title": "Patch-based 3D Natural Scene Generation from a Single Example", "abstract": " We target a 3D generative model for general natural scenes that are typically\nunique and intricate. Lacking the necessary volumes of training data, along\nwith the difficulties of having ad hoc designs in presence of varying scene\ncharacteristics, renders existing setups intractable. Inspired by classical\npatch-based image models, we advocate for synthesizing 3D scenes at the patch\nlevel, given a single example. At the core of this work lies important\nalgorithmic designs w.r.t the scene representation and generative patch\nnearest-neighbor module, that address unique challenges arising from lifting\nclassical 2D patch-based framework to 3D generation. These design choices, on a\ncollective level, contribute to a robust, effective, and efficient model that\ncan generate high-quality general natural scenes with both realistic geometric\nstructure and visual appearance, in large quantities and varieties, as\ndemonstrated upon a variety of exemplar scenes.\n", "date": "2023-04-25", "categories": []} {"id": "2304.13153", "title": "LumiGAN: Unconditional Generation of Relightable 3D Human Faces", "abstract": " Unsupervised learning of 3D human faces from unstructured 2D image data is an\nactive research area. While recent works have achieved an impressive level of\nphotorealism, they commonly lack control of lighting, which prevents the\ngenerated assets from being deployed in novel environments. To this end, we\nintroduce LumiGAN, an unconditional Generative Adversarial Network (GAN) for 3D\nhuman faces with a physically based lighting module that enables relighting\nunder novel illumination at inference time. Unlike prior work, LumiGAN can\ncreate realistic shadow effects using an efficient visibility formulation that\nis learned in a self-supervised manner. LumiGAN generates plausible physical\nproperties for relightable faces, including surface normals, diffuse albedo,\nand specular tint without any ground truth data. In addition to relightability,\nwe demonstrate significantly improved geometry generation compared to\nstate-of-the-art non-relightable 3D GANs and notably better photorealism than\nexisting relightable GANs.\n", "date": "2023-04-25", "categories": []} {"id": "2304.12439", "title": "TextMesh: Generation of Realistic 3D Meshes From Text Prompts", "abstract": " The ability to generate highly realistic 2D images from mere text prompts has\nrecently made huge progress in terms of speed and quality, thanks to the advent\nof image diffusion models. Naturally, the question arises if this can be also\nachieved in the generation of 3D content from such text prompts. To this end, a\nnew line of methods recently emerged trying to harness diffusion models,\ntrained on 2D images, for supervision of 3D model generation using view\ndependent prompts. While achieving impressive results, these methods, however,\nhave two major drawbacks. First, rather than commonly used 3D meshes, they\ninstead generate neural radiance fields (NeRFs), making them impractical for\nmost real applications. Second, these approaches tend to produce over-saturated\nmodels, giving the output a cartoonish looking effect. Therefore, in this work\nwe propose a novel method for generation of highly realistic-looking 3D meshes.\nTo this end, we extend NeRF to employ an SDF backbone, leading to improved 3D\nmesh extraction. In addition, we propose a novel way to finetune the mesh\ntexture, removing the effect of high saturation and improving the details of\nthe output 3D mesh.\n", "date": "2023-04-24", "categories": []} {"id": "2212.01406", "title": "ClipFace: Text-guided Editing of Textured 3D Morphable Models", "abstract": " We propose ClipFace, a novel self-supervised approach for text-guided editing\nof textured 3D morphable model of faces. Specifically, we employ user-friendly\nlanguage prompts to enable control of the expressions as well as appearance of\n3D faces. We leverage the geometric expressiveness of 3D morphable models,\nwhich inherently possess limited controllability and texture expressivity, and\ndevelop a self-supervised generative model to jointly synthesize expressive,\ntextured, and articulated faces in 3D. We enable high-quality texture\ngeneration for 3D faces by adversarial self-supervised training, guided by\ndifferentiable rendering against collections of real RGB images. Controllable\nediting and manipulation are given by language prompts to adapt texture and\nexpression of the 3D morphable model. To this end, we propose a neural network\nthat predicts both texture and expression latent codes of the morphable model.\nOur model is trained in a self-supervised fashion by exploiting differentiable\nrendering and losses based on a pre-trained CLIP model. Once trained, our model\njointly predicts face textures in UV-space, along with expression parameters to\ncapture both geometry and texture changes in facial expressions in a single\nforward pass. We further show the applicability of our method to generate\ntemporally changing textures for a given animation sequence.\n", "date": "2022-12-02", "categories": []} {"id": "2304.09463", "title": "HyperStyle3D: Text-Guided 3D Portrait Stylization via Hypernetworks", "abstract": " Portrait stylization is a long-standing task enabling extensive applications.\nAlthough 2D-based methods have made great progress in recent years, real-world\napplications such as metaverse and games often demand 3D content. On the other\nhand, the requirement of 3D data, which is costly to acquire, significantly\nimpedes the development of 3D portrait stylization methods. In this paper,\ninspired by the success of 3D-aware GANs that bridge 2D and 3D domains with 3D\nfields as the intermediate representation for rendering 2D images, we propose a\nnovel method, dubbed HyperStyle3D, based on 3D-aware GANs for 3D portrait\nstylization. At the core of our method is a hyper-network learned to manipulate\nthe parameters of the generator in a single forward pass. It not only offers a\nstrong capacity to handle multiple styles with a single model, but also enables\nflexible fine-grained stylization that affects only texture, shape, or local\npart of the portrait. While the use of 3D-aware GANs bypasses the requirement\nof 3D data, we further alleviate the necessity of style images with the CLIP\nmodel being the stylization guidance. We conduct an extensive set of\nexperiments across the style, attribute, and shape, and meanwhile, measure the\n3D consistency. These experiments demonstrate the superior capability of our\nHyperStyle3D model in rendering 3D-consistent images in diverse styles,\ndeforming the face shape, and editing various attributes.\n", "date": "2023-04-19", "categories": []} {"id": "2304.09787", "title": "NeuralField-LDM: Scene Generation with Hierarchical Latent Diffusion\n Models", "abstract": " Automatically generating high-quality real world 3D scenes is of enormous\ninterest for applications such as virtual reality and robotics simulation.\nTowards this goal, we introduce NeuralField-LDM, a generative model capable of\nsynthesizing complex 3D environments. We leverage Latent Diffusion Models that\nhave been successfully utilized for efficient high-quality 2D content creation.\nWe first train a scene auto-encoder to express a set of image and pose pairs as\na neural field, represented as density and feature voxel grids that can be\nprojected to produce novel views of the scene. To further compress this\nrepresentation, we train a latent-autoencoder that maps the voxel grids to a\nset of latent representations. A hierarchical diffusion model is then fit to\nthe latents to complete the scene generation pipeline. We achieve a substantial\nimprovement over existing state-of-the-art scene generation models.\nAdditionally, we show how NeuralField-LDM can be used for a variety of 3D\ncontent creation applications, including conditional scene generation, scene\ninpainting and scene style manipulation.\n", "date": "2023-04-19", "categories": []} {"id": "2303.08133", "title": "MeshDiffusion: Score-based Generative 3D Mesh Modeling", "abstract": " We consider the task of generating realistic 3D shapes, which is useful for a\nvariety of applications such as automatic scene generation and physical\nsimulation. Compared to other 3D representations like voxels and point clouds,\nmeshes are more desirable in practice, because (1) they enable easy and\narbitrary manipulation of shapes for relighting and simulation, and (2) they\ncan fully leverage the power of modern graphics pipelines which are mostly\noptimized for meshes. Previous scalable methods for generating meshes typically\nrely on sub-optimal post-processing, and they tend to produce overly-smooth or\nnoisy surfaces without fine-grained geometric details. To overcome these\nshortcomings, we take advantage of the graph structure of meshes and use a\nsimple yet very effective generative modeling method to generate 3D meshes.\nSpecifically, we represent meshes with deformable tetrahedral grids, and then\ntrain a diffusion model on this direct parametrization. We demonstrate the\neffectiveness of our model on multiple generative tasks.\n", "date": "2023-03-14", "categories": []} {"id": "2304.07918", "title": "Likelihood-Based Generative Radiance Field with Latent Space\n Energy-Based Model for 3D-Aware Disentangled Image Representation", "abstract": " We propose the NeRF-LEBM, a likelihood-based top-down 3D-aware 2D image\ngenerative model that incorporates 3D representation via Neural Radiance Fields\n(NeRF) and 2D imaging process via differentiable volume rendering. The model\nrepresents an image as a rendering process from 3D object to 2D image and is\nconditioned on some latent variables that account for object characteristics\nand are assumed to follow informative trainable energy-based prior models. We\npropose two likelihood-based learning frameworks to train the NeRF-LEBM: (i)\nmaximum likelihood estimation with Markov chain Monte Carlo-based inference and\n(ii) variational inference with the reparameterization trick. We study our\nmodels in the scenarios with both known and unknown camera poses. Experiments\non several benchmark datasets demonstrate that the NeRF-LEBM can infer 3D\nobject structures from 2D images, generate 2D images with novel views and\nobjects, learn from incomplete 2D images, and learn from 2D images with known\nor unknown camera poses.\n", "date": "2023-04-16", "categories": []} {"id": "2210.12315", "title": "Diffusion Motion: Generate Text-Guided 3D Human Motion by Diffusion\n Model", "abstract": " We propose a simple and novel method for generating 3D human motion from\ncomplex natural language sentences, which describe different velocity,\ndirection and composition of all kinds of actions. Different from existing\nmethods that use classical generative architecture, we apply the Denoising\nDiffusion Probabilistic Model to this task, synthesizing diverse motion results\nunder the guidance of texts. The diffusion model converts white noise into\nstructured 3D motion by a Markov process with a series of denoising steps and\nis efficiently trained by optimizing a variational lower bound. To achieve the\ngoal of text-conditioned image synthesis, we use the classifier-free guidance\nstrategy to fuse text embedding into the model during training. Our experiments\ndemonstrate that our model achieves competitive results on HumanML3D test set\nquantitatively and can generate more visually natural and diverse examples. We\nalso show with experiments that our model is capable of zero-shot generation of\nmotions for unseen text guidance.\n", "date": "2022-10-22", "categories": []} {"id": "2207.09446", "title": "ShapeCrafter: A Recursive Text-Conditioned 3D Shape Generation Model", "abstract": " We present ShapeCrafter, a neural network for recursive text-conditioned 3D\nshape generation. Existing methods to generate text-conditioned 3D shapes\nconsume an entire text prompt to generate a 3D shape in a single step. However,\nhumans tend to describe shapes recursively-we may start with an initial\ndescription and progressively add details based on intermediate results. To\ncapture this recursive process, we introduce a method to generate a 3D shape\ndistribution, conditioned on an initial phrase, that gradually evolves as more\nphrases are added. Since existing datasets are insufficient for training this\napproach, we present Text2Shape++, a large dataset of 369K shape-text pairs\nthat supports recursive shape generation. To capture local details that are\noften used to refine shape descriptions, we build on top of vector-quantized\ndeep implicit functions that generate a distribution of high-quality shapes.\nResults show that our method can generate shapes consistent with text\ndescriptions, and shapes evolve gradually as more phrases are added. Our method\nsupports shape editing, extrapolation, and can enable new applications in\nhuman-machine collaboration for creative design.\n", "date": "2022-07-19", "categories": []} {"id": "2302.01579", "title": "Semantic 3D-aware Portrait Synthesis and Manipulation Based on\n Compositional Neural Radiance Field", "abstract": " Recently 3D-aware GAN methods with neural radiance field have developed\nrapidly. However, current methods model the whole image as an overall neural\nradiance field, which limits the partial semantic editability of synthetic\nresults. Since NeRF renders an image pixel by pixel, it is possible to split\nNeRF in the spatial dimension. We propose a Compositional Neural Radiance Field\n(CNeRF) for semantic 3D-aware portrait synthesis and manipulation. CNeRF\ndivides the image by semantic regions and learns an independent neural radiance\nfield for each region, and finally fuses them and renders the complete image.\nThus we can manipulate the synthesized semantic regions independently, while\nfixing the other parts unchanged. Furthermore, CNeRF is also designed to\ndecouple shape and texture within each semantic region. Compared to\nstate-of-the-art 3D-aware GAN methods, our approach enables fine-grained\nsemantic region manipulation, while maintaining high-quality 3D-consistent\nsynthesis. The ablation studies show the effectiveness of the structure and\nloss function used by our method. In addition real image inversion and cartoon\nportrait 3D editing experiments demonstrate the application potential of our\nmethod.\n", "date": "2023-02-03", "categories": []} {"id": "2304.03932", "title": "3D GANs and Latent Space: A comprehensive survey", "abstract": " Generative Adversarial Networks (GANs) have emerged as a significant player\nin generative modeling by mapping lower-dimensional random noise to\nhigher-dimensional spaces. These networks have been used to generate\nhigh-resolution images and 3D objects. The efficient modeling of 3D objects and\nhuman faces is crucial in the development process of 3D graphical environments\nsuch as games or simulations. 3D GANs are a new type of generative model used\nfor 3D reconstruction, point cloud reconstruction, and 3D semantic scene\ncompletion. The choice of distribution for noise is critical as it represents\nthe latent space. Understanding a GAN's latent space is essential for\nfine-tuning the generated samples, as demonstrated by the morphing of\nsemantically meaningful parts of images. In this work, we explore the latent\nspace and 3D GANs, examine several GAN variants and training methods to gain\ninsights into improving 3D GAN training, and suggest potential future\ndirections for further research.\n", "date": "2023-04-08", "categories": []} {"id": "2212.12952", "title": "Neural Shape Compiler: A Unified Framework for Transforming between\n Text, Point Cloud, and Program", "abstract": " 3D shapes have complementary abstractions from low-level geometry to\npart-based hierarchies to languages, which convey different levels of\ninformation. This paper presents a unified framework to translate between pairs\nof shape abstractions: $\\textit{Text}$ $\\Longleftrightarrow$ $\\textit{Point\nCloud}$ $\\Longleftrightarrow$ $\\textit{Program}$. We propose $\\textbf{Neural\nShape Compiler}$ to model the abstraction transformation as a conditional\ngeneration process. It converts 3D shapes of three abstract types into unified\ndiscrete shape code, transforms each shape code into code of other abstract\ntypes through the proposed $\\textit{ShapeCode Transformer}$, and decodes them\nto output the target shape abstraction. Point Cloud code is obtained in a\nclass-agnostic way by the proposed $\\textit{Point}$VQVAE. On Text2Shape,\nShapeGlot, ABO, Genre, and Program Synthetic datasets, Neural Shape Compiler\nshows strengths in $\\textit{Text}$ $\\Longrightarrow$ $\\textit{Point Cloud}$,\n$\\textit{Point Cloud}$ $\\Longrightarrow$ $\\textit{Text}$, $\\textit{Point\nCloud}$ $\\Longrightarrow$ $\\textit{Program}$, and Point Cloud Completion tasks.\nAdditionally, Neural Shape Compiler benefits from jointly training on all\nheterogeneous data and tasks.\n", "date": "2022-12-25", "categories": []} {"id": "2304.03526", "title": "Lift3D: Synthesize 3D Training Data by Lifting 2D GAN to 3D Generative\n Radiance Field", "abstract": " This work explores the use of 3D generative models to synthesize training\ndata for 3D vision tasks. The key requirements of the generative models are\nthat the generated data should be photorealistic to match the real-world\nscenarios, and the corresponding 3D attributes should be aligned with given\nsampling labels. However, we find that the recent NeRF-based 3D GANs hardly\nmeet the above requirements due to their designed generation pipeline and the\nlack of explicit 3D supervision. In this work, we propose Lift3D, an inverted\n2D-to-3D generation framework to achieve the data generation objectives. Lift3D\nhas several merits compared to prior methods: (1) Unlike previous 3D GANs that\nthe output resolution is fixed after training, Lift3D can generalize to any\ncamera intrinsic with higher resolution and photorealistic output. (2) By\nlifting well-disentangled 2D GAN to 3D object NeRF, Lift3D provides explicit 3D\ninformation of generated objects, thus offering accurate 3D annotations for\ndownstream tasks. We evaluate the effectiveness of our framework by augmenting\nautonomous driving datasets. Experimental results demonstrate that our data\ngeneration framework can effectively improve the performance of 3D object\ndetectors. Project page: https://len-li.github.io/lift3d-web.\n", "date": "2023-04-07", "categories": []} {"id": "2304.02827", "title": "DITTO-NeRF: Diffusion-based Iterative Text To Omni-directional 3D Model", "abstract": " The increasing demand for high-quality 3D content creation has motivated the\ndevelopment of automated methods for creating 3D object models from a single\nimage and/or from a text prompt. However, the reconstructed 3D objects using\nstate-of-the-art image-to-3D methods still exhibit low correspondence to the\ngiven image and low multi-view consistency. Recent state-of-the-art text-to-3D\nmethods are also limited, yielding 3D samples with low diversity per prompt\nwith long synthesis time. To address these challenges, we propose DITTO-NeRF, a\nnovel pipeline to generate a high-quality 3D NeRF model from a text prompt or a\nsingle image. Our DITTO-NeRF consists of constructing high-quality partial 3D\nobject for limited in-boundary (IB) angles using the given or text-generated 2D\nimage from the frontal view and then iteratively reconstructing the remaining\n3D NeRF using inpainting latent diffusion model. We propose progressive 3D\nobject reconstruction schemes in terms of scales (low to high resolution),\nangles (IB angles initially to outer-boundary (OB) later), and masks (object to\nbackground boundary) in our DITTO-NeRF so that high-quality information on IB\ncan be propagated into OB. Our DITTO-NeRF outperforms state-of-the-art methods\nin terms of fidelity and diversity qualitatively and quantitatively with much\nfaster training times than prior arts on image/text-to-3D such as DreamFusion,\nand NeuralLift-360.\n", "date": "2023-04-06", "categories": []} {"id": "2304.02602", "title": "Generative Novel View Synthesis with 3D-Aware Diffusion Models", "abstract": " We present a diffusion-based model for 3D-aware generative novel view\nsynthesis from as few as a single input image. Our model samples from the\ndistribution of possible renderings consistent with the input and, even in the\npresence of ambiguity, is capable of rendering diverse and plausible novel\nviews. To achieve this, our method makes use of existing 2D diffusion backbones\nbut, crucially, incorporates geometry priors in the form of a 3D feature\nvolume. This latent feature field captures the distribution over possible scene\nrepresentations and improves our method's ability to generate view-consistent\nnovel renderings. In addition to generating novel views, our method has the\nability to autoregressively synthesize 3D-consistent sequences. We demonstrate\nstate-of-the-art results on synthetic renderings and room-scale scenes; we also\nshow compelling results for challenging, real-world objects.\n", "date": "2023-04-05", "categories": []} {"id": "2302.12798", "title": "3D Generative Model Latent Disentanglement via Local Eigenprojection", "abstract": " Designing realistic digital humans is extremely complex. Most data-driven\ngenerative models used to simplify the creation of their underlying geometric\nshape do not offer control over the generation of local shape attributes. In\nthis paper, we overcome this limitation by introducing a novel loss function\ngrounded in spectral geometry and applicable to different neural-network-based\ngenerative models of 3D head and body meshes. Encouraging the latent variables\nof mesh variational autoencoders (VAEs) or generative adversarial networks\n(GANs) to follow the local eigenprojections of identity attributes, we improve\nlatent disentanglement and properly decouple the attribute creation.\nExperimental results show that our local eigenprojection disentangled (LED)\nmodels not only offer improved disentanglement with respect to the\nstate-of-the-art, but also maintain good generation capabilities with training\ntimes comparable to the vanilla implementations of the models.\n", "date": "2023-02-24", "categories": []} {"id": "2304.01900", "title": "PODIA-3D: Domain Adaptation of 3D Generative Model Across Large Domain\n Gap Using Pose-Preserved Text-to-Image Diffusion", "abstract": " Recently, significant advancements have been made in 3D generative models,\nhowever training these models across diverse domains is challenging and\nrequires an huge amount of training data and knowledge of pose distribution.\nText-guided domain adaptation methods have allowed the generator to be adapted\nto the target domains using text prompts, thereby obviating the need for\nassembling numerous data. Recently, DATID-3D presents impressive quality of\nsamples in text-guided domain, preserving diversity in text by leveraging\ntext-to-image diffusion. However, adapting 3D generators to domains with\nsignificant domain gaps from the source domain still remains challenging due to\nissues in current text-to-image diffusion models as following: 1) shape-pose\ntrade-off in diffusion-based translation, 2) pose bias, and 3) instance bias in\nthe target domain, resulting in inferior 3D shapes, low text-image\ncorrespondence, and low intra-domain diversity in the generated samples. To\naddress these issues, we propose a novel pipeline called PODIA-3D, which uses\npose-preserved text-to-image diffusion-based domain adaptation for 3D\ngenerative models. We construct a pose-preserved text-to-image diffusion model\nthat allows the use of extremely high-level noise for significant domain\nchanges. We also propose specialized-to-general sampling strategies to improve\nthe details of the generated samples. Moreover, to overcome the instance bias,\nwe introduce a text-guided debiasing method that improves intra-domain\ndiversity. Consequently, our method successfully adapts 3D generators across\nsignificant domain gaps. Our qualitative results and user study demonstrates\nthat our approach outperforms existing 3D text-guided domain adaptation methods\nin terms of text-image correspondence, realism, diversity of rendered images,\nand sense of depth of 3D shapes in the generated samples\n", "date": "2023-04-04", "categories": []} {"id": "2211.16431", "title": "NeuralLift-360: Lifting An In-the-wild 2D Photo to A 3D Object with\n 360{\\deg} Views", "abstract": " Virtual reality and augmented reality (XR) bring increasing demand for 3D\ncontent. However, creating high-quality 3D content requires tedious work that a\nhuman expert must do. In this work, we study the challenging task of lifting a\nsingle image to a 3D object and, for the first time, demonstrate the ability to\ngenerate a plausible 3D object with 360{\\deg} views that correspond well with\nthe given reference image. By conditioning on the reference image, our model\ncan fulfill the everlasting curiosity for synthesizing novel views of objects\nfrom images. Our technique sheds light on a promising direction of easing the\nworkflows for 3D artists and XR designers. We propose a novel framework, dubbed\nNeuralLift-360, that utilizes a depth-aware neural radiance representation\n(NeRF) and learns to craft the scene guided by denoising diffusion models. By\nintroducing a ranking loss, our NeuralLift-360 can be guided with rough depth\nestimation in the wild. We also adopt a CLIP-guided sampling strategy for the\ndiffusion prior to provide coherent guidance. Extensive experiments demonstrate\nthat our NeuralLift-360 significantly outperforms existing state-of-the-art\nbaselines. Project page: https://vita-group.github.io/NeuralLift-360/\n", "date": "2022-11-29", "categories": []} {"id": "2211.17260", "title": "SinGRAF: Learning a 3D Generative Radiance Field for a Single Scene", "abstract": " Generative models have shown great promise in synthesizing photorealistic 3D\nobjects, but they require large amounts of training data. We introduce SinGRAF,\na 3D-aware generative model that is trained with a few input images of a single\nscene. Once trained, SinGRAF generates different realizations of this 3D scene\nthat preserve the appearance of the input while varying scene layout. For this\npurpose, we build on recent progress in 3D GAN architectures and introduce a\nnovel progressive-scale patch discrimination approach during training. With\nseveral experiments, we demonstrate that the results produced by SinGRAF\noutperform the closest related works in both quality and diversity by a large\nmargin.