{"id": "2606.20542", "title": "CalTennis: Large Multi-View Tennis Video Dataset and Benchmark of Monocular-to-3D Pose Estimation", "abstract": "The Caltech Tennis Dataset (CalTennis) is a large-scale video benchmark for evaluating monocular-to-3D pose estimation in the wild. CalTennis comprises over 11 million frames (51 hours) of tennis practice and match play from 40 players, captured with 2-6 synchronized cameras at 60 Hz. It is 10 times larger than existing in-the-wild human motion video datasets and 3 times larger than existing MOCAP-ground-truthed datasets, and it is the first large-scale benchmark to provide synchronized multi-view recordings of expert athletic motion. The multi-view setup enables inexpensive, label-free evaluation of monocular-to-3D pose estimation algorithms. We describe a simple, standardized protocol that enables data collection without specialized equipment or expertise, along with fully automated video calibration and synchronization. Benchmarking state-of-the-art monocular-to-3D pose methods on CalTennis, we find that while 3D joint angle recovery is now quite accurate, all models struggle to estimate depth and foot contact consistently. We further propose two novel performance metrics, footwork and stability, as well as qualitatively study body shape inconsistency. These metrics expose previously underexplored failure modes and point to concrete opportunities for improvement in pose estimation and action analysis.", "date": "2026-06-18", "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": "2606.13700", "title": "C-MambaPose: A Physics-Informed Complex Mamba Framework for Cross-Environment WiFi Human Pose Estimation", "abstract": "Human pose estimation (HPE) utilizing wireless WiFi signals has emerged as a promising technology owing to its device-free nature, privacy preservation, and robustness against occlusion and poor lighting. However, existing methods often overlook the physical complex phase information of WiFi signals and fail to generalize across diverse environments due to severe domain shifts. In this paper, we present C-MambaPose, a physics-informed complex-valued Mamba-GraFormer hybrid framework for robust cross-environment WiFi-based 3D HPE. Our framework first sanitizes raw WiFi Channel State Information (CSI) phase errors and constructs a phase-preserving complex-valued representation. We then employ a Spatiotemporal Complex Mamba encoder with a dynamic selective receptive field to capture fine-grained phase dynamics. A cross-attention joint-query mapper maps the unstructured sequence tokens to human joints, which are decoded by a Graph Convolutional Network (GCN) to predict anatomically coherent 3D coordinates. Extensive evaluations on the MM-Fi dataset show that C-MambaPose achieves competitive or superior performance to state-of-the-art baselines across all settings, setting a new state-of-the-art specifically on the challenging cross-environment split, requiring only 3.78 M parameters-an 83.1\\% reduction compared to GraphPose-Fi~\\cite{chen2026graph} and an 85.7\\% reduction compared to MetaFi++~\\cite{zhou2023metafi++}, while maintaining a comparable size to DT-Pose~\\cite{chen2025towards} (which is only 18\\% smaller) but achieving significantly superior performance without requiring any pretraining. Our code is publicly available at https://github.com/phucngvinuni/cmampose.git.", "date": "2026-06-05", "categories": []}
{"id": "2606.12988", "title": "A Machine Learning Framework for Real-Time Personalized Ergonomic Pose Analysis", "abstract": "This paper introduces a new methodology for real-time prediction of ergonomic and non-ergonomic human poses using volumetric video data in three dimensions. Although the methodology was designed for ergonomic assessments, it can be adapted to other applications requiring real-time analysis of human posture. One aspect that makes this system stand out is its ability to analyze 3D point clouds during the assessment, enabling computation from multiple angles. This overcomes a critical limitation of cameras which provide often a fixed viewpoint, thereby restricting the data available for a thorough postural evaluation, especially when occlusions occur. The system continuously and automatically performs pose inference using the chosen perspective on the real-time streaming data; however, only the poses manually selected and labeled by the user are used to train the personalized deep learning classifier. The methodology has been refined through a case study in which RGB-D cameras captured subjects performing load-lifting tasks, enabling real-time skeletal labeling. The model was trained on this data and, following the training phase, performs inference on new streaming data in real time. This research offers a scalable and pragmatic approach for real-time ergonomic evaluation by combining state-of-the-art 3D data technologies and traditional 2D pose estimation algorithms. It addresses the increasing need for safety and health monitoring in workplace environments, marking a notable contribution to the domain.", "date": "2026-06-11", "categories": []}
{"id": "2606.11578", "title": "Contactless 3D Human Body Measurement Using Depth Cameras for Smart Health Monitoring", "abstract": "Contactless body measurement technologies are becoming increasingly significant for smart health monitoring, digital health applications, and remote patient assessment. Traditional anthropometric measurements typically necessitate physical contact and trained personnel, which may constrain scalability in remote healthcare settings. In this study, we introduce a depth camera-based framework for estimating human body measurements utilizing 3D point cloud data. An Orbbec Astra 2 depth camera was employed to capture RGB images, depth maps, and 3D point clouds of participants. The captured point cloud was processed using Python-based tools, including Open3D, NumPy, and OpenCV, to segment the human body from the background. Key anthropometric measurements, such as height and arm span, were computed. The measurements were obtained through a combination of spatial filtering and landmark selection on the 3D point cloud, followed by the projection of the computed measurements onto the corresponding RGB image using camera intrinsic parameters. In addition to linear measurements, the approximate body volume and visible surface area were estimated using voxel-based occupancy analysis and mesh-based surface reconstruction methods. The experimental results from a single depth capture demonstrated that accurate body measurements and geometric estimates could be obtained from depth camera data without physical contact. This study provides a foundation for future real-time systems that integrate depth sensing with intelligent health monitoring and generative AI models for smart healthcare applications.", "date": "2026-06-10", "categories": []}
{"id": "2606.07419", "title": "DisPOSE: Projected Polystochastic Diffusion for Self-Supervised Multi-View 3D Human Pose Estimation", "abstract": "Recovering 3D human poses for multiple individuals from different camera views is a fundamental bottleneck for analyzing interacting behaviors. Existing self-supervised approaches leverage synthetic catalogues of 3D poses; however, this leads to poor generalization in real-world scenarios due to distribution shifts. We therefore introduce DisPOSE, a self-supervised framework that approximates the inherently discrete multi-view person-assignment problem as a generative diffusion process over the space of polystochastic tensors. By employing differentiable Sinkhorn projections during denoising, our model learns to guide solutions toward valid and feasible assignments based on 2D image priors. The complete 3D skeletons of localized individuals are then regressed using a Hypergraph-Convolutional Decoder that explicitly models relational structures and articulated joints across multiple views. The proposed approach outperforms current state-of-the-art self-supervised methods on standard datasets and demonstrates strong performance on a newly proposed benchmark featuring highly occluded scenes from surgical operating rooms. Our diffusion-based localization demonstrates high label efficiency, retaining 99% of its performance with only 10% of the pseudo-labels. Notably, disentangling the assignment and root regression components while maintaining differentiability makes DisPOSE nearly agnostic to different camera arrangements.", "date": "2026-06-05", "categories": []}
{"id": "2601.09698", "title": "COMPOSE: Hypergraph Cover Optimization for Multi-view 3D Human Pose Estimation", "abstract": "3D human pose estimation from sparse multi-view camera rigs is an essential task for numerous applications, including action recognition, sports analysis, and human-robot interaction. While learned methods dominate the field on benchmarks, they require large annotated datasets; training-free optimization-based methods remain promising as they circumvent 3D supervision by solving a correspondence problem across views from 2D detections. Existing combinatorial formulations rely on pairwise associations to model this correspondence problem and enforce global consistency across views only as a downstream constraint. However, reconciling locally plausible pairwise matches becomes brittle under occlusion and noisy detections, where local errors propagate globally. We propose COMPOSE, which recasts multi-view 3D human pose estimation as a weighted exact-cover optimization over a hypergraph of person hypotheses. Our formulation replaces pairwise association and post-hoc consistency enforcement with a single global combinatorial objective. To address the exponentially large candidate space, we introduce a geometric pruning strategy alongside two complementary solvers: an exact Integer Linear Programming formulation and a scalable relaxation via Belief Propagation. Without any 3D supervision, COMPOSE improves average precision by up to 31 points over the best optimization-based method and 13 points over self-supervised learned methods, demonstrating the effectiveness of higher-order combinatorial association for training-free multi-view 3D human pose estimation.", "date": "2026-01-14", "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": "2605.00242", "title": "MAEPose: Self-Supervised Spatiotemporal Learning for Human Pose Estimation on mmWave Video", "abstract": "Millimetre-wave (mmWave) radar offers a more privacy-preserving alternative to RGB-based human pose estimation. However, existing methods typically rely on pre-extracted intermediate representations such as sparse point clouds or spectrogram images, where the rich spatiotemporal information naturally present in radar video streams is discarded for model learning, while such signal processing adds system complexity. In addition, existing solutions are mainly conducted in an end-to-end supervised manner without leveraging unlabelled raw video streams to learn generalized representations. In this study, we present MAEPose, a masked autoencoding-based human pose estimation approach that operates directly on mmWave spectrogram videos. MAEPose learns spatiotemporal motion-aware generalized representations from unlabelled radar video, and leverages its heatmap decoder for multi-frame pose estimation predictions. We evaluate it across three datasets based on leave-one-person-out cross-validation with rigorous statistical testing. MAEPose consistently outperforms state-of-the-art baselines by up to 22.1% in MPJPE p<0.05, and maintains robust accuracy under zero-shot bystander interference with only a 6.5% error increase. Ablation studies confirm that both the pre-training and the heatmap decoder contribute substantially, while modality analysis indicates that leveraging Range-Doppler video as input achieves better pose estimation performance than Range-Azimuth or their fusion, with lower computational cost.", "date": "2026-04-30", "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": "2507.11075", "title": "Joint angle based learning to refine kinematic human pose estimation", "abstract": "Marker-free human pose estimation (HPE) has found increasing applications in various fields. Current HPE suffers from occasional errors in keypoint recognition and random fluctuation in keypoint trajectories when analyzing kinematic human poses. The performance of existing deep learning-based models for HPE refinement is considerably limited by inaccurate training datasets in which the keypoints are manually annotated. This paper proposed a novel method to overcome the difficulty, in which the key techniques include: (i) A robust joint angle-based description of kinematic human poses; (ii) Approximating temporal variation of joint angles using high order Fourier series to get reliable \"ground truth\"; (iii) A bidirectional recurrent network is designed as a post-processing module to refine the estimation of single image-based HPE models. Trained with the high-quality dataset constructed using our method, the network demonstrates outstanding performance to correct wrongly recognized joints and smooth their spatiotemporal trajectories. Tests show that joint angle-based refinement (JAR) outperforms the state-of-the-art HPE refinement network in challenging cases like figure skating and breaking. JAR also demonstrates great potential to rectify existing datasets.", "date": "2025-07-15", "categories": []}
{"id": "2605.31551", "title": "SMART: SMPLest-X Mesh Adaptation and RAFT Tracking for Soccer Pose Estimation", "abstract": "We present our approach to the FIFA Skeletal Tracking Challenge 2026, which requires estimating 3D world-space poses of soccer players from broadcast video. Our method finetunes SMPLest-X (ViT-H, 687 M parameters) via a stratified clip split, multi-task depth supervision, and broadcast augmentation, paired with a RAFT dense optical flow camera tracker, foot-plane anchoring, and two-pass temporal smoothing. Against the FIFA baseline score of 1.053 on the validation set, SMART achieves 0.647, a 38.6% improvement; on the held-out test set, SMART scores 0.593 (Global MPJPE: 0.324 m, Local MPJPE: 0.054 m).", "date": "2026-05-29", "categories": []}
{"id": "2605.29953", "title": "Mesh-Aware Epipolar Matching for Multi-View Multi-Person 3D Pose Estimation in Basketball", "abstract": "Multi-view multi-person 3D pose estimation in team sports scenarios remains challenging due to player occlusions, appearance similarity caused by team uniforms, and the scarcity of annotated multi-view data, all of which limit the effectiveness and generalization capability of learning-based methods. In contrast, the performance of training-free approaches is inherently constrained by the accuracy of 2D keypoint detection and the robustness of cross-view association. To address these challenges, we propose Mesh-Aware Epipolar Matching (MAEM), a training-free framework for multi-view multi-person 3D pose estimation. Our method employs a monocular 3D human mesh recovery model as the frontend and introduces a two-stage epipolar matching strategy based on the recovered mesh outputs. Specifically, the proposed framework combines disjoint-set-union-based clustering with per-joint triangulation to achieve robust cross-view association and accurate 3D pose reconstruction. Experiments on two public multi-view basketball datasets demonstrate that MAEM consistently outperforms existing training-free association baselines while achieving competitive RGB-only performance in both indoor and outdoor basketball scenarios. MAEM achieves MPJPE/PA-MPJPE scores of 59.8/40.7 mm on SportCenter EPFL and 74.0/51.8 mm on Human-M3 Basketball, highlighting the effectiveness of dense mesh geometry for cross-view association without requiring target-domain training or fine-tuning.", "date": "2026-05-28", "categories": []}
{"id": "2508.02806", "title": "PyCAT4: A Hierarchical Vision Transformer-based Framework for 3D Human Pose Estimation", "abstract": "Recently, a significant improvement in the accuracy of 3D human pose estimation has been achieved by combining convolutional neural networks (CNNs) with pyramid grid alignment feedback loops. Additionally, innovative breakthroughs have been made in the field of computer vision through the adoption of Transformer-based temporal analysis architectures. Given these advancements, this study aims to deeply optimize and improve the existing Pymaf network architecture. The main innovations of this paper include: (1) Introducing a Transformer feature extraction network layer based on self-attention mechanisms to enhance the capture of low-level features; (2) Enhancing the understanding and capture of temporal signals in video sequences through feature temporal fusion techniques; (3) Implementing spatial pyramid structures to achieve multi-scale feature fusion, effectively balancing feature representations differences across different scales. The new PyCAT4 model obtained in this study is validated through experiments on the COCO and 3DPW datasets. The results demonstrate that the proposed improvement strategies significantly enhance the network's detection capability in human pose estimation, further advancing the development of human pose estimation technology.", "date": "2025-08-04", "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": "2605.26879", "title": "Natural Human Motion Recovery by Aligning High-Order Temporal Dynamics from Monocular Videos", "abstract": "Human motion recovered from monocular videos often appears overly smooth or dynamically inconsistent, even when joint positions are numerically accurate. We observe that this limitation stems from the absence of reliable high-order temporal cues -- velocity and acceleration -- which are essential for reconstructing motion that exhibits realistic momentum, timing, and high-frequency detail. We introduce HTD-Refine, a post-processing framework that augments existing Human Motion Recovery (HMR) pipelines using explicitly estimated high-order temporal dynamics. At the core of our system is PVA-Net, a temporal transformer that infers per-joint 2D positions, 3D velocities, and 3D accelerations directly from a monocular video. These predicted dynamics serve as soft yet informative constraints in a global optimization procedure that refines world-space trajectories, significantly reducing jitter, suppressing over-smoothing, and restoring physically plausible motion. Extensive experiments on challenging in-the-wild benchmarks show that HTD-Refine consistently improves state-of-the-art HMR methods, yielding more accurate global trajectories and substantially more natural motion dynamics. Our results highlight the critical role of high-order temporal modeling in advancing monocular human motion recovery.", "date": "2026-05-26", "categories": []}
{"id": "2601.16763", "title": "Flow Matching for Probabilistic Monocular 3D Human Pose Estimation", "abstract": "Recovering 3D human poses from a monocular camera view is a highly ill-posed problem due to the depth ambiguity. Earlier studies on 3D human pose lifting from 2D often contain incorrect-yet-overconfident 3D estimations. To mitigate the problem, emerging probabilistic approaches treat the 3D estimations as a distribution, taking into account the uncertainty measurement of the poses. Falling in a similar category, we proposed FMPose, a probabilistic 3D human pose estimation method based on the flow matching generative approach. Conditioned on the 2D cues, the flow matching scheme learns the optimal transport from a simple source distribution to the plausible 3D human pose distribution via continuous normalizing flows. The 2D lifting condition is modeled via graph convolutional networks, leveraging the learnable connections between human body joints as the graph structure for feature aggregation. While trade-offs between processing time and precision exist, already in the equal-accuracy comparison, FMPose exhibits significantly faster processing time than the diffusion model, and also offers another faster and more accurate configuration. Experimental results show major improvements of our FMPose over current state-of-the-art methods on two common benchmarks for 3D human pose estimation, namely Human3.6M, MPI-INF-3DHP. Additionally, FMPose shows competitive performance on the more challenging 3DPW dataset. The code implementation is available at https://github.com/cuongle1206/FMPose", "date": "2026-01-23", "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": "2512.20538", "title": "AlignPose: Generalizable 6D Pose Estimation via Multi-view Feature-metric Alignment", "abstract": "Single-view RGB model-based object pose estimation methods achieve strong generalization but are fundamentally limited by depth ambiguity, clutter, and occlusions. Multi-view pose estimation methods have the potential to solve these issues, but existing works rely on precise single-view pose estimates or lack generalization to unseen objects. We address these challenges via the following three contributions. First, we introduce AlignPose, a 6D object pose estimation method that aggregates information from multiple extrinsically calibrated RGB views and does not require any object-specific training or symmetry annotation. Second, the key component of this approach is a new multi-view feature-metric refinement specifically designed for object pose. It optimizes a single, consistent world-frame object pose by minimizing the feature discrepancy between on-the-fly rendered object features and observed image features across all views simultaneously. Third, we report extensive experiments on six datasets (YCB-V, T-LESS, HouseCat6D, ITODD-MV, IPD, XYZ-IBD) using the BOP benchmark evaluation and show that AlignPose outperforms other published methods, especially on challenging industrial datasets where multiple views are readily available in practice.", "date": "2025-12-23", "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": "2508.03578", "title": "RadProPoser: Probabilistic Radar Tensor Human Pose Estimation That Knows Its Limits", "abstract": "Radar-based human pose estimation enables privacy-preserving motion tracking for ambient intelligence, yet the noisy nature of radar sensing makes uncertainty quantification essential. We present RadProPoser, an end-to-end probabilistic framework that predicts three-dimensional body joints with per-joint uncertainties from raw radar tensor data. Using a variational encoder-decoder with spectral attention that fuses real and imaginary radar components across temporal frames, we model aleatoric uncertainty through learnable Gaussian and Laplace distributions. Trained on a new benchmark dataset with optical motion-capture ground truth, our method achieves 6.425 cm mean per-joint position error. The model outputs per-joint aleatoric uncertainties, and isotonic recalibration yields calibrated total uncertainty with expected calibration error of 0.027. Since spectral attention operates on individual radar tensor components, extending to multi-radar configurations requires only concatenating additional input streams. On the HuPR benchmark with dual orthogonal radars, this achieves 5.042 cm MPJPE. The framework runs at 89 frames per second (FPS) on an NVIDIA RTX 3090, exceeding the 15 Hz radar frame rate.", "date": "2025-08-05", "categories": []}
{"id": "2605.20889", "title": "Map-Mono-Ego: Map-Grounded Global Human Pose Estimation from Monocular Egocentric Video", "abstract": "Monocular egocentric human pose estimation is essential for ubiquitous activity monitoring. However, understanding the user's absolute location within the environment remains a challenge. Existing methods primarily focus on relative motion from an initial position, and tend not to account for the wearer's absolute location within an environment. Furthermore, inherent scale ambiguity in monocular vision leads to severe translational drift, limiting long-term tracking without specialized multi-sensor hardware. To address this, we propose MapMonoEgo, a novel framework achieving globally consistent human pose estimation solely from a monocular camera by leveraging a pre-scanned 3D point cloud. We also introduce AIST-Living dataset, a new dataset pairing egocentric video with ground-truth motion in a scanned environment. Experiments demonstrate that our approach significantly outperforms the state-of-the-art baseline, proving its utility for practical monitoring tasks without specialized hardware.", "date": "2026-05-20", "categories": []}
{"id": "2605.19620", "title": "B\\'ezier Degradation Modeling for LiDAR-based Human Motion Capture", "abstract": "LiDAR-based 3D human motion capture has broad applications in fields such as autonomous driving and robotics, where accurate motion reconstruction is crucial. However, existing methods often struggle with unstable inputs and severe occlusions, leading to jittery or even failed pose predictions. To address these challenges, we propose BMLiCap, a coarse-to-fine framework that models motion using temporally compressible B\\'ezier curves. By reducing control points through a trajectory-preserving strategy, we obtain a coherent and learning-friendly motion representation. To reconstruct human actions from LiDAR point-cloud cues, we design a progressive motion-reconstruction module. Specifically, a Time-scale Motion Transformer (TMT) is introduced to predict motion curves at multiple temporal scales, and a Multi-level Motion Aggregator (MMA) is utilized to adaptively fuse the multi-scale curves to recover detailed, temporally coherent poses, effectively bridging observation gaps caused by occlusions and noise. Across four mainstream benchmarks LiDARHuman26M, FreeMotion, NoiseMotion, and SLOPER4D, BMLiCap achieves state-of-the-art accuracy and temporal continuity in complex scenes, demonstrating its ability to compensate for severe occlusions and reduce prediction jitter.", "date": "2026-05-19", "categories": []}
{"id": "2303.11675", "title": "ReBaR: Reference-Based Reasoning for Robust Pose Estimation from Monocular Images", "abstract": "R}easoning for Robust Human Pose and Shape Estimation), designed to estimate human body shape and pose from single-view images. ReBaR effectively addresses the challenges of occlusions and depth ambiguity by learning reference features for part regression reasoning. Our approach starts by extracting features from both body and part regions using an attention-guided mechanism. Subsequently, these features are used to encode additional part-body dependencies for individual part regression, with part features serving as queries and the body feature as a reference. This reference-based reasoning allows our network to infer the spatial relationships of occluded parts with the body, utilizing visible parts and body reference information. ReBaR outperforms contemporary methods on three benchmark datasets and still maintains competitive advantages among recent new approaches. Demonstrating significant improvement in handling depth ambiguity and occlusion. These results strongly support the effectiveness of our reference-based framework for estimating human body shape and pose from single-view images.", "date": "2023-03-21", "categories": []}
{"id": "2605.14854", "title": "FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery", "abstract": "Human Mesh Recovery (HMR) is fundamentally ambiguous: under occlusion or weak depth cues, multiple 3D bodies can explain the same image evidence. This ambiguity is not uniform across the body, as torso pose and root structure are often relatively well constrained, whereas distal articulations such as the arms and legs are more uncertain. Building on this observation, we propose FactorizedHMR, a two-stage framework that treats these two regimes differently. A deterministic regression module first recovers a stable torso-root anchor, and a probabilistic flow-matching module then completes the remaining non-torso articulation. To make this completion reliable, we combine a composite target representation with geometry-aware supervision and feature-aware classifier-free guidance, preserving the torso-root anchor while improving single-reference recovery of ambiguity-prone articulation. We also introduce a synthetic data pipeline that provides the paired image-camera-motion supervision under diverse viewpoints. Across camera-space and world-space benchmarks, FactorizedHMR remains competitive with strong baselines, with the clearest gains in occlusion-heavy recovery and drift-sensitive world-space metrics.", "date": "2026-05-14", "categories": []}
{"id": "2605.17638", "title": "TouchMap-OR: Multi-View 3D Mapping of Hand-Surface Contacts", "abstract": "Hand-surface interactions between clinicians, patients, and medical equipment play a central role in pathogen transmission during medical procedures. However, these interactions remain largely unobserved, as current infection-prevention practices rely on manual observation and cannot reconstruct detailed contact histories. In this work we formulate the problem of identity-resolved hand-surface interaction reconstruction in operating rooms and introduce TouchMap-OR, a multi-view RGB-D vision system that models clinicians, articulated hand geometry, and the semantic structure of the clinical environment to infer when and where contacts occur. The system reconstructs globally consistent multi-person 3D skeleton tracks across cameras while estimating articulated MANO hand meshes from RGB observations aligned to depth data. Multi-view hand reconstructions are fused and associated with tracked clinicians to obtain consistent left and right hand trajectories. A semantic 3D model of the operating room is built from multi-view segmentation and depth fusion, enabling reconstructed hand trajectories to be mapped to specific surfaces, including medical equipment, movable objects, and patient body sites. Temporal hand-surface proximity is used to infer contact episodes describing which clinician touched which surface and when. We evaluate TouchMap-OR on recordings from three real anesthesia inductions with manually annotated contact events. TouchMap-OR achieves 0.75 binary contact F1, outperforming tracking-based baselines while maintaining comparable multi-person tracking accuracy and achieving 0.96 identity attribution accuracy.", "date": "2026-05-17", "categories": []}
{"id": "2605.10100", "title": "HYPERPOSE: Hyperbolic Kinematic Phase-Space Attention for 3D Human Pose Estimation", "abstract": "We introduce HYPERPOSE, a novel 3D human pose estimation framework that performs spatio-temporal reasoning entirely within the Lorentz model of hyperbolic space $\\mathbb{H}^d$ to natively preserve the hierarchical tree topology of the human skeleton. Current state-of-the-art pose estimators aim to capture complex joint dynamics by relying on transformers and graph convolutional networks. Since these architectures operate exclusively in Euclidean space which fundamentally mismatches the inherent tree structure of the human body, these methods inevitably suffer from exponential volume distortion and struggle to maintain structural coherence. To this end, we depart from flat spaces and aim to improve geometric fidelity with Hyperbolic Kinematic Phase-Space Attention (HKPSA), natively embedding complex joint relationships without distortion, alongside a multi-scale windowed hyperbolic attention mechanism that efficiently models temporal dynamics in $O(TW)$ complexity. Furthermore, to overcome the well-known instability of training non-Euclidean manifolds, HYPERPOSE introduces a novel Riemannian loss suite and an uncertainty-weighted curriculum, enforcing physical geodesic constraints like bone length and velocity consistency. Extensive evaluations on the Human3.6M and MPI-INF-3DHP datasets demonstrate that HYPERPOSE achieves state-of-the-art structural and temporal coherence, significantly reducing both volume distortion and velocity error, while establishing new state-of-the-art benchmarks in overall positional accuracy.", "date": "2026-05-11", "categories": []}
{"id": "2605.15583", "title": "Unsupervised 3D Human Pose Estimation via Conditional Multi-view Ancestral Sampling", "abstract": "We propose a method of estimating a 3D human pose from a single view without 3D supervision. The key to our method is to leverage the 2D diffusion priors of motion diffusion models (MDMs) pre-trained on large 2D human pose datasets. Specifically, we extend multi-view ancestral sampling of diffusion models to the task of 2D-3D lifting of human pose. To this end, we newly propose a conditional multi-view ancestral sampling (cMAS) that optimizes the 3D pose such that its multi-view projections follow the manifold in 2D MDM noise space, while conditioning the 3D pose to match the given 2D poses and anatomical constraints of humans. Experiments on the Yoga dataset demonstrate that our method achieves better cross-domain performance compared to state-of-the-art supervised and unsupervised 3D pose estimation methods, including extreme human poses where 3D supervision is unavailable. Code is available at: https://github.com/asaa0001/c-MAS.", "date": "2026-05-15", "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": "2605.14525", "title": "From Sparse to Dense: Spatio-Temporal Fusion for Multi-View 3D Human Pose Estimation with DenseWarper", "abstract": "In multi-view 3D human pose estimation, models typically rely on images captured simultaneously from different camera views to predict a pose at a specific moment. While providing accurate spatial information, this traditional approach often overlooks the rich temporal dependencies between adjacent frames. We propose a novel 3D human pose estimation input method: the sparse interleaved input to address this. This method leverages images captured from different camera views at various time points (e.g., View 1 at time $t$ and View 2 at time $t+\\delta$), allowing our model to capture rich spatio-temporal information and effectively boost performance. More importantly, this approach offers two key advantages: First, it can theoretically increase the output pose frame rate by N times with N cameras, thereby breaking through single-view frame rate limitations and enhancing the temporal resolution of the production. Second, using a sparse subset of available frames, our method can reduce data redundancy and simultaneously achieve better performance. We introduce the DenseWarper model, which leverages epipolar geometry for efficient spatio-temporal heatmap exchange. We conducted extensive experiments on the Human3.6M and MPI-INF-3DHP datasets. Results demonstrate that our method, utilizing only sparse interleaved images as input, outperforms traditional dense multi-view input approaches and achieves state-of-the-art performance. The source code for this work is available at: https://github.com/lingli1724/DenseWarper-ICLR2026", "date": "2026-05-14", "categories": []}
{"id": "2510.02043", "title": "Zero-shot Human Pose Estimation using Diffusion-based Inverse solvers", "abstract": "Pose estimation refers to tracking a human's full body posture, including their head, torso, arms, and legs. The problem is challenging in practical settings where the number of body sensors are limited. Past work has shown promising results using conditional diffusion models, where the pose prediction is conditioned on both <location, rotation> measurements from the sensors. Unfortunately, nearly all these approaches generalize poorly across users, primarly because location measurements are highly influenced by the body size of the user. In this paper, we formulate pose estimation as an inverse problem and design an algorithm capable of zero-shot generalization. Our idea utilizes a pre-trained diffusion model and conditions it on rotational measurements alone; the priors from this model are then guided by a likelihood term, derived from the measured locations. Thus, given any user, our proposed InPose method generatively estimates the highly likely sequence of poses that best explains the sparse on-body measurements.", "date": "2025-10-02", "categories": []}
{"id": "2605.05680", "title": "MotionGRPO: Overcoming Low Intra-Group Diversity in GRPO-Based Egocentric Motion Recovery", "abstract": "This paper studies full-body 3D human motion recovery from head-mounted device signals. Existing diffusion-based methods often rely on global distribution matching, leading to local joint reconstruction errors. We propose MotionGRPO, a novel framework leveraging reinforcement learning post-training to inject fine-grained guidance into the diffusion process. Technically, we model diffusion sampling as a Markov decision process optimized via Group Relative Policy Optimization (GRPO). To this end, we introduce a hybrid reward mechanism that combines a learned conditioned perceptual model for global visual plausibility and explicit constraints for local joint precision. Our key technical insight is that policy optimization in diffusion-based recovery suffers from vanishing gradients due to limited intra-group sample diversity. To address this, we further introduce a noise-injection strategy that explicitly increases sample variance and stabilizes learning. Extensive experiments demonstrate that MotionGRPO achieves state-of-the-art performance with superior visual fidelity", "date": "2026-05-07", "categories": []}
