{"id": "2606.15966", "title": "VEPHand: View-Efficient Photometric Hand Performance Capture at Scale", "abstract": "Robust, high-fidelity 3D hand capture, while fundamental to digital human creation, remains challenging with practical multi-view systems that balance rich photometry with the geometric ambiguities of reconstruction arising from limited viewpoint density. This paper presents an end-to-end pipeline for dynamic hand performance capture and registration, specifically designed for view-efficient setups ($\\sim$20 views). We address key challenges with two primary innovations. First, to overcome reconstruction difficulties like limited view overlap and background clutter, our mask-free neural method robustly extracts detailed hand geometry and appearance from unmasked images using scene parameterization and scenario-specific density regularization. Second, addressing registration challenges such as accurately capturing non-linear skin deformations and ensuring plausible results during severe self-contact, we propose a physics-inspired framework. It aligns reconstructions to a personalized hand model by optimizing intrinsic volumetric offsets within its canonical tetrahedral mesh, alongside pose parameters. This approach, supported by robust losses and optimization, captures fine surface deformations, ensures plausible results under severe articulation and self-contact, and demonstrates strong tolerance to input noise. We demonstrate the scalability and robustness of our automated pipeline on an extensive dataset of over 12,000 sequences, from which we also derive a large-scale, high-quality synthetic 2D/3D hand dataset for training downstream tasks. This showcases its effectiveness for single hands, intricate two-hand interactions, and natural hand-object manipulations. Our method achieves state-of-the-art reconstruction fidelity in view-efficient, unmasked scenarios and highly accurate registration. Our project page are available at https://vephand.github.io/.", "date": "2026-06-14", "categories": []}
{"id": "2407.18245", "title": "VGGHeads: 3D Multi Head Alignment with a Large-Scale Synthetic Dataset", "abstract": "Human head detection, keypoint estimation, and 3D head model fitting are essential tasks with many applications. However, traditional real-world datasets often suffer from bias, privacy, and ethical concerns, and they have been recorded in laboratory environments, which makes it difficult for trained models to generalize. Here, we introduce \\method -- a large-scale synthetic dataset generated with diffusion models for human head detection and 3D mesh estimation. Our dataset comprises over 1 million high-resolution images, each annotated with detailed 3D head meshes, facial landmarks, and bounding boxes. Using this dataset, we introduce a new model architecture capable of simultaneous head detection and head mesh reconstruction from a single image in a single step. Through extensive experimental evaluations, we demonstrate that models trained on our synthetic data achieve strong performance on real images. Furthermore, the versatility of our dataset makes it applicable across a broad spectrum of tasks, offering a general and comprehensive representation of human heads.", "date": "2024-07-25", "categories": []}
{"id": "2606.19156", "title": "Hand-4DGS: Feed-Forward 3D Gaussian Splatting for 4D Hand Reconstruction from Egocentric Videos", "abstract": "Dynamic 3D hand reconstruction from egocentric videos is essential for next-generation computing platforms such as AR/VR and AI glasses. Despite its importance, most prior works focus either on multi-view 3D hand reconstruction or on 4D human body reconstruction. Egocentric 4D hand reconstruction remains challenging due to fast head motion, rapid hand dynamics, severe occlusions, and inherent ambiguity from single-view observations. To address these challenges, we introduce Hand-4DGS, the first feed-forward framework for reconstructing dynamic 4D hands directly from egocentric videos, enabling both fast (~60 FPS) inference and strong generalization. Our approach incorporates a mesh-guided representation for structural priors and temporal convolutions to model dynamic motion. We evaluate our framework on two challenging egocentric datasets, H2O and ARCTIC, and demonstrate significant improvements over baselines. Our method benefits from the generalization capability of feed-forward networks and effective 2D image supervision through Gaussian splatting, without requiring expensive 3D hand pose ground-truth annotations.", "date": "2026-06-17", "categories": []}
{"id": "2606.10550", "title": "LentiAvatar: Pseudo-Multiview Reconstruction and Subpixel Prism Rendering for Real-Time Stereoscopic Communication", "abstract": "Real-time stereoscopic video communication has long been a goal of immersive telepresence, yet practical systems still require specialized capture rigs or reduce remote users to a single portrait view. We present LentiAvatar, a Gaussian head-avatar system that connects monocular avatar capture with subpixel-encoded glasses-free lenticular display for real-time autostereoscopic communication. From a monocular portrait video, LentiAvatar reconstructs a controllable head avatar and optimizes it for the lateral viewing zones induced by the display. The method uses natural head turns as pseudo-multiview (PMV) supervision to constrain regions that are otherwise weakly observed in monocular training, including hair, ears, jaw contours, and neck boundaries. Reliable side frames are yaw-binned, aligned to virtual cameras, and supervised within a strict head-and-hair domain; contour-aware losses and staged regularization further suppress ghosting, alpha leakage, and depth instability while preserving lateral detail. At runtime, LentiAvatar renders 32 virtual views and encodes them into a 4K lenticular raster with calibrated subpixel-routing masks. The live-tracker prototype sustains 10.65 FPS, and a subject-specific distilled driver raises the same display pipeline to 38.49 FPS.", "date": "2026-06-09", "categories": []}
{"id": "2606.13655", "title": "Flex4DHuman: Flexible Multi-view Video Diffusion for 4D Human Reconstruction", "abstract": "We present Flex4DHuman, a multi-view video diffusion model that transforms a monocular or sparse multi-view video of a dynamic subject into synchronized dense multi-view videos using only relative camera-pose conditioning. Unlike prior human-centric methods that rely on skeletons, depth maps, normals, or rendered target-view geometry, Flex4DHuman requires no explicit geometry priors and instead conditions generation through relative camera-pose positional encoding. The generated videos can be directly ingested by downstream reconstruction pipelines to create dynamic 4D Gaussian splats. Built on the Wan 2.1 1.3B text-to-video model, Flex4DHuman preserves the backbone architecture and encodes camera and view information through a five-axis positional encoding that extends spatio-temporal RoPE with view indices and continuous SE(3) relative camera geometry. A three-stage curriculum progressively trains the model for pose following, flexible reference-to-target view generation, and temporal rollout. To support temporal rollout, we train with clean historical target-view tokens. We also add multi-view captions to enable test-time text control. Combined with an off-the-shelf 4D Gaussian Splatting stage, our framework lifts monocular static-camera videos into dynamic 4D Gaussian splats. Experiments on DNA-Rendering and ActorsHQ show that Flex4DHuman surpasses prior state-of-the-art methods, while the same formulation generalizes to animal categories after mixed human-animal training. These capabilities make Flex4DHuman a practical step toward scalable 4D content creation from casual monocular videos for simulation, gaming, AR/VR, and video re-shooting.", "date": "2026-06-11", "categories": []}
{"id": "2606.14841", "title": "Multi-HMR 2: Multi-Person Camera-Centric Human Detection, Mesh Recovery and Tracking", "abstract": "Most advances in human mesh recovery (HMR) have focused on pelvis-centered recovery, overlooking metric 3D localization and detection accuracy in the camera coordinate system - two key factors for real-world applications such as human-robot interaction and social scene understanding. Current evaluation protocols often ignore these aspects, emphasizing per-person, root-centered recovery rather than camera-space perception. As a result, existing approaches rely on fixed camera assumptions or handcrafted post-processing, limiting their robustness and practical deployment. We introduce Multi-HMR 2, a simple yet robust DETR-based framework for Multi-person Camera-centric Human detection, mesh Recovery, and tracking. Multi-HMR 2 predicts a scene-consistent camera together with human meshes, enabling metric 3D localization without ground-truth intrinsics. Moreover, by distilling image-based memory features from SAM2, Multi-HMR 2 extends to tracking, achieving consistent identity association without video supervision. Despite its conceptual simplicity - no handcrafted components, no video input, and no ground-truth cameras - Multi-HMR 2 achieves state-of-the-art pelvis-centered performance while substantially improving detection accuracy and metric 3D localization.", "date": "2026-06-12", "categories": []}
{"id": "2605.05367", "title": "Tamaththul3D: High-Fidelity 3D Saudi Sign Language Avatars from Monocular Video", "abstract": "Existing 3D sign language avatar reconstruction methods are developed and evaluated exclusively on Western sign languages, and no 3D parametric annotations exist for any Arabic Sign Language dataset, a gap that blocks the development of avatar-based accessibility applications for the Arab Deaf community. We release the first SMPL-X parametric annotations for the Ishara-500 Saudi Sign Language dataset, enabling quantitative evaluation and downstream sign language generation for Arabic Sign Language. We introduce Tamaththul3D, a reconstruction pipeline that aligns hand and body estimates through geometric inverse kinematics on the forearm chain followed by 2D-supervised shoulder refinement. The closed-form integration is decoupled from the specific choice of body and hand estimators: any SMPL-X-compatible body estimator and any MANO-compatible hand estimator can be substituted, as we demonstrate by swapping each module independently. Tamaththul3D achieves up to 32% lower hand error than prior methods, runs 32x faster than the strongest baseline, and generalizes across five typologically distinct sign languages without dataset-specific adaptation.", "date": "2026-05-06", "categories": []}
{"id": "2606.05912", "title": "Self-Learning Expression Deformations for Data-Efficient Gaussian Avatars", "abstract": "Modeling dynamic facial expressions using 3D Gaussian representations remains challenging due to their unstructured nature. Conventional Gaussian avatar pipelines require extensive multiview and sequential expression data, limiting scalability and accessibility. In this work, we introduce Self-Adaptive Gaussian Expression (SAGE), a framework for self-learning expression-induced Gaussian deformations that enables high-fidelity, animatable avatars from minimal input data. Our method jointly optimizes 2D Gaussian surfels and a Signed Distance Field (SDF) to enforce compact, surface-aligned Gaussian distributions, while a self-supervised expression learning phase replaces long training sequences with geometric and appearance consistency constraints. This design allows flexible deployment across multiple reconstruction regimes: in the multiview setting, only a single frame (timestep) is required instead of thousands; in the monocular setting, only head rotations are needed without expression sequences; and in the one-shot setting, no pretraining or priors are necessary. Experiments demonstrate that our approach achieves reconstruction and animation quality comparable to state-of-the-art methods, while reducing data requirements by several orders of magnitude. Our results highlight the potential of self-supervised Gaussian deformation learning as a step toward accessible, data-efficient avatar creation.", "date": "2026-06-04", "categories": []}
{"id": "2605.18102", "title": "DanceHMR: Hand-Aware Whole-Body Human Mesh Recovery from Monocular Videos", "abstract": "Monocular video human mesh recovery is essential for digital humans, avatar animation, and embodied simulation, where both temporal stability and expressive whole-body motion are required. Existing video HMR methods produce coherent body motion but often overlook detailed hand articulation, while image-based whole-body methods recover SMPL-X meshes independently per frame, often leading to jittery and inaccurate hand motion. We present a temporally coherent whole-body HMR framework for challenging in-the-wild monocular videos. Our model unifies body context and part-specific hand observations through residual body-hand fusion, enabling stable body motion and detailed hand recovery within a single temporal architecture. We further introduce close-up-aware augmentation to improve robustness under upper-body framing. Experiments on whole-body and body-only benchmarks demonstrate improved hand reconstruction and competitive body accuracy. Our method also produces temporally stable and 2D-consistent SMPL-X motion in challenging real-world videos.", "date": "2026-05-18", "categories": []}
{"id": "2606.03420", "title": "PHAF-Personalized Hand Avatars in a Flash", "abstract": "We present PHAF-Personalized Hand Avatars in a Flash, a personalized photo-realistic hand avatar which provides high quality multi-view renders from just two images (dorsal and palmar views).Unlike slow optimization-based techniques, PHAF generates fast personalized textures for real-time deployment on edge devices. Our approach combines semantic guided mesh alignment and densified texture extraction to transfer high-frequency details efficiently. A view-based inpainting network refines textures ensuring smooth, continuous appearance. PHAF generalizes to novel viewpoints and leverages a parametric hand model for accurate articulations, making it compatible with standard graphics engines. Experiments show it is comparable to existing methods in visual fidelity while drastically reducing texture generation time by 30 times, enabling practical AR/VR applications.", "date": "2026-06-02", "categories": []}
{"id": "2606.03506", "title": "AvatarMix: Identity-Preserving Cross-Avatar Composition for Outfit Personalization", "abstract": "Existing 3D avatar outfit transfer methods face distinct challenges: approaches that lift 2D edits to 3D often suffer from outfit or identity quality degradation, while those that separately model body and clothing layers are prone to intersection artifacts. We introduce AvatarMix, a compositional paradigm that bypasses these issues by directly composing the head and body from two high-fidelity Gaussian avatars. While this paradigm inherently preserves outfit quality and avoids intersections, it introduces challenges in creating a seamless join and maintaining appearance fidelity after body reshaping. To this end, we propose a two-tier refinement strategy: SeamFix, a localized diffusion module that refines hair and neck to ensure an artifact-free join, and an optional full-body refinement, FullbodyFix, that restores garment appearance when retargeting degrades the clothed body. Both operate on renders from an already 3D-consistent Gaussian avatar, which limits multi-view artifacts compared to 2D-to-3D lifting. To preserve the user's body identity, our mesh-based Gaussian representation enables the adaptation of a robust mesh retargeting technique, precisely reshaping the clothed body to the user's physique and robustly handling diverse body shapes. Extensive experiments demonstrate that our method achieves state-of-the-art results in outfit fidelity and identity preservation, providing a new perspective for realistic 3D outfit personalization. Project page: https://larsph.github.io/avatarmix/", "date": "2026-06-02", "categories": []}
{"id": "2606.01493", "title": "Splatshot: 3D Face Avatar Generation from a Single Unconstrained Photo", "abstract": "Reconstructing a photorealistic 3D face avatar from a single unconstrained photograph is challenging: feed-forward 3D Gaussian Splatting (3DGS) models degrade on out-of-distribution inputs, while pretrained diffusion models produce high-fidelity images but lack multi-view consistency. We observe that these paradigms are fundamentally complementary: explicit 3D representations guarantee geometric consistency, whereas 2D diffusion priors ensure photorealism. Building on this, we propose SplatShot, a training-free framework that couples these representations directly within the denoising process. Given a base 3DGS face model and a single reference image, we jointly denoise all target views using a per-step 3D feedback loop. At each timestep, we predict clean images from the noisy latents, refit the 3DGS to these multi-view predictions, and back-propagate the photometric discrepancy between the 3DGS re-renderings and 2D predictions into the noise estimate. This steers the sampling trajectory toward strictly 3D-coherent, identity-faithful outputs. Experiments on diverse in-the-wild images demonstrate that SplatShot produces 3D avatars with superior identity preservation, photorealism, and multi-view consistency.", "date": "2026-05-31", "categories": []}
{"id": "2606.02350", "title": "TROPHIES: Temporal Reconstruction of Places, Humans, and Cameras from Multi-view Videos", "abstract": "Reconstructing humans and their surrounding environments in a globally consistent 4D space is essential for comprehensive perception. However, prior works typically assume single-view inputs or decouple humans, scenes, and cameras, making them unable to recover coherent geometry, stable motion, and physically aligned trajectories. These limitations motivate us to introduce a new task: unified human-scene-camera reconstruction from multi-view videos, which aims to jointly estimate dynamic humans, static scenes, and camera poses in one global coordinate frame. We propose TROPHIES--Temporal Reconstruction of Places, Humans, and Cameras from Multi-view Videos-a unified framework tailored for this task. TROPHIES features a Human Branch that models humans through temporal and spatial reasoning, and a Scene Branch that reconstructs static geometry with human-aware attention. A global alignment and optimization module couples both branches by enforcing scale consistency, contact priors, and cross-view temporal coherence. Experiments on EgoHuman and EgoExo4D demonstrate that TROPHIES achieves globally aligned, physically plausible 4D reconstructions and consistently outperforms existing paradigms in both global fidelity and human-scene consistency.", "date": "2026-06-01", "categories": []}
{"id": "2606.02573", "title": "HumanNOVA: Photorealistic, Universal and Rapid 3D Human Avatar Modeling from a Single Image", "abstract": "In this paper, we present HumanNOVA, a photorealistic, universal, and rapid model for generating 3D human avatars from a single RGB image. Achieving both photorealism and generalization is challenging due to the scarcity of diverse, high-quality 3D human data. To address this, we build a scalable data generation pipeline that follows two strategies. The first one is to leverage existing rigged assets and animate them with extensive poses from daily life. The second strategy is to utilize existing multi-camera captures of humans and employ fitting to generate more diverse views for training. These two strategies enable us to scale up to 100k assets, significantly enhancing both the quantity and the diversity of data for robust model training. In terms of the architecture, HumanNOVA adopts a feed-forward, token-conditioned avatar modeling framework that allows fast inference in less than one second and requires no test-time optimization. Given an input image and an estimated simplified human mesh (SMPL) without detailed geometry or appearance, the model first encodes both inputs into compact token representations. These tokens then act as conditioning signals and are fused through cross-attention to construct a triplane-based 3D avatar representation. Extensive experiments on multiple benchmarks demonstrate the superiority of our approach, both quantitatively and qualitatively, as well as its robustness under diverse input image conditions. Project page at https://HumanNOVA.github.io .", "date": "2026-06-01", "categories": []}
{"id": "2605.31283", "title": "Topologically Consistent Multi-view 3D Head Reconstruction via Coarse-Guided Layered Surface Sampling", "abstract": "We present SHELLS (Semantic Head Estimation via Layered Local Sampling), an efficient feed-forward framework for 3D head reconstruction in dense semantic correspondence from multi-view images. Existing methods typically refine vertices independently via localized feature volumes. This approach couples memory-intensive feature sampling to mesh resolution, which limits scalability for dense topologies (> 10k vertices) and introduces surface noise. In contrast, SHELLS decouples feature extraction from mesh resolution via a hierarchical sampling strategy. We extract multi-view features using a DINOv2 backbone with LoRA adaptation, projectively sample a sparse global feature cloud, and predict an intermediate coarse mesh. This coarse prior guides the construction of layered, surface-aware sampling shells that serve as a discrete search space for the final reconstruction. SHELLS maintains surface consistency while using 88% less inference GPU memory (2.4GB vs. 20GB) than volumetric baselines. It reduces median registration error by 21% to 29% with a 3.5x inference speedup (0.08s vs. 0.29s) for 18k-vertex meshes. Notably, our model is trained exclusively on synthetic data yet generalizes effectively to real-world captures, eliminating the need for the costly, pre-registered multi-view datasets common in prior work.", "date": "2026-05-29", "categories": []}
{"id": "2605.25861", "title": "MuNet: A Mutualistic Network for Joint 3D Human Mesh Recovery and 3D Clothed Human Reconstruction from Single Images", "abstract": "3D human mesh recovery and 3D clothed human reconstruction are inherently related, yet they have long been studied in isolation, thereby overlooking the potential gains of joint optimization. To overcome this limitation, we propose to address these two tasks within a unified framework, which allows their mutual dependencies to be effectively exploited. Building on this idea, we propose MuNet, a mutualistic network for joint 3D human mesh recovery and 3D clothed human reconstruction from single images. First, we adopt 2-manifold graphs as a unified representation for all 3D models, enabling consistent modeling across 3D human mesh recovery and clothed human reconstruction. Second, we design an end-to-end graph convolutional network that progressively deforms an initial graph into a 3D human mesh and refines it into a detailed 3D clothed human model. Third, we introduce a mutualistic mechanism that allows reciprocal interaction between the two tasks {during training}, where 3D human mesh recovery provides guidance for 3D clothed human reconstruction, and reconstruction feedback refines the 3D human mesh recovery. We extensively evaluate MuNet on six benchmark datasets for 3D human mesh recovery and 3D clothed human reconstruction, including Human3.6M, 3DPW, MPI-INF-3DHP, THuman2.0, CAPE, and RenderPeople. Experimental results demonstrate that MuNet achieves state-of-the-art performance on both tasks across all datasets. The code of MuNet is released for research purposes at https://github.com/starVisionTeam/MuNet.", "date": "2026-05-25", "categories": []}
{"id": "2605.25418", "title": "Generating 3D models from sketches of human faces using a combined approach of Convolutional Neural Networks, Procedural Modeling, and Contour Mapping", "abstract": "Generating 3D models from face sketches is an active topic of research in Computer Graphics due to its potential to tremendously facilitate the modeling of faces for both professional 3D arists and novices. Motivated by the observation that facial expressions are responsible for significantly altering and shaping the contours in our faces, we combine both expression detection and 3D model generation in our approach. The result is a novel approach to generating 3D models from sketches which relies on three components: Convolutional Neural Networks, a parametric 3D face model (Valley Girl), and Active Snake Contours. For the first time in the literature, CNNs are trained (using our own generated dataset) to detect the expression in the given sketch through detecting the active FACS Action Units. The expression is then duplicated on Valley Girl to obtain a 3D model with a similar expression. Active Snake Contours are then used to find the transforms needed to close the gaps between that model and the given sketch.", "date": "2026-05-25", "categories": []}
{"id": "2605.25751", "title": "SplitAvatar: One-shot Head Avatar with Autoregressive Gaussian Splitting", "abstract": "3D Gaussian Splatting (3DGS) provides an efficient method for high-quality scene reconstruction using anisotropic Gaussians. Recently, 3DGS-based methods have significantly improved the rendering quality of human avatars while enabling real-time performance. However, existing methods suffer from a magnitude mismatch in the number of Gaussians generated by image-based and 3DMM-based approaches. This discrepancy results in reconstructed expressions that lack fine-grained detail. In this paper, we introduce a novel method for reconstructing an animatable head avatar from a single image. We propose a Graph splitting network to progressively generate Gaussians from coarse to fine using an autoregressive architecture. To address the graph inconsistency caused by split Gaussians, we employ a mesh topology extension method to align the GNN's connectivity with the increased Gaussian count. Furthermore, we introduce a novel density control method that includes a gating mechanism that generates soft masks for Gaussians, preventing over-densification after the splitting operation. This allows for dynamic control over Gaussian density across different facial regions. For smooth and rapid training, we employ a delayed filtering strategy to avoid re-computing the graph topology during training. Experimental results demonstrate that our autoregressive structure effectively improves expression representation ability by progressively splitting Gaussians. This process, enabled by the GNN-guided splitting, synthesizes more precise facial details and achieves higher reconstruction quality.", "date": "2026-05-25", "categories": []}
{"id": "2605.23064", "title": "Millimeter-wave Imaging for Anthropometric Body Measurement", "abstract": "Body shape and circumferences are clinically informative biomarkers for risk stratification, including measures such as waist to hip ratio, limb and trunk girths, yet conventional tools such as manual tape measures and optical scanners often require undressing and sustained poses. These demands slow workflows, compromise dignity, and exclude many older adults and people with limited mobility. To make measurement fast and contactless, we leverage millimeter-wave (mmWave) radar, which preserves privacy and operates through typical clothing, enabling quick full-body acquisition. In this work, we present a new optimization-based framework to recover 3D human shape and extract a comprehensive set of anthropometric measurements from volumetric mmWave data. Our method introduces a weighted registration pipeline that fits a parametric body model (SMPL) directly to the noisy mmWave point cloud. The core of our contribution is a vertex-weighting strategy that modulates a Chamfer energy function for reliable surface alignment and noise elimination. We further stabilize the fit by incorporating a foot-ground plane constraint and pose priors, optimizing directly for the SMPL parameters. Together, these components enable a fast, privacy preserving workflow that delivers high fidelity body shape and measurements through clothing without cameras or disrobing and with minimal cooperation, supporting frequent risk oriented assessments in clinics and care facilities for patients of all ages and mobility levels.", "date": "2026-05-21", "categories": []}
{"id": "2605.23555", "title": "Generator-Refiner-Examiner: A Tri-Module Data Augmentation Framework for 3D Human Avatar Learning from Monocular Videos", "abstract": "This paper addresses the challenge of reconstructing photorealistic and animatable 3D human avatars from monocular videos. While existing methods rely on combining per-subject optimization with generic human priors, they often fail to capture fine-grained details when training frames are limited. To mitigate this data scarcity, we propose TrioMan, a systematic tri-module framework for augmented 3D avatar learning. Our approach comprises three synergistic components. The Generator creates diverse unseen samples by imposing Gaussian perturbations on pose and camera. The Refiner improves the quality of generated data through one-step diffusion guided by texture and geometry cues. The Examiner selects subject-consistent samples using a dual-branch attention-based similarity evaluation. Experiments on the X-Humans and NeuMan benchmarks show that TrioMan outperforms state-of-the-art methods.", "date": "2026-05-22", "categories": []}
{"id": "2601.07603", "title": "UIKA: Fast Universal Head Avatar from Pose-Free Images", "abstract": "We present UIKA, a feed-forward animatable Gaussian head model from an arbitrary number of pose-free inputs, including a single image, multi-view captures, and smartphone-captured videos. Unlike the traditional avatar method, which requires a studio-level multi-view capture system and reconstructs a human-specific model through a long-time optimization process, we rethink the task through the lenses of model representation, network design, and data preparation. First, we introduce a UV-guided avatar modeling strategy, in which each input image is associated with a pixel-wise facial correspondence estimation. Such correspondence estimation allows us to reproject each valid pixel color from screen space to UV space, which is independent of camera pose and character expression. Furthermore, we design learnable UV tokens on which the attention mechanism can be applied at both the screen and UV levels. The learned UV tokens can be decoded into canonical Gaussian attributes using aggregated UV information from all input views. To train our large avatar model, we additionally prepare a large-scale, identity-rich synthetic training dataset. Our method significantly outperforms existing approaches in both monocular and multi-view settings.", "date": "2026-01-12", "categories": []}
{"id": "2605.02784", "title": "HumanSplatHMR: Closing the Loop Between Human Mesh Recovery and Gaussian Splatting Avatar", "abstract": "Accurately recovering human pose and appearance from video is an essential component of scene reconstruction, with applications to motion capture, motion prediction, virtual reality, and digital twinning. Despite significant interest in building realistic human avatars from video, this paper demonstrates that existing methods do not accurately recover the 3D geometry of humans. ViT-based approaches are not consistently reliable and can overfit to 2D views, while NeRF- and Gaussian Splatting-based avatars treat pose and appearance separately, limiting rendering generalization to new poses. To resolve these shortcomings, this paper proposes HumanSplatHMR, a joint optimization framework that refines 3D human poses while simultaneously learning a high-fidelity avatar for novel-view and novel-pose synthesis. Our key insight is to close the loop between geometric pose estimation and differentiable rendering. Unlike prior human avatar methods that rely on accurate human pose obtained through motion capture systems or offline refinement, which are impractical in in-the-wild scenarios, our approach uses only human mesh estimates from a state-of-the-art human pose estimator to better reflect real-world conditions. Therefore, instead of using the human pose only as a deformation prior, HumanSplatHMR backpropagates photometric, segmentation, and depth losses through a differentiable renderer to the pose parameters and global position. This coupling refines the global 3D pose over time, improving accuracy and alignment while producing better renderings from novel views. Experiments show consistent improvements over pose recovery baselines that omit image-level refinement and avatar baselines that decouple pose estimation from avatar reconstruction.", "date": "2026-05-04", "categories": []}
{"id": "2504.19584", "title": "ShowMak3r: Compositional TV Show Reconstruction", "abstract": "Reconstructing dynamic radiance fields from video clips is challenging, especially when entertainment videos like TV shows are given. Many challenges make the reconstruction difficult due to (1) actors occluding with each other and having diverse facial expressions, (2) cluttered stages, and (3) small baseline views or sudden shot changes. To address these issues, we present ShowMak3r, a comprehensive reconstruction pipeline that allows the editing of scenes like how video clips are made in a production control room. In ShowMak3r, a 3DLocator module locates recovered actors on the stage using depth prior and estimates unseen human poses via interpolation. The proposed ShotMatcher module then tracks the actors under shot changes. Furthermore, ShowMak3r introduces a face-fitting network that dynamically recovers the actors' expressions. Experiments on Sitcoms3D dataset show that our pipeline can reassemble TV show scenes with new cameras at different timestamps. We also demonstrate that ShowMak3r enables interesting applications such as synthetic shot-making, actor relocation, insertion, deletion, and pose manipulation. Project page : https://nstar1125.github.io/showmak3r", "date": "2025-04-28", "categories": []}
{"id": "2605.20185", "title": "PiG-Avatar: Hierarchical Neural-Field-Guided Gaussian Avatars", "abstract": "Existing Gaussian avatar methods typically parameterize geometry on a body-template surface, which entangles the avatar's representation space with the template's deformation space and limits the capture of layered, off-body, and non-rigid clothing geometry. We present PiG-Avatar, which addresses this limitation by using the parametric body model solely for kinematic transport, while representing the avatar as Gaussians anchored in a volumetric canonical space governed by a continuous neural field. This decouples representation from template topology, avoiding the geometric constraints of surface-based parameterizations. Kinematic coherence is maintained through 3D barycentric anchor transport, which guides motion without constraining geometry and allows anchors to deviate freely from the template surface, yielding dense, stable temporal surface correspondences by construction. To make this unconstrained formulation tractable, we introduce dual-level spatially coherent optimization, combining Sobolev-preconditioned neural-field updates with a novel KNN-based preconditioning of canonical anchor geometry. Together, these mechanisms induce an emergent self-organization of anchor density: anchors migrate toward regions of high curvature, appearance variation, and non-coherent motion without explicit heuristics. As a result, complex clothing geometry and layered surfaces emerge as natural, high-fidelity outputs. This single representation further supports hierarchical reconstruction across multiple levels of detail, with coarse-level supervision propagating to finer levels through the shared field and coupled anchor graph. On established benchmarks featuring subjects with complex clothing and challenging non-rigid motion, PiG-Avatar achieves state-of-the-art rendering quality, generalizes robustly to imperfect body model initialization, and renders in real time across all detail levels.", "date": "2026-05-19", "categories": []}
{"id": "2605.21001", "title": "DAMA: Disentangled Body-Anchored Gaussians for Controllable Multi-Layered Avatars", "abstract": "Existing 3D clothed avatar reconstruction methods achieve high visual fidelity but ignore geometric structure and physical plausibility. They either model clothed humans as a single deformable surface or attempt garment disentanglement without enforcing geometric constraints, resulting in ambiguous garment boundaries and no control over stacking or layer ordering. To address these limitations, we introduce DAMA (Disentangled body-Anchored Gaussians for Controllable Multi-layered Avatars), a 3D avatar reconstruction method that produces physically plausible clothed avatars through a dedicated representation and reconstruction method. At the representation level, we bind Gaussians to SMPL-X faces using barycentric in-plane coordinates and a positive normal offset. Based on this parameterization, the reconstruction method lifts 2D segmentations to body-anchored Gaussians, refines layers using topology-guided correction, and jointly optimizes geometry and appearance. DAMA is the first Gaussian avatar reconstruction method from multi-view images to achieve physically plausible layering, clean garment separation, and explicit stacking control. On the full 4D-DRESS dataset (82 scans), it achieves state-of-the-art performance in geometry reconstruction, garment separation, penetration rate, and penetration depth. The representation further supports user-defined garment reordering and fast conversion of body-conforming garments to simulation-ready meshes. Project Page: https://danieleskandar.github.io/dama/", "date": "2026-05-20", "categories": []}
{"id": "2605.17014", "title": "RHINO: Reconstructing Human Interactions with Novel Objects from Monocular Videos", "abstract": "Reconstructing people, objects, and their interactions in 3D is a long-standing goal for intelligent systems. Often the input is RGB video from a moving camera, making the task ill-posed; depth is ambiguous, humans and objects occlude each other, and camera and object motion entangle to create apparent motion. Most prior work addresses humans or objects in isolation, ignoring their interplay, or assumes known 3D shapes or cameras, which is impractical for real-world applications. We develop RHINO (Reconstructing Human Interactions with Novel Objects), a three-step framework that recovers in 3D a human, novel (unseen) manipulated object, and static scene in a common world frame from a monocular RGB video. First, we leverage 3D-aware foundation models to obtain cues that stabilize Structure-from-Motion (SfM) even for low-texture regions; this yields a coarse shape and apparent motion of a manipulated object from foreground pixels, and a coarse scene shape and camera motion from background pixels. Second, we estimate a human in the camera frame via an off-the-shelf method, and subtract the camera motion from apparent motion to extract the object motion; this registers the human, object, and coarse scene shapes into a common world frame. Third, we refine shapes using a compositional neural field with per-component signed-distance fields. The latter further enables differentiable contact priors that attract surfaces while penalizing interpenetration, improving the physical plausibility of the final reconstruction. For evaluation, we capture a new dataset of handheld monocular videos synchronized with a volumetric 4D capture stage, providing ground-truth shape and camera motion. RHINO outperforms state-of-the-art baselines on novel-view synthesis and 4D reconstruction. Ablations show that each stage contributes substantially. Code and data are available at https://lxxue.github.io/RHINO.", "date": "2026-05-16", "categories": []}
{"id": "2605.15320", "title": "FFAvatar: Few-Shot, Feed-Forward, and Generalizable Avatar Reconstruction", "abstract": "Avatar reconstruction has traditionally relied on per-subject optimization that requires hours of computation or on expensive preprocessing that limits scalability. We introduce FFAvatar, a generalizable feed-forward framework that reconstructs high-quality, animatable 3D Gaussian head avatars from few-shot unposed portrait images in seconds. FFAvatar fuses information from multiple source images into a unified canonical Gaussian representation through Multi-View Query-Former, which is animated via FLAME parameters predicted end-to-end directly from pixels, eliminating the overhead of offline FLAME extraction. We further propose a three-stage training curriculum that achieves both broad generalization and high-fidelity reconstruction: (i) scalable pretraining on extensive monocular video data with over 1M identities to learn strong generalizable priors; (ii) multi-view fine-tuning on a small but high-quality dataset of 360-degree captures to enhance geometric fidelity and extreme-view awareness; and (iii) optional personalization that adapts to specific identities for maximum fidelity within 500 optimization steps. Extensive experiments demonstrate that FFAvatar sets a new standard for identity preservation, geometric consistency, and animation fidelity. On the NeRSemble benchmark, it outperforms the state-of-the-art LAM by a substantial 5.5 PSNR gain. Furthermore, FFAvatar enables real-time deployment, reconstructing avatars in 2 seconds without personalization and 10 seconds with personalization, while supporting 49 FPS animation on a single NVIDIA A100 GPU.", "date": "2026-05-14", "categories": []}
{"id": "2605.04554", "title": "InterMesh: Explicit Interaction-Aware End-to-End Multi-Person Human Mesh Recovery", "abstract": "Humans constantly interact with their surroundings. Existing end-to-end multi-person human mesh recovery methods, typically based on the DETR framework, capture inter-human relationships through self-attention across all human queries. However, these approaches model interactions only implicitly and lack explicit reasoning about how humans interact with objects and with each other. In this paper, we propose InterMesh, a simple yet effective framework that explicitly incorporates human-environment interaction information into human mesh recovery pipeline. By leveraging a human-object interaction detector, InterMesh enriches query representations with structured interaction semantics, enabling more accurate pose and shape estimation. We design lightweight modules, Contextual Interaction Encoder and Interaction-Guided Refiner, to integrate these features into existing HMR architectures with minimal overhead. We validate our approach through extensive experiments on 3DPW, MuPoTS, CMU Panoptic, Hi4D, and CHI3D datasets, demonstrating remarkable improvements over state-of-the-art methods. Notably, InterMesh reduces MPJPE by 9.9% on CMU Panoptic and 8.2% on Hi4D, highlighting its effectiveness in scenarios with complex human-object and inter-human interactions. Code and models are released at https://github.com/Kelly510/InterMesh.", "date": "2026-05-06", "categories": []}
{"id": "2605.13854", "title": "Contrastive Multi-Modal Hypergraph Reasoning for 3D Crowd Mesh Recovery", "abstract": "Multi-person 3D reconstruction is pivotal for real-world interaction analysis, yet remains challenging due to severe occlusions and depth ambiguity. Current approaches typically rely on single-modality inputs, which inherently lack geometric guidance. Furthermore, these methods often reconstruct subjects in isolation, neglecting the collective group context essential for resolving ambiguities in crowded scenes. To address these limitations, we propose Contrastive Multi-modal Hypergraph Reasoning to synergize semantic, geometric, and pose cues for crowd reconstruction. We first initialize robust node representations by combining RGB features, geometric priors, and occlusion-aware incomplete poses. Additionally, we introduce a pelvis depth indicator as a global spatial anchor, aligning visual features with a metric-scale-agnostic depth ordering. Subsequently, we construct a shared-topology hypergraph that moves beyond pairwise constraints to model higher-order crowd dynamics. To improve feature fusion, we design a hypergraph-based contrastive learning scheme that jointly enhances intra-modal discriminability and enforces cross-modal orthogonality. This mechanism enables the network to propagate global context effectively, allowing it to infer missing information even under severe occlusion. Extensive experiments on the Panoptic and GigaCrowd benchmarks confirm that our method achieves new state-of-the-art performance. Code and pre-trained models are available at https://github.com/SunMH-try/CoMHR.", "date": "2026-04-01", "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": "2604.19923", "title": "UniCon3R: Unified Contact-aware 4D Human-Scene Reconstruction from Monocular Video", "abstract": "We introduce UniCon3R, a unified feed-forward framework for online human-scene 4D reconstruction from monocular video. Current feed-forward human-scene reconstruction methods suffer from artifacts, where bodies float above the ground or penetrate parts of the scene. A key reason is the lack of effective interaction modelling between the human and the environment. Our goal is to exploit contact between the human and the scene during inference to actively improve the human mesh reconstruction. To that end, we explicitly model interaction by inferring 4D contact from the human pose and scene geometry and use the contact as a corrective cue for generating the pose. This enables UniCon3R to jointly recover scene geometry and spatially aligned 4D humans within the scene. Experiments on standard human-centric video benchmarks show that UniCon3R outperforms state-of-the-art baselines on physical plausibility and global human motion estimation while preserving fast, feed-forward inference speeds. The results validate our central claim: contact serves as a powerful internal prior, thus establishing a new paradigm for physically grounded joint human-scene reconstruction. Project page is available at https://surtantheta.github.io/UniCon3R .", "date": "2026-04-21", "categories": []}
{"id": "2605.09856", "title": "MoPO: Incorporating Motion Prior for Occluded Human Mesh Recovery", "abstract": "Although recent studies have made remarkable progress in human mesh recovery, they still exhibit limited robustness to occlusions and often produce inaccurate poses and severe motion jitter due to the insufficient spatial features for occluded body parts. Inspired by the rapid advancements in human motion prediction, we discover that compared to occluded image features, pose sequence inherently contains reliable motion prior for estimating occluded body parts. In this paper, we incorporate Motion Prior for Occluded human mesh recovery, called MoPO. Our MoPO mainly consists of two components: 1) The motion de-occlusion module, where we propose a spatial-temporal occlusion detector to detect joint visibility, and then we propose a lightweight motion predictor to complete the occluded body parts by predicting the most plausible joint positions based on history poses. 2) The motion-aware fusion and refinement module, which fuses the completed joint sequence with image features to estimate human shape and initial human pose. Moreover, the completed joint sequence is further used to refine the final human pose through inverse kinematics, which provides the occlusion-free motion prior for regressing human poses. Extensive experiments demonstrate that MoPO achieves state-of-the-art performance on both occlusion-specific and standard benchmarks, significantly enhancing the accuracy and temporal consistency of occluded human mesh recovery. Our code and demo can be found in the supplementary material.", "date": "2026-05-11", "categories": []}
{"id": "2605.09956", "title": "SDTalk: Structured Facial Priors and Dual-Branch Motion Fields for Generalizable Gaussian Talking Head Synthesis", "abstract": "High-quality, real-time talking head synthesis remains a fundamental challenge in computer vision. Existing reconstruction- and rendering-based methods typically rely on identity-specific models, limiting cross-identity generalization. To address this issue, we propose SDTalk, a one-shot 3D Gaussian Splatting (3DGS)-based framework that generalizes to unseen identities without personalized training or fine-tuning. Our framework comprises two modules with a two-stage training strategy. In the first stage, we incorporate structured facial priors into the reconstruction module and separately predict 3DGS parameters for visible and occluded regions, enabling complete head reconstruction from a single image. In the second stage, we introduce a dual-branch motion field to model coarse and fine facial dynamics, improving detail fidelity and lip synchronization. Experiments demonstrate that SDTalk surpasses existing methods in both visual quality and inference efficiency.", "date": "2026-05-11", "categories": []}
{"id": "2605.04035", "title": "Large-Scale High-Quality 3D Gaussian Head Reconstruction from Multi-View Captures", "abstract": "We propose HeadsUp, a scalable feed-forward method for reconstructing high-quality 3D Gaussian heads from large-scale multi-camera setups. Our method employs an efficient encoder-decoder architecture that compresses input views into a compact latent representation. This latent representation is then decoded into a set of UV-parameterized 3D Gaussians anchored to a neutral head template. This UV representation decouples the number of 3D Gaussians from the number and resolution of input images, enabling training with many high-resolution input views. We train and evaluate our model on an internal dataset with more than 10,000 subjects, which is an order of magnitude larger than existing multi-view human head datasets. HeadsUp achieves state-of-the-art reconstruction quality and generalizes to novel identities without test-time optimization. We extensively analyze the scaling behavior of our model across identities, views, and model capacity, revealing practical insights for quality-compute trade-offs. Finally, we highlight the strength of our latent space by showcasing two downstream applications: generating novel 3D identities and animating the 3D heads with expression blendshapes.", "date": "2026-05-05", "categories": []}
{"id": "2604.08548", "title": "ETCH-X: Robustify Expressive Body Fitting to Clothed Humans with Composable Datasets", "abstract": "Human body fitting, which aligns parametric body models such as SMPL to raw 3D point clouds of clothed humans, serves as a crucial first step for downstream tasks like animation and texturing. An effective fitting method should be both locally expressive-capturing fine details such as hands and facial features-and globally robust to handle real-world challenges, including clothing dynamics, pose variations, and noisy or partial inputs. Existing approaches typically excel in only one aspect, lacking an all-in-one solution. We upgrade ETCH to ETCH-X, which leverages a tightness-aware fitting paradigm to filter out clothing dynamics (\"undress\"), extends expressiveness with SMPL-X, and replaces explicit sparse markers (which are highly sensitive to partial data) with implicit dense correspondences (\"dense fit\") for more robust and fine-grained body fitting. Our disentangled \"undress\" and \"dense fit\" modular stages enable separate and scalable training on composable data sources, including diverse simulated garments (CLOTH3D), large-scale full-body motions (AMASS), and fine-grained hand gestures (InterHand2.6M), improving outfit generalization and pose robustness of both bodies and hands. Our approach achieves robust and expressive fitting across diverse clothing, poses, and levels of input completeness, delivering a substantial performance improvement over ETCH on both: 1) seen data, such as 4D-Dress (MPJPE-All, 33.0% ) and CAPE (V2V-Hands, 35.8% ), and 2) unseen data, such as BEDLAM2.0 (MPJPE-All, 80.8% ; V2V-All, 80.5% ). Code and models will be released at https://xiaobenli00.github.io/ETCH-X/.", "date": "2026-04-09", "categories": []}
{"id": "2605.04524", "title": "High-Fidelity Single-Image Head Modeling with Industry-Grade Topology", "abstract": "We present a single-image head mesh reconstruction framework that addresses the longstanding challenge of simultaneously preserving facial identity and producing industry-grade topology. Our framework adopts a coarse-to-fine optimization pipeline that refines a rigged template across three stages -- rig, joint, and vertex -- achieving stable convergence and consistent topology. To mitigate the ill-posed nature of single-image 3D face reconstruction and ensure identity preservation, we employ a normal consistency objective jointly with landmark alignment. To further preserve local surface structure and enforce topological regularity, we introduce geometry-aware constraints based on Gaussian curvature and conformal consistency, along with auxiliary regularizations that correct fine artifacts such as lip seams and eyelid discontinuities. Our hierarchical optimization with geometry-aware regularization yields meshes with semantically meaningful edge flow and industry-grade topology. After geometry reconstruction, we extract UV-space texture and normal maps to preserve appearance details for visualization and downstream use. In a user study with 22 professional technical artists, our results were assessed as approaching industry-grade usability, and 95% of participants ranked our method as the top-performing approach, underscoring its effectiveness for real-world digital human production.", "date": "2026-05-06", "categories": []}
{"id": "2605.04728", "title": "Anny-Fit: All-Age Human Mesh Recovery", "abstract": "Recovering 3D human pose and shape from a single image remains a cornerstone of human-centric vision, yet most methods assume adult subjects and optimize each person independently. These assumptions fail in real-world, all-age scenes, where body proportions and depth must be resolved jointly. We introduce Anny-Fit, a multi-person, camera-space optimization framework for all-age 3D human mesh recovery (HMR). Unlike existing per-person fitting methods, Anny-Fit jointly optimizes all individuals directly in the camera coordinate system, enforcing global spatial consistency. At the core of our approach is the use of multiple forms of expert knowledge -- including metric depth maps, instance segmentation, 2D keypoints, and, VLM-derived semantic attributes such as age and gender -- each obtained from dedicated off-the-shelf networks. These complementary signals jointly guide the optimization, constraining the depth-scale ambiguity characteristic of all-age scenes. Across diverse datasets, Anny-Fit consistently improves 2D reprojection accuracy (+13 to 16), relative depth ordering (+6 to 7), 3D estimation error (-9 to -29) and shape estimation (+25 to +82), producing more coherent scenes. Finally, we show that VLM-based semantic knowledge can be distilled into an HMR model via the pseudo-ground-truth annotations produced by Anny-Fit on training data, enabling it to learn semantically meaningful shape parameters while improving HMR performance. Our approach bridges adult-only and all-age modeling by enabling zero-shot adaptation of adult-trained HMR pipelines to the full age spectrum without retraining. Code is publicly available at https://github.com/naver/anny-fit.", "date": "2026-05-06", "categories": []}
{"id": "2605.03996", "title": "3D Human Face Reconstruction with 3DMM face model from RGB image", "abstract": "Nowadays as convolution neural networks demonstrate its powerful problem-solving ability in the area of image processing, efforts have been made to reconstruct detailed face shapes from 2D face images or videos. However, to make the full use of CNN, a large number of labeled data is required to train the network. Coarse morphable face model has been used to synthesize labeled data. However, it is hard for coarse morphable face models to generate photo-realistic data with detail such as wrinkles. In this project, we present a pipeline that reconstructs a human face 3D model from a single RGB image. The pipeline includes face detection, landmark detection, regression of 3DMM model parameters, and soft rendering. Mentor: Zhipeng Fan (Email: zf606@nyu.edu) Code Repository: https://github.com/SeVEnMY/3d-face- reconstruction Code Reference: https://github.com/sicxu/Deep3DFaceRecon pytorch", "date": "2026-05-05", "categories": []}
{"id": "2601.06035", "title": "Investigating Anthropometric Fidelity in SAM 3D Body", "abstract": "The release of SAM 3D Body is a recent development in human mesh recovery, demonstrating improved performance in producing clean, topologically coherent meshes from single images. By leveraging the Momentum Human Rig (MHR), it achieves robustness to occlusion and diverse poses. However, our evaluation reveals a specific and consistent limitation: the model struggles to reconstruct detailed anthropometric deviations, particularly in populations exhibiting distinctive morphological alterations such as geriatric muscle atrophy, scoliosis, or pregnancy, even when these features are prominent in the input image. In this paper, we investigate this phenomenon not as a failure of the model's capacity, but as a byproduct of the \"perception-distortion trade-off\". We posit that the architectural reliance on the low-dimensional parametric MHR representation, combined with semantic-invariant conditioning (DINOv3) and annotation-based alignment, creates a pervasive \"regression to the mean\" effect. We analyze these mechanisms to understand why individual biological details are smoothed out. Furthermore, we state our contributions by proposing specific, constructive pathways for future work, such as implicit-explicit hybrid representations and Medical-in-the-Loop alignment, to extend the baseline performance of SAM 3D Body into the high-precision medical domain.", "date": "2025-12-02", "categories": []}
{"id": "2605.01450", "title": "Registration-Free Learnable Multi-View Capture of Faces in Dense Semantic Correspondence", "abstract": "Recent frameworks like ToFu and TEMPEH provide an automated alternative to classical registration pipelines by predicting 3D meshes in dense semantic correspondence directly from calibrated multi-view images. However, these learning-based methods rely on the slow, manual registration pipelines they aim to replace for their training supervision. We overcome this limitation with MOCHI (Multi-view Optimizable Correspondence of Heads from Images), a multi-view 3D face prediction framework trained without requiring registered training data. MOCHI eliminates the registration data dependency by enforcing topological consistency through a pseudo-linear inverse kinematic solver. Semantic alignment is guided by dense keypoints from a 2D landmark predictor trained exclusively on synthetic data. Our analysis further reveals that standard point-to-surface distances induce training instabilities and visual artifacts in registration-free settings. We propose pointmap- and normal-based losses instead, which provide smoother gradients and superior reconstruction fidelity. Finally, we introduce a test-time optimization scheme that refines network weights over a few dozen iterations. This approach bridges the gap between feed-forward efficiency and iterative optimization precision, allowing MOCHI to outperform traditional labor-intensive pipelines in both reconstruction accuracy and visual quality. Code and model are public at: https://filby89.github.io/mochi.", "date": "2026-05-02", "categories": []}
{"id": "2605.01746", "title": "Profile-Specific 3DMM Regression from a Single Lateral Face Image", "abstract": "Single-image 3D face reconstruction is a core problem in computer vision, with important clinical applications such as cephalometric landmark analysis in orthodontics. Traditionally, this analysis relies on lateral X-ray imaging; however, frequent X-ray exposure is impractical due to radiation concerns. While recent research has explored detecting landmarks from lateral RGB images as an alternative, existing methods typically rely on 2D features such as the eyes, mouth, ears, and boundary silhouettes, failing to fully exploit the underlying 3D facial geometry spanning the facial profile and jawline, which is essential for accurate diagnosis. Meanwhile, although 3D face reconstruction from frontal views has seen significant progress, most learning-based 3D morphable model (3DMM) regressors are developed and benchmarked on near-frontal images, where appearance cues are abundant. In extreme profile views (yaw $\\approx 90^\\circ$), much of the face is occluded, and the available signal is dominated by boundary cues, making accurate 3D reconstruction challenging. In this paper, we bridge this gap with geometry-conditioned synthetic data and a simple profile-specific FLAME regression baseline for single lateral images. We introduce ProfileSynth, a dataset created by sampling FLAME shape and pose parameters in extreme yaw ranges and generating photorealistic profile images using a diffusion model conditioned on depth and normal maps. We further study a profile-specific baseline with visibility-aware jawline regularization. Our framework provides a practical baseline for \"profile $\\times$ 3DMM\" reconstruction and a promising foundation for more accurate, non-invasive cephalometric analysis from lateral RGB images.", "date": "2026-05-03", "categories": []}
{"id": "2605.01854", "title": "High-Fidelity Mobile Avatars with Pruned Local Blendshapes", "abstract": "We propose a method to reconstruct high-fidelity human avatars from multi-view video that can run on mobile devices. Many works can model high-quality Gaussian-based full-body avatars from multi-view video. However, these methods require heavy computation to obtain pose-dependent appearance, making deployment on mobile devices very difficult. Recent methods distill from pretrained models and model pose-dependent nonlinear Gaussian attributes by linearly combining global pose features with blendshapes. Although they can run on mobile devices, they suffer some loss of detail. We observe that nearby Gaussians are often highly correlated within a local region of the body, and can be linearly modeled with less error. Therefore, we use local linear blendshapes in small body parts to capture global nonlinear changes of Gaussian attributes. To further reduce computation and model size, we propose to remove blendshapes for Gaussians whose attributes change little, yielding a minimal blendshape representation. Our method is an end-to-end training method without a pretrained model. To make it run on multiple devices, we implement our method using WebGPU. Experiments show that our method can render high-quality human avatars with better details, and can reach 120 FPS at 2K resolution on mobile devices.", "date": "2026-05-03", "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": "2604.28025", "title": "ResiHMR: Residual-Limb Aware Single-Image 3D Human Mesh Recovery for Individuals with Limb Loss", "abstract": "Single-image human mesh recovery provides a compact 3D, person-centric representation that supports analysis, animation, AR and VR, rehabilitation, and human-computer interaction. However, prevailing systems impose an intact-limb prior and degrade on people with limb loss, because fixed-topology models cannot represent residual limbs. In this work, we present ResiHMR, a residual-limb aware framework for single-image 3D human modeling. ResiHMR adopts residual-limb keypoints and introduces two components: (i) a topology-adaptive Residual Anchor-Factor Optimization module that constrains estimation to the observed kinematic subgraph of anatomically valid structures, and (ii) a geometry-based Residual-Limb Reconstruction module that estimates residual-limb boundaries and convex limb-termination geometry. These components introduce topology-aware optimization and explicit termination geometry as tools for human mesh recovery under non-standard limb anatomy. Unlike joint-removal methods in a fixed topology, ResiHMR explicitly reconstructs residual-limb surfaces and aligns optimization with limb-loss topology, which better matches prosthetic biomechanics and real-world use. To the best of our knowledge, this is the first single-image HMR system that explicitly reconstructs residual-limb surfaces and performs topology-adaptive optimization for individuals with limb loss. On a curated dataset of real-world images with limb loss, ResiHMR improves reconstruction quality under both SMPLify-X and HSMR backbones, reducing intact-joint 2D MPJPE from 41.32 to 37.40 with SMPLify-X and residual-limb 2D MPJPE from 73.61 to 23.19 with HSMR.", "date": "2026-04-30", "categories": []}
{"id": "2604.25466", "title": "Generalizable Human Gaussian Splatting via Multi-view Semantic Consistency", "abstract": "Recently, generalizable human Gaussian splatting from sparse-view inputs has been actively studied for the photorealistic human rendering. Most existing methods rely on explicit geometric constraints or predefined structural representations to accurately position 3D Gaussians. Although these approaches have shown the remarkable progress in this field, they still suffer from inconsistent feature representations across multi-view inputs due to complex articulations of the human body and limited overlaps between different views. To address this problem, we propose a novel method to accurately localize 3D Gaussians and ultimately improve the quality of human rendering. The key idea is to unproject latent embeddings encoded from each viewpoint into a shared 3D space through predicted depth maps and recalibrate them belonging to the same body part based on cross-view attention. This helps the model resolve the spatial ambiguity occurring in highly textured regions as well as occluded body parts, thus leading to the accurate localization of 3D Gaussians. Experimental results on benchmark datasets show that the proposed method efficiently improves the performance of generalizable human Gaussian splatting from sparse-view inputs.", "date": "2026-04-28", "categories": []}
{"id": "2604.22827", "title": "DGHMesh: A Large-scale Dual-radar mmWave Dataset and Generalization-focused Benchmark for Human Mesh Reconstruction", "abstract": "Millimeter-wave (mmWave) radar has shown great potential for contactless, privacy-preserving, and robust human sensing, yet existing mmWave-based human mesh reconstruction (HMR) studies are still limited by the lack of benchmarks for generalization analysis under configuration shifts and fair comparison of different algorithms. To address the limitation, we present DGHMesh, a large-scale dual-radar mmWave dataset and generalization-focused benchmark for HMR. It contains data from 15 subjects performing 8 actions, with 360,000 synchronized frames collected from FMCW radar, SFCW radar, RGB images, and high-precision 3D HMR annotations. In addition, the dataset provides synchronized raw I/Q data from both radar modalities and accurately calibrated radar spatial positions. The benchmark is designed to evaluate HMR methods under diverse measurement configurations, including human position shifts, human orientation shifts, subarray size variations, and cross-subject settings. Based on DGHMesh, we also propose mmPTM, a query-based multi-radar fusion framework that jointly exploits point clouds and imaging tubes for HMR. Extensive experiments are conducted against representative baselines under different settings. The results demonstrate that mmPTM consistently achieves outstanding accuracy and competitive generalization capability across multiple sub-benchmarks, validating the effectiveness of multi-radar fusion and the practical value of the proposed dataset and benchmark for mmWave-based HMR research. DGHMesh and mmPTM are publicly available at https://github.com/SPIresearch/DGHMesh.(The complete benchmark and code will be released after paper publication)", "date": "2026-04-19", "categories": []}
{"id": "2604.22865", "title": "MeshLAM: Feed-Forward One-Shot Animatable Textured Mesh Avatar Reconstruction", "abstract": "We introduce MeshLAM, a feed-forward framework for one-shot animatable mesh head reconstruction that generates high-fidelity, animatable 3D head avatars from a single image. Unlike previous work that relies on time-consuming test-time optimization or extensive multi-view data, our method produces complete mesh representations with inherent animatability from a single image in a single forward pass. Our approach employs a dual shape and texture map architecture that simultaneously processes mesh vertices and texture map with extracted image features from a shared transformer backbone, allowing for coherent shape carving and appearance modeling. To prevent mesh collapse and ensure topological integrity during feed-forward deformation, we propose an iterative GRU-based decoding mechanism with progressive geometry deformation and texture refinement, coupled with a novel reprojection-based texture guidance mechanism that anchors appearance learning to the input image. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in reconstruction quality, animation capability, and computational efficiency. Project page at https://meshlam.github.io.", "date": "2026-04-23", "categories": []}
{"id": "2511.02777", "title": "PercHead: Perceptual Head Model for Single-Image 3D Head Reconstruction & Editing", "abstract": "We present PercHead, a model for single-image 3D head reconstruction and disentangled 3D editing - two tasks that are inherently challenging due to ambiguity in plausible explanations for the same input. At the heart of our approach lies our novel perceptual loss based on DINOv2 and SAM 2.1. Unlike widely-adopted low-level losses like LPIPS, SSIM or L1, we rely on deep visual understanding of images and the resulting generalized supervision signals. We show that our new loss can be a drop-in replacement for standard losses and used to improve visual quality in high-frequency areas. We base our model architecture on Vision Transformers (ViTs), allowing us to decouple the 3D representation from the 2D input. We train our method on multi-view images for view-consistency and in-the-wild images for strong transferability to new environments. Our model achieves state-of-the-art performance in novel-view synthesis and, furthermore, exhibits exceptional robustness to extreme viewing angles. We also extend our base model to disentangled 3D editing by swapping the encoder and fine-tuning the network. A segmentation map controls geometry and either a text prompt or a reference image specifies appearance. We highlight the intuitive and powerful 3D editing capabilities through an interactive GUI. Project Page: https://antoniooroz.github.io/PercHead Video: https://www.youtube.com/watch?v=4hFybgTk4kE", "date": "2025-11-04", "categories": []}
{"id": "2604.21575", "title": "OmniFit: Multi-modal 3D Body Fitting via Scale-agnostic Dense Landmark Prediction", "abstract": "Fitting an underlying body model to 3D clothed human assets has been extensively studied, yet most approaches focus on either single-modal inputs such as point clouds or multi-view images alone, often requiring a known metric scale. This constraint is frequently impractical, especially for AI-generated assets where scale distortion is common. We propose OmniFit, a method that can seamlessly handle diverse multi-modal inputs, including full scans, partial depth observations, and image captures, while remaining scale-agnostic for both real and synthetic assets. Our key innovation is a simple yet effective conditional transformer decoder that directly maps surface points to dense body landmarks, which are then used for SMPL-X parameter fitting. In addition, an optional plug-and-play image adapter incorporates visual cues to compensate for missing geometric information. We further introduce a dedicated scale predictor that rescales subjects to canonical body proportions. OmniFit substantially outperforms state-of-the-art methods by 57.1 to 80.9 percent across daily and loose clothing scenarios. To the best of our knowledge, it is the first body fitting method to surpass multi-view optimization baselines and the first to achieve millimeter-level accuracy on the CAPE and 4D-DRESS benchmarks.", "date": "2026-04-23", "categories": []}
{"id": "2604.21712", "title": "Discriminative-Generative Synergy for Occlusion Robust 3D Human Mesh Recovery", "abstract": "3D human mesh recovery from monocular RGB images aims to estimate anatomically plausible 3D human models for downstream applications, but remains challenging under partial or severe occlusions. Regression-based methods are efficient yet often produce implausible or inaccurate results in unconstrained scenarios, while diffusion-based methods provide strong generative priors for occluded regions but may weaken fidelity to rare poses due to over-reliance on generation. To address these limitations, we propose a brain-inspired synergistic framework that integrates the discriminative power of vision transformers with the generative capability of conditional diffusion models. Specifically, the ViT-based pathway extracts deterministic visual cues from visible regions, while the diffusion-based pathway synthesizes structurally coherent human body representations. To effectively bridge the two pathways, we design a diverse-consistent feature learning module to align discriminative features with generative priors, and a cross-attention multi-level fusion mechanism to enable bidirectional interaction across semantic levels. Experiments on standard benchmarks demonstrate that our method achieves superior performance on key metrics and shows strong robustness in complex real-world scenarios.", "date": "2026-04-20", "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": "2604.19624", "title": "GRAFT: Geometric Refinement and Fitting Transformer for Human Scene Reconstruction", "abstract": "Reconstructing physically plausible 3D human-scene interactions (HSI) from a single image currently presents a trade-off: optimization based methods offer accurate contact but are slow (~20s), while feed-forward approaches are fast yet lack explicit interaction reasoning, producing floating and interpenetration artifacts.\n  Our key insight is that geometry-based human--scene fitting can be amortized into fast feed-forward inference. We present GRAFT (Geometric Refinement And Fitting Transformer), a learned HSI prior that predicts Interaction Gradients: corrective parameter updates that iteratively refine human meshes by reasoning about their 3D relationship to the surrounding scene.\n  GRAFT encodes the interaction state into compact body-anchored tokens, each grounded in the scene geometry via Geometric Probes that capture spatial relationships with nearby surfaces.\n  A lightweight transformer recurrently updates human meshes and re-probes the scene, ensuring the final pose aligns with both learned priors and observed geometry.\n  GRAFT operates either as an end-to-end reconstructor using image features, or with geometry alone as a transferable plug-and-play HSI prior that improves feed-forward methods without retraining.\n  Experiments show GRAFT improves interaction quality by up to 113% over state-of-the-art feed-forward methods and matches optimization-based interaction quality at ${\\sim}50{\\times}$ lower runtime, while generalizing seamlessly to in-the-wild multi-person scenes and being preferred in 64.8% of three-way user study. Project page: https://pradyumnaym.github.io/graft .", "date": "2026-04-21", "categories": []}
{"id": "2604.19702", "title": "Face Anything: 4D Face Reconstruction from Any Image Sequence", "abstract": "Accurate reconstruction and tracking of dynamic human faces from image sequences is challenging because non-rigid deformations, expression changes, and viewpoint variations occur simultaneously, creating significant ambiguity in geometry and correspondence estimation. We present a unified method for high-fidelity 4D facial reconstruction based on canonical facial point prediction, a representation that assigns each pixel a normalized facial coordinate in a shared canonical space. This formulation transforms dense tracking and dynamic reconstruction into a canonical reconstruction problem, enabling temporally consistent geometry and reliable correspondences within a single feed-forward model. By jointly predicting depth and canonical coordinates, our method enables accurate depth estimation, temporally stable reconstruction, dense 3D geometry, and robust facial point tracking within a single architecture. We implement this formulation using a transformer-based model that jointly predicts depth and canonical facial coordinates, trained using multi-view geometry data that non-rigidly warps into the canonical space. Extensive experiments on image and video benchmarks demonstrate state-of-the-art performance across reconstruction and tracking tasks, achieving approximately 3$\\times$ lower correspondence error and faster inference than prior dynamic reconstruction methods, while improving depth accuracy by 16%. These results highlight canonical facial point prediction as an effective foundation for unified feed-forward 4D facial reconstruction.", "date": "2026-04-21", "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": "2604.17818", "title": "AnyLift: Scaling Motion Reconstruction from Internet Videos via 2D Diffusion", "abstract": "Reconstructing 3D human motion and human-object interactions (HOI) from Internet videos is a fundamental step toward building large-scale datasets of human behavior. Existing methods struggle to recover globally consistent 3D motion under dynamic cameras, especially for motion types underrepresented in current motion-capture datasets, and face additional difficulty recovering coherent human-object interactions in 3D. We introduce a two-stage framework leveraging 2D diffusion that reconstructs 3D human motion and HOI from Internet videos. In the first stage, we synthesize multi-view 2D motion data for each domain, leveraging 2D keypoints extracted from Internet videos to incorporate human motions that rarely appear in existing MoCap datasets. In the second stage, a camera-conditioned multi-view 2D motion diffusion model is trained on the domain-specific synthetic data to recover 3D human motion and 3D HOI in the world space. We demonstrate the effectiveness of our method on Internet videos featuring challenging motions such as gymnastics, as well as in-the-wild HOI videos, and show that it outperforms prior work in producing realistic human motion and human-object interaction.", "date": "2026-04-20", "categories": []}
{"id": "2604.17959", "title": "Chatting about Upper-Body Expressive Human Pose and Shape Estimation", "abstract": "Expressive Human Pose and Shape Estimation (EHPS) plays a crucial role in various AR/VR applications and has witnessed significant progress in recent years. However, current state-of-the-art methods still struggle with accurate parameter estimation for facial and hand regions and exhibit limited generalization to wild images. To address these challenges, we present CoEvoer, a novel one-stage synergistic cross-dependency transformer framework tailored for upper-body EHPS. CoEvoer enables explicit feature-level interaction across different body parts, allowing for mutual enhancement through contextual information exchange. Specifically, larger and more easily estimated regions such as the torso provide global semantics and positional priors to guide the estimation of finer, more complex regions like the face and hands. Conversely, the localized details captured in facial and hand regions help refine and calibrate adjacent body parts. To the best of our knowledge, CoEvoer is the first framework designed specifically for upper-body EHPS, with the goal of capturing the strong coupling and semantic dependencies among the face, hands, and torso through joint parameter regression. Extensive experiments demonstrate that CoEvoer achieves state-of-the-art performance on upper-body benchmarks and exhibits strong generalization capability even on unseen wild images.", "date": "2026-04-20", "categories": []}
{"id": "2604.18583", "title": "MUA: Mobile Ultra-detailed Animatable Avatars", "abstract": "Building photorealistic, animatable full-body digital humans remains a longstanding challenge in computer graphics and vision. Recent advances in animatable avatar modeling have largely progressed along two directions: improving the fidelity of dynamic geometry and appearance, or reducing computational complexity to enable deployment on resource-constrained platforms, e.g., VR headsets. However, existing approaches fail to achieve both goals simultaneously: Ultra-high-fidelity avatars typically require substantial computation on server-class GPUs, whereas lightweight avatars often suffer from limited surface dynamics, reduced appearance details, and noticeable artifacts. To bridge this gap, we propose a novel animatable avatar representation, termed Wavelet-guided Multi-level Spatial Factorized Blendshapes, and a corresponding distillation pipeline that transfers motion-aware clothing dynamics and fine-grained appearance details from a pre-trained ultra-high-quality avatar model into a compact, efficient representation. By coupling multi-level wavelet spectral decomposition with low-rank structural factorization in texture space, our method achieves up to 2000X lower computational cost and a 10X smaller model size than the original high-quality teacher avatar model, while preserving visually plausible dynamics and appearance details closely resemble those of the teacher model. Extensive comparisons with state-of-the-art methods show that our approach significantly outperforms existing avatar approaches designed for mobile settings and achieves comparable or superior rendering quality to most approaches that can only run on servers. Importantly, our representation substantially improves the practicality of high-fidelity avatars for immersive applications, achieving over 180 FPS on a desktop PC and real-time native on-device performance at 24 FPS on a standalone Meta Quest 3.", "date": "2026-04-20", "categories": []}
{"id": "2604.14541", "title": "Giving Faces Their Feelings Back: Explicit Emotion Control for Feedforward Single-Image 3D Head Avatars", "abstract": "We present a framework for explicit emotion control in feed-forward, single-image 3D head avatar reconstruction. Unlike existing pipelines where emotion is implicitly entangled with geometry or appearance, we treat emotion as a first-class control signal that can be manipulated independently and consistently across identities. Our method injects emotion into existing feed-forward architectures via a dual-path modulation mechanism without modifying their core design. Geometry modulation performs emotion-conditioned normalization in the original parametric space, disentangling emotional state from speech-driven articulation, while appearance modulation captures identity-aware, emotion-dependent visual cues beyond geometry. To enable learning under this setting, we construct a time-synchronized, emotion-consistent multi-identity dataset by transferring aligned emotional dynamics across identities. Integrated into multiple state-of-the-art backbones, our framework preserves reconstruction and reenactment fidelity while enabling controllable emotion transfer, disentangled manipulation, and smooth emotion interpolation, advancing expressive and scalable 3D head avatars.", "date": "2026-04-16", "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": "2604.13171", "title": "3DRealHead: Few-Shot Detailed Head Avatar", "abstract": "The human face is central to communication. For immersive applications, the digital presence of a person should mirror the physical reality, capturing the users idiosyncrasies and detailed facial expressions. However, current 3D head avatar methods often struggle to faithfully reproduce the identity and facial expressions, despite having multi-view data or learned priors. Learning priors that capture the diversity of human appearances, especially, for regions with highly person-specific features, like the mouth and teeth region is challenging as the underlying training data is limited. In addition, many of the avatar methods are purely relying on 3D morphable model-based expression control which strongly limits expressivity. To address these challenges, we are introducing 3DRealHead, a few-shot head avatar reconstruction method with a novel expression control signal that is extracted from a monocular video stream of the subject. Specifically, the subject can take a few pictures of themselves, recover a 3D head avatar and drive it with a consumer-level webcam. The avatar reconstruction is enabled via a novel few-shot inversion process of a 3D human head prior which is represented as a Style U-Net that emits 3D Gaussian primitives which can be rendered under novel views. The prior is learned on the NeRSemble dataset. For animating the avatar, the U-Net is conditioned on 3DMM-based facial expression signals, as well as features of the mouth region extracted from the driving video. These additional mouth features allow us to recover facial expressions that cannot be represented by the 3DMM leading to a higher expressivity and closer resemblance to the physical reality.", "date": "2026-04-14", "categories": []}
{"id": "2604.13581", "title": "SocialMirror: Reconstructing 3D Human Interaction Behaviors from Monocular Videos with Semantic and Geometric Guidance", "abstract": "Accurately reconstructing human behavior in close-interaction scenarios is crucial for enabling realistic virtual interactions in augmented reality, precise motion analysis in sports, and natural collaborative behavior in human-robot tasks. Reliable reconstruction in these contexts significantly enhances the realism and effectiveness of AI-driven interactive applications. However, human reconstruction from monocular videos in close-interaction scenarios remains challenging due to severe mutual occlusions, leading local motion ambiguity, disrupted temporal continuity and spatial relationship error. In this paper, we propose SocialMirror, a diffusion-based framework that integrates semantic and geometric cues to effectively address these issues. Specifically, we first leverage high-level interaction descriptions generated by a vision-language model to guide a semantic-guided motion infiller, hallucinating occluded bodies and resolving local pose ambiguities. Next, we propose a sequence-level temporal refiner that enforces smooth, jitter-free motions, while incorporating geometric constraints during sampling to ensure plausible contact and spatial relationships. Evaluations on multiple interaction benchmarks show that SocialMirror achieves state-of-the-art performance in reconstructing interactive human meshes, demonstrating strong generalization across unseen datasets and in-the-wild scenarios. The code will be released upon publication.", "date": "2026-04-15", "categories": []}
{"id": "2604.13856", "title": "Any3DAvatar: Fast and High-Quality Full-Head 3D Avatar Reconstruction from Single Portrait Image", "abstract": "Reconstructing a complete 3D head from a single portrait remains challenging because existing methods still face a sharp quality-speed trade-off: high-fidelity pipelines often rely on multi-stage processing and per-subject optimization, while fast feed-forward models struggle with complete geometry and fine appearance details. To bridge this gap, we propose Any3DAvatar, a fast and high-quality method for single-image 3D Gaussian head avatar generation, whose fastest setting reconstructs a full head in under one second while preserving high-fidelity geometry and texture. First, we build AnyHead, a unified data suite that combines identity diversity, dense multi-view supervision, and realistic accessories, filling the main gaps of existing head data in coverage, full-head geometry, and complex appearance. Second, rather than sampling unstructured noise, we initialize from a Pl\\\"ucker-aware structured 3D Gaussian scaffold and perform one-step conditional denoising, formulating full-head reconstruction into a single forward pass while retaining high fidelity. Third, we introduce auxiliary view-conditioned appearance supervision on the same latent tokens alongside 3D Gaussian reconstruction, improving novel-view texture details at zero extra inference cost. Experiments show that Any3DAvatar outperforms prior single-image full-head reconstruction methods in rendering fidelity while remaining substantially faster.", "date": "2026-04-15", "categories": []}
{"id": "2604.13918", "title": "PartNerFace: Part-based Neural Radiance Fields for Animatable Facial Avatar Reconstruction", "abstract": "We present PartNerFace, a part-based neural radiance fields approach, for reconstructing animatable facial avatar from monocular RGB videos. Existing solutions either simply condition the implicit network with the morphable model parameters or learn an imaginary canonical radiance field, making them fail to generalize to unseen facial expressions and capture fine-scale motion details. To address these challenges, we first apply inverse skinning based on a parametric head model to map an observed point to the canonical space, and then model fine-scale motions with a part-based deformation field. Our key insight is that the deformation of different facial parts should be modeled differently. Specifically, our part-based deformation field consists of multiple local MLPs to adaptively partition the canonical space into different parts, where the deformation of a 3D point is computed by aggregating the prediction of all local MLPs by a soft-weighting mechanism. Extensive experiments demonstrate that our method generalizes well to unseen expressions and is capable of modeling fine-scale facial motions, outperforming state-of-the-art methods both quantitatively and qualitatively.", "date": "2026-04-15", "categories": []}
{"id": "2604.12894", "title": "Representing 3D Faces with Learnable B-Spline Volumes", "abstract": "We present CUBE (Control-based Unified B-spline Encoding), a new geometric representation for human faces that combines B-spline volumes with learned features, and demonstrate its use as a decoder for 3D scan registration and monocular 3D face reconstruction. Unlike existing B-spline representations with 3D control points, CUBE is parametrized by a lattice (e.g., 8 x 8 x 8) of high-dimensional control features, increasing the model's expressivity. These features define a continuous, two-stage mapping from a 3D parametric domain to 3D Euclidean space via an intermediate feature space. First, high-dimensional control features are locally blended using the B-spline bases, yielding a high-dimensional feature vector whose first three values define a 3D base mesh. A small MLP then processes this feature vector to predict a residual displacement from the base shape, yielding the final refined 3D coordinates. To reconstruct 3D surfaces in dense semantic correspondence, CUBE is queried at 3D coordinates sampled from a fixed template mesh. Crucially, CUBE retains the local support property of traditional B-spline representations, enabling local surface editing by updating individual control features. We demonstrate the strengths of this representation by training transformer-based encoders to predict CUBE's control features from unstructured point clouds and monocular images, achieving state-of-the-art scan registration results compared to recent baselines.", "date": "2026-04-14", "categories": []}
{"id": "2604.09835", "title": "F3G-Avatar : Face Focused Full-body Gaussian Avatar", "abstract": "Existing full-body Gaussian avatar methods primarily optimize global reconstruction quality and often fail to preserve fine-grained facial geometry and expression details. This challenge arises from limited facial representational capacity that causes difficulties in modeling high-frequency pose-dependent deformations. To address this, we propose F3G-Avatar, a full-body, face-aware avatar synthesis method that reconstructs animatable human representations from multi-view RGB video and regressed pose/shape parameters. Starting from a clothed Momentum Human Rig (MHR) template, front/back positional maps are rendered and decoded into 3D Gaussians through a two-branch architecture: a body branch that captures pose-dependent non-rigid deformations and a face-focused deformation branch that refines head geometry and appearance. The predicted Gaussians are fused, posed with linear blend skinning (LBS), and rendered with differentiable Gaussian splatting. Training combines reconstruction and perceptual objectives with a face-specific adversarial loss to enhance realism in close-up views. Experiments demonstrate strong rendering quality, with face-view performance reaching PSNR/SSIM/LPIPS of 26.243/0.964/0.084 on the AvatarReX dataset. Ablations further highlight contributions of the MHR template and the face-focused deformation. F3G-Avatar provides a practical, high-quality pipeline for realistic, animatable full-body avatar synthesis.", "date": "2026-04-10", "categories": []}
{"id": "2604.10259", "title": "Real-Time Human Reconstruction and Animation using Feed-Forward Gaussian Splatting", "abstract": "We present a generalizable feed-forward Gaussian splatting framework for human 3D reconstruction and real-time animation that operates directly on multi-view RGB images and their associated SMPL-X poses. Unlike prior methods that rely on depth supervision, fixed input views, UV map, or repeated feed-forward inference for each target view or pose, our approach predicts, in a canonical pose, a set of 3D Gaussian primitives associated with each SMPL-X vertex. One Gaussian is regularized to remain close to the SMPL-X surface, providing a strong geometric prior and stable correspondence to the parametric body model, while an additional small set of unconstrained Gaussians per vertex allows the representation to capture geometric structures that deviate from the parametric surface, such as clothing and hair. In contrast to recent approaches such as HumanRAM, which require repeated network inference to synthesize novel poses, our method produces an animatable human representation from a single forward pass; by explicitly associating Gaussian primitives with SMPL-X vertices, the reconstructed model can be efficiently animated via linear blend skinning without further network evaluation. We evaluate our method on the THuman 2.1, AvatarReX and THuman 4.0 datasets, where it achieves reconstruction quality comparable to state-of-the-art methods while uniquely supporting real-time animation and interactive applications. Code and pre-trained models are available at https://github.com/Devdoot57/HumanGS .", "date": "2026-04-11", "categories": []}
{"id": "2603.19929", "title": "RAM: Recover Any 3D Human Motion in-the-Wild", "abstract": "RAM incorporates a motion-aware semantic tracker with adaptive Kalman filtering to achieve robust identity association under severe occlusions and dynamic interactions. A memory-augmented Temporal HMR module further enhances human motion reconstruction by injecting spatio-temporal priors for consistent and smooth motion estimation. Moreover, a lightweight Predictor module forecasts future poses to maintain reconstruction continuity, while a gated combiner adaptively fuses reconstructed and predicted features to ensure coherence and robustness. Experiments on in-the-wild multi-person benchmarks such as PoseTrack and 3DPW, demonstrate that RAM substantially outperforms previous state-of-the-art in both Zero-shot tracking stability and 3D accuracy, offering a generalizable paradigm for markerless 3D human motion capture in-the-wild.", "date": "2026-03-20", "categories": []}
{"id": "2604.08943", "title": "MASS: Mesh-inellipse Aligned Deformable Surfel Splatting for Hand Reconstruction and Rendering from Egocentric Monocular Video", "abstract": "Reconstructing high-fidelity 3D hands from egocentric monocular videos remains a challenge due to the limitations in capturing high-resolution geometry, hand-object interactions, and complex objects on hands. Additionally, existing methods often incur high computational costs, making them impractical for real-time applications. In this work, we propose Mesh-inellipse Aligned deformable Surfel Splatting (MASS) to address these challenges by leveraging a deformable 2D Gaussian Surfel representation. We introduce the mesh-aligned Steiner Inellipse and fractal densification for mesh-to-surfel conversion that initiates high-resolution 2D Gaussian surfels from coarse parametric hand meshes, providing surface representation with photorealistic rendering potential. Second, we propose Gaussian Surfel Deformation, which enables efficient modeling of hand deformations and personalized features by predicting residual updates to surfel attributes and introducing an opacity mask to refine geometry and texture without adaptive density control. In addition, we propose a two-stage training strategy and a novel binding loss to improve the optimization robustness and reconstruction quality. Extensive experiments on the ARCTIC dataset, the Hand Appearance dataset, and the Interhand2.6M dataset demonstrate that our model achieves superior reconstruction performance compared to state-of-the-art methods.", "date": "2026-04-10", "categories": []}
{"id": "2604.09324", "title": "Structure-Aware Fine-Grained Gaussian Splatting for Expressive Avatar Reconstruction", "abstract": "Reconstructing photorealistic and topology-aware human avatars from monocular videos remains a significant challenge in the fields of computer vision and graphics. While existing 3D human avatar modeling approaches can effectively capture body motion, they often fail to accurately model fine details such as hand movements and facial expressions. To address this, we propose Structure-aware Fine-grained Gaussian Splatting (SFGS), a novel method for reconstructing expressive and coherent full-body 3D human avatars from a monocular video sequence. The SFGS use both spatial-only triplane and time-aware hexplane to capture dynamic features across consecutive frames. A structure-aware gaussian module is designed to capture pose-dependent details in a spatially coherent manner and improve pose and texture expression. To better model hand deformations, we also propose a residual refinement module based on fine-grained hand reconstruction. Our method requires only a single-stage training and outperforms state-of-the-art baselines in both quantitative and qualitative evaluations, generating high-fidelity avatars with natural motion and fine details. The code is on Github: https://github.com/Su245811YZ/SFGS", "date": "2026-04-10", "categories": []}
{"id": "2504.13378", "title": "SMPL-GPTexture: Dual-View 3D Human Texture Estimation using Text-to-Image Generation Models", "abstract": "Generating high-quality, photorealistic textures for 3D human avatars remains a fundamental yet challenging task in computer vision and multimedia field. However, real paired front and back images of human subjects are rarely available with privacy, ethical and cost of acquisition, which restricts scalability of the data. Additionally, learning priors from image inputs using deep generative models, such as GANs or diffusion models, to infer unseen regions such as the human back often leads to artifacts, structural inconsistencies, or loss of fine-grained detail. To address these issues, we present SMPL-GPTexture (skinned multi-person linear model - general purpose Texture), a novel pipeline that takes natural language prompts as input and leverages a state-of-the-art text-to-image generation model to produce paired high-resolution front and back images of a human subject as the starting point for texture estimation. Using the generated paired dual-view images, we first employ a human mesh recovery model to obtain a robust 2D-to-3D SMPL alignment between image pixels and the 3D model's UV coordinates for each views. Second, we use an inverted rasterization technique that explicitly projects the observed colour from the input images into the UV space, thereby producing accurate, complete texture maps. Finally, we apply a diffusion-based inpainting module to fill in the missing regions, and the fusion mechanism then combines these results into a unified full texture map. Extensive experiments shows that our SMPL-GPTexture can generate high resolution texture aligned with user's prompts.", "date": "2025-04-17", "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": "2604.06467", "title": "PhysHead: Simulation-Ready Gaussian Head Avatars", "abstract": "Realistic digital avatars require expressive and dynamic hair motion; however, most existing head avatar methods assume rigid hair movement. These methods often fail to disentangle hair from the head, representing it as a simple outer shell and failing to capture its natural volumetric behavior. In this paper, we address these limitations by introducing PhysHead, a hybrid representation for animatable head avatars with realistic hair dynamics learned from multi-view video. At the core is a 3D Gaussian-based layered representation of the head. Our approach combines a 3D parametric mesh for the head with strand-based hair, which can be directly simulated using physics engines. For the appearance model, we employ Gaussian primitives attached to both the head mesh and hair segments. This representation enables the creation of photorealistic head avatars with dynamic hair behavior, such as wind-blown motion, overcoming the constraints of rigid hair in existing methods. However, these animation capabilities also require new training schemes. In particular, we propose the use of VLM-based models to generate appearance of regions that are occluded in the dynamic training sequences. In quantitative and qualitative studies, we demonstrate the capabilities of the proposed model and compare it with existing baselines. We show that our method can synthesize physically plausible hair motion besides expression and camera control.", "date": "2026-04-07", "categories": []}
{"id": "2604.02320", "title": "Large-scale Codec Avatars: The Unreasonable Effectiveness of Large-scale Avatar Pretraining", "abstract": "High-quality 3D avatar modeling faces a critical trade-off between fidelity and generalization. On the one hand, multi-view studio data enables high-fidelity modeling of humans with precise control over expressions and poses, but it struggles to generalize to real-world data due to limited scale and the domain gap between the studio environment and the real world. On the other hand, recent large-scale avatar models trained on millions of in-the-wild samples show promise for generalization across a wide range of identities, yet the resulting avatars are often of low-quality due to inherent 3D ambiguities. To address this, we present Large-Scale Codec Avatars (LCA), a high-fidelity, full-body 3D avatar model that generalizes to world-scale populations in a feedforward manner, enabling efficient inference. Inspired by the success of large language models and vision foundation models, we present, for the first time, a pre/post-training paradigm for 3D avatar modeling at scale: we pretrain on 1M in-the-wild videos to learn broad priors over appearance and geometry, then post-train on high-quality curated data to enhance expressivity and fidelity. LCA generalizes across hair styles, clothing, and demographics while providing precise, fine-grained facial expressions and finger-level articulation control, with strong identity preservation. Notably, we observe emergent generalization to relightability and loose garment support to unconstrained inputs, and zero-shot robustness to stylized imagery, despite the absence of direct supervision.", "date": "2026-04-02", "categories": []}
{"id": "2604.05436", "title": "Human Interaction-Aware 3D Reconstruction from a Single Image", "abstract": "Reconstructing textured 3D human models from a single image is fundamental for AR/VR and digital human applications. However, existing methods mostly focus on single individuals and thus fail in multi-human scenes, where naive composition of individual reconstructions often leads to artifacts such as unrealistic overlaps, missing geometry in occluded regions, and distorted interactions. These limitations highlight the need for approaches that incorporate group-level context and interaction priors. We introduce a holistic method that explicitly models both group- and instance-level information. To mitigate perspective-induced geometric distortions, we first transform the input into a canonical orthographic space. Our primary component, Human Group-Instance Multi-View Diffusion (HUG-MVD), then generates complete multi-view normals and images by jointly modeling individuals and group context to resolve occlusions and proximity. Subsequently, the Human Group-Instance Geometric Reconstruction (HUG-GR) module optimizes the geometry by leveraging explicit, physics-based interaction priors to enforce physical plausibility and accurately model inter-human contact. Finally, the multi-view images are fused into a high-fidelity texture. Together, these components form our complete framework, HUG3D. Extensive experiments show that HUG3D significantly outperforms both single-human and existing multi-human methods, producing physically plausible, high-fidelity 3D reconstructions of interacting people from a single image. Project page: https://jongheean11.github.io/HUG3D_project", "date": "2026-04-07", "categories": []}
{"id": "2604.05794", "title": "EfficientMonoHair: Fast Strand-Level Reconstruction from Monocular Video via Multi-View Direction Fusion", "abstract": "Strand-level hair geometry reconstruction is a fundamental problem in virtual human modeling and the digitization of hairstyles. However, existing methods still suffer from a significant trade-off between accuracy and efficiency. Implicit neural representations can capture the global hair shape but often fail to preserve fine-grained strand details, while explicit optimization-based approaches achieve high-fidelity reconstructions at the cost of heavy computation and poor scalability. To address this issue, we propose EfficientMonoHair, a fast and accurate framework that combines the implicit neural network with multi-view geometric fusion for strand-level reconstruction from monocular video. Our method introduces a fusion-patch-based multi-view optimization that reduces the number of optimization iterations for point cloud direction, as well as a novel parallel hair-growing strategy that relaxes voxel occupancy constraints, allowing large-scale strand tracing to remain stable and robust even under inaccurate or noisy orientation fields. Extensive experiments on representative real-world hairstyles demonstrate that our method can robustly reconstruct high-fidelity strand geometries with accuracy. On synthetic benchmarks, our method achieves reconstruction quality comparable to state-of-the-art methods, while improving runtime efficiency by nearly an order of magnitude.", "date": "2026-04-07", "categories": []}
{"id": "2511.22553", "title": "Bringing Your Portrait to 3D Presence", "abstract": "We present a unified framework for reconstructing animatable 3D human avatars from a single portrait across head, half-body, and full-body inputs. Our method tackles three bottlenecks: pose- and framing-sensitive feature representations, limited scalable data, and unreliable proxy-mesh estimation. We introduce a Dual-UV representation that maps image features to a canonical UV space via Core-UV and Shell-UV branches, eliminating pose- and framing-induced token shifts. We also build a factorized synthetic data manifold combining 2D generative diversity with geometry-consistent 3D renderings, supported by a training scheme that improves realism and identity consistency. A robust proxy-mesh tracker maintains stability under partial visibility. Together, these components enable strong in-the-wild generalization. Trained only on half-body synthetic data, our model achieves state-of-the-art head and upper-body reconstruction and competitive full-body results. Extensive experiments and analyses further validate the effectiveness of our approach.", "date": "2025-11-27", "categories": []}
{"id": "2604.01447", "title": "Better Rigs, Not Bigger Networks: A Body Model Ablation for Gaussian Avatars", "abstract": "Recent 3D Gaussian splatting methods built atop SMPL achieve remarkable visual fidelity while continually increasing the complexity of the overall training architecture. We demonstrate that much of this complexity is unnecessary: by replacing SMPL with the Momentum Human Rig (MHR), estimated via SAM-3D-Body, a minimal pipeline with no learned deformations or pose-dependent corrections achieves the highest reported PSNR and competitive or superior LPIPS and SSIM on PeopleSnapshot and ZJU-MoCap. To disentangle pose estimation quality from body model representational capacity, we perform two controlled ablations: translating SAM-3D-Body meshes to SMPL-X, and translating the original dataset's SMPL poses into MHR both retrained under identical conditions. These ablations confirm that body model expressiveness has been a primary bottleneck in avatar reconstruction, with both mesh representational capacity and pose estimation quality contributing meaningfully to the full pipeline's gains.", "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": "2604.02867", "title": "HairOrbit: Multi-view Aware 3D Hair Modeling from Single Portraits", "abstract": "Reconstructing strand-level 3D hair from a single-view image is highly challenging, especially when preserving consistent and realistic attributes in unseen regions. Existing methods rely on limited frontal-view cues and small-scale/style-restricted synthetic data, often failing to produce satisfactory results in invisible regions. In this work, we propose a novel framework that leverages the strong 3D priors of video generation models to transform single-view hair reconstruction into a calibrated multi-view reconstruction task. To balance reconstruction quality and efficiency for the reformulated multi-view task, we further introduce a neural orientation extractor trained on sparse real-image annotations for better full-view orientation estimation. In addition, we design a two-stage strand-growing algorithm based on a hybrid implicit field to synthesize the 3D strand curves with fine-grained details at a relatively fast speed. Extensive experiments demonstrate that our method achieves state-of-the-art performance on single-view 3D hair strand reconstruction on a diverse range of hair portraits in both visible and invisible regions.", "date": "2026-04-03", "categories": []}
{"id": "2603.20391", "title": "Monocular Models are Strong Learners for Multi-View Human Mesh Recovery", "abstract": "Multi-view human mesh recovery (HMR) is broadly deployed in diverse domains where high accuracy and strong generalization are essential. Existing approaches can be broadly grouped into geometry-based and learning-based methods. However, geometry-based methods (e.g., triangulation) rely on cumbersome camera calibration, while learning-based approaches often generalize poorly to unseen camera configurations due to the lack of multi-view training data, limiting their performance in real-world scenarios. To enable calibration-free reconstruction that generalizes to arbitrary camera setups, we propose a training-free framework that leverages pretrained single-view HMR models as strong priors, eliminating the need for multi-view training data. Our method first constructs a robust and consistent multi-view initialization from single-view predictions, and then refines it via test-time optimization guided by multi-view consistency and anatomical constraints. Extensive experiments demonstrate state-of-the-art performance on standard benchmarks, surpassing multi-view models trained with explicit multi-view supervision.", "date": "2026-03-20", "categories": []}
{"id": "2604.00928", "title": "Autoregressive Appearance Prediction for 3D Gaussian Avatars", "abstract": "A photorealistic and immersive human avatar experience demands capturing fine, person-specific details such as cloth and hair dynamics, subtle facial expressions, and characteristic motion patterns. Achieving this requires large, high-quality datasets, which often introduce ambiguities and spurious correlations when very similar poses correspond to different appearances. Models that fit these details during training can overfit and produce unstable, abrupt appearance changes for novel poses. We propose a 3D Gaussian Splatting avatar model with a spatial MLP backbone that is conditioned on both pose and an appearance latent. The latent is learned during training by an encoder, yielding a compact representation that improves reconstruction quality and helps disambiguate pose-driven renderings. At driving time, our predictor autoregressively infers the latent, producing temporally smooth appearance evolution and improved stability. Overall, our method delivers a robust and practical path to high-fidelity, stable avatar driving.", "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": "2506.09919", "title": "MetricHMSR:Metric Human Mesh and Scene Recovery from Monocular Images", "abstract": "We introduce MetricHMSR, a novel framework for recovering metric human meshes and 3D scenes from a single monocular image. Existing methods struggle to recover metric scale due to monocular scale ambiguity and weak-perspective camera assumptions. Moreover, their fully coupled feature representations make it difficult to disentangle local pose from global translation, often requiring multi-stage pipelines that introduce accumulated errors. To address these challenges, we propose MetricHMR (Metric Human Mesh Recovery), which incorporates a bounding camera ray map representation to provide explicit metric cues for human reconstruction,together with a Human Mixture-of-Experts (HumanMoE) that dynamically routes image features to specialized experts, enabling the disentangled perception of local human pose and global metric position. Leveraging the recovered metric human as a geometric anchor, we further refine monocular metric depth estimation to achieve more accurate 3D alignment between humans and scenes.Comprehensive experiments demonstrate that our method achieves state-of-the-art performance on both human mesh recovery and metric human-scene reconstruction. Project Page: https://Metaverse-AI-Lab-THU.github.io/MetricHMSR.", "date": "2025-06-11", "categories": []}
{"id": "2512.12378", "title": "M4Human: A Large-Scale Multimodal mmWave Radar Benchmark for Human Mesh Reconstruction", "abstract": "Human mesh reconstruction (HMR) provides direct insights into body-environment interaction, which enables various immersive applications. While existing large-scale HMR datasets rely heavily on line-of-sight RGB input, vision-based sensing is limited by occlusion, lighting variation, and privacy concerns. To overcome these limitations, recent efforts have explored radio-frequency (RF) mmWave radar for privacy-preserving indoor human sensing. However, current radar datasets are constrained by sparse skeleton labels, limited scale, and simple in-place actions. To advance the HMR research community, we introduce M4Human, the current largest-scale (661K-frame) ($9\\times$ prior largest) multimodal benchmark, featuring high-resolution mmWave radar, RGB, and depth data. M4Human provides both raw radar tensors (RT) and processed radar point clouds (RPC) to enable research across different levels of RF signal granularity. M4Human includes high-quality motion capture (MoCap) annotations with 3D meshes and global trajectories, and spans 20 subjects and 50 diverse actions, including in-place, sit-in-place, and free-space sports or rehabilitation movements. We establish benchmarks on both RT and RPC modalities, as well as multimodal fusion with RGB-D modalities. Extensive results highlight the significance of M4Human for radar-based human modeling while revealing persistent challenges under fast, unconstrained motion. The dataset and code will be released after the paper publication.", "date": "2025-12-13", "categories": []}
{"id": "2603.01506", "title": "OMG-Avatar: One-shot Multi-LOD Gaussian Head Avatar", "abstract": "We propose OMG-Avatar, a novel One-shot method that leverages a Multi-LOD (Level-of-Detail) Gaussian representation for animatable 3D head reconstruction from a single image in 0.2s. Our method enables LOD head avatar modeling using a unified model that accommodates diverse hardware capabilities and inference speed requirements. To capture both global and local facial characteristics, we employ a transformer-based architecture for global feature extraction and projection-based sampling for local feature acquisition. These features are effectively fused under the guidance of a depth buffer, ensuring occlusion plausibility. We further introduce a coarse-to-fine learning paradigm to support Level-of-Detail functionality and enhance the perception of hierarchical details. To address the limitations of 3DMMs in modeling non-head regions such as the shoulders, we introduce a multi-region decomposition scheme in which the head and shoulders are predicted separately and then integrated through cross-region combination. Extensive experiments demonstrate that OMG-Avatar outperforms state-of-the-art methods in reconstruction quality, reenactment performance, and computational efficiency. The project homepage is https://human3daigc.github.io/OMGAvatar_project_page/ .", "date": "2026-03-02", "categories": []}
{"id": "2603.21332", "title": "EmoTaG: Emotion-Aware Talking Head Synthesis on Gaussian Splatting with Few-Shot Personalization", "abstract": "Audio-driven 3D talking head synthesis has advanced rapidly with Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). By leveraging rich pre-trained priors, few-shot methods enable instant personalization from just a few seconds of video. However, under expressive facial motion, existing few-shot approaches often suffer from geometric instability and audio-emotion mismatch, highlighting the need for more effective emotion-aware motion modeling. In this work, we present EmoTaG, a few-shot emotion-aware 3D talking head synthesis framework built on the Pretrain-and-Adapt paradigm. Our key insight is to reformulate motion prediction in a structured FLAME parameter space rather than directly deforming 3D Gaussians, thereby introducing explicit geometric priors that improve motion stability. Building upon this, we propose a Gated Residual Motion Network (GRMN), which captures emotional prosody from audio while supplementing head pose and upper-face cues absent from audio, enabling expressive and coherent motion generation. Extensive experiments demonstrate that EmoTaG achieves state-of-the-art performance in emotional expressiveness, lip synchronization, visual realism, and motion stability.", "date": "2026-03-22", "categories": []}
{"id": "2502.02468", "title": "High-Fidelity Human Avatars from Laptop Webcams using Edge Compute", "abstract": "Photo-realistic human avatars have broad applications, yet high-fidelity avatar generation has traditionally required expensive professional camera rigs and extensive artistic labor. Recent research has enabled constructing them automatically from smartphones with RGB and IR sensors, however, these new methods still rely on high-resolution cameras on modern smartphones and often require offloading the processing to powerful servers with GPUs. Modern applications such as video conferencing call for the ability to generate these avatars from consumer-grade laptop webcams using limited compute available on-device. In this work, we develop a novel method based on 3D morphable models, landmark detection, photorealistic texture GANs, and differentiable rendering to tackle the problem of low webcam image quality and edge computation. We build an automatic system to generate high-fidelity animatable avatars under these limitations, leveraging the compute capabilities of AMD mobile processors.", "date": "2025-02-04", "categories": []}
{"id": "2602.21100", "title": "Skullptor: High Fidelity 3D Head Reconstruction in Seconds with Multi-View Normal Prediction", "abstract": "Reconstructing high-fidelity 3D head geometry from images is critical for a wide range of applications, yet existing methods face fundamental limitations. Traditional photogrammetry achieves exceptional detail but requires extensive camera arrays (25-200+ views), substantial computation, and manual cleanup in challenging areas like facial hair. Recent alternatives present a fundamental trade-off: foundation models enable efficient single-image reconstruction but lack fine geometric detail, while optimization-based methods achieve higher fidelity but require dense views and expensive computation. We bridge this gap with a hybrid approach that combines the strengths of both paradigms. Our method introduces a multi-view surface normal prediction model that extends monocular foundation models with cross-view attention to produce geometrically consistent normals in a feed-forward pass. We then leverage these predictions as strong geometric priors within an inverse rendering optimization framework to recover high-frequency surface details. Our approach outperforms state-of-the-art single-image and multi-view methods, achieving high-fidelity reconstruction on par with dense-view photogrammetry while reducing camera requirements and computational cost.", "date": "2026-02-24", "categories": []}
{"id": "2507.02803", "title": "HyperGaussians: High-Dimensional Gaussian Splatting for High-Fidelity Animatable Face Avatars", "abstract": "We introduce HyperGaussians, a novel extension of 3D Gaussian Splatting for high-quality animatable face avatars. Creating such detailed face avatars from videos is a challenging problem and has numerous applications in augmented and virtual reality. While tremendous successes have been achieved for static faces, animatable avatars from monocular videos still fall in the uncanny valley. The de facto standard, 3D Gaussian Splatting (3DGS), represents a face through a collection of 3D Gaussian primitives. 3DGS excels at rendering static faces, but the state-of-the-art still struggles with nonlinear deformations, complex lighting effects, and fine details. While most related works focus on predicting better Gaussian parameters from expression codes, we rethink the 3D Gaussian representation itself and how to make it more expressive. Our insights lead to a novel extension of 3D Gaussians to high-dimensional multivariate Gaussians, dubbed 'HyperGaussians'. The higher dimensionality increases expressivity through conditioning on a learnable local embedding. However, splatting HyperGaussians is computationally expensive because it requires inverting a high-dimensional covariance matrix. We solve this by reparameterizing the covariance matrix, dubbed the 'inverse covariance trick'. This trick boosts the efficiency so that HyperGaussians can be seamlessly integrated into existing models. To demonstrate this, we plug in HyperGaussians into the state-of-the-art in fast monocular face avatars: FlashAvatar. Our evaluation on 19 subjects from 4 face datasets shows that HyperGaussians outperform 3DGS numerically and visually, particularly for high-frequency details like eyeglass frames, teeth, complex facial movements, and specular reflections.", "date": "2025-07-03", "categories": []}
{"id": "2603.23997", "title": "HGGT: Robust and Flexible 3D Hand Mesh Reconstruction from Uncalibrated Images", "abstract": "Recovering high-fidelity 3D hand geometry from images is a critical task in computer vision, holding significant value for domains such as robotics, animation and VR/AR. Crucially, scalable applications demand both accuracy and deployment flexibility, requiring the ability to leverage massive amounts of unstructured image data from the internet or enable deployment on consumer-grade RGB cameras without complex calibration. However, current methods face a dilemma. While single-view approaches are easy to deploy, they suffer from depth ambiguity and occlusion. Conversely, multi-view systems resolve these uncertainties but typically demand fixed, calibrated setups, limiting their real-world utility. To bridge this gap, we draw inspiration from 3D foundation models that learn explicit geometry directly from visual data. By reformulating hand reconstruction from arbitrary views as a visual-geometry grounded task, we propose a feed-forward architecture that, for the first time in literature, jointly infers 3D hand meshes and camera poses from uncalibrated views. Extensive evaluations show that our approach outperforms state-of-the-art benchmarks and demonstrates strong generalization to uncalibrated, in-the-wild scenarios. Here is the link of our project page: https://lym29.github.io/HGGT/.", "date": "2026-03-25", "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": "2603.23345", "title": "FHAvatar: Fast and High-Fidelity Reconstruction of Face-and-Hair Composable 3D Head Avatar from Few Casual Captures", "abstract": "We present FHAvatar, a novel framework for reconstructing 3D Gaussian avatars with composable face and hair components from an arbitrary number of views. Unlike previous approaches that couple facial and hair representations within a unified modeling process, we explicitly decouple two components in texture space by representing the face with planar Gaussians and the hair with strand-based Gaussians. To overcome the limitations of existing methods that rely on dense multi-view captures or costly per-identity optimization, we propose an aggregated transformer backbone to learn geometry-aware cross-view priors and head-hair structural coherence from multi-view datasets, enabling effective and efficient feature extraction and fusion from few casual captures. Extensive quantitative and qualitative experiments demonstrate that FHAvatar achieves state-of-the-art reconstruction quality from only a few observations of new identities within minutes, while supporting real-time animation, convenient hairstyle transfer, and stylized editing, broadening the accessibility and applicability of digital avatar creation.", "date": "2026-03-24", "categories": []}
{"id": "2509.24817", "title": "UP2You: Fast Reconstruction of Yourself from Unconstrained Photo Collections", "abstract": "We present UP2You, the first tuning-free solution for reconstructing high-fidelity 3D clothed portraits from extremely unconstrained in-the-wild 2D photos. Unlike previous approaches that require \"clean\" inputs (e.g., full-body images with minimal occlusions, or well-calibrated cross-view captures), UP2You directly processes raw, unstructured photographs, which may vary significantly in pose, viewpoint, cropping, and occlusion. Instead of compressing data into tokens for slow online text-to-3D optimization, we introduce a data rectifier paradigm that efficiently converts unconstrained inputs into clean, orthogonal multi-view images in a single forward pass within seconds, simplifying the 3D reconstruction. Central to UP2You is a pose-correlated feature aggregation module (PCFA), that selectively fuses information from multiple reference images w.r.t. target poses, enabling better identity preservation and nearly constant memory footprint, with more observations. We also introduce a perceiver-based multi-reference shape predictor, removing the need for pre-captured body templates. Extensive experiments on 4D-Dress, PuzzleIOI, and in-the-wild captures demonstrate that UP2You consistently surpasses previous methods in both geometric accuracy (Chamfer-15%, P2S-18% on PuzzleIOI) and texture fidelity (PSNR-21%, LPIPS-46% on 4D-Dress). UP2You is efficient (1.5 minutes per person), and versatile (supports arbitrary pose control, and training-free multi-garment 3D virtual try-on), making it practical for real-world scenarios where humans are casually captured. Both models and code will be released to facilitate future research on this underexplored task. Project Page: https://zcai0612.github.io/UP2You", "date": "2025-09-29", "categories": []}
{"id": "2603.17779", "title": "CrowdGaussian: Reconstructing High-Fidelity 3D Gaussians for Human Crowd from a Single Image", "abstract": "Single-view 3D human reconstruction has garnered significant attention in recent years. Despite numerous advancements, prior research has concentrated on reconstructing 3D models from clear, close-up images of individual subjects, often yielding subpar results in the more prevalent multi-person scenarios. Reconstructing 3D human crowd models is a highly intricate task, laden with challenges such as: 1) extensive occlusions, 2) low clarity, and 3) numerous and various appearances. To address this task, we propose CrowdGaussian, a unified framework that directly reconstructs multi-person 3D Gaussian Splatting (3DGS) representations from single-image inputs. To handle occlusions, we devise a self-supervised adaptation pipeline that enables the pretrained large human model to reconstruct complete 3D humans with plausible geometry and appearance from heavily occluded inputs. Furthermore, we introduce Self-Calibrated Learning (SCL). This training strategy enables single-step diffusion models to adaptively refine coarse renderings to optimal quality by blending identity-preserving samples with clean/corrupted image pairs. The outputs can be distilled back to enhance the quality of multi-person 3DGS representations. Extensive experiments demonstrate that CrowdGaussian generates photorealistic, geometrically coherent reconstructions of multi-person scenes.", "date": "2026-03-18", "categories": []}
{"id": "2512.09162", "title": "GTAvatar: Bridging Gaussian Splatting and Texture Mapping for Relightable and Editable Gaussian Avatars", "abstract": "Recent advancements in Gaussian Splatting have enabled increasingly accurate reconstruction of photorealistic head avatars, opening the door to numerous applications in visual effects, videoconferencing, and virtual reality. This, however, comes with the lack of intuitive editability offered by traditional triangle mesh-based methods. In contrast, we propose a method that combines the accuracy and fidelity of 2D Gaussian Splatting with the intuitiveness of UV texture mapping. By embedding each canonical Gaussian primitive's local frame into a patch in the UV space of a template mesh in a computationally efficient manner, we reconstruct continuous editable material head textures from a single monocular video on a conventional UV domain. Furthermore, we leverage an efficient physically based reflectance model to enable relighting and editing of these intrinsic material maps. Through extensive comparisons with state-of-the-art methods, we demonstrate the accuracy of our reconstructions, the quality of our relighting results, and the ability to provide intuitive controls for modifying an avatar's appearance and geometry via texture mapping without additional optimization.", "date": "2025-12-09", "categories": []}
{"id": "2603.12789", "title": "Coherent Human-Scene Reconstruction from Multi-Person Multi-View Video in a Single Pass", "abstract": "Recent advances in 3D foundation models have led to growing interest in reconstructing humans and their surrounding environments. However, most existing approaches focus on monocular inputs, and extending them to multi-view settings requires additional overhead modules or preprocessed data. To this end, we present CHROMM, a unified framework that jointly estimates cameras, scene point clouds, and human meshes from multi-person multi-view videos without relying on external modules or preprocessing. We integrate strong geometric and human priors from Pi3X and Multi-HMR into a single trainable neural network architecture, and introduce a scale adjustment module to solve the scale discrepancy between humans and the scene. We also introduce a multi-view fusion strategy to aggregate per-view estimates into a single representation at test-time. Finally, we propose a geometry-based multi-person association method, which is more robust than appearance-based approaches. Experiments on EMDB, RICH, EgoHumans, and EgoExo4D show that CHROMM achieves competitive performance in global human motion and multi-view pose estimation while running over 8x faster than prior optimization-based multi-view approaches. Project page: https://nstar1125.github.io/chromm.", "date": "2026-03-13", "categories": []}
{"id": "2603.17178", "title": "Patient4D: Temporally Consistent Patient Body Mesh Recovery from Monocular Operating Room Video", "abstract": "Recovering a dense 3D body mesh from monocular video remains challenging under occlusion from draping and continuously moving camera viewpoints. This configuration arises in surgical augmented reality (AR), where an anesthetized patient lies under surgical draping while a surgeon's head-mounted camera continuously changes viewpoint. Existing human mesh recovery (HMR) methods are typically trained on upright, moving subjects captured from relatively stable cameras, leading to performance degradation under such conditions. To address this, we present Patient4D, a stationarity-constrained reconstruction pipeline that explicitly exploits the stationarity prior. The pipeline combines image-level foundation models for perception with lightweight geometric mechanisms that enforce temporal consistency across frames. Two key components enable robust reconstruction: Pose Locking, which anchors pose parameters using stable keyframes, and Rigid Fallback, which recovers meshes under severe occlusion through silhouette-guided rigid alignment. Together, these mechanisms stabilize predictions while remaining compatible with off-the-shelf HMR models. We evaluate Patient4D on 4,680 synthetic surgical sequences and three public HMR video benchmarks. Under surgical drape occlusion, Patient4D achieves a 0.75 mean IoU, reducing failure frames from 30.5% to 1.3% compared to the best baseline. Our findings demonstrate that exploiting stationarity priors can substantially improve monocular reconstruction in clinical AR scenarios.", "date": "2026-03-17", "categories": []}
{"id": "2603.17355", "title": "OnlineHMR: Video-based Online World-Grounded Human Mesh Recovery", "abstract": "Human mesh recovery (HMR) models 3D human body from monocular videos, with recent works extending it to world-coordinate human trajectory and motion reconstruction. However, most existing methods remain offline, relying on future frames or global optimization, which limits their applicability in interactive feedback and perception-action loop scenarios such as AR/VR and telepresence. To address this, we propose OnlineHMR, a fully online framework that jointly satisfies four essential criteria of online processing, including system-level causality, faithfulness, temporal consistency, and efficiency. Built upon a two-branch architecture, OnlineHMR enables streaming inference via a causal key-value cache design and a curated sliding-window learning strategy. Meanwhile, a human-centric incremental SLAM provides online world-grounded alignment under physically plausible trajectory correction. Experimental results show that our method achieves performance comparable to existing chunk-based approaches on the standard EMDB benchmark and highly dynamic custom videos, while uniquely supporting online processing. Page and code are available at https://tsukasane.github.io/Video-OnlineHMR/.", "date": "2026-03-18", "categories": []}
{"id": "2603.17975", "title": "AHOY! Animatable Humans under Occlusion from YouTube Videos with Gaussian Splatting and Video Diffusion Priors", "abstract": "We present AHOY, a method for reconstructing complete, animatable 3D Gaussian avatars from in-the-wild monocular video despite heavy occlusion. Existing methods assume unoccluded input-a fully visible subject, often in a canonical pose-excluding the vast majority of real-world footage where people are routinely occluded by furniture, objects, or other people. Reconstructing from such footage poses fundamental challenges: large body regions may never be observed, and multi-view supervision per pose is unavailable. We address these challenges with four contributions: (i) a hallucination-as-supervision pipeline that uses identity-finetuned diffusion models to generate dense supervision for previously unobserved body regions; (ii) a two-stage canonical-to-pose-dependent architecture that bootstraps from sparse observations to full pose-dependent Gaussian maps; (iii) a map-pose/LBS-pose decoupling that absorbs multi-view inconsistencies from the generated data; (iv) a head/body split supervision strategy that preserves facial identity. We evaluate on YouTube videos and on multi-view capture data with significant occlusion and demonstrate state-of-the-art reconstruction quality. We also demonstrate that the resulting avatars are robust enough to be animated with novel poses and composited into 3DGS scenes captured using cell-phone video. Our project page is available at https://miraymen.github.io/ahoy/", "date": "2026-03-18", "categories": []}
{"id": "2312.03442", "title": "High-Quality Facial Geometry and Appearance Capture at Home", "abstract": "Facial geometry and appearance capture have demonstrated tremendous success in 3D scanning real humans in studios. Recent works propose to democratize this technique while keeping the results high quality. However, they are still inconvenient for daily usage. In addition, they focus on an easier problem of only capturing facial skin. This paper proposes a novel method for high-quality face capture, featuring an easy-to-use system and the capability to model the complete face with skin, mouth interior, hair, and eyes. We reconstruct facial geometry and appearance from a single co-located smartphone flashlight sequence captured in a dim room where the flashlight is the dominant light source (e.g. rooms with curtains or at night). To model the complete face, we propose a novel hybrid representation to effectively model both eyes and other facial regions, along with novel techniques to learn it from images. We apply a combined lighting model to compactly represent real illuminations and exploit a morphable face albedo model as a reflectance prior to disentangle diffuse and specular. Experiments show that our method can capture high-quality 3D relightable scans.", "date": "2023-12-06", "categories": []}
{"id": "2603.16447", "title": "ProgressiveAvatars: Progressive Animatable 3D Gaussian Avatars", "abstract": "In practical real-time XR and telepresence applications, network and computing resources fluctuate frequently. Therefore, a progressive 3D representation is needed. To this end, we propose ProgressiveAvatars, a progressive avatar representation built on a hierarchy of 3D Gaussians grown by adaptive implicit subdivision on a template mesh. 3D Gaussians are defined in face-local coordinates to remain animatable under varying expressions and head motion across multiple detail levels. The hierarchy expands when screen-space signals indicate a lack of detail, allocating resources to important areas. Leveraging importance ranking, ProgressiveAvatars supports incremental loading and rendering, adding new Gaussians as they arrive while preserving previous content, thus achieving smooth quality improvements across varying bandwidths. ProgressiveAvatars enables progressive delivery and progressive rendering under fluctuating network bandwidth and varying compute and memory resources.", "date": "2026-03-17", "categories": []}
{"id": "2506.13766", "title": "LHM++: An Efficient Large Human Reconstruction Model for Pose-free Images to 3D", "abstract": "Reconstructing animatable 3D humans from casually captured images of articulated subjects without camera or pose information is highly practical but remains challenging due to view misalignment, occlusions, and the absence of structural priors. In this work, we present LHM++, an efficient large-scale human reconstruction model that generates high-quality, animatable 3D avatars within seconds from one or multiple pose-free images. At its core is an Encoder-Decoder Point-Image Transformer architecture that progressively encodes and decodes 3D geometric point features to improve efficiency, while fusing hierarchical 3D point features with image features through multimodal attention. The fused features are decoded into 3D Gaussian splats to recover detailed geometry and appearance. To further enhance visual fidelity, we introduce a lightweight 3D-aware neural animation renderer that refines the rendering quality of reconstructed avatars in real time. Extensive experiments show that our method produces high-fidelity, animatable 3D humans without requiring camera or pose annotations. Our code and project page are available at https://lingtengqiu.github.io/LHM++/", "date": "2025-06-16", "categories": []}
{"id": "2511.20157", "title": "SKEL-CF: Coarse-to-Fine Biomechanical Skeleton and Surface Mesh Recovery", "abstract": "Parametric 3D human models such as SMPL have driven significant advances in human pose and shape estimation, yet their simplified kinematics limit biomechanical realism. The recently proposed SKEL model addresses this limitation by re-rigging SMPL with an anatomically accurate skeleton. However, estimating SKEL parameters directly remains challenging due to limited training data, perspective ambiguities, and the inherent complexity of human articulation. We introduce SKEL-CF, a coarse-to-fine framework for SKEL parameter estimation. SKEL-CF employs a transformer-based encoder-decoder architecture, where the encoder predicts coarse camera and SKEL parameters, and the decoder progressively refines them in successive layers. To ensure anatomically consistent supervision, we convert the existing SMPL-based dataset 4DHuman into a SKEL-aligned version, 4DHuman-SKEL, providing high-quality training data for SKEL estimation. In addition, to mitigate depth and scale ambiguities, we explicitly incorporate camera modeling into the SKEL-CF pipeline and demonstrate its importance across diverse viewpoints. Extensive experiments validate the effectiveness of the proposed design. On the challenging MOYO dataset, SKEL-CF achieves 85.0 MPJPE / 51.4 PA-MPJPE, significantly outperforming the previous SKEL-based state-of-the-art HSMR (104.5 / 79.6). These results establish SKEL-CF as a scalable and anatomically faithful framework for human motion analysis, facilitating the use of computer vision techniques in biomechanics-related analysis. Our implementation is available on the project page: https://pokerman8.github.io/SKEL-CF/.", "date": "2025-11-25", "categories": []}
{"id": "2603.14249", "title": "OAHuman: Occlusion-Aware 3D Human Reconstruction from Monocular Images", "abstract": "Monocular 3D human reconstruction in real-world scenarios remains highly challenging due to frequent occlusions from surrounding objects, people, or image truncation. Such occlusions lead to missing geometry and unreliable appearance cues, severely degrading the completeness and realism of reconstructed human models. Although recent neural implicit methods achieve impressive results on clean inputs, they struggle under occlusion due to entangled modeling of shape and texture. In this paper, we propose OAHuman, an occlusion-aware framework that explicitly decouples geometry reconstruction and texture synthesis for robust 3D human modeling from a single RGB image. The core innovation lies in the decoupling-perception paradigm, which addresses the fundamental issue of geometry-texture cross-contamination in occluded regions. Our framework ensures that geometry reconstruction is perceptually reinforced even in occluded areas, isolating it from texture interference. In parallel, texture synthesis is learned exclusively from visible regions, preventing texture errors from being transferred to the occluded areas. This decoupling approach enables OAHuman to achieve robust and high-fidelity reconstruction under occlusion, which has been a long-standing challenge in the field. Extensive experiments on occlusion-rich benchmarks demonstrate that OAHuman achieves superior performance in terms of structural completeness, surface detail, and texture realism, significantly improving monocular 3D human reconstruction under occlusion conditions.", "date": "2026-03-15", "categories": []}
{"id": "2603.14435", "title": "End-to-End Spatial-Temporal Transformer for Real-time 4D HOI Reconstruction", "abstract": "Monocular 4D human-object interaction (HOI) reconstruction - recovering a moving human and a manipulated object from a single RGB video - remains challenging due to depth ambiguity and frequent occlusions. Existing methods often rely on multi-stage pipelines or iterative optimization, leading to high inference latency, failing to meet real-time requirements, and susceptibility to error accumulation. To address these limitations, we propose THO, an end-to-end Spatial-Temporal Transformer that predicts human motion and coordinated object motion in a forward fashion from the given video and 3D template. THO achieves this by leveraging spatial-temporal HOI tuple priors. Spatial priors exploit contact-region proximity to infer occluded object features from human cues, while temporal priors capture cross-frame kinematic correlations to refine object representations and enforce physical coherence. Extensive experiments demonstrate that THO operates at an inference speed of 31.5 FPS on a single RTX 4090 GPU, achieving a >600x speedup over prior optimization-based methods while simultaneously improving reconstruction accuracy and temporal consistency. The project page is available at: https://nianheng.github.io/THO-project/", "date": "2026-03-15", "categories": []}
{"id": "2603.14772", "title": "Zero-Shot Reconstruction of Animatable 3D Avatars with Cloth Dynamics from a Single Image", "abstract": "Existing single-image 3D human avatar methods primarily rely on rigid joint transformations, limiting their ability to model realistic cloth dynamics. We present DynaAvatar, a zero-shot framework that reconstructs animatable 3D human avatars with motion-dependent cloth dynamics from a single image. Trained on large-scale multi-person motion datasets, DynaAvatar employs a Transformer-based feed-forward architecture that directly predicts dynamic 3D Gaussian deformations without subject-specific optimization. To overcome the scarcity of dynamic captures, we introduce a static-to-dynamic knowledge transfer strategy: a Transformer pretrained on large-scale static captures provides strong geometric and appearance priors, which are efficiently adapted to motion-dependent deformations through lightweight LoRA fine-tuning on dynamic captures. We further propose the DynaFlow loss, an optical flow-guided objective that provides reliable motion-direction geometric cues for cloth dynamics in rendered space. Finally, we reannotate the missing or noisy SMPL-X fittings in existing dynamic capture datasets, as most public dynamic capture datasets contain incomplete or unreliable fittings that are unsuitable for training high-quality 3D avatar reconstruction models. Experiments demonstrate that DynaAvatar produces visually rich and generalizable animations, outperforming prior methods.", "date": "2026-03-16", "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": "2603.15603", "title": "Fast SAM 3D Body: Accelerating SAM 3D Body for Real-Time Full-Body Human Mesh Recovery", "abstract": "SAM 3D Body (3DB) achieves state-of-the-art accuracy in monocular 3D human mesh recovery, yet its inference latency of several seconds per image precludes real-time application. We present Fast SAM 3D Body, a training-free acceleration framework that reformulates the 3DB inference pathway to achieve interactive rates. By decoupling serial spatial dependencies and applying architecture-aware pruning, we enable parallelized multi-crop feature extraction and streamlined transformer decoding. Moreover, to extract the joint-level kinematics (SMPL) compatible with existing humanoid control and policy learning frameworks, we replace the iterative mesh fitting with a direct feedforward mapping, accelerating this specific conversion by over 10,000x. Overall, our framework delivers up to a 10.9x end-to-end speedup while maintaining on-par reconstruction fidelity, even surpassing 3DB on benchmarks such as LSPET. We demonstrate its utility by deploying Fast SAM 3D Body in a vision-only teleoperation system that-unlike methods reliant on wearable IMUs-enables real-time humanoid control and the direct collection of manipulation policies from a single RGB stream.", "date": "2026-03-16", "categories": []}
{"id": "2603.15612", "title": "HSImul3R: Physics-in-the-Loop Reconstruction of Simulation-Ready Human-Scene Interactions", "abstract": "We present HSImul3R, a unified framework for simulation-ready 3D reconstruction of human-scene interactions (HSI) from casual captures, including sparse-view images and monocular videos. Existing methods suffer from a perception-simulation gap: visually plausible reconstructions often violate physical constraints, leading to instability in physics engines and failure in embodied AI applications. To bridge this gap, we introduce a physically-grounded bi-directional optimization pipeline that treats the physics simulator as an active supervisor to jointly refine human dynamics and scene geometry. In the forward direction, we employ Scene-targeted Reinforcement Learning to optimize human motion under dual supervision of motion fidelity and contact stability. In the reverse direction, we propose Direct Simulation Reward Optimization, which leverages simulation feedback on gravitational stability and interaction success to refine scene geometry. We further present HSIBench, a new benchmark with diverse objects and interaction scenarios. Extensive experiments demonstrate that HSImul3R produces the first stable, simulation-ready HSI reconstructions and can be directly deployed to real-world humanoid robots.", "date": "2026-03-16", "categories": []}
{"id": "2601.02267", "title": "DiffProxy: Multi-View Human Mesh Recovery via Diffusion-Generated Dense Proxies", "abstract": "Precise human mesh recovery (HMR) from multi-view images remains challenging: end-to-end methods produce entangled errors hard to localize, while fitting-based methods rely on sparse keypoints that provide limited surface constraints. We observe that the true bottleneck lies in the quality of intermediate representations, and that dense pixel-to-surface correspondences can be effectively generated by repurposing pre-trained diffusion models with rich visual priors. We propose DiffProxy, a Stable-Diffusion-based framework trained on large-scale synthetic data with pixel-perfect annotations. A multi-conditional proxy generator predicts dense correspondences from multi-view images, providing uniform surface constraints that enable precise fitting. Hand refinement feeds enlarged hand crops alongside full-body images for fine-grained detail, while test-time scaling exploits diffusion stochasticity to estimate per-pixel uncertainty. Trained only on synthetic data, DiffProxy achieves state-of-the-art results on five diverse real-world benchmarks. Project page: https://wrk226.github.io/DiffProxy.html", "date": "2026-01-05", "categories": []}
{"id": "2602.23951", "title": "AHAP: Reconstructing Arbitrary Humans from Arbitrary Perspectives with Geometric Priors", "abstract": "Reconstructing 3D humans from images captured at multiple perspectives typically requires pre-calibration, like using checkerboards or MVS algorithms, which limits scalability and applicability in diverse real-world scenarios. In this work, we present AHAP (Reconstructing Arbitrary Humans from Arbitrary Perspectives), a feed-forward framework for reconstructing arbitrary humans from arbitrary camera perspectives without requiring camera calibration. Our core lies in the effective fusion of multi-view geometry to assist human association, reconstruction and localization. Specifically, we use a Cross-View Identity Association module through learnable person queries and soft assignment, supervised by contrastive learning to resolve cross-view human identity association. A Human Head fuses cross-view features and scene context for SMPL prediction, guided by cross-view reprojection losses to enforce body pose consistency. Additionally, multi-view geometry eliminates the depth ambiguity inherent in monocular methods, providing more precise 3D human localization through multi-view triangulation. Experiments on EgoHumans and EgoExo4D demonstrate that AHAP achieves competitive performance on both world-space human reconstruction and camera pose estimation, while being 180$\\times$ faster than optimization-based approaches.", "date": "2026-02-27", "categories": []}