\n", "date": "2022-11-30", "categories": []} {"id": "2212.14704", "title": "Dream3D: Zero-Shot Text-to-3D Synthesis Using 3D Shape Prior and\n Text-to-Image Diffusion Models", "abstract": " Recent CLIP-guided 3D optimization methods, such as DreamFields and\nPureCLIPNeRF, have achieved impressive results in zero-shot text-to-3D\nsynthesis. However, due to scratch training and random initialization without\nprior knowledge, these methods often fail to generate accurate and faithful 3D\nstructures that conform to the input text. In this paper, we make the first\nattempt to introduce explicit 3D shape priors into the CLIP-guided 3D\noptimization process. Specifically, we first generate a high-quality 3D shape\nfrom the input text in the text-to-shape stage as a 3D shape prior. We then use\nit as the initialization of a neural radiance field and optimize it with the\nfull prompt. To address the challenging text-to-shape generation task, we\npresent a simple yet effective approach that directly bridges the text and\nimage modalities with a powerful text-to-image diffusion model. To narrow the\nstyle domain gap between the images synthesized by the text-to-image diffusion\nmodel and shape renderings used to train the image-to-shape generator, we\nfurther propose to jointly optimize a learnable text prompt and fine-tune the\ntext-to-image diffusion model for rendering-style image generation. Our method,\nDream3D, is capable of generating imaginative 3D content with superior visual\nquality and shape accuracy compared to state-of-the-art methods.\n", "date": "2022-12-28", "categories": []} {"id": "2303.14184", "title": "Make-It-3D: High-Fidelity 3D Creation from A Single Image with Diffusion\n Prior", "abstract": " In this work, we investigate the problem of creating high-fidelity 3D content\nfrom only a single image. This is inherently challenging: it essentially\ninvolves estimating the underlying 3D geometry while simultaneously\nhallucinating unseen textures. To address this challenge, we leverage prior\nknowledge from a well-trained 2D diffusion model to act as 3D-aware supervision\nfor 3D creation. Our approach, Make-It-3D, employs a two-stage optimization\npipeline: the first stage optimizes a neural radiance field by incorporating\nconstraints from the reference image at the frontal view and diffusion prior at\nnovel views; the second stage transforms the coarse model into textured point\nclouds and further elevates the realism with diffusion prior while leveraging\nthe high-quality textures from the reference image. Extensive experiments\ndemonstrate that our method outperforms prior works by a large margin,\nresulting in faithful reconstructions and impressive visual quality. Our method\npresents the first attempt to achieve high-quality 3D creation from a single\nimage for general objects and enables various applications such as text-to-3D\ncreation and texture editing.\n", "date": "2023-03-24", "categories": []} {"id": "2304.00838", "title": "MetaHead: An Engine to Create Realistic Digital Head", "abstract": " Collecting and labeling training data is one important step for\nlearning-based methods because the process is time-consuming and biased. For\nface analysis tasks, although some generative models can be used to generate\nface data, they can only achieve a subset of generation diversity,\nreconstruction accuracy, 3D consistency, high-fidelity visual quality, and easy\neditability. One recent related work is the graphics-based generative method,\nbut it can only render low realism head with high computation cost. In this\npaper, we propose MetaHead, a unified and full-featured controllable digital\nhead engine, which consists of a controllable head radiance field(MetaHead-F)\nto super-realistically generate or reconstruct view-consistent 3D controllable\ndigital heads and a generic top-down image generation framework LabelHead to\ngenerate digital heads consistent with the given customizable feature labels.\nExperiments validate that our controllable digital head engine achieves the\nstate-of-the-art generation visual quality and reconstruction accuracy.\nMoreover, the generated labeled data can assist real training data and\nsignificantly surpass the labeled data generated by graphics-based methods in\nterms of training effect.\n", "date": "2023-04-03", "categories": []} {"id": "2211.16374", "title": "DATID-3D: Diversity-Preserved Domain Adaptation Using Text-to-Image\n Diffusion for 3D Generative Model", "abstract": " Recent 3D generative models have achieved remarkable performance in\nsynthesizing high resolution photorealistic images with view consistency and\ndetailed 3D shapes, but training them for diverse domains is challenging since\nit requires massive training images and their camera distribution information.\nText-guided domain adaptation methods have shown impressive performance on\nconverting the 2D generative model on one domain into the models on other\ndomains with different styles by leveraging the CLIP (Contrastive\nLanguage-Image Pre-training), rather than collecting massive datasets for those\ndomains. However, one drawback of them is that the sample diversity in the\noriginal generative model is not well-preserved in the domain-adapted\ngenerative models due to the deterministic nature of the CLIP text encoder.\nText-guided domain adaptation will be even more challenging for 3D generative\nmodels not only because of catastrophic diversity loss, but also because of\ninferior text-image correspondence and poor image quality. Here we propose\nDATID-3D, a domain adaptation method tailored for 3D generative models using\ntext-to-image diffusion models that can synthesize diverse images per text\nprompt without collecting additional images and camera information for the\ntarget domain. Unlike 3D extensions of prior text-guided domain adaptation\nmethods, our novel pipeline was able to fine-tune the state-of-the-art 3D\ngenerator of the source domain to synthesize high resolution, multi-view\nconsistent images in text-guided targeted domains without additional data,\noutperforming the existing text-guided domain adaptation methods in diversity\nand text-image correspondence. Furthermore, we propose and demonstrate diverse\n3D image manipulations such as one-shot instance-selected adaptation and\nsingle-view manipulated 3D reconstruction to fully enjoy diversity in text.\n", "date": "2022-11-29", "categories": []} {"id": "2303.17905", "title": "3D-aware Image Generation using 2D Diffusion Models", "abstract": " In this paper, we introduce a novel 3D-aware image generation method that\nleverages 2D diffusion models. We formulate the 3D-aware image generation task\nas multiview 2D image set generation, and further to a sequential\nunconditional-conditional multiview image generation process. This allows us to\nutilize 2D diffusion models to boost the generative modeling power of the\nmethod. Additionally, we incorporate depth information from monocular depth\nestimators to construct the training data for the conditional diffusion model\nusing only still images. We train our method on a large-scale dataset, i.e.,\nImageNet, which is not addressed by previous methods. It produces high-quality\nimages that significantly outperform prior methods. Furthermore, our approach\nshowcases its capability to generate instances with large view angles, even\nthough the training images are diverse and unaligned, gathered from\n\"in-the-wild\" real-world environments.\n", "date": "2023-03-31", "categories": []} {"id": "2303.18193", "title": "GVP: Generative Volumetric Primitives", "abstract": " Advances in 3D-aware generative models have pushed the boundary of image\nsynthesis with explicit camera control. To achieve high-resolution image\nsynthesis, several attempts have been made to design efficient generators, such\nas hybrid architectures with both 3D and 2D components. However, such a design\ncompromises multiview consistency, and the design of a pure 3D generator with\nhigh resolution is still an open problem. In this work, we present Generative\nVolumetric Primitives (GVP), the first pure 3D generative model that can sample\nand render 512-resolution images in real-time. GVP jointly models a number of\nvolumetric primitives and their spatial information, both of which can be\nefficiently generated via a 2D convolutional network. The mixture of these\nprimitives naturally captures the sparsity and correspondence in the 3D volume.\nThe training of such a generator with a high degree of freedom is made possible\nthrough a knowledge distillation technique. Experiments on several datasets\ndemonstrate superior efficiency and 3D consistency of GVP over the\nstate-of-the-art.\n", "date": "2023-03-31", "categories": []} {"id": "2303.17015", "title": "HyperDiffusion: Generating Implicit Neural Fields with Weight-Space\n Diffusion", "abstract": " Implicit neural fields, typically encoded by a multilayer perceptron (MLP)\nthat maps from coordinates (e.g., xyz) to signals (e.g., signed distances),\nhave shown remarkable promise as a high-fidelity and compact representation.\nHowever, the lack of a regular and explicit grid structure also makes it\nchallenging to apply generative modeling directly on implicit neural fields in\norder to synthesize new data. To this end, we propose HyperDiffusion, a novel\napproach for unconditional generative modeling of implicit neural fields.\nHyperDiffusion operates directly on MLP weights and generates new neural\nimplicit fields encoded by synthesized MLP parameters. Specifically, a\ncollection of MLPs is first optimized to faithfully represent individual data\nsamples. Subsequently, a diffusion process is trained in this MLP weight space\nto model the underlying distribution of neural implicit fields. HyperDiffusion\nenables diffusion modeling over a implicit, compact, and yet high-fidelity\nrepresentation of complex signals across 3D shapes and 4D mesh animations\nwithin one single unified framework.\n", "date": "2023-03-29", "categories": []} {"id": "2303.05371", "title": "3DGen: Triplane Latent Diffusion for Textured Mesh Generation", "abstract": " Latent diffusion models for image generation have crossed a quality threshold\nwhich enabled them to achieve mass adoption. Recently, a series of works have\nmade advancements towards replicating this success in the 3D domain,\nintroducing techniques such as point cloud VAE, triplane representation, neural\nimplicit surfaces and differentiable rendering based training. We take another\nstep along this direction, combining these developments in a two-step pipeline\nconsisting of 1) a triplane VAE which can learn latent representations of\ntextured meshes and 2) a conditional diffusion model which generates the\ntriplane features. For the first time this architecture allows conditional and\nunconditional generation of high quality textured or untextured 3D meshes\nacross multiple diverse categories in a few seconds on a single GPU. It\noutperforms previous work substantially on image-conditioned and unconditional\ngeneration on mesh quality as well as texture generation. Furthermore, we\ndemonstrate the scalability of our model to large datasets for increased\nquality and diversity. We will release our code and trained models.\n", "date": "2023-03-09", "categories": []} {"id": "2303.14167", "title": "UrbanGIRAFFE: Representing Urban Scenes as Compositional Generative\n Neural Feature Fields", "abstract": " Generating photorealistic images with controllable camera pose and scene\ncontents is essential for many applications including AR/VR and simulation.\nDespite the fact that rapid progress has been made in 3D-aware generative\nmodels, most existing methods focus on object-centric images and are not\napplicable to generating urban scenes for free camera viewpoint control and\nscene editing. To address this challenging task, we propose UrbanGIRAFFE, which\nuses a coarse 3D panoptic prior, including the layout distribution of\nuncountable stuff and countable objects, to guide a 3D-aware generative model.\nOur model is compositional and controllable as it breaks down the scene into\nstuff, objects, and sky. Using stuff prior in the form of semantic voxel grids,\nwe build a conditioned stuff generator that effectively incorporates the coarse\nsemantic and geometry information. The object layout prior further allows us to\nlearn an object generator from cluttered scenes. With proper loss functions,\nour approach facilitates photorealistic 3D-aware image synthesis with diverse\ncontrollability, including large camera movement, stuff editing, and object\nmanipulation. We validate the effectiveness of our model on both synthetic and\nreal-world datasets, including the challenging KITTI-360 dataset.\n", "date": "2023-03-24", "categories": []} {"id": "2303.15539", "title": "OmniAvatar: Geometry-Guided Controllable 3D Head Synthesis", "abstract": " We present OmniAvatar, a novel geometry-guided 3D head synthesis model\ntrained from in-the-wild unstructured images that is capable of synthesizing\ndiverse identity-preserved 3D heads with compelling dynamic details under full\ndisentangled control over camera poses, facial expressions, head shapes,\narticulated neck and jaw poses. To achieve such high level of disentangled\ncontrol, we first explicitly define a novel semantic signed distance function\n(SDF) around a head geometry (FLAME) conditioned on the control parameters.\nThis semantic SDF allows us to build a differentiable volumetric correspondence\nmap from the observation space to a disentangled canonical space from all the\ncontrol parameters. We then leverage the 3D-aware GAN framework (EG3D) to\nsynthesize detailed shape and appearance of 3D full heads in the canonical\nspace, followed by a volume rendering step guided by the volumetric\ncorrespondence map to output into the observation space. To ensure the control\naccuracy on the synthesized head shapes and expressions, we introduce a\ngeometry prior loss to conform to head SDF and a control loss to conform to the\nexpression code. Further, we enhance the temporal realism with dynamic details\nconditioned upon varying expressions and joint poses. Our model can synthesize\nmore preferable identity-preserved 3D heads with compelling dynamic details\ncompared to the state-of-the-art methods both qualitatively and quantitatively.\nWe also provide an ablation study to justify many of our system design choices.\n", "date": "2023-03-27", "categories": []} {"id": "2303.15780", "title": "Instruct 3D-to-3D: Text Instruction Guided 3D-to-3D conversion", "abstract": " We propose a high-quality 3D-to-3D conversion method, Instruct 3D-to-3D. Our\nmethod is designed for a novel task, which is to convert a given 3D scene to\nanother scene according to text instructions. Instruct 3D-to-3D applies\npretrained Image-to-Image diffusion models for 3D-to-3D conversion. This\nenables the likelihood maximization of each viewpoint image and high-quality 3D\ngeneration. In addition, our proposed method explicitly inputs the source 3D\nscene as a condition, which enhances 3D consistency and controllability of how\nmuch of the source 3D scene structure is reflected. We also propose dynamic\nscaling, which allows the intensity of the geometry transformation to be\nadjusted. We performed quantitative and qualitative evaluations and showed that\nour proposed method achieves higher quality 3D-to-3D conversions than baseline\nmethods.\n", "date": "2023-03-28", "categories": []} {"id": "2303.15805", "title": "StarNet: Style-Aware 3D Point Cloud Generation", "abstract": " This paper investigates an open research task of reconstructing and\ngenerating 3D point clouds. Most existing works of 3D generative models\ndirectly take the Gaussian prior as input for the decoder to generate 3D point\nclouds, which fail to learn disentangled latent codes, leading noisy\ninterpolated results. Most of the GAN-based models fail to discriminate the\nlocal geometries, resulting in the point clouds generated not evenly\ndistributed at the object surface, hence degrading the point cloud generation\nquality. Moreover, prevailing methods adopt computation-intensive frameworks,\nsuch as flow-based models and Markov chains, which take plenty of time and\nresources in the training phase. To resolve these limitations, this paper\nproposes a unified style-aware network architecture combining both point-wise\ndistance loss and adversarial loss, StarNet which is able to reconstruct and\ngenerate high-fidelity and even 3D point clouds using a mapping network that\ncan effectively disentangle the Gaussian prior from input's high-level\nattributes in the mapped latent space to generate realistic interpolated\nobjects. Experimental results demonstrate that our framework achieves\ncomparable state-of-the-art performance on various metrics in the point cloud\nreconstruction and generation tasks, but is more lightweight in model size,\nrequires much fewer parameters and less time for model training.\n", "date": "2023-03-28", "categories": []} {"id": "2303.15892", "title": "Head3D: Complete 3D Head Generation via Tri-plane Feature Distillation", "abstract": " Head generation with diverse identities is an important task in computer\nvision and computer graphics, widely used in multimedia applications. However,\ncurrent full head generation methods require a large number of 3D scans or\nmulti-view images to train the model, resulting in expensive data acquisition\ncost. To address this issue, we propose Head3D, a method to generate full 3D\nheads with limited multi-view images. Specifically, our approach first extracts\nfacial priors represented by tri-planes learned in EG3D, a 3D-aware generative\nmodel, and then proposes feature distillation to deliver the 3D frontal faces\ninto complete heads without compromising head integrity. To mitigate the domain\ngap between the face and head models, we present dual-discriminators to guide\nthe frontal and back head generation, respectively. Our model achieves\ncost-efficient and diverse complete head generation with photo-realistic\nrenderings and high-quality geometry representations. Extensive experiments\ndemonstrate the effectiveness of our proposed Head3D, both qualitatively and\nquantitatively.\n", "date": "2023-03-28", "categories": []} {"id": "2211.10440", "title": "Magic3D: High-Resolution Text-to-3D Content Creation", "abstract": " DreamFusion has recently demonstrated the utility of a pre-trained\ntext-to-image diffusion model to optimize Neural Radiance Fields (NeRF),\nachieving remarkable text-to-3D synthesis results. However, the method has two\ninherent limitations: (a) extremely slow optimization of NeRF and (b)\nlow-resolution image space supervision on NeRF, leading to low-quality 3D\nmodels with a long processing time. In this paper, we address these limitations\nby utilizing a two-stage optimization framework. First, we obtain a coarse\nmodel using a low-resolution diffusion prior and accelerate with a sparse 3D\nhash grid structure. Using the coarse representation as the initialization, we\nfurther optimize a textured 3D mesh model with an efficient differentiable\nrenderer interacting with a high-resolution latent diffusion model. Our method,\ndubbed Magic3D, can create high quality 3D mesh models in 40 minutes, which is\n2x faster than DreamFusion (reportedly taking 1.5 hours on average), while also\nachieving higher resolution. User studies show 61.7% raters to prefer our\napproach over DreamFusion. Together with the image-conditioned generation\ncapabilities, we provide users with new ways to control 3D synthesis, opening\nup new avenues to various creative applications.\n", "date": "2022-11-18", "categories": []} {"id": "2212.01206", "title": "DiffRF: Rendering-Guided 3D Radiance Field Diffusion", "abstract": " We introduce DiffRF, a novel approach for 3D radiance field synthesis based\non denoising diffusion probabilistic models. While existing diffusion-based\nmethods operate on images, latent codes, or point cloud data, we are the first\nto directly generate volumetric radiance fields. To this end, we propose a 3D\ndenoising model which directly operates on an explicit voxel grid\nrepresentation. However, as radiance fields generated from a set of posed\nimages can be ambiguous and contain artifacts, obtaining ground truth radiance\nfield samples is non-trivial. We address this challenge by pairing the\ndenoising formulation with a rendering loss, enabling our model to learn a\ndeviated prior that favours good image quality instead of trying to replicate\nfitting errors like floating artifacts. In contrast to 2D-diffusion models, our\nmodel learns multi-view consistent priors, enabling free-view synthesis and\naccurate shape generation. Compared to 3D GANs, our diffusion-based approach\nnaturally enables conditional generation such as masked completion or\nsingle-view 3D synthesis at inference time.\n", "date": "2022-12-02", "categories": []} {"id": "2301.02700", "title": "3DAvatarGAN: Bridging Domains for Personalized Editable Avatars", "abstract": " Modern 3D-GANs synthesize geometry and texture by training on large-scale\ndatasets with a consistent structure. Training such models on stylized,\nartistic data, with often unknown, highly variable geometry, and camera\ninformation has not yet been shown possible. Can we train a 3D GAN on such\nartistic data, while maintaining multi-view consistency and texture quality? To\nthis end, we propose an adaptation framework, where the source domain is a\npre-trained 3D-GAN, while the target domain is a 2D-GAN trained on artistic\ndatasets. We then distill the knowledge from a 2D generator to the source 3D\ngenerator. To do that, we first propose an optimization-based method to align\nthe distributions of camera parameters across domains. Second, we propose\nregularizations necessary to learn high-quality texture, while avoiding\ndegenerate geometric solutions, such as flat shapes. Third, we show a\ndeformation-based technique for modeling exaggerated geometry of artistic\ndomains, enabling -- as a byproduct -- personalized geometric editing. Finally,\nwe propose a novel inversion method for 3D-GANs linking the latent spaces of\nthe source and the target domains. Our contributions -- for the first time --\nallow for the generation, editing, and animation of personalized artistic 3D\navatars on artistic datasets.\n", "date": "2023-01-06", "categories": []} {"id": "2303.13508", "title": "DreamBooth3D: Subject-Driven Text-to-3D Generation", "abstract": " We present DreamBooth3D, an approach to personalize text-to-3D generative\nmodels from as few as 3-6 casually captured images of a subject. Our approach\ncombines recent advances in personalizing text-to-image models (DreamBooth)\nwith text-to-3D generation (DreamFusion). We find that naively combining these\nmethods fails to yield satisfactory subject-specific 3D assets due to\npersonalized text-to-image models overfitting to the input viewpoints of the\nsubject. We overcome this through a 3-stage optimization strategy where we\njointly leverage the 3D consistency of neural radiance fields together with the\npersonalization capability of text-to-image models. Our method can produce\nhigh-quality, subject-specific 3D assets with text-driven modifications such as\nnovel poses, colors and attributes that are not seen in any of the input images\nof the subject.