{"id": "2605.07552", "title": "VIMCAN: Visual-Inertial 3D Human Pose Estimation with Hybrid Mamba-Cross-Attention Network", "abstract": "The rapid advances in deep learning have significantly enhanced the accuracy of multimodal 3D human pose estimation (HPE). However, the state-of-the-art (SOTA) HPE pipelines still rely on Transformers, whose quadratic complexity makes real-time processing for long sequences impractical. Mamba addresses this issue through selective state-space modeling, enabling efficient sequence processing without sacrificing representational power. Nevertheless, it struggles to capture complex spatial dependencies in multimodal settings. To bridge this gap, we propose VIMCAN, a hybrid architecture that combines the efficient sequence modeling of Mamba with the spatial reasoning of Cross-Attention, and performs robust visual-inertial fusion and human pose estimation between RGB keypoints and wearable IMU data. By leveraging Mamba's dynamic parameterization for temporal modeling and Attention for spatial dependency extraction, VIMCAN achieves superior accuracy, with mean per-joint position errors (MPJPE) of 17.2 mm on TotalCapture and 45.3 mm on 3DPW. VIMCAN outperforms prior Transformer-based and other SOTA approaches while supporting real-time inference at over 60 frames per second on consumer-grade hardware. The source code is available at \\href{https://github.com/Eddieyzp/VIMCAN}{this GitHub repository}.", "date": "2026-05-08", "categories": []}
{"id": "2605.08806", "title": "L2A: Learning to Accumulate Pose History for Accurate 3D Human Pose Estimation", "abstract": "Existing 2D-3D lifting human pose estimation methods have achieved strong performance. But the utilization of historical pose representations across network depth was overlooked. In current pipelines, information is propagated through fixed residual connections, which restricts effective reuse of early-layer features such as fine-grained spatial structures and short-term motion cues. However, naively incorporating historical features across layers is non-trivial. We further identify that maintaining a consistent representation space across layers is a prerequisite for effective cross-layer feature aggregation. To address this issue, we propose a history-aware framework that enables effective network cross-layer history feature utilization. Specifically, we adopt a spatial-temporal parallel Transformer backbone to prevent alternating spatial-temporal transformations during sequential processing, thereby maintaining a consistent representation space. Building upon this, we introduce a History Pose Accumulation (HPA) mechanism that adaptively aggregates features from all preceding layers to enhance current representations. Furthermore, we propose a Layer Pose History Aggregation (LPA) module that transforms layer pose features into a compact and structured form, reducing redundancy and enabling more stable aggregation. Extensive experiments demonstrate that our approach achieves state-of-the-art performance on benchmarks.", "date": "2026-05-09", "categories": []}
{"id": "2605.12198", "title": "Enhancing Domain Generalization in 3D Human Pose Estimation through Controllable Generative Augmentation", "abstract": "Pedestrian motion, due to its causal nature, is strongly influenced by domain gaps arising from discrepancies between training and testing data distributions. Focusing on 3D human pose estimation, this work presents a controllable human pose generation framework that synthesizes diverse video data by systematically varying poses, backgrounds, and camera viewpoints. This generative augmentation enriches training datasets, enhances model generalization, and alleviates the limitations of existing methods in handling domain discrepancies. By leveraging both indoor/real-world and outdoor/virtual datasets, we perform cross-domain data fusion and controllable video generation to construct enriched training data, tailored to realistic deployment settings. Extensive experiments show that the augmented datasets significantly improve model performance on unseen scenarios and datasets, validating the effectiveness of the proposed approach.", "date": "2026-05-12", "categories": []}
{"id": "2605.08530", "title": "A Two-Stage Motion-Aware Framework for mmWave-based Human Mesh Recovery", "abstract": "Millimeter-wave (mmWave) radar has emerged as a promising sensing modality for human perception due to its robustness under challenging environmental conditions and strong privacy-preserving properties. However, recovering accurate 3D human body meshes from radar observations remains difficult due to severe signal clutter and the inherently partial nature of radar measurements. Previous works typically adopt end-to-end frameworks that directly regress human body parameters from raw radar data, without decoupling signal interpretation from geometric reasoning or exploiting temporal motion cues, limiting learning performance. To address this, we propose a two-stage framework for radar-based human body reconstruction. First, we introduce a human reflection extraction module that performs coarse-to-fine localization and voxel-wise segmentation to produce a confidence-weighted radar volume encoding voxel-level human likelihood. Second, we design a motion-aware mesh recovery network that reconstructs the human body by jointly modeling per-frame geometry and inter-frame dynamics using a dual-branch architecture. Extensive experiments demonstrate that the proposed method outperforms existing approaches while maintaining computational efficiency.", "date": "2026-05-08", "categories": []}
{"id": "2605.08606", "title": "Egocentric Whole-Body Human Mesh Recovery with Prior-Guided Learning", "abstract": "Egocentric human mesh recovery (HMR) from monocular head-mounted cameras is increasingly important for AR/VR applications, but remains challenging due to the lack of reliable ground-truth (GT) annotations based on parametric human body models such as SMPL and SMPL-X for real egocentric images. Existing egocentric HMR methods typically rely on pseudo-GT and focus on body pose estimation, which limits their ability to recover fine-grained whole-body details such as hands and face. We study egocentric whole-body human mesh recovery and propose a prior-guided learning framework that reconstructs whole-body meshes from a single egocentric image. We construct more accurate optimization-based pseudo-GT aligned with 3D joint supervision, and leverage multiple priors by adapting an exocentric HMR foundation model together with a diffusion-based pose prior. A deterministic undistortion module is further adopted to handle fisheye distortions in egocentric images. Experiments across multiple egocentric benchmarks demonstrate improved whole-body reconstruction compared to state-of-the-art methods, and show that our optimization-based pseudo-GT is substantially more accurate than existing regression-based pseudo-GT. To facilitate reproducibility, the code and dataset annotations are publicly available at https://github.com/naso06/EgoSMPLX.", "date": "2026-05-09", "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": "2312.06409", "title": "LiCamPose: Combining Multi-View LiDAR and RGB Cameras for Robust Single-timestamp 3D Human Pose Estimation", "abstract": "Several methods have been proposed to estimate 3D human pose from multi-view images, achieving satisfactory performance on public datasets collected under relatively simple conditions. However, there are limited approaches studying extracting 3D human skeletons from multimodal inputs, such as RGB and point cloud data. To address this gap, we introduce LiCamPose, a pipeline that integrates multi-view RGB and sparse point cloud information to estimate robust 3D human poses via single frame. We demonstrate the effectiveness of the volumetric architecture in combining these modalities. Furthermore, to circumvent the need for manually labeled 3D human pose annotations, we develop a synthetic dataset generator for pretraining and design an unsupervised domain adaptation strategy to train a 3D human pose estimator without manual annotations. To validate the generalization capability of our method, LiCamPose is evaluated on four datasets, including two public datasets, one synthetic dataset, and one challenging self-collected dataset named BasketBall, covering diverse scenarios. The results demonstrate that LiCamPose exhibits great generalization performance and significant application potential. The code, generator, and datasets will be made available upon acceptance of this paper.", "date": "2023-12-11", "categories": []}
{"id": "2605.05390", "title": "LAMP: Localization Aware Multi-camera People Tracking in Metric 3D World", "abstract": "Tracking 3D human motion from egocentric multi-camera headset is challenged by severe egomotion, partial visibility or occlusions and lack of training data. Existing methods designed for monocular video often require static or slowly-moving cameras and cannot efficiently leverage multi-view, calibrated and localized input. This makes them brittle and prone to fail on dynamic egocentric captures. We propose LAMP (Localization Aware Multi-camera People Tracking): a novel, simple framework to solve this via early disentanglement of observer and target motion. LAMP introduces a two-step process. First, we leverage the known device 6 DoF motion and calibration to convert detected 2D body keypoints from all cameras over a temporal window into a unified 3D world reference frame. Second, an end-to-end-trained spatio-temporal transformer fits 3D human motion directly to this 3D ray cloud. This \"lift-then-fit\" approach allows LAMP to learn and leverage a natural human motion prior in the world-space, as well as providing an elegant framework to flexibly incorporate information from multiple temporally asynchronous, partially observing and moving cameras. LAMP achieves state-of-the-art results on monocular benchmarks, while significantly outperforming baselines for our targeted egocentric setting.", "date": "2026-05-06", "categories": []}
{"id": "2604.26620", "title": "SnapPose3D: Diffusion-Based Single-Frame 2D-to-3D Lifting of Human Poses", "abstract": "Depth ambiguity and joint uncertainty are the two main obstacles in obtaining accurate human pose predictions by 2D-to-3D lifting methods proposed in the literature. In particular, these issues are caused by 2D joint locations that can be mapped to multiple 3D positions, inducing multiple possible final poses. Following these considerations, we propose leveraging diffusion-based models generation capability to predict multiple hypotheses and aggregate them in a final accurate pose. Therefore, we introduce SnapPose3D, a pose-lifting framework trained deterministically to denoise 3D poses conditioned on both visual context and 2D pose features. SnapPose3D adopts a probabilistic approach during inference, generating multiple hypotheses through random sampling from a unit Gaussian distribution. Unlike most previous methods that address pose ambiguity by processing temporal sequences, SnapPose3D uses single frames as input, avoiding tracking and limiting computational cost, data acquisition complexity, and the need for online, real-time applications. We extensively evaluate SnapPose3D on well-known benchmarks for the 3D human pose estimation task showing its ability to generate and aggregate accurate hypotheses that lead to state-of-the-art results.", "date": "2026-04-29", "categories": []}
{"id": "2604.22830", "title": "2D Pre-Training for 3D Pose Estimation", "abstract": "Pre-training is a general method that is used in a range of deep learning tasks. By first training a model on one task, and then further training on the downstream task used for final evaluation, the model is forced to learn a more general understanding of the input data. While pre-training has been applied to 3D Human Pose Estimation (HPE) previously, the scope of datasets used is typically very limited to some strong benchmarks, like Human3.6M. Therefore, in this project, we expand the scope of an existing 3D HPE scheme to be compatible with additional 2D and 3D HPE datasets, like Occlusion Person. We perform an extensive study on how aspects of 2D pre-training, such as model size, affect downstream performance, and to what extent pre-training can help the model generalize to different datasets. Experimental results show that 2D pre-training consistently outperforms training on 3D data alone, particularly in terms of computational efficiency. Finally, using MPII and Human3.6M, we are able to obtain an MPJPE score of under 64.5mm.", "date": "2026-04-19", "categories": []}
{"id": "2604.23651", "title": "Geometry-Conditioned Diffusion for Occlusion-Robust In-Bed Pose Estimation", "abstract": "Robust in-bed human pose estimation under blanket occlusion remains challenging due to the scarcity of reliable labeled training data for heavily covered poses. Existing approaches rely on multi-modal sensing or image-to-image translation frameworks that remain conditioned on visible source imagery, limiting scalability and pose diversity. In this work, we reformulate occlusion-aware augmentation as a geometry-conditioned generative modeling task. We conduct a systematic comparison of deterministic masking, unpaired translation, paired diffusion-based translation, and a proposed pose-conditioned Latent Diffusion Model (Pose-LDM). Unlike image-guided methods, Pose-LDM synthesizes blanket-covered images directly from skeletal keypoints, eliminating dependence on paired supervision and pixel-level source-image conditioning while enabling generation from arbitrary pose inputs. All augmentation strategies are evaluated through their impact on downstream pose estimation under a fixed backbone. Pose- LDM achieves the highest strict localization accuracy under severe occlusion while maintaining overall detection performance comparable to paired diffusion models, approaching the performance of fully supervised training. These results demonstrate that geometry-conditioned diffusion provides an effective and supervision-efficient pathway toward occlusion-robust inbed pose estimation without modifying the sensing pipeline. The code is available at: github.com/navidTerraNova/ GeoDiffPose.", "date": "2026-04-26", "categories": []}
{"id": "2604.24312", "title": "Unconstrained Multi-view Human Pose Estimation with Algebraic Priors", "abstract": "Recovering 3D human pose from multi-view imagery typically relies on precise camera calibration, which is often unavailable in real-world scenarios, thereby severely limiting the applicability of existing methods. To overcome this challenge, we propose an unconstrained framework that synergizes deep neural networks, algebraic priors, and temporal dynamics for uncalibrated multi-view human pose estimation. First, we introduce the Triangulation with Transformer Regressor (TTR), which reformulates classical triangulation into a data-driven token fusion process to bypass the dependency on explicit camera parameters. Second, to explicitly embed the inherent algebraic relations of the multi-view variety into the learning process, we propose the Gr\\\"{o}bner basis Corrector (GC). This pioneering loss formulation enforces constraints derived from the multi-view variety to ensure the neural predictions strictly adhere to the laws of projective geometry. Finally, we devise the Temporal Equivariant Rectifier (TER), which exploits the equivariance property of human motion to impose temporal coherence and structural consistency, effectively mitigating scale ambiguity in uncalibrated settings. Extensive evaluations on standard benchmarks demonstrate that our framework establishes a new state-of-the-art for uncalibrated multi-view human pose estimation. Notably, our approach significantly closes the performance gap between calibration-free methods and fully calibrated oracles.", "date": "2026-04-27", "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": "2604.17688", "title": "Dual-stream Spatio-Temporal GCN-Transformer Network for 3D Human Pose Estimation", "abstract": "3D human pose estimation is a classic and important research direction in the field of computer vision. In recent years, Transformer-based methods have made significant progress in lifting 2D to 3D human pose estimation. However, these methods primarily focus on modeling global temporal and spatial relationships, neglecting local skeletal relationships and the information interaction between different channels. Therefore, we have proposed a novel method,the Dual-stream Spatio-temporal GCN-Transformer Network (MixTGFormer). This method models the spatial and temporal relationships of human skeletons simultaneously through two parallel channels, achieving effective fusion of global and local features. The core of MixTGFormer is composed of stacked Mixformers. Specifically, the Mixformer includes the Mixformer Block and the Squeeze-and-Excitation Layer ( SE Layer). It first extracts and fuses various information of human skeletons through two parallel Mixformer Blocks with different modes. Then, it further supplements the fused information through the SE Layer. The Mixformer Block integrates Graph Convolutional Networks (GCN) into the Transformer, enhancing both local and global information utilization. Additionally, we further implement its temporal and spatial forms to extract both spatial and temporal relationships. We extensively evaluated our model on two benchmark datasets (Human3.6M and MPI-INF-3DHP). The experimental results showed that, compared to other methods, our MixTGFormer achieved state-of-the-art results, with P1 errors of 37.6mm and 15.7mm on these datasets, respectively.", "date": "2026-04-20", "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": "2604.13581", "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": "2604.09715", "title": "MuPPet: Multi-person 2D-to-3D Pose Lifting", "abstract": "Multi-person social interactions are inherently built on coherence and relationships among all individuals within the group, making multi-person localization and body pose estimation essential to understanding these social dynamics. One promising approach is 2D-to-3D pose lifting which provides a 3D human pose consisting of rich spatial details by building on the significant advances in 2D pose estimation. However, the existing 2D-to-3D pose lifting methods often neglect inter-person relationships or cannot handle varying group sizes, limiting their effectiveness in multi-person settings. We propose MuPPet, a novel multi-person 2D-to-3D pose lifting framework that explicitly models inter-person correlations. To leverage these inter-person dependencies, our approach introduces Person Encoding to structure individual representations, Permutation Augmentation to enhance training diversity, and Dynamic Multi-Person Attention to adaptively model correlations between individuals. Extensive experiments on group interaction datasets demonstrate MuPPet significantly outperforms state-of-the-art single- and multi-person 2D-to-3D pose lifting methods, and improves robustness in occlusion scenarios. Our findings highlight the importance of modeling inter-person correlations, paving the way for accurate and socially-aware 3D pose estimation. Our code is available at: https://github.com/Thomas-Markhorst/MuPPet", "date": "2026-04-08", "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": "2604.08921", "title": "TAIHRI: Task-Aware 3D Human Keypoints Localization for Close-Range Human-Robot Interaction", "abstract": "Accurate 3D human keypoints localization is a critical technology enabling robots to achieve natural and safe physical interaction with users. Conventional 3D human keypoints estimation methods primarily focus on the whole-body reconstruction quality relative to the root joint. However, in practical human-robot interaction (HRI) scenarios, robots are more concerned with the precise metric-scale spatial localization of task-relevant body parts under the egocentric camera 3D coordinate. We propose TAIHRI, the first Vision-Language Model (VLM) tailored for close-range HRI perception, capable of understanding users' motion commands and directing the robot's attention to the most task-relevant keypoints. By quantizing 3D keypoints into a finite interaction space, TAIHRI precisely localize the 3D spatial coordinates of critical body parts by 2D keypoint reasoning via next token prediction, and seamlessly adapt to downstream tasks such as natural language control or global space human mesh recovery. Experiments on egocentric interaction benchmarks demonstrate that TAIHRI achieves superior estimation accuracy for task-critical body parts. We believe TAIHRI opens new research avenues in the field of embodied human-robot interaction. Code is available at: https://github.com/Tencent/TAIHRI.", "date": "2026-04-10", "categories": []}
{"id": "2604.08543", "title": "E-3DPSM: A State Machine for Event-Based Egocentric 3D Human Pose Estimation", "abstract": "Event cameras offer multiple advantages in monocular egocentric 3D human pose estimation from head-mounted devices, such as millisecond temporal resolution, high dynamic range, and negligible motion blur. Existing methods effectively leverage these properties, but suffer from low 3D estimation accuracy, insufficient in many applications (e.g., immersive VR/AR). This is due to the design not being fully tailored towards event streams (e.g., their asynchronous and continuous nature), leading to high sensitivity to self-occlusions and temporal jitter in the estimates. This paper rethinks the setting and introduces E-3DPSM, an event-driven continuous pose state machine for event-based egocentric 3D human pose estimation. E-3DPSM aligns continuous human motion with fine-grained event dynamics; it evolves latent states and predicts continuous changes in 3D joint positions associated with observed events, which are fused with direct 3D human pose predictions, leading to stable and drift-free final 3D pose reconstructions. E-3DPSM runs in real-time at 80 Hz on a single workstation and sets a new state of the art in experiments on two benchmarks, improving accuracy by up to 19% (MPJPE) and temporal stability by up to 2.7x. See our project page for the source code and trained models.", "date": "2026-04-09", "categories": []}
{"id": "2508.07112", "title": "AugLift: Depth-Aware Input Reparameterization Improves Domain Generalization in 2D-to-3D Pose Lifting", "abstract": "Lifting-based 3D human pose estimation infers 3D joints from 2D keypoints but generalizes poorly because $(x,y)$ coordinates alone are an ill-posed, sparse representation that discards geometric information modern foundation models can recover. We propose \\emph{AugLift}, which changes the representation format of lifting from 2D coordinates to a 6D geometric descriptor via two modules: (1) an \\emph{Uncertainty-Aware Depth Descriptor} (UADD) -- a compact tuple $(c, d, d_{\\min}, d_{\\max})$ extracted from a confidence-scaled neighborhood of an off-the-shelf monocular depth map -- and (2) a scale normalization component that handles train/test distance shifts. AugLift requires no new sensors, no new data collection, and no architectural changes beyond widening the input layer; because it operates at the representation level, it is composable with any lifting architecture or domain generalization technique.\n  In the detection setting, AugLift reduces cross-dataset MPJPE by $10.1$% on average across four datasets and four lifting architectures while improving in-distribution accuracy by $4.0$%; post-hoc analysis shows gains concentrate on novel poses and occluded joints. In the ground-truth 2D setting, combining AugLift with PoseAug's differentiable domain generalization achieves state-of-the-art cross-dataset performance ($62.4$\\,mm on 3DHP, $92.6$\\,mm on 3DPW; $14.5$% and $22.2$% over PoseAug), demonstrating that foundation-model depth provides genuine geometric signal complementary to explicit 3D augmentation. Code will be made publicly available.", "date": "2025-08-09", "categories": []}
{"id": "2501.09411", "title": "Towards Robust and Realistic Human Pose Estimation via WiFi Signals", "abstract": "Robust WiFi-based human pose estimation (HPE) is a challenging task that bridges discrete and subtle WiFi signals to human skeletons. We revisit this problem and reveal two critical yet overlooked issues: 1) cross-domain gap, i.e., due to significant discrepancies in pose distributions between source and target domains; and 2) structural fidelity gap, i.e., predicted skeletal poses manifest distorted topology, usually with misplaced joints and disproportionate bone lengths. This paper fills these gaps by reformulating the task into a novel two-phase framework dubbed DT-Pose: Domain-consistent representation learning and Topology-constrained Pose decoding. Concretely, we first propose a temporal consistency contrastive learning strategy with uniformity regularization, integrated into a self-supervised masked pretraining paradigm. This design facilitates robust learning of domain-consistent and motion-discriminative WiFi representations while mitigating potential mode collapse caused by signal sparsity. Beyond this, we introduce an effective hybrid decoding architecture that incorporates explicit skeletal topology constraints. By compensating for the inherent absence of spatial priors in WiFi semantic vectors, the decoder enables structured modeling of both adjacent and overarching joint relationships, producing more realistic pose predictions. Extensive experiments conducted on various benchmark datasets highlight the superior performance of our method in tackling these fundamental challenges in 2D/3D WiFi-based HPE tasks. The associated code is released at https://github.com/cseeyangchen/DT-Pose.", "date": "2025-01-16", "categories": []}
{"id": "2603.14507", "title": "Expanding mmWave Datasets for Human Pose Estimation with Unlabeled Data and LiDAR Datasets", "abstract": "Current millimeter-wave (mmWave) datasets for human pose estimation (HPE) are scarce and lack diversity in both point cloud (PC) attributes and human poses, hindering the generalization ability of their trained models. On the other hand, unlabeled mmWave HPE data and diverse LiDAR HPE datasets are readily available. We propose EMDUL, a novel approach to expand the volume and diversity of an existing mmWave dataset using unlabeled mmWave data and LiDAR datasets. EMDUL consists of two independent modules, namely a pseudo-label estimator to annotate unlabeled mmWave data, and a closed-form converter that translates an annotated LiDAR PC to its mmWave counterpart. Expanding the original dataset with both LiDAR-converted and pseudo-labeled mmWave PCs significantly boosts the performance and generalization ability of all the examined HPE models, reducing 15.1% and 18.9% error for in-domain and out-of-domain settings, respectively. Code is available at https://github.com/Shimmer93/EMDUL.", "date": "2026-03-15", "categories": []}
{"id": "2604.03299", "title": "MoViD: View-Invariant 3D Human Pose Estimation via Motion-View Disentanglement", "abstract": "3D human pose estimation is a key enabling technology for applications such as healthcare monitoring, human-robot collaboration, and immersive gaming, but real-world deployment remains challenged by viewpoint variations. Existing methods struggle to generalize to unseen camera viewpoints, require large amounts of training data, and suffer from high inference latency. We propose MoViD, a viewpoint-invariant 3D human pose estimation framework that disentangles viewpoint information from motion features. The key idea is to extract viewpoint information from intermediate pose features and leverage it to enhance both the robustness and efficiency of pose estimation. MoViD introduces a view estimator that models key joint relationships to predict viewpoint information, and an orthogonal projection module to disentangle motion and view features, further enhanced through physics-grounded contrastive alignment across views. For real-time edge deployment, MoViD employs a frame-by-frame inference pipeline with a view-aware strategy that adaptively activates flip refinement based on the estimated viewpoint. Evaluations on nine public datasets and newly collected multiview UAV and gait analysis datasets show that MoViD reduces pose estimation error by over 24.2\\% compared to state-of-the-art methods, maintains robust performance under severe occlusions with 60\\% less training data, and achieves real-time inference at 15 FPS on NVIDIA edge devices.", "date": "2026-03-29", "categories": []}
{"id": "2604.03652", "title": "Motion-Adaptive Multi-Scale Temporal Modelling with Skeleton-Constrained Spatial Graphs for Efficient 3D Human Pose Estimation", "abstract": "Accurate 3D human pose estimation from monocular videos requires effective modelling of complex spatial and temporal dependencies. However, existing methods often face challenges in efficiency and adaptability when modelling spatial and temporal dependencies, particularly under dense attention or fixed modelling schemes. In this work, we propose MASC-Pose, a Motion-Adaptive multi-scale temporal modelling framework with Skeleton-Constrained spatial graphs for efficient 3D human pose estimation. Specifically, it introduces an Adaptive Multi-scale Temporal Modelling (AMTM) module to adaptively capture heterogeneous motion dynamics at different temporal scales, together with a Skeleton-constrained Adaptive GCN (SAGCN) for joint-specific spatial interaction modelling. By jointly enabling adaptive temporal reasoning and efficient spatial aggregation, our method achieves strong accuracy with high computational efficiency. Extensive experiments on Human3.6M and MPI-INF-3DHP datasets demonstrate the effectiveness of our approach.", "date": "2026-04-04", "categories": []}
{"id": "2604.01322", "title": "Human Pose Estimation in Trampoline Gymnastics: Improving Performance Using a New Synthetic Dataset", "abstract": "Trampoline gymnastics involves extreme human poses and uncommon viewpoints, on which state-of-the art pose estimation models tend to under-perform. We demonstrate that this problem can be addressed by fine-tuning a pose estimation model on a dataset of synthetic trampoline poses (STP). STP is generated from motion capture recordings of trampoline routines. We develop a pipeline to fit noisy motion capture data to a parametric human model, then generate multiview realistic images. We use this data to fine-tune a ViTPose model, and test it on real multi-view trampoline images. The resulting model exhibits accuracy improvements in 2D which translates to improved 3D triangulation. In 2D, we obtain state-of-the-art results on such challenging data, bridging the performance gap between common and extreme poses. In 3D, we reduce the MPJPE by 12.5 mm with our best model, which represents an improvement of 19.6% compared to the pretrained ViTPose model.", "date": "2026-04-01", "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": "2603.28763", "title": "PoseDreamer: Scalable and Photorealistic Human Data Generation Pipeline with Diffusion Models", "abstract": "Acquiring labeled datasets for 3D human mesh estimation is challenging due to depth ambiguities and the inherent difficulty of annotating 3D geometry from monocular images. Existing datasets are either real, with manually annotated 3D geometry and limited scale, or synthetic, rendered from 3D engines that provide precise labels but suffer from limited photorealism, low diversity, and high production costs. In this work, we explore a third path: generated data. We introduce PoseDreamer, a novel pipeline that leverages diffusion models to generate large-scale synthetic datasets with 3D mesh annotations. Our approach combines controllable image generation with Direct Preference Optimization for control alignment, curriculum-based hard sample mining, and multi-stage quality filtering. Together, these components naturally maintain correspondence between 3D labels and generated images, while prioritizing challenging samples to maximize dataset utility. Using PoseDreamer, we generate more than 500,000 high-quality synthetic samples, achieving a 76% improvement in image-quality metrics compared to rendering-based datasets. Models trained on PoseDreamer achieve performance comparable to or superior to those trained on real-world and traditional synthetic datasets. In addition, combining PoseDreamer with synthetic datasets results in better performance than combining real-world and synthetic datasets, demonstrating the complementary nature of our dataset. We will release the full dataset and generation code.", "date": "2026-03-30", "categories": []}
{"id": "2603.26447", "title": "Meta-Learned Adaptive Optimization for Robust Human Mesh Recovery with Uncertainty-Aware Parameter Updates", "abstract": "Human mesh recovery from single images remains challenging due to inherent depth ambiguity and limited generalization across domains. While recent methods combine regression and optimization approaches, they struggle with poor initialization for test-time refinement and inefficient parameter updates during optimization. We propose a novel meta-learning framework that trains models to produce optimization-friendly initializations while incorporating uncertainty-aware adaptive updates during test-time refinement. Our approach introduces three key innovations: (1) a meta-learning strategy that simulates test-time optimization during training to learn better parameter initializations, (2) a selective parameter caching mechanism that identifies and freezes converged joints to reduce computational overhead, and (3) distribution-based adaptive updates that sample parameter changes from learned distributions, enabling robust exploration while quantifying uncertainty. Additionally, we employ stochastic approximation techniques to handle intractable gradients in complex loss landscapes. Extensive experiments on standard benchmarks demonstrate that our method achieves state-of-the-art performance, reducing MPJPE by 10.3 on 3DPW and 8.0 on Human3.6M compared to strong baselines. Our approach shows superior domain adaptation capabilities with minimal performance degradation across different environmental conditions, while providing meaningful uncertainty estimates that correlate with actual prediction errors. Combining meta-learning and adaptive optimization enables accurate mesh recovery and robust generalization to challenging scenarios.", "date": "2026-03-27", "categories": []}
{"id": "2603.25175", "title": "AG-EgoPose: Leveraging Action-Guided Motion and Kinematic Joint Encoding for Egocentric 3D Pose Estimation", "abstract": "Egocentric 3D human pose estimation remains challenging due to severe perspective distortion, limited body visibility, and complex camera motion inherent in first-person viewpoints. Existing methods typically rely on single-frame analysis or limited temporal fusion, which fails to effectively leverage the rich motion context available in egocentric videos. We introduce AG-EgoPose, a novel dual-stream framework that integrates short- and long-range motion context with fine-grained spatial cues for robust pose estimation from fisheye camera input. Our framework features two parallel streams: A spatial stream uses a weight-sharing ResNet-18 encoder-decoder to generate 2D joint heatmaps and corresponding joint-specific spatial feature tokens. Simultaneously, a temporal stream uses a ResNet-50 backbone to extract visual features, which are then processed by an action recognition backbone to capture the motion dynamics. These complementary representations are fused and refined in a transformer decoder with learnable joint tokens, which allows for the joint-level integration of spatial and temporal evidence while maintaining anatomical constraints. Experiments on real-world datasets demonstrate that AG-EgoPose achieves state-of-the-art performance in both quantitative and qualitative metrics. Code is available at: https://github.com/Mushfiq5647/AG-EgoPose.", "date": "2026-03-26", "categories": []}
{"id": "2603.23370", "title": "Object Pose Transformer: Unifying Unseen Object Pose Estimation", "abstract": "Learning model-free object pose estimation for unseen instances remains a fundamental challenge in 3D vision. Existing methods typically fall into two disjoint paradigms: category-level approaches predict absolute poses in a canonical space but rely on predefined taxonomies, while relative pose methods estimate cross-view transformations but cannot recover single-view absolute pose. In this work, we propose Object Pose Transformer (\\ours{}), a unified feed-forward framework that bridges these paradigms through task factorization within a single model. \\ours{} jointly predicts depth, point maps, camera parameters, and normalized object coordinates (NOCS) from RGB inputs, enabling both category-level absolute SA(3) pose and unseen-object relative SE(3) pose. Our approach leverages contrastive object-centric latent embeddings for canonicalization without requiring semantic labels at inference time, and uses point maps as a camera-space representation to enable multi-view relative geometric reasoning. Through cross-frame feature interaction and shared object embeddings, our model leverages relative geometric consistency across views to improve absolute pose estimation, reducing ambiguity in single-view predictions. Furthermore, \\ours{} is camera-agnostic, learning camera intrinsics on-the-fly and supporting optional depth input for metric-scale recovery, while remaining fully functional in RGB-only settings. Extensive experiments on diverse benchmarks (NOCS, HouseCat6D, Omni6DPose, Toyota-Light) demonstrate state-of-the-art performance in both absolute and relative pose estimation tasks within a single unified architecture.", "date": "2026-03-24", "categories": []}
{"id": "2603.09418", "title": "CIGPose: Causal Intervention Graph Neural Network for Whole-Body Pose Estimation", "abstract": "State-of-the-art whole-body pose estimators often lack robustness, producing anatomically implausible predictions in challenging scenes. We posit this failure stems from spurious correlations learned from visual context, a problem we formalize using a Structural Causal Model (SCM). The SCM identifies visual context as a confounder that creates a non-causal backdoor path, corrupting the model's reasoning. We introduce the Causal Intervention Graph Pose (CIGPose) framework to address this by approximating the true causal effect between visual evidence and pose. The core of CIGPose is a novel Causal Intervention Module: it first identifies confounded keypoint representations via predictive uncertainty and then replaces them with learned, context-invariant canonical embeddings. These deconfounded embeddings are processed by a hierarchical graph neural network that reasons over the human skeleton at both local and global semantic levels to enforce anatomical plausibility. Extensive experiments show CIGPose achieves a new state-of-the-art on COCO-WholeBody. Notably, our CIGPose-x model achieves 67.0\\% AP, surpassing prior methods that rely on extra training data. With the additional UBody dataset, CIGPose-x is further boosted to 67.5\\% AP, demonstrating superior robustness and data efficiency. The codes and models are publicly available at https://github.com/53mins/CIGPose.", "date": "2026-03-10", "categories": []}