{"id": "2602.24161", "title": "GeoDiff4D: Geometry-Aware Diffusion for 4D Head Avatar Reconstruction", "abstract": "Reconstructing photorealistic and animatable 4D head avatars from a single portrait image remains a fundamental challenge in computer vision. While diffusion models have enabled remarkable progress in image and video generation for avatar reconstruction, existing methods primarily rely on 2D priors and struggle to achieve consistent 3D geometry. We propose a novel framework that leverages geometry-aware diffusion to learn strong geometry priors for high-fidelity head avatar reconstruction. Our approach jointly synthesizes portrait images and corresponding surface normals, while a pose-free expression encoder captures implicit expression representations. Both synthesized images and expression latents are incorporated into 3D Gaussian-based avatars, enabling photorealistic rendering with accurate geometry. Extensive experiments demonstrate that our method substantially outperforms state-of-the-art approaches in visual quality, expression fidelity, and cross-identity generalization, while supporting real-time rendering.", "date": "2026-02-27", "categories": []}
{"id": "2603.12063", "title": "NBAvatar: Neural Billboards Avatars with Realistic Hand-Face Interaction", "abstract": "We present NBAvatar - a method for realistic rendering of head avatars handling non-rigid deformations caused by hand-face interaction. We introduce a novel representation for animated avatars by combining the training of oriented planar primitives with neural rendering. Such a combination of explicit and implicit representations enables NBAvatar to handle temporally and pose-consistent geometry, along with fine-grained appearance details provided by the neural rendering technique. In our experiments, we demonstrate that NBAvatar implicitly learns color transformations caused by face-hand interactions and surpasses existing approaches in terms of novel-view and novel-pose rendering quality. Specifically, NBAvatar achieves up to 30% LPIPS reduction under high-resolution megapixel rendering compared to Gaussian-based avatar methods, while also improving PSNR and SSIM, and achieves higher structural similarity compared to the state-of-the-art hand-face interaction method InteractAvatar.", "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": "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": "2603.07043", "title": "Fine-Grained 3D Facial Reconstruction for Micro-Expressions", "abstract": "Recent advances in 3D facial expression reconstruction have demonstrated remarkable performance in capturing macro-expressions, yet the reconstruction of micro-expressions remains unexplored. This novel task is particularly challenging due to the subtle, transient, and low-intensity nature of micro-expressions, which complicate the extraction of stable and discriminative features essential for accurate reconstruction. In this paper, we propose a fine-grained micro-expression reconstruction method that integrates a global dynamic feature capturing stable facial motion patterns with a locally-enriched feature incorporating multiple informative cues from 2D motions, facial priors and 3D facial geometry. Specifically, we devise a plug-and-play dynamic-encoded module to extract micro-expression feature for global facial action, allowing it to leverage prior knowledge from abundant macro-expression data to mitigate the scarcity of micro-expression data. Subsequently, a dynamic-guided mesh deformation module is designed for extracting aggregated local features from dense optical flow, sparse landmark cues and facial mesh geometry, which adaptively refines fine-grained facial micro-expression without compromising global 3D geometry. Extensive experiments on micro-expression datasets demonstrate that our method consistently outperforms state-of-the-art methods in both geometric accuracy and perceptual detail.", "date": "2026-03-07", "categories": []}
{"id": "2411.16758", "title": "Motion-Aware Animatable Gaussian Avatars Deblurring", "abstract": "The creation of 3D human avatars from multi-view videos is a significant yet challenging task in computer vision. However, existing techniques rely on high-quality, sharp images as input, which are often impractical to obtain in real-world scenarios due to variations in human motion speed and intensity. This paper introduces a novel method for directly reconstructing sharp 3D human Gaussian avatars from blurry videos. The proposed approach incorporates a 3D-aware, physics-based model of blur formation caused by human motion, together with a 3D human motion model designed to resolve ambiguities in motion-induced blur. This framework enables the joint optimization of the avatar representation and motion parameters from a coarse initialization. Comprehensive benchmarks are established using both a synthetic dataset and a real-world dataset captured with a 360-degree synchronous hybrid-exposure camera system. Extensive evaluations demonstrate the effectiveness of the model across diverse conditions. Codes Available: https://github.com/MyNiuuu/MAD-Avatar", "date": "2024-11-24", "categories": []}
{"id": "2511.19854", "title": "STAvatar: Soft Binding and Temporal Density Control for Monocular 3D Head Avatars Reconstruction", "abstract": "Reconstructing high-fidelity and animatable 3D head avatars from monocular videos remains a challenging yet essential task. Existing methods based on 3D Gaussian Splatting typically bind Gaussians to mesh triangles and model deformations solely via Linear Blend Skinning, which results in rigid motion and limited expressiveness. Moreover, they lack specialized strategies to handle frequently occluded regions (e.g., mouth interiors, eyelids). To address these limitations, we propose STAvatar, which consists of two key components: (1) a UV-Adaptive Soft Binding framework that leverages both image-based and geometric priors to learn per-Gaussian feature offsets within the UV space. This UV representation supports dynamic resampling, ensuring full compatibility with Adaptive Density Control (ADC) and enhanced adaptability to shape and textural variations. (2) a Temporal ADC strategy, which first clusters structurally similar frames to facilitate more targeted computation of the densification criterion. It further introduces a novel fused perceptual error as clone criterion to jointly capture geometric and textural discrepancies, encouraging densification in regions requiring finer details. Extensive experiments on four benchmark datasets demonstrate that STAvatar achieves state-of-the-art reconstruction performance, especially in capturing fine-grained details and reconstructing frequently occluded regions.", "date": "2025-11-25", "categories": []}
{"id": "2603.04290", "title": "Gaussian Wardrobe: Compositional 3D Gaussian Avatars for Free-Form Virtual Try-On", "abstract": "We introduce Gaussian Wardrobe, a novel framework to digitalize compositional 3D neural avatars from multi-view videos. Existing methods for 3D neural avatars typically treat the human body and clothing as an inseparable entity. However, this paradigm fails to capture the dynamics of complex free-form garments and limits the reuse of clothing across different individuals. To overcome these problems, we develop a novel, compositional 3D Gaussian representation to build avatars from multiple layers of free-form garments. The core of our method is decomposing neural avatars into bodies and layers of shape-agnostic neural garments. To achieve this, our framework learns to disentangle each garment layer from multi-view videos and canonicalizes it into a shape-independent space. In experiments, our method models photorealistic avatars with high-fidelity dynamics, achieving new state-of-the-art performance on novel pose synthesis benchmarks. In addition, we demonstrate that the learned compositional garments contribute to a versatile digital wardrobe, enabling a practical virtual try-on application where clothing can be freely transferred to new subjects. Project page: https://ait.ethz.ch/gaussianwardrobe", "date": "2026-03-04", "categories": []}
{"id": "2603.04958", "title": "Revisiting an Old Perspective Projection for Monocular 3D Morphable Models Regression", "abstract": "We introduce a novel camera model for monocular 3D Morphable Model (3DMM) regression methods that effectively captures the perspective distortion effect commonly seen in close-up facial images.\n  Fitting 3D morphable models to video is a key technique in content creation. In particular, regression-based approaches have produced fast and accurate results by matching the rendered output of the morphable model to the target image. These methods typically achieve stable performance with orthographic projection, which eliminates the ambiguity between focal length and object distance. However, this simplification makes them unsuitable for close-up footage, such as that captured with head-mounted cameras.\n  We extend orthographic projection with a new shrinkage parameter, incorporating a pseudo-perspective effect while preserving the stability of the original projection. We present several techniques that allow finetuning of existing models, and demonstrate the effectiveness of our modification through both quantitative and qualitative comparisons using a custom dataset recorded with head-mounted cameras.", "date": "2026-03-05", "categories": []}
{"id": "2603.04993", "title": "MultiGO++: Monocular 3D Clothed Human Reconstruction via Geometry-Texture Collaboration", "abstract": "Monocular 3D clothed human reconstruction aims to generate a complete and realistic textured 3D avatar from a single image. Existing methods are commonly trained under multi-view supervision with annotated geometric priors, and during inference, these priors are estimated by the pre-trained network from the monocular input. These methods are constrained by three key limitations: texturally by unavailability of training data, geometrically by inaccurate external priors, and systematically by biased single-modality supervision, all leading to suboptimal reconstruction. To address these issues, we propose a novel reconstruction framework, named MultiGO++, which achieves effective systematic geometry-texture collaboration. It consists of three core parts: (1) A multi-source texture synthesis strategy that constructs 15,000+ 3D textured human scans to improve the performance on texture quality estimation in challenge scenarios; (2) A region-aware shape extraction module that extracts and interacts features of each body region to obtain geometry information and a Fourier geometry encoder that mitigates the modality gap to achieve effective geometry learning; (3) A dual reconstruction U-Net that leverages geometry-texture collaborative features to refine and generate high-fidelity textured 3D human meshes. Extensive experiments on two benchmarks and many in-the-wild cases show the superiority of our method over state-of-the-art approaches.", "date": "2026-03-05", "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": "2510.06219", "title": "Human3R: Everyone Everywhere All at Once", "abstract": "We present Human3R, a unified, feed-forward framework for online 4D human-scene reconstruction, in the world frame, from casually captured monocular videos. Unlike previous approaches that rely on multi-stage pipelines, iterative contact-aware refinement between humans and scenes, and heavy dependencies, e.g., human detection, depth estimation, and SLAM pre-processing, Human3R jointly recovers global multi-person SMPL-X bodies (\"everyone\"), dense 3D scene (\"everywhere\"), and camera trajectories in a single forward pass (\"all-at-once\"). Our method builds upon the 4D online reconstruction model CUT3R, and uses parameter-efficient visual prompt tuning, to strive to preserve CUT3R's rich spatiotemporal priors, while enabling direct readout of multiple SMPL-X bodies. Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, in real-time (15 FPS) with a low memory footprint (8 GB). Extensive experiments demonstrate that Human3R delivers state-of-the-art or competitive performance across tasks, including global human motion estimation, local human mesh recovery, video depth estimation, and camera pose estimation, with a single unified model. We hope that Human3R will serve as a simple yet strong baseline, which can be easily adapted for downstream applications. Code, models and 4D interactive demos are available at https://fanegg.github.io/Human3R/.", "date": "2025-10-07", "categories": []}
{"id": "2603.02499", "title": "Biomechanically Accurate Gait Analysis: A 3d Human Reconstruction Framework for Markerless Estimation of Gait Parameters", "abstract": "This paper presents a biomechanically interpretable framework for gait analysis using 3D human reconstruction from video data. Unlike conventional keypoint based approaches, the proposed method extracts biomechanically meaningful markers analogous to motion capture systems and integrates them within OpenSim for joint kinematic estimation. To evaluate performance, both spatiotemporal and kinematic gait parameters were analysed against reference marker-based data. Results indicate strong agreement with marker-based measurements, with considerable improvements when compared with pose-estimation methods alone. The proposed framework offers a scalable, markerless, and interpretable approach for accurate gait assessment, supporting broader clinical and real world deployment of vision based biomechanics", "date": "2026-03-03", "categories": []}
{"id": "2603.02619", "title": "Direct Reward Fine-Tuning on Poses for Single Image to 3D Human in the Wild", "abstract": "Single-view 3D human reconstruction has achieved remarkable progress through the adoption of multi-view diffusion models, yet the recovered 3D humans often exhibit unnatural poses. This phenomenon becomes pronounced when reconstructing 3D humans with dynamic or challenging poses, which we attribute to the limited scale of available 3D human datasets with diverse poses. To address this limitation, we introduce DrPose, Direct Reward fine-tuning algorithm on Poses, which enables post-training of a multi-view diffusion model on diverse poses without requiring expensive 3D human assets. DrPose trains a model using only human poses paired with single-view images, employing a direct reward fine-tuning to maximize PoseScore, which is our proposed differentiable reward that quantifies consistency between a generated multi-view latent image and a ground-truth human pose. This optimization is conducted on DrPose15K, a novel dataset that was constructed from an existing human motion dataset and a pose-conditioned video generative model. Constructed from abundant human pose sequence data, DrPose15K exhibits a broader pose distribution compared to existing 3D human datasets. We validate our approach through evaluation on conventional benchmark datasets, in-the-wild images, and a newly constructed benchmark, with a particular focus on assessing performance on challenging human poses. Our results demonstrate consistent qualitative and quantitative improvements across all benchmarks. Project page: https://seunguk-do.github.io/drpose.", "date": "2026-03-03", "categories": []}
{"id": "2603.02893", "title": "Intrinsic Geometry-Appearance Consistency Optimization for Sparse-View Gaussian Splatting", "abstract": "3D human reconstruction from a single image is a challenging problem and has been exclusively studied in the literature. Recently, some methods have resorted to diffusion models for guidance, optimizing a 3D representation via Score Distillation Sampling(SDS) or generating a back-view image for facilitating reconstruction. However, these methods tend to produce unsatisfactory artifacts (\\textit{e.g.} flattened human structure or over-smoothing results caused by inconsistent priors from multiple views) and struggle with real-world generalization in the wild. In this work, we present \\emph{MVD-HuGaS}, enabling free-view 3D human rendering from a single image via a multi-view human diffusion model. We first generate multi-view images from the single reference image with an enhanced multi-view diffusion model, which is well fine-tuned on high-quality 3D human datasets to incorporate 3D geometry priors and human structure priors. To infer accurate camera poses from the sparse generated multi-view images for reconstruction, an alignment module is introduced to facilitate joint optimization of 3D Gaussians and camera poses. Furthermore, we propose a depth-based Facial Distortion Mitigation module to refine the generated facial regions, thereby improving the overall fidelity of the reconstruction. Finally, leveraging the refined multi-view images, along with their accurate camera poses, MVD-HuGaS optimizes the 3D Gaussians of the target human for high-fidelity free-view renderings. Extensive experiments on Thuman2.0 and 2K2K datasets show that the proposed MVD-HuGaS achieves state-of-the-art performance on single-view 3D human rendering.", "date": "2026-03-03", "categories": []}
{"id": "2603.03265", "title": "DuoMo: Dual Motion Diffusion for World-Space Human Reconstruction", "abstract": "We present DuoMo, a generative method that recovers human motion in world-space coordinates from unconstrained videos with noisy or incomplete observations. Reconstructing such motion requires solving a fundamental trade-off: generalizing from diverse and noisy video inputs while maintaining global motion consistency. Our approach addresses this problem by factorizing motion learning into two diffusion models. The camera-space model first estimates motion from videos in camera coordinates. The world-space model then lifts this initial estimate into world coordinates and refines it to be globally consistent. Together, the two models can reconstruct motion across diverse scenes and trajectories, even from highly noisy or incomplete observations. Moreover, our formulation is general, generating the motion of mesh vertices directly and bypassing parametric models. DuoMo achieves state-of-the-art performance. On EMDB, our method obtains a 16% reduction in world-space reconstruction error while maintaining low foot skating. On RICH, it obtains a 30% reduction in world-space error. Project page: https://yufu-wang.github.io/duomo/", "date": "2026-03-03", "categories": []}
{"id": "2508.19754", "title": "FastAvatar: Towards Unified and Fast 3D Avatar Reconstruction with Large Gaussian Reconstruction Transformers", "abstract": "Despite significant progress in 3D avatar reconstruction, it still faces challenges such as high time complexity, sensitivity to data quality, and low data utilization. We propose FastAvatar, a feedforward 3D avatar framework capable of flexibly leveraging diverse daily recordings (e.g., a single image, multi-view observations, or monocular video) to reconstruct a high-quality 3D Gaussian Splatting (3DGS) model within seconds, using only a single unified model. The core of FastAvatar is a Large Gaussian Reconstruction Transformer (LGRT) featuring three key designs: First, a 3DGS transformer aggregating multi-frame cues while injecting initial 3D prompt to predict the corresponding registered canonical 3DGS representations; Second, multi-granular guidance encoding (camera pose, expression coefficient, head pose) mitigating animation-induced misalignment for variable-length inputs; Third, incremental Gaussian aggregation via landmark tracking and sliced fusion losses. Integrating these features, FastAvatar enables incremental reconstruction, i.e., improving quality with more observations without wasting input data as in previous works. This yields a quality-speed-tunable paradigm for highly usable 3D avatar modeling. Extensive experiments show that FastAvatar has a higher quality and highly competitive speed compared to existing methods.", "date": "2025-08-27", "categories": []}
{"id": "2602.24043", "title": "Spatio-Temporal Garment Reconstruction Using Diffusion Mapping via Pattern Coordinates", "abstract": "Reconstructing 3D clothed humans from monocular images and videos is a fundamental problem with applications in virtual try-on, avatar creation, and mixed reality. Despite significant progress in human body recovery, accurately reconstructing garment geometry, particularly for loose-fitting clothing, remains an open challenge. We propose a unified framework for high-fidelity 3D garment reconstruction from both single images and video sequences. Our approach combines Implicit Sewing Patterns (ISP) with a generative diffusion model to learn expressive garment shape priors in 2D UV space. Leveraging these priors, we introduce a mapping model that establishes correspondences between image pixels, UV pattern coordinates, and 3D geometry, enabling accurate and detailed garment reconstruction from single images. We further extend this formulation to dynamic reconstruction by introducing a spatio-temporal diffusion scheme with test-time guidance to enforce long-range temporal consistency. We also develop analytic projection-based constraints that preserve image-aligned geometry in visible regions while enforcing coherent completion in occluded areas over time. Although trained exclusively on synthetically simulated cloth data, our method generalizes well to real-world imagery and consistently outperforms existing approaches on both tight- and loose-fitting garments. The reconstructed garments preserve fine geometric detail while exhibiting realistic dynamic motion, supporting downstream applications such as texture editing, garment retargeting, and animation.", "date": "2026-02-27", "categories": []}
{"id": "2602.24148", "title": "HumanOrbit: 3D Human Reconstruction as 360{\\deg} Orbit Generation", "abstract": "We present a method for generating a full 360{\\deg} orbit video around a person from a single input image. Existing methods typically adapt image-based diffusion models for multi-view synthesis, but yield inconsistent results across views and with the original identity. In contrast, recent video diffusion models have demonstrated their ability in generating photorealistic results that align well with the given prompts. Inspired by these results, we propose HumanOrbit, a video diffusion model for multi-view human image generation. Our approach enables the model to synthesize continuous camera rotations around the subject, producing geometrically consistent novel views while preserving the appearance and identity of the person. Using the generated multi-view frames, we further propose a reconstruction pipeline that recovers a textured mesh of the subject. Experimental results validate the effectiveness of HumanOrbit for multi-view image generation and that the reconstructed 3D models exhibit superior completeness and fidelity compared to those from state-of-the-art baselines.", "date": "2026-02-27", "categories": []}
{"id": "2602.20556", "title": "WildGHand: Learning Anti-Perturbation Gaussian Hand Avatars from Monocular In-the-Wild Videos", "abstract": "Despite recent progress in 3D hand reconstruction from monocular videos, most existing methods rely on data captured in well-controlled environments and therefore degrade in real-world settings with severe perturbations, such as hand-object interactions, extreme poses, illumination changes, and motion blur. To tackle these issues, we introduce WildGHand, an optimization-based framework that enables self-adaptive 3D Gaussian splatting on in-the-wild videos and produces high-fidelity hand avatars. WildGHand incorporates two key components: (i) a dynamic perturbation disentanglement module that explicitly represents perturbations as time-varying biases on 3D Gaussian attributes during optimization, and (ii) a perturbation-aware optimization strategy that generates per-frame anisotropic weighted masks to guide optimization. Together, these components allow the framework to identify and suppress perturbations across both spatial and temporal dimensions. We further curate a dataset of monocular hand videos captured under diverse perturbations to benchmark in-the-wild hand avatar reconstruction. Extensive experiments on this dataset and two public datasets demonstrate that WildGHand achieves state-of-the-art performance and substantially improves over its base model across multiple metrics (e.g., up to a $15.8\\%$ relative gain in PSNR and a $23.1\\%$ relative reduction in LPIPS). Our implementation and dataset are available at https://github.com/XuanHuang0/WildGHand.", "date": "2026-02-24", "categories": []}
{"id": "2602.19180", "title": "VLM-Guided Group Preference Alignment for Diffusion-based Human Mesh Recovery", "abstract": "Human mesh recovery (HMR) from a single RGB image is inherently ambiguous, as multiple 3D poses can correspond to the same 2D observation. Recent diffusion-based methods tackle this by generating various hypotheses, but often sacrifice accuracy. They yield predictions that are either physically implausible or drift from the input image, especially under occlusion or in cluttered, in-the-wild scenes. To address this, we introduce a dual-memory augmented HMR critique agent with self-reflection to produce context-aware quality scores for predicted meshes. These scores distill fine-grained cues about 3D human motion structure, physical feasibility, and alignment with the input image. We use these scores to build a group-wise HMR preference dataset. Leveraging this dataset, we propose a group preference alignment framework for finetuning diffusion-based HMR models. This process injects the rich preference signals into the model, guiding it to generate more physically plausible and image-consistent human meshes. Extensive experiments demonstrate that our method achieves superior performance compared to state-of-the-art approaches.", "date": "2026-02-22", "categories": []}
{"id": "2602.19679", "title": "TeHOR: Text-Guided 3D Human and Object Reconstruction with Textures", "abstract": "Joint reconstruction of 3D human and object from a single image is an active research area, with pivotal applications in robotics and digital content creation. Despite recent advances, existing approaches suffer from two fundamental limitations. First, their reconstructions rely heavily on physical contact information, which inherently cannot capture non-contact human-object interactions, such as gazing at or pointing toward an object. Second, the reconstruction process is primarily driven by local geometric proximity, neglecting the human and object appearances that provide global context crucial for understanding holistic interactions. To address these issues, we introduce TeHOR, a framework built upon two core designs. First, beyond contact information, our framework leverages text descriptions of human-object interactions to enforce semantic alignment between the 3D reconstruction and its textual cues, enabling reasoning over a wider spectrum of interactions, including non-contact cases. Second, we incorporate appearance cues of the 3D human and object into the alignment process to capture holistic contextual information, thereby ensuring visually plausible reconstructions. As a result, our framework produces accurate and semantically coherent reconstructions, achieving state-of-the-art performance.", "date": "2026-02-23", "categories": []}
{"id": "2307.05000", "title": "Neural Point-based Volumetric Avatar: Surface-guided Neural Points for Efficient and Photorealistic Volumetric Head Avatar", "abstract": "Rendering photorealistic and dynamically moving human heads is crucial for ensuring a pleasant and immersive experience in AR/VR and video conferencing applications. However, existing methods often struggle to model challenging facial regions (e.g., mouth interior, eyes, hair/beard), resulting in unrealistic and blurry results. In this paper, we propose {\\fullname} ({\\name}), a method that adopts the neural point representation as well as the neural volume rendering process and discards the predefined connectivity and hard correspondence imposed by mesh-based approaches. Specifically, the neural points are strategically constrained around the surface of the target expression via a high-resolution UV displacement map, achieving increased modeling capacity and more accurate control. We introduce three technical innovations to improve the rendering and training efficiency: a patch-wise depth-guided (shading point) sampling strategy, a lightweight radiance decoding process, and a Grid-Error-Patch (GEP) ray sampling strategy during training. By design, our {\\name} is better equipped to handle topologically changing regions and thin structures while also ensuring accurate expression control when animating avatars. Experiments conducted on three subjects from the Multiface dataset demonstrate the effectiveness of our designs, outperforming previous state-of-the-art methods, especially in handling challenging facial regions.", "date": "2023-07-11", "categories": []}
{"id": "2602.15989", "title": "SAM 3D Body: Robust Full-Body Human Mesh Recovery", "abstract": "We introduce SAM 3D Body (3DB), a promptable model for single-image full-body 3D human mesh recovery (HMR) that demonstrates state-of-the-art performance, with strong generalization and consistent accuracy in diverse in-the-wild conditions. 3DB estimates the human pose of the body, feet, and hands. It is the first model to use a new parametric mesh representation, Momentum Human Rig (MHR), which decouples skeletal structure and surface shape. 3DB employs an encoder-decoder architecture and supports auxiliary prompts, including 2D keypoints and masks, enabling user-guided inference similar to the SAM family of models. We derive high-quality annotations from a multi-stage annotation pipeline that uses various combinations of manual keypoint annotation, differentiable optimization, multi-view geometry, and dense keypoint detection. Our data engine efficiently selects and processes data to ensure data diversity, collecting unusual poses and rare imaging conditions. We present a new evaluation dataset organized by pose and appearance categories, enabling nuanced analysis of model behavior. Our experiments demonstrate superior generalization and substantial improvements over prior methods in both qualitative user preference studies and traditional quantitative analysis. Both 3DB and MHR are open-source.", "date": "2026-02-17", "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": "2602.09736", "title": "Toward Fine-Grained Facial Control in 3D Talking Head Generation", "abstract": "Audio-driven talking head generation is a core component of digital avatars, and 3D Gaussian Splatting has shown strong performance in real-time rendering of high-fidelity talking heads. However, achieving precise control over fine-grained facial movements remains a significant challenge, particularly due to lip-synchronization inaccuracies and facial jitter, both of which can contribute to the uncanny valley effect. To address these challenges, we propose Fine-Grained 3D Gaussian Splatting (FG-3DGS), a novel framework that enables temporally consistent and high-fidelity talking head generation. Our method introduces a frequency-aware disentanglement strategy to explicitly model facial regions based on their motion characteristics. Low-frequency regions, such as the cheeks, nose, and forehead, are jointly modeled using a standard MLP, while high-frequency regions, including the eyes and mouth, are captured separately using a dedicated network guided by facial area masks. The predicted motion dynamics, represented as Gaussian deltas, are applied to the static Gaussians to generate the final head frames, which are rendered via a rasterizer using frame-specific camera parameters. Additionally, a high-frequency-refined post-rendering alignment mechanism, learned from large-scale audio-video pairs by a pretrained model, is incorporated to enhance per-frame generation and achieve more accurate lip synchronization. Extensive experiments on widely used datasets for talking head generation demonstrate that our method outperforms recent state-of-the-art approaches in producing high-fidelity, lip-synced talking head videos.", "date": "2026-02-10", "categories": []}
{"id": "2602.09918", "title": "SARS: A Novel Face and Body Shape and Appearance Aware 3D Reconstruction System extends Morphable Models", "abstract": "Morphable Models (3DMMs) are a type of morphable model that takes 2D images as inputs and recreates the structure and physical appearance of 3D objects, especially human faces and bodies. 3DMM combines identity and expression blendshapes with a basic face mesh to create a detailed 3D model. The variability in the 3D Morphable models can be controlled by tuning diverse parameters. They are high-level image descriptors, such as shape, texture, illumination, and camera parameters. Previous research in 3D human reconstruction concentrated solely on global face structure or geometry, ignoring face semantic features such as age, gender, and facial landmarks characterizing facial boundaries, curves, dips, and wrinkles. In order to accommodate changes in these high-level facial characteristics, this work introduces a shape and appearance-aware 3D reconstruction system (named SARS by us), a c modular pipeline that extracts body and face information from a single image to properly rebuild the 3D model of the human full body.", "date": "2026-02-10", "categories": []}
{"id": "2511.05403", "title": "PALM: A Dataset and Baseline for Learning Multi-subject Hand Prior", "abstract": "The ability to grasp objects, signal with gestures, and share emotion through touch all stem from the unique capabilities of human hands. Yet creating high-quality personalized hand avatars from images remains challenging due to complex geometry, appearance, and articulation, particularly under unconstrained lighting and limited views. Progress has also been limited by the lack of datasets that jointly provide accurate 3D geometry, high-resolution multiview imagery, and a diverse population of subjects. To address this, we present PALM, a large-scale dataset comprising 13k high-quality hand scans from 263 subjects and 90k multi-view images, capturing rich variation in skin tone, age, and geometry. To show its utility, we present a baseline PALM-Net, a multi-subject prior over hand geometry and material properties learned via physically based inverse rendering, enabling realistic, relightable single-image hand avatar personalization. PALM's scale and diversity make it a valuable real-world resource for hand modeling and related research.", "date": "2025-11-07", "categories": []}
{"id": "2602.06122", "title": "From Blurry to Believable: Enhancing Low-quality Talking Heads with 3D Generative Priors", "abstract": "Creating high-fidelity, animatable 3D talking heads is crucial for immersive applications, yet often hindered by the prevalence of low-quality image or video sources, which yield poor 3D reconstructions. In this paper, we introduce SuperHead, a novel framework for enhancing low-resolution, animatable 3D head avatars. The core challenge lies in synthesizing high-quality geometry and textures, while ensuring both 3D and temporal consistency during animation and preserving subject identity. Despite recent progress in image, video and 3D-based super-resolution (SR), existing SR techniques are ill-equipped to handle dynamic 3D inputs. To address this, SuperHead leverages the rich priors from pre-trained 3D generative models via a novel dynamics-aware 3D inversion scheme. This process optimizes the latent representation of the generative model to produce a super-resolved 3D Gaussian Splatting (3DGS) head model, which is subsequently rigged to an underlying parametric head model (e.g., FLAME) for animation. The inversion is jointly supervised using a sparse collection of upscaled 2D face renderings and corresponding depth maps, captured from diverse facial expressions and camera viewpoints, to ensure realism under dynamic facial motions. Experiments demonstrate that SuperHead generates avatars with fine-grained facial details under dynamic motions, significantly outperforming baseline methods in visual quality.", "date": "2026-02-05", "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": "2601.22693", "title": "PEAR: Pixel-aligned Expressive humAn mesh Recovery", "abstract": "Reconstructing detailed 3D human meshes from a single in-the-wild image remains a fundamental challenge in computer vision. Existing SMPLX-based methods often suffer from slow inference, produce only coarse body poses, and exhibit misalignments or unnatural artifacts in fine-grained regions such as the face and hands. These issues make current approaches difficult to apply to downstream tasks. To address these challenges, we propose PEAR-a fast and robust framework for pixel-aligned expressive human mesh recovery. PEAR explicitly tackles three major limitations of existing methods: slow inference, inaccurate localization of fine-grained human pose details, and insufficient facial expression capture. Specifically, to enable real-time SMPLX parameter inference, we depart from prior designs that rely on high resolution inputs or multi-branch architectures. Instead, we adopt a clean and unified ViT-based model capable of recovering coarse 3D human geometry. To compensate for the loss of fine-grained details caused by this simplified architecture, we introduce pixel-level supervision to optimize the geometry, significantly improving the reconstruction accuracy of fine-grained human details. To make this approach practical, we further propose a modular data annotation strategy that enriches the training data and enhances the robustness of the model. Overall, PEAR is a preprocessing-free framework that can simultaneously infer EHM-s (SMPLX and scaled-FLAME) parameters at over 100 FPS. Extensive experiments on multiple benchmark datasets demonstrate that our method achieves substantial improvements in pose estimation accuracy compared to previous SMPLX-based approaches. Project page: https://wujh2001.github.io/PEAR", "date": "2026-01-30", "categories": []}
{"id": "2602.05190", "title": "PoseGaussian: Pose-Driven Novel View Synthesis for Robust 3D Human Reconstruction", "abstract": "We propose PoseGaussian, a pose-guided Gaussian Splatting framework for high-fidelity human novel view synthesis. Human body pose serves a dual purpose in our design: as a structural prior, it is fused with a color encoder to refine depth estimation; as a temporal cue, it is processed by a dedicated pose encoder to enhance temporal consistency across frames. These components are integrated into a fully differentiable, end-to-end trainable pipeline. Unlike prior works that use pose only as a condition or for warping, PoseGaussian embeds pose signals into both geometric and temporal stages to improve robustness and generalization. It is specifically designed to address challenges inherent in dynamic human scenes, such as articulated motion and severe self-occlusion. Notably, our framework achieves real-time rendering at 100 FPS, maintaining the efficiency of standard Gaussian Splatting pipelines. We validate our approach on ZJU-MoCap, THuman2.0, and in-house datasets, demonstrating state-of-the-art performance in perceptual quality and structural accuracy (PSNR 30.86, SSIM 0.979, LPIPS 0.028).", "date": "2026-02-05", "categories": []}
{"id": "2602.05572", "title": "ShapeGaussian: High-Fidelity 4D Human Reconstruction in Monocular Videos via Vision Priors", "abstract": "We introduce ShapeGaussian, a high-fidelity, template-free method for 4D human reconstruction from casual monocular videos. Generic reconstruction methods lacking robust vision priors, such as 4DGS, struggle to capture high-deformation human motion without multi-view cues. While template-based approaches, primarily relying on SMPL, such as HUGS, can produce photorealistic results, they are highly susceptible to errors in human pose estimation, often leading to unrealistic artifacts. In contrast, ShapeGaussian effectively integrates template-free vision priors to achieve both high-fidelity and robust scene reconstructions. Our method follows a two-step pipeline: first, we learn a coarse, deformable geometry using pretrained models that estimate data-driven priors, providing a foundation for reconstruction. Then, we refine this geometry using a neural deformation model to capture fine-grained dynamic details. By leveraging 2D vision priors, we mitigate artifacts from erroneous pose estimation in template-based methods and employ multiple reference frames to resolve the invisibility issue of 2D keypoints in a template-free manner. Extensive experiments demonstrate that ShapeGaussian surpasses template-based methods in reconstruction accuracy, achieving superior visual quality and robustness across diverse human motions in casual monocular videos.", "date": "2026-02-05", "categories": []}
{"id": "2602.04257", "title": "Depth-Guided Metric-Aware Temporal Consistency for Monocular Video Human Mesh Recovery", "abstract": "Monocular video human mesh recovery faces fundamental challenges in maintaining metric consistency and temporal stability due to inherent depth ambiguities and scale uncertainties. While existing methods rely primarily on RGB features and temporal smoothing, they struggle with depth ordering, scale drift, and occlusion-induced instabilities. We propose a comprehensive depth-guided framework that achieves metric-aware temporal consistency through three synergistic components: A Depth-Guided Multi-Scale Fusion module that adaptively integrates geometric priors with RGB features via confidence-aware gating; A Depth-guided Metric-Aware Pose and Shape (D-MAPS) estimator that leverages depth-calibrated bone statistics for scale-consistent initialization; A Motion-Depth Aligned Refinement (MoDAR) module that enforces temporal coherence through cross-modal attention between motion dynamics and geometric cues. Our method achieves superior results on three challenging benchmarks, demonstrating significant improvements in robustness against heavy occlusion and spatial accuracy while maintaining computational efficiency.", "date": "2026-02-04", "categories": []}
{"id": "2602.04317", "title": "JOintGS: Joint Optimization of Cameras, Bodies and 3D Gaussians for In-the-Wild Monocular Reconstruction", "abstract": "Reconstructing high-fidelity animatable 3D human avatars from monocular RGB videos remains challenging, particularly in unconstrained in-the-wild scenarios where camera parameters and human poses from off-the-shelf methods (e.g., COLMAP, HMR2.0) are often inaccurate. Splatting (3DGS) advances demonstrate impressive rendering quality and real-time performance, they critically depend on precise camera calibration and pose annotations, limiting their applicability in real-world settings. We present JOintGS, a unified framework that jointly optimizes camera extrinsics, human poses, and 3D Gaussian representations from coarse initialization through a synergistic refinement mechanism. Our key insight is that explicit foreground-background disentanglement enables mutual reinforcement: static background Gaussians anchor camera estimation via multi-view consistency; refined cameras improve human body alignment through accurate temporal correspondence; optimized human poses enhance scene reconstruction by removing dynamic artifacts from static constraints. We further introduce a temporal dynamics module to capture fine-grained pose-dependent deformations and a residual color field to model illumination variations. Extensive experiments on NeuMan and EMDB datasets demonstrate that JOintGS achieves superior reconstruction quality, with 2.1~dB PSNR improvement over state-of-the-art methods on NeuMan dataset, while maintaining real-time rendering. Notably, our method shows significantly enhanced robustness to noisy initialization compared to the baseline.Our source code is available at https://github.com/MiliLab/JOintGS.", "date": "2026-02-04", "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": "2601.21376", "title": "Towards Geometry-Aware and Motion-Guided Video Human Mesh Recovery", "abstract": "Existing video-based 3D Human Mesh Recovery (HMR) methods often produce physically implausible results, stemming from their reliance on flawed intermediate 3D pose anchors and their inability to effectively model complex spatiotemporal dynamics. To overcome these deep-rooted architectural problems, we introduce HMRMamba, a new paradigm for HMR that pioneers the use of Structured State Space Models (SSMs) for their efficiency and long-range modeling prowess. Our framework is distinguished by two core contributions. First, the Geometry-Aware Lifting Module, featuring a novel dual-scan Mamba architecture, creates a robust foundation for reconstruction. It directly grounds the 2D-to-3D pose lifting process with geometric cues from image features, producing a highly reliable 3D pose sequence that serves as a stable anchor. Second, the Motion-guided Reconstruction Network leverages this anchor to explicitly process kinematic patterns over time. By injecting this crucial temporal awareness, it significantly enhances the final mesh's coherence and robustness, particularly under occlusion and motion blur. Comprehensive evaluations on 3DPW, MPI-INF-3DHP, and Human3.6M benchmarks confirm that HMRMamba sets a new state-of-the-art, outperforming existing methods in both reconstruction accuracy and temporal consistency while offering superior computational efficiency.", "date": "2026-01-29", "categories": []}
{"id": "2601.18851", "title": "SelfieAvatar: Real-time Head Avatar reenactment from a Selfie Video", "abstract": "Head avatar reenactment focuses on creating animatable personal avatars from monocular videos, serving as a foundational element for applications like social signal understanding, gaming, human-machine interaction, and computer vision. Recent advances in 3D Morphable Model (3DMM)-based facial reconstruction methods have achieved remarkable high-fidelity face estimation. However, on the one hand, they struggle to capture the entire head, including non-facial regions and background details in real time, which is an essential aspect for producing realistic, high-fidelity head avatars. On the other hand, recent approaches leveraging generative adversarial networks (GANs) for head avatar generation from videos can achieve high-quality reenactments but encounter limitations in reproducing fine-grained head details, such as wrinkles and hair textures. In addition, existing methods generally rely on a large amount of training data, and rarely focus on using only a simple selfie video to achieve avatar reenactment. To address these challenges, this study introduces a method for detailed head avatar reenactment using a selfie video. The approach combines 3DMMs with a StyleGAN-based generator. A detailed reconstruction model is proposed, incorporating mixed loss functions for foreground reconstruction and avatar image generation during adversarial training to recover high-frequency details. Qualitative and quantitative evaluations on self-reenactment and cross-reenactment tasks demonstrate that the proposed method achieves superior head avatar reconstruction with rich and intricate textures compared to existing approaches.", "date": "2026-01-26", "categories": []}
{"id": "2207.06400", "title": "PyMAF-X: Towards Well-aligned Full-body Model Regression from Monocular Images", "abstract": "We present PyMAF-X, a regression-based approach to recovering parametric full-body models from monocular images. This task is very challenging since minor parametric deviation may lead to noticeable misalignment between the estimated mesh and the input image. Moreover, when integrating part-specific estimations into the full-body model, existing solutions tend to either degrade the alignment or produce unnatural wrist poses. To address these issues, we propose a Pyramidal Mesh Alignment Feedback (PyMAF) loop in our regression network for well-aligned human mesh recovery and extend it as PyMAF-X for the recovery of expressive full-body models. The core idea of PyMAF is to leverage a feature pyramid and rectify the predicted parameters explicitly based on the mesh-image alignment status. Specifically, given the currently predicted parameters, mesh-aligned evidence will be extracted from finer-resolution features accordingly and fed back for parameter rectification. To enhance the alignment perception, an auxiliary dense supervision is employed to provide mesh-image correspondence guidance while spatial alignment attention is introduced to enable the awareness of the global contexts for our network. When extending PyMAF for full-body mesh recovery, an adaptive integration strategy is proposed in PyMAF-X to produce natural wrist poses while maintaining the well-aligned performance of the part-specific estimations. The efficacy of our approach is validated on several benchmark datasets for body, hand, face, and full-body mesh recovery, where PyMAF and PyMAF-X effectively improve the mesh-image alignment and achieve new state-of-the-art results. The project page with code and video results can be found at https://zhanghongwen.cn/pymaf-x.", "date": "2022-07-13", "categories": []}
{"id": "2304.03167", "title": "CloSET: Modeling Clothed Humans on Continuous Surface with Explicit Template Decomposition", "abstract": "Creating animatable avatars from static scans requires the modeling of clothing deformations in different poses. Existing learning-based methods typically add pose-dependent deformations upon a minimally-clothed mesh template or a learned implicit template, which have limitations in capturing details or hinder end-to-end learning. In this paper, we revisit point-based solutions and propose to decompose explicit garment-related templates and then add pose-dependent wrinkles to them. In this way, the clothing deformations are disentangled such that the pose-dependent wrinkles can be better learned and applied to unseen poses. Additionally, to tackle the seam artifact issues in recent state-of-the-art point-based methods, we propose to learn point features on a body surface, which establishes a continuous and compact feature space to capture the fine-grained and pose-dependent clothing geometry. To facilitate the research in this field, we also introduce a high-quality scan dataset of humans in real-world clothing. Our approach is validated on two existing datasets and our newly introduced dataset, showing better clothing deformation results in unseen poses. The project page with code and dataset can be found at https://zhanghongwen.cn/closet.", "date": "2023-04-06", "categories": []}
{"id": "2411.18624", "title": "GeneMAN: Generalizable Single-Image 3D Human Reconstruction from Multi-Source Human Data", "abstract": "Given a single in-the-wild human photo, it remains a challenging task to reconstruct a high-fidelity 3D human model. Existing methods face difficulties including a) the varying body proportions captured by in-the-wild human images; b) diverse personal belongings within the shot; and c) ambiguities in human postures and inconsistency in human textures. In addition, the scarcity of high-quality human data intensifies the challenge. To address these problems, we propose a Generalizable image-to-3D huMAN reconstruction framework, dubbed GeneMAN, building upon a comprehensive multi-source collection of high-quality human data, including 3D scans, multi-view videos, single photos, and our generated synthetic human data. GeneMAN encompasses three key modules. 1) Without relying on parametric human models (e.g., SMPL), GeneMAN first trains a human-specific text-to-image diffusion model and a view-conditioned diffusion model, serving as GeneMAN 2D human prior and 3D human prior for reconstruction, respectively. 2) With the help of the pretrained human prior models, the Geometry Initialization-&-Sculpting pipeline is leveraged to recover high-quality 3D human geometry given a single image. 3) To achieve high-fidelity 3D human textures, GeneMAN employs the Multi-Space Texture Refinement pipeline, consecutively refining textures in the latent and the pixel spaces. Extensive experimental results demonstrate that GeneMAN could generate high-quality 3D human models from a single image input, outperforming prior state-of-the-art methods. Notably, GeneMAN could reveal much better generalizability in dealing with in-the-wild images, often yielding high-quality 3D human models in natural poses with common items, regardless of the body proportions in the input images.", "date": "2024-11-27", "categories": []}
{"id": "2601.18633", "title": "Splat-Portrait: Generalizing Talking Heads with Gaussian Splatting", "abstract": "Talking Head Generation aims at synthesizing natural-looking talking videos from speech and a single portrait image. Previous 3D talking head generation methods have relied on domain-specific heuristics such as warping-based facial motion representation priors to animate talking motions, yet still produce inaccurate 3D avatar reconstructions, thus undermining the realism of generated animations. We introduce Splat-Portrait, a Gaussian-splatting-based method that addresses the challenges of 3D head reconstruction and lip motion synthesis. Our approach automatically learns to disentangle a single portrait image into a static 3D reconstruction represented as static Gaussian Splatting, and a predicted whole-image 2D background. It then generates natural lip motion conditioned on input audio, without any motion driven priors. Training is driven purely by 2D reconstruction and score-distillation losses, without 3D supervision nor landmarks. Experimental results demonstrate that Splat-Portrait exhibits superior performance on talking head generation and novel view synthesis, achieving better visual quality compared to previous works. Our project code and supplementary documents are public available at https://github.com/stonewalking/Splat-portrait.", "date": "2026-01-26", "categories": []}
{"id": "2505.10072", "title": "ToonifyGB: StyleGAN-based Gaussian Blendshapes for 3D Stylized Head Avatars", "abstract": "The introduction of 3D Gaussian blendshapes has enabled the real-time reconstruction of animatable head avatars from monocular video. Toonify, a StyleGAN-based method, has become widely used for facial image stylization. To extend Toonify for synthesizing diverse stylized 3D head avatars using Gaussian blendshapes, we propose an efficient two-stage framework, ToonifyGB. In Stage 1 (stylized video generation), we adopt an improved StyleGAN to generate the stylized video from the input video frames, which overcomes the limitation of cropping aligned faces at a fixed resolution as preprocessing for normal StyleGAN. This process provides a more stable stylized video, which enables Gaussian blendshapes to better capture the high-frequency details of the video frames, facilitating the synthesis of high-quality animations in the next stage. In Stage 2 (Gaussian blendshapes synthesis), our method learns a stylized neutral head model and a set of expression blendshapes from the generated stylized video. By combining the neutral head model with expression blendshapes, ToonifyGB can efficiently render stylized avatars with arbitrary expressions. We validate the effectiveness of ToonifyGB on benchmark datasets using two representative styles: Arcane and Pixar.", "date": "2025-05-15", "categories": []}
{"id": "2601.14844", "title": "CAG-Avatar: Cross-Attention Guided Gaussian Avatars for High-Fidelity Head Reconstruction", "abstract": "Creating high-fidelity, real-time drivable 3D head avatars is a core challenge in digital animation. While 3D Gaussian Splashing (3D-GS) offers unprecedented rendering speed and quality, current animation techniques often rely on a \"one-size-fits-all\" global tuning approach, where all Gaussian primitives are uniformly driven by a single expression code. This simplistic approach fails to unravel the distinct dynamics of different facial regions, such as deformable skin versus rigid teeth, leading to significant blurring and distortion artifacts. We introduce Conditionally-Adaptive Gaussian Avatars (CAG-Avatar), a framework that resolves this key limitation. At its core is a Conditionally Adaptive Fusion Module built on cross-attention. This mechanism empowers each 3D Gaussian to act as a query, adaptively extracting relevant driving signals from the global expression code based on its canonical position. This \"tailor-made\" conditioning strategy drastically enhances the modeling of fine-grained, localized dynamics. Our experiments confirm a significant improvement in reconstruction fidelity, particularly for challenging regions such as teeth, while preserving real-time rendering performance.", "date": "2026-01-21", "categories": []}
{"id": "2601.12736", "title": "KaoLRM: Repurposing Pre-trained Large Reconstruction Models for Parametric 3D Face Reconstruction", "abstract": "We propose KaoLRM to re-target the learned prior of the Large Reconstruction Model (LRM) for parametric 3D face reconstruction from single-view images. Parametric 3D Morphable Models (3DMMs) have been widely used for facial reconstruction due to their compact and interpretable parameterization, yet existing 3DMM regressors often exhibit poor consistency across varying viewpoints. To address this, we harness the pre-trained 3D prior of LRM and incorporate FLAME-based 2D Gaussian Splatting into LRM's rendering pipeline. Specifically, KaoLRM projects LRM's pre-trained triplane features into the FLAME parameter space to recover geometry, and models appearance via 2D Gaussian primitives that are tightly coupled to the FLAME mesh. The rich prior enables the FLAME regressor to be aware of the 3D structure, leading to accurate and robust reconstructions under self-occlusions and diverse viewpoints. Experiments on both controlled and in-the-wild benchmarks demonstrate that KaoLRM achieves superior reconstruction accuracy and cross-view consistency, while existing methods remain sensitive to viewpoint variations. The code is released at https://github.com/CyberAgentAILab/KaoLRM.", "date": "2026-01-19", "categories": []}
{"id": "2601.12770", "title": "Generalizable and Animatable 3D Full-Head Gaussian Avatar from a Single Image", "abstract": "Building 3D animatable head avatars from a single image is an important yet challenging problem. Existing methods generally collapse under large camera pose variations, compromising the realism of 3D avatars. In this work, we propose a new framework to tackle the novel setting of one-shot 3D full-head animatable avatar reconstruction in a single feed-forward pass, enabling real-time animation and simultaneous 360$^\\circ$ rendering views. To facilitate efficient animation control, we model 3D head avatars with Gaussian primitives embedded on the surface of a parametric face model within the UV space. To obtain knowledge of full-head geometry and textures, we leverage rich 3D full-head priors within a pretrained 3D generative adversarial network (GAN) for global full-head feature extraction and multi-view supervision. To increase the fidelity of the 3D reconstruction of the input image, we take advantage of the symmetric nature of the UV space and human faces to fuse local fine-grained input image features with the global full-head textures. Extensive experiments demonstrate the effectiveness of our method, achieving high-quality 3D full-head modeling as well as real-time animation, thereby improving the realism of 3D talking avatars.", "date": "2026-01-19", "categories": []}
{"id": "2601.13148", "title": "ICo3D: An Interactive Conversational 3D Virtual Human", "abstract": "This work presents Interactive Conversational 3D Virtual Human (ICo3D), a method for generating an interactive, conversational, and photorealistic 3D human avatar. Based on multi-view captures of a subject, we create an animatable 3D face model and a dynamic 3D body model, both rendered by splatting Gaussian primitives. Once merged together, they represent a lifelike virtual human avatar suitable for real-time user interactions. We equip our avatar with an LLM for conversational ability. During conversation, the audio speech of the avatar is used as a driving signal to animate the face model, enabling precise synchronization. We describe improvements to our dynamic Gaussian models that enhance photorealism: SWinGS++ for body reconstruction and HeadGaS++ for face reconstruction, and provide as well a solution to merge the separate face and body models without artifacts. We also present a demo of the complete system, showcasing several use cases of real-time conversation with the 3D avatar. Our approach offers a fully integrated virtual avatar experience, supporting both oral and written form interactions in immersive environments. ICo3D is applicable to a wide range of fields, including gaming, virtual assistance, and personalized education, among others. Project page: https://ico3d.github.io/", "date": "2026-01-19", "categories": []}
{"id": "2412.05961", "title": "FOF-X: Towards Real-time Detailed Human Reconstruction from a Single Image", "abstract": "We introduce FOF-X for real-time reconstruction of detailed human geometry from a single image. Balancing real-time speed against high-quality results is a persistent challenge, mainly due to the high computational demands of existing 3D representations. To address this, we propose Fourier Occupancy Field (FOF), an efficient 3D representation by learning the Fourier series. The core of FOF is to factorize a 3D occupancy field into a 2D vector field, retaining topology and spatial relationships within the 3D domain while facilitating compatibility with 2D convolutional neural networks. Such a representation bridges the gap between 3D and 2D domains, enabling the integration of human parametric models as priors and enhancing the reconstruction robustness. Based on FOF, we design a new reconstruction framework, FOF-X, to avoid the performance degradation caused by texture and lighting. This enables our real-time reconstruction system to better handle the domain gap between training images and real images. Additionally, in FOF-X, we enhance the inter-conversion algorithms between FOF and mesh representations with a Laplacian constraint and an automaton-based discontinuity matcher, improving both quality and robustness. We validate the strengths of our approach on different datasets and real-captured data, where FOF-X achieves new state-of-the-art results. The code has already been released for research purposes at https://cic.tju.edu.cn/faculty/likun/projects/FOFX/index.html.", "date": "2024-12-08", "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": "2601.10200", "title": "ELITE: Efficient Gaussian Head Avatar from a Monocular Video via Learned Initialization and TEst-time Generative Adaptation", "abstract": "We introduce ELITE, an Efficient Gaussian head avatar synthesis from a monocular video via Learned Initialization and TEst-time generative adaptation. Prior works rely either on a 3D data prior or a 2D generative prior to compensate for missing visual cues in monocular videos. However, 3D data prior methods often struggle to generalize in-the-wild, while 2D generative prior methods are computationally heavy and prone to identity hallucination. We identify a complementary synergy between these two priors and design an efficient system that achieves high-fidelity animatable avatar synthesis with strong in-the-wild generalization. Specifically, we introduce a feed-forward Mesh2Gaussian Prior Model (MGPM) that enables fast initialization of a Gaussian avatar. To further bridge the domain gap at test time, we design a test-time generative adaptation stage, leveraging both real and synthetic images as supervision. Unlike previous full diffusion denoising strategies that are slow and hallucination-prone, we propose a rendering-guided single-step diffusion enhancer that restores missing visual details, grounded on Gaussian avatar renderings. Our experiments demonstrate that ELITE produces visually superior avatars to prior works, even for challenging expressions, while achieving 60x faster synthesis than the 2D generative prior method.", "date": "2026-01-15", "categories": []}
{"id": "2601.07518", "title": "Mon3tr: Monocular 3D Telepresence with Pre-built Gaussian Avatars as Amortization", "abstract": "Immersive telepresence aims to transform human interaction in AR/VR applications by enabling lifelike full-body holographic representations for enhanced remote collaboration. However, existing systems rely on hardware-intensive multi-camera setups and demand high bandwidth for volumetric streaming, limiting their real-time performance on mobile devices. To overcome these challenges, we propose Mon3tr, a novel Monocular 3D telepresence framework that integrates 3D Gaussian splatting (3DGS) based parametric human modeling into telepresence for the first time. Mon3tr adopts an amortized computation strategy, dividing the process into a one-time offline multi-view reconstruction phase to build a user-specific avatar and a monocular online inference phase during live telepresence sessions. A single monocular RGB camera is used to capture body motions and facial expressions in real time to drive the 3DGS-based parametric human model, significantly reducing system complexity and cost. The extracted motion and appearance features are transmitted at < 0.2 Mbps over WebRTC's data channel, allowing robust adaptation to network fluctuations. On the receiver side, e.g., Meta Quest 3, we develop a lightweight 3DGS attribute deformation network to dynamically generate corrective 3DGS attribute adjustments on the pre-built avatar, synthesizing photorealistic motion and appearance at ~ 60 FPS. Extensive experiments demonstrate the state-of-the-art performance of our method, achieving a PSNR of > 28 dB for novel poses, an end-to-end latency of ~ 80 ms, and > 1000x bandwidth reduction compared to point-cloud streaming, while supporting real-time operation from monocular inputs across diverse scenarios. Our demos can be found at https://mon3tr3d.github.io.", "date": "2026-01-12", "categories": []}
{"id": "2601.05853", "title": "LayerGS: Decomposition and Inpainting of Layered 3D Human Avatars via 2D Gaussian Splatting", "abstract": "We propose a novel framework for decomposing arbitrarily posed humans into animatable multi-layered 3D human avatars, separating the body and garments. Conventional single-layer reconstruction methods lock clothing to one identity, while prior multi-layer approaches struggle with occluded regions. We overcome both limitations by encoding each layer as a set of 2D Gaussians for accurate geometry and photorealistic rendering, and inpainting hidden regions with a pretrained 2D diffusion model via score-distillation sampling (SDS). Our three-stage training strategy first reconstructs the coarse canonical garment via single-layer reconstruction, followed by multi-layer training to jointly recover the inner-layer body and outer-layer garment details. Experiments on two 3D human benchmark datasets (4D-Dress, Thuman2.0) show that our approach achieves better rendering quality and layer decomposition and recomposition than the previous state-of-the-art, enabling realistic virtual try-on under novel viewpoints and poses, and advancing practical creation of high-fidelity 3D human assets for immersive applications. Our code is available at https://github.com/RockyXu66/LayerGS", "date": "2026-01-09", "categories": []}
{"id": "2502.01045", "title": "WonderHuman: Hallucinating Unseen Parts in Dynamic 3D Human Reconstruction", "abstract": "In this paper, we present WonderHuman to reconstruct dynamic human avatars from a monocular video for high-fidelity novel view synthesis. Previous dynamic human avatar reconstruction methods typically require the input video to have full coverage of the observed human body. However, in daily practice, one typically has access to limited viewpoints, such as monocular front-view videos, making it a cumbersome task for previous methods to reconstruct the unseen parts of the human avatar. To tackle the issue, we present WonderHuman, which leverages 2D generative diffusion model priors to achieve high-quality, photorealistic reconstructions of dynamic human avatars from monocular videos, including accurate rendering of unseen body parts. Our approach introduces a Dual-Space Optimization technique, applying Score Distillation Sampling (SDS) in both canonical and observation spaces to ensure visual consistency and enhance realism in dynamic human reconstruction. Additionally, we present a View Selection strategy and Pose Feature Injection to enforce the consistency between SDS predictions and observed data, ensuring pose-dependent effects and higher fidelity in the reconstructed avatar. In the experiments, our method achieves SOTA performance in producing photorealistic renderings from the given monocular video, particularly for those challenging unseen parts. The project page and source code can be found at https://wyiguanw.github.io/WonderHuman/.", "date": "2025-02-03", "categories": []}
{"id": "2601.03319", "title": "CaricatureGS: Exaggerating 3D Gaussian Splatting Faces With Gaussian Curvature", "abstract": "A photorealistic and controllable 3D caricaturization framework for faces is introduced. We start with an intrinsic Gaussian curvature-based surface exaggeration technique, which, when coupled with texture, tends to produce over-smoothed renders. To address this, we resort to 3D Gaussian Splatting (3DGS), which has recently been shown to produce realistic free-viewpoint avatars. Given a multiview sequence, we extract a FLAME mesh, solve a curvature-weighted Poisson equation, and obtain its exaggerated form. However, directly deforming the Gaussians yields poor results, necessitating the synthesis of pseudo-ground-truth caricature images by warping each frame to its exaggerated 2D representation using local affine transformations. We then devise a training scheme that alternates real and synthesized supervision, enabling a single Gaussian collection to represent both natural and exaggerated avatars. This scheme improves fidelity, supports local edits, and allows continuous control over the intensity of the caricature. In order to achieve real-time deformations, an efficient interpolation between the original and exaggerated surfaces is introduced. We further analyze and show that it has a bounded deviation from closed-form solutions. In both quantitative and qualitative evaluations, our results outperform prior work, delivering photorealistic, geometry-controlled caricature avatars.", "date": "2026-01-06", "categories": []}
{"id": "2511.09827", "title": "AHA! Animating Human Avatars in Diverse Scenes with Gaussian Splatting", "abstract": "We present a novel framework for animating humans in 3D scenes using 3D Gaussian Splatting (3DGS), a neural scene representation that has recently achieved state-of-the-art photorealistic results for novel-view synthesis but remains under-explored for human-scene animation and interaction. Unlike existing animation pipelines that use meshes or point clouds as the underlying 3D representation, our approach introduces the use of 3DGS as the 3D representation for animating humans in scenes. By representing humans and scenes as Gaussians, our approach allows geometry-consistent free-viewpoint rendering of humans interacting with 3D scenes. Our key insight is that rendering can be decoupled from motion synthesis, and each sub-problem can be addressed independently without the need for paired human-scene data. Central to our method is a Gaussian-aligned motion module that synthesizes motion without explicit scene geometry, using opacity-based cues and projected Gaussian structures to guide human placement and pose alignment. To ensure natural interactions, we further propose a human-scene Gaussian refinement optimization that enforces realistic contact and navigation. We evaluate our approach on scenes from Scannet++ and the SuperSplat library, and on avatars reconstructed from sparse and dense multi-view human capture. Finally, we demonstrate that our framework enables novel applications such as geometry-consistent free-viewpoint rendering of edited monocular RGB videos with newly animated humans, showcasing the unique advantages of 3DGS for monocular video-based human animation. To assess the full quality of our results, we encourage readers to view the supplementary material available at https://miraymen.github.io/aha/ .", "date": "2025-11-13", "categories": []}
{"id": "2601.01222", "title": "UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass", "abstract": "We present UniSH, a unified, feed-forward framework for joint metric-scale 3D scene and human reconstruction. A key challenge in this domain is the scarcity of large-scale, annotated real-world data, forcing a reliance on synthetic datasets. This reliance introduces a significant sim-to-real domain gap, leading to poor generalization, low-fidelity human geometry, and poor alignment on in-the-wild videos. To address this, we propose an innovative training paradigm that effectively leverages unlabeled in-the-wild data. Our framework bridges strong, disparate priors from scene reconstruction and HMR, and is trained with two core components: (1) a robust distillation strategy to refine human surface details by distilling high-frequency details from an expert depth model, and (2) a two-stage supervision scheme, which first learns coarse localization on synthetic data, then fine-tunes on real data by directly optimizing the geometric correspondence between the SMPL mesh and the human point cloud. This approach enables our feed-forward model to jointly recover high-fidelity scene geometry, human point clouds, camera parameters, and coherent, metric-scale SMPL bodies, all in a single forward pass. Extensive experiments demonstrate that our model achieves state-of-the-art performance on human-centric scene reconstruction and delivers highly competitive results on global human motion estimation, comparing favorably against both optimization-based frameworks and HMR-only methods. Project page: https://murphylmf.github.io/UniSH/", "date": "2026-01-03", "categories": []}
{"id": "2601.02098", "title": "InpaintHuman: Reconstructing Occluded Humans with Multi-Scale UV Mapping and Identity-Preserving Diffusion Inpainting", "abstract": "Reconstructing complete and animatable 3D human avatars from monocular videos remains challenging, particularly under severe occlusions. While 3D Gaussian Splatting has enabled photorealistic human rendering, existing methods struggle with incomplete observations, often producing corrupted geometry and temporal inconsistencies. We present InpaintHuman, a novel method for generating high-fidelity, complete, and animatable avatars from occluded monocular videos. Our approach introduces two key innovations: (i) a multi-scale UV-parameterized representation with hierarchical coarse-to-fine feature interpolation, enabling robust reconstruction of occluded regions while preserving geometric details; and (ii) an identity-preserving diffusion inpainting module that integrates textual inversion with semantic-conditioned guidance for subject-specific, temporally coherent completion. Unlike SDS-based methods, our approach employs direct pixel-level supervision to ensure identity fidelity. Experiments on synthetic benchmarks (PeopleSnapshot, ZJU-MoCap) and real-world scenarios (OcMotion) demonstrate competitive performance with consistent improvements in reconstruction quality across diverse poses and viewpoints.", "date": "2026-01-05", "categories": []}
{"id": "2601.00328", "title": "Joint Geometry-Appearance Human Reconstruction in a Unified Latent Space via Bridge Diffusion", "abstract": "Achieving consistent and high-fidelity geometry and appearance reconstruction of 3D digital humans from a single RGB image is inherently a challenging task. Existing studies typically resort to decoupled pipelines for geometry estimation and appearance synthesis, often hindering unified reconstruction and causing inconsistencies. This paper introduces \\textbf{JGA-LBD}, a novel framework that unifies the modeling of geometry and appearance into a joint latent representation and formulates the generation process as bridge diffusion. Observing that directly integrating heterogeneous input conditions (e.g., depth maps, SMPL models) leads to substantial training difficulties, we unify all conditions into the 3D Gaussian representations, which can be further compressed into a unified latent space through a shared sparse variational autoencoder (VAE). Subsequently, the specialized form of bridge diffusion enables to start with a partial observation of the target latent code and solely focuses on inferring the missing components. Finally, a dedicated decoding module extracts the complete 3D human geometric structure and renders novel views from the inferred latent representation. Experiments demonstrate that JGA-LBD outperforms current state-of-the-art approaches in terms of both geometry fidelity and appearance quality, including challenging in-the-wild scenarios. Our code will be made publicly available at https://github.com/haiantyz/JGA-LBD.", "date": "2026-01-01", "categories": []}
{"id": "2512.21054", "title": "DexAvatar: 3D Sign Language Reconstruction with Hand and Body Pose Priors", "abstract": "The trend in sign language generation is centered around data-driven generative methods that require vast amounts of precise 2D and 3D human pose data to achieve an acceptable generation quality. However, currently, most sign language datasets are video-based and limited to automatically reconstructed 2D human poses (i.e., keypoints) and lack accurate 3D information. Furthermore, existing state-of-the-art for automatic 3D human pose estimation from sign language videos is prone to self-occlusion, noise, and motion blur effects, resulting in poor reconstruction quality. In response to this, we introduce DexAvatar, a novel framework to reconstruct bio-mechanically accurate fine-grained hand articulations and body movements from in-the-wild monocular sign language videos, guided by learned 3D hand and body priors. DexAvatar achieves strong performance in the SGNify motion capture dataset, the only benchmark available for this task, reaching an improvement of 35.11% in the estimation of body and hand poses compared to the state-of-the-art. The official website of this work is: https://github.com/kaustesseract/DexAvatar.", "date": "2025-12-24", "categories": []}
{"id": "2512.21099", "title": "TexAvatars : Hybrid Texel-3D Representations for Stable Rigging of Photorealistic Gaussian Head Avatars", "abstract": "Constructing drivable and photorealistic 3D head avatars has become a central task in AR/XR, enabling immersive and expressive user experiences. With the emergence of high-fidelity and efficient representations such as 3D Gaussians, recent works have pushed toward ultra-detailed head avatars. Existing approaches typically fall into two categories: rule-based analytic rigging or neural network-based deformation fields. While effective in constrained settings, both approaches often fail to generalize to unseen expressions and poses, particularly in extreme reenactment scenarios. Other methods constrain Gaussians to the global texel space of 3DMMs to reduce rendering complexity. However, these texel-based avatars tend to underutilize the underlying mesh structure. They apply minimal analytic deformation and rely heavily on neural regressors and heuristic regularization in UV space, which weakens geometric consistency and limits extrapolation to complex, out-of-distribution deformations. To address these limitations, we introduce TexAvatars, a hybrid avatar representation that combines the explicit geometric grounding of analytic rigging with the spatial continuity of texel space. Our approach predicts local geometric attributes in UV space via CNNs, but drives 3D deformation through mesh-aware Jacobians, enabling smooth and semantically meaningful transitions across triangle boundaries. This hybrid design separates semantic modeling from geometric control, resulting in improved generalization, interpretability, and stability. Furthermore, TexAvatars captures fine-grained expression effects, including muscle-induced wrinkles, glabellar lines, and realistic mouth cavity geometry, with high fidelity. Our method achieves state-of-the-art performance under extreme pose and expression variations, demonstrating strong generalization in challenging head reenactment settings.", "date": "2025-12-24", "categories": []}
{"id": "2503.14736", "title": "HandSCS: Structural Coordinate Space for Animatable Hand Gaussian Splatting", "abstract": "Creating animatable hand avatars from multi-view images requires modeling complex articulations and maintaining structural consistency across poses in real time. We present HandSCS, a structure-guided 3D Gaussian Splatting framework for high-fidelity hand animation. Unlike existing approaches that condition all Gaussians on the same global pose parameters, which are inadequate for highly articulated hands, HandSCS equips each Gaussian with explicit structural guidance from both intra-pose and inter-pose perspectives. To establish intra-pose structural guidance, we introduce a Structural Coordinate Space (SCS), which bridges the gap between sparse bones and dense Gaussians through hybrid static-dynamic coordinate basis and angular-radial descriptors. To improve cross-pose coherence, we further introduce an Inter-pose Consistency Loss that promotes consistent Gaussian attributes under similar articulations. Together, these components achieve high-fidelity results with consistent fine details, even in challenging high-deformation and self-contact regions. Experiments on the InterHand2.6M dataset demonstrate that HandSCS achieves state-of-the-art performance in hand avatar animation, confirming the effectiveness of explicit structural modeling.", "date": "2025-03-18", "categories": []}
{"id": "2511.03589", "title": "Human Mesh Modeling for Anny Body", "abstract": "Parametric body models provide the structural basis for many human-centric tasks, yet existing models often rely on costly 3D scans and learned shape spaces that are proprietary and demographically narrow. We introduce Anny, a simple, fully differentiable, and scan-free human body model grounded in anthropometric knowledge from the MakeHuman community. Anny defines a continuous, interpretable shape space, where phenotype parameters (e.g. gender, age, height, weight) control blendshapes spanning a wide range of human forms--across ages (from infants to elders), body types, and proportions. Calibrated using WHO population statistics, it provides realistic and demographically grounded human shape variation within a single unified model. Thanks to its openness and semantic control, Anny serves as a versatile foundation for 3D human modeling--supporting millimeter-accurate scan fitting, controlled synthetic data generation, and Human Mesh Recovery (HMR). We further introduce Anny-One, a collection of 800k photorealistic images generated with Anny, showing that despite its simplicity, HMR models trained with Anny can match the performance of those trained with scan-based body models. The Anny body model and its code are released under the Apache 2.0 license, making Anny an accessible foundation for human-centric 3D modeling.", "date": "2025-11-05", "categories": []}
{"id": "2512.17545", "title": "ClothHMR: 3D Mesh Recovery of Humans in Diverse Clothing from Single Image", "abstract": "With 3D data rapidly emerging as an important form of multimedia information, 3D human mesh recovery technology has also advanced accordingly. However, current methods mainly focus on handling humans wearing tight clothing and perform poorly when estimating body shapes and poses under diverse clothing, especially loose garments. To this end, we make two key insights: (1) tailoring clothing to fit the human body can mitigate the adverse impact of clothing on 3D human mesh recovery, and (2) utilizing human visual information from large foundational models can enhance the generalization ability of the estimation. Based on these insights, we propose ClothHMR, to accurately recover 3D meshes of humans in diverse clothing. ClothHMR primarily consists of two modules: clothing tailoring (CT) and FHVM-based mesh recovering (MR). The CT module employs body semantic estimation and body edge prediction to tailor the clothing, ensuring it fits the body silhouette. The MR module optimizes the initial parameters of the 3D human mesh by continuously aligning the intermediate representations of the 3D mesh with those inferred from the foundational human visual model (FHVM). ClothHMR can accurately recover 3D meshes of humans wearing diverse clothing, precisely estimating their body shapes and poses. Experimental results demonstrate that ClothHMR significantly outperforms existing state-of-the-art methods across benchmark datasets and in-the-wild images. Additionally, a web application for online fashion and shopping powered by ClothHMR is developed, illustrating that ClothHMR can effectively serve real-world usage scenarios. The code and model for ClothHMR are available at: \\url{https://github.com/starVisionTeam/ClothHMR}.", "date": "2025-12-19", "categories": []}
{"id": "2512.17717", "title": "FlexAvatar: Flexible Large Reconstruction Model for Animatable Gaussian Head Avatars with Detailed Deformation", "abstract": "We present FlexAvatar, a flexible large reconstruction model for high-fidelity 3D head avatars with detailed dynamic deformation from single or sparse images, without requiring camera poses or expression labels. It leverages a transformer-based reconstruction model with structured head query tokens as canonical anchor to aggregate flexible input-number-agnostic, camera-pose-free and expression-free inputs into a robust canonical 3D representation. For detailed dynamic deformation, we introduce a lightweight UNet decoder conditioned on UV-space position maps, which can produce detailed expression-dependent deformations in real time. To better capture rare but critical expressions like wrinkles and bared teeth, we also adopt a data distribution adjustment strategy during training to balance the distribution of these expressions in the training set. Moreover, a lightweight 10-second refinement can further enhances identity-specific details in extreme identities without affecting deformation quality. Extensive experiments demonstrate that our FlexAvatar achieves superior 3D consistency, detailed dynamic realism compared with previous methods, providing a practical solution for animatable 3D avatar creation.", "date": "2025-12-19", "categories": []}
{"id": "2512.17773", "title": "Pix2NPHM: Learning to Regress NPHM Reconstructions From a Single Image", "abstract": "Neural Parametric Head Models (NPHMs) are a recent advancement over mesh-based 3d morphable models (3DMMs) to facilitate high-fidelity geometric detail. However, fitting NPHMs to visual inputs is notoriously challenging due to the expressive nature of their underlying latent space. To this end, we propose Pix2NPHM, a vision transformer (ViT) network that directly regresses NPHM parameters, given a single image as input. Compared to existing approaches, the neural parametric space allows our method to reconstruct more recognizable facial geometry and accurate facial expressions. For broad generalization, we exploit domain-specific ViTs as backbones, which are pretrained on geometric prediction tasks. We train Pix2NPHM on a mixture of 3D data, including a total of over 100K NPHM registrations that enable direct supervision in SDF space, and large-scale 2D video datasets, for which normal estimates serve as pseudo ground truth geometry. Pix2NPHM not only allows for 3D reconstructions at interactive frame rates, it is also possible to improve geometric fidelity by a subsequent inference-time optimization against estimated surface normals and canonical point maps. As a result, we achieve unprecedented face reconstruction quality that can run at scale on in-the-wild data.", "date": "2025-12-19", "categories": []}
{"id": "2512.16397", "title": "Using Gaussian Splats to Create High-Fidelity Facial Geometry and Texture", "abstract": "We leverage increasingly popular three-dimensional neural representations in order to construct a unified and consistent explanation of a collection of uncalibrated images of the human face. Our approach utilizes Gaussian Splatting, since it is more explicit and thus more amenable to constraints than NeRFs. We leverage segmentation annotations to align the semantic regions of the face, facilitating the reconstruction of a neutral pose from only 11 images (as opposed to requiring a long video). We soft constrain the Gaussians to an underlying triangulated surface in order to provide a more structured Gaussian Splat reconstruction, which in turn informs subsequent perturbations to increase the accuracy of the underlying triangulated surface. The resulting triangulated surface can then be used in a standard graphics pipeline. In addition, and perhaps most impactful, we show how accurate geometry enables the Gaussian Splats to be transformed into texture space where they can be treated as a view-dependent neural texture. This allows one to use high visual fidelity Gaussian Splatting on any asset in a scene without the need to modify any other asset or any other aspect (geometry, lighting, renderer, etc.) of the graphics pipeline. We utilize a relightable Gaussian model to disentangle texture from lighting in order to obtain a delit high-resolution albedo texture that is also readily usable in a standard graphics pipeline. The flexibility of our system allows for training with disparate images, even with incompatible lighting, facilitating robust regularization. Finally, we demonstrate the efficacy of our approach by illustrating its use in a text-driven asset creation pipeline.", "date": "2025-12-18", "categories": []}
{"id": "2512.15711", "title": "Gaussian Pixel Codec Avatars: A Hybrid Representation for Efficient Rendering", "abstract": "We present Gaussian Pixel Codec Avatars (GPiCA), photorealistic head avatars that can be generated from multi-view images and efficiently rendered on mobile devices. GPiCA utilizes a unique hybrid representation that combines a triangle mesh and anisotropic 3D Gaussians. This combination maximizes memory and rendering efficiency while maintaining a photorealistic appearance. The triangle mesh is highly efficient in representing surface areas like facial skin, while the 3D Gaussians effectively handle non-surface areas such as hair and beard. To this end, we develop a unified differentiable rendering pipeline that treats the mesh as a semi-transparent layer within the volumetric rendering paradigm of 3D Gaussian Splatting. We train neural networks to decode a facial expression code into three components: a 3D face mesh, an RGBA texture, and a set of 3D Gaussians. These components are rendered simultaneously in a unified rendering engine. The networks are trained using multi-view image supervision. Our results demonstrate that GPiCA achieves the realism of purely Gaussian-based avatars while matching the rendering performance of mesh-based avatars.", "date": "2025-12-17", "categories": []}
{"id": "2503.09342", "title": "GASPACHO: Gaussian Splatting for Controllable Humans and Objects", "abstract": "We present GASPACHO, a method for generating photorealistic, controllable renderings of human-object interactions from multi-view RGB video. Unlike prior work that reconstructs only the human and treats objects as background, GASPACHO simultaneously recovers animatable templates for both the human and the interacting object as distinct sets of Gaussians, thereby allowing for controllable renderings of novel human object interactions in different poses from novel-camera viewpoints. We introduce a novel formulation that learns object Gaussians on an underlying 2D surface manifold rather than in 3D volume, yielding sharper, fine-grained object details for dynamic object reconstruction. We further propose a contact constraint in Gaussian space that regularizes human-object relations and enables natural, physically plausible animation. Across three benchmarks - BEHAVE, NeuralDome, and DNA-Rendering - GASPACHO achieves high-quality reconstructions under heavy occlusion and supports controllable synthesis of novel human-object interactions. We also demonstrate that our method allows for composition of humans and objects in 3D scenes and for the first time showcase that neural rendering can be used for the controllable generation of photoreal humans interacting with dynamic objects in diverse scenes. Our results are available at: https://miraymen.github.io/gaspacho/", "date": "2025-03-12", "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": "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": "2512.09335", "title": "Relightable and Dynamic Gaussian Avatar Reconstruction from Monocular Video", "abstract": "Modeling relightable and animatable human avatars from monocular video is a long-standing and challenging task. Recently, Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) methods have been employed to reconstruct the avatars. However, they often produce unsatisfactory photo-realistic results because of insufficient geometrical details related to body motion, such as clothing wrinkles. In this paper, we propose a 3DGS-based human avatar modeling framework, termed as Relightable and Dynamic Gaussian Avatar (RnD-Avatar), that presents accurate pose-variant deformation for high-fidelity geometrical details. To achieve this, we introduce dynamic skinning weights that define the human avatar's articulation based on pose while also learning additional deformations induced by body motion. We also introduce a novel regularization to capture fine geometric details under sparse visual cues. Furthermore, we present a new multi-view dataset with varied lighting conditions to evaluate relight. Our framework enables realistic rendering of novel poses and views while supporting photo-realistic lighting effects under arbitrary lighting conditions. Our method achieves state-of-the-art performance in novel view synthesis, novel pose rendering, and relighting.", "date": "2025-12-10", "categories": []}
{"id": "2512.06368", "title": "HuPrior3R: Incorporating Human Priors for Better 3D Dynamic Reconstruction from Monocular Videos", "abstract": "Monocular dynamic video reconstruction faces significant challenges in dynamic human scenes due to geometric inconsistencies and resolution degradation issues. Existing methods lack 3D human structural understanding, producing geometrically inconsistent results with distorted limb proportions and unnatural human-object fusion, while memory-constrained downsampling causes human boundary drift toward background geometry. To address these limitations, we propose to incorporate hybrid geometric priors that combine SMPL human body models with monocular depth estimation. Our approach leverages structured human priors to maintain surface consistency while capturing fine-grained geometric details in human regions. We introduce HuPrior3R, featuring a hierarchical pipeline with refinement components that processes full-resolution images for overall scene geometry, then applies strategic cropping and cross-attention fusion for human-specific detail enhancement. The method integrates SMPL priors through a Feature Fusion Module to ensure geometrically plausible reconstruction while preserving fine-grained human boundaries. Extensive experiments on TUM Dynamics and GTA-IM datasets demonstrate superior performance in dynamic human reconstruction.", "date": "2025-12-06", "categories": []}
{"id": "2512.07720", "title": "ViSA: 3D-Aware Video Shading for Real-Time Upper-Body Avatar Creation", "abstract": "Generating high-fidelity upper-body 3D avatars from one-shot input image remains a significant challenge. Current 3D avatar generation methods, which rely on large reconstruction models, are fast and capable of producing stable body structures, but they often suffer from artifacts such as blurry textures and stiff, unnatural motion. In contrast, generative video models show promising performance by synthesizing photorealistic and dynamic results, but they frequently struggle with unstable behavior, including body structural errors and identity drift. To address these limitations, we propose a novel approach that combines the strengths of both paradigms. Our framework employs a 3D reconstruction model to provide robust structural and appearance priors, which in turn guides a real-time autoregressive video diffusion model for rendering. This process enables the model to synthesize high-frequency, photorealistic details and fluid dynamics in real time, effectively reducing texture blur and motion stiffness while preventing the structural inconsistencies common in video generation methods. By uniting the geometric stability of 3D reconstruction with the generative capabilities of video models, our method produces high-fidelity digital avatars with realistic appearance and dynamic, temporally coherent motion. Experiments demonstrate that our approach significantly reduces artifacts and achieves substantial improvements in visual quality over leading methods, providing a robust and efficient solution for real-time applications such as gaming and virtual reality. Project page: https://lhyfst.github.io/visa", "date": "2025-12-08", "categories": []}
{"id": "2512.05259", "title": "Age-Inclusive 3D Human Mesh Recovery for Action-Preserving Data Anonymization", "abstract": "While three-dimensional (3D) shape and pose estimation is a highly researched area that has yielded significant advances, the resulting methods, despite performing well for the adult population, generally fail to generalize effectively to children and infants. This paper addresses this challenge by introducing AionHMR, a comprehensive framework designed to bridge this domain gap. We propose an optimization-based method that extends a top-performing model by incorporating the SMPL-A body model, enabling the concurrent and accurate modeling of adults, children, and infants. Leveraging this approach, we generated pseudo-ground-truth annotations for publicly available child and infant image databases. Using these new training data, we then developed and trained a specialized transformer-based deep learning model capable of real-time 3D age-inclusive human reconstruction. Extensive experiments demonstrate that our methods significantly improve shape and pose estimation for children and infants without compromising accuracy on adults. Importantly, our reconstructed meshes serve as privacy-preserving substitutes for raw images, retaining essential action, pose, and geometry information while enabling anonymized datasets release. As a demonstration, we introduce the 3D-BabyRobot dataset, a collection of action-preserving 3D reconstructions of children interacting with robots. This work bridges a crucial domain gap and establishes a foundation for inclusive, privacy-aware, and age-diverse 3D human modeling.", "date": "2025-12-04", "categories": []}
{"id": "2512.03593", "title": "CloseUpAvatar: High-Fidelity Animatable Full-Body Avatars with Mixture of Multi-Scale Textures", "abstract": "We present a CloseUpAvatar - a novel approach for articulated human avatar representation dealing with more general camera motions, while preserving rendering quality for close-up views. CloseUpAvatar represents an avatar as a set of textured planes with two sets of learnable textures for low and high-frequency detail. The method automatically switches to high-frequency textures only for cameras positioned close to the avatar's surface and gradually reduces their impact as the camera moves farther away. Such parametrization of the avatar enables CloseUpAvatar to adjust rendering quality based on camera distance ensuring realistic rendering across a wider range of camera orientations than previous approaches. We provide experiments using the ActorsHQ dataset with high-resolution input images. CloseUpAvatar demonstrates both qualitative and quantitative improvements over existing methods in rendering from novel wide range camera positions, while maintaining high FPS by limiting the number of required primitives.", "date": "2025-12-03", "categories": []}
{"id": "2512.01444", "title": "FastAnimate: Towards Learnable Template Construction and Pose Deformation for Fast 3D Human Avatar Animation", "abstract": "3D human avatar animation aims at transforming a human avatar from an arbitrary initial pose to a specified target pose using deformation algorithms. Existing approaches typically divide this task into two stages: canonical template construction and target pose deformation. However, current template construction methods demand extensive skeletal rigging and often produce artifacts for specific poses. Moreover, target pose deformation suffers from structural distortions caused by Linear Blend Skinning (LBS), which significantly undermines animation realism. To address these problems, we propose a unified learning-based framework to address both challenges in two phases. For the former phase, to overcome the inefficiencies and artifacts during template construction, we leverage a U-Net architecture that decouples texture and pose information in a feed-forward process, enabling fast generation of a human template. For the latter phase, we propose a data-driven refinement technique that enhances structural integrity. Extensive experiments show that our model delivers consistent performance across diverse poses with an optimal balance between efficiency and quality,surpassing state-of-the-art (SOTA) methods.", "date": "2025-12-01", "categories": []}
{"id": "2511.21245", "title": "FIELDS: Face reconstruction with accurate Inference of Expression using Learning with Direct Supervision", "abstract": "Facial expressions convey the bulk of emotional information in human communication, yet existing 3D face reconstruction methods often miss subtle affective details due to reliance on 2D supervision and lack of 3D ground truth. We propose FIELDS (Face reconstruction with accurate Inference of Expression using Learning with Direct Supervision) to address these limitations by extending self-supervised 2D image consistency cues with direct 3D expression parameter supervision and an auxiliary emotion recognition branch. Our encoder is guided by authentic expression parameters from spontaneous 4D facial scans, while an intensity-aware emotion loss encourages the 3D expression parameters to capture genuine emotion content without exaggeration. This dual-supervision strategy bridges the 2D/3D domain gap and mitigates expression-intensity bias, yielding high-fidelity 3D reconstructions that preserve subtle emotional cues. From a single image, FIELDS produces emotion-rich face models with highly realistic expressions, significantly improving in-the-wild facial expression recognition performance without sacrificing naturalness.", "date": "2025-11-26", "categories": []}
{"id": "2508.18389", "title": "FastAvatar: Instant 3D Gaussian Splatting for Faces from Single Unconstrained Poses", "abstract": "We present FastAvatar, a fast and robust algorithm for single-image 3D face reconstruction using 3D Gaussian Splatting (3DGS). Given a single input image from an arbitrary pose, FastAvatar recovers a high-quality, full-head 3DGS avatar in approximately 3 seconds on a single NVIDIA A100 GPU. We use a two-stage design: a feed-forward encoder-decoder predicts coarse face geometry by regressing Gaussian structure from a pose-invariant identity embedding, and a lightweight test-time refinement stage then optimizes the appearance parameters for photorealistic rendering. This hybrid strategy combines the speed and stability of direct prediction with the accuracy of optimization, enabling strong identity preservation even under extreme input poses. FastAvatar achieves state-of-the-art reconstruction quality (24.01 dB PSNR, 0.91 SSIM) while running over 600x faster than existing per-subject optimization methods (e.g., FlashAvatar, GaussianAvatars, GASP). Once reconstructed, our avatars support photorealistic novel-view synthesis and FLAME-guided expression animation, enabling controllable reenactment from a single image. By jointly offering high fidelity, robustness to pose, and rapid reconstruction, FastAvatar significantly broadens the applicability of 3DGS-based facial avatars.", "date": "2025-08-25", "categories": []}
{"id": "2511.20366", "title": "VGGTFace: Topologically Consistent Facial Geometry Reconstruction in the Wild", "abstract": "Reconstructing topologically consistent facial geometry is crucial for the digital avatar creation pipelines. Existing methods either require tedious manual efforts, lack generalization to in-the-wild data, or are constrained by the limited expressiveness of 3D Morphable Models. To address these limitations, we propose VGGTFace, an automatic approach that innovatively applies the 3D foundation model, i.e. VGGT, for topologically consistent facial geometry reconstruction from in-the-wild multi-view images captured by everyday users. Our key insight is that, by leveraging VGGT, our method naturally inherits strong generalization ability and expressive power from its large-scale training and point map representation. However, it is unclear how to reconstruct a topologically consistent mesh from VGGT, as the topology information is missing in its prediction. To this end, we augment VGGT with Pixel3DMM for injecting topology information via pixel-aligned UV values. In this manner, we convert the pixel-aligned point map of VGGT to a point cloud with topology. Tailored to this point cloud with known topology, we propose a novel Topology-Aware Bundle Adjustment strategy to fuse them, where we construct a Laplacian energy for the Bundle Adjustment objective. Our method achieves high-quality reconstruction in 10 seconds for 16 views on a single NVIDIA RTX 4090. Experiments demonstrate state-of-the-art results on benchmarks and impressive generalization to in-the-wild data. Code is available at https://github.com/grignarder/vggtface.", "date": "2025-11-25", "categories": []}
{"id": "2502.17852", "title": "Sketch-1-to-3: One Single Sketch to 3D Detailed Face Reconstruction", "abstract": "3D face reconstruction from a single sketch is a critical yet underexplored task with significant practical applications. The primary challenges stem from the substantial modality gap between 2D sketches and 3D facial structures, including: (1) accurately extracting facial keypoints from 2D sketches; (2) preserving diverse facial expressions and fine-grained texture details; and (3) training a high-performing model with limited data. In this paper, we propose Sketch-1-to-3, a novel framework for realistic 3D face reconstruction from a single sketch, to address these challenges. Specifically, we first introduce the Geometric Contour and Texture Detail (GCTD) module, which enhances the extraction of geometric contours and texture details from facial sketches. Additionally, we design a deep learning architecture with a domain adaptation module and a tailored loss function to align sketches with the 3D facial space, enabling high-fidelity expression and texture reconstruction. To facilitate evaluation and further research, we construct SketchFaces, a real hand-drawn facial sketch dataset, and Syn-SketchFaces, a synthetic facial sketch dataset. Extensive experiments demonstrate that Sketch-1-to-3 achieves state-of-the-art performance in sketch-based 3D face reconstruction.", "date": "2025-02-25", "categories": []}
{"id": "2505.21502", "title": "Generalizable and Relightable Gaussian Splatting for Human Novel View Synthesis", "abstract": "We propose GRGS, a generalizable and relightable 3D Gaussian framework for high-fidelity human novel view synthesis under diverse lighting conditions. Unlike existing methods that rely on per-character optimization or ignore physical constraints, GRGS adopts a feed-forward, fully supervised strategy projecting geometry, material, and illumination cues from multi-view 2D observations into 3D Gaussian representations. To recover accurate geometry under diverse lighting conditions, we introduce a Lighting-robust Geometry Refinement (LGR) module trained on synthetically relit data to predict precise depth and surface normals. Based on the high-quality geometry, a Physically Grounded Neural Rendering (PGNR) module is further proposed to integrate neural prediction with physics-based shading, supporting editable relighting with shadows and indirect illumination. Moreover, we design a 2D-to-3D projection training scheme leveraging differentiable supervision from ambient occlusion, direct, and indirect lighting maps, alleviating the computational cost of ray tracing. Extensive experiments demonstrate that GRGS achieves superior visual quality, geometric consistency, and generalization across characters and lighting conditions.", "date": "2025-05-27", "categories": []}
{"id": "2511.13282", "title": "Towards Metric-Aware Multi-Person Mesh Recovery by Jointly Optimizing Human Crowd in Camera Space", "abstract": "Multi-person human mesh recovery from a single image is a challenging task, hindered by the scarcity of in-the-wild training data. Prevailing in-the-wild human mesh pseudo-ground-truth (pGT) generation pipelines are single-person-centric, where each human is processed individually without joint optimization. This oversight leads to a lack of scene-level consistency, producing individuals with conflicting depths and scales within the same image. To address this, we introduce Depth-conditioned Translation Optimization (DTO), a novel optimization-based method that jointly refines the camera-space translations of all individuals in a crowd. By leveraging anthropometric priors on human height and depth cues from a monocular depth estimator, DTO solves for a scene-consistent placement of all subjects within a principled Maximum a posteriori (MAP) framework. Applying DTO to the 4D-Humans dataset, we construct DTO-Humans, a new large-scale pGT dataset of 0.56M high-quality, scene-consistent multi-person images, featuring dense crowds with an average of 4.8 persons per image. Furthermore, we propose Metric-Aware HMR, an end-to-end network that directly estimates human mesh and camera parameters in metric scale. This is enabled by a camera branch and a relative metric loss that enforces plausible relative scales. Extensive experiments demonstrate that our method achieves state-of-the-art performance on relative depth reasoning and human mesh recovery. Code is available at: https://github.com/gouba2333/MA-HMR.", "date": "2025-11-17", "categories": []}
{"id": "2511.16673", "title": "NoPo-Avatar: Generalizable and Animatable Avatars from Sparse Inputs without Human Poses", "abstract": "We tackle the task of recovering an animatable 3D human avatar from a single or a sparse set of images. For this task, beyond a set of images, many prior state-of-the-art methods use accurate \"ground-truth\" camera poses and human poses as input to guide reconstruction at test-time. We show that pose-dependent reconstruction degrades results significantly if pose estimates are noisy. To overcome this, we introduce NoPo-Avatar, which reconstructs avatars solely from images, without any pose input. By removing the dependence of test-time reconstruction on human poses, NoPo-Avatar is not affected by noisy human pose estimates, making it more widely applicable. Experiments on challenging THuman2.0, XHuman, and HuGe100K data show that NoPo-Avatar outperforms existing baselines in practical settings (without ground-truth poses) and delivers comparable results in lab settings (with ground-truth poses).", "date": "2025-11-20", "categories": []}
{"id": "2504.20607", "title": "EfficientHuman: Efficient Training and Reconstruction of Moving Human using Articulated 2D Gaussian", "abstract": "3D Gaussian Splatting (3DGS) has been recognized as a pioneering technique in scene reconstruction and novel view synthesis. Recent work on reconstructing the 3D human body using 3DGS attempts to leverage prior information on human pose to enhance rendering quality and improve training speed. However, it struggles to effectively fit dynamic surface planes due to multi-view inconsistency and redundant Gaussians. This inconsistency arises because Gaussian ellipsoids cannot accurately represent the surfaces of dynamic objects, which hinders the rapid reconstruction of the dynamic human body. Meanwhile, the prevalence of redundant Gaussians means that the training time of these works is still not ideal for quickly fitting a dynamic human body. To address these, we propose EfficientHuman, a model that quickly accomplishes the dynamic reconstruction of the human body using Articulated 2D Gaussian while ensuring high rendering quality. The key innovation involves encoding Gaussian splats as Articulated 2D Gaussian surfels in canonical space and then transforming them to pose space via Linear Blend Skinning (LBS) to achieve efficient pose transformations. Unlike 3D Gaussians, Articulated 2D Gaussian surfels can quickly conform to the dynamic human body while ensuring view-consistent geometries. Additionally, we introduce a pose calibration module and an LBS optimization module to achieve precise fitting of dynamic human poses, enhancing the model's performance. Extensive experiments on the ZJU-MoCap dataset demonstrate that EfficientHuman achieves rapid 3D dynamic human reconstruction in less than a minute on average, which is 20 seconds faster than the current state-of-the-art method, while also reducing the number of redundant Gaussians.", "date": "2025-04-29", "categories": []}
{"id": "2511.12935", "title": "PFAvatar: Pose-Fusion 3D Personalized Avatar Reconstruction from Real-World Outfit-of-the-Day Photos", "abstract": "We propose PFAvatar (Pose-Fusion Avatar), a new method that reconstructs high-quality 3D avatars from Outfit of the Day(OOTD) photos, which exhibit diverse poses, occlusions, and complex backgrounds. Our method consists of two stages: (1) fine-tuning a pose-aware diffusion model from few-shot OOTD examples and (2) distilling a 3D avatar represented by a neural radiance field (NeRF). In the first stage, unlike previous methods that segment images into assets (e.g., garments, accessories) for 3D assembly, which is prone to inconsistency, we avoid decomposition and directly model the full-body appearance. By integrating a pre-trained ControlNet for pose estimation and a novel Condition Prior Preservation Loss (CPPL), our method enables end-to-end learning of fine details while mitigating language drift in few-shot training. Our method completes personalization in just 5 minutes, achieving a 48x speed-up compared to previous approaches. In the second stage, we introduce a NeRF-based avatar representation optimized by canonical SMPL-X space sampling and Multi-Resolution 3D-SDS. Compared to mesh-based representations that suffer from resolution-dependent discretization and erroneous occluded geometry, our continuous radiance field can preserve high-frequency textures (e.g., hair) and handle occlusions correctly through transmittance. Experiments demonstrate that PFAvatar outperforms state-of-the-art methods in terms of reconstruction fidelity, detail preservation, and robustness to occlusions/truncations, advancing practical 3D avatar generation from real-world OOTD albums. In addition, the reconstructed 3D avatar supports downstream applications such as virtual try-on, animation, and human video reenactment, further demonstrating the versatility and practical value of our approach.", "date": "2025-11-17", "categories": []}
{"id": "2507.15979", "title": "Dream, Lift, Animate: From Single Images to Animatable Gaussian Avatars", "abstract": "We introduce Dream, Lift, Animate (DLA), a novel framework that reconstructs animatable 3D human avatars from a single image. This is achieved by leveraging multi-view generation, 3D Gaussian lifting, and pose-aware UV-space mapping of 3D Gaussians. Given an image, we first dream plausible multi-views using a video diffusion model, capturing rich geometric and appearance details. These views are then lifted into unstructured 3D Gaussians. To enable animation, we propose a transformer-based encoder that models global spatial relationships and projects these Gaussians into a structured latent representation aligned with the UV space of a parametric body model. This latent code is decoded into UV-space Gaussians that can be animated via body-driven deformation and rendered conditioned on pose and viewpoint. By anchoring Gaussians to the UV manifold, our method ensures consistency during animation while preserving fine visual details. DLA enables real-time rendering and intuitive editing without requiring post-processing. Our method outperforms state-of-the-art approaches on the ActorsHQ and 4D-Dress datasets in both perceptual quality and photometric accuracy. By combining the generative strengths of video diffusion models with a pose-aware UV-space Gaussian mapping, DLA bridges the gap between unstructured 3D representations and high-fidelity, animation-ready avatars.", "date": "2025-07-21", "categories": []}
{"id": "2508.04505", "title": "MonoCloth: Reconstruction and Animation of Cloth-Decoupled Human Avatars from Monocular Videos", "abstract": "Reconstructing realistic 3D human avatars from monocular videos is a challenging task due to the limited geometric information and complex non-rigid motion involved. We present MonoCloth, a new method for reconstructing and animating clothed human avatars from monocular videos. To overcome the limitations of monocular input, we introduce a part-based decomposition strategy that separates the avatar into body, face, hands, and clothing. This design reflects the varying levels of reconstruction difficulty and deformation complexity across these components. Specifically, we focus on detailed geometry recovery for the face and hands. For clothing, we propose a dedicated cloth simulation module that captures garment deformation using temporal motion cues and geometric constraints. Experimental results demonstrate that MonoCloth improves both visual reconstruction quality and animation realism compared to existing methods. Furthermore, thanks to its part-based design, MonoCloth also supports additional tasks such as clothing transfer, underscoring its versatility and practical utility.", "date": "2025-08-06", "categories": []}
{"id": "2510.12660", "title": "On the Use of Hierarchical Vision Foundation Models for Low-Cost Human Mesh Recovery and Pose Estimation", "abstract": "In this work, we aim to develop simple and efficient models for human mesh recovery (HMR) and its predecessor task, human pose estimation (HPE). State-of-the-art HMR methods, such as HMR2.0 and its successors, rely on large, non-hierarchical vision transformers as encoders, which are inherited from the corresponding HPE models like ViTPose. To establish baselines across varying computational budgets, we first construct three lightweight HMR2.0 variants by adapting the corresponding ViTPose models. In addition, we propose leveraging the early stages of hierarchical vision foundation models (VFMs), including Swin Transformer, GroupMixFormer, and VMamba, as encoders. This design is motivated by the observation that intermediate stages of hierarchical VFMs produce feature maps with resolutions comparable to or higher than those of non-hierarchical counterparts. We conduct a comprehensive evaluation of 27 hierarchical-VFM-based HMR and HPE models, demonstrating that using only the first two or three stages achieves performance on par with full-stage models. Moreover, we show that the resulting truncated models exhibit better trade-offs between accuracy and computational efficiency compared to existing lightweight alternatives. The source code is available at https://github.com/nttcom/TruncHierVFM.", "date": "2025-10-14", "categories": []}
{"id": "2506.06271", "title": "BecomingLit: Relightable Gaussian Avatars with Hybrid Neural Shading", "abstract": "We introduce BecomingLit, a novel method for reconstructing relightable, high-resolution head avatars that can be rendered from novel viewpoints at interactive rates. Therefore, we propose a new low-cost light stage capture setup, tailored specifically towards capturing faces. Using this setup, we collect a novel dataset consisting of diverse multi-view sequences of numerous subjects under varying illumination conditions and facial expressions. By leveraging our new dataset, we introduce a new relightable avatar representation based on 3D Gaussian primitives that we animate with a parametric head model and an expression-dependent dynamics module. We propose a new hybrid neural shading approach, combining a neural diffuse BRDF with an analytical specular term. Our method reconstructs disentangled materials from our dynamic light stage recordings and enables all-frequency relighting of our avatars with both point lights and environment maps. In addition, our avatars can easily be animated and controlled from monocular videos. We validate our approach in extensive experiments on our dataset, where we consistently outperform existing state-of-the-art methods in relighting and reenactment by a significant margin.", "date": "2025-06-06", "categories": []}
{"id": "2511.09147", "title": "PressTrack-HMR: Pressure-Based Top-Down Multi-Person Global Human Mesh Recovery", "abstract": "Multi-person global human mesh recovery (HMR) is crucial for understanding crowd dynamics and interactions. Traditional vision-based HMR methods sometimes face limitations in real-world scenarios due to mutual occlusions, insufficient lighting, and privacy concerns. Human-floor tactile interactions offer an occlusion-free and privacy-friendly alternative for capturing human motion. Existing research indicates that pressure signals acquired from tactile mats can effectively estimate human pose in single-person scenarios. However, when multiple individuals walk randomly on the mat simultaneously, how to distinguish intermingled pressure signals generated by different persons and subsequently acquire individual temporal pressure data remains a pending challenge for extending pressure-based HMR to the multi-person situation. In this paper, we present \\textbf{PressTrack-HMR}, a top-down pipeline that recovers multi-person global human meshes solely from pressure signals. This pipeline leverages a tracking-by-detection strategy to first identify and segment each individual's pressure signal from the raw pressure data, and subsequently performs HMR for each extracted individual signal. Furthermore, we build a multi-person interaction pressure dataset \\textbf{MIP}, which facilitates further research into pressure-based human motion analysis in multi-person scenarios. Experimental results demonstrate that our method excels in multi-person HMR using pressure data, with 89.2 $mm$ MPJPE and 112.6 $mm$ WA-MPJPE$_{100}$, and these showcase the potential of tactile mats for ubiquitous, privacy-preserving multi-person action recognition. Our dataset & code are available at https://github.com/Jiayue-Yuan/PressTrack-HMR.", "date": "2025-11-12", "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": "2506.01802", "title": "UMA: Ultra-detailed Human Avatars via Multi-level Surface Alignment", "abstract": "Learning an animatable and clothed human avatar model with vivid dynamics and photorealistic appearance from multi-view videos is an important foundational research problem in computer graphics and vision. Fueled by recent advances in implicit representations, the quality of the animatable avatars has achieved an unprecedented level by attaching the implicit representation to drivable human template meshes. However, they usually fail to preserve the highest level of detail, particularly apparent when the virtual camera is zoomed in and when rendering at 4K resolution and higher. We argue that this limitation stems from inaccurate surface tracking, specifically, depth misalignment and surface drift between character geometry and the ground truth surface, which forces the detailed appearance model to compensate for geometric errors. To address this, we propose a latent deformation model and supervising the 3D deformation of the animatable character using guidance from foundational 2D video point trackers, which offer improved robustness to shading and surface variations, and are less prone to local minima than differentiable rendering. To mitigate the drift over time and lack of 3D awareness of 2D point trackers, we introduce a cascaded training strategy that generates consistent 3D point tracks by anchoring point tracks to the rendered avatar, which ultimately supervises our avatar at the vertex and texel level. To validate the effectiveness of our approach, we introduce a novel dataset comprising five multi-view video sequences, each over 10 minutes in duration, captured using 40 calibrated 6K-resolution cameras, featuring subjects dressed in clothing with challenging texture patterns and wrinkle deformations. Our approach demonstrates significantly improved performance in rendering quality and geometric accuracy over the prior state of the art.", "date": "2025-06-02", "categories": []}
{"id": "2511.00468", "title": "HumanCrafter: Synergizing Generalizable Human Reconstruction and Semantic 3D Segmentation", "abstract": "Recent advances in generative models have achieved high-fidelity in 3D human reconstruction, yet their utility for specific tasks (e.g., human 3D segmentation) remains constrained. We propose HumanCrafter, a unified framework that enables the joint modeling of appearance and human-part semantics from a single image in a feed-forward manner. Specifically, we integrate human geometric priors in the reconstruction stage and self-supervised semantic priors in the segmentation stage. To address labeled 3D human datasets scarcity, we further develop an interactive annotation procedure for generating high-quality data-label pairs. Our pixel-aligned aggregation enables cross-task synergy, while the multi-task objective simultaneously optimizes texture modeling fidelity and semantic consistency. Extensive experiments demonstrate that HumanCrafter surpasses existing state-of-the-art methods in both 3D human-part segmentation and 3D human reconstruction from a single image.", "date": "2025-11-01", "categories": []}
{"id": "2412.04955", "title": "MixedGaussianAvatar: Realistically and Geometrically Accurate Head Avatar via Mixed 2D-3D Gaussians", "abstract": "Reconstructing high-fidelity 3D head avatars is crucial in various applications such as virtual reality. The pioneering methods reconstruct realistic head avatars with Neural Radiance Fields (NeRF), which have been limited by training and rendering speed. Recent methods based on 3D Gaussian Splatting (3DGS) significantly improve the efficiency of training and rendering. However, the surface inconsistency of 3DGS results in subpar geometric accuracy; later, 2DGS uses 2D surfels to enhance geometric accuracy at the expense of rendering fidelity. To leverage the benefits of both 2DGS and 3DGS, we propose a novel method named MixedGaussianAvatar for realistically and geometrically accurate head avatar reconstruction. Our main idea is to utilize 2D Gaussians to reconstruct the surface of the 3D head, ensuring geometric accuracy. We attach the 2D Gaussians to the triangular mesh of the FLAME model and connect additional 3D Gaussians to those 2D Gaussians where the rendering quality of 2DGS is inadequate, creating a mixed 2D-3D Gaussian representation. These 2D-3D Gaussians can then be animated using FLAME parameters. We further introduce a progressive training strategy that first trains the 2D Gaussians and then fine-tunes the mixed 2D-3D Gaussians. We use a unified mixed Gaussian representation to integrate the two modalities of 2D image and 3D mesh. Furthermore, the comprehensive experiments demonstrate the superiority of MixedGaussianAvatar. The code will be released.", "date": "2024-12-06", "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": "2510.14081", "title": "Capture, Canonicalize, Splat: Zero-Shot 3D Gaussian Avatars from Unstructured Phone Images", "abstract": "We present a novel, zero-shot pipeline for creating hyperrealistic, identity-preserving 3D avatars from a few unstructured phone images. Existing methods face several challenges: single-view approaches suffer from geometric inconsistencies and hallucinations, degrading identity preservation, while models trained on synthetic data fail to capture high-frequency details like skin wrinkles and fine hair, limiting realism. Our method introduces two key contributions: (1) a generative canonicalization module that processes multiple unstructured views into a standardized, consistent representation, and (2) a transformer-based model trained on a new, large-scale dataset of high-fidelity Gaussian splatting avatars derived from dome captures of real people. This \"Capture, Canonicalize, Splat\" pipeline produces static quarter-body avatars with compelling realism and robust identity preservation from unstructured photos.", "date": "2025-10-15", "categories": []}
{"id": "2510.22140", "title": "STG-Avatar: Animatable Human Avatars via Spacetime Gaussian", "abstract": "Realistic animatable human avatars from monocular videos are crucial for advancing human-robot interaction and enhancing immersive virtual experiences. While recent research on 3DGS-based human avatars has made progress, it still struggles with accurately representing detailed features of non-rigid objects (e.g., clothing deformations) and dynamic regions (e.g., rapidly moving limbs). To address these challenges, we present STG-Avatar, a 3DGS-based framework for high-fidelity animatable human avatar reconstruction. Specifically, our framework introduces a rigid-nonrigid coupled deformation framework that synergistically integrates Spacetime Gaussians (STG) with linear blend skinning (LBS). In this hybrid design, LBS enables real-time skeletal control by driving global pose transformations, while STG complements it through spacetime adaptive optimization of 3D Gaussians. Furthermore, we employ optical flow to identify high-dynamic regions and guide the adaptive densification of 3D Gaussians in these regions. Experimental results demonstrate that our method consistently outperforms state-of-the-art baselines in both reconstruction quality and operational efficiency, achieving superior quantitative metrics while retaining real-time rendering capabilities. Our code is available at https://github.com/jiangguangan/STG-Avatar", "date": "2025-10-25", "categories": []}
{"id": "2510.18256", "title": "Hyperbolic Space Learning Method Leveraging Temporal Motion Priors for Human Mesh Recovery", "abstract": "3D human meshes show a natural hierarchical structure (like torso-limbs-fingers). But existing video-based 3D human mesh recovery methods usually learn mesh features in Euclidean space. It's hard to catch this hierarchical structure accurately. So wrong human meshes are reconstructed. To solve this problem, we propose a hyperbolic space learning method leveraging temporal motion prior for recovering 3D human meshes from videos. First, we design a temporal motion prior extraction module. This module extracts the temporal motion features from the input 3D pose sequences and image feature sequences respectively. Then it combines them into the temporal motion prior. In this way, it can strengthen the ability to express features in the temporal motion dimension. Since data representation in non-Euclidean space has been proved to effectively capture hierarchical relationships in real-world datasets (especially in hyperbolic space), we further design a hyperbolic space optimization learning strategy. This strategy uses the temporal motion prior information to assist learning, and uses 3D pose and pose motion information respectively in the hyperbolic space to optimize and learn the mesh features. Then, we combine the optimized results to get an accurate and smooth human mesh. Besides, to make the optimization learning process of human meshes in hyperbolic space stable and effective, we propose a hyperbolic mesh optimization loss. Extensive experimental results on large publicly available datasets indicate superiority in comparison with most state-of-the-art.", "date": "2025-10-21", "categories": []}
{"id": "2510.18267", "title": "Latent-Info and Low-Dimensional Learning for Human Mesh Recovery and Parallel Optimization", "abstract": "Existing 3D human mesh recovery methods often fail to fully exploit the latent information (e.g., human motion, shape alignment), leading to issues with limb misalignment and insufficient local details in the reconstructed human mesh (especially in complex scenes). Furthermore, the performance improvement gained by modelling mesh vertices and pose node interactions using attention mechanisms comes at a high computational cost. To address these issues, we propose a two-stage network for human mesh recovery based on latent information and low dimensional learning. Specifically, the first stage of the network fully excavates global (e.g., the overall shape alignment) and local (e.g., textures, detail) information from the low and high-frequency components of image features and aggregates this information into a hybrid latent frequency domain feature. This strategy effectively extracts latent information. Subsequently, utilizing extracted hybrid latent frequency domain features collaborates to enhance 2D poses to 3D learning. In the second stage, with the assistance of hybrid latent features, we model the interaction learning between the rough 3D human mesh template and the 3D pose, optimizing the pose and shape of the human mesh. Unlike existing mesh pose interaction methods, we design a low-dimensional mesh pose interaction method through dimensionality reduction and parallel optimization that significantly reduces computational costs without sacrificing reconstruction accuracy. Extensive experimental results on large publicly available datasets indicate superiority compared to the most state-of-the-art.", "date": "2025-10-21", "categories": []}
{"id": "2510.16463", "title": "HGC-Avatar: Hierarchical Gaussian Compression for Streamable Dynamic 3D Avatars", "abstract": "Recent advances in 3D Gaussian Splatting (3DGS) have enabled fast, photorealistic rendering of dynamic 3D scenes, showing strong potential in immersive communication. However, in digital human encoding and transmission, the compression methods based on general 3DGS representations are limited by the lack of human priors, resulting in suboptimal bitrate efficiency and reconstruction quality at the decoder side, which hinders their application in streamable 3D avatar systems. We propose HGC-Avatar, a novel Hierarchical Gaussian Compression framework designed for efficient transmission and high-quality rendering of dynamic avatars. Our method disentangles the Gaussian representation into a structural layer, which maps poses to Gaussians via a StyleUNet-based generator, and a motion layer, which leverages the SMPL-X model to represent temporal pose variations compactly and semantically. This hierarchical design supports layer-wise compression, progressive decoding, and controllable rendering from diverse pose inputs such as video sequences or text. Since people are most concerned with facial realism, we incorporate a facial attention mechanism during StyleUNet training to preserve identity and expression details under low-bitrate constraints. Experimental results demonstrate that HGC-Avatar provides a streamable solution for rapid 3D avatar rendering, while significantly outperforming prior methods in both visual quality and compression efficiency.", "date": "2025-10-18", "categories": []}
{"id": "2510.17181", "title": "Capturing Head Avatar with Hand Contacts from a Monocular Video", "abstract": "Photorealistic 3D head avatars are vital for telepresence, gaming, and VR. However, most methods focus solely on facial regions, ignoring natural hand-face interactions, such as a hand resting on the chin or fingers gently touching the cheek, which convey cognitive states like pondering. In this work, we present a novel framework that jointly learns detailed head avatars and the non-rigid deformations induced by hand-face interactions.\n  There are two principal challenges in this task. First, naively tracking hand and face separately fails to capture their relative poses. To overcome this, we propose to combine depth order loss with contact regularization during pose tracking, ensuring correct spatial relationships between the face and hand. Second, no publicly available priors exist for hand-induced deformations, making them non-trivial to learn from monocular videos. To address this, we learn a PCA basis specific to hand-induced facial deformations from a face-hand interaction dataset. This reduces the problem to estimating a compact set of PCA parameters rather than a full spatial deformation field. Furthermore, inspired by physics-based simulation, we incorporate a contact loss that provides additional supervision, significantly reducing interpenetration artifacts and enhancing the physical plausibility of the results.\n  We evaluate our approach on RGB(D) videos captured by an iPhone. Additionally, to better evaluate the reconstructed geometry, we construct a synthetic dataset of avatars with various types of hand interactions. We show that our method can capture better appearance and more accurate deforming geometry of the face than SOTA surface reconstruction methods.", "date": "2025-10-20", "categories": []}
{"id": "2403.11111", "title": "Diffusion Models are Efficient Data Generators for Human Mesh Recovery", "abstract": "Despite remarkable progress having been made on the problem of 3D human pose and shape estimation (HPS), current state-of-the-art methods rely heavily on either confined indoor mocap datasets or datasets generated by a rendering engine using computer graphics (CG). Both categories of datasets exhibit inadequacies in furnishing adequate human identities and authentic in-the-wild background scenes, which are crucial for accurately simulating real-world distributions. In this work, we show that synthetic data created by generative models is complementary to CG-rendered data for achieving remarkable generalization performance on diverse real-world scenes. We propose an effective data generation pipeline based on recent diffusion models, termed HumanWild, which can effortlessly generate human images and corresponding 3D mesh annotations. Specifically, we first collect a large-scale human-centric dataset with comprehensive annotations, e.g, text captions, the depth map, and surface normal images. To generate a wide variety of human images with initial labels, we train a customized, multi-condition ControlNet model. The key to this process is using a 3D parametric model, e.g, SMPL-X, to create various condition inputs easily. Our data generation pipeline is both flexible and customizable, making it adaptable to multiple real-world tasks, such as human interaction in complex scenes and humans captured by wide-angle lenses. By relying solely on generative models, we can produce large-scale, in-the-wild human images with high-quality annotations, significantly reducing the need for manual image collection and annotation. The generated dataset encompasses a wide range of viewpoints, environments, and human identities, ensuring its versatility across different scenarios. We hope that our work could pave the way for scaling up 3D human recovery to in-the-wild scenes.", "date": "2024-03-17", "categories": []}
{"id": "2503.15671", "title": "CHROME: Clothed Human Reconstruction with Occlusion-Resilience and Multiview-Consistency from a Single Image", "abstract": "Reconstructing clothed humans from a single image is a fundamental task in computer vision with wide-ranging applications. Although existing monocular clothed human reconstruction solutions have shown promising results, they often rely on the assumption that the human subject is in an occlusion-free environment. Thus, when encountering in-the-wild occluded images, these algorithms produce multiview inconsistent and fragmented reconstructions. Additionally, most algorithms for monocular 3D human reconstruction leverage geometric priors such as SMPL annotations for training and inference, which are extremely challenging to acquire in real-world applications. To address these limitations, we propose CHROME: Clothed Human Reconstruction with Occlusion-Resilience and Multiview-ConsistEncy from a Single Image, a novel pipeline designed to reconstruct occlusion-resilient 3D humans with multiview consistency from a single occluded image, without requiring either ground-truth geometric prior annotations or 3D supervision. Specifically, CHROME leverages a multiview diffusion model to first synthesize occlusion-free human images from the occluded input, compatible with off-the-shelf pose control to explicitly enforce cross-view consistency during synthesis. A 3D reconstruction model is then trained to predict a set of 3D Gaussians conditioned on both the occluded input and synthesized views, aligning cross-view details to produce a cohesive and accurate 3D representation. CHROME achieves significant improvements in terms of both novel view synthesis (upto 3 db PSNR) and geometric reconstruction under challenging conditions.", "date": "2025-03-19", "categories": []}
{"id": "2504.12292", "title": "SHeaP: Self-Supervised Head Geometry Predictor Learned via 2D Gaussians", "abstract": "Accurate, real-time 3D reconstruction of human heads from monocular images and videos underlies numerous visual applications. As 3D ground truth data is hard to come by at scale, previous methods have sought to learn from abundant 2D videos in a self-supervised manner. Typically, this involves the use of differentiable mesh rendering, which is effective but faces limitations. To improve on this, we propose SHeaP (Self-supervised Head Geometry Predictor Learned via 2D Gaussians). Given a source image, we predict a 3DMM mesh and a set of Gaussians that are rigged to this mesh. We then reanimate this rigged head avatar to match a target frame, and backpropagate photometric losses to both the 3DMM and Gaussian prediction networks. We find that using Gaussians for rendering substantially improves the effectiveness of this self-supervised approach. Training solely on 2D data, our method surpasses existing self-supervised approaches in geometric evaluations on the NoW benchmark for neutral faces and a new benchmark for non-neutral expressions. Our method also produces highly expressive meshes, outperforming state-of-the-art in emotion classification.", "date": "2025-04-16", "categories": []}
{"id": "2510.15342", "title": "SHARE: Scene-Human Aligned Reconstruction", "abstract": "Animating realistic character interactions with the surrounding environment is important for autonomous agents in gaming, AR/VR, and robotics. However, current methods for human motion reconstruction struggle with accurately placing humans in 3D space. We introduce Scene-Human Aligned REconstruction (SHARE), a technique that leverages the scene geometry's inherent spatial cues to accurately ground human motion reconstruction. Each reconstruction relies solely on a monocular RGB video from a stationary camera. SHARE first estimates a human mesh and segmentation mask for every frame, alongside a scene point map at keyframes. It iteratively refines the human's positions at these keyframes by comparing the human mesh against the human point map extracted from the scene using the mask. Crucially, we also ensure that non-keyframe human meshes remain consistent by preserving their relative root joint positions to keyframe root joints during optimization. Our approach enables more accurate 3D human placement while reconstructing the surrounding scene, facilitating use cases on both curated datasets and in-the-wild web videos. Extensive experiments demonstrate that SHARE outperforms existing methods.", "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": "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": "2510.07723", "title": "SyncHuman: Synchronizing 2D and 3D Generative Models for Single-view Human Reconstruction", "abstract": "Photorealistic 3D full-body human reconstruction from a single image is a critical yet challenging task for applications in films and video games due to inherent ambiguities and severe self-occlusions. While recent approaches leverage SMPL estimation and SMPL-conditioned image generative models to hallucinate novel views, they suffer from inaccurate 3D priors estimated from SMPL meshes and have difficulty in handling difficult human poses and reconstructing fine details. In this paper, we propose SyncHuman, a novel framework that combines 2D multiview generative model and 3D native generative model for the first time, enabling high-quality clothed human mesh reconstruction from single-view images even under challenging human poses. Multiview generative model excels at capturing fine 2D details but struggles with structural consistency, whereas 3D native generative model generates coarse yet structurally consistent 3D shapes. By integrating the complementary strengths of these two approaches, we develop a more effective generation framework. Specifically, we first jointly fine-tune the multiview generative model and the 3D native generative model with proposed pixel-aligned 2D-3D synchronization attention to produce geometrically aligned 3D shapes and 2D multiview images. To further improve details, we introduce a feature injection mechanism that lifts fine details from 2D multiview images onto the aligned 3D shapes, enabling accurate and high-fidelity reconstruction. Extensive experiments demonstrate that SyncHuman achieves robust and photo-realistic 3D human reconstruction, even for images with challenging poses. Our method outperforms baseline methods in geometric accuracy and visual fidelity, demonstrating a promising direction for future 3D generation models.", "date": "2025-10-09", "categories": []}
{"id": "2510.10492", "title": "Towards Efficient 3D Gaussian Human Avatar Compression: A Prior-Guided Framework", "abstract": "This paper proposes an efficient 3D avatar coding framework that leverages compact human priors and canonical-to-target transformation to enable high-quality 3D human avatar video compression at ultra-low bit rates. The framework begins by training a canonical Gaussian avatar using articulated splatting in a network-free manner, which serves as the foundation for avatar appearance modeling. Simultaneously, a human-prior template is employed to capture temporal body movements through compact parametric representations. This decomposition of appearance and temporal evolution minimizes redundancy, enabling efficient compression: the canonical avatar is shared across the sequence, requiring compression only once, while the temporal parameters, consisting of just 94 parameters per frame, are transmitted with minimal bit-rate. For each frame, the target human avatar is generated by deforming canonical avatar via Linear Blend Skinning transformation, facilitating temporal coherent video reconstruction and novel view synthesis. Experimental results demonstrate that the proposed method significantly outperforms conventional 2D/3D codecs and existing learnable dynamic 3D Gaussian splatting compression method in terms of rate-distortion performance on mainstream multi-view human video datasets, paving the way for seamless immersive multimedia experiences in meta-verse applications.", "date": "2025-10-12", "categories": []}
{"id": "2510.10793", "title": "ImHead: A Large-scale Implicit Morphable Model for Localized Head Modeling", "abstract": "Over the last years, 3D morphable models (3DMMs) have emerged as a state-of-the-art methodology for modeling and generating expressive 3D avatars. However, given their reliance on a strict topology, along with their linear nature, they struggle to represent complex full-head shapes. Following the advent of deep implicit functions, we propose imHead, a novel implicit 3DMM that not only models expressive 3D head avatars but also facilitates localized editing of the facial features. Previous methods directly divided the latent space into local components accompanied by an identity encoding to capture the global shape variations, leading to expensive latent sizes. In contrast, we retain a single compact identity space and introduce an intermediate region-specific latent representation to enable local edits. To train imHead, we curate a large-scale dataset of 4K distinct identities, making a step-towards large scale 3D head modeling. Under a series of experiments we demonstrate the expressive power of the proposed model to represent diverse identities and expressions outperforming previous approaches. Additionally, the proposed approach provides an interpretable solution for 3D face manipulation, allowing the user to make localized edits.", "date": "2025-10-12", "categories": []}
{"id": "2510.10868", "title": "FastHMR: Accelerating Human Mesh Recovery via Token and Layer Merging with Diffusion Decoding", "abstract": "Recent transformer-based models for 3D Human Mesh Recovery (HMR) have achieved strong performance but often suffer from high computational cost and complexity due to deep transformer architectures and redundant tokens. In this paper, we introduce two HMR-specific merging strategies: Error-Constrained Layer Merging (ECLM) and Mask-guided Token Merging (Mask-ToMe). ECLM selectively merges transformer layers that have minimal impact on the Mean Per Joint Position Error (MPJPE), while Mask-ToMe focuses on merging background tokens that contribute little to the final prediction. To further address the potential performance drop caused by merging, we propose a diffusion-based decoder that incorporates temporal context and leverages pose priors learned from large-scale motion capture datasets. Experiments across multiple benchmarks demonstrate that our method achieves up to 2.3x speed-up while slightly improving performance over the baseline.", "date": "2025-10-13", "categories": []}
{"id": "2510.11649", "title": "PhySIC: Physically Plausible 3D Human-Scene Interaction and Contact from a Single Image", "abstract": "Reconstructing metrically accurate humans and their surrounding scenes from a single image is crucial for virtual reality, robotics, and comprehensive 3D scene understanding. However, existing methods struggle with depth ambiguity, occlusions, and physically inconsistent contacts. To address these challenges, we introduce PhySIC, a framework for physically plausible Human-Scene Interaction and Contact reconstruction. PhySIC recovers metrically consistent SMPL-X human meshes, dense scene surfaces, and vertex-level contact maps within a shared coordinate frame from a single RGB image. Starting from coarse monocular depth and body estimates, PhySIC performs occlusion-aware inpainting, fuses visible depth with unscaled geometry for a robust metric scaffold, and synthesizes missing support surfaces like floors. A confidence-weighted optimization refines body pose, camera parameters, and global scale by jointly enforcing depth alignment, contact priors, interpenetration avoidance, and 2D reprojection consistency. Explicit occlusion masking safeguards invisible regions against implausible configurations. PhySIC is efficient, requiring only 9 seconds for joint human-scene optimization and under 27 seconds end-to-end. It naturally handles multiple humans, enabling reconstruction of diverse interactions. Empirically, PhySIC outperforms single-image baselines, reducing mean per-vertex scene error from 641 mm to 227 mm, halving PA-MPJPE to 42 mm, and improving contact F1 from 0.09 to 0.51. Qualitative results show realistic foot-floor interactions, natural seating, and plausible reconstructions of heavily occluded furniture. By converting a single image into a physically plausible 3D human-scene pair, PhySIC advances scalable 3D scene understanding. Our implementation is publicly available at https://yuxuan-xue.com/physic.", "date": "2025-10-13", "categories": []}
{"id": "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": "2509.07552", "title": "PanoLAM: Large Avatar Model for Gaussian Full-Head Synthesis from One-shot Unposed Image", "abstract": "We present a feed-forward framework for Gaussian full-head synthesis from a single unposed image. Unlike previous work that relies on time-consuming GAN inversion and test-time optimization, our framework can reconstruct the Gaussian full-head model given a single unposed image in a single forward pass. This enables fast reconstruction and rendering during inference. To mitigate the lack of large-scale 3D head assets, we propose a large-scale synthetic dataset from trained 3D GANs and train our framework using only synthetic data. For efficient high-fidelity generation, we introduce a coarse-to-fine Gaussian head generation pipeline, where sparse points from the FLAME model interact with the image features by transformer blocks for feature extraction and coarse shape reconstruction, which are then densified for high-fidelity reconstruction. To fully leverage the prior knowledge residing in pretrained 3D GANs for effective reconstruction, we propose a dual-branch framework that effectively aggregates the structured spherical triplane feature and unstructured point-based features for more effective Gaussian head reconstruction. Experimental results show the effectiveness of our framework towards existing work. Project page at: https://panolam.github.io/.", "date": "2025-09-09", "categories": []}