\n", "date": "2023-03-23", "categories": []} {"id": "2303.14297", "title": "AgileGAN3D: Few-Shot 3D Portrait Stylization by Augmented Transfer\n Learning", "abstract": " While substantial progresses have been made in automated 2D portrait\nstylization, admirable 3D portrait stylization from a single user photo remains\nto be an unresolved challenge. One primary obstacle here is the lack of high\nquality stylized 3D training data. In this paper, we propose a novel framework\n\\emph{AgileGAN3D} that can produce 3D artistically appealing and personalized\nportraits with detailed geometry. New stylization can be obtained with just a\nfew (around 20) unpaired 2D exemplars. We achieve this by first leveraging\nexisting 2D stylization capabilities, \\emph{style prior creation}, to produce a\nlarge amount of augmented 2D style exemplars. These augmented exemplars are\ngenerated with accurate camera pose labels, as well as paired real face images,\nwhich prove to be critical for the downstream 3D stylization task. Capitalizing\non the recent advancement of 3D-aware GAN models, we perform \\emph{guided\ntransfer learning} on a pretrained 3D GAN generator to produce\nmulti-view-consistent stylized renderings. In order to achieve 3D GAN inversion\nthat can preserve subject's identity well, we incorporate \\emph{multi-view\nconsistency loss} in the training of our encoder. Our pipeline demonstrates\nstrong capability in turning user photos into a diverse range of 3D artistic\nportraits. Both qualitative results and quantitative evaluations have been\nconducted to show the superior performance of our method. Code and pretrained\nmodels will be released for reproduction purpose.\n", "date": "2023-03-24", "categories": []} {"id": "2303.14700", "title": "Learning Versatile 3D Shape Generation with Improved AR Models", "abstract": " Auto-Regressive (AR) models have achieved impressive results in 2D image\ngeneration by modeling joint distributions in the grid space. While this\napproach has been extended to the 3D domain for powerful shape generation, it\nstill has two limitations: expensive computations on volumetric grids and\nambiguous auto-regressive order along grid dimensions. To overcome these\nlimitations, we propose the Improved Auto-regressive Model (ImAM) for 3D shape\ngeneration, which applies discrete representation learning based on a latent\nvector instead of volumetric grids. Our approach not only reduces computational\ncosts but also preserves essential geometric details by learning the joint\ndistribution in a more tractable order. Moreover, thanks to the simplicity of\nour model architecture, we can naturally extend it from unconditional to\nconditional generation by concatenating various conditioning inputs, such as\npoint clouds, categories, images, and texts. Extensive experiments demonstrate\nthat ImAM can synthesize diverse and faithful shapes of multiple categories,\nachieving state-of-the-art performance.\n", "date": "2023-03-26", "categories": []} {"id": "2303.14706", "title": "BlobGAN-3D: A Spatially-Disentangled 3D-Aware Generative Model for\n Indoor Scenes", "abstract": " 3D-aware image synthesis has attracted increasing interest as it models the\n3D nature of our real world. However, performing realistic object-level editing\nof the generated images in the multi-object scenario still remains a challenge.\nRecently, a 2D GAN termed BlobGAN has demonstrated great multi-object editing\ncapabilities on real-world indoor scene datasets. In this work, we propose\nBlobGAN-3D, which is a 3D-aware improvement of the original 2D BlobGAN. We\nenable explicit camera pose control while maintaining the disentanglement for\nindividual objects in the scene by extending the 2D blobs into 3D blobs. We\nkeep the object-level editing capabilities of BlobGAN and in addition allow\nflexible control over the 3D location of the objects in the scene. We test our\nmethod on real-world indoor datasets and show that our method can achieve\ncomparable image quality compared to the 2D BlobGAN and other 3D-aware GAN\nbaselines while being able to enable camera pose control and object-level\nediting in the challenging multi-object real-world scenarios.\n", "date": "2023-03-26", "categories": []} {"id": "2303.15437", "title": "FaceLit: Neural 3D Relightable Faces", "abstract": " We propose a generative framework, FaceLit, capable of generating a 3D face\nthat can be rendered at various user-defined lighting conditions and views,\nlearned purely from 2D images in-the-wild without any manual annotation. Unlike\nexisting works that require careful capture setup or human labor, we rely on\noff-the-shelf pose and illumination estimators. With these estimates, we\nincorporate the Phong reflectance model in the neural volume rendering\nframework. Our model learns to generate shape and material properties of a face\nsuch that, when rendered according to the natural statistics of pose and\nillumination, produces photorealistic face images with multiview 3D and\nillumination consistency. Our method enables photorealistic generation of faces\nwith explicit illumination and view controls on multiple datasets - FFHQ,\nMetFaces and CelebA-HQ. We show state-of-the-art photorealism among 3D aware\nGANs on FFHQ dataset achieving an FID score of 3.5.\n", "date": "2023-03-27", "categories": []} {"id": "2301.07702", "title": "Learning 3D-aware Image Synthesis with Unknown Pose Distribution", "abstract": " Existing methods for 3D-aware image synthesis largely depend on the 3D pose\ndistribution pre-estimated on the training set. An inaccurate estimation may\nmislead the model into learning faulty geometry. This work proposes PoF3D that\nfrees generative radiance fields from the requirements of 3D pose priors. We\nfirst equip the generator with an efficient pose learner, which is able to\ninfer a pose from a latent code, to approximate the underlying true pose\ndistribution automatically. We then assign the discriminator a task to learn\npose distribution under the supervision of the generator and to differentiate\nreal and synthesized images with the predicted pose as the condition. The\npose-free generator and the pose-aware discriminator are jointly trained in an\nadversarial manner. Extensive results on a couple of datasets confirm that the\nperformance of our approach, regarding both image quality and geometry quality,\nis on par with state of the art. To our best knowledge, PoF3D demonstrates the\nfeasibility of learning high-quality 3D-aware image synthesis without using 3D\npose priors for the first time.\n", "date": "2023-01-18", "categories": []} {"id": "2303.12218", "title": "Compositional 3D Scene Generation using Locally Conditioned Diffusion", "abstract": " Designing complex 3D scenes has been a tedious, manual process requiring\ndomain expertise. Emerging text-to-3D generative models show great promise for\nmaking this task more intuitive, but existing approaches are limited to\nobject-level generation. We introduce \\textbf{locally conditioned diffusion} as\nan approach to compositional scene diffusion, providing control over semantic\nparts using text prompts and bounding boxes while ensuring seamless transitions\nbetween these parts. We demonstrate a score distillation sampling--based\ntext-to-3D synthesis pipeline that enables compositional 3D scene generation at\na higher fidelity than relevant baselines.\n", "date": "2023-03-21", "categories": []} {"id": "2303.13071", "title": "PanoHead: Geometry-Aware 3D Full-Head Synthesis in 360$^{\\circ}$", "abstract": " Synthesis and reconstruction of 3D human head has gained increasing interests\nin computer vision and computer graphics recently. Existing state-of-the-art 3D\ngenerative adversarial networks (GANs) for 3D human head synthesis are either\nlimited to near-frontal views or hard to preserve 3D consistency in large view\nangles. We propose PanoHead, the first 3D-aware generative model that enables\nhigh-quality view-consistent image synthesis of full heads in $360^\\circ$ with\ndiverse appearance and detailed geometry using only in-the-wild unstructured\nimages for training. At its core, we lift up the representation power of recent\n3D GANs and bridge the data alignment gap when training from in-the-wild images\nwith widely distributed views. Specifically, we propose a novel two-stage\nself-adaptive image alignment for robust 3D GAN training. We further introduce\na tri-grid neural volume representation that effectively addresses front-face\nand back-head feature entanglement rooted in the widely-adopted tri-plane\nformulation. Our method instills prior knowledge of 2D image segmentation in\nadversarial learning of 3D neural scene structures, enabling compositable head\nsynthesis in diverse backgrounds. Benefiting from these designs, our method\nsignificantly outperforms previous 3D GANs, generating high-quality 3D heads\nwith accurate geometry and diverse appearances, even with long wavy and afro\nhairstyles, renderable from arbitrary poses. Furthermore, we show that our\nsystem can reconstruct full 3D heads from single input images for personalized\nrealistic 3D avatars.\n", "date": "2023-03-23", "categories": []} {"id": "2303.13273", "title": "TAPS3D: Text-Guided 3D Textured Shape Generation from Pseudo Supervision", "abstract": " In this paper, we investigate an open research task of generating\ncontrollable 3D textured shapes from the given textual descriptions. Previous\nworks either require ground truth caption labeling or extensive optimization\ntime. To resolve these issues, we present a novel framework, TAPS3D, to train a\ntext-guided 3D shape generator with pseudo captions. Specifically, based on\nrendered 2D images, we retrieve relevant words from the CLIP vocabulary and\nconstruct pseudo captions using templates. Our constructed captions provide\nhigh-level semantic supervision for generated 3D shapes. Further, in order to\nproduce fine-grained textures and increase geometry diversity, we propose to\nadopt low-level image regularization to enable fake-rendered images to align\nwith the real ones. During the inference phase, our proposed model can generate\n3D textured shapes from the given text without any additional optimization. We\nconduct extensive experiments to analyze each of our proposed components and\nshow the efficacy of our framework in generating high-fidelity 3D textured and\ntext-relevant shapes.\n", "date": "2023-03-23", "categories": []} {"id": "2303.13450", "title": "Set-the-Scene: Global-Local Training for Generating Controllable NeRF\n Scenes", "abstract": " Recent breakthroughs in text-guided image generation have led to remarkable\nprogress in the field of 3D synthesis from text. By optimizing neural radiance\nfields (NeRF) directly from text, recent methods are able to produce remarkable\nresults. Yet, these methods are limited in their control of each object's\nplacement or appearance, as they represent the scene as a whole. This can be a\nmajor issue in scenarios that require refining or manipulating objects in the\nscene. To remedy this deficit, we propose a novel GlobalLocal training\nframework for synthesizing a 3D scene using object proxies. A proxy represents\nthe object's placement in the generated scene and optionally defines its coarse\ngeometry. The key to our approach is to represent each object as an independent\nNeRF. We alternate between optimizing each NeRF on its own and as part of the\nfull scene. Thus, a complete representation of each object can be learned,\nwhile also creating a harmonious scene with style and lighting match. We show\nthat using proxies allows a wide variety of editing options, such as adjusting\nthe placement of each independent object, removing objects from a scene, or\nrefining an object. Our results show that Set-the-Scene offers a powerful\nsolution for scene synthesis and manipulation, filling a crucial gap in\ncontrollable text-to-3D synthesis.\n", "date": "2023-03-23", "categories": []} {"id": "2303.13515", "title": "Persistent Nature: A Generative Model of Unbounded 3D Worlds", "abstract": " Despite increasingly realistic image quality, recent 3D image generative\nmodels often operate on 3D volumes of fixed extent with limited camera motions.\nWe investigate the task of unconditionally synthesizing unbounded nature\nscenes, enabling arbitrarily large camera motion while maintaining a persistent\n3D world model. Our scene representation consists of an extendable, planar\nscene layout grid, which can be rendered from arbitrary camera poses via a 3D\ndecoder and volume rendering, and a panoramic skydome. Based on this\nrepresentation, we learn a generative world model solely from single-view\ninternet photos. Our method enables simulating long flights through 3D\nlandscapes, while maintaining global scene consistency--for instance, returning\nto the starting point yields the same view of the scene. Our approach enables\nscene extrapolation beyond the fixed bounds of current 3D generative models,\nwhile also supporting a persistent, camera-independent world representation\nthat stands in contrast to auto-regressive 3D prediction models. Our project\npage: https://chail.github.io/persistent-nature/.\n", "date": "2023-03-23", "categories": []} {"id": "2212.04493", "title": "SDFusion: Multimodal 3D Shape Completion, Reconstruction, and Generation", "abstract": " In this work, we present a novel framework built to simplify 3D asset\ngeneration for amateur users. To enable interactive generation, our method\nsupports a variety of input modalities that can be easily provided by a human,\nincluding images, text, partially observed shapes and combinations of these,\nfurther allowing to adjust the strength of each input. At the core of our\napproach is an encoder-decoder, compressing 3D shapes into a compact latent\nrepresentation, upon which a diffusion model is learned. To enable a variety of\nmulti-modal inputs, we employ task-specific encoders with dropout followed by a\ncross-attention mechanism. Due to its flexibility, our model naturally supports\na variety of tasks, outperforming prior works on shape completion, image-based\n3D reconstruction, and text-to-3D. Most interestingly, our model can combine\nall these tasks into one swiss-army-knife tool, enabling the user to perform\nshape generation using incomplete shapes, images, and textual descriptions at\nthe same time, providing the relative weights for each input and facilitating\ninteractivity. Despite our approach being shape-only, we further show an\nefficient method to texture the generated shape using large-scale text-to-image\nmodels.\n", "date": "2022-12-08", "categories": []} {"id": "2303.09554", "title": "PartNeRF: Generating Part-Aware Editable 3D Shapes without 3D\n Supervision", "abstract": " Impressive progress in generative models and implicit representations gave\nrise to methods that can generate 3D shapes of high quality. However, being\nable to locally control and edit shapes is another essential property that can\nunlock several content creation applications. Local control can be achieved\nwith part-aware models, but existing methods require 3D supervision and cannot\nproduce textures. In this work, we devise PartNeRF, a novel part-aware\ngenerative model for editable 3D shape synthesis that does not require any\nexplicit 3D supervision. Our model generates objects as a set of locally\ndefined NeRFs, augmented with an affine transformation. This enables several\nediting operations such as applying transformations on parts, mixing parts from\ndifferent objects etc. To ensure distinct, manipulable parts we enforce a hard\nassignment of rays to parts that makes sure that the color of each ray is only\ndetermined by a single NeRF. As a result, altering one part does not affect the\nappearance of the others. Evaluations on various ShapeNet categories\ndemonstrate the ability of our model to generate editable 3D objects of\nimproved fidelity, compared to previous part-based generative approaches that\nrequire 3D supervision or models relying on NeRFs.\n", "date": "2023-03-16", "categories": []} {"id": "2303.11396", "title": "Text2Tex: Text-driven Texture Synthesis via Diffusion Models", "abstract": " We present Text2Tex, a novel method for generating high-quality textures for\n3D meshes from the given text prompts. Our method incorporates inpainting into\na pre-trained depth-aware image diffusion model to progressively synthesize\nhigh resolution partial textures from multiple viewpoints. To avoid\naccumulating inconsistent and stretched artifacts across views, we dynamically\nsegment the rendered view into a generation mask, which represents the\ngeneration status of each visible texel. This partitioned view representation\nguides the depth-aware inpainting model to generate and update partial textures\nfor the corresponding regions. Furthermore, we propose an automatic view\nsequence generation scheme to determine the next best view for updating the\npartial texture. Extensive experiments demonstrate that our method\nsignificantly outperforms the existing text-driven approaches and GAN-based\nmethods.\n", "date": "2023-03-20", "categories": []} {"id": "2211.12131", "title": "DiffDreamer: Towards Consistent Unsupervised Single-view Scene\n Extrapolation with Conditional Diffusion Models", "abstract": " Scene extrapolation -- the idea of generating novel views by flying into a\ngiven image -- is a promising, yet challenging task. For each predicted frame,\na joint inpainting and 3D refinement problem has to be solved, which is ill\nposed and includes a high level of ambiguity. Moreover, training data for\nlong-range scenes is difficult to obtain and usually lacks sufficient views to\ninfer accurate camera poses. We introduce DiffDreamer, an unsupervised\nframework capable of synthesizing novel views depicting a long camera\ntrajectory while training solely on internet-collected images of nature scenes.\nUtilizing the stochastic nature of the guided denoising steps, we train the\ndiffusion models to refine projected RGBD images but condition the denoising\nsteps on multiple past and future frames for inference. We demonstrate that\nimage-conditioned diffusion models can effectively perform long-range scene\nextrapolation while preserving consistency significantly better than prior\nGAN-based methods. DiffDreamer is a powerful and efficient solution for scene\nextrapolation, producing impressive results despite limited supervision.\nProject page: https://primecai.github.io/diffdreamer.\n", "date": "2022-11-22", "categories": []} {"id": "2303.10406", "title": "3DQD: Generalized Deep 3D Shape Prior via Part-Discretized Diffusion\n Process", "abstract": " We develop a generalized 3D shape generation prior model, tailored for\nmultiple 3D tasks including unconditional shape generation, point cloud\ncompletion, and cross-modality shape generation, etc. On one hand, to precisely\ncapture local fine detailed shape information, a vector quantized variational\nautoencoder (VQ-VAE) is utilized to index local geometry from a compactly\nlearned codebook based on a broad set of task training data. On the other hand,\na discrete diffusion generator is introduced to model the inherent structural\ndependencies among different tokens. In the meantime, a multi-frequency fusion\nmodule (MFM) is developed to suppress high-frequency shape feature\nfluctuations, guided by multi-frequency contextual information. The above\ndesigns jointly equip our proposed 3D shape prior model with high-fidelity,\ndiverse features as well as the capability of cross-modality alignment, and\nextensive experiments have demonstrated superior performances on various 3D\nshape generation tasks.\n", "date": "2023-03-18", "categories": []} {"id": "2303.11235", "title": "FullFormer: Generating Shapes Inside Shapes", "abstract": " Implicit generative models have been widely employed to model 3D data and\nhave recently proven to be successful in encoding and generating high-quality\n3D shapes. This work builds upon these models and alleviates current\nlimitations by presenting the first implicit generative model that facilitates\nthe generation of complex 3D shapes with rich internal geometric details. To\nachieve this, our model uses unsigned distance fields to represent nested 3D\nsurfaces allowing learning from non-watertight mesh data. We propose a\ntransformer-based autoregressive model for 3D shape generation that leverages\ncontext-rich tokens from vector quantized shape embeddings. The generated\ntokens are decoded into an unsigned distance field which is rendered into a\nnovel 3D shape exhibiting a rich internal structure. We demonstrate that our\nmodel achieves state-of-the-art point cloud generation results on popular\nclasses of 'Cars', 'Planes', and 'Chairs' of the ShapeNet dataset.\nAdditionally, we curate a dataset that exclusively comprises shapes with\nrealistic internal details from the `Cars' class of ShapeNet and demonstrate\nour method's efficacy in generating these shapes with internal geometry.\n", "date": "2023-03-20", "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": "2211.13757", "title": "Diffusion-SDF: Conditional Generative Modeling of Signed Distance\n Functions", "abstract": " Probabilistic diffusion models have achieved state-of-the-art results for\nimage synthesis, inpainting, and text-to-image tasks. However, they are still\nin the early stages of generating complex 3D shapes. This work proposes\nDiffusion-SDF, a generative model for shape completion, single-view\nreconstruction, and reconstruction of real-scanned point clouds. We use neural\nsigned distance functions (SDFs) as our 3D representation to parameterize the\ngeometry of various signals (e.g., point clouds, 2D images) through neural\nnetworks. Neural SDFs are implicit functions and diffusing them amounts to\nlearning the reversal of their neural network weights, which we solve using a\ncustom modulation module. Extensive experiments show that our method is capable\nof both realistic unconditional generation and conditional generation from\npartial inputs. This work expands the domain of diffusion models from learning\n2D, explicit representations, to 3D, implicit representations.\n", "date": "2022-11-24", "categories": []} {"id": "2303.07938", "title": "Controllable Mesh Generation Through Sparse Latent Point Diffusion\n Models", "abstract": " Mesh generation is of great value in various applications involving computer\ngraphics and virtual content, yet designing generative models for meshes is\nchallenging due to their irregular data structure and inconsistent topology of\nmeshes in the same category. In this work, we design a novel sparse latent\npoint diffusion model for mesh generation. Our key insight is to regard point\nclouds as an intermediate representation of meshes, and model the distribution\nof point clouds instead. While meshes can be generated from point clouds via\ntechniques like Shape as Points (SAP), the challenges of directly generating\nmeshes can be effectively avoided. To boost the efficiency and controllability\nof our mesh generation method, we propose to further encode point clouds to a\nset of sparse latent points with point-wise semantic meaningful features, where\ntwo DDPMs are trained in the space of sparse latent points to respectively\nmodel the distribution of the latent point positions and features at these\nlatent points. We find that sampling in this latent space is faster than\ndirectly sampling dense point clouds. Moreover, the sparse latent points also\nenable us to explicitly control both the overall structures and local details\nof the generated meshes. Extensive experiments are conducted on the ShapeNet\ndataset, where our proposed sparse latent point diffusion model achieves\nsuperior performance in terms of generation quality and controllability when\ncompared to existing methods.\n", "date": "2023-03-14", "categories": []} {"id": "2211.16927", "title": "3D GAN Inversion with Facial Symmetry Prior", "abstract": " Recently, a surge of high-quality 3D-aware GANs have been proposed, which\nleverage the generative power of neural rendering. It is natural to associate\n3D GANs with GAN inversion methods to project a real image into the generator's\nlatent space, allowing free-view consistent synthesis and editing, referred as\n3D GAN inversion. Although with the facial prior preserved in pre-trained 3D\nGANs, reconstructing a 3D portrait with only one monocular image is still an\nill-pose problem. The straightforward application of 2D GAN inversion methods\nfocuses on texture similarity only while ignoring the correctness of 3D\ngeometry shapes. It may raise geometry collapse effects, especially when\nreconstructing a side face under an extreme pose. Besides, the synthetic\nresults in novel views are prone to be blurry. In this work, we propose a novel\nmethod to promote 3D GAN inversion by introducing facial symmetry prior. We\ndesign a pipeline and constraints to make full use of the pseudo auxiliary view\nobtained via image flipping, which helps obtain a robust and reasonable\ngeometry shape during the inversion process. To enhance texture fidelity in\nunobserved viewpoints, pseudo labels from depth-guided 3D warping can provide\nextra supervision. We design constraints aimed at filtering out conflict areas\nfor optimization in asymmetric situations. Comprehensive quantitative and\nqualitative evaluations on image reconstruction and editing demonstrate the\nsuperiority of our method.