{"id": "2603.20323", "title": "NCSTR: Node-Centric Decoupled Spatio-Temporal Reasoning for Video-based Human Pose Estimation", "abstract": "Video-based human pose estimation remains challenged by motion blur, occlusion, and complex spatiotemporal dynamics. Existing methods often rely on heatmaps or implicit spatio-temporal feature aggregation, which limits joint topology expressiveness and weakens cross-frame consistency. To address these problems, we propose a novel node-centric framework that explicitly integrates visual, temporal, and structural reasoning for accurate pose estimation. First, we design a visuo-temporal velocity-based joint embedding that fuses sub-pixel joint cues and inter-frame motion to build appearance- and motion-aware representations. Then, we introduce an attention-driven pose-query encoder, which applies attention over joint-wise heatmaps and frame-wise features to map the joint representations into a pose-aware node space, generating image-conditioned joint-aware node embeddings. Building upon these node embeddings, we propose a dual-branch decoupled spatio-temporal attention graph that models temporal propagation and spatial constraint reasoning in specialized local and global branches. Finally, a node-space expert fusion module is proposed to adaptively fuse the complementary outputs from both branches, integrating local and global cues for final joint predictions. Extensive experiments on three widely used video pose benchmarks demonstrate that our method outperforms state-of-the-art methods. The results highlight the value of explicit node-centric reasoning, offering a new perspective for advancing video-based human pose estimation.", "date": "2026-03-20", "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": "2603.16343", "title": "Learning Human-Object Interaction for 3D Human Pose Estimation from LiDAR Point Clouds", "abstract": "Understanding humans from LiDAR point clouds is one of the most critical tasks in autonomous driving due to its close relationships with pedestrian safety, yet it remains challenging in the presence of diverse human-object interactions and cluttered backgrounds. Nevertheless, existing methods largely overlook the potential of leveraging human-object interactions to build robust 3D human pose estimation frameworks. There are two major challenges that motivate the incorporation of human-object interaction. First, human-object interactions introduce spatial ambiguity between human and object points, which often leads to erroneous 3D human keypoint predictions in interaction regions. Second, there exists severe class imbalance in the number of points between interacting and non-interacting body parts, with the interaction-frequent regions such as hand and foot being sparsely observed in LiDAR data. To address these challenges, we propose a Human-Object Interaction Learning (HOIL) framework for robust 3D human pose estimation from LiDAR point clouds. To mitigate the spatial ambiguity issue, we present human-object interaction-aware contrastive learning (HOICL) that effectively enhances feature discrimination between human and object points, particularly in interaction regions. To alleviate the class imbalance issue, we introduce contact-aware part-guided pooling (CPPool) that adaptively reallocates representational capacity by compressing overrepresented points while preserving informative points from interacting body parts. In addition, we present an optional contact-based temporal refinement that refines erroneous per-frame keypoint estimates using contact cues over time. As a result, our HOIL effectively leverages human-object interaction to resolve spatial ambiguity and class imbalance in interaction regions. Codes will be released.", "date": "2026-03-17", "categories": []}
{"id": "2501.05264", "title": "Towards Balanced Multi-Modal Learning in 3D Human Pose Estimation", "abstract": "3D human pose estimation (3D HPE) has emerged as a prominent research topic, particularly in the realm of RGB-based methods. However, the use of RGB images is often limited by issues such as occlusion and privacy constraints. Consequently, multi-modal sensing, which leverages non-intrusive sensors, is gaining increasing attention. Nevertheless, multi-modal 3D HPE still faces challenges, including modality imbalance. In this work, we introduce a novel balanced multi-modal learning method for 3D HPE, which harnesses the power of RGB, LiDAR, mmWave, and WiFi. Specifically, we propose a Shapley value-based contribution algorithm to assess the contribution of each modality and detect modality imbalance. To address this imbalance, we design a modality learning regulation strategy that decelerates the learning process during the early stages of training. We conduct extensive experiments on the widely adopted multi-modal dataset, MM-Fi, demonstrating the superiority of our approach in enhancing 3D pose estimation under complex conditions. Our source code is available at https://github.com/MICLAB-BUPT/AWC.", "date": "2025-01-09", "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": "2603.14726", "title": "Enhancing Hands in 3D Whole-Body Pose Estimation with Conditional Hands Modulator", "abstract": "Accurately recovering hand poses within the body context remains a major challenge in 3D whole-body pose estimation. This difficulty arises from a fundamental supervision gap: whole-body pose estimators are trained on full-body datasets with limited hand diversity, while hand-only estimators, trained on hand-centric datasets, excel at detailed finger articulation but lack global body awareness. To address this, we propose Hand4Whole++, a modular framework that leverages the strengths of both pre-trained whole-body and hand pose estimators. We introduce CHAM (Conditional Hands Modulator), a lightweight module that modulates the whole-body feature stream using hand-specific features extracted from a pre-trained hand pose estimator. This modulation enables the whole-body model to predict wrist orientations that are both accurate and coherent with the upper-body kinematic structure, without retraining the full-body model. In parallel, we directly incorporate finger articulations and hand shapes predicted by the hand pose estimator, aligning them to the full-body mesh via differentiable rigid alignment. This design allows Hand4Whole++ to combine globally consistent body reasoning with fine-grained hand detail. Extensive experiments demonstrate that Hand4Whole++ substantially improves hand accuracy and enhances overall full-body pose quality.", "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": "2503.03222", "title": "Mocap-2-to-3: Multi-view Lifting for Monocular Motion Recovery with 2D Pretraining", "abstract": "Human motion recovery for real-world interaction demands both precise action details and metric-scale trajectories. Recovering absolute human pose from monocular input presents a viable solution, but faces two main challenges: (1) models' reliance on 3D training data from constrained environments limits their out-of-distribution generalization; and (2) the inherent difficulty of estimating metric-scale poses from monocular observations. This paper introduces Mocap-2-to-3, a novel framework that differs from prior HMR methods by recovering absolute poses from monocular input and leveraging abundant 2D data to enhance 3D motion recovery. To effectively utilize the action priors and diversity in large-scale 2D datasets, we reformulate 3D motion as a multi-view synthesis process and divide the training into two stages: a single-view diffusion model is first pre-trained on extensive 2D data, followed by multi-view fine-tuning on 3D data, thus achieving a combination of strong priors and geometric constraints. Furthermore, to recover absolute poses, we introduce a novel human motion representation that decouples the learning of local pose and global movements, while encoding ground geometric priors to accelerate convergence, thereby yielding more precise positioning in the physical world. Experiments on in-the-wild benchmarks show that our method outperforms state-of-the-art approaches in both camera-space motion realism and world-grounded human positioning, while exhibiting strong generalization capability.", "date": "2025-03-05", "categories": []}
{"id": "2509.24980", "title": "SDPose: Exploiting Diffusion Priors for Out-of-Domain and Robust Pose Estimation", "abstract": "Pre-trained diffusion models provide rich latent features across U-Net levels and are emerging as powerful vision backbones. While prior works such as Marigold and Lotus repurpose diffusion priors for dense geometric perception tasks such as depth and surface normal estimation, their potential for cross-domain human pose estimation remains largely unexplored. Through a systematic analysis of latent features from different upsampling levels of the Stable Diffusion U-Net, we identify the levels that deliver the strongest robustness and cross-domain generalization for pose estimation. Building on these findings, we propose \\textbf{SDPose}, which (i) extracts U-Net features from the selected upsampling blocks, (ii) fuses them with a lightweight feature aggregation module to form a robust representation, and (iii) jointly optimizes keypoint heatmap supervision with an auxiliary latent reconstruction loss to regularize training and preserve the pre-trained generative prior. To evaluate cross-domain generalization and robustness, we construct COCO-OOD, a COCO-based benchmark with four subsets: three style-transferred splits to assess domain shift, and one corruption split (noise, weather, digital artifacts, and blur) to test robustness. With a shorter fine-tuning schedule, SDPose achieves performance comparable to Sapiens on COCO, surpasses Sapiens-1B on COCO-WholeBody, and establishes new state-of-the-art results on HumanArt and COCO-OOD.", "date": "2025-09-29", "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": "2512.04862", "title": "Contact-Aware Refinement of Human Pose Pseudo-Ground Truth via Bioimpedance Sensing", "abstract": "Capturing accurate 3D human pose in the wild would provide valuable data for training pose estimation and motion generation methods. While video-based estimation approaches have become increasingly accurate, they often fail in common scenarios involving self-contact, such as a hand touching the face. In contrast, wearable bioimpedance sensing can cheaply and unobtrusively measure ground-truth skin-to-skin contact. Consequently, we propose a novel framework that combines visual pose estimators with bioimpedance sensing to capture the 3D pose of people by taking self-contact into account. Our method, BioTUCH, initializes the pose using an off-the-shelf estimator and introduces contact-aware pose optimization during measured self-contact: reprojection error and deviations from the input estimate are minimized while enforcing vertex proximity constraints. We validate our approach using a new dataset of synchronized RGB video, bioimpedance measurements, and 3D motion capture. Testing with three input pose estimators, we demonstrate an average of 11.7% improvement in reconstruction accuracy. We also present a miniature wearable bioimpedance sensor that enables efficient large-scale collection of contact-aware training data for improving pose estimation and generation using BioTUCH. Code and data are available at biotuch.is.tue.mpg.de", "date": "2025-12-04", "categories": []}
{"id": "2602.20792", "title": "SIMSPINE: A Biomechanics-Aware Simulation Framework for 3D Spine Motion Annotation and Benchmarking", "abstract": "Modeling spinal motion is fundamental to understanding human biomechanics, yet remains underexplored in computer vision due to the spine's complex multi-joint kinematics and the lack of large-scale 3D annotations. We present a biomechanics-aware keypoint simulation framework that augments existing human pose datasets with anatomically consistent 3D spinal keypoints derived from musculoskeletal modeling. Using this framework, we create the first open dataset, named SIMSPINE, which provides sparse vertebra-level 3D spinal annotations for natural full-body motions in indoor multi-camera capture without external restraints. With 2.14 million frames, this enables data-driven learning of vertebral kinematics from subtle posture variations and bridges the gap between musculoskeletal simulation and computer vision. In addition, we release pretrained baselines covering fine-tuned 2D detectors, monocular 3D pose lifting models, and multi-view reconstruction pipelines, establishing a unified benchmark for biomechanically valid spine motion estimation. Specifically, our 2D spine baselines improve the state-of-the-art from 0.63 to 0.80 AUC in controlled environments, and from 0.91 to 0.93 AP for in-the-wild spine tracking. Together, the simulation framework and SIMSPINE dataset advance research in vision-based biomechanics, motion analysis, and digital human modeling by enabling reproducible, anatomically grounded 3D spine estimation under natural conditions.", "date": "2026-02-24", "categories": []}
{"id": "2306.03538", "title": "SDR-GAIN: A High Real-Time Occluded Pedestrian Pose Completion Method for Autonomous Driving", "abstract": "With the advancement of vision-based autonomous driving technology, pedestrian detection have become an important component for improving traffic safety and driving system robustness. Nevertheless, in complex traffic scenarios, conventional pose estimation approaches frequently fail to accurately reconstruct occluded keypoints, primarily due to obstructions caused by vehicles, vegetation, or architectural elements. To address this issue, we propose a novel real-time occluded pedestrian pose completion framework termed Separation and Dimensionality Reduction-based Generative Adversarial Imputation Nets (SDR-GAIN). Unlike previous approaches that train visual models to distinguish occlusion patterns, SDR-GAIN aims to learn human pose directly from the numerical distribution of keypoint coordinates and interpolate missing positions. It employs a self-supervised adversarial learning paradigm to train lightweight generators with residual structures for the imputation of missing pose keypoints. Additionally, it integrates multiple pose standardization techniques to alleviate the difficulty of the learning process. Experiments conducted on the COCO and JAAD datasets demonstrate that SDR-GAIN surpasses conventional machine learning and Transformer-based missing data interpolation algorithms in accurately recovering occluded pedestrian keypoints, while simultaneously achieving microsecond-level real-time inference.", "date": "2023-06-06", "categories": []}
{"id": "2603.09681", "title": "Improving 3D Foot Motion Reconstruction in Markerless Monocular Human Motion Capture", "abstract": "State-of-the-art methods can recover accurate overall 3D human body motion from in-the-wild videos. However, they often fail to capture fine-grained articulations, especially in the feet, which are critical for applications such as gait analysis and animation. This limitation results from training datasets with inaccurate foot annotations and limited foot motion diversity. We address this gap with FootMR, a Foot Motion Refinement method that refines foot motion estimated by an existing human recovery model through lifting 2D foot keypoint sequences to 3D. By avoiding direct image input, FootMR circumvents inaccurate image-3D annotation pairs and can instead leverage large-scale motion capture data. To resolve ambiguities of 2D-to-3D lifting, FootMR incorporates knee and foot motion as context and predicts only residual foot motion. Generalization to extreme foot poses is further improved by representing joints in global rather than parent-relative rotations and applying extensive data augmentation. To support evaluation of foot motion reconstruction, we introduce MOOF, a 2D dataset of complex foot movements. Experiments on MOOF, MOYO, and RICH show that FootMR outperforms state-of-the-art methods, reducing ankle joint angle error on MOYO by up to 30% over the best video-based approach.", "date": "2026-03-10", "categories": []}
{"id": "2503.17788", "title": "From 2D Alignment to 3D Plausibility: Unifying Heterogeneous 2D Priors and Penetration-Free Diffusion for Occlusion-Robust Two-Hand Reconstruction", "abstract": "Two-hand reconstruction from monocular images is hampered by complex poses and severe occlusions, which often cause interaction misalignment and two-hand penetration. We address this by decoupling the problem into 2D structural alignment and 3D spatial interaction alignment, each handled by a tailored component. For 2D alignment, we pioneer the attempt to unify heterogeneous structural priors (keypoints, segmentation, and depth) from vision foundation models as complementary structured guidance for two-hand recovery. Instead of extracting priors prediction as explicit inputs, we propose a fusion-alignment encoder that absorbs their structural knowledge implicitly, achieving foundation-level guidance without foundation-level cost. For 3D spatial alignment, we propose a two-hand penetration-free diffusion model that learns a generative mapping from interpenetrated poses to realistic, collision-free configurations. Guided by collision gradients during denoising, the model converges toward the manifold of valid two-hand interactions, preserving geometric and kinematic coherence. This generative formulation approach enables physically credible reconstructions even under occlusion or ambiguous visual input. Extensive experiments on InterHand2.6M and HIC show state-of-the-art or leading performance in interaction alignment and penetration suppression. Project: https://gaogehan.github.io/A2P/", "date": "2025-03-22", "categories": []}
{"id": "2508.21363", "title": "Efficient Diffusion-Based 3D Human Pose Estimation with Hierarchical Temporal Pruning", "abstract": "Diffusion models have demonstrated strong capabilities in generating high-fidelity 3D human poses, yet their iterative nature and multi-hypothesis requirements incur substantial computational cost. In this paper, we propose an Efficient Diffusion-Based 3D Human Pose Estimation framework with a Hierarchical Temporal Pruning (HTP) strategy, which dynamically prunes redundant pose tokens across both frame and semantic levels while preserving critical motion dynamics. HTP operates in a staged, top-down manner: (1) Temporal Correlation-Enhanced Pruning (TCEP) identifies essential frames by analyzing inter-frame motion correlations through adaptive temporal graph construction; (2) Sparse-Focused Temporal MHSA (SFT MHSA) leverages the resulting frame-level sparsity to reduce attention computation, focusing on motion-relevant tokens; and (3) Mask-Guided Pose Token Pruner (MGPTP) performs fine-grained semantic pruning via clustering, retaining only the most informative pose tokens. Experiments on Human3.6M and MPI-INF-3DHP show that HTP reduces training MACs by 38.5\\%, inference MACs by 56.8\\%, and improves inference speed by an average of 81.1\\% compared to prior diffusion-based methods, while achieving state-of-the-art performance.", "date": "2025-08-29", "categories": []}
{"id": "2603.08551", "title": "mmGAT: Pose Estimation by Graph Attention with Mutual Features from mmWave Radar Point Cloud", "abstract": "Pose estimation and human action recognition (HAR) are pivotal technologies spanning various domains. While the image-based pose estimation and HAR are widely admired for their superior performance, they lack in privacy protection and suboptimal performance in low-light and dark environments. This paper exploits the capabilities of millimeter-wave (mmWave) radar technology for human pose estimation by processing radar data with Graph Neural Network (GNN) architecture, coupled with the attention mechanism. Our goal is to capture the finer details of the radar point cloud to improve the pose estimation performance. To this end, we present a unique feature extraction technique that exploits the full potential of the GNN processing method for pose estimation. Our model mmGAT demonstrates remarkable performance on two publicly available benchmark mmWave datasets and establishes new state of the art results in most scenarios in terms of human pose estimation. Our approach achieves a noteworthy reduction of pose estimation mean per joint position error (MPJPE) by 35.6% and PA-MPJPE by 14.1% from the current state of the art benchmark within this domain.", "date": "2026-03-09", "categories": []}
{"id": "2512.06306", "title": "Exploiting Spatiotemporal Properties for Efficient Event-Driven Human Pose Estimation", "abstract": "Human pose estimation focuses on predicting body keypoints to analyze human motion. Currently, most pose estimation tasks rely on conventional RGB cameras. In contrast, event cameras provide high temporal resolution and low latency, enabling robust estimation under challenging conditions and opening up new possibilities for pose estimation. However, most existing methods convert event streams into dense event frames, which adds extra computation and sacrifices the high temporal resolution of the event signal. In this work, we aim to exploit the spatiotemporal properties of event streams based on point cloud-based framework, designed to enhance human pose estimation performance while maintaining computational efficiency. We design Event Temporal Slicing Convolution module to capture short-term dependencies across event slices, and combine it with Event Slice Sequencing module for structured temporal modeling. We further propose an edge-enhanced point cloud-based event representation to enhance spatial edge information under sparse event conditions to further improve performance. Experiments on the DHP19 dataset show our proposed method consistently improves performance across three representative point cloud backbones: PointNet, DGCNN, and Point Transformer, with an average MPJPE reduction of 4%.", "date": "2025-12-06", "categories": []}
{"id": "2503.21692", "title": "RapidPoseTriangulation: Multi-view Multi-person Whole-body Human Pose Triangulation in a Millisecond", "abstract": "The integration of multi-view imaging and pose estimation represents a significant advance in computer vision applications, offering new possibilities for understanding human movement and interactions. This work presents a new algorithm that improves multi-view multi-person pose estimation, focusing on fast triangulation speeds and good generalization capabilities. The approach extends to whole-body pose estimation, capturing details from facial expressions to finger movements across multiple individuals and viewpoints. Adaptability to different settings is demonstrated through strong performance across unseen datasets and configurations. To support further progress in this field, all of this work is publicly accessible.", "date": "2025-03-27", "categories": []}
{"id": "2603.04090", "title": "EgoPoseFormer v2: Accurate Egocentric Human Motion Estimation for AR/VR", "abstract": "Egocentric human motion estimation is essential for AR/VR experiences, yet remains challenging due to limited body coverage from the egocentric viewpoint, frequent occlusions, and scarce labeled data. We present EgoPoseFormer v2, a method that addresses these challenges through two key contributions: (1) a transformer-based model for temporally consistent and spatially grounded body pose estimation, and (2) an auto-labeling system that enables the use of large unlabeled datasets for training. Our model is fully differentiable, introduces identity-conditioned queries, multi-view spatial refinement, causal temporal attention, and supports both keypoints and parametric body representations under a constant compute budget. The auto-labeling system scales learning to tens of millions of unlabeled frames via uncertainty-aware semi-supervised training. The system follows a teacher-student schema to generate pseudo-labels and guide training with uncertainty distillation, enabling the model to generalize to different environments. On the EgoBody3M benchmark, with a 0.8 ms latency on GPU, our model outperforms two state-of-the-art methods by 12.2% and 19.4% in accuracy, and reduces temporal jitter by 22.2% and 51.7%. Furthermore, our auto-labeling system further improves the wrist MPJPE by 13.1%.", "date": "2026-03-04", "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": "2603.03265", "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": "2602.18726", "title": "WiCompass: Oracle-driven Data Scaling for mmWave Human Pose Estimation", "abstract": "Millimeter-wave Human Pose Estimation (mmWave HPE) promises privacy but suffers from poor generalization under distribution shifts. We demonstrate that brute-force data scaling is ineffective for out-of-distribution (OOD) robustness; efficiency and coverage are the true bottlenecks. To address this, we introduce WiCompass, a coverage-aware data-collection framework. WiCompass leverages large-scale motion-capture corpora to build a universal pose space ``oracle'' that quantifies dataset redundancy and identifies underrepresented motions. Guided by this oracle, WiCompass employs a closed-loop policy to prioritize collecting informative missing samples. Experiments show that WiCompass consistently improves OOD accuracy at matched budgets and exhibits superior scaling behavior compared to conventional collection strategies. By shifting focus from brute-force scaling to coverage-aware data acquisition, this work offers a practical path toward robust mmWave sensing.", "date": "2026-02-21", "categories": []}
{"id": "2603.02130", "title": "Stereo-Inertial Poser: Towards Metric-Accurate Shape-Aware Motion Capture Using Sparse IMUs and a Single Stereo Camera", "abstract": "Recent advancements in visual-inertial motion capture systems have demonstrated the potential of combining monocular cameras with sparse inertial measurement units (IMUs) as cost-effective solutions, which effectively mitigate occlusion and drift issues inherent in single-modality systems. However, they are still limited by metric inaccuracies in global translations stemming from monocular depth ambiguity, and shape-agnostic local motion estimations that ignore anthropometric variations. We present Stereo-Inertial Poser, a real-time motion capture system that leverages a single stereo camera and six IMUs to estimate metric-accurate and shape-aware 3D human motion. By replacing the monocular RGB with stereo vision, our system resolves depth ambiguity through calibrated baseline geometry, enabling direct 3D keypoint extraction and body shape parameter estimation. IMU data and visual cues are fused for predicting drift-compensated joint positions and root movements, while a novel shape-aware fusion module dynamically harmonizes anthropometry variations with global translations. Our end-to-end pipeline achieves over 200 FPS without optimization-based post-processing, enabling real-time deployment. Quantitative evaluations across various datasets demonstrate state-of-the-art performance. Qualitative results show our method produces drift-free global translation under a long recording time and reduces foot-skating effects.", "date": "2026-03-02", "categories": []}
{"id": "2602.23618", "title": "Egocentric Visibility-Aware Human Pose Estimation", "abstract": "Egocentric human pose estimation (HPE) using a head-mounted device is crucial for various VR and AR applications, but it faces significant challenges due to keypoint invisibility. Nevertheless, none of the existing egocentric HPE datasets provide keypoint visibility annotations, and the existing methods often overlook the invisibility problem, treating visible and invisible keypoints indiscriminately during estimation. As a result, their capacity to accurately predict visible keypoints is compromised. In this paper, we first present Eva-3M, a large-scale egocentric visibility-aware HPE dataset comprising over 3.0M frames, with 435K of them annotated with keypoint visibility labels. Additionally, we augment the existing EMHI dataset with keypoint visibility annotations to further facilitate the research in this direction. Furthermore, we propose EvaPose, a novel egocentric visibility-aware HPE method that explicitly incorporates visibility information to enhance pose estimation accuracy. Extensive experiments validate the significant value of ground-truth visibility labels in egocentric HPE settings, and demonstrate that our EvaPose achieves state-of-the-art performance in both Eva-3M and EMHI datasets.", "date": "2026-02-27", "categories": []}
{"id": "2409.20469", "title": "PoseAdapt: Sustainable Human Pose Estimation via Continual Learning Benchmarks and Toolkit", "abstract": "Human pose estimators are typically retrained from scratch or naively fine-tuned whenever keypoint sets, sensing modalities, or deployment domains change--an inefficient, compute-intensive practice that rarely matches field constraints. We present PoseAdapt, an open-source framework and benchmark suite for continual pose model adaptation. PoseAdapt defines domain-incremental and class-incremental tracks that simulate realistic changes in density, lighting, and sensing modality, as well as skeleton growth. The toolkit supports two workflows: (i) Strategy Benchmarking, which lets researchers implement continual learning (CL) methods as plugins and evaluate them under standardized protocols; and (ii) Model Adaptation, which allows practitioners to adapt strong pretrained models to new tasks with minimal supervision. We evaluate representative regularization-based methods in single-step and sequential settings. Benchmarks enforce a fixed lightweight backbone, no access to past data, and tight per-step budgets. This isolates adaptation strategy effects, highlighting the difficulty of maintaining accuracy under strict resource limits. PoseAdapt connects modern CL techniques with practical pose estimation needs, enabling adaptable models that improve over time without repeated full retraining.", "date": "2024-09-30", "categories": []}
{"id": "2602.20051", "title": "SEAL-pose: Enhancing 3D Human Pose Estimation via a Learned Loss for Structural Consistency", "abstract": "3D human pose estimation (HPE) is characterized by intricate local and global dependencies among joints. Conventional supervised losses are limited in capturing these correlations because they treat each joint independently. Previous studies have attempted to promote structural consistency through manually designed priors or rule-based constraints; however, these approaches typically require manual specification and are often non-differentiable, limiting their use as end-to-end training objectives. We propose SEAL-pose, a data-driven framework in which a learnable loss-net trains a pose-net by evaluating structural plausibility. Rather than relying on hand-crafted priors, our joint-graph-based design enables the loss-net to learn complex structural dependencies directly from data. Extensive experiments on three 3D HPE benchmarks with eight backbones show that SEAL-pose reduces per-joint errors and improves pose plausibility compared with the corresponding backbones across all settings. Beyond improving each backbone, SEAL-pose also outperforms models with explicit structural constraints, despite not enforcing any such constraints. Finally, we analyze the relationship between the loss-net and structural consistency, and evaluate SEAL-pose in cross-dataset and in-the-wild settings.", "date": "2026-02-23", "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": "2602.13901", "title": "RPGD: RANSAC-P3P Gradient Descent for Extrinsic Calibration in 3D Human Pose Estimation", "abstract": "In this paper, we propose RPGD (RANSAC-P3P Gradient Descent), a human-pose-driven extrinsic calibration framework that robustly aligns MoCap-based 3D skeletal data with monocular or multi-view RGB cameras using only natural human motion. RPGD formulates extrinsic calibration as a coarse-to-fine problem tailored to human poses, combining the global robustness of RANSAC-P3P with Gradient-Descent-based refinement. We evaluate RPGD on three large-scale public 3D HPE datasets as well as on a self-collected in-the-wild dataset. Experimental results demonstrate that RPGD consistently recovers extrinsic parameters with accuracy comparable to the provided ground truth, achieving sub-pixel MPJPE reprojection error even in challenging, noisy settings. These results indicate that RPGD provides a practical and automatic solution for reliable extrinsic calibration of large-scale 3D HPE dataset collection.", "date": "2026-02-14", "categories": []}
{"id": "2504.10350", "title": "Benchmarking 3D Human Pose Estimation Models under Occlusions", "abstract": "Human Pose Estimation (HPE) involves detecting and localizing keypoints on the human body from visual data. In 3D HPE, occlusions, where parts of the body are not visible in the image, pose a significant challenge for accurate pose reconstruction. This paper presents a benchmark on the robustness of 3D HPE models under realistic occlusion conditions, involving combinations of occluded keypoints commonly observed in real-world scenarios. We evaluate nine state-of-the-art 2D-to-3D HPE models, spanning convolutional, transformer-based, graph-based, and diffusion-based architectures, using the BlendMimic3D dataset, a synthetic dataset with ground-truth 2D/3D annotations and occlusion labels. All models were originally trained on Human3.6M and tested here without retraining to assess their generalization. We introduce a protocol that simulates occlusion by adding noise into 2D keypoints based on real detector behavior, and conduct both global and per-joint sensitivity analyses. Our findings reveal that all models exhibit notable performance degradation under occlusion, with diffusion-based models underperforming despite their stochastic nature. Additionally, a per-joint occlusion analysis identifies consistent vulnerability in distal joints (e.g., wrists, feet) across models. Overall, this work highlights critical limitations of current 3D HPE models in handling occlusions, and provides insights for improving real-world robustness.", "date": "2025-04-14", "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": "2602.05590", "title": "EgoPoseVR: Spatiotemporal Multi-Modal Reasoning for Egocentric Full-Body Pose in Virtual Reality", "abstract": "Immersive virtual reality (VR) applications demand accurate, temporally coherent full-body pose tracking. Recent head-mounted camera-based approaches show promise in egocentric pose estimation, but encounter challenges when applied to VR head-mounted displays (HMDs), including temporal instability, inaccurate lower-body estimation, and the lack of real-time performance. To address these limitations, we present EgoPoseVR, an end-to-end framework for accurate egocentric full-body pose estimation in VR that integrates headset motion cues with egocentric RGB-D observations through a dual-modality fusion pipeline. A spatiotemporal encoder extracts frame- and joint-level representations, which are fused via cross-attention to fully exploit complementary motion cues across modalities. A kinematic optimization module then imposes constraints from HMD signals, enhancing the accuracy and stability of pose estimation. To facilitate training and evaluation, we introduce a large-scale synthetic dataset of over 1.8 million temporally aligned HMD and RGB-D frames across diverse VR scenarios. Experimental results show that EgoPoseVR outperforms state-of-the-art egocentric pose estimation models. A user study in real-world scenes further shows that EgoPoseVR achieved significantly higher subjective ratings in accuracy, stability, embodiment, and intention for future use compared to baseline methods. These results show that EgoPoseVR enables robust full-body pose tracking, offering a practical solution for accurate VR embodiment without requiring additional body-worn sensors or room-scale tracking systems.", "date": "2026-02-05", "categories": []}
{"id": "2602.05755", "title": "FMPose3D: monocular 3D pose estimation via flow matching", "abstract": "Monocular 3D pose estimation is fundamentally ill-posed due to depth ambiguity and occlusions, thereby motivating probabilistic methods that generate multiple plausible 3D pose hypotheses. In particular, diffusion-based models have recently demonstrated strong performance, but their iterative denoising process typically requires many timesteps for each prediction, making inference computationally expensive. In contrast, we leverage Flow Matching (FM) to learn a velocity field defined by an Ordinary Differential Equation (ODE), enabling efficient generation of 3D pose samples with only a few integration steps. We propose a novel generative pose estimation framework, FMPose3D, that formulates 3D pose estimation as a conditional distribution transport problem. It continuously transports samples from a standard Gaussian prior to the distribution of plausible 3D poses conditioned only on 2D inputs. Although ODE trajectories are deterministic, FMPose3D naturally generates various pose hypotheses by sampling different noise seeds. To obtain a single accurate prediction from those hypotheses, we further introduce a Reprojection-based Posterior Expectation Aggregation (RPEA) module, which approximates the Bayesian posterior expectation over 3D hypotheses. FMPose3D surpasses existing methods on the widely used human pose estimation benchmarks Human3.6M and MPI-INF-3DHP, and further achieves state-of-the-art performance on the 3D animal pose datasets Animal3D and CtrlAni3D, demonstrating strong performance across both 3D pose domains. The code is available at https://github.com/AdaptiveMotorControlLab/FMPose3D.", "date": "2026-02-05", "categories": []}