{"id": "2407.15212", "title": "Surfel-based Gaussian Inverse Rendering for Fast and Relightable Dynamic Human Reconstruction from Monocular Video", "abstract": "Efficient and accurate reconstruction of a relightable, dynamic clothed human avatar from a monocular video is crucial for the entertainment industry. This paper presents SGIA (Surfel-based Gaussian Inverse Avatar), which introduces efficient training and rendering for relightable dynamic human reconstruction. SGIA advances previous Gaussian Avatar methods by comprehensively modeling Physically-Based Rendering (PBR) properties for clothed human avatars, allowing for the manipulation of avatars into novel poses under diverse lighting conditions. Specifically, our approach integrates pre-integration and image-based lighting for fast light calculations that surpass the performance of existing implicit-based techniques. To address challenges related to material lighting disentanglement and accurate geometry reconstruction, we propose an innovative occlusion approximation strategy and a progressive training approach. Extensive experiments demonstrate that SGIA not only achieves highly accurate physical properties but also significantly enhances the realistic relighting of dynamic human avatars, providing a substantial speed advantage. We exhibit more results in our project page: https://GS-IA.github.io.", "date": "2024-07-21", "categories": []}
{"id": "2510.06046", "title": "GLVD: Guided Learned Vertex Descent", "abstract": "Existing 3D face modeling methods usually depend on 3D Morphable Models, which inherently constrain the representation capacity to fixed shape priors. Optimization-based approaches offer high-quality reconstructions but tend to be computationally expensive. In this work, we introduce GLVD, a hybrid method for 3D face reconstruction from few-shot images that extends Learned Vertex Descent (LVD) by integrating per-vertex neural field optimization with global structural guidance from dynamically predicted 3D keypoints. By incorporating relative spatial encoding, GLVD iteratively refines mesh vertices without requiring dense 3D supervision. This enables expressive and adaptable geometry reconstruction while maintaining computational efficiency. GLVD achieves state-of-the-art performance in single-view settings and remains highly competitive in multi-view scenarios, all while substantially reducing inference time.", "date": "2025-10-07", "categories": []}
{"id": "2404.05606", "title": "Reconstructing Topology-Consistent Face Mesh by Volume Rendering from Multi-View Images", "abstract": "Industrial 3D face assets creation typically reconstructs topology-consistent face meshes from multi-view images for downstream production. However, high-quality reconstruction usually requires manual processing or specific capture settings. Recently NeRF has shown great advantages in 3D reconstruction, by representing scenes as density and radiance fields and utilizing neural volume rendering for novel view synthesis. Inspired by this, we introduce a novel method which combines explicit mesh with neural volume rendering to optimize geometry of an artist-made template face mesh from multi-view images while keeping the topology unchanged. Our method derives density fields from meshes using distance fields as an intermediary and encodes radiance field in compact tri-planes. To improve convergence, several adaptions tailored for meshes are introduced to the volume rendering. Experiments demonstrate that our method achieves superior reconstruction quality compared to previous approaches, validating the feasibility of integrating mesh and neural volume rendering.", "date": "2024-04-08", "categories": []}
{"id": "2510.02566", "title": "PhysHMR: Learning Humanoid Control Policies from Vision for Physically Plausible Human Motion Reconstruction", "abstract": "Reconstructing physically plausible human motion from monocular videos remains a challenging problem in computer vision and graphics. Existing methods primarily focus on kinematics-based pose estimation, often leading to unrealistic results due to the lack of physical constraints. To address such artifacts, prior methods have typically relied on physics-based post-processing following the initial kinematics-based motion estimation. However, this two-stage design introduces error accumulation, ultimately limiting the overall reconstruction quality. In this paper, we present PhysHMR, a unified framework that directly learns a visual-to-action policy for humanoid control in a physics-based simulator, enabling motion reconstruction that is both physically grounded and visually aligned with the input video. A key component of our approach is the pixel-as-ray strategy, which lifts 2D keypoints into 3D spatial rays and transforms them into global space. These rays are incorporated as policy inputs, providing robust global pose guidance without depending on noisy 3D root predictions. This soft global grounding, combined with local visual features from a pretrained encoder, allows the policy to reason over both detailed pose and global positioning. To overcome the sample inefficiency of reinforcement learning, we further introduce a distillation scheme that transfers motion knowledge from a mocap-trained expert to the vision-conditioned policy, which is then refined using physically motivated reinforcement learning rewards. Extensive experiments demonstrate that PhysHMR produces high-fidelity, physically plausible motion across diverse scenarios, outperforming prior approaches in both visual accuracy and physical realism.", "date": "2025-10-02", "categories": []}
{"id": "2311.11971", "title": "LiDAR-HMR: 3D Human Mesh Recovery from LiDAR", "abstract": "In recent years, point cloud perception tasks have been garnering increasing attention. This paper presents the first attempt to estimate 3D human body mesh from sparse LiDAR point clouds. We found that the major challenge in estimating human pose and mesh from point clouds lies in the sparsity, noise, and incompletion of LiDAR point clouds. Facing these challenges, we propose an effective sparse-to-dense reconstruction scheme to reconstruct 3D human mesh. This involves estimating a sparse representation of a human (3D human pose) and gradually reconstructing the body mesh. To better leverage the 3D structural information of point clouds, we employ a cascaded graph transformer (graphormer) to introduce point cloud features during sparse-to-dense reconstruction. Experimental results on three publicly available databases demonstrate the effectiveness of the proposed approach. Code: https://github.com/soullessrobot/LiDAR-HMR/", "date": "2023-11-20", "categories": []}
{"id": "2510.01619", "title": "MPMAvatar: Learning 3D Gaussian Avatars with Accurate and Robust Physics-Based Dynamics", "abstract": "While there has been significant progress in the field of 3D avatar creation from visual observations, modeling physically plausible dynamics of humans with loose garments remains a challenging problem. Although a few existing works address this problem by leveraging physical simulation, they suffer from limited accuracy or robustness to novel animation inputs. In this work, we present MPMAvatar, a framework for creating 3D human avatars from multi-view videos that supports highly realistic, robust animation, as well as photorealistic rendering from free viewpoints. For accurate and robust dynamics modeling, our key idea is to use a Material Point Method-based simulator, which we carefully tailor to model garments with complex deformations and contact with the underlying body by incorporating an anisotropic constitutive model and a novel collision handling algorithm. We combine this dynamics modeling scheme with our canonical avatar that can be rendered using 3D Gaussian Splatting with quasi-shadowing, enabling high-fidelity rendering for physically realistic animations. In our experiments, we demonstrate that MPMAvatar significantly outperforms the existing state-of-the-art physics-based avatar in terms of (1) dynamics modeling accuracy, (2) rendering accuracy, and (3) robustness and efficiency. Additionally, we present a novel application in which our avatar generalizes to unseen interactions in a zero-shot manner-which was not achievable with previous learning-based methods due to their limited simulation generalizability. Our project page is at: https://KAISTChangmin.github.io/MPMAvatar/", "date": "2025-10-02", "categories": []}
{"id": "2510.00506", "title": "Affordance-Guided Diffusion Prior for 3D Hand Reconstruction", "abstract": "How can we reconstruct 3D hand poses when large portions of the hand are heavily occluded by itself or by objects? Humans often resolve such ambiguities by leveraging contextual knowledge -- such as affordances, where an object's shape and function suggest how the object is typically grasped. Inspired by this observation, we propose a generative prior for hand pose refinement guided by affordance-aware textual descriptions of hand-object interactions (HOI). Our method employs a diffusion-based generative model that learns the distribution of plausible hand poses conditioned on affordance descriptions, which are inferred from a large vision-language model (VLM). This enables the refinement of occluded regions into more accurate and functionally coherent hand poses. Extensive experiments on HOGraspNet, a 3D hand-affordance dataset with severe occlusions, demonstrate that our affordance-guided refinement significantly improves hand pose estimation over both recent regression methods and diffusion-based refinement lacking contextual reasoning.", "date": "2025-10-01", "categories": []}
{"id": "2509.26621", "title": "HART: Human Aligned Reconstruction Transformer", "abstract": "We introduce HART, a unified framework for sparse-view human reconstruction. Given a small set of uncalibrated RGB images of a person as input, it outputs a watertight clothed mesh, the aligned SMPL-X body mesh, and a Gaussian-splat representation for photorealistic novel-view rendering. Prior methods for clothed human reconstruction either optimize parametric templates, which overlook loose garments and human-object interactions, or train implicit functions under simplified camera assumptions, limiting applicability in real scenes. In contrast, HART predicts per-pixel 3D point maps, normals, and body correspondences, and employs an occlusion-aware Poisson reconstruction to recover complete geometry, even in self-occluded regions. These predictions also align with a parametric SMPL-X body model, ensuring that reconstructed geometry remains consistent with human structure while capturing loose clothing and interactions. These human-aligned meshes initialize Gaussian splats to further enable sparse-view rendering. While trained on only 2.3K synthetic scans, HART achieves state-of-the-art results: Chamfer Distance improves by 18-23 percent for clothed-mesh reconstruction, PA-V2V drops by 6-27 percent for SMPL-X estimation, LPIPS decreases by 15-27 percent for novel-view synthesis on a wide range of datasets. These results suggest that feed-forward transformers can serve as a scalable model for robust human reconstruction in real-world settings. Code and models will be released.", "date": "2025-09-30", "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": "2509.10024", "title": "Hierarchical MLANet: Multi-level Attention for 3D Face Reconstruction From Single Images", "abstract": "Recovering 3D face models from 2D in-the-wild images has gained considerable attention in the computer vision community due to its wide range of potential applications. However, the lack of ground-truth labeled datasets and the complexity of real-world environments remain significant challenges. In this chapter, we propose a convolutional neural network-based approach, the Hierarchical Multi-Level Attention Network (MLANet), for reconstructing 3D face models from single in-the-wild images. Our model predicts detailed facial geometry, texture, pose, and illumination parameters from a single image. Specifically, we employ a pre-trained hierarchical backbone network and introduce multi-level attention mechanisms at different stages of 2D face image feature extraction. A semi-supervised training strategy is employed, incorporating 3D Morphable Model (3DMM) parameters from publicly available datasets along with a differentiable renderer, enabling an end-to-end training process. Extensive experiments, including both comparative and ablation studies, were conducted on two benchmark datasets, AFLW2000-3D and MICC Florence, focusing on 3D face reconstruction and 3D face alignment tasks. The effectiveness of the proposed method was evaluated both quantitatively and qualitatively.", "date": "2025-09-12", "categories": []}
{"id": "2509.24209", "title": "Forge4D: Feed-Forward 4D Human Reconstruction and Interpolation from Uncalibrated Sparse-view Videos", "abstract": "Instant reconstruction of dynamic 3D humans from uncalibrated sparse-view videos is critical for numerous downstream applications. Existing methods, however, are either limited by the slow reconstruction speeds or incapable of generating novel-time representations. To address these challenges, we propose Forge4D, a feed-forward 4D human reconstruction and interpolation model that efficiently reconstructs temporally aligned representations from uncalibrated sparse-view videos, enabling both novel view and novel time synthesis. Our model simplifies the 4D reconstruction and interpolation problem as a joint task of streaming 3D Gaussian reconstruction and dense motion prediction. For the task of streaming 3D Gaussian reconstruction, we first reconstruct static 3D Gaussians from uncalibrated sparse-view images and then introduce learnable state tokens to enforce temporal consistency in a memory-friendly manner by interactively updating shared information across different timestamps. For novel time synthesis, we design a novel motion prediction module to predict dense motions for each 3D Gaussian between two adjacent frames, coupled with an occlusion-aware Gaussian fusion process to interpolate 3D Gaussians at arbitrary timestamps. To overcome the lack of the ground truth for dense motion supervision, we formulate dense motion prediction as a dense point matching task and introduce a self-supervised retargeting loss to optimize this module. An additional occlusion-aware optical flow loss is introduced to ensure motion consistency with plausible human movement, providing stronger regularization. Extensive experiments demonstrate the effectiveness of our model on both in-domain and out-of-domain datasets. Project page and code at: https://zhenliuzju.github.io/huyingdong/Forge4D.", "date": "2025-09-29", "categories": []}
{"id": "2509.21859", "title": "SRHand: Super-Resolving Hand Images and 3D Shapes via View/Pose-aware Neural Image Representations and Explicit 3D Meshes", "abstract": "Reconstructing detailed hand avatars plays a crucial role in various applications. While prior works have focused on capturing high-fidelity hand geometry, they heavily rely on high-resolution multi-view image inputs and struggle to generalize on low-resolution images. Multi-view image super-resolution methods have been proposed to enforce 3D view consistency. These methods, however, are limited to static objects/scenes with fixed resolutions and are not applicable to articulated deformable hands. In this paper, we propose SRHand (Super-Resolution Hand), the method for reconstructing detailed 3D geometry as well as textured images of hands from low-resolution images. SRHand leverages the advantages of implicit image representation with explicit hand meshes. Specifically, we introduce a geometric-aware implicit image function (GIIF) that learns detailed hand prior by upsampling the coarse input images. By jointly optimizing the implicit image function and explicit 3D hand shapes, our method preserves multi-view and pose consistency among upsampled hand images, and achieves fine-detailed 3D reconstruction (wrinkles, nails). In experiments using the InterHand2.6M and Goliath datasets, our method significantly outperforms state-of-the-art image upsampling methods adapted to hand datasets, and 3D hand reconstruction methods, quantitatively and qualitatively. Project page: https://yunminjin2.github.io/projects/srhand", "date": "2025-09-26", "categories": []}
{"id": "2509.19939", "title": "AJAHR: Amputated Joint Aware 3D Human Mesh Recovery", "abstract": "Existing human mesh recovery methods assume a standard human body structure, overlooking diverse anatomical conditions such as limb loss. This assumption introduces bias when applied to individuals with amputations - a limitation further exacerbated by the scarcity of suitable datasets. To address this gap, we propose Amputated Joint Aware 3D Human Mesh Recovery (AJAHR), which is an adaptive pose estimation framework that improves mesh reconstruction for individuals with limb loss. Our model integrates a body-part amputation classifier, jointly trained with the mesh recovery network, to detect potential amputations. We also introduce Amputee 3D (A3D), which is a synthetic dataset offering a wide range of amputee poses for robust training. While maintaining competitive performance on non-amputees, our approach achieves state-of-the-art results for amputated individuals. Additional materials can be found at the project webpage.", "date": "2025-09-24", "categories": []}
{"id": "2509.12197", "title": "3D Human Pose and Shape Estimation from LiDAR Point Clouds: A Review", "abstract": "In this paper, we present a comprehensive review of 3D human pose estimation and human mesh recovery from in-the-wild LiDAR point clouds. We compare existing approaches across several key dimensions, and propose a structured taxonomy to classify these methods. Following this taxonomy, we analyze each method's strengths, limitations, and design choices. In addition, (i) we perform a quantitative comparison of the three most widely used datasets, detailing their characteristics; (ii) we compile unified definitions of all evaluation metrics; and (iii) we establish benchmark tables for both tasks on these datasets to enable fair comparisons and promote progress in the field. We also outline open challenges and research directions critical for advancing LiDAR-based 3D human understanding. Moreover, we maintain an accompanying webpage that organizes papers according to our taxonomy and continuously update it with new studies: https://github.com/valeoai/3D-Human-Pose-Shape-Estimation-from-LiDAR", "date": "2025-09-15", "categories": []}
{"id": "2509.18566", "title": "Event-guided 3D Gaussian Splatting for Dynamic Human and Scene Reconstruction", "abstract": "Reconstructing dynamic humans together with static scenes from monocular videos remains difficult, especially under fast motion, where RGB frames suffer from motion blur. Event cameras exhibit distinct advantages, e.g., microsecond temporal resolution, making them a superior sensing choice for dynamic human reconstruction. Accordingly, we present a novel event-guided human-scene reconstruction framework that jointly models human and scene from a single monocular event camera via 3D Gaussian Splatting. Specifically, a unified set of 3D Gaussians carries a learnable semantic attribute; only Gaussians classified as human undergo deformation for animation, while scene Gaussians stay static. To combat blur, we propose an event-guided loss that matches simulated brightness changes between consecutive renderings with the event stream, improving local fidelity in fast-moving regions. Our approach removes the need for external human masks and simplifies managing separate Gaussian sets. On two benchmark datasets, ZJU-MoCap-Blur and MMHPSD-Blur, it delivers state-of-the-art human-scene reconstruction, with notable gains over strong baselines in PSNR/SSIM and reduced LPIPS, especially for high-speed subjects.", "date": "2025-09-23", "categories": []}
{"id": "2509.18924", "title": "Audio-Driven Universal Gaussian Head Avatars", "abstract": "We introduce the first method for audio-driven universal photorealistic avatar synthesis, combining a person-agnostic speech model with our novel Universal Head Avatar Prior (UHAP). UHAP is trained on cross-identity multi-view videos. In particular, our UHAP is supervised with neutral scan data, enabling it to capture the identity-specific details at high fidelity. In contrast to previous approaches, which predominantly map audio features to geometric deformations only while ignoring audio-dependent appearance variations, our universal speech model directly maps raw audio inputs into the UHAP latent expression space. This expression space inherently encodes, both, geometric and appearance variations. For efficient personalization to new subjects, we employ a monocular encoder, which enables lightweight regression of dynamic expression variations across video frames. By accounting for these expression-dependent changes, it enables the subsequent model fine-tuning stage to focus exclusively on capturing the subject's global appearance and geometry. Decoding these audio-driven expression codes via UHAP generates highly realistic avatars with precise lip synchronization and nuanced expressive details, such as eyebrow movement, gaze shifts, and realistic mouth interior appearance as well as motion. Extensive evaluations demonstrate that our method is not only the first generalizable audio-driven avatar model that can account for detailed appearance modeling and rendering, but it also outperforms competing (geometry-only) methods across metrics measuring lip-sync accuracy, quantitative image quality, and perceptual realism.", "date": "2025-09-23", "categories": []}
{"id": "2509.14739", "title": "FMGS-Avatar: Mesh-Guided 2D Gaussian Splatting with Foundation Model Priors for 3D Monocular Avatar Reconstruction", "abstract": "Reconstructing high-fidelity animatable human avatars from monocular videos remains challenging due to insufficient geometric information in single-view observations. While recent 3D Gaussian Splatting methods have shown promise, they struggle with surface detail preservation due to the free-form nature of 3D Gaussian primitives. To address both the representation limitations and information scarcity, we propose a novel method, \\textbf{FMGS-Avatar}, that integrates two key innovations. First, we introduce Mesh-Guided 2D Gaussian Splatting, where 2D Gaussian primitives are attached directly to template mesh faces with constrained position, rotation, and movement, enabling superior surface alignment and geometric detail preservation. Second, we leverage foundation models trained on large-scale datasets, such as Sapiens, to complement the limited visual cues from monocular videos. However, when distilling multi-modal prior knowledge from foundation models, conflicting optimization objectives can emerge as different modalities exhibit distinct parameter sensitivities. We address this through a coordinated training strategy with selective gradient isolation, enabling each loss component to optimize its relevant parameters without interference. Through this combination of enhanced representation and coordinated information distillation, our approach significantly advances 3D monocular human avatar reconstruction. Experimental evaluation demonstrates superior reconstruction quality compared to existing methods, with notable gains in geometric accuracy and appearance fidelity while providing rich semantic information. Additionally, the distilled prior knowledge within a shared canonical space naturally enables spatially and temporally consistent rendering under novel views and poses.", "date": "2025-09-18", "categories": []}
{"id": "2509.05582", "title": "Reconstruction and Reenactment Separated Method for Realistic Gaussian Head", "abstract": "In this paper, we explore a reconstruction and reenactment separated framework for 3D Gaussians head, which requires only a single portrait image as input to generate controllable avatar. Specifically, we developed a large-scale one-shot gaussian head generator built upon WebSSL and employed a two-stage training approach that significantly enhances the capabilities of generalization and high-frequency texture reconstruction. During inference, an ultra-lightweight gaussian avatar driven by control signals enables high frame-rate rendering, achieving 90 FPS at a resolution of 512x512. We further demonstrate that the proposed framework follows the scaling law, whereby increasing the parameter scale of the reconstruction module leads to improved performance. Moreover, thanks to the separation design, driving efficiency remains unaffected. Finally, extensive quantitative and qualitative experiments validate that our approach outperforms current state-of-the-art methods.", "date": "2025-09-06", "categories": []}
{"id": "2509.11763", "title": "MSMA: Multi-Scale Feature Fusion For Multi-Attribute 3D Face Reconstruction From Unconstrained Images", "abstract": "Reconstructing 3D face from a single unconstrained image remains a challenging problem due to diverse conditions in unconstrained environments. Recently, learning-based methods have achieved notable results by effectively capturing complex facial structures and details across varying conditions. Consequently, many existing approaches employ projection-based losses between generated and input images to constrain model training. However, learning-based methods for 3D face reconstruction typically require substantial amounts of 3D facial data, which is difficult and costly to obtain. Consequently, to reduce reliance on labeled 3D face datasets, many existing approaches employ projection-based losses between generated and input images to constrain model training. Nonetheless, despite these advancements, existing approaches frequently struggle to capture detailed and multi-scale features under diverse facial attributes and conditions, leading to incomplete or less accurate reconstructions. In this paper, we propose a Multi-Scale Feature Fusion with Multi-Attribute (MSMA) framework for 3D face reconstruction from unconstrained images. Our method integrates multi-scale feature fusion with a focus on multi-attribute learning and leverages a large-kernel attention module to enhance the precision of feature extraction across scales, enabling accurate 3D facial parameter estimation from a single 2D image. Comprehensive experiments on the MICC Florence, Facewarehouse and custom-collect datasets demonstrate that our approach achieves results on par with current state-of-the-art methods, and in some instances, surpasses SOTA performance across challenging conditions.", "date": "2025-09-15", "categories": []}
{"id": "2509.13013", "title": "Dream3DAvatar: Text-Controlled 3D Avatar Reconstruction from a Single Image", "abstract": "With the rapid advancement of 3D representation techniques and generative models, substantial progress has been made in reconstructing full-body 3D avatars from a single image. However, this task remains fundamentally ill-posedness due to the limited information available from monocular input, making it difficult to control the geometry and texture of occluded regions during generation. To address these challenges, we redesign the reconstruction pipeline and propose Dream3DAvatar, an efficient and text-controllable two-stage framework for 3D avatar generation. In the first stage, we develop a lightweight, adapter-enhanced multi-view generation model. Specifically, we introduce the Pose-Adapter to inject SMPL-X renderings and skeletal information into SDXL, enforcing geometric and pose consistency across views. To preserve facial identity, we incorporate ID-Adapter-G, which injects high-resolution facial features into the generation process. Additionally, we leverage BLIP2 to generate high-quality textual descriptions of the multi-view images, enhancing text-driven controllability in occluded regions. In the second stage, we design a feedforward Transformer model equipped with a multi-view feature fusion module to reconstruct high-fidelity 3D Gaussian Splat representations (3DGS) from the generated images. Furthermore, we introduce ID-Adapter-R, which utilizes a gating mechanism to effectively fuse facial features into the reconstruction process, improving high-frequency detail recovery. Extensive experiments demonstrate that our method can generate realistic, animation-ready 3D avatars without any post-processing and consistently outperforms existing baselines across multiple evaluation metrics.", "date": "2025-09-16", "categories": []}
{"id": "2502.20220", "title": "Avat3r: Large Animatable Gaussian Reconstruction Model for High-fidelity 3D Head Avatars", "abstract": "Traditionally, creating photo-realistic 3D head avatars requires a studio-level multi-view capture setup and expensive optimization during test-time, limiting the use of digital human doubles to the VFX industry or offline renderings.\n  To address this shortcoming, we present Avat3r, which regresses a high-quality and animatable 3D head avatar from just a few input images, vastly reducing compute requirements during inference. More specifically, we make Large Reconstruction Models animatable and learn a powerful prior over 3D human heads from a large multi-view video dataset. For better 3D head reconstructions, we employ position maps from DUSt3R and generalized feature maps from the human foundation model Sapiens. To animate the 3D head, our key discovery is that simple cross-attention to an expression code is already sufficient. Finally, we increase robustness by feeding input images with different expressions to our model during training, enabling the reconstruction of 3D head avatars from inconsistent inputs, e.g., an imperfect phone capture with accidental movement, or frames from a monocular video.\n  We compare Avat3r with current state-of-the-art methods for few-input and single-input scenarios, and find that our method has a competitive advantage in both tasks. Finally, we demonstrate the wide applicability of our proposed model, creating 3D head avatars from images of different sources, smartphone captures, single images, and even out-of-domain inputs like antique busts.\n  Project website: https://tobias-kirschstein.github.io/avat3r/", "date": "2025-02-27", "categories": []}
{"id": "2507.08137", "title": "Occlusion-Aware Temporally Consistent Amodal Completion for 3D Human-Object Interaction Reconstruction", "abstract": "We introduce a novel framework for reconstructing dynamic human-object interactions from monocular video that overcomes challenges associated with occlusions and temporal inconsistencies. Traditional 3D reconstruction methods typically assume static objects or full visibility of dynamic subjects, leading to degraded performance when these assumptions are violated-particularly in scenarios where mutual occlusions occur. To address this, our framework leverages amodal completion to infer the complete structure of partially obscured regions. Unlike conventional approaches that operate on individual frames, our method integrates temporal context, enforcing coherence across video sequences to incrementally refine and stabilize reconstructions. This template-free strategy adapts to varying conditions without relying on predefined models, significantly enhancing the recovery of intricate details in dynamic scenes. We validate our approach using 3D Gaussian Splatting on challenging monocular videos, demonstrating superior precision in handling occlusions and maintaining temporal stability compared to existing techniques.", "date": "2025-07-10", "categories": []}
{"id": "2509.11411", "title": "On the Skinning of Gaussian Avatars", "abstract": "Radiance field-based methods have recently been used to reconstruct human avatars, showing that we can significantly downscale the systems needed for creating animated human avatars. Although this progress has been initiated by neural radiance fields, their slow rendering and backward mapping from the observation space to the canonical space have been the main challenges. With Gaussian splatting overcoming both challenges, a new family of approaches has emerged that are faster to train and render, while also straightforward to implement using forward skinning from the canonical to the observation space. However, the linear blend skinning required for the deformation of the Gaussians does not provide valid results for their non-linear rotation properties. To address such artifacts, recent works use mesh properties to rotate the non-linear Gaussian properties or train models to predict corrective offsets. Instead, we propose a weighted rotation blending approach that leverages quaternion averaging. This leads to simpler vertex-based Gaussians that can be efficiently animated and integrated in any engine by only modifying the linear blend skinning technique, and using any Gaussian rasterizer.", "date": "2025-09-14", "categories": []}
{"id": "2509.07774", "title": "HairGS: Hair Strand Reconstruction based on 3D Gaussian Splatting", "abstract": "Human hair reconstruction is a challenging problem in computer vision, with growing importance for applications in virtual reality and digital human modeling. Recent advances in 3D Gaussians Splatting (3DGS) provide efficient and explicit scene representations that naturally align with the structure of hair strands. In this work, we extend the 3DGS framework to enable strand-level hair geometry reconstruction from multi-view images. Our multi-stage pipeline first reconstructs detailed hair geometry using a differentiable Gaussian rasterizer, then merges individual Gaussian segments into coherent strands through a novel merging scheme, and finally refines and grows the strands under photometric supervision.\n  While existing methods typically evaluate reconstruction quality at the geometric level, they often neglect the connectivity and topology of hair strands. To address this, we propose a new evaluation metric that serves as a proxy for assessing topological accuracy in strand reconstruction. Extensive experiments on both synthetic and real-world datasets demonstrate that our method robustly handles a wide range of hairstyles and achieves efficient reconstruction, typically completing within one hour.\n  The project page can be found at: https://yimin-pan.github.io/hair-gs/", "date": "2025-09-09", "categories": []}
{"id": "2412.08511", "title": "NF3DM: Combining Neural Fields and Deformation Models for 3D Non-Rigid Motion Reconstruction", "abstract": "We introduce a novel, data-driven approach for reconstructing temporally coherent 3D motion from unstructured and potentially partial observations of non-rigidly deforming shapes. Our goal is to achieve high-fidelity motion reconstructions for shapes that undergo near-isometric deformations, such as humans wearing loose clothing. The key novelty of our work lies in its ability to combine implicit shape representations with explicit mesh-based deformation models, enabling detailed and temporally coherent motion reconstructions without relying on parametric shape models or decoupling shape and motion. Each frame is represented as a neural field decoded from a feature space where observations over time are fused, hence preserving geometric details present in the input data. Temporal coherence is enforced with a near-isometric deformation constraint between adjacent frames that applies to the underlying surface in the neural field. Our method outperforms state-of-the-art approaches, as demonstrated by its application to human and animal motion sequences reconstructed from monocular depth videos.", "date": "2024-12-11", "categories": []}
{"id": "2509.04145", "title": "Hyper Diffusion Avatars: Dynamic Human Avatar Generation using Network Weight Space Diffusion", "abstract": "Creating human avatars is a highly desirable yet challenging task. Recent advancements in radiance field rendering have achieved unprecedented photorealism and real-time performance for personalized dynamic human avatars. However, these approaches are typically limited to person-specific rendering models trained on multi-view video data for a single individual, limiting their ability to generalize across different identities. On the other hand, generative approaches leveraging prior knowledge from pre-trained 2D diffusion models can produce cartoonish, static human avatars, which are animated through simple skeleton-based articulation. Therefore, the avatars generated by these methods suffer from lower rendering quality compared to person-specific rendering methods and fail to capture pose-dependent deformations such as cloth wrinkles. In this paper, we propose a novel approach that unites the strengths of person-specific rendering and diffusion-based generative modeling to enable dynamic human avatar generation with both high photorealism and realistic pose-dependent deformations. Our method follows a two-stage pipeline: first, we optimize a set of person-specific UNets, with each network representing a dynamic human avatar that captures intricate pose-dependent deformations. In the second stage, we train a hyper diffusion model over the optimized network weights. During inference, our method generates network weights for real-time, controllable rendering of dynamic human avatars. Using a large-scale, cross-identity, multi-view video dataset, we demonstrate that our approach outperforms state-of-the-art human avatar generation methods.", "date": "2025-09-04", "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": "2509.02466", "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": "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": "2412.12765", "title": "Monocular Facial Appearance Capture in the Wild", "abstract": "We present a new method for reconstructing the appearance properties of human faces from a lightweight capture procedure in an unconstrained environment. Our method recovers the surface geometry, diffuse albedo, specular intensity and specular roughness from a monocular video containing a simple head rotation in-the-wild. Notably, we make no simplifying assumptions on the environment lighting, and we explicitly take visibility and occlusions into account. As a result, our method can produce facial appearance maps that approach the fidelity of studio-based multi-view captures, but with a far easier and cheaper procedure.", "date": "2024-12-17", "categories": []}
{"id": "2509.00403", "title": "DevilSight: Augmenting Monocular Human Avatar Reconstruction through a Virtual Perspective", "abstract": "We present a novel framework to reconstruct human avatars from monocular videos. Recent approaches have struggled either to capture the fine-grained dynamic details from the input or to generate plausible details at novel viewpoints, which mainly stem from the limited representational capacity of the avatar model and insufficient observational data. To overcome these challenges, we propose to leverage the advanced video generative model, Human4DiT, to generate the human motions from alternative perspective as an additional supervision signal. This approach not only enriches the details in previously unseen regions but also effectively regularizes the avatar representation to mitigate artifacts. Furthermore, we introduce two complementary strategies to enhance video generation: To ensure consistent reproduction of human motion, we inject the physical identity into the model through video fine-tuning. For higher-resolution outputs with finer details, a patch-based denoising algorithm is employed. Experimental results demonstrate that our method outperforms recent state-of-the-art approaches and validate the effectiveness of our proposed strategies.", "date": "2025-08-30", "categories": []}
{"id": "2509.01469", "title": "Im2Haircut: Single-view Strand-based Hair Reconstruction for Human Avatars", "abstract": "We present a novel approach for 3D hair reconstruction from single photographs based on a global hair prior combined with local optimization. Capturing strand-based hair geometry from single photographs is challenging due to the variety and geometric complexity of hairstyles and the lack of ground truth training data. Classical reconstruction methods like multi-view stereo only reconstruct the visible hair strands, missing the inner structure of hairstyles and hampering realistic hair simulation. To address this, existing methods leverage hairstyle priors trained on synthetic data. Such data, however, is limited in both quantity and quality since it requires manual work from skilled artists to model the 3D hairstyles and create near-photorealistic renderings. To address this, we propose a novel approach that uses both, real and synthetic data to learn an effective hairstyle prior. Specifically, we train a transformer-based prior model on synthetic data to obtain knowledge of the internal hairstyle geometry and introduce real data in the learning process to model the outer structure. This training scheme is able to model the visible hair strands depicted in an input image, while preserving the general 3D structure of hairstyles. We exploit this prior to create a Gaussian-splatting-based reconstruction method that creates hairstyles from one or more images. Qualitative and quantitative comparisons with existing reconstruction pipelines demonstrate the effectiveness and superior performance of our method for capturing detailed hair orientation, overall silhouette, and backside consistency. For additional results and code, please refer to https://im2haircut.is.tue.mpg.de.", "date": "2025-09-01", "categories": []}
{"id": "2509.01681", "title": "GaussianGAN: Real-Time Photorealistic controllable Human Avatars", "abstract": "Photorealistic and controllable human avatars have gained popularity in the research community thanks to rapid advances in neural rendering, providing fast and realistic synthesis tools. However, a limitation of current solutions is the presence of noticeable blurring. To solve this problem, we propose GaussianGAN, an animatable avatar approach developed for photorealistic rendering of people in real-time. We introduce a novel Gaussian splatting densification strategy to build Gaussian points from the surface of cylindrical structures around estimated skeletal limbs. Given the camera calibration, we render an accurate semantic segmentation with our novel view segmentation module. Finally, a UNet generator uses the rendered Gaussian splatting features and the segmentation maps to create photorealistic digital avatars. Our method runs in real-time with a rendering speed of 79 FPS. It outperforms previous methods regarding visual perception and quality, achieving a state-of-the-art results in terms of a pixel fidelity of 32.94db on the ZJU Mocap dataset and 33.39db on the Thuman4 dataset.", "date": "2025-09-01", "categories": []}
{"id": "2509.02141", "title": "GRMM: Real-Time High-Fidelity Gaussian Morphable Head Model with Learned Residuals", "abstract": "3D Morphable Models (3DMMs) enable controllable facial geometry and expression editing for reconstruction, animation, and AR/VR, but traditional PCA-based mesh models are limited in resolution, detail, and photorealism. Neural volumetric methods improve realism but remain too slow for interactive use. Recent Gaussian Splatting (3DGS) based facial models achieve fast, high-quality rendering but still depend solely on a mesh-based 3DMM prior for expression control, limiting their ability to capture fine-grained geometry, expressions, and full-head coverage. We introduce GRMM, the first full-head Gaussian 3D morphable model that augments a base 3DMM with residual geometry and appearance components, additive refinements that recover high-frequency details such as wrinkles, fine skin texture, and hairline variations. GRMM provides disentangled control through low-dimensional, interpretable parameters (e.g., identity shape, facial expressions) while separately modelling residuals that capture subject- and expression-specific detail beyond the base model's capacity. Coarse decoders produce vertex-level mesh deformations, fine decoders represent per-Gaussian appearance, and a lightweight CNN refines rasterised images for enhanced realism, all while maintaining 75 FPS real-time rendering. To learn consistent, high-fidelity residuals, we present EXPRESS-50, the first dataset with 60 aligned expressions across 50 identities, enabling robust disentanglement of identity and expression in Gaussian-based 3DMMs. Across monocular 3D face reconstruction, novel-view synthesis, and expression transfer, GRMM surpasses state-of-the-art methods in fidelity and expression accuracy while delivering interactive real-time performance.", "date": "2025-09-02", "categories": []}
{"id": "2508.20623", "title": "AvatarBack: Back-Head Generation for Complete 3D Avatars from Front-View Images", "abstract": "Recent advances in Gaussian Splatting have significantly boosted the reconstruction of head avatars, enabling high-quality facial modeling by representing an 3D avatar as a collection of 3D Gaussians. However, existing methods predominantly rely on frontal-view images, leaving the back-head poorly constructed. This leads to geometric inconsistencies, structural blurring, and reduced realism in the rear regions, ultimately limiting the fidelity of reconstructed avatars. To address this challenge, we propose AvatarBack, a novel plug-and-play framework specifically designed to reconstruct complete and consistent 3D Gaussian avatars by explicitly modeling the missing back-head regions. AvatarBack integrates two core technical innovations,i.e., the Subject-specific Generator (SSG) and the Adaptive Spatial Alignment Strategy (ASA). The former leverages a generative prior to synthesize identity-consistent, plausible back-view pseudo-images from sparse frontal inputs, providing robust multi-view supervision. To achieve precise geometric alignment between these synthetic views and the 3D Gaussian representation, the later employs learnable transformation matrices optimized during training, effectively resolving inherent pose and coordinate discrepancies. Extensive experiments on NeRSemble and K-hairstyle datasets, evaluated using geometric, photometric, and GPT-4o-based perceptual metrics, demonstrate that AvatarBack significantly enhances back-head reconstruction quality while preserving frontal fidelity. Moreover, the reconstructed avatars maintain consistent visual realism under diverse motions and remain fully animatable.", "date": "2025-08-28", "categories": []}
{"id": "2508.19688", "title": "SAT: Supervisor Regularization and Animation Augmentation for Two-process Monocular Texture 3D Human Reconstruction", "abstract": "Monocular texture 3D human reconstruction aims to create a complete 3D digital avatar from just a single front-view human RGB image. However, the geometric ambiguity inherent in a single 2D image and the scarcity of 3D human training data are the main obstacles limiting progress in this field. To address these issues, current methods employ prior geometric estimation networks to derive various human geometric forms, such as the SMPL model and normal maps. However, they struggle to integrate these modalities effectively, leading to view inconsistencies, such as facial distortions. To this end, we propose a two-process 3D human reconstruction framework, SAT, which seamlessly learns various prior geometries in a unified manner and reconstructs high-quality textured 3D avatars as the final output. To further facilitate geometry learning, we introduce a Supervisor Feature Regularization module. By employing a multi-view network with the same structure to provide intermediate features as training supervision, these varied geometric priors can be better fused. To tackle data scarcity and further improve reconstruction quality, we also propose an Online Animation Augmentation module. By building a one-feed-forward animation network, we augment a massive number of samples from the original 3D human data online for model training. Extensive experiments on two benchmarks show the superiority of our approach compared to state-of-the-art methods.", "date": "2025-08-27", "categories": []}
{"id": "2508.18944", "title": "PanoHair: Detailed Hair Strand Synthesis on Volumetric Heads", "abstract": "Achieving realistic hair strand synthesis is essential for creating lifelike digital humans, but producing high-fidelity hair strand geometry remains a significant challenge. Existing methods require a complex setup for data acquisition, involving multi-view images captured in constrained studio environments. Additionally, these methods have longer hair volume estimation and strand synthesis times, which hinder efficiency. We introduce PanoHair, a model that estimates head geometry as signed distance fields using knowledge distillation from a pre-trained generative teacher model for head synthesis. Our approach enables the prediction of semantic segmentation masks and 3D orientations specifically for the hair region of the estimated geometry. Our method is generative and can generate diverse hairstyles with latent space manipulations. For real images, our approach involves an inversion process to infer latent codes and produces visually appealing hair strands, offering a streamlined alternative to complex multi-view data acquisition setups. Given the latent code, PanoHair generates a clean manifold mesh for the hair region in under 5 seconds, along with semantic and orientation maps, marking a significant improvement over existing methods, as demonstrated in our experiments.", "date": "2025-08-26", "categories": []}
{"id": "2508.15500", "title": "MExECON: Multi-view Extended Explicit Clothed humans Optimized via Normal integration", "abstract": "This work presents MExECON, a novel pipeline for 3D reconstruction of clothed human avatars from sparse multi-view RGB images. Building on the single-view method ECON, MExECON extends its capabilities to leverage multiple viewpoints, improving geometry and body pose estimation. At the core of the pipeline is the proposed Joint Multi-view Body Optimization (JMBO) algorithm, which fits a single SMPL-X body model jointly across all input views, enforcing multi-view consistency. The optimized body model serves as a low-frequency prior that guides the subsequent surface reconstruction, where geometric details are added via normal map integration. MExECON integrates normal maps from both front and back views to accurately capture fine-grained surface details such as clothing folds and hairstyles. All multi-view gains are achieved without requiring any network re-training. Experimental results show that MExECON consistently improves fidelity over the single-view baseline and achieves competitive performance compared to modern few-shot 3D reconstruction methods.", "date": "2025-08-21", "categories": []}
{"id": "2508.15767", "title": "ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human Modeling", "abstract": "Parametric body models offer expressive 3D representation of humans across a wide range of poses, shapes, and facial expressions, typically derived by learning a basis over registered 3D meshes. However, existing human mesh modeling approaches struggle to capture detailed variations across diverse body poses and shapes, largely due to limited training data diversity and restrictive modeling assumptions. Moreover, the common paradigm first optimizes the external body surface using a linear basis, then regresses internal skeletal joints from surface vertices. This approach introduces problematic dependencies between internal skeleton and outer soft tissue, limiting direct control over body height and bone lengths. To address these issues, we present ATLAS, a high-fidelity body model learned from 600k high-resolution scans captured using 240 synchronized cameras. Unlike previous methods, we explicitly decouple the shape and skeleton bases by grounding our mesh representation in the human skeleton. This decoupling enables enhanced shape expressivity, fine-grained customization of body attributes, and keypoint fitting independent of external soft-tissue characteristics. ATLAS outperforms existing methods by fitting unseen subjects in diverse poses more accurately, and quantitative evaluations show that our non-linear pose correctives more effectively capture complex poses compared to linear models.", "date": "2025-08-21", "categories": []}
{"id": "2508.14892", "title": "Snap-Snap: Taking Two Images to Reconstruct 3D Human Gaussians in Milliseconds", "abstract": "Reconstructing 3D human bodies from sparse views has been an appealing topic, which is crucial to broader the related applications. In this paper, we propose a quite challenging but valuable task to reconstruct the human body from only two images, i.e., the front and back view, which can largely lower the barrier for users to create their own 3D digital humans. The main challenges lie in the difficulty of building 3D consistency and recovering missing information from the highly sparse input. We redesign a geometry reconstruction model based on foundation reconstruction models to predict consistent point clouds even input images have scarce overlaps with extensive human data training. Furthermore, an enhancement algorithm is applied to supplement the missing color information, and then the complete human point clouds with colors can be obtained, which are directly transformed into 3D Gaussians for better rendering quality. Experiments show that our method can reconstruct the entire human in 190 ms on a single NVIDIA RTX 4090, with two images at a resolution of 1024x1024, demonstrating state-of-the-art performance on the THuman2.0 and cross-domain datasets. Additionally, our method can complete human reconstruction even with images captured by low-cost mobile devices, reducing the requirements for data collection. Demos and code are available at https://hustvl.github.io/Snap-Snap/.", "date": "2025-08-20", "categories": []}
{"id": "2303.13397", "title": "DiffMesh: A Motion-aware Diffusion Framework for Human Mesh Recovery from Videos", "abstract": "Human mesh recovery (HMR) provides rich human body information for various real-world applications. While image-based HMR methods have achieved impressive results, they often struggle to recover humans in dynamic scenarios, leading to temporal inconsistencies and non-smooth 3D motion predictions due to the absence of human motion. In contrast, video-based approaches leverage temporal information to mitigate this issue. In this paper, we present DiffMesh, an innovative motion-aware Diffusion-like framework for video-based HMR. DiffMesh establishes a bridge between diffusion models and human motion, efficiently generating accurate and smooth output mesh sequences by incorporating human motion within the forward process and reverse process in the diffusion model. Extensive experiments are conducted on the widely used datasets (Human3.6M \\cite{h36m_pami} and 3DPW \\cite{pw3d2018}), which demonstrate the effectiveness and efficiency of our DiffMesh. Visual comparisons in real-world scenarios further highlight DiffMesh's suitability for practical applications.", "date": "2023-03-23", "categories": []}
{"id": "2504.10486", "title": "DNF-Avatar: Distilling Neural Fields for Real-time Animatable Avatar Relighting", "abstract": "Creating relightable and animatable human avatars from monocular videos is a rising research topic with a range of applications, e.g. virtual reality, sports, and video games. Previous works utilize neural fields together with physically based rendering (PBR), to estimate geometry and disentangle appearance properties of human avatars. However, one drawback of these methods is the slow rendering speed due to the expensive Monte Carlo ray tracing. To tackle this problem, we proposed to distill the knowledge from implicit neural fields (teacher) to explicit 2D Gaussian splatting (student) representation to take advantage of the fast rasterization property of Gaussian splatting. To avoid ray-tracing, we employ the split-sum approximation for PBR appearance. We also propose novel part-wise ambient occlusion probes for shadow computation. Shadow prediction is achieved by querying these probes only once per pixel, which paves the way for real-time relighting of avatars. These techniques combined give high-quality relighting results with realistic shadow effects. Our experiments demonstrate that the proposed student model achieves comparable or even better relighting results with our teacher model while being 370 times faster at inference time, achieving a 67 FPS rendering speed.", "date": "2025-04-14", "categories": []}
{"id": "2508.13537", "title": "EAvatar: Expression-Aware Head Avatar Reconstruction with Generative Geometry Priors", "abstract": "High-fidelity head avatar reconstruction plays a crucial role in AR/VR, gaming, and multimedia content creation. Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated effectiveness in modeling complex geometry with real-time rendering capability and are now widely used in high-fidelity head avatar reconstruction tasks. However, existing 3DGS-based methods still face significant challenges in capturing fine-grained facial expressions and preserving local texture continuity, especially in highly deformable regions. To mitigate these limitations, we propose a novel 3DGS-based framework termed EAvatar for head reconstruction that is both expression-aware and deformation-aware. Our method introduces a sparse expression control mechanism, where a small number of key Gaussians are used to influence the deformation of their neighboring Gaussians, enabling accurate modeling of local deformations and fine-scale texture transitions. Furthermore, we leverage high-quality 3D priors from pretrained generative models to provide a more reliable facial geometry, offering structural guidance that improves convergence stability and shape accuracy during training. Experimental results demonstrate that our method produces more accurate and visually coherent head reconstructions with improved expression controllability and detail fidelity.", "date": "2025-08-19", "categories": []}
{"id": "2508.09597", "title": "SVG-Head: Hybrid Surface-Volumetric Gaussians for High-Fidelity Head Reconstruction and Real-Time Editing", "abstract": "Creating high-fidelity and editable head avatars is a pivotal challenge in computer vision and graphics, boosting many AR/VR applications. While recent advancements have achieved photorealistic renderings and plausible animation, head editing, especially real-time appearance editing, remains challenging due to the implicit representation and entangled modeling of the geometry and global appearance. To address this, we propose Surface-Volumetric Gaussian Head Avatar (SVG-Head), a novel hybrid representation that explicitly models the geometry with 3D Gaussians bound on a FLAME mesh and leverages disentangled texture images to capture the global appearance. Technically, it contains two types of Gaussians, in which surface Gaussians explicitly model the appearance of head avatars using learnable texture images, facilitating real-time texture editing, while volumetric Gaussians enhance the reconstruction quality of non-Lambertian regions (e.g., lips and hair). To model the correspondence between 3D world and texture space, we provide a mesh-aware Gaussian UV mapping method, which leverages UV coordinates given by the FLAME mesh to obtain sharp texture images and real-time rendering speed. A hierarchical optimization strategy is further designed to pursue the optimal performance in both reconstruction quality and editing flexibility. Experiments on the NeRSemble dataset show that SVG-Head not only generates high-fidelity rendering results, but also is the first method to obtain explicit texture images for Gaussian head avatars and support real-time appearance editing.", "date": "2025-08-13", "categories": []}
{"id": "2508.09629", "title": "Enhancing Monocular 3D Hand Reconstruction with Learned Texture Priors", "abstract": "We revisit the role of texture in monocular 3D hand reconstruction, not as an afterthought for photorealism, but as a dense, spatially grounded cue that can actively support pose and shape estimation. Our observation is simple: even in high-performing models, the overlay between predicted hand geometry and image appearance is often imperfect, suggesting that texture alignment may be an underused supervisory signal. We propose a lightweight texture module that embeds per-pixel observations into UV texture space and enables a novel dense alignment loss between predicted and observed hand appearances. Our approach assumes access to a differentiable rendering pipeline and a model that maps images to 3D hand meshes with known topology, allowing us to back-project a textured hand onto the image and perform pixel-based alignment. The module is self-contained and easily pluggable into existing reconstruction pipelines. To isolate and highlight the value of texture-guided supervision, we augment HaMeR, a high-performing yet unadorned transformer architecture for 3D hand pose estimation. The resulting system improves both accuracy and realism, demonstrating the value of appearance-guided alignment in hand reconstruction.", "date": "2025-08-13", "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": "2508.08178", "title": "3D Human Mesh Estimation from Single View RGBD", "abstract": "Despite significant progress in 3D human mesh estimation from RGB images; RGBD cameras, offering additional depth data, remain underutilized. In this paper, we present a method for accurate 3D human mesh estimation from a single RGBD view, leveraging the affordability and widespread adoption of RGBD cameras for real-world applications. A fully supervised approach for this problem, requires a dataset with RGBD image and 3D mesh label pairs. However, collecting such a dataset is costly and challenging, hence, existing datasets are small, and limited in pose and shape diversity. To overcome this data scarcity, we leverage existing Motion Capture (MoCap) datasets. We first obtain complete 3D meshes from the body models found in MoCap datasets, and create partial, single-view versions of them by projection to a virtual camera. This simulates the depth data provided by an RGBD camera from a single viewpoint. Then, we train a masked autoencoder to complete the partial, single-view mesh. During inference, our method, which we name as M$^3$ for ``Masked Mesh Modeling'', matches the depth values coming from the sensor to vertices of a template human mesh, which creates a partial, single-view mesh. We effectively recover parts of the 3D human body mesh model that are not visible, resulting in a full body mesh. M$^3$ achieves 16.8 mm and 22.0 mm per-vertex-error (PVE) on the SURREAL and CAPE datasets, respectively; outperforming existing methods that use full-body point clouds as input. We obtain a competitive 70.9 PVE on the BEHAVE dataset, outperforming a recently published RGB based method by 18.4 mm, highlighting the usefulness of depth data. Code will be released.", "date": "2025-08-11", "categories": []}
{"id": "2507.23597", "title": "MoGA: 3D Generative Avatar Prior for Monocular Gaussian Avatar Reconstruction", "abstract": "We present MoGA, a novel method to reconstruct high-fidelity 3D Gaussian avatars from a single-view image. The main challenge lies in inferring unseen appearance and geometric details while ensuring 3D consistency and realism. Most previous methods rely on 2D diffusion models to synthesize unseen views; however, these generated views are sparse and inconsistent, resulting in unrealistic 3D artifacts and blurred appearance. To address these limitations, we leverage a generative avatar model, that can generate diverse 3D avatars by sampling deformed Gaussians from a learned prior distribution. Due to limited 3D training data, such a 3D model alone cannot capture all image details of unseen identities. Consequently, we integrate it as a prior, ensuring 3D consistency by projecting input images into its latent space and enforcing additional 3D appearance and geometric constraints. Our novel approach formulates Gaussian avatar creation as model inversion by fitting the generative avatar to synthetic views from 2D diffusion models. The generative avatar provides an initialization for model fitting, enforces 3D regularization, and helps in refining pose. Experiments show that our method surpasses state-of-the-art techniques and generalizes well to real-world scenarios. Our Gaussian avatars are also inherently animatable. For code, see https://zj-dong.github.io/MoGA/.", "date": "2025-07-31", "categories": []}
{"id": "2503.10624", "title": "ETCH: Generalizing Body Fitting to Clothed Humans via Equivariant Tightness", "abstract": "Fitting a body to a 3D clothed human point cloud is a common yet challenging task. Traditional optimization-based approaches use multi-stage pipelines that are sensitive to pose initialization, while recent learning-based methods often struggle with generalization across diverse poses and garment types. We propose Equivariant Tightness Fitting for Clothed Humans, or ETCH, a novel pipeline that estimates cloth-to-body surface mapping through locally approximate SE(3) equivariance, encoding tightness as displacement vectors from the cloth surface to the underlying body. Following this mapping, pose-invariant body features regress sparse body markers, simplifying clothed human fitting into an inner-body marker fitting task. Extensive experiments on CAPE and 4D-Dress show that ETCH significantly outperforms state-of-the-art methods -- both tightness-agnostic and tightness-aware -- in body fitting accuracy on loose clothing (16.7% ~ 69.5%) and shape accuracy (average 49.9%). Our equivariant tightness design can even reduce directional errors by (67.2% ~ 89.8%) in one-shot (or out-of-distribution) settings (~ 1% data). Qualitative results demonstrate strong generalization of ETCH, regardless of challenging poses, unseen shapes, loose clothing, and non-rigid dynamics. We will release the code and models soon for research purposes at https://boqian-li.github.io/ETCH/.", "date": "2025-03-13", "categories": []}
{"id": "2411.16768", "title": "Sequential Gaussian Avatars with Hierarchical Motion Context", "abstract": "The emergence of neural rendering has significantly advanced the rendering quality of 3D human avatars, with the recently popular 3DGS technique enabling real-time performance. However, SMPL-driven 3DGS human avatars still struggle to capture fine appearance details due to the complex mapping from pose to appearance during fitting. In this paper, we propose SeqAvatar, which excavates the explicit 3DGS representation to better model human avatars based on a hierarchical motion context. Specifically, we utilize a coarse-to-fine motion conditions that incorporate both the overall human skeleton and fine-grained vertex motions for non-rigid deformation. To enhance the robustness of the proposed motion conditions, we adopt a spatio-temporal multi-scale sampling strategy to hierarchically integrate more motion clues to model human avatars. Extensive experiments demonstrate that our method significantly outperforms 3DGS-based approaches and renders human avatars orders of magnitude faster than the latest NeRF-based models that incorporate temporal context, all while delivering performance that is at least comparable or even superior. Project page: https://zezeaaa.github.io/projects/SeqAvatar/", "date": "2024-11-25", "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": "2508.01218", "title": "MoGaFace: Momentum-Guided and Texture-Aware Gaussian Avatars for Consistent Facial Geometry", "abstract": "Existing 3D head avatar reconstruction methods adopt a two-stage process, relying on tracked FLAME meshes derived from facial landmarks, followed by Gaussian-based rendering. However, misalignment between the estimated mesh and target images often leads to suboptimal rendering quality and loss of fine visual details. In this paper, we present MoGaFace, a novel 3D head avatar modeling framework that continuously refines facial geometry and texture attributes throughout the Gaussian rendering process. To address the misalignment between estimated FLAME meshes and target images, we introduce the Momentum-Guided Consistent Geometry module, which incorporates a momentum-updated expression bank and an expression-aware correction mechanism to ensure temporal and multi-view consistency. Additionally, we propose Latent Texture Attention, which encodes compact multi-view features into head-aware representations, enabling geometry-aware texture refinement via integration into Gaussians. Extensive experiments show that MoGaFace achieves high-fidelity head avatar reconstruction and significantly improves novel-view synthesis quality, even under inaccurate mesh initialization and unconstrained real-world settings.", "date": "2025-08-02", "categories": []}
{"id": "2508.01381", "title": "ReMu: Reconstructing Multi-layer 3D Clothed Human from Image Layers", "abstract": "The reconstruction of multi-layer 3D garments typically requires expensive multi-view capture setups and specialized 3D editing efforts. To support the creation of life-like clothed human avatars, we introduce ReMu for reconstructing multi-layer clothed humans in a new setup, Image Layers, which captures a subject wearing different layers of clothing with a single RGB camera. To reconstruct physically plausible multi-layer 3D garments, a unified 3D representation is necessary to model these garments in a layered manner. Thus, we first reconstruct and align each garment layer in a shared coordinate system defined by the canonical body pose. Afterwards, we introduce a collision-aware optimization process to address interpenetration and further refine the garment boundaries leveraging implicit neural fields. It is worth noting that our method is template-free and category-agnostic, which enables the reconstruction of 3D garments in diverse clothing styles. Through our experiments, we show that our method reconstructs nearly penetration-free 3D clothed humans and achieves competitive performance compared to category-specific methods. Project page: https://eth-ait.github.io/ReMu/", "date": "2025-08-02", "categories": []}
{"id": "2412.17812", "title": "FaceLift: Learning Generalizable Single Image 3D Face Reconstruction from Synthetic Heads", "abstract": "We present FaceLift, a novel feed-forward approach for generalizable high-quality 360-degree 3D head reconstruction from a single image. Our pipeline first employs a multi-view latent diffusion model to generate consistent side and back views from a single facial input, which then feeds into a transformer-based reconstructor that produces a comprehensive 3D Gaussian splats representation. Previous methods for monocular 3D face reconstruction often lack full view coverage or view consistency due to insufficient multi-view supervision. We address this by creating a high-quality synthetic head dataset that enables consistent supervision across viewpoints. To bridge the domain gap between synthetic training data and real-world images, we propose a simple yet effective technique that ensures the view generation process maintains fidelity to the input by learning to reconstruct the input image alongside the view generation. Despite being trained exclusively on synthetic data, our method demonstrates remarkable generalization to real-world images. Through extensive qualitative and quantitative evaluations, we show that FaceLift outperforms state-of-the-art 3D face reconstruction methods on identity preservation, detail recovery, and rendering quality.", "date": "2024-12-23", "categories": []}
{"id": "2505.03351", "title": "GUAVA: Generalizable Upper Body 3D Gaussian Avatar", "abstract": "Reconstructing a high-quality, animatable 3D human avatar with expressive facial and hand motions from a single image has gained significant attention due to its broad application potential. 3D human avatar reconstruction typically requires multi-view or monocular videos and training on individual IDs, which is both complex and time-consuming. Furthermore, limited by SMPLX's expressiveness, these methods often focus on body motion but struggle with facial expressions. To address these challenges, we first introduce an expressive human model (EHM) to enhance facial expression capabilities and develop an accurate tracking method. Based on this template model, we propose GUAVA, the first framework for fast animatable upper-body 3D Gaussian avatar reconstruction. We leverage inverse texture mapping and projection sampling techniques to infer Ubody (upper-body) Gaussians from a single image. The rendered images are refined through a neural refiner. Experimental results demonstrate that GUAVA significantly outperforms previous methods in rendering quality and offers significant speed improvements, with reconstruction times in the sub-second range (0.1s), and supports real-time animation and rendering.", "date": "2025-05-06", "categories": []}
{"id": "2507.17332", "title": "PARTE: Part-Guided Texturing for 3D Human Reconstruction from a Single Image", "abstract": "The misaligned human texture across different human parts is one of the main limitations of existing 3D human reconstruction methods. Each human part, such as a jacket or pants, should maintain a distinct texture without blending into others. The structural coherence of human parts serves as a crucial cue to infer human textures in the invisible regions of a single image. However, most existing 3D human reconstruction methods do not explicitly exploit such part segmentation priors, leading to misaligned textures in their reconstructions. In this regard, we present PARTE, which utilizes 3D human part information as a key guide to reconstruct 3D human textures. Our framework comprises two core components. First, to infer 3D human part information from a single image, we propose a 3D part segmentation module (PartSegmenter) that initially reconstructs a textureless human surface and predicts human part labels based on the textureless surface. Second, to incorporate part information into texture reconstruction, we introduce a part-guided texturing module (PartTexturer), which acquires prior knowledge from a pre-trained image generation network on texture alignment of human parts. Extensive experiments demonstrate that our framework achieves state-of-the-art quality in 3D human reconstruction. The project page is available at https://hygenie1228.github.io/PARTE/.", "date": "2025-07-23", "categories": []}
{"id": "2507.21311", "title": "VoluMe -- Authentic 3D Video Calls from Live Gaussian Splat Prediction", "abstract": "Virtual 3D meetings offer the potential to enhance copresence, increase engagement and thus improve effectiveness of remote meetings compared to standard 2D video calls. However, representing people in 3D meetings remains a challenge; existing solutions achieve high quality by using complex hardware, making use of fixed appearance via enrolment, or by inverting a pre-trained generative model. These approaches lead to constraints that are unwelcome and ill-fitting for videoconferencing applications. We present the first method to predict 3D Gaussian reconstructions in real time from a single 2D webcam feed, where the 3D representation is not only live and realistic, but also authentic to the input video. By conditioning the 3D representation on each video frame independently, our reconstruction faithfully recreates the input video from the captured viewpoint (a property we call authenticity), while generalizing realistically to novel viewpoints. Additionally, we introduce a stability loss to obtain reconstructions that are temporally stable on video sequences. We show that our method delivers state-of-the-art accuracy in visual quality and stability metrics compared to existing methods, and demonstrate our approach in live one-to-one 3D meetings using only a standard 2D camera and display. This demonstrates that our approach can allow anyone to communicate volumetrically, via a method for 3D videoconferencing that is not only highly accessible, but also realistic and authentic.", "date": "2025-07-28", "categories": []}
{"id": "2507.18758", "title": "Learning Efficient and Generalizable Human Representation with Human Gaussian Model", "abstract": "Modeling animatable human avatars from videos is a long-standing and challenging problem. While conventional methods require per-instance optimization, recent feed-forward methods have been proposed to generate 3D Gaussians with a learnable network. However, these methods predict Gaussians for each frame independently, without fully capturing the relations of Gaussians from different timestamps. To address this, we propose Human Gaussian Graph to model the connection between predicted Gaussians and human SMPL mesh, so that we can leverage information from all frames to recover an animatable human representation. Specifically, the Human Gaussian Graph contains dual layers where Gaussians are the first layer nodes and mesh vertices serve as the second layer nodes. Based on this structure, we further propose the intra-node operation to aggregate various Gaussians connected to one mesh vertex, and inter-node operation to support message passing among mesh node neighbors. Experimental results on novel view synthesis and novel pose animation demonstrate the efficiency and generalization of our method.", "date": "2025-07-24", "categories": []}
{"id": "2507.18155", "title": "GeoAvatar: Adaptive Geometrical Gaussian Splatting for 3D Head Avatar", "abstract": "Despite recent progress in 3D head avatar generation, balancing identity preservation, i.e., reconstruction, with novel poses and expressions, i.e., animation, remains a challenge. Existing methods struggle to adapt Gaussians to varying geometrical deviations across facial regions, resulting in suboptimal quality. To address this, we propose GeoAvatar, a framework for adaptive geometrical Gaussian Splatting. GeoAvatar leverages Adaptive Pre-allocation Stage (APS), an unsupervised method that segments Gaussians into rigid and flexible sets for adaptive offset regularization. Then, based on mouth anatomy and dynamics, we introduce a novel mouth structure and the part-wise deformation strategy to enhance the animation fidelity of the mouth. Finally, we propose a regularization loss for precise rigging between Gaussians and 3DMM faces. Moreover, we release DynamicFace, a video dataset with highly expressive facial motions. Extensive experiments show the superiority of GeoAvatar compared to state-of-the-art methods in reconstruction and novel animation scenarios.", "date": "2025-07-24", "categories": []}
{"id": "2503.17032", "title": "TaoAvatar: Real-Time Lifelike Full-Body Talking Avatars for Augmented Reality via 3D Gaussian Splatting", "abstract": "Realistic 3D full-body talking avatars hold great potential in AR, with applications ranging from e-commerce live streaming to holographic communication. Despite advances in 3D Gaussian Splatting (3DGS) for lifelike avatar creation, existing methods struggle with fine-grained control of facial expressions and body movements in full-body talking tasks. Additionally, they often lack sufficient details and cannot run in real-time on mobile devices. We present TaoAvatar, a high-fidelity, lightweight, 3DGS-based full-body talking avatar driven by various signals. Our approach starts by creating a personalized clothed human parametric template that binds Gaussians to represent appearances. We then pre-train a StyleUnet-based network to handle complex pose-dependent non-rigid deformation, which can capture high-frequency appearance details but is too resource-intensive for mobile devices. To overcome this, we \"bake\" the non-rigid deformations into a lightweight MLP-based network using a distillation technique and develop blend shapes to compensate for details. Extensive experiments show that TaoAvatar achieves state-of-the-art rendering quality while running in real-time across various devices, maintaining 90 FPS on high-definition stereo devices such as the Apple Vision Pro.", "date": "2025-03-21", "categories": []}
{"id": "2507.17029", "title": "StreamME: Simplify 3D Gaussian Avatar within Live Stream", "abstract": "We propose StreamME, a method focuses on fast 3D avatar reconstruction. The StreamME synchronously records and reconstructs a head avatar from live video streams without any pre-cached data, enabling seamless integration of the reconstructed appearance into downstream applications. This exceptionally fast training strategy, which we refer to as on-the-fly training, is central to our approach. Our method is built upon 3D Gaussian Splatting (3DGS), eliminating the reliance on MLPs in deformable 3DGS and relying solely on geometry, which significantly improves the adaptation speed to facial expression. To further ensure high efficiency in on-the-fly training, we introduced a simplification strategy based on primary points, which distributes the point clouds more sparsely across the facial surface, optimizing points number while maintaining rendering quality. Leveraging the on-the-fly training capabilities, our method protects the facial privacy and reduces communication bandwidth in VR system or online conference. Additionally, it can be directly applied to downstream application such as animation, toonify, and relighting. Please refer to our project page for more details: https://songluchuan.github.io/StreamME/.", "date": "2025-07-22", "categories": []}
{"id": "2503.21313", "title": "HORT: Monocular Hand-held Objects Reconstruction with Transformers", "abstract": "Reconstructing hand-held objects in 3D from monocular images remains a significant challenge in computer vision. Most existing approaches rely on implicit 3D representations, which produce overly smooth reconstructions and are time-consuming to generate explicit 3D shapes. While more recent methods directly reconstruct point clouds with diffusion models, the multi-step denoising makes high-resolution reconstruction inefficient. To address these limitations, we propose a transformer-based model to efficiently reconstruct dense 3D point clouds of hand-held objects. Our method follows a coarse-to-fine strategy, first generating a sparse point cloud from the image and progressively refining it into a dense representation using pixel-aligned image features. To enhance reconstruction accuracy, we integrate image features with 3D hand geometry to jointly predict the object point cloud and its pose relative to the hand. Our model is trained end-to-end for optimal performance. Experimental results on both synthetic and real datasets demonstrate that our method achieves state-of-the-art accuracy with much faster inference speed, while generalizing well to in-the-wild images.", "date": "2025-03-27", "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": "2507.13648", "title": "EPSilon: Efficient Point Sampling for Lightening of Hybrid-based 3D Avatar Generation", "abstract": "The rapid advancement of neural radiance fields (NeRF) has paved the way to generate animatable human avatars from a monocular video. However, the sole usage of NeRF suffers from a lack of details, which results in the emergence of hybrid representation that utilizes SMPL-based mesh together with NeRF representation. While hybrid-based models show photo-realistic human avatar generation qualities, they suffer from extremely slow inference due to their deformation scheme: to be aligned with the mesh, hybrid-based models use the deformation based on SMPL skinning weights, which needs high computational costs on each sampled point. We observe that since most of the sampled points are located in empty space, they do not affect the generation quality but result in inference latency with deformation. In light of this observation, we propose EPSilon, a hybrid-based 3D avatar generation scheme with novel efficient point sampling strategies that boost both training and inference. In EPSilon, we propose two methods to omit empty points at rendering; empty ray omission (ERO) and empty interval omission (EIO). In ERO, we wipe out rays that progress through the empty space. Then, EIO narrows down the sampling interval on the ray, which wipes out the region not occupied by either clothes or mesh. The delicate sampling scheme of EPSilon enables not only great computational cost reduction during deformation but also the designation of the important regions to be sampled, which enables a single-stage NeRF structure without hierarchical sampling. Compared to existing methods, EPSilon maintains the generation quality while using only 3.9% of sampled points and achieves around 20 times faster inference, together with 4 times faster training convergence. We provide video results on https://github.com/seungjun-moon/epsilon.", "date": "2025-07-18", "categories": []}
{"id": "2507.08981", "title": "Video Inference for Human Mesh Recovery with Vision Transformer", "abstract": "Human Mesh Recovery (HMR) from an image is a challenging problem because of the inherent ambiguity of the task. Existing HMR methods utilized either temporal information or kinematic relationships to achieve higher accuracy, but there is no method using both. Hence, we propose \"Video Inference for Human Mesh Recovery with Vision Transformer (HMR-ViT)\" that can take into account both temporal and kinematic information. In HMR-ViT, a Temporal-kinematic Feature Image is constructed using feature vectors obtained from video frames by an image encoder. When generating the feature image, we use a Channel Rearranging Matrix (CRM) so that similar kinematic features could be located spatially close together. The feature image is then further encoded using Vision Transformer, and the SMPL pose and shape parameters are finally inferred using a regression network. Extensive evaluation on the 3DPW and Human3.6M datasets indicates that our method achieves a competitive performance in HMR.", "date": "2025-07-11", "categories": []}
{"id": "2507.10542", "title": "ScaffoldAvatar: High-Fidelity Gaussian Avatars with Patch Expressions", "abstract": "Generating high-fidelity real-time animated sequences of photorealistic 3D head avatars is important for many graphics applications, including immersive telepresence and movies. This is a challenging problem particularly when rendering digital avatar close-ups for showing character's facial microfeatures and expressions. To capture the expressive, detailed nature of human heads, including skin furrowing and finer-scale facial movements, we propose to couple locally-defined facial expressions with 3D Gaussian splatting to enable creating ultra-high fidelity, expressive and photorealistic 3D head avatars. In contrast to previous works that operate on a global expression space, we condition our avatar's dynamics on patch-based local expression features and synthesize 3D Gaussians at a patch level. In particular, we leverage a patch-based geometric 3D face model to extract patch expressions and learn how to translate these into local dynamic skin appearance and motion by coupling the patches with anchor points of Scaffold-GS, a recent hierarchical scene representation. These anchors are then used to synthesize 3D Gaussians on-the-fly, conditioned by patch-expressions and viewing direction. We employ color-based densification and progressive training to obtain high-quality results and faster convergence for high resolution 3K training images. By leveraging patch-level expressions, ScaffoldAvatar consistently achieves state-of-the-art performance with visually natural motion, while encompassing diverse facial expressions and styles in real time.", "date": "2025-07-14", "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": "2502.12080", "title": "HumanGif: Single-View Human Diffusion with Generative Prior", "abstract": "Previous 3D human creation methods have made significant progress in synthesizing view-consistent and temporally aligned results from sparse-view images or monocular videos. However, it remains challenging to produce perpetually realistic, view-consistent, and temporally coherent human avatars from a single image, as limited information is available in the single-view input setting. Motivated by the success of 2D character animation, we propose HumanGif, a single-view human diffusion model with generative prior. Specifically, we formulate the single-view-based 3D human novel view and pose synthesis as a single-view-conditioned human diffusion process, utilizing generative priors from foundational diffusion models to complement the missing information. To ensure fine-grained and consistent novel view and pose synthesis, we introduce a Human NeRF module in HumanGif to learn spatially aligned features from the input image, implicitly capturing the relative camera and human pose transformation. Furthermore, we introduce an image-level loss during optimization to bridge the gap between latent and image spaces in diffusion models. Extensive experiments on RenderPeople, DNA-Rendering, THuman 2.1, and TikTok datasets demonstrate that HumanGif achieves the best perceptual performance, with better generalizability for novel view and pose synthesis.", "date": "2025-02-17", "categories": []}
{"id": "2506.22044", "title": "Few-Shot Identity Adaptation for 3D Talking Heads via Global Gaussian Field", "abstract": "Reconstruction and rendering-based talking head synthesis methods achieve high-quality results with strong identity preservation but are limited by their dependence on identity-specific models. Each new identity requires training from scratch, incurring high computational costs and reduced scalability compared to generative model-based approaches. To overcome this limitation, we propose FIAG, a novel 3D speaking head synthesis framework that enables efficient identity-specific adaptation using only a few training footage. FIAG incorporates Global Gaussian Field, which supports the representation of multiple identities within a shared field, and Universal Motion Field, which captures the common motion dynamics across diverse identities. Benefiting from the shared facial structure information encoded in the Global Gaussian Field and the general motion priors learned in the motion field, our framework enables rapid adaptation from canonical identity representations to specific ones with minimal data. Extensive comparative and ablation experiments demonstrate that our method outperforms existing state-of-the-art approaches, validating both the effectiveness and generalizability of the proposed framework. Code is available at: \\textit{https://github.com/gme-hong/FIAG}.", "date": "2025-06-27", "categories": []}
{"id": "2410.08082", "title": "ToMiE: Towards Explicit Exoskeleton for the Reconstruction of Complicated 3D Human Avatars", "abstract": "In this paper, we highlight a critical yet often overlooked factor in most 3D human tasks, namely modeling complicated 3D human with with hand-held objects or loose-fitting clothing. It is known that the parameterized formulation of SMPL is able to fit human skin; while hand-held objects and loose-fitting clothing, are difficult to get modeled within the unified framework, since their movements are usually decoupled with the human body. To enhance the capability of SMPL skeleton in response to this situation, we propose a growth strategy that enables the joint tree of the skeleton to expand adaptively. Specifically, our method, called ToMiE, consists of parent joints localization and external joints optimization. For parent joints localization, we employ a gradient-based approach guided by both LBS blending weights and motion kernels. Once the external joints are obtained, we proceed to optimize their transformations in SE(3) across different frames, enabling rendering and explicit animation. ToMiE manages to outperform other methods across various cases with hand-held objects and loose-fitting clothing, not only in rendering quality but also by offering free animation of grown joints, thereby enhancing the expressive ability of SMPL skeleton for a broader range of applications.", "date": "2024-10-10", "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": "2501.13335", "title": "Deblur-Avatar: Animatable Avatars from Motion-Blurred Monocular Videos", "abstract": "We introduce a novel framework for modeling high-fidelity, animatable 3D human avatars from motion-blurred monocular video inputs. Motion blur is prevalent in real-world dynamic video capture, especially due to human movements in 3D human avatar modeling. Existing methods either (1) assume sharp image inputs, failing to address the detail loss introduced by motion blur, or (2) mainly consider blur by camera movements, neglecting the human motion blur which is more common in animatable avatars. Our proposed approach integrates a human movement-based motion blur model into 3D Gaussian Splatting (3DGS). By explicitly modeling human motion trajectories during exposure time, we jointly optimize the trajectories and 3D Gaussians to reconstruct sharp, high-quality human avatars. We employ a pose-dependent fusion mechanism to distinguish moving body regions, optimizing both blurred and sharp areas effectively. Extensive experiments on synthetic and real-world datasets demonstrate that our method significantly outperforms existing methods in rendering quality and quantitative metrics, producing sharp avatar reconstructions and enabling real-time rendering under challenging motion blur conditions.", "date": "2025-01-23", "categories": []}
{"id": "2504.18215", "title": "Unify3D: An Augmented Holistic End-to-end Monocular 3D Human Reconstruction via Anatomy Shaping and Twins Negotiating", "abstract": "Monocular 3D clothed human reconstruction aims to create a complete 3D avatar from a single image. To tackle the human geometry lacking in one RGB image, current methods typically resort to a preceding model for an explicit geometric representation. For the reconstruction itself, focus is on modeling both it and the input image. This routine is constrained by the preceding model, and overlooks the integrity of the reconstruction task. To address this, this paper introduces a novel paradigm that treats human reconstruction as a holistic process, utilizing an end-to-end network for direct prediction from 2D image to 3D avatar, eliminating any explicit intermediate geometry display. Based on this, we further propose a novel reconstruction framework consisting of two core components: the Anatomy Shaping Extraction module, which captures implicit shape features taking into account the specialty of human anatomy, and the Twins Negotiating Reconstruction U-Net, which enhances reconstruction through feature interaction between two U-Nets of different modalities. Moreover, we propose a Comic Data Augmentation strategy and construct 15k+ 3D human scans to bolster model performance in more complex case input. Extensive experiments on two test sets and many in-the-wild cases show the superiority of our method over SOTA methods. Our demos can be found in : https://e2e3dgsrecon.github.io/e2e3dgsrecon/.", "date": "2025-04-25", "categories": []}
{"id": "2506.12793", "title": "SMPL Normal Map Is All You Need for Single-view Textured Human Reconstruction", "abstract": "Single-view textured human reconstruction aims to reconstruct a clothed 3D digital human by inputting a monocular 2D image. Existing approaches include feed-forward methods, limited by scarce 3D human data, and diffusion-based methods, prone to erroneous 2D hallucinations. To address these issues, we propose a novel SMPL normal map Equipped 3D Human Reconstruction (SEHR) framework, integrating a pretrained large 3D reconstruction model with human geometry prior. SEHR performs single-view human reconstruction without using a preset diffusion model in one forward propagation. Concretely, SEHR consists of two key components: SMPL Normal Map Guidance (SNMG) and SMPL Normal Map Constraint (SNMC). SNMG incorporates SMPL normal maps into an auxiliary network to provide improved body shape guidance. SNMC enhances invisible body parts by constraining the model to predict an extra SMPL normal Gaussians. Extensive experiments on two benchmark datasets demonstrate that SEHR outperforms existing state-of-the-art methods.", "date": "2025-06-15", "categories": []}
{"id": "2506.13233", "title": "High-Quality Facial Albedo Generation for 3D Face Reconstruction from a Single Image using a Coarse-to-Fine Approach", "abstract": "Facial texture generation is crucial for high-fidelity 3D face reconstruction from a single image. However, existing methods struggle to generate UV albedo maps with high-frequency details. To address this challenge, we propose a novel end-to-end coarse-to-fine approach for UV albedo map generation. Our method first utilizes a UV Albedo Parametric Model (UVAPM), driven by low-dimensional coefficients, to generate coarse albedo maps with skin tones and low-frequency texture details. To capture high-frequency details, we train a detail generator using a decoupled albedo map dataset, producing high-resolution albedo maps. Extensive experiments demonstrate that our method can generate high-fidelity textures from a single image, outperforming existing methods in terms of texture quality and realism. The code and pre-trained model are publicly available at https://github.com/MVIC-DAI/UVAPM, facilitating reproducibility and further research.", "date": "2025-06-16", "categories": []}
{"id": "2506.06645", "title": "Parametric Gaussian Human Model: Generalizable Prior for Efficient and Realistic Human Avatar Modeling", "abstract": "Photorealistic and animatable human avatars are a key enabler for virtual/augmented reality, telepresence, and digital entertainment. While recent advances in 3D Gaussian Splatting (3DGS) have greatly improved rendering quality and efficiency, existing methods still face fundamental challenges, including time-consuming per-subject optimization and poor generalization under sparse monocular inputs. In this work, we present the Parametric Gaussian Human Model (PGHM), a generalizable and efficient framework that integrates human priors into 3DGS for fast and high-fidelity avatar reconstruction from monocular videos. PGHM introduces two core components: (1) a UV-aligned latent identity map that compactly encodes subject-specific geometry and appearance into a learnable feature tensor; and (2) a disentangled Multi-Head U-Net that predicts Gaussian attributes by decomposing static, pose-dependent, and view-dependent components via conditioned decoders. This design enables robust rendering quality under challenging poses and viewpoints, while allowing efficient subject adaptation without requiring multi-view capture or long optimization time. Experiments show that PGHM is significantly more efficient than optimization-from-scratch methods, requiring only approximately 20 minutes per subject to produce avatars with comparable visual quality, thereby demonstrating its practical applicability for real-world monocular avatar creation.", "date": "2025-06-07", "categories": []}
{"id": "2506.07628", "title": "HuSc3D: Human Sculpture dataset for 3D object reconstruction", "abstract": "3D scene reconstruction from 2D images is one of the most important tasks in computer graphics. Unfortunately, existing datasets and benchmarks concentrate on idealized synthetic or meticulously captured realistic data. Such benchmarks fail to convey the inherent complexities encountered in newly acquired real-world scenes. In such scenes especially those acquired outside, the background is often dynamic, and by popular usage of cell phone cameras, there might be discrepancies in, e.g., white balance. To address this gap, we present HuSc3D, a novel dataset specifically designed for rigorous benchmarking of 3D reconstruction models under realistic acquisition challenges. Our dataset uniquely features six highly detailed, fully white sculptures characterized by intricate perforations and minimal textural and color variation. Furthermore, the number of images per scene varies significantly, introducing the additional challenge of limited training data for some instances alongside scenes with a standard number of views. By evaluating popular 3D reconstruction methods on this diverse dataset, we demonstrate the distinctiveness of HuSc3D in effectively differentiating model performance, particularly highlighting the sensitivity of methods to fine geometric details, color ambiguity, and varying data availability--limitations often masked by more conventional datasets.", "date": "2025-06-09", "categories": []}
{"id": "2506.04351", "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": "2412.13058", "title": "CondiMen: Conditional Multi-Person Mesh Recovery", "abstract": "Multi-person human mesh recovery (HMR) consists in detecting all individuals in a given input image, and predicting the body shape, pose, and 3D location for each detected person. The dominant approaches to this task rely on neural networks trained to output a single prediction for each detected individual. In contrast, we propose CondiMen, a method that outputs a joint parametric distribution over likely poses, body shapes, intrinsics and distances to the camera, using a Bayesian network. This approach offers several advantages. First, a probability distribution can handle some inherent ambiguities of this task -- such as the uncertainty between a person's size and their distance to the camera, or simply the loss of information when projecting 3D data onto the 2D image plane. Second, the output distribution can be combined with additional information to produce better predictions, by using e.g. known camera or body shape parameters, or by exploiting multi-view observations. Third, one can efficiently extract the most likely predictions from the output distribution, making our proposed approach suitable for real-time applications. Empirically we find that our model i) achieves performance on par with or better than the state-of-the-art, ii) captures uncertainties and correlations inherent in pose estimation and iii) can exploit additional information at test time, such as multi-view consistency or body shape priors. CondiMen spices up the modeling of ambiguity, using just the right ingredients on hand.", "date": "2024-12-17", "categories": []}
{"id": "2506.03118", "title": "HumanRAM: Feed-forward Human Reconstruction and Animation Model using Transformers", "abstract": "3D human reconstruction and animation are long-standing topics in computer graphics and vision. However, existing methods typically rely on sophisticated dense-view capture and/or time-consuming per-subject optimization procedures. To address these limitations, we propose HumanRAM, a novel feed-forward approach for generalizable human reconstruction and animation from monocular or sparse human images. Our approach integrates human reconstruction and animation into a unified framework by introducing explicit pose conditions, parameterized by a shared SMPL-X neural texture, into transformer-based large reconstruction models (LRM). Given monocular or sparse input images with associated camera parameters and SMPL-X poses, our model employs scalable transformers and a DPT-based decoder to synthesize realistic human renderings under novel viewpoints and novel poses. By leveraging the explicit pose conditions, our model simultaneously enables high-quality human reconstruction and high-fidelity pose-controlled animation. Experiments show that HumanRAM significantly surpasses previous methods in terms of reconstruction accuracy, animation fidelity, and generalization performance on real-world datasets. Video results are available at https://zju3dv.github.io/humanram/.", "date": "2025-06-03", "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": "2505.23085", "title": "GeoMan: Temporally Consistent Human Geometry Estimation using Image-to-Video Diffusion", "abstract": "Estimating accurate and temporally consistent 3D human geometry from videos is a challenging problem in computer vision. Existing methods, primarily optimized for single images, often suffer from temporal inconsistencies and fail to capture fine-grained dynamic details. To address these limitations, we present GeoMan, a novel architecture designed to produce accurate and temporally consistent depth and normal estimations from monocular human videos. GeoMan addresses two key challenges: the scarcity of high-quality 4D training data and the need for metric depth estimation to accurately model human size. To overcome the first challenge, GeoMan employs an image-based model to estimate depth and normals for the first frame of a video, which then conditions a video diffusion model, reframing video geometry estimation task as an image-to-video generation problem. This design offloads the heavy lifting of geometric estimation to the image model and simplifies the video model's role to focus on intricate details while using priors learned from large-scale video datasets. Consequently, GeoMan improves temporal consistency and generalizability while requiring minimal 4D training data. To address the challenge of accurate human size estimation, we introduce a root-relative depth representation that retains critical human-scale details and is easier to be estimated from monocular inputs, overcoming the limitations of traditional affine-invariant and metric depth representations. GeoMan achieves state-of-the-art performance in both qualitative and quantitative evaluations, demonstrating its effectiveness in overcoming longstanding challenges in 3D human geometry estimation from videos.", "date": "2025-05-29", "categories": []}
{"id": "2505.19049", "title": "Disentangled Human Body Representation Based on Unsupervised Semantic-Aware Learning", "abstract": "In recent years, more and more attention has been paid to the learning of 3D human representation. However, the complexity of lots of hand-defined human body constraints and the absence of supervision data limit that the existing works controllably and accurately represent the human body in views of semantics and representation ability. In this paper, we propose a human body representation with controllable fine-grained semantics and high precison of reconstruction in an unsupervised learning framework. In particularly, we design a whole-aware skeleton-grouped disentangle strategy to learn a correspondence between geometric semantical measurement of body and latent codes, which facilitates the control of shape and posture of human body by modifying latent coding paramerers. With the help of skeleton-grouped whole-aware encoder and unsupervised disentanglement losses, our representation model is learned by an unsupervised manner. Besides, a based-template residual learning scheme is injected into the encoder to ease of learning human body latent parameter in complicated body shape and pose spaces. Because of the geometrically meaningful latent codes, it can be used in a wide range of applications, from human body pose transfer to bilinear latent code interpolation. Further more, a part-aware decoder is utlized to promote the learning of controllable fine-grained semantics. The experimental results on public 3D human datasets show that the method has the ability of precise reconstruction.", "date": "2025-05-25", "categories": []}
{"id": "2411.06232", "title": "RCR: Robust Crowd Reconstruction with Upright Space from a Single Large-scene Image", "abstract": "This paper focuses on spatially consistent hundreds of human pose and shape reconstruction from a single large-scene image with various human scales under arbitrary camera FoVs (Fields of View). Due to the small and highly varying 2D human scales, depth ambiguity, and perspective distortion, no existing methods can achieve globally consistent reconstruction with correct reprojection. To address these challenges, we first propose a new concept, Human-scene Virtual Interaction Point (HVIP), to convert the complex 3D human localization into 2D-pixel localization. We then extend it to RCR (Robust Crowd Reconstruction), which achieves globally consistent reconstruction and stable generalization on different camera FoVs without test-time optimization. To perceive humans in varying pixel sizes, we propose an Iterative Ground-aware Cropping to automatically crop the image and then merge the results. To eliminate the influence of the camera and cropping process during the reconstruction, we introduce a canonical Upright 3D Space and the corresponding Upright 2D Space. To link the canonical space and the camera space, we propose the Upright Normalization, which transforms the local crop input into the Upright 2D Space, and transforms the output from the Upright 3D Space into the unified camera space. Besides, we contribute two benchmark datasets, LargeCrowd and SynCrowd, for evaluating crowd reconstruction in large scenes. Experimental results demonstrate the effectiveness of the proposed method. The source code and data will be publicly available for research purposes.", "date": "2024-11-09", "categories": []}
{"id": "2505.18025", "title": "3D Face Reconstruction Error Decomposed: A Modular Benchmark for Fair and Fast Method Evaluation", "abstract": "Computing the standard benchmark metric for 3D face reconstruction, namely geometric error, requires a number of steps, such as mesh cropping, rigid alignment, or point correspondence. Current benchmark tools are monolithic (they implement a specific combination of these steps), even though there is no consensus on the best way to measure error. We present a toolkit for a Modularized 3D Face reconstruction Benchmark (M3DFB), where the fundamental components of error computation are segregated and interchangeable, allowing one to quantify the effect of each. Furthermore, we propose a new component, namely correction, and present a computationally efficient approach that penalizes for mesh topology inconsistency. Using this toolkit, we test 16 error estimators with 10 reconstruction methods on two real and two synthetic datasets. Critically, the widely used ICP-based estimator provides the worst benchmarking performance, as it significantly alters the true ranking of the top-5 reconstruction methods. Notably, the correlation of ICP with the true error can be as low as 0.41. Moreover, non-rigid alignment leads to significant improvement (correlation larger than 0.90), highlighting the importance of annotating 3D landmarks on datasets. Finally, the proposed correction scheme, together with non-rigid warping, leads to an accuracy on a par with the best non-rigid ICP-based estimators, but runs an order of magnitude faster. Our open-source codebase is designed for researchers to easily compare alternatives for each component, thus helping accelerating progress in benchmarking for 3D face reconstruction and, furthermore, supporting the improvement of learned reconstruction methods, which depend on accurate error estimation for effective training.", "date": "2025-05-23", "categories": []}
{"id": "2412.17806", "title": "Reconstructing People, Places, and Cameras", "abstract": "We present \"Humans and Structure from Motion\" (HSfM), a method for jointly reconstructing multiple human meshes, scene point clouds, and camera parameters in a metric world coordinate system from a sparse set of uncalibrated multi-view images featuring people. Our approach combines data-driven scene reconstruction with the traditional Structure-from-Motion (SfM) framework to achieve more accurate scene reconstruction and camera estimation, while simultaneously recovering human meshes. In contrast to existing scene reconstruction and SfM methods that lack metric scale information, our method estimates approximate metric scale by leveraging a human statistical model. Furthermore, it reconstructs multiple human meshes within the same world coordinate system alongside the scene point cloud, effectively capturing spatial relationships among individuals and their positions in the environment. We initialize the reconstruction of humans, scenes, and cameras using robust foundational models and jointly optimize these elements. This joint optimization synergistically improves the accuracy of each component. We compare our method to existing approaches on two challenging benchmarks, EgoHumans and EgoExo4D, demonstrating significant improvements in human localization accuracy within the world coordinate frame (reducing error from 3.51m to 1.04m in EgoHumans and from 2.9m to 0.56m in EgoExo4D). Notably, our results show that incorporating human data into the SfM pipeline improves camera pose estimation (e.g., increasing RRA@15 by 20.3% on EgoHumans). Additionally, qualitative results show that our approach improves overall scene reconstruction quality. Our code is available at: https://github.com/hongsukchoi/HSfM_RELEASE", "date": "2024-12-23", "categories": []}
{"id": "2505.15385", "title": "EVA: Expressive Virtual Avatars from Multi-view Videos", "abstract": "With recent advancements in neural rendering and motion capture algorithms, remarkable progress has been made in photorealistic human avatar modeling, unlocking immense potential for applications in virtual reality, augmented reality, remote communication, and industries such as gaming, film, and medicine. However, existing methods fail to provide complete, faithful, and expressive control over human avatars due to their entangled representation of facial expressions and body movements. In this work, we introduce Expressive Virtual Avatars (EVA), an actor-specific, fully controllable, and expressive human avatar framework that achieves high-fidelity, lifelike renderings in real time while enabling independent control of facial expressions, body movements, and hand gestures. Specifically, our approach designs the human avatar as a two-layer model: an expressive template geometry layer and a 3D Gaussian appearance layer. First, we present an expressive template tracking algorithm that leverages coarse-to-fine optimization to accurately recover body motions, facial expressions, and non-rigid deformation parameters from multi-view videos. Next, we propose a novel decoupled 3D Gaussian appearance model designed to effectively disentangle body and facial appearance. Unlike unified Gaussian estimation approaches, our method employs two specialized and independent modules to model the body and face separately. Experimental results demonstrate that EVA surpasses state-of-the-art methods in terms of rendering quality and expressiveness, validating its effectiveness in creating full-body avatars. This work represents a significant advancement towards fully drivable digital human models, enabling the creation of lifelike digital avatars that faithfully replicate human geometry and appearance.", "date": "2025-05-21", "categories": []}
{"id": "2407.10707", "title": "Interactive Rendering of Relightable and Animatable Gaussian Avatars", "abstract": "Creating relightable and animatable avatars from multi-view or monocular videos is a challenging task for digital human creation and virtual reality applications. Previous methods rely on neural radiance fields or ray tracing, resulting in slow training and rendering processes. By utilizing Gaussian Splatting, we propose a simple and efficient method to decouple body materials and lighting from sparse-view or monocular avatar videos, so that the avatar can be rendered simultaneously under novel viewpoints, poses, and lightings at interactive frame rates (6.9 fps). Specifically, we first obtain the canonical body mesh using a signed distance function and assign attributes to each mesh vertex. The Gaussians in the canonical space then interpolate from nearby body mesh vertices to obtain the attributes. We subsequently deform the Gaussians to the posed space using forward skinning, and combine the learnable environment light with the Gaussian attributes for shading computation. To achieve fast shadow modeling, we rasterize the posed body mesh from dense viewpoints to obtain the visibility. Our approach is not only simple but also fast enough to allow interactive rendering of avatar animation under environmental light changes. Experiments demonstrate that, compared to previous works, our method can render higher quality results at a faster speed on both synthetic and real datasets.", "date": "2024-07-15", "categories": []}
{"id": "2505.05022", "title": "SOAP: Style-Omniscient Animatable Portraits", "abstract": "Creating animatable 3D avatars from a single image remains challenging due to style limitations (realistic, cartoon, anime) and difficulties in handling accessories or hairstyles. While 3D diffusion models advance single-view reconstruction for general objects, outputs often lack animation controls or suffer from artifacts because of the domain gap. We propose SOAP, a style-omniscient framework to generate rigged, topology-consistent avatars from any portrait. Our method leverages a multiview diffusion model trained on 24K 3D heads with multiple styles and an adaptive optimization pipeline to deform the FLAME mesh while maintaining topology and rigging via differentiable rendering. The resulting textured avatars support FACS-based animation, integrate with eyeballs and teeth, and preserve details like braided hair or accessories. Extensive experiments demonstrate the superiority of our method over state-of-the-art techniques for both single-view head modeling and diffusion-based generation of Image-to-3D. Our code and data are publicly available for research purposes at https://github.com/TingtingLiao/soap.", "date": "2025-05-08", "categories": []}
{"id": "2505.12005", "title": "CHRIS: Clothed Human Reconstruction with Side View Consistency", "abstract": "Creating a realistic clothed human from a single-view RGB image is crucial for applications like mixed reality and filmmaking. Despite some progress in recent years, mainstream methods often fail to fully utilize side-view information, as the input single-view image contains front-view information only. This leads to globally unrealistic topology and local surface inconsistency in side views. To address these, we introduce Clothed Human Reconstruction with Side View Consistency, namely CHRIS, which consists of 1) A Side-View Normal Discriminator that enhances global visual reasonability by distinguishing the generated side-view normals from the ground truth ones; 2) A Multi-to-One Gradient Computation (M2O) that ensures local surface consistency. M2O calculates the gradient of a sampling point by integrating the gradients of the nearby points, effectively acting as a smooth operation. Experimental results demonstrate that CHRIS achieves state-of-the-art performance on public benchmarks and outperforms the prior work.", "date": "2025-05-17", "categories": []}
{"id": "2405.03689", "title": "Pose Priors from Language Models", "abstract": "Language is often used to describe physical interaction, yet most 3D human pose estimation methods overlook this rich source of information. We bridge this gap by leveraging large multimodal models (LMMs) as priors for reconstructing contact poses, offering a scalable alternative to traditional methods that rely on human annotations or motion capture data. Our approach extracts contact-relevant descriptors from an LMM and translates them into tractable losses to constrain 3D human pose optimization. Despite its simplicity, our method produces compelling reconstructions for both two-person interactions and self-contact scenarios, accurately capturing the semantics of physical and social interactions. Our results demonstrate that LMMs can serve as powerful tools for contact prediction and pose estimation, offering an alternative to costly manual human annotations or motion capture data. Our code is publicly available at https://prosepose.github.io.", "date": "2024-05-06", "categories": []}
{"id": "2503.11792", "title": "StyleMorpheus: A Style-Based 3D-Aware Morphable Face Model", "abstract": "For 3D face modeling, the recently developed 3D-aware neural rendering methods are able to render photorealistic face images with arbitrary viewing directions. The training of the parametric controllable 3D-aware face models, however, still relies on a large-scale dataset that is lab-collected. To address this issue, this paper introduces \"StyleMorpheus\", the first style-based neural 3D Morphable Face Model (3DMM) that is trained on in-the-wild images. It inherits 3DMM's disentangled controllability (over face identity, expression, and appearance) but without the need for accurately reconstructed explicit 3D shapes. StyleMorpheus employs an auto-encoder structure. The encoder aims at learning a representative disentangled parametric code space and the decoder improves the disentanglement using shape and appearance-related style codes in the different sub-modules of the network. Furthermore, we fine-tune the decoder through style-based generative adversarial learning to achieve photorealistic 3D rendering quality. The proposed style-based design enables StyleMorpheus to achieve state-of-the-art 3D-aware face reconstruction results, while also allowing disentangled control of the reconstructed face. Our model achieves real-time rendering speed, allowing its use in virtual reality applications. We also demonstrate the capability of the proposed style-based design in face editing applications such as style mixing and color editing. Project homepage: https://github.com/ubc-3d-vision-lab/StyleMorpheus.", "date": "2025-03-14", "categories": []}
{"id": "2504.08353", "title": "Single View Garment Reconstruction Using Diffusion Mapping Via Pattern Coordinates", "abstract": "Reconstructing 3D clothed humans from images is fundamental to applications like virtual try-on, avatar creation, and mixed reality. While recent advances have enhanced human body recovery, accurate reconstruction of garment geometry -- especially for loose-fitting clothing -- remains an open challenge. We present a novel method for high-fidelity 3D garment reconstruction from single images that bridges 2D and 3D representations. Our approach combines Implicit Sewing Patterns (ISP) with a generative diffusion model to learn rich garment shape priors in a 2D UV space. A key innovation is our mapping model that establishes correspondences between 2D image pixels, UV pattern coordinates, and 3D geometry, enabling joint optimization of both 3D garment meshes and the corresponding 2D patterns by aligning learned priors with image observations. Despite training exclusively on synthetically simulated cloth data, our method generalizes effectively to real-world images, outperforming existing approaches on both tight- and loose-fitting garments. The reconstructed garments maintain physical plausibility while capturing fine geometric details, enabling downstream applications including garment retargeting and texture manipulation.", "date": "2025-04-11", "categories": []}
{"id": "2505.07333", "title": "Link to the Past: Temporal Propagation for Fast 3D Human Reconstruction from Monocular Video", "abstract": "Fast 3D clothed human reconstruction from monocular video remains a significant challenge in computer vision, particularly in balancing computational efficiency with reconstruction quality. Current approaches are either focused on static image reconstruction but too computationally intensive, or achieve high quality through per-video optimization that requires minutes to hours of processing, making them unsuitable for real-time applications. To this end, we present TemPoFast3D, a novel method that leverages temporal coherency of human appearance to reduce redundant computation while maintaining reconstruction quality. Our approach is a \"plug-and play\" solution that uniquely transforms pixel-aligned reconstruction networks to handle continuous video streams by maintaining and refining a canonical appearance representation through efficient coordinate mapping. Extensive experiments demonstrate that TemPoFast3D matches or exceeds state-of-the-art methods across standard metrics while providing high-quality textured reconstruction across diverse pose and appearance, with a maximum speed of 12 FPS.", "date": "2025-05-12", "categories": []}
{"id": "2505.05672", "title": "TeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling", "abstract": "Sparse volumetric reconstruction and rendering via 3D Gaussian splatting have recently enabled animatable 3D head avatars that are rendered under arbitrary viewpoints with impressive photorealism. Today, such photoreal avatars are seen as a key component in emerging applications in telepresence, extended reality, and entertainment. Building a photoreal avatar requires estimating the complex non-rigid motion of different facial components as seen in input video images; due to inaccurate motion estimation, animatable models typically present a loss of fidelity and detail when compared to their non-animatable counterparts, built from an individual facial expression. Also, recent state-of-the-art models are often affected by memory limitations that reduce the number of 3D Gaussians used for modeling, leading to lower detail and quality. To address these problems, we present a new high-detail 3D head avatar model that improves upon the state of the art, largely increasing the number of 3D Gaussians and modeling quality for rendering at 4K resolution. Our high-quality model is reconstructed from multiview input video and builds on top of a mesh-based 3D morphable model, which provides a coarse deformation layer for the head. Photoreal appearance is modelled by 3D Gaussians embedded within the continuous UVD tangent space of this mesh, allowing for more effective densification where most needed. Additionally, these Gaussians are warped by a novel UVD deformation field to capture subtle, localized motion. Our key contribution is the novel deformable Gaussian encoding and overall fitting procedure that allows our head model to preserve appearance detail, while capturing facial motion and other transient high-frequency features such as skin wrinkling.", "date": "2025-05-08", "categories": []}
{"id": "2406.06050", "title": "Generalizable Human Gaussians from Single-View Image", "abstract": "  In this work, we tackle the task of learning 3D human Gaussians from a single\nimage, focusing on recovering detailed appearance and geometry including\nunobserved regions. We introduce a single-view generalizable Human Gaussian\nModel (HGM), which employs a novel generate-then-refine pipeline with the\nguidance from human body prior and diffusion prior. Our approach uses a\nControlNet to refine rendered back-view images from coarse predicted human\nGaussians, then uses the refined image along with the input image to\nreconstruct refined human Gaussians. To mitigate the potential generation of\nunrealistic human poses and shapes, we incorporate human priors from the SMPL-X\nmodel as a dual branch, propagating image features from the SMPL-X volume to\nthe image Gaussians using sparse convolution and attention mechanisms. Given\nthat the initial SMPL-X estimation might be inaccurate, we gradually refine it\nwith our HGM model. We validate our approach on several publicly available\ndatasets. Our method surpasses previous methods in both novel view synthesis\nand surface reconstruction. Our approach also exhibits strong generalization\nfor cross-dataset evaluation and in-the-wild images.\n", "date": "2024-06-10", "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": "2505.01838", "title": "MVHumanNet++: A Large-scale Dataset of Multi-view Daily Dressing Human\n  Captures with Richer Annotations for 3D Human Digitization", "abstract": "  In this era, the success of large language models and text-to-image models\ncan be attributed to the driving force of large-scale datasets. However, in the\nrealm of 3D vision, while significant progress has been achieved in\nobject-centric tasks through large-scale datasets like Objaverse and MVImgNet,\nhuman-centric tasks have seen limited advancement, largely due to the absence\nof a comparable large-scale human dataset. To bridge this gap, we present\nMVHumanNet++, a dataset that comprises multi-view human action sequences of\n4,500 human identities. The primary focus of our work is on collecting human\ndata that features a large number of diverse identities and everyday clothing\nusing multi-view human capture systems, which facilitates easily scalable data\ncollection. Our dataset contains 9,000 daily outfits, 60,000 motion sequences\nand 645 million frames with extensive annotations, including human masks,\ncamera parameters, 2D and 3D keypoints, SMPL/SMPLX parameters, and\ncorresponding textual descriptions. Additionally, the proposed MVHumanNet++\ndataset is enhanced with newly processed normal maps and depth maps,\nsignificantly expanding its applicability and utility for advanced\nhuman-centric research. To explore the potential of our proposed MVHumanNet++\ndataset in various 2D and 3D visual tasks, we conducted several pilot studies\nto demonstrate the performance improvements and effective applications enabled\nby the scale provided by MVHumanNet++. As the current largest-scale 3D human\ndataset, we hope that the release of MVHumanNet++ dataset with annotations will\nfoster further innovations in the domain of 3D human-centric tasks at scale.\nMVHumanNet++ is publicly available at\nhttps://kevinlee09.github.io/research/MVHumanNet++/.\n", "date": "2025-05-03", "categories": []}
{"id": "2505.02108", "title": "SignSplat: Rendering Sign Language via Gaussian Splatting", "abstract": "  State-of-the-art approaches for conditional human body rendering via Gaussian\nsplatting typically focus on simple body motions captured from many views. This\nis often in the context of dancing or walking. However, for more complex use\ncases, such as sign language, we care less about large body motion and more\nabout subtle and complex motions of the hands and face. The problems of\nbuilding high fidelity models are compounded by the complexity of capturing\nmulti-view data of sign. The solution is to make better use of sequence data,\nensuring that we can overcome the limited information from only a few views by\nexploiting temporal variability. Nevertheless, learning from sequence-level\ndata requires extremely accurate and consistent model fitting to ensure that\nappearance is consistent across complex motions. We focus on how to achieve\nthis, constraining mesh parameters to build an accurate Gaussian splatting\nframework from few views capable of modelling subtle human motion. We leverage\nregularization techniques on the Gaussian parameters to mitigate overfitting\nand rendering artifacts. Additionally, we propose a new adaptive control method\nto densify Gaussians and prune splat points on the mesh surface. To demonstrate\nthe accuracy of our approach, we render novel sequences of sign language video,\nbuilding on neural machine translation approaches to sign stitching. On\nbenchmark datasets, our approach achieves state-of-the-art performance; and on\nhighly articulated and complex sign language motion, we significantly\noutperform competing approaches.\n", "date": "2025-05-04", "categories": []}
{"id": "2312.04106", "title": "Do Not DeepFake Me: Privacy-Preserving Neural 3D Head Reconstruction\n  Without Sensitive Images", "abstract": "  While 3D head reconstruction is widely used for modeling, existing neural\nreconstruction approaches rely on high-resolution multi-view images, posing\nnotable privacy issues. Individuals are particularly sensitive to facial\nfeatures, and facial image leakage can lead to many malicious activities, such\nas unauthorized tracking and deepfake. In contrast, geometric data is less\nsusceptible to misuse due to its complex processing requirements, and absence\nof facial texture features. In this paper, we propose a novel two-stage 3D\nfacial reconstruction method aimed at avoiding exposure to sensitive facial\ninformation while preserving detailed geometric accuracy. Our approach first\nuses non-sensitive rear-head images for initial geometry and then refines this\ngeometry using processed privacy-removed gradient images. Extensive experiments\nshow that the resulting geometry is comparable to methods using full images,\nwhile the process is resistant to DeepFake applications and facial recognition\n(FR) systems, thereby proving its effectiveness in privacy protection.\n", "date": "2023-12-07", "categories": []}
{"id": "2505.00421", "title": "Real-Time Animatable 2DGS-Avatars with Detail Enhancement from Monocular\n  Videos", "abstract": "  High-quality, animatable 3D human avatar reconstruction from monocular videos\noffers significant potential for reducing reliance on complex hardware, making\nit highly practical for applications in game development, augmented reality,\nand social media. However, existing methods still face substantial challenges\nin capturing fine geometric details and maintaining animation stability,\nparticularly under dynamic or complex poses. To address these issues, we\npropose a novel real-time framework for animatable human avatar reconstruction\nbased on 2D Gaussian Splatting (2DGS). By leveraging 2DGS and global SMPL pose\nparameters, our framework not only aligns positional and rotational\ndiscrepancies but also enables robust and natural pose-driven animation of the\nreconstructed avatars. Furthermore, we introduce a Rotation Compensation\nNetwork (RCN) that learns rotation residuals by integrating local geometric\nfeatures with global pose parameters. This network significantly improves the\nhandling of non-rigid deformations and ensures smooth, artifact-free pose\ntransitions during animation. Experimental results demonstrate that our method\nsuccessfully reconstructs realistic and highly animatable human avatars from\nmonocular videos, effectively preserving fine-grained details while ensuring\nstable and natural pose variation. Our approach surpasses current\nstate-of-the-art methods in both reconstruction quality and animation\nrobustness on public benchmarks.\n", "date": "2025-05-01", "categories": []}
{"id": "2505.00615", "title": "Pixel3DMM: Versatile Screen-Space Priors for Single-Image 3D Face\n  Reconstruction", "abstract": "  We address the 3D reconstruction of human faces from a single RGB image. To\nthis end, we propose Pixel3DMM, a set of highly-generalized vision transformers\nwhich predict per-pixel geometric cues in order to constrain the optimization\nof a 3D morphable face model (3DMM). We exploit the latent features of the DINO\nfoundation model, and introduce a tailored surface normal and uv-coordinate\nprediction head. We train our model by registering three high-quality 3D face\ndatasets against the FLAME mesh topology, which results in a total of over\n1,000 identities and 976K images. For 3D face reconstruction, we propose a\nFLAME fitting opitmization that solves for the 3DMM parameters from the\nuv-coordinate and normal estimates. To evaluate our method, we introduce a new\nbenchmark for single-image face reconstruction, which features high diversity\nfacial expressions, viewing angles, and ethnicities. Crucially, our benchmark\nis the first to evaluate both posed and neutral facial geometry. Ultimately,\nour method outperforms the most competitive baselines by over 15% in terms of\ngeometric accuracy for posed facial expressions.\n", "date": "2025-05-01", "categories": []}
{"id": "2504.12909", "title": "Real-time High-fidelity Gaussian Human Avatars with Position-based\n  Interpolation of Spatially Distributed MLPs", "abstract": "  Many works have succeeded in reconstructing Gaussian human avatars from\nmulti-view videos. However, they either struggle to capture pose-dependent\nappearance details with a single MLP, or rely on a computationally intensive\nneural network to reconstruct high-fidelity appearance but with rendering\nperformance degraded to non-real-time. We propose a novel Gaussian human avatar\nrepresentation that can reconstruct high-fidelity pose-dependence appearance\nwith details and meanwhile can be rendered in real time. Our Gaussian avatar is\nempowered by spatially distributed MLPs which are explicitly located on\ndifferent positions on human body. The parameters stored in each Gaussian are\nobtained by interpolating from the outputs of its nearby MLPs based on their\ndistances. To avoid undesired smooth Gaussian property changing during\ninterpolation, for each Gaussian we define a set of Gaussian offset basis, and\na linear combination of basis represents the Gaussian property offsets relative\nto the neutral properties. Then we propose to let the MLPs output a set of\ncoefficients corresponding to the basis. In this way, although Gaussian\ncoefficients are derived from interpolation and change smoothly, the Gaussian\noffset basis is learned freely without constraints. The smoothly varying\ncoefficients combined with freely learned basis can still produce distinctly\ndifferent Gaussian property offsets, allowing the ability to learn\nhigh-frequency spatial signals. We further use control points to constrain the\nGaussians distributed on a surface layer rather than allowing them to be\nirregularly distributed inside the body, to help the human avatar generalize\nbetter when animated under novel poses. Compared to the state-of-the-art\nmethod, our method achieves better appearance quality with finer details while\nthe rendering speed is significantly faster under novel views and novel poses.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.19390", "title": "HumMorph: Generalized Dynamic Human Neural Fields from Few Views", "abstract": "  We introduce HumMorph, a novel generalized approach to free-viewpoint\nrendering of dynamic human bodies with explicit pose control. HumMorph renders\na human actor in any specified pose given a few observed views (starting from\njust one) in arbitrary poses. Our method enables fast inference as it relies\nonly on feed-forward passes through the model. We first construct a coarse\nrepresentation of the actor in the canonical T-pose, which combines visual\nfeatures from individual partial observations and fills missing information\nusing learned prior knowledge. The coarse representation is complemented by\nfine-grained pixel-aligned features extracted directly from the observed views,\nwhich provide high-resolution appearance information. We show that HumMorph is\ncompetitive with the state-of-the-art when only a single input view is\navailable, however, we achieve results with significantly better visual quality\ngiven just 2 monocular observations. Moreover, previous generalized methods\nassume access to accurate body shape and pose parameters obtained using\nsynchronized multi-camera setups. In contrast, we consider a more practical\nscenario where these body parameters are noisily estimated directly from the\nobserved views. Our experimental results demonstrate that our architecture is\nmore robust to errors in the noisy parameters and clearly outperforms the state\nof the art in this setting.\n", "date": "2025-04-27", "categories": []}
{"id": "2503.08224", "title": "HRAvatar: High-Quality and Relightable Gaussian Head Avatar", "abstract": "  Reconstructing animatable and high-quality 3D head avatars from monocular\nvideos, especially with realistic relighting, is a valuable task. However, the\nlimited information from single-view input, combined with the complex head\nposes and facial movements, makes this challenging. Previous methods achieve\nreal-time performance by combining 3D Gaussian Splatting with a parametric head\nmodel, but the resulting head quality suffers from inaccurate face tracking and\nlimited expressiveness of the deformation model. These methods also fail to\nproduce realistic effects under novel lighting conditions. To address these\nissues, we propose HRAvatar, a 3DGS-based method that reconstructs\nhigh-fidelity, relightable 3D head avatars. HRAvatar reduces tracking errors\nthrough end-to-end optimization and better captures individual facial\ndeformations using learnable blendshapes and learnable linear blend skinning.\nAdditionally, it decomposes head appearance into several physical properties\nand incorporates physically-based shading to account for environmental\nlighting. Extensive experiments demonstrate that HRAvatar not only reconstructs\nsuperior-quality heads but also achieves realistic visual effects under varying\nlighting conditions.\n", "date": "2025-03-11", "categories": []}
{"id": "2504.14967", "title": "3D Gaussian Head Avatars with Expressive Dynamic Appearances by Compact\n  Tensorial Representations", "abstract": "  Recent studies have combined 3D Gaussian and 3D Morphable Models (3DMM) to\nconstruct high-quality 3D head avatars. In this line of research, existing\nmethods either fail to capture the dynamic textures or incur significant\noverhead in terms of runtime speed or storage space. To this end, we propose a\nnovel method that addresses all the aforementioned demands. In specific, we\nintroduce an expressive and compact representation that encodes texture-related\nattributes of the 3D Gaussians in the tensorial format. We store appearance of\nneutral expression in static tri-planes, and represents dynamic texture details\nfor different expressions using lightweight 1D feature lines, which are then\ndecoded into opacity offset relative to the neutral face. We further propose\nadaptive truncated opacity penalty and class-balanced sampling to improve\ngeneralization across different expressions. Experiments show this design\nenables accurate face dynamic details capturing while maintains real-time\nrendering and significantly reduces storage costs, thus broadening the\napplicability to more scenarios.\n", "date": "2025-04-21", "categories": []}
{"id": "2504.15179", "title": "FaceCraft4D: Animated 3D Facial Avatar Generation from a Single Image", "abstract": "  We present a novel framework for generating high-quality, animatable 4D\navatar from a single image. While recent advances have shown promising results\nin 4D avatar creation, existing methods either require extensive multiview data\nor struggle with shape accuracy and identity consistency. To address these\nlimitations, we propose a comprehensive system that leverages shape, image, and\nvideo priors to create full-view, animatable avatars. Our approach first\nobtains initial coarse shape through 3D-GAN inversion. Then, it enhances\nmultiview textures using depth-guided warping signals for cross-view\nconsistency with the help of the image diffusion model. To handle expression\nanimation, we incorporate a video prior with synchronized driving signals\nacross viewpoints. We further introduce a Consistent-Inconsistent training to\neffectively handle data inconsistencies during 4D reconstruction. Experimental\nresults demonstrate that our method achieves superior quality compared to the\nprior art, while maintaining consistency across different viewpoints and\nexpressions.\n", "date": "2025-04-21", "categories": []}
{"id": "2410.11682", "title": "SurFhead: Affine Rig Blending for Geometrically Accurate 2D Gaussian\n  Surfel Head Avatars", "abstract": "  Recent advancements in head avatar rendering using Gaussian primitives have\nachieved significantly high-fidelity results. Although precise head geometry is\ncrucial for applications like mesh reconstruction and relighting, current\nmethods struggle to capture intricate geometric details and render unseen poses\ndue to their reliance on similarity transformations, which cannot handle\nstretch and shear transforms essential for detailed deformations of geometry.\nTo address this, we propose SurFhead, a novel method that reconstructs riggable\nhead geometry from RGB videos using 2D Gaussian surfels, which offer\nwell-defined geometric properties, such as precise depth from fixed ray\nintersections and normals derived from their surface orientation, making them\nadvantageous over 3D counterparts. SurFhead ensures high-fidelity rendering of\nboth normals and images, even in extreme poses, by leveraging classical\nmesh-based deformation transfer and affine transformation interpolation.\nSurFhead introduces precise geometric deformation and blends surfels through\npolar decomposition of transformations, including those affecting normals. Our\nkey contribution lies in bridging classical graphics techniques, such as\nmesh-based deformation, with modern Gaussian primitives, achieving\nstate-of-the-art geometry reconstruction and rendering quality. Unlike previous\navatar rendering approaches, SurFhead enables efficient reconstruction driven\nby Gaussian primitives while preserving high-fidelity geometry.\n", "date": "2024-10-15", "categories": []}
{"id": "2504.13167", "title": "ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from\n  Monocular Videos", "abstract": "  Creating a photorealistic scene and human reconstruction from a single\nmonocular in-the-wild video figures prominently in the perception of a\nhuman-centric 3D world. Recent neural rendering advances have enabled holistic\nhuman-scene reconstruction but require pre-calibrated camera and human poses,\nand days of training time. In this work, we introduce a novel unified framework\nthat simultaneously performs camera tracking, human pose estimation and\nhuman-scene reconstruction in an online fashion. 3D Gaussian Splatting is\nutilized to learn Gaussian primitives for humans and scenes efficiently, and\nreconstruction-based camera tracking and human pose estimation modules are\ndesigned to enable holistic understanding and effective disentanglement of pose\nand appearance. Specifically, we design a human deformation module to\nreconstruct the details and enhance generalizability to out-of-distribution\nposes faithfully. Aiming to learn the spatial correlation between human and\nscene accurately, we introduce occlusion-aware human silhouette rendering and\nmonocular geometric priors, which further improve reconstruction quality.\nExperiments on the EMDB and NeuMan datasets demonstrate superior or on-par\nperformance with existing methods in camera tracking, human pose estimation,\nnovel view synthesis and runtime. Our project page is at\nhttps://eth-ait.github.io/ODHSR.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.12999", "title": "GSAC: Leveraging Gaussian Splatting for Photorealistic Avatar Creation\n  with Unity Integration", "abstract": "  Photorealistic avatars have become essential for immersive applications in\nvirtual reality (VR) and augmented reality (AR), enabling lifelike interactions\nin areas such as training simulations, telemedicine, and virtual collaboration.\nThese avatars bridge the gap between the physical and digital worlds, improving\nthe user experience through realistic human representation. However, existing\navatar creation techniques face significant challenges, including high costs,\nlong creation times, and limited utility in virtual applications. Manual\nmethods, such as MetaHuman, require extensive time and expertise, while\nautomatic approaches, such as NeRF-based pipelines often lack efficiency,\ndetailed facial expression fidelity, and are unable to be rendered at a speed\nsufficent for real-time applications. By involving several cutting-edge modern\ntechniques, we introduce an end-to-end 3D Gaussian Splatting (3DGS) avatar\ncreation pipeline that leverages monocular video input to create a scalable and\nefficient photorealistic avatar directly compatible with the Unity game engine.\nOur pipeline incorporates a novel Gaussian splatting technique with customized\npreprocessing that enables the user of \"in the wild\" monocular video capture,\ndetailed facial expression reconstruction and embedding within a fully rigged\navatar model. Additionally, we present a Unity-integrated Gaussian Splatting\nAvatar Editor, offering a user-friendly environment for VR/AR application\ndevelopment. Experimental results validate the effectiveness of our\npreprocessing pipeline in standardizing custom data for 3DGS training and\ndemonstrate the versatility of Gaussian avatars in Unity, highlighting the\nscalability and practicality of our approach.\n", "date": "2025-04-17", "categories": []}
{"id": "2409.04196", "title": "GST: Precise 3D Human Body from a Single Image with Gaussian Splatting\n  Transformers", "abstract": "  Reconstructing posed 3D human models from monocular images has important\napplications in the sports industry, including performance tracking, injury\nprevention and virtual training. In this work, we combine 3D human pose and\nshape estimation with 3D Gaussian Splatting (3DGS), a representation of the\nscene composed of a mixture of Gaussians. This allows training or fine-tuning a\nhuman model predictor on multi-view images alone, without 3D ground truth.\nPredicting such mixtures for a human from a single input image is challenging\ndue to self-occlusions and dependence on articulations, while also needing to\nretain enough flexibility to accommodate a variety of clothes and poses. Our\nkey observation is that the vertices of standardized human meshes (such as\nSMPL) can provide an adequate spatial density and approximate initial position\nfor the Gaussians. We can then train a transformer model to jointly predict\ncomparatively small adjustments to these positions, as well as the other 3DGS\nattributes and the SMPL parameters. We show empirically that this combination\n(using only multi-view supervision) can achieve near real-time inference of 3D\nhuman models from a single image without expensive diffusion models or 3D\npoints supervision, thus making it ideal for the sport industry at any level.\nMore importantly, rendering is an effective auxiliary objective to refine 3D\npose estimation by accounting for clothes and other geometric variations. The\ncode is available at https://github.com/prosperolo/GST.\n", "date": "2024-09-06", "categories": []}
{"id": "2312.01632", "title": "GaussianHead: High-fidelity Head Avatars with Learnable Gaussian\n  Derivation", "abstract": "  Constructing vivid 3D head avatars for given subjects and realizing a series\nof animations on them is valuable yet challenging. This paper presents\nGaussianHead, which models the actional human head with anisotropic 3D\nGaussians. In our framework, a motion deformation field and multi-resolution\ntri-plane are constructed respectively to deal with the head's dynamic geometry\nand complex texture. Notably, we impose an exclusive derivation scheme on each\nGaussian, which generates its multiple doppelgangers through a set of learnable\nparameters for position transformation. With this design, we can compactly and\naccurately encode the appearance information of Gaussians, even those fitting\nthe head's particular components with sophisticated structures. In addition, an\ninherited derivation strategy for newly added Gaussians is adopted to\nfacilitate training acceleration. Extensive experiments show that our method\ncan produce high-fidelity renderings, outperforming state-of-the-art approaches\nin reconstruction, cross-identity reenactment, and novel view synthesis tasks.\nOur code is available at: https://github.com/chiehwangs/gaussian-head.\n", "date": "2023-12-04", "categories": []}
{"id": "2411.19824", "title": "SAT-HMR: Real-Time Multi-Person 3D Mesh Estimation via Scale-Adaptive\n  Tokens", "abstract": "  We propose a one-stage framework for real-time multi-person 3D human mesh\nestimation from a single RGB image. While current one-stage methods, which\nfollow a DETR-style pipeline, achieve state-of-the-art (SOTA) performance with\nhigh-resolution inputs, we observe that this particularly benefits the\nestimation of individuals in smaller scales of the image (e.g., those far from\nthe camera), but at the cost of significantly increased computation overhead.\nTo address this, we introduce scale-adaptive tokens that are dynamically\nadjusted based on the relative scale of each individual in the image within the\nDETR framework. Specifically, individuals in smaller scales are processed at\nhigher resolutions, larger ones at lower resolutions, and background regions\nare further distilled. These scale-adaptive tokens more efficiently encode the\nimage features, facilitating subsequent decoding to regress the human mesh,\nwhile allowing the model to allocate computational resources more effectively\nand focus on more challenging cases. Experiments show that our method preserves\nthe accuracy benefits of high-resolution processing while substantially\nreducing computational cost, achieving real-time inference with performance\ncomparable to SOTA methods.\n", "date": "2024-11-29", "categories": []}
{"id": "2409.17671", "title": "Leveraging Anthropometric Measurements to Improve Human Mesh Estimation\n  and Ensure Consistent Body Shapes", "abstract": "  The basic body shape (i.e., the body shape in T-pose) of a person does not\nchange within a single video. However, most SOTA human mesh estimation (HME)\nmodels output a slightly different, thus inconsistent basic body shape for each\nvideo frame. Furthermore, we find that SOTA 3D human pose estimation (HPE)\nmodels outperform HME models regarding the precision of the estimated 3D\nkeypoint positions. We solve the problem of inconsistent body shapes by\nleveraging anthropometric measurements like taken by tailors from humans. We\ncreate a model called A2B that converts given anthropometric measurements to\nbasic body shape parameters of human mesh models. We obtain superior and\nconsistent human meshes by combining the A2B model results with the keypoints\nof 3D HPE models using inverse kinematics. We evaluate our approach on\nchallenging datasets like ASPset or fit3D, where we can lower the MPJPE by over\n30 mm compared to SOTA HME models. Further, replacing estimates of the body\nshape parameters from existing HME models with A2B results not only increases\nthe performance of these HME models, but also guarantees consistent body\nshapes.\n", "date": "2024-09-26", "categories": []}