\n", "date": "2022-11-30", "categories": []} {"id": "2211.11208", "title": "Next3D: Generative Neural Texture Rasterization for 3D-Aware Head\n Avatars", "abstract": " 3D-aware generative adversarial networks (GANs) synthesize high-fidelity and\nmulti-view-consistent facial images using only collections of single-view 2D\nimagery. Towards fine-grained control over facial attributes, recent efforts\nincorporate 3D Morphable Face Model (3DMM) to describe deformation in\ngenerative radiance fields either explicitly or implicitly. Explicit methods\nprovide fine-grained expression control but cannot handle topological changes\ncaused by hair and accessories, while implicit ones can model varied topologies\nbut have limited generalization caused by the unconstrained deformation fields.\nWe propose a novel 3D GAN framework for unsupervised learning of generative,\nhigh-quality and 3D-consistent facial avatars from unstructured 2D images. To\nachieve both deformation accuracy and topological flexibility, we propose a 3D\nrepresentation called Generative Texture-Rasterized Tri-planes. The proposed\nrepresentation learns Generative Neural Textures on top of parametric mesh\ntemplates and then projects them into three orthogonal-viewed feature planes\nthrough rasterization, forming a tri-plane feature representation for volume\nrendering. In this way, we combine both fine-grained expression control of\nmesh-guided explicit deformation and the flexibility of implicit volumetric\nrepresentation. We further propose specific modules for modeling mouth interior\nwhich is not taken into account by 3DMM. Our method demonstrates\nstate-of-the-art 3D-aware synthesis quality and animation ability through\nextensive experiments. Furthermore, serving as 3D prior, our animatable 3D\nrepresentation boosts multiple applications including one-shot facial avatars\nand 3D-aware stylization.\n", "date": "2022-11-21", "categories": []} {"id": "2303.06821", "title": "SDF-3DGAN: A 3D Object Generative Method Based on Implicit Signed\n Distance Function", "abstract": " In this paper, we develop a new method, termed SDF-3DGAN, for 3D object\ngeneration and 3D-Aware image synthesis tasks, which introduce implicit Signed\nDistance Function (SDF) as the 3D object representation method in the\ngenerative field. We apply SDF for higher quality representation of 3D object\nin space and design a new SDF neural renderer, which has higher efficiency and\nhigher accuracy. To train only on 2D images, we first generate the objects,\nwhich are represented by SDF, from Gaussian distribution. Then we render them\nto 2D images and use them to apply GAN training method together with 2D images\nin the dataset. In the new rendering method, we relieve all the potential of\nSDF mathematical property to alleviate computation pressure in the previous SDF\nneural renderer. In specific, our new SDF neural renderer can solve the problem\nof sampling ambiguity when the number of sampling point is not enough, \\ie use\nthe less points to finish higher quality sampling task in the rendering\npipeline. And in this rendering pipeline, we can locate the surface easily.\nTherefore, we apply normal loss on it to control the smoothness of generated\nobject surface, which can make our method enjoy the much higher generation\nquality. Quantitative and qualitative experiments conducted on public\nbenchmarks demonstrate favorable performance against the state-of-the-art\nmethods in 3D object generation task and 3D-Aware image synthesis task. Our\ncodes will be released at https://github.com/lutao2021/SDF-3DGAN.\n", "date": "2023-03-13", "categories": []} {"id": "2303.02688", "title": "Text2Face: A Multi-Modal 3D Face Model", "abstract": " We present the first 3D morphable modelling approach, whereby 3D face shape\ncan be directly and completely defined using a textual prompt. Building on work\nin multi-modal learning, we extend the FLAME head model to a common\nimage-and-text latent space. This allows for direct 3D Morphable Model (3DMM)\nparameter generation and therefore shape manipulation from textual\ndescriptions. Our method, Text2Face, has many applications; for example:\ngenerating police photofits where the input is already in natural language. It\nfurther enables multi-modal 3DMM image fitting to sketches and sculptures, as\nwell as images.\n", "date": "2023-03-05", "categories": []} {"id": "2303.01416", "title": "3D generation on ImageNet", "abstract": " Existing 3D-from-2D generators are typically designed for well-curated\nsingle-category datasets, where all the objects have (approximately) the same\nscale, 3D location, and orientation, and the camera always points to the center\nof the scene. This makes them inapplicable to diverse, in-the-wild datasets of\nnon-alignable scenes rendered from arbitrary camera poses. In this work, we\ndevelop a 3D generator with Generic Priors (3DGP): a 3D synthesis framework\nwith more general assumptions about the training data, and show that it scales\nto very challenging datasets, like ImageNet. Our model is based on three new\nideas. First, we incorporate an inaccurate off-the-shelf depth estimator into\n3D GAN training via a special depth adaptation module to handle the\nimprecision. Then, we create a flexible camera model and a regularization\nstrategy for it to learn its distribution parameters during training. Finally,\nwe extend the recent ideas of transferring knowledge from pre-trained\nclassifiers into GANs for patch-wise trained models by employing a simple\ndistillation-based technique on top of the discriminator. It achieves more\nstable training than the existing methods and speeds up the convergence by at\nleast 40%. We explore our model on four datasets: SDIP Dogs 256x256, SDIP\nElephants 256x256, LSUN Horses 256x256, and ImageNet 256x256, and demonstrate\nthat 3DGP outperforms the recent state-of-the-art in terms of both texture and\ngeometry quality. Code and visualizations:\nhttps://snap-research.github.io/3dgp.\n", "date": "2023-03-02", "categories": []} {"id": "2303.00167", "title": "Sketch2Cloth: Sketch-based 3D Garment Generation with Unsigned Distance\n Fields", "abstract": " 3D model reconstruction from a single image has achieved great progress with\nthe recent deep generative models. However, the conventional reconstruction\napproaches with template mesh deformation and implicit fields have difficulty\nin reconstructing non-watertight 3D mesh models, such as garments. In contrast\nto image-based modeling, the sketch-based approach can help users generate 3D\nmodels to meet the design intentions from hand-drawn sketches. In this study,\nwe propose Sketch2Cloth, a sketch-based 3D garment generation system using the\nunsigned distance fields from the user's sketch input. Sketch2Cloth first\nestimates the unsigned distance function of the target 3D model from the sketch\ninput, and extracts the mesh from the estimated field with Marching Cubes. We\nalso provide the model editing function to modify the generated mesh. We\nverified the proposed Sketch2Cloth with quantitative evaluations on garment\ngeneration and editing with a state-of-the-art approach.\n", "date": "2023-03-01", "categories": []} {"id": "2209.04145", "title": "ISS: Image as Stepping Stone for Text-Guided 3D Shape Generation", "abstract": " Text-guided 3D shape generation remains challenging due to the absence of\nlarge paired text-shape data, the substantial semantic gap between these two\nmodalities, and the structural complexity of 3D shapes. This paper presents a\nnew framework called Image as Stepping Stone (ISS) for the task by introducing\n2D image as a stepping stone to connect the two modalities and to eliminate the\nneed for paired text-shape data. Our key contribution is a two-stage\nfeature-space-alignment approach that maps CLIP features to shapes by\nharnessing a pre-trained single-view reconstruction (SVR) model with multi-view\nsupervisions: first map the CLIP image feature to the detail-rich shape space\nin the SVR model, then map the CLIP text feature to the shape space and\noptimize the mapping by encouraging CLIP consistency between the input text and\nthe rendered images. Further, we formulate a text-guided shape stylization\nmodule to dress up the output shapes with novel textures. Beyond existing works\non 3D shape generation from text, our new approach is general for creating\nshapes in a broad range of categories, without requiring paired text-shape\ndata. Experimental results manifest that our approach outperforms the\nstate-of-the-arts and our baselines in terms of fidelity and consistency with\ntext. Further, our approach can stylize the generated shapes with both\nrealistic and fantasy structures and textures.\n", "date": "2022-09-09", "categories": []} {"id": "2302.10663", "title": "RealFusion: 360{\\deg} Reconstruction of Any Object from a Single Image", "abstract": " We consider the problem of reconstructing a full 360{\\deg} photographic model\nof an object from a single image of it. We do so by fitting a neural radiance\nfield to the image, but find this problem to be severely ill-posed. We thus\ntake an off-the-self conditional image generator based on diffusion and\nengineer a prompt that encourages it to \"dream up\" novel views of the object.\nUsing an approach inspired by DreamFields and DreamFusion, we fuse the given\ninput view, the conditional prior, and other regularizers in a final,\nconsistent reconstruction. We demonstrate state-of-the-art reconstruction\nresults on benchmark images when compared to prior methods for monocular 3D\nreconstruction of objects. Qualitatively, our reconstructions provide a\nfaithful match of the input view and a plausible extrapolation of its\nappearance and 3D shape, including to the side of the object not visible in the\nimage.\n", "date": "2023-02-21", "categories": []} {"id": "2302.10237", "title": "SceneHGN: Hierarchical Graph Networks for 3D Indoor Scene Generation\n with Fine-Grained Geometry", "abstract": " 3D indoor scenes are widely used in computer graphics, with applications\nranging from interior design to gaming to virtual and augmented reality. They\nalso contain rich information, including room layout, as well as furniture\ntype, geometry, and placement. High-quality 3D indoor scenes are highly\ndemanded while it requires expertise and is time-consuming to design\nhigh-quality 3D indoor scenes manually. Existing research only addresses\npartial problems: some works learn to generate room layout, and other works\nfocus on generating detailed structure and geometry of individual furniture\nobjects. However, these partial steps are related and should be addressed\ntogether for optimal synthesis. We propose SCENEHGN, a hierarchical graph\nnetwork for 3D indoor scenes that takes into account the full hierarchy from\nthe room level to the object level, then finally to the object part level.\nTherefore for the first time, our method is able to directly generate plausible\n3D room content, including furniture objects with fine-grained geometry, and\ntheir layout. To address the challenge, we introduce functional regions as\nintermediate proxies between the room and object levels to make learning more\nmanageable. To ensure plausibility, our graph-based representation incorporates\nboth vertical edges connecting child nodes with parent nodes from different\nlevels, and horizontal edges encoding relationships between nodes at the same\nlevel. Extensive experiments demonstrate that our method produces superior\ngeneration results, even when comparing results of partial steps with\nalternative methods that can only achieve these. We also demonstrate that our\nmethod is effective for various applications such as part-level room editing,\nroom interpolation, and room generation by arbitrary room boundaries.\n", "date": "2023-02-16", "categories": []} {"id": "2302.09486", "title": "LC-NeRF: Local Controllable Face Generation in Neural Randiance Field", "abstract": " 3D face generation has achieved high visual quality and 3D consistency thanks\nto the development of neural radiance fields (NeRF). Recently, to generate and\nedit 3D faces with NeRF representation, some methods are proposed and achieve\ngood results in decoupling geometry and texture. The latent codes of these\ngenerative models affect the whole face, and hence modifications to these codes\ncause the entire face to change. However, users usually edit a local region\nwhen editing faces and do not want other regions to be affected. Since changes\nto the latent code affect global generation results, these methods do not allow\nfor fine-grained control of local facial regions. To improve local\ncontrollability in NeRF-based face editing, we propose LC-NeRF, which is\ncomposed of a Local Region Generators Module and a Spatial-Aware Fusion Module,\nallowing for local geometry and texture control of local facial regions.\nQualitative and quantitative evaluations show that our method provides better\nlocal editing than state-of-the-art face editing methods. Our method also\nperforms well in downstream tasks, such as text-driven facial image editing.\n", "date": "2023-02-19", "categories": []} {"id": "2302.06833", "title": "VQ3D: Learning a 3D-Aware Generative Model on ImageNet", "abstract": " Recent work has shown the possibility of training generative models of 3D\ncontent from 2D image collections on small datasets corresponding to a single\nobject class, such as human faces, animal faces, or cars. However, these models\nstruggle on larger, more complex datasets. To model diverse and unconstrained\nimage collections such as ImageNet, we present VQ3D, which introduces a\nNeRF-based decoder into a two-stage vector-quantized autoencoder. Our Stage 1\nallows for the reconstruction of an input image and the ability to change the\ncamera position around the image, and our Stage 2 allows for the generation of\nnew 3D scenes. VQ3D is capable of generating and reconstructing 3D-aware images\nfrom the 1000-class ImageNet dataset of 1.2 million training images. We achieve\nan ImageNet generation FID score of 16.8, compared to 69.8 for the next best\nbaseline method.\n", "date": "2023-02-14", "categories": []} {"id": "2206.08547", "title": "Texture Generation Using A Graph Generative Adversarial Network And\n Differentiable Rendering", "abstract": " Novel photo-realistic texture synthesis is an important task for generating\nnovel scenes, including asset generation for 3D simulations. However, to date,\nthese methods predominantly generate textured objects in 2D space. If we rely\non 2D object generation, then we need to make a computationally expensive\nforward pass each time we change the camera viewpoint or lighting. Recent work\nthat can generate textures in 3D requires 3D component segmentation that is\nexpensive to acquire. In this work, we present a novel conditional generative\narchitecture that we call a graph generative adversarial network (GGAN) that\ncan generate textures in 3D by learning object component information in an\nunsupervised way. In this framework, we do not need an expensive forward pass\nwhenever the camera viewpoint or lighting changes, and we do not need expensive\n3D part information for training, yet the model can generalize to unseen 3D\nmeshes and generate appropriate novel 3D textures. We compare this approach\nagainst state-of-the-art texture generation methods and demonstrate that the\nGGAN obtains significantly better texture generation quality (according to\nFrechet inception distance). We release our model source code as open source.\n", "date": "2022-06-17", "categories": []} {"id": "2302.01721", "title": "TEXTure: Text-Guided Texturing of 3D Shapes", "abstract": " In this paper, we present TEXTure, a novel method for text-guided generation,\nediting, and transfer of textures for 3D shapes. Leveraging a pretrained\ndepth-to-image diffusion model, TEXTure applies an iterative scheme that paints\na 3D model from different viewpoints. Yet, while depth-to-image models can\ncreate plausible textures from a single viewpoint, the stochastic nature of the\ngeneration process can cause many inconsistencies when texturing an entire 3D\nobject. To tackle these problems, we dynamically define a trimap partitioning\nof the rendered image into three progression states, and present a novel\nelaborated diffusion sampling process that uses this trimap representation to\ngenerate seamless textures from different views. We then show that one can\ntransfer the generated texture maps to new 3D geometries without requiring\nexplicit surface-to-surface mapping, as well as extract semantic textures from\na set of images without requiring any explicit reconstruction. Finally, we show\nthat TEXTure can be used to not only generate new textures but also edit and\nrefine existing textures using either a text prompt or user-provided scribbles.\nWe demonstrate that our TEXTuring method excels at generating, transferring,\nand editing textures through extensive evaluation, and further close the gap\nbetween 2D image generation and 3D texturing.\n", "date": "2023-02-03", "categories": []} {"id": "2302.00190", "title": "Neural Wavelet-domain Diffusion for 3D Shape Generation, Inversion, and\n Manipulation", "abstract": " This paper presents a new approach for 3D shape generation, inversion, and\nmanipulation, through a direct generative modeling on a continuous implicit\nrepresentation in wavelet domain. Specifically, we propose a compact wavelet\nrepresentation with a pair of coarse and detail coefficient volumes to\nimplicitly represent 3D shapes via truncated signed distance functions and\nmulti-scale biorthogonal wavelets. Then, we design a pair of neural networks: a\ndiffusion-based generator to produce diverse shapes in the form of the coarse\ncoefficient volumes and a detail predictor to produce compatible detail\ncoefficient volumes for introducing fine structures and details. Further, we\nmay jointly train an encoder network to learn a latent space for inverting\nshapes, allowing us to enable a rich variety of whole-shape and region-aware\nshape manipulations. Both quantitative and qualitative experimental results\nmanifest the compelling shape generation, inversion, and manipulation\ncapabilities of our approach over the state-of-the-art methods.\n", "date": "2023-02-01", "categories": []} {"id": "2301.11280", "title": "Text-To-4D Dynamic Scene Generation", "abstract": " We present MAV3D (Make-A-Video3D), a method for generating three-dimensional\ndynamic scenes from text descriptions. Our approach uses a 4D dynamic Neural\nRadiance Field (NeRF), which is optimized for scene appearance, density, and\nmotion consistency by querying a Text-to-Video (T2V) diffusion-based model. The\ndynamic video output generated from the provided text can be viewed from any\ncamera location and angle, and can be composited into any 3D environment. MAV3D\ndoes not require any 3D or 4D data and the T2V model is trained only on\nText-Image pairs and unlabeled videos. We demonstrate the effectiveness of our\napproach using comprehensive quantitative and qualitative experiments and show\nan improvement over previously established internal baselines. To the best of\nour knowledge, our method is the first to generate 3D dynamic scenes given a\ntext description.\n", "date": "2023-01-26", "categories": []} {"id": "2301.05747", "title": "Laser: Latent Set Representations for 3D Generative Modeling", "abstract": " NeRF provides unparalleled fidelity of novel view synthesis: rendering a 3D\nscene from an arbitrary viewpoint. NeRF requires training on a large number of\nviews that fully cover a scene, which limits its applicability. While these\nissues can be addressed by learning a prior over scenes in various forms,\nprevious approaches have been either applied to overly simple scenes or\nstruggling to render unobserved parts. We introduce Laser-NV: a generative\nmodel which achieves high modelling capacity, and which is based on a\nset-valued latent representation modelled by normalizing flows. Similarly to\nprevious amortized approaches, Laser-NV learns structure from multiple scenes\nand is capable of fast, feed-forward inference from few views. To encourage\nhigher rendering fidelity and consistency with observed views, Laser-NV further\nincorporates a geometry-informed attention mechanism over the observed views.\nLaser-NV further produces diverse and plausible completions of occluded parts\nof a scene while remaining consistent with observations. Laser-NV shows\nstate-of-the-art novel-view synthesis quality when evaluated on ShapeNet and on\na novel simulated City dataset, which features high uncertainty in the\nunobserved regions of the scene.\n", "date": "2023-01-13", "categories": []} {"id": "2301.06015", "title": "Diffusion-based Generation, Optimization, and Planning in 3D Scenes", "abstract": " We introduce SceneDiffuser, a conditional generative model for 3D scene\nunderstanding. SceneDiffuser provides a unified model for solving\nscene-conditioned generation, optimization, and planning. In contrast to prior\nworks, SceneDiffuser is intrinsically scene-aware, physics-based, and\ngoal-oriented. With an iterative sampling strategy, SceneDiffuser jointly\nformulates the scene-aware generation, physics-based optimization, and\ngoal-oriented planning via a diffusion-based denoising process in a fully\ndifferentiable fashion. Such a design alleviates the discrepancies among\ndifferent modules and the posterior collapse of previous scene-conditioned\ngenerative models. We evaluate SceneDiffuser with various 3D scene\nunderstanding tasks, including human pose and motion generation, dexterous\ngrasp generation, path planning for 3D navigation, and motion planning for\nrobot arms. The results show significant improvements compared with previous\nmodels, demonstrating the tremendous potential of SceneDiffuser for the broad\ncommunity of 3D scene understanding.\n", "date": "2023-01-15", "categories": []} {"id": "2206.14314", "title": "Generative Neural Articulated Radiance Fields", "abstract": " Unsupervised learning of 3D-aware generative adversarial networks (GANs)\nusing only collections of single-view 2D photographs has very recently made\nmuch progress. These 3D GANs, however, have not been demonstrated for human\nbodies and the generated radiance fields of existing frameworks are not\ndirectly editable, limiting their applicability in downstream tasks. We propose\na solution to these challenges by developing a 3D GAN framework that learns to\ngenerate radiance fields of human bodies or faces in a canonical pose and warp\nthem using an explicit deformation field into a desired body pose or facial\nexpression. Using our framework, we demonstrate the first high-quality radiance\nfield generation results for human bodies. Moreover, we show that our\ndeformation-aware training procedure significantly improves the quality of\ngenerated bodies or faces when editing their poses or facial expressions\ncompared to a 3D GAN that is not trained with explicit deformations.\n", "date": "2022-06-28", "categories": []} {"id": "2301.00527", "title": "Diffusion Probabilistic Models for Scene-Scale 3D Categorical Data", "abstract": " In this paper, we learn a diffusion model to generate 3D data on a\nscene-scale. Specifically, our model crafts a 3D scene consisting of multiple\nobjects, while recent diffusion research has focused on a single object. To\nrealize our goal, we represent a scene with discrete class labels, i.e.,\ncategorical distribution, to assign multiple objects into semantic categories.\nThus, we extend discrete diffusion models to learn scene-scale categorical\ndistributions. In addition, we validate that a latent diffusion model can\nreduce computation costs for training and deploying. To the best of our\nknowledge, our work is the first to apply discrete and latent diffusion for 3D\ncategorical data on a scene-scale. We further propose to perform semantic scene\ncompletion (SSC) by learning a conditional distribution using our diffusion\nmodel, where the condition is a partial observation in a sparse point cloud. In\nexperiments, we empirically show that our diffusion models not only generate\nreasonable scenes, but also perform the scene completion task better than a\ndiscriminative model. Our code and models are available at\nhttps://github.com/zoomin-lee/scene-scale-diffusion\n", "date": "2023-01-02", "categories": []} {"id": "2212.11984", "title": "DisCoScene: Spatially Disentangled Generative Radiance Fields for\n Controllable 3D-aware Scene Synthesis", "abstract": " Existing 3D-aware image synthesis approaches mainly focus on generating a\nsingle canonical object and show limited capacity in composing a complex scene\ncontaining a variety of objects. This work presents DisCoScene: a 3Daware\ngenerative model for high-quality and controllable scene synthesis. The key\ningredient of our method is a very abstract object-level representation (i.e.,\n3D bounding boxes without semantic annotation) as the scene layout prior, which\nis simple to obtain, general to describe various scene contents, and yet\ninformative to disentangle objects and background. Moreover, it serves as an\nintuitive user control for scene editing. Based on such a prior, the proposed\nmodel spatially disentangles the whole scene into object-centric generative\nradiance fields by learning on only 2D images with the global-local\ndiscrimination. Our model obtains the generation fidelity and editing\nflexibility of individual objects while being able to efficiently compose\nobjects and the background into a complete scene. We demonstrate\nstate-of-the-art performance on many scene datasets, including the challenging\nWaymo outdoor dataset. Project page:\nhttps://snap-research.github.io/discoscene/\n", "date": "2022-12-22", "categories": []} {"id": "2212.08751", "title": "Point-E: A System for Generating 3D Point Clouds from Complex Prompts", "abstract": " While recent work on text-conditional 3D object generation has shown\npromising results, the state-of-the-art methods typically require multiple\nGPU-hours to produce a single sample. This is in stark contrast to\nstate-of-the-art generative image models, which produce samples in a number of\nseconds or minutes. In this paper, we explore an alternative method for 3D\nobject generation which produces 3D models in only 1-2 minutes on a single GPU.