{"id": "2602.03064", "title": "JRDB-Pose3D: A Multi-person 3D Human Pose and Shape Estimation Dataset for Robotics", "abstract": "Real-world scenes are inherently crowded. Hence, estimating 3D poses of all nearby humans, tracking their movements over time, and understanding their activities within social and environmental contexts are essential for many applications, such as autonomous driving, robot perception, robot navigation, and human-robot interaction. However, most existing 3D human pose estimation datasets primarily focus on single-person scenes or are collected in controlled laboratory environments, which restricts their relevance to real-world applications. To bridge this gap, we introduce JRDB-Pose3D, which captures multi-human indoor and outdoor environments from a mobile robotic platform. JRDB-Pose3D provides rich 3D human pose annotations for such complex and dynamic scenes, including SMPL-based pose annotations with consistent body-shape parameters and track IDs for each individual over time. JRDB-Pose3D contains, on average, 5-10 human poses per frame, with some scenes featuring up to 35 individuals simultaneously. The proposed dataset presents unique challenges, including frequent occlusions, truncated bodies, and out-of-frame body parts, which closely reflect real-world environments. Moreover, JRDB-Pose3D inherits all available annotations from the JRDB dataset, such as 2D pose, information about social grouping, activities, and interactions, full-scene semantic masks with consistent human- and object-level tracking, and detailed annotations for each individual, such as age, gender, and race, making it a holistic dataset for a wide range of downstream perception and human-centric understanding tasks.", "date": "2026-02-03", "categories": []}
{"id": "2602.03126", "title": "Flexible Geometric Guidance for Probabilistic Human Pose Estimation with Diffusion Models", "abstract": "3D human pose estimation from 2D images is a challenging problem due to depth ambiguity and occlusion. Because of these challenges the task is underdetermined, where there exists multiple -- possibly infinite -- poses that are plausible given the image. Despite this, many prior works assume the existence of a deterministic mapping and estimate a single pose given an image. Furthermore, methods based on machine learning require a large amount of paired 2D-3D data to train and suffer from generalization issues to unseen scenarios. To address both of these issues, we propose a framework for pose estimation using diffusion models, which enables sampling from a probability distribution over plausible poses which are consistent with a 2D image. Our approach falls under the guidance framework for conditional generation, and guides samples from an unconditional diffusion model, trained only on 3D data, using the gradients of the heatmaps from a 2D keypoint detector. We evaluate our method on the Human 3.6M dataset under best-of-$m$ multiple hypothesis evaluation, showing state-of-the-art performance among methods which do not require paired 2D-3D data for training. We additionally evaluate the generalization ability using the MPI-INF-3DHP and 3DPW datasets and demonstrate competitive performance. Finally, we demonstrate the flexibility of our framework by using it for novel tasks including pose generation and pose completion, without the need to train bespoke conditional models. We make code available at https://github.com/fsnelgar/diffusion_pose .", "date": "2026-02-03", "categories": []}
{"id": "2602.01095", "title": "PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space", "abstract": "3D human pose lifting from a single RGB image is a challenging task in 3D vision. Existing methods typically establish a direct joint-to-joint mapping from 2D to 3D poses based on 2D features. This formulation suffers from two fundamental limitations: inevitable error propagation from input predicted 2D pose to 3D predictions and inherent difficulties in handling self-occlusion cases. In this paper, we propose PandaPose, a 3D human pose lifting approach via propagating 2D pose prior to 3D anchor space as the unified intermediate representation. Specifically, our 3D anchor space comprises: (1) Joint-wise 3D anchors in the canonical coordinate system, providing accurate and robust priors to mitigate 2D pose estimation inaccuracies. (2) Depth-aware joint-wise feature lifting that hierarchically integrates depth information to resolve self-occlusion ambiguities. (3) The anchor-feature interaction decoder that incorporates 3D anchors with lifted features to generate unified anchor queries encapsulating joint-wise 3D anchor set, visual cues and geometric depth information. The anchor queries are further employed to facilitate anchor-to-joint ensemble prediction. Experiments on three well-established benchmarks (i.e., Human3.6M, MPI-INF-3DHP and 3DPW) demonstrate the superiority of our proposition. The substantial reduction in error by $14.7\\%$ compared to SOTA methods on the challenging conditions of Human3.6M and qualitative comparisons further showcase the effectiveness and robustness of our approach.", "date": "2026-02-01", "categories": []}
{"id": "2408.16536", "title": "Are Pose Estimators Ready for the Open World? STAGE: A GenAI Toolkit for Auditing 3D Human Pose Estimators", "abstract": "For safety-critical applications, it is crucial to audit 3D human pose estimators before deployment. Will the system break down if the weather or the clothing changes? Is it robust regarding gender and age? To answer these questions and more, we need controlled studies with images that differ in a single attribute, but real benchmarks cannot provide such pairs. We thus present STAGE, a GenAI data toolkit for auditing 3D human pose estimators. For STAGE, we develop the first GenAI image creator with accurate 3D pose control and propose a novel evaluation strategy to isolate and quantify the effects of single factors such as gender, ethnicity, age, clothing, location, and weather. Enabled by STAGE, we generate a series of benchmarks to audit, for the first time, the sensitivity of popular pose estimators towards such factors. Our results show that natural variations can severely degrade pose estimator performance, raising doubts about their readiness for open-world deployment. We aim to highlight these robustness issues and establish STAGE as a benchmark to quantify them.", "date": "2024-08-28", "categories": []}
{"id": "2505.02287", "title": "Continuous Normalizing Flows for Uncertainty-Aware Human Pose Estimation", "abstract": "Human Pose Estimation (HPE) is increasingly important for applications like virtual reality and motion analysis, yet current methods struggle with balancing accuracy, computational efficiency, and reliable uncertainty quantification (UQ). Traditional regression-based methods assume fixed distributions, which might lead to poor UQ. Heatmap-based methods effectively model the output distribution using likelihood heatmaps, however, they demand significant resources. To address this, we propose Continuous Flow Residual Estimation (CFRE), an integration of Continuous Normalizing Flows (CNFs) into regression-based models, which allows for dynamic distribution adaptation. Through extensive experiments, we show that CFRE leads to better accuracy and uncertainty quantification with retained computational efficiency on both 2D and 3D human pose estimation tasks.", "date": "2025-05-04", "categories": []}
{"id": "2601.19519", "title": "Mocap Anywhere: Towards Pairwise-Distance based Motion Capture in the Wild (for the Wild)", "abstract": "We introduce a novel motion capture system that reconstructs full-body 3D motion using only sparse pairwise distance (PWD) measurements from body-mounted(UWB) sensors. Using time-of-flight ranging between wireless nodes, our method eliminates the need for external cameras, enabling robust operation in uncontrolled and outdoor environments. Unlike traditional optical or inertial systems, our approach is shape-invariant and resilient to environmental constraints such as lighting and magnetic interference. At the core of our system is Wild-Poser (WiP for short), a compact, real-time Transformer-based architecture that directly predicts 3D joint positions from noisy or corrupted PWD measurements, which can later be used for joint rotation reconstruction via learned methods. WiP generalizes across subjects of varying morphologies, including non-human species, without requiring individual body measurements or shape fitting. Operating in real time, WiP achieves low joint position error and demonstrates accurate 3D motion reconstruction for both human and animal subjects in-the-wild. Our empirical analysis highlights its potential for scalable, low-cost, and general purpose motion capture in real-world settings.", "date": "2026-01-27", "categories": []}
{"id": "2601.19580", "title": "QuaMo: Quaternion Motions for Vision-based 3D Human Kinematics Capture", "abstract": "Vision-based 3D human motion capture from videos remains a challenge in computer vision. Traditional 3D pose estimation approaches often ignore the temporal consistency between frames, causing implausible and jittery motion. The emerging field of kinematics-based 3D motion capture addresses these issues by estimating the temporal transitioning between poses instead. A major drawback in current kinematics approaches is their reliance on Euler angles. Despite their simplicity, Euler angles suffer from discontinuity that leads to unstable motion reconstructions, especially in online settings where trajectory refinement is unavailable. Contrarily, quaternions have no discontinuity and can produce continuous transitions between poses. In this paper, we propose QuaMo, a novel Quaternion Motions method using quaternion differential equations (QDE) for human kinematics capture. We utilize the state-space model, an effective system for describing real-time kinematics estimations, with quaternion state and the QDE describing quaternion velocity. The corresponding angular acceleration is computed from a meta-PD controller with a novel acceleration enhancement that adaptively regulates the control signals as the human quickly changes to a new pose. Unlike previous work, our QDE is solved under the quaternion unit-sphere constraint that results in more accurate estimations. Experimental results show that our novel formulation of the QDE with acceleration enhancement accurately estimates 3D human kinematics with no discontinuity and minimal implausibilities. QuaMo outperforms comparable state-of-the-art methods on multiple datasets, namely Human3.6M, Fit3D, SportsPose and AIST. The code is available at https://github.com/cuongle1206/QuaMo", "date": "2026-01-27", "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": "2601.12252", "title": "Breaking Coordinate Overfitting: Geometry-Aware WiFi Sensing for Cross-Layout 3D Pose Estimation", "abstract": "WiFi-based 3D human pose estimation offers a low-cost and privacy-preserving alternative to vision-based systems for smart interaction. However, existing approaches rely on visual 3D poses as supervision and directly regress CSI to a camera-based coordinate system. We find that this practice leads to coordinate overfitting: models memorize deployment-specific WiFi transceiver layouts rather than only learning activity-relevant representations, resulting in severe generalization failures. To address this challenge, we present PerceptAlign, the first geometry-conditioned framework for WiFi-based cross-layout pose estimation. PerceptAlign introduces a lightweight coordinate unification procedure that aligns WiFi and vision measurements in a shared 3D space using only two checkerboards and a few photos. Within this unified space, it encodes calibrated transceiver positions into high-dimensional embeddings and fuses them with CSI features, making the model explicitly aware of device geometry as a conditional variable. This design forces the network to disentangle human motion from deployment layouts, enabling robust and, for the first time, layout-invariant WiFi pose estimation. To support systematic evaluation, we construct the largest cross-domain 3D WiFi pose estimation dataset to date, comprising 21 subjects, 5 scenes, 18 actions, and 7 device layouts. Experiments show that PerceptAlign reduces in-domain error by 12.3% and cross-domain error by more than 60% compared to state-of-the-art baselines. These results establish geometry-conditioned learning as a viable path toward scalable and practical WiFi sensing.", "date": "2026-01-18", "categories": []}
{"id": "2601.13913", "title": "On the Role of Rotation Equivariance in Monocular 3D Human Pose Estimation", "abstract": "Estimating 3D from 2D is one of the central tasks in computer vision. In this work, we consider the monocular setting, i.e. single-view input, for 3D human pose estimation (HPE). Here, the task is to predict a 3D point set of human skeletal joints from a single 2D input image. While by definition this is an ill-posed problem, recent work has presented methods that solve it with up to several-centimetre error. Typically, these methods employ a two-step approach, where the first step is to detect the 2D skeletal joints in the input image, followed by the step of 2D-to-3D lifting. We find that common lifting models fail when encountering a rotated input. We argue that learning a single human pose along with its in-plane rotations is considerably easier and more geometrically grounded than directly learning a point-to-point mapping. Furthermore, our intuition is that endowing the model with the notion of rotation equivariance without explicitly constraining its parameter space should lead to a more straightforward learning process than one with equivariance by design. Utilising the common HPE benchmarks, we confirm that the 2D rotation equivariance per se improves the model performance on human poses akin to rotations in the image plane, and can be efficiently and straightforwardly learned by augmentation, outperforming state-of-the-art equivariant-by-design methods.", "date": "2026-01-20", "categories": []}
{"id": "2512.23860", "title": "Lifelong Domain Adaptive 3D Human Pose Estimation", "abstract": "3D Human Pose Estimation (3D HPE) is vital in various applications, from person re-identification and action recognition to virtual reality. However, the reliance on annotated 3D data collected in controlled environments poses challenges for generalization to diverse in-the-wild scenarios. Existing domain adaptation (DA) paradigms like general DA and source-free DA for 3D HPE overlook the issues of non-stationary target pose datasets. To address these challenges, we propose a novel task named lifelong domain adaptive 3D HPE. To our knowledge, we are the first to introduce the lifelong domain adaptation to the 3D HPE task. In this lifelong DA setting, the pose estimator is pretrained on the source domain and subsequently adapted to distinct target domains. Moreover, during adaptation to the current target domain, the pose estimator cannot access the source and all the previous target domains. The lifelong DA for 3D HPE involves overcoming challenges in adapting to current domain poses and preserving knowledge from previous domains, particularly combating catastrophic forgetting. We present an innovative Generative Adversarial Network (GAN) framework, which incorporates 3D pose generators, a 2D pose discriminator, and a 3D pose estimator. This framework effectively mitigates domain shifts and aligns original and augmented poses. Moreover, we construct a novel 3D pose generator paradigm, integrating pose-aware, temporal-aware, and domain-aware knowledge to enhance the current domain's adaptation and alleviate catastrophic forgetting on previous domains. Our method demonstrates superior performance through extensive experiments on diverse domain adaptive 3D HPE datasets.", "date": "2025-12-29", "categories": []}
{"id": "2512.12718", "title": "Spinal Line Detection for Posture Evaluation through Train-ing-free 3D Human Body Reconstruction with 2D Depth Images", "abstract": "The spinal angle is an important indicator of body balance. It is important to restore the 3D shape of the human body and estimate the spine center line. Existing mul-ti-image-based body restoration methods require expensive equipment and complex pro-cedures, and single image-based body restoration methods have limitations in that it is difficult to accurately estimate the internal structure such as the spine center line due to occlusion and viewpoint limitation. This study proposes a method to compensate for the shortcomings of the multi-image-based method and to solve the limitations of the sin-gle-image method. We propose a 3D body posture analysis system that integrates depth images from four directions to restore a 3D human model and automatically estimate the spine center line. Through hierarchical matching of global and fine registration, restora-tion to noise and occlusion is performed. Also, the Adaptive Vertex Reduction is applied to maintain the resolution and shape reliability of the mesh, and the accuracy and stabil-ity of spinal angle estimation are simultaneously secured by using the Level of Detail en-semble. The proposed method achieves high-precision 3D spine registration estimation without relying on training data or complex neural network models, and the verification confirms the improvement of matching quality.", "date": "2025-12-14", "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": "2601.00625", "title": "RePose: A Real-Time 3D Human Pose Estimation and Biomechanical Analysis Framework for Rehabilitation", "abstract": "We propose a real-time 3D human pose estimation and motion analysis method termed RePose for rehabilitation training. It is capable of real-time monitoring and evaluation of patients'motion during rehabilitation, providing immediate feedback and guidance to assist patients in executing rehabilitation exercises correctly. Firstly, we introduce a unified pipeline for end-to-end real-time human pose estimation and motion analysis using RGB video input from multiple cameras which can be applied to the field of rehabilitation training. The pipeline can help to monitor and correct patients'actions, thus aiding them in regaining muscle strength and motor functions. Secondly, we propose a fast tracking method for medical rehabilitation scenarios with multiple-person interference, which requires less than 1ms for tracking for a single frame. Additionally, we modify SmoothNet for real-time posture estimation, effectively reducing pose estimation errors and restoring the patient's true motion state, making it visually smoother. Finally, we use Unity platform for real-time monitoring and evaluation of patients' motion during rehabilitation, and to display the muscle stress conditions to assist patients with their rehabilitation training.", "date": "2026-01-02", "categories": []}
{"id": "2512.22228", "title": "KAN-FPN-Stem:A KAN-Enhanced Feature Pyramid Stem for Boosting ViT-based Pose Estimation", "abstract": "Vision Transformers (ViT) have demonstrated significant promise in dense prediction tasks such as pose estimation. However, their performance is frequently constrained by the overly simplistic front-end designs employed in models like ViTPose. This naive patchification mechanism struggles to effectively handle multi-scale variations and results in irreversible information loss during the initial feature extraction phase. To overcome this limitation, we introduce a novel KAN-enhanced FPN-Stem architecture. Through rigorous ablation studies, we first identified that the true bottleneck for performance improvement lies not in plug-and-play attention modules (e.g., CBAM), but in the post-fusion non-linear smoothing step within the FPN. Guided by this insight, our core innovation is to retain the classic \"upsample-and-add\" fusion stream of the FPN, but replace its terminal, standard linear 3x3 smoothing convolution with a powerful KAN-based convolutional layer. Leveraging its superior non-linear modeling capabilities, this KAN-based layer adaptively learns and rectifies the \"artifacts\" generated during the multi-scale fusion process. Extensive experiments on the COCO dataset demonstrate that our KAN-FPN-Stem achieves a significant performance boost of up to +2.0 AP over the lightweight ViTPose-S baseline. This work not only delivers a plug-and-play, high-performance module but, more importantly, reveals that: the performance bottleneck in ViT front-end often lies not in 'feature refinement' (Attention), but in the quality of 'feature fusion' (Fusion). Furthermore, it provides an effective path to address this bottleneck through the introduction of the KAN operator.", "date": "2025-12-23", "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": "2512.17545", "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": "2512.16494", "title": "PoseMoE: Mixture-of-Experts Network for Monocular 3D Human Pose Estimation", "abstract": "The lifting-based methods have dominated monocular 3D human pose estimation by leveraging detected 2D poses as intermediate representations. The 2D component of the final 3D human pose benefits from the detected 2D poses, whereas its depth counterpart must be estimated from scratch. The lifting-based methods encode the detected 2D pose and unknown depth in an entangled feature space, explicitly introducing depth uncertainty to the detected 2D pose, thereby limiting overall estimation accuracy. This work reveals that the depth representation is pivotal for the estimation process. Specifically, when depth is in an initial, completely unknown state, jointly encoding depth features with 2D pose features is detrimental to the estimation process. In contrast, when depth is initially refined to a more dependable state via network-based estimation, encoding it together with 2D pose information is beneficial. To address this limitation, we present a Mixture-of-Experts network for monocular 3D pose estimation named PoseMoE. Our approach introduces: (1) A mixture-of-experts network where specialized expert modules refine the well-detected 2D pose features and learn the depth features. This mixture-of-experts design disentangles the feature encoding process for 2D pose and depth, therefore reducing the explicit influence of uncertain depth features on 2D pose features. (2) A cross-expert knowledge aggregation module is proposed to aggregate cross-expert spatio-temporal contextual information. This step enhances features through bidirectional mapping between 2D pose and depth. Extensive experiments show that our proposed PoseMoE outperforms the conventional lifting-based methods on three widely used datasets: Human3.6M, MPI-INF-3DHP, and 3DPW.", "date": "2025-12-18", "categories": []}
{"id": "2512.16791", "title": "KineST: A Kinematics-guided Spatiotemporal State Space Model for Human Motion Tracking from Sparse Signals", "abstract": "Full-body motion tracking plays an essential role in AR/VR applications, bridging physical and virtual interactions. However, it is challenging to reconstruct realistic and diverse full-body poses based on sparse signals obtained by head-mounted displays, which are the main devices in AR/VR scenarios. Existing methods for pose reconstruction often incur high computational costs or rely on separately modeling spatial and temporal dependencies, making it difficult to balance accuracy, temporal coherence, and efficiency. To address this problem, we propose KineST, a novel kinematics-guided state space model, which effectively extracts spatiotemporal dependencies while integrating local and global pose perception. The innovation comes from two core ideas. Firstly, in order to better capture intricate joint relationships, the scanning strategy within the State Space Duality framework is reformulated into kinematics-guided bidirectional scanning, which embeds kinematic priors. Secondly, a mixed spatiotemporal representation learning approach is employed to tightly couple spatial and temporal contexts, balancing accuracy and smoothness. Additionally, a geometric angular velocity loss is introduced to impose physically meaningful constraints on rotational variations for further improving motion stability. Extensive experiments demonstrate that KineST has superior performance in both accuracy and temporal consistency within a lightweight framework. Project page: https://kaka-1314.github.io/KineST/", "date": "2025-12-18", "categories": []}
{"id": "2512.15212", "title": "From Camera to World: A Plug-and-Play Module for Human Mesh Transformation", "abstract": "Reconstructing accurate 3D human meshes in the world coordinate system from in-the-wild images remains challenging due to the lack of camera rotation information. While existing methods achieve promising results in the camera coordinate system by assuming zero camera rotation, this simplification leads to significant errors when transforming the reconstructed mesh to the world coordinate system. To address this challenge, we propose Mesh-Plug, a plug-and-play module that accurately transforms human meshes from camera coordinates to world coordinates. Our key innovation lies in a human-centered approach that leverages both RGB images and depth maps rendered from the initial mesh to estimate camera rotation parameters, eliminating the dependency on environmental cues. Specifically, we first train a camera rotation prediction module that focuses on the human body's spatial configuration to estimate camera pitch angle. Then, by integrating the predicted camera parameters with the initial mesh, we design a mesh adjustment module that simultaneously refines the root joint orientation and body pose. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods on the benchmark datasets SPEC-SYN and SPEC-MTP.", "date": "2025-12-17", "categories": []}
{"id": "2512.15488", "title": "RUMPL: Ray-Based Transformers for Universal Multi-View 2D to 3D Human Pose Lifting", "abstract": "Estimating 3D human poses from 2D images remains challenging due to occlusions and projective ambiguity. Multi-view learning-based approaches mitigate these issues but often fail to generalize to real-world scenarios, as large-scale multi-view datasets with 3D ground truth are scarce and captured under constrained conditions. To overcome this limitation, recent methods rely on 2D pose estimation combined with 2D-to-3D pose lifting trained on synthetic data. Building on our previous MPL framework, we propose RUMPL, a transformer-based 3D pose lifter that introduces a 3D ray-based representation of 2D keypoints. This formulation makes the model independent of camera calibration and the number of views, enabling universal deployment across arbitrary multi-view configurations without retraining or fine-tuning. A new View Fusion Transformer leverages learned fused-ray tokens to aggregate information along rays, further improving multi-view consistency. Extensive experiments demonstrate that RUMPL reduces MPJPE by up to 53% compared to triangulation and over 60% compared to transformer-based image-representation baselines. Results on new benchmarks, including in-the-wild multi-view and multi-person datasets, confirm its robustness and scalability. The framework's source code is available at https://github.com/aghasemzadeh/OpenRUMPL", "date": "2025-12-17", "categories": []}
{"id": "2512.14162", "title": "FastDDHPose: Towards Unified, Efficient, and Disentangled 3D Human Pose Estimation", "abstract": "Recent approaches for monocular 3D human pose estimation (3D HPE) have achieved leading performance by directly regressing 3D poses from 2D keypoint sequences. Despite the rapid progress in 3D HPE, existing methods are typically trained and evaluated under disparate frameworks, lacking a unified framework for fair comparison. To address these limitations, we propose Fast3DHPE, a modular framework that facilitates rapid reproduction and flexible development of new methods. By standardizing training and evaluation protocols, Fast3DHPE enables fair comparison across 3D human pose estimation methods while significantly improving training efficiency. Within this framework, we introduce FastDDHPose, a Disentangled Diffusion-based 3D Human Pose Estimation method which leverages the strong latent distribution modeling capability of diffusion models to explicitly model the distributions of bone length and bone direction while avoiding further amplification of hierarchical error accumulation. Moreover, we design an efficient Kinematic-Hierarchical Spatial and Temporal Denoiser that encourages the model to focus on kinematic joint hierarchies while avoiding unnecessary modeling of overly complex joint topologies. Extensive experiments on Human3.6M and MPI-INF-3DHP show that the Fast3DHPE framework enables fair comparison of all methods while significantly improving training efficiency. Within this unified framework, FastDDHPose achieves state-of-the-art performance with strong generalization and robustness in in-the-wild scenarios. The framework and models will be released at: https://github.com/Andyen512/Fast3DHPE", "date": "2025-12-16", "categories": []}
{"id": "2401.08930", "title": "PADS: Plug-and-Play 3D Human Pose Analysis via Diffusion Generative Modeling", "abstract": "Diffusion models have demonstrated impressive capabilities in modeling complex data distributions and are increasingly applied in various generative tasks. In this work, we propose Pose Analysis by Diffusion Synthesis PADS, a unified generative modeling framework for 3D human pose analysis. PADS first learns a task-agnostic 3D pose prior via unconditional diffusion synthesis and then performs training-free adaptation to a wide range of pose analysis tasks, including 3D pose estimation, denoising, completion, etc., through a posterior sampling scheme. By formulating each task as an inverse problem with a known forward operator, PADS injects task-specific constraints during inference while keeping the pose prior fixed. This plug-and-play framework removes the need for task-specific supervision or retraining, offering flexibility and scalability across diverse conditions. Extensive experiments on different benchmarks showcase the superior performance against both learning-based and optimization-based baselines, demonstrating the effectiveness and generalization capability of our method.", "date": "2024-01-17", "categories": []}
{"id": "2512.10321", "title": "Point2Pose: A Generative Framework for 3D Human Pose Estimation with Multi-View Point Cloud Dataset", "abstract": "We propose a novel generative approach for 3D human pose estimation. 3D human pose estimation poses several key challenges due to the complex geometry of the human body, self-occluding joints, and the requirement for large-scale real-world motion datasets. To address these challenges, we introduce Point2Pose, a framework that effectively models the distribution of human poses conditioned on sequential point cloud and pose history. Specifically, we employ a spatio-temporal point cloud encoder and a pose feature encoder to extract joint-wise features, followed by an attention-based generative regressor. Additionally, we present a large-scale indoor dataset MVPose3D, which contains multiple modalities, including IMU data of non-trivial human motions, dense multi-view point clouds, and RGB images. Experimental results show that the proposed method outperforms the baseline models, demonstrating its superior performance across various datasets.", "date": "2025-12-11", "categories": []}
{"id": "2512.08406", "title": "SAM-Body4D: Training-Free 4D Human Body Mesh Recovery from Videos", "abstract": "Human Mesh Recovery (HMR) aims to reconstruct 3D human pose and shape from 2D observations and is fundamental to human-centric understanding in real-world scenarios. While recent image-based HMR methods such as SAM 3D Body achieve strong robustness on in-the-wild images, they rely on per-frame inference when applied to videos, leading to temporal inconsistency and degraded performance under occlusions. We address these issues without extra training by leveraging the inherent human continuity in videos. We propose SAM-Body4D, a training-free framework for temporally consistent and occlusion-robust HMR from videos. We first generate identity-consistent masklets using a promptable video segmentation model, then refine them with an Occlusion-Aware module to recover missing regions. The refined masklets guide SAM 3D Body to produce consistent full-body mesh trajectories, while a padding-based parallel strategy enables efficient multi-human inference. Experimental results demonstrate that SAM-Body4D achieves improved temporal stability and robustness in challenging in-the-wild videos, without any retraining. Our code and demo are available at: https://github.com/gaomingqi/sam-body4d.", "date": "2025-12-09", "categories": []}
{"id": "2512.06783", "title": "Physics Informed Human Posture Estimation Based on 3D Landmarks from Monocular RGB-Videos", "abstract": "Applications providing automated coaching for physical training are increasing in popularity, for example physical therapy. These applications rely on accurate and robust pose estimation using monocular video streams. State-of-the-art models like BlazePose excel in real-time pose tracking, but their lack of anatomical constraints indicates improvement potential by including physical knowledge. We present a real-time post-processing algorithm fusing the strengths of BlazePose 3D and 2D estimations using a weighted optimization, penalizing deviations from expected bone length and biomechanical models. Bone length estimations are refined to the individual anatomy using a Kalman filter with adapting measurement trust. Evaluation using the Physio2.2M dataset shows a 10.2 percent reduction in 3D MPJPE and a 16.6 percent decrease in errors of angles between body segments compared to BlazePose 3D estimation. Our method provides a robust, anatomically consistent pose estimation based on a computationally efficient video-to-3D pose estimation, suitable for automated physiotherapy, healthcare, and sports coaching on consumer-level laptops and mobile devices. The refinement runs on the backend with anonymized data only.", "date": "2025-12-07", "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": "2511.08294", "title": "SkelSplat: Robust Multi-view 3D Human Pose Estimation with Differentiable Gaussian Rendering", "abstract": "Accurate 3D human pose estimation is fundamental for applications such as augmented reality and human-robot interaction. State-of-the-art multi-view methods learn to fuse predictions across views by training on large annotated datasets, leading to poor generalization when the test scenario differs. To overcome these limitations, we propose SkelSplat, a novel framework for multi-view 3D human pose estimation based on differentiable Gaussian rendering. Human pose is modeled as a skeleton of 3D Gaussians, one per joint, optimized via differentiable rendering to enable seamless fusion of arbitrary camera views without 3D ground-truth supervision. Since Gaussian Splatting was originally designed for dense scene reconstruction, we propose a novel one-hot encoding scheme that enables independent optimization of human joints. SkelSplat outperforms approaches that do not rely on 3D ground truth in Human3.6M and CMU, while reducing the cross-dataset error up to 47.8% compared to learning-based methods. Experiments on Human3.6M-Occ and Occlusion-Person demonstrate robustness to occlusions, without scenario-specific fine-tuning. Our project page is available here: https://skelsplat.github.io.", "date": "2025-11-11", "categories": []}
{"id": "2511.13208", "title": "End-to-End Multi-Person Pose Estimation with Pose-Aware Video Transformer", "abstract": "Existing multi-person video pose estimation methods typically adopt a two-stage pipeline: detecting individuals in each frame, followed by temporal modeling for single person pose estimation. This design relies on heuristic operations such as detection, RoI cropping, and non-maximum suppression (NMS), limiting both accuracy and efficiency. In this paper, we present a fully end-to-end framework for multi-person 2D pose estimation in videos, effectively eliminating heuristic operations. A key challenge is to associate individuals across frames under complex and overlapping temporal trajectories. To address this, we introduce a novel Pose-Aware Video transformEr Network (PAVE-Net), which features a spatial encoder to model intra-frame relations and a spatiotemporal pose decoder to capture global dependencies across frames. To achieve accurate temporal association, we propose a pose-aware attention mechanism that enables each pose query to selectively aggregate features corresponding to the same individual across consecutive frames. Additionally, we explicitly model spatiotemporal dependencies among pose keypoints to improve accuracy. Notably, our approach is the first end-to-end method for multi-frame 2D human pose estimation. Extensive experiments show that PAVE-Net substantially outperforms prior image-based end-to-end methods, achieving a 6.0 mAP improvement on PoseTrack2017, and delivers accuracy competitive with state-of-the-art two-stage video based approaches, while offering significant gains in efficiency. Project page: https://github.com/zgspose/PAVENet.", "date": "2025-11-17", "categories": []}