{"id": "2412.10209", "title": "GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view\n  Diffusion", "abstract": "  We propose a novel approach for reconstructing animatable 3D Gaussian avatars\nfrom monocular videos captured by commodity devices like smartphones.\nPhotorealistic 3D head avatar reconstruction from such recordings is\nchallenging due to limited observations, which leaves unobserved regions\nunder-constrained and can lead to artifacts in novel views. To address this\nproblem, we introduce a multi-view head diffusion model, leveraging its priors\nto fill in missing regions and ensure view consistency in Gaussian splatting\nrenderings. To enable precise viewpoint control, we use normal maps rendered\nfrom FLAME-based head reconstruction, which provides pixel-aligned inductive\nbiases. We also condition the diffusion model on VAE features extracted from\nthe input image to preserve facial identity and appearance details. For\nGaussian avatar reconstruction, we distill multi-view diffusion priors by using\niteratively denoised images as pseudo-ground truths, effectively mitigating\nover-saturation issues. To further improve photorealism, we apply latent\nupsampling priors to refine the denoised latent before decoding it into an\nimage. We evaluate our method on the NeRSemble dataset, showing that GAF\noutperforms previous state-of-the-art methods in novel view synthesis.\nFurthermore, we demonstrate higher-fidelity avatar reconstructions from\nmonocular videos captured on commodity devices.\n", "date": "2024-12-13", "categories": []}
{"id": "2504.07949", "title": "InteractAvatar: Modeling Hand-Face Interaction in Photorealistic Avatars\n  with Deformable Gaussians", "abstract": "  With the rising interest from the community in digital avatars coupled with\nthe importance of expressions and gestures in communication, modeling natural\navatar behavior remains an important challenge across many industries such as\nteleconferencing, gaming, and AR/VR. Human hands are the primary tool for\ninteracting with the environment and essential for realistic human behavior\nmodeling, yet existing 3D hand and head avatar models often overlook the\ncrucial aspect of hand-body interactions, such as between hand and face. We\npresent InteracttAvatar, the first model to faithfully capture the\nphotorealistic appearance of dynamic hand and non-rigid hand-face interactions.\nOur novel Dynamic Gaussian Hand model, combining template model and 3D Gaussian\nSplatting as well as a dynamic refinement module, captures pose-dependent\nchange, e.g. the fine wrinkles and complex shadows that occur during\narticulation. Importantly, our hand-face interaction module models the subtle\ngeometry and appearance dynamics that underlie common gestures. Through\nexperiments of novel view synthesis, self reenactment and cross-identity\nreenactment, we demonstrate that InteracttAvatar can reconstruct hand and\nhand-face interactions from monocular or multiview videos with high-fidelity\ndetails and be animated with novel poses.\n", "date": "2025-04-10", "categories": []}
{"id": "2411.19942", "title": "FreeCloth: Free-form Generation Enhances Challenging Clothed Human\n  Modeling", "abstract": "  Achieving realistic animated human avatars requires accurate modeling of\npose-dependent clothing deformations. Existing learning-based methods heavily\nrely on the Linear Blend Skinning (LBS) of minimally-clothed human models like\nSMPL to model deformation. However, they struggle to handle loose clothing,\nsuch as long dresses, where the canonicalization process becomes ill-defined\nwhen the clothing is far from the body, leading to disjointed and fragmented\nresults. To overcome this limitation, we propose FreeCloth, a novel hybrid\nframework to model challenging clothed humans. Our core idea is to use\ndedicated strategies to model different regions, depending on whether they are\nclose to or distant from the body. Specifically, we segment the human body into\nthree categories: unclothed, deformed, and generated. We simply replicate\nunclothed regions that require no deformation. For deformed regions close to\nthe body, we leverage LBS to handle the deformation. As for the generated\nregions, which correspond to loose clothing areas, we introduce a novel\nfree-form, part-aware generator to model them, as they are less affected by\nmovements. This free-form generation paradigm brings enhanced flexibility and\nexpressiveness to our hybrid framework, enabling it to capture the intricate\ngeometric details of challenging loose clothing, such as skirts and dresses.\nExperimental results on the benchmark dataset featuring loose clothing\ndemonstrate that FreeCloth achieves state-of-the-art performance with superior\nvisual fidelity and realism, particularly in the most challenging cases.\n", "date": "2024-11-29", "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": "2502.17796", "title": "LAM: Large Avatar Model for One-shot Animatable Gaussian Head", "abstract": "  We present LAM, an innovative Large Avatar Model for animatable Gaussian head\nreconstruction from a single image. Unlike previous methods that require\nextensive training on captured video sequences or rely on auxiliary neural\nnetworks for animation and rendering during inference, our approach generates\nGaussian heads that are immediately animatable and renderable. Specifically,\nLAM creates an animatable Gaussian head in a single forward pass, enabling\nreenactment and rendering without additional networks or post-processing steps.\nThis capability allows for seamless integration into existing rendering\npipelines, ensuring real-time animation and rendering across a wide range of\nplatforms, including mobile phones. The centerpiece of our framework is the\ncanonical Gaussian attributes generator, which utilizes FLAME canonical points\nas queries. These points interact with multi-scale image features through a\nTransformer to accurately predict Gaussian attributes in the canonical space.\nThe reconstructed canonical Gaussian avatar can then be animated utilizing\nstandard linear blend skinning (LBS) with corrective blendshapes as the FLAME\nmodel did and rendered in real-time on various platforms. Our experimental\nresults demonstrate that LAM outperforms state-of-the-art methods on existing\nbenchmarks. Our code and video are available at\nhttps://aigc3d.github.io/projects/LAM/\n", "date": "2025-02-25", "categories": []}
{"id": "2503.19207", "title": "FRESA: Feedforward Reconstruction of Personalized Skinned Avatars from\n  Few Images", "abstract": "  We present a novel method for reconstructing personalized 3D human avatars\nwith realistic animation from only a few images. Due to the large variations in\nbody shapes, poses, and cloth types, existing methods mostly require hours of\nper-subject optimization during inference, which limits their practical\napplications. In contrast, we learn a universal prior from over a thousand\nclothed humans to achieve instant feedforward generation and zero-shot\ngeneralization. Specifically, instead of rigging the avatar with shared\nskinning weights, we jointly infer personalized avatar shape, skinning weights,\nand pose-dependent deformations, which effectively improves overall geometric\nfidelity and reduces deformation artifacts. Moreover, to normalize pose\nvariations and resolve coupled ambiguity between canonical shapes and skinning\nweights, we design a 3D canonicalization process to produce pixel-aligned\ninitial conditions, which helps to reconstruct fine-grained geometric details.\nWe then propose a multi-frame feature aggregation to robustly reduce artifacts\nintroduced in canonicalization and fuse a plausible avatar preserving\nperson-specific identities. Finally, we train the model in an end-to-end\nframework on a large-scale capture dataset, which contains diverse human\nsubjects paired with high-quality 3D scans. Extensive experiments show that our\nmethod generates more authentic reconstruction and animation than\nstate-of-the-arts, and can be directly generalized to inputs from casually\ntaken phone photos. Project page and code is available at\nhttps://github.com/rongakowang/FRESA.\n", "date": "2025-03-24", "categories": []}
{"id": "2504.03006", "title": "DiSRT-In-Bed: Diffusion-Based Sim-to-Real Transfer Framework for In-Bed\n  Human Mesh Recovery", "abstract": "  In-bed human mesh recovery can be crucial and enabling for several healthcare\napplications, including sleep pattern monitoring, rehabilitation support, and\npressure ulcer prevention. However, it is difficult to collect large real-world\nvisual datasets in this domain, in part due to privacy and expense constraints,\nwhich in turn presents significant challenges for training and deploying deep\nlearning models. Existing in-bed human mesh estimation methods often rely\nheavily on real-world data, limiting their ability to generalize across\ndifferent in-bed scenarios, such as varying coverings and environmental\nsettings. To address this, we propose a Sim-to-Real Transfer Framework for\nin-bed human mesh recovery from overhead depth images, which leverages\nlarge-scale synthetic data alongside limited or no real-world samples. We\nintroduce a diffusion model that bridges the gap between synthetic data and\nreal data to support generalization in real-world in-bed pose and body\ninference scenarios. Extensive experiments and ablation studies validate the\neffectiveness of our framework, demonstrating significant improvements in\nrobustness and adaptability across diverse healthcare scenarios.\n", "date": "2025-04-03", "categories": []}
{"id": "2504.01559", "title": "RealityAvatar: Towards Realistic Loose Clothing Modeling in Animatable\n  3D Gaussian Avatars", "abstract": "  Modeling animatable human avatars from monocular or multi-view videos has\nbeen widely studied, with recent approaches leveraging neural radiance fields\n(NeRFs) or 3D Gaussian Splatting (3DGS) achieving impressive results in\nnovel-view and novel-pose synthesis. However, existing methods often struggle\nto accurately capture the dynamics of loose clothing, as they primarily rely on\nglobal pose conditioning or static per-frame representations, leading to\noversmoothing and temporal inconsistencies in non-rigid regions. To address\nthis, We propose RealityAvatar, an efficient framework for high-fidelity\ndigital human modeling, specifically targeting loosely dressed avatars. Our\nmethod leverages 3D Gaussian Splatting to capture complex clothing deformations\nand motion dynamics while ensuring geometric consistency. By incorporating a\nmotion trend module and a latentbone encoder, we explicitly model\npose-dependent deformations and temporal variations in clothing behavior.\nExtensive experiments on benchmark datasets demonstrate the effectiveness of\nour approach in capturing fine-grained clothing deformations and motion-driven\nshape variations. Our method significantly enhances structural fidelity and\nperceptual quality in dynamic human reconstruction, particularly in non-rigid\nregions, while achieving better consistency across temporal frames.\n", "date": "2025-04-02", "categories": []}
{"id": "2407.04545", "title": "Gaussian Eigen Models for Human Heads", "abstract": "  Current personalized neural head avatars face a trade-off: lightweight models\nlack detail and realism, while high-quality, animatable avatars require\nsignificant computational resources, making them unsuitable for commodity\ndevices. To address this gap, we introduce Gaussian Eigen Models (GEM), which\nprovide high-quality, lightweight, and easily controllable head avatars. GEM\nutilizes 3D Gaussian primitives for representing the appearance combined with\nGaussian splatting for rendering. Building on the success of mesh-based 3D\nmorphable face models (3DMM), we define GEM as an ensemble of linear eigenbases\nfor representing the head appearance of a specific subject. In particular, we\nconstruct linear bases to represent the position, scale, rotation, and opacity\nof the 3D Gaussians. This allows us to efficiently generate Gaussian primitives\nof a specific head shape by a linear combination of the basis vectors, only\nrequiring a low-dimensional parameter vector that contains the respective\ncoefficients. We propose to construct these linear bases (GEM) by distilling\nhigh-quality compute-intense CNN-based Gaussian avatar models that can generate\nexpression-dependent appearance changes like wrinkles. These high-quality\nmodels are trained on multi-view videos of a subject and are distilled using a\nseries of principal component analyses. Once we have obtained the bases that\nrepresent the animatable appearance space of a specific human, we learn a\nregressor that takes a single RGB image as input and predicts the\nlow-dimensional parameter vector that corresponds to the shown facial\nexpression. In a series of experiments, we compare GEM's self-reenactment and\ncross-person reenactment results to state-of-the-art 3D avatar methods,\ndemonstrating GEM's higher visual quality and better generalization to new\nexpressions.\n", "date": "2024-07-05", "categories": []}
{"id": "2411.11903", "title": "DiHuR: Diffusion-Guided Generalizable Human Reconstruction", "abstract": "  We introduce DiHuR, a novel Diffusion-guided model for generalizable Human 3D\nReconstruction and view synthesis from sparse, minimally overlapping images.\nWhile existing generalizable human radiance fields excel at novel view\nsynthesis, they often struggle with comprehensive 3D reconstruction. Similarly,\ndirectly optimizing implicit Signed Distance Function (SDF) fields from\nsparse-view images typically yields poor results due to limited overlap. To\nenhance 3D reconstruction quality, we propose using learnable tokens associated\nwith SMPL vertices to aggregate sparse view features and then to guide SDF\nprediction. These tokens learn a generalizable prior across different\nidentities in training datasets, leveraging the consistent projection of SMPL\nvertices onto similar semantic areas across various human identities. This\nconsistency enables effective knowledge transfer to unseen identities during\ninference. Recognizing SMPL's limitations in capturing clothing details, we\nincorporate a diffusion model as an additional prior to fill in missing\ninformation, particularly for complex clothing geometries. Our method\nintegrates two key priors in a coherent manner: the prior from generalizable\nfeed-forward models and the 2D diffusion prior, and it requires only multi-view\nimage training, without 3D supervision. DiHuR demonstrates superior performance\nin both within-dataset and cross-dataset generalization settings, as validated\non THuman, ZJU-MoCap, and HuMMan datasets compared to existing methods.\n", "date": "2024-11-16", "categories": []}
{"id": "2501.06903", "title": "Synthetic Prior for Few-Shot Drivable Head Avatar Inversion", "abstract": "  We present SynShot, a novel method for the few-shot inversion of a drivable\nhead avatar based on a synthetic prior. We tackle three major challenges.\nFirst, training a controllable 3D generative network requires a large number of\ndiverse sequences, for which pairs of images and high-quality tracked meshes\nare not always available. Second, the use of real data is strictly regulated\n(e.g., under the General Data Protection Regulation, which mandates frequent\ndeletion of models and data to accommodate a situation when a participant's\nconsent is withdrawn). Synthetic data, free from these constraints, is an\nappealing alternative. Third, state-of-the-art monocular avatar models struggle\nto generalize to new views and expressions, lacking a strong prior and often\noverfitting to a specific viewpoint distribution. Inspired by machine learning\nmodels trained solely on synthetic data, we propose a method that learns a\nprior model from a large dataset of synthetic heads with diverse identities,\nexpressions, and viewpoints. With few input images, SynShot fine-tunes the\npretrained synthetic prior to bridge the domain gap, modeling a photorealistic\nhead avatar that generalizes to novel expressions and viewpoints. We model the\nhead avatar using 3D Gaussian splatting and a convolutional encoder-decoder\nthat outputs Gaussian parameters in UV texture space. To account for the\ndifferent modeling complexities over parts of the head (e.g., skin vs hair), we\nembed the prior with explicit control for upsampling the number of per-part\nprimitives. Compared to SOTA monocular and GAN-based methods, SynShot\nsignificantly improves novel view and expression synthesis.\n", "date": "2025-01-12", "categories": []}
{"id": "2503.21886", "title": "Refined Geometry-guided Head Avatar Reconstruction from Monocular RGB\n  Video", "abstract": "  High-fidelity reconstruction of head avatars from monocular videos is highly\ndesirable for virtual human applications, but it remains a challenge in the\nfields of computer graphics and computer vision. In this paper, we propose a\ntwo-phase head avatar reconstruction network that incorporates a refined 3D\nmesh representation. Our approach, in contrast to existing methods that rely on\ncoarse template-based 3D representations derived from 3DMM, aims to learn a\nrefined mesh representation suitable for a NeRF that captures complex facial\nnuances. In the first phase, we train 3DMM-stored NeRF with an initial mesh to\nutilize geometric priors and integrate observations across frames using a\nconsistent set of latent codes. In the second phase, we leverage a novel mesh\nrefinement procedure based on an SDF constructed from the density field of the\ninitial NeRF. To mitigate the typical noise in the NeRF density field without\ncompromising the features of the 3DMM, we employ Laplace smoothing on the\ndisplacement field. Subsequently, we apply a second-phase training with these\nrefined meshes, directing the learning process of the network towards capturing\nintricate facial details. Our experiments demonstrate that our method further\nenhances the NeRF rendering based on the initial mesh and achieves performance\nsuperior to state-of-the-art methods in reconstructing high-fidelity head\navatars with such input.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.21268", "title": "ClimbingCap: Multi-Modal Dataset and Method for Rock Climbing in World\n  Coordinate", "abstract": "  Human Motion Recovery (HMR) research mainly focuses on ground-based motions\nsuch as running. The study on capturing climbing motion, an off-ground motion,\nis sparse. This is partly due to the limited availability of climbing motion\ndatasets, especially large-scale and challenging 3D labeled datasets. To\naddress the insufficiency of climbing motion datasets, we collect AscendMotion,\na large-scale well-annotated, and challenging climbing motion dataset. It\nconsists of 412k RGB, LiDAR frames, and IMU measurements, including the\nchallenging climbing motions of 22 skilled climbing coaches across 12 different\nrock walls. Capturing the climbing motions is challenging as it requires\nprecise recovery of not only the complex pose but also the global position of\nclimbers. Although multiple global HMR methods have been proposed, they cannot\nfaithfully capture climbing motions. To address the limitations of HMR methods\nfor climbing, we propose ClimbingCap, a motion recovery method that\nreconstructs continuous 3D human climbing motion in a global coordinate system.\nOne key insight is to use the RGB and LiDAR modalities to separately\nreconstruct motions in camera coordinates and global coordinates and to\noptimize them jointly. We demonstrate the quality of the AscendMotion dataset\nand present promising results from ClimbingCap. The AscendMotion dataset and\nsource code release publicly at \\href{this\nlink}{http://www.lidarhumanmotion.net/climbingcap/}\n", "date": "2025-03-27", "categories": []}
{"id": "2503.21751", "title": "Reconstructing Humans with a Biomechanically Accurate Skeleton", "abstract": "  In this paper, we introduce a method for reconstructing 3D humans from a\nsingle image using a biomechanically accurate skeleton model. To achieve this,\nwe train a transformer that takes an image as input and estimates the\nparameters of the model. Due to the lack of training data for this task, we\nbuild a pipeline to produce pseudo ground truth model parameters for single\nimages and implement a training procedure that iteratively refines these pseudo\nlabels. Compared to state-of-the-art methods for 3D human mesh recovery, our\nmodel achieves competitive performance on standard benchmarks, while it\nsignificantly outperforms them in settings with extreme 3D poses and\nviewpoints. Additionally, we show that previous reconstruction methods\nfrequently violate joint angle limits, leading to unnatural rotations. In\ncontrast, our approach leverages the biomechanically plausible degrees of\nfreedom making more realistic joint rotation estimates. We validate our\napproach across multiple human pose estimation benchmarks. We make the code,\nmodels and data available at: https://isshikihugh.github.io/HSMR/\n", "date": "2025-03-27", "categories": []}
{"id": "2503.19906", "title": "AvatarArtist: Open-Domain 4D Avatarization", "abstract": "  This work focuses on open-domain 4D avatarization, with the purpose of\ncreating a 4D avatar from a portrait image in an arbitrary style. We select\nparametric triplanes as the intermediate 4D representation and propose a\npractical training paradigm that takes advantage of both generative adversarial\nnetworks (GANs) and diffusion models. Our design stems from the observation\nthat 4D GANs excel at bridging images and triplanes without supervision yet\nusually face challenges in handling diverse data distributions. A robust 2D\ndiffusion prior emerges as the solution, assisting the GAN in transferring its\nexpertise across various domains. The synergy between these experts permits the\nconstruction of a multi-domain image-triplane dataset, which drives the\ndevelopment of a general 4D avatar creator. Extensive experiments suggest that\nour model, AvatarArtist, is capable of producing high-quality 4D avatars with\nstrong robustness to various source image domains. The code, the data, and the\nmodels will be made publicly available to facilitate future studies.\n", "date": "2025-03-25", "categories": []}
{"id": "2210.12241", "title": "FIND: An Unsupervised Implicit 3D Model of Articulated Human Feet", "abstract": "  In this paper we present a high fidelity and articulated 3D human foot model.\nThe model is parameterised by a disentangled latent code in terms of shape,\ntexture and articulated pose. While high fidelity models are typically created\nwith strong supervision such as 3D keypoint correspondences or\npre-registration, we focus on the difficult case of little to no annotation. To\nthis end, we make the following contributions: (i) we develop a Foot Implicit\nNeural Deformation field model, named FIND, capable of tailoring explicit\nmeshes at any resolution i.e. for low or high powered devices; (ii) an approach\nfor training our model in various modes of weak supervision with progressively\nbetter disentanglement as more labels, such as pose categories, are provided;\n(iii) a novel unsupervised part-based loss for fitting our model to 2D images\nwhich is better than traditional photometric or silhouette losses; (iv)\nfinally, we release a new dataset of high resolution 3D human foot scans,\nFoot3D. On this dataset, we show our model outperforms a strong PCA\nimplementation trained on the same data in terms of shape quality and part\ncorrespondences, and that our novel unsupervised part-based loss improves\ninference on images.\n", "date": "2022-10-21", "categories": []}
{"id": "2412.14963", "title": "IDOL: Instant Photorealistic 3D Human Creation from a Single Image", "abstract": "  Creating a high-fidelity, animatable 3D full-body avatar from a single image\nis a challenging task due to the diverse appearance and poses of humans and the\nlimited availability of high-quality training data. To achieve fast and\nhigh-quality human reconstruction, this work rethinks the task from the\nperspectives of dataset, model, and representation. First, we introduce a\nlarge-scale HUman-centric GEnerated dataset, HuGe100K, consisting of 100K\ndiverse, photorealistic sets of human images. Each set contains 24-view frames\nin specific human poses, generated using a pose-controllable\nimage-to-multi-view model. Next, leveraging the diversity in views, poses, and\nappearances within HuGe100K, we develop a scalable feed-forward transformer\nmodel to predict a 3D human Gaussian representation in a uniform space from a\ngiven human image. This model is trained to disentangle human pose, body shape,\nclothing geometry, and texture. The estimated Gaussians can be animated without\npost-processing. We conduct comprehensive experiments to validate the\neffectiveness of the proposed dataset and method. Our model demonstrates the\nability to efficiently reconstruct photorealistic humans at 1K resolution from\na single input image using a single GPU instantly. Additionally, it seamlessly\nsupports various applications, as well as shape and texture editing tasks.\nProject page: https://yiyuzhuang.github.io/IDOL/.\n", "date": "2024-12-19", "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": "2503.19373", "title": "DeClotH: Decomposable 3D Cloth and Human Body Reconstruction from a\n  Single Image", "abstract": "  Most existing methods of 3D clothed human reconstruction from a single image\ntreat the clothed human as a single object without distinguishing between cloth\nand human body. In this regard, we present DeClotH, which separately\nreconstructs 3D cloth and human body from a single image. This task remains\nlargely unexplored due to the extreme occlusion between cloth and the human\nbody, making it challenging to infer accurate geometries and textures.\nMoreover, while recent 3D human reconstruction methods have achieved impressive\nresults using text-to-image diffusion models, directly applying such an\napproach to this problem often leads to incorrect guidance, particularly in\nreconstructing 3D cloth. To address these challenges, we propose two core\ndesigns in our framework. First, to alleviate the occlusion issue, we leverage\n3D template models of cloth and human body as regularizations, which provide\nstrong geometric priors to prevent erroneous reconstruction by the occlusion.\nSecond, we introduce a cloth diffusion model specifically designed to provide\ncontextual information about cloth appearance, thereby enhancing the\nreconstruction of 3D cloth. Qualitative and quantitative experiments\ndemonstrate that our proposed approach is highly effective in reconstructing\nboth 3D cloth and the human body. More qualitative results are provided at\nhttps://hygenie1228.github.io/DeClotH/.\n", "date": "2025-03-25", "categories": []}
{"id": "2409.10141", "title": "PSHuman: Photorealistic Single-image 3D Human Reconstruction using\n  Cross-Scale Multiview Diffusion and Explicit Remeshing", "abstract": "  Detailed and photorealistic 3D human modeling is essential for various\napplications and has seen tremendous progress. However, full-body\nreconstruction from a monocular RGB image remains challenging due to the\nill-posed nature of the problem and sophisticated clothing topology with\nself-occlusions. In this paper, we propose PSHuman, a novel framework that\nexplicitly reconstructs human meshes utilizing priors from the multiview\ndiffusion model. It is found that directly applying multiview diffusion on\nsingle-view human images leads to severe geometric distortions, especially on\ngenerated faces. To address it, we propose a cross-scale diffusion that models\nthe joint probability distribution of global full-body shape and local facial\ncharacteristics, enabling detailed and identity-preserved novel-view generation\nwithout any geometric distortion. Moreover, to enhance cross-view body shape\nconsistency of varied human poses, we condition the generative model on\nparametric models like SMPL-X, which provide body priors and prevent unnatural\nviews inconsistent with human anatomy. Leveraging the generated multi-view\nnormal and color images, we present SMPLX-initialized explicit human carving to\nrecover realistic textured human meshes efficiently. Extensive experimental\nresults and quantitative evaluations on CAPE and THuman2.1 datasets demonstrate\nPSHumans superiority in geometry details, texture fidelity, and generalization\ncapability.\n", "date": "2024-09-16", "categories": []}
{"id": "2503.00068", "title": "PI-HMR: Towards Robust In-bed Temporal Human Shape Reconstruction with\n  Contact Pressure Sensing", "abstract": "  Long-term in-bed monitoring benefits automatic and real-time health\nmanagement within healthcare, and the advancement of human shape reconstruction\ntechnologies further enhances the representation and visualization of users'\nactivity patterns. However, existing technologies are primarily based on visual\ncues, facing serious challenges in non-light-of-sight and privacy-sensitive\nin-bed scenes. Pressure-sensing bedsheets offer a promising solution for\nreal-time motion reconstruction. Yet, limited exploration in model designs and\ndata have hindered its further development. To tackle these issues, we propose\na general framework that bridges gaps in data annotation and model design.\nFirstly, we introduce SMPLify-IB, an optimization method that overcomes the\ndepth ambiguity issue in top-view scenarios through gravity constraints,\nenabling generating high-quality 3D human shape annotations for in-bed\ndatasets. Then we present PI-HMR, a temporal-based human shape estimator to\nregress meshes from pressure sequences. By integrating multi-scale feature\nfusion with high-pressure distribution and spatial position priors, PI-HMR\noutperforms SOTA methods with 17.01mm Mean-Per-Joint-Error decrease. This work\nprovides a whole\n", "date": "2025-02-27", "categories": []}
{"id": "2411.15604", "title": "FATE: Full-head Gaussian Avatar with Textural Editing from Monocular\n  Video", "abstract": "  Reconstructing high-fidelity, animatable 3D head avatars from effortlessly\ncaptured monocular videos is a pivotal yet formidable challenge. Although\nsignificant progress has been made in rendering performance and manipulation\ncapabilities, notable challenges remain, including incomplete reconstruction\nand inefficient Gaussian representation. To address these challenges, we\nintroduce FATE, a novel method for reconstructing an editable full-head avatar\nfrom a single monocular video. FATE integrates a sampling-based densification\nstrategy to ensure optimal positional distribution of points, improving\nrendering efficiency. A neural baking technique is introduced to convert\ndiscrete Gaussian representations into continuous attribute maps, facilitating\nintuitive appearance editing. Furthermore, we propose a universal completion\nframework to recover non-frontal appearance, culminating in a\n360$^\\circ$-renderable 3D head avatar. FATE outperforms previous approaches in\nboth qualitative and quantitative evaluations, achieving state-of-the-art\nperformance. To the best of our knowledge, FATE is the first animatable and\n360$^\\circ$ full-head monocular reconstruction method for a 3D head avatar.\n", "date": "2024-11-23", "categories": []}
{"id": "2503.14943", "title": "3D Engine-ready Photorealistic Avatars via Dynamic Textures", "abstract": "  As the digital and physical worlds become more intertwined, there has been a\nlot of interest in digital avatars that closely resemble their real-world\ncounterparts. Current digitization methods used in 3D production pipelines\nrequire costly capture setups, making them impractical for mass usage among\ncommon consumers. Recent academic literature has found success in\nreconstructing humans from limited data using implicit representations (e.g.,\nvoxels used in NeRFs), which are able to produce impressive videos. However,\nthese methods are incompatible with traditional rendering pipelines, making it\ndifficult to use them in applications such as games. In this work, we propose\nan end-to-end pipeline that builds explicitly-represented photorealistic 3D\navatars using standard 3D assets. Our key idea is the use of\ndynamically-generated textures to enhance the realism and visually mask\ndeficiencies in the underlying mesh geometry. This allows for seamless\nintegration with current graphics pipelines while achieving comparable visual\nquality to state-of-the-art 3D avatar generation methods.\n", "date": "2025-03-19", "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": "2502.19739", "title": "LUCAS: Layered Universal Codec Avatars", "abstract": "  Photorealistic 3D head avatar reconstruction faces critical challenges in\nmodeling dynamic face-hair interactions and achieving cross-identity\ngeneralization, particularly during expressions and head movements. We present\nLUCAS, a novel Universal Prior Model (UPM) for codec avatar modeling that\ndisentangles face and hair through a layered representation. Unlike previous\nUPMs that treat hair as an integral part of the head, our approach separates\nthe modeling of the hairless head and hair into distinct branches. LUCAS is the\nfirst to introduce a mesh-based UPM, facilitating real-time rendering on\ndevices. Our layered representation also improves the anchor geometry for\nprecise and visually appealing Gaussian renderings. Experimental results\nindicate that LUCAS outperforms existing single-mesh and Gaussian-based avatar\nmodels in both quantitative and qualitative assessments, including evaluations\non held-out subjects in zero-shot driving scenarios. LUCAS demonstrates\nsuperior dynamic performance in managing head pose changes, expression\ntransfer, and hairstyle variations, thereby advancing the state-of-the-art in\n3D head avatar reconstruction.\n", "date": "2025-02-27", "categories": []}
{"id": "2312.06085", "title": "SFDM: Robust Decomposition of Geometry and Reflectance for Realistic\n  Face Rendering from Sparse-view Images", "abstract": "  In this study, we introduce a novel two-stage technique for decomposing and\nreconstructing facial features from sparse-view images, a task made challenging\nby the unique geometry and complex skin reflectance of each individual. To\nsynthesize 3D facial models more realistically, we endeavor to decouple key\nfacial attributes from the RGB color, including geometry, diffuse reflectance,\nand specular reflectance. Specifically, we design a Sparse-view Face\nDecomposition Model (SFDM): 1) In the first stage, we create a general facial\ntemplate from a wide array of individual faces, encapsulating essential\ngeometric and reflectance characteristics. 2) Guided by this template, we\nrefine a specific facial model for each individual in the second stage,\nconsidering the interaction between geometry and reflectance, as well as the\neffects of subsurface scattering on the skin. With these advances, our method\ncan reconstruct high-quality facial representations from as few as three\nimages. The comprehensive evaluation and comparison reveal that our approach\noutperforms existing methods by effectively disentangling geometric and\nreflectance components, significantly enhancing the quality of synthesized\nnovel views, and paving the way for applications in facial relighting and\nreflectance editing.\n", "date": "2023-12-11", "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": "2503.12492", "title": "Geometry-Aware Face Reconstruction Under Occluded Scenes", "abstract": "  Recently, deep learning-based 3D face reconstruction methods have\ndemonstrated promising advancements in terms of quality and efficiency.\nNevertheless, these techniques face challenges in effectively handling occluded\nscenes and fail to capture intricate geometric facial details. Inspired by the\nprinciples of GANs and bump mapping, we have successfully addressed these\nissues. Our approach aims to deliver comprehensive 3D facial reconstructions,\neven in the presence of occlusions.While maintaining the overall shape's\nrobustness, we introduce a mid-level shape refinement to the fundamental\nstructure. Furthermore, we illustrate how our method adeptly extends to\ngenerate plausible details for obscured facial regions. We offer numerous\nexamples that showcase the effectiveness of our framework in producing\nrealistic results, where traditional methods often struggle. To substantiate\nthe superior adaptability of our approach, we have conducted extensive\nexperiments in the context of general 3D face reconstruction tasks, serving as\nconcrete evidence of its regulatory prowess compared to manual occlusion\nremoval methods.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12494", "title": "Learning Contour-Guided 3D Face Reconstruction with Occlusions", "abstract": "  Recently, deep learning-based 3D face reconstruction methods have\ndemonstrated promising advancements in terms of quality and efficiency.\nNevertheless, these techniques face challenges in effectively handling occluded\nscenes and fail to capture intricate geometric facial details. Inspired by the\nprinciples of GANs and bump mapping, we have successfully addressed these\nissues. Our approach aims to deliver comprehensive 3D facial reconstructions,\neven in the presence of occlusions.While maintaining the overall shape's\nrobustness, we introduce a mid-level shape refinement to the fundamental\nstructure. Furthermore, we illustrate how our method adeptly extends to\ngenerate plausible details for obscured facial regions. We offer numerous\nexamples that showcase the effectiveness of our framework in producing\nrealistic results, where traditional methods often struggle. To substantiate\nthe superior adaptability of our approach, we have conducted extensive\nexperiments in the context of general 3D face reconstruction tasks, serving as\nconcrete evidence of its regulatory prowess compared to manual occlusion\nremoval methods.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12751", "title": "R3-Avatar: Record and Retrieve Temporal Codebook for Reconstructing\n  Photorealistic Human Avatars", "abstract": "  We present R3-Avatar, incorporating a temporal codebook, to overcome the\ninability of human avatars to be both animatable and of high-fidelity rendering\nquality. Existing video-based reconstruction of 3D human avatars either focuses\nsolely on rendering, lacking animation support, or learns a pose-appearance\nmapping for animating, which degrades under limited training poses or complex\nclothing. In this paper, we adopt a \"record-retrieve-reconstruct\" strategy that\nensures high-quality rendering from novel views while mitigating degradation in\nnovel poses. Specifically, disambiguating timestamps record temporal appearance\nvariations in a codebook, ensuring high-fidelity novel-view rendering, while\nnovel poses retrieve corresponding timestamps by matching the most similar\ntraining poses for augmented appearance. Our R3-Avatar outperforms cutting-edge\nvideo-based human avatar reconstruction, particularly in overcoming visual\nquality degradation in extreme scenarios with limited training human poses and\ncomplex clothing.\n", "date": "2025-03-17", "categories": []}
{"id": "2503.12886", "title": "RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head\n  Avatars", "abstract": "  We present Reduced Gaussian Blendshapes Avatar (RGBAvatar), a method for\nreconstructing photorealistic, animatable head avatars at speeds sufficient for\non-the-fly reconstruction. Unlike prior approaches that utilize linear bases\nfrom 3D morphable models (3DMM) to model Gaussian blendshapes, our method maps\ntracked 3DMM parameters into reduced blendshape weights with an MLP, leading to\na compact set of blendshape bases. The learned compact base composition\neffectively captures essential facial details for specific individuals, and\ndoes not rely on the fixed base composition weights of 3DMM, leading to\nenhanced reconstruction quality and higher efficiency. To further expedite the\nreconstruction process, we develop a novel color initialization estimation\nmethod and a batch-parallel Gaussian rasterization process, achieving\nstate-of-the-art quality with training throughput of about 630 images per\nsecond. Moreover, we propose a local-global sampling strategy that enables\ndirect on-the-fly reconstruction, immediately reconstructing the model as video\nstreams in real time while achieving quality comparable to offline settings.\nOur source code is available at https://github.com/gapszju/RGBAvatar.\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": "2404.04104", "title": "SMIRK: 3D Facial Expressions through Analysis-by-Neural-Synthesis", "abstract": "  While existing methods for 3D face reconstruction from in-the-wild images\nexcel at recovering the overall face shape, they commonly miss subtle, extreme,\nasymmetric, or rarely observed expressions. We improve upon these methods with\nSMIRK (Spatial Modeling for Image-based Reconstruction of Kinesics), which\nfaithfully reconstructs expressive 3D faces from images. We identify two key\nlimitations in existing methods: shortcomings in their self-supervised training\nformulation, and a lack of expression diversity in the training images. For\ntraining, most methods employ differentiable rendering to compare a predicted\nface mesh with the input image, along with a plethora of additional loss\nfunctions. This differentiable rendering loss not only has to provide\nsupervision to optimize for 3D face geometry, camera, albedo, and lighting,\nwhich is an ill-posed optimization problem, but the domain gap between\nrendering and input image further hinders the learning process. Instead, SMIRK\nreplaces the differentiable rendering with a neural rendering module that,\ngiven the rendered predicted mesh geometry, and sparsely sampled pixels of the\ninput image, generates a face image. As the neural rendering gets color\ninformation from sampled image pixels, supervising with neural rendering-based\nreconstruction loss can focus solely on the geometry. Further, it enables us to\ngenerate images of the input identity with varying expressions while training.\nThese are then utilized as input to the reconstruction model and used as\nsupervision with ground truth geometry. This effectively augments the training\ndata and enhances the generalization for diverse expressions. Our qualitative,\nquantitative and particularly our perceptual evaluations demonstrate that SMIRK\nachieves the new state-of-the art performance on accurate expression\nreconstruction. Project webpage: https://georgeretsi.github.io/smirk/.\n", "date": "2024-04-05", "categories": []}
{"id": "2409.04851", "title": "AdaptiveFusion: Adaptive Multi-Modal Multi-View Fusion for 3D Human Body\n  Reconstruction", "abstract": "  Recent advancements in sensor technology and deep learning have led to\nsignificant progress in 3D human body reconstruction. However, most existing\napproaches rely on data from a specific sensor, which can be unreliable due to\nthe inherent limitations of individual sensing modalities. Additionally,\nexisting multi-modal fusion methods generally require customized designs based\non the specific sensor combinations or setups, which limits the flexibility and\ngenerality of these methods. Furthermore, conventional point-image\nprojection-based and Transformer-based fusion networks are susceptible to the\ninfluence of noisy modalities and sensor poses. To address these limitations\nand achieve robust 3D human body reconstruction in various conditions, we\npropose AdaptiveFusion, a generic adaptive multi-modal multi-view fusion\nframework that can effectively incorporate arbitrary combinations of\nuncalibrated sensor inputs. By treating different modalities from various\nviewpoints as equal tokens, and our handcrafted modality sampling module by\nleveraging the inherent flexibility of Transformer models, AdaptiveFusion is\nable to cope with arbitrary numbers of inputs and accommodate noisy modalities\nwith only a single training network. Extensive experiments on large-scale human\ndatasets demonstrate the effectiveness of AdaptiveFusion in achieving\nhigh-quality 3D human body reconstruction in various environments. In addition,\nour method achieves superior accuracy compared to state-of-the-art fusion\nmethods.\n", "date": "2024-09-07", "categories": []}
{"id": "2503.10625", "title": "LHM: Large Animatable Human Reconstruction Model from a Single Image in\n  Seconds", "abstract": "  Animatable 3D human reconstruction from a single image is a challenging\nproblem due to the ambiguity in decoupling geometry, appearance, and\ndeformation. Recent advances in 3D human reconstruction mainly focus on static\nhuman modeling, and the reliance of using synthetic 3D scans for training\nlimits their generalization ability. Conversely, optimization-based video\nmethods achieve higher fidelity but demand controlled capture conditions and\ncomputationally intensive refinement processes. Motivated by the emergence of\nlarge reconstruction models for efficient static reconstruction, we propose LHM\n(Large Animatable Human Reconstruction Model) to infer high-fidelity avatars\nrepresented as 3D Gaussian splatting in a feed-forward pass. Our model\nleverages a multimodal transformer architecture to effectively encode the human\nbody positional features and image features with attention mechanism, enabling\ndetailed preservation of clothing geometry and texture. To further boost the\nface identity preservation and fine detail recovery, we propose a head feature\npyramid encoding scheme to aggregate multi-scale features of the head regions.\nExtensive experiments demonstrate that our LHM generates plausible animatable\nhuman in seconds without post-processing for face and hands, outperforming\nexisting methods in both reconstruction accuracy and generalization ability.\n", "date": "2025-03-13", "categories": []}
{"id": "2407.02165", "title": "WildAvatar: Learning In-the-wild 3D Avatars from the Web", "abstract": "  Existing research on avatar creation is typically limited to laboratory\ndatasets, which require high costs against scalability and exhibit insufficient\nrepresentation of the real world. On the other hand, the web abounds with\noff-the-shelf real-world human videos, but these videos vary in quality and\nrequire accurate annotations for avatar creation. To this end, we propose an\nautomatic annotating pipeline with filtering protocols to curate these humans\nfrom the web. Our pipeline surpasses state-of-the-art methods on the EMDB\nbenchmark, and the filtering protocols boost verification metrics on web\nvideos. We then curate WildAvatar, a web-scale in-the-wild human avatar\ncreation dataset extracted from YouTube, with $10000+$ different human subjects\nand scenes. WildAvatar is at least $10\\times$ richer than previous datasets for\n3D human avatar creation and closer to the real world. To explore its\npotential, we demonstrate the quality and generalizability of avatar creation\nmethods on WildAvatar. We will publicly release our code, data source links and\nannotations to push forward 3D human avatar creation and other related fields\nfor real-world applications.\n", "date": "2024-07-02", "categories": []}
{"id": "2410.01425", "title": "EVA-Gaussian: 3D Gaussian-based Real-time Human Novel View Synthesis\n  under Diverse Multi-view Camera Settings", "abstract": "  Feed-forward based 3D Gaussian Splatting methods have demonstrated\nexceptional capability in real-time novel view synthesis for human models.\nHowever, current approaches are confined to either dense viewpoint\nconfigurations or restricted image resolutions. These limitations hinder their\nflexibility in free-viewpoint rendering across a wide range of camera view\nangle discrepancies, and also restrict their ability to recover fine-grained\nhuman details in real time using commonly available GPUs. To address these\nchallenges, we propose a novel pipeline named EVA-Gaussian for 3D human novel\nview synthesis across diverse multi-view camera settings. Specifically, we\nfirst design an Efficient Cross-View Attention (EVA) module to effectively fuse\ncross-view information under high resolution inputs and sparse view settings,\nwhile minimizing temporal and computational overhead. Additionally, we\nintroduce a feature refinement mechianism to predict the attributes of the 3D\nGaussians and assign a feature value to each Gaussian, enabling the correction\nof artifacts caused by geometric inaccuracies in position estimation and\nenhancing overall visual fidelity. Experimental results on the THuman2.0 and\nTHumansit datasets showcase the superiority of EVA-Gaussian in rendering\nquality across diverse camera settings. Project page:\nhttps://zhenliuzju.github.io/huyingdong/EVA-Gaussian.\n", "date": "2024-10-02", "categories": []}
{"id": "2503.09293", "title": "Better Together: Unified Motion Capture and 3D Avatar Reconstruction", "abstract": "  We present Better Together, a method that simultaneously solves the human\npose estimation problem while reconstructing a photorealistic 3D human avatar\nfrom multi-view videos. While prior art usually solves these problems\nseparately, we argue that joint optimization of skeletal motion with a 3D\nrenderable body model brings synergistic effects, i.e. yields more precise\nmotion capture and improved visual quality of real-time rendering of avatars.\nTo achieve this, we introduce a novel animatable avatar with 3D Gaussians\nrigged on a personalized mesh and propose to optimize the motion sequence with\ntime-dependent MLPs that provide accurate and temporally consistent pose\nestimates. We first evaluate our method on highly challenging yoga poses and\ndemonstrate state-of-the-art accuracy on multi-view human pose estimation,\nreducing error by 35% on body joints and 45% on hand joints compared to\nkeypoint-based methods. At the same time, our method significantly boosts the\nvisual quality of animatable avatars (+2dB PSNR on novel view synthesis) on\ndiverse challenging subjects.\n", "date": "2025-03-12", "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": "2503.08218", "title": "MVD-HuGaS: Human Gaussians from a Single Image via 3D Human Multi-view\n  Diffusion Prior", "abstract": "  3D human reconstruction from a single image is a challenging problem and has\nbeen exclusively studied in the literature. Recently, some methods have\nresorted to diffusion models for guidance, optimizing a 3D representation via\nScore Distillation Sampling(SDS) or generating one back-view image for\nfacilitating reconstruction. However, these methods tend to produce\nunsatisfactory artifacts (\\textit{e.g.} flattened human structure or\nover-smoothing results caused by inconsistent priors from multiple views) and\nstruggle with real-world generalization in the wild. In this work, we present\n\\emph{MVD-HuGaS}, enabling free-view 3D human rendering from a single image via\na multi-view human diffusion model. We first generate multi-view images from\nthe single reference image with an enhanced multi-view diffusion model, which\nis well fine-tuned on high-quality 3D human datasets to incorporate 3D geometry\npriors and human structure priors. To infer accurate camera poses from the\nsparse generated multi-view images for reconstruction, an alignment module is\nintroduced to facilitate joint optimization of 3D Gaussians and camera poses.\nFurthermore, we propose a depth-based Facial Distortion Mitigation module to\nrefine the generated facial regions, thereby improving the overall fidelity of\nthe reconstruction.Finally, leveraging the refined multi-view images, along\nwith their accurate camera poses, MVD-HuGaS optimizes the 3D Gaussians of the\ntarget human for high-fidelity free-view renderings. Extensive experiments on\nThuman2.0 and 2K2K datasets show that the proposed MVD-HuGaS achieves\nstate-of-the-art performance on single-view 3D human rendering.\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": "2311.02892", "title": "Human as Points: Explicit Point-based 3D Human Reconstruction from\n  Single-view RGB Images", "abstract": "  The latest trends in the research field of single-view human reconstruction\ndevote to learning deep implicit functions constrained by explicit body shape\npriors. Despite the remarkable performance improvements compared with\ntraditional processing pipelines, existing learning approaches still show\ndifferent aspects of limitations in terms of flexibility, generalizability,\nrobustness, and/or representation capability. To comprehensively address the\nabove issues, in this paper, we investigate an explicit point-based human\nreconstruction framework called HaP, which adopts point clouds as the\nintermediate representation of the target geometric structure. Technically, our\napproach is featured by fully-explicit point cloud estimation, manipulation,\ngeneration, and refinement in the 3D geometric space, instead of an implicit\nlearning process that can be ambiguous and less controllable. The overall\nworkflow is carefully organized with dedicated designs of the corresponding\nspecialized learning components as well as processing procedures. Extensive\nexperiments demonstrate that our framework achieves quantitative performance\nimprovements of 20% to 40% over current state-of-the-art methods, and better\nqualitative results. Our promising results may indicate a paradigm rollback to\nthe fully-explicit and geometry-centric algorithm design, which enables to\nexploit various powerful point cloud modeling architectures and processing\ntechniques. We will make our code and data publicly available at\nhttps://github.com/yztang4/HaP.\n", "date": "2023-11-06", "categories": []}
{"id": "2502.18150", "title": "Realistic Clothed Human and Object Joint Reconstruction from a Single\n  Image", "abstract": "  Recent approaches to jointly reconstruct 3D humans and objects from a single\nRGB image represent 3D shapes with template-based or coarse models, which fail\nto capture details of loose clothing on human bodies. In this paper, we\nintroduce a novel implicit approach for jointly reconstructing realistic 3D\nclothed humans and objects from a monocular view. For the first time, we model\nboth the human and the object with an implicit representation, allowing to\ncapture more realistic details such as clothing. This task is extremely\nchallenging due to human-object occlusions and the lack of 3D information in 2D\nimages, often leading to poor detail reconstruction and depth ambiguity. To\naddress these problems, we propose a novel attention-based neural implicit\nmodel that leverages image pixel alignment from both the input human-object\nimage for a global understanding of the human-object scene and from local\nseparate views of the human and object images to improve realism with, for\nexample, clothing details. Additionally, the network is conditioned on semantic\nfeatures derived from an estimated human-object pose prior, which provides 3D\nspatial information about the shared space of humans and objects. To handle\nhuman occlusion caused by objects, we use a generative diffusion model that\ninpaints the occluded regions, recovering otherwise lost details. For training\nand evaluation, we introduce a synthetic dataset featuring rendered scenes of\ninter-occluded 3D human scans and diverse objects. Extensive evaluation on both\nsynthetic and real-world datasets demonstrates the superior quality of the\nproposed human-object reconstructions over competitive methods.\n", "date": "2025-02-25", "categories": []}
{"id": "2503.06089", "title": "Fish2Mesh Transformer: 3D Human Mesh Recovery from Egocentric Vision", "abstract": "  Egocentric human body estimation allows for the inference of user body pose\nand shape from a wearable camera's first-person perspective. Although research\nhas used pose estimation techniques to overcome self-occlusions and image\ndistortions caused by head-mounted fisheye images, similar advances in 3D human\nmesh recovery (HMR) techniques have been limited. We introduce Fish2Mesh, a\nfisheye-aware transformer-based model designed for 3D egocentric human mesh\nrecovery. We propose an egocentric position embedding block to generate an\nego-specific position table for the Swin Transformer to reduce fisheye image\ndistortion. Our model utilizes multi-task heads for SMPL parametric regression\nand camera translations, estimating 3D and 2D joints as auxiliary loss to\nsupport model training. To address the scarcity of egocentric camera data, we\ncreate a training dataset by employing the pre-trained 4D-Human model and\nthird-person cameras for weak supervision. Our experiments demonstrate that\nFish2Mesh outperforms previous state-of-the-art 3D HMR models.\n", "date": "2025-03-08", "categories": []}
{"id": "2503.06154", "title": "SRM-Hair: Single Image Head Mesh Reconstruction via 3D Morphable Hair", "abstract": "  3D Morphable Models (3DMMs) have played a pivotal role as a fundamental\nrepresentation or initialization for 3D avatar animation and reconstruction.\nHowever, extending 3DMMs to hair remains challenging due to the difficulty of\nenforcing vertex-level consistent semantic meaning across hair shapes. This\npaper introduces a novel method, Semantic-consistent Ray Modeling of Hair\n(SRM-Hair), for making 3D hair morphable and controlled by coefficients. The\nkey contribution lies in semantic-consistent ray modeling, which extracts\nordered hair surface vertices and exhibits notable properties such as\nadditivity for hairstyle fusion, adaptability, flipping, and thickness\nmodification. We collect a dataset of over 250 high-fidelity real hair scans\npaired with 3D face data to serve as a prior for the 3D morphable hair. Based\non this, SRM-Hair can reconstruct a hair mesh combined with a 3D head from a\nsingle image. Note that SRM-Hair produces an independent hair mesh,\nfacilitating applications in virtual avatar creation, realistic animation, and\nhigh-fidelity hair rendering. Both quantitative and qualitative experiments\ndemonstrate that SRM-Hair achieves state-of-the-art performance in 3D mesh\nreconstruction. Our project is available at\nhttps://github.com/wang-zidu/SRM-Hair\n", "date": "2025-03-08", "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": "2503.02375", "title": "mmDEAR: mmWave Point Cloud Density Enhancement for Accurate Human Body\n  Reconstruction", "abstract": "  Millimeter-wave (mmWave) radar offers robust sensing capabilities in diverse\nenvironments, making it a highly promising solution for human body\nreconstruction due to its privacy-friendly and non-intrusive nature. However,\nthe significant sparsity of mmWave point clouds limits the estimation accuracy.\nTo overcome this challenge, we propose a two-stage deep learning framework that\nenhances mmWave point clouds and improves human body reconstruction accuracy.\nOur method includes a mmWave point cloud enhancement module that densifies the\nraw data by leveraging temporal features and a multi-stage completion network,\nfollowed by a 2D-3D fusion module that extracts both 2D and 3D motion features\nto refine SMPL parameters. The mmWave point cloud enhancement module learns the\ndetailed shape and posture information from 2D human masks in single-view\nimages. However, image-based supervision is involved only during the training\nphase, and the inference relies solely on sparse point clouds to maintain\nprivacy. Experiments on multiple datasets demonstrate that our approach\noutperforms state-of-the-art methods, with the enhanced point clouds further\nimproving performance when integrated into existing models.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.02452", "title": "2DGS-Avatar: Animatable High-fidelity Clothed Avatar via 2D Gaussian\n  Splatting", "abstract": "  Real-time rendering of high-fidelity and animatable avatars from monocular\nvideos remains a challenging problem in computer vision and graphics. Over the\npast few years, the Neural Radiance Field (NeRF) has made significant progress\nin rendering quality but behaves poorly in run-time performance due to the low\nefficiency of volumetric rendering. Recently, methods based on 3D Gaussian\nSplatting (3DGS) have shown great potential in fast training and real-time\nrendering. However, they still suffer from artifacts caused by inaccurate\ngeometry. To address these problems, we propose 2DGS-Avatar, a novel approach\nbased on 2D Gaussian Splatting (2DGS) for modeling animatable clothed avatars\nwith high-fidelity and fast training performance. Given monocular RGB videos as\ninput, our method generates an avatar that can be driven by poses and rendered\nin real-time. Compared to 3DGS-based methods, our 2DGS-Avatar retains the\nadvantages of fast training and rendering while also capturing detailed,\ndynamic, and photo-realistic appearances. We conduct abundant experiments on\npopular datasets such as AvatarRex and THuman4.0, demonstrating impressive\nperformance in both qualitative and quantitative metrics.\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": "2410.21629", "title": "OFER: Occluded Face Expression Reconstruction", "abstract": "  Reconstructing 3D face models from a single image is an inherently ill-posed\nproblem, which becomes even more challenging in the presence of occlusions. In\naddition to fewer available observations, occlusions introduce an extra source\nof ambiguity where multiple reconstructions can be equally valid. Despite the\nubiquity of the problem, very few methods address its multi-hypothesis nature.\nIn this paper we introduce OFER, a novel approach for single-image 3D face\nreconstruction that can generate plausible, diverse, and expressive 3D faces,\neven under strong occlusions. Specifically, we train two diffusion models to\ngenerate the shape and expression coefficients of a face parametric model,\nconditioned on the input image. This approach captures the multi-modal nature\nof the problem, generating a distribution of solutions as output. However, to\nmaintain consistency across diverse expressions, the challenge is to select the\nbest matching shape. To achieve this, we propose a novel ranking mechanism that\nsorts the outputs of the shape diffusion network based on predicted shape\naccuracy scores. We evaluate our method using standard benchmarks and introduce\nCO-545, a new protocol and dataset designed to assess the accuracy of\nexpressive faces under occlusion. Our results show improved performance over\nocclusion-based methods, while also enabling the generation of diverse\nexpressions for a given image.\n", "date": "2024-10-29", "categories": []}
{"id": "2502.10982", "title": "TEASER: Token Enhanced Spatial Modeling for Expressions Reconstruction", "abstract": "  3D facial reconstruction from a single in-the-wild image is a crucial task in\nhuman-centered computer vision tasks. While existing methods can recover\naccurate facial shapes, there remains significant space for improvement in\nfine-grained expression capture. Current approaches struggle with irregular\nmouth shapes, exaggerated expressions, and asymmetrical facial movements. We\npresent TEASER (Token EnhAnced Spatial modeling for Expressions\nReconstruction), which addresses these challenges and enhances 3D facial\ngeometry performance. TEASER tackles two main limitations of existing methods:\ninsufficient photometric loss for self-reconstruction and inaccurate\nlocalization of subtle expressions. We introduce a multi-scale tokenizer to\nextract facial appearance information. Combined with a neural renderer, these\ntokens provide precise geometric guidance for expression reconstruction.\nFurthermore, TEASER incorporates a pose-dependent landmark loss to further\nimprove geometric performances. Our approach not only significantly enhances\nexpression reconstruction quality but also offers interpretable tokens suitable\nfor various downstream applications, such as photorealistic facial video\ndriving, expression transfer, and identity swapping. Quantitative and\nqualitative experimental results across multiple datasets demonstrate that\nTEASER achieves state-of-the-art performance in precise expression\nreconstruction.\n", "date": "2025-02-16", "categories": []}
{"id": "2503.01610", "title": "Vid2Avatar-Pro: Authentic Avatar from Videos in the Wild via Universal\n  Prior", "abstract": "  We present Vid2Avatar-Pro, a method to create photorealistic and animatable\n3D human avatars from monocular in-the-wild videos. Building a high-quality\navatar that supports animation with diverse poses from a monocular video is\nchallenging because the observation of pose diversity and view points is\ninherently limited. The lack of pose variations typically leads to poor\ngeneralization to novel poses, and avatars can easily overfit to limited input\nview points, producing artifacts and distortions from other views. In this\nwork, we address these limitations by leveraging a universal prior model (UPM)\nlearned from a large corpus of multi-view clothed human performance capture\ndata. We build our representation on top of expressive 3D Gaussians with\ncanonical front and back maps shared across identities. Once the UPM is learned\nto accurately reproduce the large-scale multi-view human images, we fine-tune\nthe model with an in-the-wild video via inverse rendering to obtain a\npersonalized photorealistic human avatar that can be faithfully animated to\nnovel human motions and rendered from novel views. The experiments show that\nour approach based on the learned universal prior sets a new state-of-the-art\nin monocular avatar reconstruction by substantially outperforming existing\napproaches relying only on heuristic regularization or a shape prior of\nminimally clothed bodies (e.g., SMPL) on publicly available datasets.\n", "date": "2025-03-03", "categories": []}
{"id": "2502.19441", "title": "AniGaussian: Animatable Gaussian Avatar with Pose-guided Deformation", "abstract": "  Recent advancements in Gaussian-based human body reconstruction have achieved\nnotable success in creating animatable avatars. However, there are ongoing\nchallenges to fully exploit the SMPL model's prior knowledge and enhance the\nvisual fidelity of these models to achieve more refined avatar reconstructions.\nIn this paper, we introduce AniGaussian which addresses the above issues with\ntwo insights. First, we propose an innovative pose guided deformation strategy\nthat effectively constrains the dynamic Gaussian avatar with SMPL pose\nguidance, ensuring that the reconstructed model not only captures the detailed\nsurface nuances but also maintains anatomical correctness across a wide range\nof motions. Second, we tackle the expressiveness limitations of Gaussian models\nin representing dynamic human bodies. We incorporate rigid-based priors from\nprevious works to enhance the dynamic transform capabilities of the Gaussian\nmodel. Furthermore, we introduce a split-with-scale strategy that significantly\nimproves geometry quality. The ablative study experiment demonstrates the\neffectiveness of our innovative model design. Through extensive comparisons\nwith existing methods, AniGaussian demonstrates superior performance in both\nqualitative result and quantitative metrics.\n", "date": "2025-02-24", "categories": []}
{"id": "2502.20387", "title": "InsTaG: Learning Personalized 3D Talking Head from Few-Second Video", "abstract": "  Despite exhibiting impressive performance in synthesizing lifelike\npersonalized 3D talking heads, prevailing methods based on radiance fields\nsuffer from high demands for training data and time for each new identity. This\npaper introduces InsTaG, a 3D talking head synthesis framework that allows a\nfast learning of realistic personalized 3D talking head from few training data.\nBuilt upon a lightweight 3DGS person-specific synthesizer with universal motion\npriors, InsTaG achieves high-quality and fast adaptation while preserving\nhigh-level personalization and efficiency. As preparation, we first propose an\nIdentity-Free Pre-training strategy that enables the pre-training of the\nperson-specific model and encourages the collection of universal motion priors\nfrom long-video data corpus. To fully exploit the universal motion priors to\nlearn an unseen new identity, we then present a Motion-Aligned Adaptation\nstrategy to adaptively align the target head to the pre-trained field, and\nconstrain a robust dynamic head structure under few training data. Experiments\ndemonstrate our outstanding performance and efficiency under various data\nscenarios to render high-quality personalized talking heads.\n", "date": "2025-02-27", "categories": []}
{"id": "2404.08449", "title": "OccGaussian: 3D Gaussian Splatting for Occluded Human Rendering", "abstract": "  Rendering dynamic 3D human from monocular videos is crucial for various\napplications such as virtual reality and digital entertainment. Most methods\nassume the people is in an unobstructed scene, while various objects may cause\nthe occlusion of body parts in real-life scenarios. Previous method utilizing\nNeRF for surface rendering to recover the occluded areas, but it requiring more\nthan one day to train and several seconds to render, failing to meet the\nrequirements of real-time interactive applications. To address these issues, we\npropose OccGaussian based on 3D Gaussian Splatting, which can be trained within\n6 minutes and produces high-quality human renderings up to 160 FPS with\noccluded input. OccGaussian initializes 3D Gaussian distributions in the\ncanonical space, and we perform occlusion feature query at occluded regions,\nthe aggregated pixel-align feature is extracted to compensate for the missing\ninformation. Then we use Gaussian Feature MLP to further process the feature\nalong with the occlusion-aware loss functions to better perceive the occluded\narea. Extensive experiments both in simulated and real-world occlusions,\ndemonstrate that our method achieves comparable or even superior performance\ncompared to the state-of-the-art method. And we improving training and\ninference speeds by 250x and 800x, respectively. Our code will be available for\nresearch purposes.\n", "date": "2024-04-12", "categories": []}
{"id": "2403.11453", "title": "Hybrid Explicit Representation for Ultra-Realistic Head Avatars", "abstract": "  We introduce a novel approach to creating ultra-realistic head avatars and\nrendering them in real-time (>30fps at $2048 \\times 1334$ resolution). First,\nwe propose a hybrid explicit representation that combines the advantages of two\nprimitive-based efficient rendering techniques. UV-mapped 3D mesh is utilized\nto capture sharp and rich textures on smooth surfaces, while 3D Gaussian\nSplatting is employed to represent complex geometric structures. In the\npipeline of modeling an avatar, after tracking parametric models based on\ncaptured multi-view RGB videos, our goal is to simultaneously optimize the\ntexture and opacity map of mesh, as well as a set of 3D Gaussian splats\nlocalized and rigged onto the mesh facets. Specifically, we perform\n$\\alpha$-blending on the color and opacity values based on the merged and\nre-ordered z-buffer from the rasterization results of mesh and 3DGS. This\nprocess involves the mesh and 3DGS adaptively fitting the captured visual\ninformation to outline a high-fidelity digital avatar. To avoid artifacts\ncaused by Gaussian splats crossing the mesh facets, we design a stable hybrid\ndepth sorting strategy. Experiments illustrate that our modeled results exceed\nthose of state-of-the-art approaches.\n", "date": "2024-03-18", "categories": []}
{"id": "2502.06367", "title": "FOCUS -- Multi-View Foot Reconstruction From Synthetically Trained Dense\n  Correspondences", "abstract": "  Surface reconstruction from multiple, calibrated images is a challenging task\n- often requiring a large number of collected images with significant overlap.\nWe look at the specific case of human foot reconstruction. As with previous\nsuccessful foot reconstruction work, we seek to extract rich per-pixel geometry\ncues from multi-view RGB images, and fuse these into a final 3D object. Our\nmethod, FOCUS, tackles this problem with 3 main contributions: (i) SynFoot2, an\nextension of an existing synthetic foot dataset to include a new data type:\ndense correspondence with the parameterized foot model FIND; (ii) an\nuncertainty-aware dense correspondence predictor trained on our synthetic\ndataset; (iii) two methods for reconstructing a 3D surface from dense\ncorrespondence predictions: one inspired by Structure-from-Motion, and one\noptimization-based using the FIND model. We show that our reconstruction\nachieves state-of-the-art reconstruction quality in a few-view setting,\nperforming comparably to state-of-the-art when many views are available, and\nruns substantially faster. We release our synthetic dataset to the research\ncommunity. Code is available at: https://github.com/OllieBoyne/FOCUS\n", "date": "2025-02-10", "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": "2502.09425", "title": "A 3D Facial Reconstruction Evaluation Methodology: Comparing Smartphone\n  Scans with Deep Learning Based Methods Using Geometry and Morphometry\n  Criteria", "abstract": "  Three-dimensional (3D) facial shape analysis has gained interest due to its\npotential clinical applications. However, the high cost of advanced 3D facial\nacquisition systems limits their widespread use, driving the development of\nlow-cost acquisition and reconstruction methods. This study introduces a novel\nevaluation methodology that goes beyond traditional geometry-based benchmarks\nby integrating morphometric shape analysis techniques, providing a statistical\nframework for assessing facial morphology preservation. As a case study, we\ncompare smartphone-based 3D scans with state-of-the-art deep learning\nreconstruction methods from 2D images, using high-end stereophotogrammetry\nmodels as ground truth. This methodology enables a quantitative assessment of\nglobal and local shape differences, offering a biologically meaningful\nvalidation approach for low-cost 3D facial acquisition and reconstruction\ntechniques.\n", "date": "2025-02-13", "categories": []}
{"id": "2502.09617", "title": "LIFe-GoM: Generalizable Human Rendering with Learned Iterative Feedback\n  Over Multi-Resolution Gaussians-on-Mesh", "abstract": "  Generalizable rendering of an animatable human avatar from sparse inputs\nrelies on data priors and inductive biases extracted from training on large\ndata to avoid scene-specific optimization and to enable fast reconstruction.\nThis raises two main challenges: First, unlike iterative gradient-based\nadjustment in scene-specific optimization, generalizable methods must\nreconstruct the human shape representation in a single pass at inference time.\nSecond, rendering is preferably computationally efficient yet of high\nresolution. To address both challenges we augment the recently proposed dual\nshape representation, which combines the benefits of a mesh and Gaussian\npoints, in two ways. To improve reconstruction, we propose an iterative\nfeedback update framework, which successively improves the canonical human\nshape representation during reconstruction. To achieve computationally\nefficient yet high-resolution rendering, we study a coupled-multi-resolution\nGaussians-on-Mesh representation. We evaluate the proposed approach on the\nchallenging THuman2.0, XHuman and AIST++ data. Our approach reconstructs an\nanimatable representation from sparse inputs in less than 1s, renders views\nwith 95.1FPS at $1024 \\times 1024$, and achieves PSNR/LPIPS*/FID of\n24.65/110.82/51.27 on THuman2.0, outperforming the state-of-the-art in\nrendering quality.\n", "date": "2025-02-13", "categories": []}
{"id": "2502.07844", "title": "The establishment of static digital humans and the integration with\n  spinal models", "abstract": "  Adolescent idiopathic scoliosis (AIS), a prevalent spinal deformity,\nsignificantly affects individuals' health and quality of life. Conventional\nimaging techniques, such as X - rays, computed tomography (CT), and magnetic\nresonance imaging (MRI), offer static views of the spine. However, they are\nrestricted in capturing the dynamic changes of the spine and its interactions\nwith overall body motion. Therefore, developing new techniques to address these\nlimitations has become extremely important. Dynamic digital human modeling\nrepresents a major breakthrough in digital medicine. It enables a three -\ndimensional (3D) view of the spine as it changes during daily activities,\nassisting clinicians in detecting deformities that might be missed in static\nimaging. Although dynamic modeling holds great potential, constructing an\naccurate static digital human model is a crucial initial step for high -\nprecision simulations. In this study, our focus is on constructing an accurate\nstatic digital human model integrating the spine, which is vital for subsequent\ndynamic digital human research on AIS. First, we generate human point - cloud\ndata by combining the 3D Gaussian method with the Skinned Multi - Person Linear\n(SMPL) model from the patient's multi - view images. Then, we fit a standard\nskeletal model to the generated human model. Next, we align the real spine\nmodel reconstructed from CT images with the standard skeletal model. We\nvalidated the resulting personalized spine model using X - ray data from six\nAIS patients, with Cobb angles (used to measure the severity of scoliosis) as\nevaluation metrics. The results indicate that the model's error was within 1\ndegree of the actual measurements. This study presents an important method for\nconstructing digital humans.\n", "date": "2025-02-11", "categories": []}
{"id": "2310.08784", "title": "Implicit Shape and Appearance Priors for Few-Shot Full Head\n  Reconstruction", "abstract": "  Recent advancements in learning techniques that employ coordinate-based\nneural representations have yielded remarkable results in multi-view 3D\nreconstruction tasks. However, these approaches often require a substantial\nnumber of input views (typically several tens) and computationally intensive\noptimization procedures to achieve their effectiveness. In this paper, we\naddress these limitations specifically for the problem of few-shot full 3D head\nreconstruction. We accomplish this by incorporating a probabilistic shape and\nappearance prior into coordinate-based representations, enabling faster\nconvergence and improved generalization when working with only a few input\nimages (even as low as a single image). During testing, we leverage this prior\nto guide the fitting process of a signed distance function using a\ndifferentiable renderer. By incorporating the statistical prior alongside\nparallelizable ray tracing and dynamic caching strategies, we achieve an\nefficient and accurate approach to few-shot full 3D head reconstruction.\nMoreover, we extend the H3DS dataset, which now comprises 60 high-resolution 3D\nfull head scans and their corresponding posed images and masks, which we use\nfor evaluation purposes. By leveraging this dataset, we demonstrate the\nremarkable capabilities of our approach in achieving state-of-the-art results\nin geometry reconstruction while being an order of magnitude faster than\nprevious approaches.\n", "date": "2023-10-12", "categories": []}
{"id": "2311.08581", "title": "Drivable 3D Gaussian Avatars", "abstract": "  We present Drivable 3D Gaussian Avatars (D3GA), a multi-layered 3D\ncontrollable model for human bodies that utilizes 3D Gaussian primitives\nembedded into tetrahedral cages. The advantage of using cages compared to\ncommonly employed linear blend skinning (LBS) is that primitives like 3D\nGaussians are naturally re-oriented and their kernels are stretched via the\ndeformation gradients of the encapsulating tetrahedron. Additional offsets are\nmodeled for the tetrahedron vertices, effectively decoupling the\nlow-dimensional driving poses from the extensive set of primitives to be\nrendered. This separation is achieved through the localized influence of each\ntetrahedron on 3D Gaussians, resulting in improved optimization. Using the\ncage-based deformation model, we introduce a compositional pipeline that\ndecomposes an avatar into layers, such as garments, hands, or faces, improving\nthe modeling of phenomena like garment sliding. These parts can be conditioned\non different driving signals, such as keypoints for facial expressions or\njoint-angle vectors for garments and the body. Our experiments on two\nmulti-view datasets with varied body shapes, clothes, and motions show\nhigher-quality results. They surpass PSNR and SSIM metrics of other SOTA\nmethods using the same data while offering greater flexibility and compactness.\n", "date": "2023-11-14", "categories": []}
{"id": "2502.07140", "title": "Few-Shot Multi-Human Neural Rendering Using Geometry Constraints", "abstract": "  We present a method for recovering the shape and radiance of a scene\nconsisting of multiple people given solely a few images. Multi-human scenes are\ncomplex due to additional occlusion and clutter. For single-human settings,\nexisting approaches using implicit neural representations have achieved\nimpressive results that deliver accurate geometry and appearance. However, it\nremains challenging to extend these methods for estimating multiple humans from\nsparse views. We propose a neural implicit reconstruction method that addresses\nthe inherent challenges of this task through the following contributions:\nFirst, we propose to use geometry constraints by exploiting pre-computed meshes\nusing a human body model (SMPL). Specifically, we regularize the signed\ndistances using the SMPL mesh and leverage bounding boxes for improved\nrendering. Second, we propose a ray regularization scheme to minimize rendering\ninconsistencies, and a saturation regularization for robust optimization in\nvariable illumination. Extensive experiments on both real and synthetic\ndatasets demonstrate the benefits of our approach and show state-of-the-art\nperformance against existing neural reconstruction methods.\n", "date": "2025-02-11", "categories": []}
{"id": "2408.10588", "title": "DEGAS: Detailed Expressions on Full-Body Gaussian Avatars", "abstract": "  Although neural rendering has made significant advances in creating lifelike,\nanimatable full-body and head avatars, incorporating detailed expressions into\nfull-body avatars remains largely unexplored. We present DEGAS, the first 3D\nGaussian Splatting (3DGS)-based modeling method for full-body avatars with rich\nfacial expressions. Trained on multiview videos of a given subject, our method\nlearns a conditional variational autoencoder that takes both the body motion\nand facial expression as driving signals to generate Gaussian maps in the UV\nlayout. To drive the facial expressions, instead of the commonly used 3D\nMorphable Models (3DMMs) in 3D head avatars, we propose to adopt the expression\nlatent space trained solely on 2D portrait images, bridging the gap between 2D\ntalking faces and 3D avatars. Leveraging the rendering capability of 3DGS and\nthe rich expressiveness of the expression latent space, the learned avatars can\nbe reenacted to reproduce photorealistic rendering images with subtle and\naccurate facial expressions. Experiments on an existing dataset and our newly\nproposed dataset of full-body talking avatars demonstrate the efficacy of our\nmethod. We also propose an audio-driven extension of our method with the help\nof 2D talking faces, opening new possibilities for interactive AI agents.\n", "date": "2024-08-20", "categories": []}
{"id": "2502.03836", "title": "Adapting Human Mesh Recovery with Vision-Language Feedback", "abstract": "  Human mesh recovery can be approached using either regression-based or\noptimization-based methods. Regression models achieve high pose accuracy but\nstruggle with model-to-image alignment due to the lack of explicit 2D-3D\ncorrespondences. In contrast, optimization-based methods align 3D models to 2D\nobservations but are prone to local minima and depth ambiguity. In this work,\nwe leverage large vision-language models (VLMs) to generate interactive body\npart descriptions, which serve as implicit constraints to enhance 3D perception\nand limit the optimization space. Specifically, we formulate monocular human\nmesh recovery as a distribution adaptation task by integrating both 2D\nobservations and language descriptions. To bridge the gap between text and 3D\npose signals, we first train a text encoder and a pose VQ-VAE, aligning texts\nto body poses in a shared latent space using contrastive learning.\nSubsequently, we employ a diffusion-based framework to refine the initial\nparameters guided by gradients derived from both 2D observations and text\ndescriptions. Finally, the model can produce poses with accurate 3D perception\nand image consistency. Experimental results on multiple benchmarks validate its\neffectiveness. The code will be made publicly available.\n", "date": "2025-02-06", "categories": []}
{"id": "2412.02403", "title": "3D Face Reconstruction From Radar Images", "abstract": "  The 3D reconstruction of faces gains wide attention in computer vision and is\nused in many fields of application, for example, animation, virtual reality,\nand even forensics. This work is motivated by monitoring patients in sleep\nlaboratories. Due to their unique characteristics, sensors from the radar\ndomain have advantages compared to optical sensors, namely penetration of\nelectrically non-conductive materials and independence of light. These\nadvantages of radar signals unlock new applications and require adaptation of\n3D reconstruction frameworks. We propose a novel model-based method for 3D\nreconstruction from radar images. We generate a dataset of synthetic radar\nimages with a physics-based but non-differentiable radar renderer. This dataset\nis used to train a CNN-based encoder to estimate the parameters of a 3D\nmorphable face model. Whilst the encoder alone already leads to strong\nreconstructions of synthetic data, we extend our reconstruction in an\nAnalysis-by-Synthesis fashion to a model-based autoencoder. This is enabled by\nlearning the rendering process in the decoder, which acts as an object-specific\ndifferentiable radar renderer. Subsequently, the combination of both network\nparts is trained to minimize both, the loss of the parameters and the loss of\nthe resulting reconstructed radar image. This leads to the additional benefit,\nthat at test time the parameters can be further optimized by finetuning the\nautoencoder unsupervised on the image loss. We evaluated our framework on\ngenerated synthetic face images as well as on real radar images with 3D ground\ntruth of four individuals.\n", "date": "2024-12-03", "categories": []}
{"id": "2311.14323", "title": "BinaryHPE: 3D Human Pose and Shape Estimation via Binarization", "abstract": "  3D human pose and shape estimation (HPE) aims to reconstruct the 3D human\nbody, face, and hands from a single image. Although powerful deep learning\nmodels have achieved accurate estimation in this task, they require enormous\nmemory and computational resources. Consequently, these methods can hardly be\ndeployed on resource-limited edge devices. In this work, we propose BinaryHPE,\na novel binarization method designed to estimate the 3D human body, face, and\nhands parameters efficiently. Specifically, we propose a novel binary backbone\ncalled Binarized Dual Residual Network (BiDRN), designed to retain as much\nfull-precision information as possible. Furthermore, we propose the Binarized\nBoxNet, an efficient sub-network for predicting face and hands bounding boxes,\nwhich further reduces model redundancy. Comprehensive quantitative and\nqualitative experiments demonstrate the effectiveness of BinaryHPE, which has a\nsignificant improvement over state-of-the-art binarization algorithms.\nMoreover, our BinaryHPE achieves comparable performance with the full-precision\nmethod Hand4Whole while using only 22.1% parameters and 14.8% operations. We\nwill release all the code and pretrained models.\n", "date": "2023-11-24", "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": "2501.15616", "title": "IPVTON: Image-based 3D Virtual Try-on with Image Prompt Adapter", "abstract": "  Given a pair of images depicting a person and a garment separately,\nimage-based 3D virtual try-on methods aim to reconstruct a 3D human model that\nrealistically portrays the person wearing the desired garment. In this paper,\nwe present IPVTON, a novel image-based 3D virtual try-on framework. IPVTON\nemploys score distillation sampling with image prompts to optimize a hybrid 3D\nhuman representation, integrating target garment features into diffusion priors\nthrough an image prompt adapter. To avoid interference with non-target areas,\nwe leverage mask-guided image prompt embeddings to focus the image features on\nthe try-on regions. Moreover, we impose geometric constraints on the 3D model\nwith a pseudo silhouette generated by ControlNet, ensuring that the clothed 3D\nhuman model retains the shape of the source identity while accurately wearing\nthe target garments. Extensive qualitative and quantitative experiments\ndemonstrate that IPVTON outperforms previous methods in image-based 3D virtual\ntry-on tasks, excelling in both geometry and texture.\n", "date": "2025-01-26", "categories": []}
{"id": "2501.14726", "title": "Relightable Full-Body Gaussian Codec Avatars", "abstract": "  We propose Relightable Full-Body Gaussian Codec Avatars, a new approach for\nmodeling relightable full-body avatars with fine-grained details including face\nand hands. The unique challenge for relighting full-body avatars lies in the\nlarge deformations caused by body articulation and the resulting impact on\nappearance caused by light transport. Changes in body pose can dramatically\nchange the orientation of body surfaces with respect to lights, resulting in\nboth local appearance changes due to changes in local light transport\nfunctions, as well as non-local changes due to occlusion between body parts. To\naddress this, we decompose the light transport into local and non-local\neffects. Local appearance changes are modeled using learnable zonal harmonics\nfor diffuse radiance transfer. Unlike spherical harmonics, zonal harmonics are\nhighly efficient to rotate under articulation. This allows us to learn diffuse\nradiance transfer in a local coordinate frame, which disentangles the local\nradiance transfer from the articulation of the body. To account for non-local\nappearance changes, we introduce a shadow network that predicts shadows given\nprecomputed incoming irradiance on a base mesh. This facilitates the learning\nof non-local shadowing between the body parts. Finally, we use a deferred\nshading approach to model specular radiance transfer and better capture\nreflections and highlights such as eye glints. We demonstrate that our approach\nsuccessfully models both the local and non-local light transport required for\nrelightable full-body avatars, with a superior generalization ability under\nnovel illumination conditions and unseen poses.\n", "date": "2025-01-24", "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": "2501.07104", "title": "RMAvatar: Photorealistic Human Avatar Reconstruction from Monocular\n  Video Based on Rectified Mesh-embedded Gaussians", "abstract": "  We introduce RMAvatar, a novel human avatar representation with Gaussian\nsplatting embedded on mesh to learn clothed avatar from a monocular video. We\nutilize the explicit mesh geometry to represent motion and shape of a virtual\nhuman and implicit appearance rendering with Gaussian Splatting. Our method\nconsists of two main modules: Gaussian initialization module and Gaussian\nrectification module. We embed Gaussians into triangular faces and control\ntheir motion through the mesh, which ensures low-frequency motion and surface\ndeformation of the avatar. Due to the limitations of LBS formula, the human\nskeleton is hard to control complex non-rigid transformations. We then design a\npose-related Gaussian rectification module to learn fine-detailed non-rigid\ndeformations, further improving the realism and expressiveness of the avatar.\nWe conduct extensive experiments on public datasets, RMAvatar shows\nstate-of-the-art performance on both rendering quality and quantitative\nevaluations. Please see our project page at https://rm-avatar.github.io.\n", "date": "2025-01-13", "categories": []}
{"id": "2501.01589", "title": "D$^3$-Human: Dynamic Disentangled Digital Human from Monocular Video", "abstract": "  We introduce D$^3$-Human, a method for reconstructing Dynamic Disentangled\nDigital Human geometry from monocular videos. Past monocular video human\nreconstruction primarily focuses on reconstructing undecoupled clothed human\nbodies or only reconstructing clothing, making it difficult to apply directly\nin applications such as animation production. The challenge in reconstructing\ndecoupled clothing and body lies in the occlusion caused by clothing over the\nbody. To this end, the details of the visible area and the plausibility of the\ninvisible area must be ensured during the reconstruction process. Our proposed\nmethod combines explicit and implicit representations to model the decoupled\nclothed human body, leveraging the robustness of explicit representations and\nthe flexibility of implicit representations. Specifically, we reconstruct the\nvisible region as SDF and propose a novel human manifold signed distance field\n(hmSDF) to segment the visible clothing and visible body, and then merge the\nvisible and invisible body. Extensive experimental results demonstrate that,\ncompared with existing reconstruction schemes, D$^3$-Human can achieve\nhigh-quality decoupled reconstruction of the human body wearing different\nclothing, and can be directly applied to clothing transfer and animation.\n", "date": "2025-01-03", "categories": []}
{"id": "2412.19848", "title": "Generative Landmarks Guided Eyeglasses Removal 3D Face Reconstruction", "abstract": "  Single-view 3D face reconstruction is a fundamental Computer Vision problem\nof extraordinary difficulty. Current systems often assume the input is\nunobstructed faces which makes their method not suitable for in-the-wild\nconditions. We present a method for performing a 3D face that removes\neyeglasses from a single image. Existing facial reconstruction methods fail to\nremove eyeglasses automatically for generating a photo-realistic 3D face\n\"in-the-wild\".The innovation of our method lies in a process for identifying\nthe eyeglasses area robustly and remove it intelligently. In this work, we\nestimate the 2D face structure of the reasonable position of the eyeglasses\narea, which is used for the construction of 3D texture. An excellent\nanti-eyeglasses face reconstruction method should ensure the authenticity of\nthe output, including the topological structure between the eyes, nose, and\nmouth. We achieve this via a deep learning architecture that performs direct\nregression of a 3DMM representation of the 3D facial geometry from a single 2D\nimage. We also demonstrate how the related face parsing task can be\nincorporated into the proposed framework and help improve reconstruction\nquality. We conduct extensive experiments on existing 3D face reconstruction\ntasks as concrete examples to demonstrate the method's superior regulation\nability over existing methods often break down.\n", "date": "2024-12-25", "categories": []}
{"id": "2412.19849", "title": "3D Face Reconstruction With Geometry Details From a Single Color Image\n  Under Occluded Scenes", "abstract": "  3D face reconstruction technology aims to generate a face stereo model\nnaturally and realistically. Previous deep face reconstruction approaches are\ntypically designed to generate convincing textures and cannot generalize well\nto multiple occluded scenarios simultaneously. By introducing bump mapping, we\nsuccessfully added mid-level details to coarse 3D faces. More innovatively, our\nmethod takes into account occlusion scenarios. Thus on top of common 3D face\nreconstruction approaches, we in this paper propose a unified framework to\nhandle multiple types of obstruction simultaneously (e.g., hair, palms and\nglasses et al.).Extensive experiments and comparisons demonstrate that our\nmethod can generate high-quality reconstruction results with geometry details\nfrom captured facial images under occluded scenes.\n", "date": "2024-12-25", "categories": []}
{"id": "2412.18785", "title": "Simultaneously Recovering Multi-Person Meshes and Multi-View Cameras\n  with Human Semantics", "abstract": "  Dynamic multi-person mesh recovery has broad applications in sports\nbroadcasting, virtual reality, and video games. However, current multi-view\nframeworks rely on a time-consuming camera calibration procedure. In this work,\nwe focus on multi-person motion capture with uncalibrated cameras, which mainly\nfaces two challenges: one is that inter-person interactions and occlusions\nintroduce inherent ambiguities for both camera calibration and motion capture;\nthe other is that a lack of dense correspondences can be used to constrain\nsparse camera geometries in a dynamic multi-person scene. Our key idea is to\nincorporate motion prior knowledge to simultaneously estimate camera parameters\nand human meshes from noisy human semantics. We first utilize human information\nfrom 2D images to initialize intrinsic and extrinsic parameters. Thus, the\napproach does not rely on any other calibration tools or background features.\nThen, a pose-geometry consistency is introduced to associate the detected\nhumans from different views. Finally, a latent motion prior is proposed to\nrefine the camera parameters and human motions. Experimental results show that\naccurate camera parameters and human motions can be obtained through a one-step\nreconstruction. The code are publicly available\nat~\\url{https://github.com/boycehbz/DMMR}.\n", "date": "2024-12-25", "categories": []}
{"id": "2412.18920", "title": "Generative Face Parsing Map Guided 3D Face Reconstruction Under Occluded\n  Scenes", "abstract": "  Over the past few years, single-view 3D face reconstruction methods can\nproduce beautiful 3D models. Nevertheless,the input of these works is\nunobstructed faces.We describe a system designed to reconstruct convincing face\ntexture in the case of occlusion.Motivated by parsing facial features,we\npropose a complete face parsing map generation method guided by landmarks.We\nestimate the 2D face structure of the reasonable position of the occlusion\narea,which is used for the construction of 3D texture.An excellent\nanti-occlusion face reconstruction method should ensure the authenticity of the\noutput,including the topological structure between the eyes,nose, and mouth. We\nextensively tested our method and its components, qualitatively demonstrating\nthe rationality of our estimated facial structure. We conduct extensive\nexperiments on general 3D face reconstruction tasks as concrete examples to\ndemonstrate the method's superior regulation ability over existing methods\noften break down.We further provide numerous quantitative examples showing that\nour method advances both the quality and the robustness of 3D face\nreconstruction under occlusion scenes.\n", "date": "2024-12-25", "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": "2403.16207", "title": "Skull-to-Face: Anatomy-Guided 3D Facial Reconstruction and Editing", "abstract": "  Deducing the 3D face from a skull is a challenging task in forensic science\nand archaeology. This paper proposes an end-to-end 3D face reconstruction\npipeline and an exploration method that can conveniently create textured,\nrealistic faces that match the given skull. To this end, we propose a\ntissue-guided face creation and adaptation scheme. With the help of the\nstate-of-the-art text-to-image diffusion model and parametric face model, we\nfirst generate an initial reference 3D face, whose biological profile aligns\nwith the given skull. Then, with the help of tissue thickness distribution, we\nmodify these initial faces to match the skull through a latent optimization\nprocess. The joint distribution of tissue thickness is learned on a set of\nskull landmarks using a collection of scanned skull-face pairs. We also develop\nan efficient face adaptation tool to allow users to interactively adjust tissue\nthickness either globally or at local regions to explore different plausible\nfaces. Experiments conducted on a real skull-face dataset demonstrated the\neffectiveness of our proposed pipeline in terms of reconstruction accuracy,\ndiversity, and stability. Our project page is\nhttps://xmlyqing00.github.io/skull-to-face-page.\n", "date": "2024-03-24", "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": "2412.09545", "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": "2412.13983", "title": "GraphAvatar: Compact Head Avatars with GNN-Generated 3D Gaussians", "abstract": "  Rendering photorealistic head avatars from arbitrary viewpoints is crucial\nfor various applications like virtual reality. Although previous methods based\non Neural Radiance Fields (NeRF) can achieve impressive results, they lack\nfidelity and efficiency. Recent methods using 3D Gaussian Splatting (3DGS) have\nimproved rendering quality and real-time performance but still require\nsignificant storage overhead. In this paper, we introduce a method called\nGraphAvatar that utilizes Graph Neural Networks (GNN) to generate 3D Gaussians\nfor the head avatar. Specifically, GraphAvatar trains a geometric GNN and an\nappearance GNN to generate the attributes of the 3D Gaussians from the tracked\nmesh. Therefore, our method can store the GNN models instead of the 3D\nGaussians, significantly reducing the storage overhead to just 10MB. To reduce\nthe impact of face-tracking errors, we also present a novel graph-guided\noptimization module to refine face-tracking parameters during training.\nFinally, we introduce a 3D-aware enhancer for post-processing to enhance the\nrendering quality. We conduct comprehensive experiments to demonstrate the\nadvantages of GraphAvatar, surpassing existing methods in visual fidelity and\nstorage consumption. The ablation study sheds light on the trade-offs between\nrendering quality and model size. The code will be released at:\nhttps://github.com/ucwxb/GraphAvatar\n", "date": "2024-12-18", "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": "2412.10430", "title": "Unsupervised Cross-Domain Regression for Fine-grained 3D Game Character\n  Reconstruction", "abstract": "  With the rise of the ``metaverse'' and the rapid development of games, it has\nbecome more and more critical to reconstruct characters in the virtual world\nfaithfully. The immersive experience is one of the most central themes of the\n``metaverse'', while the reducibility of the avatar is the crucial point.\nMeanwhile, the game is the carrier of the metaverse, in which players can\nfreely edit the facial appearance of the game character. In this paper, we\npropose a simple but powerful cross-domain framework that can reconstruct\nfine-grained 3D game characters from single-view images in an end-to-end\nmanner. Different from the previous methods, which do not resolve the\ncross-domain gap, we propose an effective regressor that can greatly reduce the\ndiscrepancy between the real-world domain and the game domain. To figure out\nthe drawbacks of no ground truth, our unsupervised framework has accomplished\nthe knowledge transfer of the target domain. Additionally, an innovative\ncontrastive loss is proposed to solve the instance-wise disparity, which keeps\nthe person-specific details of the reconstructed character. In contrast, an\nauxiliary 3D identity-aware extractor is activated to make the results of our\nmodel more impeccable. Then a large set of physically meaningful facial\nparameters is generated robustly and exquisitely. Experiments demonstrate that\nour method yields state-of-the-art performance in 3D game character\nreconstruction.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.11325", "title": "Sonicmesh: Enhancing 3D Human Mesh Reconstruction in Vision-Impaired\n  Environments With Acoustic Signals", "abstract": "  3D Human Mesh Reconstruction (HMR) from 2D RGB images faces challenges in\nenvironments with poor lighting, privacy concerns, or occlusions. These\nweaknesses of RGB imaging can be complemented by acoustic signals, which are\nwidely available, easy to deploy, and capable of penetrating obstacles.\nHowever, no existing methods effectively combine acoustic signals with RGB data\nfor robust 3D HMR. The primary challenges include the low-resolution images\ngenerated by acoustic signals and the lack of dedicated processing backbones.\nWe introduce SonicMesh, a novel approach combining acoustic signals with RGB\nimages to reconstruct 3D human mesh. To address the challenges of low\nresolution and the absence of dedicated processing backbones in images\ngenerated by acoustic signals, we modify an existing method, HRNet, for\neffective feature extraction. We also integrate a universal feature embedding\ntechnique to enhance the precision of cross-dimensional feature alignment,\nenabling SonicMesh to achieve high accuracy. Experimental results demonstrate\nthat SonicMesh accurately reconstructs 3D human mesh in challenging\nenvironments such as occlusions, non-line-of-sight scenarios, and poor\nlighting.\n", "date": "2024-12-15", "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": "2412.12093", "title": "CAP4D: Creating Animatable 4D Portrait Avatars with Morphable Multi-View\n  Diffusion Models", "abstract": "  Reconstructing photorealistic and dynamic portrait avatars from images is\nessential to many applications including advertising, visual effects, and\nvirtual reality. Depending on the application, avatar reconstruction involves\ndifferent capture setups and constraints $-$ for example, visual effects\nstudios use camera arrays to capture hundreds of reference images, while\ncontent creators may seek to animate a single portrait image downloaded from\nthe internet. As such, there is a large and heterogeneous ecosystem of methods\nfor avatar reconstruction. Techniques based on multi-view stereo or neural\nrendering achieve the highest quality results, but require hundreds of\nreference images. Recent generative models produce convincing avatars from a\nsingle reference image, but visual fidelity yet lags behind multi-view\ntechniques. Here, we present CAP4D: an approach that uses a morphable\nmulti-view diffusion model to reconstruct photoreal 4D (dynamic 3D) portrait\navatars from any number of reference images (i.e., one to 100) and animate and\nrender them in real time. Our approach demonstrates state-of-the-art\nperformance for single-, few-, and multi-image 4D portrait avatar\nreconstruction, and takes steps to bridge the gap in visual fidelity between\nsingle-image and multi-view reconstruction techniques.\n", "date": "2024-12-16", "categories": []}
{"id": "2407.00574", "title": "Humans as Checkerboards: Calibrating Camera Motion Scale for\n  World-Coordinate Human Mesh Recovery", "abstract": "  Accurate camera motion estimation is essential for recovering global human\nmotion in world coordinates from RGB video inputs. SLAM is widely used for\nestimating camera trajectory and point cloud, but monocular SLAM does so only\nup to an unknown scale factor. Previous works estimate the scale factor through\noptimization, but this is unreliable and time-consuming. This paper presents an\noptimization-free scale calibration framework, Human as Checkerboard (HAC). HAC\ninnovatively leverages the human body predicted by human mesh recovery model as\na calibration reference. Specifically, it uses the absolute depth of\nhuman-scene contact joints as references to calibrate the corresponding\nrelative scene depth from SLAM. HAC benefits from geometric priors encoded in\nhuman mesh recovery models to estimate the SLAM scale and achieves precise\nglobal human motion estimation. Simple yet powerful, our method sets a new\nstate-of-the-art performance for global human mesh estimation tasks, reducing\nmotion errors by 50% over prior local-to-global methods while using 100$\\times$\nless inference time than optimization-based methods. Project page:\nhttps://martayang.github.io/HAC.\n", "date": "2024-06-30", "categories": []}
{"id": "2412.08684", "title": "Coherent3D: Coherent 3D Portrait Video Reconstruction via Triplane\n  Fusion", "abstract": "  Recent breakthroughs in single-image 3D portrait reconstruction have enabled\ntelepresence systems to stream 3D portrait videos from a single camera in\nreal-time, democratizing telepresence. However, per-frame 3D reconstruction\nexhibits temporal inconsistency and forgets the user's appearance. On the other\nhand, self-reenactment methods can render coherent 3D portraits by driving a 3D\navatar built from a single reference image, but fail to faithfully preserve the\nuser's per-frame appearance (e.g., instantaneous facial expression and\nlighting). As a result, none of these two frameworks is an ideal solution for\ndemocratized 3D telepresence. In this work, we address this dilemma and propose\na novel solution that maintains both coherent identity and dynamic per-frame\nappearance to enable the best possible realism. To this end, we propose a new\nfusion-based method that takes the best of both worlds by fusing a canonical 3D\nprior from a reference view with dynamic appearance from per-frame input views,\nproducing temporally stable 3D videos with faithful reconstruction of the\nuser's per-frame appearance. Trained only using synthetic data produced by an\nexpression-conditioned 3D GAN, our encoder-based method achieves both\nstate-of-the-art 3D reconstruction and temporal consistency on in-studio and\nin-the-wild datasets. https://research.nvidia.com/labs/amri/projects/coherent3d\n", "date": "2024-12-11", "categories": []}
{"id": "2412.08524", "title": "Learning to Decouple the Lights for 3D Face Texture Modeling", "abstract": "  Existing research has made impressive strides in reconstructing human facial\nshapes and textures from images with well-illuminated faces and minimal\nexternal occlusions. Nevertheless, it remains challenging to recover accurate\nfacial textures from scenarios with complicated illumination affected by\nexternal occlusions, e.g. a face that is partially obscured by items such as a\nhat. Existing works based on the assumption of single and uniform illumination\ncannot correctly process these data. In this work, we introduce a novel\napproach to model 3D facial textures under such unnatural illumination. Instead\nof assuming single illumination, our framework learns to imitate the unnatural\nillumination as a composition of multiple separate light conditions combined\nwith learned neural representations, named Light Decoupling. According to\nexperiments on both single images and video sequences, we demonstrate the\neffectiveness of our approach in modeling facial textures under challenging\nillumination affected by occlusions. Please check\nhttps://tianxinhuang.github.io/projects/Deface for our videos and codes.\n", "date": "2024-12-11", "categories": []}
{"id": "2407.07532", "title": "Neural Localizer Fields for Continuous 3D Human Pose and Shape\n  Estimation", "abstract": "  With the explosive growth of available training data, single-image 3D human\nmodeling is ahead of a transition to a data-centric paradigm. A key to\nsuccessfully exploiting data scale is to design flexible models that can be\nsupervised from various heterogeneous data sources produced by different\nresearchers or vendors. To this end, we propose a simple yet powerful paradigm\nfor seamlessly unifying different human pose and shape-related tasks and\ndatasets. Our formulation is centered on the ability -- both at training and\ntest time -- to query any arbitrary point of the human volume, and obtain its\nestimated location in 3D. We achieve this by learning a continuous neural field\nof body point localizer functions, each of which is a differently parameterized\n3D heatmap-based convolutional point localizer (detector). For generating\nparametric output, we propose an efficient post-processing step for fitting\nSMPL-family body models to nonparametric joint and vertex predictions. With\nthis approach, we can naturally exploit differently annotated data sources\nincluding mesh, 2D/3D skeleton and dense pose, without having to convert\nbetween them, and thereby train large-scale 3D human mesh and skeleton\nestimation models that considerably outperform the state-of-the-art on several\npublic benchmarks including 3DPW, EMDB, EHF, SSP-3D and AGORA.\n", "date": "2024-07-10", "categories": []}
{"id": "2412.07739", "title": "GASP: Gaussian Avatars with Synthetic Priors", "abstract": "  Gaussian Splatting has changed the game for real-time photo-realistic\nrendering. One of the most popular applications of Gaussian Splatting is to\ncreate animatable avatars, known as Gaussian Avatars. Recent works have pushed\nthe boundaries of quality and rendering efficiency but suffer from two main\nlimitations. Either they require expensive multi-camera rigs to produce avatars\nwith free-view rendering, or they can be trained with a single camera but only\nrendered at high quality from this fixed viewpoint. An ideal model would be\ntrained using a short monocular video or image from available hardware, such as\na webcam, and rendered from any view. To this end, we propose GASP: Gaussian\nAvatars with Synthetic Priors. To overcome the limitations of existing\ndatasets, we exploit the pixel-perfect nature of synthetic data to train a\nGaussian Avatar prior. By fitting this prior model to a single photo or video\nand fine-tuning it, we get a high-quality Gaussian Avatar, which supports\n360$^\\circ$ rendering. Our prior is only required for fitting, not inference,\nenabling real-time application. Through our method, we obtain high-quality,\nanimatable Avatars from limited data which can be animated and rendered at\n70fps on commercial hardware. See our project page\n(https://microsoft.github.io/GASP/) for results.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.04433", "title": "PBDyG: Position Based Dynamic Gaussians for Motion-Aware Clothed Human\n  Avatars", "abstract": "  This paper introduces a novel clothed human model that can be learned from\nmultiview RGB videos, with a particular emphasis on recovering physically\naccurate body and cloth movements. Our method, Position Based Dynamic Gaussians\n(PBDyG), realizes ``movement-dependent'' cloth deformation via physical\nsimulation, rather than merely relying on ``pose-dependent'' rigid\ntransformations. We model the clothed human holistically but with two distinct\nphysical entities in contact: clothing modeled as 3D Gaussians, which are\nattached to a skinned SMPL body that follows the movement of the person in the\ninput videos. The articulation of the SMPL body also drives physically-based\nsimulation of the clothes' Gaussians to transform the avatar to novel poses. In\norder to run position based dynamics simulation, physical properties including\nmass and material stiffness are estimated from the RGB videos through Dynamic\n3D Gaussian Splatting. Experiments demonstrate that our method not only\naccurately reproduces appearance but also enables the reconstruction of avatars\nwearing highly deformable garments, such as skirts or coats, which have been\nchallenging to reconstruct using existing methods.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.03103", "title": "MultiGO: Towards Multi-level Geometry Learning for Monocular 3D Textured\n  Human Reconstruction", "abstract": "  This paper investigates the research task of reconstructing the 3D clothed\nhuman body from a monocular image. Due to the inherent ambiguity of single-view\ninput, existing approaches leverage pre-trained SMPL(-X) estimation models or\ngenerative models to provide auxiliary information for human reconstruction.\nHowever, these methods capture only the general human body geometry and\noverlook specific geometric details, leading to inaccurate skeleton\nreconstruction, incorrect joint positions, and unclear cloth wrinkles. In\nresponse to these issues, we propose a multi-level geometry learning framework.\nTechnically, we design three key components: skeleton-level enhancement,\njoint-level augmentation, and wrinkle-level refinement modules. Specifically,\nwe effectively integrate the projected 3D Fourier features into a Gaussian\nreconstruction model, introduce perturbations to improve joint depth estimation\nduring training, and refine the human coarse wrinkles by resembling the\nde-noising process of diffusion model. Extensive quantitative and qualitative\nexperiments on two out-of-distribution test sets show the superior performance\nof our approach compared to state-of-the-art (SOTA) methods.\n", "date": "2024-12-04", "categories": []}
{"id": "2412.02684", "title": "AniGS: Animatable Gaussian Avatar from a Single Image with Inconsistent\n  Gaussian Reconstruction", "abstract": "  Generating animatable human avatars from a single image is essential for\nvarious digital human modeling applications. Existing 3D reconstruction methods\noften struggle to capture fine details in animatable models, while generative\napproaches for controllable animation, though avoiding explicit 3D modeling,\nsuffer from viewpoint inconsistencies in extreme poses and computational\ninefficiencies. In this paper, we address these challenges by leveraging the\npower of generative models to produce detailed multi-view canonical pose\nimages, which help resolve ambiguities in animatable human reconstruction. We\nthen propose a robust method for 3D reconstruction of inconsistent images,\nenabling real-time rendering during inference. Specifically, we adapt a\ntransformer-based video generation model to generate multi-view canonical pose\nimages and normal maps, pretraining on a large-scale video dataset to improve\ngeneralization. To handle view inconsistencies, we recast the reconstruction\nproblem as a 4D task and introduce an efficient 3D modeling approach using 4D\nGaussian Splatting. Experiments demonstrate that our method achieves\nphotorealistic, real-time animation of 3D human avatars from in-the-wild\nimages, showcasing its effectiveness and generalization capability.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.00845", "title": "SAGA: Surface-Aligned Gaussian Avatar", "abstract": "  This paper presents a Surface-Aligned Gaussian representation for creating\nanimatable human avatars from monocular videos,aiming at improving the novel\nview and pose synthesis performance while ensuring fast training and real-time\nrendering. Recently,3DGS has emerged as a more efficient and expressive\nalternative to NeRF, and has been used for creating dynamic human avatars.\nHowever,when applied to the severely ill-posed task of monocular dynamic\nreconstruction, the Gaussians tend to overfit the constantly changing regions\nsuch as clothes wrinkles or shadows since these regions cannot provide\nconsistent supervision, resulting in noisy geometry and abrupt deformation that\ntypically fail to generalize under novel views and poses.To address these\nlimitations, we present SAGA,i.e.,Surface-Aligned Gaussian Avatar,which aligns\nthe Gaussians with a mesh to enforce well-defined geometry and consistent\ndeformation, thereby improving generalization under novel views and poses.\nUnlike existing strict alignment methods that suffer from limited expressive\npower and low realism,SAGA employs a two-stage alignment strategy where the\nGaussians are first adhered on while then detached from the mesh, thus\nfacilitating both good geometry and high expressivity. In the Adhered Stage, we\nimprove the flexibility of Adhered-on-Mesh Gaussians by allowing them to flow\non the mesh, in contrast to existing methods that rigidly bind Gaussians to\nfixed location. In the second Detached Stage, we introduce a Gaussian-Mesh\nAlignment regularization, which allows us to unleash the expressivity by\ndetaching the Gaussians but maintain the geometric alignment by minimizing\ntheir location and orientation offsets from the bound triangles. Finally, since\nthe Gaussians may drift outside the bound triangles during optimization, an\nefficient Walking-on-Mesh strategy is proposed to dynamically update the bound\ntriangles.\n", "date": "2024-12-01", "categories": []}
{"id": "2412.01106", "title": "One Shot, One Talk: Whole-body Talking Avatar from a Single Image", "abstract": "  Building realistic and animatable avatars still requires minutes of\nmulti-view or monocular self-rotating videos, and most methods lack precise\ncontrol over gestures and expressions. To push this boundary, we address the\nchallenge of constructing a whole-body talking avatar from a single image. We\npropose a novel pipeline that tackles two critical issues: 1) complex dynamic\nmodeling and 2) generalization to novel gestures and expressions. To achieve\nseamless generalization, we leverage recent pose-guided image-to-video\ndiffusion models to generate imperfect video frames as pseudo-labels. To\novercome the dynamic modeling challenge posed by inconsistent and noisy\npseudo-videos, we introduce a tightly coupled 3DGS-mesh hybrid avatar\nrepresentation and apply several key regularizations to mitigate\ninconsistencies caused by imperfect labels. Extensive experiments on diverse\nsubjects demonstrate that our method enables the creation of a photorealistic,\nprecisely animatable, and expressive whole-body talking avatar from just a\nsingle image.\n", "date": "2024-12-02", "categories": []}
{"id": "2412.01179", "title": "Dual-Branch Graph Transformer Network for 3D Human Mesh Reconstruction\n  from Video", "abstract": "  Human Mesh Reconstruction (HMR) from monocular video plays an important role\nin human-robot interaction and collaboration. However, existing video-based\nhuman mesh reconstruction methods face a trade-off between accurate\nreconstruction and smooth motion. These methods design networks based on either\nRNNs or attention mechanisms to extract local temporal correlations or global\ntemporal dependencies, but the lack of complementary long-term information and\nlocal details limits their performance. To address this problem, we propose a\n\\textbf{D}ual-branch \\textbf{G}raph \\textbf{T}ransformer network for 3D human\nmesh \\textbf{R}econstruction from video, named DGTR. DGTR employs a dual-branch\nnetwork including a Global Motion Attention (GMA) branch and a Local Details\nRefine (LDR) branch to parallelly extract long-term dependencies and local\ncrucial information, helping model global human motion and local human details\n(e.g., local motion, tiny movement). Specifically, GMA utilizes a global\ntransformer to model long-term human motion. LDR combines modulated graph\nconvolutional networks and the transformer framework to aggregate local\ninformation in adjacent frames and extract crucial information of human\ndetails. Experiments demonstrate that our DGTR outperforms state-of-the-art\nvideo-based methods in reconstruction accuracy and maintains competitive motion\nsmoothness. Moreover, DGTR utilizes fewer parameters and FLOPs, which validate\nthe effectiveness and efficiency of the proposed DGTR. Code is publicly\navailable at\n\\href{https://github.com/TangTao-PKU/DGTR}{\\textcolor{myBlue}{https://github.com/TangTao-PKU/DGTR}}.\n", "date": "2024-12-02", "categories": []}
{"id": "2411.01512", "title": "InstantGeoAvatar: Effective Geometry and Appearance Modeling of\n  Animatable Avatars from Monocular Video", "abstract": "  We present InstantGeoAvatar, a method for efficient and effective learning\nfrom monocular video of detailed 3D geometry and appearance of animatable\nimplicit human avatars. Our key observation is that the optimization of a hash\ngrid encoding to represent a signed distance function (SDF) of the human\nsubject is fraught with instabilities and bad local minima. We thus propose a\nprincipled geometry-aware SDF regularization scheme that seamlessly fits into\nthe volume rendering pipeline and adds negligible computational overhead. Our\nregularization scheme significantly outperforms previous approaches for\ntraining SDFs on hash grids. We obtain competitive results in geometry\nreconstruction and novel view synthesis in as little as five minutes of\ntraining time, a significant reduction from the several hours required by\nprevious work. InstantGeoAvatar represents a significant leap forward towards\nachieving interactive reconstruction of virtual avatars.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.18675", "title": "GaussianSpeech: Audio-Driven Gaussian Avatars", "abstract": "  We introduce GaussianSpeech, a novel approach that synthesizes high-fidelity\nanimation sequences of photo-realistic, personalized 3D human head avatars from\nspoken audio. To capture the expressive, detailed nature of human heads,\nincluding skin furrowing and finer-scale facial movements, we propose to couple\nspeech signal with 3D Gaussian splatting to create realistic, temporally\ncoherent motion sequences. We propose a compact and efficient 3DGS-based avatar\nrepresentation that generates expression-dependent color and leverages wrinkle-\nand perceptually-based losses to synthesize facial details, including wrinkles\nthat occur with different expressions. To enable sequence modeling of 3D\nGaussian splats with audio, we devise an audio-conditioned transformer model\ncapable of extracting lip and expression features directly from audio input.\nDue to the absence of high-quality datasets of talking humans in correspondence\nwith audio, we captured a new large-scale multi-view dataset of audio-visual\nsequences of talking humans with native English accents and diverse facial\ngeometry. GaussianSpeech consistently achieves state-of-the-art performance\nwith visually natural motion at real time rendering rates, while encompassing\ndiverse facial expressions and styles.\n", "date": "2024-11-27", "categories": []}
{"id": "2411.15732", "title": "DynamicAvatars: Accurate Dynamic Facial Avatars Reconstruction and\n  Precise Editing with Diffusion Models", "abstract": "  Generating and editing dynamic 3D head avatars are crucial tasks in virtual\nreality and film production. However, existing methods often suffer from facial\ndistortions, inaccurate head movements, and limited fine-grained editing\ncapabilities. To address these challenges, we present DynamicAvatars, a dynamic\nmodel that generates photorealistic, moving 3D head avatars from video clips\nand parameters associated with facial positions and expressions. Our approach\nenables precise editing through a novel prompt-based editing model, which\nintegrates user-provided prompts with guiding parameters derived from large\nlanguage models (LLMs). To achieve this, we propose a dual-tracking framework\nbased on Gaussian Splatting and introduce a prompt preprocessing module to\nenhance editing stability. By incorporating a specialized GAN algorithm and\nconnecting it to our control module, which generates precise guiding parameters\nfrom LLMs, we successfully address the limitations of existing methods.\nAdditionally, we develop a dynamic editing strategy that selectively utilizes\nspecific training datasets to improve the efficiency and adaptability of the\nmodel for dynamic editing tasks.\n", "date": "2024-11-24", "categories": []}
{"id": "2411.16289", "title": "Utilizing Uncertainty in 2D Pose Detectors for Probabilistic 3D Human\n  Mesh Recovery", "abstract": "  Monocular 3D human pose and shape estimation is an inherently ill-posed\nproblem due to depth ambiguities, occlusions, and truncations. Recent\nprobabilistic approaches learn a distribution over plausible 3D human meshes by\nmaximizing the likelihood of the ground-truth pose given an image. We show that\nthis objective function alone is not sufficient to best capture the full\ndistributions. Instead, we propose to additionally supervise the learned\ndistributions by minimizing the distance to distributions encoded in heatmaps\nof a 2D pose detector. Moreover, we reveal that current methods often generate\nincorrect hypotheses for invisible joints which is not detected by the\nevaluation protocols. We demonstrate that person segmentation masks can be\nutilized during training to significantly decrease the number of invalid\nsamples and introduce two metrics to evaluate it. Our normalizing flow-based\napproach predicts plausible 3D human mesh hypotheses that are consistent with\nthe image evidence while maintaining high diversity for ambiguous body parts.\nExperiments on 3DPW and EMDB show that we outperform other state-of-the-art\nprobabilistic methods. Code is available for research purposes at\nhttps://github.com/twehrbein/humr.\n", "date": "2024-11-25", "categories": []}
{"id": "2312.03029", "title": "HHAvatar: Gaussian Head Avatar with Dynamic Hairs", "abstract": "  Creating high-fidelity 3D head avatars has always been a research hotspot,\nbut it remains a great challenge under lightweight sparse view setups. In this\npaper, we propose HHAvatar represented by controllable 3D Gaussians for\nhigh-fidelity head avatar with dynamic hair modeling. We first use 3D Gaussians\nto represent the appearance of the head, and then jointly optimize neutral 3D\nGaussians and a fully learned MLP-based deformation field to capture complex\nexpressions. The two parts benefit each other, thereby our method can model\nfine-grained dynamic details while ensuring expression accuracy. Furthermore,\nwe devise a well-designed geometry-guided initialization strategy based on\nimplicit SDF and Deep Marching Tetrahedra for the stability and convergence of\nthe training procedure. To address the problem of dynamic hair modeling, we\nintroduce a hybrid head model into our avatar representation based Gaussian\nHead Avatar and a training method that considers timing information and an\nocclusion perception module to model the non-rigid motion of hair. Experiments\nshow that our approach outperforms other state-of-the-art sparse-view methods,\nachieving ultra high-fidelity rendering quality at 2K resolution even under\nexaggerated expressions and driving hairs reasonably with the motion of the\nhead\n", "date": "2023-12-05", "categories": []}
{"id": "2408.09663", "title": "3D-Consistent Human Avatars with Sparse Inputs via Gaussian Splatting\n  and Contrastive Learning", "abstract": "  Existing approaches for human avatar generation--both NeRF-based and 3D\nGaussian Splatting (3DGS) based--struggle with maintaining 3D consistency and\nexhibit degraded detail reconstruction, particularly when training with sparse\ninputs. To address this challenge, we propose CHASE, a novel framework that\nachieves dense-input-level performance using only sparse inputs through two key\ninnovations: cross-pose intrinsic 3D consistency supervision and 3D geometry\ncontrastive learning. Building upon prior skeleton-driven approaches that\ncombine rigid deformation with non-rigid cloth dynamics, we first establish\nbaseline avatars with fundamental 3D consistency. To enhance 3D consistency\nunder sparse inputs, we introduce a Dynamic Avatar Adjustment (DAA) module,\nwhich refines deformed Gaussians by leveraging similar poses from the training\nset. By minimizing the rendering discrepancy between adjusted Gaussians and\nreference poses, DAA provides additional supervision for avatar reconstruction.\nWe further maintain global 3D consistency through a novel geometry-aware\ncontrastive learning strategy. While designed for sparse inputs, CHASE\nsurpasses state-of-the-art methods across both full and sparse settings on\nZJU-MoCap and H36M datasets, demonstrating that our enhanced 3D consistency\nleads to superior rendering quality.\n", "date": "2024-08-19", "categories": []}
{"id": "2408.09665", "title": "Topology-aware Human Avatars with Semantically-guided Gaussian Splatting", "abstract": "  Reconstructing photo-realistic and topology-aware animatable human avatars\nfrom monocular videos remains challenging in computer vision and graphics.\nRecently, methods using 3D Gaussians to represent the human body have emerged,\noffering faster optimization and real-time rendering. However, due to ignoring\nthe crucial role of human body semantic information which represents the\nexplicit topological and intrinsic structure within human body, they fail to\nachieve fine-detail reconstruction of human avatars. To address this issue, we\npropose SG-GS, which uses semantics-embedded 3D Gaussians, skeleton-driven\nrigid deformation, and non-rigid cloth dynamics deformation to create\nphoto-realistic human avatars. We then design a Semantic Human-Body Annotator\n(SHA) which utilizes SMPL's semantic prior for efficient body part semantic\nlabeling. The generated labels are used to guide the optimization of semantic\nattributes of Gaussian. To capture the explicit topological structure of the\nhuman body, we employ a 3D network that integrates both topological and\ngeometric associations for human avatar deformation. We further implement three\nkey strategies to enhance the semantic accuracy of 3D Gaussians and rendering\nquality: semantic projection with 2D regularization, semantic-guided density\nregularization and semantic-aware regularization with neighborhood consistency.\nExtensive experiments demonstrate that SG-GS achieves state-of-the-art geometry\nand appearance reconstruction performance.\n", "date": "2024-08-19", "categories": []}
{"id": "2411.10582", "title": "Motion Diffusion-Guided 3D Global HMR from a Dynamic Camera", "abstract": "  Motion capture technologies have transformed numerous fields, from the film\nand gaming industries to sports science and healthcare, by providing a tool to\ncapture and analyze human movement in great detail. The holy grail in the topic\nof monocular global human mesh and motion reconstruction (GHMR) is to achieve\naccuracy on par with traditional multi-view capture on any monocular videos\ncaptured with a dynamic camera, in-the-wild. This is a challenging task as the\nmonocular input has inherent depth ambiguity, and the moving camera adds\nadditional complexity as the rendered human motion is now a product of both\nhuman and camera movement. Not accounting for this confusion, existing GHMR\nmethods often output motions that are unrealistic, e.g. unaccounted root\ntranslation of the human causes foot sliding. We present DiffOpt, a novel 3D\nglobal HMR method using Diffusion Optimization. Our key insight is that recent\nadvances in human motion generation, such as the motion diffusion model (MDM),\ncontain a strong prior of coherent human motion. The core of our method is to\noptimize the initial motion reconstruction using the MDM prior. This step can\nlead to more globally coherent human motion. Our optimization jointly optimizes\nthe motion prior loss and reprojection loss to correctly disentangle the human\nand camera motions. We validate DiffOpt with video sequences from the\nElectromagnetic Database of Global 3D Human Pose and Shape in the Wild (EMDB)\nand Egobody, and demonstrate superior global human motion recovery capability\nover other state-of-the-art global HMR methods most prominently in long video\nsettings.\n", "date": "2024-11-15", "categories": []}
{"id": "2411.11214", "title": "DeforHMR: Vision Transformer with Deformable Cross-Attention for 3D\n  Human Mesh Recovery", "abstract": "  Human Mesh Recovery (HMR) is an important yet challenging problem with\napplications across various domains including motion capture, augmented\nreality, and biomechanics. Accurately predicting human pose parameters from a\nsingle image remains a challenging 3D computer vision task. In this work, we\nintroduce DeforHMR, a novel regression-based monocular HMR framework designed\nto enhance the prediction of human pose parameters using deformable attention\ntransformers. DeforHMR leverages a novel query-agnostic deformable\ncross-attention mechanism within the transformer decoder to effectively regress\nthe visual features extracted from a frozen pretrained vision transformer (ViT)\nencoder. The proposed deformable cross-attention mechanism allows the model to\nattend to relevant spatial features more flexibly and in a data-dependent\nmanner. Equipped with a transformer decoder capable of spatially-nuanced\nattention, DeforHMR achieves state-of-the-art performance for single-frame\nregression-based methods on the widely used 3D HMR benchmarks 3DPW and RICH. By\npushing the boundary on the field of 3D human mesh recovery through deformable\nattention, we introduce an new, effective paradigm for decoding local spatial\ninformation from large pretrained vision encoders in computer vision.\n", "date": "2024-11-18", "categories": []}
{"id": "2411.09952", "title": "GGAvatar: Reconstructing Garment-Separated 3D Gaussian Splatting Avatars\n  from Monocular Video", "abstract": "  Avatar modelling has broad applications in human animation and virtual\ntry-ons. Recent advancements in this field have focused on high-quality and\ncomprehensive human reconstruction but often overlook the separation of\nclothing from the body. To bridge this gap, this paper introduces GGAvatar\n(Garment-separated 3D Gaussian Splatting Avatar), which relies on monocular\nvideos. Through advanced parameterized templates and unique phased training,\nthis model effectively achieves decoupled, editable, and realistic\nreconstruction of clothed humans. Comparative evaluations with other costly\nmodels confirm GGAvatar's superior quality and efficiency in modelling both\nclothed humans and separable garments. The paper also showcases applications in\nclothing editing, as illustrated in Figure 1, highlighting the model's benefits\nand the advantages of effective disentanglement. The code is available at\nhttps://github.com/J-X-Chen/GGAvatar/.\n", "date": "2024-11-15", "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": "2411.08128", "title": "CameraHMR: Aligning People with Perspective", "abstract": "  We address the challenge of accurate 3D human pose and shape estimation from\nmonocular images. The key to accuracy and robustness lies in high-quality\ntraining data. Existing training datasets containing real images with pseudo\nground truth (pGT) use SMPLify to fit SMPL to sparse 2D joint locations,\nassuming a simplified camera with default intrinsics. We make two contributions\nthat improve pGT accuracy. First, to estimate camera intrinsics, we develop a\nfield-of-view prediction model (HumanFoV) trained on a dataset of images\ncontaining people. We use the estimated intrinsics to enhance the 4D-Humans\ndataset by incorporating a full perspective camera model during SMPLify\nfitting. Second, 2D joints provide limited constraints on 3D body shape,\nresulting in average-looking bodies. To address this, we use the BEDLAM dataset\nto train a dense surface keypoint detector. We apply this detector to the\n4D-Humans dataset and modify SMPLify to fit the detected keypoints, resulting\nin significantly more realistic body shapes. Finally, we upgrade the HMR2.0\narchitecture to include the estimated camera parameters. We iterate model\ntraining and SMPLify fitting initialized with the previously trained model.\nThis leads to more accurate pGT and a new model, CameraHMR, with\nstate-of-the-art accuracy. Code and pGT are available for research purposes.\n", "date": "2024-11-12", "categories": []}
{"id": "2211.07955", "title": "IntegratedPIFu: Integrated Pixel Aligned Implicit Function for\n  Single-view Human Reconstruction", "abstract": "  We propose IntegratedPIFu, a new pixel aligned implicit model that builds on\nthe foundation set by PIFuHD. IntegratedPIFu shows how depth and human parsing\ninformation can be predicted and capitalised upon in a pixel-aligned implicit\nmodel. In addition, IntegratedPIFu introduces depth oriented sampling, a novel\ntraining scheme that improve any pixel aligned implicit model ability to\nreconstruct important human features without noisy artefacts. Lastly,\nIntegratedPIFu presents a new architecture that, despite using less model\nparameters than PIFuHD, is able to improves the structural correctness of\nreconstructed meshes. Our results show that IntegratedPIFu significantly\noutperforms existing state of the arts methods on single view human\nreconstruction. Our code has been made available online.\n", "date": "2022-11-15", "categories": []}
{"id": "2402.19197", "title": "Fine Structure-Aware Sampling: A New Sampling Training Scheme for\n  Pixel-Aligned Implicit Models in Single-View Human Reconstruction", "abstract": "  Pixel-aligned implicit models, such as PIFu, PIFuHD, and ICON, are used for\nsingle-view clothed human reconstruction. These models need to be trained using\na sampling training scheme. Existing sampling training schemes either fail to\ncapture thin surfaces (e.g. ears, fingers) or cause noisy artefacts in\nreconstructed meshes. To address these problems, we introduce Fine\nStructured-Aware Sampling (FSS), a new sampling training scheme to train\npixel-aligned implicit models for single-view human reconstruction. FSS\nresolves the aforementioned problems by proactively adapting to the thickness\nand complexity of surfaces. In addition, unlike existing sampling training\nschemes, FSS shows how normals of sample points can be capitalized in the\ntraining process to improve results. Lastly, to further improve the training\nprocess, FSS proposes a mesh thickness loss signal for pixel-aligned implicit\nmodels. It becomes computationally feasible to introduce this loss once a\nslight reworking of the pixel-aligned implicit function framework is carried\nout. Our results show that our methods significantly outperform SOTA methods\nqualitatively and quantitatively. Our code is publicly available at\nhttps://github.com/kcyt/FSS.\n", "date": "2024-02-29", "categories": []}
{"id": "2411.04249", "title": "PocoLoco: A Point Cloud Diffusion Model of Human Shape in Loose Clothing", "abstract": "  Modeling a human avatar that can plausibly deform to articulations is an\nactive area of research. We present PocoLoco -- the first template-free,\npoint-based, pose-conditioned generative model for 3D humans in loose clothing.\nWe motivate our work by noting that most methods require a parametric model of\nthe human body to ground pose-dependent deformations. Consequently, they are\nrestricted to modeling clothing that is topologically similar to the naked body\nand do not extend well to loose clothing. The few methods that attempt to model\nloose clothing typically require either canonicalization or a\nUV-parameterization and need to address the challenging problem of explicitly\nestimating correspondences for the deforming clothes. In this work, we\nformulate avatar clothing deformation as a conditional point-cloud generation\ntask within the denoising diffusion framework. Crucially, our framework\noperates directly on unordered point clouds, eliminating the need for a\nparametric model or a clothing template. This also enables a variety of\npractical applications, such as point-cloud completion and pose-based editing\n-- important features for virtual human animation. As current datasets for\nhuman avatars in loose clothing are far too small for training diffusion\nmodels, we release a dataset of two subjects performing various poses in loose\nclothing with a total of 75K point clouds. By contributing towards tackling the\nchallenging task of effectively modeling loose clothing and expanding the\navailable data for training these models, we aim to set the stage for further\ninnovation in digital humans. The source code is available at\nhttps://github.com/sidsunny/pocoloco .\n", "date": "2024-11-06", "categories": []}
{"id": "2409.18057", "title": "LightAvatar: Efficient Head Avatar as Dynamic Neural Light Field", "abstract": "  Recent works have shown that neural radiance fields (NeRFs) on top of\nparametric models have reached SOTA quality to build photorealistic head\navatars from a monocular video. However, one major limitation of the NeRF-based\navatars is the slow rendering speed due to the dense point sampling of NeRF,\npreventing them from broader utility on resource-constrained devices. We\nintroduce LightAvatar, the first head avatar model based on neural light fields\n(NeLFs). LightAvatar renders an image from 3DMM parameters and a camera pose\nvia a single network forward pass, without using mesh or volume rendering. The\nproposed approach, while being conceptually appealing, poses a significant\nchallenge towards real-time efficiency and training stability. To resolve them,\nwe introduce dedicated network designs to obtain proper representations for the\nNeLF model and maintain a low FLOPs budget. Meanwhile, we tap into a\ndistillation-based training strategy that uses a pretrained avatar model as\nteacher to synthesize abundant pseudo data for training. A warping field\nnetwork is introduced to correct the fitting error in the real data so that the\nmodel can learn better. Extensive experiments suggest that our method can\nachieve new SOTA image quality quantitatively or qualitatively, while being\nsignificantly faster than the counterparts, reporting 174.1 FPS (512x512\nresolution) on a consumer-grade GPU (RTX3090) with no customized optimization.\n", "date": "2024-09-26", "categories": []}
{"id": "2411.04399", "title": "ProGraph: Temporally-alignable Probability Guided Graph Topological\n  Modeling for 3D Human Reconstruction", "abstract": "  Current 3D human motion reconstruction methods from monocular videos rely on\nfeatures within the current reconstruction window, leading to distortion and\ndeformations in the human structure under local occlusions or blurriness in\nvideo frames. To estimate realistic 3D human mesh sequences based on incomplete\nfeatures, we propose Temporally-alignable Probability Guided Graph Topological\nModeling for 3D Human Reconstruction (ProGraph). For missing parts recovery, we\nexploit the explicit topological-aware probability distribution across the\nentire motion sequence. To restore the complete human, Graph Topological\nModeling (GTM) learns the underlying topological structure, focusing on the\nrelationships inherent in the individual parts. Next, to generate blurred\nmotion parts, Temporal-alignable Probability Distribution (TPDist) utilizes the\nGTM to predict features based on distribution. This interactive mechanism\nfacilitates motion consistency, allowing the restoration of human parts.\nFurthermore, Hierarchical Human Loss (HHLoss) constrains the probability\ndistribution errors of inter-frame features during topological structure\nvariation. Our Method achieves superior results than other SOTA methods in\naddressing occlusions and blurriness on 3DPW.\n", "date": "2024-11-07", "categories": []}
{"id": "2409.16147", "title": "Gaussian Deja-vu: Creating Controllable 3D Gaussian Head-Avatars with\n  Enhanced Generalization and Personalization Abilities", "abstract": "  Recent advancements in 3D Gaussian Splatting (3DGS) have unlocked significant\npotential for modeling 3D head avatars, providing greater flexibility than\nmesh-based methods and more efficient rendering compared to NeRF-based\napproaches. Despite these advancements, the creation of controllable 3DGS-based\nhead avatars remains time-intensive, often requiring tens of minutes to hours.\nTo expedite this process, we here introduce the \"Gaussian Deja-vu\" framework,\nwhich first obtains a generalized model of the head avatar and then\npersonalizes the result. The generalized model is trained on large 2D\n(synthetic and real) image datasets. This model provides a well-initialized 3D\nGaussian head that is further refined using a monocular video to achieve the\npersonalized head avatar. For personalizing, we propose learnable\nexpression-aware rectification blendmaps to correct the initial 3D Gaussians,\nensuring rapid convergence without the reliance on neural networks. Experiments\ndemonstrate that the proposed method meets its objectives. It outperforms\nstate-of-the-art 3D Gaussian head avatars in terms of photorealistic quality as\nwell as reduces training time consumption to at least a quarter of the existing\nmethods, producing the avatar in minutes.\n", "date": "2024-09-23", "categories": []}
{"id": "2411.03047", "title": "GarVerseLOD: High-Fidelity 3D Garment Reconstruction from a Single\n  In-the-Wild Image using a Dataset with Levels of Details", "abstract": "  Neural implicit functions have brought impressive advances to the\nstate-of-the-art of clothed human digitization from multiple or even single\nimages. However, despite the progress, current arts still have difficulty\ngeneralizing to unseen images with complex cloth deformation and body poses. In\nthis work, we present GarVerseLOD, a new dataset and framework that paves the\nway to achieving unprecedented robustness in high-fidelity 3D garment\nreconstruction from a single unconstrained image. Inspired by the recent\nsuccess of large generative models, we believe that one key to addressing the\ngeneralization challenge lies in the quantity and quality of 3D garment data.\nTowards this end, GarVerseLOD collects 6,000 high-quality cloth models with\nfine-grained geometry details manually created by professional artists. In\naddition to the scale of training data, we observe that having disentangled\ngranularities of geometry can play an important role in boosting the\ngeneralization capability and inference accuracy of the learned model. We hence\ncraft GarVerseLOD as a hierarchical dataset with levels of details (LOD),\nspanning from detail-free stylized shape to pose-blended garment with\npixel-aligned details. This allows us to make this highly under-constrained\nproblem tractable by factorizing the inference into easier tasks, each narrowed\ndown with smaller searching space. To ensure GarVerseLOD can generalize well to\nin-the-wild images, we propose a novel labeling paradigm based on conditional\ndiffusion models to generate extensive paired images for each garment model\nwith high photorealism. We evaluate our method on a massive amount of\nin-the-wild images. Experimental results demonstrate that GarVerseLOD can\ngenerate standalone garment pieces with significantly better quality than prior\napproaches. Project page: https://garverselod.github.io/\n", "date": "2024-11-05", "categories": []}
{"id": "2411.03086", "title": "HFGaussian: Learning Generalizable Gaussian Human with Integrated Human\n  Features", "abstract": "  Recent advancements in radiance field rendering show promising results in 3D\nscene representation, where Gaussian splatting-based techniques emerge as\nstate-of-the-art due to their quality and efficiency. Gaussian splatting is\nwidely used for various applications, including 3D human representation.\nHowever, previous 3D Gaussian splatting methods either use parametric body\nmodels as additional information or fail to provide any underlying structure,\nlike human biomechanical features, which are essential for different\napplications. In this paper, we present a novel approach called HFGaussian that\ncan estimate novel views and human features, such as the 3D skeleton, 3D key\npoints, and dense pose, from sparse input images in real time at 25 FPS. The\nproposed method leverages generalizable Gaussian splatting technique to\nrepresent the human subject and its associated features, enabling efficient and\ngeneralizable reconstruction. By incorporating a pose regression network and\nthe feature splatting technique with Gaussian splatting, HFGaussian\ndemonstrates improved capabilities over existing 3D human methods, showcasing\nthe potential of 3D human representations with integrated biomechanics. We\nthoroughly evaluate our HFGaussian method against the latest state-of-the-art\ntechniques in human Gaussian splatting and pose estimation, demonstrating its\nreal-time, state-of-the-art performance.\n", "date": "2024-11-05", "categories": []}
{"id": "2403.18784", "title": "SplatFace: Gaussian Splat Face Reconstruction Leveraging an Optimizable\n  Surface", "abstract": "  We present SplatFace, a novel Gaussian splatting framework designed for 3D\nhuman face reconstruction without reliance on accurate pre-determined geometry.\nOur method is designed to simultaneously deliver both high-quality novel view\nrendering and accurate 3D mesh reconstructions. We incorporate a generic 3D\nMorphable Model (3DMM) to provide a surface geometric structure, making it\npossible to reconstruct faces with a limited set of input images. We introduce\na joint optimization strategy that refines both the Gaussians and the morphable\nsurface through a synergistic non-rigid alignment process. A novel distance\nmetric, splat-to-surface, is proposed to improve alignment by considering both\nthe Gaussian position and covariance. The surface information is also utilized\nto incorporate a world-space densification process, resulting in superior\nreconstruction quality. Our experimental analysis demonstrates that the\nproposed method is competitive with both other Gaussian splatting techniques in\nnovel view synthesis and other 3D reconstruction methods in producing 3D face\nmeshes with high geometric precision.\n", "date": "2024-03-27", "categories": []}