\nOur method first generates a single synthetic view using a text-to-image\ndiffusion model, and then produces a 3D point cloud using a second diffusion\nmodel which conditions on the generated image. While our method still falls\nshort of the state-of-the-art in terms of sample quality, it is one to two\norders of magnitude faster to sample from, offering a practical trade-off for\nsome use cases. We release our pre-trained point cloud diffusion models, as\nwell as evaluation code and models, at https://github.com/openai/point-e.\n", "date": "2022-12-16", "categories": []} {"id": "2208.02946", "title": "Learning to Generate 3D Shapes from a Single Example", "abstract": " Existing generative models for 3D shapes are typically trained on a large 3D\ndataset, often of a specific object category. In this paper, we investigate the\ndeep generative model that learns from only a single reference 3D shape.\nSpecifically, we present a multi-scale GAN-based model designed to capture the\ninput shape's geometric features across a range of spatial scales. To avoid\nlarge memory and computational cost induced by operating on the 3D volume, we\nbuild our generator atop the tri-plane hybrid representation, which requires\nonly 2D convolutions. We train our generative model on a voxel pyramid of the\nreference shape, without the need of any external supervision or manual\nannotation. Once trained, our model can generate diverse and high-quality 3D\nshapes possibly of different sizes and aspect ratios. The resulting shapes\npresent variations across different scales, and at the same time retain the\nglobal structure of the reference shape. Through extensive evaluation, both\nqualitative and quantitative, we demonstrate that our model can generate 3D\nshapes of various types.\n", "date": "2022-08-05", "categories": []} {"id": "2206.10535", "title": "EpiGRAF: Rethinking training of 3D GANs", "abstract": " A very recent trend in generative modeling is building 3D-aware generators\nfrom 2D image collections. To induce the 3D bias, such models typically rely on\nvolumetric rendering, which is expensive to employ at high resolutions. During\nthe past months, there appeared more than 10 works that address this scaling\nissue by training a separate 2D decoder to upsample a low-resolution image (or\na feature tensor) produced from a pure 3D generator. But this solution comes at\na cost: not only does it break multi-view consistency (i.e. shape and texture\nchange when the camera moves), but it also learns the geometry in a low\nfidelity. In this work, we show that it is possible to obtain a high-resolution\n3D generator with SotA image quality by following a completely different route\nof simply training the model patch-wise. We revisit and improve this\noptimization scheme in two ways. First, we design a location- and scale-aware\ndiscriminator to work on patches of different proportions and spatial\npositions. Second, we modify the patch sampling strategy based on an annealed\nbeta distribution to stabilize training and accelerate the convergence. The\nresulted model, named EpiGRAF, is an efficient, high-resolution, pure 3D\ngenerator, and we test it on four datasets (two introduced in this work) at\n$256^2$ and $512^2$ resolutions. It obtains state-of-the-art image quality,\nhigh-fidelity geometry and trains ${\\approx} 2.5 \\times$ faster than the\nupsampler-based counterparts. Project website:\nhttps://universome.github.io/epigraf.\n", "date": "2022-06-21", "categories": []} {"id": "2212.00842", "title": "3D-LDM: Neural Implicit 3D Shape Generation with Latent Diffusion Models", "abstract": " Diffusion models have shown great promise for image generation, beating GANs\nin terms of generation diversity, with comparable image quality. However, their\napplication to 3D shapes has been limited to point or voxel representations\nthat can in practice not accurately represent a 3D surface. We propose a\ndiffusion model for neural implicit representations of 3D shapes that operates\nin the latent space of an auto-decoder. This allows us to generate diverse and\nhigh quality 3D surfaces. We additionally show that we can condition our model\non images or text to enable image-to-3D generation and text-to-3D generation\nusing CLIP embeddings. Furthermore, adding noise to the latent codes of\nexisting shapes allows us to explore shape variations.\n", "date": "2022-12-01", "categories": []} {"id": "2206.08361", "title": "Controllable 3D Face Synthesis with Conditional Generative Occupancy\n Fields", "abstract": " Capitalizing on the recent advances in image generation models, existing\ncontrollable face image synthesis methods are able to generate high-fidelity\nimages with some levels of controllability, e.g., controlling the shapes,\nexpressions, textures, and poses of the generated face images. However, these\nmethods focus on 2D image generative models, which are prone to producing\ninconsistent face images under large expression and pose changes. In this\npaper, we propose a new NeRF-based conditional 3D face synthesis framework,\nwhich enables 3D controllability over the generated face images by imposing\nexplicit 3D conditions from 3D face priors. At its core is a conditional\nGenerative Occupancy Field (cGOF) that effectively enforces the shape of the\ngenerated face to commit to a given 3D Morphable Model (3DMM) mesh. To achieve\naccurate control over fine-grained 3D face shapes of the synthesized image, we\nadditionally incorporate a 3D landmark loss as well as a volume warping loss\ninto our synthesis algorithm. Experiments validate the effectiveness of the\nproposed method, which is able to generate high-fidelity face images and shows\nmore precise 3D controllability than state-of-the-art 2D-based controllable\nface synthesis methods. Find code and demo at\nhttps://keqiangsun.github.io/projects/cgof.\n", "date": "2022-06-16", "categories": []} {"id": "2212.05321", "title": "HumanGen: Generating Human Radiance Fields with Explicit Priors", "abstract": " Recent years have witnessed the tremendous progress of 3D GANs for generating\nview-consistent radiance fields with photo-realism. Yet, high-quality\ngeneration of human radiance fields remains challenging, partially due to the\nlimited human-related priors adopted in existing methods. We present HumanGen,\na novel 3D human generation scheme with detailed geometry and\n$\\text{360}^{\\circ}$ realistic free-view rendering. It explicitly marries the\n3D human generation with various priors from the 2D generator and 3D\nreconstructor of humans through the design of \"anchor image\". We introduce a\nhybrid feature representation using the anchor image to bridge the latent space\nof HumanGen with the existing 2D generator. We then adopt a pronged design to\ndisentangle the generation of geometry and appearance. With the aid of the\nanchor image, we adapt a 3D reconstructor for fine-grained details synthesis\nand propose a two-stage blending scheme to boost appearance generation.\nExtensive experiments demonstrate our effectiveness for state-of-the-art 3D\nhuman generation regarding geometry details, texture quality, and free-view\nperformance. Notably, HumanGen can also incorporate various off-the-shelf 2D\nlatent editing methods, seamlessly lifting them into 3D.\n", "date": "2022-12-10", "categories": []} {"id": "2212.06135", "title": "Rodin: A Generative Model for Sculpting 3D Digital Avatars Using\n Diffusion", "abstract": " This paper presents a 3D generative model that uses diffusion models to\nautomatically generate 3D digital avatars represented as neural radiance\nfields. A significant challenge in generating such avatars is that the memory\nand processing costs in 3D are prohibitive for producing the rich details\nrequired for high-quality avatars. To tackle this problem we propose the\nroll-out diffusion network (Rodin), which represents a neural radiance field as\nmultiple 2D feature maps and rolls out these maps into a single 2D feature\nplane within which we perform 3D-aware diffusion. The Rodin model brings the\nmuch-needed computational efficiency while preserving the integrity of\ndiffusion in 3D by using 3D-aware convolution that attends to projected\nfeatures in the 2D feature plane according to their original relationship in\n3D. We also use latent conditioning to orchestrate the feature generation for\nglobal coherence, leading to high-fidelity avatars and enabling their semantic\nediting based on text prompts. Finally, we use hierarchical synthesis to\nfurther enhance details. The 3D avatars generated by our model compare\nfavorably with those produced by existing generative techniques. We can\ngenerate highly detailed avatars with realistic hairstyles and facial hair like\nbeards. We also demonstrate 3D avatar generation from image or text as well as\ntext-guided editability.\n", "date": "2022-12-12", "categories": []} {"id": "2212.04360", "title": "MIME: Human-Aware 3D Scene Generation", "abstract": " Generating realistic 3D worlds occupied by moving humans has many\napplications in games, architecture, and synthetic data creation. But\ngenerating such scenes is expensive and labor intensive. Recent work generates\nhuman poses and motions given a 3D scene. Here, we take the opposite approach\nand generate 3D indoor scenes given 3D human motion. Such motions can come from\narchival motion capture or from IMU sensors worn on the body, effectively\nturning human movement in a \"scanner\" of the 3D world. Intuitively, human\nmovement indicates the free-space in a room and human contact indicates\nsurfaces or objects that support activities such as sitting, lying or touching.\nWe propose MIME (Mining Interaction and Movement to infer 3D Environments),\nwhich is a generative model of indoor scenes that produces furniture layouts\nthat are consistent with the human movement. MIME uses an auto-regressive\ntransformer architecture that takes the already generated objects in the scene\nas well as the human motion as input, and outputs the next plausible object. To\ntrain MIME, we build a dataset by populating the 3D FRONT scene dataset with 3D\nhumans. Our experiments show that MIME produces more diverse and plausible 3D\nscenes than a recent generative scene method that does not know about human\nmovement. Code and data will be available for research at\nhttps://mime.is.tue.mpg.de.\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": "2212.01747", "title": "Fast Point Cloud Generation with Straight Flows", "abstract": " Diffusion models have emerged as a powerful tool for point cloud generation.\nA key component that drives the impressive performance for generating\nhigh-quality samples from noise is iteratively denoise for thousands of steps.\nWhile beneficial, the complexity of learning steps has limited its applications\nto many 3D real-world. To address this limitation, we propose Point Straight\nFlow (PSF), a model that exhibits impressive performance using one step. Our\nidea is based on the reformulation of the standard diffusion model, which\noptimizes the curvy learning trajectory into a straight path. Further, we\ndevelop a distillation strategy to shorten the straight path into one step\nwithout a performance loss, enabling applications to 3D real-world with latency\nconstraints. We perform evaluations on multiple 3D tasks and find that our PSF\nperforms comparably to the standard diffusion model, outperforming other\nefficient 3D point cloud generation methods. On real-world applications such as\npoint cloud completion and training-free text-guided generation in a\nlow-latency setup, PSF performs favorably.\n", "date": "2022-12-04", "categories": []} {"id": "2212.00774", "title": "Score Jacobian Chaining: Lifting Pretrained 2D Diffusion Models for 3D\n Generation", "abstract": " A diffusion model learns to predict a vector field of gradients. We propose\nto apply chain rule on the learned gradients, and back-propagate the score of a\ndiffusion model through the Jacobian of a differentiable renderer, which we\ninstantiate to be a voxel radiance field. This setup aggregates 2D scores at\nmultiple camera viewpoints into a 3D score, and repurposes a pretrained 2D\nmodel for 3D data generation. We identify a technical challenge of distribution\nmismatch that arises in this application, and propose a novel estimation\nmechanism to resolve it. We run our algorithm on several off-the-shelf\ndiffusion image generative models, including the recently released Stable\nDiffusion trained on the large-scale LAION dataset.\n", "date": "2022-12-01", "categories": []} {"id": "2211.16677", "title": "3D Neural Field Generation using Triplane Diffusion", "abstract": " Diffusion models have emerged as the state-of-the-art for image generation,\namong other tasks. Here, we present an efficient diffusion-based model for\n3D-aware generation of neural fields. Our approach pre-processes training data,\nsuch as ShapeNet meshes, by converting them to continuous occupancy fields and\nfactoring them into a set of axis-aligned triplane feature representations.\nThus, our 3D training scenes are all represented by 2D feature planes, and we\ncan directly train existing 2D diffusion models on these representations to\ngenerate 3D neural fields with high quality and diversity, outperforming\nalternative approaches to 3D-aware generation. Our approach requires essential\nmodifications to existing triplane factorization pipelines to make the\nresulting features easy to learn for the diffusion model. We demonstrate\nstate-of-the-art results on 3D generation on several object classes from\nShapeNet.\n", "date": "2022-11-30", "categories": []} {"id": "2211.16798", "title": "Dr.3D: Adapting 3D GANs to Artistic Drawings", "abstract": " While 3D GANs have recently demonstrated the high-quality synthesis of\nmulti-view consistent images and 3D shapes, they are mainly restricted to\nphoto-realistic human portraits. This paper aims to extend 3D GANs to a\ndifferent, but meaningful visual form: artistic portrait drawings. However,\nextending existing 3D GANs to drawings is challenging due to the inevitable\ngeometric ambiguity present in drawings. To tackle this, we present Dr.3D, a\nnovel adaptation approach that adapts an existing 3D GAN to artistic drawings.\nDr.3D is equipped with three novel components to handle the geometric\nambiguity: a deformation-aware 3D synthesis network, an alternating adaptation\nof pose estimation and image synthesis, and geometric priors. Experiments show\nthat our approach can successfully adapt 3D GANs to drawings and enable\nmulti-view consistent semantic editing of drawings.\n", "date": "2022-11-30", "categories": []} {"id": "2211.15662", "title": "High-fidelity 3D GAN Inversion by Pseudo-multi-view Optimization", "abstract": " We present a high-fidelity 3D generative adversarial network (GAN) inversion\nframework that can synthesize photo-realistic novel views while preserving\nspecific details of the input image. High-fidelity 3D GAN inversion is\ninherently challenging due to the geometry-texture trade-off in 3D inversion,\nwhere overfitting to a single view input image often damages the estimated\ngeometry during the latent optimization. To solve this challenge, we propose a\nnovel pipeline that builds on the pseudo-multi-view estimation with visibility\nanalysis. We keep the original textures for the visible parts and utilize\ngenerative priors for the occluded parts. Extensive experiments show that our\napproach achieves advantageous reconstruction and novel view synthesis quality\nover state-of-the-art methods, even for images with out-of-distribution\ntextures. The proposed pipeline also enables image attribute editing with the\ninverted latent code and 3D-aware texture modification. Our approach enables\nhigh-fidelity 3D rendering from a single image, which is promising for various\napplications of AI-generated 3D content.\n", "date": "2022-11-28", "categories": []} {"id": "2211.14589", "title": "AvatarGen: A 3D Generative Model for Animatable Human Avatars", "abstract": " Unsupervised generation of 3D-aware clothed humans with various appearances\nand controllable geometries is important for creating virtual human avatars and\nother AR/VR applications. Existing methods are either limited to rigid object\nmodeling, or not generative and thus unable to generate high-quality virtual\nhumans and animate them. In this work, we propose AvatarGen, the first method\nthat enables not only geometry-aware clothed human synthesis with high-fidelity\nappearances but also disentangled human animation controllability, while only\nrequiring 2D images for training. Specifically, we decompose the generative 3D\nhuman synthesis into pose-guided mapping and canonical representation with\npredefined human pose and shape, such that the canonical representation can be\nexplicitly driven to different poses and shapes with the guidance of a 3D\nparametric human model SMPL. AvatarGen further introduces a deformation network\nto learn non-rigid deformations for modeling fine-grained geometric details and\npose-dependent dynamics. To improve the geometry quality of the generated human\navatars, it leverages the signed distance field as geometric proxy, which\nallows more direct regularization from the 3D geometric priors of SMPL.\nBenefiting from these designs, our method can generate animatable 3D human\navatars with high-quality appearance and geometry modeling, significantly\noutperforming previous 3D GANs. Furthermore, it is competent for many\napplications, e.g., single-view reconstruction, re-animation, and text-guided\nsynthesis/editing. Code and pre-trained model will be available at\nhttp://jeff95.me/projects/avatargen.html.\n", "date": "2022-11-26", "categories": []} {"id": "2211.14902", "title": "3inGAN: Learning a 3D Generative Model from Images of a Self-similar\n Scene", "abstract": " We introduce 3inGAN, an unconditional 3D generative model trained from 2D\nimages of a single self-similar 3D scene. Such a model can be used to produce\n3D \"remixes\" of a given scene, by mapping spatial latent codes into a 3D\nvolumetric representation, which can subsequently be rendered from arbitrary\nviews using physically based volume rendering. By construction, the generated\nscenes remain view-consistent across arbitrary camera configurations, without\nany flickering or spatio-temporal artifacts. During training, we employ a\ncombination of 2D, obtained through differentiable volume tracing, and 3D\nGenerative Adversarial Network (GAN) losses, across multiple scales, enforcing\nrealism on both its 3D structure and the 2D renderings. We show results on\nsemi-stochastic scenes of varying scale and complexity, obtained from real and\nsynthetic sources. We demonstrate, for the first time, the feasibility of\nlearning plausible view-consistent 3D scene variations from a single exemplar\nscene and provide qualitative and quantitative comparisons against recent\nrelated methods.\n", "date": "2022-11-27", "categories": []} {"id": "2211.09770", "title": "3DLatNav: Navigating Generative Latent Spaces for Semantic-Aware 3D\n Object Manipulation", "abstract": " 3D generative models have been recently successful in generating realistic 3D\nobjects in the form of point clouds. However, most models do not offer\ncontrollability to manipulate the shape semantics of component object parts\nwithout extensive semantic attribute labels or other reference point clouds.\nMoreover, beyond the ability to perform simple latent vector arithmetic or\ninterpolations, there is a lack of understanding of how part-level semantics of\n3D shapes are encoded in their corresponding generative latent spaces. In this\npaper, we propose 3DLatNav; a novel approach to navigating pretrained\ngenerative latent spaces to enable controlled part-level semantic manipulation\nof 3D objects. First, we propose a part-level weakly-supervised shape semantics\nidentification mechanism using latent representations of 3D shapes. Then, we\ntransfer that knowledge to a pretrained 3D object generative latent space to\nunravel disentangled embeddings to represent different shape semantics of\ncomponent parts of an object in the form of linear subspaces, despite the\nunavailability of part-level labels during the training. Finally, we utilize\nthose identified subspaces to show that controllable 3D object part\nmanipulation can be achieved by applying the proposed framework to any\npretrained 3D generative model. With two novel quantitative metrics to evaluate\nthe consistency and localization accuracy of part-level manipulations, we show\nthat 3DLatNav outperforms existing unsupervised latent disentanglement methods\nin identifying latent directions that encode part-level shape semantics of 3D\nobjects. With multiple ablation studies and testing on state-of-the-art\ngenerative models, we show that 3DLatNav can implement controlled part-level\nsemantic manipulations on an input point cloud while preserving other features\nand the realistic nature of the object.\n", "date": "2022-11-17", "categories": []} {"id": "2211.08702", "title": "PointInverter: Point Cloud Reconstruction and Editing via a Generative\n Model with Shape Priors", "abstract": " In this paper, we propose a new method for mapping a 3D point cloud to the\nlatent space of a 3D generative adversarial network. Our generative model for\n3D point clouds is based on SP-GAN, a state-of-the-art sphere-guided 3D point\ncloud generator. We derive an efficient way to encode an input 3D point cloud\nto the latent space of the SP-GAN. Our point cloud encoder can resolve the\npoint ordering issue during inversion, and thus can determine the\ncorrespondences between points in the generated 3D point cloud and those in the\ncanonical sphere used by the generator. We show that our method outperforms\nprevious GAN inversion methods for 3D point clouds, achieving state-of-the-art\nresults both quantitatively and qualitatively. Our code is available at\nhttps://github.com/hkust-vgd/point_inverter.\n", "date": "2022-11-16", "categories": []} {"id": "2211.07818", "title": "AgileAvatar: Stylized 3D Avatar Creation via Cascaded Domain Bridging", "abstract": " Stylized 3D avatars have become increasingly prominent in our modern life.\nCreating these avatars manually usually involves laborious selection and\nadjustment of continuous and discrete parameters and is time-consuming for\naverage users. Self-supervised approaches to automatically create 3D avatars\nfrom user selfies promise high quality with little annotation cost but fall\nshort in application to stylized avatars due to a large style domain gap. We\npropose a novel self-supervised learning framework to create high-quality\nstylized 3D avatars with a mix of continuous and discrete parameters. Our\ncascaded domain bridging framework first leverages a modified portrait\nstylization approach to translate input selfies into stylized avatar renderings\nas the targets for desired 3D avatars. Next, we find the best parameters of the\navatars to match the stylized avatar renderings through a differentiable\nimitator we train to mimic the avatar graphics engine. To ensure we can\neffectively optimize the discrete parameters, we adopt a cascaded\nrelaxation-and-search pipeline. We use a human preference study to evaluate how\nwell our method preserves user identity compared to previous work as well as\nmanual creation. Our results achieve much higher preference scores than\nprevious work and close to those of manual creation. We also provide an\nablation study to justify the design choices in our pipeline.