{"id": "2504.08718", "title": "EMO-X: Efficient Multi-Person Pose and Shape Estimation in One-Stage", "abstract": "Expressive Human Pose and Shape Estimation (EHPS) aims to jointly estimate human pose, hand gesture, and facial expression from monocular images. Existing methods predominantly rely on Transformer-based architectures, which suffer from quadratic complexity in self-attention, leading to substantial computational overhead, especially in multi-person scenarios. Recently, Mamba has emerged as a promising alternative to Transformers due to its efficient global modeling capability. However, it remains limited in capturing fine-grained local dependencies, which are essential for precise EHPS. To address these issues, we propose EMO-X, the Efficient Multi-person One-stage model for multi-person EHPS. Specifically, we explore a Scan-based Global-Local Decoder (SGLD) that integrates global context with skeleton-aware local features to iteratively enhance human tokens. Our EMO-X leverages the superior global modeling capability of Mamba and designs a local bidirectional scan mechanism for skeleton-aware local refinement. Comprehensive experiments demonstrate that EMO-X strikes an excellent balance between efficiency and accuracy. Notably, it achieves a significant reduction in computational complexity, requiring 69.8% less inference time compared to state-of-the-art (SOTA) methods, while outperforming most of them in accuracy.", "date": "2025-04-11", "categories": []}
{"id": "2511.19105", "title": "Graph-based 3D Human Pose Estimation using WiFi Signals", "abstract": "WiFi-based human pose estimation (HPE) has attracted increasing attention due to its resilience to occlusion and privacy-preserving compared to camera-based methods. However, existing WiFi-based HPE approaches often employ regression networks that directly map WiFi channel state information (CSI) to 3D joint coordinates, ignoring the inherent topological relationships among human joints. In this paper, we present GraphPose-Fi, a graph-based framework that explicitly models skeletal topology for WiFi-based 3D HPE. Our framework comprises a CNN encoder shared across antennas for subcarrier-time feature extraction, a lightweight attention module that adaptively reweights features over time and across antennas, and a graph-based regression head that combines GCN layers with self-attention to capture local topology and global dependencies. Our proposed method significantly outperforms existing methods on the MM-Fi dataset in various settings. The source code is available at: https://github.com/Cirrick/GraphPose-Fi.", "date": "2025-11-24", "categories": []}
{"id": "2511.19326", "title": "MonoMSK: Monocular 3D Musculoskeletal Dynamics Estimation", "abstract": "Reconstructing biomechanically realistic 3D human motion - recovering both kinematics (motion) and kinetics (forces) - is a critical challenge. While marker-based systems are lab-bound and slow, popular monocular methods use oversimplified, anatomically inaccurate models (e.g., SMPL) and ignore physics, fundamentally limiting their biomechanical fidelity. In this work, we introduce MonoMSK, a hybrid framework that bridges data-driven learning and physics-based simulation for biomechanically realistic 3D human motion estimation from monocular video. MonoMSK jointly recovers both kinematics (motions) and kinetics (forces and torques) through an anatomically accurate musculoskeletal model. By integrating transformer-based inverse dynamics with differentiable forward kinematics and dynamics layers governed by ODE-based simulation, MonoMSK establishes a physics-regulated inverse-forward loop that enforces biomechanical causality and physical plausibility. A novel forward-inverse consistency loss further aligns motion reconstruction with the underlying kinetic reasoning. Experiments on BML-MoVi, BEDLAM, and OpenCap show that MonoMSK significantly outperforms state-of-the-art methods in kinematic accuracy, while for the first time enabling precise monocular kinetics estimation.", "date": "2025-11-24", "categories": []}
{"id": "2511.15874", "title": "WALDO: Where Unseen Model-based 6D Pose Estimation Meets Occlusion", "abstract": "Accurate 6D object pose estimation is vital for robotics, augmented reality, and scene understanding. For seen objects, high accuracy is often attainable via per-object fine-tuning but generalizing to unseen objects remains a challenge. To address this problem, past arts assume access to CAD models at test time and typically follow a multi-stage pipeline to estimate poses: detect and segment the object, propose an initial pose, and then refine it. Under occlusion, however, the early-stage of such pipelines are prone to errors, which can propagate through the sequential processing, and consequently degrade the performance. To remedy this shortcoming, we propose four novel extensions to model-based 6D pose estimation methods: (i) a dynamic non-uniform dense sampling strategy that focuses computation on visible regions, reducing occlusion-induced errors; (ii) a multi-hypothesis inference mechanism that retains several confidence-ranked pose candidates, mitigating brittle single-path failures; (iii) iterative refinement to progressively improve pose accuracy; and (iv) series of occlusion-focused training augmentations that strengthen robustness and generalization. Furthermore, we propose a new weighted by visibility metric for evaluation under occlusion to minimize the bias in the existing protocols. Via extensive empirical evaluations, we show that our proposed approach achieves more than 5% improvement in accuracy on ICBIN and more than 2% on BOP dataset benchmarks, while achieving approximately 3 times faster inference.", "date": "2025-11-19", "categories": []}
{"id": "2511.14394", "title": "BEDLAM2.0: Synthetic Humans and Cameras in Motion", "abstract": "Inferring 3D human motion from video remains a challenging problem with many applications. While traditional methods estimate the human in image coordinates, many applications require human motion to be estimated in world coordinates. This is particularly challenging when there is both human and camera motion. Progress on this topic has been limited by the lack of rich video data with ground truth human and camera movement. We address this with BEDLAM2.0, a new dataset that goes beyond the popular BEDLAM dataset in important ways. In addition to introducing more diverse and realistic cameras and camera motions, BEDLAM2.0 increases diversity and realism of body shape, motions, clothing, hair, and 3D environments. Additionally, it adds shoes, which were missing in BEDLAM. BEDLAM has become a key resource for training 3D human pose and motion regressors today and we show that BEDLAM2.0 is significantly better, particularly for training methods that estimate humans in world coordinates. We compare state-of-the art methods trained on BEDLAM and BEDLAM2.0, and find that BEDLAM2.0 significantly improves accuracy over BEDLAM. For research purposes, we provide the rendered videos, ground truth body parameters, and camera motions. We also provide the 3D assets to which we have rights and links to those from third parties.", "date": "2025-11-18", "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": "2510.26196", "title": "Sketch2PoseNet: Efficient and Generalized Sketch to 3D Human Pose Prediction", "abstract": "3D human pose estimation from sketches has broad applications in computer animation and film production. Unlike traditional human pose estimation, this task presents unique challenges due to the abstract and disproportionate nature of sketches. Previous sketch-to-pose methods, constrained by the lack of large-scale sketch-3D pose annotations, primarily relied on optimization with heuristic rules-an approach that is both time-consuming and limited in generalizability. To address these challenges, we propose a novel approach leveraging a \"learn from synthesis\" strategy. First, a diffusion model is trained to synthesize sketch images from 2D poses projected from 3D human poses, mimicking disproportionate human structures in sketches. This process enables the creation of a synthetic dataset, SKEP-120K, consisting of 120k accurate sketch-3D pose annotation pairs across various sketch styles. Building on this synthetic dataset, we introduce an end-to-end data-driven framework for estimating human poses and shapes from diverse sketch styles. Our framework combines existing 2D pose detectors and generative diffusion priors for sketch feature extraction with a feed-forward neural network for efficient 2D pose estimation. Multiple heuristic loss functions are incorporated to guarantee geometric coherence between the derived 3D poses and the detected 2D poses while preserving accurate self-contacts. Qualitative, quantitative, and subjective evaluations collectively show that our model substantially surpasses previous ones in both estimation accuracy and speed for sketch-to-pose tasks.", "date": "2025-10-30", "categories": []}
{"id": "2511.12030", "title": "VPHO: Joint Visual-Physical Cue Learning and Aggregation for Hand-Object Pose Estimation", "abstract": "Estimating the 3D poses of hands and objects from a single RGB image is a fundamental yet challenging problem, with broad applications in augmented reality and human-computer interaction. Existing methods largely rely on visual cues alone, often producing results that violate physical constraints such as interpenetration or non-contact. Recent efforts to incorporate physics reasoning typically depend on post-optimization or non-differentiable physics engines, which compromise visual consistency and end-to-end trainability. To overcome these limitations, we propose a novel framework that jointly integrates visual and physical cues for hand-object pose estimation. This integration is achieved through two key ideas: 1) joint visual-physical cue learning: The model is trained to extract 2D visual cues and 3D physical cues, thereby enabling more comprehensive representation learning for hand-object interactions; 2) candidate pose aggregation: A novel refinement process that aggregates multiple diffusion-generated candidate poses by leveraging both visual and physical predictions, yielding a final estimate that is visually consistent and physically plausible. Extensive experiments demonstrate that our method significantly outperforms existing state-of-the-art approaches in both pose accuracy and physical plausibility.", "date": "2025-11-15", "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": "2409.00449", "title": "LangPose: Language-Aligned Motion for Robust 3D Human Pose Estimation", "abstract": "2D-to-3D human pose lifting is an ill-posed problem due to depth ambiguity and occlusion. Existing methods relying on spatial and temporal consistency alone are insufficient to resolve these problems especially in the presence of significant occlusions or high dynamic actions. Semantic information, however, offers a complementary signal that can help disambiguate such cases. To this end, we propose LangPose, a framework that leverages action knowledge by aligning motion embeddings with text embeddings of fine-grained action labels. LangPose operates in two stages: pretraining and fine-tuning. In the pretraining stage, the model simultaneously learns to recognize actions and reconstruct 3D poses from masked and noisy 2D poses. During the fine-tuning stage, the model is further refined using real-world 3D human pose estimation datasets without action labels. Additionally, our framework incorporates masked body parts and masked time windows in motion modeling, encouraging the model to leverage semantic information when spatial and temporal consistency is unreliable. Experiments demonstrate the effectiveness of LangPose, achieving SOTA level performance in 3D pose estimation on public datasets, including Human3.6M and MPI-INF-3DHP. Specifically, LangPose achieves an MPJPE of 36.7mm on Human3.6M with detected 2D poses as input and 15.5mm on MPI-INF-3DHP with ground-truth 2D poses as input.", "date": "2024-08-31", "categories": []}
{"id": "2511.08809", "title": "Adaptive graph Kolmogorov-Arnold network for 3D human pose estimation", "abstract": "Graph convolutional network (GCN)-based methods have shown strong performance in 3D human pose estimation by leveraging the natural graph structure of the human skeleton. However, their local receptive field limits their ability to capture long-range dependencies essential for handling occlusions and depth ambiguities. They also exhibit spectral bias, which prioritizes low-frequency components while struggling to model high-frequency details. In this paper, we introduce PoseKAN, an adaptive graph Kolmogorov-Arnold Network (KAN), framework that extends KANs to graph-based learning for 2D-to-3D pose lifting from a single image. Unlike GCNs that use fixed activation functions, KANs employ learnable functions on graph edges, allowing data-driven, adaptive feature transformations. This enhances the model's adaptability and expressiveness, making it more expressive in learning complex pose variations. Our model employs multi-hop feature aggregation, ensuring the body joints can leverage information from both local and distant neighbors, leading to improved spatial awareness. It also incorporates residual PoseKAN blocks for deeper feature refinement, and a global response normalization for improved feature selectivity and contrast. Extensive experiments on benchmark datasets demonstrate the competitive performance of our model against state-of-the-art methods.", "date": "2025-11-11", "categories": []}
{"id": "2511.08872", "title": "SasMamba: A Lightweight Structure-Aware Stride State Space Model for 3D Human Pose Estimation", "abstract": "Recently, the Mamba architecture based on State Space Models (SSMs) has gained attention in 3D human pose estimation due to its linear complexity and strong global modeling capability. However, existing SSM-based methods typically apply manually designed scan operations to flatten detected 2D pose sequences into purely temporal sequences, either locally or globally. This approach disrupts the inherent spatial structure of human poses and entangles spatial and temporal features, making it difficult to capture complex pose dependencies. To address these limitations, we propose the Skeleton Structure-Aware Stride SSM (SAS-SSM), which first employs a structure-aware spatiotemporal convolution to dynamically capture essential local interactions between joints, and then applies a stride-based scan strategy to construct multi-scale global structural representations. This enables flexible modeling of both local and global pose information while maintaining linear computational complexity. Built upon SAS-SSM, our model SasMamba achieves competitive 3D pose estimation performance with significantly fewer parameters compared to existing hybrid models. The source code is available at https://hucui2022.github.io/sasmamba_proj/.", "date": "2025-11-12", "categories": []}
{"id": "2511.09502", "title": "DreamPose3D: Hallucinative Diffusion with Prompt Learning for 3D Human Pose Estimation", "abstract": "Accurate 3D human pose estimation remains a critical yet unresolved challenge, requiring both temporal coherence across frames and fine-grained modeling of joint relationships. However, most existing methods rely solely on geometric cues and predict each 3D pose independently, which limits their ability to resolve ambiguous motions and generalize to real-world scenarios. Inspired by how humans understand and anticipate motion, we introduce DreamPose3D, a diffusion-based framework that combines action-aware reasoning with temporal imagination for 3D pose estimation. DreamPose3D dynamically conditions the denoising process using task-relevant action prompts extracted from 2D pose sequences, capturing high-level intent. To model the structural relationships between joints effectively, we introduce a representation encoder that incorporates kinematic joint affinity into the attention mechanism. Finally, a hallucinative pose decoder predicts temporally coherent 3D pose sequences during training, simulating how humans mentally reconstruct motion trajectories to resolve ambiguity in perception. Extensive experiments on benchmarked Human3.6M and MPI-3DHP datasets demonstrate state-of-the-art performance across all metrics. To further validate DreamPose3D's robustness, we tested it on a broadcast baseball dataset, where it demonstrated strong performance despite ambiguous and noisy 2D inputs, effectively handling temporal consistency and intent-driven motion variations.", "date": "2025-11-12", "categories": []}
{"id": "2511.08387", "title": "RAPTR: Radar-based 3D Pose Estimation using Transformer", "abstract": "Radar-based indoor 3D human pose estimation typically relied on fine-grained 3D keypoint labels, which are costly to obtain especially in complex indoor settings involving clutter, occlusions, or multiple people. In this paper, we propose \\textbf{RAPTR} (RAdar Pose esTimation using tRansformer) under weak supervision, using only 3D BBox and 2D keypoint labels which are considerably easier and more scalable to collect. Our RAPTR is characterized by a two-stage pose decoder architecture with a pseudo-3D deformable attention to enhance (pose/joint) queries with multi-view radar features: a pose decoder estimates initial 3D poses with a 3D template loss designed to utilize the 3D BBox labels and mitigate depth ambiguities; and a joint decoder refines the initial poses with 2D keypoint labels and a 3D gravity loss. Evaluated on two indoor radar datasets, RAPTR outperforms existing methods, reducing joint position error by $34.3\\%$ on HIBER and $76.9\\%$ on MMVR. Our implementation is available at https://github.com/merlresearch/radar-pose-transformer.", "date": "2025-11-11", "categories": []}
{"id": "2408.17168", "title": "EMHI: A Multimodal Egocentric Human Motion Dataset with HMD and Body-Worn IMUs", "abstract": "Egocentric human pose estimation (HPE) using wearable sensors is essential for VR/AR applications. Most methods rely solely on either egocentric-view images or sparse Inertial Measurement Unit (IMU) signals, leading to inaccuracies due to self-occlusion in images or the sparseness and drift of inertial sensors. Most importantly, the lack of real-world datasets containing both modalities is a major obstacle to progress in this field. To overcome the barrier, we propose EMHI, a multimodal \\textbf{E}gocentric human \\textbf{M}otion dataset with \\textbf{H}ead-Mounted Display (HMD) and body-worn \\textbf{I}MUs, with all data collected under the real VR product suite. Specifically, EMHI provides synchronized stereo images from downward-sloping cameras on the headset and IMU data from body-worn sensors, along with pose annotations in SMPL format. This dataset consists of 885 sequences captured by 58 subjects performing 39 actions, totaling about 28.5 hours of recording. We evaluate the annotations by comparing them with optical marker-based SMPL fitting results. To substantiate the reliability of our dataset, we introduce MEPoser, a new baseline method for multimodal egocentric HPE, which employs a multimodal fusion encoder, temporal feature encoder, and MLP-based regression heads. The experiments on EMHI show that MEPoser outperforms existing single-modal methods and demonstrates the value of our dataset in solving the problem of egocentric HPE. We believe the release of EMHI and the method could advance the research of egocentric HPE and expedite the practical implementation of this technology in VR/AR products.", "date": "2024-08-30", "categories": []}
{"id": "2510.24464", "title": "Kineo: Calibration-Free Metric Motion Capture From Sparse RGB Cameras", "abstract": "Markerless multiview motion capture is often constrained by the need for precise camera calibration, limiting accessibility for non-experts and in-the-wild captures. Existing calibration-free approaches mitigate this requirement but suffer from high computational cost and reduced reconstruction accuracy.\n  We present Kineo, a fully automatic, calibration-free pipeline for markerless motion capture from videos captured by unsynchronized, uncalibrated, consumer-grade RGB cameras. Kineo leverages 2D keypoints from off-the-shelf detectors to simultaneously calibrate cameras, including Brown-Conrady distortion coefficients, and reconstruct 3D keypoints and dense scene point maps at metric scale. A confidence-driven spatio-temporal keypoint sampling strategy, combined with graph-based global optimization, ensures robust calibration at a fixed computational cost independent of sequence length. We further introduce a pairwise reprojection consensus score to quantify 3D reconstruction reliability for downstream tasks.\n  Evaluations on EgoHumans and Human3.6M demonstrate substantial improvements over prior calibration-free methods. Compared to previous state-of-the-art approaches, Kineo reduces camera translation error by approximately 83-85%, camera angular error by 86-92%, and world mean-per-joint error (W-MPJPE) by 83-91%.\n  Kineo is also efficient in real-world scenarios, processing multi-view sequences faster than their duration in specific configuration (e.g., 36min to process 1h20min of footage). The full pipeline and evaluation code are openly released to promote reproducibility and practical adoption at https://liris-xr.github.io/kineo/.", "date": "2025-10-28", "categories": []}
{"id": "2511.01756", "title": "HGFreNet: Hop-hybrid GraphFomer for 3D Human Pose Estimation with Trajectory Consistency in Frequency Domain", "abstract": "2D-to-3D human pose lifting is a fundamental challenge for 3D human pose estimation in monocular video, where graph convolutional networks (GCNs) and attention mechanisms have proven to be inherently suitable for encoding the spatial-temporal correlations of skeletal joints. However, depth ambiguity and errors in 2D pose estimation lead to incoherence in the 3D trajectory. Previous studies have attempted to restrict jitters in the time domain, for instance, by constraining the differences between adjacent frames while neglecting the global spatial-temporal correlations of skeletal joint motion. To tackle this problem, we design HGFreNet, a novel GraphFormer architecture with hop-hybrid feature aggregation and 3D trajectory consistency in the frequency domain. Specifically, we propose a hop-hybrid graph attention (HGA) module and a Transformer encoder to model global joint spatial-temporal correlations. The HGA module groups all $k$-hop neighbors of a skeletal joint into a hybrid group to enlarge the receptive field and applies the attention mechanism to discover the latent correlations of these groups globally. We then exploit global temporal correlations by constraining trajectory consistency in the frequency domain. To provide 3D information for depth inference across frames and maintain coherence over time, a preliminary network is applied to estimate the 3D pose. Extensive experiments were conducted on two standard benchmark datasets: Human3.6M and MPI-INF-3DHP. The results demonstrate that the proposed HGFreNet outperforms state-of-the-art (SOTA) methods in terms of positional accuracy and temporal consistency.", "date": "2025-11-03", "categories": []}
{"id": "2510.19078", "title": "UniHPR: Unified Human Pose Representation via Singular Value Contrastive Learning", "abstract": "In recent years, there has been a growing interest in developing effective alignment pipelines to generate unified representations from different modalities for multi-modal fusion and generation. As an important component of Human-Centric applications, Human Pose representations are critical in many downstream tasks, such as Human Pose Estimation, Action Recognition, Human-Computer Interaction, Object tracking, etc. Human Pose representations or embeddings can be extracted from images, 2D keypoints, 3D skeletons, mesh models, and lots of other modalities. Yet, there are limited instances where the correlation among all of those representations has been clearly researched using a contrastive paradigm. In this paper, we propose UniHPR, a unified Human Pose Representation learning pipeline, which aligns Human Pose embeddings from images, 2D and 3D human poses. To align more than two data representations at the same time, we propose a novel singular value-based contrastive learning loss, which better aligns different modalities and further boosts performance. To evaluate the effectiveness of the aligned representation, we choose 2D and 3D Human Pose Estimation (HPE) as our evaluation tasks. In our evaluation, with a simple 3D human pose decoder, UniHPR achieves remarkable performance metrics: MPJPE 49.9mm on the Human3.6M dataset and PA-MPJPE 51.6mm on the 3DPW dataset with cross-domain evaluation. Meanwhile, we are able to achieve 2D and 3D pose retrieval with our unified human pose representations in Human3.6M dataset, where the retrieval error is 9.24mm in MPJPE.", "date": "2025-10-21", "categories": []}
{"id": "2510.19475", "title": "PRGCN: A Graph Memory Network for Cross-Sequence Pattern Reuse in 3D Human Pose Estimation", "abstract": "Monocular 3D human pose estimation remains a fundamentally ill-posed inverse problem due to the inherent depth ambiguity in 2D-to-3D lifting. While contemporary video-based methods leverage temporal context to enhance spatial reasoning, they operate under a critical paradigm limitation: processing each sequence in isolation, thereby failing to exploit the strong structural regularities and repetitive motion patterns that pervade human movement across sequences. This work introduces the Pattern Reuse Graph Convolutional Network (PRGCN), a novel framework that formalizes pose estimation as a problem of pattern retrieval and adaptation. At its core, PRGCN features a graph memory bank that learns and stores a compact set of pose prototypes, encoded as relational graphs, which are dynamically retrieved via an attention mechanism to provide structured priors. These priors are adaptively fused with hard-coded anatomical constraints through a memory-driven graph convolution, ensuring geometrical plausibility. To underpin this retrieval process with robust spatiotemporal features, we design a dual-stream hybrid architecture that synergistically combines the linear-complexity, local temporal modeling of Mamba-based state-space models with the global relational capacity of self-attention. Extensive evaluations on Human3.6M and MPI-INF-3DHP benchmarks demonstrate that PRGCN establishes a new state-of-the-art, achieving an MPJPE of 37.1mm and 13.4mm, respectively, while exhibiting enhanced cross-domain generalization capability. Our work posits that the long-overlooked mechanism of cross-sequence pattern reuse is pivotal to advancing the field, shifting the paradigm from per-sequence optimization towards cumulative knowledge learning.", "date": "2025-10-22", "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": "2501.15763", "title": "NanoHTNet: Nano Human Topology Network for Efficient 3D Human Pose Estimation", "abstract": "The widespread application of 3D human pose estimation (HPE) is limited by resource-constrained edge devices, requiring more efficient models. A key approach to enhancing efficiency involves designing networks based on the structural characteristics of input data. However, effectively utilizing the structural priors in human skeletal inputs remains challenging. To address this, we leverage both explicit and implicit spatio-temporal priors of the human body through innovative model design and a pre-training proxy task. First, we propose a Nano Human Topology Network (NanoHTNet), a tiny 3D HPE network with stacked Hierarchical Mixers to capture explicit features. Specifically, the spatial Hierarchical Mixer efficiently learns the human physical topology across multiple semantic levels, while the temporal Hierarchical Mixer with discrete cosine transform and low-pass filtering captures local instantaneous movements and global action coherence. Moreover, Efficient Temporal-Spatial Tokenization (ETST) is introduced to enhance spatio-temporal interaction and reduce computational complexity significantly. Second, PoseCLR is proposed as a general pre-training method based on contrastive learning for 3D HPE, aimed at extracting implicit representations of human topology. By aligning 2D poses from diverse viewpoints in the proxy task, PoseCLR aids 3D HPE encoders like NanoHTNet in more effectively capturing the high-dimensional features of the human body, leading to further performance improvements. Extensive experiments verify that NanoHTNet with PoseCLR outperforms other state-of-the-art methods in efficiency, making it ideal for deployment on edge devices like the Jetson Nano. Code and models are available at https://github.com/vefalun/NanoHTNet.", "date": "2025-01-27", "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": "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": "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": "2510.11017", "title": "High-Resolution Spatiotemporal Modeling with Global-Local State Space Models for Video-Based Human Pose Estimation", "abstract": "Modeling high-resolution spatiotemporal representations, including both global dynamic contexts (e.g., holistic human motion tendencies) and local motion details (e.g., high-frequency changes of keypoints), is essential for video-based human pose estimation (VHPE). Current state-of-the-art methods typically unify spatiotemporal learning within a single type of modeling structure (convolution or attention-based blocks), which inherently have difficulties in balancing global and local dynamic modeling and may bias the network to one of them, leading to suboptimal performance. Moreover, existing VHPE models suffer from quadratic complexity when capturing global dependencies, limiting their applicability especially for high-resolution sequences. Recently, the state space models (known as Mamba) have demonstrated significant potential in modeling long-range contexts with linear complexity; however, they are restricted to 1D sequential data. In this paper, we present a novel framework that extends Mamba from two aspects to separately learn global and local high-resolution spatiotemporal representations for VHPE. Specifically, we first propose a Global Spatiotemporal Mamba, which performs 6D selective space-time scan and spatial- and temporal-modulated scan merging to efficiently extract global representations from high-resolution sequences. We further introduce a windowed space-time scan-based Local Refinement Mamba to enhance the high-frequency details of localized keypoint motions. Extensive experiments on four benchmark datasets demonstrate that the proposed model outperforms state-of-the-art VHPE approaches while achieving better computational trade-offs.", "date": "2025-10-13", "categories": []}
{"id": "2505.21309", "title": "Spectral Compression Transformer with Line Pose Graph for Monocular 3D Human Pose Estimation", "abstract": "Transformer-based 3D human pose estimation methods suffer from high computational costs due to the quadratic complexity of self-attention with respect to sequence length. Additionally, pose sequences often contain significant redundancy between frames. However, recent methods typically fail to improve model capacity while effectively eliminating sequence redundancy. In this work, we introduce the Spectral Compression Transformer (SCT) to reduce sequence length and accelerate computation. The SCT encoder treats hidden features between blocks as Temporal Feature Signals (TFS) and applies the Discrete Cosine Transform, a Fourier transform-based technique, to determine the spectral components to be retained. By filtering out certain high-frequency noise components, SCT compresses the sequence length and reduces redundancy. To further enrich the input sequence with prior structural information, we propose the Line Pose Graph (LPG) based on line graph theory. The LPG generates skeletal position information that complements the input 2D joint positions, thereby improving the model's performance. Finally, we design a dual-stream network architecture to effectively model spatial joint relationships and the compressed motion trajectory within the pose sequence. Extensive experiments on two benchmark datasets (i.e., Human3.6M and MPI-INF-3DHP) demonstrate that our model achieves state-of-the-art performance with improved computational efficiency. For example, on the Human3.6M dataset, our method achieves an MPJPE of 37.7mm while maintaining a low computational cost. Furthermore, we perform ablation studies on each module to assess its effectiveness. The code and models will be released.", "date": "2025-05-27", "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": "2510.02264", "title": "Paving the Way Towards Kinematic Assessment Using Monocular Video: A Preclinical Benchmark of State-of-the-Art Deep-Learning-Based 3D Human Pose Estimators Against Inertial Sensors in Daily Living Activities", "abstract": "Advances in machine learning and wearable sensors offer new opportunities for capturing and analyzing human movement outside specialized laboratories. Accurate assessment of human movement under real-world conditions is essential for telemedicine, sports science, and rehabilitation. This preclinical benchmark compares monocular video-based 3D human pose estimation models with inertial measurement units (IMUs), leveraging the VIDIMU dataset containing a total of 13 clinically relevant daily activities which were captured using both commodity video cameras and five IMUs. During this initial study only healthy subjects were recorded, so results cannot be generalized to pathological cohorts. Joint angles derived from state-of-the-art deep learning frameworks (MotionAGFormer, MotionBERT, MMPose 2D-to-3D pose lifting, and NVIDIA BodyTrack) were evaluated against joint angles computed from IMU data using OpenSim inverse kinematics following the Human3.6M dataset format with 17 keypoints. Among them, MotionAGFormer demonstrated superior performance, achieving the lowest overall RMSE ($9.27\\deg \\pm 4.80\\deg$) and MAE ($7.86\\deg \\pm 4.18\\deg$), as well as the highest Pearson correlation ($0.86 \\pm 0.15$) and the highest coefficient of determination $R^{2}$ ($0.67 \\pm 0.28$). The results reveal that both technologies are viable for out-of-the-lab kinematic assessment. However, they also highlight key trade-offs between video- and sensor-based approaches including costs, accessibility, and precision. This study clarifies where off-the-shelf video models already provide clinically promising kinematics in healthy adults and where they lag behind IMU-based estimates while establishing valuable guidelines for researchers and clinicians seeking to develop robust, cost-effective, and user-friendly solutions for telehealth and remote patient monitoring.", "date": "2025-10-02", "categories": []}
{"id": "2509.25739", "title": "LieHMR: Autoregressive Human Mesh Recovery with $SO(3)$ Diffusion", "abstract": "We tackle the problem of Human Mesh Recovery (HMR) from a single RGB image, formulating it as an image-conditioned human pose and shape generation. While recovering 3D human pose from 2D observations is inherently ambiguous, most existing approaches have regressed a single deterministic output. Probabilistic methods attempt to address this by generating multiple plausible outputs to model the ambiguity. However, these methods often exhibit a trade-off between accuracy and sample diversity, and their single predictions are not competitive with state-of-the-art deterministic models. To overcome these limitations, we propose a novel approach that models well-aligned distribution to 2D observations. In particular, we introduce $SO(3)$ diffusion model, which generates the distribution of pose parameters represented as 3D rotations unconditional and conditional to image observations via conditioning dropout. Our model learns the hierarchical structure of human body joints using the transformer. Instead of using transformer as a denoising model, the time-independent transformer extracts latent vectors for the joints and a small MLP-based denoising model learns the per-joint distribution conditioned on the latent vector. We experimentally demonstrate and analyze that our model predicts accurate pose probability distribution effectively.", "date": "2025-09-30", "categories": []}
{"id": "2509.23376", "title": "UniPose: Unified Cross-modality Pose Prior Propagation towards RGB-D data for Weakly Supervised 3D Human Pose Estimation", "abstract": "In this paper, we present UniPose, a unified cross-modality pose prior propagation method for weakly supervised 3D human pose estimation (HPE) using unannotated single-view RGB-D sequences (RGB, depth, and point cloud data). UniPose transfers 2D HPE annotations from large-scale RGB datasets (e.g., MS COCO) to the 3D domain via self-supervised learning on easily acquired RGB-D sequences, eliminating the need for labor-intensive 3D keypoint annotations. This approach bridges the gap between 2D and 3D domains without suffering from issues related to multi-view camera calibration or synthetic-to-real data shifts. During training, UniPose leverages off-the-shelf 2D pose estimations as weak supervision for point cloud networks, incorporating spatial-temporal constraints like body symmetry and joint motion. The 2D-to-3D back-projection loss and cross-modality interaction further enhance this process. By treating the point cloud network's 3D HPE results as pseudo ground truth, our anchor-to-joint prediction method performs 3D lifting on RGB and depth networks, making it more robust against inaccuracies in 2D HPE results compared to state-of-the-art methods. Experiments on CMU Panoptic and ITOP datasets show that UniPose achieves comparable performance to fully supervised methods. Incorporating large-scale unlabeled data (e.g., NTU RGB+D 60) enhances its performance under challenging conditions, demonstrating its potential for practical applications. Our proposed 3D lifting method also achieves state-of-the-art results.", "date": "2025-09-27", "categories": []}