{"id": "2410.23800", "title": "SOAR: Self-Occluded Avatar Recovery from a Single Video In the Wild", "abstract": "  Self-occlusion is common when capturing people in the wild, where the\nperformer do not follow predefined motion scripts. This challenges existing\nmonocular human reconstruction systems that assume full body visibility. We\nintroduce Self-Occluded Avatar Recovery (SOAR), a method for complete human\nreconstruction from partial observations where parts of the body are entirely\nunobserved. SOAR leverages structural normal prior and generative diffusion\nprior to address such an ill-posed reconstruction problem. For structural\nnormal prior, we model human with an reposable surfel model with well-defined\nand easily readable shapes. For generative diffusion prior, we perform an\ninitial reconstruction and refine it using score distillation. On various\nbenchmarks, we show that SOAR performs favorably than state-of-the-art\nreconstruction and generation methods, and on-par comparing to concurrent\nworks. Additional video results and code are available at\nhttps://soar-avatar.github.io/.\n", "date": "2024-10-31", "categories": []}
{"id": "2410.24223", "title": "URAvatar: Universal Relightable Gaussian Codec Avatars", "abstract": "  We present a new approach to creating photorealistic and relightable head\navatars from a phone scan with unknown illumination. The reconstructed avatars\ncan be animated and relit in real time with the global illumination of diverse\nenvironments. Unlike existing approaches that estimate parametric reflectance\nparameters via inverse rendering, our approach directly models learnable\nradiance transfer that incorporates global light transport in an efficient\nmanner for real-time rendering. However, learning such a complex light\ntransport that can generalize across identities is non-trivial. A phone scan in\na single environment lacks sufficient information to infer how the head would\nappear in general environments. To address this, we build a universal\nrelightable avatar model represented by 3D Gaussians. We train on hundreds of\nhigh-quality multi-view human scans with controllable point lights.\nHigh-resolution geometric guidance further enhances the reconstruction accuracy\nand generalization. Once trained, we finetune the pretrained model on a phone\nscan using inverse rendering to obtain a personalized relightable avatar. Our\nexperiments establish the efficacy of our design, outperforming existing\napproaches while retaining real-time rendering capability.\n", "date": "2024-10-31", "categories": []}
{"id": "2406.12459", "title": "HumanSplat: Generalizable Single-Image Human Gaussian Splatting with\n  Structure Priors", "abstract": "  Despite recent advancements in high-fidelity human reconstruction techniques,\nthe requirements for densely captured images or time-consuming per-instance\noptimization significantly hinder their applications in broader scenarios. To\ntackle these issues, we present HumanSplat which predicts the 3D Gaussian\nSplatting properties of any human from a single input image in a generalizable\nmanner. In particular, HumanSplat comprises a 2D multi-view diffusion model and\na latent reconstruction transformer with human structure priors that adeptly\nintegrate geometric priors and semantic features within a unified framework. A\nhierarchical loss that incorporates human semantic information is further\ndesigned to achieve high-fidelity texture modeling and better constrain the\nestimated multiple views. Comprehensive experiments on standard benchmarks and\nin-the-wild images demonstrate that HumanSplat surpasses existing\nstate-of-the-art methods in achieving photorealistic novel-view synthesis.\n", "date": "2024-06-18", "categories": []}
{"id": "2409.18364", "title": "Multi-hypotheses Conditioned Point Cloud Diffusion for 3D Human\n  Reconstruction from Occluded Images", "abstract": "  3D human shape reconstruction under severe occlusion due to human-object or\nhuman-human interaction is a challenging problem. Parametric models i.e.,\nSMPL(-X), which are based on the statistics across human shapes, can represent\nwhole human body shapes but are limited to minimally-clothed human shapes.\nImplicit-function-based methods extract features from the parametric models to\nemploy prior knowledge of human bodies and can capture geometric details such\nas clothing and hair. However, they often struggle to handle misaligned\nparametric models and inpaint occluded regions given a single RGB image. In\nthis work, we propose a novel pipeline, MHCDIFF, Multi-hypotheses Conditioned\nPoint Cloud Diffusion, composed of point cloud diffusion conditioned on\nprobabilistic distributions for pixel-aligned detailed 3D human reconstruction\nunder occlusion. Compared to previous implicit-function-based methods, the\npoint cloud diffusion model can capture the global consistent features to\ngenerate the occluded regions, and the denoising process corrects the\nmisaligned SMPL meshes. The core of MHCDIFF is extracting local features from\nmultiple hypothesized SMPL(-X) meshes and aggregating the set of features to\ncondition the diffusion model. In the experiments on CAPE and MultiHuman\ndatasets, the proposed method outperforms various SOTA methods based on SMPL,\nimplicit functions, point cloud diffusion, and their combined, under synthetic\nand real occlusions. Our code is publicly available at\nhttps://donghwankim0101.github.io/projects/mhcdiff/ .\n", "date": "2024-09-27", "categories": []}
{"id": "2407.15070", "title": "GPHM: Gaussian Parametric Head Model for Monocular Head Avatar\n  Reconstruction", "abstract": "  Creating high-fidelity 3D human head avatars is crucial for applications in\nVR/AR, digital human, and film production. Recent advances have leveraged\nmorphable face models to generate animated head avatars from easily accessible\ndata, representing varying identities and expressions within a low-dimensional\nparametric space. However, existing methods often struggle with modeling\ncomplex appearance details, e.g., hairstyles, and suffer from low rendering\nquality and efficiency. In this paper we introduce a novel approach, 3D\nGaussian Parametric Head Model, which employs 3D Gaussians to accurately\nrepresent the complexities of the human head, allowing precise control over\nboth identity and expression. The Gaussian model can handle intricate details,\nenabling realistic representations of varying appearances and complex\nexpressions. Furthermore, we presents a well-designed training framework to\nensure smooth convergence, providing a robust guarantee for learning the rich\ncontent. Our method achieves high-quality, photo-realistic rendering with\nreal-time efficiency, making it a valuable contribution to the field of\nparametric head models. Finally, we apply the 3D Gaussian Parametric Head Model\nto monocular video or few-shot head avatar reconstruction tasks, which enables\ninstant reconstruction of high-quality 3D head avatars even when input data is\nextremely limited, surpassing previous methods in terms of reconstruction\nquality and training speed.\n", "date": "2024-07-21", "categories": []}
{"id": "2410.17741", "title": "Efficient Neural Implicit Representation for 3D Human Reconstruction", "abstract": "  High-fidelity digital human representations are increasingly in demand in the\ndigital world, particularly for interactive telepresence, AR/VR, 3D graphics,\nand the rapidly evolving metaverse. Even though they work well in small spaces,\nconventional methods for reconstructing 3D human motion frequently require the\nuse of expensive hardware and have high processing costs. This study presents\nHumanAvatar, an innovative approach that efficiently reconstructs precise human\navatars from monocular video sources. At the core of our methodology, we\nintegrate the pre-trained HuMoR, a model celebrated for its proficiency in\nhuman motion estimation. This is adeptly fused with the cutting-edge neural\nradiance field technology, Instant-NGP, and the state-of-the-art articulated\nmodel, Fast-SNARF, to enhance the reconstruction fidelity and speed. By\ncombining these two technologies, a system is created that can render quickly\nand effectively while also providing estimation of human pose parameters that\nare unmatched in accuracy. We have enhanced our system with an advanced\nposture-sensitive space reduction technique, which optimally balances rendering\nquality with computational efficiency. In our detailed experimental analysis\nusing both artificial and real-world monocular videos, we establish the\nadvanced performance of our approach. HumanAvatar consistently equals or\nsurpasses contemporary leading-edge reconstruction techniques in quality.\nFurthermore, it achieves these complex reconstructions in minutes, a fraction\nof the time typically required by existing methods. Our models achieve a\ntraining speed that is 110X faster than that of State-of-The-Art (SoTA)\nNeRF-based models. Our technique performs noticeably better than SoTA dynamic\nhuman NeRF methods if given an identical runtime limit. HumanAvatar can provide\neffective visuals after only 30 seconds of training.\n", "date": "2024-10-23", "categories": []}
{"id": "2410.16009", "title": "3D-GANTex: 3D Face Reconstruction with StyleGAN3-based Multi-View Images\n  and 3DDFA based Mesh Generation", "abstract": "  Geometry and texture estimation from a single face image is an ill-posed\nproblem since there is very little information to work with. The problem\nfurther escalates when the face is rotated at a different angle. This paper\ntries to tackle this problem by introducing a novel method for texture\nestimation from a single image by first using StyleGAN and 3D Morphable Models.\nThe method begins by generating multi-view faces using the latent space of GAN.\nThen 3DDFA trained on 3DMM estimates a 3D face mesh as well as a\nhigh-resolution texture map that is consistent with the estimated face shape.\nThe result shows that the generated mesh is of high quality with near to\naccurate texture representation.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.16337", "title": "Disambiguating Monocular Reconstruction of 3D Clothed Human with\n  Spatial-Temporal Transformer", "abstract": "  Reconstructing 3D clothed humans from monocular camera data is highly\nchallenging due to viewpoint limitations and image ambiguity. While implicit\nfunction-based approaches, combined with prior knowledge from parametric\nmodels, have made significant progress, there are still two notable problems.\nFirstly, the back details of human models are ambiguous due to viewpoint\ninvisibility. The quality of the back details depends on the back normal map\npredicted by a convolutional neural network (CNN). However, the CNN lacks\nglobal information awareness for comprehending the back texture, resulting in\nexcessively smooth back details. Secondly, a single image suffers from local\nambiguity due to lighting conditions and body movement. However, implicit\nfunctions are highly sensitive to pixel variations in ambiguous regions. To\naddress these ambiguities, we propose the Spatial-Temporal Transformer (STT)\nnetwork for 3D clothed human reconstruction. A spatial transformer is employed\nto extract global information for normal map prediction. The establishment of\nglobal correlations facilitates the network in comprehending the holistic\ntexture and shape of the human body. Simultaneously, to compensate for local\nambiguity in images, a temporal transformer is utilized to extract temporal\nfeatures from adjacent frames. The incorporation of temporal features can\nenhance the accuracy of input features in implicit networks. Furthermore, to\nobtain more accurate temporal features, joint tokens are employed to establish\nlocal correspondences between frames. Experimental results on the Adobe and\nMonoPerfCap datasets have shown that our method outperforms state-of-the-art\nmethods and maintains robust generalization even under low-light outdoor\nconditions.\n", "date": "2024-10-21", "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": "2410.15582", "title": "ARTS: Semi-Analytical Regressor using Disentangled Skeletal\n  Representations for Human Mesh Recovery from Videos", "abstract": "  Although existing video-based 3D human mesh recovery methods have made\nsignificant progress, simultaneously estimating human pose and shape from\nlow-resolution image features limits their performance. These image features\nlack sufficient spatial information about the human body and contain various\nnoises (e.g., background, lighting, and clothing), which often results in\ninaccurate pose and inconsistent motion. Inspired by the rapid advance in human\npose estimation, we discover that compared to image features, skeletons\ninherently contain accurate human pose and motion. Therefore, we propose a\nnovel semiAnalytical Regressor using disenTangled Skeletal representations for\nhuman mesh recovery from videos, called ARTS. Specifically, a skeleton\nestimation and disentanglement module is proposed to estimate the 3D skeletons\nfrom a video and decouple them into disentangled skeletal representations\n(i.e., joint position, bone length, and human motion). Then, to fully utilize\nthese representations, we introduce a semi-analytical regressor to estimate the\nparameters of the human mesh model. The regressor consists of three modules:\nTemporal Inverse Kinematics (TIK), Bone-guided Shape Fitting (BSF), and\nMotion-Centric Refinement (MCR). TIK utilizes joint position to estimate\ninitial pose parameters and BSF leverages bone length to regress bone-aligned\nshape parameters. Finally, MCR combines human motion representation with image\nfeatures to refine the initial human model parameters. Extensive experiments\ndemonstrate that our ARTS surpasses existing state-of-the-art video-based\nmethods in both per-frame accuracy and temporal consistency on popular\nbenchmarks: 3DPW, MPI-INF-3DHP, and Human3.6M. Code is available at\nhttps://github.com/TangTao-PKU/ARTS.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.09690", "title": "FAMOUS: High-Fidelity Monocular 3D Human Digitization Using View\n  Synthesis", "abstract": "  The advancement in deep implicit modeling and articulated models has\nsignificantly enhanced the process of digitizing human figures in 3D from just\na single image. While state-of-the-art methods have greatly improved geometric\nprecision, the challenge of accurately inferring texture remains, particularly\nin obscured areas such as the back of a person in frontal-view images. This\nlimitation in texture prediction largely stems from the scarcity of large-scale\nand diverse 3D datasets, whereas their 2D counterparts are abundant and easily\naccessible. To address this issue, our paper proposes leveraging extensive 2D\nfashion datasets to enhance both texture and shape prediction in 3D human\ndigitization. We incorporate 2D priors from the fashion dataset to learn the\noccluded back view, refined with our proposed domain alignment strategy. We\nthen fuse this information with the input image to obtain a fully textured mesh\nof the given person. Through extensive experimentation on standard 3D human\nbenchmarks, we demonstrate the superior performance of our approach in terms of\nboth texture and geometry. Code and dataset is available at\nhttps://github.com/humansensinglab/FAMOUS.\n", "date": "2024-10-13", "categories": []}
{"id": "2410.08840", "title": "Learning Interaction-aware 3D Gaussian Splatting for One-shot Hand\n  Avatars", "abstract": "  In this paper, we propose to create animatable avatars for interacting hands\nwith 3D Gaussian Splatting (GS) and single-image inputs. Existing GS-based\nmethods designed for single subjects often yield unsatisfactory results due to\nlimited input views, various hand poses, and occlusions. To address these\nchallenges, we introduce a novel two-stage interaction-aware GS framework that\nexploits cross-subject hand priors and refines 3D Gaussians in interacting\nareas. Particularly, to handle hand variations, we disentangle the 3D\npresentation of hands into optimization-based identity maps and learning-based\nlatent geometric features and neural texture maps. Learning-based features are\ncaptured by trained networks to provide reliable priors for poses, shapes, and\ntextures, while optimization-based identity maps enable efficient one-shot\nfitting of out-of-distribution hands. Furthermore, we devise an\ninteraction-aware attention module and a self-adaptive Gaussian refinement\nmodule. These modules enhance image rendering quality in areas with intra- and\ninter-hand interactions, overcoming the limitations of existing GS-based\nmethods. Our proposed method is validated via extensive experiments on the\nlarge-scale InterHand2.6M dataset, and it significantly improves the\nstate-of-the-art performance in image quality. Project Page:\n\\url{https://github.com/XuanHuang0/GuassianHand}.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.07971", "title": "Generalizable and Animatable Gaussian Head Avatar", "abstract": "  In this paper, we propose Generalizable and Animatable Gaussian head Avatar\n(GAGAvatar) for one-shot animatable head avatar reconstruction. Existing\nmethods rely on neural radiance fields, leading to heavy rendering consumption\nand low reenactment speeds. To address these limitations, we generate the\nparameters of 3D Gaussians from a single image in a single forward pass. The\nkey innovation of our work is the proposed dual-lifting method, which produces\nhigh-fidelity 3D Gaussians that capture identity and facial details.\nAdditionally, we leverage global image features and the 3D morphable model to\nconstruct 3D Gaussians for controlling expressions. After training, our model\ncan reconstruct unseen identities without specific optimizations and perform\nreenactment rendering at real-time speeds. Experiments show that our method\nexhibits superior performance compared to previous methods in terms of\nreconstruction quality and expression accuracy. We believe our method can\nestablish new benchmarks for future research and advance applications of\ndigital avatars. Code and demos are available\nhttps://github.com/xg-chu/GAGAvatar.\n", "date": "2024-10-10", "categories": []}
{"id": "2404.01053", "title": "HAHA: Highly Articulated Gaussian Human Avatars with Textured Mesh Prior", "abstract": "  We present HAHA - a novel approach for animatable human avatar generation\nfrom monocular input videos. The proposed method relies on learning the\ntrade-off between the use of Gaussian splatting and a textured mesh for\nefficient and high fidelity rendering. We demonstrate its efficiency to animate\nand render full-body human avatars controlled via the SMPL-X parametric model.\nOur model learns to apply Gaussian splatting only in areas of the SMPL-X mesh\nwhere it is necessary, like hair and out-of-mesh clothing. This results in a\nminimal number of Gaussians being used to represent the full avatar, and\nreduced rendering artifacts. This allows us to handle the animation of small\nbody parts such as fingers that are traditionally disregarded. We demonstrate\nthe effectiveness of our approach on two open datasets: SnapshotPeople and\nX-Humans. Our method demonstrates on par reconstruction quality to the\nstate-of-the-art on SnapshotPeople, while using less than a third of Gaussians.\nHAHA outperforms previous state-of-the-art on novel poses from X-Humans both\nquantitatively and qualitatively.\n", "date": "2024-04-01", "categories": []}
{"id": "2409.20563", "title": "DressRecon: Freeform 4D Human Reconstruction from Monocular Video", "abstract": "  We present a method to reconstruct time-consistent human body models from\nmonocular videos, focusing on extremely loose clothing or handheld object\ninteractions. Prior work in human reconstruction is either limited to tight\nclothing with no object interactions, or requires calibrated multi-view\ncaptures or personalized template scans which are costly to collect at scale.\nOur key insight for high-quality yet flexible reconstruction is the careful\ncombination of generic human priors about articulated body shape (learned from\nlarge-scale training data) with video-specific articulated \"bag-of-bones\"\ndeformation (fit to a single video via test-time optimization). We accomplish\nthis by learning a neural implicit model that disentangles body versus clothing\ndeformations as separate motion model layers. To capture subtle geometry of\nclothing, we leverage image-based priors such as human body pose, surface\nnormals, and optical flow during optimization. The resulting neural fields can\nbe extracted into time-consistent meshes, or further optimized as explicit 3D\nGaussians for high-fidelity interactive rendering. On datasets with highly\nchallenging clothing deformations and object interactions, DressRecon yields\nhigher-fidelity 3D reconstructions than prior art. Project page:\nhttps://jefftan969.github.io/dressrecon/\n", "date": "2024-09-30", "categories": []}
{"id": "2410.00630", "title": "Cafca: High-quality Novel View Synthesis of Expressive Faces from Casual\n  Few-shot Captures", "abstract": "  Volumetric modeling and neural radiance field representations have\nrevolutionized 3D face capture and photorealistic novel view synthesis.\nHowever, these methods often require hundreds of multi-view input images and\nare thus inapplicable to cases with less than a handful of inputs. We present a\nnovel volumetric prior on human faces that allows for high-fidelity expressive\nface modeling from as few as three input views captured in the wild. Our key\ninsight is that an implicit prior trained on synthetic data alone can\ngeneralize to extremely challenging real-world identities and expressions and\nrender novel views with fine idiosyncratic details like wrinkles and eyelashes.\nWe leverage a 3D Morphable Face Model to synthesize a large training set,\nrendering each identity with different expressions, hair, clothing, and other\nassets. We then train a conditional Neural Radiance Field prior on this\nsynthetic dataset and, at inference time, fine-tune the model on a very sparse\nset of real images of a single subject. On average, the fine-tuning requires\nonly three inputs to cross the synthetic-to-real domain gap. The resulting\npersonalized 3D model reconstructs strong idiosyncratic facial expressions and\noutperforms the state-of-the-art in high-quality novel view synthesis of faces\nfrom sparse inputs in terms of perceptual and photo-metric quality.\n", "date": "2024-10-01", "categories": []}
{"id": "2409.15269", "title": "ReLoo: Reconstructing Humans Dressed in Loose Garments from Monocular\n  Video in the Wild", "abstract": "  While previous years have seen great progress in the 3D reconstruction of\nhumans from monocular videos, few of the state-of-the-art methods are able to\nhandle loose garments that exhibit large non-rigid surface deformations during\narticulation. This limits the application of such methods to humans that are\ndressed in standard pants or T-shirts. Our method, ReLoo, overcomes this\nlimitation and reconstructs high-quality 3D models of humans dressed in loose\ngarments from monocular in-the-wild videos. To tackle this problem, we first\nestablish a layered neural human representation that decomposes clothed humans\ninto a neural inner body and outer clothing. On top of the layered neural\nrepresentation, we further introduce a non-hierarchical virtual bone\ndeformation module for the clothing layer that can freely move, which allows\nthe accurate recovery of non-rigidly deforming loose clothing. A global\noptimization jointly optimizes the shape, appearance, and deformations of the\nhuman body and clothing via multi-layer differentiable volume rendering. To\nevaluate ReLoo, we record subjects with dynamically deforming garments in a\nmulti-view capture studio. This evaluation, both on existing and our novel\ndataset, demonstrates ReLoo's clear superiority over prior art on both indoor\ndatasets and in-the-wild videos.\n", "date": "2024-09-23", "categories": []}
{"id": "2409.20530", "title": "Dual Encoder GAN Inversion for High-Fidelity 3D Head Reconstruction from\n  Single Images", "abstract": "  3D GAN inversion aims to project a single image into the latent space of a 3D\nGenerative Adversarial Network (GAN), thereby achieving 3D geometry\nreconstruction. While there exist encoders that achieve good results in 3D GAN\ninversion, they are predominantly built on EG3D, which specializes in\nsynthesizing near-frontal views and is limiting in synthesizing comprehensive\n3D scenes from diverse viewpoints. In contrast to existing approaches, we\npropose a novel framework built on PanoHead, which excels in synthesizing\nimages from a 360-degree perspective. To achieve realistic 3D modeling of the\ninput image, we introduce a dual encoder system tailored for high-fidelity\nreconstruction and realistic generation from different viewpoints. Accompanying\nthis, we propose a stitching framework on the triplane domain to get the best\npredictions from both. To achieve seamless stitching, both encoders must output\nconsistent results despite being specialized for different tasks. For this\nreason, we carefully train these encoders using specialized losses, including\nan adversarial loss based on our novel occlusion-aware triplane discriminator.\nExperiments reveal that our approach surpasses the existing encoder training\nmethods qualitatively and quantitatively. Please visit the project page:\nhttps://berkegokmen1.github.io/dual-enc-3d-gan-inv.\n", "date": "2024-09-30", "categories": []}
{"id": "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": "2409.17280", "title": "Disco4D: Disentangled 4D Human Generation and Animation from a Single\n  Image", "abstract": "  We present \\textbf{Disco4D}, a novel Gaussian Splatting framework for 4D\nhuman generation and animation from a single image. Different from existing\nmethods, Disco4D distinctively disentangles clothings (with Gaussian models)\nfrom the human body (with SMPL-X model), significantly enhancing the generation\ndetails and flexibility. It has the following technical innovations.\n\\textbf{1)} Disco4D learns to efficiently fit the clothing Gaussians over the\nSMPL-X Gaussians. \\textbf{2)} It adopts diffusion models to enhance the 3D\ngeneration process, \\textit{e.g.}, modeling occluded parts not visible in the\ninput image. \\textbf{3)} It learns an identity encoding for each clothing\nGaussian to facilitate the separation and extraction of clothing assets.\nFurthermore, Disco4D naturally supports 4D human animation with vivid dynamics.\nExtensive experiments demonstrate the superiority of Disco4D on 4D human\ngeneration and animation tasks. Our visualizations can be found in\n\\url{https://disco-4d.github.io/}.\n", "date": "2024-09-25", "categories": []}
{"id": "2409.16666", "title": "TalkinNeRF: Animatable Neural Fields for Full-Body Talking Humans", "abstract": "  We introduce a novel framework that learns a dynamic neural radiance field\n(NeRF) for full-body talking humans from monocular videos. Prior work\nrepresents only the body pose or the face. However, humans communicate with\ntheir full body, combining body pose, hand gestures, as well as facial\nexpressions. In this work, we propose TalkinNeRF, a unified NeRF-based network\nthat represents the holistic 4D human motion. Given a monocular video of a\nsubject, we learn corresponding modules for the body, face, and hands, that are\ncombined together to generate the final result. To capture complex finger\narticulation, we learn an additional deformation field for the hands. Our\nmulti-identity representation enables simultaneous training for multiple\nsubjects, as well as robust animation under completely unseen poses. It can\nalso generalize to novel identities, given only a short video as input. We\ndemonstrate state-of-the-art performance for animating full-body talking\nhumans, with fine-grained hand articulation and facial expressions.\n", "date": "2024-09-25", "categories": []}
{"id": "2409.16863", "title": "Towards Unified 3D Hair Reconstruction from Single-View Portraits", "abstract": "  Single-view 3D hair reconstruction is challenging, due to the wide range of\nshape variations among diverse hairstyles. Current state-of-the-art methods are\nspecialized in recovering un-braided 3D hairs and often take braided styles as\ntheir failure cases, because of the inherent difficulty to define priors for\ncomplex hairstyles, whether rule-based or data-based. We propose a novel\nstrategy to enable single-view 3D reconstruction for a variety of hair types\nvia a unified pipeline. To achieve this, we first collect a large-scale\nsynthetic multi-view hair dataset SynMvHair with diverse 3D hair in both\nbraided and un-braided styles, and learn two diffusion priors specialized on\nhair. Then we optimize 3D Gaussian-based hair from the priors with two\nspecially designed modules, i.e. view-wise and pixel-wise Gaussian refinement.\nOur experiments demonstrate that reconstructing braided and un-braided 3D hair\nfrom single-view images via a unified approach is possible and our method\nachieves the state-of-the-art performance in recovering complex hairstyles. It\nis worth to mention that our method shows good generalization ability to real\nimages, although it learns hair priors from synthetic data.\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": "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": "2409.14249", "title": "End to End Face Reconstruction via Differentiable PnP", "abstract": "  This is a challenge report of the ECCV 2022 WCPA Challenge, Face\nReconstruction Track. Inside this report is a brief explanation of how we\naccomplish this challenge. We design a two-branch network to accomplish this\ntask, whose roles are Face Reconstruction and Face Landmark Detection. The\nformer outputs canonical 3D face coordinates. The latter outputs pixel\ncoordinates, i.e. 2D mapping of 3D coordinates with head pose and perspective\nprojection. In addition, we utilize a differentiable PnP (Perspective-n-Points)\nlayer to finetune the outputs of the two branch. Our method achieves very\ncompetitive quantitative results on the MVP-Human dataset and wins a $3^{rd}$\nprize in the challenge.\n", "date": "2024-09-21", "categories": []}
{"id": "2409.14778", "title": "Human Hair Reconstruction with Strand-Aligned 3D Gaussians", "abstract": "  We introduce a new hair modeling method that uses a dual representation of\nclassical hair strands and 3D Gaussians to produce accurate and realistic\nstrand-based reconstructions from multi-view data. In contrast to recent\napproaches that leverage unstructured Gaussians to model human avatars, our\nmethod reconstructs the hair using 3D polylines, or strands. This fundamental\ndifference allows the use of the resulting hairstyles out-of-the-box in modern\ncomputer graphics engines for editing, rendering, and simulation. Our 3D\nlifting method relies on unstructured Gaussians to generate multi-view ground\ntruth data to supervise the fitting of hair strands. The hairstyle itself is\nrepresented in the form of the so-called strand-aligned 3D Gaussians. This\nrepresentation allows us to combine strand-based hair priors, which are\nessential for realistic modeling of the inner structure of hairstyles, with the\ndifferentiable rendering capabilities of 3D Gaussian Splatting. Our method,\nnamed Gaussian Haircut, is evaluated on synthetic and real scenes and\ndemonstrates state-of-the-art performance in the task of strand-based hair\nreconstruction.\n", "date": "2024-09-23", "categories": []}
{"id": "2409.11951", "title": "GaussianHeads: End-to-End Learning of Drivable Gaussian Head Avatars\n  from Coarse-to-fine Representations", "abstract": "  Real-time rendering of human head avatars is a cornerstone of many computer\ngraphics applications, such as augmented reality, video games, and films, to\nname a few. Recent approaches address this challenge with computationally\nefficient geometry primitives in a carefully calibrated multi-view setup.\nAlbeit producing photorealistic head renderings, it often fails to represent\ncomplex motion changes such as the mouth interior and strongly varying head\nposes. We propose a new method to generate highly dynamic and deformable human\nhead avatars from multi-view imagery in real-time. At the core of our method is\na hierarchical representation of head models that allows to capture the complex\ndynamics of facial expressions and head movements. First, with rich facial\nfeatures extracted from raw input frames, we learn to deform the coarse facial\ngeometry of the template mesh. We then initialize 3D Gaussians on the deformed\nsurface and refine their positions in a fine step. We train this coarse-to-fine\nfacial avatar model along with the head pose as a learnable parameter in an\nend-to-end framework. This enables not only controllable facial animation via\nvideo inputs, but also high-fidelity novel view synthesis of challenging facial\nexpressions, such as tongue deformations and fine-grained teeth structure under\nlarge motion changes. Moreover, it encourages the learned head avatar to\ngeneralize towards new facial expressions and head poses at inference time. We\ndemonstrate the performance of our method with comparisons against the related\nmethods on different datasets, spanning challenging facial expression sequences\nacross multiple identities. We also show the potential application of our\napproach by demonstrating a cross-identity facial performance transfer\napplication.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.09740", "title": "VGG-Tex: A Vivid Geometry-Guided Facial Texture Estimation Model for\n  High Fidelity Monocular 3D Face Reconstruction", "abstract": "  3D face reconstruction from monocular images has promoted the development of\nvarious applications such as augmented reality. Though existing methods have\nmade remarkable progress, most of them emphasize geometric reconstruction,\nwhile overlooking the importance of texture prediction. To address this issue,\nwe propose VGG-Tex, a novel Vivid Geometry-Guided Facial Texture Estimation\nmodel designed for High Fidelity Monocular 3D Face Reconstruction. The core of\nthis approach is leveraging 3D parametric priors to enhance the outcomes of 2D\nUV texture estimation. Specifically, VGG-Tex includes a Facial Attributes\nEncoding Module, a Geometry-Guided Texture Generator, and a Visibility-Enhanced\nTexture Completion Module. These components are responsible for extracting\nparametric priors, generating initial textures, and refining texture details,\nrespectively. Based on the geometry-texture complementarity principle, VGG-Tex\nalso introduces a Texture-guided Geometry Refinement Module to further balance\nthe overall fidelity of the reconstructed 3D faces, along with corresponding\nlosses. Comprehensive experiments demonstrate that our method significantly\nimproves texture reconstruction performance compared to existing\nstate-of-the-art methods.\n", "date": "2024-09-15", "categories": []}
{"id": "2405.14869", "title": "PuzzleAvatar: Assembling 3D Avatars from Personal Albums", "abstract": "  Generating personalized 3D avatars is crucial for AR/VR. However, recent\ntext-to-3D methods that generate avatars for celebrities or fictional\ncharacters, struggle with everyday people. Methods for faithful reconstruction\ntypically require full-body images in controlled settings. What if a user could\njust upload their personal \"OOTD\" (Outfit Of The Day) photo collection and get\na faithful avatar in return? The challenge is that such casual photo\ncollections contain diverse poses, challenging viewpoints, cropped views, and\nocclusion (albeit with a consistent outfit, accessories and hairstyle). We\naddress this novel \"Album2Human\" task by developing PuzzleAvatar, a novel model\nthat generates a faithful 3D avatar (in a canonical pose) from a personal OOTD\nalbum, while bypassing the challenging estimation of body and camera pose. To\nthis end, we fine-tune a foundational vision-language model (VLM) on such\nphotos, encoding the appearance, identity, garments, hairstyles, and\naccessories of a person into (separate) learned tokens and instilling these\ncues into the VLM. In effect, we exploit the learned tokens as \"puzzle pieces\"\nfrom which we assemble a faithful, personalized 3D avatar. Importantly, we can\ncustomize avatars by simply inter-changing tokens. As a benchmark for this new\ntask, we collect a new dataset, called PuzzleIOI, with 41 subjects in a total\nof nearly 1K OOTD configurations, in challenging partial photos with paired\nground-truth 3D bodies. Evaluation shows that PuzzleAvatar not only has high\nreconstruction accuracy, outperforming TeCH and MVDreamBooth, but also a unique\nscalability to album photos, and strong robustness. Our code and data are\npublicly available for research purpose at https://puzzleavatar.is.tue.mpg.de/\n", "date": "2024-05-23", "categories": []}
{"id": "2405.19331", "title": "NPGA: Neural Parametric Gaussian Avatars", "abstract": "  The creation of high-fidelity, digital versions of human heads is an\nimportant stepping stone in the process of further integrating virtual\ncomponents into our everyday lives. Constructing such avatars is a challenging\nresearch problem, due to a high demand for photo-realism and real-time\nrendering performance. In this work, we propose Neural Parametric Gaussian\nAvatars (NPGA), a data-driven approach to create high-fidelity, controllable\navatars from multi-view video recordings. We build our method around 3D\nGaussian splatting for its highly efficient rendering and to inherit the\ntopological flexibility of point clouds. In contrast to previous work, we\ncondition our avatars' dynamics on the rich expression space of neural\nparametric head models (NPHM), instead of mesh-based 3DMMs. To this end, we\ndistill the backward deformation field of our underlying NPHM into forward\ndeformations which are compatible with rasterization-based rendering. All\nremaining fine-scale, expression-dependent details are learned from the\nmulti-view videos. For increased representational capacity of our avatars, we\npropose per-Gaussian latent features that condition each primitives dynamic\nbehavior. To regularize this increased dynamic expressivity, we propose\nLaplacian terms on the latent features and predicted dynamics. We evaluate our\nmethod on the public NeRSemble dataset, demonstrating that NPGA significantly\noutperforms the previous state-of-the-art avatars on the self-reenactment task\nby 2.6 PSNR. Furthermore, we demonstrate accurate animation capabilities from\nreal-world monocular videos.\n", "date": "2024-05-29", "categories": []}
{"id": "2409.07984", "title": "SPARK: Self-supervised Personalized Real-time Monocular Face Capture", "abstract": "  Feedforward monocular face capture methods seek to reconstruct posed faces\nfrom a single image of a person. Current state of the art approaches have the\nability to regress parametric 3D face models in real-time across a wide range\nof identities, lighting conditions and poses by leveraging large image datasets\nof human faces. These methods however suffer from clear limitations in that the\nunderlying parametric face model only provides a coarse estimation of the face\nshape, thereby limiting their practical applicability in tasks that require\nprecise 3D reconstruction (aging, face swapping, digital make-up, ...). In this\npaper, we propose a method for high-precision 3D face capture taking advantage\nof a collection of unconstrained videos of a subject as prior information. Our\nproposal builds on a two stage approach. We start with the reconstruction of a\ndetailed 3D face avatar of the person, capturing both precise geometry and\nappearance from a collection of videos. We then use the encoder from a\npre-trained monocular face reconstruction method, substituting its decoder with\nour personalized model, and proceed with transfer learning on the video\ncollection. Using our pre-estimated image formation model, we obtain a more\nprecise self-supervision objective, enabling improved expression and pose\nalignment. This results in a trained encoder capable of efficiently regressing\npose and expression parameters in real-time from previously unseen images,\nwhich combined with our personalized geometry model yields more accurate and\nhigh fidelity mesh inference. Through extensive qualitative and quantitative\nevaluation, we showcase the superiority of our final model as compared to\nstate-of-the-art baselines, and demonstrate its generalization ability to\nunseen pose, expression and lighting.\n", "date": "2024-09-12", "categories": []}
{"id": "2311.17460", "title": "W-HMR: Monocular Human Mesh Recovery in World Space with Weak-Supervised\n  Calibration", "abstract": "  Previous methods for 3D human motion recovery from monocular images often\nfall short due to reliance on camera coordinates, leading to inaccuracies in\nreal-world applications. The limited availability and diversity of focal length\nlabels further exacerbate misalignment issues in reconstructed 3D human bodies.\nTo address these challenges, we introduce W-HMR, a weak-supervised calibration\nmethod that predicts \"reasonable\" focal lengths based on body distortion\ninformation, eliminating the need for precise focal length labels. Our approach\nenhances 2D supervision precision and recovery accuracy. Additionally, we\npresent the OrientCorrect module, which corrects body orientation for plausible\nreconstructions in world space, avoiding the error accumulation associated with\ninaccurate camera rotation predictions. Our contributions include a novel\nweak-supervised camera calibration technique, an effective orientation\ncorrection module, and a decoupling strategy that significantly improves the\ngeneralizability and accuracy of human motion recovery in both camera and world\ncoordinates. The robustness of W-HMR is validated through extensive experiments\non various datasets, showcasing its superiority over existing methods. Codes\nand demos have been made available on the project page\nhttps://yw0208.github.io/w-hmr/.\n", "date": "2023-11-29", "categories": []}
{"id": "2403.17346", "title": "TRAM: Global Trajectory and Motion of 3D Humans from in-the-wild Videos", "abstract": "  We propose TRAM, a two-stage method to reconstruct a human's global\ntrajectory and motion from in-the-wild videos. TRAM robustifies SLAM to recover\nthe camera motion in the presence of dynamic humans and uses the scene\nbackground to derive the motion scale. Using the recovered camera as a\nmetric-scale reference frame, we introduce a video transformer model (VIMO) to\nregress the kinematic body motion of a human. By composing the two motions, we\nachieve accurate recovery of 3D humans in the world space, reducing global\nmotion errors by a large margin from prior work.\nhttps://yufu-wang.github.io/tram4d/\n", "date": "2024-03-26", "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": "2403.05055", "title": "MUC: Mixture of Uncalibrated Cameras for Robust 3D Human Body\n  Reconstruction", "abstract": "  Multiple cameras can provide comprehensive multi-view video coverage of a\nperson. Fusing this multi-view data is crucial for tasks like behavioral\nanalysis, although it traditionally requires camera calibration, a process that\nis often complex. Moreover, previous studies have overlooked the challenges\nposed by self-occlusion under multiple views and the continuity of human body\nshape estimation. In this study, we introduce a method to reconstruct the 3D\nhuman body from multiple uncalibrated camera views. Initially, we utilize a\npre-trained human body encoder to process each camera view individually,\nenabling the reconstruction of human body models and parameters for each view\nalong with predicted camera positions. Rather than merely averaging the models\nacross views, we develop a neural network trained to assign weights to\nindividual views for all human body joints, based on the estimated distribution\nof joint distances from each camera. Additionally, we focus on the mesh surface\nof the human body for dynamic fusion, allowing for the seamless integration of\nfacial expressions and body shape into a unified human body model. Our method\nhas shown excellent performance in reconstructing the human body on two public\ndatasets, advancing beyond previous work from the SMPL model to the SMPL-X\nmodel. This extension incorporates more complex hand poses and facial\nexpressions, enhancing the detail and accuracy of the reconstructions.\nCrucially, it supports the flexible ad-hoc deployment of any number of cameras,\noffering significant potential for various applications. Our code is available\nat https://github.com/AbsterZhu/MUC.\n", "date": "2024-03-08", "categories": []}
{"id": "2408.14211", "title": "MagicMan: Generative Novel View Synthesis of Humans with 3D-Aware\n  Diffusion and Iterative Refinement", "abstract": "  Existing works in single-image human reconstruction suffer from weak\ngeneralizability due to insufficient training data or 3D inconsistencies for a\nlack of comprehensive multi-view knowledge. In this paper, we introduce\nMagicMan, a human-specific multi-view diffusion model designed to generate\nhigh-quality novel view images from a single reference image. As its core, we\nleverage a pre-trained 2D diffusion model as the generative prior for\ngeneralizability, with the parametric SMPL-X model as the 3D body prior to\npromote 3D awareness. To tackle the critical challenge of maintaining\nconsistency while achieving dense multi-view generation for improved 3D human\nreconstruction, we first introduce hybrid multi-view attention to facilitate\nboth efficient and thorough information interchange across different views.\nAdditionally, we present a geometry-aware dual branch to perform concurrent\ngeneration in both RGB and normal domains, further enhancing consistency via\ngeometry cues. Last but not least, to address ill-shaped issues arising from\ninaccurate SMPL-X estimation that conflicts with the reference image, we\npropose a novel iterative refinement strategy, which progressively optimizes\nSMPL-X accuracy while enhancing the quality and consistency of the generated\nmulti-views. Extensive experimental results demonstrate that our method\nsignificantly outperforms existing approaches in both novel view synthesis and\nsubsequent 3D human reconstruction tasks.\n", "date": "2024-08-26", "categories": []}
{"id": "2310.18279", "title": "FOUND: Foot Optimization with Uncertain Normals for Surface Deformation\n  Using Synthetic Data", "abstract": "  Surface reconstruction from multi-view images is a challenging task, with\nsolutions often requiring a large number of sampled images with high overlap.\nWe seek to develop a method for few-view reconstruction, for the case of the\nhuman foot. To solve this task, we must extract rich geometric cues from RGB\nimages, before carefully fusing them into a final 3D object. Our FOUND approach\ntackles this, with 4 main contributions: (i) SynFoot, a synthetic dataset of\n50,000 photorealistic foot images, paired with ground truth surface normals and\nkeypoints; (ii) an uncertainty-aware surface normal predictor trained on our\nsynthetic dataset; (iii) an optimization scheme for fitting a generative foot\nmodel to a series of images; and (iv) a benchmark dataset of calibrated images\nand high resolution ground truth geometry. We show that our normal predictor\noutperforms all off-the-shelf equivalents significantly on real images, and our\noptimization scheme outperforms state-of-the-art photogrammetry pipelines,\nespecially for a few-view setting. We release our synthetic dataset and\nbaseline 3D scans to the research community.\n", "date": "2023-10-27", "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": "2401.08398", "title": "High-Quality Mesh Blendshape Generation from Face Videos via Neural\n  Inverse Rendering", "abstract": "  Readily editable mesh blendshapes have been widely used in animation\npipelines, while recent advancements in neural geometry and appearance\nrepresentations have enabled high-quality inverse rendering. Building upon\nthese observations, we introduce a novel technique that reconstructs mesh-based\nblendshape rigs from single or sparse multi-view videos, leveraging\nstate-of-the-art neural inverse rendering. We begin by constructing a\ndeformation representation that parameterizes vertex displacements into\ndifferential coordinates with tetrahedral connections, allowing for\nhigh-quality vertex deformation on high-resolution meshes. By constructing a\nset of semantic regulations in this representation, we achieve joint\noptimization of blendshapes and expression coefficients. Furthermore, to enable\na user-friendly multi-view setup with unsynchronized cameras, we propose a\nneural regressor to model time-varying motion parameters. This approach\nimplicitly considers the time difference across multiple cameras, enhancing the\naccuracy of motion modeling. Experiments demonstrate that, with the flexible\ninput of single or sparse multi-view videos, we reconstruct personalized\nhigh-fidelity blendshapes. These blendshapes are both geometrically and\nsemantically accurate, and they are compatible with industrial animation\npipelines. Code and data are available at\nhttps://github.com/grignarder/high-quality-blendshape-generation.\n", "date": "2024-01-16", "categories": []}
{"id": "2402.00827", "title": "GaussianStyle: Gaussian Head Avatar via StyleGAN", "abstract": "  Existing methods like Neural Radiation Fields (NeRF) and 3D Gaussian\nSplatting (3DGS) have made significant strides in facial attribute control such\nas facial animation and components editing, yet they struggle with fine-grained\nrepresentation and scalability in dynamic head modeling. To address these\nlimitations, we propose GaussianStyle, a novel framework that integrates the\nvolumetric strengths of 3DGS with the powerful implicit representation of\nStyleGAN. The GaussianStyle preserves structural information, such as\nexpressions and poses, using Gaussian points, while projecting the implicit\nvolumetric representation into StyleGAN to capture high-frequency details and\nmitigate the over-smoothing commonly observed in neural texture rendering.\nExperimental outcomes indicate that our method achieves state-of-the-art\nperformance in reenactment, novel view synthesis, and animation.\n", "date": "2024-02-01", "categories": []}
{"id": "2408.08086", "title": "Single-image coherent reconstruction of objects and humans", "abstract": "  Existing methods for reconstructing objects and humans from a monocular image\nsuffer from severe mesh collisions and performance limitations for interacting\noccluding objects. This paper introduces a method to obtain a globally\nconsistent 3D reconstruction of interacting objects and people from a single\nimage. Our contributions include: 1) an optimization framework, featuring a\ncollision loss, tailored to handle human-object and human-human interactions,\nensuring spatially coherent scene reconstruction; and 2) a novel technique to\nrobustly estimate 6 degrees of freedom (DOF) poses, specifically for heavily\noccluded objects, exploiting image inpainting. Notably, our proposed method\noperates effectively on images from real-world scenarios, without necessitating\nscene or object-level 3D supervision. Extensive qualitative and quantitative\nevaluation against existing methods demonstrates a significant reduction in\ncollisions in the final reconstructions of scenes with multiple interacting\nhumans and objects and a more coherent scene reconstruction.\n", "date": "2024-08-15", "categories": []}
{"id": "2312.05826", "title": "R2Human: Real-Time 3D Human Appearance Rendering from a Single Image", "abstract": "  Rendering 3D human appearance from a single image in real-time is crucial for\nachieving holographic communication and immersive VR/AR. Existing methods\neither rely on multi-camera setups or are constrained to offline operations. In\nthis paper, we propose R2Human, the first approach for real-time inference and\nrendering of photorealistic 3D human appearance from a single image. The core\nof our approach is to combine the strengths of implicit texture fields and\nexplicit neural rendering with our novel representation, namely Z-map. Based on\nthis, we present an end-to-end network that performs high-fidelity color\nreconstruction of visible areas and provides reliable color inference for\noccluded regions. To further enhance the 3D perception ability of our network,\nwe leverage the Fourier occupancy field as a prior for generating the texture\nfield and providing a sampling surface in the rendering stage. We also propose\na consistency loss and a spatial fusion strategy to ensure the multi-view\ncoherence. Experimental results show that our method outperforms the\nstate-of-the-art methods on both synthetic data and challenging real-world\nimages, in real-time. The project page can be found at\nhttp://cic.tju.edu.cn/faculty/likun/projects/R2Human.\n", "date": "2023-12-10", "categories": []}
{"id": "2312.02902", "title": "HeadGaS: Real-Time Animatable Head Avatars via 3D Gaussian Splatting", "abstract": "  3D head animation has seen major quality and runtime improvements over the\nlast few years, particularly empowered by the advances in differentiable\nrendering and neural radiance fields. Real-time rendering is a highly desirable\ngoal for real-world applications. We propose HeadGaS, a model that uses 3D\nGaussian Splats (3DGS) for 3D head reconstruction and animation. In this paper\nwe introduce a hybrid model that extends the explicit 3DGS representation with\na base of learnable latent features, which can be linearly blended with\nlow-dimensional parameters from parametric head models to obtain\nexpression-dependent color and opacity values. We demonstrate that HeadGaS\ndelivers state-of-the-art results in real-time inference frame rates,\nsurpassing baselines by up to 2dB, while accelerating rendering speed by over\nx10.\n", "date": "2023-12-05", "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": "2408.01218", "title": "S2TD-Face: Reconstruct a Detailed 3D Face with Controllable Texture from\n  a Single Sketch", "abstract": "  3D textured face reconstruction from sketches applicable in many scenarios\nsuch as animation, 3D avatars, artistic design, missing people search, etc., is\na highly promising but underdeveloped research topic. On the one hand, the\nstylistic diversity of sketches leads to existing sketch-to-3D-face methods\nonly being able to handle pose-limited and realistically shaded sketches. On\nthe other hand, texture plays a vital role in representing facial appearance,\nyet sketches lack this information, necessitating additional texture control in\nthe reconstruction process. This paper proposes a novel method for\nreconstructing controllable textured and detailed 3D faces from sketches, named\nS2TD-Face. S2TD-Face introduces a two-stage geometry reconstruction framework\nthat directly reconstructs detailed geometry from the input sketch. To keep\ngeometry consistent with the delicate strokes of the sketch, we propose a novel\nsketch-to-geometry loss that ensures the reconstruction accurately fits the\ninput features like dimples and wrinkles. Our training strategies do not rely\non hard-to-obtain 3D face scanning data or labor-intensive hand-drawn sketches.\nFurthermore, S2TD-Face introduces a texture control module utilizing text\nprompts to select the most suitable textures from a library and seamlessly\nintegrate them into the geometry, resulting in a 3D detailed face with\ncontrollable texture. S2TD-Face surpasses existing state-of-the-art methods in\nextensive quantitative and qualitative experiments. Our project is available at\nhttps://github.com/wang-zidu/S2TD-Face .\n", "date": "2024-08-02", "categories": []}
{"id": "2408.00296", "title": "Head360: Learning a Parametric 3D Full-Head for Free-View Synthesis in\n  360{\\deg}", "abstract": "  Creating a 360{\\deg} parametric model of a human head is a very challenging\ntask. While recent advancements have demonstrated the efficacy of leveraging\nsynthetic data for building such parametric head models, their performance\nremains inadequate in crucial areas such as expression-driven animation,\nhairstyle editing, and text-based modifications. In this paper, we build a\ndataset of artist-designed high-fidelity human heads and propose to create a\nnovel parametric 360{\\deg} renderable parametric head model from it. Our scheme\ndecouples the facial motion/shape and facial appearance, which are represented\nby a classic parametric 3D mesh model and an attached neural texture,\nrespectively. We further propose a training method for decompositing hairstyle\nand facial appearance, allowing free-swapping of the hairstyle. A novel\ninversion fitting method is presented based on single image input with high\ngeneralization and fidelity. To the best of our knowledge, our model is the\nfirst parametric 3D full-head that achieves 360{\\deg} free-view synthesis,\nimage-based fitting, appearance editing, and animation within a single model.\nExperiments show that facial motions and appearances are well disentangled in\nthe parametric space, leading to SOTA performance in rendering and animating\nquality. The code and SynHead100 dataset are released at\nhttps://nju-3dv.github.io/projects/Head360.\n", "date": "2024-08-01", "categories": []}
{"id": "2407.20566", "title": "Monocular Human-Object Reconstruction in the Wild", "abstract": "  Learning the prior knowledge of the 3D human-object spatial relation is\ncrucial for reconstructing human-object interaction from images and\nunderstanding how humans interact with objects in 3D space. Previous works\nlearn this prior from datasets collected in controlled environments, but due to\nthe diversity of domains, they struggle to generalize to real-world scenarios.\nTo overcome this limitation, we present a 2D-supervised method that learns the\n3D human-object spatial relation prior purely from 2D images in the wild. Our\nmethod utilizes a flow-based neural network to learn the prior distribution of\nthe 2D human-object keypoint layout and viewports for each image in the\ndataset. The effectiveness of the prior learned from 2D images is demonstrated\non the human-object reconstruction task by applying the prior to tune the\nrelative pose between the human and the object during the post-optimization\nstage. To validate and benchmark our method on in-the-wild images, we collect\nthe WildHOI dataset from the YouTube website, which consists of various\ninteractions with 8 objects in real-world scenarios. We conduct the experiments\non the indoor BEHAVE dataset and the outdoor WildHOI dataset. The results show\nthat our method achieves almost comparable performance with fully 3D supervised\nmethods on the BEHAVE dataset, even if we have only utilized the 2D layout\ninformation, and outperforms previous methods in terms of generality and\ninteraction diversity on in-the-wild images.\n", "date": "2024-07-30", "categories": []}
{"id": "2407.21047", "title": "PAV: Personalized Head Avatar from Unstructured Video Collection", "abstract": "  We propose PAV, Personalized Head Avatar for the synthesis of human faces\nunder arbitrary viewpoints and facial expressions. PAV introduces a method that\nlearns a dynamic deformable neural radiance field (NeRF), in particular from a\ncollection of monocular talking face videos of the same character under various\nappearance and shape changes. Unlike existing head NeRF methods that are\nlimited to modeling such input videos on a per-appearance basis, our method\nallows for learning multi-appearance NeRFs, introducing appearance embedding\nfor each input video via learnable latent neural features attached to the\nunderlying geometry. Furthermore, the proposed appearance-conditioned density\nformulation facilitates the shape variation of the character, such as facial\nhair and soft tissues, in the radiance field prediction. To the best of our\nknowledge, our approach is the first dynamic deformable NeRF framework to model\nappearance and shape variations in a single unified network for\nmulti-appearances of the same subject. We demonstrate experimentally that PAV\noutperforms the baseline method in terms of visual rendering quality in our\nquantitative and qualitative studies on various subjects.\n", "date": "2024-07-22", "categories": []}
{"id": "2407.21686", "title": "Expressive Whole-Body 3D Gaussian Avatar", "abstract": "  Facial expression and hand motions are necessary to express our emotions and\ninteract with the world. Nevertheless, most of the 3D human avatars modeled\nfrom a casually captured video only support body motions without facial\nexpressions and hand motions.In this work, we present ExAvatar, an expressive\nwhole-body 3D human avatar learned from a short monocular video. We design\nExAvatar as a combination of the whole-body parametric mesh model (SMPL-X) and\n3D Gaussian Splatting (3DGS). The main challenges are 1) a limited diversity of\nfacial expressions and poses in the video and 2) the absence of 3D\nobservations, such as 3D scans and RGBD images. The limited diversity in the\nvideo makes animations with novel facial expressions and poses non-trivial. In\naddition, the absence of 3D observations could cause significant ambiguity in\nhuman parts that are not observed in the video, which can result in noticeable\nartifacts under novel motions. To address them, we introduce our hybrid\nrepresentation of the mesh and 3D Gaussians. Our hybrid representation treats\neach 3D Gaussian as a vertex on the surface with pre-defined connectivity\ninformation (i.e., triangle faces) between them following the mesh topology of\nSMPL-X. It makes our ExAvatar animatable with novel facial expressions by\ndriven by the facial expression space of SMPL-X. In addition, by using\nconnectivity-based regularizers, we significantly reduce artifacts in novel\nfacial expressions and poses.\n", "date": "2024-07-31", "categories": []}
{"id": "2407.19593", "title": "Bridging the Gap: Studio-like Avatar Creation from a Monocular Phone\n  Capture", "abstract": "  Creating photorealistic avatars for individuals traditionally involves\nextensive capture sessions with complex and expensive studio devices like the\nLightStage system. While recent strides in neural representations have enabled\nthe generation of photorealistic and animatable 3D avatars from quick phone\nscans, they have the capture-time lighting baked-in, lack facial details and\nhave missing regions in areas such as the back of the ears. Thus, they lag in\nquality compared to studio-captured avatars. In this paper, we propose a method\nthat bridges this gap by generating studio-like illuminated texture maps from\nshort, monocular phone captures. We do this by parameterizing the phone texture\nmaps using the $W^+$ space of a StyleGAN2, enabling near-perfect\nreconstruction. Then, we finetune a StyleGAN2 by sampling in the $W^+$\nparameterized space using a very small set of studio-captured textures as an\nadversarial training signal. To further enhance the realism and accuracy of\nfacial details, we super-resolve the output of the StyleGAN2 using carefully\ndesigned diffusion model that is guided by image gradients of the\nphone-captured texture map. Once trained, our method excels at producing\nstudio-like facial texture maps from casual monocular smartphone videos.\nDemonstrating its capabilities, we showcase the generation of photorealistic,\nuniformly lit, complete avatars from monocular phone captures. The project page\ncan be found at http://shahrukhathar.github.io/2024/07/22/Bridging.html\n", "date": "2024-07-28", "categories": []}
{"id": "2309.17448", "title": "SMPLer-X: Scaling Up Expressive Human Pose and Shape Estimation", "abstract": "  Expressive human pose and shape estimation (EHPS) unifies body, hands, and\nface motion capture with numerous applications. Despite encouraging progress,\ncurrent state-of-the-art methods still depend largely on a confined set of\ntraining datasets. In this work, we investigate scaling up EHPS towards the\nfirst generalist foundation model (dubbed SMPLer-X), with up to ViT-Huge as the\nbackbone and training with up to 4.5M instances from diverse data sources. With\nbig data and the large model, SMPLer-X exhibits strong performance across\ndiverse test benchmarks and excellent transferability to even unseen\nenvironments. 1) For the data scaling, we perform a systematic investigation on\n32 EHPS datasets, including a wide range of scenarios that a model trained on\nany single dataset cannot handle. More importantly, capitalizing on insights\nobtained from the extensive benchmarking process, we optimize our training\nscheme and select datasets that lead to a significant leap in EHPS\ncapabilities. 2) For the model scaling, we take advantage of vision\ntransformers to study the scaling law of model sizes in EHPS. Moreover, our\nfinetuning strategy turn SMPLer-X into specialist models, allowing them to\nachieve further performance boosts. Notably, our foundation model SMPLer-X\nconsistently delivers state-of-the-art results on seven benchmarks such as\nAGORA (107.2 mm NMVE), UBody (57.4 mm PVE), EgoBody (63.6 mm PVE), and EHF\n(62.3 mm PVE without finetuning). Homepage:\nhttps://caizhongang.github.io/projects/SMPLer-X/\n", "date": "2023-09-29", "categories": []}
{"id": "2311.16482", "title": "Animatable 3D Gaussian: Fast and High-Quality Reconstruction of Multiple\n  Human Avatars", "abstract": "  Neural radiance fields are capable of reconstructing high-quality drivable\nhuman avatars but are expensive to train and render and not suitable for\nmulti-human scenes with complex shadows. To reduce consumption, we propose\nAnimatable 3D Gaussian, which learns human avatars from input images and poses.\nWe extend 3D Gaussians to dynamic human scenes by modeling a set of skinned 3D\nGaussians and a corresponding skeleton in canonical space and deforming 3D\nGaussians to posed space according to the input poses. We introduce a\nmulti-head hash encoder for pose-dependent shape and appearance and a\ntime-dependent ambient occlusion module to achieve high-quality reconstructions\nin scenes containing complex motions and dynamic shadows. On both novel view\nsynthesis and novel pose synthesis tasks, our method achieves higher\nreconstruction quality than InstantAvatar with less training time (1/60), less\nGPU memory (1/4), and faster rendering speed (7x). Our method can be easily\nextended to multi-human scenes and achieve comparable novel view synthesis\nresults on a scene with ten people in only 25 seconds of training.\n", "date": "2023-11-27", "categories": []}
{"id": "2312.11587", "title": "Relightable Neural Actor with Intrinsic Decomposition and Pose Control", "abstract": "  Creating a controllable and relightable digital avatar from multi-view video\nwith fixed illumination is a very challenging problem since humans are highly\narticulated, creating pose-dependent appearance effects, and skin as well as\nclothing require space-varying BRDF modeling. Existing works on creating\nanimatible avatars either do not focus on relighting at all, require controlled\nillumination setups, or try to recover a relightable avatar from very low cost\nsetups, i.e. a single RGB video, at the cost of severely limited result\nquality, e.g. shadows not even being modeled. To address this, we propose\nRelightable Neural Actor, a new video-based method for learning a pose-driven\nneural human model that can be relighted, allows appearance editing, and models\npose-dependent effects such as wrinkles and self-shadows. Importantly, for\ntraining, our method solely requires a multi-view recording of the human under\na known, but static lighting condition. To tackle this challenging problem, we\nleverage an implicit geometry representation of the actor with a drivable\ndensity field that models pose-dependent deformations and derive a dynamic\nmapping between 3D and UV spaces, where normal, visibility, and materials are\neffectively encoded. To evaluate our approach in real-world scenarios, we\ncollect a new dataset with four identities recorded under different light\nconditions, indoors and outdoors, providing the first benchmark of its kind for\nhuman relighting, and demonstrating state-of-the-art relighting results for\nnovel human poses.\n", "date": "2023-12-18", "categories": []}
{"id": "2407.18034", "title": "AttentionHand: Text-driven Controllable Hand Image Generation for 3D\n  Hand Reconstruction in the Wild", "abstract": "  Recently, there has been a significant amount of research conducted on 3D\nhand reconstruction to use various forms of human-computer interaction.\nHowever, 3D hand reconstruction in the wild is challenging due to extreme lack\nof in-the-wild 3D hand datasets. Especially, when hands are in complex pose\nsuch as interacting hands, the problems like appearance similarity, self-handed\noccclusion and depth ambiguity make it more difficult. To overcome these\nissues, we propose AttentionHand, a novel method for text-driven controllable\nhand image generation. Since AttentionHand can generate various and numerous\nin-the-wild hand images well-aligned with 3D hand label, we can acquire a new\n3D hand dataset, and can relieve the domain gap between indoor and outdoor\nscenes. Our method needs easy-to-use four modalities (i.e, an RGB image, a hand\nmesh image from 3D label, a bounding box, and a text prompt). These modalities\nare embedded into the latent space by the encoding phase. Then, through the\ntext attention stage, hand-related tokens from the given text prompt are\nattended to highlight hand-related regions of the latent embedding. After the\nhighlighted embedding is fed to the visual attention stage, hand-related\nregions in the embedding are attended by conditioning global and local hand\nmesh images with the diffusion-based pipeline. In the decoding phase, the final\nfeature is decoded to new hand images, which are well-aligned with the given\nhand mesh image and text prompt. As a result, AttentionHand achieved\nstate-of-the-art among text-to-hand image generation models, and the\nperformance of 3D hand mesh reconstruction was improved by additionally\ntraining with hand images generated by AttentionHand.\n", "date": "2024-07-25", "categories": []}
{"id": "2402.14654", "title": "Multi-HMR: Multi-Person Whole-Body Human Mesh Recovery in a Single Shot", "abstract": "  We present Multi-HMR, a strong sigle-shot model for multi-person 3D human\nmesh recovery from a single RGB image. Predictions encompass the whole body,\ni.e., including hands and facial expressions, using the SMPL-X parametric model\nand 3D location in the camera coordinate system. Our model detects people by\npredicting coarse 2D heatmaps of person locations, using features produced by a\nstandard Vision Transformer (ViT) backbone. It then predicts their whole-body\npose, shape and 3D location using a new cross-attention module called the Human\nPrediction Head (HPH), with one query attending to the entire set of features\nfor each detected person. As direct prediction of fine-grained hands and facial\nposes in a single shot, i.e., without relying on explicit crops around body\nparts, is hard to learn from existing data, we introduce CUFFS, the Close-Up\nFrames of Full-Body Subjects dataset, containing humans close to the camera\nwith diverse hand poses. We show that incorporating it into the training data\nfurther enhances predictions, particularly for hands. Multi-HMR also optionally\naccounts for camera intrinsics, if available, by encoding camera ray directions\nfor each image token. This simple design achieves strong performance on\nwhole-body and body-only benchmarks simultaneously: a ViT-S backbone on\n$448{\\times}448$ images already yields a fast and competitive model, while\nlarger models and higher resolutions obtain state-of-the-art results.\n", "date": "2024-02-22", "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": "2407.07284", "title": "MIGS: Multi-Identity Gaussian Splatting via Tensor Decomposition", "abstract": "  We introduce MIGS (Multi-Identity Gaussian Splatting), a novel method that\nlearns a single neural representation for multiple identities, using only\nmonocular videos. Recent 3D Gaussian Splatting (3DGS) approaches for human\navatars require per-identity optimization. However, learning a multi-identity\nrepresentation presents advantages in robustly animating humans under arbitrary\nposes. We propose to construct a high-order tensor that combines all the\nlearnable 3DGS parameters for all the training identities. By assuming a\nlow-rank structure and factorizing the tensor, we model the complex rigid and\nnon-rigid deformations of multiple subjects in a unified network, significantly\nreducing the total number of parameters. Our proposed approach leverages\ninformation from all the training identities and enables robust animation under\nchallenging unseen poses, outperforming existing approaches. It can also be\nextended to learn unseen identities.\n", "date": "2024-07-10", "categories": []}
{"id": "2407.12307", "title": "Weakly-Supervised 3D Hand Reconstruction with Knowledge Prior and\n  Uncertainty Guidance", "abstract": "  Fully-supervised monocular 3D hand reconstruction is often difficult because\ncapturing the requisite 3D data entails deploying specialized equipment in a\ncontrolled environment. We introduce a weakly-supervised method that avoids\nsuch requirements by leveraging fundamental principles well-established in the\nunderstanding of the human hand's unique structure and functionality.\nSpecifically, we systematically study hand knowledge from different sources,\nincluding biomechanics, functional anatomy, and physics. We effectively\nincorporate these valuable foundational insights into 3D hand reconstruction\nmodels through an appropriate set of differentiable training losses. This\nenables training solely with readily-obtainable 2D hand landmark annotations\nand eliminates the need for expensive 3D supervision. Moreover, we explicitly\nmodel the uncertainty that is inherent in image observations. We enhance the\ntraining process by exploiting a simple yet effective Negative Log Likelihood\n(NLL) loss that incorporates uncertainty into the loss function. Through\nextensive experiments, we demonstrate that our method significantly outperforms\nstate-of-the-art weakly-supervised methods. For example, our method achieves\nnearly a 21\\% performance improvement on the widely adopted FreiHAND dataset.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.12777", "title": "Generalizable Human Gaussians for Sparse View Synthesis", "abstract": "  Recent progress in neural rendering has brought forth pioneering methods,\nsuch as NeRF and Gaussian Splatting, which revolutionize view rendering across\nvarious domains like AR/VR, gaming, and content creation. While these methods\nexcel at interpolating {\\em within the training data}, the challenge of\ngeneralizing to new scenes and objects from very sparse views persists.\nSpecifically, modeling 3D humans from sparse views presents formidable hurdles\ndue to the inherent complexity of human geometry, resulting in inaccurate\nreconstructions of geometry and textures. To tackle this challenge, this paper\nleverages recent advancements in Gaussian Splatting and introduces a new method\nto learn generalizable human Gaussians that allows photorealistic and accurate\nview-rendering of a new human subject from a limited set of sparse views in a\nfeed-forward manner. A pivotal innovation of our approach involves\nreformulating the learning of 3D Gaussian parameters into a regression process\ndefined on the 2D UV space of a human template, which allows leveraging the\nstrong geometry prior and the advantages of 2D convolutions. In addition, a\nmulti-scaffold is proposed to effectively represent the offset details. Our\nmethod outperforms recent methods on both within-dataset generalization as well\nas cross-dataset generalization settings.\n", "date": "2024-07-17", "categories": []}
{"id": "2403.19160", "title": "Within the Dynamic Context: Inertia-aware 3D Human Modeling with Pose\n  Sequence", "abstract": "  Neural rendering techniques have significantly advanced 3D human body\nmodeling. However, previous approaches often overlook dynamics induced by\nfactors such as motion inertia, leading to challenges in scenarios like abrupt\nstops after rotation, where the pose remains static while the appearance\nchanges. This limitation arises from reliance on a single pose as conditional\ninput, resulting in ambiguity in mapping one pose to multiple appearances. In\nthis study, we elucidate that variations in human appearance depend not only on\nthe current frame's pose condition but also on past pose states. Therefore, we\nintroduce Dyco, a novel method utilizing the delta pose sequence representation\nfor non-rigid deformations and canonical space to effectively model temporal\nappearance variations. To prevent a decrease in the model's generalization\nability to novel poses, we further propose low-dimensional global context to\nreduce unnecessary inter-body part dependencies and a quantization operation to\nmitigate overfitting of the delta pose sequence by the model. To validate the\neffectiveness of our approach, we collected a novel dataset named I3D-Human,\nwith a focus on capturing temporal changes in clothing appearance under\napproximate poses. Through extensive experiments on both I3D-Human and existing\ndatasets, our approach demonstrates superior qualitative and quantitative\nperformance. In addition, our inertia-aware 3D human method can unprecedentedly\nsimulate appearance changes caused by inertia at different velocities.\n", "date": "2024-03-28", "categories": []}
{"id": "2407.11174", "title": "iHuman: Instant Animatable Digital Humans From Monocular Videos", "abstract": "  Personalized 3D avatars require an animatable representation of digital\nhumans. Doing so instantly from monocular videos offers scalability to broad\nclass of users and wide-scale applications. In this paper, we present a fast,\nsimple, yet effective method for creating animatable 3D digital humans from\nmonocular videos. Our method utilizes the efficiency of Gaussian splatting to\nmodel both 3D geometry and appearance. However, we observed that naively\noptimizing Gaussian splats results in inaccurate geometry, thereby leading to\npoor animations. This work achieves and illustrates the need of accurate 3D\nmesh-type modelling of the human body for animatable digitization through\nGaussian splats. This is achieved by developing a novel pipeline that benefits\nfrom three key aspects: (a) implicit modelling of surface's displacements and\nthe color's spherical harmonics; (b) binding of 3D Gaussians to the respective\ntriangular faces of the body template; (c) a novel technique to render normals\nfollowed by their auxiliary supervision. Our exhaustive experiments on three\ndifferent benchmark datasets demonstrates the state-of-the-art results of our\nmethod, in limited time settings. In fact, our method is faster by an order of\nmagnitude (in terms of training time) than its closest competitor. At the same\ntime, we achieve superior rendering and 3D reconstruction performance under the\nchange of poses.\n", "date": "2024-07-15", "categories": []}
{"id": "2401.16810", "title": "An Embeddable Implicit IUVD Representation for Part-based 3D Human\n  Surface Reconstruction", "abstract": "  To reconstruct a 3D human surface from a single image, it is crucial to\nsimultaneously consider human pose, shape, and clothing details. Recent\napproaches have combined parametric body models (such as SMPL), which capture\nbody pose and shape priors, with neural implicit functions that flexibly learn\nclothing details. However, this combined representation introduces additional\ncomputation, e.g. signed distance calculation in 3D body feature extraction,\nleading to redundancy in the implicit query-and-infer process and failing to\npreserve the underlying body shape prior. To address these issues, we propose a\nnovel IUVD-Feedback representation, consisting of an IUVD occupancy function\nand a feedback query algorithm. This representation replaces the time-consuming\nsigned distance calculation with a simple linear transformation in the IUVD\nspace, leveraging the SMPL UV maps. Additionally, it reduces redundant query\npoints through a feedback mechanism, leading to more reasonable 3D body\nfeatures and more effective query points, thereby preserving the parametric\nbody prior. Moreover, the IUVD-Feedback representation can be embedded into any\nexisting implicit human reconstruction pipeline without requiring modifications\nto the trained neural networks. Experiments on the THuman2.0 dataset\ndemonstrate that the proposed IUVD-Feedback representation improves the\nrobustness of results and achieves three times faster acceleration in the\nquery-and-infer process. Furthermore, this representation holds potential for\ngenerative applications by leveraging its inherent semantic information from\nthe parametric body model.\n", "date": "2024-01-30", "categories": []}
{"id": "2406.00440", "title": "Topo4D: Topology-Preserving Gaussian Splatting for High-Fidelity 4D Head\n  Capture", "abstract": "  4D head capture aims to generate dynamic topological meshes and corresponding\ntexture maps from videos, which is widely utilized in movies and games for its\nability to simulate facial muscle movements and recover dynamic textures in\npore-squeezing. The industry often adopts the method involving multi-view\nstereo and non-rigid alignment. However, this approach is prone to errors and\nheavily reliant on time-consuming manual processing by artists. To simplify\nthis process, we propose Topo4D, a novel framework for automatic geometry and\ntexture generation, which optimizes densely aligned 4D heads and 8K texture\nmaps directly from calibrated multi-view time-series images. Specifically, we\nfirst represent the time-series faces as a set of dynamic 3D Gaussians with\nfixed topology in which the Gaussian centers are bound to the mesh vertices.\nAfterward, we perform alternative geometry and texture optimization\nframe-by-frame for high-quality geometry and texture learning while maintaining\ntemporal topology stability. Finally, we can extract dynamic facial meshes in\nregular wiring arrangement and high-fidelity textures with pore-level details\nfrom the learned Gaussians. Extensive experiments show that our method achieves\nsuperior results than the current SOTA face reconstruction methods both in the\nquality of meshes and textures. Project page:\nhttps://xuanchenli.github.io/Topo4D/.\n", "date": "2024-06-01", "categories": []}
{"id": "2407.04345", "title": "CanonicalFusion: Generating Drivable 3D Human Avatars from Multiple\n  Images", "abstract": "  We present a novel framework for reconstructing animatable human avatars from\nmultiple images, termed CanonicalFusion. Our central concept involves\nintegrating individual reconstruction results into the canonical space. To be\nspecific, we first predict Linear Blend Skinning (LBS) weight maps and depth\nmaps using a shared-encoder-dual-decoder network, enabling direct\ncanonicalization of the 3D mesh from the predicted depth maps. Here, instead of\npredicting high-dimensional skinning weights, we infer compressed skinning\nweights, i.e., 3-dimensional vector, with the aid of pre-trained MLP networks.\nWe also introduce a forward skinning-based differentiable rendering scheme to\nmerge the reconstructed results from multiple images. This scheme refines the\ninitial mesh by reposing the canonical mesh via the forward skinning and by\nminimizing photometric and geometric errors between the rendered and the\npredicted results. Our optimization scheme considers the position and color of\nvertices as well as the joint angles for each image, thereby mitigating the\nnegative effects of pose errors. We conduct extensive experiments to\ndemonstrate the effectiveness of our method and compare our CanonicalFusion\nwith state-of-the-art methods. Our source codes are available at\nhttps://github.com/jsshin98/CanonicalFusion.\n", "date": "2024-07-05", "categories": []}
{"id": "2407.09694", "title": "Divide and Fuse: Body Part Mesh Recovery from Partially Visible Human\n  Images", "abstract": "  We introduce a novel bottom-up approach for human body mesh reconstruction,\nspecifically designed to address the challenges posed by partial visibility and\nocclusion in input images. Traditional top-down methods, relying on whole-body\nparametric models like SMPL, falter when only a small part of the human is\nvisible, as they require visibility of most of the human body for accurate mesh\nreconstruction. To overcome this limitation, our method employs a \"Divide and\nFuse (D&F)\" strategy, reconstructing human body parts independently before\nfusing them, thereby ensuring robustness against occlusions. We design Human\nPart Parametric Models (HPPM) that independently reconstruct the mesh from a\nfew shape and global-location parameters, without inter-part dependency. A\nspecially designed fusion module then seamlessly integrates the reconstructed\nparts, even when only a few are visible. We harness a large volume of\nground-truth SMPL data to train our parametric mesh models. To facilitate the\ntraining and evaluation of our method, we have established benchmark datasets\nfeaturing images of partially visible humans with HPPM annotations. Our\nexperiments, conducted on these benchmark datasets, demonstrate the\neffectiveness of our D&F method, particularly in scenarios with substantial\ninvisibility, where traditional approaches struggle to maintain reconstruction\nquality.\n", "date": "2024-07-12", "categories": []}
{"id": "2407.10193", "title": "GRAPE: Generalizable and Robust Multi-view Facial Capture", "abstract": "  Deep learning-based multi-view facial capture methods have shown impressive\naccuracy while being several orders of magnitude faster than a traditional mesh\nregistration pipeline. However, the existing systems (e.g. TEMPEH) are strictly\nrestricted to inference on the data captured by the same camera array used to\ncapture their training data. In this study, we aim to improve the\ngeneralization ability so that a trained model can be readily used for\ninference (i.e. capture new data) on a different camera array. To this end, we\npropose a more generalizable initialization module to extract the camera\narray-agnostic 3D feature, including a visual hull-based head localization and\na visibility-aware 3D feature aggregation module enabled by the visual hull. In\naddition, we propose an ``update-by-disagreement'' learning strategy to better\nhandle data noise (e.g. inaccurate registration, scan noise) by discarding\npotentially inaccurate supervision signals during training. The resultant\ngeneralizable and robust topologically consistent multi-view facial capture\nsystem (GRAPE) can be readily used to capture data on a different camera array,\nreducing great effort on data collection and processing. Experiments on the\nFaMoS and FaceScape datasets demonstrate the effectiveness of the proposed\nmethod.\n", "date": "2024-07-14", "categories": []}
{"id": "2407.10586", "title": "COSMU: Complete 3D human shape from monocular unconstrained images", "abstract": "  We present a novel framework to reconstruct complete 3D human shapes from a\ngiven target image by leveraging monocular unconstrained images. The objective\nof this work is to reproduce high-quality details in regions of the\nreconstructed human body that are not visible in the input target. The proposed\nmethodology addresses the limitations of existing approaches for reconstructing\n3D human shapes from a single image, which cannot reproduce shape details in\noccluded body regions. The missing information of the monocular input can be\nrecovered by using multiple views captured from multiple cameras. However,\nmulti-view reconstruction methods necessitate accurately calibrated and\nregistered images, which can be challenging to obtain in real-world scenarios.\nGiven a target RGB image and a collection of multiple uncalibrated and\nunregistered images of the same individual, acquired using a single camera, we\npropose a novel framework to generate complete 3D human shapes. We introduce a\nnovel module to generate 2D multi-view normal maps of the person registered\nwith the target input image. The module consists of body part-based reference\nselection and body part-based registration. The generated 2D normal maps are\nthen processed by a multi-view attention-based neural implicit model that\nestimates an implicit representation of the 3D shape, ensuring the reproduction\nof details in both observed and occluded regions. Extensive experiments\ndemonstrate that the proposed approach estimates higher quality details in the\nnon-visible regions of the 3D clothed human shapes compared to related methods,\nwithout using parametric models.\n", "date": "2024-07-15", "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": "2312.05210", "title": "IntrinsicAvatar: Physically Based Inverse Rendering of Dynamic Humans\n  from Monocular Videos via Explicit Ray Tracing", "abstract": "  We present IntrinsicAvatar, a novel approach to recovering the intrinsic\nproperties of clothed human avatars including geometry, albedo, material, and\nenvironment lighting from only monocular videos. Recent advancements in\nhuman-based neural rendering have enabled high-quality geometry and appearance\nreconstruction of clothed humans from just monocular videos. However, these\nmethods bake intrinsic properties such as albedo, material, and environment\nlighting into a single entangled neural representation. On the other hand, only\na handful of works tackle the problem of estimating geometry and disentangled\nappearance properties of clothed humans from monocular videos. They usually\nachieve limited quality and disentanglement due to approximations of secondary\nshading effects via learned MLPs. In this work, we propose to model secondary\nshading effects explicitly via Monte-Carlo ray tracing. We model the rendering\nprocess of clothed humans as a volumetric scattering process, and combine ray\ntracing with body articulation. Our approach can recover high-quality geometry,\nalbedo, material, and lighting properties of clothed humans from a single\nmonocular video, without requiring supervised pre-training using ground truth\nmaterials. Furthermore, since we explicitly model the volumetric scattering\nprocess and ray tracing, our model naturally generalizes to novel poses,\nenabling animation of the reconstructed avatar in novel lighting conditions.\n", "date": "2023-12-08", "categories": []}
{"id": "2401.09386", "title": "Tri$^{2}$-plane: Thinking Head Avatar via Feature Pyramid", "abstract": "  Recent years have witnessed considerable achievements in facial avatar\nreconstruction with neural volume rendering. Despite notable advancements, the\nreconstruction of complex and dynamic head movements from monocular videos\nstill suffers from capturing and restoring fine-grained details. In this work,\nwe propose a novel approach, named Tri$^2$-plane, for monocular photo-realistic\nvolumetric head avatar reconstructions. Distinct from the existing works that\nrely on a single tri-plane deformation field for dynamic facial modeling, the\nproposed Tri$^2$-plane leverages the principle of feature pyramids and three\ntop-to-down lateral connections tri-planes for details improvement. It samples\nand renders facial details at multiple scales, transitioning from the entire\nface to specific local regions and then to even more refined sub-regions.\nMoreover, we incorporate a camera-based geometry-aware sliding window method as\nan augmentation in training, which improves the robustness beyond the canonical\nspace, with a particular improvement in cross-identity generation capabilities.\nExperimental outcomes indicate that the Tri$^2$-plane not only surpasses\nexisting methodologies but also achieves superior performance across\nquantitative and qualitative assessments. The project website is:\n\\url{https://songluchuan.github.io/Tri2Plane.github.io/}.\n", "date": "2024-01-17", "categories": []}
{"id": "2407.08414", "title": "MeshAvatar: Learning High-quality Triangular Human Avatars from\n  Multi-view Videos", "abstract": "  We present a novel pipeline for learning high-quality triangular human\navatars from multi-view videos. Recent methods for avatar learning are\ntypically based on neural radiance fields (NeRF), which is not compatible with\ntraditional graphics pipeline and poses great challenges for operations like\nediting or synthesizing under different environments. To overcome these\nlimitations, our method represents the avatar with an explicit triangular mesh\nextracted from an implicit SDF field, complemented by an implicit material\nfield conditioned on given poses. Leveraging this triangular avatar\nrepresentation, we incorporate physics-based rendering to accurately decompose\ngeometry and texture. To enhance both the geometric and appearance details, we\nfurther employ a 2D UNet as the network backbone and introduce pseudo normal\nground-truth as additional supervision. Experiments show that our method can\nlearn triangular avatars with high-quality geometry reconstruction and\nplausible material decomposition, inherently supporting editing, manipulation\nor relighting operations.\n", "date": "2024-07-11", "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": "2407.05324", "title": "PICA: Physics-Integrated Clothed Avatar", "abstract": "  We introduce PICA, a novel representation for high-fidelity animatable\nclothed human avatars with physics-accurate dynamics, even for loose clothing.\nPrevious neural rendering-based representations of animatable clothed humans\ntypically employ a single model to represent both the clothing and the\nunderlying body. While efficient, these approaches often fail to accurately\nrepresent complex garment dynamics, leading to incorrect deformations and\nnoticeable rendering artifacts, especially for sliding or loose garments.\nFurthermore, previous works represent garment dynamics as pose-dependent\ndeformations and facilitate novel pose animations in a data-driven manner. This\noften results in outcomes that do not faithfully represent the mechanics of\nmotion and are prone to generating artifacts in out-of-distribution poses. To\naddress these issues, we adopt two individual 3D Gaussian Splatting (3DGS)\nmodels with different deformation characteristics, modeling the human body and\nclothing separately. This distinction allows for better handling of their\nrespective motion characteristics. With this representation, we integrate a\ngraph neural network (GNN)-based clothed body physics simulation module to\nensure an accurate representation of clothing dynamics. Our method, through its\ncarefully designed features, achieves high-fidelity rendering of clothed human\nbodies in complex and novel driving poses, significantly outperforming previous\nmethods under the same settings.\n", "date": "2024-07-07", "categories": []}
{"id": "2404.15264", "title": "TalkingGaussian: Structure-Persistent 3D Talking Head Synthesis via\n  Gaussian Splatting", "abstract": "  Radiance fields have demonstrated impressive performance in synthesizing\nlifelike 3D talking heads. However, due to the difficulty in fitting steep\nappearance changes, the prevailing paradigm that presents facial motions by\ndirectly modifying point appearance may lead to distortions in dynamic regions.\nTo tackle this challenge, we introduce TalkingGaussian, a deformation-based\nradiance fields framework for high-fidelity talking head synthesis. Leveraging\nthe point-based Gaussian Splatting, facial motions can be represented in our\nmethod by applying smooth and continuous deformations to persistent Gaussian\nprimitives, without requiring to learn the difficult appearance change like\nprevious methods. Due to this simplification, precise facial motions can be\nsynthesized while keeping a highly intact facial feature. Under such a\ndeformation paradigm, we further identify a face-mouth motion inconsistency\nthat would affect the learning of detailed speaking motions. To address this\nconflict, we decompose the model into two branches separately for the face and\ninside mouth areas, therefore simplifying the learning tasks to help\nreconstruct more accurate motion and structure of the mouth region. Extensive\nexperiments demonstrate that our method renders high-quality lip-synchronized\ntalking head videos, with better facial fidelity and higher efficiency compared\nwith previous methods.\n", "date": "2024-04-23", "categories": []}
{"id": "2402.18844", "title": "Deep learning for 3D human pose estimation and mesh recovery: A survey", "abstract": "  3D human pose estimation and mesh recovery have attracted widespread research\ninterest in many areas, such as computer vision, autonomous driving, and\nrobotics. Deep learning on 3D human pose estimation and mesh recovery has\nrecently thrived, with numerous methods proposed to address different problems\nin this area. In this paper, to stimulate future research, we present a\ncomprehensive review of recent progress over the past five years in deep\nlearning methods for this area by delving into over 200 references. To the best\nof our knowledge, this survey is arguably the first to comprehensively cover\ndeep learning methods for 3D human pose estimation, including both\nsingle-person and multi-person approaches, as well as human mesh recovery,\nencompassing methods based on explicit models and implicit representations. We\nalso present comparative results on several publicly available datasets,\ntogether with insightful observations and inspiring future research directions.\nA regularly updated project page can be found at\nhttps://github.com/liuyangme/SOTA-3DHPE-HMR.\n", "date": "2024-02-29", "categories": []}
{"id": "2407.03204", "title": "Expressive Gaussian Human Avatars from Monocular RGB Video", "abstract": "  Nuanced expressiveness, particularly through fine-grained hand and facial\nexpressions, is pivotal for enhancing the realism and vitality of digital human\nrepresentations. In this work, we focus on investigating the expressiveness of\nhuman avatars when learned from monocular RGB video; a setting that introduces\nnew challenges in capturing and animating fine-grained details. To this end, we\nintroduce EVA, a drivable human model that meticulously sculpts fine details\nbased on 3D Gaussians and SMPL-X, an expressive parametric human model. Focused\non enhancing expressiveness, our work makes three key contributions. First, we\nhighlight the critical importance of aligning the SMPL-X model with RGB frames\nfor effective avatar learning. Recognizing the limitations of current SMPL-X\nprediction methods for in-the-wild videos, we introduce a plug-and-play module\nthat significantly ameliorates misalignment issues. Second, we propose a\ncontext-aware adaptive density control strategy, which is adaptively adjusting\nthe gradient thresholds to accommodate the varied granularity across body\nparts. Last but not least, we develop a feedback mechanism that predicts\nper-pixel confidence to better guide the learning of 3D Gaussians. Extensive\nexperiments on two benchmarks demonstrate the superiority of our framework both\nquantitatively and qualitatively, especially on the fine-grained hand and\nfacial details. See the project website at \\url{https://evahuman.github.io}\n", "date": "2024-07-03", "categories": []}
{"id": "2303.11726", "title": "3D Human Mesh Estimation from Virtual Markers", "abstract": "  Inspired by the success of volumetric 3D pose estimation, some recent human\nmesh estimators propose to estimate 3D skeletons as intermediate\nrepresentations, from which, the dense 3D meshes are regressed by exploiting\nthe mesh topology. However, body shape information is lost in extracting\nskeletons, leading to mediocre performance. The advanced motion capture systems\nsolve the problem by placing dense physical markers on the body surface, which\nallows to extract realistic meshes from their non-rigid motions. However, they\ncannot be applied to wild images without markers. In this work, we present an\nintermediate representation, named virtual markers, which learns 64 landmark\nkeypoints on the body surface based on the large-scale mocap data in a\ngenerative style, mimicking the effects of physical markers. The virtual\nmarkers can be accurately detected from wild images and can reconstruct the\nintact meshes with realistic shapes by simple interpolation. Our approach\noutperforms the state-of-the-art methods on three datasets. In particular, it\nsurpasses the existing methods by a notable margin on the SURREAL dataset,\nwhich has diverse body shapes. Code is available at\nhttps://github.com/ShirleyMaxx/VirtualMarker\n", "date": "2023-03-21", "categories": []}
{"id": "2406.19070", "title": "FAGhead: Fully Animate Gaussian Head from Monocular Videos", "abstract": "  High-fidelity reconstruction of 3D human avatars has a wild application in\nvisual reality. In this paper, we introduce FAGhead, a method that enables\nfully controllable human portraits from monocular videos. We explicit the\ntraditional 3D morphable meshes (3DMM) and optimize the neutral 3D Gaussians to\nreconstruct with complex expressions. Furthermore, we employ a novel\nPoint-based Learnable Representation Field (PLRF) with learnable Gaussian point\npositions to enhance reconstruction performance. Meanwhile, to effectively\nmanage the edges of avatars, we introduced the alpha rendering to supervise the\nalpha value of each pixel. Extensive experimental results on the open-source\ndatasets and our capturing datasets demonstrate that our approach is able to\ngenerate high-fidelity 3D head avatars and fully control the expression and\npose of the virtual avatars, which is outperforming than existing works.\n", "date": "2024-06-27", "categories": []}
{"id": "2404.09819", "title": "3D Face Tracking from 2D Video through Iterative Dense UV to Image Flow", "abstract": "  When working with 3D facial data, improving fidelity and avoiding the uncanny\nvalley effect is critically dependent on accurate 3D facial performance\ncapture. Because such methods are expensive and due to the widespread\navailability of 2D videos, recent methods have focused on how to perform\nmonocular 3D face tracking. However, these methods often fall short in\ncapturing precise facial movements due to limitations in their network\narchitecture, training, and evaluation processes. Addressing these challenges,\nwe propose a novel face tracker, FlowFace, that introduces an innovative 2D\nalignment network for dense per-vertex alignment. Unlike prior work, FlowFace\nis trained on high-quality 3D scan annotations rather than weak supervision or\nsynthetic data. Our 3D model fitting module jointly fits a 3D face model from\none or many observations, integrating existing neutral shape priors for\nenhanced identity and expression disentanglement and per-vertex deformations\nfor detailed facial feature reconstruction. Additionally, we propose a novel\nmetric and benchmark for assessing tracking accuracy. Our method exhibits\nsuperior performance on both custom and publicly available benchmarks. We\nfurther validate the effectiveness of our tracker by generating high-quality 3D\ndata from 2D videos, which leads to performance gains on downstream tasks.\n", "date": "2024-04-15", "categories": []}
{"id": "2401.12900", "title": "PSAvatar: A Point-based Shape Model for Real-Time Head Avatar Animation\n  with 3D Gaussian Splatting", "abstract": "  Despite much progress, achieving real-time high-fidelity head avatar\nanimation is still difficult and existing methods have to trade-off between\nspeed and quality. 3DMM based methods often fail to model non-facial structures\nsuch as eyeglasses and hairstyles, while neural implicit models suffer from\ndeformation inflexibility and rendering inefficiency. Although 3D Gaussian has\nbeen demonstrated to possess promising capability for geometry representation\nand radiance field reconstruction, applying 3D Gaussian in head avatar creation\nremains a major challenge since it is difficult for 3D Gaussian to model the\nhead shape variations caused by changing poses and expressions. In this paper,\nwe introduce PSAvatar, a novel framework for animatable head avatar creation\nthat utilizes discrete geometric primitive to create a parametric morphable\nshape model and employs 3D Gaussian for fine detail representation and high\nfidelity rendering. The parametric morphable shape model is a Point-based\nMorphable Shape Model (PMSM) which uses points instead of meshes for 3D\nrepresentation to achieve enhanced representation flexibility. The PMSM first\nconverts the FLAME mesh to points by sampling on the surfaces as well as off\nthe meshes to enable the reconstruction of not only surface-like structures but\nalso complex geometries such as eyeglasses and hairstyles. By aligning these\npoints with the head shape in an analysis-by-synthesis manner, the PMSM makes\nit possible to utilize 3D Gaussian for fine detail representation and\nappearance modeling, thus enabling the creation of high-fidelity avatars. We\nshow that PSAvatar can reconstruct high-fidelity head avatars of a variety of\nsubjects and the avatars can be animated in real-time ($\\ge$ 25 fps at a\nresolution of 512 $\\times$ 512 ).\n", "date": "2024-01-23", "categories": []}
{"id": "2405.12477", "title": "Gaussian Control with Hierarchical Semantic Graphs in 3D Human Recovery", "abstract": "  Although 3D Gaussian Splatting (3DGS) has recently made progress in 3D human\nreconstruction, it primarily relies on 2D pixel-level supervision, overlooking\nthe geometric complexity and topological relationships of different body parts.\nTo address this gap, we introduce the Hierarchical Graph Human Gaussian Control\n(HUGS) framework for achieving high-fidelity 3D human reconstruction. Our\napproach involves leveraging explicitly semantic priors of body parts to ensure\nthe consistency of geometric topology, thereby enabling the capture of the\ncomplex geometrical and topological associations among body parts.\nAdditionally, we disentangle high-frequency features from global human features\nto refine surface details in body parts. Extensive experiments demonstrate that\nour method exhibits superior performance in human body reconstruction,\nparticularly in enhancing surface details and accurately reconstructing body\npart junctions. Codes are available at https://wanghongsheng01.github.io/HUGS/.\n", "date": "2024-05-21", "categories": []}
{"id": "2405.12806", "title": "MOSS: Motion-based 3D Clothed Human Synthesis from Monocular Video", "abstract": "  Single-view clothed human reconstruction holds a central position in virtual\nreality applications, especially in contexts involving intricate human motions.\nIt presents notable challenges in achieving realistic clothing deformation.\nCurrent methodologies often overlook the influence of motion on surface\ndeformation, resulting in surfaces lacking the constraints imposed by global\nmotion. To overcome these limitations, we introduce an innovative framework,\nMotion-Based 3D Clo}thed Humans Synthesis (MOSS), which employs kinematic\ninformation to achieve motion-aware Gaussian split on the human surface. Our\nframework consists of two modules: Kinematic Gaussian Locating Splatting (KGAS)\nand Surface Deformation Detector (UID). KGAS incorporates matrix-Fisher\ndistribution to propagate global motion across the body surface. The density\nand rotation factors of this distribution explicitly control the Gaussians,\nthereby enhancing the realism of the reconstructed surface. Additionally, to\naddress local occlusions in single-view, based on KGAS, UID identifies\nsignificant surfaces, and geometric reconstruction is performed to compensate\nfor these deformations. Experimental results demonstrate that MOSS achieves\nstate-of-the-art visual quality in 3D clothed human synthesis from monocular\nvideos. Notably, we improve the Human NeRF and the Gaussian Splatting by 33.94%\nand 16.75% in LPIPS* respectively. Codes are available at\nhttps://wanghongsheng01.github.io/MOSS/.\n", "date": "2024-05-21", "categories": []}
{"id": "2312.02173", "title": "TailorMe: Self-Supervised Learning of an Anatomically Constrained\n  Volumetric Human Shape Model", "abstract": "  Human shape spaces have been extensively studied, as they are a core element\nof human shape and pose inference tasks. Classic methods for creating a human\nshape model register a surface template mesh to a database of 3D scans and use\ndimensionality reduction techniques, such as Principal Component Analysis, to\nlearn a compact representation. While these shape models enable global shape\nmodifications by correlating anthropometric measurements with the learned\nsubspace, they only provide limited localized shape control. We instead\nregister a volumetric anatomical template, consisting of skeleton bones and\nsoft tissue, to the surface scans of the CAESAR database. We further enlarge\nour training data to the full Cartesian product of all skeletons and all soft\ntissues using physically plausible volumetric deformation transfer. This data\nis then used to learn an anatomically constrained volumetric human shape model\nin a self-supervised fashion. The resulting TailorMe model enables shape\nsampling, localized shape manipulation, and fast inference from given surface\nscans.\n", "date": "2023-11-03", "categories": []}
{"id": "2304.11542", "title": "KBody: Towards general, robust, and aligned monocular whole-body\n  estimation", "abstract": "  KBody is a method for fitting a low-dimensional body model to an image. It\nfollows a predict-and-optimize approach, relying on data-driven model estimates\nfor the constraints that will be used to solve for the body's parameters.\nAcknowledging the importance of high quality correspondences, it leverages\n``virtual joints\" to improve fitting performance, disentangles the optimization\nbetween the pose and shape parameters, and integrates asymmetric distance\nfields to strike a balance in terms of pose and shape capturing capacity, as\nwell as pixel alignment. We also show that generative model inversion offers a\nstrong appearance prior that can be used to complete partial human images and\nused as a building block for generalized and robust monocular body fitting.\nProject page: https://zokin.github.io/KBody.\n", "date": "2023-04-23", "categories": []}
{"id": "2406.10180", "title": "MeshPose: Unifying DensePose and 3D Body Mesh reconstruction", "abstract": "  DensePose provides a pixel-accurate association of images with 3D mesh\ncoordinates, but does not provide a 3D mesh, while Human Mesh Reconstruction\n(HMR) systems have high 2D reprojection error, as measured by DensePose\nlocalization metrics. In this work we introduce MeshPose to jointly tackle\nDensePose and HMR. For this we first introduce new losses that allow us to use\nweak DensePose supervision to accurately localize in 2D a subset of the mesh\nvertices ('VertexPose'). We then lift these vertices to 3D, yielding a low-poly\nbody mesh ('MeshPose'). Our system is trained in an end-to-end manner and is\nthe first HMR method to attain competitive DensePose accuracy, while also being\nlightweight and amenable to efficient inference, making it suitable for\nreal-time AR applications.\n", "date": "2024-06-14", "categories": []}
{"id": "2406.07163", "title": "FaceGPT: Self-supervised Learning to Chat about 3D Human Faces", "abstract": "  We introduce FaceGPT, a self-supervised learning framework for Large\nVision-Language Models (VLMs) to reason about 3D human faces from images and\ntext. Typical 3D face reconstruction methods are specialized algorithms that\nlack semantic reasoning capabilities. FaceGPT overcomes this limitation by\nembedding the parameters of a 3D morphable face model (3DMM) into the token\nspace of a VLM, enabling the generation of 3D faces from both textual and\nvisual inputs. FaceGPT is trained in a self-supervised manner as a model-based\nautoencoder from in-the-wild images. In particular, the hidden state of LLM is\nprojected into 3DMM parameters and subsequently rendered as 2D face image to\nguide the self-supervised learning process via image-based reconstruction.\nWithout relying on expensive 3D annotations of human faces, FaceGPT obtains a\ndetailed understanding about 3D human faces, while preserving the capacity to\nunderstand general user instructions. Our experiments demonstrate that FaceGPT\nnot only achieves high-quality 3D face reconstructions but also retains the\nability for general-purpose visual instruction following. Furthermore, FaceGPT\nlearns fully self-supervised to generate 3D faces based on complex textual\ninputs, which opens a new direction in human face analysis.\n", "date": "2024-06-11", "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": "2406.04253", "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": "2406.02074", "title": "FaceCom: Towards High-fidelity 3D Facial Shape Completion via\n  Optimization and Inpainting Guidance", "abstract": "  We propose FaceCom, a method for 3D facial shape completion, which delivers\nhigh-fidelity results for incomplete facial inputs of arbitrary forms. Unlike\nend-to-end shape completion methods based on point clouds or voxels, our\napproach relies on a mesh-based generative network that is easy to optimize,\nenabling it to handle shape completion for irregular facial scans. We first\ntrain a shape generator on a mixed 3D facial dataset containing 2405\nidentities. Based on the incomplete facial input, we fit complete faces using\nan optimization approach under image inpainting guidance. The completion\nresults are refined through a post-processing step. FaceCom demonstrates the\nability to effectively and naturally complete facial scan data with varying\nmissing regions and degrees of missing areas. Our method can be used in medical\nprosthetic fabrication and the registration of deficient scanning data. Our\nexperimental results demonstrate that FaceCom achieves exceptional performance\nin fitting and shape completion tasks. The code is available at\nhttps://github.com/dragonylee/FaceCom.git.\n", "date": "2024-06-04", "categories": []}
{"id": "2311.18729", "title": "Portrait4D: Learning One-Shot 4D Head Avatar Synthesis using Synthetic\n  Data", "abstract": "  Existing one-shot 4D head synthesis methods usually learn from monocular\nvideos with the aid of 3DMM reconstruction, yet the latter is evenly\nchallenging which restricts them from reasonable 4D head synthesis. We present\na method to learn one-shot 4D head synthesis via large-scale synthetic data.\nThe key is to first learn a part-wise 4D generative model from monocular images\nvia adversarial learning, to synthesize multi-view images of diverse identities\nand full motions as training data; then leverage a transformer-based animatable\ntriplane reconstructor to learn 4D head reconstruction using the synthetic\ndata. A novel learning strategy is enforced to enhance the generalizability to\nreal images by disentangling the learning process of 3D reconstruction and\nreenactment. Experiments demonstrate our superiority over the prior art.\n", "date": "2023-11-30", "categories": []}
{"id": "2405.20786", "title": "Stratified Avatar Generation from Sparse Observations", "abstract": "  Estimating 3D full-body avatars from AR/VR devices is essential for creating\nimmersive experiences in AR/VR applications. This task is challenging due to\nthe limited input from Head Mounted Devices, which capture only sparse\nobservations from the head and hands. Predicting the full-body avatars,\nparticularly the lower body, from these sparse observations presents\nsignificant difficulties. In this paper, we are inspired by the inherent\nproperty of the kinematic tree defined in the Skinned Multi-Person Linear\n(SMPL) model, where the upper body and lower body share only one common\nancestor node, bringing the potential of decoupled reconstruction. We propose a\nstratified approach to decouple the conventional full-body avatar\nreconstruction pipeline into two stages, with the reconstruction of the upper\nbody first and a subsequent reconstruction of the lower body conditioned on the\nprevious stage. To implement this straightforward idea, we leverage the latent\ndiffusion model as a powerful probabilistic generator, and train it to follow\nthe latent distribution of decoupled motions explored by a VQ-VAE\nencoder-decoder model. Extensive experiments on AMASS mocap dataset demonstrate\nour state-of-the-art performance in the reconstruction of full-body motions.\n", "date": "2024-05-30", "categories": []}
{"id": "2406.00637", "title": "Representing Animatable Avatar via Factorized Neural Fields", "abstract": "  For reconstructing high-fidelity human 3D models from monocular videos, it is\ncrucial to maintain consistent large-scale body shapes along with finely\nmatched subtle wrinkles. This paper explores the observation that the per-frame\nrendering results can be factorized into a pose-independent component and a\ncorresponding pose-dependent equivalent to facilitate frame consistency. Pose\nadaptive textures can be further improved by restricting frequency bands of\nthese two components. In detail, pose-independent outputs are expected to be\nlow-frequency, while highfrequency information is linked to pose-dependent\nfactors. We achieve a coherent preservation of both coarse body contours across\nthe entire input video and finegrained texture features that are time variant\nwith a dual-branch network with distinct frequency components. The first branch\ntakes coordinates in canonical space as input, while the second branch\nadditionally considers features outputted by the first branch and pose\ninformation of each frame. Our network integrates the information predicted by\nboth branches and utilizes volume rendering to generate photo-realistic 3D\nhuman images. Through experiments, we demonstrate that our network surpasses\nthe neural radiance fields (NeRF) based state-of-the-art methods in preserving\nhigh-frequency details and ensuring consistent body contours.\n", "date": "2024-06-02", "categories": []}
{"id": "2406.01595", "title": "MultiPly: Reconstruction of Multiple People from Monocular Video in the\n  Wild", "abstract": "  We present MultiPly, a novel framework to reconstruct multiple people in 3D\nfrom monocular in-the-wild videos. Reconstructing multiple individuals moving\nand interacting naturally from monocular in-the-wild videos poses a challenging\ntask. Addressing it necessitates precise pixel-level disentanglement of\nindividuals without any prior knowledge about the subjects. Moreover, it\nrequires recovering intricate and complete 3D human shapes from short video\nsequences, intensifying the level of difficulty. To tackle these challenges, we\nfirst define a layered neural representation for the entire scene, composited\nby individual human and background models. We learn the layered neural\nrepresentation from videos via our layer-wise differentiable volume rendering.\nThis learning process is further enhanced by our hybrid instance segmentation\napproach which combines the self-supervised 3D segmentation and the promptable\n2D segmentation module, yielding reliable instance segmentation supervision\neven under close human interaction. A confidence-guided optimization\nformulation is introduced to optimize the human poses and shape/appearance\nalternately. We incorporate effective objectives to refine human poses via\nphotometric information and impose physically plausible constraints on human\ndynamics, leading to temporally consistent 3D reconstructions with high\nfidelity. The evaluation of our method shows the superiority over prior art on\npublicly available datasets and in-the-wild videos.\n", "date": "2024-06-03", "categories": []}
{"id": "2312.06740", "title": "MonoNPHM: Dynamic Head Reconstruction from Monocular Videos", "abstract": "  We present Monocular Neural Parametric Head Models (MonoNPHM) for dynamic 3D\nhead reconstructions from monocular RGB videos. To this end, we propose a\nlatent appearance space that parameterizes a texture field on top of a neural\nparametric model. We constrain predicted color values to be correlated with the\nunderlying geometry such that gradients from RGB effectively influence latent\ngeometry codes during inverse rendering. To increase the representational\ncapacity of our expression space, we augment our backward deformation field\nwith hyper-dimensions, thus improving color and geometry representation in\ntopologically challenging expressions. Using MonoNPHM as a learned prior, we\napproach the task of 3D head reconstruction using signed distance field based\nvolumetric rendering. By numerically inverting our backward deformation field,\nwe incorporated a landmark loss using facial anchor points that are closely\ntied to our canonical geometry representation. To evaluate the task of dynamic\nface reconstruction from monocular RGB videos we record 20 challenging Kinect\nsequences under casual conditions. MonoNPHM outperforms all baselines with a\nsignificant margin, and makes an important step towards easily accessible\nneural parametric face models through RGB tracking.\n", "date": "2023-12-11", "categories": []}
{"id": "2405.19609", "title": "SMPLX-Lite: A Realistic and Drivable Avatar Benchmark with Rich Geometry\n  and Texture Annotations", "abstract": "  Recovering photorealistic and drivable full-body avatars is crucial for\nnumerous applications, including virtual reality, 3D games, and tele-presence.\nMost methods, whether reconstruction or generation, require large numbers of\nhuman motion sequences and corresponding textured meshes. To easily learn a\ndrivable avatar, a reasonable parametric body model with unified topology is\nparamount. However, existing human body datasets either have images or textured\nmodels and lack parametric models which fit clothes well. We propose a new\nparametric model SMPLX-Lite-D, which can fit detailed geometry of the scanned\nmesh while maintaining stable geometry in the face, hand and foot regions. We\npresent SMPLX-Lite dataset, the most comprehensive clothing avatar dataset with\nmulti-view RGB sequences, keypoints annotations, textured scanned meshes, and\ntextured SMPLX-Lite-D models. With the SMPLX-Lite dataset, we train a\nconditional variational autoencoder model that takes human pose and facial\nkeypoints as input, and generates a photorealistic drivable human avatar.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.19659", "title": "CSANet: Channel Spatial Attention Network for Robust 3D Face Alignment\n  and Reconstruction", "abstract": "  Our project proposes an end-to-end 3D face alignment and reconstruction\nnetwork. The backbone of our model is built by Bottle-Neck structure via\nDepth-wise Separable Convolution. We integrate Coordinate Attention mechanism\nand Spatial Group-wise Enhancement to extract more representative features. For\nmore stable training process and better convergence, we jointly use Wing loss\nand the Weighted Parameter Distance Cost to learn parameters for 3D Morphable\nmodel and 3D vertices. Our proposed model outperforms all baseline models both\nquantitatively and qualitatively.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.19833", "title": "KITRO: Refining Human Mesh by 2D Clues and Kinematic-tree Rotation", "abstract": "  2D keypoints are commonly used as an additional cue to refine estimated 3D\nhuman meshes. Current methods optimize the pose and shape parameters with a\nreprojection loss on the provided 2D keypoints. Such an approach, while simple\nand intuitive, has limited effectiveness because the optimal solution is hard\nto find in ambiguous parameter space and may sacrifice depth. Additionally,\ndivergent gradients from distal joints complicate and deviate the refinement of\nproximal joints in the kinematic chain. To address these, we introduce\nKinematic-Tree Rotation (KITRO), a novel mesh refinement strategy that\nexplicitly models depth and human kinematic-tree structure. KITRO treats\nrefinement from a bone-wise perspective. Unlike previous methods which perform\ngradient-based optimizations, our method calculates bone directions in closed\nform. By accounting for the 2D pose, bone length, and parent joint's depth, the\ncalculation results in two possible directions for each child joint. We then\nuse a decision tree to trace binary choices for all bones along the human\nskeleton's kinematic-tree to select the most probable hypothesis. Our\nexperiments across various datasets and baseline models demonstrate that KITRO\nsignificantly improves 3D joint estimation accuracy and achieves an ideal 2D\nfit simultaneously. Our code available at: https://github.com/MartaYang/KITRO.\n", "date": "2024-05-30", "categories": []}
{"id": "2312.03704", "title": "Relightable Gaussian Codec Avatars", "abstract": "  The fidelity of relighting is bounded by both geometry and appearance\nrepresentations. For geometry, both mesh and volumetric approaches have\ndifficulty modeling intricate structures like 3D hair geometry. For appearance,\nexisting relighting models are limited in fidelity and often too slow to render\nin real-time with high-resolution continuous environments. In this work, we\npresent Relightable Gaussian Codec Avatars, a method to build high-fidelity\nrelightable head avatars that can be animated to generate novel expressions.\nOur geometry model based on 3D Gaussians can capture 3D-consistent\nsub-millimeter details such as hair strands and pores on dynamic face\nsequences. To support diverse materials of human heads such as the eyes, skin,\nand hair in a unified manner, we present a novel relightable appearance model\nbased on learnable radiance transfer. Together with global illumination-aware\nspherical harmonics for the diffuse components, we achieve real-time relighting\nwith all-frequency reflections using spherical Gaussians. This appearance model\ncan be efficiently relit under both point light and continuous illumination. We\nfurther improve the fidelity of eye reflections and enable explicit gaze\ncontrol by introducing relightable explicit eye models. Our method outperforms\nexisting approaches without compromising real-time performance. We also\ndemonstrate real-time relighting of avatars on a tethered consumer VR headset,\nshowcasing the efficiency and fidelity of our avatars.\n", "date": "2023-12-06", "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": "2311.16096", "title": "Animatable and Relightable Gaussians for High-fidelity Human Avatar\n  Modeling", "abstract": "  Modeling animatable human avatars from RGB videos is a long-standing and\nchallenging problem. Recent works usually adopt MLP-based neural radiance\nfields (NeRF) to represent 3D humans, but it remains difficult for pure MLPs to\nregress pose-dependent garment details. To this end, we introduce Animatable\nGaussians, a new avatar representation that leverages powerful 2D CNNs and 3D\nGaussian splatting to create high-fidelity avatars. To associate 3D Gaussians\nwith the animatable avatar, we learn a parametric template from the input\nvideos, and then parameterize the template on two front & back canonical\nGaussian maps where each pixel represents a 3D Gaussian. The learned template\nis adaptive to the wearing garments for modeling looser clothes like dresses.\nSuch template-guided 2D parameterization enables us to employ a powerful\nStyleGAN-based CNN to learn the pose-dependent Gaussian maps for modeling\ndetailed dynamic appearances. Furthermore, we introduce a pose projection\nstrategy for better generalization given novel poses. To tackle the realistic\nrelighting of animatable avatars, we introduce physically-based rendering into\nthe avatar representation for decomposing avatar materials and environment\nillumination. Overall, our method can create lifelike avatars with dynamic,\nrealistic, generalized and relightable appearances. Experiments show that our\nmethod outperforms other state-of-the-art approaches.\n", "date": "2023-11-27", "categories": []}
{"id": "2312.02222", "title": "InvertAvatar: Incremental GAN Inversion for Generalized Head Avatars", "abstract": "  While high fidelity and efficiency are central to the creation of digital\nhead avatars, recent methods relying on 2D or 3D generative models often\nexperience limitations such as shape distortion, expression inaccuracy, and\nidentity flickering. Additionally, existing one-shot inversion techniques fail\nto fully leverage multiple input images for detailed feature extraction. We\npropose a novel framework, \\textbf{Incremental 3D GAN Inversion}, that enhances\navatar reconstruction performance using an algorithm designed to increase the\nfidelity from multiple frames, resulting in improved reconstruction quality\nproportional to frame count. Our method introduces a unique animatable 3D GAN\nprior with two crucial modifications for enhanced expression controllability\nalongside an innovative neural texture encoder that categorizes texture feature\nspaces based on UV parameterization. Differentiating from traditional\ntechniques, our architecture emphasizes pixel-aligned image-to-image\ntranslation, mitigating the need to learn correspondences between observation\nand canonical spaces. Furthermore, we incorporate ConvGRU-based recurrent\nnetworks for temporal data aggregation from multiple frames, boosting geometry\nand texture detail reconstruction. The proposed paradigm demonstrates\nstate-of-the-art performance on one-shot and few-shot avatar animation tasks.\nCode will be available at https://github.com/XChenZ/invertAvatar.\n", "date": "2023-12-03", "categories": []}
{"id": "2405.14855", "title": "Synergistic Global-space Camera and Human Reconstruction from Videos", "abstract": "  Remarkable strides have been made in reconstructing static scenes or human\nbodies from monocular videos. Yet, the two problems have largely been\napproached independently, without much synergy. Most visual SLAM methods can\nonly reconstruct camera trajectories and scene structures up to scale, while\nmost HMR methods reconstruct human meshes in metric scale but fall short in\nreasoning with cameras and scenes. This work introduces Synergistic Camera and\nHuman Reconstruction (SynCHMR) to marry the best of both worlds. Specifically,\nwe design Human-aware Metric SLAM to reconstruct metric-scale camera poses and\nscene point clouds using camera-frame HMR as a strong prior, addressing depth,\nscale, and dynamic ambiguities. Conditioning on the dense scene recovered, we\nfurther learn a Scene-aware SMPL Denoiser to enhance world-frame HMR by\nincorporating spatio-temporal coherency and dynamic scene constraints.\nTogether, they lead to consistent reconstructions of camera trajectories, human\nmeshes, and dense scene point clouds in a common world frame. Project page:\nhttps://paulchhuang.github.io/synchmr\n", "date": "2024-05-23", "categories": []}
{"id": "2405.12069", "title": "Gaussian Head & Shoulders: High Fidelity Neural Upper Body Avatars with\n  Anchor Gaussian Guided Texture Warping", "abstract": "  By equipping the most recent 3D Gaussian Splatting representation with head\n3D morphable models (3DMM), existing methods manage to create head avatars with\nhigh fidelity. However, most existing methods only reconstruct a head without\nthe body, substantially limiting their application scenarios. We found that\nnaively applying Gaussians to model the clothed chest and shoulders tends to\nresult in blurry reconstruction and noisy floaters under novel poses. This is\nbecause of the fundamental limitation of Gaussians and point clouds -- each\nGaussian or point can only have a single directional radiance without spatial\nvariance, therefore an unnecessarily large number of them is required to\nrepresent complicated spatially varying texture, even for simple geometry. In\ncontrast, we propose to model the body part with a neural texture that consists\nof coarse and pose-dependent fine colors. To properly render the body texture\nfor each view and pose without accurate geometry nor UV mapping, we optimize\nanother sparse set of Gaussians as anchors that constrain the neural warping\nfield that maps image plane coordinates to the texture space. We demonstrate\nthat Gaussian Head & Shoulders can fit the high-frequency details on the\nclothed upper body with high fidelity and potentially improve the accuracy and\nfidelity of the head region. We evaluate our method with casual phone-captured\nand internet videos and show our method archives superior reconstruction\nquality and robustness in both self and cross reenactment tasks. To fully\nutilize the efficient rendering speed of Gaussian splatting, we additionally\npropose an accelerated inference method of our trained model without\nMulti-Layer Perceptron (MLP) queries and reach a stable rendering speed of\naround 130 FPS for any subjects.\n", "date": "2024-05-20", "categories": []}
{"id": "2405.12724", "title": "RemoCap: Disentangled Representation Learning for Motion Capture", "abstract": "  Reconstructing 3D human bodies from realistic motion sequences remains a\nchallenge due to pervasive and complex occlusions. Current methods struggle to\ncapture the dynamics of occluded body parts, leading to model penetration and\ndistorted motion. RemoCap leverages Spatial Disentanglement (SD) and Motion\nDisentanglement (MD) to overcome these limitations. SD addresses occlusion\ninterference between the target human body and surrounding objects. It achieves\nthis by disentangling target features along the dimension axis. By aligning\nfeatures based on their spatial positions in each dimension, SD isolates the\ntarget object's response within a global window, enabling accurate capture\ndespite occlusions. The MD module employs a channel-wise temporal shuffling\nstrategy to simulate diverse scene dynamics. This process effectively\ndisentangles motion features, allowing RemoCap to reconstruct occluded parts\nwith greater fidelity. Furthermore, this paper introduces a sequence velocity\nloss that promotes temporal coherence. This loss constrains inter-frame\nvelocity errors, ensuring the predicted motion exhibits realistic consistency.\nExtensive comparisons with state-of-the-art (SOTA) methods on benchmark\ndatasets demonstrate RemoCap's superior performance in 3D human body\nreconstruction. On the 3DPW dataset, RemoCap surpasses all competitors,\nachieving the best results in MPVPE (81.9), MPJPE (72.7), and PA-MPJPE (44.1)\nmetrics. Codes are available at https://wanghongsheng01.github.io/RemoCap/.\n", "date": "2024-05-21", "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": "2405.11270", "title": "HR Human: Modeling Human Avatars with Triangular Mesh and\n  High-Resolution Textures from Videos", "abstract": "  Recently, implicit neural representation has been widely used to generate\nanimatable human avatars. However, the materials and geometry of those\nrepresentations are coupled in the neural network and hard to edit, which\nhinders their application in traditional graphics engines. We present a\nframework for acquiring human avatars that are attached with high-resolution\nphysically-based material textures and triangular mesh from monocular video.\nOur method introduces a novel information fusion strategy to combine the\ninformation from the monocular video and synthesize virtual multi-view images\nto tackle the sparsity of the input view. We reconstruct humans as deformable\nneural implicit surfaces and extract triangle mesh in a well-behaved pose as\nthe initial mesh of the next stage. In addition, we introduce an approach to\ncorrect the bias for the boundary and size of the coarse mesh extracted.\nFinally, we adapt prior knowledge of the latent diffusion model at\nsuper-resolution in multi-view to distill the decomposed texture. Experiments\nshow that our approach outperforms previous representations in terms of high\nfidelity, and this explicit result supports deployment on common renderers.\n", "date": "2024-05-18", "categories": []}
{"id": "2405.11993", "title": "GGAvatar: Geometric Adjustment of Gaussian Head Avatar", "abstract": "  We propose GGAvatar, a novel 3D avatar representation designed to robustly\nmodel dynamic head avatars with complex identities and deformations. GGAvatar\nemploys a coarse-to-fine structure, featuring two core modules: Neutral\nGaussian Initialization Module and Geometry Morph Adjuster. Neutral Gaussian\nInitialization Module pairs Gaussian primitives with deformable triangular\nmeshes, employing an adaptive density control strategy to model the geometric\nstructure of the target subject with neutral expressions. Geometry Morph\nAdjuster introduces deformation bases for each Gaussian in global space,\ncreating fine-grained low-dimensional representations of deformation behaviors\nto address the Linear Blend Skinning formula's limitations effectively.\nExtensive experiments show that GGAvatar can produce high-fidelity renderings,\noutperforming state-of-the-art methods in visual quality and quantitative\nmetrics.\n", "date": "2024-05-20", "categories": []}
{"id": "2309.04750", "title": "Mirror-Aware Neural Humans", "abstract": "  Human motion capture either requires multi-camera systems or is unreliable\nwhen using single-view input due to depth ambiguities. Meanwhile, mirrors are\nreadily available in urban environments and form an affordable alternative by\nrecording two views with only a single camera. However, the mirror setting\nposes the additional challenge of handling occlusions of real and mirror image.\nGoing beyond existing mirror approaches for 3D human pose estimation, we\nutilize mirrors for learning a complete body model, including shape and dense\nappearance. Our main contributions are extending articulated neural radiance\nfields to include a notion of a mirror, making it sample-efficient over\npotential occlusion regions. Together, our contributions realize a\nconsumer-level 3D motion capture system that starts from off-the-shelf 2D poses\nby automatically calibrating the camera, estimating mirror orientation, and\nsubsequently lifting 2D keypoint detections to 3D skeleton pose that is used to\ncondition the mirror-aware NeRF. We empirically demonstrate the benefit of\nlearning a body model and accounting for occlusion in challenging mirror\nscenes.\n", "date": "2023-09-09", "categories": []}
{"id": "2405.07319", "title": "LayGA: Layered Gaussian Avatars for Animatable Clothing Transfer", "abstract": "  Animatable clothing transfer, aiming at dressing and animating garments\nacross characters, is a challenging problem. Most human avatar works entangle\nthe representations of the human body and clothing together, which leads to\ndifficulties for virtual try-on across identities. What's worse, the entangled\nrepresentations usually fail to exactly track the sliding motion of garments.\nTo overcome these limitations, we present Layered Gaussian Avatars (LayGA), a\nnew representation that formulates body and clothing as two separate layers for\nphotorealistic animatable clothing transfer from multi-view videos. Our\nrepresentation is built upon the Gaussian map-based avatar for its excellent\nrepresentation power of garment details. However, the Gaussian map produces\nunstructured 3D Gaussians distributed around the actual surface. The absence of\na smooth explicit surface raises challenges in accurate garment tracking and\ncollision handling between body and garments. Therefore, we propose two-stage\ntraining involving single-layer reconstruction and multi-layer fitting. In the\nsingle-layer reconstruction stage, we propose a series of geometric constraints\nto reconstruct smooth surfaces and simultaneously obtain the segmentation\nbetween body and clothing. Next, in the multi-layer fitting stage, we train two\nseparate models to represent body and clothing and utilize the reconstructed\nclothing geometries as 3D supervision for more accurate garment tracking.\nFurthermore, we propose geometry and rendering layers for both high-quality\ngeometric reconstruction and high-fidelity rendering. Overall, the proposed\nLayGA realizes photorealistic animations and virtual try-on, and outperforms\nother baseline methods. Our project page is\nhttps://jsnln.github.io/layga/index.html.\n", "date": "2024-05-12", "categories": []}
{"id": "2405.07933", "title": "Authentic Hand Avatar from a Phone Scan via Universal Hand Model", "abstract": "  The authentic 3D hand avatar with every identifiable information, such as\nhand shapes and textures, is necessary for immersive experiences in AR/VR. In\nthis paper, we present a universal hand model (UHM), which 1) can universally\nrepresent high-fidelity 3D hand meshes of arbitrary identities (IDs) and 2) can\nbe adapted to each person with a short phone scan for the authentic hand\navatar. For effective universal hand modeling, we perform tracking and modeling\nat the same time, while previous 3D hand models perform them separately. The\nconventional separate pipeline suffers from the accumulated errors from the\ntracking stage, which cannot be recovered in the modeling stage. On the other\nhand, ours does not suffer from the accumulated errors while having a much more\nconcise overall pipeline. We additionally introduce a novel image matching loss\nfunction to address a skin sliding during the tracking and modeling, while\nexisting works have not focused on it much. Finally, using learned priors from\nour UHM, we effectively adapt our UHM to each person's short phone scan for the\nauthentic hand avatar.\n", "date": "2024-05-13", "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": "2402.04101", "title": "VRMM: A Volumetric Relightable Morphable Head Model", "abstract": "  In this paper, we introduce the Volumetric Relightable Morphable Model\n(VRMM), a novel volumetric and parametric facial prior for 3D face modeling.\nWhile recent volumetric prior models offer improvements over traditional\nmethods like 3D Morphable Models (3DMMs), they face challenges in model\nlearning and personalized reconstructions. Our VRMM overcomes these by\nemploying a novel training framework that efficiently disentangles and encodes\nlatent spaces of identity, expression, and lighting into low-dimensional\nrepresentations. This framework, designed with self-supervised learning,\nsignificantly reduces the constraints for training data, making it more\nfeasible in practice. The learned VRMM offers relighting capabilities and\nencompasses a comprehensive range of expressions. We demonstrate the\nversatility and effectiveness of VRMM through various applications like avatar\ngeneration, facial reconstruction, and animation. Additionally, we address the\ncommon issue of overfitting in generative volumetric models with a novel\nprior-preserving personalization framework based on VRMM. Such an approach\nenables high-quality 3D face reconstruction from even a single portrait input.\nOur experiments showcase the potential of VRMM to significantly enhance the\nfield of 3D face modeling.\n", "date": "2024-02-06", "categories": []}
{"id": "2308.10638", "title": "SCULPT: Shape-Conditioned Unpaired Learning of Pose-dependent Clothed\n  and Textured Human Meshes", "abstract": "  We present SCULPT, a novel 3D generative model for clothed and textured 3D\nmeshes of humans. Specifically, we devise a deep neural network that learns to\nrepresent the geometry and appearance distribution of clothed human bodies.\nTraining such a model is challenging, as datasets of textured 3D meshes for\nhumans are limited in size and accessibility. Our key observation is that there\nexist medium-sized 3D scan datasets like CAPE, as well as large-scale 2D image\ndatasets of clothed humans and multiple appearances can be mapped to a single\ngeometry. To effectively learn from the two data modalities, we propose an\nunpaired learning procedure for pose-dependent clothed and textured human\nmeshes. Specifically, we learn a pose-dependent geometry space from 3D scan\ndata. We represent this as per vertex displacements w.r.t. the SMPL model.\nNext, we train a geometry conditioned texture generator in an unsupervised way\nusing the 2D image data. We use intermediate activations of the learned\ngeometry model to condition our texture generator. To alleviate entanglement\nbetween pose and clothing type, and pose and clothing appearance, we condition\nboth the texture and geometry generators with attribute labels such as clothing\ntypes for the geometry, and clothing colors for the texture generator. We\nautomatically generated these conditioning labels for the 2D images based on\nthe visual question answering model BLIP and CLIP. We validate our method on\nthe SCULPT dataset, and compare to state-of-the-art 3D generative models for\nclothed human bodies. Our code and data can be found at\nhttps://sculpt.is.tue.mpg.de.\n", "date": "2023-08-21", "categories": []}
{"id": "2404.19398", "title": "3D Gaussian Blendshapes for Head Avatar Animation", "abstract": "  We introduce 3D Gaussian blendshapes for modeling photorealistic head\navatars. Taking a monocular video as input, we learn a base head model of\nneutral expression, along with a group of expression blendshapes, each of which\ncorresponds to a basis expression in classical parametric face models. Both the\nneutral model and expression blendshapes are represented as 3D Gaussians, which\ncontain a few properties to depict the avatar appearance. The avatar model of\nan arbitrary expression can be effectively generated by combining the neutral\nmodel and expression blendshapes through linear blending of Gaussians with the\nexpression coefficients. High-fidelity head avatar animations can be\nsynthesized in real time using Gaussian splatting. Compared to state-of-the-art\nmethods, our Gaussian blendshape representation better captures high-frequency\ndetails exhibited in input video, and achieves superior rendering performance.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.00794", "title": "Coherent 3D Portrait Video Reconstruction via Triplane Fusion", "abstract": "  Recent breakthroughs in single-image 3D portrait reconstruction have enabled\ntelepresence systems to stream 3D portrait videos from a single camera in\nreal-time, potentially democratizing telepresence. However, per-frame 3D\nreconstruction exhibits temporal inconsistency and forgets the user's\nappearance. On the other hand, self-reenactment methods can render coherent 3D\nportraits by driving a personalized 3D prior, but fail to faithfully\nreconstruct the user's per-frame appearance (e.g., facial expressions and\nlighting). In this work, we recognize the need to maintain both coherent\nidentity and dynamic per-frame appearance to enable the best possible realism.\nTo this end, we propose a new fusion-based method that fuses a personalized 3D\nsubject prior with per-frame information, producing temporally stable 3D videos\nwith faithful reconstruction of the user's per-frame appearances. Trained only\nusing synthetic data produced by an expression-conditioned 3D GAN, our\nencoder-based method achieves both state-of-the-art 3D reconstruction accuracy\nand temporal consistency on in-studio and in-the-wild datasets.\n", "date": "2024-05-01", "categories": []}
{"id": "2404.16012", "title": "GaussianTalker: Real-Time High-Fidelity Talking Head Synthesis with\n  Audio-Driven 3D Gaussian Splatting", "abstract": "  We propose GaussianTalker, a novel framework for real-time generation of\npose-controllable talking heads. It leverages the fast rendering capabilities\nof 3D Gaussian Splatting (3DGS) while addressing the challenges of directly\ncontrolling 3DGS with speech audio. GaussianTalker constructs a canonical 3DGS\nrepresentation of the head and deforms it in sync with the audio. A key insight\nis to encode the 3D Gaussian attributes into a shared implicit feature\nrepresentation, where it is merged with audio features to manipulate each\nGaussian attribute. This design exploits the spatial-aware features and\nenforces interactions between neighboring points. The feature embeddings are\nthen fed to a spatial-audio attention module, which predicts frame-wise offsets\nfor the attributes of each Gaussian. It is more stable than previous\nconcatenation or multiplication approaches for manipulating the numerous\nGaussians and their intricate parameters. Experimental results showcase\nGaussianTalker's superiority in facial fidelity, lip synchronization accuracy,\nand rendering speed compared to previous methods. Specifically, GaussianTalker\nachieves a remarkable rendering speed up to 120 FPS, surpassing previous\nbenchmarks. Our code is made available at\nhttps://github.com/KU-CVLAB/GaussianTalker/ .\n", "date": "2024-04-24", "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": "2404.15276", "title": "SMPLer: Taming Transformers for Monocular 3D Human Shape and Pose\n  Estimation", "abstract": "  Existing Transformers for monocular 3D human shape and pose estimation\ntypically have a quadratic computation and memory complexity with respect to\nthe feature length, which hinders the exploitation of fine-grained information\nin high-resolution features that is beneficial for accurate reconstruction. In\nthis work, we propose an SMPL-based Transformer framework (SMPLer) to address\nthis issue. SMPLer incorporates two key ingredients: a decoupled attention\noperation and an SMPL-based target representation, which allow effective\nutilization of high-resolution features in the Transformer. In addition, based\non these two designs, we also introduce several novel modules including a\nmulti-scale attention and a joint-aware attention to further boost the\nreconstruction performance. Extensive experiments demonstrate the effectiveness\nof SMPLer against existing 3D human shape and pose estimation methods both\nquantitatively and qualitatively. Notably, the proposed algorithm achieves an\nMPJPE of 45.2 mm on the Human3.6M dataset, improving upon Mesh Graphormer by\nmore than 10% with fewer than one-third of the parameters. Code and pretrained\nmodels are available at https://github.com/xuxy09/SMPLer.\n", "date": "2024-04-23", "categories": []}