\n", "date": "2022-11-15", "categories": []} {"id": "2211.06583", "title": "3D-Aware Encoding for Style-based Neural Radiance Fields", "abstract": " We tackle the task of NeRF inversion for style-based neural radiance fields,\n(e.g., StyleNeRF). In the task, we aim to learn an inversion function to\nproject an input image to the latent space of a NeRF generator and then\nsynthesize novel views of the original image based on the latent code. Compared\nwith GAN inversion for 2D generative models, NeRF inversion not only needs to\n1) preserve the identity of the input image, but also 2) ensure 3D consistency\nin generated novel views. This requires the latent code obtained from the\nsingle-view image to be invariant across multiple views. To address this new\nchallenge, we propose a two-stage encoder for style-based NeRF inversion. In\nthe first stage, we introduce a base encoder that converts the input image to a\nlatent code. To ensure the latent code is view-invariant and is able to\nsynthesize 3D consistent novel view images, we utilize identity contrastive\nlearning to train the base encoder. Second, to better preserve the identity of\nthe input image, we introduce a refining encoder to refine the latent code and\nadd finer details to the output image. Importantly note that the novelty of\nthis model lies in the design of its first-stage encoder which produces the\nclosest latent code lying on the latent manifold and thus the refinement in the\nsecond stage would be close to the NeRF manifold. Through extensive\nexperiments, we demonstrate that our proposed two-stage encoder qualitatively\nand quantitatively exhibits superiority over the existing encoders for\ninversion in both image reconstruction and novel-view rendering.\n", "date": "2022-11-12", "categories": []} {"id": "2211.07600", "title": "Latent-NeRF for Shape-Guided Generation of 3D Shapes and Textures", "abstract": " Text-guided image generation has progressed rapidly in recent years,\ninspiring major breakthroughs in text-guided shape generation. Recently, it has\nbeen shown that using score distillation, one can successfully text-guide a\nNeRF model to generate a 3D object. We adapt the score distillation to the\npublicly available, and computationally efficient, Latent Diffusion Models,\nwhich apply the entire diffusion process in a compact latent space of a\npretrained autoencoder. As NeRFs operate in image space, a naive solution for\nguiding them with latent score distillation would require encoding to the\nlatent space at each guidance step. Instead, we propose to bring the NeRF to\nthe latent space, resulting in a Latent-NeRF. Analyzing our Latent-NeRF, we\nshow that while Text-to-3D models can generate impressive results, they are\ninherently unconstrained and may lack the ability to guide or enforce a\nspecific 3D structure. To assist and direct the 3D generation, we propose to\nguide our Latent-NeRF using a Sketch-Shape: an abstract geometry that defines\nthe coarse structure of the desired object. Then, we present means to integrate\nsuch a constraint directly into a Latent-NeRF. This unique combination of text\nand shape guidance allows for increased control over the generation process. We\nalso show that latent score distillation can be successfully applied directly\non 3D meshes. This allows for generating high-quality textures on a given\ngeometry. Our experiments validate the power of our different forms of guidance\nand the efficiency of using latent rendering. Implementation is available at\nhttps://github.com/eladrich/latent-nerf\n", "date": "2022-11-14", "categories": []} {"id": "2206.07695", "title": "VoxGRAF: Fast 3D-Aware Image Synthesis with Sparse Voxel Grids", "abstract": " State-of-the-art 3D-aware generative models rely on coordinate-based MLPs to\nparameterize 3D radiance fields. While demonstrating impressive results,\nquerying an MLP for every sample along each ray leads to slow rendering.\nTherefore, existing approaches often render low-resolution feature maps and\nprocess them with an upsampling network to obtain the final image. Albeit\nefficient, neural rendering often entangles viewpoint and content such that\nchanging the camera pose results in unwanted changes of geometry or appearance.\nMotivated by recent results in voxel-based novel view synthesis, we investigate\nthe utility of sparse voxel grid representations for fast and 3D-consistent\ngenerative modeling in this paper. Our results demonstrate that monolithic MLPs\ncan indeed be replaced by 3D convolutions when combining sparse voxel grids\nwith progressive growing, free space pruning and appropriate regularization. To\nobtain a compact representation of the scene and allow for scaling to higher\nvoxel resolutions, our model disentangles the foreground object (modeled in 3D)\nfrom the background (modeled in 2D). In contrast to existing approaches, our\nmethod requires only a single forward pass to generate a full 3D scene. It\nhence allows for efficient rendering from arbitrary viewpoints while yielding\n3D consistent results with high visual fidelity.\n", "date": "2022-06-15", "categories": []} {"id": "2210.17344", "title": "gCoRF: Generative Compositional Radiance Fields", "abstract": " 3D generative models of objects enable photorealistic image synthesis with 3D\ncontrol. Existing methods model the scene as a global scene representation,\nignoring the compositional aspect of the scene. Compositional reasoning can\nenable a wide variety of editing applications, in addition to enabling\ngeneralizable 3D reasoning. In this paper, we present a compositional\ngenerative model, where each semantic part of the object is represented as an\nindependent 3D representation learned from only in-the-wild 2D data. We start\nwith a global generative model (GAN) and learn to decompose it into different\nsemantic parts using supervision from 2D segmentation masks. We then learn to\ncomposite independently sampled parts in order to create coherent global\nscenes. Different parts can be independently sampled while keeping the rest of\nthe object fixed. We evaluate our method on a wide variety of objects and parts\nand demonstrate editing applications.\n", "date": "2022-10-31", "categories": []} {"id": "2209.05434", "title": "3DFaceShop: Explicitly Controllable 3D-Aware Portrait Generation", "abstract": " In contrast to the traditional avatar creation pipeline which is a costly\nprocess, contemporary generative approaches directly learn the data\ndistribution from photographs. While plenty of works extend unconditional\ngenerative models and achieve some levels of controllability, it is still\nchallenging to ensure multi-view consistency, especially in large poses. In\nthis work, we propose a network that generates 3D-aware portraits while being\ncontrollable according to semantic parameters regarding pose, identity,\nexpression and illumination. Our network uses neural scene representation to\nmodel 3D-aware portraits, whose generation is guided by a parametric face model\nthat supports explicit control. While the latent disentanglement can be further\nenhanced by contrasting images with partially different attributes, there still\nexists noticeable inconsistency in non-face areas, e.g., hair and background,\nwhen animating expressions. Wesolve this by proposing a volume blending\nstrategy in which we form a composite output by blending dynamic and static\nareas, with two parts segmented from the jointly learned semantic field. Our\nmethod outperforms prior arts in extensive experiments, producing realistic\nportraits with vivid expression in natural lighting when viewed from free\nviewpoints. It also demonstrates generalization ability to real images as well\nas out-of-domain data, showing great promise in real applications.\n", "date": "2022-09-12", "categories": []} {"id": "2210.01202", "title": "SinGRAV: Learning a Generative Radiance Volume from a Single Natural\n Scene", "abstract": " We present a 3D generative model for general natural scenes. Lacking\nnecessary volumes of 3D data characterizing the target scene, we propose to\nlearn from a single scene. Our key insight is that a natural scene often\ncontains multiple constituents whose geometry, texture, and spatial\narrangements follow some clear patterns, but still exhibit rich variations over\ndifferent regions within the same scene. This suggests localizing the learning\nof a generative model on substantial local regions. Hence, we exploit a\nmulti-scale convolutional network, which possesses the spatial locality bias in\nnature, to learn from the statistics of local regions at multiple scales within\na single scene. In contrast to existing methods, our learning setup bypasses\nthe need to collect data from many homogeneous 3D scenes for learning common\nfeatures. We coin our method SinGRAV, for learning a Generative RAdiance Volume\nfrom a Single natural scene. We demonstrate the ability of SinGRAV in\ngenerating plausible and diverse variations from a single scene, the merits of\nSinGRAV over state-of-the-art generative neural scene methods, as well as the\nversatility of SinGRAV by its use in a variety of applications, spanning 3D\nscene editing, composition, and animation. Code and data will be released to\nfacilitate further research.\n", "date": "2022-10-03", "categories": []} {"id": "2210.06978", "title": "LION: Latent Point Diffusion Models for 3D Shape Generation", "abstract": " Denoising diffusion models (DDMs) have shown promising results in 3D point\ncloud synthesis. To advance 3D DDMs and make them useful for digital artists,\nwe require (i) high generation quality, (ii) flexibility for manipulation and\napplications such as conditional synthesis and shape interpolation, and (iii)\nthe ability to output smooth surfaces or meshes. To this end, we introduce the\nhierarchical Latent Point Diffusion Model (LION) for 3D shape generation. LION\nis set up as a variational autoencoder (VAE) with a hierarchical latent space\nthat combines a global shape latent representation with a point-structured\nlatent space. For generation, we train two hierarchical DDMs in these latent\nspaces. The hierarchical VAE approach boosts performance compared to DDMs that\noperate on point clouds directly, while the point-structured latents are still\nideally suited for DDM-based modeling. Experimentally, LION achieves\nstate-of-the-art generation performance on multiple ShapeNet benchmarks.\nFurthermore, our VAE framework allows us to easily use LION for different\nrelevant tasks: LION excels at multimodal shape denoising and voxel-conditioned\nsynthesis, and it can be adapted for text- and image-driven 3D generation. We\nalso demonstrate shape autoencoding and latent shape interpolation, and we\naugment LION with modern surface reconstruction techniques to generate smooth\n3D meshes. We hope that LION provides a powerful tool for artists working with\n3D shapes due to its high-quality generation, flexibility, and surface\nreconstruction. Project page and code: https://nv-tlabs.github.io/LION.\n", "date": "2022-10-12", "categories": []} {"id": "2210.05735", "title": "TetGAN: A Convolutional Neural Network for Tetrahedral Mesh Generation", "abstract": " We present TetGAN, a convolutional neural network designed to generate\ntetrahedral meshes. We represent shapes using an irregular tetrahedral grid\nwhich encodes an occupancy and displacement field. Our formulation enables\ndefining tetrahedral convolution, pooling, and upsampling operations to\nsynthesize explicit mesh connectivity with variable topological genus. The\nproposed neural network layers learn deep features over each tetrahedron and\nlearn to extract patterns within spatial regions across multiple scales. We\nillustrate the capabilities of our technique to encode tetrahedral meshes into\na semantically meaningful latent-space which can be used for shape editing and\nsynthesis. Our project page is at https://threedle.github.io/tetGAN/.\n", "date": "2022-10-11", "categories": []} {"id": "2210.05825", "title": "Controllable Radiance Fields for Dynamic Face Synthesis", "abstract": " Recent work on 3D-aware image synthesis has achieved compelling results using\nadvances in neural rendering. However, 3D-aware synthesis of face dynamics\nhasn't received much attention. Here, we study how to explicitly control\ngenerative model synthesis of face dynamics exhibiting non-rigid motion (e.g.,\nfacial expression change), while simultaneously ensuring 3D-awareness. For this\nwe propose a Controllable Radiance Field (CoRF): 1) Motion control is achieved\nby embedding motion features within the layered latent motion space of a\nstyle-based generator; 2) To ensure consistency of background, motion features\nand subject-specific attributes such as lighting, texture, shapes, albedo, and\nidentity, a face parsing net, a head regressor and an identity encoder are\nincorporated. On head image/video data we show that CoRFs are 3D-aware while\nenabling editing of identity, viewing directions, and motion.\n", "date": "2022-10-11", "categories": []} {"id": "2210.06465", "title": "AniFaceGAN: Animatable 3D-Aware Face Image Generation for Video Avatars", "abstract": " Although 2D generative models have made great progress in face image\ngeneration and animation, they often suffer from undesirable artifacts such as\n3D inconsistency when rendering images from different camera viewpoints. This\nprevents them from synthesizing video animations indistinguishable from real\nones. Recently, 3D-aware GANs extend 2D GANs for explicit disentanglement of\ncamera pose by leveraging 3D scene representations. These methods can well\npreserve the 3D consistency of the generated images across different views, yet\nthey cannot achieve fine-grained control over other attributes, among which\nfacial expression control is arguably the most useful and desirable for face\nanimation. In this paper, we propose an animatable 3D-aware GAN for multiview\nconsistent face animation generation. The key idea is to decompose the 3D\nrepresentation of the 3D-aware GAN into a template field and a deformation\nfield, where the former represents different identities with a canonical\nexpression, and the latter characterizes expression variations of each\nidentity. To achieve meaningful control over facial expressions via\ndeformation, we propose a 3D-level imitative learning scheme between the\ngenerator and a parametric 3D face model during adversarial training of the\n3D-aware GAN. This helps our method achieve high-quality animatable face image\ngeneration with strong visual 3D consistency, even though trained with only\nunstructured 2D images. Extensive experiments demonstrate our superior\nperformance over prior works. Project page:\nhttps://yuewuhkust.github.io/AniFaceGAN\n", "date": "2022-10-12", "categories": []} {"id": "2210.04072", "title": "Flow-based GAN for 3D Point Cloud Generation from a Single Image", "abstract": " Generating a 3D point cloud from a single 2D image is of great importance for\n3D scene understanding applications. To reconstruct the whole 3D shape of the\nobject shown in the image, the existing deep learning based approaches use\neither explicit or implicit generative modeling of point clouds, which,\nhowever, suffer from limited quality. In this work, we aim to alleviate this\nissue by introducing a hybrid explicit-implicit generative modeling scheme,\nwhich inherits the flow-based explicit generative models for sampling point\nclouds with arbitrary resolutions while improving the detailed 3D structures of\npoint clouds by leveraging the implicit generative adversarial networks (GANs).\nWe evaluate on the large-scale synthetic dataset ShapeNet, with the\nexperimental results demonstrating the superior performance of the proposed\nmethod. In addition, the generalization ability of our method is demonstrated\nby performing on cross-category synthetic images as well as by testing on real\nimages from PASCAL3D+ dataset.\n", "date": "2022-10-08", "categories": []} {"id": "2210.04888", "title": "EVA3D: Compositional 3D Human Generation from 2D Image Collections", "abstract": " Inverse graphics aims to recover 3D models from 2D observations. Utilizing\ndifferentiable rendering, recent 3D-aware generative models have shown\nimpressive results of rigid object generation using 2D images. However, it\nremains challenging to generate articulated objects, like human bodies, due to\ntheir complexity and diversity in poses and appearances. In this work, we\npropose, EVA3D, an unconditional 3D human generative model learned from 2D\nimage collections only. EVA3D can sample 3D humans with detailed geometry and\nrender high-quality images (up to 512x256) without bells and whistles (e.g.\nsuper resolution). At the core of EVA3D is a compositional human NeRF\nrepresentation, which divides the human body into local parts. Each part is\nrepresented by an individual volume. This compositional representation enables\n1) inherent human priors, 2) adaptive allocation of network parameters, 3)\nefficient training and rendering. Moreover, to accommodate for the\ncharacteristics of sparse 2D human image collections (e.g. imbalanced pose\ndistribution), we propose a pose-guided sampling strategy for better GAN\nlearning. Extensive experiments validate that EVA3D achieves state-of-the-art\n3D human generation performance regarding both geometry and texture quality.\nNotably, EVA3D demonstrates great potential and scalability to\n\"inverse-graphics\" diverse human bodies with a clean framework.\n", "date": "2022-10-10", "categories": []} {"id": "2210.03158", "title": "Neural Volumetric Mesh Generator", "abstract": " Deep generative models have shown success in generating 3D shapes with\ndifferent representations. In this work, we propose Neural Volumetric Mesh\nGenerator(NVMG) which can generate novel and high-quality volumetric meshes.\nUnlike the previous 3D generative model for point cloud, voxel, and implicit\nsurface, the volumetric mesh representation is a ready-to-use representation in\nindustry with details on both the surface and interior. Generating this such\nhighly-structured data thus brings a significant challenge. We first propose a\ndiffusion-based generative model to tackle this problem by generating voxelized\nshapes with close-to-reality outlines and structures. We can simply obtain a\ntetrahedral mesh as a template with the voxelized shape. Further, we use a\nvoxel-conditional neural network to predict the smooth implicit surface\nconditioned on the voxels, and progressively project the tetrahedral mesh to\nthe predicted surface under regularizations. The regularization terms are\ncarefully designed so that they can (1) get rid of the defects like flipping\nand high distortion; (2) force the regularity of the interior and surface\nstructure during the deformation procedure for a high-quality final mesh. As\nshown in the experiments, our pipeline can generate high-quality artifact-free\nvolumetric and surface meshes from random noise or a reference image without\nany post-processing. Compared with the state-of-the-art voxel-to-mesh\ndeformation method, we show more robustness and better performance when taking\ngenerated voxels as input.\n", "date": "2022-10-06", "categories": []} {"id": "2210.03007", "title": "XDGAN: Multi-Modal 3D Shape Generation in 2D Space", "abstract": " Generative models for 2D images has recently seen tremendous progress in\nquality, resolution and speed as a result of the efficiency of 2D convolutional\narchitectures. However it is difficult to extend this progress into the 3D\ndomain since most current 3D representations rely on custom network components.\nThis paper addresses a central question: Is it possible to directly leverage 2D\nimage generative models to generate 3D shapes instead? To answer this, we\npropose XDGAN, an effective and fast method for applying 2D image GAN\narchitectures to the generation of 3D object geometry combined with additional\nsurface attributes, like color textures and normals. Specifically, we propose a\nnovel method to convert 3D shapes into compact 1-channel geometry images and\nleverage StyleGAN3 and image-to-image translation networks to generate 3D\nobjects in 2D space. The generated geometry images are quick to convert to 3D\nmeshes, enabling real-time 3D object synthesis, visualization and interactive\nediting. Moreover, the use of standard 2D architectures can help bring more 2D\nadvances into the 3D realm. We show both quantitatively and qualitatively that\nour method is highly effective at various tasks such as 3D shape generation,\nsingle view reconstruction and shape manipulation, while being significantly\nfaster and more flexible compared to recent 3D generative models.\n", "date": "2022-10-06", "categories": []} {"id": "2209.15172", "title": "Understanding Pure CLIP Guidance for Voxel Grid NeRF Models", "abstract": " We explore the task of text to 3D object generation using CLIP. Specifically,\nwe use CLIP for guidance without access to any datasets, a setting we refer to\nas pure CLIP guidance. While prior work has adopted this setting, there is no\nsystematic study of mechanics for preventing adversarial generations within\nCLIP. We illustrate how different image-based augmentations prevent the\nadversarial generation problem, and how the generated results are impacted. We\ntest different CLIP model architectures and show that ensembling different\nmodels for guidance can prevent adversarial generations within bigger models\nand generate sharper results. Furthermore, we implement an implicit voxel grid\nmodel to show how neural networks provide an additional layer of\nregularization, resulting in better geometrical structure and coherency of\ngenerated objects. Compared to prior work, we achieve more coherent results\nwith higher memory efficiency and faster training speeds.\n", "date": "2022-09-30", "categories": []} {"id": "2209.15637", "title": "Improving 3D-aware Image Synthesis with A Geometry-aware Discriminator", "abstract": " 3D-aware image synthesis aims at learning a generative model that can render\nphoto-realistic 2D images while capturing decent underlying 3D shapes. A\npopular solution is to adopt the generative adversarial network (GAN) and\nreplace the generator with a 3D renderer, where volume rendering with neural\nradiance field (NeRF) is commonly used. Despite the advancement of synthesis\nquality, existing methods fail to obtain moderate 3D shapes. We argue that,\nconsidering the two-player game in the formulation of GANs, only making the\ngenerator 3D-aware is not enough. In other words, displacing the generative\nmechanism only offers the capability, but not the guarantee, of producing\n3D-aware images, because the supervision of the generator primarily comes from\nthe discriminator. To address this issue, we propose GeoD through learning a\ngeometry-aware discriminator to improve 3D-aware GANs. Concretely, besides\ndifferentiating real and fake samples from the 2D image space, the\ndiscriminator is additionally asked to derive the geometry information from the\ninputs, which is then applied as the guidance of the generator. Such a simple\nyet effective design facilitates learning substantially more accurate 3D\nshapes. Extensive experiments on various generator architectures and training\ndatasets verify the superiority of GeoD over state-of-the-art alternatives.\nMoreover, our approach is registered as a general framework such that a more\ncapable discriminator (i.e., with a third task of novel view synthesis beyond\ndomain classification and geometry extraction) can further assist the generator\nwith a better multi-view consistency.\n", "date": "2022-09-30", "categories": []} {"id": "2209.14988", "title": "DreamFusion: Text-to-3D using 2D Diffusion", "abstract": " Recent breakthroughs in text-to-image synthesis have been driven by diffusion\nmodels trained on billions of image-text pairs. Adapting this approach to 3D\nsynthesis would require large-scale datasets of labeled 3D data and efficient\narchitectures for denoising 3D data, neither of which currently exist. In this\nwork, we circumvent these limitations by using a pretrained 2D text-to-image\ndiffusion model to perform text-to-3D synthesis. We introduce a loss based on\nprobability density distillation that enables the use of a 2D diffusion model\nas a prior for optimization of a parametric image generator. Using this loss in\na DeepDream-like procedure, we optimize a randomly-initialized 3D model (a\nNeural Radiance Field, or NeRF) via gradient descent such that its 2D\nrenderings from random angles achieve a low loss. The resulting 3D model of the\ngiven text can be viewed from any angle, relit by arbitrary illumination, or\ncomposited into any 3D environment. Our approach requires no 3D training data\nand no modifications to the image diffusion model, demonstrating the\neffectiveness of pretrained image diffusion models as priors.