{"id": "2509.23393", "title": "Generative Modeling of Shape-Dependent Self-Contact Human Poses", "abstract": "One can hardly model self-contact of human poses without considering underlying body shapes. For example, the pose of rubbing a belly for a person with a low BMI leads to penetration of the hand into the belly for a person with a high BMI. Despite its relevance, existing self-contact datasets lack the variety of self-contact poses and precise body shapes, limiting conclusive analysis between self-contact poses and shapes. To address this, we begin by introducing the first extensive self-contact dataset with precise body shape registration, Goliath-SC, consisting of 383K self-contact poses across 130 subjects. Using this dataset, we propose generative modeling of self-contact prior conditioned by body shape parameters, based on a body-part-wise latent diffusion with self-attention. We further incorporate this prior into single-view human pose estimation while refining estimated poses to be in contact. Our experiments suggest that shape conditioning is vital to the successful modeling of self-contact pose distribution, hence improving single-view pose estimation in self-contact.", "date": "2025-09-27", "categories": []}
{"id": "2509.23455", "title": "3DPCNet: Pose Canonicalization for Robust Viewpoint-Invariant 3D Kinematic Analysis from Monocular RGB cameras", "abstract": "Monocular 3D pose estimators produce camera-centered skeletons, creating view-dependent kinematic signals that complicate comparative analysis in applications such as health and sports science. We present 3DPCNet, a compact, estimator-agnostic module that operates directly on 3D joint coordinates to rectify any input pose into a consistent, body-centered canonical frame. Its hybrid encoder fuses local skeletal features from a graph convolutional network with global context from a transformer via a gated cross-attention mechanism. From this representation, the model predicts a continuous 6D rotation that is mapped to an $SO(3)$ matrix to align the pose. We train the model in a self-supervised manner on the MM-Fi dataset using synthetically rotated poses, guided by a composite loss ensuring both accurate rotation and pose reconstruction. On the MM-Fi benchmark, 3DPCNet reduces the mean rotation error from over 20$^{\\circ}$ to 3.4$^{\\circ}$ and the Mean Per Joint Position Error from ~64 mm to 47 mm compared to a geometric baseline. Qualitative evaluations on the TotalCapture dataset further demonstrate that our method produces acceleration signals from video that show strong visual correspondence to ground-truth IMU sensor data, confirming that our module removes viewpoint variability to enable physically plausible motion analysis.", "date": "2025-09-27", "categories": []}
{"id": "2505.20611", "title": "Mamba-Driven Topology Fusion for Monocular 3D Human Pose Estimation", "abstract": "Transformer-based methods for 3D human pose estimation face significant computational challenges due to the quadratic growth of self-attention mechanism complexity with sequence length. Recently, the Mamba model has substantially reduced computational overhead and demonstrated outstanding performance in modeling long sequences by leveraging state space model (SSM). However, the ability of SSM to process sequential data is not suitable for 3D joint sequences with topological structures, and the causal convolution structure in Mamba also lacks insight into local joint relationships. To address these issues, we propose the Mamba-Driven Topology Fusion framework in this paper. Specifically, the proposed Bone Aware Module infers the direction and length of bone vectors in the spherical coordinate system, providing effective topological guidance for the Mamba model in processing joint sequences. Furthermore, we enhance the convolutional structure within the Mamba model by integrating forward and backward graph convolutional network, enabling it to better capture local joint dependencies. Finally, we design a Spatiotemporal Refinement Module to model both temporal and spatial relationships within the sequence. Through the incorporation of skeletal topology, our approach effectively alleviates Mamba's limitations in capturing human structural relationships. We conduct extensive experiments on the Human3.6M and MPI-INF-3DHP datasets for testing and comparison, and the results show that the proposed method greatly reduces computational cost while achieving higher accuracy. Ablation studies further demonstrate the effectiveness of each proposed module. The code and models will be released.", "date": "2025-05-27", "categories": []}
{"id": "2409.03887", "title": "The Influence of Faulty Labels in Data Sets on Human Pose Estimation", "abstract": "In this study we provide empirical evidence demonstrating that the quality of training data impacts model performance in Human Pose Estimation (HPE). Inaccurate labels in widely used data sets, ranging from minor errors to severe mislabeling, can negatively influence learning and distort performance metrics. We perform an in-depth analysis of popular HPE data sets to show the extent and nature of label inaccuracies. Our findings suggest that accounting for the impact of faulty labels will facilitate the development of more robust and accurate HPE models for a variety of real-world applications. We show improved performance with cleansed data.", "date": "2024-09-05", "categories": []}
{"id": "2504.12699", "title": "Unsupervised Cross-Domain 3D Human Pose Estimation via Pseudo-Label-Guided Global Transforms", "abstract": "Existing 3D human pose estimation methods often suffer in performance, when applied to cross-scenario inference, due to domain shifts in characteristics such as camera viewpoint, position, posture, and body size. Among these factors, camera viewpoints and locations have been shown to contribute significantly to the domain gap by influencing the global positions of human poses. To address this, we propose a novel framework that explicitly conducts global transformations between pose positions in the camera coordinate systems of source and target domains. We start with a Pseudo-Label Generation Module that is applied to the 2D poses of the target dataset to generate pseudo-3D poses. Then, a Global Transformation Module leverages a human-centered coordinate system as a novel bridging mechanism to seamlessly align the positional orientations of poses across disparate domains, ensuring consistent spatial referencing. To further enhance generalization, a Pose Augmentor is incorporated to address variations in human posture and body size. This process is iterative, allowing refined pseudo-labels to progressively improve guidance for domain adaptation. Our method is evaluated on various cross-dataset benchmarks, including Human3.6M, MPI-INF-3DHP, and 3DPW. The proposed method outperforms state-of-the-art approaches and even outperforms the target-trained model.", "date": "2025-04-17", "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": "2505.10888", "title": "PoseBench3D: A Cross-Dataset Analysis Framework for 3D Human Pose Estimation via Pose Lifting Networks", "abstract": "Reliable three-dimensional human pose estimation (3D HPE) remains challenging due to the differences in viewpoints, environments, and camera conventions among datasets. As a result, methods that achieve near-optimal in-dataset accuracy often degrade on unseen datasets. In practice, however, systems must adapt to diverse viewpoints, environments, and camera setups--conditions that differ significantly from those encountered during training, which is often the case in real-world scenarios. Measuring cross-dataset performance is a vital process, but extremely labor-intensive when done manually for human pose estimation. To address these challenges, we automate this evaluation using PoseBench3D, a standardized testing framework that enables consistent and fair cross-dataset comparisons on previously unseen data. PoseBench3D streamlines testing across four widely used 3D HPE datasets via a single, configurable interface. Using this framework, we re-evaluate 18 methods and report over 100 cross-dataset results under Protocol 1: MPJPE and Protocol 2: PA-MPJPE, revealing systematic generalization gaps and the impact of common preprocessing and dataset setup choices. The PoseBench3D code is found at: https://github.com/bryanjvela/PoseBench3D", "date": "2025-05-16", "categories": []}
{"id": "2508.06757", "title": "VOccl3D: A Video Benchmark Dataset for 3D Human Pose and Shape Estimation under real Occlusions", "abstract": "Human pose and shape (HPS) estimation methods have been extensively studied, with many demonstrating high zero-shot performance on in-the-wild images and videos. However, these methods often struggle in challenging scenarios involving complex human poses or significant occlusions. Although some studies address 3D human pose estimation under occlusion, they typically evaluate performance on datasets that lack realistic or substantial occlusions, e.g., most existing datasets introduce occlusions with random patches over the human or clipart-style overlays, which may not reflect real-world challenges. To bridge this gap in realistic occlusion datasets, we introduce a novel benchmark dataset, VOccl3D, a Video-based human Occlusion dataset with 3D body pose and shape annotations. Inspired by works such as AGORA and BEDLAM, we constructed this dataset using advanced computer graphics rendering techniques, incorporating diverse real-world occlusion scenarios, clothing textures, and human motions. Additionally, we fine-tuned recent HPS methods, CLIFF and BEDLAM-CLIFF, on our dataset, demonstrating significant qualitative and quantitative improvements across multiple public datasets, as well as on the test split of our dataset, while comparing its performance with other state-of-the-art methods. Furthermore, we leveraged our dataset to enhance human detection performance under occlusion by fine-tuning an existing object detector, YOLO11, thus leading to a robust end-to-end HPS estimation system under occlusions. Overall, this dataset serves as a valuable resource for future research aimed at benchmarking methods designed to handle occlusions, offering a more realistic alternative to existing occlusion datasets. See the Project page for code and dataset:https://yashgarg98.github.io/VOccl3D-dataset/", "date": "2025-08-09", "categories": []}
{"id": "2509.07385", "title": "Parse Graph-Based Visual-Language Interaction for Human Pose Estimation", "abstract": "Parse graphs boost human pose estimation (HPE) by integrating context and hierarchies, yet prior work mostly focuses on single modality modeling, ignoring the potential of multimodal fusion. Notably, language offers rich HPE priors like spatial relations for occluded scenes, but existing visual-language fusion via global feature integration weakens occluded region responses and causes alignment and location failures. To address this issue, we propose Parse Graph-based Visual-Language interaction (PGVL) with a core novel Guided Module (GM). In PGVL, low-level nodes focus on local features, maximizing the maintenance of responses in occluded areas and high-level nodes integrate global features to infer occluded or invisible parts. GM enables high semantic nodes to guide the feature update of low semantic nodes that have undergone cross attention. It ensuring effective fusion of diverse information. PGVL includes top-down decomposition and bottom-up composition. In the first stage, modality specific parse graphs are constructed. Next stage. recursive bidirectional cross-attention is used, purified by GM. We also design network based on PGVL. The PGVL and our network is validated on major pose estimation datasets. We will release the code soon.", "date": "2025-09-09", "categories": []}
{"id": "2509.06607", "title": "From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans", "abstract": "Great progress has been made in estimating 3D human pose and shape from images and video by training neural networks to directly regress the parameters of parametric human models like SMPL. However, existing body models have simplified kinematic structures that do not correspond to the true joint locations and articulations in the human skeletal system, limiting their potential use in biomechanics. On the other hand, methods for estimating biomechanically accurate skeletal motion typically rely on complex motion capture systems and expensive optimization methods. What is needed is a parametric 3D human model with a biomechanically accurate skeletal structure that can be easily posed. To that end, we develop SKEL, which re-rigs the SMPL body model with a biomechanics skeleton. To enable this, we need training data of skeletons inside SMPL meshes in diverse poses.\n  We build such a dataset by optimizing biomechanically accurate skeletons inside SMPL meshes from AMASS sequences. We then learn a regressor from SMPL mesh vertices to the optimized joint locations and bone rotations. Finally, we re-parametrize the SMPL mesh with the new kinematic parameters. The resulting SKEL model is animatable like SMPL but with fewer, and biomechanically-realistic, degrees of freedom. We show that SKEL has more biomechanically accurate joint locations than SMPL, and the bones fit inside the body surface better than previous methods. By fitting SKEL to SMPL meshes we are able to \"upgrade\" existing human pose and shape datasets to include biomechanical parameters. SKEL provides a new tool to enable biomechanics in the wild, while also providing vision and graphics researchers with a better constrained and more realistic model of human articulation. The model, code, and data are available for research at https://skel.is.tue.mpg.de..", "date": "2025-09-08", "categories": []}
{"id": "2509.06956", "title": "H$_{2}$OT: Hierarchical Hourglass Tokenizer for Efficient Video Pose Transformers", "abstract": "Transformers have been successfully applied in the field of video-based 3D human pose estimation. However, the high computational costs of these video pose transformers (VPTs) make them impractical on resource-constrained devices. In this paper, we present a hierarchical plug-and-play pruning-and-recovering framework, called Hierarchical Hourglass Tokenizer (H$_{2}$OT), for efficient transformer-based 3D human pose estimation from videos. H$_{2}$OT begins with progressively pruning pose tokens of redundant frames and ends with recovering full-length sequences, resulting in a few pose tokens in the intermediate transformer blocks and thus improving the model efficiency. It works with two key modules, namely, a Token Pruning Module (TPM) and a Token Recovering Module (TRM). TPM dynamically selects a few representative tokens to eliminate the redundancy of video frames, while TRM restores the detailed spatio-temporal information based on the selected tokens, thereby expanding the network output to the original full-length temporal resolution for fast inference. Our method is general-purpose: it can be easily incorporated into common VPT models on both seq2seq and seq2frame pipelines while effectively accommodating different token pruning and recovery strategies. In addition, our H$_{2}$OT reveals that maintaining the full pose sequence is unnecessary, and a few pose tokens of representative frames can achieve both high efficiency and estimation accuracy. Extensive experiments on multiple benchmark datasets demonstrate both the effectiveness and efficiency of the proposed method. Code and models are available at https://github.com/NationalGAILab/HoT.", "date": "2025-09-08", "categories": []}
{"id": "2406.18453", "title": "Towards Human-Level 3D Relative Pose Estimation: Generalizable, Training-Free, with Single Reference", "abstract": "Humans can easily deduce the relative pose of a previously unseen object, without labeling or training, given only a single query-reference image pair. This is arguably achieved by incorporating i) 3D/2.5D shape perception from a single image, ii) render-and-compare simulation, and iii) rich semantic cue awareness to furnish (coarse) reference-query correspondence. Motivated by this, we propose a novel 3D generalizable relative pose estimation method by elaborating 3D/2.5D shape perception with a 2.5D shape from an RGB-D reference, fulfilling the render-and-compare paradigm with an off-the-shelf differentiable renderer, and leveraging the semantic cues from a pretrained model like DINOv2. Specifically, our differentiable renderer takes the 2.5D rotatable mesh textured by the RGB and the semantic maps (obtained by DINOv2 from the RGB input), then renders new RGB and semantic maps (with back-surface culling) under a novel rotated view. The refinement loss comes from comparing the rendered RGB and semantic maps with the query ones, back-propagating the gradients through the differentiable renderer to refine the 3D relative pose. As a result, \\emph{our method can be readily applied to unseen objects, given only a single RGB-D reference, without labeling or training}. Extensive experiments on LineMOD, LM-O, and YCB-V show that our training-free method significantly outperforms the state-of-the-art supervised methods, especially under the rigorous \\texttt{Acc@5/10/15}$^\\circ$ metrics and the challenging cross-dataset settings.", "date": "2024-06-26", "categories": []}
{"id": "2509.00649", "title": "MV-SSM: Multi-View State Space Modeling for 3D Human Pose Estimation", "abstract": "While significant progress has been made in single-view 3D human pose estimation, multi-view 3D human pose estimation remains challenging, particularly in terms of generalizing to new camera configurations. Existing attention-based transformers often struggle to accurately model the spatial arrangement of keypoints, especially in occluded scenarios. Additionally, they tend to overfit specific camera arrangements and visual scenes from training data, resulting in substantial performance drops in new settings. In this study, we introduce a novel Multi-View State Space Modeling framework, named MV-SSM, for robustly estimating 3D human keypoints. We explicitly model the joint spatial sequence at two distinct levels: the feature level from multi-view images and the person keypoint level. We propose a Projective State Space (PSS) block to learn a generalized representation of joint spatial arrangements using state space modeling. Moreover, we modify Mamba's traditional scanning into an effective Grid Token-guided Bidirectional Scanning (GTBS), which is integral to the PSS block. Multiple experiments demonstrate that MV-SSM achieves strong generalization, outperforming state-of-the-art methods: +10.8 on AP25 (+24%) on the challenging three-camera setting in CMU Panoptic, +7.0 on AP25 (+13%) on varying camera arrangements, and +15.3 PCP (+38%) on Campus A1 in cross-dataset evaluations. Project Website: https://aviralchharia.github.io/MV-SSM", "date": "2025-08-31", "categories": []}
{"id": "2509.01095", "title": "An End-to-End Framework for Video Multi-Person Pose Estimation", "abstract": "Video-based human pose estimation models aim to address scenarios that cannot be effectively solved by static image models such as motion blur, out-of-focus and occlusion. Most existing approaches consist of two stages: detecting human instances in each image frame and then using a temporal model for single-person pose estimation. This approach separates the spatial and temporal dimensions and cannot capture the global spatio-temporal context between spatial instances for end-to-end optimization. In addition, it relies on separate detectors and complex post-processing such as RoI cropping and NMS, which reduces the inference efficiency of the video scene. To address the above problems, we propose VEPE (Video End-to-End Pose Estimation), a simple and flexible framework for end-to-end pose estimation in video. The framework utilizes three crucial spatio-temporal Transformer components: the Spatio-Temporal Pose Encoder (STPE), the Spatio-Temporal Deformable Memory Encoder (STDME), and the Spatio-Temporal Pose Decoder (STPD). These components are designed to effectively utilize temporal context for optimizing human body pose estimation. Furthermore, to reduce the mismatch problem during the cross-frame pose query matching process, we propose an instance consistency mechanism, which aims to enhance the consistency and discrepancy of the cross-frame instance query and realize the instance tracking function, which in turn accurately guides the pose query to perform cross-frame matching. Extensive experiments on the Posetrack dataset show that our approach outperforms most two-stage models and improves inference efficiency by 300%.", "date": "2025-09-01", "categories": []}
{"id": "2508.21257", "title": "PHD: Personalized 3D Human Body Fitting with Point Diffusion", "abstract": "We introduce PHD, a novel approach for personalized 3D human mesh recovery (HMR) and body fitting that leverages user-specific shape information to improve pose estimation accuracy from videos. Traditional HMR methods are designed to be user-agnostic and optimized for generalization. While these methods often refine poses using constraints derived from the 2D image to improve alignment, this process compromises 3D accuracy by failing to jointly account for person-specific body shapes and the plausibility of 3D poses. In contrast, our pipeline decouples this process by first calibrating the user's body shape and then employing a personalized pose fitting process conditioned on that shape. To achieve this, we develop a body shape-conditioned 3D pose prior, implemented as a Point Diffusion Transformer, which iteratively guides the pose fitting via a Point Distillation Sampling loss. This learned 3D pose prior effectively mitigates errors arising from an over-reliance on 2D constraints. Consequently, our approach improves not only pelvis-aligned pose accuracy but also absolute pose accuracy -- an important metric often overlooked by prior work. Furthermore, our method is highly data-efficient, requiring only synthetic data for training, and serves as a versatile plug-and-play module that can be seamlessly integrated with existing 3D pose estimators to enhance their performance. Project page: https://phd-pose.github.io/", "date": "2025-08-28", "categories": []}
{"id": "2508.20920", "title": "COMETH: Convex Optimization for Multiview Estimation and Tracking of Humans", "abstract": "In the era of Industry 5.0, monitoring human activity is essential for ensuring both ergonomic safety and overall well-being. While multi-camera centralized setups improve pose estimation accuracy, they often suffer from high computational costs and bandwidth requirements, limiting scalability and real-time applicability. Distributing processing across edge devices can reduce network bandwidth and computational load. On the other hand, the constrained resources of edge devices lead to accuracy degradation, and the distribution of computation leads to temporal and spatial inconsistencies. We address this challenge by proposing COMETH (Convex Optimization for Multiview Estimation and Tracking of Humans), a lightweight algorithm for real-time multi-view human pose fusion that relies on three concepts: it integrates kinematic and biomechanical constraints to increase the joint positioning accuracy; it employs convex optimization-based inverse kinematics for spatial fusion; and it implements a state observer to improve temporal consistency. We evaluate COMETH on both public and industrial datasets, where it outperforms state-of-the-art methods in localization, detection, and tracking accuracy. The proposed fusion pipeline enables accurate and scalable human motion tracking, making it well-suited for industrial and safety-critical applications. The code is publicly available at https://github.com/PARCO-LAB/COMETH.", "date": "2025-08-28", "categories": []}
{"id": "2508.17239", "title": "PersPose: 3D Human Pose Estimation with Perspective Encoding and Perspective Rotation", "abstract": "Monocular 3D human pose estimation (HPE) methods estimate the 3D positions of joints from individual images. Existing 3D HPE approaches often use the cropped image alone as input for their models. However, the relative depths of joints cannot be accurately estimated from cropped images without the corresponding camera intrinsics, which determine the perspective relationship between 3D objects and the cropped images. In this work, we introduce Perspective Encoding (PE) to encode the camera intrinsics of the cropped images. Moreover, since the human subject can appear anywhere within the original image, the perspective relationship between the 3D scene and the cropped image differs significantly, which complicates model fitting. Additionally, the further the human subject deviates from the image center, the greater the perspective distortions in the cropped image. To address these issues, we propose Perspective Rotation (PR), a transformation applied to the original image that centers the human subject, thereby reducing perspective distortions and alleviating the difficulty of model fitting. By incorporating PE and PR, we propose a novel 3D HPE framework, PersPose. Experimental results demonstrate that PersPose achieves state-of-the-art (SOTA) performance on the 3DPW, MPI-INF-3DHP, and Human3.6M datasets. For example, on the in-the-wild dataset 3DPW, PersPose achieves an MPJPE of 60.1 mm, 7.54% lower than the previous SOTA approach. Code is available at: https://github.com/KenAdamsJoseph/PersPose.", "date": "2025-08-24", "categories": []}
{"id": "2508.18238", "title": "PriorFormer: A Transformer for Real-time Monocular 3D Human Pose Estimation with Versatile Geometric Priors", "abstract": "This paper proposes a new lightweight Transformer-based lifter that maps short sequences of human 2D joint positions to 3D poses using a single camera. The proposed model takes as input geometric priors including segment lengths and camera intrinsics and is designed to operate in both calibrated and uncalibrated settings. To this end, a masking mechanism enables the model to ignore missing priors during training and inference. This yields a single versatile network that can adapt to different deployment scenarios, from fully calibrated lab environments to in-the-wild monocular videos without calibration.  The model was trained using 3D keypoints from AMASS dataset with corresponding 2D synthetic data generated by sampling random camera poses and intrinsics. It was then compared to an expert model trained, only on complete priors, and the validation was done by conducting an ablation study. Results show that both, camera and segment length priors, improve performance and that the versatile model outperforms the expert, even when all priors are available, and maintains high accuracy when priors are missing. Overall the average 3D joint center positions estimation accuracy was as low as 36mm improving state of the art by half a centimeter and at a much lower computational cost. Indeed, the proposed model runs in 380$\\mu$s on GPU and 1800$\\mu$s on CPU, making it suitable for deployment on embedded platforms and low-power devices.", "date": "2025-08-21", "categories": []}
{"id": "2503.11652", "title": "Bring Your Rear Cameras for Egocentric 3D Human Pose Estimation", "abstract": "Egocentric 3D human pose estimation has been actively studied using cameras installed in front of a head-mounted device (HMD). While frontal placement is the optimal and the only option for some tasks, such as hand tracking, it remains unclear if the same holds for full-body tracking due to self-occlusion and limited field-of-view coverage. Notably, even the state-of-the-art methods often fail to estimate accurate 3D poses in many scenarios, such as when HMD users tilt their heads upward -- a common motion in human activities. A key limitation of existing HMD designs is their neglect of the back of the body, despite its potential to provide crucial 3D reconstruction cues. Hence, this paper investigates the usefulness of rear cameras for full-body tracking. We also show that simply adding rear views to the frontal inputs is not optimal for existing methods due to their dependence on individual 2D joint detectors without effective multi-view integration. To address this issue, we propose a new transformer-based method that refines 2D joint heatmap estimation with multi-view information and heatmap uncertainty, thereby improving 3D pose tracking. Also, we introduce two new large-scale datasets, Ego4View-Syn and Ego4View-RW, for a rear-view evaluation. Our experiments show that the new camera configurations with back views provide superior support for 3D pose tracking compared to only frontal placements. The proposed method achieves significant improvement over the current state of the art (>10% on MPJPE). The source code, trained models, and datasets are available on our project page at https://4dqv.mpi-inf.mpg.de/EgoRear/.", "date": "2025-03-14", "categories": []}
{"id": "2508.14431", "title": "HyperDiff: Hypergraph Guided Diffusion Model for 3D Human Pose Estimation", "abstract": "Monocular 3D human pose estimation (HPE) often encounters challenges such as depth ambiguity and occlusion during the 2D-to-3D lifting process. Additionally, traditional methods may overlook multi-scale skeleton features when utilizing skeleton structure information, which can negatively impact the accuracy of pose estimation. To address these challenges, this paper introduces a novel 3D pose estimation method, HyperDiff, which integrates diffusion models with HyperGCN. The diffusion model effectively captures data uncertainty, alleviating depth ambiguity and occlusion. Meanwhile, HyperGCN, serving as a denoiser, employs multi-granularity structures to accurately model high-order correlations between joints. This improves the model's denoising capability especially for complex poses. Experimental results demonstrate that HyperDiff achieves state-of-the-art performance on the Human3.6M and MPI-INF-3DHP datasets and can flexibly adapt to varying computational resources to balance performance and efficiency.", "date": "2025-08-20", "categories": []}
{"id": "2407.10099", "title": "STGFormer: Spatio-Temporal GraphFormer for 3D Human Pose Estimation in Video", "abstract": "The current methods of video-based 3D human pose estimation have achieved significant progress.However, they still face pressing challenges, such as the underutilization of spatiotemporal bodystructure features in transformers and the inadequate granularity of spatiotemporal interaction modeling in graph convolutional networks, which leads to pervasive depth ambiguity in monocular 3D human pose estimation. To address these limitations, this paper presents the Spatio-Temporal GraphFormer framework (STGFormer) for 3D human pose estimation in videos. First, we introduce a Spatio-Temporal criss-cross Graph (STG) attention mechanism, designed to more effectively leverage the inherent graph priors of the human body within continuous sequence distributions while capturing spatiotemporal long-range dependencies. Next, we present a dual-path Modulated Hop-wise Regular GCN (MHR-GCN) to independently process temporal and spatial dimensions in parallel, preserving features rich in temporal dynamics and the original or high-dimensional representations of spatial structures. Furthermore, the module leverages modulation to optimize parameter efficiency and incorporates spatiotemporal hop-wise skip connections to capture higher-order information. Finally, we demonstrate that our method achieves state-of-the-art performance on the Human3.6M and MPIINF-3DHP datasets.", "date": "2024-07-14", "categories": []}
{"id": "2508.13562", "title": "Learnable SMPLify: A Neural Solution for Optimization-Free Human Pose Inverse Kinematics", "abstract": "In 3D human pose and shape estimation, SMPLify remains a robust baseline that solves inverse kinematics (IK) through iterative optimization. However, its high computational cost limits its practicality. Recent advances across domains have shown that replacing iterative optimization with data-driven neural networks can achieve significant runtime improvements without sacrificing accuracy. Motivated by this trend, we propose Learnable SMPLify, a neural framework that replaces the iterative fitting process in SMPLify with a single-pass regression model. The design of our framework targets two core challenges in neural IK: data construction and generalization. To enable effective training, we propose a temporal sampling strategy that constructs initialization-target pairs from sequential frames. To improve generalization across diverse motions and unseen poses, we propose a human-centric normalization scheme and residual learning to narrow the solution space. Learnable SMPLify supports both sequential inference and plug-in post-processing to refine existing image-based estimators. Extensive experiments demonstrate that our method establishes itself as a practical and simple baseline: it achieves nearly 200x faster runtime compared to SMPLify, generalizes well to unseen 3DPW and RICH, and operates in a model-agnostic manner when used as a plug-in tool on LucidAction. The code is available at https://github.com/Charrrrrlie/Learnable-SMPLify.", "date": "2025-08-19", "categories": []}
{"id": "2508.12644", "title": "DyCrowd: Towards Dynamic Crowd Reconstruction from a Large-scene Video", "abstract": "3D reconstruction of dynamic crowds in large scenes has become increasingly important for applications such as city surveillance and crowd analysis. However, current works attempt to reconstruct 3D crowds from a static image, causing a lack of temporal consistency and inability to alleviate the typical impact caused by occlusions. In this paper, we propose DyCrowd, the first framework for spatio-temporally consistent 3D reconstruction of hundreds of individuals' poses, positions and shapes from a large-scene video. We design a coarse-to-fine group-guided motion optimization strategy for occlusion-robust crowd reconstruction in large scenes. To address temporal instability and severe occlusions, we further incorporate a VAE (Variational Autoencoder)-based human motion prior along with a segment-level group-guided optimization. The core of our strategy leverages collective crowd behavior to address long-term dynamic occlusions. By jointly optimizing the motion sequences of individuals with similar motion segments and combining this with the proposed Asynchronous Motion Consistency (AMC) loss, we enable high-quality unoccluded motion segments to guide the motion recovery of occluded ones, ensuring robust and plausible motion recovery even in the presence of temporal desynchronization and rhythmic inconsistencies. Additionally, in order to fill the gap of no existing well-annotated large-scene video dataset, we contribute a virtual benchmark dataset, VirtualCrowd, for evaluating dynamic crowd reconstruction from large-scene videos. Experimental results demonstrate that the proposed method achieves state-of-the-art performance in the large-scene dynamic crowd reconstruction task. The code and dataset will be available for research purposes.", "date": "2025-08-18", "categories": []}
{"id": "2410.03174", "title": "Efficient High-Resolution Visual Representation Learning with State Space Model for Human Pose Estimation", "abstract": "Capturing long-range dependencies while preserving high-resolution visual representations is crucial for dense prediction tasks such as human pose estimation. Vision Transformers (ViTs) have advanced global modeling through self-attention but suffer from quadratic computational complexity with respect to token count, limiting their efficiency and scalability to high-resolution inputs, especially on mobile and resource-constrained devices. State Space Models (SSMs), exemplified by Mamba, offer an efficient alternative by combining global receptive fields with linear computational complexity, enabling scalable and resource-friendly sequence modeling. However, when applied to dense prediction tasks, existing visual SSMs face key limitations: weak spatial inductive bias, long-range forgetting from hidden state decay, and low-resolution outputs that hinder fine-grained localization. To address these issues, we propose the Dynamic Visual State Space (DVSS) block, which augments visual state space models with multi-scale convolutional operations to enhance local spatial representations and strengthen spatial inductive biases. Through architectural exploration and theoretical analysis, we incorporate deformable operation into the DVSS block, identifying it as an efficient and effective mechanism to enhance semantic aggregation and mitigate long-range forgetting via input-dependent, adaptive spatial sampling. We embed DVSS into a multi-branch high-resolution architecture to build HRVMamba, a novel model for efficient high-resolution representation learning. Extensive experiments on human pose estimation, image classification, and semantic segmentation show that HRVMamba performs competitively against leading CNN-, ViT-, and SSM-based baselines. Code is available at https://github.com/zhanghao5201/PoseVMamba.", "date": "2024-10-04", "categories": []}