{"id": "2404.13862", "title": "PGAHum: Prior-Guided Geometry and Appearance Learning for High-Fidelity\n  Animatable Human Reconstruction", "abstract": "  Recent techniques on implicit geometry representation learning and neural\nrendering have shown promising results for 3D clothed human reconstruction from\nsparse video inputs. However, it is still challenging to reconstruct detailed\nsurface geometry and even more difficult to synthesize photorealistic novel\nviews with animated human poses. In this work, we introduce PGAHum, a\nprior-guided geometry and appearance learning framework for high-fidelity\nanimatable human reconstruction. We thoroughly exploit 3D human priors in three\nkey modules of PGAHum to achieve high-quality geometry reconstruction with\nintricate details and photorealistic view synthesis on unseen poses. First, a\nprior-based implicit geometry representation of 3D human, which contains a\ndelta SDF predicted by a tri-plane network and a base SDF derived from the\nprior SMPL model, is proposed to model the surface details and the body shape\nin a disentangled manner. Second, we introduce a novel prior-guided sampling\nstrategy that fully leverages the prior information of the human pose and body\nto sample the query points within or near the body surface. By avoiding\nunnecessary learning in the empty 3D space, the neural rendering can recover\nmore appearance details. Last, we propose a novel iterative backward\ndeformation strategy to progressively find the correspondence for the query\npoint in observation space. A skinning weights prediction model is learned\nbased on the prior provided by the SMPL model to achieve the iterative backward\nLBS deformation. Extensive quantitative and qualitative comparisons on various\ndatasets are conducted and the results demonstrate the superiority of our\nframework. Ablation studies also verify the effectiveness of each scheme for\ngeometry and appearance learning.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.14199", "title": "Generalizable Neural Human Renderer", "abstract": "  While recent advancements in animatable human rendering have achieved\nremarkable results, they require test-time optimization for each subject which\ncan be a significant limitation for real-world applications. To address this,\nwe tackle the challenging task of learning a Generalizable Neural Human\nRenderer (GNH), a novel method for rendering animatable humans from monocular\nvideo without any test-time optimization. Our core method focuses on\ntransferring appearance information from the input video to the output image\nplane by utilizing explicit body priors and multi-view geometry. To render the\nsubject in the intended pose, we utilize a straightforward CNN-based image\nrenderer, foregoing the more common ray-sampling or rasterizing-based rendering\nmodules. Our GNH achieves remarkable generalizable, photorealistic rendering\nwith unseen subjects with a three-stage process. We quantitatively and\nqualitatively demonstrate that GNH significantly surpasses current\nstate-of-the-art methods, notably achieving a 31.3% improvement in LPIPS.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.14410", "title": "Guess The Unseen: Dynamic 3D Scene Reconstruction from Partial 2D\n  Glimpses", "abstract": "  In this paper, we present a method to reconstruct the world and multiple\ndynamic humans in 3D from a monocular video input. As a key idea, we represent\nboth the world and multiple humans via the recently emerging 3D Gaussian\nSplatting (3D-GS) representation, enabling to conveniently and efficiently\ncompose and render them together. In particular, we address the scenarios with\nseverely limited and sparse observations in 3D human reconstruction, a common\nchallenge encountered in the real world. To tackle this challenge, we introduce\na novel approach to optimize the 3D-GS representation in a canonical space by\nfusing the sparse cues in the common space, where we leverage a pre-trained 2D\ndiffusion model to synthesize unseen views while keeping the consistency with\nthe observed 2D appearances. We demonstrate our method can reconstruct\nhigh-quality animatable 3D humans in various challenging examples, in the\npresence of occlusion, image crops, few-shot, and extremely sparse\nobservations. After reconstruction, our method is capable of not only rendering\nthe scene in any novel views at arbitrary time instances, but also editing the\n3D scene by removing individual humans or applying different motions for each\nhuman. Through various experiments, we demonstrate the quality and efficiency\nof our methods over alternative existing approaches.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.04876", "title": "HiLo: Detailed and Robust 3D Clothed Human Reconstruction with High-and\n  Low-Frequency Information of Parametric Models", "abstract": "  Reconstructing 3D clothed human involves creating a detailed geometry of\nindividuals in clothing, with applications ranging from virtual try-on, movies,\nto games. To enable practical and widespread applications, recent advances\npropose to generate a clothed human from an RGB image. However, they struggle\nto reconstruct detailed and robust avatars simultaneously. We empirically find\nthat the high-frequency (HF) and low-frequency (LF) information from a\nparametric model has the potential to enhance geometry details and improve\nrobustness to noise, respectively. Based on this, we propose HiLo, namely\nclothed human reconstruction with high- and low-frequency information, which\ncontains two components. 1) To recover detailed geometry using HF information,\nwe propose a progressive HF Signed Distance Function to enhance the detailed 3D\ngeometry of a clothed human. We analyze that our progressive learning manner\nalleviates large gradients that hinder model convergence. 2) To achieve robust\nreconstruction against inaccurate estimation of the parametric model by using\nLF information, we propose a spatial interaction implicit function. This\nfunction effectively exploits the complementary spatial information from a\nlow-resolution voxel grid of the parametric model. Experimental results\ndemonstrate that HiLo outperforms the state-of-the-art methods by 10.43% and\n9.54% in terms of Chamfer distance on the Thuman2.0 and CAPE datasets,\nrespectively. Additionally, HiLo demonstrates robustness to noise from the\nparametric model, challenging poses, and various clothing styles.\n", "date": "2024-04-07", "categories": []}
{"id": "2311.18635", "title": "DiffusionAvatars: Deferred Diffusion for High-fidelity 3D Head Avatars", "abstract": "  DiffusionAvatars synthesizes a high-fidelity 3D head avatar of a person,\noffering intuitive control over both pose and expression. We propose a\ndiffusion-based neural renderer that leverages generic 2D priors to produce\ncompelling images of faces. For coarse guidance of the expression and head\npose, we render a neural parametric head model (NPHM) from the target\nviewpoint, which acts as a proxy geometry of the person. Additionally, to\nenhance the modeling of intricate facial expressions, we condition\nDiffusionAvatars directly on the expression codes obtained from NPHM via\ncross-attention. Finally, to synthesize consistent surface details across\ndifferent viewpoints and expressions, we rig learnable spatial features to the\nhead's surface via TriPlane lookup in NPHM's canonical space. We train\nDiffusionAvatars on RGB videos and corresponding fitted NPHM meshes of a person\nand test the obtained avatars in both self-reenactment and animation scenarios.\nOur experiments demonstrate that DiffusionAvatars generates temporally\nconsistent and visually appealing videos for novel poses and expressions of a\nperson, outperforming existing approaches.\n", "date": "2023-11-30", "categories": []}
{"id": "2312.00311", "title": "3D Face Reconstruction with the Geometric Guidance of Facial Part\n  Segmentation", "abstract": "  3D Morphable Models (3DMMs) provide promising 3D face reconstructions in\nvarious applications. However, existing methods struggle to reconstruct faces\nwith extreme expressions due to deficiencies in supervisory signals, such as\nsparse or inaccurate landmarks. Segmentation information contains effective\ngeometric contexts for face reconstruction. Certain attempts intuitively depend\non differentiable renderers to compare the rendered silhouettes of\nreconstruction with segmentation, which is prone to issues like local optima\nand gradient instability. In this paper, we fully utilize the facial part\nsegmentation geometry by introducing Part Re-projection Distance Loss (PRDL).\nSpecifically, PRDL transforms facial part segmentation into 2D points and\nre-projects the reconstruction onto the image plane. Subsequently, by\nintroducing grid anchors and computing different statistical distances from\nthese anchors to the point sets, PRDL establishes geometry descriptors to\noptimize the distribution of the point sets for face reconstruction. PRDL\nexhibits a clear gradient compared to the renderer-based methods and presents\nstate-of-the-art reconstruction performance in extensive quantitative and\nqualitative experiments. Our project is available at\nhttps://github.com/wang-zidu/3DDFA-V3 .\n", "date": "2023-12-01", "categories": []}
{"id": "2404.11291", "title": "Closely Interactive Human Reconstruction with Proxemics and\n  Physics-Guided Adaption", "abstract": "  Existing multi-person human reconstruction approaches mainly focus on\nrecovering accurate poses or avoiding penetration, but overlook the modeling of\nclose interactions. In this work, we tackle the task of reconstructing closely\ninteractive humans from a monocular video. The main challenge of this task\ncomes from insufficient visual information caused by depth ambiguity and severe\ninter-person occlusion. In view of this, we propose to leverage knowledge from\nproxemic behavior and physics to compensate the lack of visual information.\nThis is based on the observation that human interaction has specific patterns\nfollowing the social proxemics. Specifically, we first design a latent\nrepresentation based on Vector Quantised-Variational AutoEncoder (VQ-VAE) to\nmodel human interaction. A proxemics and physics guided diffusion model is then\nintroduced to denoise the initial distribution. We design the diffusion model\nas dual branch with each branch representing one individual such that the\ninteraction can be modeled via cross attention. With the learned priors of\nVQ-VAE and physical constraint as the additional information, our proposed\napproach is capable of estimating accurate poses that are also proxemics and\nphysics plausible. Experimental results on Hi4D, 3DPW, and CHI3D demonstrate\nthat our method outperforms existing approaches. The code is available at\n\\url{https://github.com/boycehbz/HumanInteraction}.\n", "date": "2024-04-17", "categories": []}
{"id": "2312.02155", "title": "GPS-Gaussian: Generalizable Pixel-wise 3D Gaussian Splatting for\n  Real-time Human Novel View Synthesis", "abstract": "  We present a new approach, termed GPS-Gaussian, for synthesizing novel views\nof a character in a real-time manner. The proposed method enables 2K-resolution\nrendering under a sparse-view camera setting. Unlike the original Gaussian\nSplatting or neural implicit rendering methods that necessitate per-subject\noptimizations, we introduce Gaussian parameter maps defined on the source views\nand regress directly Gaussian Splatting properties for instant novel view\nsynthesis without any fine-tuning or optimization. To this end, we train our\nGaussian parameter regression module on a large amount of human scan data,\njointly with a depth estimation module to lift 2D parameter maps to 3D space.\nThe proposed framework is fully differentiable and experiments on several\ndatasets demonstrate that our method outperforms state-of-the-art methods while\nachieving an exceeding rendering speed.\n", "date": "2023-12-04", "categories": []}
{"id": "2404.08504", "title": "3D Human Scan With A Moving Event Camera", "abstract": "  Capturing a 3D human body is one of the important tasks in computer vision\nwith a wide range of applications such as virtual reality and sports analysis.\nHowever, conventional frame cameras are limited by their temporal resolution\nand dynamic range, which imposes constraints in real-world application setups.\nEvent cameras have the advantages of high temporal resolution and high dynamic\nrange (HDR), but the development of event-based methods is necessary to handle\ndata with different characteristics. This paper proposes a novel event-based\nmethod for 3D pose estimation and human mesh recovery. Prior work on\nevent-based human mesh recovery require frames (images) as well as event data.\nThe proposed method solely relies on events; it carves 3D voxels by moving the\nevent camera around a stationary body, reconstructs the human pose and mesh by\nattenuated rays, and fit statistical body models, preserving high-frequency\ndetails. The experimental results show that the proposed method outperforms\nconventional frame-based methods in the estimation accuracy of both pose and\nbody mesh. We also demonstrate results in challenging situations where a\nconventional camera has motion blur. This is the first to demonstrate\nevent-only human mesh recovery, and we hope that it is the first step toward\nachieving robust and accurate 3D human body scanning from vision sensors.\nhttps://florpeng.github.io/event-based-human-scan/\n", "date": "2024-04-12", "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": "2312.05941", "title": "ASH: Animatable Gaussian Splats for Efficient and Photoreal Human\n  Rendering", "abstract": "  Real-time rendering of photorealistic and controllable human avatars stands\nas a cornerstone in Computer Vision and Graphics. While recent advances in\nneural implicit rendering have unlocked unprecedented photorealism for digital\navatars, real-time performance has mostly been demonstrated for static scenes\nonly. To address this, we propose ASH, an animatable Gaussian splatting\napproach for photorealistic rendering of dynamic humans in real-time. We\nparameterize the clothed human as animatable 3D Gaussians, which can be\nefficiently splatted into image space to generate the final rendering. However,\nnaively learning the Gaussian parameters in 3D space poses a severe challenge\nin terms of compute. Instead, we attach the Gaussians onto a deformable\ncharacter model, and learn their parameters in 2D texture space, which allows\nleveraging efficient 2D convolutional architectures that easily scale with the\nrequired number of Gaussians. We benchmark ASH with competing methods on\npose-controllable avatars, demonstrating that our method outperforms existing\nreal-time methods by a large margin and shows comparable or even better results\nthan offline methods.\n", "date": "2023-12-10", "categories": []}
{"id": "2401.08570", "title": "RoHM: Robust Human Motion Reconstruction via Diffusion", "abstract": "  We propose RoHM, an approach for robust 3D human motion reconstruction from\nmonocular RGB(-D) videos in the presence of noise and occlusions. Most previous\napproaches either train neural networks to directly regress motion in 3D or\nlearn data-driven motion priors and combine them with optimization at test\ntime. The former do not recover globally coherent motion and fail under\nocclusions; the latter are time-consuming, prone to local minima, and require\nmanual tuning. To overcome these shortcomings, we exploit the iterative,\ndenoising nature of diffusion models. RoHM is a novel diffusion-based motion\nmodel that, conditioned on noisy and occluded input data, reconstructs\ncomplete, plausible motions in consistent global coordinates. Given the\ncomplexity of the problem -- requiring one to address different tasks\n(denoising and infilling) in different solution spaces (local and global\nmotion) -- we decompose it into two sub-tasks and learn two models, one for\nglobal trajectory and one for local motion. To capture the correlations between\nthe two, we then introduce a novel conditioning module, combining it with an\niterative inference scheme. We apply RoHM to a variety of tasks -- from motion\nreconstruction and denoising to spatial and temporal infilling. Extensive\nexperiments on three popular datasets show that our method outperforms\nstate-of-the-art approaches qualitatively and quantitatively, while being\nfaster at test time. The code is available at\nhttps://sanweiliti.github.io/ROHM/ROHM.html.\n", "date": "2024-01-16", "categories": []}
{"id": "2404.07991", "title": "GoMAvatar: Efficient Animatable Human Modeling from Monocular Video\n  Using Gaussians-on-Mesh", "abstract": "  We introduce GoMAvatar, a novel approach for real-time, memory-efficient,\nhigh-quality animatable human modeling. GoMAvatar takes as input a single\nmonocular video to create a digital avatar capable of re-articulation in new\nposes and real-time rendering from novel viewpoints, while seamlessly\nintegrating with rasterization-based graphics pipelines. Central to our method\nis the Gaussians-on-Mesh representation, a hybrid 3D model combining rendering\nquality and speed of Gaussian splatting with geometry modeling and\ncompatibility of deformable meshes. We assess GoMAvatar on ZJU-MoCap data and\nvarious YouTube videos. GoMAvatar matches or surpasses current monocular human\nmodeling algorithms in rendering quality and significantly outperforms them in\ncomputational efficiency (43 FPS) while being memory-efficient (3.63 MB per\nsubject).\n", "date": "2024-04-11", "categories": []}
{"id": "2404.07094", "title": "MoCap-to-Visual Domain Adaptation for Efficient Human Mesh Estimation\n  from 2D Keypoints", "abstract": "  This paper presents Key2Mesh, a model that takes a set of 2D human pose\nkeypoints as input and estimates the corresponding body mesh. Since this\nprocess does not involve any visual (i.e. RGB image) data, the model can be\ntrained on large-scale motion capture (MoCap) datasets, thereby overcoming the\nscarcity of image datasets with 3D labels. To enable the model's application on\nRGB images, we first run an off-the-shelf 2D pose estimator to obtain the 2D\nkeypoints, and then feed these 2D keypoints to Key2Mesh. To improve the\nperformance of our model on RGB images, we apply an adversarial domain\nadaptation (DA) method to bridge the gap between the MoCap and visual domains.\nCrucially, our DA method does not require 3D labels for visual data, which\nenables adaptation to target sets without the need for costly labels. We\nevaluate Key2Mesh for the task of estimating 3D human meshes from 2D keypoints,\nin the absence of RGB and mesh label pairs. Our results on widely used H3.6M\nand 3DPW datasets show that Key2Mesh sets the new state-of-the-art by\noutperforming other models in PA-MPJPE for both datasets, and in MPJPE and PVE\nfor the 3DPW dataset. Thanks to our model's simple architecture, it operates at\nleast 12x faster than the prior state-of-the-art model, LGD. Additional\nqualitative samples and code are available on the project website:\nhttps://key2mesh.github.io/.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.04421", "title": "PhysAvatar: Learning the Physics of Dressed 3D Avatars from Visual\n  Observations", "abstract": "  Modeling and rendering photorealistic avatars is of crucial importance in\nmany applications. Existing methods that build a 3D avatar from visual\nobservations, however, struggle to reconstruct clothed humans. We introduce\nPhysAvatar, a novel framework that combines inverse rendering with inverse\nphysics to automatically estimate the shape and appearance of a human from\nmulti-view video data along with the physical parameters of the fabric of their\nclothes. For this purpose, we adopt a mesh-aligned 4D Gaussian technique for\nspatio-temporal mesh tracking as well as a physically based inverse renderer to\nestimate the intrinsic material properties. PhysAvatar integrates a physics\nsimulator to estimate the physical parameters of the garments using\ngradient-based optimization in a principled manner. These novel capabilities\nenable PhysAvatar to create high-quality novel-view renderings of avatars\ndressed in loose-fitting clothes under motions and lighting conditions not seen\nin the training data. This marks a significant advancement towards modeling\nphotorealistic digital humans using physically based inverse rendering with\nphysics in the loop. Our project website is at:\nhttps://qingqing-zhao.github.io/PhysAvatar\n", "date": "2024-04-05", "categories": []}
{"id": "2312.01068", "title": "DPHMs: Diffusion Parametric Head Models for Depth-based Tracking", "abstract": "  We introduce Diffusion Parametric Head Models (DPHMs), a generative model\nthat enables robust volumetric head reconstruction and tracking from monocular\ndepth sequences. While recent volumetric head models, such as NPHMs, can now\nexcel in representing high-fidelity head geometries, tracking and\nreconstructing heads from real-world single-view depth sequences remains very\nchallenging, as the fitting to partial and noisy observations is\nunderconstrained. To tackle these challenges, we propose a latent\ndiffusion-based prior to regularize volumetric head reconstruction and\ntracking. This prior-based regularizer effectively constrains the identity and\nexpression codes to lie on the underlying latent manifold which represents\nplausible head shapes. To evaluate the effectiveness of the diffusion-based\nprior, we collect a dataset of monocular Kinect sequences consisting of various\ncomplex facial expression motions and rapid transitions. We compare our method\nto state-of-the-art tracking methods and demonstrate improved head identity\nreconstruction as well as robust expression tracking.\n", "date": "2023-12-02", "categories": []}
{"id": "2312.06704", "title": "SIFU: Side-view Conditioned Implicit Function for Real-world Usable\n  Clothed Human Reconstruction", "abstract": "  Creating high-quality 3D models of clothed humans from single images for\nreal-world applications is crucial. Despite recent advancements, accurately\nreconstructing humans in complex poses or with loose clothing from in-the-wild\nimages, along with predicting textures for unseen areas, remains a significant\nchallenge. A key limitation of previous methods is their insufficient prior\nguidance in transitioning from 2D to 3D and in texture prediction. In response,\nwe introduce SIFU (Side-view Conditioned Implicit Function for Real-world\nUsable Clothed Human Reconstruction), a novel approach combining a Side-view\nDecoupling Transformer with a 3D Consistent Texture Refinement pipeline.SIFU\nemploys a cross-attention mechanism within the transformer, using SMPL-X\nnormals as queries to effectively decouple side-view features in the process of\nmapping 2D features to 3D. This method not only improves the precision of the\n3D models but also their robustness, especially when SMPL-X estimates are not\nperfect. Our texture refinement process leverages text-to-image diffusion-based\nprior to generate realistic and consistent textures for invisible views.\nThrough extensive experiments, SIFU surpasses SOTA methods in both geometry and\ntexture reconstruction, showcasing enhanced robustness in complex scenarios and\nachieving an unprecedented Chamfer and P2S measurement. Our approach extends to\npractical applications such as 3D printing and scene building, demonstrating\nits broad utility in real-world scenarios. Project page\nhttps://river-zhang.github.io/SIFU-projectpage/ .\n", "date": "2023-12-10", "categories": []}
{"id": "2312.07063", "title": "Template Free Reconstruction of Human-object Interaction with Procedural\n  Interaction Generation", "abstract": "  Reconstructing human-object interaction in 3D from a single RGB image is a\nchallenging task and existing data driven methods do not generalize beyond the\nobjects present in the carefully curated 3D interaction datasets. Capturing\nlarge-scale real data to learn strong interaction and 3D shape priors is very\nexpensive due to the combinatorial nature of human-object interactions. In this\npaper, we propose ProciGen (Procedural interaction Generation), a method to\nprocedurally generate datasets with both, plausible interaction and diverse\nobject variation. We generate 1M+ human-object interaction pairs in 3D and\nleverage this large-scale data to train our HDM (Hierarchical Diffusion Model),\na novel method to reconstruct interacting human and unseen objects, without any\ntemplates. Our HDM is an image-conditioned diffusion model that learns both\nrealistic interaction and highly accurate human and object shapes. Experiments\nshow that our HDM trained with ProciGen significantly outperforms prior methods\nthat requires template meshes and that our dataset allows training methods with\nstrong generalization ability to unseen object instances. Our code and data are\nreleased.\n", "date": "2023-12-12", "categories": []}
{"id": "2404.04673", "title": "Neural-ABC: Neural Parametric Models for Articulated Body with Clothes", "abstract": "  In this paper, we introduce Neural-ABC, a novel parametric model based on\nneural implicit functions that can represent clothed human bodies with\ndisentangled latent spaces for identity, clothing, shape, and pose. Traditional\nmesh-based representations struggle to represent articulated bodies with\nclothes due to the diversity of human body shapes and clothing styles, as well\nas the complexity of poses. Our proposed model provides a unified framework for\nparametric modeling, which can represent the identity, clothing, shape and pose\nof the clothed human body. Our proposed approach utilizes the power of neural\nimplicit functions as the underlying representation and integrates\nwell-designed structures to meet the necessary requirements. Specifically, we\nrepresent the underlying body as a signed distance function and clothing as an\nunsigned distance function, and they can be uniformly represented as unsigned\ndistance fields. Different types of clothing do not require predefined\ntopological structures or classifications, and can follow changes in the\nunderlying body to fit the body. Additionally, we construct poses using a\ncontrollable articulated structure. The model is trained on both open and newly\nconstructed datasets, and our decoupling strategy is carefully designed to\nensure optimal performance. Our model excels at disentangling clothing and\nidentity in different shape and poses while preserving the style of the\nclothing. We demonstrate that Neural-ABC fits new observations of different\ntypes of clothing. Compared to other state-of-the-art parametric models,\nNeural-ABC demonstrates powerful advantages in the reconstruction of clothed\nhuman bodies, as evidenced by fitting raw scans, depth maps and images. We show\nthat the attributes of the fitted results can be further edited by adjusting\ntheir identities, clothing, shape and pose codes.\n", "date": "2024-04-06", "categories": []}
{"id": "2404.04819", "title": "Joint Reconstruction of 3D Human and Object via Contact-Based Refinement\n  Transformer", "abstract": "  Human-object contact serves as a strong cue to understand how humans\nphysically interact with objects. Nevertheless, it is not widely explored to\nutilize human-object contact information for the joint reconstruction of 3D\nhuman and object from a single image. In this work, we present a novel joint 3D\nhuman-object reconstruction method (CONTHO) that effectively exploits contact\ninformation between humans and objects. There are two core designs in our\nsystem: 1) 3D-guided contact estimation and 2) contact-based 3D human and\nobject refinement. First, for accurate human-object contact estimation, CONTHO\ninitially reconstructs 3D humans and objects and utilizes them as explicit 3D\nguidance for contact estimation. Second, to refine the initial reconstructions\nof 3D human and object, we propose a novel contact-based refinement Transformer\nthat effectively aggregates human features and object features based on the\nestimated human-object contact. The proposed contact-based refinement prevents\nthe learning of erroneous correlation between human and object, which enables\naccurate 3D reconstruction. As a result, our CONTHO achieves state-of-the-art\nperformance in both human-object contact estimation and joint reconstruction of\n3D human and object. The code is publicly available at\nhttps://github.com/dqj5182/CONTHO_RELEASE.\n", "date": "2024-04-07", "categories": []}
{"id": "2308.04868", "title": "InstantAvatar: Efficient 3D Head Reconstruction via Surface Rendering", "abstract": "  Recent advances in full-head reconstruction have been obtained by optimizing\na neural field through differentiable surface or volume rendering to represent\na single scene. While these techniques achieve an unprecedented accuracy, they\ntake several minutes, or even hours, due to the expensive optimization process\nrequired. In this work, we introduce InstantAvatar, a method that recovers\nfull-head avatars from few images (down to just one) in a few seconds on\ncommodity hardware. In order to speed up the reconstruction process, we propose\na system that combines, for the first time, a voxel-grid neural field\nrepresentation with a surface renderer. Notably, a naive combination of these\ntwo techniques leads to unstable optimizations that do not converge to valid\nsolutions. In order to overcome this limitation, we present a novel statistical\nmodel that learns a prior distribution over 3D head signed distance functions\nusing a voxel-grid based architecture. The use of this prior model, in\ncombination with other design choices, results into a system that achieves 3D\nhead reconstructions with comparable accuracy as the state-of-the-art with a\n100x speed-up.\n", "date": "2023-08-09", "categories": []}
{"id": "2312.09228", "title": "3DGS-Avatar: Animatable Avatars via Deformable 3D Gaussian Splatting", "abstract": "  We introduce an approach that creates animatable human avatars from monocular\nvideos using 3D Gaussian Splatting (3DGS). Existing methods based on neural\nradiance fields (NeRFs) achieve high-quality novel-view/novel-pose image\nsynthesis but often require days of training, and are extremely slow at\ninference time. Recently, the community has explored fast grid structures for\nefficient training of clothed avatars. Albeit being extremely fast at training,\nthese methods can barely achieve an interactive rendering frame rate with\naround 15 FPS. In this paper, we use 3D Gaussian Splatting and learn a\nnon-rigid deformation network to reconstruct animatable clothed human avatars\nthat can be trained within 30 minutes and rendered at real-time frame rates\n(50+ FPS). Given the explicit nature of our representation, we further\nintroduce as-isometric-as-possible regularizations on both the Gaussian mean\nvectors and the covariance matrices, enhancing the generalization of our model\non highly articulated unseen poses. Experimental results show that our method\nachieves comparable and even better performance compared to state-of-the-art\napproaches on animatable avatar creation from a monocular input, while being\n400x and 250x faster in training and inference, respectively.\n", "date": "2023-12-14", "categories": []}
{"id": "2403.02561", "title": "Semantic Human Mesh Reconstruction with Textures", "abstract": "  The field of 3D detailed human mesh reconstruction has made significant\nprogress in recent years. However, current methods still face challenges when\nused in industrial applications due to unstable results, low-quality meshes,\nand a lack of UV unwrapping and skinning weights. In this paper, we present\nSHERT, a novel pipeline that can reconstruct semantic human meshes with\ntextures and high-precision details. SHERT applies semantic- and normal-based\nsampling between the detailed surface (e.g. mesh and SDF) and the corresponding\nSMPL-X model to obtain a partially sampled semantic mesh and then generates the\ncomplete semantic mesh by our specifically designed self-supervised completion\nand refinement networks. Using the complete semantic mesh as a basis, we employ\na texture diffusion model to create human textures that are driven by both\nimages and texts. Our reconstructed meshes have stable UV unwrapping,\nhigh-quality triangle meshes, and consistent semantic information. The given\nSMPL-X model provides semantic information and shape priors, allowing SHERT to\nperform well even with incorrect and incomplete inputs. The semantic\ninformation also makes it easy to substitute and animate different body parts\nsuch as the face, body, and hands. Quantitative and qualitative experiments\ndemonstrate that SHERT is capable of producing high-fidelity and robust\nsemantic meshes that outperform state-of-the-art methods.\n", "date": "2024-03-05", "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": "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": "2401.06415", "title": "3D Reconstruction of Interacting Multi-Person in Clothing from a Single\n  Image", "abstract": "  This paper introduces a novel pipeline to reconstruct the geometry of\ninteracting multi-person in clothing on a globally coherent scene space from a\nsingle image. The main challenge arises from the occlusion: a part of a human\nbody is not visible from a single view due to the occlusion by others or the\nself, which introduces missing geometry and physical implausibility (e.g.,\npenetration). We overcome this challenge by utilizing two human priors for\ncomplete 3D geometry and surface contacts. For the geometry prior, an encoder\nlearns to regress the image of a person with missing body parts to the latent\nvectors; a decoder decodes these vectors to produce 3D features of the\nassociated geometry; and an implicit network combines these features with a\nsurface normal map to reconstruct a complete and detailed 3D humans. For the\ncontact prior, we develop an image-space contact detector that outputs a\nprobability distribution of surface contacts between people in 3D. We use these\npriors to globally refine the body poses, enabling the penetration-free and\naccurate reconstruction of interacting multi-person in clothing on the scene\nspace. The results demonstrate that our method is complete, globally coherent,\nand physically plausible compared to existing methods.\n", "date": "2024-01-12", "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": "2403.16080", "title": "PKU-DyMVHumans: A Multi-View Video Benchmark for High-Fidelity Dynamic\n  Human Modeling", "abstract": "  High-quality human reconstruction and photo-realistic rendering of a dynamic\nscene is a long-standing problem in computer vision and graphics. Despite\nconsiderable efforts invested in developing various capture systems and\nreconstruction algorithms, recent advancements still struggle with loose or\noversized clothing and overly complex poses. In part, this is due to the\nchallenges of acquiring high-quality human datasets. To facilitate the\ndevelopment of these fields, in this paper, we present PKU-DyMVHumans, a\nversatile human-centric dataset for high-fidelity reconstruction and rendering\nof dynamic human scenarios from dense multi-view videos. It comprises 8.2\nmillion frames captured by more than 56 synchronized cameras across diverse\nscenarios. These sequences comprise 32 human subjects across 45 different\nscenarios, each with a high-detailed appearance and realistic human motion.\nInspired by recent advancements in neural radiance field (NeRF)-based scene\nrepresentations, we carefully set up an off-the-shelf framework that is easy to\nprovide those state-of-the-art NeRF-based implementations and benchmark on\nPKU-DyMVHumans dataset. It is paving the way for various applications like\nfine-grained foreground/background decomposition, high-quality human\nreconstruction and photo-realistic novel view synthesis of a dynamic scene.\nExtensive studies are performed on the benchmark, demonstrating new\nobservations and challenges that emerge from using such high-fidelity dynamic\ndata.\n", "date": "2024-03-24", "categories": []}
{"id": "2404.01225", "title": "SurMo: Surface-based 4D Motion Modeling for Dynamic Human Rendering", "abstract": "  Dynamic human rendering from video sequences has achieved remarkable progress\nby formulating the rendering as a mapping from static poses to human images.\nHowever, existing methods focus on the human appearance reconstruction of every\nsingle frame while the temporal motion relations are not fully explored. In\nthis paper, we propose a new 4D motion modeling paradigm, SurMo, that jointly\nmodels the temporal dynamics and human appearances in a unified framework with\nthree key designs: 1) Surface-based motion encoding that models 4D human\nmotions with an efficient compact surface-based triplane. It encodes both\nspatial and temporal motion relations on the dense surface manifold of a\nstatistical body template, which inherits body topology priors for\ngeneralizable novel view synthesis with sparse training observations. 2)\nPhysical motion decoding that is designed to encourage physical motion learning\nby decoding the motion triplane features at timestep t to predict both spatial\nderivatives and temporal derivatives at the next timestep t+1 in the training\nstage. 3) 4D appearance decoding that renders the motion triplanes into images\nby an efficient volumetric surface-conditioned renderer that focuses on the\nrendering of body surfaces with motion learning conditioning. Extensive\nexperiments validate the state-of-the-art performance of our new paradigm and\nillustrate the expressiveness of surface-based motion triplanes for rendering\nhigh-fidelity view-consistent humans with fast motions and even\nmotion-dependent shadows. Our project page is at:\nhttps://taohuumd.github.io/projects/SurMo/\n", "date": "2024-04-01", "categories": []}
{"id": "2404.01424", "title": "DPMesh: Exploiting Diffusion Prior for Occluded Human Mesh Recovery", "abstract": "  The recovery of occluded human meshes presents challenges for current methods\ndue to the difficulty in extracting effective image features under severe\nocclusion. In this paper, we introduce DPMesh, an innovative framework for\noccluded human mesh recovery that capitalizes on the profound diffusion prior\nabout object structure and spatial relationships embedded in a pre-trained\ntext-to-image diffusion model. Unlike previous methods reliant on conventional\nbackbones for vanilla feature extraction, DPMesh seamlessly integrates the\npre-trained denoising U-Net with potent knowledge as its image backbone and\nperforms a single-step inference to provide occlusion-aware information. To\nenhance the perception capability for occluded poses, DPMesh incorporates\nwell-designed guidance via condition injection, which produces effective\ncontrols from 2D observations for the denoising U-Net. Furthermore, we explore\na dedicated noisy key-point reasoning approach to mitigate disturbances arising\nfrom occlusion and crowded scenarios. This strategy fully unleashes the\nperceptual capability of the diffusion prior, thereby enhancing accuracy.\nExtensive experiments affirm the efficacy of our framework, as we outperform\nstate-of-the-art methods on both occlusion-specific and standard datasets. The\npersuasive results underscore its ability to achieve precise and robust 3D\nhuman mesh recovery, particularly in challenging scenarios involving occlusion\nand crowded scenes.\n", "date": "2024-04-01", "categories": []}
{"id": "2311.15672", "title": "HAVE-FUN: Human Avatar Reconstruction from Few-Shot Unconstrained Images", "abstract": "  As for human avatar reconstruction, contemporary techniques commonly\nnecessitate the acquisition of costly data and struggle to achieve satisfactory\nresults from a small number of casual images. In this paper, we investigate\nthis task from a few-shot unconstrained photo album. The reconstruction of\nhuman avatars from such data sources is challenging because of limited data\namount and dynamic articulated poses. For handling dynamic data, we integrate a\nskinning mechanism with deep marching tetrahedra (DMTet) to form a drivable\ntetrahedral representation, which drives arbitrary mesh topologies generated by\nthe DMTet for the adaptation of unconstrained images. To effectively mine\ninstructive information from few-shot data, we devise a two-phase optimization\nmethod with few-shot reference and few-shot guidance. The former focuses on\naligning avatar identity with reference images, while the latter aims to\ngenerate plausible appearances for unseen regions. Overall, our framework,\ncalled HaveFun, can undertake avatar reconstruction, rendering, and animation.\nExtensive experiments on our developed benchmarks demonstrate that HaveFun\nexhibits substantially superior performance in reconstructing the human body\nand hand. Project website: https://seanchenxy.github.io/HaveFunWeb/.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15855", "title": "SiTH: Single-view Textured Human Reconstruction with Image-Conditioned\n  Diffusion", "abstract": "  A long-standing goal of 3D human reconstruction is to create lifelike and\nfully detailed 3D humans from single-view images. The main challenge lies in\ninferring unknown body shapes, appearances, and clothing details in areas not\nvisible in the images. To address this, we propose SiTH, a novel pipeline that\nuniquely integrates an image-conditioned diffusion model into a 3D mesh\nreconstruction workflow. At the core of our method lies the decomposition of\nthe challenging single-view reconstruction problem into generative\nhallucination and reconstruction subproblems. For the former, we employ a\npowerful generative diffusion model to hallucinate unseen back-view appearance\nbased on the input images. For the latter, we leverage skinned body meshes as\nguidance to recover full-body texture meshes from the input and back-view\nimages. SiTH requires as few as 500 3D human scans for training while\nmaintaining its generality and robustness to diverse images. Extensive\nevaluations on two 3D human benchmarks, including our newly created one,\nhighlighted our method's superior accuracy and perceptual quality in 3D\ntextured human reconstruction. Our code and evaluation benchmark are available\nat https://ait.ethz.ch/sith\n", "date": "2023-11-27", "categories": []}
{"id": "2312.02134", "title": "GaussianAvatar: Towards Realistic Human Avatar Modeling from a Single\n  Video via Animatable 3D Gaussians", "abstract": "  We present GaussianAvatar, an efficient approach to creating realistic human\navatars with dynamic 3D appearances from a single video. We start by\nintroducing animatable 3D Gaussians to explicitly represent humans in various\nposes and clothing styles. Such an explicit and animatable representation can\nfuse 3D appearances more efficiently and consistently from 2D observations. Our\nrepresentation is further augmented with dynamic properties to support\npose-dependent appearance modeling, where a dynamic appearance network along\nwith an optimizable feature tensor is designed to learn the\nmotion-to-appearance mapping. Moreover, by leveraging the differentiable motion\ncondition, our method enables a joint optimization of motions and appearances\nduring avatar modeling, which helps to tackle the long-standing issue of\ninaccurate motion estimation in monocular settings. The efficacy of\nGaussianAvatar is validated on both the public dataset and our collected\ndataset, demonstrating its superior performances in terms of appearance quality\nand rendering efficiency.\n", "date": "2023-12-04", "categories": []}
{"id": "2404.00301", "title": "Monocular Identity-Conditioned Facial Reflectance Reconstruction", "abstract": "  Recent 3D face reconstruction methods have made remarkable advancements, yet\nthere remain huge challenges in monocular high-quality facial reflectance\nreconstruction. Existing methods rely on a large amount of light-stage captured\ndata to learn facial reflectance models. However, the lack of subject diversity\nposes challenges in achieving good generalization and widespread applicability.\nIn this paper, we learn the reflectance prior in image space rather than UV\nspace and present a framework named ID2Reflectance. Our framework can directly\nestimate the reflectance maps of a single image while using limited reflectance\ndata for training. Our key insight is that reflectance data shares facial\nstructures with RGB faces, which enables obtaining expressive facial prior from\ninexpensive RGB data thus reducing the dependency on reflectance data. We first\nlearn a high-quality prior for facial reflectance. Specifically, we pretrain\nmulti-domain facial feature codebooks and design a codebook fusion method to\nalign the reflectance and RGB domains. Then, we propose an identity-conditioned\nswapping module that injects facial identity from the target image into the\npre-trained autoencoder to modify the identity of the source reflectance image.\nFinally, we stitch multi-view swapped reflectance images to obtain renderable\nassets. Extensive experiments demonstrate that our method exhibits excellent\ngeneralization capability and achieves state-of-the-art facial reflectance\nreconstruction results for in-the-wild faces. Our project page is\nhttps://xingyuren.github.io/id2reflectance/.\n", "date": "2024-03-30", "categories": []}
{"id": "2404.00485", "title": "DiffHuman: Probabilistic Photorealistic 3D Reconstruction of Humans", "abstract": "  We present DiffHuman, a probabilistic method for photorealistic 3D human\nreconstruction from a single RGB image. Despite the ill-posed nature of this\nproblem, most methods are deterministic and output a single solution, often\nresulting in a lack of geometric detail and blurriness in unseen or uncertain\nregions. In contrast, DiffHuman predicts a probability distribution over 3D\nreconstructions conditioned on an input 2D image, which allows us to sample\nmultiple detailed 3D avatars that are consistent with the image. DiffHuman is\nimplemented as a conditional diffusion model that denoises pixel-aligned 2D\nobservations of an underlying 3D shape representation. During inference, we may\nsample 3D avatars by iteratively denoising 2D renders of the predicted 3D\nrepresentation. Furthermore, we introduce a generator neural network that\napproximates rendering with considerably reduced runtime (55x speed up),\nresulting in a novel dual-branch diffusion framework. Our experiments show that\nDiffHuman can produce diverse and detailed reconstructions for the parts of the\nperson that are unseen or uncertain in the input image, while remaining\ncompetitive with the state-of-the-art when reconstructing visible surfaces.\n", "date": "2024-03-30", "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": "2312.02214", "title": "FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding", "abstract": "  We propose FlashAvatar, a novel and lightweight 3D animatable avatar\nrepresentation that could reconstruct a digital avatar from a short monocular\nvideo sequence in minutes and render high-fidelity photo-realistic images at\n300FPS on a consumer-grade GPU. To achieve this, we maintain a uniform 3D\nGaussian field embedded in the surface of a parametric face model and learn\nextra spatial offset to model non-surface regions and subtle facial details.\nWhile full use of geometric priors can capture high-frequency facial details\nand preserve exaggerated expressions, proper initialization can help reduce the\nnumber of Gaussians, thus enabling super-fast rendering speed. Extensive\nexperimental results demonstrate that FlashAvatar outperforms existing works\nregarding visual quality and personalized details and is almost an order of\nmagnitude faster in rendering speed. Project page:\nhttps://ustc3dv.github.io/FlashAvatar/\n", "date": "2023-12-03", "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": "2311.17113", "title": "Human Gaussian Splatting: Real-time Rendering of Animatable Avatars", "abstract": "  This work addresses the problem of real-time rendering of photorealistic\nhuman body avatars learned from multi-view videos. While the classical\napproaches to model and render virtual humans generally use a textured mesh,\nrecent research has developed neural body representations that achieve\nimpressive visual quality. However, these models are difficult to render in\nreal-time and their quality degrades when the character is animated with body\nposes different than the training observations. We propose an animatable human\nmodel based on 3D Gaussian Splatting, that has recently emerged as a very\nefficient alternative to neural radiance fields. The body is represented by a\nset of gaussian primitives in a canonical space which is deformed with a coarse\nto fine approach that combines forward skinning and local non-rigid refinement.\nWe describe how to learn our Human Gaussian Splatting (HuGS) model in an\nend-to-end fashion from multi-view observations, and evaluate it against the\nstate-of-the-art approaches for novel pose synthesis of clothed body. Our\nmethod achieves 1.5 dB PSNR improvement over the state-of-the-art on THuman4\ndataset while being able to render in real-time (80 fps for 512x512\nresolution).\n", "date": "2023-11-28", "categories": []}
{"id": "2312.02069", "title": "GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians", "abstract": "  We introduce GaussianAvatars, a new method to create photorealistic head\navatars that are fully controllable in terms of expression, pose, and\nviewpoint. The core idea is a dynamic 3D representation based on 3D Gaussian\nsplats that are rigged to a parametric morphable face model. This combination\nfacilitates photorealistic rendering while allowing for precise animation\ncontrol via the underlying parametric model, e.g., through expression transfer\nfrom a driving sequence or by manually changing the morphable model parameters.\nWe parameterize each splat by a local coordinate frame of a triangle and\noptimize for explicit displacement offset to obtain a more accurate geometric\nrepresentation. During avatar reconstruction, we jointly optimize for the\nmorphable model parameters and Gaussian splat parameters in an end-to-end\nfashion. We demonstrate the animation capabilities of our photorealistic avatar\nin several challenging scenarios. For instance, we show reenactments from a\ndriving video, where our method outperforms existing works by a significant\nmargin.\n", "date": "2023-12-04", "categories": []}
{"id": "2403.05218", "title": "3D Face Reconstruction Using A Spectral-Based Graph Convolution Encoder", "abstract": "  Monocular 3D face reconstruction plays a crucial role in avatar generation,\nwith significant demand in web-related applications such as generating virtual\nfinancial advisors in FinTech. Current reconstruction methods predominantly\nrely on deep learning techniques and employ 2D self-supervision as a means to\nguide model learning. However, these methods encounter challenges in capturing\nthe comprehensive 3D structural information of the face due to the utilization\nof 2D images for model training purposes. To overcome this limitation and\nenhance the reconstruction of 3D structural features, we propose an innovative\napproach that integrates existing 2D features with 3D features to guide the\nmodel learning process. Specifically, we introduce the 3D-ID Loss, which\nleverages the high-dimensional structure features extracted from a\nSpectral-Based Graph Convolution Encoder applied to the facial mesh. This\napproach surpasses the sole reliance on the 3D information provided by the\nfacial mesh vertices coordinates. Our model is trained using 2D-3D data pairs\nfrom a combination of datasets and achieves state-of-the-art performance on the\nNoW benchmark.\n", "date": "2024-03-08", "categories": []}
{"id": "2403.17631", "title": "AniArtAvatar: Animatable 3D Art Avatar from a Single Image", "abstract": "  We present a novel approach for generating animatable 3D-aware art avatars\nfrom a single image, with controllable facial expressions, head poses, and\nshoulder movements. Unlike previous reenactment methods, our approach utilizes\na view-conditioned 2D diffusion model to synthesize multi-view images from a\nsingle art portrait with a neutral expression. With the generated colors and\nnormals, we synthesize a static avatar using an SDF-based neural surface. For\navatar animation, we extract control points, transfer the motion with these\npoints, and deform the implicit canonical space. Firstly, we render the front\nimage of the avatar, extract the 2D landmarks, and project them to the 3D space\nusing a trained SDF network. We extract 3D driving landmarks using 3DMM and\ntransfer the motion to the avatar landmarks. To animate the avatar pose, we\nmanually set the body height and bound the head and torso of an avatar with two\ncages. The head and torso can be animated by transforming the two cages. Our\napproach is a one-shot pipeline that can be applied to various styles.\nExperiments demonstrate that our method can generate high-quality 3D art\navatars with desired control over different motions.\n", "date": "2024-03-26", "categories": []}
{"id": "2403.17934", "title": "AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation", "abstract": "  Expressive human pose and shape estimation (a.k.a. 3D whole-body mesh\nrecovery) involves the human body, hand, and expression estimation. Most\nexisting methods have tackled this task in a two-stage manner, first detecting\nthe human body part with an off-the-shelf detection model and inferring the\ndifferent human body parts individually. Despite the impressive results\nachieved, these methods suffer from 1) loss of valuable contextual information\nvia cropping, 2) introducing distractions, and 3) lacking inter-association\namong different persons and body parts, inevitably causing performance\ndegradation, especially for crowded scenes. To address these issues, we\nintroduce a novel all-in-one-stage framework, AiOS, for multiple expressive\nhuman pose and shape recovery without an additional human detection step.\nSpecifically, our method is built upon DETR, which treats multi-person\nwhole-body mesh recovery task as a progressive set prediction problem with\nvarious sequential detection. We devise the decoder tokens and extend them to\nour task. Specifically, we first employ a human token to probe a human location\nin the image and encode global features for each instance, which provides a\ncoarse location for the later transformer block. Then, we introduce a\njoint-related token to probe the human joint in the image and encoder a\nfine-grained local feature, which collaborates with the global feature to\nregress the whole-body mesh. This straightforward but effective model\noutperforms previous state-of-the-art methods by a 9% reduction in NMVE on\nAGORA, a 30% reduction in PVE on EHF, a 10% reduction in PVE on ARCTIC, and a\n3% reduction in PVE on EgoBody.\n", "date": "2024-03-26", "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": "2401.08503", "title": "Real3D-Portrait: One-shot Realistic 3D Talking Portrait Synthesis", "abstract": "  One-shot 3D talking portrait generation aims to reconstruct a 3D avatar from\nan unseen image, and then animate it with a reference video or audio to\ngenerate a talking portrait video. The existing methods fail to simultaneously\nachieve the goals of accurate 3D avatar reconstruction and stable talking face\nanimation. Besides, while the existing works mainly focus on synthesizing the\nhead part, it is also vital to generate natural torso and background segments\nto obtain a realistic talking portrait video. To address these limitations, we\npresent Real3D-Potrait, a framework that (1) improves the one-shot 3D\nreconstruction power with a large image-to-plane model that distills 3D prior\nknowledge from a 3D face generative model; (2) facilitates accurate\nmotion-conditioned animation with an efficient motion adapter; (3) synthesizes\nrealistic video with natural torso movement and switchable background using a\nhead-torso-background super-resolution model; and (4) supports one-shot\naudio-driven talking face generation with a generalizable audio-to-motion\nmodel. Extensive experiments show that Real3D-Portrait generalizes well to\nunseen identities and generates more realistic talking portrait videos compared\nto previous methods. Video samples and source code are available at\nhttps://real3dportrait.github.io .\n", "date": "2024-01-16", "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": "2310.03615", "title": "Animatable Virtual Humans: Learning pose-dependent human representations\n  in UV space for interactive performance synthesis", "abstract": "  We propose a novel representation of virtual humans for highly realistic\nreal-time animation and rendering in 3D applications. We learn pose dependent\nappearance and geometry from highly accurate dynamic mesh sequences obtained\nfrom state-of-the-art multiview-video reconstruction. Learning pose-dependent\nappearance and geometry from mesh sequences poses significant challenges, as it\nrequires the network to learn the intricate shape and articulated motion of a\nhuman body. However, statistical body models like SMPL provide valuable\na-priori knowledge which we leverage in order to constrain the dimension of the\nsearch space enabling more efficient and targeted learning and define\npose-dependency. Instead of directly learning absolute pose-dependent geometry,\nwe learn the difference between the observed geometry and the fitted SMPL\nmodel. This allows us to encode both pose-dependent appearance and geometry in\nthe consistent UV space of the SMPL model. This approach not only ensures a\nhigh level of realism but also facilitates streamlined processing and rendering\nof virtual humans in real-time scenarios.\n", "date": "2023-10-05", "categories": []}
{"id": "2402.16607", "title": "GVA: Reconstructing Vivid 3D Gaussian Avatars from Monocular Videos", "abstract": "  In this paper, we present a novel method that facilitates the creation of\nvivid 3D Gaussian avatars from monocular video inputs (GVA). Our innovation\nlies in addressing the intricate challenges of delivering high-fidelity human\nbody reconstructions and aligning 3D Gaussians with human skin surfaces\naccurately. The key contributions of this paper are twofold. Firstly, we\nintroduce a pose refinement technique to improve hand and foot pose accuracy by\naligning normal maps and silhouettes. Precise pose is crucial for correct shape\nand appearance reconstruction. Secondly, we address the problems of unbalanced\naggregation and initialization bias that previously diminished the quality of\n3D Gaussian avatars, through a novel surface-guided re-initialization method\nthat ensures accurate alignment of 3D Gaussian points with avatar surfaces.\nExperimental results demonstrate that our proposed method achieves\nhigh-fidelity and vivid 3D Gaussian avatar reconstruction. Extensive\nexperimental analyses validate the performance qualitatively and\nquantitatively, demonstrating that it achieves state-of-the-art performance in\nphoto-realistic novel view synthesis while offering fine-grained control over\nthe human body and hand pose. Project page: https://3d-aigc.github.io/GVA/.\n", "date": "2024-02-26", "categories": []}
{"id": "2403.10357", "title": "ANIM: Accurate Neural Implicit Model for Human Reconstruction from a\n  single RGB-D image", "abstract": "  Recent progress in human shape learning, shows that neural implicit models\nare effective in generating 3D human surfaces from limited number of views, and\neven from a single RGB image. However, existing monocular approaches still\nstruggle to recover fine geometric details such as face, hands or cloth\nwrinkles. They are also easily prone to depth ambiguities that result in\ndistorted geometries along the camera optical axis. In this paper, we explore\nthe benefits of incorporating depth observations in the reconstruction process\nby introducing ANIM, a novel method that reconstructs arbitrary 3D human shapes\nfrom single-view RGB-D images with an unprecedented level of accuracy. Our\nmodel learns geometric details from both multi-resolution pixel-aligned and\nvoxel-aligned features to leverage depth information and enable spatial\nrelationships, mitigating depth ambiguities. We further enhance the quality of\nthe reconstructed shape by introducing a depth-supervision strategy, which\nimproves the accuracy of the signed distance field estimation of points that\nlie on the reconstructed surface. Experiments demonstrate that ANIM outperforms\nstate-of-the-art works that use RGB, surface normals, point cloud or RGB-D data\nas input. In addition, we introduce ANIM-Real, a new multi-modal dataset\ncomprising high-quality scans paired with consumer-grade RGB-D camera, and our\nprotocol to fine-tune ANIM, enabling high-quality reconstruction from\nreal-world human capture.\n", "date": "2024-03-15", "categories": []}
{"id": "2403.12473", "title": "PostoMETRO: Pose Token Enhanced Mesh Transformer for Robust 3D Human\n  Mesh Recovery", "abstract": "  With the recent advancements in single-image-based human mesh recovery, there\nis a growing interest in enhancing its performance in certain extreme\nscenarios, such as occlusion, while maintaining overall model accuracy.\nAlthough obtaining accurately annotated 3D human poses under occlusion is\nchallenging, there is still a wealth of rich and precise 2D pose annotations\nthat can be leveraged. However, existing works mostly focus on directly\nleveraging 2D pose coordinates to estimate 3D pose and mesh. In this paper, we\npresent PostoMETRO($\\textbf{Pos}$e $\\textbf{to}$ken enhanced $\\textbf{ME}$sh\n$\\textbf{TR}$ansf$\\textbf{O}$rmer), which integrates occlusion-resilient 2D\npose representation into transformers in a token-wise manner. Utilizing a\nspecialized pose tokenizer, we efficiently condense 2D pose data to a compact\nsequence of pose tokens and feed them to the transformer together with the\nimage tokens. This process not only ensures a rich depiction of texture from\nthe image but also fosters a robust integration of pose and image information.\nSubsequently, these combined tokens are queried by vertex and joint tokens to\ndecode 3D coordinates of mesh vertices and human joints. Facilitated by the\nrobust pose token representation and the effective combination, we are able to\nproduce more precise 3D coordinates, even under extreme scenarios like\nocclusion. Experiments on both standard and occlusion-specific benchmarks\ndemonstrate the effectiveness of PostoMETRO. Qualitative results further\nillustrate the clarity of how 2D pose can help 3D reconstruction. Code will be\nmade available.\n", "date": "2024-03-19", "categories": []}
{"id": "2403.11589", "title": "UV Gaussians: Joint Learning of Mesh Deformation and Gaussian Textures\n  for Human Avatar Modeling", "abstract": "  Reconstructing photo-realistic drivable human avatars from multi-view image\nsequences has been a popular and challenging topic in the field of computer\nvision and graphics. While existing NeRF-based methods can achieve high-quality\nnovel view rendering of human models, both training and inference processes are\ntime-consuming. Recent approaches have utilized 3D Gaussians to represent the\nhuman body, enabling faster training and rendering. However, they undermine the\nimportance of the mesh guidance and directly predict Gaussians in 3D space with\ncoarse mesh guidance. This hinders the learning procedure of the Gaussians and\ntends to produce blurry textures. Therefore, we propose UV Gaussians, which\nmodels the 3D human body by jointly learning mesh deformations and 2D UV-space\nGaussian textures. We utilize the embedding of UV map to learn Gaussian\ntextures in 2D space, leveraging the capabilities of powerful 2D networks to\nextract features. Additionally, through an independent Mesh network, we\noptimize pose-dependent geometric deformations, thereby guiding Gaussian\nrendering and significantly enhancing rendering quality. We collect and process\na new dataset of human motion, which includes multi-view images, scanned\nmodels, parametric model registration, and corresponding texture maps.\nExperimental results demonstrate that our method achieves state-of-the-art\nsynthesis of novel view and novel pose. The code and data will be made\navailable on the homepage https://alex-jyj.github.io/UV-Gaussians/ once the\npaper is accepted.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.11634", "title": "Personalized 3D Human Pose and Shape Refinement", "abstract": "  Recently, regression-based methods have dominated the field of 3D human pose\nand shape estimation. Despite their promising results, a common issue is the\nmisalignment between predictions and image observations, often caused by minor\njoint rotation errors that accumulate along the kinematic chain. To address\nthis issue, we propose to construct dense correspondences between initial human\nmodel estimates and the corresponding images that can be used to refine the\ninitial predictions. To this end, we utilize renderings of the 3D models to\npredict per-pixel 2D displacements between the synthetic renderings and the RGB\nimages. This allows us to effectively integrate and exploit appearance\ninformation of the persons. Our per-pixel displacements can be efficiently\ntransformed to per-visible-vertex displacements and then used for 3D model\nrefinement by minimizing a reprojection loss. To demonstrate the effectiveness\nof our approach, we refine the initial 3D human mesh predictions of multiple\nmodels using different refinement procedures on 3DPW and RICH. We show that our\napproach not only consistently leads to better image-model alignment, but also\nto improved 3D accuracy.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.12028", "title": "Ultraman: Single Image 3D Human Reconstruction with Ultra Speed and\n  Detail", "abstract": "  3D human body reconstruction has been a challenge in the field of computer\nvision. Previous methods are often time-consuming and difficult to capture the\ndetailed appearance of the human body. In this paper, we propose a new method\ncalled \\emph{Ultraman} for fast reconstruction of textured 3D human models from\na single image. Compared to existing techniques, \\emph{Ultraman} greatly\nimproves the reconstruction speed and accuracy while preserving high-quality\ntexture details. We present a set of new frameworks for human reconstruction\nconsisting of three parts, geometric reconstruction, texture generation and\ntexture mapping. Firstly, a mesh reconstruction framework is used, which\naccurately extracts 3D human shapes from a single image. At the same time, we\npropose a method to generate a multi-view consistent image of the human body\nbased on a single image. This is finally combined with a novel texture mapping\nmethod to optimize texture details and ensure color consistency during\nreconstruction. Through extensive experiments and evaluations, we demonstrate\nthe superior performance of \\emph{Ultraman} on various standard datasets. In\naddition, \\emph{Ultraman} outperforms state-of-the-art methods in terms of\nhuman rendering quality and speed. Upon acceptance of the article, we will make\nthe code and data publicly available.\n", "date": "2024-03-18", "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": "2403.10335", "title": "NECA: Neural Customizable Human Avatar", "abstract": "  Human avatar has become a novel type of 3D asset with various applications.\nIdeally, a human avatar should be fully customizable to accommodate different\nsettings and environments. In this work, we introduce NECA, an approach capable\nof learning versatile human representation from monocular or sparse-view\nvideos, enabling granular customization across aspects such as pose, shadow,\nshape, lighting and texture. The core of our approach is to represent humans in\ncomplementary dual spaces and predict disentangled neural fields of geometry,\nalbedo, shadow, as well as an external lighting, from which we are able to\nderive realistic rendering with high-frequency details via volumetric\nrendering. Extensive experiments demonstrate the advantage of our method over\nthe state-of-the-art methods in photorealistic rendering, as well as various\nediting tasks such as novel pose synthesis and relighting. The code is\navailable at https://github.com/iSEE-Laboratory/NECA.\n", "date": "2024-03-15", "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": "2401.12175", "title": "Template-Free Single-View 3D Human Digitalization with Diffusion-Guided\n  LRM", "abstract": "  Reconstructing 3D humans from a single image has been extensively\ninvestigated. However, existing approaches often fall short on capturing fine\ngeometry and appearance details, hallucinating occluded parts with plausible\ndetails, and achieving generalization across unseen and in-the-wild datasets.\nWe present Human-LRM, a diffusion-guided feed-forward model that predicts the\nimplicit field of a human from a single image. Leveraging the power of the\nstate-of-the-art reconstruction model (i.e., LRM) and generative model (i.e\nStable Diffusion), our method is able to capture human without any template\nprior, e.g., SMPL, and effectively enhance occluded parts with rich and\nrealistic details. Our approach first uses a single-view LRM model with an\nenhanced geometry decoder to get the triplane NeRF representation. The novel\nview renderings from the triplane NeRF provide strong geometry and color prior,\nfrom which we generate photo-realistic details for the occluded parts using a\ndiffusion model. The generated multiple views then enable reconstruction with\nhigh-quality geometry and appearance, leading to superior overall performance\ncomparing to all existing human reconstruction methods.\n", "date": "2024-01-22", "categories": []}
{"id": "2403.09063", "title": "Distribution and Depth-Aware Transformers for 3D Human Mesh Recovery", "abstract": "  Precise Human Mesh Recovery (HMR) with in-the-wild data is a formidable\nchallenge and is often hindered by depth ambiguities and reduced precision.\nExisting works resort to either pose priors or multi-modal data such as\nmulti-view or point cloud information, though their methods often overlook the\nvaluable scene-depth information inherently present in a single image.\nMoreover, achieving robust HMR for out-of-distribution (OOD) data is\nexceedingly challenging due to inherent variations in pose, shape and depth.\nConsequently, understanding the underlying distribution becomes a vital\nsubproblem in modeling human forms. Motivated by the need for unambiguous and\nrobust human modeling, we introduce Distribution and depth-aware human mesh\nrecovery (D2A-HMR), an end-to-end transformer architecture meticulously\ndesigned to minimize the disparity between distributions and incorporate\nscene-depth leveraging prior depth information. Our approach demonstrates\nsuperior performance in handling OOD data in certain scenarios while\nconsistently achieving competitive results against state-of-the-art HMR methods\non controlled datasets.\n", "date": "2024-03-14", "categories": []}
{"id": "2403.09623", "title": "Score-Guided Diffusion for 3D Human Recovery", "abstract": "  We present Score-Guided Human Mesh Recovery (ScoreHMR), an approach for\nsolving inverse problems for 3D human pose and shape reconstruction. These\ninverse problems involve fitting a human body model to image observations,\ntraditionally solved through optimization techniques. ScoreHMR mimics model\nfitting approaches, but alignment with the image observation is achieved\nthrough score guidance in the latent space of a diffusion model. The diffusion\nmodel is trained to capture the conditional distribution of the human model\nparameters given an input image. By guiding its denoising process with a\ntask-specific score, ScoreHMR effectively solves inverse problems for various\napplications without the need for retraining the task-agnostic diffusion model.\nWe evaluate our approach on three settings/applications. These are: (i)\nsingle-frame model fitting; (ii) reconstruction from multiple uncalibrated\nviews; (iii) reconstructing humans in video sequences. ScoreHMR consistently\noutperforms all optimization baselines on popular benchmarks across all\nsettings. We make our code and models available at the\nhttps://statho.github.io/ScoreHMR.\n", "date": "2024-03-14", "categories": []}
{"id": "2311.10356", "title": "Garment Recovery with Shape and Deformation Priors", "abstract": "  While modeling people wearing tight-fitting clothing has made great strides\nin recent years, loose-fitting clothing remains a challenge. We propose a\nmethod that delivers realistic garment models from real-world images,\nregardless of garment shape or deformation. To this end, we introduce a fitting\napproach that utilizes shape and deformation priors learned from synthetic data\nto accurately capture garment shapes and deformations, including large ones.\nNot only does our approach recover the garment geometry accurately, it also\nyields models that can be directly used by downstream applications such as\nanimation and simulation.\n", "date": "2023-11-17", "categories": []}
{"id": "2208.10769", "title": "PIFu for the Real World: A Self-supervised Framework to Reconstruct\n  Dressed Human from Single-view Images", "abstract": "  It is very challenging to accurately reconstruct sophisticated human geometry\ncaused by various poses and garments from a single image. Recently, works based\non pixel-aligned implicit function (PIFu) have made a big step and achieved\nstate-of-the-art fidelity on image-based 3D human digitization. However, the\ntraining of PIFu relies heavily on expensive and limited 3D ground truth data\n(i.e. synthetic data), thus hindering its generalization to more diverse real\nworld images. In this work, we propose an end-to-end self-supervised network\nnamed SelfPIFu to utilize abundant and diverse in-the-wild images, resulting in\nlargely improved reconstructions when tested on unconstrained in-the-wild\nimages. At the core of SelfPIFu is the depth-guided volume-/surface-aware\nsigned distance fields (SDF) learning, which enables self-supervised learning\nof a PIFu without access to GT mesh. The whole framework consists of a normal\nestimator, a depth estimator, and a SDF-based PIFu and better utilizes extra\ndepth GT during training. Extensive experiments demonstrate the effectiveness\nof our self-supervised framework and the superiority of using depth as input.\nOn synthetic data, our Intersection-Over-Union (IoU) achieves to 93.5%, 18%\nhigher compared with PIFuHD. For in-the-wild images, we conduct user studies on\nthe reconstructed results, the selection rate of our results is over 68%\ncompared with other state-of-the-art methods.\n", "date": "2022-08-23", "categories": []}
{"id": "2403.05087", "title": "SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded\n  Gaussian Splatting", "abstract": "  We present SplattingAvatar, a hybrid 3D representation of photorealistic\nhuman avatars with Gaussian Splatting embedded on a triangle mesh, which\nrenders over 300 FPS on a modern GPU and 30 FPS on a mobile device. We\ndisentangle the motion and appearance of a virtual human with explicit mesh\ngeometry and implicit appearance modeling with Gaussian Splatting. The\nGaussians are defined by barycentric coordinates and displacement on a triangle\nmesh as Phong surfaces. We extend lifted optimization to simultaneously\noptimize the parameters of the Gaussians while walking on the triangle mesh.\nSplattingAvatar is a hybrid representation of virtual humans where the mesh\nrepresents low-frequency motion and surface deformation, while the Gaussians\ntake over the high-frequency geometry and detailed appearance. Unlike existing\ndeformation methods that rely on an MLP-based linear blend skinning (LBS) field\nfor motion, we control the rotation and translation of the Gaussians directly\nby mesh, which empowers its compatibility with various animation techniques,\ne.g., skeletal animation, blend shapes, and mesh editing. Trainable from\nmonocular videos for both full-body and head avatars, SplattingAvatar shows\nstate-of-the-art rendering quality across multiple datasets.\n", "date": "2024-03-08", "categories": []}
{"id": "2403.01733", "title": "3D Hand Reconstruction via Aggregating Intra and Inter Graphs Guided by\n  Prior Knowledge for Hand-Object Interaction Scenario", "abstract": "  Recently, 3D hand reconstruction has gained more attention in human-computer\ncooperation, especially for hand-object interaction scenario. However, it still\nremains huge challenge due to severe hand-occlusion caused by interaction,\nwhich contain the balance of accuracy and physical plausibility, highly\nnonlinear mapping of model parameters and occlusion feature enhancement. To\novercome these issues, we propose a 3D hand reconstruction network combining\nthe benefits of model-based and model-free approaches to balance accuracy and\nphysical plausibility for hand-object interaction scenario. Firstly, we present\na novel MANO pose parameters regression module from 2D joints directly, which\navoids the process of highly nonlinear mapping from abstract image feature and\nno longer depends on accurate 3D joints. Moreover, we further propose a\nvertex-joint mutual graph-attention model guided by MANO to jointly refine hand\nmeshes and joints, which model the dependencies of vertex-vertex and\njoint-joint and capture the correlation of vertex-joint for aggregating\nintra-graph and inter-graph node features respectively. The experimental\nresults demonstrate that our method achieves a competitive performance on\nrecently benchmark datasets HO3DV2 and Dex-YCB, and outperforms all only\nmodel-base approaches and model-free approaches.\n", "date": "2024-03-04", "categories": []}
{"id": "2402.18032", "title": "Human Shape and Clothing Estimation", "abstract": "  Human shape and clothing estimation has gained significant prominence in\nvarious domains, including online shopping, fashion retail, augmented reality\n(AR), virtual reality (VR), and gaming. The visual representation of human\nshape and clothing has become a focal point for computer vision researchers in\nrecent years. This paper presents a comprehensive survey of the major works in\nthe field, focusing on four key aspects: human shape estimation, fashion\ngeneration, landmark detection, and attribute recognition. For each of these\ntasks, the survey paper examines recent advancements, discusses their strengths\nand limitations, and qualitative differences in approaches and outcomes. By\nexploring the latest developments in human shape and clothing estimation, this\nsurvey aims to provide a comprehensive understanding of the field and inspire\nfuture research in this rapidly evolving domain.\n", "date": "2024-02-28", "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": "2402.17364", "title": "Learning Dynamic Tetrahedra for High-Quality Talking Head Synthesis", "abstract": "  Recent works in implicit representations, such as Neural Radiance Fields\n(NeRF), have advanced the generation of realistic and animatable head avatars\nfrom video sequences. These implicit methods are still confronted by visual\nartifacts and jitters, since the lack of explicit geometric constraints poses a\nfundamental challenge in accurately modeling complex facial deformations. In\nthis paper, we introduce Dynamic Tetrahedra (DynTet), a novel hybrid\nrepresentation that encodes explicit dynamic meshes by neural networks to\nensure geometric consistency across various motions and viewpoints. DynTet is\nparameterized by the coordinate-based networks which learn signed distance,\ndeformation, and material texture, anchoring the training data into a\npredefined tetrahedra grid. Leveraging Marching Tetrahedra, DynTet efficiently\ndecodes textured meshes with a consistent topology, enabling fast rendering\nthrough a differentiable rasterizer and supervision via a pixel loss. To\nenhance training efficiency, we incorporate classical 3D Morphable Models to\nfacilitate geometry learning and define a canonical space for simplifying\ntexture learning. These advantages are readily achievable owing to the\neffective geometric representation employed in DynTet. Compared with prior\nworks, DynTet demonstrates significant improvements in fidelity, lip\nsynchronization, and real-time performance according to various metrics. Beyond\nproducing stable and visually appealing synthesis videos, our method also\noutputs the dynamic meshes which is promising to enable many emerging\napplications.\n", "date": "2024-02-27", "categories": []}
{"id": "2402.16806", "title": "Multi-Human Mesh Recovery with Transformers", "abstract": "  Conventional approaches to human mesh recovery predominantly employ a\nregion-based strategy. This involves initially cropping out a human-centered\nregion as a preprocessing step, with subsequent modeling focused on this\nzoomed-in image. While effective for single figures, this pipeline poses\nchallenges when dealing with images featuring multiple individuals, as\ndifferent people are processed separately, often leading to inaccuracies in\nrelative positioning. Despite the advantages of adopting a whole-image-based\napproach to address this limitation, early efforts in this direction have\nfallen short in performance compared to recent region-based methods. In this\nwork, we advocate for this under-explored area of modeling all people at once,\nemphasizing its potential for improved accuracy in multi-person scenarios\nthrough considering all individuals simultaneously and leveraging the overall\ncontext and interactions. We introduce a new model with a streamlined\ntransformer-based design, featuring three critical design choices: multi-scale\nfeature incorporation, focused attention mechanisms, and relative joint\nsupervision. Our proposed model demonstrates a significant performance\nimprovement, surpassing state-of-the-art region-based and whole-image-based\nmethods on various benchmarks involving multiple individuals.\n", "date": "2024-02-26", "categories": []}
{"id": "2402.11909", "title": "One2Avatar: Generative Implicit Head Avatar For Few-shot User Adaptation", "abstract": "  Traditional methods for constructing high-quality, personalized head avatars\nfrom monocular videos demand extensive face captures and training time, posing\na significant challenge for scalability. This paper introduces a novel approach\nto create high quality head avatar utilizing only a single or a few images per\nuser. We learn a generative model for 3D animatable photo-realistic head avatar\nfrom a multi-view dataset of expressions from 2407 subjects, and leverage it as\na prior for creating personalized avatar from few-shot images. Different from\nprevious 3D-aware face generative models, our prior is built with a\n3DMM-anchored neural radiance field backbone, which we show to be more\neffective for avatar creation through auto-decoding based on few-shot inputs.\nWe also handle unstable 3DMM fitting by jointly optimizing the 3DMM fitting and\ncamera calibration that leads to better few-shot adaptation. Our method\ndemonstrates compelling results and outperforms existing state-of-the-art\nmethods for few-shot avatar adaptation, paving the way for more efficient and\npersonalized avatar creation.\n", "date": "2024-02-19", "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": "2402.03723", "title": "Rig3DGS: Creating Controllable Portraits from Casual Monocular Videos", "abstract": "  Creating controllable 3D human portraits from casual smartphone videos is\nhighly desirable due to their immense value in AR/VR applications. The recent\ndevelopment of 3D Gaussian Splatting (3DGS) has shown improvements in rendering\nquality and training efficiency. However, it still remains a challenge to\naccurately model and disentangle head movements and facial expressions from a\nsingle-view capture to achieve high-quality renderings. In this paper, we\nintroduce Rig3DGS to address this challenge. We represent the entire scene,\nincluding the dynamic subject, using a set of 3D Gaussians in a canonical\nspace. Using a set of control signals, such as head pose and expressions, we\ntransform them to the 3D space with learned deformations to generate the\ndesired rendering. Our key innovation is a carefully designed deformation\nmethod which is guided by a learnable prior derived from a 3D morphable model.\nThis approach is highly efficient in training and effective in controlling\nfacial expressions, head positions, and view synthesis across various captures.\nWe demonstrate the effectiveness of our learned deformation through extensive\nquantitative and qualitative experiments. The project page can be found at\nhttp://shahrukhathar.github.io/2024/02/05/Rig3DGS.html\n", "date": "2024-02-06", "categories": []}
{"id": "2401.02383", "title": "Survey of 3D Human Body Pose and Shape Estimation Methods for\n  Contemporary Dance Applications", "abstract": "  3D human body shape and pose estimation from RGB images is a challenging\nproblem with potential applications in augmented/virtual reality, healthcare\nand fitness technology and virtual retail. Recent solutions have focused on\nthree types of inputs: i) single images, ii) multi-view images and iii) videos.\nIn this study, we surveyed and compared 3D body shape and pose estimation\nmethods for contemporary dance and performing arts, with a special focus on\nhuman body pose and dressing, camera viewpoint, illumination conditions and\nbackground conditions. We demonstrated that multi-frame methods, such as PHALP,\nprovide better results than single-frame method for pose estimation when\ndancers are performing contemporary dances.\n", "date": "2024-01-04", "categories": []}
{"id": "2401.09720", "title": "GaussianBody: Clothed Human Reconstruction via 3d Gaussian Splatting", "abstract": "  In this work, we propose a novel clothed human reconstruction method called\nGaussianBody, based on 3D Gaussian Splatting. Compared with the costly neural\nradiance based models, 3D Gaussian Splatting has recently demonstrated great\nperformance in terms of training time and rendering quality. However, applying\nthe static 3D Gaussian Splatting model to the dynamic human reconstruction\nproblem is non-trivial due to complicated non-rigid deformations and rich cloth\ndetails. To address these challenges, our method considers explicit pose-guided\ndeformation to associate dynamic Gaussians across the canonical space and the\nobservation space, introducing a physically-based prior with regularized\ntransformations helps mitigate ambiguity between the two spaces. During the\ntraining process, we further propose a pose refinement strategy to update the\npose regression for compensating the inaccurate initial estimation and a\nsplit-with-scale mechanism to enhance the density of regressed point clouds.\nThe experiments validate that our method can achieve state-of-the-art\nphotorealistic novel-view rendering results with high-quality details for\ndynamic clothed human bodies, along with explicit geometry reconstruction.\n", "date": "2024-01-18", "categories": []}
{"id": "2401.15348", "title": "AniDress: Animatable Loose-Dressed Avatar from Sparse Views Using\n  Garment Rigging Model", "abstract": "  Recent communities have seen significant progress in building photo-realistic\nanimatable avatars from sparse multi-view videos. However, current workflows\nstruggle to render realistic garment dynamics for loose-fitting characters as\nthey predominantly rely on naked body models for human modeling while leaving\nthe garment part un-modeled. This is mainly due to that the deformations\nyielded by loose garments are highly non-rigid, and capturing such deformations\noften requires dense views as supervision. In this paper, we introduce\nAniDress, a novel method for generating animatable human avatars in loose\nclothes using very sparse multi-view videos (4-8 in our setting). To allow the\ncapturing and appearance learning of loose garments in such a situation, we\nemploy a virtual bone-based garment rigging model obtained from physics-based\nsimulation data. Such a model allows us to capture and render complex garment\ndynamics through a set of low-dimensional bone transformations. Technically, we\ndevelop a novel method for estimating temporal coherent garment dynamics from a\nsparse multi-view video. To build a realistic rendering for unseen garment\nstatus using coarse estimations, a pose-driven deformable neural radiance field\nconditioned on both body and garment motions is introduced, providing explicit\ncontrol of both parts. At test time, the new garment poses can be captured from\nunseen situations, derived from a physics-based or neural network-based\nsimulator to drive unseen garment dynamics. To evaluate our approach, we create\na multi-view dataset that captures loose-dressed performers with diverse\nmotions. Experiments show that our method is able to render natural garment\ndynamics that deviate highly from the body and generalize well to both unseen\nviews and poses, surpassing the performance of existing methods. The code and\ndata will be publicly available.\n", "date": "2024-01-27", "categories": []}
{"id": "2401.14565", "title": "TIFu: Tri-directional Implicit Function for High-Fidelity 3D Character\n  Reconstruction", "abstract": "  Recent advances in implicit function-based approaches have shown promising\nresults in 3D human reconstruction from a single RGB image. However, these\nmethods are not sufficient to extend to more general cases, often generating\ndragged or disconnected body parts, particularly for animated characters. We\nargue that these limitations stem from the use of the existing point-level 3D\nshape representation, which lacks holistic 3D context understanding.\nVoxel-based reconstruction methods are more suitable for capturing the entire\n3D space at once, however, these methods are not practical for high-resolution\nreconstructions due to their excessive memory usage. To address these\nchallenges, we introduce Tri-directional Implicit Function (TIFu), which is a\nvector-level representation that increases global 3D consistencies while\nsignificantly reducing memory usage compared to voxel representations. We also\nintroduce a new algorithm in 3D reconstruction at an arbitrary resolution by\naggregating vectors along three orthogonal axes, resolving inherent problems\nwith regressing fixed dimension of vectors. Our approach achieves\nstate-of-the-art performances in both our self-curated character dataset and\nthe benchmark 3D human dataset. We provide both quantitative and qualitative\nanalyses to support our findings.\n", "date": "2024-01-25", "categories": []}
{"id": "2401.12979", "title": "GALA: Generating Animatable Layered Assets from a Single Scan", "abstract": "  We present GALA, a framework that takes as input a single-layer clothed 3D\nhuman mesh and decomposes it into complete multi-layered 3D assets. The outputs\ncan then be combined with other assets to create novel clothed human avatars\nwith any pose. Existing reconstruction approaches often treat clothed humans as\na single-layer of geometry and overlook the inherent compositionality of humans\nwith hairstyles, clothing, and accessories, thereby limiting the utility of the\nmeshes for downstream applications. Decomposing a single-layer mesh into\nseparate layers is a challenging task because it requires the synthesis of\nplausible geometry and texture for the severely occluded regions. Moreover,\neven with successful decomposition, meshes are not normalized in terms of poses\nand body shapes, failing coherent composition with novel identities and poses.\nTo address these challenges, we propose to leverage the general knowledge of a\npretrained 2D diffusion model as geometry and appearance prior for humans and\nother assets. We first separate the input mesh using the 3D surface\nsegmentation extracted from multi-view 2D segmentations. Then we synthesize the\nmissing geometry of different layers in both posed and canonical spaces using a\nnovel pose-guided Score Distillation Sampling (SDS) loss. Once we complete\ninpainting high-fidelity 3D geometry, we also apply the same SDS loss to its\ntexture to obtain the complete appearance including the initially occluded\nregions. Through a series of decomposition steps, we obtain multiple layers of\n3D assets in a shared canonical space normalized in terms of poses and human\nshapes, hence supporting effortless composition to novel identities and\nreanimation with novel poses. Our experiments demonstrate the effectiveness of\nour approach for decomposition, canonicalization, and composition tasks\ncompared to existing solutions.\n", "date": "2024-01-23", "categories": []}
{"id": "2401.09677", "title": "Eye Motion Matters for 3D Face Reconstruction", "abstract": "  Recent advances in single-image 3D face reconstruction have shown remarkable\nprogress in various applications. Nevertheless, prevailing techniques tend to\nprioritize the global facial contour and expression, often neglecting the\nnuanced dynamics of the eye region. In response, we introduce an Eye Landmark\nAdjustment Module, complemented by a Local Dynamic Loss, designed to capture\nthe dynamic features of the eyes area. Our module allows for flexible\nadjustment of landmarks, resulting in accurate recreation of various eye\nstates. In this paper, we present a comprehensive evaluation of our approach,\nconducting extensive experiments on two datasets. The results underscore the\nsuperior performance of our approach, highlighting its significant\ncontributions in addressing this particular challenge.\n", "date": "2024-01-18", "categories": []}
{"id": "2401.10215", "title": "GPAvatar: Generalizable and Precise Head Avatar from Image(s)", "abstract": "  Head avatar reconstruction, crucial for applications in virtual reality,\nonline meetings, gaming, and film industries, has garnered substantial\nattention within the computer vision community. The fundamental objective of\nthis field is to faithfully recreate the head avatar and precisely control\nexpressions and postures. Existing methods, categorized into 2D-based warping,\nmesh-based, and neural rendering approaches, present challenges in maintaining\nmulti-view consistency, incorporating non-facial information, and generalizing\nto new identities. In this paper, we propose a framework named GPAvatar that\nreconstructs 3D head avatars from one or several images in a single forward\npass. The key idea of this work is to introduce a dynamic point-based\nexpression field driven by a point cloud to precisely and effectively capture\nexpressions. Furthermore, we use a Multi Tri-planes Attention (MTA) fusion\nmodule in the tri-planes canonical field to leverage information from multiple\ninput images. The proposed method achieves faithful identity reconstruction,\nprecise expression control, and multi-view consistency, demonstrating promising\nresults for free-viewpoint rendering and novel view synthesis.\n", "date": "2024-01-18", "categories": []}
{"id": "2401.05334", "title": "URHand: Universal Relightable Hands", "abstract": "  Existing photorealistic relightable hand models require extensive\nidentity-specific observations in different views, poses, and illuminations,\nand face challenges in generalizing to natural illuminations and novel\nidentities. To bridge this gap, we present URHand, the first universal\nrelightable hand model that generalizes across viewpoints, poses,\nilluminations, and identities. Our model allows few-shot personalization using\nimages captured with a mobile phone, and is ready to be photorealistically\nrendered under novel illuminations. To simplify the personalization process\nwhile retaining photorealism, we build a powerful universal relightable prior\nbased on neural relighting from multi-view images of hands captured in a light\nstage with hundreds of identities. The key challenge is scaling the\ncross-identity training while maintaining personalized fidelity and sharp\ndetails without compromising generalization under natural illuminations. To\nthis end, we propose a spatially varying linear lighting model as the neural\nrenderer that takes physics-inspired shading as input feature. By removing\nnon-linear activations and bias, our specifically designed lighting model\nexplicitly keeps the linearity of light transport. This enables single-stage\ntraining from light-stage data while generalizing to real-time rendering under\narbitrary continuous illuminations across diverse identities. In addition, we\nintroduce the joint learning of a physically based model and our neural\nrelighting model, which further improves fidelity and generalization. Extensive\nexperiments show that our approach achieves superior performance over existing\nmethods in terms of both quality and generalizability. We also demonstrate\nquick personalization of URHand from a short phone scan of an unseen identity.\n", "date": "2024-01-10", "categories": []}
{"id": "2401.01730", "title": "STAF: 3D Human Mesh Recovery from Video with Spatio-Temporal Alignment\n  Fusion", "abstract": "  The recovery of 3D human mesh from monocular images has significantly been\ndeveloped in recent years. However, existing models usually ignore spatial and\ntemporal information, which might lead to mesh and image misalignment and\ntemporal discontinuity. For this reason, we propose a novel Spatio-Temporal\nAlignment Fusion (STAF) model. As a video-based model, it leverages coherence\nclues from human motion by an attention-based Temporal Coherence Fusion Module\n(TCFM). As for spatial mesh-alignment evidence, we extract fine-grained local\ninformation through predicted mesh projection on the feature maps. Based on the\nspatial features, we further introduce a multi-stage adjacent Spatial Alignment\nFusion Module (SAFM) to enhance the feature representation of the target frame.\nIn addition to the above, we propose an Average Pooling Module (APM) to allow\nthe model to focus on the entire input sequence rather than just the target\nframe. This method can remarkably improve the smoothness of recovery results\nfrom video. Extensive experiments on 3DPW, MPII3D, and H36M demonstrate the\nsuperiority of STAF. We achieve a state-of-the-art trade-off between precision\nand smoothness. Our code and more video results are on the project page\nhttps://yw0208.github.io/staf/\n", "date": "2024-01-03", "categories": []}
{"id": "2401.00979", "title": "3D Visibility-aware Generalizable Neural Radiance Fields for Interacting\n  Hands", "abstract": "  Neural radiance fields (NeRFs) are promising 3D representations for scenes,\nobjects, and humans. However, most existing methods require multi-view inputs\nand per-scene training, which limits their real-life applications. Moreover,\ncurrent methods focus on single-subject cases, leaving scenes of interacting\nhands that involve severe inter-hand occlusions and challenging view variations\nremain unsolved. To tackle these issues, this paper proposes a generalizable\nvisibility-aware NeRF (VA-NeRF) framework for interacting hands. Specifically,\ngiven an image of interacting hands as input, our VA-NeRF first obtains a\nmesh-based representation of hands and extracts their corresponding geometric\nand textural features. Subsequently, a feature fusion module that exploits the\nvisibility of query points and mesh vertices is introduced to adaptively merge\nfeatures of both hands, enabling the recovery of features in unseen areas.\nAdditionally, our VA-NeRF is optimized together with a novel discriminator\nwithin an adversarial learning paradigm. In contrast to conventional\ndiscriminators that predict a single real/fake label for the synthesized image,\nthe proposed discriminator generates a pixel-wise visibility map, providing\nfine-grained supervision for unseen areas and encouraging the VA-NeRF to\nimprove the visual quality of synthesized images. Experiments on the\nInterhand2.6M dataset demonstrate that our proposed VA-NeRF outperforms\nconventional NeRFs significantly. Project Page:\n\\url{https://github.com/XuanHuang0/VANeRF}.\n", "date": "2024-01-02", "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": "2303.09541", "title": "Diffusion-HPC: Synthetic Data Generation for Human Mesh Recovery in\n  Challenging Domains", "abstract": "  Recent text-to-image generative models have exhibited remarkable abilities in\ngenerating high-fidelity and photo-realistic images. However, despite the\nvisually impressive results, these models often struggle to preserve plausible\nhuman structure in the generations. Due to this reason, while generative models\nhave shown promising results in aiding downstream image recognition tasks by\ngenerating large volumes of synthetic data, they are not suitable for improving\ndownstream human pose perception and understanding. In this work, we propose a\nDiffusion model with Human Pose Correction (Diffusion-HPC), a text-conditioned\nmethod that generates photo-realistic images with plausible posed humans by\ninjecting prior knowledge about human body structure. Our generated images are\naccompanied by 3D meshes that serve as ground truths for improving Human Mesh\nRecovery tasks, where a shortage of 3D training data has long been an issue.\nFurthermore, we show that Diffusion-HPC effectively improves the realism of\nhuman generations under varying conditioning strategies.\n", "date": "2023-03-16", "categories": []}
{"id": "2310.01291", "title": "3DHR-Co: A Collaborative Test-time Refinement Framework for In-the-Wild\n  3D Human-Body Reconstruction Task", "abstract": "  The field of 3D human-body reconstruction (abbreviated as 3DHR) that utilizes\nparametric pose and shape representations has witnessed significant\nadvancements in recent years. However, the application of 3DHR techniques to\nhandle real-world, diverse scenes, known as in-the-wild data, still faces\nlimitations. The primary challenge arises as curating accurate 3D human pose\nground truth (GT) for in-the-wild scenes is still difficult to obtain due to\nvarious factors. Recent test-time refinement approaches on 3DHR leverage\ninitial 2D off-the-shelf human keypoints information to support the lack of 3D\nsupervision on in-the-wild data. However, we observed that additional 2D\nsupervision alone could cause the overfitting issue on common 3DHR backbones,\nmaking the 3DHR test-time refinement task seem intractable. We answer this\nchallenge by proposing a strategy that complements 3DHR test-time refinement\nwork under a collaborative approach. Specifically, we initially apply a\npre-adaptation approach that works by collaborating various 3DHR models in a\nsingle framework to directly improve their initial outputs. This approach is\nthen further combined with the test-time adaptation work under specific\nsettings that minimize the overfitting issue to further boost the 3DHR\nperformance. The whole framework is termed as 3DHR-Co, and on the experiment\nsides, we showed that the proposed work can significantly enhance the scores of\ncommon classic 3DHR backbones up to -34 mm pose error suppression, putting them\namong the top list on the in-the-wild benchmark data. Such achievement shows\nthat our approach helps unveil the true potential of the common classic 3DHR\nbackbones. Based on these findings, we further investigate various settings on\nthe proposed framework to better elaborate the capability of our collaborative\napproach in the 3DHR task.\n", "date": "2023-10-02", "categories": []}
{"id": "2312.16842", "title": "Dynamic Appearance Modeling of Clothed 3D Human Avatars using a Single\n  Camera", "abstract": "  The appearance of a human in clothing is driven not only by the pose but also\nby its temporal context, i.e., motion. However, such context has been largely\nneglected by existing monocular human modeling methods whose neural networks\noften struggle to learn a video of a person with large dynamics due to the\nmotion ambiguity, i.e., there exist numerous geometric configurations of\nclothes that are dependent on the context of motion even for the same pose. In\nthis paper, we introduce a method for high-quality modeling of clothed 3D human\navatars using a video of a person with dynamic movements. The main challenge\ncomes from the lack of 3D ground truth data of geometry and its temporal\ncorrespondences. We address this challenge by introducing a novel compositional\nhuman modeling framework that takes advantage of both explicit and implicit\nhuman modeling. For explicit modeling, a neural network learns to generate\npoint-wise shape residuals and appearance features of a 3D body model by\ncomparing its 2D rendering results and the original images. This explicit model\nallows for the reconstruction of discriminative 3D motion features from UV\nspace by encoding their temporal correspondences. For implicit modeling, an\nimplicit network combines the appearance and 3D motion features to decode\nhigh-fidelity clothed 3D human avatars with motion-dependent geometry and\ntexture. The experiments show that our method can generate a large variation of\nsecondary motion in a physically plausible way.\n", "date": "2023-12-28", "categories": []}
{"id": "2312.17192", "title": "HISR: Hybrid Implicit Surface Representation for Photorealistic 3D Human\n  Reconstruction", "abstract": "  Neural reconstruction and rendering strategies have demonstrated\nstate-of-the-art performances due, in part, to their ability to preserve high\nlevel shape details. Existing approaches, however, either represent objects as\nimplicit surface functions or neural volumes and still struggle to recover\nshapes with heterogeneous materials, in particular human skin, hair or clothes.\nTo this aim, we present a new hybrid implicit surface representation to model\nhuman shapes. This representation is composed of two surface layers that\nrepresent opaque and translucent regions on the clothed human body. We segment\ndifferent regions automatically using visual cues and learn to reconstruct two\nsigned distance functions (SDFs). We perform surface-based rendering on opaque\nregions (e.g., body, face, clothes) to preserve high-fidelity surface normals\nand volume rendering on translucent regions (e.g., hair). Experiments\ndemonstrate that our approach obtains state-of-the-art results on 3D human\nreconstructions, and also shows competitive performances on other objects.\n", "date": "2023-12-28", "categories": []}
{"id": "2306.06406", "title": "D3L: Decomposition of 3D Rotation and Lift from 2D Joint to 3D for Human\n  Mesh Recovery", "abstract": "  Existing methods for 3D human mesh recovery always directly estimate SMPL\nparameters, which involve both joint rotations and shape parameters. However,\nthese methods present rotation semantic ambiguity, rotation error accumulation,\nand shape estimation overfitting, which also leads to errors in the estimated\npose. Additionally, these methods have not efficiently leveraged the\nadvancements in another hot topic, human pose estimation. To address these\nissues, we propose a novel approach, Decomposition of 3D Rotation and Lift from\n2D Joint to 3D mesh (D3L). We disentangle 3D joint rotation into bone direction\nand bone twist direction so that the human mesh recovery task is broken down\ninto estimation of pose, twist, and shape, which can be handled independently.\nThen we design a 2D-to-3D lifting network for estimating twist direction and 3D\njoint position from 2D joint position sequences and introduce a nonlinear\noptimization method for fitting shape parameters and bone directions. Our\napproach can leverage human pose estimation methods, and avoid pose errors\nintroduced by shape estimation overfitting. We conduct experiments on the\nHuman3.6M dataset and demonstrate improved performance compared to existing\nmethods by a large margin.\n", "date": "2023-06-10", "categories": []}
{"id": "2312.15059", "title": "Deformable 3D Gaussian Splatting for Animatable Human Avatars", "abstract": "  Recent advances in neural radiance fields enable novel view synthesis of\nphoto-realistic images in dynamic settings, which can be applied to scenarios\nwith human animation. Commonly used implicit backbones to establish accurate\nmodels, however, require many input views and additional annotations such as\nhuman masks, UV maps and depth maps. In this work, we propose ParDy-Human\n(Parameterized Dynamic Human Avatar), a fully explicit approach to construct a\ndigital avatar from as little as a single monocular sequence. ParDy-Human\nintroduces parameter-driven dynamics into 3D Gaussian Splatting where 3D\nGaussians are deformed by a human pose model to animate the avatar. Our method\nis composed of two parts: A first module that deforms canonical 3D Gaussians\naccording to SMPL vertices and a consecutive module that further takes their\ndesigned joint encodings and predicts per Gaussian deformations to deal with\ndynamics beyond SMPL vertex deformations. Images are then synthesized by a\nrasterizer. ParDy-Human constitutes an explicit model for realistic dynamic\nhuman avatars which requires significantly fewer training views and images. Our\navatars learning is free of additional annotations such as masks and can be\ntrained with variable backgrounds while inferring full-resolution images\nefficiently even on consumer hardware. We provide experimental evidence to show\nthat ParDy-Human outperforms state-of-the-art methods on ZJU-MoCap and\nTHUman4.0 datasets both quantitatively and visually.\n", "date": "2023-12-22", "categories": []}
{"id": "2312.15258", "title": "Human101: Training 100+FPS Human Gaussians in 100s from 1 View", "abstract": "  Reconstructing the human body from single-view videos plays a pivotal role in\nthe virtual reality domain. One prevalent application scenario necessitates the\nrapid reconstruction of high-fidelity 3D digital humans while simultaneously\nensuring real-time rendering and interaction. Existing methods often struggle\nto fulfill both requirements. In this paper, we introduce Human101, a novel\nframework adept at producing high-fidelity dynamic 3D human reconstructions\nfrom 1-view videos by training 3D Gaussians in 100 seconds and rendering in\n100+ FPS. Our method leverages the strengths of 3D Gaussian Splatting, which\nprovides an explicit and efficient representation of 3D humans. Standing apart\nfrom prior NeRF-based pipelines, Human101 ingeniously applies a Human-centric\nForward Gaussian Animation method to deform the parameters of 3D Gaussians,\nthereby enhancing rendering speed (i.e., rendering 1024-resolution images at an\nimpressive 60+ FPS and rendering 512-resolution images at 100+ FPS).\nExperimental results indicate that our approach substantially eclipses current\nmethods, clocking up to a 10 times surge in frames per second and delivering\ncomparable or superior rendering quality. Code and demos will be released at\nhttps://github.com/longxiang-ai/Human101.\n", "date": "2023-12-23", "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": "2312.13091", "title": "MoSAR: Monocular Semi-Supervised Model for Avatar Reconstruction using\n  Differentiable Shading", "abstract": "  Reconstructing an avatar from a portrait image has many applications in\nmultimedia, but remains a challenging research problem. Extracting reflectance\nmaps and geometry from one image is ill-posed: recovering geometry is a\none-to-many mapping problem and reflectance and light are difficult to\ndisentangle. Accurate geometry and reflectance can be captured under the\ncontrolled conditions of a light stage, but it is costly to acquire large\ndatasets in this fashion. Moreover, training solely with this type of data\nleads to poor generalization with in-the-wild images. This motivates the\nintroduction of MoSAR, a method for 3D avatar generation from monocular images.\nWe propose a semi-supervised training scheme that improves generalization by\nlearning from both light stage and in-the-wild datasets. This is achieved using\na novel differentiable shading formulation. We show that our approach\neffectively disentangles the intrinsic face parameters, producing relightable\navatars. As a result, MoSAR estimates a richer set of skin reflectance maps,\nand generates more realistic avatars than existing state-of-the-art methods. We\nalso introduce a new dataset, named FFHQ-UV-Intrinsics, the first public\ndataset providing intrinsic face attributes at scale (diffuse, specular,\nambient occlusion and translucency maps) for a total of 10k subjects. The\nproject website and the dataset are available on the following link:\nhttps://ubisoft-laforge.github.io/character/mosar/\n", "date": "2023-12-20", "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": "2312.13770", "title": "3D Points Splatting for Real-Time Dynamic Hand Reconstruction", "abstract": "  We present 3D Points Splatting Hand Reconstruction (3D-PSHR), a real-time and\nphoto-realistic hand reconstruction approach. We propose a self-adaptive\ncanonical points upsampling strategy to achieve high-resolution hand geometry\nrepresentation. This is followed by a self-adaptive deformation that deforms\nthe hand from the canonical space to the target pose, adapting to the dynamic\nchanging of canonical points which, in contrast to the common practice of\nsubdividing the MANO model, offers greater flexibility and results in improved\ngeometry fitting. To model texture, we disentangle the appearance color into\nthe intrinsic albedo and pose-aware shading, which are learned through a\nContext-Attention module. Moreover, our approach allows the geometric and the\nappearance models to be trained simultaneously in an end-to-end manner. We\ndemonstrate that our method is capable of producing animatable, photorealistic\nand relightable hand reconstructions using multiple datasets, including\nmonocular videos captured with handheld smartphones and large-scale multi-view\nvideos featuring various hand poses. We also demonstrate that our approach\nachieves real-time rendering speeds while simultaneously maintaining superior\nperformance compared to existing state-of-the-art methods.\n", "date": "2023-12-21", "categories": []}
{"id": "2312.12096", "title": "DLCA-Recon: Dynamic Loose Clothing Avatar Reconstruction from Monocular\n  Videos", "abstract": "  Reconstructing a dynamic human with loose clothing is an important but\ndifficult task. To address this challenge, we propose a method named DLCA-Recon\nto create human avatars from monocular videos. The distance from loose clothing\nto the underlying body rapidly changes in every frame when the human freely\nmoves and acts. Previous methods lack effective geometric initialization and\nconstraints for guiding the optimization of deformation to explain this\ndramatic change, resulting in the discontinuous and incomplete reconstruction\nsurface. To model the deformation more accurately, we propose to initialize an\nestimated 3D clothed human in the canonical space, as it is easier for\ndeformation fields to learn from the clothed human than from SMPL. With both\nrepresentations of explicit mesh and implicit SDF, we utilize the physical\nconnection information between consecutive frames and propose a dynamic\ndeformation field (DDF) to optimize deformation fields. DDF accounts for\ncontributive forces on loose clothing to enhance the interpretability of\ndeformations and effectively capture the free movement of loose clothing.\nMoreover, we propagate SMPL skinning weights to each individual and refine pose\nand skinning weights during the optimization to improve skinning\ntransformation. Based on more reasonable initialization and DDF, we can\nsimulate real-world physics more accurately. Extensive experiments on public\nand our own datasets validate that our method can produce superior results for\nhumans with loose clothing compared to the SOTA methods.\n", "date": "2023-12-19", "categories": []}
{"id": "2312.12877", "title": "Relightable and Animatable Neural Avatars from Videos", "abstract": "  Lightweight creation of 3D digital avatars is a highly desirable but\nchallenging task. With only sparse videos of a person under unknown\nillumination, we propose a method to create relightable and animatable neural\navatars, which can be used to synthesize photorealistic images of humans under\nnovel viewpoints, body poses, and lighting. The key challenge here is to\ndisentangle the geometry, material of the clothed body, and lighting, which\nbecomes more difficult due to the complex geometry and shadow changes caused by\nbody motions. To solve this ill-posed problem, we propose novel techniques to\nbetter model the geometry and shadow changes. For geometry change modeling, we\npropose an invertible deformation field, which helps to solve the inverse\nskinning problem and leads to better geometry quality. To model the spatial and\ntemporal varying shading cues, we propose a pose-aware part-wise light\nvisibility network to estimate light occlusion. Extensive experiments on\nsynthetic and real datasets show that our approach reconstructs high-quality\ngeometry and generates realistic shadows under different body poses. Code and\ndata are available at\n\\url{https://wenbin-lin.github.io/RelightableAvatar-page/}.\n", "date": "2023-12-20", "categories": []}
{"id": "2301.06020", "title": "Delving Deep into Pixel Alignment Feature for Accurate Multi-view Human\n  Mesh Recovery", "abstract": "  Regression-based methods have shown high efficiency and effectiveness for\nmulti-view human mesh recovery. The key components of a typical regressor lie\nin the feature extraction of input views and the fusion of multi-view features.\nIn this paper, we present Pixel-aligned Feedback Fusion (PaFF) for accurate yet\nefficient human mesh recovery from multi-view images. PaFF is an iterative\nregression framework that performs feature extraction and fusion alternately.\nAt each iteration, PaFF extracts pixel-aligned feedback features from each\ninput view according to the reprojection of the current estimation and fuses\nthem together with respect to each vertex of the downsampled mesh. In this way,\nour regressor can not only perceive the misalignment status of each view from\nthe feedback features but also correct the mesh parameters more effectively\nbased on the feature fusion on mesh vertices. Additionally, our regressor\ndisentangles the global orientation and translation of the body mesh from the\nestimation of mesh parameters such that the camera parameters of input views\ncan be better utilized in the regression process. The efficacy of our method is\nvalidated in the Human3.6M dataset via comprehensive ablation experiments,\nwhere PaFF achieves 33.02 MPJPE and brings significant improvements over the\nprevious best solutions by more than 29%. The project page with code and video\nresults can be found at https://kairobo.github.io/PaFF/.\n", "date": "2023-01-15", "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": "2312.10346", "title": "MMBaT: A Multi-task Framework for mmWave-based Human Body Reconstruction\n  and Translation Prediction", "abstract": "  Human body reconstruction with Millimeter Wave (mmWave) radar point clouds\nhas gained significant interest due to its ability to work in adverse\nenvironments and its capacity to mitigate privacy concerns associated with\ntraditional camera-based solutions. Despite pioneering efforts in this field,\ntwo challenges persist. Firstly, raw point clouds contain massive noise points,\nusually caused by the ambient objects and multi-path effects of Radio Frequency\n(RF) signals. Recent approaches typically rely on prior knowledge or elaborate\npreprocessing methods, limiting their applicability. Secondly, even after noise\nremoval, the sparse and inconsistent body-related points pose an obstacle to\naccurate human body reconstruction. To address these challenges, we introduce\nmmBaT, a novel multi-task deep learning framework that concurrently estimates\nthe human body and predicts body translations in subsequent frames to extract\nbody-related point clouds. Our method is evaluated on two public datasets that\nare collected with different radar devices and noise levels. A comprehensive\ncomparison against other state-of-the-art methods demonstrates our method has a\nsuperior reconstruction performance and generalization ability from noisy raw\ndata, even when compared to methods provided with body-related point clouds.\n", "date": "2023-12-16", "categories": []}
{"id": "2312.09527", "title": "TIFace: Improving Facial Reconstruction through Tensorial Radiance\n  Fields and Implicit Surfaces", "abstract": "  This report describes the solution that secured the first place in the \"View\nSynthesis Challenge for Human Heads (VSCHH)\" at the ICCV 2023 workshop. Given\nthe sparse view images of human heads, the objective of this challenge is to\nsynthesize images from novel viewpoints. Due to the complexity of textures on\nthe face and the impact of lighting, the baseline method TensoRF yields results\nwith significant artifacts, seriously affecting facial reconstruction. To\naddress this issue, we propose TI-Face, which improves facial reconstruction\nthrough tensorial radiance fields (T-Face) and implicit surfaces (I-Face),\nrespectively. Specifically, we employ an SAM-based approach to obtain the\nforeground mask, thereby filtering out intense lighting in the background.\nAdditionally, we design mask-based constraints and sparsity constraints to\neliminate rendering artifacts effectively. The experimental results demonstrate\nthe effectiveness of the proposed improvements and superior performance of our\nmethod on face reconstruction. The code will be available at\nhttps://github.com/RuijieZhu94/TI-Face.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.08679", "title": "A Local Appearance Model for Volumetric Capture of Diverse Hairstyle", "abstract": "  Hair plays a significant role in personal identity and appearance, making it\nan essential component of high-quality, photorealistic avatars. Existing\napproaches either focus on modeling the facial region only or rely on\npersonalized models, limiting their generalizability and scalability. In this\npaper, we present a novel method for creating high-fidelity avatars with\ndiverse hairstyles. Our method leverages the local similarity across different\nhairstyles and learns a universal hair appearance prior from multi-view\ncaptures of hundreds of people. This prior model takes 3D-aligned features as\ninput and generates dense radiance fields conditioned on a sparse point cloud\nwith color. As our model splits different hairstyles into local primitives and\nbuilds prior at that level, it is capable of handling various hair topologies.\nThrough experiments, we demonstrate that our model captures a diverse range of\nhairstyles and generalizes well to challenging new hairstyles. Empirical\nresults show that our method improves the state-of-the-art approaches in\ncapturing and generating photorealistic, personalized avatars with complete\nhair.\n", "date": "2023-12-14", "categories": []}
{"id": "2312.08863", "title": "HeadRecon: High-Fidelity 3D Head Reconstruction from Monocular Video", "abstract": "  Recently, the reconstruction of high-fidelity 3D head models from static\nportrait image has made great progress. However, most methods require\nmulti-view or multi-illumination information, which therefore put forward high\nrequirements for data acquisition. In this paper, we study the reconstruction\nof high-fidelity 3D head models from arbitrary monocular videos. Non-rigid\nstructure from motion (NRSFM) methods have been widely used to solve such\nproblems according to the two-dimensional correspondence between different\nframes. However, the inaccurate correspondence caused by high-complex hair\nstructures and various facial expression changes would heavily influence the\nreconstruction accuracy. To tackle these problems, we propose a prior-guided\ndynamic implicit neural network. Specifically, we design a two-part dynamic\ndeformation field to transform the current frame space to the canonical one. We\nfurther model the head geometry in the canonical space with a learnable signed\ndistance field (SDF) and optimize it using the volumetric rendering with the\nguidance of two-main head priors to improve the reconstruction accuracy and\nrobustness. Extensive ablation studies and comparisons with state-of-the-art\nmethods demonstrate the effectiveness and robustness of our proposed method.\n", "date": "2023-12-14", "categories": []}
{"id": "2312.07538", "title": "Anatomically Constrained Implicit Face Models", "abstract": "  Coordinate based implicit neural representations have gained rapid popularity\nin recent years as they have been successfully used in image, geometry and\nscene modeling tasks. In this work, we present a novel use case for such\nimplicit representations in the context of learning anatomically constrained\nface models. Actor specific anatomically constrained face models are the state\nof the art in both facial performance capture and performance retargeting.\nDespite their practical success, these anatomical models are slow to evaluate\nand often require extensive data capture to be built. We propose the anatomical\nimplicit face model; an ensemble of implicit neural networks that jointly learn\nto model the facial anatomy and the skin surface with high-fidelity, and can\nreadily be used as a drop in replacement to conventional blendshape models.\nGiven an arbitrary set of skin surface meshes of an actor and only a neutral\nshape with estimated skull and jaw bones, our method can recover a dense\nanatomical substructure which constrains every point on the facial surface. We\ndemonstrate the usefulness of our approach in several tasks ranging from shape\nfitting, shape editing, and performance retargeting.\n", "date": "2023-12-12", "categories": []}
{"id": "2303.10275", "title": "MoRF: Mobile Realistic Fullbody Avatars from a Monocular Video", "abstract": "  We present a system to create Mobile Realistic Fullbody (MoRF) avatars. MoRF\navatars are rendered in real-time on mobile devices, learned from monocular\nvideos, and have high realism. We use SMPL-X as a proxy geometry and render it\nwith DNR (neural texture and image-2-image network). We improve on prior work,\nby overfitting per-frame warping fields in the neural texture space, allowing\nto better align the training signal between different frames. We also refine\nSMPL-X mesh fitting procedure to improve the overall avatar quality. In the\ncomparisons to other monocular video-based avatar systems, MoRF avatars achieve\nhigher image sharpness and temporal consistency. Participants of our user study\nalso preferred avatars generated by MoRF.\n", "date": "2023-03-17", "categories": []}
{"id": "2312.05311", "title": "360{\\deg} Volumetric Portrait Avatar", "abstract": "  We propose 360{\\deg} Volumetric Portrait (3VP) Avatar, a novel method for\nreconstructing 360{\\deg} photo-realistic portrait avatars of human subjects\nsolely based on monocular video inputs. State-of-the-art monocular avatar\nreconstruction methods rely on stable facial performance capturing. However,\nthe common usage of 3DMM-based facial tracking has its limits; side-views can\nhardly be captured and it fails, especially, for back-views, as required inputs\nlike facial landmarks or human parsing masks are missing. This results in\nincomplete avatar reconstructions that only cover the frontal hemisphere. In\ncontrast to this, we propose a template-based tracking of the torso, head and\nfacial expressions which allows us to cover the appearance of a human subject\nfrom all sides. Thus, given a sequence of a subject that is rotating in front\nof a single camera, we train a neural volumetric representation based on neural\nradiance fields. A key challenge to construct this representation is the\nmodeling of appearance changes, especially, in the mouth region (i.e., lips and\nteeth). We, therefore, propose a deformation-field-based blend basis which\nallows us to interpolate between different appearance states. We evaluate our\napproach on captured real-world data and compare against state-of-the-art\nmonocular reconstruction methods. In contrast to those, our method is the first\nmonocular technique that reconstructs an entire 360{\\deg} avatar.\n", "date": "2023-12-08", "categories": []}
{"id": "2312.03461", "title": "HiFi4G: High-Fidelity Human Performance Rendering via Compact Gaussian\n  Splatting", "abstract": "  We have recently seen tremendous progress in photo-real human modeling and\nrendering. Yet, efficiently rendering realistic human performance and\nintegrating it into the rasterization pipeline remains challenging. In this\npaper, we present HiFi4G, an explicit and compact Gaussian-based approach for\nhigh-fidelity human performance rendering from dense footage. Our core\nintuition is to marry the 3D Gaussian representation with non-rigid tracking,\nachieving a compact and compression-friendly representation. We first propose a\ndual-graph mechanism to obtain motion priors, with a coarse deformation graph\nfor effective initialization and a fine-grained Gaussian graph to enforce\nsubsequent constraints. Then, we utilize a 4D Gaussian optimization scheme with\nadaptive spatial-temporal regularizers to effectively balance the non-rigid\nprior and Gaussian updating. We also present a companion compression scheme\nwith residual compensation for immersive experiences on various platforms. It\nachieves a substantial compression rate of approximately 25 times, with less\nthan 2MB of storage per frame. Extensive experiments demonstrate the\neffectiveness of our approach, which significantly outperforms existing\napproaches in terms of optimization speed, rendering quality, and storage\noverhead.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.04558", "title": "MonoGaussianAvatar: Monocular Gaussian Point-based Head Avatar", "abstract": "  The ability to animate photo-realistic head avatars reconstructed from\nmonocular portrait video sequences represents a crucial step in bridging the\ngap between the virtual and real worlds. Recent advancements in head avatar\ntechniques, including explicit 3D morphable meshes (3DMM), point clouds, and\nneural implicit representation have been exploited for this ongoing research.\nHowever, 3DMM-based methods are constrained by their fixed topologies,\npoint-based approaches suffer from a heavy training burden due to the extensive\nquantity of points involved, and the last ones suffer from limitations in\ndeformation flexibility and rendering efficiency. In response to these\nchallenges, we propose MonoGaussianAvatar (Monocular Gaussian Point-based Head\nAvatar), a novel approach that harnesses 3D Gaussian point representation\ncoupled with a Gaussian deformation field to learn explicit head avatars from\nmonocular portrait videos. We define our head avatars with Gaussian points\ncharacterized by adaptable shapes, enabling flexible topology. These points\nexhibit movement with a Gaussian deformation field in alignment with the target\npose and expression of a person, facilitating efficient deformation.\nAdditionally, the Gaussian points have controllable shape, size, color, and\nopacity combined with Gaussian splatting, allowing for efficient training and\nrendering. Experiments demonstrate the superior performance of our method,\nwhich achieves state-of-the-art results among previous methods.\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": "2312.02963", "title": "MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human\n  Captures", "abstract": "  In this era, the success of large language models and text-to-image models\ncan be attributed to the driving force of large-scale datasets. However, in the\nrealm of 3D vision, while remarkable progress has been made with models trained\non large-scale synthetic and real-captured object data like Objaverse and\nMVImgNet, a similar level of progress has not been observed in the domain of\nhuman-centric tasks partially due to the lack of a large-scale human dataset.\nExisting datasets of high-fidelity 3D human capture continue to be mid-sized\ndue to the significant challenges in acquiring large-scale high-quality 3D\nhuman data. To bridge this gap, we present MVHumanNet, a dataset that comprises\nmulti-view human action sequences of 4,500 human identities. The primary focus\nof our work is on collecting human data that features a large number of diverse\nidentities and everyday clothing using a multi-view human capture system, which\nfacilitates easily scalable data collection. Our dataset contains 9,000 daily\noutfits, 60,000 motion sequences and 645 million frames with extensive\nannotations, including human masks, camera parameters, 2D and 3D keypoints,\nSMPL/SMPLX parameters, and corresponding textual descriptions. To explore the\npotential of MVHumanNet in various 2D and 3D visual tasks, we conducted pilot\nstudies on view-consistent action recognition, human NeRF reconstruction,\ntext-driven view-unconstrained human image generation, as well as 2D\nview-unconstrained human image and 3D avatar generation. Extensive experiments\ndemonstrate the performance improvements and effective applications enabled by\nthe scale provided by MVHumanNet. As the current largest-scale 3D human\ndataset, we hope that the release of MVHumanNet data with annotations will\nfoster further innovations in the domain of 3D human-centric tasks at scale.\n", "date": "2023-12-05", "categories": []}
{"id": "2312.02973", "title": "GauHuman: Articulated Gaussian Splatting from Monocular Human Videos", "abstract": "  We present, GauHuman, a 3D human model with Gaussian Splatting for both fast\ntraining (1 ~ 2 minutes) and real-time rendering (up to 189 FPS), compared with\nexisting NeRF-based implicit representation modelling frameworks demanding\nhours of training and seconds of rendering per frame. Specifically, GauHuman\nencodes Gaussian Splatting in the canonical space and transforms 3D Gaussians\nfrom canonical space to posed space with linear blend skinning (LBS), in which\neffective pose and LBS refinement modules are designed to learn fine details of\n3D humans under negligible computational cost. Moreover, to enable fast\noptimization of GauHuman, we initialize and prune 3D Gaussians with 3D human\nprior, while splitting/cloning via KL divergence guidance, along with a novel\nmerge operation for further speeding up. Extensive experiments on ZJU_Mocap and\nMonoCap datasets demonstrate that GauHuman achieves state-of-the-art\nperformance quantitatively and qualitatively with fast training and real-time\nrendering speed. Notably, without sacrificing rendering quality, GauHuman can\nfast model the 3D human performer with ~13k 3D Gaussians.\n", "date": "2023-12-05", "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": "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": "2311.18448", "title": "HOLD: Category-agnostic 3D Reconstruction of Interacting Hands and\n  Objects from Video", "abstract": "  Since humans interact with diverse objects every day, the holistic 3D capture\nof these interactions is important to understand and model human behaviour.\nHowever, most existing methods for hand-object reconstruction from RGB either\nassume pre-scanned object templates or heavily rely on limited 3D hand-object\ndata, restricting their ability to scale and generalize to more unconstrained\ninteraction settings. To this end, we introduce HOLD -- the first\ncategory-agnostic method that reconstructs an articulated hand and object\njointly from a monocular interaction video. We develop a compositional\narticulated implicit model that can reconstruct disentangled 3D hand and object\nfrom 2D images. We also further incorporate hand-object constraints to improve\nhand-object poses and consequently the reconstruction quality. Our method does\nnot rely on 3D hand-object annotations while outperforming fully-supervised\nbaselines in both in-the-lab and challenging in-the-wild settings. Moreover, we\nqualitatively show its robustness in reconstructing from in-the-wild videos.\nCode: https://github.com/zc-alexfan/hold\n", "date": "2023-11-30", "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": "2311.17123", "title": "ConTex-Human: Free-View Rendering of Human from a Single Image with\n  Texture-Consistent Synthesis", "abstract": "  In this work, we propose a method to address the challenge of rendering a 3D\nhuman from a single image in a free-view manner. Some existing approaches could\nachieve this by using generalizable pixel-aligned implicit fields to\nreconstruct a textured mesh of a human or by employing a 2D diffusion model as\nguidance with the Score Distillation Sampling (SDS) method, to lift the 2D\nimage into 3D space. However, a generalizable implicit field often results in\nan over-smooth texture field, while the SDS method tends to lead to a\ntexture-inconsistent novel view with the input image. In this paper, we\nintroduce a texture-consistent back view synthesis module that could transfer\nthe reference image content to the back view through depth and text-guided\nattention injection. Moreover, to alleviate the color distortion that occurs in\nthe side region, we propose a visibility-aware patch consistency regularization\nfor texture mapping and refinement combined with the synthesized back view\ntexture. With the above techniques, we could achieve high-fidelity and\ntexture-consistent human rendering from a single image. Experiments conducted\non both real and synthetic data demonstrate the effectiveness of our method and\nshow that our approach outperforms previous baseline methods.\n", "date": "2023-11-28", "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": "2311.17910", "title": "HUGS: Human Gaussian Splats", "abstract": "  Recent advances in neural rendering have improved both training and rendering\ntimes by orders of magnitude. While these methods demonstrate state-of-the-art\nquality and speed, they are designed for photogrammetry of static scenes and do\nnot generalize well to freely moving humans in the environment. In this work,\nwe introduce Human Gaussian Splats (HUGS) that represents an animatable human\ntogether with the scene using 3D Gaussian Splatting (3DGS). Our method takes\nonly a monocular video with a small number of (50-100) frames, and it\nautomatically learns to disentangle the static scene and a fully animatable\nhuman avatar within 30 minutes. We utilize the SMPL body model to initialize\nthe human Gaussians. To capture details that are not modeled by SMPL (e.g.\ncloth, hairs), we allow the 3D Gaussians to deviate from the human body model.\nUtilizing 3D Gaussians for animated humans brings new challenges, including the\nartifacts created when articulating the Gaussians. We propose to jointly\noptimize the linear blend skinning weights to coordinate the movements of\nindividual Gaussians during animation. Our approach enables novel-pose\nsynthesis of human and novel view synthesis of both the human and the scene. We\nachieve state-of-the-art rendering quality with a rendering speed of 60 FPS\nwhile being ~100x faster to train over previous work. Our code will be\nannounced here: https://github.com/apple/ml-hugs\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": "2311.05521", "title": "BakedAvatar: Baking Neural Fields for Real-Time Head Avatar Synthesis", "abstract": "  Synthesizing photorealistic 4D human head avatars from videos is essential\nfor VR/AR, telepresence, and video game applications. Although existing Neural\nRadiance Fields (NeRF)-based methods achieve high-fidelity results, the\ncomputational expense limits their use in real-time applications. To overcome\nthis limitation, we introduce BakedAvatar, a novel representation for real-time\nneural head avatar synthesis, deployable in a standard polygon rasterization\npipeline. Our approach extracts deformable multi-layer meshes from learned\nisosurfaces of the head and computes expression-, pose-, and view-dependent\nappearances that can be baked into static textures for efficient rasterization.\nWe thus propose a three-stage pipeline for neural head avatar synthesis, which\nincludes learning continuous deformation, manifold, and radiance fields,\nextracting layered meshes and textures, and fine-tuning texture details with\ndifferential rasterization. Experimental results demonstrate that our\nrepresentation generates synthesis results of comparable quality to other\nstate-of-the-art methods while significantly reducing the inference time\nrequired. We further showcase various head avatar synthesis results from\nmonocular videos, including view synthesis, face reenactment, expression\nediting, and pose editing, all at interactive frame rates.\n", "date": "2023-11-09", "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": "2303.06226", "title": "NeRFlame: FLAME-based conditioning of NeRF for 3D face rendering", "abstract": "  Traditional 3D face models are based on mesh representations with texture.\nOne of the most important models is FLAME (Faces Learned with an Articulated\nModel and Expressions), which produces meshes of human faces that are fully\ncontrollable. Unfortunately, such models have problems with capturing geometric\nand appearance details. In contrast to mesh representation, the neural radiance\nfield (NeRF) produces extremely sharp renders. However, implicit methods are\nhard to animate and do not generalize well to unseen expressions. It is not\ntrivial to effectively control NeRF models to obtain face manipulation.\n  The present paper proposes a novel approach, named NeRFlame, which combines\nthe strengths of both NeRF and FLAME methods. Our method enables high-quality\nrendering capabilities of NeRF while also offering complete control over the\nvisual appearance, similar to FLAME. In contrast to traditional NeRF-based\nstructures that use neural networks for RGB color and volume density modeling,\nour approach utilizes the FLAME mesh as a distinct density volume.\nConsequently, color values exist only in the vicinity of the FLAME mesh. This\nFLAME framework is seamlessly incorporated into the NeRF architecture for\npredicting RGB colors, enabling our model to explicitly represent volume\ndensity and implicitly capture RGB colors.\n", "date": "2023-03-10", "categories": []}
{"id": "2311.13404", "title": "Animatable 3D Gaussians for High-fidelity Synthesis of Human Motions", "abstract": "  We present a novel animatable 3D Gaussian model for rendering high-fidelity\nfree-view human motions in real time. Compared to existing NeRF-based methods,\nthe model owns better capability in synthesizing high-frequency details without\nthe jittering problem across video frames. The core of our model is a novel\naugmented 3D Gaussian representation, which attaches each Gaussian with a\nlearnable code. The learnable code serves as a pose-dependent appearance\nembedding for refining the erroneous appearance caused by geometric\ntransformation of Gaussians, based on which an appearance refinement model is\nlearned to produce residual Gaussian properties to match the appearance in\ntarget pose. To force the Gaussians to learn the foreground human only without\nbackground interference, we further design a novel alpha loss to explicitly\nconstrain the Gaussians within the human body. We also propose to jointly\noptimize the human joint parameters to improve the appearance accuracy. The\nanimatable 3D Gaussian model can be learned with shallow MLPs, so new human\nmotions can be synthesized in real time (66 fps on avarage). Experiments show\nthat our model has superior performance over NeRF-based methods.\n", "date": "2023-11-22", "categories": []}
{"id": "2311.15171", "title": "HumanRecon: Neural Reconstruction of Dynamic Human Using Geometric Cues\n  and Physical Priors", "abstract": "  Recent methods for dynamic human reconstruction have attained promising\nreconstruction results. Most of these methods rely only on RGB color\nsupervision without considering explicit geometric constraints. This leads to\nexisting human reconstruction techniques being more prone to overfitting to\ncolor and causes geometrically inherent ambiguities, especially in the sparse\nmulti-view setup.\n  Motivated by recent advances in the field of monocular geometry prediction,\nwe consider the geometric constraints of estimated depth and normals in the\nlearning of neural implicit representation for dynamic human reconstruction. As\na geometric regularization, this provides reliable yet explicit supervision\ninformation, and improves reconstruction quality. We also exploit several\nbeneficial physical priors, such as adding noise into view direction and\nmaximizing the density on the human surface. These priors ensure the color\nrendered along rays to be robust to view direction and reduce the inherent\nambiguities of density estimated along rays. Experimental results demonstrate\nthat depth and normal cues, predicted by human-specific monocular estimators,\ncan provide effective supervision signals and render more accurate images.\nFinally, we also show that the proposed physical priors significantly reduce\noverfitting and improve the overall quality of novel view synthesis. Our code\nis available\nat:~\\href{https://github.com/PRIS-CV/HumanRecon}{https://github.com/PRIS-CV/HumanRecon}.\n", "date": "2023-11-26", "categories": []}
{"id": "2311.16042", "title": "Weakly-Supervised 3D Reconstruction of Clothed Humans via Normal Maps", "abstract": "  We present a novel deep learning-based approach to the 3D reconstruction of\nclothed humans using weak supervision via 2D normal maps. Given a single RGB\nimage or multiview images, our network infers a signed distance function (SDF)\ndiscretized on a tetrahedral mesh surrounding the body in a rest pose.\nSubsequently, inferred pose and camera parameters are used to generate a normal\nmap from the SDF. A key aspect of our approach is the use of Marching\nTetrahedra to (uniquely) compute a triangulated surface from the SDF on the\ntetrahedral mesh, facilitating straightforward differentiation (and thus\nbackpropagation). Thus, given only ground truth normal maps (with no volumetric\ninformation ground truth information), we can train the network to produce SDF\nvalues from corresponding RGB images. Optionally, an additional multiview loss\nleads to improved results. We demonstrate the efficacy of our approach for both\nnetwork inference and 3D reconstruction.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.16099", "title": "GART: Gaussian Articulated Template Models", "abstract": "  We introduce Gaussian Articulated Template Model GART, an explicit,\nefficient, and expressive representation for non-rigid articulated subject\ncapturing and rendering from monocular videos. GART utilizes a mixture of\nmoving 3D Gaussians to explicitly approximate a deformable subject's geometry\nand appearance. It takes advantage of a categorical template model prior (SMPL,\nSMAL, etc.) with learnable forward skinning while further generalizing to more\ncomplex non-rigid deformations with novel latent bones. GART can be\nreconstructed via differentiable rendering from monocular videos in seconds or\nminutes and rendered in novel poses faster than 150fps.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.13655", "title": "GAN-Avatar: Controllable Personalized GAN-based Human Head Avatar", "abstract": "  Digital humans and, especially, 3D facial avatars have raised a lot of\nattention in the past years, as they are the backbone of several applications\nlike immersive telepresence in AR or VR. Despite the progress, facial avatars\nreconstructed from commodity hardware are incomplete and miss out on parts of\nthe side and back of the head, severely limiting the usability of the avatar.\nThis limitation in prior work stems from their requirement of face tracking,\nwhich fails for profile and back views. To address this issue, we propose to\nlearn person-specific animatable avatars from images without assuming to have\naccess to precise facial expression tracking. At the core of our method, we\nleverage a 3D-aware generative model that is trained to reproduce the\ndistribution of facial expressions from the training data. To train this\nappearance model, we only assume to have a collection of 2D images with the\ncorresponding camera parameters. For controlling the model, we learn a mapping\nfrom 3DMM facial expression parameters to the latent space of the generative\nmodel. This mapping can be learned by sampling the latent space of the\nappearance model and reconstructing the facial parameters from a normalized\nfrontal view, where facial expression estimation performs well. With this\nscheme, we decouple 3D appearance reconstruction and animation control to\nachieve high fidelity in image synthesis. In a series of experiments, we\ncompare our proposed technique to state-of-the-art monocular methods and show\nsuperior quality while not requiring expression tracking of the training data.\n", "date": "2023-11-22", "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": "2309.13101", "title": "Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene\n  Reconstruction", "abstract": "  Implicit neural representation has paved the way for new approaches to\ndynamic scene reconstruction and rendering. Nonetheless, cutting-edge dynamic\nneural rendering methods rely heavily on these implicit representations, which\nfrequently struggle to capture the intricate details of objects in the scene.\nFurthermore, implicit methods have difficulty achieving real-time rendering in\ngeneral dynamic scenes, limiting their use in a variety of tasks. To address\nthe issues, we propose a deformable 3D Gaussians Splatting method that\nreconstructs scenes using 3D Gaussians and learns them in canonical space with\na deformation field to model monocular dynamic scenes. We also introduce an\nannealing smoothing training mechanism with no extra overhead, which can\nmitigate the impact of inaccurate poses on the smoothness of time interpolation\ntasks in real-world datasets. Through a differential Gaussian rasterizer, the\ndeformable 3D Gaussians not only achieve higher rendering quality but also\nreal-time rendering speed. Experiments show that our method outperforms\nexisting methods significantly in terms of both rendering quality and speed,\nmaking it well-suited for tasks such as novel-view synthesis, time\ninterpolation, and real-time rendering.\n", "date": "2023-09-22", "categories": []}
{"id": "2311.10812", "title": "SplatArmor: Articulated Gaussian splatting for animatable humans from\n  monocular RGB videos", "abstract": "  We propose SplatArmor, a novel approach for recovering detailed and\nanimatable human models by `armoring' a parameterized body model with 3D\nGaussians. Our approach represents the human as a set of 3D Gaussians within a\ncanonical space, whose articulation is defined by extending the skinning of the\nunderlying SMPL geometry to arbitrary locations in the canonical space. To\naccount for pose-dependent effects, we introduce a SE(3) field, which allows us\nto capture both the location and anisotropy of the Gaussians. Furthermore, we\npropose the use of a neural color field to provide color regularization and 3D\nsupervision for the precise positioning of these Gaussians. We show that\nGaussian splatting provides an interesting alternative to neural rendering\nbased methods by leverging a rasterization primitive without facing any of the\nnon-differentiability and optimization challenges typically faced in such\napproaches. The rasterization paradigms allows us to leverage forward skinning,\nand does not suffer from the ambiguities associated with inverse skinning and\nwarping. We show compelling results on the ZJU MoCap and People Snapshot\ndatasets, which underscore the effectiveness of our method for controllable\nhuman synthesis.\n", "date": "2023-11-17", "categories": []}
{"id": "2311.11614", "title": "Semantic-Preserved Point-based Human Avatar", "abstract": "  To enable realistic experience in AR/VR and digital entertainment, we present\nthe first point-based human avatar model that embodies the entirety expressive\nrange of digital humans. We employ two MLPs to model pose-dependent deformation\nand linear skinning (LBS) weights. The representation of appearance relies on a\ndecoder and the features that attached to each point. In contrast to\nalternative implicit approaches, the oriented points representation not only\nprovides a more intuitive way to model human avatar animation but also\nsignificantly reduces both training and inference time. Moreover, we propose a\nnovel method to transfer semantic information from the SMPL-X model to the\npoints, which enables to better understand human body movements. By leveraging\nthe semantic information of points, we can facilitate virtual try-on and human\navatar composition through exchanging the points of same category across\ndifferent subjects. Experimental results demonstrate the efficacy of our\npresented method.\n", "date": "2023-11-20", "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": "2311.06443", "title": "CVTHead: One-shot Controllable Head Avatar with Vertex-feature\n  Transformer", "abstract": "  Reconstructing personalized animatable head avatars has significant\nimplications in the fields of AR/VR. Existing methods for achieving explicit\nface control of 3D Morphable Models (3DMM) typically rely on multi-view images\nor videos of a single subject, making the reconstruction process complex.\nAdditionally, the traditional rendering pipeline is time-consuming, limiting\nreal-time animation possibilities. In this paper, we introduce CVTHead, a novel\napproach that generates controllable neural head avatars from a single\nreference image using point-based neural rendering. CVTHead considers the\nsparse vertices of mesh as the point set and employs the proposed\nVertex-feature Transformer to learn local feature descriptors for each vertex.\nThis enables the modeling of long-range dependencies among all the vertices.\nExperimental results on the VoxCeleb dataset demonstrate that CVTHead achieves\ncomparable performance to state-of-the-art graphics-based methods. Moreover, it\nenables efficient rendering of novel human heads with various expressions, head\nposes, and camera views. These attributes can be explicitly controlled using\nthe coefficients of 3DMMs, facilitating versatile and realistic animation in\nreal-time scenarios.\n", "date": "2023-11-11", "categories": []}
{"id": "2307.13639", "title": "Fake It Without Making It: Conditioned Face Generation for Accurate 3D\n  Face Reconstruction", "abstract": "  Accurate 3D face reconstruction from 2D images is an enabling technology with\napplications in healthcare, security, and creative industries. However, current\nstate-of-the-art methods either rely on supervised training with very limited\n3D data or self-supervised training with 2D image data. To bridge this gap, we\npresent a method to generate a large-scale synthesised dataset of 250K\nphotorealistic images and their corresponding shape parameters and depth maps,\nwhich we call SynthFace. Our synthesis method conditions Stable Diffusion on\ndepth maps sampled from the FLAME 3D Morphable Model (3DMM) of the human face,\nallowing us to generate a diverse set of shape-consistent facial images that is\ndesigned to be balanced in race and gender. We further propose ControlFace, a\ndeep neural network, trained on SynthFace, which achieves competitive\nperformance on the NoW benchmark, without requiring 3D supervision or manual 3D\nasset creation. The complete SynthFace dataset will be made publicly available\nupon publication.\n", "date": "2023-07-25", "categories": []}