\n", "date": "2022-09-29", "categories": []} {"id": "2209.11163", "title": "GET3D: A Generative Model of High Quality 3D Textured Shapes Learned\n from Images", "abstract": " As several industries are moving towards modeling massive 3D virtual worlds,\nthe need for content creation tools that can scale in terms of the quantity,\nquality, and diversity of 3D content is becoming evident. In our work, we aim\nto train performant 3D generative models that synthesize textured meshes which\ncan be directly consumed by 3D rendering engines, thus immediately usable in\ndownstream applications. Prior works on 3D generative modeling either lack\ngeometric details, are limited in the mesh topology they can produce, typically\ndo not support textures, or utilize neural renderers in the synthesis process,\nwhich makes their use in common 3D software non-trivial. In this work, we\nintroduce GET3D, a Generative model that directly generates Explicit Textured\n3D meshes with complex topology, rich geometric details, and high-fidelity\ntextures. We bridge recent success in the differentiable surface modeling,\ndifferentiable rendering as well as 2D Generative Adversarial Networks to train\nour model from 2D image collections. GET3D is able to generate high-quality 3D\ntextured meshes, ranging from cars, chairs, animals, motorbikes and human\ncharacters to buildings, achieving significant improvements over previous\nmethods.\n", "date": "2022-09-22", "categories": []} {"id": "2209.08725", "title": "Neural Wavelet-domain Diffusion for 3D Shape Generation", "abstract": " This paper presents a new approach for 3D shape generation, enabling direct\ngenerative modeling on a continuous implicit representation in wavelet domain.\nSpecifically, we propose a compact wavelet representation with a pair of coarse\nand detail coefficient volumes to implicitly represent 3D shapes via truncated\nsigned distance functions and multi-scale biorthogonal wavelets, and formulate\na pair of neural networks: a generator based on the diffusion model to produce\ndiverse shapes in the form of coarse coefficient volumes; and a detail\npredictor to further produce compatible detail coefficient volumes for\nenriching the generated shapes with fine structures and details. Both\nquantitative and qualitative experimental results manifest the superiority of\nour approach in generating diverse and high-quality shapes with complex\ntopology and structures, clean surfaces, and fine details, exceeding the 3D\ngeneration capabilities of the state-of-the-art models.\n", "date": "2022-09-19", "categories": []} {"id": "2209.05710", "title": "MDM: Molecular Diffusion Model for 3D Molecule Generation", "abstract": " Molecule generation, especially generating 3D molecular geometries from\nscratch (i.e., 3D \\textit{de novo} generation), has become a fundamental task\nin drug designs. Existing diffusion-based 3D molecule generation methods could\nsuffer from unsatisfactory performances, especially when generating large\nmolecules. At the same time, the generated molecules lack enough diversity.\nThis paper proposes a novel diffusion model to address those two challenges.\n First, interatomic relations are not in molecules' 3D point cloud\nrepresentations. Thus, it is difficult for existing generative models to\ncapture the potential interatomic forces and abundant local constraints. To\ntackle this challenge, we propose to augment the potential interatomic forces\nand further involve dual equivariant encoders to encode interatomic forces of\ndifferent strengths. Second, existing diffusion-based models essentially shift\nelements in geometry along the gradient of data density. Such a process lacks\nenough exploration in the intermediate steps of the Langevin dynamics. To\naddress this issue, we introduce a distributional controlling variable in each\ndiffusion/reverse step to enforce thorough explorations and further improve\ngeneration diversity.\n Extensive experiments on multiple benchmarks demonstrate that the proposed\nmodel significantly outperforms existing methods for both unconditional and\nconditional generation tasks. We also conduct case studies to help understand\nthe physicochemical properties of the generated molecules.\n", "date": "2022-09-13", "categories": []} {"id": "2209.04183", "title": "Generative Deformable Radiance Fields for Disentangled Image Synthesis\n of Topology-Varying Objects", "abstract": " 3D-aware generative models have demonstrated their superb performance to\ngenerate 3D neural radiance fields (NeRF) from a collection of monocular 2D\nimages even for topology-varying object categories. However, these methods\nstill lack the capability to separately control the shape and appearance of the\nobjects in the generated radiance fields. In this paper, we propose a\ngenerative model for synthesizing radiance fields of topology-varying objects\nwith disentangled shape and appearance variations. Our method generates\ndeformable radiance fields, which builds the dense correspondence between the\ndensity fields of the objects and encodes their appearances in a shared\ntemplate field. Our disentanglement is achieved in an unsupervised manner\nwithout introducing extra labels to previous 3D-aware GAN training. We also\ndevelop an effective image inversion scheme for reconstructing the radiance\nfield of an object in a real monocular image and manipulating its shape and\nappearance. Experiments show that our method can successfully learn the\ngenerative model from unstructured monocular images and well disentangle the\nshape and appearance for objects (e.g., chairs) with large topological\nvariance. The model trained on synthetic data can faithfully reconstruct the\nreal object in a given single image and achieve high-quality texture and shape\nediting results.\n", "date": "2022-09-09", "categories": []} {"id": "2208.14263", "title": "Controllable 3D Generative Adversarial Face Model via Disentangling\n Shape and Appearance", "abstract": " 3D face modeling has been an active area of research in computer vision and\ncomputer graphics, fueling applications ranging from facial expression transfer\nin virtual avatars to synthetic data generation. Existing 3D deep learning\ngenerative models (e.g., VAE, GANs) allow generating compact face\nrepresentations (both shape and texture) that can model non-linearities in the\nshape and appearance space (e.g., scatter effects, specularities, etc.).\nHowever, they lack the capability to control the generation of subtle\nexpressions. This paper proposes a new 3D face generative model that can\ndecouple identity and expression and provides granular control over\nexpressions. In particular, we propose using a pair of supervised auto-encoder\nand generative adversarial networks to produce high-quality 3D faces, both in\nterms of appearance and shape. Experimental results in the generation of 3D\nfaces learned with holistic expression labels, or Action Unit labels, show how\nwe can decouple identity and expression; gaining fine-control over expressions\nwhile preserving identity.\n", "date": "2022-08-30", "categories": []} {"id": "2208.00561", "title": "AvatarGen: a 3D Generative Model for Animatable Human Avatars", "abstract": " Unsupervised generation of clothed virtual humans with various appearance and\nanimatable poses is important for creating 3D human avatars and other AR/VR\napplications. Existing methods are either limited to rigid object modeling, or\nnot generative and thus unable to synthesize high-quality virtual humans and\nanimate them. In this work, we propose AvatarGen, the first method that enables\nnot only non-rigid human generation with diverse appearance but also full\ncontrol over poses and viewpoints, while only requiring 2D images for training.\nSpecifically, it extends the recent 3D GANs to clothed human generation by\nutilizing a coarse human body model as a proxy to warp the observation space\ninto a standard avatar under a canonical space. To model non-rigid dynamics, it\nintroduces a deformation network to learn pose-dependent deformations in the\ncanonical space. To improve geometry quality of the generated human avatars, it\nleverages signed distance field as geometric representation, which allows more\ndirect regularization from the body model on the geometry learning. Benefiting\nfrom these designs, our method can generate animatable human avatars with\nhigh-quality appearance and geometry modeling, significantly outperforming\nprevious 3D GANs. Furthermore, it is competent for many applications, e.g.,\nsingle-view reconstruction, reanimation, and text-guided synthesis. Code and\npre-trained model will be available.\n", "date": "2022-08-01", "categories": []} {"id": "2207.13751", "title": "GAUDI: A Neural Architect for Immersive 3D Scene Generation", "abstract": " We introduce GAUDI, a generative model capable of capturing the distribution\nof complex and realistic 3D scenes that can be rendered immersively from a\nmoving camera. We tackle this challenging problem with a scalable yet powerful\napproach, where we first optimize a latent representation that disentangles\nradiance fields and camera poses. This latent representation is then used to\nlearn a generative model that enables both unconditional and conditional\ngeneration of 3D scenes. Our model generalizes previous works that focus on\nsingle objects by removing the assumption that the camera pose distribution can\nbe shared across samples. We show that GAUDI obtains state-of-the-art\nperformance in the unconditional generative setting across multiple datasets\nand allows for conditional generation of 3D scenes given conditioning variables\nlike sparse image observations or text that describes the scene.\n", "date": "2022-07-27", "categories": []} {"id": "2207.10257", "title": "Injecting 3D Perception of Controllable NeRF-GAN into StyleGAN for\n Editable Portrait Image Synthesis", "abstract": " Over the years, 2D GANs have achieved great successes in photorealistic\nportrait generation. However, they lack 3D understanding in the generation\nprocess, thus they suffer from multi-view inconsistency problem. To alleviate\nthe issue, many 3D-aware GANs have been proposed and shown notable results, but\n3D GANs struggle with editing semantic attributes. The controllability and\ninterpretability of 3D GANs have not been much explored. In this work, we\npropose two solutions to overcome these weaknesses of 2D GANs and 3D-aware\nGANs. We first introduce a novel 3D-aware GAN, SURF-GAN, which is capable of\ndiscovering semantic attributes during training and controlling them in an\nunsupervised manner. After that, we inject the prior of SURF-GAN into StyleGAN\nto obtain a high-fidelity 3D-controllable generator. Unlike existing\nlatent-based methods allowing implicit pose control, the proposed\n3D-controllable StyleGAN enables explicit pose control over portrait\ngeneration. This distillation allows direct compatibility between 3D control\nand many StyleGAN-based techniques (e.g., inversion and stylization), and also\nbrings an advantage in terms of computational resources. Our codes are\navailable at https://github.com/jgkwak95/SURF-GAN.\n", "date": "2022-07-21", "categories": []} {"id": "2207.11795", "title": "Cross-Modal 3D Shape Generation and Manipulation", "abstract": " Creating and editing the shape and color of 3D objects require tremendous\nhuman effort and expertise. Compared to direct manipulation in 3D interfaces,\n2D interactions such as sketches and scribbles are usually much more natural\nand intuitive for the users. In this paper, we propose a generic multi-modal\ngenerative model that couples the 2D modalities and implicit 3D representations\nthrough shared latent spaces. With the proposed model, versatile 3D generation\nand manipulation are enabled by simply propagating the editing from a specific\n2D controlling modality through the latent spaces. For example, editing the 3D\nshape by drawing a sketch, re-colorizing the 3D surface via painting color\nscribbles on the 2D rendering, or generating 3D shapes of a certain category\ngiven one or a few reference images. Unlike prior works, our model does not\nrequire re-training or fine-tuning per editing task and is also conceptually\nsimple, easy to implement, robust to input domain shifts, and flexible to\ndiverse reconstruction on partial 2D inputs. We evaluate our framework on two\nrepresentative 2D modalities of grayscale line sketches and rendered color\nimages, and demonstrate that our method enables various shape manipulation and\ngeneration tasks with these 2D modalities.\n", "date": "2022-07-24", "categories": []} {"id": "2207.10642", "title": "Generative Multiplane Images: Making a 2D GAN 3D-Aware", "abstract": " What is really needed to make an existing 2D GAN 3D-aware? To answer this\nquestion, we modify a classical GAN, i.e., StyleGANv2, as little as possible.\nWe find that only two modifications are absolutely necessary: 1) a multiplane\nimage style generator branch which produces a set of alpha maps conditioned on\ntheir depth; 2) a pose-conditioned discriminator. We refer to the generated\noutput as a 'generative multiplane image' (GMPI) and emphasize that its\nrenderings are not only high-quality but also guaranteed to be view-consistent,\nwhich makes GMPIs different from many prior works. Importantly, the number of\nalpha maps can be dynamically adjusted and can differ between training and\ninference, alleviating memory concerns and enabling fast training of GMPIs in\nless than half a day at a resolution of $1024^2$. Our findings are consistent\nacross three challenging and common high-resolution datasets, including FFHQ,\nAFHQv2, and MetFaces.\n", "date": "2022-07-21", "categories": []} {"id": "2207.04632", "title": "SkexGen: Autoregressive Generation of CAD Construction Sequences with\n Disentangled Codebooks", "abstract": " We present SkexGen, a novel autoregressive generative model for\ncomputer-aided design (CAD) construction sequences containing\nsketch-and-extrude modeling operations. Our model utilizes distinct Transformer\narchitectures to encode topological, geometric, and extrusion variations of\nconstruction sequences into disentangled codebooks. Autoregressive Transformer\ndecoders generate CAD construction sequences sharing certain properties\nspecified by the codebook vectors. Extensive experiments demonstrate that our\ndisentangled codebook representation generates diverse and high-quality CAD\nmodels, enhances user control, and enables efficient exploration of the design\nspace. The code is available at https://samxuxiang.github.io/skexgen.\n", "date": "2022-07-11", "categories": []} {"id": "2206.12055", "title": "SDF-StyleGAN: Implicit SDF-Based StyleGAN for 3D Shape Generation", "abstract": " We present a StyleGAN2-based deep learning approach for 3D shape generation,\ncalled SDF-StyleGAN, with the aim of reducing visual and geometric\ndissimilarity between generated shapes and a shape collection. We extend\nStyleGAN2 to 3D generation and utilize the implicit signed distance function\n(SDF) as the 3D shape representation, and introduce two novel global and local\nshape discriminators that distinguish real and fake SDF values and gradients to\nsignificantly improve shape geometry and visual quality. We further complement\nthe evaluation metrics of 3D generative models with the shading-image-based\nFr\\'echet inception distance (FID) scores to better assess visual quality and\nshape distribution of the generated shapes. Experiments on shape generation\ndemonstrate the superior performance of SDF-StyleGAN over the state-of-the-art.\nWe further demonstrate the efficacy of SDF-StyleGAN in various tasks based on\nGAN inversion, including shape reconstruction, shape completion from partial\npoint clouds, single-view image-based shape generation, and shape style\nediting. Extensive ablation studies justify the efficacy of our framework\ndesign. Our code and trained models are available at\nhttps://github.com/Zhengxinyang/SDF-StyleGAN.\n", "date": "2022-06-24", "categories": []} {"id": "2206.04942", "title": "Neural Template: Topology-aware Reconstruction and Disentangled\n Generation of 3D Meshes", "abstract": " This paper introduces a novel framework called DTNet for 3D mesh\nreconstruction and generation via Disentangled Topology. Beyond previous works,\nwe learn a topology-aware neural template specific to each input then deform\nthe template to reconstruct a detailed mesh while preserving the learned\ntopology. One key insight is to decouple the complex mesh reconstruction into\ntwo sub-tasks: topology formulation and shape deformation. Thanks to the\ndecoupling, DT-Net implicitly learns a disentangled representation for the\ntopology and shape in the latent space. Hence, it can enable novel disentangled\ncontrols for supporting various shape generation applications, e.g., remix the\ntopologies of 3D objects, that are not achievable by previous reconstruction\nworks. Extensive experimental results demonstrate that our method is able to\nproduce high-quality meshes, particularly with diverse topologies, as compared\nwith the state-of-the-art methods.\n", "date": "2022-06-10", "categories": []} {"id": "2606.23489", "title": "MeshFlow: Mesh Generation with Equivariant Flow Matching", "abstract": "Meshes are among the most common 3D scene representations, but directly generating meshes is challenging because the representation contains important symmetries, including permutation invariance of faces and vertices. MeshFlow learns to generate triangle meshes directly as triangle soups, avoiding the need to serialize meshes into long autoregressive sequences. We adopt equivariant optimal-transport flow matching models that respect the key symmetries of triangle soups: arbitrary permutations of faces and permutations of the vertices within each face.\n Toward this goal, we propose a simple yet effective modification to the Diffusion Transformer architecture, resulting in a scalable network capable of modeling a velocity field while maintaining the desired equivariance. We further introduce an optimal-transport-based training objective that improves convergence by eliminating supervision signals that violate these symmetries. MeshFlow achieves mesh quality comparable to state-of-the-art autoregressive mesh generators while providing about an 18$\\times$ speedup during inference. Project page is at https://qiisun.github.io/MeshFlow/.", "date": "2026-06-22", "categories": []} {"id": "2606.23514", "title": "Arbor: Explicit Geometric Conditioning for Controllable 3D Asset Generation", "abstract": "Text and image conditioned 3D models now generate convincing assets, but they still offer little direct control over the space an object should occupy or avoid. In authoring, this spatial intent is often known before generation starts. A chair should fit a seating envelope, a prop should leave clearance for motion, or a part should expose a contact surface. Prompts and image views are poor carriers for such constraints, requiring the need for an explicit control interface.\n We present Arbor, a trainable attachment for text conditioned latent 3D generation. Arbor introduces constraint meshes as a native 3D control interface. The interface uses hull regions where geometry should exist, avoidance regions that should remain empty, and touch regions the object should contact. Unlike completion or whole object scaffold control, these meshes are not target evidence. They are local typed requirements and can include regions where no surface should appear. Arbor keeps this signal as geometry by converting constraint meshes into tokens and learning a routed attachment inside a frozen denoiser. Each latent region can therefore receive the part of the constraint that matters for its spatial location.\n We evaluate Arbor on automatic and artist curated control benchmarks with hull, avoidance, and touch constraints, and compare the metric trends to a user preference study. Even without dedicated compliance losses, Arbor improves constraint obedience while preserving object quality and variation under fixed constraints.", "date": "2026-06-22", "categories": []} {"id": "2606.24138", "title": "Sat2City v2: Native 3D City Asset Generation from a Single Satellite Image", "abstract": "Generating explicit 3D city assets from a single satellite image is important for digital twins, urban simulation, and geospatial intelligence. Unlike satellite-to-street-view synthesis, the task requires a reusable textured mesh with plausible geometry and controllable appearance rather than a 3D proxy optimized only for rendering a small set of images or videos. The ICCV Sat2City framework made a first step by conditioning cascaded sparse-voxel latent diffusion on satellite-derived height maps, but its appearance was random, its training data were synthetic, and its task-specific VAE did not scale well to noisy real-world reconstructions. We present Sat2City v2, a journal extension that adapts a pretrained native structured-latent 3D foundation model to weakly aligned satellite images and textured meshes. We build a real-world dataset with 16,241 satellite-mesh pairs across 24 regions in 9 cities. Instead of learning a 3D representation from noisy city meshes, Sat2City v2 encodes each mesh into a pretrained native 3D latent space, fine-tunes a satellite-conditioned geometry flow, and uses the decoded shape to anchor satellite-conditioned texturing. This retains Sat2City's geometry-to-appearance cascade while enabling appearance-controllable generation from the satellite input. Experiments on metric-scale DSM reconstruction and generative city-asset benchmarks for geometry and appearance show that Sat2City v2 achieves the best overall performance among evaluated baselines. Overall, Sat2City v2 advances satellite-to-city generation from rendering-oriented 3D proxies to explicit textured mesh assets, supported by, to the best of our knowledge, the first documented satellite-mesh paired dataset collected from matched geographic crops for this asset-level task. Project page: https://ai4city-hkust.github.io/Sat2City-v2/", "date": "2026-06-23", "categories": []} {"id": "2606.24206", "title": "Inclusive Interactive Collisions for Multi-View Consistent Compositional 3D Generation", "abstract": "Recent breakthroughs in 3D generation have advanced notably with the development of text-to-image diffusion model. However, existing methods remain two practical challenges: (1) They primarily generate single 3D object, but struggle to generate multi-object compositional 3D assets due to the lack of the modeling for Gaussian primitives in reasonable interactions. (2) They often suffer from cross-view inconsistency during 3D optimization, as Score Distillation Sampling inherently performs on each single view, inevitably resulting in cross-view hallucinations. To solve above issues, we propose I2C-3D, a novel optimization-based method to generate multi-view consistent compositional 3D assets with reasonable interactions. Specifically, we propose an Inclusive Interactive Collisions strategy to guide Gaussian primitives appearing in reasonable interaction regions naturally, thereby ensuring objects in the compositional scene interact in a physically plausible and visually coherent way. Additionally, to enhance multi-view consistency, Multi-View Adaptive Score Distillation Sampling is devised to distill multi-view consistency prior and layout prior from pre-trained diffusion model by modulating attention map of instance token and spatial token across viewpoints. Benefiting from above elaborate designs, I2C-3D not only generates high-fidelity multi-view consistent compositional 3D assets but also supports 3D editing flexibly, facilitating complex scene generation. Extensive experiments demonstrate our I2C-3D outperforms existing methods in generation quality and multi-view consistency.", "date": "2026-06-23", "categories": []} {"id": "2606.24232", "title": "FiCA: Feed-forward instant Gaussian Codec Avatars from a Single Portrait Image", "abstract": "We introduce FiCA, a Feed-forward, instant Gaussian Codec Avatar generation pipeline that creates lifelike avatars from a single portrait image. Generating a photorealistic and drivable avatar from just a single image is significantly challenging due to the limited visual information available to accurately infer the 3D appearance and geometry of human heads. To address this, we develop a novel system that combines human-centric vision foundation models with a diffusion model. This system is designed to fully exploit partial visual observations to generate lifelike human avatars. Our proposed diffusion model learns a generative mapping from these partial observations to complete and authentic 3D mesh reconstruction. Additionally, we introduce a feed-forward mesh refinement network that enhances the fidelity and identity preservation of the generated avatars, eliminating the need for person-specific test-time optimization. By leveraging a universal prior model that decodes a generated mesh into a set of 3D Gaussians, we generate a photorealistic 3D Gaussian avatar, capable of being driven with novel expressions in real-time. Our experiments demonstrate that the avatars generated by our feed-forward approach faithfully represent diverse identities and surpass the visual quality of avatars produced by recent competing methods.", "date": "2026-06-23", "categories": []} {"id": "2606.24257", "title": "3DCarGen: Scalable 3D Car Generation via 3D-consistent Multi-view Synthesis", "abstract": "High-quality 3D vehicle assets are essential for autonomous driving simulation. Although