{"id": "2508.10897", "title": "Human-in-Context: Unified Cross-Domain 3D Human Motion Modeling via In-Context Learning", "abstract": "This paper aims to model 3D human motion across domains, where a single model is expected to handle multiple modalities, tasks, and datasets. Existing cross-domain models often rely on domain-specific components and multi-stage training, which limits their practicality and scalability. To overcome these challenges, we propose a new setting to train a unified cross-domain model through a single process, eliminating the need for domain-specific components and multi-stage training. We first introduce Pose-in-Context (PiC), which leverages in-context learning to create a pose-centric cross-domain model. While PiC generalizes across multiple pose-based tasks and datasets, it encounters difficulties with modality diversity, prompting strategy, and contextual dependency handling. We thus propose Human-in-Context (HiC), an extension of PiC that broadens generalization across modalities, tasks, and datasets. HiC combines pose and mesh representations within a unified framework, expands task coverage, and incorporates larger-scale datasets. Additionally, HiC introduces a max-min similarity prompt sampling strategy to enhance generalization across diverse domains and a network architecture with dual-branch context injection for improved handling of contextual dependencies. Extensive experimental results show that HiC performs better than PiC in terms of generalization, data scale, and performance across a wide range of domains. These results demonstrate the potential of HiC for building a unified cross-domain 3D human motion model with improved flexibility and scalability. The source codes and models are available at https://github.com/BradleyWang0416/Human-in-Context.", "date": "2025-08-14", "categories": []}
{"id": "2506.15290", "title": "Human Motion Capture from Loose and Sparse Inertial Sensors with Garment-aware Diffusion Models", "abstract": "Motion capture using sparse inertial sensors has shown great promise due to its portability and lack of occlusion issues compared to camera-based tracking. Existing approaches typically assume that IMU sensors are tightly attached to the human body. However, this assumption often does not hold in real-world scenarios. In this paper, we present Garment Inertial Poser (GaIP), a method for estimating full-body poses from sparse and loosely attached IMU sensors. We first simulate IMU recordings using an existing garment-aware human motion dataset. Our transformer-based diffusion models synthesize loose IMU data and estimate human poses from this challenging loose IMU data. We also demonstrate that incorporating garment-related parameters during training on loose IMU data effectively maintains expressiveness and enhances the ability to capture variations introduced by looser or tighter garments. Our experiments show that our diffusion methods trained on simulated and synthetic data outperform state-of-the-art inertial full-body pose estimators, both quantitatively and qualitatively, opening up a promising direction for future research on motion capture from such realistic sensor placements.", "date": "2025-06-18", "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": "2508.02056", "title": "StarPose: 3D Human Pose Estimation via Spatial-Temporal Autoregressive Diffusion", "abstract": "Monocular 3D human pose estimation remains a challenging task due to inherent depth ambiguities and occlusions. Compared to traditional methods based on Transformers or Convolutional Neural Networks (CNNs), recent diffusion-based approaches have shown superior performance, leveraging their probabilistic nature and high-fidelity generation capabilities. However, these methods often fail to account for the spatial and temporal correlations across predicted frames, resulting in limited temporal consistency and inferior accuracy in predicted 3D pose sequences. To address these shortcomings, this paper proposes StarPose, an autoregressive diffusion framework that effectively incorporates historical 3D pose predictions and spatial-temporal physical guidance to significantly enhance both the accuracy and temporal coherence of pose predictions. Unlike existing approaches, StarPose models the 2D-to-3D pose mapping as an autoregressive diffusion process. By synergically integrating previously predicted 3D poses with 2D pose inputs via a Historical Pose Integration Module (HPIM), the framework generates rich and informative historical pose embeddings that guide subsequent denoising steps, ensuring temporally consistent predictions. In addition, a fully plug-and-play Spatial-Temporal Physical Guidance (STPG) mechanism is tailored to refine the denoising process in an iterative manner, which further enforces spatial anatomical plausibility and temporal motion dynamics, rendering robust and realistic pose estimates. Extensive experiments on benchmark datasets demonstrate that StarPose outperforms state-of-the-art methods, achieving superior accuracy and temporal consistency in 3D human pose estimation. Code is available at https://github.com/wileychan/StarPose.", "date": "2025-08-04", "categories": []}
{"id": "2506.01691", "title": "SteerPose: Simultaneous Extrinsic Camera Calibration and Matching from Articulation", "abstract": "Can freely moving humans or animals themselves serve as calibration targets for multi-camera systems while simultaneously estimating their correspondences across views? We humans can solve this problem by mentally rotating the observed 2D poses and aligning them with those in the target views. Inspired by this cognitive ability, we propose SteerPose, a neural network that performs this rotation of 2D poses into another view. By integrating differentiable matching, SteerPose simultaneously performs extrinsic camera calibration and correspondence search within a single unified framework. We also introduce a novel geometric consistency loss that explicitly ensures that the estimated rotation and correspondences result in a valid translation estimation. Experimental results on diverse in-the-wild datasets of humans and animals validate the effectiveness and robustness of the proposed method. Furthermore, we demonstrate that our method can reconstruct the 3D poses of novel animals in multi-camera setups by leveraging off-the-shelf 2D pose estimators and our class-agnostic model.", "date": "2025-06-02", "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": "2508.00599", "title": "DPoser-X: Diffusion Model as Robust 3D Whole-body Human Pose Prior", "abstract": "We present DPoser-X, a diffusion-based prior model for 3D whole-body human poses. Building a versatile and robust full-body human pose prior remains challenging due to the inherent complexity of articulated human poses and the scarcity of high-quality whole-body pose datasets. To address these limitations, we introduce a Diffusion model as body Pose prior (DPoser) and extend it to DPoser-X for expressive whole-body human pose modeling. Our approach unifies various pose-centric tasks as inverse problems, solving them through variational diffusion sampling. To enhance performance on downstream applications, we introduce a novel truncated timestep scheduling method specifically designed for pose data characteristics. We also propose a masked training mechanism that effectively combines whole-body and part-specific datasets, enabling our model to capture interdependencies between body parts while avoiding overfitting to specific actions. Extensive experiments demonstrate DPoser-X's robustness and versatility across multiple benchmarks for body, hand, face, and full-body pose modeling. Our model consistently outperforms state-of-the-art alternatives, establishing a new benchmark for whole-body human pose prior modeling.", "date": "2025-08-01", "categories": []}
{"id": "2412.01562", "title": "Detection, Pose Estimation and Segmentation for Multiple Bodies: Closing the Virtuous Circle", "abstract": "Human pose estimation methods work well on isolated people but struggle with multiple-bodies-in-proximity scenarios. Previous work has addressed this problem by conditioning pose estimation by detected bounding boxes or keypoints, but overlooked instance masks. We propose to iteratively enforce mutual consistency of bounding boxes, instance masks, and poses. The introduced BBox-Mask-Pose (BMP) method uses three specialized models that improve each other's output in a closed loop. All models are adapted for mutual conditioning, which improves robustness in multi-body scenes. MaskPose, a new mask-conditioned pose estimation model, is the best among top-down approaches on OCHuman. BBox-Mask-Pose pushes SOTA on OCHuman dataset in all three tasks - detection, instance segmentation, and pose estimation. It also achieves SOTA performance on COCO pose estimation. The method is especially good in scenes with large instances overlap, where it improves detection by 39% over the baseline detector. With small specialized models and faster runtime, BMP is an effective alternative to large human-centered foundational models. Code and models are available on https://MiraPurkrabek.github.io/BBox-Mask-Pose.", "date": "2024-12-02", "categories": []}
{"id": "2402.02339", "title": "Uncertainty-Aware Testing-Time Optimization for 3D Human Pose Estimation", "abstract": "Although data-driven methods have achieved success in 3D human pose estimation, they often suffer from domain gaps and exhibit limited generalization. In contrast, optimization-based methods excel in fine-tuning for specific cases but are generally inferior to data-driven methods in overall performance. We observe that previous optimization-based methods commonly rely on a projection constraint, which only ensures alignment in 2D space, potentially leading to the overfitting problem. To address this, we propose an Uncertainty-Aware testing-time Optimization (UAO) framework, which keeps the prior information of the pre-trained model and alleviates the overfitting problem using the uncertainty of joints. Specifically, during the training phase, we design an effective 2D-to-3D network for estimating the corresponding 3D pose while quantifying the uncertainty of each 3D joint. For optimization during testing, the proposed optimization framework freezes the pre-trained model and optimizes only a latent state. Projection loss is then employed to ensure the generated poses are well aligned in 2D space for high-quality optimization. Furthermore, we utilize the uncertainty of each joint to determine how much each joint is allowed for optimization. The effectiveness and superiority of the proposed framework are validated through extensive experiments on challenging datasets: Human3.6M, MPI-INF-3DHP, and 3DPW. Notably, our approach outperforms the previous best result by a large margin of 5.5\\% on Human3.6M. Code is available at \\href{https://github.com/xiu-cs/UAO-Pose3D}{https://github.com/xiu-cs/UAO-Pose3D}.", "date": "2024-02-04", "categories": []}
{"id": "2507.19852", "title": "A Structure-aware and Motion-adaptive Framework for 3D Human Pose Estimation with Mamba", "abstract": "Recent Mamba-based methods for the pose-lifting task tend to model joint dependencies by 2D-to-1D mapping with diverse scanning strategies. Though effective, they struggle to model intricate joint connections and uniformly process all joint motion trajectories while neglecting the intrinsic differences across motion characteristics. In this work, we propose a structure-aware and motion-adaptive framework to capture spatial joint topology along with diverse motion dynamics independently, named as SAMA. Specifically, SAMA consists of a Structure-aware State Integrator (SSI) and a Motion-adaptive State Modulator (MSM). The Structure-aware State Integrator is tasked with leveraging dynamic joint relationships to fuse information at both the joint feature and state levels in the state space, based on pose topology rather than sequential state transitions. The Motion-adaptive State Modulator is responsible for joint-specific motion characteristics recognition, thus applying tailored adjustments to diverse motion patterns across different joints. Through the above key modules, our algorithm enables structure-aware and motion-adaptive pose lifting. Extensive experiments across multiple benchmarks demonstrate that our algorithm achieves advanced results with fewer computational costs.", "date": "2025-07-26", "categories": []}
{"id": "2507.20763", "title": "KASportsFormer: Kinematic Anatomy Enhanced Transformer for 3D Human Pose Estimation on Short Sports Scene Video", "abstract": "Recent transformer based approaches have demonstrated impressive performance in solving real-world 3D human pose estimation problems. Albeit these approaches achieve fruitful results on benchmark datasets, they tend to fall short of sports scenarios where human movements are more complicated than daily life actions, as being hindered by motion blur, occlusions, and domain shifts. Moreover, due to the fact that critical motions in a sports game often finish in moments of time (e.g., shooting), the ability to focus on momentary actions is becoming a crucial factor in sports analysis, where current methods appear to struggle with instantaneous scenarios. To overcome these limitations, we introduce KASportsFormer, a novel transformer based 3D pose estimation framework for sports that incorporates a kinematic anatomy-informed feature representation and integration module. In which the inherent kinematic motion information is extracted with the Bone Extractor (BoneExt) and Limb Fuser (LimbFus) modules and encoded in a multimodal manner. This improved the capability of comprehending sports poses in short videos. We evaluate our method through two representative sports scene datasets: SportsPose and WorldPose. Experimental results show that our proposed method achieves state-of-the-art results with MPJPE errors of 58.0mm and 34.3mm, respectively. Our code and models are available at: https://github.com/jw0r1n/KASportsFormer", "date": "2025-07-28", "categories": []}
{"id": "2507.16850", "title": "Toward a Real-Time Framework for Accurate Monocular 3D Human Pose Estimation with Geometric Priors", "abstract": "Monocular 3D human pose estimation remains a challenging and ill-posed problem, particularly in real-time settings and unconstrained environments. While direct imageto-3D approaches require large annotated datasets and heavy models, 2D-to-3D lifting offers a more lightweight and flexible alternative-especially when enhanced with prior knowledge. In this work, we propose a framework that combines real-time 2D keypoint detection with geometry-aware 2D-to-3D lifting, explicitly leveraging known camera intrinsics and subject-specific anatomical priors. Our approach builds on recent advances in self-calibration and biomechanically-constrained inverse kinematics to generate large-scale, plausible 2D-3D training pairs from MoCap and synthetic datasets. We discuss how these ingredients can enable fast, personalized, and accurate 3D pose estimation from monocular images without requiring specialized hardware. This proposal aims to foster discussion on bridging data-driven learning and model-based priors to improve accuracy, interpretability, and deployability of 3D human motion capture on edge devices in the wild.", "date": "2025-07-21", "categories": []}
{"id": "2507.17406", "title": "Physics-based Human Pose Estimation from a Single Moving RGB Camera", "abstract": "Most monocular and physics-based human pose tracking methods, while achieving state-of-the-art results, suffer from artifacts when the scene does not have a strictly flat ground plane or when the camera is moving. Moreover, these methods are often evaluated on in-the-wild real world videos without ground-truth data or on synthetic datasets, which fail to model the real world light transport, camera motion, and pose-induced appearance and geometry changes. To tackle these two problems, we introduce MoviCam, the first non-synthetic dataset containing ground-truth camera trajectories of a dynamically moving monocular RGB camera, scene geometry, and 3D human motion with human-scene contact labels. Additionally, we propose PhysDynPose, a physics-based method that incorporates scene geometry and physical constraints for more accurate human motion tracking in case of camera motion and non-flat scenes. More precisely, we use a state-of-the-art kinematics estimator to obtain the human pose and a robust SLAM method to capture the dynamic camera trajectory, enabling the recovery of the human pose in the world frame. We then refine the kinematic pose estimate using our scene-aware physics optimizer. From our new benchmark, we found that even state-of-the-art methods struggle with this inherently challenging setting, i.e. a moving camera and non-planar environments, while our method robustly estimates both human and camera poses in world coordinates.", "date": "2025-07-23", "categories": []}
{"id": "2507.12905", "title": "AthleticsPose: Authentic Sports Motion Dataset on Athletic Field and Evaluation of Monocular 3D Pose Estimation Ability", "abstract": "Monocular 3D pose estimation is a promising, flexible alternative to costly motion capture systems for sports analysis. However, its practical application is hindered by two factors: a lack of realistic sports datasets and unclear reliability for sports tasks. To address these challenges, we introduce the AthleticsPose dataset, a new public dataset featuring ``real'' motions captured from 23 athletes performing various athletics events on an athletic field. Using this dataset, we trained a representative 3D pose estimation model and performed a comprehensive evaluation. Our results show that the model trained on AthleticsPose significantly outperforms a baseline model trained on an imitated sports motion dataset, reducing MPJPE by approximately 75 %. These results show the importance of training on authentic sports motion data, as models based on imitated motions do not effectively transfer to real-world motions. Further analysis reveals that estimation accuracy is sensitive to camera view and subject scale. In case studies of kinematic indicators, the model demonstrated the potential to capture individual differences in knee angles but struggled with higher-speed metrics, such as knee-drive velocity, due to prediction biases. This work provides the research community with a valuable dataset and clarifies the potential and practical limitations of using monocular 3D pose estimation for sports motion analysis. Our dataset, code, and checkpoints are available at https://github.com/SZucchini/AthleticsPose.", "date": "2025-07-17", "categories": []}
{"id": "2503.07499", "title": "AthletePose3D: A Benchmark Dataset for 3D Human Pose Estimation and Kinematic Validation in Athletic Movements", "abstract": "Human pose estimation is a critical task in computer vision and sports biomechanics, with applications spanning sports science, rehabilitation, and biomechanical research. While significant progress has been made in monocular 3D pose estimation, current datasets often fail to capture the complex, high-acceleration movements typical of competitive sports. In this work, we introduce AthletePose3D, a novel dataset designed to address this gap. AthletePose3D includes 12 types of sports motions across various disciplines, with approximately 1.3 million frames and 165 thousand individual postures, specifically capturing high-speed, high-acceleration athletic movements. We evaluate state-of-the-art (SOTA) monocular 2D and 3D pose estimation models on the dataset, revealing that models trained on conventional datasets perform poorly on athletic motions. However, fine-tuning these models on AthletePose3D notably reduces the SOTA model mean per joint position error (MPJPE) from 214mm to 65mm-a reduction of over 69%. We also validate the kinematic accuracy of monocular pose estimations through waveform analysis, highlighting strong correlations in joint angle estimations but limitations in velocity estimation. Our work provides a comprehensive evaluation of monocular pose estimation models in the context of sports, contributing valuable insights for advancing monocular pose estimation techniques in high-performance sports environments. The dataset, code, and model checkpoints are available at: https://github.com/calvinyeungck/AthletePose3D", "date": "2025-03-10", "categories": []}
{"id": "2503.14097", "title": "SCJD: Sparse Correlation and Joint Distillation for Efficient 3D Human Pose Estimation", "abstract": "Existing 3D Human Pose Estimation (HPE) methods achieve high accuracy but suffer from computational overhead and slow inference, while knowledge distillation methods fail to address spatial relationships between joints and temporal correlations in multi-frame inputs. In this paper, we propose Sparse Correlation and Joint Distillation (SCJD), a novel framework that balances efficiency and accuracy for 3D HPE. SCJD introduces Sparse Correlation Input Sequence Downsampling to reduce redundancy in student network inputs while preserving inter-frame correlations. For effective knowledge transfer, we propose Dynamic Joint Spatial Attention Distillation, which includes Dynamic Joint Embedding Distillation to enhance the student's feature representation using the teacher's multi-frame context feature, and Adjacent Joint Attention Distillation to improve the student network's focus on adjacent joint relationships for better spatial understanding. Additionally, Temporal Consistency Distillation aligns the temporal correlations between teacher and student networks through upsampling and global supervision. Extensive experiments demonstrate that SCJD achieves state-of-the-art performance. Code is available at https://github.com/wileychan/SCJD.", "date": "2025-03-18", "categories": []}
{"id": "2504.08175", "title": "Multi-person Physics-based Pose Estimation for Combat Sports", "abstract": "We propose a novel framework for accurate 3D human pose estimation in combat sports using sparse multi-camera setups. Our method integrates robust multi-view 2D pose tracking via a transformer-based top-down approach, employing epipolar geometry constraints and long-term video object segmentation for consistent identity tracking across views. Initial 3D poses are obtained through weighted triangulation and spline smoothing, followed by kinematic optimization to refine pose accuracy. We further enhance pose realism and robustness by introducing a multi-person physics-based trajectory optimization step, effectively addressing challenges such as rapid motions, occlusions, and close interactions. Experimental results on diverse datasets, including a new benchmark of elite boxing footage, demonstrate state-of-the-art performance. Additionally, we release comprehensive annotated video datasets to advance future research in multi-person pose estimation for combat sports.", "date": "2025-04-11", "categories": []}
{"id": "2507.02565", "title": "Reconstructing Close Human Interaction with Appearance and Proxemics Reasoning", "abstract": "Due to visual ambiguities and inter-person occlusions, existing human pose estimation methods cannot recover plausible close interactions from in-the-wild videos. Even state-of-the-art large foundation models~(\\eg, SAM) cannot accurately distinguish human semantics in such challenging scenarios. In this work, we find that human appearance can provide a straightforward cue to address these obstacles. Based on this observation, we propose a dual-branch optimization framework to reconstruct accurate interactive motions with plausible body contacts constrained by human appearances, social proxemics, and physical laws. Specifically, we first train a diffusion model to learn the human proxemic behavior and pose prior knowledge. The trained network and two optimizable tensors are then incorporated into a dual-branch optimization framework to reconstruct human motions and appearances. Several constraints based on 3D Gaussians, 2D keypoints, and mesh penetrations are also designed to assist the optimization. With the proxemics prior and diverse constraints, our method is capable of estimating accurate interactions from in-the-wild videos captured in complex environments. We further build a dataset with pseudo ground-truth interaction annotations, which may promote future research on pose estimation and human behavior understanding. Experimental results on several benchmarks demonstrate that our method outperforms existing approaches. The code and data are available at https://www.buzhenhuang.com/works/CloseApp.html.", "date": "2025-07-03", "categories": []}
{"id": "2403.04444", "title": "Disentangled Diffusion-Based 3D Human Pose Estimation with Hierarchical Spatial and Temporal Denoiser", "abstract": "Recently, diffusion-based methods for monocular 3D human pose estimation have achieved state-of-the-art (SOTA) performance by directly regressing the 3D joint coordinates from the 2D pose sequence. Although some methods decompose the task into bone length and bone direction prediction based on the human anatomical skeleton to explicitly incorporate more human body prior constraints, the performance of these methods is significantly lower than that of the SOTA diffusion-based methods. This can be attributed to the tree structure of the human skeleton. Direct application of the disentangled method could amplify the accumulation of hierarchical errors, propagating through each hierarchy. Meanwhile, the hierarchical information has not been fully explored by the previous methods. To address these problems, a Disentangled Diffusion-based 3D Human Pose Estimation method with Hierarchical Spatial and Temporal Denoiser is proposed, termed DDHPose. In our approach: (1) We disentangle the 3D pose and diffuse the bone length and bone direction during the forward process of the diffusion model to effectively model the human pose prior. A disentanglement loss is proposed to supervise diffusion model learning. (2) For the reverse process, we propose Hierarchical Spatial and Temporal Denoiser (HSTDenoiser) to improve the hierarchical modeling of each joint. Our HSTDenoiser comprises two components: the Hierarchical-Related Spatial Transformer (HRST) and the Hierarchical-Related Temporal Transformer (HRTT). HRST exploits joint spatial information and the influence of the parent joint on each joint for spatial modeling, while HRTT utilizes information from both the joint and its hierarchical adjacent joints to explore the hierarchical temporal correlations among joints. Code and models are available at https://github.com/Andyen512/DDHPose", "date": "2024-03-07", "categories": []}
{"id": "2506.19747", "title": "Systematic Comparison of Projection Methods for Monocular 3D Human Pose Estimation on Fisheye Images", "abstract": "Fisheye cameras offer robots the ability to capture human movements across a wider field of view (FOV) than standard pinhole cameras, making them particularly useful for applications in human-robot interaction and automotive contexts. However, accurately detecting human poses in fisheye images is challenging due to the curved distortions inherent to fisheye optics. While various methods for undistorting fisheye images have been proposed, their effectiveness and limitations for poses that cover a wide FOV has not been systematically evaluated in the context of absolute human pose estimation from monocular fisheye images. To address this gap, we evaluate the impact of pinhole, equidistant and double sphere camera models, as well as cylindrical projection methods, on 3D human pose estimation accuracy. We find that in close-up scenarios, pinhole projection is inadequate, and the optimal projection method varies with the FOV covered by the human pose. The usage of advanced fisheye models like the double sphere model significantly enhances 3D human pose estimation accuracy. We propose a heuristic for selecting the appropriate projection model based on the detection bounding box to enhance prediction quality. Additionally, we introduce and evaluate on our novel dataset FISHnCHIPS, which features 3D human skeleton annotations in fisheye images, including images from unconventional angles, such as extreme close-ups, ground-mounted cameras, and wide-FOV poses, available at: https://www.vision.rwth-aachen.de/fishnchips", "date": "2025-06-24", "categories": []}
{"id": "2506.16061", "title": "STAR-Pose: Efficient Low-Resolution Video Human Pose Estimation via Spatial-Temporal Adaptive Super-Resolution", "abstract": "Human pose estimation in low-resolution videos presents a fundamental challenge in computer vision. Conventional methods either assume high-quality inputs or employ computationally expensive cascaded processing, which limits their deployment in resource-constrained environments. We propose STAR-Pose, a spatial-temporal adaptive super-resolution framework specifically designed for video-based human pose estimation. Our method features a novel spatial-temporal Transformer with LeakyReLU-modified linear attention, which efficiently captures long-range temporal dependencies. Moreover, it is complemented by an adaptive fusion module that integrates parallel CNN branch for local texture enhancement. We also design a pose-aware compound loss to achieve task-oriented super-resolution. This loss guides the network to reconstruct structural features that are most beneficial for keypoint localization, rather than optimizing purely for visual quality. Extensive experiments on several mainstream video HPE datasets demonstrate that STAR-Pose outperforms existing approaches. It achieves up to 5.2% mAP improvement under extremely low-resolution (64x48) conditions while delivering 2.8x to 4.4x faster inference than cascaded approaches.", "date": "2025-06-19", "categories": []}
{"id": "2506.14596", "title": "PoseGRAF: Geometric-Reinforced Adaptive Fusion for Monocular 3D Human Pose Estimation", "abstract": "Existing monocular 3D pose estimation methods primarily rely on joint positional features, while overlooking intrinsic directional and angular correlations within the skeleton. As a result, they often produce implausible poses under joint occlusions or rapid motion changes. To address these challenges, we propose the PoseGRAF framework. We first construct a dual graph convolutional structure that separately processes joint and bone graphs, effectively capturing their local dependencies. A Cross-Attention module is then introduced to model interdependencies between bone directions and joint features. Building upon this, a dynamic fusion module is designed to adaptively integrate both feature types by leveraging the relational dependencies between joints and bones. An improved Transformer encoder is further incorporated in a residual manner to generate the final output. Experimental results on the Human3.6M and MPI-INF-3DHP datasets show that our method exceeds state-of-the-art approaches. Additional evaluations on in-the-wild videos further validate its generalizability. The code is publicly available at https://github.com/iCityLab/PoseGRAF.", "date": "2025-06-17", "categories": []}
{"id": "2407.03817", "title": "Markerless Multi-view 3D Human Pose Estimation: a survey", "abstract": "3D human pose estimation involves reconstructing the human skeleton by detecting the body joints. Accurate and efficient solutions are required for several real-world applications including animation, human-robot interaction, surveillance, and sports. However, challenges such as occlusions, 2D pose mismatches, random camera perspectives, and limited 3D labelled data have been hampering the models' performance and limiting their deployment in real-world scenarios. The higher availability of cameras has led researchers to explore multi-view solutions to take advantage of the different perspectives to reconstruct the pose.\n  Most existing reviews have mainly focused on monocular 3D human pose estimation, so a comprehensive survey on multi-view approaches has been missing since 2012. According to the reviewed articles, the majority of the existing methods are fully-supervised approaches based on geometric constraints, which are often limited by 2D pose mismatches. To mitigate this, researchers have proposed incorporating temporal consistency or depth information. Alternatively, working directly with 3D features has been shown to completely overcome this issue, albeit at the cost of increased computational complexity. Additionally, models with lower levels of supervision have been identified to help address challenges such as annotated data scarcity and generalisation to new setups. Therefore, no method currently addresses all challenges associated with 3D pose reconstruction, and a trade-off between complexity and performance exists. Further research is needed to develop approaches capable of quickly inferring a highly accurate 3D pose with bearable computation cost. Techniques such as active learning, low-supervision methods, temporal consistency, view selection, depth information estimation, and multi-modal approaches are strategies to consider when developing a new method for this task.", "date": "2024-07-04", "categories": []}
{"id": "2506.03169", "title": "Improvement of human health lifespan with hybrid group pose estimation methods", "abstract": "Human beings rely heavily on estimation of poses in order to access their body movements. Human pose estimation methods take advantage of computer vision advances in order to track human body movements in real life applications. This comes from videos which are recorded through available devices. These para-digms provide potential to make human movement measurement more accessible to users. The consumers of pose estimation movements believe that human poses content tend to supplement available videos. This has increased pose estimation software usage to estimate human poses. In order to address this problem, we develop hybrid-ensemble-based group pose estimation method to improve human health. This proposed hybrid-ensemble-based group pose estimation method aims to detect multi-person poses using modified group pose estimation and modified real time pose estimation. This ensemble allows fusion of performance of stated methods in real time. The input poses from images are fed into individual meth-ods. The pose transformation method helps to identify relevant features for en-semble to perform training effectively. After this, customized pre-trained hybrid ensemble is trained on public benchmarked datasets which is being evaluated through test datasets. The effectiveness and viability of proposed method is estab-lished based on comparative analysis of group pose estimation methods and ex-periments conducted on benchmarked datasets. It provides best optimized results in real-time pose estimation. It makes pose estimation method more robust to oc-clusion and improves dense regression accuracy. These results have affirmed po-tential application of this method in several real-time situations with improvement in human health life span", "date": "2025-05-28", "categories": []}
{"id": "2506.02853", "title": "Learning Pyramid-structured Long-range Dependencies for 3D Human Pose Estimation", "abstract": "Action coordination in human structure is indispensable for the spatial constraints of 2D joints to recover 3D pose. Usually, action coordination is represented as a long-range dependence among body parts. However, there are two main challenges in modeling long-range dependencies. First, joints should not only be constrained by other individual joints but also be modulated by the body parts. Second, existing methods make networks deeper to learn dependencies between non-linked parts. They introduce uncorrelated noise and increase the model size. In this paper, we utilize a pyramid structure to better learn potential long-range dependencies. It can capture the correlation across joints and groups, which complements the context of the human sub-structure. In an effective cross-scale way, it captures the pyramid-structured long-range dependence. Specifically, we propose a novel Pyramid Graph Attention (PGA) module to capture long-range cross-scale dependencies. It concatenates information from various scales into a compact sequence, and then computes the correlation between scales in parallel. Combining PGA with graph convolution modules, we develop a Pyramid Graph Transformer (PGFormer) for 3D human pose estimation, which is a lightweight multi-scale transformer architecture. It encapsulates human sub-structures into self-attention by pooling. Extensive experiments show that our approach achieves lower error and smaller model size than state-of-the-art methods on Human3.6M and MPI-INF-3DHP datasets. The code is available at https://github.com/MingjieWe/PGFormer.", "date": "2025-06-03", "categories": []}
{"id": "2505.24411", "title": "PCIE_Pose Solution for EgoExo4D Pose and Proficiency Estimation Challenge", "abstract": "This report introduces our team's (PCIE_EgoPose) solutions for the EgoExo4D Pose and Proficiency Estimation Challenges at CVPR2025. Focused on the intricate task of estimating 21 3D hand joints from RGB egocentric videos, which are complicated by subtle movements and frequent occlusions, we developed the Hand Pose Vision Transformer (HP-ViT+). This architecture synergizes a Vision Transformer and a CNN backbone, using weighted fusion to refine the hand pose predictions. For the EgoExo4D Body Pose Challenge, we adopted a multimodal spatio-temporal feature integration strategy to address the complexities of body pose estimation across dynamic contexts. Our methods achieved remarkable performance: 8.31 PA-MPJPE in the Hand Pose Challenge and 11.25 MPJPE in the Body Pose Challenge, securing championship titles in both competitions. We extended our pose estimation solutions to the Proficiency Estimation task, applying core technologies such as transformer-based architectures. This extension enabled us to achieve a top-1 accuracy of 0.53, a SOTA result, in the Demonstrator Proficiency Estimation competition.", "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": "2502.10724", "title": "Semantics-aware Test-time Adaptation for 3D Human Pose Estimation", "abstract": "This work highlights a semantics misalignment in 3D human pose estimation. For the task of test-time adaptation, the misalignment manifests as overly smoothed and unguided predictions. The smoothing settles predictions towards some average pose. Furthermore, when there are occlusions or truncations, the adaptation becomes fully unguided. To this end, we pioneer the integration of a semantics-aware motion prior for the test-time adaptation of 3D pose estimation. We leverage video understanding and a well-structured motion-text space to adapt the model motion prediction to adhere to video semantics during test time. Additionally, we incorporate a missing 2D pose completion based on the motion-text similarity. The pose completion strengthens the motion prior's guidance for occlusions and truncations. Our method significantly improves state-of-the-art 3D human pose estimation TTA techniques, with more than 12% decrease in PA-MPJPE on 3DPW and 3DHP.", "date": "2025-02-15", "categories": []}