{"id": "2311.02461", "title": "SPHEAR: Spherical Head Registration for Complete Statistical 3D Modeling", "abstract": "  We present \\emph{SPHEAR}, an accurate, differentiable parametric statistical\n3D human head model, enabled by a novel 3D registration method based on\nspherical embeddings. We shift the paradigm away from the classical Non-Rigid\nRegistration methods, which operate under various surface priors, increasing\nreconstruction fidelity and minimizing required human intervention.\nAdditionally, SPHEAR is a \\emph{complete} model that allows not only to sample\ndiverse synthetic head shapes and facial expressions, but also gaze directions,\nhigh-resolution color textures, surface normal maps, and hair cuts represented\nin detail, as strands. SPHEAR can be used for automatic realistic visual data\ngeneration, semantic annotation, and general reconstruction tasks. Compared to\nstate-of-the-art approaches, our components are fast and memory efficient, and\nexperiments support the validity of our design choices and the accuracy of\nregistration, reconstruction and generation techniques.\n", "date": "2023-11-04", "categories": []}
{"id": "2311.01214", "title": "High-Quality Animatable Dynamic Garment Reconstruction from Monocular\n  Videos", "abstract": "  Much progress has been made in reconstructing garments from an image or a\nvideo. However, none of existing works meet the expectations of digitizing\nhigh-quality animatable dynamic garments that can be adjusted to various unseen\nposes. In this paper, we propose the first method to recover high-quality\nanimatable dynamic garments from monocular videos without depending on scanned\ndata. To generate reasonable deformations for various unseen poses, we propose\na learnable garment deformation network that formulates the garment\nreconstruction task as a pose-driven deformation problem. To alleviate the\nambiguity estimating 3D garments from monocular videos, we design a\nmulti-hypothesis deformation module that learns spatial representations of\nmultiple plausible deformations. Experimental results on several public\ndatasets demonstrate that our method can reconstruct high-quality dynamic\ngarments with coherent surface details, which can be easily animated under\nunseen poses. The code will be provided for research purposes.\n", "date": "2023-11-02", "categories": []}
{"id": "2311.00457", "title": "Single-view 3D Scene Reconstruction with High-fidelity Shape and Texture", "abstract": "  Reconstructing detailed 3D scenes from single-view images remains a\nchallenging task due to limitations in existing approaches, which primarily\nfocus on geometric shape recovery, overlooking object appearances and fine\nshape details. To address these challenges, we propose a novel framework for\nsimultaneous high-fidelity recovery of object shapes and textures from\nsingle-view images. Our approach utilizes the proposed Single-view neural\nimplicit Shape and Radiance field (SSR) representations to leverage both\nexplicit 3D shape supervision and volume rendering of color, depth, and surface\nnormal images. To overcome shape-appearance ambiguity under partial\nobservations, we introduce a two-stage learning curriculum incorporating both\n3D and 2D supervisions. A distinctive feature of our framework is its ability\nto generate fine-grained textured meshes while seamlessly integrating rendering\ncapabilities into the single-view 3D reconstruction model. This integration\nenables not only improved textured 3D object reconstruction by 27.7% and 11.6%\non the 3D-FRONT and Pix3D datasets, respectively, but also supports the\nrendering of images from novel viewpoints. Beyond individual objects, our\napproach facilitates composing object-level representations into flexible scene\nrepresentations, thereby enabling applications such as holistic scene\nunderstanding and 3D scene editing. We conduct extensive experiments to\ndemonstrate the effectiveness of our method.\n", "date": "2023-11-01", "categories": []}
{"id": "2211.14485", "title": "FastHuman: Reconstructing High-Quality Clothed Human in Minutes", "abstract": "  We propose an approach for optimizing high-quality clothed human body shapes\nin minutes, using multi-view posed images. While traditional neural rendering\nmethods struggle to disentangle geometry and appearance using only rendering\nloss, and are computationally intensive, our method uses a mesh-based patch\nwarping technique to ensure multi-view photometric consistency, and sphere\nharmonics (SH) illumination to refine geometric details efficiently. We employ\noriented point clouds' shape representation and SH shading, which significantly\nreduces optimization and rendering times compared to implicit methods. Our\napproach has demonstrated promising results on both synthetic and real-world\ndatasets, making it an effective solution for rapidly generating high-quality\nhuman body shapes. Project page\n\\href{https://l1346792580123.github.io/nccsfs/}{https://l1346792580123.github.io/nccsfs/}\n", "date": "2022-11-26", "categories": []}
{"id": "2308.14847", "title": "NSF: Neural Surface Fields for Human Modeling from Monocular Depth", "abstract": "  Obtaining personalized 3D animatable avatars from a monocular camera has\nseveral real world applications in gaming, virtual try-on, animation, and\nVR/XR, etc. However, it is very challenging to model dynamic and fine-grained\nclothing deformations from such sparse data. Existing methods for modeling 3D\nhumans from depth data have limitations in terms of computational efficiency,\nmesh coherency, and flexibility in resolution and topology. For instance,\nreconstructing shapes using implicit functions and extracting explicit meshes\nper frame is computationally expensive and cannot ensure coherent meshes across\nframes. Moreover, predicting per-vertex deformations on a pre-designed human\ntemplate with a discrete surface lacks flexibility in resolution and topology.\nTo overcome these limitations, we propose a novel method Neural Surface Fields\nfor modeling 3D clothed humans from monocular depth. NSF defines a neural field\nsolely on the base surface which models a continuous and flexible displacement\nfield. NSF can be adapted to the base surface with different resolution and\ntopology without retraining at inference time. Compared to existing approaches,\nour method eliminates the expensive per-frame surface extraction while\nmaintaining mesh coherency, and is capable of reconstructing meshes with\narbitrary resolution without retraining. To foster research in this direction,\nwe release our code in project page at: https://yuxuan-xue.com/nsf.\n", "date": "2023-08-28", "categories": []}
{"id": "2310.19580", "title": "A Perceptual Shape Loss for Monocular 3D Face Reconstruction", "abstract": "  Monocular 3D face reconstruction is a wide-spread topic, and existing\napproaches tackle the problem either through fast neural network inference or\noffline iterative reconstruction of face geometry. In either case\ncarefully-designed energy functions are minimized, commonly including loss\nterms like a photometric loss, a landmark reprojection loss, and others. In\nthis work we propose a new loss function for monocular face capture, inspired\nby how humans would perceive the quality of a 3D face reconstruction given a\nparticular image. It is widely known that shading provides a strong indicator\nfor 3D shape in the human visual system. As such, our new 'perceptual' shape\nloss aims to judge the quality of a 3D face estimate using only shading cues.\nOur loss is implemented as a discriminator-style neural network that takes an\ninput face image and a shaded render of the geometry estimate, and then\npredicts a score that perceptually evaluates how well the shaded render matches\nthe given image. This 'critic' network operates on the RGB image and geometry\nrender alone, without requiring an estimate of the albedo or illumination in\nthe scene. Furthermore, our loss operates entirely in image space and is thus\nagnostic to mesh topology. We show how our new perceptual shape loss can be\ncombined with traditional energy terms for monocular 3D face optimization and\ndeep neural network regression, improving upon current state-of-the-art\nresults.\n", "date": "2023-10-30", "categories": []}
{"id": "2310.17519", "title": "FLARE: Fast Learning of Animatable and Relightable Mesh Avatars", "abstract": "  Our goal is to efficiently learn personalized animatable 3D head avatars from\nvideos that are geometrically accurate, realistic, relightable, and compatible\nwith current rendering systems. While 3D meshes enable efficient processing and\nare highly portable, they lack realism in terms of shape and appearance. Neural\nrepresentations, on the other hand, are realistic but lack compatibility and\nare slow to train and render. Our key insight is that it is possible to\nefficiently learn high-fidelity 3D mesh representations via differentiable\nrendering by exploiting highly-optimized methods from traditional computer\ngraphics and approximating some of the components with neural networks. To that\nend, we introduce FLARE, a technique that enables the creation of animatable\nand relightable mesh avatars from a single monocular video. First, we learn a\ncanonical geometry using a mesh representation, enabling efficient\ndifferentiable rasterization and straightforward animation via learned\nblendshapes and linear blend skinning weights. Second, we follow\nphysically-based rendering and factor observed colors into intrinsic albedo,\nroughness, and a neural representation of the illumination, allowing the\nlearned avatars to be relit in novel scenes. Since our input videos are\ncaptured on a single device with a narrow field of view, modeling the\nsurrounding environment light is non-trivial. Based on the split-sum\napproximation for modeling specular reflections, we address this by\napproximating the pre-filtered environment map with a multi-layer perceptron\n(MLP) modulated by the surface roughness, eliminating the need to explicitly\nmodel the light. We demonstrate that our mesh-based avatar formulation,\ncombined with learned deformation, material, and lighting MLPs, produces\navatars with high-quality geometry and appearance, while also being efficient\nto train and render compared to existing approaches.\n", "date": "2023-10-26", "categories": []}
{"id": "2308.13369", "title": "Distribution-Aligned Diffusion for Human Mesh Recovery", "abstract": "  Recovering a 3D human mesh from a single RGB image is a challenging task due\nto depth ambiguity and self-occlusion, resulting in a high degree of\nuncertainty. Meanwhile, diffusion models have recently seen much success in\ngenerating high-quality outputs by progressively denoising noisy inputs.\nInspired by their capability, we explore a diffusion-based approach for human\nmesh recovery, and propose a Human Mesh Diffusion (HMDiff) framework which\nframes mesh recovery as a reverse diffusion process. We also propose a\nDistribution Alignment Technique (DAT) that infuses prior distribution\ninformation into the mesh distribution diffusion process, and provides useful\nprior knowledge to facilitate the mesh recovery task. Our method achieves\nstate-of-the-art performance on three widely used datasets. Project page:\nhttps://gongjia0208.github.io/HMDiff/.\n", "date": "2023-08-25", "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": "2309.13524", "title": "Global-correlated 3D-decoupling Transformer for Clothed Avatar\n  Reconstruction", "abstract": "  Reconstructing 3D clothed human avatars from single images is a challenging\ntask, especially when encountering complex poses and loose clothing. Current\nmethods exhibit limitations in performance, largely attributable to their\ndependence on insufficient 2D image features and inconsistent query methods.\nOwing to this, we present the Global-correlated 3D-decoupling Transformer for\nclothed Avatar reconstruction (GTA), a novel transformer-based architecture\nthat reconstructs clothed human avatars from monocular images. Our approach\nleverages transformer architectures by utilizing a Vision Transformer model as\nan encoder for capturing global-correlated image features. Subsequently, our\ninnovative 3D-decoupling decoder employs cross-attention to decouple tri-plane\nfeatures, using learnable embeddings as queries for cross-plane generation. To\neffectively enhance feature fusion with the tri-plane 3D feature and human body\nprior, we propose a hybrid prior fusion strategy combining spatial and\nprior-enhanced queries, leveraging the benefits of spatial localization and\nhuman body prior knowledge. Comprehensive experiments on CAPE and THuman2.0\ndatasets illustrate that our method outperforms state-of-the-art approaches in\nboth geometry and texture reconstruction, exhibiting high robustness to\nchallenging poses and loose clothing, and producing higher-resolution textures.\nCodes will be available at https://github.com/River-Zhang/GTA.\n", "date": "2023-09-24", "categories": []}
{"id": "2310.14237", "title": "High-Quality 3D Face Reconstruction with Affine Convolutional Networks", "abstract": "  Recent works based on convolutional encoder-decoder architecture and 3DMM\nparameterization have shown great potential for canonical view reconstruction\nfrom a single input image. Conventional CNN architectures benefit from\nexploiting the spatial correspondence between the input and output pixels.\nHowever, in 3D face reconstruction, the spatial misalignment between the input\nimage (e.g. face) and the canonical/UV output makes the feature\nencoding-decoding process quite challenging. In this paper, to tackle this\nproblem, we propose a new network architecture, namely the Affine Convolution\nNetworks, which enables CNN based approaches to handle spatially\nnon-corresponding input and output images and maintain high-fidelity quality\noutput at the same time. In our method, an affine transformation matrix is\nlearned from the affine convolution layer for each spatial location of the\nfeature maps. In addition, we represent 3D human heads in UV space with\nmultiple components, including diffuse maps for texture representation,\nposition maps for geometry representation, and light maps for recovering more\ncomplex lighting conditions in the real world. All the components can be\ntrained without any manual annotations. Our method is parametric-free and can\ngenerate high-quality UV maps at resolution of 512 x 512 pixels, while previous\napproaches normally generate 256 x 256 pixels or smaller. Our code will be\nreleased once the paper got accepted.\n", "date": "2023-10-22", "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": "2302.03397", "title": "AniPixel: Towards Animatable Pixel-Aligned Human Avatar", "abstract": "  Although human reconstruction typically results in human-specific avatars,\nrecent 3D scene reconstruction techniques utilizing pixel-aligned features show\npromise in generalizing to new scenes. Applying these techniques to human\navatar reconstruction can result in a volumetric avatar with generalizability\nbut limited animatability due to rendering only being possible for static\nrepresentations. In this paper, we propose AniPixel, a novel animatable and\ngeneralizable human avatar reconstruction method that leverages pixel-aligned\nfeatures for body geometry prediction and RGB color blending. Technically, to\nalign the canonical space with the target space and the observation space, we\npropose a bidirectional neural skinning field based on skeleton-driven\ndeformation to establish the target-to-canonical and canonical-to-observation\ncorrespondences. Then, we disentangle the canonical body geometry into a\nnormalized neutral-sized body and a subject-specific residual for better\ngeneralizability. As the geometry and appearance are closely related, we\nintroduce pixel-aligned features to facilitate the body geometry prediction and\ndetailed surface normals to reinforce the RGB color blending. We also devise a\npose-dependent and view direction-related shading module to represent the local\nillumination variance. Experiments show that AniPixel renders comparable novel\nviews while delivering better novel pose animation results than\nstate-of-the-art methods.\n", "date": "2023-02-07", "categories": []}
{"id": "2308.08857", "title": "D-IF: Uncertainty-aware Human Digitization via Implicit Distribution\n  Field", "abstract": "  Realistic virtual humans play a crucial role in numerous industries, such as\nmetaverse, intelligent healthcare, and self-driving simulation. But creating\nthem on a large scale with high levels of realism remains a challenge. The\nutilization of deep implicit function sparks a new era of image-based 3D\nclothed human reconstruction, enabling pixel-aligned shape recovery with fine\ndetails. Subsequently, the vast majority of works locate the surface by\nregressing the deterministic implicit value for each point. However, should all\npoints be treated equally regardless of their proximity to the surface? In this\npaper, we propose replacing the implicit value with an adaptive uncertainty\ndistribution, to differentiate between points based on their distance to the\nsurface. This simple ``value to distribution'' transition yields significant\nimprovements on nearly all the baselines. Furthermore, qualitative results\ndemonstrate that the models trained using our uncertainty distribution loss,\ncan capture more intricate wrinkles, and realistic limbs. Code and models are\navailable for research purposes at https://github.com/psyai-net/D-IF_release.\n", "date": "2023-08-17", "categories": []}
{"id": "2310.01228", "title": "Reconstructing 3D Human Pose from RGB-D Data with Occlusions", "abstract": "  We propose a new method to reconstruct the 3D human body from RGB-D images\nwith occlusions. The foremost challenge is the incompleteness of the RGB-D data\ndue to occlusions between the body and the environment, leading to implausible\nreconstructions that suffer from severe human-scene penetration. To reconstruct\na semantically and physically plausible human body, we propose to reduce the\nsolution space based on scene information and prior knowledge. Our key idea is\nto constrain the solution space of the human body by considering the occluded\nbody parts and visible body parts separately: modeling all plausible poses\nwhere the occluded body parts do not penetrate the scene, and constraining the\nvisible body parts using depth data. Specifically, the first component is\nrealized by a neural network that estimates the candidate region named the\n\"free zone\", a region carved out of the open space within which it is safe to\nsearch for poses of the invisible body parts without concern for penetration.\nThe second component constrains the visible body parts using the \"truncated\nshadow volume\" of the scanned body point cloud. Furthermore, we propose to use\na volume matching strategy, which yields better performance than surface\nmatching, to match the human body with the confined region. We conducted\nexperiments on the PROX dataset, and the results demonstrate that our method\nproduces more accurate and plausible results compared with other methods.\n", "date": "2023-10-02", "categories": []}
{"id": "2310.05917", "title": "Drivable Avatar Clothing: Faithful Full-Body Telepresence with Dynamic\n  Clothing Driven by Sparse RGB-D Input", "abstract": "  Clothing is an important part of human appearance but challenging to model in\nphotorealistic avatars. In this work we present avatars with dynamically moving\nloose clothing that can be faithfully driven by sparse RGB-D inputs as well as\nbody and face motion. We propose a Neural Iterative Closest Point (N-ICP)\nalgorithm that can efficiently track the coarse garment shape given sparse\ndepth input. Given the coarse tracking results, the input RGB-D images are then\nremapped to texel-aligned features, which are fed into the drivable avatar\nmodels to faithfully reconstruct appearance details. We evaluate our method\nagainst recent image-driven synthesis baselines, and conduct a comprehensive\nanalysis of the N-ICP algorithm. We demonstrate that our method can generalize\nto a novel testing environment, while preserving the ability to produce\nhigh-fidelity and faithful clothing dynamics and appearance.\n", "date": "2023-10-09", "categories": []}
{"id": "2310.06275", "title": "High-Fidelity 3D Head Avatars Reconstruction through Spatially-Varying\n  Expression Conditioned Neural Radiance Field", "abstract": "  One crucial aspect of 3D head avatar reconstruction lies in the details of\nfacial expressions. Although recent NeRF-based photo-realistic 3D head avatar\nmethods achieve high-quality avatar rendering, they still encounter challenges\nretaining intricate facial expression details because they overlook the\npotential of specific expression variations at different spatial positions when\nconditioning the radiance field. Motivated by this observation, we introduce a\nnovel Spatially-Varying Expression (SVE) conditioning. The SVE can be obtained\nby a simple MLP-based generation network, encompassing both spatial positional\nfeatures and global expression information. Benefiting from rich and diverse\ninformation of the SVE at different positions, the proposed SVE-conditioned\nneural radiance field can deal with intricate facial expressions and achieve\nrealistic rendering and geometry details of high-fidelity 3D head avatars.\nAdditionally, to further elevate the geometric and rendering quality, we\nintroduce a new coarse-to-fine training strategy, including a geometry\ninitialization strategy at the coarse stage and an adaptive importance sampling\nstrategy at the fine stage. Extensive experiments indicate that our method\noutperforms other state-of-the-art (SOTA) methods in rendering and geometry\nquality on mobile phone-collected and public datasets.\n", "date": "2023-10-10", "categories": []}
{"id": "2310.06332", "title": "CrowdRec: 3D Crowd Reconstruction from Single Color Images", "abstract": "  This is a technical report for the GigaCrowd challenge. Reconstructing 3D\ncrowds from monocular images is a challenging problem due to mutual occlusions,\nserver depth ambiguity, and complex spatial distribution. Since no large-scale\n3D crowd dataset can be used to train a robust model, the current multi-person\nmesh recovery methods can hardly achieve satisfactory performance in crowded\nscenes. In this paper, we exploit the crowd features and propose a\ncrowd-constrained optimization to improve the common single-person method on\ncrowd images. To avoid scale variations, we first detect human bounding-boxes\nand 2D poses from the original images with off-the-shelf detectors. Then, we\ntrain a single-person mesh recovery network using existing in-the-wild image\ndatasets. To promote a more reasonable spatial distribution, we further propose\na crowd constraint to refine the single-person network parameters. With the\noptimization, we can obtain accurate body poses and shapes with reasonable\nabsolute positions from a large-scale crowd image using a single-person\nbackbone. The code will be publicly available\nat~\\url{https://github.com/boycehbz/CrowdRec}.\n", "date": "2023-10-10", "categories": []}
{"id": "2310.06577", "title": "SketchBodyNet: A Sketch-Driven Multi-faceted Decoder Network for 3D\n  Human Reconstruction", "abstract": "  Reconstructing 3D human shapes from 2D images has received increasing\nattention recently due to its fundamental support for many high-level 3D\napplications. Compared with natural images, freehand sketches are much more\nflexible to depict various shapes, providing a high potential and valuable way\nfor 3D human reconstruction. However, such a task is highly challenging. The\nsparse abstract characteristics of sketches add severe difficulties, such as\narbitrariness, inaccuracy, and lacking image details, to the already badly\nill-posed problem of 2D-to-3D reconstruction. Although current methods have\nachieved great success in reconstructing 3D human bodies from a single-view\nimage, they do not work well on freehand sketches. In this paper, we propose a\nnovel sketch-driven multi-faceted decoder network termed SketchBodyNet to\naddress this task. Specifically, the network consists of a backbone and three\nseparate attention decoder branches, where a multi-head self-attention module\nis exploited in each decoder to obtain enhanced features, followed by a\nmulti-layer perceptron. The multi-faceted decoders aim to predict the camera,\nshape, and pose parameters, respectively, which are then associated with the\nSMPL model to reconstruct the corresponding 3D human mesh. In learning,\nexisting 3D meshes are projected via the camera parameters into 2D synthetic\nsketches with joints, which are combined with the freehand sketches to optimize\nthe model. To verify our method, we collect a large-scale dataset of about 26k\nfreehand sketches and their corresponding 3D meshes containing various poses of\nhuman bodies from 14 different angles. Extensive experimental results\ndemonstrate our SketchBodyNet achieves superior performance in reconstructing\n3D human meshes from freehand sketches.\n", "date": "2023-10-10", "categories": []}
{"id": "2304.09423", "title": "ASM: Adaptive Skinning Model for High-Quality 3D Face Modeling", "abstract": "  The research fields of parametric face model and 3D face reconstruction have\nbeen extensively studied. However, a critical question remains unanswered: how\nto tailor the face model for specific reconstruction settings. We argue that\nreconstruction with multi-view uncalibrated images demands a new model with\nstronger capacity. Our study shifts attention from data-dependent 3D Morphable\nModels (3DMM) to an understudied human-designed skinning model. We propose\nAdaptive Skinning Model (ASM), which redefines the skinning model with more\ncompact and fully tunable parameters. With extensive experiments, we\ndemonstrate that ASM achieves significantly improved capacity than 3DMM, with\nthe additional advantage of model size and easy implementation for new\ntopology. We achieve state-of-the-art performance with ASM for multi-view\nreconstruction on the Florence MICC Coop benchmark. Our quantitative analysis\ndemonstrates the importance of a high-capacity model for fully exploiting\nabundant information from multi-view input in reconstruction. Furthermore, our\nmodel with physical-semantic parameters can be directly utilized for real-world\napplications, such as in-game avatar creation. As a result, our work opens up\nnew research direction for parametric face model and facilitates future\nresearch on multi-view reconstruction.\n", "date": "2023-04-19", "categories": []}
{"id": "2310.03205", "title": "A Large-Scale 3D Face Mesh Video Dataset via Neural Re-parameterized\n  Optimization", "abstract": "  We propose NeuFace, a 3D face mesh pseudo annotation method on videos via\nneural re-parameterized optimization. Despite the huge progress in 3D face\nreconstruction methods, generating reliable 3D face labels for in-the-wild\ndynamic videos remains challenging. Using NeuFace optimization, we annotate the\nper-view/-frame accurate and consistent face meshes on large-scale face videos,\ncalled the NeuFace-dataset. We investigate how neural re-parameterization helps\nto reconstruct image-aligned facial details on 3D meshes via gradient analysis.\nBy exploiting the naturalness and diversity of 3D faces in our dataset, we\ndemonstrate the usefulness of our dataset for 3D face-related tasks: improving\nthe reconstruction accuracy of an existing 3D face reconstruction model and\nlearning 3D facial motion prior. Code and datasets will be available at\nhttps://neuface-dataset.github.io.\n", "date": "2023-10-04", "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": "2309.16859", "title": "Preface: A Data-driven Volumetric Prior for Few-shot Ultra\n  High-resolution Face Synthesis", "abstract": "  NeRFs have enabled highly realistic synthesis of human faces including\ncomplex appearance and reflectance effects of hair and skin. These methods\ntypically require a large number of multi-view input images, making the process\nhardware intensive and cumbersome, limiting applicability to unconstrained\nsettings. We propose a novel volumetric human face prior that enables the\nsynthesis of ultra high-resolution novel views of subjects that are not part of\nthe prior's training distribution. This prior model consists of an\nidentity-conditioned NeRF, trained on a dataset of low-resolution multi-view\nimages of diverse humans with known camera calibration. A simple sparse\nlandmark-based 3D alignment of the training dataset allows our model to learn a\nsmooth latent space of geometry and appearance despite a limited number of\ntraining identities. A high-quality volumetric representation of a novel\nsubject can be obtained by model fitting to 2 or 3 camera views of arbitrary\nresolution. Importantly, our method requires as few as two views of casually\ncaptured images as input at inference time.\n", "date": "2023-09-28", "categories": []}
{"id": "2309.17128", "title": "HAvatar: High-fidelity Head Avatar via Facial Model Conditioned Neural\n  Radiance Field", "abstract": "  The problem of modeling an animatable 3D human head avatar under light-weight\nsetups is of significant importance but has not been well solved. Existing 3D\nrepresentations either perform well in the realism of portrait images synthesis\nor the accuracy of expression control, but not both. To address the problem, we\nintroduce a novel hybrid explicit-implicit 3D representation, Facial Model\nConditioned Neural Radiance Field, which integrates the expressiveness of NeRF\nand the prior information from the parametric template. At the core of our\nrepresentation, a synthetic-renderings-based condition method is proposed to\nfuse the prior information from the parametric model into the implicit field\nwithout constraining its topological flexibility. Besides, based on the hybrid\nrepresentation, we properly overcome the inconsistent shape issue presented in\nexisting methods and improve the animation stability. Moreover, by adopting an\noverall GAN-based architecture using an image-to-image translation network, we\nachieve high-resolution, realistic and view-consistent synthesis of dynamic\nhead appearance. Experiments demonstrate that our method can achieve\nstate-of-the-art performance for 3D head avatar animation compared with\nprevious methods.\n", "date": "2023-09-29", "categories": []}
{"id": "2309.16189", "title": "Cloth2Body: Generating 3D Human Body Mesh from 2D Clothing", "abstract": "  In this paper, we define and study a new Cloth2Body problem which has a goal\nof generating 3D human body meshes from a 2D clothing image. Unlike the\nexisting human mesh recovery problem, Cloth2Body needs to address new and\nemerging challenges raised by the partial observation of the input and the high\ndiversity of the output. Indeed, there are three specific challenges. First,\nhow to locate and pose human bodies into the clothes. Second, how to\neffectively estimate body shapes out of various clothing types. Finally, how to\ngenerate diverse and plausible results from a 2D clothing image. To this end,\nwe propose an end-to-end framework that can accurately estimate 3D body mesh\nparameterized by pose and shape from a 2D clothing image. Along this line, we\nfirst utilize Kinematics-aware Pose Estimation to estimate body pose\nparameters. 3D skeleton is employed as a proxy followed by an inverse\nkinematics module to boost the estimation accuracy. We additionally design an\nadaptive depth trick to align the re-projected 3D mesh better with 2D clothing\nimage by disentangling the effects of object size and camera extrinsic. Next,\nwe propose Physics-informed Shape Estimation to estimate body shape parameters.\n3D shape parameters are predicted based on partial body measurements estimated\nfrom RGB image, which not only improves pixel-wise human-cloth alignment, but\nalso enables flexible user editing. Finally, we design Evolution-based pose\ngeneration method, a skeleton transplanting method inspired by genetic\nalgorithms to generate diverse reasonable poses during inference. As shown by\nexperimental results on both synthetic and real-world data, the proposed\nframework achieves state-of-the-art performance and can effectively recover\nnatural and diverse 3D body meshes from 2D images that align well with\nclothing.\n", "date": "2023-09-28", "categories": []}
{"id": "2308.07551", "title": "FLAME-based Multi-View 3D Face Reconstruction", "abstract": "  At present, face 3D reconstruction has broad application prospects in various\nfields, but the research on it is still in the development stage. In this\npaper, we hope to achieve better face 3D reconstruction quality by combining\nmulti-view training framework with face parametric model Flame, propose a\nmulti-view training and testing model MFNet (Multi-view Flame Network). We\nbuild a self-supervised training framework and implement constraints such as\nmulti-view optical flow loss function and face landmark loss, and finally\nobtain a complete MFNet. We propose innovative implementations of multi-view\noptical flow loss and the covisible mask. We test our model on AFLW and\nfacescape datasets and also take pictures of our faces to reconstruct 3D faces\nwhile simulating actual scenarios as much as possible, which achieves good\nresults. Our work mainly addresses the problem of combining parametric models\nof faces with multi-view face 3D reconstruction and explores the implementation\nof a Flame based multi-view training and testing framework for contributing to\nthe field of face 3D reconstruction.\n", "date": "2023-08-15", "categories": []}
{"id": "2209.05070", "title": "mmBody Benchmark: 3D Body Reconstruction Dataset and Analysis for\n  Millimeter Wave Radar", "abstract": "  Millimeter Wave (mmWave) Radar is gaining popularity as it can work in\nadverse environments like smoke, rain, snow, poor lighting, etc. Prior work has\nexplored the possibility of reconstructing 3D skeletons or meshes from the\nnoisy and sparse mmWave Radar signals. However, it is unclear how accurately we\ncan reconstruct the 3D body from the mmWave signals across scenes and how it\nperforms compared with cameras, which are important aspects needed to be\nconsidered when either using mmWave radars alone or combining them with\ncameras. To answer these questions, an automatic 3D body annotation system is\nfirst designed and built up with multiple sensors to collect a large-scale\ndataset. The dataset consists of synchronized and calibrated mmWave radar point\nclouds and RGB(D) images in different scenes and skeleton/mesh annotations for\nhumans in the scenes. With this dataset, we train state-of-the-art methods with\ninputs from different sensors and test them in various scenarios. The results\ndemonstrate that 1) despite the noise and sparsity of the generated point\nclouds, the mmWave radar can achieve better reconstruction accuracy than the\nRGB camera but worse than the depth camera; 2) the reconstruction from the\nmmWave radar is affected by adverse weather conditions moderately while the\nRGB(D) camera is severely affected. Further, analysis of the dataset and the\nresults shadow insights on improving the reconstruction from the mmWave radar\nand the combination of signals from different sensors.\n", "date": "2022-09-12", "categories": []}
{"id": "2309.12183", "title": "ORTexME: Occlusion-Robust Human Shape and Pose via Temporal Average\n  Texture and Mesh Encoding", "abstract": "  In 3D human shape and pose estimation from a monocular video, models trained\nwith limited labeled data cannot generalize well to videos with occlusion,\nwhich is common in the wild videos. The recent human neural rendering\napproaches focusing on novel view synthesis initialized by the off-the-shelf\nhuman shape and pose methods have the potential to correct the initial human\nshape. However, the existing methods have some drawbacks such as, erroneous in\nhandling occlusion, sensitive to inaccurate human segmentation, and ineffective\nloss computation due to the non-regularized opacity field. To address these\nproblems, we introduce ORTexME, an occlusion-robust temporal method that\nutilizes temporal information from the input video to better regularize the\noccluded body parts. While our ORTexME is based on NeRF, to determine the\nreliable regions for the NeRF ray sampling, we utilize our novel average\ntexture learning approach to learn the average appearance of a person, and to\ninfer a mask based on the average texture. In addition, to guide the\nopacity-field updates in NeRF to suppress blur and noise, we propose the use of\nhuman body mesh. The quantitative evaluation demonstrates that our method\nachieves significant improvement on the challenging multi-person 3DPW dataset,\nwhere our method achieves 1.8 P-MPJPE error reduction. The SOTA rendering-based\nmethods fail and enlarge the error up to 5.6 on the same dataset.\n", "date": "2023-09-21", "categories": []}
{"id": "2210.01346", "title": "ImmFusion: Robust mmWave-RGB Fusion for 3D Human Body Reconstruction in\n  All Weather Conditions", "abstract": "  3D human reconstruction from RGB images achieves decent results in good\nweather conditions but degrades dramatically in rough weather. Complementary,\nmmWave radars have been employed to reconstruct 3D human joints and meshes in\nrough weather. However, combining RGB and mmWave signals for robust all-weather\n3D human reconstruction is still an open challenge, given the sparse nature of\nmmWave and the vulnerability of RGB images. In this paper, we present\nImmFusion, the first mmWave-RGB fusion solution to reconstruct 3D human bodies\nin all weather conditions robustly. Specifically, our ImmFusion consists of\nimage and point backbones for token feature extraction and a Transformer module\nfor token fusion. The image and point backbones refine global and local\nfeatures from original data, and the Fusion Transformer Module aims for\neffective information fusion of two modalities by dynamically selecting\ninformative tokens. Extensive experiments on a large-scale dataset, mmBody,\ncaptured in various environments demonstrate that ImmFusion can efficiently\nutilize the information of two modalities to achieve a robust 3D human body\nreconstruction in all weather conditions. In addition, our method's accuracy is\nsignificantly superior to that of state-of-the-art Transformer-based\nLiDAR-camera fusion methods.\n", "date": "2022-10-04", "categories": []}
{"id": "2309.11357", "title": "3D Face Reconstruction: the Road to Forensics", "abstract": "  3D face reconstruction algorithms from images and videos are applied to many\nfields, from plastic surgery to the entertainment sector, thanks to their\nadvantageous features. However, when looking at forensic applications, 3D face\nreconstruction must observe strict requirements that still make its possible\nrole in bringing evidence to a lawsuit unclear. An extensive investigation of\nthe constraints, potential, and limits of its application in forensics is still\nmissing. Shedding some light on this matter is the goal of the present survey,\nwhich starts by clarifying the relation between forensic applications and\nbiometrics, with a focus on face recognition. Therefore, it provides an\nanalysis of the achievements of 3D face reconstruction algorithms from\nsurveillance videos and mugshot images and discusses the current obstacles that\nseparate 3D face reconstruction from an active role in forensic applications.\nFinally, it examines the underlying data sets, with their advantages and\nlimitations, while proposing alternatives that could substitute or complement\nthem.\n", "date": "2023-09-20", "categories": []}
{"id": "2304.10528", "title": "Generalizing Neural Human Fitting to Unseen Poses With Articulated SE(3)\n  Equivariance", "abstract": "  We address the problem of fitting a parametric human body model (SMPL) to\npoint cloud data. Optimization-based methods require careful initialization and\nare prone to becoming trapped in local optima. Learning-based methods address\nthis but do not generalize well when the input pose is far from those seen\nduring training. For rigid point clouds, remarkable generalization has been\nachieved by leveraging SE(3)-equivariant networks, but these methods do not\nwork on articulated objects. In this work we extend this idea to human bodies\nand propose ArtEq, a novel part-based SE(3)-equivariant neural architecture for\nSMPL model estimation from point clouds. Specifically, we learn a part\ndetection network by leveraging local SO(3) invariance, and regress shape and\npose using articulated SE(3) shape-invariant and pose-equivariant networks, all\ntrained end-to-end. Our novel pose regression module leverages the\npermutation-equivariant property of self-attention layers to preserve\nrotational equivariance. Experimental results show that ArtEq generalizes to\nposes not seen during training, outperforming state-of-the-art methods by ~44%\nin terms of body reconstruction accuracy, without requiring an optimization\nrefinement step. Furthermore, ArtEq is three orders of magnitude faster during\ninference than prior work and has 97.3% fewer parameters. The code and model\nare available for research purposes at https://arteq.is.tue.mpg.de.\n", "date": "2023-04-20", "categories": []}
{"id": "2304.06024", "title": "Probabilistic Human Mesh Recovery in 3D Scenes from Egocentric Views", "abstract": "  Automatic perception of human behaviors during social interactions is crucial\nfor AR/VR applications, and an essential component is estimation of plausible\n3D human pose and shape of our social partners from the egocentric view. One of\nthe biggest challenges of this task is severe body truncation due to close\nsocial distances in egocentric scenarios, which brings large pose ambiguities\nfor unseen body parts. To tackle this challenge, we propose a novel\nscene-conditioned diffusion method to model the body pose distribution.\nConditioned on the 3D scene geometry, the diffusion model generates bodies in\nplausible human-scene interactions, with the sampling guided by a physics-based\ncollision score to further resolve human-scene inter-penetrations. The\nclassifier-free training enables flexible sampling with different conditions\nand enhanced diversity. A visibility-aware graph convolution model guided by\nper-joint visibility serves as the diffusion denoiser to incorporate\ninter-joint dependencies and per-body-part control. Extensive evaluations show\nthat our method generates bodies in plausible interactions with 3D scenes,\nachieving both superior accuracy for visible joints and diversity for invisible\nbody parts. The code is available at\nhttps://sanweiliti.github.io/egohmr/egohmr.html.\n", "date": "2023-04-12", "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": "2305.03027", "title": "NeRSemble: Multi-view Radiance Field Reconstruction of Human Heads", "abstract": "  We focus on reconstructing high-fidelity radiance fields of human heads,\ncapturing their animations over time, and synthesizing re-renderings from novel\nviewpoints at arbitrary time steps. To this end, we propose a new multi-view\ncapture setup composed of 16 calibrated machine vision cameras that record\ntime-synchronized images at 7.1 MP resolution and 73 frames per second. With\nour setup, we collect a new dataset of over 4700 high-resolution,\nhigh-framerate sequences of more than 220 human heads, from which we introduce\na new human head reconstruction benchmark. The recorded sequences cover a wide\nrange of facial dynamics, including head motions, natural expressions,\nemotions, and spoken language. In order to reconstruct high-fidelity human\nheads, we propose Dynamic Neural Radiance Fields using Hash Ensembles\n(NeRSemble). We represent scene dynamics by combining a deformation field and\nan ensemble of 3D multi-resolution hash encodings. The deformation field allows\nfor precise modeling of simple scene movements, while the ensemble of hash\nencodings helps to represent complex dynamics. As a result, we obtain radiance\nfield representations of human heads that capture motion over time and\nfacilitate re-rendering of arbitrary novel viewpoints. In a series of\nexperiments, we explore the design choices of our method and demonstrate that\nour approach outperforms state-of-the-art dynamic radiance field approaches by\na significant margin.\n", "date": "2023-05-04", "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": "2309.06441", "title": "Learning Disentangled Avatars with Hybrid 3D Representations", "abstract": "  Tremendous efforts have been made to learn animatable and photorealistic\nhuman avatars. Towards this end, both explicit and implicit 3D representations\nare heavily studied for a holistic modeling and capture of the whole human\n(e.g., body, clothing, face and hair), but neither representation is an optimal\nchoice in terms of representation efficacy since different parts of the human\navatar have different modeling desiderata. For example, meshes are generally\nnot suitable for modeling clothing and hair. Motivated by this, we present\nDisentangled Avatars~(DELTA), which models humans with hybrid explicit-implicit\n3D representations. DELTA takes a monocular RGB video as input, and produces a\nhuman avatar with separate body and clothing/hair layers. Specifically, we\ndemonstrate two important applications for DELTA. For the first one, we\nconsider the disentanglement of the human body and clothing and in the second,\nwe disentangle the face and hair. To do so, DELTA represents the body or face\nwith an explicit mesh-based parametric 3D model and the clothing or hair with\nan implicit neural radiance field. To make this possible, we design an\nend-to-end differentiable renderer that integrates meshes into volumetric\nrendering, enabling DELTA to learn directly from monocular videos without any\n3D supervision. Finally, we show that how these two applications can be easily\ncombined to model full-body avatars, such that the hair, face, body and\nclothing can be fully disentangled yet jointly rendered. Such a disentanglement\nenables hair and clothing transfer to arbitrary body shapes. We empirically\nvalidate the effectiveness of DELTA's disentanglement by demonstrating its\npromising performance on disentangled reconstruction, virtual clothing try-on\nand hairstyle transfer. To facilitate future research, we also release an\nopen-sourced pipeline for the study of hybrid human avatar modeling.\n", "date": "2023-09-12", "categories": []}
{"id": "2209.04436", "title": "Multi-NeuS: 3D Head Portraits from Single Image with Neural Implicit\n  Functions", "abstract": "  We present an approach for the reconstruction of textured 3D meshes of human\nheads from one or few views. Since such few-shot reconstruction is\nunderconstrained, it requires prior knowledge which is hard to impose on\ntraditional 3D reconstruction algorithms. In this work, we rely on the recently\nintroduced 3D representation $\\unicode{x2013}$ neural implicit functions\n$\\unicode{x2013}$ which, being based on neural networks, allows to naturally\nlearn priors about human heads from data, and is directly convertible to\ntextured mesh. Namely, we extend NeuS, a state-of-the-art neural implicit\nfunction formulation, to represent multiple objects of a class (human heads in\nour case) simultaneously. The underlying neural net architecture is designed to\nlearn the commonalities among these objects and to generalize to unseen ones.\nOur model is trained on just a hundred smartphone videos and does not require\nany scanned 3D data. Afterwards, the model can fit novel heads in the few-shot\nor one-shot modes with good results.\n", "date": "2022-09-07", "categories": []}
{"id": "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": "2309.04247", "title": "Towards Practical Capture of High-Fidelity Relightable Avatars", "abstract": "  In this paper, we propose a novel framework, Tracking-free Relightable Avatar\n(TRAvatar), for capturing and reconstructing high-fidelity 3D avatars. Compared\nto previous methods, TRAvatar works in a more practical and efficient setting.\nSpecifically, TRAvatar is trained with dynamic image sequences captured in a\nLight Stage under varying lighting conditions, enabling realistic relighting\nand real-time animation for avatars in diverse scenes. Additionally, TRAvatar\nallows for tracking-free avatar capture and obviates the need for accurate\nsurface tracking under varying illumination conditions. Our contributions are\ntwo-fold: First, we propose a novel network architecture that explicitly builds\non and ensures the satisfaction of the linear nature of lighting. Trained on\nsimple group light captures, TRAvatar can predict the appearance in real-time\nwith a single forward pass, achieving high-quality relighting effects under\nilluminations of arbitrary environment maps. Second, we jointly optimize the\nfacial geometry and relightable appearance from scratch based on image\nsequences, where the tracking is implicitly learned. This tracking-free\napproach brings robustness for establishing temporal correspondences between\nframes under different lighting conditions. Extensive qualitative and\nquantitative experiments demonstrate that our framework achieves superior\nperformance for photorealistic avatar animation and relighting.\n", "date": "2023-09-08", "categories": []}
{"id": "2206.02194", "title": "FOF: Learning Fourier Occupancy Field for Monocular Real-time Human\n  Reconstruction", "abstract": "  The advent of deep learning has led to significant progress in monocular\nhuman reconstruction. However, existing representations, such as parametric\nmodels, voxel grids, meshes and implicit neural representations, have\ndifficulties achieving high-quality results and real-time speed at the same\ntime. In this paper, we propose Fourier Occupancy Field (FOF), a novel\npowerful, efficient and flexible 3D representation, for monocular real-time and\naccurate human reconstruction. The FOF represents a 3D object with a 2D field\northogonal to the view direction where at each 2D position the occupancy field\nof the object along the view direction is compactly represented with the first\nfew terms of Fourier series, which retains the topology and neighborhood\nrelation in the 2D domain. A FOF can be stored as a multi-channel image, which\nis compatible with 2D convolutional neural networks and can bridge the gap\nbetween 3D geometries and 2D images. The FOF is very flexible and extensible,\ne.g., parametric models can be easily integrated into a FOF as a prior to\ngenerate more robust results. Based on FOF, we design the first 30+FPS\nhigh-fidelity real-time monocular human reconstruction framework. We\ndemonstrate the potential of FOF on both public dataset and real captured data.\nThe code will be released for research purposes.\n", "date": "2022-06-05", "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": "2304.02013", "title": "NPC: Neural Point Characters from Video", "abstract": "  High-fidelity human 3D models can now be learned directly from videos,\ntypically by combining a template-based surface model with neural\nrepresentations. However, obtaining a template surface requires expensive\nmulti-view capture systems, laser scans, or strictly controlled conditions.\nPrevious methods avoid using a template but rely on a costly or ill-posed\nmapping from observation to canonical space. We propose a hybrid point-based\nrepresentation for reconstructing animatable characters that does not require\nan explicit surface model, while being generalizable to novel poses. For a\ngiven video, our method automatically produces an explicit set of 3D points\nrepresenting approximate canonical geometry, and learns an articulated\ndeformation model that produces pose-dependent point transformations. The\npoints serve both as a scaffold for high-frequency neural features and an\nanchor for efficiently mapping between observation and canonical space. We\ndemonstrate on established benchmarks that our representation overcomes\nlimitations of prior work operating in either canonical or in observation\nspace. Moreover, our automatic point extraction approach enables learning\nmodels of human and animal characters alike, matching the performance of the\nmethods using rigged surface templates despite being more general. Project\nwebsite: https://lemonatsu.github.io/npc/\n", "date": "2023-04-04", "categories": []}
{"id": "2305.20091", "title": "Humans in 4D: Reconstructing and Tracking Humans with Transformers", "abstract": "  We present an approach to reconstruct humans and track them over time. At the\ncore of our approach, we propose a fully \"transformerized\" version of a network\nfor human mesh recovery. This network, HMR 2.0, advances the state of the art\nand shows the capability to analyze unusual poses that have in the past been\ndifficult to reconstruct from single images. To analyze video, we use 3D\nreconstructions from HMR 2.0 as input to a tracking system that operates in 3D.\nThis enables us to deal with multiple people and maintain identities through\nocclusion events. Our complete approach, 4DHumans, achieves state-of-the-art\nresults for tracking people from monocular video. Furthermore, we demonstrate\nthe effectiveness of HMR 2.0 on the downstream task of action recognition,\nachieving significant improvements over previous pose-based action recognition\napproaches. Our code and models are available on the project website:\nhttps://shubham-goel.github.io/4dhumans/.\n", "date": "2023-05-31", "categories": []}
{"id": "2308.15844", "title": "Reconstructing Groups of People with Hypergraph Relational Reasoning", "abstract": "  Due to the mutual occlusion, severe scale variation, and complex spatial\ndistribution, the current multi-person mesh recovery methods cannot produce\naccurate absolute body poses and shapes in large-scale crowded scenes. To\naddress the obstacles, we fully exploit crowd features for reconstructing\ngroups of people from a monocular image. A novel hypergraph relational\nreasoning network is proposed to formulate the complex and high-order relation\ncorrelations among individuals and groups in the crowd. We first extract\ncompact human features and location information from the original\nhigh-resolution image. By conducting the relational reasoning on the extracted\nindividual features, the underlying crowd collectiveness and interaction\nrelationship can provide additional group information for the reconstruction.\nFinally, the updated individual features and the localization information are\nused to regress human meshes in camera coordinates. To facilitate the network\ntraining, we further build pseudo ground-truth on two crowd datasets, which may\nalso promote future research on pose estimation and human behavior\nunderstanding in crowded scenes. The experimental results show that our\napproach outperforms other baseline methods both in crowded and common\nscenarios. The code and datasets are publicly available at\nhttps://github.com/boycehbz/GroupRec.\n", "date": "2023-08-30", "categories": []}
{"id": "2308.13903", "title": "Disjoint Pose and Shape for 3D Face Reconstruction", "abstract": "  Existing methods for 3D face reconstruction from a few casually captured\nimages employ deep learning based models along with a 3D Morphable Model(3DMM)\nas face geometry prior. Structure From Motion(SFM), followed by Multi-View\nStereo (MVS), on the other hand, uses dozens of high-resolution images to\nreconstruct accurate 3D faces.However, it produces noisy and stretched-out\nresults with only two views available. In this paper, taking inspiration from\nboth these methods, we propose an end-to-end pipeline that disjointly solves\nfor pose and shape to make the optimization stable and accurate. We use a face\nshape prior to estimate face pose and use stereo matching followed by a 3DMM to\nsolve for the shape. The proposed method achieves end-to-end topological\nconsistency, enables iterative face pose refinement procedure, and show\nremarkable improvement on both quantitative and qualitative results over\nexisting state-of-the-art methods.\n", "date": "2023-08-26", "categories": []}
{"id": "2303.13796", "title": "Zolly: Zoom Focal Length Correctly for Perspective-Distorted Human Mesh\n  Reconstruction", "abstract": "  As it is hard to calibrate single-view RGB images in the wild, existing 3D\nhuman mesh reconstruction (3DHMR) methods either use a constant large focal\nlength or estimate one based on the background environment context, which can\nnot tackle the problem of the torso, limb, hand or face distortion caused by\nperspective camera projection when the camera is close to the human body. The\nnaive focal length assumptions can harm this task with the incorrectly\nformulated projection matrices. To solve this, we propose Zolly, the first\n3DHMR method focusing on perspective-distorted images. Our approach begins with\nanalysing the reason for perspective distortion, which we find is mainly caused\nby the relative location of the human body to the camera center. We propose a\nnew camera model and a novel 2D representation, termed distortion image, which\ndescribes the 2D dense distortion scale of the human body. We then estimate the\ndistance from distortion scale features rather than environment context\nfeatures. Afterwards, we integrate the distortion feature with image features\nto reconstruct the body mesh. To formulate the correct projection matrix and\nlocate the human body position, we simultaneously use perspective and\nweak-perspective projection loss. Since existing datasets could not handle this\ntask, we propose the first synthetic dataset PDHuman and extend two real-world\ndatasets tailored for this task, all containing perspective-distorted human\nimages. Extensive experiments show that Zolly outperforms existing\nstate-of-the-art methods on both perspective-distorted datasets and the\nstandard benchmark (3DPW).\n", "date": "2023-03-24", "categories": []}
{"id": "2303.11225", "title": "HiFace: High-Fidelity 3D Face Reconstruction by Learning Static and\n  Dynamic Details", "abstract": "  3D Morphable Models (3DMMs) demonstrate great potential for reconstructing\nfaithful and animatable 3D facial surfaces from a single image. The facial\nsurface is influenced by the coarse shape, as well as the static detail (e,g.,\nperson-specific appearance) and dynamic detail (e.g., expression-driven\nwrinkles). Previous work struggles to decouple the static and dynamic details\nthrough image-level supervision, leading to reconstructions that are not\nrealistic. In this paper, we aim at high-fidelity 3D face reconstruction and\npropose HiFace to explicitly model the static and dynamic details.\nSpecifically, the static detail is modeled as the linear combination of a\ndisplacement basis, while the dynamic detail is modeled as the linear\ninterpolation of two displacement maps with polarized expressions. We exploit\nseveral loss functions to jointly learn the coarse shape and fine details with\nboth synthetic and real-world datasets, which enable HiFace to reconstruct\nhigh-fidelity 3D shapes with animatable details. Extensive quantitative and\nqualitative experiments demonstrate that HiFace presents state-of-the-art\nreconstruction quality and faithfully recovers both the static and dynamic\ndetails. Our project page can be found at https://project-hiface.github.io.\n", "date": "2023-03-20", "categories": []}
{"id": "2305.06077", "title": "Relightify: Relightable 3D Faces from a Single Image via Diffusion\n  Models", "abstract": "  Following the remarkable success of diffusion models on image generation,\nrecent works have also demonstrated their impressive ability to address a\nnumber of inverse problems in an unsupervised way, by properly constraining the\nsampling process based on a conditioning input. Motivated by this, in this\npaper, we present the first approach to use diffusion models as a prior for\nhighly accurate 3D facial BRDF reconstruction from a single image. We start by\nleveraging a high-quality UV dataset of facial reflectance (diffuse and\nspecular albedo and normals), which we render under varying illumination\nsettings to simulate natural RGB textures and, then, train an unconditional\ndiffusion model on concatenated pairs of rendered textures and reflectance\ncomponents. At test time, we fit a 3D morphable model to the given image and\nunwrap the face in a partial UV texture. By sampling from the diffusion model,\nwhile retaining the observed texture part intact, the model inpaints not only\nthe self-occluded areas but also the unknown reflectance components, in a\nsingle sequence of denoising steps. In contrast to existing methods, we\ndirectly acquire the observed texture from the input image, thus, resulting in\nmore faithful and consistent reflectance estimation. Through a series of\nqualitative and quantitative comparisons, we demonstrate superior performance\nin both texture completion as well as reflectance reconstruction tasks.\n", "date": "2023-05-10", "categories": []}
{"id": "2308.11184", "title": "ReFit: Recurrent Fitting Network for 3D Human Recovery", "abstract": "  We present Recurrent Fitting (ReFit), a neural network architecture for\nsingle-image, parametric 3D human reconstruction. ReFit learns a\nfeedback-update loop that mirrors the strategy of solving an inverse problem\nthrough optimization. At each iterative step, it reprojects keypoints from the\nhuman model to feature maps to query feedback, and uses a recurrent-based\nupdater to adjust the model to fit the image better. Because ReFit encodes\nstrong knowledge of the inverse problem, it is faster to train than previous\nregression models. At the same time, ReFit improves state-of-the-art\nperformance on standard benchmarks. Moreover, ReFit applies to other\noptimization settings, such as multi-view fitting and single-view shape\nfitting. Project website: https://yufu-wang.github.io/refit_humans/\n", "date": "2023-08-22", "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": "2308.08545", "title": "TeCH: Text-guided Reconstruction of Lifelike Clothed Humans", "abstract": "  Despite recent research advancements in reconstructing clothed humans from a\nsingle image, accurately restoring the \"unseen regions\" with high-level details\nremains an unsolved challenge that lacks attention. Existing methods often\ngenerate overly smooth back-side surfaces with a blurry texture. But how to\neffectively capture all visual attributes of an individual from a single image,\nwhich are sufficient to reconstruct unseen areas (e.g., the back view)?\nMotivated by the power of foundation models, TeCH reconstructs the 3D human by\nleveraging 1) descriptive text prompts (e.g., garments, colors, hairstyles)\nwhich are automatically generated via a garment parsing model and Visual\nQuestion Answering (VQA), 2) a personalized fine-tuned Text-to-Image diffusion\nmodel (T2I) which learns the \"indescribable\" appearance. To represent\nhigh-resolution 3D clothed humans at an affordable cost, we propose a hybrid 3D\nrepresentation based on DMTet, which consists of an explicit body shape grid\nand an implicit distance field. Guided by the descriptive prompts +\npersonalized T2I diffusion model, the geometry and texture of the 3D humans are\noptimized through multi-view Score Distillation Sampling (SDS) and\nreconstruction losses based on the original observation. TeCH produces\nhigh-fidelity 3D clothed humans with consistent & delicate texture, and\ndetailed full-body geometry. Quantitative and qualitative experiments\ndemonstrate that TeCH outperforms the state-of-the-art methods in terms of\nreconstruction accuracy and rendering quality. The code will be publicly\navailable for research purposes at https://huangyangyi.github.io/TeCH\n", "date": "2023-08-16", "categories": []}
{"id": "2308.10305", "title": "Co-Evolution of Pose and Mesh for 3D Human Body Estimation from Video", "abstract": "  Despite significant progress in single image-based 3D human mesh recovery,\naccurately and smoothly recovering 3D human motion from a video remains\nchallenging. Existing video-based methods generally recover human mesh by\nestimating the complex pose and shape parameters from coupled image features,\nwhose high complexity and low representation ability often result in\ninconsistent pose motion and limited shape patterns. To alleviate this issue,\nwe introduce 3D pose as the intermediary and propose a Pose and Mesh\nCo-Evolution network (PMCE) that decouples this task into two parts: 1)\nvideo-based 3D human pose estimation and 2) mesh vertices regression from the\nestimated 3D pose and temporal image feature. Specifically, we propose a\ntwo-stream encoder that estimates mid-frame 3D pose and extracts a temporal\nimage feature from the input image sequence. In addition, we design a\nco-evolution decoder that performs pose and mesh interactions with the\nimage-guided Adaptive Layer Normalization (AdaLN) to make pose and mesh fit the\nhuman body shape. Extensive experiments demonstrate that the proposed PMCE\noutperforms previous state-of-the-art methods in terms of both per-frame\naccuracy and temporal consistency on three benchmark datasets: 3DPW, Human3.6M,\nand MPI-INF-3DHP. Our code is available at https://github.com/kasvii/PMCE.\n", "date": "2023-08-20", "categories": []}
{"id": "2308.10334", "title": "Coordinate Transformer: Achieving Single-stage Multi-person Mesh\n  Recovery from Videos", "abstract": "  Multi-person 3D mesh recovery from videos is a critical first step towards\nautomatic perception of group behavior in virtual reality, physical therapy and\nbeyond. However, existing approaches rely on multi-stage paradigms, where the\nperson detection and tracking stages are performed in a multi-person setting,\nwhile temporal dynamics are only modeled for one person at a time.\nConsequently, their performance is severely limited by the lack of inter-person\ninteractions in the spatial-temporal mesh recovery, as well as by detection and\ntracking defects. To address these challenges, we propose the Coordinate\ntransFormer (CoordFormer) that directly models multi-person spatial-temporal\nrelations and simultaneously performs multi-mesh recovery in an end-to-end\nmanner. Instead of partitioning the feature map into coarse-scale patch-wise\ntokens, CoordFormer leverages a novel Coordinate-Aware Attention to preserve\npixel-level spatial-temporal coordinate information. Additionally, we propose a\nsimple, yet effective Body Center Attention mechanism to fuse position\ninformation. Extensive experiments on the 3DPW dataset demonstrate that\nCoordFormer significantly improves the state-of-the-art, outperforming the\npreviously best results by 4.2%, 8.8% and 4.7% according to the MPJPE, PAMPJPE,\nand PVE metrics, respectively, while being 40% faster than recent video-based\napproaches. The released code can be found at\nhttps://github.com/Li-Hao-yuan/CoordFormer.\n", "date": "2023-08-20", "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": "2303.13855", "title": "Deformable Model-Driven Neural Rendering for High-Fidelity 3D\n  Reconstruction of Human Heads Under Low-View Settings", "abstract": "  Reconstructing 3D human heads in low-view settings presents technical\nchallenges, mainly due to the pronounced risk of overfitting with limited views\nand high-frequency signals. To address this, we propose geometry decomposition\nand adopt a two-stage, coarse-to-fine training strategy, allowing for\nprogressively capturing high-frequency geometric details. We represent 3D human\nheads using the zero level-set of a combined signed distance field, comprising\na smooth template, a non-rigid deformation, and a high-frequency displacement\nfield. The template captures features that are independent of both identity and\nexpression and is co-trained with the deformation network across multiple\nindividuals with sparse and randomly selected views. The displacement field,\ncapturing individual-specific details, undergoes separate training for each\nperson. Our network training does not require 3D supervision or object masks.\nExperimental results demonstrate the effectiveness and robustness of our\ngeometry decomposition and two-stage training strategy. Our method outperforms\nexisting neural rendering approaches in terms of reconstruction accuracy and\nnovel view synthesis under low-view settings. Moreover, the pre-trained\ntemplate serves a good initialization for our model when encountering unseen\nindividuals.\n", "date": "2023-03-24", "categories": []}
{"id": "2306.16736", "title": "GraMMaR: Ground-aware Motion Model for 3D Human Motion Reconstruction", "abstract": "  Demystifying complex human-ground interactions is essential for accurate and\nrealistic 3D human motion reconstruction from RGB videos, as it ensures\nconsistency between the humans and the ground plane. Prior methods have modeled\nhuman-ground interactions either implicitly or in a sparse manner, often\nresulting in unrealistic and incorrect motions when faced with noise and\nuncertainty. In contrast, our approach explicitly represents these interactions\nin a dense and continuous manner. To this end, we propose a novel Ground-aware\nMotion Model for 3D Human Motion Reconstruction, named GraMMaR, which jointly\nlearns the distribution of transitions in both pose and interaction between\nevery joint and ground plane at each time step of a motion sequence. It is\ntrained to explicitly promote consistency between the motion and distance\nchange towards the ground. After training, we establish a joint optimization\nstrategy that utilizes GraMMaR as a dual-prior, regularizing the optimization\ntowards the space of plausible ground-aware motions. This leads to realistic\nand coherent motion reconstruction, irrespective of the assumed or learned\nground plane. Through extensive evaluation on the AMASS and AIST++ datasets,\nour model demonstrates good generalization and discriminating abilities in\nchallenging cases including complex and ambiguous human-ground interactions.\nThe code will be available at https://github.com/xymsh/GraMMaR.\n", "date": "2023-06-29", "categories": []}
{"id": "2307.16377", "title": "JOTR: 3D Joint Contrastive Learning with Transformers for Occluded Human\n  Mesh Recovery", "abstract": "  In this study, we focus on the problem of 3D human mesh recovery from a\nsingle image under obscured conditions. Most state-of-the-art methods aim to\nimprove 2D alignment technologies, such as spatial averaging and 2D joint\nsampling. However, they tend to neglect the crucial aspect of 3D alignment by\nimproving 3D representations. Furthermore, recent methods struggle to separate\nthe target human from occlusion or background in crowded scenes as they\noptimize the 3D space of target human with 3D joint coordinates as local\nsupervision. To address these issues, a desirable method would involve a\nframework for fusing 2D and 3D features and a strategy for optimizing the 3D\nspace globally. Therefore, this paper presents 3D JOint contrastive learning\nwith TRansformers (JOTR) framework for handling occluded 3D human mesh\nrecovery. Our method includes an encoder-decoder transformer architecture to\nfuse 2D and 3D representations for achieving 2D$\\&$3D aligned results in a\ncoarse-to-fine manner and a novel 3D joint contrastive learning approach for\nadding explicitly global supervision for the 3D feature space. The contrastive\nlearning approach includes two contrastive losses: joint-to-joint contrast for\nenhancing the similarity of semantically similar voxels (i.e., human joints),\nand joint-to-non-joint contrast for ensuring discrimination from others (e.g.,\nocclusions and background). Qualitative and quantitative analyses demonstrate\nthat our method outperforms state-of-the-art competitors on both\nocclusion-specific and standard benchmarks, significantly improving the\nreconstruction of occluded humans.\n", "date": "2023-07-31", "categories": []}
{"id": "2308.07903", "title": "Relightable and Animatable Neural Avatar from Sparse-View Video", "abstract": "  This paper tackles the challenge of creating relightable and animatable\nneural avatars from sparse-view (or even monocular) videos of dynamic humans\nunder unknown illumination. Compared to studio environments, this setting is\nmore practical and accessible but poses an extremely challenging ill-posed\nproblem. Previous neural human reconstruction methods are able to reconstruct\nanimatable avatars from sparse views using deformed Signed Distance Fields\n(SDF) but cannot recover material parameters for relighting. While\ndifferentiable inverse rendering-based methods have succeeded in material\nrecovery of static objects, it is not straightforward to extend them to dynamic\nhumans as it is computationally intensive to compute pixel-surface intersection\nand light visibility on deformed SDFs for inverse rendering. To solve this\nchallenge, we propose a Hierarchical Distance Query (HDQ) algorithm to\napproximate the world space distances under arbitrary human poses.\nSpecifically, we estimate coarse distances based on a parametric human model\nand compute fine distances by exploiting the local deformation invariance of\nSDF. Based on the HDQ algorithm, we leverage sphere tracing to efficiently\nestimate the surface intersection and light visibility. This allows us to\ndevelop the first system to recover animatable and relightable neural avatars\nfrom sparse view (or monocular) inputs. Experiments demonstrate that our\napproach is able to produce superior results compared to state-of-the-art\nmethods. Our code will be released for reproducibility.\n", "date": "2023-08-15", "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": "2303.12791", "title": "SHERF: Generalizable Human NeRF from a Single Image", "abstract": "  Existing Human NeRF methods for reconstructing 3D humans typically rely on\nmultiple 2D images from multi-view cameras or monocular videos captured from\nfixed camera views. However, in real-world scenarios, human images are often\ncaptured from random camera angles, presenting challenges for high-quality 3D\nhuman reconstruction. In this paper, we propose SHERF, the first generalizable\nHuman NeRF model for recovering animatable 3D humans from a single input image.\nSHERF extracts and encodes 3D human representations in canonical space,\nenabling rendering and animation from free views and poses. To achieve\nhigh-fidelity novel view and pose synthesis, the encoded 3D human\nrepresentations should capture both global appearance and local fine-grained\ntextures. To this end, we propose a bank of 3D-aware hierarchical features,\nincluding global, point-level, and pixel-aligned features, to facilitate\ninformative encoding. Global features enhance the information extracted from\nthe single input image and complement the information missing from the partial\n2D observation. Point-level features provide strong clues of 3D human\nstructure, while pixel-aligned features preserve more fine-grained details. To\neffectively integrate the 3D-aware hierarchical feature bank, we design a\nfeature fusion transformer. Extensive experiments on THuman, RenderPeople,\nZJU_MoCap, and HuMMan datasets demonstrate that SHERF achieves state-of-the-art\nperformance, with better generalizability for novel view and pose synthesis.\n", "date": "2023-03-22", "categories": []}
{"id": "2308.06554", "title": "Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh\n  Reconstruction", "abstract": "  Despite recent advances in 3D human mesh reconstruction, domain gap between\ntraining and test data is still a major challenge. Several prior works tackle\nthe domain gap problem via test-time adaptation that fine-tunes a network\nrelying on 2D evidence (e.g., 2D human keypoints) from test images. However,\nthe high reliance on 2D evidence during adaptation causes two major issues.\nFirst, 2D evidence induces depth ambiguity, preventing the learning of accurate\n3D human geometry. Second, 2D evidence is noisy or partially non-existent\nduring test time, and such imperfect 2D evidence leads to erroneous adaptation.\nTo overcome the above issues, we introduce CycleAdapt, which cyclically adapts\ntwo networks: a human mesh reconstruction network (HMRNet) and a human motion\ndenoising network (MDNet), given a test video. In our framework, to alleviate\nhigh reliance on 2D evidence, we fully supervise HMRNet with generated 3D\nsupervision targets by MDNet. Our cyclic adaptation scheme progressively\nelaborates the 3D supervision targets, which compensate for imperfect 2D\nevidence. As a result, our CycleAdapt achieves state-of-the-art performance\ncompared to previous test-time adaptation methods. The codes are available at\nhttps://github.com/hygenie1228/CycleAdapt_RELEASE.\n", "date": "2023-08-12", "categories": []}
{"id": "2308.05925", "title": "CaPhy: Capturing Physical Properties for Animatable Human Avatars", "abstract": "  We present CaPhy, a novel method for reconstructing animatable human avatars\nwith realistic dynamic properties for clothing. Specifically, we aim for\ncapturing the geometric and physical properties of the clothing from real\nobservations. This allows us to apply novel poses to the human avatar with\nphysically correct deformations and wrinkles of the clothing. To this end, we\ncombine unsupervised training with physics-based losses and 3D-supervised\ntraining using scanned data to reconstruct a dynamic model of clothing that is\nphysically realistic and conforms to the human scans. We also optimize the\nphysical parameters of the underlying physical model from the scans by\nintroducing gradient constraints of the physics-based losses. In contrast to\nprevious work on 3D avatar reconstruction, our method is able to generalize to\nnovel poses with realistic dynamic cloth deformations. Experiments on several\nsubjects demonstrate that our method can estimate the physical properties of\nthe garments, resulting in superior quantitative and qualitative results\ncompared with previous methods.\n", "date": "2023-08-11", "categories": []}
{"id": "2211.10705", "title": "TORE: Token Reduction for Efficient Human Mesh Recovery with Transformer", "abstract": "  In this paper, we introduce a set of simple yet effective TOken REduction\n(TORE) strategies for Transformer-based Human Mesh Recovery from monocular\nimages. Current SOTA performance is achieved by Transformer-based structures.\nHowever, they suffer from high model complexity and computation cost caused by\nredundant tokens. We propose token reduction strategies based on two important\naspects, i.e., the 3D geometry structure and 2D image feature, where we\nhierarchically recover the mesh geometry with priors from body structure and\nconduct token clustering to pass fewer but more discriminative image feature\ntokens to the Transformer. Our method massively reduces the number of tokens\ninvolved in high-complexity interactions in the Transformer. This leads to a\nsignificantly reduced computational cost while still achieving competitive or\neven higher accuracy in shape recovery. Extensive experiments across a wide\nrange of benchmarks validate the superior effectiveness of the proposed method.\nWe further demonstrate the generalizability of our method on hand mesh\nrecovery. Visit our project page at\nhttps://frank-zy-dou.github.io/projects/Tore/index.html.\n", "date": "2022-11-19", "categories": []}
{"id": "2307.11074", "title": "Learning Dense UV Completion for Human Mesh Recovery", "abstract": "  Human mesh reconstruction from a single image is challenging in the presence\nof occlusion, which can be caused by self, objects, or other humans. Existing\nmethods either fail to separate human features accurately or lack proper\nsupervision for feature completion. In this paper, we propose Dense Inpainting\nHuman Mesh Recovery (DIMR), a two-stage method that leverages dense\ncorrespondence maps to handle occlusion. Our method utilizes a dense\ncorrespondence map to separate visible human features and completes human\nfeatures on a structured UV map dense human with an attention-based feature\ncompletion module. We also design a feature inpainting training procedure that\nguides the network to learn from unoccluded features. We evaluate our method on\nseveral datasets and demonstrate its superior performance under heavily\noccluded scenarios compared to other methods. Extensive experiments show that\nour method obviously outperforms prior SOTA methods on heavily occluded images\nand achieves comparable results on the standard benchmarks (3DPW).\n", "date": "2023-07-20", "categories": []}
{"id": "2308.05112", "title": "Explicifying Neural Implicit Fields for Efficient Dynamic Human Avatar\n  Modeling via a Neural Explicit Surface", "abstract": "  This paper proposes a technique for efficiently modeling dynamic humans by\nexplicifying the implicit neural fields via a Neural Explicit Surface (NES).\nImplicit neural fields have advantages over traditional explicit\nrepresentations in modeling dynamic 3D content from sparse observations and\neffectively representing complex geometries and appearances. Implicit neural\nfields defined in 3D space, however, are expensive to render due to the need\nfor dense sampling during volumetric rendering. Moreover, their memory\nefficiency can be further optimized when modeling sparse 3D space. To overcome\nthese issues, the paper proposes utilizing Neural Explicit Surface (NES) to\nexplicitly represent implicit neural fields, facilitating memory and\ncomputational efficiency. To achieve this, the paper creates a fully\ndifferentiable conversion between the implicit neural fields and the explicit\nrendering interface of NES, leveraging the strengths of both implicit and\nexplicit approaches. This conversion enables effective training of the hybrid\nrepresentation using implicit methods and efficient rendering by integrating\nthe explicit rendering interface with a newly proposed rasterization-based\nneural renderer that only incurs a texture color query once for the initial ray\ninteraction with the explicit surface, resulting in improved inference\nefficiency. NES describes dynamic human geometries with pose-dependent neural\nimplicit surface deformation fields and their dynamic neural textures both in\n2D space, which is a more memory-efficient alternative to traditional 3D\nmethods, reducing redundancy and computational load. The comprehensive\nexperiments show that NES performs similarly to previous 3D approaches, with\ngreatly improved rendering speed and reduced memory cost.\n", "date": "2023-08-07", "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": "2308.01854", "title": "Reconstructing Three-Dimensional Models of Interacting Humans", "abstract": "  Understanding 3d human interactions is fundamental for fine-grained scene\nanalysis and behavioural modeling. However, most of the existing models predict\nincorrect, lifeless 3d estimates, that miss the subtle human contact\naspects--the essence of the event--and are of little use for detailed\nbehavioral understanding. This paper addresses such issues with several\ncontributions: (1) we introduce models for interaction signature estimation\n(ISP) encompassing contact detection, segmentation, and 3d contact signature\nprediction; (2) we show how such components can be leveraged to ensure contact\nconsistency during 3d reconstruction; (3) we construct several large datasets\nfor learning and evaluating 3d contact prediction and reconstruction methods;\nspecifically, we introduce CHI3D, a lab-based accurate 3d motion capture\ndataset with 631 sequences containing $2,525$ contact events, $728,664$ ground\ntruth 3d poses, as well as FlickrCI3D, a dataset of $11,216$ images, with\n$14,081$ processed pairs of people, and $81,233$ facet-level surface\ncorrespondences. Finally, (4) we propose methodology for recovering the\nground-truth pose and shape of interacting people in a controlled setup and (5)\nannotate all 3d interaction motions in CHI3D with textual descriptions. Motion\ndata in multiple formats (GHUM and SMPLX parameters, Human3.6m 3d joints) is\nmade available for research purposes at \\url{https://ci3d.imar.ro}, together\nwith an evaluation server and a public benchmark.\n", "date": "2023-08-03", "categories": []}
{"id": "2308.00799", "title": "Body Knowledge and Uncertainty Modeling for Monocular 3D Human Body\n  Reconstruction", "abstract": "  While 3D body reconstruction methods have made remarkable progress recently,\nit remains difficult to acquire the sufficiently accurate and numerous 3D\nsupervisions required for training. In this paper, we propose \\textbf{KNOWN}, a\nframework that effectively utilizes body \\textbf{KNOW}ledge and\nu\\textbf{N}certainty modeling to compensate for insufficient 3D supervisions.\nKNOWN exploits a comprehensive set of generic body constraints derived from\nwell-established body knowledge. These generic constraints precisely and\nexplicitly characterize the reconstruction plausibility and enable 3D\nreconstruction models to be trained without any 3D data. Moreover, existing\nmethods typically use images from multiple datasets during training, which can\nresult in data noise (\\textit{e.g.}, inconsistent joint annotation) and data\nimbalance (\\textit{e.g.}, minority images representing unusual poses or\ncaptured from challenging camera views). KNOWN solves these problems through a\nnovel probabilistic framework that models both aleatoric and epistemic\nuncertainty. Aleatoric uncertainty is encoded in a robust Negative\nLog-Likelihood (NLL) training loss, while epistemic uncertainty is used to\nguide model refinement. Experiments demonstrate that KNOWN's body\nreconstruction outperforms prior weakly-supervised approaches, particularly on\nthe challenging minority images.\n", "date": "2023-08-01", "categories": []}
{"id": "2307.14570", "title": "Physically Plausible 3D Human-Scene Reconstruction from Monocular RGB\n  Image using an Adversarial Learning Approach", "abstract": "  Holistic 3D human-scene reconstruction is a crucial and emerging research\narea in robot perception. A key challenge in holistic 3D human-scene\nreconstruction is to generate a physically plausible 3D scene from a single\nmonocular RGB image. The existing research mainly proposes optimization-based\napproaches for reconstructing the scene from a sequence of RGB frames with\nexplicitly defined physical laws and constraints between different scene\nelements (humans and objects). However, it is hard to explicitly define and\nmodel every physical law in every scenario. This paper proposes using an\nimplicit feature representation of the scene elements to distinguish a\nphysically plausible alignment of humans and objects from an implausible one.\nWe propose using a graph-based holistic representation with an encoded physical\nrepresentation of the scene to analyze the human-object and object-object\ninteractions within the scene. Using this graphical representation, we\nadversarially train our model to learn the feasible alignments of the scene\nelements from the training data itself without explicitly defining the laws and\nconstraints between them. Unlike the existing inference-time optimization-based\napproaches, we use this adversarially trained model to produce a per-frame 3D\nreconstruction of the scene that abides by the physical laws and constraints.\nOur learning-based method achieves comparable 3D reconstruction quality to\nexisting optimization-based holistic human-scene reconstruction methods and\ndoes not need inference time optimization. This makes it better suited when\ncompared to existing methods, for potential use in robotic applications, such\nas robot navigation, etc.\n", "date": "2023-07-27", "categories": []}
{"id": "2307.13282", "title": "High-Resolution Volumetric Reconstruction for Clothed Humans", "abstract": "  We present a novel method for reconstructing clothed humans from a sparse set\nof, e.g., 1 to 6 RGB images. Despite impressive results from recent works\nemploying deep implicit representation, we revisit the volumetric approach and\ndemonstrate that better performance can be achieved with proper system design.\nThe volumetric representation offers significant advantages in leveraging 3D\nspatial context through 3D convolutions, and the notorious quantization error\nis largely negligible with a reasonably large yet affordable volume resolution,\ne.g., 512. To handle memory and computation costs, we propose a sophisticated\ncoarse-to-fine strategy with voxel culling and subspace sparse convolution. Our\nmethod starts with a discretized visual hull to compute a coarse shape and then\nfocuses on a narrow band nearby the coarse shape for refinement. Once the shape\nis reconstructed, we adopt an image-based rendering approach, which computes\nthe colors of surface points by blending input images with learned weights.\nExtensive experimental results show that our method significantly reduces the\nmean point-to-surface (P2S) precision of state-of-the-art methods by more than\n50% to achieve approximately 2mm accuracy with a 512 volume resolution.\nAdditionally, images rendered from our textured model achieve a higher peak\nsignal-to-noise ratio (PSNR) compared to state-of-the-art methods.\n", "date": "2023-07-25", "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": "2210.12003", "title": "HDHumans: A Hybrid Approach for High-fidelity Digital Humans", "abstract": "  Photo-real digital human avatars are of enormous importance in graphics, as\nthey enable immersive communication over the globe, improve gaming and\nentertainment experiences, and can be particularly beneficial for AR and VR\nsettings. However, current avatar generation approaches either fall short in\nhigh-fidelity novel view synthesis, generalization to novel motions,\nreproduction of loose clothing, or they cannot render characters at the high\nresolution offered by modern displays. To this end, we propose HDHumans, which\nis the first method for HD human character synthesis that jointly produces an\naccurate and temporally coherent 3D deforming surface and highly\nphoto-realistic images of arbitrary novel views and of motions not seen at\ntraining time. At the technical core, our method tightly integrates a classical\ndeforming character template with neural radiance fields (NeRF). Our method is\ncarefully designed to achieve a synergy between classical surface deformation\nand NeRF. First, the template guides the NeRF, which allows synthesizing novel\nviews of a highly dynamic and articulated character and even enables the\nsynthesis of novel motions. Second, we also leverage the dense pointclouds\nresulting from NeRF to further improve the deforming surface via 3D-to-3D\nsupervision. We outperform the state of the art quantitatively and\nqualitatively in terms of synthesis quality and resolution, as well as the\nquality of 3D surface reconstruction.\n", "date": "2022-10-21", "categories": []}
{"id": "2307.09153", "title": "OPHAvatars: One-shot Photo-realistic Head Avatars", "abstract": "  We propose a method for synthesizing photo-realistic digital avatars from\nonly one portrait as the reference. Given a portrait, our method synthesizes a\ncoarse talking head video using driving keypoints features. And with the coarse\nvideo, our method synthesizes a coarse talking head avatar with a deforming\nneural radiance field. With rendered images of the coarse avatar, our method\nupdates the low-quality images with a blind face restoration model. With\nupdated images, we retrain the avatar for higher quality. After several\niterations, our method can synthesize a photo-realistic animatable 3D neural\nhead avatar. The motivation of our method is deformable neural radiance field\ncan eliminate the unnatural distortion caused by the image2video method. Our\nmethod outperforms state-of-the-art methods in quantitative and qualitative\nstudies on various subjects.\n", "date": "2023-07-18", "categories": []}
{"id": "2307.05541", "title": "High Fidelity 3D Hand Shape Reconstruction via Scalable Graph Frequency\n  Decomposition", "abstract": "  Despite the impressive performance obtained by recent single-image hand\nmodeling techniques, they lack the capability to capture sufficient details of\nthe 3D hand mesh. This deficiency greatly limits their applications when\nhigh-fidelity hand modeling is required, e.g., personalized hand modeling. To\naddress this problem, we design a frequency split network to generate 3D hand\nmesh using different frequency bands in a coarse-to-fine manner. To capture\nhigh-frequency personalized details, we transform the 3D mesh into the\nfrequency domain, and propose a novel frequency decomposition loss to supervise\neach frequency component. By leveraging such a coarse-to-fine scheme, hand\ndetails that correspond to the higher frequency domain can be preserved. In\naddition, the proposed network is scalable, and can stop the inference at any\nresolution level to accommodate different hardware with varying computational\npowers. To quantitatively evaluate the performance of our method in terms of\nrecovering personalized shape details, we introduce a new evaluation metric\nnamed Mean Signal-to-Noise Ratio (MSNR) to measure the signal-to-noise ratio of\neach mesh frequency component. Extensive experiments demonstrate that our\napproach generates fine-grained details for high-fidelity 3D hand\nreconstruction, and our evaluation metric is more effective for measuring mesh\ndetails compared with traditional metrics.\n", "date": "2023-07-08", "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": "2307.03441", "title": "NOFA: NeRF-based One-shot Facial Avatar Reconstruction", "abstract": "  3D facial avatar reconstruction has been a significant research topic in\ncomputer graphics and computer vision, where photo-realistic rendering and\nflexible controls over poses and expressions are necessary for many related\napplications. Recently, its performance has been greatly improved with the\ndevelopment of neural radiance fields (NeRF). However, most existing NeRF-based\nfacial avatars focus on subject-specific reconstruction and reenactment,\nrequiring multi-shot images containing different views of the specific subject\nfor training, and the learned model cannot generalize to new identities,\nlimiting its further applications. In this work, we propose a one-shot 3D\nfacial avatar reconstruction framework that only requires a single source image\nto reconstruct a high-fidelity 3D facial avatar. For the challenges of lacking\ngeneralization ability and missing multi-view information, we leverage the\ngenerative prior of 3D GAN and develop an efficient encoder-decoder network to\nreconstruct the canonical neural volume of the source image, and further\npropose a compensation network to complement facial details. To enable\nfine-grained control over facial dynamics, we propose a deformation field to\nwarp the canonical volume into driven expressions. Through extensive\nexperimental comparisons, we achieve superior synthesis results compared to\nseveral state-of-the-art methods.\n", "date": "2023-07-07", "categories": []}
{"id": "2307.01468", "title": "Generating Animatable 3D Cartoon Faces from Single Portraits", "abstract": "  With the booming of virtual reality (VR) technology, there is a growing need\nfor customized 3D avatars. However, traditional methods for 3D avatar modeling\nare either time-consuming or fail to retain similarity to the person being\nmodeled. We present a novel framework to generate animatable 3D cartoon faces\nfrom a single portrait image. We first transfer an input real-world portrait to\na stylized cartoon image with a StyleGAN. Then we propose a two-stage\nreconstruction method to recover the 3D cartoon face with detailed texture,\nwhich first makes a coarse estimation based on template models, and then\nrefines the model by non-rigid deformation under landmark supervision. Finally,\nwe propose a semantic preserving face rigging method based on manually created\ntemplates and deformation transfer. Compared with prior arts, qualitative and\nquantitative results show that our method achieves better accuracy, aesthetics,\nand similarity criteria. Furthermore, we demonstrate the capability of\nreal-time facial animation of our 3D model.\n", "date": "2023-07-04", "categories": []}
{"id": "2306.17651", "title": "Implicit 3D Human Mesh Recovery using Consistency with Pose and Shape\n  from Unseen-view", "abstract": "  From an image of a person, we can easily infer the natural 3D pose and shape\nof the person even if ambiguity exists. This is because we have a mental model\nthat allows us to imagine a person's appearance at different viewing directions\nfrom a given image and utilize the consistency between them for inference.\nHowever, existing human mesh recovery methods only consider the direction in\nwhich the image was taken due to their structural limitations. Hence, we\npropose \"Implicit 3D Human Mesh Recovery (ImpHMR)\" that can implicitly imagine\na person in 3D space at the feature-level via Neural Feature Fields. In ImpHMR,\nfeature fields are generated by CNN-based image encoder for a given image.\nThen, the 2D feature map is volume-rendered from the feature field for a given\nviewing direction, and the pose and shape parameters are regressed from the\nfeature. To utilize consistency with pose and shape from unseen-view, if there\nare 3D labels, the model predicts results including the silhouette from an\narbitrary direction and makes it equal to the rotated ground-truth. In the case\nof only 2D labels, we perform self-supervised learning through the constraint\nthat the pose and shape parameters inferred from different directions should be\nthe same. Extensive evaluations show the efficacy of the proposed method.\n", "date": "2023-06-30", "categories": []}
{"id": "2307.00842", "title": "VINECS: Video-based Neural Character Skinning", "abstract": "  Rigging and skinning clothed human avatars is a challenging task and\ntraditionally requires a lot of manual work and expertise. Recent methods\naddressing it either generalize across different characters or focus on\ncapturing the dynamics of a single character observed under different pose\nconfigurations. However, the former methods typically predict solely static\nskinning weights, which perform poorly for highly articulated poses, and the\nlatter ones either require dense 3D character scans in different poses or\ncannot generate an explicit mesh with vertex correspondence over time. To\naddress these challenges, we propose a fully automated approach for creating a\nfully rigged character with pose-dependent skinning weights, which can be\nsolely learned from multi-view video. Therefore, we first acquire a rigged\ntemplate, which is then statically skinned. Next, a coordinate-based MLP learns\na skinning weights field parameterized over the position in a canonical pose\nspace and the respective pose. Moreover, we introduce our pose- and\nview-dependent appearance field allowing us to differentiably render and\nsupervise the posed mesh using multi-view imagery. We show that our approach\noutperforms state-of-the-art while not relying on dense 4D scans.\n", "date": "2023-07-03", "categories": []}
{"id": "2306.16615", "title": "Representation learning of vertex heatmaps for 3D human mesh\n  reconstruction from multi-view images", "abstract": "  This study addresses the problem of 3D human mesh reconstruction from\nmulti-view images. Recently, approaches that directly estimate the skinned\nmulti-person linear model (SMPL)-based human mesh vertices based on volumetric\nheatmap representation from input images have shown good performance. We show\nthat representation learning of vertex heatmaps using an autoencoder helps\nimprove the performance of such approaches. Vertex heatmap autoencoder (VHA)\nlearns the manifold of plausible human meshes in the form of latent codes using\nAMASS, which is a large-scale motion capture dataset. Body code predictor (BCP)\nutilizes the learned body prior from VHA for human mesh reconstruction from\nmulti-view images through latent code-based supervision and transfer of\npretrained weights. According to experiments on Human3.6M and LightStage\ndatasets, the proposed method outperforms previous methods and achieves\nstate-of-the-art human mesh reconstruction performance.\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": "2209.02385", "title": "Inferring Implicit 3D Representations from Human Figures on Pictorial\n  Maps", "abstract": "  In this work, we present an automated workflow to bring human figures, one of\nthe most frequently appearing entities on pictorial maps, to the third\ndimension. Our workflow is based on training data and neural networks for\nsingle-view 3D reconstruction of real humans from photos. We first let a\nnetwork consisting of fully connected layers estimate the depth coordinate of\n2D pose points. The gained 3D pose points are inputted together with 2D masks\nof body parts into a deep implicit surface network to infer 3D signed distance\nfields (SDFs). By assembling all body parts, we derive 2D depth images and body\npart masks of the whole figure for different views, which are fed into a fully\nconvolutional network to predict UV images. These UV images and the texture for\nthe given perspective are inserted into a generative network to inpaint the\ntextures for the other views. The textures are enhanced by a cartoonization\nnetwork and facial details are resynthesized by an autoencoder. Finally, the\ngenerated textures are assigned to the inferred body parts in a ray marcher. We\ntest our workflow with 12 pictorial human figures after having validated\nseveral network configurations. The created 3D models look generally promising,\nespecially when considering the challenges of silhouette-based 3D recovery and\nreal-time rendering of the implicit SDFs. Further improvement is needed to\nreduce gaps between the body parts and to add pictorial details to the\ntextures. Overall, the constructed figures may be used for animation and\nstorytelling in digital 3D maps.\n", "date": "2022-08-30", "categories": []}
{"id": "2306.10350", "title": "MA-NeRF: Motion-Assisted Neural Radiance Fields for Face Synthesis from\n  Sparse Images", "abstract": "  We address the problem of photorealistic 3D face avatar synthesis from sparse\nimages. Existing Parametric models for face avatar reconstruction struggle to\ngenerate details that originate from inputs. Meanwhile, although current\nNeRF-based avatar methods provide promising results for novel view synthesis,\nthey fail to generalize well for unseen expressions. We improve from NeRF and\npropose a novel framework that, by leveraging the parametric 3DMM models, can\nreconstruct a high-fidelity drivable face avatar and successfully handle the\nunseen expressions. At the core of our implementation are structured\ndisplacement feature and semantic-aware learning module. Our structured\ndisplacement feature will introduce the motion prior as an additional\nconstraints and help perform better for unseen expressions, by constructing\ndisplacement volume. Besides, the semantic-aware learning incorporates\nmulti-level prior, e.g., semantic embedding, learnable latent code, to lift the\nperformance to a higher level. Thorough experiments have been doen both\nquantitatively and qualitatively to demonstrate the design of our framework,\nand our method achieves much better results than the current state-of-the-arts.\n", "date": "2023-06-17", "categories": []}
{"id": "2306.14490", "title": "TaiChi Action Capture and Performance Analysis with Multi-view RGB\n  Cameras", "abstract": "  Recent advances in computer vision and deep learning have influenced the\nfield of sports performance analysis for researchers to track and reconstruct\nfreely moving humans without any marker attachment. However, there are few\nworks for vision-based motion capture and intelligent analysis for professional\nTaiChi movement. In this paper, we propose a framework for TaiChi performance\ncapture and analysis with multi-view geometry and artificial intelligence\ntechnology. The main innovative work is as follows: 1) A multi-camera system\nsuitable for TaiChi motion capture is built and the multi-view TaiChi data is\ncollected and processed; 2) A combination of traditional visual method and\nimplicit neural radiance field is proposed to achieve sparse 3D skeleton fusion\nand dense 3D surface reconstruction. 3) The normalization modeling of movement\nsequences is carried out based on motion transfer, so as to realize TaiChi\nperformance analysis for different groups. We have carried out evaluation\nexperiments, and the experimental results have shown the efficiency of our\nmethod.\n", "date": "2023-06-26", "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": "2306.08768", "title": "Generalizable One-shot Neural Head Avatar", "abstract": "  We present a method that reconstructs and animates a 3D head avatar from a\nsingle-view portrait image. Existing methods either involve time-consuming\noptimization for a specific person with multiple images, or they struggle to\nsynthesize intricate appearance details beyond the facial region. To address\nthese limitations, we propose a framework that not only generalizes to unseen\nidentities based on a single-view image without requiring person-specific\noptimization, but also captures characteristic details within and beyond the\nface area (e.g. hairstyle, accessories, etc.). At the core of our method are\nthree branches that produce three tri-planes representing the coarse 3D\ngeometry, detailed appearance of a source image, as well as the expression of a\ntarget image. By applying volumetric rendering to the combination of the three\ntri-planes followed by a super-resolution module, our method yields a high\nfidelity image of the desired identity, expression and pose. Once trained, our\nmodel enables efficient 3D head avatar reconstruction and animation via a\nsingle forward pass through a network. Experiments show that the proposed\napproach generalizes well to unseen validation datasets, surpassing SOTA\nbaseline methods by a large margin on head avatar reconstruction and animation.\n", "date": "2023-06-14", "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": "2306.07437", "title": "Instant Multi-View Head Capture through Learnable Registration", "abstract": "  Existing methods for capturing datasets of 3D heads in dense semantic\ncorrespondence are slow, and commonly address the problem in two separate\nsteps; multi-view stereo (MVS) reconstruction followed by non-rigid\nregistration. To simplify this process, we introduce TEMPEH (Towards Estimation\nof 3D Meshes from Performances of Expressive Heads) to directly infer 3D heads\nin dense correspondence from calibrated multi-view images. Registering datasets\nof 3D scans typically requires manual parameter tuning to find the right\nbalance between accurately fitting the scans surfaces and being robust to\nscanning noise and outliers. Instead, we propose to jointly register a 3D head\ndataset while training TEMPEH. Specifically, during training we minimize a\ngeometric loss commonly used for surface registration, effectively leveraging\nTEMPEH as a regularizer. Our multi-view head inference builds on a volumetric\nfeature representation that samples and fuses features from each view using\ncamera calibration information. To account for partial occlusions and a large\ncapture volume that enables head movements, we use view- and surface-aware\nfeature fusion, and a spatial transformer-based head localization module,\nrespectively. We use raw MVS scans as supervision during training, but, once\ntrained, TEMPEH directly predicts 3D heads in dense correspondence without\nrequiring scans. Predicting one head takes about 0.3 seconds with a median\nreconstruction error of 0.26 mm, 64% lower than the current state-of-the-art.\nThis enables the efficient capture of large datasets containing multiple people\nand diverse facial motions. Code, model, and data are publicly available at\nhttps://tempeh.is.tue.mpg.de.\n", "date": "2023-06-12", "categories": []}
{"id": "2306.05872", "title": "Neural Haircut: Prior-Guided Strand-Based Hair Reconstruction", "abstract": "  Generating realistic human 3D reconstructions using image or video data is\nessential for various communication and entertainment applications. While\nexisting methods achieved impressive results for body and facial regions,\nrealistic hair modeling still remains challenging due to its high mechanical\ncomplexity. This work proposes an approach capable of accurate hair geometry\nreconstruction at a strand level from a monocular video or multi-view images\ncaptured in uncontrolled lighting conditions. Our method has two stages, with\nthe first stage performing joint reconstruction of coarse hair and bust shapes\nand hair orientation using implicit volumetric representations. The second\nstage then estimates a strand-level hair reconstruction by reconciling in a\nsingle optimization process the coarse volumetric constraints with hair strand\nand hairstyle priors learned from the synthetic data. To further increase the\nreconstruction fidelity, we incorporate image-based losses into the fitting\nprocess using a new differentiable renderer. The combined system, named Neural\nHaircut, achieves high realism and personalization of the reconstructed\nhairstyles.\n", "date": "2023-06-09", "categories": []}
{"id": "2306.03576", "title": "Human 3D Avatar Modeling with Implicit Neural Representation: A Brief\n  Survey", "abstract": "  A human 3D avatar is one of the important elements in the metaverse, and the\nmodeling effect directly affects people's visual experience. However, the human\nbody has a complex topology and diverse details, so it is often expensive,\ntime-consuming, and laborious to build a satisfactory model. Recent studies\nhave proposed a novel method, implicit neural representation, which is a\ncontinuous representation method and can describe objects with arbitrary\ntopology at arbitrary resolution. Researchers have applied implicit neural\nrepresentation to human 3D avatar modeling and obtained more excellent results\nthan traditional methods. This paper comprehensively reviews the application of\nimplicit neural representation in human body modeling. First, we introduce\nthree implicit representations of occupancy field, SDF, and NeRF, and make a\nclassification of the literature investigated in this paper. Then the\napplication of implicit modeling methods in the body, hand, and head are\ncompared and analyzed respectively. Finally, we point out the shortcomings of\ncurrent work and provide available suggestions for researchers.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.03847", "title": "Learning Human Mesh Recovery in 3D Scenes", "abstract": "  We present a novel method for recovering the absolute pose and shape of a\nhuman in a pre-scanned scene given a single image. Unlike previous methods that\nperform sceneaware mesh optimization, we propose to first estimate absolute\nposition and dense scene contacts with a sparse 3D CNN, and later enhance a\npretrained human mesh recovery network by cross-attention with the derived 3D\nscene cues. Joint learning on images and scene geometry enables our method to\nreduce the ambiguity caused by depth and occlusion, resulting in more\nreasonable global postures and contacts. Encoding scene-aware cues in the\nnetwork also allows the proposed method to be optimization-free, and opens up\nthe opportunity for real-time applications. The experiments show that the\nproposed network is capable of recovering accurate and physically-plausible\nmeshes by a single forward pass and outperforms state-of-the-art methods in\nterms of both accuracy and speed.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.00547", "title": "AvatarStudio: Text-driven Editing of 3D Dynamic Human Head Avatars", "abstract": "  Capturing and editing full head performances enables the creation of virtual\ncharacters with various applications such as extended reality and media\nproduction. The past few years witnessed a steep rise in the photorealism of\nhuman head avatars. Such avatars can be controlled through different input data\nmodalities, including RGB, audio, depth, IMUs and others. While these data\nmodalities provide effective means of control, they mostly focus on editing the\nhead movements such as the facial expressions, head pose and/or camera\nviewpoint. In this paper, we propose AvatarStudio, a text-based method for\nediting the appearance of a dynamic full head avatar. Our approach builds on\nexisting work to capture dynamic performances of human heads using neural\nradiance field (NeRF) and edits this representation with a text-to-image\ndiffusion model. Specifically, we introduce an optimization strategy for\nincorporating multiple keyframes representing different camera viewpoints and\ntime stamps of a video performance into a single diffusion model. Using this\npersonalized diffusion model, we edit the dynamic NeRF by introducing\nview-and-time-aware Score Distillation Sampling (VT-SDS) following a\nmodel-based guidance approach. Our method edits the full head in a canonical\nspace, and then propagates these edits to remaining time steps via a pretrained\ndeformation network. We evaluate our method visually and numerically via a user\nstudy, and results show that our method outperforms existing approaches. Our\nexperiments validate the design choices of our method and highlight that our\nedits are genuine, personalized, as well as 3D- and time-consistent.\n", "date": "2023-06-01", "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": "2305.14236", "title": "REC-MV: REconstructing 3D Dynamic Cloth from Monocular Videos", "abstract": "  Reconstructing dynamic 3D garment surfaces with open boundaries from\nmonocular videos is an important problem as it provides a practical and\nlow-cost solution for clothes digitization. Recent neural rendering methods\nachieve high-quality dynamic clothed human reconstruction results from\nmonocular video, but these methods cannot separate the garment surface from the\nbody. Moreover, despite existing garment reconstruction methods based on\nfeature curve representation demonstrating impressive results for garment\nreconstruction from a single image, they struggle to generate temporally\nconsistent surfaces for the video input. To address the above limitations, in\nthis paper, we formulate this task as an optimization problem of 3D garment\nfeature curves and surface reconstruction from monocular video. We introduce a\nnovel approach, called REC-MV, to jointly optimize the explicit feature curves\nand the implicit signed distance field (SDF) of the garments. Then the open\ngarment meshes can be extracted via garment template registration in the\ncanonical space. Experiments on multiple casually captured datasets show that\nour approach outperforms existing methods and can produce high-quality dynamic\ngarment surfaces. The source code is available at\nhttps://github.com/GAP-LAB-CUHK-SZ/REC-MV.\n", "date": "2023-05-23", "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": "2305.16411", "title": "ZeroAvatar: Zero-shot 3D Avatar Generation from a Single Image", "abstract": "  Recent advancements in text-to-image generation have enabled significant\nprogress in zero-shot 3D shape generation. This is achieved by score\ndistillation, a methodology that uses pre-trained text-to-image diffusion\nmodels to optimize the parameters of a 3D neural presentation, e.g. Neural\nRadiance Field (NeRF). While showing promising results, existing methods are\noften not able to preserve the geometry of complex shapes, such as human\nbodies. To address this challenge, we present ZeroAvatar, a method that\nintroduces the explicit 3D human body prior to the optimization process.\nSpecifically, we first estimate and refine the parameters of a parametric human\nbody from a single image. Then during optimization, we use the posed parametric\nbody as additional geometry constraint to regularize the diffusion model as\nwell as the underlying density field. Lastly, we propose a UV-guided texture\nregularization term to further guide the completion of texture on invisible\nbody parts. We show that ZeroAvatar significantly enhances the robustness and\n3D consistency of optimization-based image-to-3D avatar generation,\noutperforming existing zero-shot image-to-3D methods.\n", "date": "2023-05-25", "categories": []}
{"id": "2208.11905", "title": "Neural Novel Actor: Learning a Generalized Animatable Neural\n  Representation for Human Actors", "abstract": "  We propose a new method for learning a generalized animatable neural human\nrepresentation from a sparse set of multi-view imagery of multiple persons. The\nlearned representation can be used to synthesize novel view images of an\narbitrary person from a sparse set of cameras, and further animate them with\nthe user's pose control. While existing methods can either generalize to new\npersons or synthesize animations with user control, none of them can achieve\nboth at the same time. We attribute this accomplishment to the employment of a\n3D proxy for a shared multi-person human model, and further the warping of the\nspaces of different poses to a shared canonical pose space, in which we learn a\nneural field and predict the person- and pose-dependent deformations, as well\nas appearance with the features extracted from input images. To cope with the\ncomplexity of the large variations in body shapes, poses, and clothing\ndeformations, we design our neural human model with disentangled geometry and\nappearance. Furthermore, we utilize the image features both at the spatial\npoint and on the surface points of the 3D proxy for predicting person- and\npose-dependent properties. Experiments show that our method significantly\noutperforms the state-of-the-arts on both tasks. The video and code are\navailable at https://talegqz.github.io/neural_novel_actor.\n", "date": "2022-08-25", "categories": []}
{"id": "2305.13353", "title": "RenderMe-360: A Large Digital Asset Library and Benchmarks Towards\n  High-fidelity Head Avatars", "abstract": "  Synthesizing high-fidelity head avatars is a central problem for computer\nvision and graphics. While head avatar synthesis algorithms have advanced\nrapidly, the best ones still face great obstacles in real-world scenarios. One\nof the vital causes is inadequate datasets -- 1) current public datasets can\nonly support researchers to explore high-fidelity head avatars in one or two\ntask directions; 2) these datasets usually contain digital head assets with\nlimited data volume, and narrow distribution over different attributes. In this\npaper, we present RenderMe-360, a comprehensive 4D human head dataset to drive\nadvance in head avatar research. It contains massive data assets, with 243+\nmillion complete head frames, and over 800k video sequences from 500 different\nidentities captured by synchronized multi-view cameras at 30 FPS. It is a\nlarge-scale digital library for head avatars with three key attributes: 1) High\nFidelity: all subjects are captured by 60 synchronized, high-resolution 2K\ncameras in 360 degrees. 2) High Diversity: The collected subjects vary from\ndifferent ages, eras, ethnicities, and cultures, providing abundant materials\nwith distinctive styles in appearance and geometry. Moreover, each subject is\nasked to perform various motions, such as expressions and head rotations, which\nfurther extend the richness of assets. 3) Rich Annotations: we provide\nannotations with different granularities: cameras' parameters, matting, scan,\n2D/3D facial landmarks, FLAME fitting, and text description.\n  Based on the dataset, we build a comprehensive benchmark for head avatar\nresearch, with 16 state-of-the-art methods performed on five main tasks: novel\nview synthesis, novel expression synthesis, hair rendering, hair editing, and\ntalking head generation. Our experiments uncover the strengths and weaknesses\nof current methods. RenderMe-360 opens the door for future exploration in head\navatars.\n", "date": "2023-05-22", "categories": []}
{"id": "2305.11101", "title": "XFormer: Fast and Accurate Monocular 3D Body Capture", "abstract": "  We present XFormer, a novel human mesh and motion capture method that\nachieves real-time performance on consumer CPUs given only monocular images as\ninput. The proposed network architecture contains two branches: a keypoint\nbranch that estimates 3D human mesh vertices given 2D keypoints, and an image\nbranch that makes predictions directly from the RGB image features. At the core\nof our method is a cross-modal transformer block that allows information to\nflow across these two branches by modeling the attention between 2D keypoint\ncoordinates and image spatial features. Our architecture is smartly designed,\nwhich enables us to train on various types of datasets including images with\n2D/3D annotations, images with 3D pseudo labels, and motion capture datasets\nthat do not have associated images. This effectively improves the accuracy and\ngeneralization ability of our system. Built on a lightweight backbone\n(MobileNetV3), our method runs blazing fast (over 30fps on a single CPU core)\nand still yields competitive accuracy. Furthermore, with an HRNet backbone,\nXFormer delivers state-of-the-art performance on Huamn3.6 and 3DPW datasets.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.09641", "title": "FitMe: Deep Photorealistic 3D Morphable Model Avatars", "abstract": "  In this paper, we introduce FitMe, a facial reflectance model and a\ndifferentiable rendering optimization pipeline, that can be used to acquire\nhigh-fidelity renderable human avatars from single or multiple images. The\nmodel consists of a multi-modal style-based generator, that captures facial\nappearance in terms of diffuse and specular reflectance, and a PCA-based shape\nmodel. We employ a fast differentiable rendering process that can be used in an\noptimization pipeline, while also achieving photorealistic facial shading. Our\noptimization process accurately captures both the facial reflectance and shape\nin high-detail, by exploiting the expressivity of the style-based latent\nrepresentation and of our shape model. FitMe achieves state-of-the-art\nreflectance acquisition and identity preservation on single \"in-the-wild\"\nfacial images, while it produces impressive scan-like results, when given\nmultiple unconstrained facial images pertaining to the same identity. In\ncontrast with recent implicit avatar reconstructions, FitMe requires only one\nminute and produces relightable mesh and texture-based avatars, that can be\nused by end-user applications.\n", "date": "2023-05-16", "categories": []}
{"id": "2304.13115", "title": "AVFace: Towards Detailed Audio-Visual 4D Face Reconstruction", "abstract": "  In this work, we present a multimodal solution to the problem of 4D face\nreconstruction from monocular videos. 3D face reconstruction from 2D images is\nan under-constrained problem due to the ambiguity of depth. State-of-the-art\nmethods try to solve this problem by leveraging visual information from a\nsingle image or video, whereas 3D mesh animation approaches rely more on audio.\nHowever, in most cases (e.g. AR/VR applications), videos include both visual\nand speech information. We propose AVFace that incorporates both modalities and\naccurately reconstructs the 4D facial and lip motion of any speaker, without\nrequiring any 3D ground truth for training. A coarse stage estimates the\nper-frame parameters of a 3D morphable model, followed by a lip refinement, and\nthen a fine stage recovers facial geometric details. Due to the temporal audio\nand video information captured by transformer-based modules, our method is\nrobust in cases when either modality is insufficient (e.g. face occlusions).\nExtensive qualitative and quantitative evaluation demonstrates the superiority\nof our method over the current state-of-the-art.\n", "date": "2023-04-25", "categories": []}
{"id": "2305.04789", "title": "AvatarReX: Real-time Expressive Full-body Avatars", "abstract": "  We present AvatarReX, a new method for learning NeRF-based full-body avatars\nfrom video data. The learnt avatar not only provides expressive control of the\nbody, hands and the face together, but also supports real-time animation and\nrendering. To this end, we propose a compositional avatar representation, where\nthe body, hands and the face are separately modeled in a way that the\nstructural prior from parametric mesh templates is properly utilized without\ncompromising representation flexibility. Furthermore, we disentangle the\ngeometry and appearance for each part. With these technical designs, we propose\na dedicated deferred rendering pipeline, which can be executed in real-time\nframerate to synthesize high-quality free-view images. The disentanglement of\ngeometry and appearance also allows us to design a two-pass training strategy\nthat combines volume rendering and surface rendering for network training. In\nthis way, patch-level supervision can be applied to force the network to learn\nsharp appearance details on the basis of geometry estimation. Overall, our\nmethod enables automatic construction of expressive full-body avatars with\nreal-time rendering capability, and can generate photo-realistic images with\ndynamic details for novel body motions and facial expressions.\n", "date": "2023-05-08", "categories": []}
{"id": "2305.03043", "title": "Single-Shot Implicit Morphable Faces with Consistent Texture\n  Parameterization", "abstract": "  There is a growing demand for the accessible creation of high-quality 3D\navatars that are animatable and customizable. Although 3D morphable models\nprovide intuitive control for editing and animation, and robustness for\nsingle-view face reconstruction, they cannot easily capture geometric and\nappearance details. Methods based on neural implicit representations, such as\nsigned distance functions (SDF) or neural radiance fields, approach\nphoto-realism, but are difficult to animate and do not generalize well to\nunseen data. To tackle this problem, we propose a novel method for constructing\nimplicit 3D morphable face models that are both generalizable and intuitive for\nediting. Trained from a collection of high-quality 3D scans, our face model is\nparameterized by geometry, expression, and texture latent codes with a learned\nSDF and explicit UV texture parameterization. Once trained, we can reconstruct\nan avatar from a single in-the-wild image by leveraging the learned prior to\nproject the image into the latent space of our model. Our implicit morphable\nface models can be used to render an avatar from novel views, animate facial\nexpressions by modifying expression codes, and edit textures by directly\npainting on the learned UV-texture maps. We demonstrate quantitatively and\nqualitatively that our method improves upon photo-realism, geometry, and\nexpression accuracy compared to state-of-the-art methods.\n", "date": "2023-05-04", "categories": []}
{"id": "2211.13206", "title": "AvatarMAV: Fast 3D Head Avatar Reconstruction Using Motion-Aware Neural\n  Voxels", "abstract": "  With NeRF widely used for facial reenactment, recent methods can recover\nphoto-realistic 3D head avatar from just a monocular video. Unfortunately, the\ntraining process of the NeRF-based methods is quite time-consuming, as MLP used\nin the NeRF-based methods is inefficient and requires too many iterations to\nconverge. To overcome this problem, we propose AvatarMAV, a fast 3D head avatar\nreconstruction method using Motion-Aware Neural Voxels. AvatarMAV is the first\nto model both the canonical appearance and the decoupled expression motion by\nneural voxels for head avatar. In particular, the motion-aware neural voxels is\ngenerated from the weighted concatenation of multiple 4D tensors. The 4D\ntensors semantically correspond one-to-one with 3DMM expression basis and share\nthe same weights as 3DMM expression coefficients. Benefiting from our novel\nrepresentation, the proposed AvatarMAV can recover photo-realistic head avatars\nin just 5 minutes (implemented with pure PyTorch), which is significantly\nfaster than the state-of-the-art facial reenactment methods. Project page:\nhttps://www.liuyebin.com/avatarmav.\n", "date": "2022-11-23", "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": "2305.01652", "title": "Humans as Light Bulbs: 3D Human Reconstruction from Thermal Reflection", "abstract": "  The relatively hot temperature of the human body causes people to turn into\nlong-wave infrared light sources. Since this emitted light has a larger\nwavelength than visible light, many surfaces in typical scenes act as infrared\nmirrors with strong specular reflections. We exploit the thermal reflections of\na person onto objects in order to locate their position and reconstruct their\npose, even if they are not visible to a normal camera. We propose an\nanalysis-by-synthesis framework that jointly models the objects, people, and\ntheir thermal reflections, which allows us to combine generative models with\ndifferentiable rendering of reflections. Quantitative and qualitative\nexperiments show our approach works in highly challenging cases, such as with\ncurved mirrors or when the person is completely unseen by a normal camera.\n", "date": "2023-05-02", "categories": []}
{"id": "2305.00121", "title": "Learning Locally Editable Virtual Humans", "abstract": "  In this paper, we propose a novel hybrid representation and end-to-end\ntrainable network architecture to model fully editable and customizable neural\navatars. At the core of our work lies a representation that combines the\nmodeling power of neural fields with the ease of use and inherent 3D\nconsistency of skinned meshes. To this end, we construct a trainable feature\ncodebook to store local geometry and texture features on the vertices of a\ndeformable body model, thus exploiting its consistent topology under\narticulation. This representation is then employed in a generative auto-decoder\narchitecture that admits fitting to unseen scans and sampling of realistic\navatars with varied appearances and geometries. Furthermore, our representation\nallows local editing by swapping local features between 3D assets. To verify\nour method for avatar creation and editing, we contribute a new high-quality\ndataset, dubbed CustomHumans, for training and evaluation. Our experiments\nquantitatively and qualitatively show that our method generates diverse\ndetailed avatars and achieves better model fitting performance compared to\nstate-of-the-art methods. Our code and dataset are available at\nhttps://custom-humans.github.io/.\n", "date": "2023-04-28", "categories": []}
{"id": "2305.00646", "title": "Overcoming the Trade-off Between Accuracy and Plausibility in 3D Hand\n  Shape Reconstruction", "abstract": "  Direct mesh fitting for 3D hand shape reconstruction is highly accurate.\nHowever, the reconstructed meshes are prone to artifacts and do not appear as\nplausible hand shapes. Conversely, parametric models like MANO ensure plausible\nhand shapes but are not as accurate as the non-parametric methods. In this\nwork, we introduce a novel weakly-supervised hand shape estimation framework\nthat integrates non-parametric mesh fitting with MANO model in an end-to-end\nfashion. Our joint model overcomes the tradeoff in accuracy and plausibility to\nyield well-aligned and high-quality 3D meshes, especially in challenging\ntwo-hand and hand-object interaction scenarios.\n", "date": "2023-05-01", "categories": []}
{"id": "2305.00936", "title": "Generating Texture for 3D Human Avatar from a Single Image using\n  Sampling and Refinement Networks", "abstract": "  There has been significant progress in generating an animatable 3D human\navatar from a single image. However, recovering texture for the 3D human avatar\nfrom a single image has been relatively less addressed. Because the generated\n3D human avatar reveals the occluded texture of the given image as it moves, it\nis critical to synthesize the occluded texture pattern that is unseen from the\nsource image. To generate a plausible texture map for 3D human avatars, the\noccluded texture pattern needs to be synthesized with respect to the visible\ntexture from the given image. Moreover, the generated texture should align with\nthe surface of the target 3D mesh. In this paper, we propose a texture\nsynthesis method for a 3D human avatar that incorporates geometry information.\nThe proposed method consists of two convolutional networks for the sampling and\nrefining process. The sampler network fills in the occluded regions of the\nsource image and aligns the texture with the surface of the target 3D mesh\nusing the geometry information. The sampled texture is further refined and\nadjusted by the refiner network. To maintain the clear details in the given\nimage, both sampled and refined texture is blended to produce the final texture\nmap. To effectively guide the sampler network to achieve its goal, we designed\na curriculum learning scheme that starts from a simple sampling task and\ngradually progresses to the task where the alignment needs to be considered. We\nconducted experiments to show that our method outperforms previous methods\nqualitatively and quantitatively.\n", "date": "2023-05-01", "categories": []}
{"id": "2304.14299", "title": "A Probabilistic Attention Model with Occlusion-aware Texture Regression\n  for 3D Hand Reconstruction from a Single RGB Image", "abstract": "  Recently, deep learning based approaches have shown promising results in 3D\nhand reconstruction from a single RGB image. These approaches can be roughly\ndivided into model-based approaches, which are heavily dependent on the model's\nparameter space, and model-free approaches, which require large numbers of 3D\nground truths to reduce depth ambiguity and struggle in weakly-supervised\nscenarios. To overcome these issues, we propose a novel probabilistic model to\nachieve the robustness of model-based approaches and reduced dependence on the\nmodel's parameter space of model-free approaches. The proposed probabilistic\nmodel incorporates a model-based network as a prior-net to estimate the prior\nprobability distribution of joints and vertices. An Attention-based Mesh\nVertices Uncertainty Regression (AMVUR) model is proposed to capture\ndependencies among vertices and the correlation between joints and mesh\nvertices to improve their feature representation. We further propose a learning\nbased occlusion-aware Hand Texture Regression model to achieve high-fidelity\ntexture reconstruction. We demonstrate the flexibility of the proposed\nprobabilistic model to be trained in both supervised and weakly-supervised\nscenarios. The experimental results demonstrate our probabilistic model's\nstate-of-the-art accuracy in 3D hand and texture reconstruction from a single\nimage in both training schemes, including in the presence of severe occlusions.\n", "date": "2023-04-27", "categories": []}
{"id": "2304.14401", "title": "ActorsNeRF: Animatable Few-shot Human Rendering with Generalizable NeRFs", "abstract": "  While NeRF-based human representations have shown impressive novel view\nsynthesis results, most methods still rely on a large number of images / views\nfor training. In this work, we propose a novel animatable NeRF called\nActorsNeRF. It is first pre-trained on diverse human subjects, and then adapted\nwith few-shot monocular video frames for a new actor with unseen poses.\nBuilding on previous generalizable NeRFs with parameter sharing using a ConvNet\nencoder, ActorsNeRF further adopts two human priors to capture the large human\nappearance, shape, and pose variations. Specifically, in the encoded feature\nspace, we will first align different human subjects in a category-level\ncanonical space, and then align the same human from different frames in an\ninstance-level canonical space for rendering. We quantitatively and\nqualitatively demonstrate that ActorsNeRF significantly outperforms the\nexisting state-of-the-art on few-shot generalization to new people and poses on\nmultiple datasets. Project Page: https://jitengmu.github.io/ActorsNeRF/\n", "date": "2023-04-27", "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": "2304.12455", "title": "Unsupervised Style-based Explicit 3D Face Reconstruction from Single\n  Image", "abstract": "  Inferring 3D object structures from a single image is an ill-posed task due\nto depth ambiguity and occlusion. Typical resolutions in the literature include\nleveraging 2D or 3D ground truth for supervised learning, as well as imposing\nhand-crafted symmetry priors or using an implicit representation to hallucinate\nnovel viewpoints for unsupervised methods. In this work, we propose a general\nadversarial learning framework for solving Unsupervised 2D to Explicit 3D Style\nTransfer (UE3DST). Specifically, we merge two architectures: the unsupervised\nexplicit 3D reconstruction network of Wu et al.\\ and the Generative Adversarial\nNetwork (GAN) named StarGAN-v2. We experiment across three facial datasets\n(Basel Face Model, 3DFAW and CelebA-HQ) and show that our solution is able to\noutperform well established solutions such as DepthNet in 3D reconstruction and\nPix2NeRF in conditional style transfer, while we also justify the individual\ncontributions of our model components via ablation. In contrast to the\naforementioned baselines, our scheme produces features for explicit 3D\nrendering, which can be manipulated and utilized in downstream tasks.\n", "date": "2023-04-24", "categories": []}
{"id": "2304.11900", "title": "Learning Visibility Field for Detailed 3D Human Reconstruction and\n  Relighting", "abstract": "  Detailed 3D reconstruction and photo-realistic relighting of digital humans\nare essential for various applications. To this end, we propose a novel\nsparse-view 3d human reconstruction framework that closely incorporates the\noccupancy field and albedo field with an additional visibility field--it not\nonly resolves occlusion ambiguity in multiview feature aggregation, but can\nalso be used to evaluate light attenuation for self-shadowed relighting. To\nenhance its training viability and efficiency, we discretize visibility onto a\nfixed set of sample directions and supply it with coupled geometric 3D depth\nfeature and local 2D image feature. We further propose a novel\nrendering-inspired loss, namely TransferLoss, to implicitly enforce the\nalignment between visibility and occupancy field, enabling end-to-end joint\ntraining. Results and extensive experiments demonstrate the effectiveness of\nthe proposed method, as it surpasses state-of-the-art in terms of\nreconstruction accuracy while achieving comparably accurate relighting to\nray-traced ground truth.\n", "date": "2023-04-24", "categories": []}
{"id": "2304.11113", "title": "Implicit Neural Head Synthesis via Controllable Local Deformation Fields", "abstract": "  High-quality reconstruction of controllable 3D head avatars from 2D videos is\nhighly desirable for virtual human applications in movies, games, and\ntelepresence. Neural implicit fields provide a powerful representation to model\n3D head avatars with personalized shape, expressions, and facial parts, e.g.,\nhair and mouth interior, that go beyond the linear 3D morphable model (3DMM).\nHowever, existing methods do not model faces with fine-scale facial features,\nor local control of facial parts that extrapolate asymmetric expressions from\nmonocular videos. Further, most condition only on 3DMM parameters with poor(er)\nlocality, and resolve local features with a global neural field. We build on\npart-based implicit shape models that decompose a global deformation field into\nlocal ones. Our novel formulation models multiple implicit deformation fields\nwith local semantic rig-like control via 3DMM-based parameters, and\nrepresentative facial landmarks. Further, we propose a local control loss and\nattention mask mechanism that promote sparsity of each learned deformation\nfield. Our formulation renders sharper locally controllable nonlinear\ndeformations than previous implicit monocular approaches, especially mouth\ninterior, asymmetric expressions, and facial details.\n", "date": "2023-04-21", "categories": []}
{"id": "2304.10482", "title": "Reconstructing Signing Avatars From Video Using Linguistic Priors", "abstract": "  Sign language (SL) is the primary method of communication for the 70 million\nDeaf people around the world. Video dictionaries of isolated signs are a core\nSL learning tool. Replacing these with 3D avatars can aid learning and enable\nAR/VR applications, improving access to technology and online media. However,\nlittle work has attempted to estimate expressive 3D avatars from SL video;\nocclusion, noise, and motion blur make this task difficult. We address this by\nintroducing novel linguistic priors that are universally applicable to SL and\nprovide constraints on 3D hand pose that help resolve ambiguities within\nisolated signs. Our method, SGNify, captures fine-grained hand pose, facial\nexpression, and body movement fully automatically from in-the-wild monocular SL\nvideos. We evaluate SGNify quantitatively by using a commercial motion-capture\nsystem to compute 3D avatars synchronized with monocular video. SGNify\noutperforms state-of-the-art 3D body-pose- and shape-estimation methods on SL\nvideos. A perceptual study shows that SGNify's 3D reconstructions are\nsignificantly more comprehensible and natural than those of previous methods\nand are on par with the source videos. Code and data are available at\n$\\href{http://sgnify.is.tue.mpg.de}{\\text{sgnify.is.tue.mpg.de}}$.\n", "date": "2023-04-20", "categories": []}
{"id": "2304.09502", "title": "Sampling is Matter: Point-guided 3D Human Mesh Reconstruction", "abstract": "  This paper presents a simple yet powerful method for 3D human mesh\nreconstruction from a single RGB image. Most recently, the non-local\ninteractions of the whole mesh vertices have been effectively estimated in the\ntransformer while the relationship between body parts also has begun to be\nhandled via the graph model. Even though those approaches have shown the\nremarkable progress in 3D human mesh reconstruction, it is still difficult to\ndirectly infer the relationship between features, which are encoded from the 2D\ninput image, and 3D coordinates of each vertex. To resolve this problem, we\npropose to design a simple feature sampling scheme. The key idea is to sample\nfeatures in the embedded space by following the guide of points, which are\nestimated as projection results of 3D mesh vertices (i.e., ground truth). This\nhelps the model to concentrate more on vertex-relevant features in the 2D\nspace, thus leading to the reconstruction of the natural human pose.\nFurthermore, we apply progressive attention masking to precisely estimate local\ninteractions between vertices even under severe occlusions. Experimental\nresults on benchmark datasets show that the proposed method efficiently\nimproves the performance of 3D human mesh reconstruction. The code and model\nare publicly available at: https://github.com/DCVL-3D/PointHMR_release.\n", "date": "2023-04-19", "categories": []}
{"id": "2212.06820", "title": "Structured 3D Features for Reconstructing Controllable Avatars", "abstract": "  We introduce Structured 3D Features, a model based on a novel implicit 3D\nrepresentation that pools pixel-aligned image features onto dense 3D points\nsampled from a parametric, statistical human mesh surface. The 3D points have\nassociated semantics and can move freely in 3D space. This allows for optimal\ncoverage of the person of interest, beyond just the body shape, which in turn,\nadditionally helps modeling accessories, hair, and loose clothing. Owing to\nthis, we present a complete 3D transformer-based attention framework which,\ngiven a single image of a person in an unconstrained pose, generates an\nanimatable 3D reconstruction with albedo and illumination decomposition, as a\nresult of a single end-to-end model, trained semi-supervised, and with no\nadditional postprocessing. We show that our S3F model surpasses the previous\nstate-of-the-art on various tasks, including monocular 3D reconstruction, as\nwell as albedo and shading estimation. Moreover, we show that the proposed\nmethodology allows novel view synthesis, relighting, and re-posing the\nreconstruction, and can naturally be extended to handle multiple input images\n(e.g. different views of a person, or the same view, in different poses, in\nvideo). Finally, we demonstrate the editing capabilities of our model for 3D\nvirtual try-on applications.\n", "date": "2022-12-13", "categories": []}
{"id": "2304.07389", "title": "Shape of You: Precise 3D shape estimations for diverse body types", "abstract": "  This paper presents Shape of You (SoY), an approach to improve the accuracy\nof 3D body shape estimation for vision-based clothing recommendation systems.\nWhile existing methods have successfully estimated 3D poses, there remains a\nlack of work in precise shape estimation, particularly for diverse human\nbodies. To address this gap, we propose two loss functions that can be readily\nintegrated into parametric 3D human reconstruction pipelines. Additionally, we\npropose a test-time optimization routine that further improves quality. Our\nmethod improves over the recent SHAPY method by 17.7% on the challenging SSP-3D\ndataset. We consider our work to be a step towards a more accurate 3D shape\nestimation system that works reliably on diverse body types and holds promise\nfor practical applications in the fashion industry.\n", "date": "2023-04-14", "categories": []}
{"id": "2304.07915", "title": "CAT-NeRF: Constancy-Aware Tx$^2$Former for Dynamic Body Modeling", "abstract": "  This paper addresses the problem of human rendering in the video with\ntemporal appearance constancy. Reconstructing dynamic body shapes with\nvolumetric neural rendering methods, such as NeRF, requires finding the\ncorrespondence of the points in the canonical and observation space, which\ndemands understanding human body shape and motion. Some methods use rigid\ntransformation, such as SE(3), which cannot precisely model each frame's unique\nmotion and muscle movements. Others generate the transformation for each frame\nwith a trainable network, such as neural blend weight field or translation\nvector field, which does not consider the appearance constancy of general body\nshape. In this paper, we propose CAT-NeRF for self-awareness of appearance\nconstancy with Tx$^2$Former, a novel way to combine two Transformer layers, to\nseparate appearance constancy and uniqueness. Appearance constancy models the\ngeneral shape across the video, and uniqueness models the unique patterns for\neach frame. We further introduce a novel Covariance Loss to limit the\ncorrelation between each pair of appearance uniquenesses to ensure the\nframe-unique pattern is maximally captured in appearance uniqueness. We assess\nour method on H36M and ZJU-MoCap and show state-of-the-art performance.\n", "date": "2023-04-16", "categories": []}
{"id": "2212.02761", "title": "Learning Neural Parametric Head Models", "abstract": "  We propose a novel 3D morphable model for complete human heads based on\nhybrid neural fields. At the core of our model lies a neural parametric\nrepresentation that disentangles identity and expressions in disjoint latent\nspaces. To this end, we capture a person's identity in a canonical space as a\nsigned distance field (SDF), and model facial expressions with a neural\ndeformation field. In addition, our representation achieves high-fidelity local\ndetail by introducing an ensemble of local fields centered around facial anchor\npoints. To facilitate generalization, we train our model on a newly-captured\ndataset of over 5200 head scans from 255 different identities using a custom\nhigh-end 3D scanning setup. Our dataset significantly exceeds comparable\nexisting datasets, both with respect to quality and completeness of geometry,\naveraging around 3.5M mesh faces per scan. Finally, we demonstrate that our\napproach outperforms state-of-the-art methods in terms of fitting error and\nreconstruction quality.\n", "date": "2022-12-06", "categories": []}
{"id": "2304.05690", "title": "HybrIK-X: Hybrid Analytical-Neural Inverse Kinematics for Whole-body\n  Mesh Recovery", "abstract": "  Recovering whole-body mesh by inferring the abstract pose and shape\nparameters from visual content can obtain 3D bodies with realistic structures.\nHowever, the inferring process is highly non-linear and suffers from image-mesh\nmisalignment, resulting in inaccurate reconstruction. In contrast, 3D keypoint\nestimation methods utilize the volumetric representation to achieve pixel-level\naccuracy but may predict unrealistic body structures. To address these issues,\nthis paper presents a novel hybrid inverse kinematics solution, HybrIK, that\nintegrates the merits of 3D keypoint estimation and body mesh recovery in a\nunified framework. HybrIK directly transforms accurate 3D joints to body-part\nrotations via twist-and-swing decomposition. The swing rotations are\nanalytically solved with 3D joints, while the twist rotations are derived from\nvisual cues through neural networks. To capture comprehensive whole-body\ndetails, we further develop a holistic framework, HybrIK-X, which enhances\nHybrIK with articulated hands and an expressive face. HybrIK-X is fast and\naccurate by solving the whole-body pose with a one-stage model. Experiments\ndemonstrate that HybrIK and HybrIK-X preserve both the accuracy of 3D joints\nand the realistic structure of the parametric human model, leading to\npixel-aligned whole-body mesh recovery. The proposed method significantly\nsurpasses the state-of-the-art methods on various benchmarks for body-only,\nhand-only, and whole-body scenarios. Code and results can be found at\nhttps://jeffli.site/HybrIK-X/\n", "date": "2023-04-12", "categories": []}
{"id": "2304.04875", "title": "Three Recipes for Better 3D Pseudo-GTs of 3D Human Mesh Estimation in\n  the Wild", "abstract": "  Recovering 3D human mesh in the wild is greatly challenging as in-the-wild\n(ITW) datasets provide only 2D pose ground truths (GTs). Recently, 3D\npseudo-GTs have been widely used to train 3D human mesh estimation networks as\nthe 3D pseudo-GTs enable 3D mesh supervision when training the networks on ITW\ndatasets. However, despite the great potential of the 3D pseudo-GTs, there has\nbeen no extensive analysis that investigates which factors are important to\nmake more beneficial 3D pseudo-GTs. In this paper, we provide three recipes to\nobtain highly beneficial 3D pseudo-GTs of ITW datasets. The main challenge is\nthat only 2D-based weak supervision is allowed when obtaining the 3D\npseudo-GTs. Each of our three recipes addresses the challenge in each aspect:\ndepth ambiguity, sub-optimality of weak supervision, and implausible\narticulation. Experimental results show that simply re-training\nstate-of-the-art networks with our new 3D pseudo-GTs elevates their performance\nto the next level without bells and whistles. The 3D pseudo-GT is publicly\navailable in https://github.com/mks0601/NeuralAnnot_RELEASE.\n", "date": "2023-04-10", "categories": []}
{"id": "2304.04897", "title": "Neural Image-based Avatars: Generalizable Radiance Fields for Human\n  Avatar Modeling", "abstract": "  We present a method that enables synthesizing novel views and novel poses of\narbitrary human performers from sparse multi-view images. A key ingredient of\nour method is a hybrid appearance blending module that combines the advantages\nof the implicit body NeRF representation and image-based rendering. Existing\ngeneralizable human NeRF methods that are conditioned on the body model have\nshown robustness against the geometric variation of arbitrary human performers.\nYet they often exhibit blurry results when generalized onto unseen identities.\nMeanwhile, image-based rendering shows high-quality results when sufficient\nobservations are available, whereas it suffers artifacts in sparse-view\nsettings. We propose Neural Image-based Avatars (NIA) that exploits the best of\nthose two methods: to maintain robustness under new articulations and\nself-occlusions while directly leveraging the available (sparse) source view\ncolors to preserve appearance details of new subject identities. Our hybrid\ndesign outperforms recent methods on both in-domain identity generalization as\nwell as challenging cross-dataset generalization settings. Also, in terms of\nthe pose generalization, our method outperforms even the per-subject optimized\nanimatable NeRF methods. The video results are available at\nhttps://youngjoongunc.github.io/nia\n", "date": "2023-04-10", "categories": []}
{"id": "2304.03903", "title": "High-Fidelity Clothed Avatar Reconstruction from a Single Image", "abstract": "  This paper presents a framework for efficient 3D clothed avatar\nreconstruction. By combining the advantages of the high accuracy of\noptimization-based methods and the efficiency of learning-based methods, we\npropose a coarse-to-fine way to realize a high-fidelity clothed avatar\nreconstruction (CAR) from a single image. At the first stage, we use an\nimplicit model to learn the general shape in the canonical space of a person in\na learning-based way, and at the second stage, we refine the surface detail by\nestimating the non-rigid deformation in the posed space in an optimization way.\nA hyper-network is utilized to generate a good initialization so that the\nconvergence o f the optimization process is greatly accelerated. Extensive\nexperiments on various datasets show that the proposed CAR successfully\nproduces high-fidelity avatars for arbitrarily clothed humans in real scenes.\n", "date": "2023-04-08", "categories": []}