multi-view diffusion-based paradigms enable controllable single-image reconstruction, they typically produce limited viewpoints and exhibit cross-view geometric inconsistencies, thereby reducing reconstruction fidelity in real-world scenarios. In this work, we introduce 3DCarGen, a scalable single-view 3D car generation framework designed for real-world images by synthesizing an arbitrary number of 3D-consistent multi-view images. Specifically, given a single image as input, we first synthesize a set of images from fixed viewpoints. These images are then fed into a feed-forward reconstruction model, resulting in a coarse 3D representation based on 3D Gaussian Splatting. Conditioned on this explicit 3D prior, our multi-view diffusion model generates 3D-consistent images from arbitrary camera viewpoints. We further extend a fast mesh reconstruction algorithm by incorporating color-normal joint optimization to recover detailed and coherent 3D vehicle models from the synthesized dense views. Extensive experiments on synthetic and real-world datasets demonstrate that our approach achieves robust geometric consistency and reconstruction fidelity compared to existing methods. Project page: https://honglixiao.github.io/3dcargen.github.io/.", "date": "2026-06-23", "categories": []} {"id": "2606.24301", "title": "MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving", "abstract": "Recovering realistic 3D vehicle models from autonomous driving scenes is crucial for synthesizing training data and building simulation environment. However, most existing vehicle generation methods fail to fully exploit multimodal sensors i.e. multi-view images and LiDAR point clouds) and rely on neural rendering based reconstruction, leading to low-quality mesh. Recently, native 3D generative models have made significant progress, yet they are not built for arbitrary multi-view inputs and often struggle with in-the-wild driving images. In this work, we present MM-TRELLIS, a multi-modal version of TRELLIS for in-the-wild 3D vehicle generation that integrates LiDAR and image sensors from autonomous driving datasets into native 3D generative models. Specifically, multi-view images are cycled as conditioning inputs, while LiDAR point clouds provide test-time guidance to ensure geometric accuracy and cross-view consistency. During denoising, we first align the guidance point cloud with the model priors, then enforce consistency between the generated geometry and the guidance point cloud. Finally, we introduce a voxel filtering strategy based on the opacity of 3D Gaussian Splatting to suppress floaters and produce clean meshes. Comprehensive experiments on Waymo dataset demonstrate our method outperforms existing methods in high-fidelity 3D vehicle generation. Code is available at https://github.com/HongliXiao/MM-TRELLIS.", "date": "2026-06-23", "categories": []} {"id": "2606.24799", "title": "OrbitForge: Text-to-3D Scene Generation via Reconstruction-Anchored Video Synthesis", "abstract": "Generic text-to-video models can be used as rich open-world scene priors. Despite the high quality of today's generated videos, they do not directly yield reliable 3D assets: camera motion is difficult to control, view coverage is partial, and frames often contain inconsistencies across time. We introduce OrbitForge, an adapter built from frozen video priors and per-prompt Gaussian Splatting reconstruction optimization that converts a single text-generated video into a canonical closed-orbit 3D Gaussian Splatting scene. We use 3D reconstruction as an anchor to improve the 3D consistency of the generated video. We obtain a preliminary 3D reconstruction from a first generated video via Deformable Gaussian Splatting with a robust MedianGS proxy. We render views from a prescribed orbit to detect missing viewpoints. OrbitForge uses the text-to-video model to complete only the missing views, and reconstructs the completed orbit into a final Gaussian Splatting scene. This design requires no task-specific video or multiview fine-tuning, avoids per-prompt score-distillation optimization, and does not progressively generate views one step at a time. We further argue that this setting demands coverage-aware evaluation: local smoothness alone rewards methods that never attempt a full orbit. On a frozen 300-prompt T3Bench-derived audit, OrbitForge reconstruction attains a 359.0-degree measured median span, raises originally unsupported-bin Q10 ImageReward from 8.07 to 16.36 relative to MedianGS-only reconstruction, while remaining competitive with VideoMV on the coverage-quality.", "date": "2026-06-23", "categories": []} {"id": "2606.24874", "title": "FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation", "abstract": "Sparse voxel representation has emerged as a scalable foundation for image-to-3D Gaussian Splatting (3DGS) generation, yet current methods struggle to preserve high-frequency visual details of input images due to two structural bottlenecks. First, they adopt discriminative 2D features optimized for semantic abstraction to construct sparse voxel latents, which suppress reconstructive cues and induce a representation bottleneck. Second, in the generation stage, standard diffusion transformers lack effective mechanisms to align dense 2D image tokens with sparse 3D voxel latents, resulting in a cross-modal correspondence bottleneck. To address these issues, we propose FLUX3D, a scalable image-to-3DGS framework that boosts both representation learning and cross-modal alignment during generation. We first revisit 2D feature selection for sparse-voxel-based 3D representation learning, propose Diffusion-Aligned Structured Latents (DA-SLAT) and couple it with a decoder-only architecture to improve 3DGS reconstruction fidelity. We also design a sparse-structure-aware diffusion framework, which integrates the Sparse-structure Multimodal Diffusion Transformer (SMDiT) and Modal-Aware Rotary Positional Embedding (MARoPE) to achieve geometry-agnostic 2D-3D alignment. Extensive benchmark experiments demonstrate that FLUX3D yields substantial improvements in appearance fidelity and significantly outperforms all state-of-the-art (SOTA) methods in generating high-quality 3DGS assets.", "date": "2026-06-23", "categories": []} {"id": "2606.24876", "title": "FLAT: Feedforward Latent Triangle Splatting for Geometrically Accurate Scene Generation", "abstract": "Generating explorable 3D scenes from a single image requires strong generative priors and accurate geometric representations suitable for downstream use. Current video diffusion models offer high-quality generation and implicitly encode multi-view geometric structure in latent space. However, existing feedforward latent scene decoders typically output volumetric 3D Gaussians that lack a well-defined surface, limiting their use in simulation or standard graphics pipelines. This motivates decoding surface-aligned primitives that are not only renderable but also closer to explicit geometric assets. We ask whether compressed video diffusion latents can be mapped directly to explicit surface primitives in a single pass. To this end, we introduce FLAT and, for the first time, show that triangle splats can be decoded directly from video diffusion latents. Compared with decoding 3D Gaussians, predicting flat primitives is notoriously more challenging due to high sensitivity to primitive orientations, oftentimes leading to poor gradient flow. FLAT solves with two key ingredients: a ray-centered rotation parameterization for triangle regression and a novel product window function that improves gradient flow during differentiable triangle rendering. On standard benchmarks, FLAT achieves significantly better geometric accuracy while maintaining competitive visual quality compared to state-of-the-art feedforward baselines. We further show that a lightweight test-time refinement step converts the predicted triangle soup into a fully opaque, game-engine-ready representation that supports real-time rendering. By evaluating 3DGS, 2DGS, and triangle splatting variants under an identical training setup, we provide the first systematic analysis of representation tradeoffs in feedforward scene generation. The project page is available at https://flat-splat.github.io", "date": "2026-06-23", "categories": []} {"id": "2606.25300", "title": "HiFiVe: High-Fidelity Vehicle Generation Leveraging Auto-Regressive 2D Generative Priors", "abstract": "Existing 3D vehicle generation methods often suffer from low geometric fidelity and blurry textures, hindering their downstream applications. While recent works adopt multi-view diffusion models for high-fidelity texture, they are often constrained by fixed viewpoints, limited resolution, and a reliance on costly fine-tuning to achieve cross-view consistency. In this paper, we propose HiFiVe, a training-free framework for high-fidelity vehicle modeling through joint texture and geometry enhancement by imposing 3D geometric constraints to anchor 2D generative priors. Specifically, we propose an auto-regressive texture refinement pipeline that progressively synthesizes high-resolution textures from arbitrary viewpoints. To ensure cross-view consistency, the coarse geometry serves as a synchronization prior, conditioning each generation step on previously synthesized frames via depth-based warping and multi-view texture fusion. Moreover, the inherent symmetry of vehicles is exploited to mitigate error accumulation. Finally, high-frequency surface details are recovered by refining the mesh geometry using normal maps estimated from the enhanced textures. Extensive experiments on synthetic and real-world vehicle datasets demonstrate that our method significantly improves both geometric detail and texture quality compared to state-of-the-art baselines. Project page: https://honglixiao.github.io/hifive.github.io/.", "date": "2026-06-24", "categories": []} {"id": "2606.27305", "title": "Sculpting NeRF Geometry: Human-Preference Fine-Tuning of a 3D-Aware Face GAN", "abstract": "Reinforcement learning from human feedback (RLHF) for 3D generation is now established across a number of works, but most existing pipelines optimise explicit surface representations, often by converting radiance fields into meshes and training heavily on surface-supervised data. We instead fine-tune a pretrained 3D-aware generative model directly from a learned reward over radiance-field density ($\\sigma$) values, with no externally supplied mesh or shape prior. The reward model requires no pretraining, trains easily on a small set of preference samples, and yields robust improvement in 3D geometry. Working on an unconditional 3D-aware face GAN (EG3D), our reward reads the continuous 3D density field of the neural radiance field (NeRF) directly and supplies a geometry-only learning signal, requiring neither text conditioning, mesh extraction, nor multi-view rendering. A density-consistency constraint keeps the 2D appearance qualitatively similar while the geometry is reshaped, at a measurable but bounded distributional cost (FID-50k rises from 4.09 to 6.66): the fine-tuned generator, trained from the preferences of a single annotator as a proof of concept, produces face geometries preferred by users in 74.4% of pairwise comparisons.", "date": "2026-06-25", "categories": []} {"id": "2606.30673", "title": "PolyFlow: Continuous Topology Embedding Flow Matching for Artist-style Mesh Generation", "abstract": "Autoregressive Transformers dominate high-quality mesh generation by producing artist-worthy topologies, yet their inherent sequential decoding induces substantial computational overhead, falling orders of magnitude slower than parallel generative models. On the other hand, while continuous diffusion and flow-matching methods support efficient parallel synthesis across a variety of domains, they cannot be directly applied to meshes: mesh connectivity is inherently discrete and incompatible with standard continuous noise injection and denoising operations. To resolve this fundamental incompatibility, we introduce a compact topology embedder that projects discrete mesh vertex positions and normals into continuous per-vertex embeddings, where the original discrete adjacency information can be faithfully recovered via spacetime distance thresholding. After pretraining and freezing this embedder, any raw mesh can be fully converted into a continuous per-vertex state space unifying position, normal, and implicit topological attributes. Built upon this novel continuous mesh representation, we present PolyFlow, a Transformer-based flow-matching framework that achieves fully parallel vertex state denoising conditioned on extracted point-cloud features. During inference, our model completes generation rapidly via an ODE solver, and supports explicit, precise control over output mesh resolution by directly specifying the target vertex count. Extensive evaluations on the Toys4K benchmark demonstrate that PolyFlow surpasses state-of-the-art autoregressive baselines in both Chamfer Distance and Hausdorff Distance.", "date": "2026-06-25", "categories": []} {"id": "2606.27659", "title": "GeoFace: Consistent Multi-View Face Generation with Geometry-Constrained Diffusion", "abstract": "We present GeoFace, a geometry-constrained multi-view diffusion framework for consistent face generation from a single input. % While recent multi-view diffusion models achieve photorealistic synthesis at the per-view level, they lack an explicit mechanism to enforce a shared 3D structure across views, often leading to inconsistent geometry across viewpoints. To address this, GeoFace proposes a unified dual-stream framework for joint generation of multi-view RGB images and 3D face geometry, where the appearance and geometry streams interact through shared attention layers. To encourage the two streams to mutually constrain each other, we introduce a geometry-guided attention alignment loss that supervises the cross-attention between appearance and geometry tokens with 3D-consistent correspondences, enabling the appearance stream to correctly reference pose-invariant geometric cues for robust alignment across viewpoints. Geometry is represented as a canonical UV position map, derived from a FLAME mesh fitted to multi-view observations, serving as a view-invariant shared constraint across all generated views. Experiments on RenderMe-360 and NeRSemble demonstrate that GeoFace consistently outperforms existing methods in both visual quality and cross-view geometric consistency, facilitating more efficient 3D reconstruction.", "date": "2026-06-26", "categories": []} {"id": "2606.27923", "title": "Home3D 1.0: A High-Fidelity Image-to-3D Asset Generation System for Interior Design", "abstract": "We present Home3D 1.0, a modular image-to-3D generation system that produces high-quality 3D assets from a single reference image, targeting interior design and e-commerce applications. Given a photograph of a furniture or decor item, the system outputs a mesh with physically-based rendering (PBR) materials, and the mesh can be decomposed into material-specific components. The pipeline is organized into four tightly coupled modules: Geometry reconstructs a watertight mesh through latent SDF modelling with a geometry VAE and a coarse-to-fine flow-matching DiT; Texture predicts multiview albedo observations, reprojects them onto the mesh, and completes unseen surface regions with a 3D texture field; Material uses MatWeaver to obtain component masks through video-based segmentation and UV-space voting, then retrieves and bakes PBR maps from a curated material library through hierarchical multi-modal matching; and Parts generates material-editable semantic part meshes with a PartVAE and PartDiT, decoding multi-head part-specific SDF fields in one pass. Each module is evaluated independently with dedicated metrics, highlighting both the current system capability and the remaining gaps toward broader deployment.", "date": "2026-06-26", "categories": []} {"id": "2606.28144", "title": "Monocular Avatar Reconstruction via Cascaded Diffusion Priors and UV-Space Differentiable Shading", "abstract": "Reconstructing high-fidelity, relightable 3D avatars from a single in-the-wild image is a challenging ill-posed problem, primarily hindered by the scarcity of high-quality PBR data and the complexity of disentangling illumination from intrinsic materials. In this paper, we present a data-efficient framework that leverages the robust priors of a unified pre-trained diffusion backbone to sequentially address texture completion, delighting, and material decomposition. Unlike existing methods that rely on fragmented pipelines or extensive proprietary datasets, we utilize cascaded Low-Rank Adaptations (LoRAs) to adapt the strong generative prior of the diffusion model for each sub-task in UV space. Specifically, we first employ an Inpainting LoRA to complete missing UV textures caused by occlusion, leveraging the model's semantic understanding to generate semantically and photometrically coherent details. Subsequently, a Light-Homogenization LoRA and a novel Cross-Intrinsic Attention mechanism are introduced to remove baked-in lighting and collaboratively synthesize pixel-aligned PBR maps (Albedo, Normal, Roughness, Specular, and Displacement). To ensure physical plausibility, we impose a UV-space differentiable BRDF shading loss during the decomposition stage, forcing the generative process to adhere to the rendering equation without the artifacts typical of rasterization-based supervision. Extensive experiments demonstrate that our method, trained on fewer than 100 real 3D scans, generates comprehensive, 4K-resolution PBR assets with superior realism and generalization compared to state-of-the-art methods, and all training code and model weights will be released upon acceptance.", "date": "2026-06-26", "categories": []} {"id": "2606.29395", "title": "NaLA: A 3D Native LLM Layout Agent for High-quality 3D Scene Generation", "abstract": "Recently, Large Language Models (LLMs) have emerged as promising layout agents for 3D scene generation. Existing layout agents still suffer from implausible layout generation because most of them convert 3D assets and 3D layouts into textual descriptions as inputs and outputs, which involves severe information loss due to the modality gap between texts and 3D assets and 3D layouts. We propose NaLA, a native 3D LLM layout Agent for high-quality 3D scene generation by placing 3D assets in the scene. For the inputs, NaLA encodes 3D scene boundaries and 3D assets directly into the LLM, preserving fine-grained geometry and enabling explicit reasoning over relationships like collisions, surface supporting, and containment. To accurately output the positions and orientations of assets, NaLA adopts a coarse-to-fine prediction mechanism that first predicts discrete poses in an autoregressive manner and then refines the discrete poses with a continuous regression. Trained on diverse layout datasets, NaLA attains strong geometric perception and layout coherence. Experiments demonstrate that NaLA outperforms prior layout agents in both generation quality and inference efficiency, with comprehensive ablation studies to verify each component's effectiveness.", "date": "2026-06-28", "categories": []} {"id": "2606.29964", "title": "Variance Reduction on the Camera Axis: Multi-View Score Distillation for 3D", "abstract": "Score distillation turns a pretrained 2D diffusion model into a 3D generator, but the per-step gradient is estimated from a single randomly chosen view: it is high-variance and blind to global shape consistency. Prior work addresses this by retraining the diffusion prior on multi-view data; this improves consistency but makes the sampling contribution inseparable from prior quality. We instead isolate the sampling axis. The per-step gradient is one noisy sample of an expectation over views; aggregating K samples per step at a fixed total UNet budget reduces variance without touching the prior. We introduce Multi-View Aggregated Score Distillation (MV-SDI), which aggregates gradients from K views per step via gradient accumulation, keeping peak memory unchanged and the 2D prior frozen, and draws views as antithetic antipodal pairs, a prior-independent geometric property, for balanced angular coverage. At a fixed 10,000-UNet-call budget, K=2 raises CLIP R-Precision from 74.8% to 83.8% and CLIP score from 0.297 to 0.312, with consistent gains on HPSv2 and ImageReward and a 0.0% divergence rate on the 43-prompt benchmark; optimization steps halve as a consequence. K=4 gives a fourfold step reduction at R-Precision 86.9% and CLIP 0.307, still well above the single-view baseline on every alignment metric. MV-SDI is compatible with gradient-based score-distillation pipelines, including Score Distillation via Inversion, and requires no retraining and no multi-view data.", "date": "2026-06-29", "categories": []} {"id": "2606.30147", "title": "T2LDM++: A Self-Conditioned Representation Guided Diffusion Model for Realistic Text-to-LiDAR Scene Generation", "abstract": "Recent progress in Text-to-Image generation benefits from large-scale Text-Image pairs. However, the scarcity of Text-LiDAR pairs often causes over-smoothed scenes and limited controllability. In this paper, we rethink the limitations of Text-LiDAR generation task, focusing on alleviating insufficient training priors and constructing controllable Text-LiDAR data. We propose a \\textbf{T}ext-\\textbf{to}-\\textbf{L}iDAR \\textbf{D}iffusion \\textbf{M}odel for LiDAR scene generation, T2LDM++, with a Self-Conditioned Representation Guidance (SCRG). Specifically, to alleviate object over-smoothing, SCRG employs a Guidance Network (GN) to provide reconstruction-based soft supervision to the Denoising Network (DN). This enables DN to learn geometry-aware representations through reconstruction guidance, leading to more accurate denoising in DDPMs. Meanwhile, through analysis and design, SCRG exhibits more effective and lightweight, while decoupled in inference, avoiding computational overhead. Furthermore, we construct two high-quality Text-LiDAR benchmarks ($>$100K samples) using a generalized strategy of geometric annotations, along with a controllability metric. Moreover, a directional position prior is designed to mitigate street distortion, further improving scene fidelity. Additionally, T2LDM++ supports multiple conditions, including (Semantic, Box, BEV, Camera)-to-LiDAR, Sparse-to-Dense, and Dense-to-Sparse generation, by learning a control encoder via frozen DN. With effective prior modeling and high-quality Text-LiDAR benchmarks, T2LDM++ can generate realistic LiDAR scenes with rich geometric details in unconditional and conditional settings.", "date": "2026-06-29", "categories": []} {"id": "2606.30429", "title": "Arko-T: A Foundation Model for Text-to-Structured 3D Generation", "abstract": "Text-to-3D systems can now synthesize a model from a single sentence, yet the result is a shape to render, not a design to edit. We present Arko-T, a 4B-parameter text-to-design model that maps natural-language intent directly into executable, parametric CAD programs. Rather than optimizing for code executability alone, Arko-T aligns every stage of the pipeline to a formal notion of design state, so that data curation, code normalization, and execution-grounded supervision all work to preserve the features, parameters, and construction logic that make a CAD artifact editable. Benchmarked against seven frontier LLMs across 12 metrics, Arko-T attains the best score on 8 and the second-best on 3 more, at roughly one-tenth the per-benchmark cost. The results suggest that targeted design-level training at moderate scale can match frontier general-purpose models on structured CAD generation.", "date": "2026-06-29", "categories": []} {"id": "2606.31777", "title": "Mesh BDF: Barycentric Dominance Field for 3D Native Mesh Generation", "abstract": "Autoregressive (AR) modeling has recently achieved remarkable progress in native 3D mesh generation, largely due to its natural ability to handle variable-length, discrete data structures. However, the inherent constraints of the AR paradigm severely restrict the generated meshes, leading to limited face counts, bounded vertex resolutions, and difficulties in supporting textures. To overcome these bottlenecks, we propose the Barycentric Dominance Field (BDF), a continuous representation defined on triangular mesh surfaces that elegantly encodes vertex topological connectivity. BDF bridges the fundamental gap between discrete mesh topology and continuous diffusion-based generative modeling by transforming connectivity into a continuous surface signal. As an intrinsic mesh property, BDF shares strong similarities with texture maps, enabling its seamless integration into existing 3D diffusion pipelines without requiring architectural modifications. Extensive experiments demonstrate that BDF empowers diffusion models to generate native meshes with significantly higher quality, greater scalability, and stronger robustness compared to state-of-the-art autoregressive methods.", "date": "2026-06-30", "categories": []}