{"id": "2304.03950", "title": "GANHead: Towards Generative Animatable Neural Head Avatars", "abstract": "  To bring digital avatars into people's lives, it is highly demanded to\nefficiently generate complete, realistic, and animatable head avatars. This\ntask is challenging, and it is difficult for existing methods to satisfy all\nthe requirements at once. To achieve these goals, we propose GANHead\n(Generative Animatable Neural Head Avatar), a novel generative head model that\ntakes advantages of both the fine-grained control over the explicit expression\nparameters and the realistic rendering results of implicit representations.\nSpecifically, GANHead represents coarse geometry, fine-gained details and\ntexture via three networks in canonical space to obtain the ability to generate\ncomplete and realistic head avatars. To achieve flexible animation, we define\nthe deformation filed by standard linear blend skinning (LBS), with the learned\ncontinuous pose and expression bases and LBS weights. This allows the avatars\nto be directly animated by FLAME parameters and generalize well to unseen poses\nand expressions. Compared to state-of-the-art (SOTA) methods, GANHead achieves\nsuperior performance on head avatar generation and raw scan fitting.\n", "date": "2023-04-08", "categories": []}
{"id": "2304.04060", "title": "Application of Self-Supervised Learning to MICA Model for Reconstructing\n  Imperfect 3D Facial Structures", "abstract": "  In this study, we emphasize the integration of a pre-trained MICA model with\nan imperfect face dataset, employing a self-supervised learning approach. We\npresent an innovative method for regenerating flawed facial structures,\nyielding 3D printable outputs that effectively support physicians in their\npatient treatment process. Our results highlight the model's capacity for\nconcealing scars and achieving comprehensive facial reconstructions without\ndiscernible scarring. By capitalizing on pre-trained models and necessitating\nonly a few hours of supplementary training, our methodology adeptly devises an\noptimal model for reconstructing damaged and imperfect facial features.\nHarnessing contemporary 3D printing technology, we institute a standardized\nprotocol for fabricating realistic, camouflaging mask models for patients in a\nlaboratory environment.\n", "date": "2023-04-08", "categories": []}
{"id": "2304.02626", "title": "Dynamic Point Fields", "abstract": "  Recent years have witnessed significant progress in the field of neural\nsurface reconstruction. While the extensive focus was put on volumetric and\nimplicit approaches, a number of works have shown that explicit graphics\nprimitives such as point clouds can significantly reduce computational\ncomplexity, without sacrificing the reconstructed surface quality. However,\nless emphasis has been put on modeling dynamic surfaces with point primitives.\nIn this work, we present a dynamic point field model that combines the\nrepresentational benefits of explicit point-based graphics with implicit\ndeformation networks to allow efficient modeling of non-rigid 3D surfaces.\nUsing explicit surface primitives also allows us to easily incorporate\nwell-established constraints such as-isometric-as-possible regularisation.\nWhile learning this deformation model is prone to local optima when trained in\na fully unsupervised manner, we propose to additionally leverage semantic\ninformation such as keypoint dynamics to guide the deformation learning. We\ndemonstrate our model with an example application of creating an expressive\nanimatable human avatar from a collection of 3D scans. Here, previous methods\nmostly rely on variants of the linear blend skinning paradigm, which\nfundamentally limits the expressivity of such models when dealing with complex\ncloth appearances such as long skirts. We show the advantages of our dynamic\npoint field framework in terms of its representational power, learning\nefficiency, and robustness to out-of-distribution novel poses.\n", "date": "2023-04-05", "categories": []}
{"id": "2304.01436", "title": "Learning Personalized High Quality Volumetric Head Avatars from\n  Monocular RGB Videos", "abstract": "  We propose a method to learn a high-quality implicit 3D head avatar from a\nmonocular RGB video captured in the wild. The learnt avatar is driven by a\nparametric face model to achieve user-controlled facial expressions and head\nposes. Our hybrid pipeline combines the geometry prior and dynamic tracking of\na 3DMM with a neural radiance field to achieve fine-grained control and\nphotorealism. To reduce over-smoothing and improve out-of-model expressions\nsynthesis, we propose to predict local features anchored on the 3DMM geometry.\nThese learnt features are driven by 3DMM deformation and interpolated in 3D\nspace to yield the volumetric radiance at a designated query point. We further\nshow that using a Convolutional Neural Network in the UV space is critical in\nincorporating spatial context and producing representative local features.\nExtensive experiments show that we are able to reconstruct high-quality\navatars, with more accurate expression-dependent details, good generalization\nto out-of-training expressions, and quantitatively superior renderings compared\nto other state-of-the-art approaches.\n", "date": "2023-04-04", "categories": []}
{"id": "2304.02001", "title": "MonoHuman: Animatable Human Neural Field from Monocular Video", "abstract": "  Animating virtual avatars with free-view control is crucial for various\napplications like virtual reality and digital entertainment. Previous studies\nhave attempted to utilize the representation power of the neural radiance field\n(NeRF) to reconstruct the human body from monocular videos. Recent works\npropose to graft a deformation network into the NeRF to further model the\ndynamics of the human neural field for animating vivid human motions. However,\nsuch pipelines either rely on pose-dependent representations or fall short of\nmotion coherency due to frame-independent optimization, making it difficult to\ngeneralize to unseen pose sequences realistically. In this paper, we propose a\nnovel framework MonoHuman, which robustly renders view-consistent and\nhigh-fidelity avatars under arbitrary novel poses. Our key insight is to model\nthe deformation field with bi-directional constraints and explicitly leverage\nthe off-the-peg keyframe information to reason the feature correlations for\ncoherent results. Specifically, we first propose a Shared Bidirectional\nDeformation module, which creates a pose-independent generalizable deformation\nfield by disentangling backward and forward deformation correspondences into\nshared skeletal motion weight and separate non-rigid motions. Then, we devise a\nForward Correspondence Search module, which queries the correspondence feature\nof keyframes to guide the rendering network. The rendered results are thus\nmulti-view consistent with high fidelity, even under challenging novel pose\nsettings. Extensive experiments demonstrate the superiority of our proposed\nMonoHuman over state-of-the-art methods.\n", "date": "2023-04-04", "categories": []}
{"id": "2301.09376", "title": "Crowd3D: Towards Hundreds of People Reconstruction from a Single Image", "abstract": "  Image-based multi-person reconstruction in wide-field large scenes is\ncritical for crowd analysis and security alert. However, existing methods\ncannot deal with large scenes containing hundreds of people, which encounter\nthe challenges of large number of people, large variations in human scale, and\ncomplex spatial distribution. In this paper, we propose Crowd3D, the first\nframework to reconstruct the 3D poses, shapes and locations of hundreds of\npeople with global consistency from a single large-scene image. The core of our\napproach is to convert the problem of complex crowd localization into pixel\nlocalization with the help of our newly defined concept, Human-scene Virtual\nInteraction Point (HVIP). To reconstruct the crowd with global consistency, we\npropose a progressive reconstruction network based on HVIP by pre-estimating a\nscene-level camera and a ground plane. To deal with a large number of persons\nand various human sizes, we also design an adaptive human-centric cropping\nscheme. Besides, we contribute a benchmark dataset, LargeCrowd, for crowd\nreconstruction in a large scene. Experimental results demonstrate the\neffectiveness of the proposed method. The code and datasets will be made\npublic.\n", "date": "2023-01-23", "categories": []}
{"id": "2304.00359", "title": "SeSDF: Self-evolved Signed Distance Field for Implicit 3D Clothed Human\n  Reconstruction", "abstract": "  We address the problem of clothed human reconstruction from a single image or\nuncalibrated multi-view images. Existing methods struggle with reconstructing\ndetailed geometry of a clothed human and often require a calibrated setting for\nmulti-view reconstruction. We propose a flexible framework which, by leveraging\nthe parametric SMPL-X model, can take an arbitrary number of input images to\nreconstruct a clothed human model under an uncalibrated setting. At the core of\nour framework is our novel self-evolved signed distance field (SeSDF) module\nwhich allows the framework to learn to deform the signed distance field (SDF)\nderived from the fitted SMPL-X model, such that detailed geometry reflecting\nthe actual clothed human can be encoded for better reconstruction. Besides, we\npropose a simple method for self-calibration of multi-view images via the\nfitted SMPL-X parameters. This lifts the requirement of tedious manual\ncalibration and largely increases the flexibility of our method. Further, we\nintroduce an effective occlusion-aware feature fusion strategy to account for\nthe most useful features to reconstruct the human model. We thoroughly evaluate\nour framework on public benchmarks, demonstrating significant superiority over\nthe state-of-the-arts both qualitatively and quantitatively.\n", "date": "2023-04-01", "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": "2304.01197", "title": "Bringing Telepresence to Every Desk", "abstract": "  In this paper, we work to bring telepresence to every desktop. Unlike\ncommercial systems, personal 3D video conferencing systems must render\nhigh-quality videos while remaining financially and computationally viable for\nthe average consumer. To this end, we introduce a capturing and rendering\nsystem that only requires 4 consumer-grade RGBD cameras and synthesizes\nhigh-quality free-viewpoint videos of users as well as their environments.\n  Experimental results show that our system renders high-quality free-viewpoint\nvideos without using object templates or heavy pre-processing. While not\nreal-time, our system is fast and does not require per-video optimizations.\nMoreover, our system is robust to complex hand gestures and clothing, and it\ncan generalize to new users. This work provides a strong basis for further\noptimization, and it will help bring telepresence to every desk in the near\nfuture. The code and dataset will be made available on our website\nhttps://mcmvmc.github.io/PersonalTelepresence/.\n", "date": "2023-04-03", "categories": []}
{"id": "2211.11734", "title": "PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body\n  Estimation", "abstract": "  We introduce PLIKS (Pseudo-Linear Inverse Kinematic Solver) for\nreconstruction of a 3D mesh of the human body from a single 2D image. Current\ntechniques directly regress the shape, pose, and translation of a parametric\nmodel from an input image through a non-linear mapping with minimal flexibility\nto any external influences. We approach the task as a model-in-the-loop\noptimization problem. PLIKS is built on a linearized formulation of the\nparametric SMPL model. Using PLIKS, we can analytically reconstruct the human\nmodel via 2D pixel-aligned vertices. This enables us with the flexibility to\nuse accurate camera calibration information when available. PLIKS offers an\neasy way to introduce additional constraints such as shape and translation. We\npresent quantitative evaluations which confirm that PLIKS achieves more\naccurate reconstruction with greater than 10% improvement compared to other\nstate-of-the-art methods with respect to the standard 3D human pose and shape\nbenchmarks while also obtaining a reconstruction error improvement of 12.9 mm\non the newer AGORA dataset.\n", "date": "2022-11-21", "categories": []}
{"id": "2302.14434", "title": "A Hierarchical Representation Network for Accurate and Detailed Face\n  Reconstruction from In-The-Wild Images", "abstract": "  Limited by the nature of the low-dimensional representational capacity of\n3DMM, most of the 3DMM-based face reconstruction (FR) methods fail to recover\nhigh-frequency facial details, such as wrinkles, dimples, etc. Some attempt to\nsolve the problem by introducing detail maps or non-linear operations, however,\nthe results are still not vivid. To this end, we in this paper present a novel\nhierarchical representation network (HRN) to achieve accurate and detailed face\nreconstruction from a single image. Specifically, we implement the geometry\ndisentanglement and introduce the hierarchical representation to fulfill\ndetailed face modeling. Meanwhile, 3D priors of facial details are incorporated\nto enhance the accuracy and authenticity of the reconstruction results. We also\npropose a de-retouching module to achieve better decoupling of the geometry and\nappearance. It is noteworthy that our framework can be extended to a multi-view\nfashion by considering detail consistency of different views. Extensive\nexperiments on two single-view and two multi-view FR benchmarks demonstrate\nthat our method outperforms the existing methods in both reconstruction\naccuracy and visual effects. Finally, we introduce a high-quality 3D face\ndataset FaceHD-100 to boost the research of high-fidelity face reconstruction.\nThe project homepage is at https://younglbw.github.io/HRN-homepage/.\n", "date": "2023-02-28", "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": "2303.15892", "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": "2303.16160", "title": "One-Stage 3D Whole-Body Mesh Recovery with Component Aware Transformer", "abstract": "  Whole-body mesh recovery aims to estimate the 3D human body, face, and hands\nparameters from a single image. It is challenging to perform this task with a\nsingle network due to resolution issues, i.e., the face and hands are usually\nlocated in extremely small regions. Existing works usually detect hands and\nfaces, enlarge their resolution to feed in a specific network to predict the\nparameter, and finally fuse the results. While this copy-paste pipeline can\ncapture the fine-grained details of the face and hands, the connections between\ndifferent parts cannot be easily recovered in late fusion, leading to\nimplausible 3D rotation and unnatural pose. In this work, we propose a\none-stage pipeline for expressive whole-body mesh recovery, named OSX, without\nseparate networks for each part. Specifically, we design a Component Aware\nTransformer (CAT) composed of a global body encoder and a local face/hand\ndecoder. The encoder predicts the body parameters and provides a high-quality\nfeature map for the decoder, which performs a feature-level upsample-crop\nscheme to extract high-resolution part-specific features and adopt\nkeypoint-guided deformable attention to estimate hand and face precisely. The\nwhole pipeline is simple yet effective without any manual post-processing and\nnaturally avoids implausible prediction. Comprehensive experiments demonstrate\nthe effectiveness of OSX. Lastly, we build a large-scale Upper-Body dataset\n(UBody) with high-quality 2D and 3D whole-body annotations. It contains persons\nwith partially visible bodies in diverse real-life scenarios to bridge the gap\nbetween the basic task and downstream applications.\n", "date": "2023-03-28", "categories": []}
{"id": "2302.14348", "title": "Im2Hands: Learning Attentive Implicit Representation of Interacting\n  Two-Hand Shapes", "abstract": "  We present Implicit Two Hands (Im2Hands), the first neural implicit\nrepresentation of two interacting hands. Unlike existing methods on two-hand\nreconstruction that rely on a parametric hand model and/or low-resolution\nmeshes, Im2Hands can produce fine-grained geometry of two hands with high\nhand-to-hand and hand-to-image coherency. To handle the shape complexity and\ninteraction context between two hands, Im2Hands models the occupancy volume of\ntwo hands - conditioned on an RGB image and coarse 3D keypoints - by two novel\nattention-based modules responsible for (1) initial occupancy estimation and\n(2) context-aware occupancy refinement, respectively. Im2Hands first learns\nper-hand neural articulated occupancy in the canonical space designed for each\nhand using query-image attention. It then refines the initial two-hand\noccupancy in the posed space to enhance the coherency between the two hand\nshapes using query-anchor attention. In addition, we introduce an optional\nkeypoint refinement module to enable robust two-hand shape estimation from\npredicted hand keypoints in a single-image reconstruction scenario. We\nexperimentally demonstrate the effectiveness of Im2Hands on two-hand\nreconstruction in comparison to related methods, where ours achieves\nstate-of-the-art results. Our code is publicly available at\nhttps://github.com/jyunlee/Im2Hands.\n", "date": "2023-02-28", "categories": []}
{"id": "2303.14092", "title": "NeuFace: Realistic 3D Neural Face Rendering from Multi-view Images", "abstract": "  Realistic face rendering from multi-view images is beneficial to various\ncomputer vision and graphics applications. Due to the complex spatially-varying\nreflectance properties and geometry characteristics of faces, however, it\nremains challenging to recover 3D facial representations both faithfully and\nefficiently in the current studies. This paper presents a novel 3D face\nrendering model, namely NeuFace, to learn accurate and physically-meaningful\nunderlying 3D representations by neural rendering techniques. It naturally\nincorporates the neural BRDFs into physically based rendering, capturing\nsophisticated facial geometry and appearance clues in a collaborative manner.\nSpecifically, we introduce an approximated BRDF integration and a simple yet\nnew low-rank prior, which effectively lower the ambiguities and boost the\nperformance of the facial BRDFs. Extensive experiments demonstrate the\nsuperiority of NeuFace in human face rendering, along with a decent\ngeneralization ability to common objects.\n", "date": "2023-03-24", "categories": []}
{"id": "2303.14471", "title": "HQ3DAvatar: High Quality Controllable 3D Head Avatar", "abstract": "  Multi-view volumetric rendering techniques have recently shown great\npotential in modeling and synthesizing high-quality head avatars. A common\napproach to capture full head dynamic performances is to track the underlying\ngeometry using a mesh-based template or 3D cube-based graphics primitives.\nWhile these model-based approaches achieve promising results, they often fail\nto learn complex geometric details such as the mouth interior, hair, and\ntopological changes over time. This paper presents a novel approach to building\nhighly photorealistic digital head avatars. Our method learns a canonical space\nvia an implicit function parameterized by a neural network. It leverages\nmultiresolution hash encoding in the learned feature space, allowing for\nhigh-quality, faster training and high-resolution rendering. At test time, our\nmethod is driven by a monocular RGB video. Here, an image encoder extracts\nface-specific features that also condition the learnable canonical space. This\nencourages deformation-dependent texture variations during training. We also\npropose a novel optical flow based loss that ensures correspondences in the\nlearned canonical space, thus encouraging artifact-free and temporally\nconsistent renderings. We show results on challenging facial expressions and\nshow free-viewpoint renderings at interactive real-time rates for medium image\nresolutions. Our method outperforms all existing approaches, both visually and\nnumerically. We will release our multiple-identity dataset to encourage further\nresearch. Our Project page is available at:\nhttps://vcai.mpi-inf.mpg.de/projects/HQ3DAvatar/\n", "date": "2023-03-25", "categories": []}
{"id": "2303.15108", "title": "High-fidelity 3D Human Digitization from Single 2K Resolution Images", "abstract": "  High-quality 3D human body reconstruction requires high-fidelity and\nlarge-scale training data and appropriate network design that effectively\nexploits the high-resolution input images. To tackle these problems, we propose\na simple yet effective 3D human digitization method called 2K2K, which\nconstructs a large-scale 2K human dataset and infers 3D human models from 2K\nresolution images. The proposed method separately recovers the global shape of\na human and its details. The low-resolution depth network predicts the global\nstructure from a low-resolution image, and the part-wise image-to-normal\nnetwork predicts the details of the 3D human body structure. The\nhigh-resolution depth network merges the global 3D shape and the detailed\nstructures to infer the high-resolution front and back side depth maps.\nFinally, an off-the-shelf mesh generator reconstructs the full 3D human model,\nwhich are available at https://github.com/SangHunHan92/2K2K. In addition, we\nalso provide 2,050 3D human models, including texture maps, 3D joints, and SMPL\nparameters for research purposes. In experiments, we demonstrate competitive\nperformance over the recent works on various datasets.\n", "date": "2023-03-27", "categories": []}
{"id": "2211.13874", "title": "FFHQ-UV: Normalized Facial UV-Texture Dataset for 3D Face Reconstruction", "abstract": "  We present a large-scale facial UV-texture dataset that contains over 50,000\nhigh-quality texture UV-maps with even illuminations, neutral expressions, and\ncleaned facial regions, which are desired characteristics for rendering\nrealistic 3D face models under different lighting conditions. The dataset is\nderived from a large-scale face image dataset namely FFHQ, with the help of our\nfully automatic and robust UV-texture production pipeline. Our pipeline\nutilizes the recent advances in StyleGAN-based facial image editing approaches\nto generate multi-view normalized face images from single-image inputs. An\nelaborated UV-texture extraction, correction, and completion procedure is then\napplied to produce high-quality UV-maps from the normalized face images.\nCompared with existing UV-texture datasets, our dataset has more diverse and\nhigher-quality texture maps. We further train a GAN-based texture decoder as\nthe nonlinear texture basis for parametric fitting based 3D face\nreconstruction. Experiments show that our method improves the reconstruction\naccuracy over state-of-the-art approaches, and more importantly, produces\nhigh-quality texture maps that are ready for realistic renderings. The dataset,\ncode, and pre-trained texture decoder are publicly available at\nhttps://github.com/csbhr/FFHQ-UV.\n", "date": "2022-11-25", "categories": []}
{"id": "2211.12499", "title": "Instant Volumetric Head Avatars", "abstract": "  We present Instant Volumetric Head Avatars (INSTA), a novel approach for\nreconstructing photo-realistic digital avatars instantaneously. INSTA models a\ndynamic neural radiance field based on neural graphics primitives embedded\naround a parametric face model. Our pipeline is trained on a single monocular\nRGB portrait video that observes the subject under different expressions and\nviews. While state-of-the-art methods take up to several days to train an\navatar, our method can reconstruct a digital avatar in less than 10 minutes on\nmodern GPU hardware, which is orders of magnitude faster than previous\nsolutions. In addition, it allows for the interactive rendering of novel poses\nand expressions. By leveraging the geometry prior of the underlying parametric\nface model, we demonstrate that INSTA extrapolates to unseen poses. In\nquantitative and qualitative studies on various subjects, INSTA outperforms\nstate-of-the-art methods regarding rendering quality and training time.\n", "date": "2022-11-22", "categories": []}
{"id": "2212.07422", "title": "ECON: Explicit Clothed humans Optimized via Normal integration", "abstract": "  The combination of deep learning, artist-curated scans, and Implicit\nFunctions (IF), is enabling the creation of detailed, clothed, 3D humans from\nimages. However, existing methods are far from perfect. IF-based methods\nrecover free-form geometry, but produce disembodied limbs or degenerate shapes\nfor novel poses or clothes. To increase robustness for these cases, existing\nwork uses an explicit parametric body model to constrain surface\nreconstruction, but this limits the recovery of free-form surfaces such as\nloose clothing that deviates from the body. What we want is a method that\ncombines the best properties of implicit representation and explicit body\nregularization. To this end, we make two key observations: (1) current networks\nare better at inferring detailed 2D maps than full-3D surfaces, and (2) a\nparametric model can be seen as a \"canvas\" for stitching together detailed\nsurface patches. Based on these, our method, ECON, has three main steps: (1) It\ninfers detailed 2D normal maps for the front and back side of a clothed person.\n(2) From these, it recovers 2.5D front and back surfaces, called d-BiNI, that\nare equally detailed, yet incomplete, and registers these w.r.t. each other\nwith the help of a SMPL-X body mesh recovered from the image. (3) It \"inpaints\"\nthe missing geometry between d-BiNI surfaces. If the face and hands are noisy,\nthey can optionally be replaced with the ones of SMPL-X. As a result, ECON\ninfers high-fidelity 3D humans even in loose clothes and challenging poses.\nThis goes beyond previous methods, according to the quantitative evaluation on\nthe CAPE and Renderpeople datasets. Perceptual studies also show that ECON's\nperceived realism is better by a large margin. Code and models are available\nfor research purposes at econ.is.tue.mpg.de\n", "date": "2022-12-14", "categories": []}
{"id": "2303.12965", "title": "Efficient Meshy Neural Fields for Animatable Human Avatars", "abstract": "  Efficiently digitizing high-fidelity animatable human avatars from videos is\na challenging and active research topic. Recent volume rendering-based neural\nrepresentations open a new way for human digitization with their friendly\nusability and photo-realistic reconstruction quality. However, they are\ninefficient for long optimization times and slow inference speed; their\nimplicit nature results in entangled geometry, materials, and dynamics of\nhumans, which are hard to edit afterward. Such drawbacks prevent their direct\napplicability to downstream applications, especially the prominent\nrasterization-based graphic ones. We present EMA, a method that Efficiently\nlearns Meshy neural fields to reconstruct animatable human Avatars. It jointly\noptimizes explicit triangular canonical mesh, spatial-varying material, and\nmotion dynamics, via inverse rendering in an end-to-end fashion. Each above\ncomponent is derived from separate neural fields, relaxing the requirement of a\ntemplate, or rigging. The mesh representation is highly compatible with the\nefficient rasterization-based renderer, thus our method only takes about an\nhour of training and can render in real-time. Moreover, only minutes of\noptimization is enough for plausible reconstruction results. The\ndisentanglement of meshes enables direct downstream applications. Extensive\nexperiments illustrate the very competitive performance and significant speed\nboost against previous methods. We also showcase applications including novel\npose synthesis, material editing, and relighting. The project page:\nhttps://xk-huang.github.io/ema/.\n", "date": "2023-03-23", "categories": []}
{"id": "2211.12782", "title": "Hand Avatar: Free-Pose Hand Animation and Rendering from Monocular Video", "abstract": "  We present HandAvatar, a novel representation for hand animation and\nrendering, which can generate smoothly compositional geometry and\nself-occlusion-aware texture. Specifically, we first develop a MANO-HD model as\na high-resolution mesh topology to fit personalized hand shapes. Sequentially,\nwe decompose hand geometry into per-bone rigid parts, and then re-compose\npaired geometry encodings to derive an across-part consistent occupancy field.\nAs for texture modeling, we propose a self-occlusion-aware shading field\n(SelF). In SelF, drivable anchors are paved on the MANO-HD surface to record\nalbedo information under a wide variety of hand poses. Moreover, directed soft\noccupancy is designed to describe the ray-to-surface relation, which is\nleveraged to generate an illumination field for the disentanglement of\npose-independent albedo and pose-dependent illumination. Trained from monocular\nvideo data, our HandAvatar can perform free-pose hand animation and rendering\nwhile at the same time achieving superior appearance fidelity. We also\ndemonstrate that HandAvatar provides a route for hand appearance editing.\nProject website: https://seanchenxy.github.io/HandAvatarWeb.\n", "date": "2022-11-23", "categories": []}
{"id": "2303.11898", "title": "Real-time volumetric rendering of dynamic humans", "abstract": "  We present a method for fast 3D reconstruction and real-time rendering of\ndynamic humans from monocular videos with accompanying parametric body fits.\nOur method can reconstruct a dynamic human in less than 3h using a single GPU,\ncompared to recent state-of-the-art alternatives that take up to 72h. These\nspeedups are obtained by using a lightweight deformation model solely based on\nlinear blend skinning, and an efficient factorized volumetric representation\nfor modeling the shape and color of the person in canonical pose. Moreover, we\npropose a novel local ray marching rendering which, by exploiting standard GPU\nhardware and without any baking or conversion of the radiance field, allows\nvisualizing the neural human on a mobile VR device at 40 frames per second with\nminimal loss of visual quality. Our experimental evaluation shows superior or\ncompetitive results with state-of-the art methods while obtaining large\ntraining speedup, using a simple model, and achieving real-time rendering.\n", "date": "2023-03-21", "categories": []}
{"id": "2211.13357", "title": "MPT: Mesh Pre-Training with Transformers for Human Pose and Mesh\n  Reconstruction", "abstract": "  Traditional methods of reconstructing 3D human pose and mesh from single\nimages rely on paired image-mesh datasets, which can be difficult and expensive\nto obtain. Due to this limitation, model scalability is constrained as well as\nreconstruction performance. Towards addressing the challenge, we introduce Mesh\nPre-Training (MPT), an effective pre-training strategy that leverages large\namounts of MoCap data to effectively perform pre-training at scale. We\nintroduce the use of MoCap-generated heatmaps as input representations to the\nmesh regression transformer and propose a Masked Heatmap Modeling approach for\nimproving pre-training performance. This study demonstrates that pre-training\nusing the proposed MPT allows our models to perform effective inference without\nrequiring fine-tuning. We further show that fine-tuning the pre-trained MPT\nmodel considerably improves the accuracy of human mesh reconstruction from\nsingle images. Experimental results show that MPT outperforms previous\nstate-of-the-art methods on Human3.6M and 3DPW datasets. As a further\napplication, we benchmark and study MPT on the task of 3D hand reconstruction,\nshowing that our generic pre-training scheme generalizes well to hand pose\nestimation and achieves promising reconstruction performance.\n", "date": "2022-11-24", "categories": []}
{"id": "2212.01160", "title": "High-Res Facial Appearance Capture from Polarized Smartphone Images", "abstract": "  We propose a novel method for high-quality facial texture reconstruction from\nRGB images using a novel capturing routine based on a single smartphone which\nwe equip with an inexpensive polarization foil. Specifically, we turn the\nflashlight into a polarized light source and add a polarization filter on top\nof the camera. Leveraging this setup, we capture the face of a subject with\ncross-polarized and parallel-polarized light. For each subject, we record two\nshort sequences in a dark environment under flash illumination with different\nlight polarization using the modified smartphone. Based on these observations,\nwe reconstruct an explicit surface mesh of the face using structure from\nmotion. We then exploit the camera and light co-location within a\ndifferentiable renderer to optimize the facial textures using an\nanalysis-by-synthesis approach. Our method optimizes for high-resolution normal\ntextures, diffuse albedo, and specular albedo using a coarse-to-fine\noptimization scheme. We show that the optimized textures can be used in a\nstandard rendering pipeline to synthesize high-quality photo-realistic 3D\ndigital humans in novel environments.\n", "date": "2022-12-02", "categories": []}
{"id": "2303.08808", "title": "Mesh Strikes Back: Fast and Efficient Human Reconstruction from RGB\n  videos", "abstract": "  Human reconstruction and synthesis from monocular RGB videos is a challenging\nproblem due to clothing, occlusion, texture discontinuities and sharpness, and\nframespecific pose changes. Many methods employ deferred rendering, NeRFs and\nimplicit methods to represent clothed humans, on the premise that mesh-based\nrepresentations cannot capture complex clothing and textures from RGB,\nsilhouettes, and keypoints alone. We provide a counter viewpoint to this\nfundamental premise by optimizing a SMPL+D mesh and an efficient,\nmulti-resolution texture representation using only RGB images, binary\nsilhouettes and sparse 2D keypoints. Experimental results demonstrate that our\napproach is more capable of capturing geometric details compared to visual\nhull, mesh-based methods. We show competitive novel view synthesis and\nimprovements in novel pose synthesis compared to NeRF-based methods, which\nintroduce noticeable, unwanted artifacts. By restricting the solution space to\nthe SMPL+D model combined with differentiable rendering, we obtain dramatic\nspeedups in compute, training times (up to 24x) and inference times (up to\n192x). Our method therefore can be used as is or as a fast initialization to\nNeRF-based methods.\n", "date": "2023-03-15", "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": "2302.14683", "title": "IntrinsicNGP: Intrinsic Coordinate based Hash Encoding for Human NeRF", "abstract": "  Recently, many works have been proposed to utilize the neural radiance field\nfor novel view synthesis of human performers. However, most of these methods\nrequire hours of training, making them difficult for practical use. To address\nthis challenging problem, we propose IntrinsicNGP, which can train from scratch\nand achieve high-fidelity results in few minutes with videos of a human\nperformer. To achieve this target, we introduce a continuous and optimizable\nintrinsic coordinate rather than the original explicit Euclidean coordinate in\nthe hash encoding module of instant-NGP. With this novel intrinsic coordinate,\nIntrinsicNGP can aggregate inter-frame information for dynamic objects with the\nhelp of proxy geometry shapes. Moreover, the results trained with the given\nrough geometry shapes can be further refined with an optimizable offset field\nbased on the intrinsic coordinate.Extensive experimental results on several\ndatasets demonstrate the effectiveness and efficiency of IntrinsicNGP. We also\nillustrate our approach's ability to edit the shape of reconstructed subjects.\n", "date": "2023-02-28", "categories": []}
{"id": "2303.05652", "title": "GATOR: Graph-Aware Transformer with Motion-Disentangled Regression for\n  Human Mesh Recovery from a 2D Pose", "abstract": "  3D human mesh recovery from a 2D pose plays an important role in various\napplications. However, it is hard for existing methods to simultaneously\ncapture the multiple relations during the evolution from skeleton to mesh,\nincluding joint-joint, joint-vertex and vertex-vertex relations, which often\nleads to implausible results. To address this issue, we propose a novel\nsolution, called GATOR, that contains an encoder of Graph-Aware Transformer\n(GAT) and a decoder with Motion-Disentangled Regression (MDR) to explore these\nmultiple relations. Specifically, GAT combines a GCN and a graph-aware\nself-attention in parallel to capture physical and hidden joint-joint\nrelations. Furthermore, MDR models joint-vertex and vertex-vertex interactions\nto explore joint and vertex relations. Based on the clustering characteristics\nof vertex offset fields, MDR regresses the vertices by composing the predicted\nbase motions. Extensive experiments show that GATOR achieves state-of-the-art\nperformance on two challenging benchmarks.\n", "date": "2023-03-10", "categories": []}
{"id": "2303.05835", "title": "You Only Train Once: Multi-Identity Free-Viewpoint Neural Human\n  Rendering from Monocular Videos", "abstract": "  We introduce You Only Train Once (YOTO), a dynamic human generation\nframework, which performs free-viewpoint rendering of different human\nidentities with distinct motions, via only one-time training from monocular\nvideos. Most prior works for the task require individualized optimization for\neach input video that contains a distinct human identity, leading to a\nsignificant amount of time and resources for the deployment, thereby impeding\nthe scalability and the overall application potential of the system. In this\npaper, we tackle this problem by proposing a set of learnable identity codes to\nexpand the capability of the framework for multi-identity free-viewpoint\nrendering, and an effective pose-conditioned code query mechanism to finely\nmodel the pose-dependent non-rigid motions. YOTO optimizes neural radiance\nfields (NeRF) by utilizing designed identity codes to condition the model for\nlearning various canonical T-pose appearances in a single shared volumetric\nrepresentation. Besides, our joint learning of multiple identities within a\nunified model incidentally enables flexible motion transfer in high-quality\nphoto-realistic renderings for all learned appearances. This capability expands\nits potential use in important applications, including Virtual Reality. We\npresent extensive experimental results on ZJU-MoCap and PeopleSnapshot to\nclearly demonstrate the effectiveness of our proposed model. YOTO shows\nstate-of-the-art performance on all evaluation metrics while showing\nsignificant benefits in training and inference efficiency as well as rendering\nquality. The code and model will be made publicly available soon.\n", "date": "2023-03-10", "categories": []}
{"id": "2303.04805", "title": "X-Avatar: Expressive Human Avatars", "abstract": "  We present X-Avatar, a novel avatar model that captures the full\nexpressiveness of digital humans to bring about life-like experiences in\ntelepresence, AR/VR and beyond. Our method models bodies, hands, facial\nexpressions and appearance in a holistic fashion and can be learned from either\nfull 3D scans or RGB-D data. To achieve this, we propose a part-aware learned\nforward skinning module that can be driven by the parameter space of SMPL-X,\nallowing for expressive animation of X-Avatars. To efficiently learn the neural\nshape and deformation fields, we propose novel part-aware sampling and\ninitialization strategies. This leads to higher fidelity results, especially\nfor smaller body parts while maintaining efficient training despite increased\nnumber of articulated bones. To capture the appearance of the avatar with\nhigh-frequency details, we extend the geometry and deformation fields with a\ntexture network that is conditioned on pose, facial expression, geometry and\nthe normals of the deformed surface. We show experimentally that our method\noutperforms strong baselines in both data domains both quantitatively and\nqualitatively on the animation task. To facilitate future research on\nexpressive avatars we contribute a new dataset, called X-Humans, containing 233\nsequences of high-quality textured scans from 20 participants, totalling 35,500\ndata frames.\n", "date": "2023-03-08", "categories": []}
{"id": "2303.03471", "title": "Refining 3D Human Texture Estimation from a Single Image", "abstract": "  Estimating 3D human texture from a single image is essential in graphics and\nvision. It requires learning a mapping function from input images of humans\nwith diverse poses into the parametric (UV) space and reasonably hallucinating\ninvisible parts. To achieve a high-quality 3D human texture estimation, we\npropose a framework that adaptively samples the input by a deformable\nconvolution where offsets are learned via a deep neural network. Additionally,\nwe describe a novel cycle consistency loss that improves view generalization.\nWe further propose to train our framework with an uncertainty-based pixel-level\nimage reconstruction loss, which enhances color fidelity. We compare our method\nagainst the state-of-the-art approaches and show significant qualitative and\nquantitative improvements.\n", "date": "2023-03-06", "categories": []}
{"id": "2211.15064", "title": "High-fidelity Facial Avatar Reconstruction from Monocular Video with\n  Generative Priors", "abstract": "  High-fidelity facial avatar reconstruction from a monocular video is a\nsignificant research problem in computer graphics and computer vision.\nRecently, Neural Radiance Field (NeRF) has shown impressive novel view\nrendering results and has been considered for facial avatar reconstruction.\nHowever, the complex facial dynamics and missing 3D information in monocular\nvideos raise significant challenges for faithful facial reconstruction. In this\nwork, we propose a new method for NeRF-based facial avatar reconstruction that\nutilizes 3D-aware generative prior. Different from existing works that depend\non a conditional deformation field for dynamic modeling, we propose to learn a\npersonalized generative prior, which is formulated as a local and low\ndimensional subspace in the latent space of 3D-GAN. We propose an efficient\nmethod to construct the personalized generative prior based on a small set of\nfacial images of a given individual. After learning, it allows for\nphoto-realistic rendering with novel views and the face reenactment can be\nrealized by performing navigation in the latent space. Our proposed method is\napplicable for different driven signals, including RGB images, 3DMM\ncoefficients, and audios. Compared with existing works, we obtain superior\nnovel view synthesis results and faithfully face reenactment performance.\n", "date": "2022-11-28", "categories": []}
{"id": "2302.10518", "title": "USR: Unsupervised Separated 3D Garment and Human Reconstruction via\n  Geometry and Semantic Consistency", "abstract": "  Dressed people reconstruction from images is a popular task with promising\napplications in the creative media and game industry. However, most existing\nmethods reconstruct the human body and garments as a whole with the supervision\nof 3D models, which hinders the downstream interaction tasks and requires\nhard-to-obtain data. To address these issues, we propose an unsupervised\nseparated 3D garments and human reconstruction model (USR), which reconstructs\nthe human body and authentic textured clothes in layers without 3D models. More\nspecifically, our method proposes a generalized surface-aware neural radiance\nfield to learn the mapping between sparse multi-view images and geometries of\nthe dressed people. Based on the full geometry, we introduce a Semantic and\nConfidence Guided Separation strategy (SCGS) to detect, segment, and\nreconstruct the clothes layer, leveraging the consistency between 2D semantic\nand 3D geometry. Moreover, we propose a Geometry Fine-tune Module to smooth\nedges. Extensive experiments on our dataset show that comparing with\nstate-of-the-art methods, USR achieves improvements on both geometry and\nappearance reconstruction while supporting generalizing to unseen people in\nreal time. Besides, we also introduce SMPL-D model to show the benefit of the\nseparated modeling of clothes and the human body that allows swapping clothes\nand virtual try-on.\n", "date": "2023-02-21", "categories": []}
{"id": "2303.00298", "title": "Capturing the motion of every joint: 3D human pose and shape estimation\n  with independent tokens", "abstract": "  In this paper we present a novel method to estimate 3D human pose and shape\nfrom monocular videos. This task requires directly recovering pixel-alignment\n3D human pose and body shape from monocular images or videos, which is\nchallenging due to its inherent ambiguity. To improve precision, existing\nmethods highly rely on the initialized mean pose and shape as prior estimates\nand parameter regression with an iterative error feedback manner. In addition,\nvideo-based approaches model the overall change over the image-level features\nto temporally enhance the single-frame feature, but fail to capture the\nrotational motion at the joint level, and cannot guarantee local temporal\nconsistency. To address these issues, we propose a novel Transformer-based\nmodel with a design of independent tokens. First, we introduce three types of\ntokens independent of the image feature: \\textit{joint rotation tokens, shape\ntoken, and camera token}. By progressively interacting with image features\nthrough Transformer layers, these tokens learn to encode the prior knowledge of\nhuman 3D joint rotations, body shape, and position information from large-scale\ndata, and are updated to estimate SMPL parameters conditioned on a given image.\nSecond, benefiting from the proposed token-based representation, we further use\na temporal model to focus on capturing the rotational temporal information of\neach joint, which is empirically conducive to preventing large jitters in local\nparts. Despite being conceptually simple, the proposed method attains superior\nperformances on the 3DPW and Human3.6M datasets. Using ResNet-50 and\nTransformer architectures, it obtains 42.0 mm error on the PA-MPJPE metric of\nthe challenging 3DPW, outperforming state-of-the-art counterparts by a large\nmargin. Code will be publicly available at\nhttps://github.com/yangsenius/INT_HMR_Model\n", "date": "2023-03-01", "categories": []}
{"id": "2212.08377", "title": "PointAvatar: Deformable Point-based Head Avatars from Videos", "abstract": "  The ability to create realistic, animatable and relightable head avatars from\ncasual video sequences would open up wide ranging applications in communication\nand entertainment. Current methods either build on explicit 3D morphable meshes\n(3DMM) or exploit neural implicit representations. The former are limited by\nfixed topology, while the latter are non-trivial to deform and inefficient to\nrender. Furthermore, existing approaches entangle lighting in the color\nestimation, thus they are limited in re-rendering the avatar in new\nenvironments. In contrast, we propose PointAvatar, a deformable point-based\nrepresentation that disentangles the source color into intrinsic albedo and\nnormal-dependent shading. We demonstrate that PointAvatar bridges the gap\nbetween existing mesh- and implicit representations, combining high-quality\ngeometry and appearance with topological flexibility, ease of deformation and\nrendering efficiency. We show that our method is able to generate animatable 3D\navatars using monocular videos from multiple sources including hand-held\nsmartphones, laptop webcams and internet videos, achieving state-of-the-art\nquality in challenging cases where previous methods fail, e.g., thin hair\nstrands, while being significantly more efficient in training than competing\nmethods.\n", "date": "2022-12-16", "categories": []}
{"id": "2302.11566", "title": "Vid2Avatar: 3D Avatar Reconstruction from Videos in the Wild via\n  Self-supervised Scene Decomposition", "abstract": "  We present Vid2Avatar, a method to learn human avatars from monocular\nin-the-wild videos. Reconstructing humans that move naturally from monocular\nin-the-wild videos is difficult. Solving it requires accurately separating\nhumans from arbitrary backgrounds. Moreover, it requires reconstructing\ndetailed 3D surface from short video sequences, making it even more\nchallenging. Despite these challenges, our method does not require any\ngroundtruth supervision or priors extracted from large datasets of clothed\nhuman scans, nor do we rely on any external segmentation modules. Instead, it\nsolves the tasks of scene decomposition and surface reconstruction directly in\n3D by modeling both the human and the background in the scene jointly,\nparameterized via two separate neural fields. Specifically, we define a\ntemporally consistent human representation in canonical space and formulate a\nglobal optimization over the background model, the canonical human shape and\ntexture, and per-frame human pose parameters. A coarse-to-fine sampling\nstrategy for volume rendering and novel objectives are introduced for a clean\nseparation of dynamic human and static background, yielding detailed and robust\n3D human geometry reconstructions. We evaluate our methods on publicly\navailable datasets and show improvements over prior art.\n", "date": "2023-02-22", "categories": []}
{"id": "2302.08504", "title": "PersonNeRF: Personalized Reconstruction from Photo Collections", "abstract": "  We present PersonNeRF, a method that takes a collection of photos of a\nsubject (e.g. Roger Federer) captured across multiple years with arbitrary body\nposes and appearances, and enables rendering the subject with arbitrary novel\ncombinations of viewpoint, body pose, and appearance. PersonNeRF builds a\ncustomized neural volumetric 3D model of the subject that is able to render an\nentire space spanned by camera viewpoint, body pose, and appearance. A central\nchallenge in this task is dealing with sparse observations; a given body pose\nis likely only observed by a single viewpoint with a single appearance, and a\ngiven appearance is only observed under a handful of different body poses. We\naddress this issue by recovering a canonical T-pose neural volumetric\nrepresentation of the subject that allows for changing appearance across\ndifferent observations, but uses a shared pose-dependent motion field across\nall observations. We demonstrate that this approach, along with regularization\nof the recovered volumetric geometry to encourage smoothness, is able to\nrecover a model that renders compelling images from novel combinations of\nviewpoint, pose, and appearance from these challenging unstructured photo\ncollections, outperforming prior work for free-viewpoint human rendering.\n", "date": "2023-02-16", "categories": []}
{"id": "2302.01058", "title": "IKOL: Inverse kinematics optimization layer for 3D human pose and shape\n  estimation via Gauss-Newton differentiation", "abstract": "  This paper presents an inverse kinematic optimization layer (IKOL) for 3D\nhuman pose and shape estimation that leverages the strength of both\noptimization- and regression-based methods within an end-to-end framework. IKOL\ninvolves a nonconvex optimization that establishes an implicit mapping from an\nimage's 3D keypoints and body shapes to the relative body-part rotations. The\n3D keypoints and the body shapes are the inputs and the relative body-part\nrotations are the solutions. However, this procedure is implicit and hard to\nmake differentiable. So, to overcome this issue, we designed a Gauss-Newton\ndifferentiation (GN-Diff) procedure to differentiate IKOL. GN-Diff iteratively\nlinearizes the nonconvex objective function to obtain Gauss-Newton directions\nwith closed form solutions. Then, an automatic differentiation procedure is\ndirectly applied to generate a Jacobian matrix for end-to-end training.\nNotably, the GN-Diff procedure works fast because it does not rely on a\ntime-consuming implicit differentiation procedure. The twist rotation and shape\nparameters are learned from the neural networks and, as a result, IKOL has a\nmuch lower computational overhead than most existing optimization-based\nmethods. Additionally, compared to existing regression-based methods, IKOL\nprovides a more accurate mesh-image correspondence. This is because it\niteratively reduces the distance between the keypoints and also enhances the\nreliability of the pose structures. Extensive experiments demonstrate the\nsuperiority of our proposed framework over a wide range of 3D human pose and\nshape estimation methods.\n", "date": "2023-02-02", "categories": []}
{"id": "2302.05486", "title": "RAFaRe: Learning Robust and Accurate Non-parametric 3D Face\n  Reconstruction from Pseudo 2D&3D Pairs", "abstract": "  We propose a robust and accurate non-parametric method for single-view 3D\nface reconstruction (SVFR). While tremendous efforts have been devoted to\nparametric SVFR, a visible gap still lies between the result 3D shape and the\nground truth. We believe there are two major obstacles: 1) the representation\nof the parametric model is limited to a certain face database; 2) 2D images and\n3D shapes in the fitted datasets are distinctly misaligned. To resolve these\nissues, a large-scale pseudo 2D\\&3D dataset is created by first rendering the\ndetailed 3D faces, then swapping the face in the wild images with the rendered\nface. These pseudo 2D&3D pairs are created from publicly available datasets\nwhich eliminate the gaps between 2D and 3D data while covering diverse\nappearances, poses, scenes, and illumination. We further propose a\nnon-parametric scheme to learn a well-generalized SVFR model from the created\ndataset, and the proposed hierarchical signed distance function turns out to be\neffective in predicting middle-scale and small-scale 3D facial geometry. Our\nmodel outperforms previous methods on FaceScape-wild/lab and MICC benchmarks\nand is well generalized to various appearances, poses, expressions, and\nin-the-wild environments. The code is released at\nhttp://github.com/zhuhao-nju/rafare .\n", "date": "2023-02-10", "categories": []}
{"id": "2207.09086", "title": "MHR-Net: Multiple-Hypothesis Reconstruction of Non-Rigid Shapes from 2D\n  Views", "abstract": "  We propose MHR-Net, a novel method for recovering Non-Rigid Shapes from\nMotion (NRSfM). MHR-Net aims to find a set of reasonable reconstructions for a\n2D view, and it also selects the most likely reconstruction from the set. To\ndeal with the challenging unsupervised generation of non-rigid shapes, we\ndevelop a new Deterministic Basis and Stochastic Deformation scheme in MHR-Net.\nThe non-rigid shape is first expressed as the sum of a coarse shape basis and a\nflexible shape deformation, then multiple hypotheses are generated with\nuncertainty modeling of the deformation part. MHR-Net is optimized with\nreprojection loss on the basis and the best hypothesis. Furthermore, we design\na new Procrustean Residual Loss, which reduces the rigid rotations between\nsimilar shapes and further improves the performance. Experiments show that\nMHR-Net achieves state-of-the-art reconstruction accuracy on Human3.6M, SURREAL\nand 300-VW datasets.\n", "date": "2022-07-19", "categories": []}
{"id": "2211.14822", "title": "Adjustable Method Based on Body Parts for Improving the Accuracy of 3D\n  Reconstruction in Visually Important Body Parts from Silhouettes", "abstract": "  This research proposes a novel adjustable algorithm for reconstructing 3D\nbody shapes from front and side silhouettes. Most recent silhouette-based\napproaches use a deep neural network trained by silhouettes and key points to\nestimate the shape parameters but cannot accurately fit the model to the body\ncontours and consequently are struggling to cover detailed body geometry,\nespecially in the torso. In addition, in most of these cases, body parts have\nthe same accuracy priority, making the optimization harder and avoiding\nreaching the optimum possible result in essential body parts, like the torso,\nwhich is visually important in most applications, such as virtual garment\nfitting. In the proposed method, we can adjust the expected accuracy for each\nbody part based on our purpose by assigning coefficients for the distance of\neach body part between the projected 3D body and 2D silhouettes. To measure\nthis distance, we first recognize the correspondent body parts using body\nsegmentation in both views. Then, we align individual body parts by 2D rigid\nregistration and match them using pairwise matching. The objective function\ntries to minimize the distance cost for the individual body parts in both views\nbased on distances and coefficients by optimizing the statistical model\nparameters. We also handle the slight variation in the degree of arms and limbs\nby matching the pose. We evaluate the proposed method with synthetic body\nmeshes from the normalized S-SCAPE. The result shows that the algorithm can\nmore accurately reconstruct visually important body parts with high\ncoefficients.\n", "date": "2022-11-27", "categories": []}
{"id": "2212.13660", "title": "NeMo: 3D Neural Motion Fields from Multiple Video Instances of the Same\n  Action", "abstract": "  The task of reconstructing 3D human motion has wideranging applications. The\ngold standard Motion capture (MoCap) systems are accurate but inaccessible to\nthe general public due to their cost, hardware and space constraints. In\ncontrast, monocular human mesh recovery (HMR) methods are much more accessible\nthan MoCap as they take single-view videos as inputs. Replacing the multi-view\nMo- Cap systems with a monocular HMR method would break the current barriers to\ncollecting accurate 3D motion thus making exciting applications like motion\nanalysis and motiondriven animation accessible to the general public. However,\nperformance of existing HMR methods degrade when the video contains challenging\nand dynamic motion that is not in existing MoCap datasets used for training.\nThis reduces its appeal as dynamic motion is frequently the target in 3D motion\nrecovery in the aforementioned applications. Our study aims to bridge the gap\nbetween monocular HMR and multi-view MoCap systems by leveraging information\nshared across multiple video instances of the same action. We introduce the\nNeural Motion (NeMo) field. It is optimized to represent the underlying 3D\nmotions across a set of videos of the same action. Empirically, we show that\nNeMo can recover 3D motion in sports using videos from the Penn Action dataset,\nwhere NeMo outperforms existing HMR methods in terms of 2D keypoint detection.\nTo further validate NeMo using 3D metrics, we collected a small MoCap dataset\nmimicking actions in Penn Action,and show that NeMo achieves better 3D\nreconstruction compared to various baselines.\n", "date": "2022-12-28", "categories": []}
{"id": "2212.10550", "title": "InstantAvatar: Learning Avatars from Monocular Video in 60 Seconds", "abstract": "  In this paper, we take a significant step towards real-world applicability of\nmonocular neural avatar reconstruction by contributing InstantAvatar, a system\nthat can reconstruct human avatars from a monocular video within seconds, and\nthese avatars can be animated and rendered at an interactive rate. To achieve\nthis efficiency we propose a carefully designed and engineered system, that\nleverages emerging acceleration structures for neural fields, in combination\nwith an efficient empty space-skipping strategy for dynamic scenes. We also\ncontribute an efficient implementation that we will make available for research\npurposes. Compared to existing methods, InstantAvatar converges 130x faster and\ncan be trained in minutes instead of hours. It achieves comparable or even\nbetter reconstruction quality and novel pose synthesis results. When given the\nsame time budget, our method significantly outperforms SoTA methods.\nInstantAvatar can yield acceptable visual quality in as little as 10 seconds\ntraining time.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.07409", "title": "Self-Supervised Geometry-Aware Encoder for Style-Based 3D GAN Inversion", "abstract": "  StyleGAN has achieved great progress in 2D face reconstruction and semantic\nediting via image inversion and latent editing. While studies over extending 2D\nStyleGAN to 3D faces have emerged, a corresponding generic 3D GAN inversion\nframework is still missing, limiting the applications of 3D face reconstruction\nand semantic editing. In this paper, we study the challenging problem of 3D GAN\ninversion where a latent code is predicted given a single face image to\nfaithfully recover its 3D shapes and detailed textures. The problem is\nill-posed: innumerable compositions of shape and texture could be rendered to\nthe current image. Furthermore, with the limited capacity of a global latent\ncode, 2D inversion methods cannot preserve faithful shape and texture at the\nsame time when applied to 3D models. To solve this problem, we devise an\neffective self-training scheme to constrain the learning of inversion. The\nlearning is done efficiently without any real-world 2D-3D training pairs but\nproxy samples generated from a 3D GAN. In addition, apart from a global latent\ncode that captures the coarse shape and texture information, we augment the\ngeneration network with a local branch, where pixel-aligned features are added\nto faithfully reconstruct face details. We further consider a new pipeline to\nperform 3D view-consistent editing. Extensive experiments show that our method\noutperforms state-of-the-art inversion methods in both shape and texture\nreconstruction quality. Code and data will be released.\n", "date": "2022-12-14", "categories": []}
{"id": "2212.05223", "title": "Progressive Multi-view Human Mesh Recovery with Self-Supervision", "abstract": "  To date, little attention has been given to multi-view 3D human mesh\nestimation, despite real-life applicability (e.g., motion capture, sport\nanalysis) and robustness to single-view ambiguities. Existing solutions\ntypically suffer from poor generalization performance to new settings, largely\ndue to the limited diversity of image-mesh pairs in multi-view training data.\nTo address this shortcoming, people have explored the use of synthetic images.\nBut besides the usual impact of visual gap between rendered and target data,\nsynthetic-data-driven multi-view estimators also suffer from overfitting to the\ncamera viewpoint distribution sampled during training which usually differs\nfrom real-world distributions. Tackling both challenges, we propose a novel\nsimulation-based training pipeline for multi-view human mesh recovery, which\n(a) relies on intermediate 2D representations which are more robust to\nsynthetic-to-real domain gap; (b) leverages learnable calibration and\ntriangulation to adapt to more diversified camera setups; and (c) progressively\naggregates multi-view information in a canonical 3D space to remove ambiguities\nin 2D representations. Through extensive benchmarking, we demonstrate the\nsuperiority of the proposed solution especially for unseen in-the-wild\nscenarios.\n", "date": "2022-12-10", "categories": []}
{"id": "2212.02765", "title": "Pixel2ISDF: Implicit Signed Distance Fields based Human Body Model from\n  Multi-view and Multi-pose Images", "abstract": "  In this report, we focus on reconstructing clothed humans in the canonical\nspace given multiple views and poses of a human as the input. To achieve this,\nwe utilize the geometric prior of the SMPLX model in the canonical space to\nlearn the implicit representation for geometry reconstruction. Based on the\nobservation that the topology between the posed mesh and the mesh in the\ncanonical space are consistent, we propose to learn latent codes on the posed\nmesh by leveraging multiple input images and then assign the latent codes to\nthe mesh in the canonical space. Specifically, we first leverage normal and\ngeometry networks to extract the feature vector for each vertex on the SMPLX\nmesh. Normal maps are adopted for better generalization to unseen images\ncompared to 2D images. Then, features for each vertex on the posed mesh from\nmultiple images are integrated by MLPs. The integrated features acting as the\nlatent code are anchored to the SMPLX mesh in the canonical space. Finally,\nlatent code for each 3D point is extracted and utilized to calculate the SDF.\nOur work for reconstructing the human shape on canonical pose achieves 3rd\nperformance on WCPA MVP-Human Body Challenge.\n", "date": "2022-12-06", "categories": []}
{"id": "2209.06423", "title": "SCULPTOR: Skeleton-Consistent Face Creation Using a Learned Parametric\n  Generator", "abstract": "  Recent years have seen growing interest in 3D human faces modelling due to\nits wide applications in digital human, character generation and animation.\nExisting approaches overwhelmingly emphasized on modeling the exterior shapes,\ntextures and skin properties of faces, ignoring the inherent correlation\nbetween inner skeletal structures and appearance. In this paper, we present\nSCULPTOR, 3D face creations with Skeleton Consistency Using a Learned\nParametric facial generaTOR, aiming to facilitate easy creation of both\nanatomically correct and visually convincing face models via a hybrid\nparametric-physical representation. At the core of SCULPTOR is LUCY, the first\nlarge-scale shape-skeleton face dataset in collaboration with plastic surgeons.\nNamed after the fossils of one of the oldest known human ancestors, our LUCY\ndataset contains high-quality Computed Tomography (CT) scans of the complete\nhuman head before and after orthognathic surgeries, critical for evaluating\nsurgery results. LUCY consists of 144 scans of 72 subjects (31 male and 41\nfemale) where each subject has two CT scans taken pre- and post-orthognathic\noperations. Based on our LUCY dataset, we learn a novel skeleton consistent\nparametric facial generator, SCULPTOR, which can create the unique and nuanced\nfacial features that help define a character and at the same time maintain\nphysiological soundness. Our SCULPTOR jointly models the skull, face geometry\nand face appearance under a unified data-driven framework, by separating the\ndepiction of a 3D face into shape blend shape, pose blend shape and facial\nexpression blend shape. SCULPTOR preserves both anatomic correctness and visual\nrealism in facial generation tasks compared with existing methods. Finally, we\nshowcase the robustness and effectiveness of SCULPTOR in various fancy\napplications unseen before.\n", "date": "2022-09-14", "categories": []}
{"id": "2208.02817", "title": "Occupancy Planes for Single-view RGB-D Human Reconstruction", "abstract": "  Single-view RGB-D human reconstruction with implicit functions is often\nformulated as per-point classification. Specifically, a set of 3D locations\nwithin the view-frustum of the camera are first projected independently onto\nthe image and a corresponding feature is subsequently extracted for each 3D\nlocation. The feature of each 3D location is then used to classify\nindependently whether the corresponding 3D point is inside or outside the\nobserved object. This procedure leads to sub-optimal results because\ncorrelations between predictions for neighboring locations are only taken into\naccount implicitly via the extracted features. For more accurate results we\npropose the occupancy planes (OPlanes) representation, which enables to\nformulate single-view RGB-D human reconstruction as occupancy prediction on\nplanes which slice through the camera's view frustum. Such a representation\nprovides more flexibility than voxel grids and enables to better leverage\ncorrelations than per-point classification. On the challenging S3D data we\nobserve a simple classifier based on the OPlanes representation to yield\ncompelling results, especially in difficult situations with partial occlusions\ndue to other objects and partial visibility, which haven't been addressed by\nprior work.\n", "date": "2022-08-04", "categories": []}
{"id": "2211.15601", "title": "Fast-SNARF: A Fast Deformer for Articulated Neural Fields", "abstract": "  Neural fields have revolutionized the area of 3D reconstruction and novel\nview synthesis of rigid scenes. A key challenge in making such methods\napplicable to articulated objects, such as the human body, is to model the\ndeformation of 3D locations between the rest pose (a canonical space) and the\ndeformed space. We propose a new articulation module for neural fields,\nFast-SNARF, which finds accurate correspondences between canonical space and\nposed space via iterative root finding. Fast-SNARF is a drop-in replacement in\nfunctionality to our previous work, SNARF, while significantly improving its\ncomputational efficiency. We contribute several algorithmic and implementation\nimprovements over SNARF, yielding a speed-up of $150\\times$. These improvements\ninclude voxel-based correspondence search, pre-computing the linear blend\nskinning function, and an efficient software implementation with CUDA kernels.\nFast-SNARF enables efficient and simultaneous optimization of shape and\nskinning weights given deformed observations without correspondences (e.g. 3D\nmeshes). Because learning of deformation maps is a crucial component in many 3D\nhuman avatar methods and since Fast-SNARF provides a computationally efficient\nsolution, we believe that this work represents a significant step towards the\npractical creation of 3D virtual humans.\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": "2211.13429", "title": "UV-Based 3D Hand-Object Reconstruction with Grasp Optimization", "abstract": "  We propose a novel framework for 3D hand shape reconstruction and hand-object\ngrasp optimization from a single RGB image. The representation of hand-object\ncontact regions is critical for accurate reconstructions. Instead of\napproximating the contact regions with sparse points, as in previous works, we\npropose a dense representation in the form of a UV coordinate map. Furthermore,\nwe introduce inference-time optimization to fine-tune the grasp and improve\ninteractions between the hand and the object. Our pipeline increases hand shape\nreconstruction accuracy and produces a vibrant hand texture. Experiments on\ndatasets such as Ho3D, FreiHAND, and DexYCB reveal that our proposed method\noutperforms the state-of-the-art.\n", "date": "2022-11-24", "categories": []}
{"id": "2211.11931", "title": "Layered-Garment Net: Generating Multiple Implicit Garment Layers from a\n  Single Image", "abstract": "  Recent research works have focused on generating human models and garments\nfrom their 2D images. However, state-of-the-art researches focus either on only\na single layer of the garment on a human model or on generating multiple\ngarment layers without any guarantee of the intersection-free geometric\nrelationship between them. In reality, people wear multiple layers of garments\nin their daily life, where an inner layer of garment could be partially covered\nby an outer one. In this paper, we try to address this multi-layer modeling\nproblem and propose the Layered-Garment Net (LGN) that is capable of generating\nintersection-free multiple layers of garments defined by implicit function\nfields over the body surface, given the person's near front-view image. With a\nspecial design of garment indication fields (GIF), we can enforce an implicit\ncovering relationship between the signed distance fields (SDF) of different\nlayers to avoid self-intersections among different garment surfaces and the\nhuman body. Experiments demonstrate the strength of our proposed LGN framework\nin generating multi-layer garments as compared to state-of-the-art methods. To\nthe best of our knowledge, LGN is the first research work to generate\nintersection-free multiple layers of garments on the human body from a single\nimage.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.10098", "title": "UnconFuse: Avatar Reconstruction from Unconstrained Images", "abstract": "  The report proposes an effective solution about 3D human body reconstruction\nfrom multiple unconstrained frames for ECCV 2022 WCPA Challenge: From Face,\nBody and Fashion to 3D Virtual avatars I (track1: Multi-View Based 3D Human\nBody Reconstruction). We reproduce the reconstruction method presented in\nMVP-Human as our baseline, and make some improvements for the particularity of\nthis challenge. We finally achieve the score 0.93 on the official testing set,\ngetting the 1st place on the leaderboard.\n", "date": "2022-11-18", "categories": []}
{"id": "2211.04753", "title": "ReFu: Refine and Fuse the Unobserved View for Detail-Preserving\n  Single-Image 3D Human Reconstruction", "abstract": "  Single-image 3D human reconstruction aims to reconstruct the 3D textured\nsurface of the human body given a single image. While implicit function-based\nmethods recently achieved reasonable reconstruction performance, they still\nbear limitations showing degraded quality in both surface geometry and texture\nfrom an unobserved view. In response, to generate a realistic textured surface,\nwe propose ReFu, a coarse-to-fine approach that refines the projected backside\nview image and fuses the refined image to predict the final human body. To\nsuppress the diffused occupancy that causes noise in projection images and\nreconstructed meshes, we propose to train occupancy probability by\nsimultaneously utilizing 2D and 3D supervisions with occupancy-based volume\nrendering. We also introduce a refinement architecture that generates\ndetail-preserving backside-view images with front-to-back warping. Extensive\nexperiments demonstrate that our method achieves state-of-the-art performance\nin 3D human reconstruction from a single image, showing enhanced geometry and\ntexture quality from an unobserved view.\n", "date": "2022-11-09", "categories": []}
{"id": "2210.13529", "title": "Multi-Person 3D Pose and Shape Estimation via Inverse Kinematics and\n  Refinement", "abstract": "  Estimating 3D poses and shapes in the form of meshes from monocular RGB\nimages is challenging. Obviously, it is more difficult than estimating 3D poses\nonly in the form of skeletons or heatmaps. When interacting persons are\ninvolved, the 3D mesh reconstruction becomes more challenging due to the\nambiguity introduced by person-to-person occlusions. To tackle the challenges,\nwe propose a coarse-to-fine pipeline that benefits from 1) inverse kinematics\nfrom the occlusion-robust 3D skeleton estimation and 2) Transformer-based\nrelation-aware refinement techniques. In our pipeline, we first obtain\nocclusion-robust 3D skeletons for multiple persons from an RGB image. Then, we\napply inverse kinematics to convert the estimated skeletons to deformable 3D\nmesh parameters. Finally, we apply the Transformer-based mesh refinement that\nrefines the obtained mesh parameters considering intra- and inter-person\nrelations of 3D meshes. Via extensive experiments, we demonstrate the\neffectiveness of our method, outperforming state-of-the-arts on 3DPW, MuPoTS\nand AGORA datasets.\n", "date": "2022-10-24", "categories": []}
{"id": "2207.00210", "title": "Neural Parameterization for Dynamic Human Head Editing", "abstract": "  Implicit radiance functions emerged as a powerful scene representation for\nreconstructing and rendering photo-realistic views of a 3D scene. These\nrepresentations, however, suffer from poor editability. On the other hand,\nexplicit representations such as polygonal meshes allow easy editing but are\nnot as suitable for reconstructing accurate details in dynamic human heads,\nsuch as fine facial features, hair, teeth, and eyes. In this work, we present\nNeural Parameterization (NeP), a hybrid representation that provides the\nadvantages of both implicit and explicit methods. NeP is capable of\nphoto-realistic rendering while allowing fine-grained editing of the scene\ngeometry and appearance. We first disentangle the geometry and appearance by\nparameterizing the 3D geometry into 2D texture space. We enable geometric\neditability by introducing an explicit linear deformation blending layer. The\ndeformation is controlled by a set of sparse key points, which can be\nexplicitly and intuitively displaced to edit the geometry. For appearance, we\ndevelop a hybrid 2D texture consisting of an explicit texture map for easy\nediting and implicit view and time-dependent residuals to model temporal and\nview variations. We compare our method to several reconstruction and editing\nbaselines. The results show that the NeP achieves almost the same level of\nrendering accuracy while maintaining high editability.\n", "date": "2022-07-01", "categories": []}
{"id": "2210.15200", "title": "Deep-MDS Framework for Recovering the 3D Shape of 2D Landmarks from a\n  Single Image", "abstract": "  In this paper, a low parameter deep learning framework utilizing the\nNon-metric Multi-Dimensional scaling (NMDS) method, is proposed to recover the\n3D shape of 2D landmarks on a human face, in a single input image. Hence, NMDS\napproach is used for the first time to establish a mapping from a 2D landmark\nspace to the corresponding 3D shape space. A deep neural network learns the\npairwise dissimilarity among 2D landmarks, used by NMDS approach, whose\nobjective is to learn the pairwise 3D Euclidean distance of the corresponding\n2D landmarks on the input image. This scheme results in a symmetric\ndissimilarity matrix, with the rank larger than 2, leading the NMDS approach\ntoward appropriately recovering the 3D shape of corresponding 2D landmarks. In\nthe case of posed images and complex image formation processes like perspective\nprojection which causes occlusion in the input image, we consider an\nautoencoder component in the proposed framework, as an occlusion removal part,\nwhich turns different input views of the human face into a profile view. The\nresults of a performance evaluation using different synthetic and real-world\nhuman face datasets, including Besel Face Model (BFM), CelebA, CoMA - FLAME,\nand CASIA-3D, indicates the comparable performance of the proposed framework,\ndespite its small number of training parameters, with the related\nstate-of-the-art and powerful 3D reconstruction methods from the literature, in\nterms of efficiency and accuracy.\n", "date": "2022-10-27", "categories": []}
{"id": "2210.13861", "title": "SUPR: A Sparse Unified Part-Based Human Representation", "abstract": "  Statistical 3D shape models of the head, hands, and fullbody are widely used\nin computer vision and graphics. Despite their wide use, we show that existing\nmodels of the head and hands fail to capture the full range of motion for these\nparts. Moreover, existing work largely ignores the feet, which are crucial for\nmodeling human movement and have applications in biomechanics, animation, and\nthe footwear industry. The problem is that previous body part models are\ntrained using 3D scans that are isolated to the individual parts. Such data\ndoes not capture the full range of motion for such parts, e.g. the motion of\nhead relative to the neck. Our observation is that full-body scans provide\nimportant information about the motion of the body parts. Consequently, we\npropose a new learning scheme that jointly trains a full-body model and\nspecific part models using a federated dataset of full-body and body-part\nscans. Specifically, we train an expressive human body model called SUPR\n(Sparse Unified Part-Based Human Representation), where each joint strictly\ninfluences a sparse set of model vertices. The factorized representation\nenables separating SUPR into an entire suite of body part models. Note that the\nfeet have received little attention and existing 3D body models have highly\nunder-actuated feet. Using novel 4D scans of feet, we train a model with an\nextended kinematic tree that captures the range of motion of the toes.\nAdditionally, feet deform due to ground contact. To model this, we include a\nnovel non-linear deformation function that predicts foot deformation\nconditioned on the foot pose, shape, and ground contact. We train SUPR on an\nunprecedented number of scans: 1.2 million body, head, hand and foot scans. We\nquantitatively compare SUPR and the separated body parts and find that our\nsuite of models generalizes better than existing models. SUPR is available at\nhttp://supr.is.tue.mpg.de\n", "date": "2022-10-25", "categories": []}
{"id": "2210.11594", "title": "Photo-realistic 360 Head Avatars in the Wild", "abstract": "  Delivering immersive, 3D experiences for human communication requires a\nmethod to obtain 360 degree photo-realistic avatars of humans. To make these\nexperiences accessible to all, only commodity hardware, like mobile phone\ncameras, should be necessary to capture the data needed for avatar creation.\nFor avatars to be rendered realistically from any viewpoint, we require\ntraining images and camera poses from all angles. However, we cannot rely on\nthere being trackable features in the foreground or background of all images\nfor use in estimating poses, especially from the side or back of the head. To\novercome this, we propose a novel landmark detector trained on synthetic data\nto estimate camera poses from 360 degree mobile phone videos of a human head\nfor use in a multi-stage optimization process which creates a photo-realistic\navatar. We perform validation experiments with synthetic data and showcase our\nmethod on 360 degree avatars trained from mobile phone videos.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.10957", "title": "3D Human Mesh Construction Leveraging Wi-Fi", "abstract": "  In this paper, we present, Wi-Mesh, a WiFi vision-based 3D human mesh\nconstruction system. Our system leverages the advances of WiFi to visualize the\nshape and deformations of the human body for 3D mesh construction. In\nparticular, it leverages multiple transmitting and receiving antennas on WiFi\ndevices to estimate the two-dimensional angle of arrival (2D AoA) of the WiFi\nsignal reflections to enable WiFi devices to see the physical environment as we\nhumans do. It then extracts only the images of the human body from the physical\nenvironment and leverages deep learning models to digitize the extracted human\nbody into a 3D mesh representation. Experimental evaluation under various\nindoor environments shows that Wi-Mesh achieves an average vertices location\nerror of 2.81cm and joint position error of 2.4cm, which is comparable to the\nsystems that utilize specialized and dedicated hardware. The proposed system\nhas the advantage of re-using the WiFi devices that already exist in the\nenvironment for potential mass adoption. It can also work in non-line of sight\n(NLoS), poor lighting conditions, and baggy clothes, where the camera-based\nsystems do not work well.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.10036", "title": "ARAH: Animatable Volume Rendering of Articulated Human SDFs", "abstract": "  Combining human body models with differentiable rendering has recently\nenabled animatable avatars of clothed humans from sparse sets of multi-view RGB\nvideos. While state-of-the-art approaches achieve realistic appearance with\nneural radiance fields (NeRF), the inferred geometry often lacks detail due to\nmissing geometric constraints. Further, animating avatars in\nout-of-distribution poses is not yet possible because the mapping from\nobservation space to canonical space does not generalize faithfully to unseen\nposes. In this work, we address these shortcomings and propose a model to\ncreate animatable clothed human avatars with detailed geometry that generalize\nwell to out-of-distribution poses. To achieve detailed geometry, we combine an\narticulated implicit surface representation with volume rendering. For\ngeneralization, we propose a novel joint root-finding algorithm for\nsimultaneous ray-surface intersection search and correspondence search. Our\nalgorithm enables efficient point sampling and accurate point canonicalization\nwhile generalizing well to unseen poses. We demonstrate that our proposed\npipeline can generate clothed avatars with high-quality pose-dependent geometry\nand appearance from a sparse set of multi-view RGB videos. Our method achieves\nstate-of-the-art performance on geometry and appearance reconstruction while\ncreating animatable avatars that generalize well to out-of-distribution poses\nbeyond the small number of training poses.\n", "date": "2022-10-18", "categories": []}
{"id": "2210.08535", "title": "Realistic, Animatable Human Reconstructions for Virtual Fit-On", "abstract": "  We present an end-to-end virtual try-on pipeline, that can fit different\nclothes on a personalized 3-D human model, reconstructed using a single RGB\nimage. Our main idea is to construct an animatable 3-D human model and try-on\ndifferent clothes in a 3-D virtual environment. The existing frame by frame\nvolumetric reconstruction of 3-D human models are highly resource-demanding and\ndo not allow clothes switching. Moreover, existing virtual fit-on systems also\nlack realism due to predominantly being 2-D or not using user's features in the\nreconstruction. These shortcomings are due to either the human body or clothing\nmodel being 2-D or not having the user's facial features in the dressed model.\nWe solve these problems by manipulating a parametric representation of the 3-D\nhuman body model and stitching a head model reconstructed from the actual\nimage. Fitting the 3-D clothing models on the parameterized human model is also\nadjustable to the body shape of the input image. Our reconstruction results, in\ncomparison with recent existing work, are more visually-pleasing.\n", "date": "2022-10-16", "categories": []}
{"id": "2210.09198", "title": "Pixel-Aligned Non-parametric Hand Mesh Reconstruction", "abstract": "  Non-parametric mesh reconstruction has recently shown significant progress in\n3D hand and body applications. In these methods, mesh vertices and edges are\nvisible to neural networks, enabling the possibility to establish a direct\nmapping between 2D image pixels and 3D mesh vertices. In this paper, we seek to\nestablish and exploit this mapping with a simple and compact architecture. The\nnetwork is designed with these considerations: 1) aggregating both local 2D\nimage features from the encoder and 3D geometric features captured in the mesh\ndecoder; 2) decoding coarse-to-fine meshes along the decoding layers to make\nthe best use of the hierarchical multi-scale information. Specifically, we\npropose an end-to-end pipeline for hand mesh recovery tasks which consists of\nthree phases: a 2D feature extractor constructing multi-scale feature maps, a\nfeature mapping module transforming local 2D image features to 3D vertex\nfeatures via 3D-to-2D projection, and a mesh decoder combining the graph\nconvolution and self-attention to reconstruct mesh. The decoder aggregate both\nlocal image features in pixels and geometric features in vertices. It also\nregresses the mesh vertices in a coarse-to-fine manner, which can leverage\nmulti-scale information. By exploiting the local connection and designing the\nmesh decoder, Our approach achieves state-of-the-art for hand mesh\nreconstruction on the public FreiHAND dataset.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.06108", "title": "Reconstructing Personalized Semantic Facial NeRF Models From Monocular\n  Video", "abstract": "  We present a novel semantic model for human head defined with neural radiance\nfield. The 3D-consistent head model consist of a set of disentangled and\ninterpretable bases, and can be driven by low-dimensional expression\ncoefficients. Thanks to the powerful representation ability of neural radiance\nfield, the constructed model can represent complex facial attributes including\nhair, wearings, which can not be represented by traditional mesh blendshape. To\nconstruct the personalized semantic facial model, we propose to define the\nbases as several multi-level voxel fields. With a short monocular RGB video as\ninput, our method can construct the subject's semantic facial NeRF model with\nonly ten to twenty minutes, and can render a photo-realistic human head image\nin tens of miliseconds with a given expression coefficient and view direction.\nWith this novel representation, we apply it to many tasks like facial\nretargeting and expression editing. Experimental results demonstrate its strong\nrepresentation ability and training/inference speed. Demo videos and released\ncode are provided in our project page:\nhttps://ustc3dv.github.io/NeRFBlendShape/\n", "date": "2022-10-12", "categories": []}
{"id": "2210.05665", "title": "HiFECap: Monocular High-Fidelity and Expressive Capture of Human\n  Performances", "abstract": "  Monocular 3D human performance capture is indispensable for many applications\nin computer graphics and vision for enabling immersive experiences. However,\ndetailed capture of humans requires tracking of multiple aspects, including the\nskeletal pose, the dynamic surface, which includes clothing, hand gestures as\nwell as facial expressions. No existing monocular method allows joint tracking\nof all these components. To this end, we propose HiFECap, a new neural human\nperformance capture approach, which simultaneously captures human pose,\nclothing, facial expression, and hands just from a single RGB video. We\ndemonstrate that our proposed network architecture, the carefully designed\ntraining strategy, and the tight integration of parametric face and hand models\nto a template mesh enable the capture of all these individual aspects.\nImportantly, our method also captures high-frequency details, such as deforming\nwrinkles on the clothes, better than the previous works. Furthermore, we show\nthat HiFECap outperforms the state-of-the-art human performance capture\napproaches qualitatively and quantitatively while for the first time capturing\nall aspects of the human.\n", "date": "2022-10-11", "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": "2210.01868", "title": "Capturing and Animation of Body and Clothing from Monocular Video", "abstract": "  While recent work has shown progress on extracting clothed 3D human avatars\nfrom a single image, video, or a set of 3D scans, several limitations remain.\nMost methods use a holistic representation to jointly model the body and\nclothing, which means that the clothing and body cannot be separated for\napplications like virtual try-on. Other methods separately model the body and\nclothing, but they require training from a large set of 3D clothed human meshes\nobtained from 3D/4D scanners or physics simulations. Our insight is that the\nbody and clothing have different modeling requirements. While the body is well\nrepresented by a mesh-based parametric 3D model, implicit representations and\nneural radiance fields are better suited to capturing the large variety in\nshape and appearance present in clothing. Building on this insight, we propose\nSCARF (Segmented Clothed Avatar Radiance Field), a hybrid model combining a\nmesh-based body with a neural radiance field. Integrating the mesh into the\nvolumetric rendering in combination with a differentiable rasterizer enables us\nto optimize SCARF directly from monocular videos, without any 3D supervision.\nThe hybrid modeling enables SCARF to (i) animate the clothed body avatar by\nchanging body poses (including hand articulation and facial expressions), (ii)\nsynthesize novel views of the avatar, and (iii) transfer clothing between\navatars in virtual try-on applications. We demonstrate that SCARF reconstructs\nclothing with higher visual quality than existing methods, that the clothing\ndeforms with changing body pose and body shape, and that clothing can be\nsuccessfully transferred between avatars of different subjects. The code and\nmodels are available at https://github.com/YadiraF/SCARF.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.01651", "title": "SelfNeRF: Fast Training NeRF for Human from Monocular Self-rotating\n  Video", "abstract": "  In this paper, we propose SelfNeRF, an efficient neural radiance field based\nnovel view synthesis method for human performance. Given monocular\nself-rotating videos of human performers, SelfNeRF can train from scratch and\nachieve high-fidelity results in about twenty minutes. Some recent works have\nutilized the neural radiance field for dynamic human reconstruction. However,\nmost of these methods need multi-view inputs and require hours of training,\nmaking it still difficult for practical use. To address this challenging\nproblem, we introduce a surface-relative representation based on\nmulti-resolution hash encoding that can greatly improve the training speed and\naggregate inter-frame information. Extensive experimental results on several\ndifferent datasets demonstrate the effectiveness and efficiency of SelfNeRF to\nchallenging monocular videos.\n", "date": "2022-10-04", "categories": []}
{"id": "2209.10621", "title": "GNPM: Geometric-Aware Neural Parametric Models", "abstract": "  We propose Geometric Neural Parametric Models (GNPM), a learned parametric\nmodel that takes into account the local structure of data to learn disentangled\nshape and pose latent spaces of 4D dynamics, using a geometric-aware\narchitecture on point clouds. Temporally consistent 3D deformations are\nestimated without the need for dense correspondences at training time, by\nexploiting cycle consistency. Besides its ability to learn dense\ncorrespondences, GNPMs also enable latent-space manipulations such as\ninterpolation and shape/pose transfer. We evaluate GNPMs on various datasets of\nclothed humans, and show that it achieves comparable performance to\nstate-of-the-art methods that require dense correspondences during training.\n", "date": "2022-09-21", "categories": []}
{"id": "2208.00571", "title": "CLIFF: Carrying Location Information in Full Frames into Human Pose and\n  Shape Estimation", "abstract": "  Top-down methods dominate the field of 3D human pose and shape estimation,\nbecause they are decoupled from human detection and allow researchers to focus\non the core problem. However, cropping, their first step, discards the location\ninformation from the very beginning, which makes themselves unable to\naccurately predict the global rotation in the original camera coordinate\nsystem. To address this problem, we propose to Carry Location Information in\nFull Frames (CLIFF) into this task. Specifically, we feed more holistic\nfeatures to CLIFF by concatenating the cropped-image feature with its bounding\nbox information. We calculate the 2D reprojection loss with a broader view of\nthe full frame, taking a projection process similar to that of the person\nprojected in the image. Fed and supervised by global-location-aware\ninformation, CLIFF directly predicts the global rotation along with more\naccurate articulated poses. Besides, we propose a pseudo-ground-truth annotator\nbased on CLIFF, which provides high-quality 3D annotations for in-the-wild 2D\ndatasets and offers crucial full supervision for regression-based methods.\nExtensive experiments on popular benchmarks show that CLIFF outperforms prior\narts by a significant margin, and reaches the first place on the AGORA\nleaderboard (the SMPL-Algorithms track). The code and data are available at\nhttps://github.com/huawei-noah/noah-research/tree/master/CLIFF.\n", "date": "2022-08-01", "categories": []}
{"id": "2209.10529", "title": "Benchmarking and Analyzing 3D Human Pose and Shape Estimation Beyond\n  Algorithms", "abstract": "  3D human pose and shape estimation (a.k.a. \"human mesh recovery\") has\nachieved substantial progress. Researchers mainly focus on the development of\nnovel algorithms, while less attention has been paid to other critical factors\ninvolved. This could lead to less optimal baselines, hindering the fair and\nfaithful evaluations of newly designed methodologies. To address this problem,\nthis work presents the first comprehensive benchmarking study from three\nunder-explored perspectives beyond algorithms. 1) Datasets. An analysis on 31\ndatasets reveals the distinct impacts of data samples: datasets featuring\ncritical attributes (i.e. diverse poses, shapes, camera characteristics,\nbackbone features) are more effective. Strategical selection and combination of\nhigh-quality datasets can yield a significant boost to the model performance.\n2) Backbones. Experiments with 10 backbones, ranging from CNNs to transformers,\nshow the knowledge learnt from a proximity task is readily transferable to\nhuman mesh recovery. 3) Training strategies. Proper augmentation techniques and\nloss designs are crucial. With the above findings, we achieve a PA-MPJPE of\n47.3 mm on the 3DPW test set with a relatively simple model. More importantly,\nwe provide strong baselines for fair comparisons of algorithms, and\nrecommendations for building effective training configurations in the future.\nCodebase is available at http://github.com/smplbody/hmr-benchmarks\n", "date": "2022-09-21", "categories": []}
{"id": "2209.08468", "title": "Human Performance Modeling and Rendering via Neural Animated Mesh", "abstract": "  We have recently seen tremendous progress in the neural advances for\nphoto-real human modeling and rendering. However, it's still challenging to\nintegrate them into an existing mesh-based pipeline for downstream\napplications. In this paper, we present a comprehensive neural approach for\nhigh-quality reconstruction, compression, and rendering of human performances\nfrom dense multi-view videos. Our core intuition is to bridge the traditional\nanimated mesh workflow with a new class of highly efficient neural techniques.\nWe first introduce a neural surface reconstructor for high-quality surface\ngeneration in minutes. It marries the implicit volumetric rendering of the\ntruncated signed distance field (TSDF) with multi-resolution hash encoding. We\nfurther propose a hybrid neural tracker to generate animated meshes, which\ncombines explicit non-rigid tracking with implicit dynamic deformation in a\nself-supervised framework. The former provides the coarse warping back into the\ncanonical space, while the latter implicit one further predicts the\ndisplacements using the 4D hash encoding as in our reconstructor. Then, we\ndiscuss the rendering schemes using the obtained animated meshes, ranging from\ndynamic texturing to lumigraph rendering under various bandwidth settings. To\nstrike an intricate balance between quality and bandwidth, we propose a\nhierarchical solution by first rendering 6 virtual views covering the performer\nand then conducting occlusion-aware neural texture blending. We demonstrate the\nefficacy of our approach in a variety of mesh-based applications and\nphoto-realistic free-view experiences on various platforms, i.e., inserting\nvirtual human performances into real environments through mobile AR or\nimmersively watching talent shows with VR headsets.\n", "date": "2022-09-18", "categories": []}
{"id": "2206.10457", "title": "Domain Adaptive 3D Pose Augmentation for In-the-wild Human Mesh Recovery", "abstract": "  The ability to perceive 3D human bodies from a single image has a multitude\nof applications ranging from entertainment and robotics to neuroscience and\nhealthcare. A fundamental challenge in human mesh recovery is in collecting the\nground truth 3D mesh targets required for training, which requires burdensome\nmotion capturing systems and is often limited to indoor laboratories. As a\nresult, while progress is made on benchmark datasets collected in these\nrestrictive settings, models fail to generalize to real-world \"in-the-wild\"\nscenarios due to distribution shifts. We propose Domain Adaptive 3D Pose\nAugmentation (DAPA), a data augmentation method that enhances the model's\ngeneralization ability in in-the-wild scenarios. DAPA combines the strength of\nmethods based on synthetic datasets by getting direct supervision from the\nsynthesized meshes, and domain adaptation methods by using ground truth 2D\nkeypoints from the target dataset. We show quantitatively that finetuning with\nDAPA effectively improves results on benchmarks 3DPW and AGORA. We further\ndemonstrate the utility of DAPA on a challenging dataset curated from videos of\nreal-world parent-child interaction.\n", "date": "2022-06-21", "categories": []}
{"id": "2209.06814", "title": "Neural Point-based Shape Modeling of Humans in Challenging Clothing", "abstract": "  Parametric 3D body models like SMPL only represent minimally-clothed people\nand are hard to extend to clothing because they have a fixed mesh topology and\nresolution. To address these limitations, recent work uses implicit surfaces or\npoint clouds to model clothed bodies. While not limited by topology, such\nmethods still struggle to model clothing that deviates significantly from the\nbody, such as skirts and dresses. This is because they rely on the body to\ncanonicalize the clothed surface by reposing it to a reference shape.\nUnfortunately, this process is poorly defined when clothing is far from the\nbody. Additionally, they use linear blend skinning to pose the body and the\nskinning weights are tied to the underlying body parts. In contrast, we model\nthe clothing deformation in a local coordinate space without canonicalization.\nWe also relax the skinning weights to let multiple body parts influence the\nsurface. Specifically, we extend point-based methods with a coarse stage, that\nreplaces canonicalization with a learned pose-independent \"coarse shape\" that\ncan capture the rough surface geometry of clothing like skirts. We then refine\nthis using a network that infers the linear blend skinning weights and pose\ndependent displacements from the coarse representation. The approach works well\nfor garments that both conform to, and deviate from, the body. We demonstrate\nthe usefulness of our approach by learning person-specific avatars from\nexamples and then show how they can be animated in new poses and motions. We\nalso show that the method can learn directly from raw scans with missing data,\ngreatly simplifying the process of creating realistic avatars. Code is\navailable for research purposes at\n{\\small\\url{https://qianlim.github.io/SkiRT}}.\n", "date": "2022-09-14", "categories": []}
{"id": "2208.05751", "title": "FDNeRF: Few-shot Dynamic Neural Radiance Fields for Face Reconstruction\n  and Expression Editing", "abstract": "  We propose a Few-shot Dynamic Neural Radiance Field (FDNeRF), the first\nNeRF-based method capable of reconstruction and expression editing of 3D faces\nbased on a small number of dynamic images. Unlike existing dynamic NeRFs that\nrequire dense images as input and can only be modeled for a single identity,\nour method enables face reconstruction across different persons with few-shot\ninputs. Compared to state-of-the-art few-shot NeRFs designed for modeling\nstatic scenes, the proposed FDNeRF accepts view-inconsistent dynamic inputs and\nsupports arbitrary facial expression editing, i.e., producing faces with novel\nexpressions beyond the input ones. To handle the inconsistencies between\ndynamic inputs, we introduce a well-designed conditional feature warping (CFW)\nmodule to perform expression conditioned warping in 2D feature space, which is\nalso identity adaptive and 3D constrained. As a result, features of different\nexpressions are transformed into the target ones. We then construct a radiance\nfield based on these view-consistent features and use volumetric rendering to\nsynthesize novel views of the modeled faces. Extensive experiments with\nquantitative and qualitative evaluation demonstrate that our method outperforms\nexisting dynamic and few-shot NeRFs on both 3D face reconstruction and\nexpression editing tasks. Code is available at\nhttps://github.com/FDNeRF/FDNeRF.\n", "date": "2022-08-11", "categories": []}
{"id": "2209.04596", "title": "Self-supervised Human Mesh Recovery with Cross-Representation Alignment", "abstract": "  Fully supervised human mesh recovery methods are data-hungry and have poor\ngeneralizability due to the limited availability and diversity of 3D-annotated\nbenchmark datasets. Recent progress in self-supervised human mesh recovery has\nbeen made using synthetic-data-driven training paradigms where the model is\ntrained from synthetic paired 2D representation (e.g., 2D keypoints and\nsegmentation masks) and 3D mesh. However, on synthetic dense correspondence\nmaps (i.e., IUV) few have been explored since the domain gap between synthetic\ntraining data and real testing data is hard to address for 2D dense\nrepresentation. To alleviate this domain gap on IUV, we propose\ncross-representation alignment utilizing the complementary information from the\nrobust but sparse representation (2D keypoints). Specifically, the alignment\nerrors between initial mesh estimation and both 2D representations are\nforwarded into regressor and dynamically corrected in the following mesh\nregression. This adaptive cross-representation alignment explicitly learns from\nthe deviations and captures complementary information: robustness from sparse\nrepresentation and richness from dense representation. We conduct extensive\nexperiments on multiple standard benchmark datasets and demonstrate competitive\nresults, helping take a step towards reducing the annotation effort needed to\nproduce state-of-the-art models in human mesh estimation.\n", "date": "2022-09-10", "categories": []}
{"id": "2209.03027", "title": "SIRA: Relightable Avatars from a Single Image", "abstract": "  Recovering the geometry of a human head from a single image, while\nfactorizing the materials and illumination is a severely ill-posed problem that\nrequires prior information to be solved. Methods based on 3D Morphable Models\n(3DMM), and their combination with differentiable renderers, have shown\npromising results. However, the expressiveness of 3DMMs is limited, and they\ntypically yield over-smoothed and identity-agnostic 3D shapes limited to the\nface region. Highly accurate full head reconstructions have recently been\nobtained with neural fields that parameterize the geometry using multilayer\nperceptrons. The versatility of these representations has also proved effective\nfor disentangling geometry, materials and lighting. However, these methods\nrequire several tens of input images. In this paper, we introduce SIRA, a\nmethod which, from a single image, reconstructs human head avatars with high\nfidelity geometry and factorized lights and surface materials. Our key\ningredients are two data-driven statistical models based on neural fields that\nresolve the ambiguities of single-view 3D surface reconstruction and appearance\nfactorization. Experiments show that SIRA obtains state of the art results in\n3D head reconstruction while at the same time it successfully disentangles the\nglobal illumination, and the diffuse and specular albedos. Furthermore, our\nreconstructions are amenable to physically-based appearance editing and head\nmodel relighting.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.03254", "title": "3D Textured Shape Recovery with Learned Geometric Priors", "abstract": "  3D textured shape recovery from partial scans is crucial for many real-world\napplications. Existing approaches have demonstrated the efficacy of implicit\nfunction representation, but they suffer from partial inputs with severe\nocclusions and varying object types, which greatly hinders their application\nvalue in the real world. This technical report presents our approach to address\nthese limitations by incorporating learned geometric priors. To this end, we\ngenerate a SMPL model from learned pose prediction and fuse it into the partial\ninput to add prior knowledge of human bodies. We also propose a novel\ncompleteness-aware bounding box adaptation for handling different levels of\nscales and partialness of partial scans.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.02600", "title": "Domain Engineering for Applied Monocular Reconstruction of Parametric\n  Faces", "abstract": "  Many modern online 3D applications and video games rely on parametric models\nof human faces for creating believable avatars. However, manually reproducing\nsomeone's facial likeness with a parametric model is difficult and\ntime-consuming. Machine Learning solution for that task is highly desirable but\nis also challenging. The paper proposes a novel approach to the so-called\nFace-to-Parameters problem (F2P for short), aiming to reconstruct a parametric\nface from a single image. The proposed method utilizes synthetic data, domain\ndecomposition, and domain adaptation to address multifaceted challenges in\nsolving the F2P. The open-sourced codebase illustrates our key observations and\nprovides means for quantitative evaluation. The presented approach proves\npractical in an industrial application; it improves accuracy and allows for\nmore efficient models training. The techniques have the potential to extend to\nother types of parametric models.\n", "date": "2022-09-06", "categories": []}
{"id": "2208.14851", "title": "Dual-Space NeRF: Learning Animatable Avatars and Scene Lighting in\n  Separate Spaces", "abstract": "  Modeling the human body in a canonical space is a common practice for\ncapturing and animation. But when involving the neural radiance field (NeRF),\nlearning a static NeRF in the canonical space is not enough because the\nlighting of the body changes when the person moves even though the scene\nlighting is constant. Previous methods alleviate the inconsistency of lighting\nby learning a per-frame embedding, but this operation does not generalize to\nunseen poses. Given that the lighting condition is static in the world space\nwhile the human body is consistent in the canonical space, we propose a\ndual-space NeRF that models the scene lighting and the human body with two MLPs\nin two separate spaces. To bridge these two spaces, previous methods mostly\nrely on the linear blend skinning (LBS) algorithm. However, the blending\nweights for LBS of a dynamic neural field are intractable and thus are usually\nmemorized with another MLP, which does not generalize to novel poses. Although\nit is possible to borrow the blending weights of a parametric mesh such as\nSMPL, the interpolation operation introduces more artifacts. In this paper, we\npropose to use the barycentric mapping, which can directly generalize to unseen\nposes and surprisingly achieves superior results than LBS with neural blending\nweights. Quantitative and qualitative results on the Human3.6M and the\nZJU-MoCap datasets show the effectiveness of our method.\n", "date": "2022-08-31", "categories": []}
{"id": "2208.12078", "title": "Learning to regulate 3D head shape by removing occluding hair from\n  in-the-wild images", "abstract": "  Recent 3D face reconstruction methods reconstruct the entire head compared to\nearlier approaches which only model the face. Although these methods accurately\nreconstruct facial features, they do not explicitly regulate the upper part of\nthe head. Extracting information about this part of the head is challenging due\nto varying degrees of occlusion by hair. We present a novel approach for\nmodeling the upper head by removing occluding hair and reconstructing the skin,\nrevealing information about the head shape. We introduce three objectives: 1) a\ndice consistency loss that enforces similarity between the overall head shape\nof the source and rendered image, 2) a scale consistency loss to ensure that\nhead shape is accurately reproduced even if the upper part of the head is not\nvisible, and 3) a 71 landmark detector trained using a moving average loss\nfunction to detect additional landmarks on the head. These objectives are used\nto train an encoder in an unsupervised manner to regress FLAME parameters from\nin-the-wild input images. Our unsupervised 3DMM model achieves state-of-the-art\nresults on popular benchmarks and can be used to infer the head shape, facial\nfeatures, and textures for direct use in animation or avatar creation.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.11251", "title": "Learnable human mesh triangulation for 3D human pose and shape\n  estimation", "abstract": "  Compared to joint position, the accuracy of joint rotation and shape\nestimation has received relatively little attention in the skinned multi-person\nlinear model (SMPL)-based human mesh reconstruction from multi-view images. The\nwork in this field is broadly classified into two categories. The first\napproach performs joint estimation and then produces SMPL parameters by fitting\nSMPL to resultant joints. The second approach regresses SMPL parameters\ndirectly from the input images through a convolutional neural network\n(CNN)-based model. However, these approaches suffer from the lack of\ninformation for resolving the ambiguity of joint rotation and shape\nreconstruction and the difficulty of network learning. To solve the\naforementioned problems, we propose a two-stage method. The proposed method\nfirst estimates the coordinates of mesh vertices through a CNN-based model from\ninput images, and acquires SMPL parameters by fitting the SMPL model to the\nestimated vertices. Estimated mesh vertices provide sufficient information for\ndetermining joint rotation and shape, and are easier to learn than SMPL\nparameters. According to experiments using Human3.6M and MPI-INF-3DHP datasets,\nthe proposed method significantly outperforms the previous works in terms of\njoint rotation and shape estimation, and achieves competitive performance in\nterms of joint location estimation.\n", "date": "2022-08-24", "categories": []}
{"id": "2208.10652", "title": "Learning Visibility for Robust Dense Human Body Estimation", "abstract": "  Estimating 3D human pose and shape from 2D images is a crucial yet\nchallenging task. While prior methods with model-based representations can\nperform reasonably well on whole-body images, they often fail when parts of the\nbody are occluded or outside the frame. Moreover, these results usually do not\nfaithfully capture the human silhouettes due to their limited representation\npower of deformable models (e.g., representing only the naked body). An\nalternative approach is to estimate dense vertices of a predefined template\nbody in the image space. Such representations are effective in localizing\nvertices within an image but cannot handle out-of-frame body parts. In this\nwork, we learn dense human body estimation that is robust to partial\nobservations. We explicitly model the visibility of human joints and vertices\nin the x, y, and z axes separately. The visibility in x and y axes help\ndistinguishing out-of-frame cases, and the visibility in depth axis corresponds\nto occlusions (either self-occlusions or occlusions by other objects). We\nobtain pseudo ground-truths of visibility labels from dense UV correspondences\nand train a neural network to predict visibility along with 3D coordinates. We\nshow that visibility can serve as 1) an additional signal to resolve depth\nordering ambiguities of self-occluded vertices and 2) a regularization term\nwhen fitting a human body model to the predictions. Extensive experiments on\nmultiple 3D human datasets demonstrate that visibility modeling significantly\nimproves the accuracy of human body estimation, especially for partial-body\ncases. Our project page with code is at: https://github.com/chhankyao/visdb.\n", "date": "2022-08-23", "categories": []}
{"id": "2208.10738", "title": "Super-resolution 3D Human Shape from a Single Low-Resolution Image", "abstract": "  We propose a novel framework to reconstruct super-resolution human shape from\na single low-resolution input image. The approach overcomes limitations of\nexisting approaches that reconstruct 3D human shape from a single image, which\nrequire high-resolution images together with auxiliary data such as surface\nnormal or a parametric model to reconstruct high-detail shape. The proposed\nframework represents the reconstructed shape with a high-detail implicit\nfunction. Analogous to the objective of 2D image super-resolution, the approach\nlearns the mapping from a low-resolution shape to its high-resolution\ncounterpart and it is applied to reconstruct 3D shape detail from\nlow-resolution images. The approach is trained end-to-end employing a novel\nloss function which estimates the information lost between a low and\nhigh-resolution representation of the same 3D surface shape. Evaluation for\nsingle image reconstruction of clothed people demonstrates that our method\nachieves high-detail surface reconstruction from low-resolution images without\nauxiliary data. Extensive experiments show that the proposed approach can\nestimate super-resolution human geometries with a significantly higher level of\ndetail than that obtained with previous approaches when applied to\nlow-resolution images.\n", "date": "2022-08-23", "categories": []}
{"id": "2208.08622", "title": "LoRD: Local 4D Implicit Representation for High-Fidelity Dynamic Human\n  Modeling", "abstract": "  Recent progress in 4D implicit representation focuses on globally controlling\nthe shape and motion with low dimensional latent vectors, which is prone to\nmissing surface details and accumulating tracking error. While many deep local\nrepresentations have shown promising results for 3D shape modeling, their 4D\ncounterpart does not exist yet. In this paper, we fill this blank by proposing\na novel Local 4D implicit Representation for Dynamic clothed human, named LoRD,\nwhich has the merits of both 4D human modeling and local representation, and\nenables high-fidelity reconstruction with detailed surface deformations, such\nas clothing wrinkles. Particularly, our key insight is to encourage the network\nto learn the latent codes of local part-level representation, capable of\nexplaining the local geometry and temporal deformations. To make the inference\nat test-time, we first estimate the inner body skeleton motion to track local\nparts at each time step, and then optimize the latent codes for each part via\nauto-decoding based on different types of observed data. Extensive experiments\ndemonstrate that the proposed method has strong capability for representing 4D\nhuman, and outperforms state-of-the-art methods on practical applications,\nincluding 4D reconstruction from sparse points, non-rigid depth fusion, both\nqualitatively and quantitatively.\n", "date": "2022-08-18", "categories": []}
{"id": "2208.08728", "title": "Neural Capture of Animatable 3D Human from Monocular Video", "abstract": "  We present a novel paradigm of building an animatable 3D human representation\nfrom a monocular video input, such that it can be rendered in any unseen poses\nand views. Our method is based on a dynamic Neural Radiance Field (NeRF) rigged\nby a mesh-based parametric 3D human model serving as a geometry proxy. Previous\nmethods usually rely on multi-view videos or accurate 3D geometry information\nas additional inputs; besides, most methods suffer from degraded quality when\ngeneralized to unseen poses. We identify that the key to generalization is a\ngood input embedding for querying dynamic NeRF: A good input embedding should\ndefine an injective mapping in the full volumetric space, guided by surface\nmesh deformation under pose variation. Based on this observation, we propose to\nembed the input query with its relationship to local surface regions spanned by\na set of geodesic nearest neighbors on mesh vertices. By including both\nposition and relative distance information, our embedding defines a\ndistance-preserved deformation mapping and generalizes well to unseen poses. To\nreduce the dependency on additional inputs, we first initialize per-frame 3D\nmeshes using off-the-shelf tools and then propose a pipeline to jointly\noptimize NeRF and refine the initial mesh. Extensive experiments show our\nmethod can synthesize plausible human rendering results under unseen poses and\nviews.\n", "date": "2022-08-18", "categories": []}
{"id": "2208.07142", "title": "Perspective Reconstruction of Human Faces by Joint Mesh and Landmark\n  Regression", "abstract": "  Even though 3D face reconstruction has achieved impressive progress, most\northogonal projection-based face reconstruction methods can not achieve\naccurate and consistent reconstruction results when the face is very close to\nthe camera due to the distortion under the perspective projection. In this\npaper, we propose to simultaneously reconstruct 3D face mesh in the world space\nand predict 2D face landmarks on the image plane to address the problem of\nperspective 3D face reconstruction. Based on the predicted 3D vertices and 2D\nlandmarks, the 6DoF (6 Degrees of Freedom) face pose can be easily estimated by\nthe PnP solver to represent perspective projection. Our approach achieves 1st\nplace on the leader-board of the ECCV 2022 WCPA challenge and our model is\nvisually robust under different identities, expressions and poses. The training\ncode and models are released to facilitate future research.\n", "date": "2022-08-15", "categories": []}
{"id": "2208.02935", "title": "Applied monocular reconstruction of parametric faces with domain\n  engineering", "abstract": "  Many modern online 3D applications and videogames rely on parametric models\nof human faces for creating believable avatars. However, manual reproduction of\nsomeone's facial likeness with a parametric model is difficult and\ntime-consuming. Machine Learning solution for that task is highly desirable but\nis also challenging. The paper proposes a novel approach to the so-called\nFace-to-Parameters problem (F2P for short), aiming to reconstruct a parametric\nface from a single image. The proposed method utilizes synthetic data, domain\ndecomposition, and domain adaptation for addressing multifaceted challenges in\nsolving the F2P. The open-sourced codebase illustrates our key observations and\nprovides means for quantitative evaluation. The presented approach proves\npractical in an industrial application; it improves accuracy and allows for\nmore efficient models training. The techniques have the potential to extend to\nother types of parametric models.\n", "date": "2022-08-05", "categories": []}
{"id": "2208.03167", "title": "Disentangling 3D Attributes from a Single 2D Image: Human Pose, Shape\n  and Garment", "abstract": "  For visual manipulation tasks, we aim to represent image content with\nsemantically meaningful features. However, learning implicit representations\nfrom images often lacks interpretability, especially when attributes are\nintertwined. We focus on the challenging task of extracting disentangled 3D\nattributes only from 2D image data. Specifically, we focus on human appearance\nand learn implicit pose, shape and garment representations of dressed humans\nfrom RGB images. Our method learns an embedding with disentangled latent\nrepresentations of these three image properties and enables meaningful\nre-assembling of features and property control through a 2D-to-3D\nencoder-decoder structure. The 3D model is inferred solely from the feature map\nin the learned embedding space. To the best of our knowledge, our method is the\nfirst to achieve cross-domain disentanglement for this highly under-constrained\nproblem. We qualitatively and quantitatively demonstrate our framework's\nability to transfer pose, shape, and garments in 3D reconstruction on virtual\ndata and show how an implicit shape loss can benefit the model's ability to\nrecover fine-grained reconstruction details.\n", "date": "2022-08-05", "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": "2207.13820", "title": "Cross-Attention of Disentangled Modalities for 3D Human Mesh Recovery\n  with Transformers", "abstract": "  Transformer encoder architectures have recently achieved state-of-the-art\nresults on monocular 3D human mesh reconstruction, but they require a\nsubstantial number of parameters and expensive computations. Due to the large\nmemory overhead and slow inference speed, it is difficult to deploy such models\nfor practical use. In this paper, we propose a novel transformer\nencoder-decoder architecture for 3D human mesh reconstruction from a single\nimage, called FastMETRO. We identify the performance bottleneck in the\nencoder-based transformers is caused by the token design which introduces high\ncomplexity interactions among input tokens. We disentangle the interactions via\nan encoder-decoder architecture, which allows our model to demand much fewer\nparameters and shorter inference time. In addition, we impose the prior\nknowledge of human body's morphological relationship via attention masking and\nmesh upsampling operations, which leads to faster convergence with higher\naccuracy. Our FastMETRO improves the Pareto-front of accuracy and efficiency,\nand clearly outperforms image-based methods on Human3.6M and 3DPW. Furthermore,\nwe validate its generalizability on FreiHAND.\n", "date": "2022-07-27", "categories": []}
{"id": "2207.11094", "title": "Visual Speech-Aware Perceptual 3D Facial Expression Reconstruction from\n  Videos", "abstract": "  The recent state of the art on monocular 3D face reconstruction from image\ndata has made some impressive advancements, thanks to the advent of Deep\nLearning. However, it has mostly focused on input coming from a single RGB\nimage, overlooking the following important factors: a) Nowadays, the vast\nmajority of facial image data of interest do not originate from single images\nbut rather from videos, which contain rich dynamic information. b) Furthermore,\nthese videos typically capture individuals in some form of verbal communication\n(public talks, teleconferences, audiovisual human-computer interactions,\ninterviews, monologues/dialogues in movies, etc). When existing 3D face\nreconstruction methods are applied in such videos, the artifacts in the\nreconstruction of the shape and motion of the mouth area are often severe,\nsince they do not match well with the speech audio.\n  To overcome the aforementioned limitations, we present the first method for\nvisual speech-aware perceptual reconstruction of 3D mouth expressions. We do\nthis by proposing a \"lipread\" loss, which guides the fitting process so that\nthe elicited perception from the 3D reconstructed talking head resembles that\nof the original video footage. We demonstrate that, interestingly, the lipread\nloss is better suited for 3D reconstruction of mouth movements compared to\ntraditional landmark losses, and even direct 3D supervision. Furthermore, the\ndevised method does not rely on any text transcriptions or corresponding audio,\nrendering it ideal for training in unlabeled datasets. We verify the efficiency\nof our method through exhaustive objective evaluations on three large-scale\ndatasets, as well as subjective evaluation with two web-based user studies.\n", "date": "2022-07-22", "categories": []}
{"id": "2207.08000", "title": "DiffuStereo: High Quality Human Reconstruction via Diffusion-based\n  Stereo Using Sparse Cameras", "abstract": "  We propose DiffuStereo, a novel system using only sparse cameras (8 in this\nwork) for high-quality 3D human reconstruction. At its core is a novel\ndiffusion-based stereo module, which introduces diffusion models, a type of\npowerful generative models, into the iterative stereo matching network. To this\nend, we design a new diffusion kernel and additional stereo constraints to\nfacilitate stereo matching and depth estimation in the network. We further\npresent a multi-level stereo network architecture to handle high-resolution (up\nto 4k) inputs without requiring unaffordable memory footprint. Given a set of\nsparse-view color images of a human, the proposed multi-level diffusion-based\nstereo network can produce highly accurate depth maps, which are then converted\ninto a high-quality 3D human model through an efficient multi-view fusion\nstrategy. Overall, our method enables automatic reconstruction of human models\nwith quality on par to high-end dense-view camera rigs, and this is achieved\nusing a much more light-weight hardware setup. Experiments show that our method\noutperforms state-of-the-art methods by a large margin both qualitatively and\nquantitatively.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.09735", "title": "CrossHuman: Learning Cross-Guidance from Multi-Frame Images for Human\n  Reconstruction", "abstract": "  We propose CrossHuman, a novel method that learns cross-guidance from\nparametric human model and multi-frame RGB images to achieve high-quality 3D\nhuman reconstruction. To recover geometry details and texture even in invisible\nregions, we design a reconstruction pipeline combined with tracking-based\nmethods and tracking-free methods. Given a monocular RGB sequence, we track the\nparametric human model in the whole sequence, the points (voxels) corresponding\nto the target frame are warped to reference frames by the parametric body\nmotion. Guided by the geometry priors of the parametric body and spatially\naligned features from RGB sequence, the robust implicit surface is fused.\nMoreover, a multi-frame transformer (MFT) and a self-supervised warp refinement\nmodule are integrated to the framework to relax the requirements of parametric\nbody and help to deal with very loose cloth. Compared with previous works, our\nCrossHuman enables high-fidelity geometry details and texture in both visible\nand invisible regions and improves the accuracy of the human reconstruction\neven under estimated inaccurate parametric human models. The experiments\ndemonstrate that our method achieves state-of-the-art (SOTA) performance.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09774", "title": "Drivable Volumetric Avatars using Texel-Aligned Features", "abstract": "  Photorealistic telepresence requires both high-fidelity body modeling and\nfaithful driving to enable dynamically synthesized appearance that is\nindistinguishable from reality. In this work, we propose an end-to-end\nframework that addresses two core challenges in modeling and driving full-body\navatars of real people. One challenge is driving an avatar while staying\nfaithful to details and dynamics that cannot be captured by a global\nlow-dimensional parameterization such as body pose. Our approach supports\ndriving of clothed avatars with wrinkles and motion that a real driving\nperformer exhibits beyond the training corpus. Unlike existing global state\nrepresentations or non-parametric screen-space approaches, we introduce\ntexel-aligned features -- a localised representation which can leverage both\nthe structural prior of a skeleton-based parametric model and observed sparse\nimage signals at the same time. Another challenge is modeling a temporally\ncoherent clothed avatar, which typically requires precise surface tracking. To\ncircumvent this, we propose a novel volumetric avatar representation by\nextending mixtures of volumetric primitives to articulated objects. By\nexplicitly incorporating articulation, our approach naturally generalizes to\nunseen poses. We also introduce a localized viewpoint conditioning, which leads\nto a large improvement in generalization of view-dependent appearance. The\nproposed volumetric representation does not require high-quality mesh tracking\nas a prerequisite and brings significant quality improvements compared to\nmesh-based counterparts. In our experiments, we carefully examine our design\nchoices and demonstrate the efficacy of our approach, outperforming the\nstate-of-the-art methods on challenging driving scenarios.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09835", "title": "UNIF: United Neural Implicit Functions for Clothed Human Reconstruction\n  and Animation", "abstract": "  We propose united implicit functions (UNIF), a part-based method for clothed\nhuman reconstruction and animation with raw scans and skeletons as the input.\nPrevious part-based methods for human reconstruction rely on ground-truth part\nlabels from SMPL and thus are limited to minimal-clothed humans. In contrast,\nour method learns to separate parts from body motions instead of part\nsupervision, thus can be extended to clothed humans and other articulated\nobjects. Our Partition-from-Motion is achieved by a bone-centered\ninitialization, a bone limit loss, and a section normal loss that ensure stable\npart division even when the training poses are limited. We also present a\nminimal perimeter loss for SDF to suppress extra surfaces and part overlapping.\nAnother core of our method is an adjacent part seaming algorithm that produces\nnon-rigid deformations to maintain the connection between parts which\nsignificantly relieves the part-based artifacts. Under this algorithm, we\nfurther propose \"Competing Parts\", a method that defines blending weights by\nthe relative position of a point to bones instead of the absolute position,\navoiding the generalization problem of neural implicit functions with inverse\nLBS (linear blend skinning). We demonstrate the effectiveness of our method by\nclothed human body reconstruction and animation on the CAPE and the ClothSeq\ndatasets.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.10053", "title": "3D Clothed Human Reconstruction in the Wild", "abstract": "  Although much progress has been made in 3D clothed human reconstruction, most\nof the existing methods fail to produce robust results from in-the-wild images,\nwhich contain diverse human poses and appearances. This is mainly due to the\nlarge domain gap between training datasets and in-the-wild datasets. The\ntraining datasets are usually synthetic ones, which contain rendered images\nfrom GT 3D scans. However, such datasets contain simple human poses and less\nnatural image appearances compared to those of real in-the-wild datasets, which\nmakes generalization of it to in-the-wild images extremely challenging. To\nresolve this issue, in this work, we propose ClothWild, a 3D clothed human\nreconstruction framework that firstly addresses the robustness on in-thewild\nimages. First, for the robustness to the domain gap, we propose a weakly\nsupervised pipeline that is trainable with 2D supervision targets of\nin-the-wild datasets. Second, we design a DensePose-based loss function to\nreduce ambiguities of the weak supervision. Extensive empirical tests on\nseveral public in-the-wild datasets demonstrate that our proposed ClothWild\nproduces much more accurate and robust results than the state-of-the-art\nmethods. The codes are available in here:\nhttps://github.com/hygenie1228/ClothWild_RELEASE.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09193", "title": "NDF: Neural Deformable Fields for Dynamic Human Modelling", "abstract": "  We propose Neural Deformable Fields (NDF), a new representation for dynamic\nhuman digitization from a multi-view video. Recent works proposed to represent\na dynamic human body with shared canonical neural radiance fields which links\nto the observation space with deformation fields estimations. However, the\nlearned canonical representation is static and the current design of the\ndeformation fields is not able to represent large movements or detailed\ngeometry changes. In this paper, we propose to learn a neural deformable field\nwrapped around a fitted parametric body model to represent the dynamic human.\nThe NDF is spatially aligned by the underlying reference surface. A neural\nnetwork is then learned to map pose to the dynamics of NDF. The proposed NDF\nrepresentation can synthesize the digitized performer with novel views and\nnovel poses with a detailed and reasonable dynamic appearance. Experiments show\nthat our method significantly outperforms recent human synthesis methods.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.06955", "title": "Learning Implicit Templates for Point-Based Clothed Human Modeling", "abstract": "  We present FITE, a First-Implicit-Then-Explicit framework for modeling human\navatars in clothing. Our framework first learns implicit surface templates\nrepresenting the coarse clothing topology, and then employs the templates to\nguide the generation of point sets which further capture pose-dependent\nclothing deformations such as wrinkles. Our pipeline incorporates the merits of\nboth implicit and explicit representations, namely, the ability to handle\nvarying topology and the ability to efficiently capture fine details. We also\npropose diffused skinning to facilitate template training especially for loose\nclothing, and projection-based pose-encoding to extract pose information from\nmesh templates without predefined UV map or connectivity. Our code is publicly\navailable at https://github.com/jsnln/fite.\n", "date": "2022-07-14", "categories": []}
{"id": "2206.08428", "title": "EyeNeRF: A Hybrid Representation for Photorealistic Synthesis, Animation\n  and Relighting of Human Eyes", "abstract": "  A unique challenge in creating high-quality animatable and relightable 3D\navatars of people is modeling human eyes. The challenge of synthesizing eyes is\nmultifold as it requires 1) appropriate representations for the various\ncomponents of the eye and the periocular region for coherent viewpoint\nsynthesis, capable of representing diffuse, refractive and highly reflective\nsurfaces, 2) disentangling skin and eye appearance from environmental\nillumination such that it may be rendered under novel lighting conditions, and\n3) capturing eyeball motion and the deformation of the surrounding skin to\nenable re-gazing. These challenges have traditionally necessitated the use of\nexpensive and cumbersome capture setups to obtain high-quality results, and\neven then, modeling of the eye region holistically has remained elusive. We\npresent a novel geometry and appearance representation that enables\nhigh-fidelity capture and photorealistic animation, view synthesis and\nrelighting of the eye region using only a sparse set of lights and cameras. Our\nhybrid representation combines an explicit parametric surface model for the\neyeball with implicit deformable volumetric representations for the periocular\nregion and the interior of the eye. This novel hybrid model has been designed\nto address the various parts of that challenging facial area - the explicit\neyeball surface allows modeling refraction and high-frequency specular\nreflection at the cornea, whereas the implicit representation is well suited to\nmodel lower-frequency skin reflection via spherical harmonics and can represent\nnon-surface structures such as hair or diffuse volumetric bodies, both of which\nare a challenge for explicit surface models. We show that for high-resolution\nclose-ups of the eye, our model can synthesize high-fidelity animated gaze from\nnovel views under unseen illumination conditions.\n", "date": "2022-06-16", "categories": []}
{"id": "2207.02031", "title": "AvatarCap: Animatable Avatar Conditioned Monocular Human Volumetric\n  Capture", "abstract": "  To address the ill-posed problem caused by partial observations in monocular\nhuman volumetric capture, we present AvatarCap, a novel framework that\nintroduces animatable avatars into the capture pipeline for high-fidelity\nreconstruction in both visible and invisible regions. Our method firstly\ncreates an animatable avatar for the subject from a small number (~20) of 3D\nscans as a prior. Then given a monocular RGB video of this subject, our method\nintegrates information from both the image observation and the avatar prior,\nand accordingly recon-structs high-fidelity 3D textured models with dynamic\ndetails regardless of the visibility. To learn an effective avatar for\nvolumetric capture from only few samples, we propose GeoTexAvatar, which\nleverages both geometry and texture supervisions to constrain the\npose-dependent dynamics in a decomposed implicit manner. An avatar-conditioned\nvolumetric capture method that involves a canonical normal fusion and a\nreconstruction network is further proposed to integrate both image observations\nand avatar dynamics for high-fidelity reconstruction in both observed and\ninvisible regions. Overall, our method enables monocular human volumetric\ncapture with detailed and pose-dependent dynamics, and the experiments show\nthat our method outperforms state of the art. Code is available at\nhttps://github.com/lizhe00/AvatarCap.\n", "date": "2022-07-05", "categories": []}
{"id": "2206.08343", "title": "Realistic One-shot Mesh-based Head Avatars", "abstract": "  We present a system for realistic one-shot mesh-based human head avatars\ncreation, ROME for short. Using a single photograph, our model estimates a\nperson-specific head mesh and the associated neural texture, which encodes both\nlocal photometric and geometric details. The resulting avatars are rigged and\ncan be rendered using a neural network, which is trained alongside the mesh and\ntexture estimators on a dataset of in-the-wild videos. In the experiments, we\nobserve that our system performs competitively both in terms of head geometry\nrecovery and the quality of renders, especially for the cross-person\nreenactment. See results https://samsunglabs.github.io/rome/\n", "date": "2022-06-16", "categories": []}
{"id": "2206.07036", "title": "Accurate 3D Body Shape Regression using Metric and Semantic Attributes", "abstract": "  While methods that regress 3D human meshes from images have progressed\nrapidly, the estimated body shapes often do not capture the true human shape.\nThis is problematic since, for many applications, accurate body shape is as\nimportant as pose. The key reason that body shape accuracy lags pose accuracy\nis the lack of data. While humans can label 2D joints, and these constrain 3D\npose, it is not so easy to \"label\" 3D body shape. Since paired data with images\nand 3D body shape are rare, we exploit two sources of information: (1) we\ncollect internet images of diverse \"fashion\" models together with a small set\nof anthropometric measurements; (2) we collect linguistic shape attributes for\na wide range of 3D body meshes and the model images. Taken together, these\ndatasets provide sufficient constraints to infer dense 3D shape. We exploit the\nanthropometric measurements and linguistic shape attributes in several novel\nways to train a neural network, called SHAPY, that regresses 3D human pose and\nshape from an RGB image. We evaluate SHAPY on public benchmarks, but note that\nthey either lack significant body shape variation, ground-truth shape, or\nclothing variation. Thus, we collect a new dataset for evaluating 3D human\nshape estimation, called HBW, containing photos of \"Human Bodies in the Wild\"\nfor which we have ground-truth 3D body scans. On this new benchmark, SHAPY\nsignificantly outperforms state-of-the-art methods on the task of 3D body shape\nestimation. This is the first demonstration that 3D body shape regression from\nimages can be trained from easy-to-obtain anthropometric measurements and\nlinguistic shape attributes. Our model and data are available at:\nshapy.is.tue.mpg.de\n", "date": "2022-06-14", "categories": []}
{"id": "2206.03373", "title": "Garment Avatars: Realistic Cloth Driving using Pattern Registration", "abstract": "  Virtual telepresence is the future of online communication. Clothing is an\nessential part of a person's identity and self-expression. Yet, ground truth\ndata of registered clothes is currently unavailable in the required resolution\nand accuracy for training telepresence models for realistic cloth animation.\nHere, we propose an end-to-end pipeline for building drivable representations\nfor clothing. The core of our approach is a multi-view patterned cloth tracking\nalgorithm capable of capturing deformations with high accuracy. We further rely\non the high-quality data produced by our tracking method to build a Garment\nAvatar: an expressive and fully-drivable geometry model for a piece of\nclothing. The resulting model can be animated using a sparse set of views and\nproduces highly realistic reconstructions which are faithful to the driving\nsignals. We demonstrate the efficacy of our pipeline on a realistic virtual\ntelepresence application, where a garment is being reconstructed from two\nviews, and a user can pick and swap garment design as they wish. In addition,\nwe show a challenging scenario when driven exclusively with body pose, our\ndrivable garment avatar is capable of producing realistic cloth geometry of\nsignificantly higher quality than the state-of-the-art.\n", "date": "2022-06-07", "categories": []}
{"id": "2606.23062", "title": "VolHuMe: a High-Resolution Large Scale Dataset of Volumetric Human Meshes", "abstract": "We introduce VolHuMe, a dataset of high-quality 4D human scans captured with a state-of-the-art volumetric studio using 64 RGB and 32 depth cameras. VolHuMe contains individual captures of 104 subjects and provides extensive ground truth, including SMPL-X, high-resolution meshes, multi-view RGB/depth images, rigged meshes, point clouds, garment segmentation, and detailed hand and facial geometry. Unlike prior datasets that primarily rely on full-body imagery, VolHuMe uses a close-range, high-resolution capture setup that preserves fine-grained body-part details, improving geometric fidelity and texture resolution. We benchmark VolHuMe on state-of-the-art methods across 3D and 4D human reconstruction tasks, showcasing the dataset's quality and exposing the limitations of current evaluation testbeds.", "date": "2026-06-22", "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": "2606.27720", "title": "Scene and Human in One World: Reconstruction in a Feedforward Pass", "abstract": "Reconstructing humans in dynamic scenes from moving monocular cameras remains challenging due to scale ambiguity, human-scene misalignment, and occlusion interference. Rather than treating human mesh recovery and scene reconstruction as separate tasks, we believe that accurate human-scene reconstruction requires the two tasks to mutually inform each other: parametric human models offer semantic structure and metric-scale priors, while scene geometry provides spatial context for human localization and alignment. Built on this insight, we introduce SHOW, a mask-promptable human mesh recovery framework that couples feed-forward 3D scene reconstruction with Human Mesh Recovery in a unified metric space. SHOW injects human semantics and scale priors from parametric human models into normalized point-map prediction, enabling metric-scale scene reconstruction from inherently scale-ambiguous monocular input. In turn, the recovered scene geometry constrains human mesh estimation, encouraging spatially consistent human placement and improved human-scene alignment. To handle complex multi-person and cluttered scenes, SHOW further incorporates a promptable masking mechanism that enables flexible target-human selection while suppressing background distractions and occlusion interference. Through joint training, the model learns both human-aware geometric features and geometry-constrained human features, producing aligned metric-scale reconstructions from monocular human-centric videos. Extensive experiments demonstrate that SHOW improves metric-scale consistency, human-scene alignment, and reconstruction accuracy under challenging camera motion, occlusion, and cluttered backgrounds.", "date": "2026-06-26", "categories": []}
{"id": "2606.28144", "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": "2606.29333", "title": "HiReFF: High-Resolution Feedforward Human Reconstruction from Uncalibrated Sparse-View Video", "abstract": "Uncalibrated volumetric video streaming for human reconstruction is essential for holographic communication and AR/VR, yet remains challenging due to the need for temporal consistency and computational efficiency from sparse-view inputs. Existing methods rely on per-scene optimization or calibrated cameras, while recent feed-forward models are limited to low-resolution (0.5K) single-frame synthesis. We present HiReFF, a feed-forward method for 2K-resolution 360{\\deg} human video reconstruction from uncalibrated sparse-view videos. Our framework decomposes the problem into two key tasks: foreground 3D Gaussian reconstruction from sparse-view videos (four views separated by 90{\\deg}) and computationally efficient high-resolution synthesis. To enable the former, we propose Scale-synchronized Camera Calibration to resolve scale ambiguity for multi-view supervision, and Gaussian-wise Foreground Masking to reconstruct clean foregrounds by modulating Gaussian parameters. For efficient high-resolution synthesis, our High-resolution Side-tuning achieves 2K rendering by augmenting the Gaussian head with supplementary features while keeping the backbone at 0.5K, drastically reducing computational overhead. Experiments demonstrate that HiReFF significantly outperforms existing methods in high-resolution streaming volumetric video reconstruction. https://iridescentjiang.github.io/HiReFF", "date": "2026-06-28", "categories": []}
{"id": "2606.30347", "title": "FFAvatar: Feed-Forward 4D Head Avatar Reconstruction from Sparse Portrait Images", "abstract": "We present FFAvatar, a Transformer-based 3D Gaussian framework for fast construction of high-quality and animatable 4D head avatars from one or more reference portrait images. Unlike existing feed-forward approaches that require a fixed number of input views, FFAvatar supports incremental reconstruction, progressively refining the avatar representation as additional reference images become available. At the core of our method is an alternating attention mechanism that disentangles identity appearance from expression and viewpoint variations, enabling the reconstruction of a canonical 3D appearance that remains consistent across poses and facial expressions. To balance visual fidelity and computational efficiency, we introduce a sparse-to-dense learning paradigm. Coarse appearance features are first learned using sparse primitives anchored to the FLAME vertex level and are subsequently densified in the UV domain to capture fine-grained geometric and texture details. We further propose a plug-and-play motion refinement module that enables subject-specific dynamic personalization by modeling residual motion beyond parametric deformation. Extensive experiments demonstrate that FFAvatar efficiently produces high-fidelity and controllable 4D head avatars, achieving superior flexibility, driving efficiency, and identity-consistent rendering across diverse expressions and viewpoints.", "date": "2026-06-29", "categories": []}
{"id": "2606.31115", "title": "JacobianAvatar: Temporally Consistent Semi-rigid Avatar Reconstruction from a Monocular Video", "abstract": "Generating realistic human avatars in complex motions--such as clothing dynamics--requires modeling of global and local deformations which remains challenging in monocular settings. We address this problem by leveraging neural Jacobian fields (NJFs) for representing semi-rigid deformations. We train self-supervised neural networks for predicting Jacobian matrices that give the pose-dependent deformations, by solving a Poisson equation. However, monocular input presents several difficulties such as self-occluded regions and invisible surfaces. To address these issues, we introduce three key components: a constrained Poisson solver, signed distance-based Jacobian regularization, and a deformation-guided residual flow loss, which together suppress boundary artifacts, recover frequently occluded regions such as armpits and thighs, and enforce temporal consistency during motion. Experiments on benchmark and in-the-wild videos demonstrate that our method generates temporally stable and geometrically coherent avatars, outperforming state-of-the-art approaches.", "date": "2026-06-30", "categories": []}
{"id": "2606.32036", "title": "PointSplat: Compact Gaussian Splatting via Human-Centric Prediction", "abstract": "Producing 3D human representations from input views on the fly is essential for immersive live streaming systems, where representation compactness is as critical as high fidelity given limited computational power and transmission bandwidth. Although recent feed-forward reconstruction methods achieve impressive quality through the view-centric prediction of 3D representations, they repeatedly encode the same subject content across multiple views, leading to significant inter-view redundancy. Our key insight is to perform predictions directly in 3D space, enabling the network to learn and produce a highly compact representation. To this end, we propose PointSplat, a novel human-centric approach that directly infers Gaussian primitives from an input point set. The proposed method first estimates a coarse geometric proxy and performs ray casting to prune redundant points and establish explicit 2D--3D correspondences. Subsequently, it employs a Point-Image Transformer to fuse appearance and geometry features, predicting Gaussian attributes in a single forward pass. This design restricts predictions to foreground regions of interest, substantially reducing the total number of Gaussians while improving novel-view rendering quality. Extensive experiments demonstrate that PointSplat achieves higher efficiency and quality while exhibiting strong robustness to variations in view count and image resolution across multiple datasets.", "date": "2026-06-30", "categories": []}