{"id": "2505.22007", "title": "Event-based Egocentric Human Pose Estimation in Dynamic Environment", "abstract": "Estimating human pose using a front-facing egocentric camera is essential for applications such as sports motion analysis, VR/AR, and AI for wearable devices. However, many existing methods rely on RGB cameras and do not account for low-light environments or motion blur. Event-based cameras have the potential to address these challenges. In this work, we introduce a novel task of human pose estimation using a front-facing event-based camera mounted on the head and propose D-EventEgo, the first framework for this task. The proposed method first estimates the head poses, and then these are used as conditions to generate body poses. However, when estimating head poses, the presence of dynamic objects mixed with background events may reduce head pose estimation accuracy. Therefore, we introduce the Motion Segmentation Module to remove dynamic objects and extract background information. Extensive experiments on our synthetic event-based dataset derived from EgoBody, demonstrate that our approach outperforms our baseline in four out of five evaluation metrics in dynamic environments.", "date": "2025-05-28", "categories": []}
{"id": "2504.06397", "title": "PromptHMR: Promptable Human Mesh Recovery", "abstract": "Human pose and shape (HPS) estimation presents challenges in diverse scenarios such as crowded scenes, person-person interactions, and single-view reconstruction. Existing approaches lack mechanisms to incorporate auxiliary \"side information\" that could enhance reconstruction accuracy in such challenging scenarios. Furthermore, the most accurate methods rely on cropped person detections and cannot exploit scene context while methods that process the whole image often fail to detect people and are less accurate than methods that use crops. While recent language-based methods explore HPS reasoning through large language or vision-language models, their metric accuracy is well below the state of the art. In contrast, we present PromptHMR, a transformer-based promptable method that reformulates HPS estimation through spatial and semantic prompts. Our method processes full images to maintain scene context and accepts multiple input modalities: spatial prompts like bounding boxes and masks, and semantic prompts like language descriptions or interaction labels. PromptHMR demonstrates robust performance across challenging scenarios: estimating people from bounding boxes as small as faces in crowded scenes, improving body shape estimation through language descriptions, modeling person-person interactions, and producing temporally coherent motions in videos. Experiments on benchmarks show that PromptHMR achieves state-of-the-art performance while offering flexible prompt-based control over the HPS estimation process.", "date": "2025-04-08", "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": "2505.10250", "title": "ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization", "abstract": "Human mesh recovery (HMR) from a single image is inherently ill-posed due to depth ambiguity and occlusions. Probabilistic methods have tried to solve this by generating numerous plausible 3D human mesh predictions, but they often exhibit misalignment with 2D image observations and weak robustness to in-the-wild images. To address these issues, we propose ADHMR, a framework that Aligns a Diffusion-based HMR model in a preference optimization manner. First, we train a human mesh prediction assessment model, HMR-Scorer, capable of evaluating predictions even for in-the-wild images without 3D annotations. We then use HMR-Scorer to create a preference dataset, where each input image has a pair of winner and loser mesh predictions. This dataset is used to finetune the base model using direct preference optimization. Moreover, HMR-Scorer also helps improve existing HMR models by data cleaning, even with fewer training samples. Extensive experiments show that ADHMR outperforms current state-of-the-art methods. Code is available at: https://github.com/shenwenhao01/ADHMR.", "date": "2025-05-15", "categories": []}
{"id": "2505.12130", "title": "Keypoints as Dynamic Centroids for Unified Human Pose and Segmentation", "abstract": "The dynamic movement of the human body presents a fundamental challenge for human pose estimation and body segmentation. State-of-the-art approaches primarily rely on combining keypoint heatmaps with segmentation masks but often struggle in scenarios involving overlapping joints or rapidly changing poses during instance-level segmentation. To address these limitations, we propose Keypoints as Dynamic Centroid (KDC), a new centroid-based representation for unified human pose estimation and instance-level segmentation. KDC adopts a bottom-up paradigm to generate keypoint heatmaps for both easily distinguishable and complex keypoints and improves keypoint detection and confidence scores by introducing KeyCentroids using a keypoint disk. It leverages high-confidence keypoints as dynamic centroids in the embedding space to generate MaskCentroids, allowing for swift clustering of pixels to specific human instances during rapid body movements in live environments. Our experimental evaluations on the CrowdPose, OCHuman, and COCO benchmarks demonstrate KDC's effectiveness and generalizability in challenging scenarios in terms of both accuracy and runtime performance. The implementation is available at: https://sites.google.com/view/niazahmad/projects/kdc.", "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": "2505.09393", "title": "UMotion: Uncertainty-driven Human Motion Estimation from Inertial and Ultra-wideband Units", "abstract": "Sparse wearable inertial measurement units (IMUs) have gained popularity for estimating 3D human motion. However, challenges such as pose ambiguity, data drift, and limited adaptability to diverse bodies persist. To address these issues, we propose UMotion, an uncertainty-driven, online fusing-all state estimation framework for 3D human shape and pose estimation, supported by six integrated, body-worn ultra-wideband (UWB) distance sensors with IMUs. UWB sensors measure inter-node distances to infer spatial relationships, aiding in resolving pose ambiguities and body shape variations when combined with anthropometric data. Unfortunately, IMUs are prone to drift, and UWB sensors are affected by body occlusions. Consequently, we develop a tightly coupled Unscented Kalman Filter (UKF) framework that fuses uncertainties from sensor data and estimated human motion based on individual body shape. The UKF iteratively refines IMU and UWB measurements by aligning them with uncertain human motion constraints in real-time, producing optimal estimates for each. Experiments on both synthetic and real-world datasets demonstrate the effectiveness of UMotion in stabilizing sensor data and the improvement over state of the art in pose accuracy.", "date": "2025-05-14", "categories": []}
{"id": "2505.06411", "title": "MAGE:A Multi-stage Avatar Generator with Sparse Observations", "abstract": "Inferring full-body poses from Head Mounted Devices, which capture only 3-joint observations from the head and wrists, is a challenging task with wide AR/VR applications. Previous attempts focus on learning one-stage motion mapping and thus suffer from an over-large inference space for unobserved body joint motions. This often leads to unsatisfactory lower-body predictions and poor temporal consistency, resulting in unrealistic or incoherent motion sequences. To address this, we propose a powerful Multi-stage Avatar GEnerator named MAGE that factorizes this one-stage direct motion mapping learning with a progressive prediction strategy. Specifically, given initial 3-joint motions, MAGE gradually inferring multi-scale body part poses at different abstract granularity levels, starting from a 6-part body representation and gradually refining to 22 joints. With decreasing abstract levels step by step, MAGE introduces more motion context priors from former prediction stages and thus improves realistic motion completion with richer constraint conditions and less ambiguity. Extensive experiments on large-scale datasets verify that MAGE significantly outperforms state-of-the-art methods with better accuracy and continuity.", "date": "2025-05-09", "categories": []}
{"id": "2505.06575", "title": "GRACE: Estimating Geometry-level 3D Human-Scene Contact from 2D Images", "abstract": "Estimating the geometry level of human-scene contact aims to ground specific contact surface points at 3D human geometries, which provides a spatial prior and bridges the interaction between human and scene, supporting applications such as human behavior analysis, embodied AI, and AR/VR. To complete the task, existing approaches predominantly rely on parametric human models (e.g., SMPL), which establish correspondences between images and contact regions through fixed SMPL vertex sequences. This actually completes the mapping from image features to an ordered sequence. However, this approach lacks consideration of geometry, limiting its generalizability in distinct human geometries. In this paper, we introduce GRACE (Geometry-level Reasoning for 3D Human-scene Contact Estimation), a new paradigm for 3D human contact estimation. GRACE incorporates a point cloud encoder-decoder architecture along with a hierarchical feature extraction and fusion module, enabling the effective integration of 3D human geometric structures with 2D interaction semantics derived from images. Guided by visual cues, GRACE establishes an implicit mapping from geometric features to the vertex space of the 3D human mesh, thereby achieving accurate modeling of contact regions. This design ensures high prediction accuracy and endows the framework with strong generalization capability across diverse human geometries. Extensive experiments on multiple benchmark datasets demonstrate that GRACE achieves state-of-the-art performance in contact estimation, with additional results further validating its robust generalization to unstructured human point clouds.", "date": "2025-05-10", "categories": []}
{"id": "2505.07249", "title": "When Dance Video Archives Challenge Computer Vision", "abstract": "The accuracy and efficiency of human body pose estimation depend on the quality of the data to be processed and of the particularities of these data. To demonstrate how dance videos can challenge pose estimation techniques, we proposed a new 3D human body pose estimation pipeline which combined up-to-date techniques and methods that had not been yet used in dance analysis. Second, we performed tests and extensive experimentations from dance video archives, and used visual analytic tools to evaluate the impact of several data parameters on human body pose. Our results are publicly available for research at https://www.couleur.org/articles/arXiv-1-2025/", "date": "2025-05-12", "categories": []}
{"id": "2311.12421", "title": "Two Views Are Better than One: Monocular 3D Pose Estimation with\n  Multiview Consistency", "abstract": "  Deducing a 3D human pose from a single 2D image is inherently challenging\nbecause multiple 3D poses can correspond to the same 2D representation. 3D data\ncan resolve this pose ambiguity, but it is expensive to record and requires an\nintricate setup that is often restricted to controlled lab environments. We\npropose a method that improves the performance of deep learning-based monocular\n3D human pose estimation models by using multiview data only during training,\nbut not during inference. We introduce a novel loss function, consistency loss,\nwhich operates on two synchronized views. This approach is simpler than\nprevious models that require 3D ground truth or intrinsic and extrinsic camera\nparameters. Our consistency loss penalizes differences in two pose sequences\nafter rigid alignment. We also demonstrate that our consistency loss\nsubstantially improves performance for fine-tuning without requiring 3D data.\nFurthermore, we show that using our consistency loss can yield state-of-the-art\nperformance when training models from scratch in a semi-supervised manner. Our\nfindings provide a simple way to capture new data, e.g in a new domain. This\ndata can be added using off-the-shelf cameras with no calibration requirements.\nWe make all our code and data publicly available.\n", "date": "2023-11-21", "categories": []}
{"id": "2505.05336", "title": "Progressive Inertial Poser: Progressive Real-Time Kinematic Chain\n  Estimation for 3D Full-Body Pose from Three IMU Sensors", "abstract": "  The motion capture system that supports full-body virtual representation is\nof key significance for virtual reality. Compared to vision-based systems,\nfull-body pose estimation from sparse tracking signals is not limited by\nenvironmental conditions or recording range. However, previous works either\nface the challenge of wearing additional sensors on the pelvis and lower-body\nor rely on external visual sensors to obtain global positions of key joints. To\nimprove the practicality of the technology for virtual reality applications, we\nestimate full-body poses using only inertial data obtained from three Inertial\nMeasurement Unit (IMU) sensors worn on the head and wrists, thereby reducing\nthe complexity of the hardware system. In this work, we propose a method called\nProgressive Inertial Poser (ProgIP) for human pose estimation, which combines\nneural network estimation with a human dynamics model, considers the\nhierarchical structure of the kinematic chain, and employs a multi-stage\nprogressive network estimation with increased depth to reconstruct full-body\nmotion in real time. The encoder combines Transformer Encoder and bidirectional\nLSTM (TE-biLSTM) to flexibly capture the temporal dependencies of the inertial\nsequence, while the decoder based on multi-layer perceptrons (MLPs) transforms\nhigh-dimensional features and accurately projects them onto Skinned\nMulti-Person Linear (SMPL) model parameters. Quantitative and qualitative\nexperimental results on multiple public datasets show that our method\noutperforms state-of-the-art methods with the same inputs, and is comparable to\nrecent works using six IMU sensors.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.04276", "title": "HDiffTG: A Lightweight Hybrid Diffusion-Transformer-GCN Architecture for\n  3D Human Pose Estimation", "abstract": "  We propose HDiffTG, a novel 3D Human Pose Estimation (3DHPE) method that\nintegrates Transformer, Graph Convolutional Network (GCN), and diffusion model\ninto a unified framework. HDiffTG leverages the strengths of these techniques\nto significantly improve pose estimation accuracy and robustness while\nmaintaining a lightweight design. The Transformer captures global\nspatiotemporal dependencies, the GCN models local skeletal structures, and the\ndiffusion model provides step-by-step optimization for fine-tuning, achieving a\ncomplementary balance between global and local features. This integration\nenhances the model's ability to handle pose estimation under occlusions and in\ncomplex scenarios. Furthermore, we introduce lightweight optimizations to the\nintegrated model and refine the objective function design to reduce\ncomputational overhead without compromising performance. Evaluation results on\nthe Human3.6M and MPI-INF-3DHP datasets demonstrate that HDiffTG achieves\nstate-of-the-art (SOTA) performance on the MPI-INF-3DHP dataset while excelling\nin both accuracy and computational efficiency. Additionally, the model exhibits\nexceptional robustness in noisy and occluded environments. Source codes and\nmodels are available at https://github.com/CirceJie/HDiffTG\n", "date": "2025-05-07", "categories": []}
{"id": "2505.03445", "title": "Polar Coordinate-Based 2D Pose Prior with Neural Distance Field", "abstract": "  Human pose capture is essential for sports analysis, enabling precise\nevaluation of athletes' movements. While deep learning-based human pose\nestimation (HPE) models from RGB videos have achieved impressive performance on\npublic datasets, their effectiveness in real-world sports scenarios is often\nhindered by motion blur, occlusions, and domain shifts across different pose\nrepresentations. Fine-tuning these models can partially alleviate such\nchallenges but typically requires large-scale annotated data and still\nstruggles to generalize across diverse sports environments. To address these\nlimitations, we propose a 2D pose prior-guided refinement approach based on\nNeural Distance Fields (NDF). Unlike existing approaches that rely solely on\nangular representations of human poses, we introduce a polar coordinate-based\nrepresentation that explicitly incorporates joint connection lengths, enabling\na more accurate correction of erroneous pose estimations. Additionally, we\ndefine a novel non-geodesic distance metric that separates angular and radial\ndiscrepancies, which we demonstrate is better suited for polar representations\nthan traditional geodesic distances. To mitigate data scarcity, we develop a\ngradient-based batch-projection augmentation strategy, which synthesizes\nrealistic pose samples through iterative refinement. Our method is evaluated on\na long jump dataset, demonstrating its ability to improve 2D pose estimation\nacross multiple pose representations, making it robust across different\ndomains. Experimental results show that our approach enhances pose plausibility\nwhile requiring only limited training data. Code is available at:\nhttps://github.com/QGAN2019/polar-NDF.\n", "date": "2025-05-06", "categories": []}
{"id": "2505.00755", "title": "P2P-Insole: Human Pose Estimation Using Foot Pressure Distribution and\n  Motion Sensors", "abstract": "  This work presents P2P-Insole, a low-cost approach for estimating and\nvisualizing 3D human skeletal data using insole-type sensors integrated with\nIMUs. Each insole, fabricated with e-textile garment techniques, costs under\nUSD 1, making it significantly cheaper than commercial alternatives and ideal\nfor large-scale production. Our approach uses foot pressure distribution,\nacceleration, and rotation data to overcome limitations, providing a\nlightweight, minimally intrusive, and privacy-aware solution. The system\nemploys a Transformer model for efficient temporal feature extraction, enriched\nby first and second derivatives in the input stream. Including multimodal\ninformation, such as accelerometers and rotational measurements, improves the\naccuracy of complex motion pattern recognition. These facts are demonstrated\nexperimentally, while error metrics show the robustness of the approach in\nvarious posture estimation tasks. This work could be the foundation for a\nlow-cost, practical application in rehabilitation, injury prevention, and\nhealth monitoring while enabling further development through sensor\noptimization and expanded datasets.\n", "date": "2025-05-01", "categories": []}
{"id": "2505.01003", "title": "3D Human Pose Estimation via Spatial Graph Order Attention and Temporal\n  Body Aware Transformer", "abstract": "  Nowadays, Transformers and Graph Convolutional Networks (GCNs) are the\nprevailing techniques for 3D human pose estimation. However, Transformer-based\nmethods either ignore the spatial neighborhood relationships between the joints\nwhen used for skeleton representations or disregard the local temporal patterns\nof the local joint movements in skeleton sequence modeling, while GCN-based\nmethods often neglect the need for pose-specific representations. To address\nthese problems, we propose a new method that exploits the graph modeling\ncapability of GCN to represent each skeleton with multiple graphs of different\norders, incorporated with a newly introduced Graph Order Attention module that\ndynamically emphasizes the most representative orders for each joint. The\nresulting spatial features of the sequence are further processed using a\nproposed temporal Body Aware Transformer that models the global body feature\ndependencies in the sequence with awareness of the local inter-skeleton feature\ndependencies of joints. Given that our 3D pose output aligns with the central\n2D pose in the sequence, we improve the self-attention mechanism to be aware of\nthe central pose while diminishing its focus gradually towards the first and\nthe last poses. Extensive experiments on Human3.6m, MPIINF-3DHP, and HumanEva-I\ndatasets demonstrate the effectiveness of the proposed method. Code and models\nare made available on Github.\n", "date": "2025-05-02", "categories": []}
{"id": "2411.18808", "title": "Lifting Motion to the 3D World via 2D Diffusion", "abstract": "  Estimating 3D motion from 2D observations is a long-standing research\nchallenge. Prior work typically requires training on datasets containing ground\ntruth 3D motions, limiting their applicability to activities well-represented\nin existing motion capture data. This dependency particularly hinders\ngeneralization to out-of-distribution scenarios or subjects where collecting 3D\nground truth is challenging, such as complex athletic movements or animal\nmotion. We introduce MVLift, a novel approach to predict global 3D motion --\nincluding both joint rotations and root trajectories in the world coordinate\nsystem -- using only 2D pose sequences for training. Our multi-stage framework\nleverages 2D motion diffusion models to progressively generate consistent 2D\npose sequences across multiple views, a key step in recovering accurate global\n3D motion. MVLift generalizes across various domains, including human poses,\nhuman-object interactions, and animal poses. Despite not requiring 3D\nsupervision, it outperforms prior work on five datasets, including those\nmethods that require 3D supervision.\n", "date": "2024-11-27", "categories": []}
{"id": "2504.19032", "title": "VISUALCENT: Visual Human Analysis using Dynamic Centroid Representation", "abstract": "  We introduce VISUALCENT, a unified human pose and instance segmentation\nframework to address generalizability and scalability limitations to multi\nperson visual human analysis. VISUALCENT leverages centroid based bottom up\nkeypoint detection paradigm and uses Keypoint Heatmap incorporating Disk\nRepresentation and KeyCentroid to identify the optimal keypoint coordinates.\nFor the unified segmentation task, an explicit keypoint is defined as a dynamic\ncentroid called MaskCentroid to swiftly cluster pixels to specific human\ninstance during rapid changes in human body movement or significantly occluded\nenvironment. Experimental results on COCO and OCHuman datasets demonstrate\nVISUALCENTs accuracy and real time performance advantages, outperforming\nexisting methods in mAP scores and execution frame rate per second. The\nimplementation is available on the project page.\n", "date": "2025-04-26", "categories": []}
{"id": "2504.18332", "title": "SSD-Poser: Avatar Pose Estimation with State Space Duality from Sparse\n  Observations", "abstract": "  The growing applications of AR/VR increase the demand for real-time full-body\npose estimation from Head-Mounted Displays (HMDs). Although HMDs provide joint\nsignals from the head and hands, reconstructing a full-body pose remains\nchallenging due to the unconstrained lower body. Recent advancements often rely\non conventional neural networks and generative models to improve performance in\nthis task, such as Transformers and diffusion models. However, these approaches\nstruggle to strike a balance between achieving precise pose reconstruction and\nmaintaining fast inference speed. To overcome these challenges, a lightweight\nand efficient model, SSD-Poser, is designed for robust full-body motion\nestimation from sparse observations. SSD-Poser incorporates a well-designed\nhybrid encoder, State Space Attention Encoders, to adapt the state space\nduality to complex motion poses and enable real-time realistic pose\nreconstruction. Moreover, a Frequency-Aware Decoder is introduced to mitigate\njitter caused by variable-frequency motion signals, remarkably enhancing the\nmotion smoothness. Comprehensive experiments on the AMASS dataset demonstrate\nthat SSD-Poser achieves exceptional accuracy and computational efficiency,\nshowing outstanding inference efficiency compared to state-of-the-art methods.\n", "date": "2025-04-25", "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": "2504.12186", "title": "CoMotion: Concurrent Multi-person 3D Motion", "abstract": "  We introduce an approach for detecting and tracking detailed 3D poses of\nmultiple people from a single monocular camera stream. Our system maintains\ntemporally coherent predictions in crowded scenes filled with difficult poses\nand occlusions. Our model performs both strong per-frame detection and a\nlearned pose update to track people from frame to frame. Rather than match\ndetections across time, poses are updated directly from a new input image,\nwhich enables online tracking through occlusion. We train on numerous image and\nvideo datasets leveraging pseudo-labeled annotations to produce a model that\nmatches state-of-the-art systems in 3D pose estimation accuracy while being\nfaster and more accurate in tracking multiple people through time. Code and\nweights are provided at https://github.com/apple/ml-comotion\n", "date": "2025-04-16", "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": "2504.09953", "title": "Efficient 2D to Full 3D Human Pose Uplifting including Joint Rotations", "abstract": "  In sports analytics, accurately capturing both the 3D locations and rotations\nof body joints is essential for understanding an athlete's biomechanics. While\nHuman Mesh Recovery (HMR) models can estimate joint rotations, they often\nexhibit lower accuracy in joint localization compared to 3D Human Pose\nEstimation (HPE) models. Recent work addressed this limitation by combining a\n3D HPE model with inverse kinematics (IK) to estimate both joint locations and\nrotations. However, IK is computationally expensive. To overcome this, we\npropose a novel 2D-to-3D uplifting model that directly estimates 3D human\nposes, including joint rotations, in a single forward pass. We investigate\nmultiple rotation representations, loss functions, and training strategies -\nboth with and without access to ground truth rotations. Our models achieve\nstate-of-the-art accuracy in rotation estimation, are 150 times faster than the\nIK-based approach, and surpass HMR models in joint localization precision.\n", "date": "2025-04-14", "categories": []}
{"id": "2504.06638", "title": "HGMamba: Enhancing 3D Human Pose Estimation with a HyperGCN-Mamba\n  Network", "abstract": "  3D human pose lifting is a promising research area that leverages estimated\nand ground-truth 2D human pose data for training. While existing approaches\nprimarily aim to enhance the performance of estimated 2D poses, they often\nstruggle when applied to ground-truth 2D pose data. We observe that achieving\naccurate 3D pose reconstruction from ground-truth 2D poses requires precise\nmodeling of local pose structures, alongside the ability to extract robust\nglobal spatio-temporal features. To address these challenges, we propose a\nnovel Hyper-GCN and Shuffle Mamba (HGMamba) block, which processes input data\nthrough two parallel streams: Hyper-GCN and Shuffle-Mamba. The Hyper-GCN stream\nmodels the human body structure as hypergraphs with varying levels of\ngranularity to effectively capture local joint dependencies. Meanwhile, the\nShuffle Mamba stream leverages a state space model to perform spatio-temporal\nscanning across all joints, enabling the establishment of global dependencies.\nBy adaptively fusing these two representations, HGMamba achieves strong global\nfeature modeling while excelling at local structure modeling. We stack multiple\nHGMamba blocks to create three variants of our model, allowing users to select\nthe most suitable configuration based on the desired speed-accuracy trade-off.\nExtensive evaluations on the Human3.6M and MPI-INF-3DHP benchmark datasets\ndemonstrate the effectiveness of our approach. HGMamba-B achieves\nstate-of-the-art results, with P1 errors of 38.65 mm and 14.33 mm on the\nrespective datasets. Code and models are available:\nhttps://github.com/HuCui2022/HGMamba\n", "date": "2025-04-09", "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": "2504.03052", "title": "Cooperative Inference for Real-Time 3D Human Pose Estimation in\n  Multi-Device Edge Networks", "abstract": "  Accurate and real-time three-dimensional (3D) pose estimation is challenging\nin resource-constrained and dynamic environments owing to its high\ncomputational complexity. To address this issue, this study proposes a novel\ncooperative inference method for real-time 3D human pose estimation in mobile\nedge computing (MEC) networks. In the proposed method, multiple end devices\nequipped with lightweight inference models employ dual confidence thresholds to\nfilter ambiguous images. Only the filtered images are offloaded to an edge\nserver with a more powerful inference model for re-evaluation, thereby\nimproving the estimation accuracy under computational and communication\nconstraints. We numerically analyze the performance of the proposed inference\nmethod in terms of the inference accuracy and end-to-end delay and formulate a\njoint optimization problem to derive the optimal confidence thresholds and\ntransmission time for each device, with the objective of minimizing the mean\nper-joint position error (MPJPE) while satisfying the required end-to-end delay\nconstraint. To solve this problem, we demonstrate that minimizing the MPJPE is\nequivalent to maximizing the sum of the inference accuracies for all devices,\ndecompose the problem into manageable subproblems, and present a low-complexity\noptimization algorithm to obtain a near-optimal solution. The experimental\nresults show that a trade-off exists between the MPJPE and end-to-end delay\ndepending on the confidence thresholds. Furthermore, the results confirm that\nthe proposed cooperative inference method achieves a significant reduction in\nthe MPJPE through the optimal selection of confidence thresholds and\ntransmission times, while consistently satisfying the end-to-end delay\nrequirement in various MEC environments.\n", "date": "2025-04-03", "categories": []}
{"id": "2504.03602", "title": "Robust Human Registration with Body Part Segmentation on Noisy Point\n  Clouds", "abstract": "  Registering human meshes to 3D point clouds is essential for applications\nsuch as augmented reality and human-robot interaction but often yields\nimprecise results due to noise and background clutter in real-world data. We\nintroduce a hybrid approach that incorporates body-part segmentation into the\nmesh fitting process, enhancing both human pose estimation and segmentation\naccuracy. Our method first assigns body part labels to individual points, which\nthen guide a two-step SMPL-X fitting: initial pose and orientation estimation\nusing body part centroids, followed by global refinement of the point cloud\nalignment. Additionally, we demonstrate that the fitted human mesh can refine\nbody part labels, leading to improved segmentation. Evaluations on the\ncluttered and noisy real-world datasets InterCap, EgoBody, and BEHAVE show that\nour approach significantly outperforms prior methods in both pose estimation\nand segmentation accuracy. Code and results are available on our project\nwebsite: https://segfit.github.io\n", "date": "2025-04-04", "categories": []}
{"id": "2504.01764", "title": "Dual-stream Transformer-GCN Model with Contextualized Representations\n  Learning for Monocular 3D Human Pose Estimation", "abstract": "  This paper introduces a novel approach to monocular 3D human pose estimation\nusing contextualized representation learning with the Transformer-GCN\ndual-stream model. Monocular 3D human pose estimation is challenged by depth\nambiguity, limited 3D-labeled training data, imbalanced modeling, and\nrestricted model generalization. To address these limitations, our work\nintroduces a groundbreaking motion pre-training method based on contextualized\nrepresentation learning. Specifically, our method involves masking 2D pose\nfeatures and utilizing a Transformer-GCN dual-stream model to learn\nhigh-dimensional representations through a self-distillation setup. By focusing\non contextualized representation learning and spatial-temporal modeling, our\napproach enhances the model's ability to understand spatial-temporal\nrelationships between postures, resulting in superior generalization.\nFurthermore, leveraging the Transformer-GCN dual-stream model, our approach\neffectively balances global and local interactions in video pose estimation.\nThe model adaptively integrates information from both the Transformer and GCN\nstreams, where the GCN stream effectively learns local relationships between\nadjacent key points and frames, while the Transformer stream captures\ncomprehensive global spatial and temporal features. Our model achieves\nstate-of-the-art performance on two benchmark datasets, with an MPJPE of 38.0mm\nand P-MPJPE of 31.9mm on Human3.6M, and an MPJPE of 15.9mm on MPI-INF-3DHP.\nFurthermore, visual experiments on public datasets and in-the-wild videos\ndemonstrate the robustness and generalization capabilities of our approach.\n", "date": "2025-04-02", "categories": []}
{"id": "2503.23331", "title": "HiPART: Hierarchical Pose AutoRegressive Transformer for Occluded 3D\n  Human Pose Estimation", "abstract": "  Existing 2D-to-3D human pose estimation (HPE) methods struggle with the\nocclusion issue by enriching information like temporal and visual cues in the\nlifting stage. In this paper, we argue that these methods ignore the limitation\nof the sparse skeleton 2D input representation, which fundamentally restricts\nthe 2D-to-3D lifting and worsens the occlusion issue. To address these, we\npropose a novel two-stage generative densification method, named Hierarchical\nPose AutoRegressive Transformer (HiPART), to generate hierarchical 2D dense\nposes from the original sparse 2D pose. Specifically, we first develop a\nmulti-scale skeleton tokenization module to quantize the highly dense 2D pose\ninto hierarchical tokens and propose a Skeleton-aware Alignment to strengthen\ntoken connections. We then develop a Hierarchical AutoRegressive Modeling\nscheme for hierarchical 2D pose generation. With generated hierarchical poses\nas inputs for 2D-to-3D lifting, the proposed method shows strong robustness in\noccluded scenarios and achieves state-of-the-art performance on the\nsingle-frame-based 3D HPE. Moreover, it outperforms numerous multi-frame\nmethods while reducing parameter and computational complexity and can also\ncomplement them to further enhance performance and robustness.\n", "date": "2025-03-30", "categories": []}
{"id": "2503.23587", "title": "PhysPose: Refining 6D Object Poses with Physical Constraints", "abstract": "  Accurate 6D object pose estimation from images is a key problem in\nobject-centric scene understanding, enabling applications in robotics,\naugmented reality, and scene reconstruction. Despite recent advances, existing\nmethods often produce physically inconsistent pose estimates, hindering their\ndeployment in real-world scenarios. We introduce PhysPose, a novel approach\nthat integrates physical reasoning into pose estimation through a\npostprocessing optimization enforcing non-penetration and gravitational\nconstraints. By leveraging scene geometry, PhysPose refines pose estimates to\nensure physical plausibility. Our approach achieves state-of-the-art accuracy\non the YCB-Video dataset from the BOP benchmark and improves over the\nstate-of-the-art pose estimation methods on the HOPE-Video dataset.\nFurthermore, we demonstrate its impact in robotics by significantly improving\nsuccess rates in a challenging pick-and-place task, highlighting the importance\nof physical consistency in real-world applications.\n", "date": "2025-03-30", "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": "2412.04456", "title": "HeatFormer: A Neural Optimizer for Multiview Human Mesh Recovery", "abstract": "  We introduce a novel method for human shape and pose recovery that can fully\nleverage multiple static views. We target fixed-multiview people monitoring,\nincluding elderly care and safety monitoring, in which calibrated cameras can\nbe installed at the corners of a room or an open space but whose configuration\nmay vary depending on the environment. Our key idea is to formulate it as\nneural optimization. We achieve this with HeatFormer, a neural optimizer that\niteratively refines the SMPL parameters given multiview images, which is\nfundamentally agonistic to the configuration of views. HeatFormer realizes this\nSMPL parameter estimation as heat map generation and alignment with a novel\ntransformer encoder and decoder. We demonstrate the effectiveness of HeatFormer\nincluding its accuracy, robustness to occlusion, and generalizability through\nan extensive set of experiments. We believe HeatFormer can serve a key role in\npassive human behavior modeling.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.10235", "title": "EnvPoser: Environment-aware Realistic Human Motion Estimation from\n  Sparse Observations with Uncertainty Modeling", "abstract": "  Estimating full-body motion using the tracking signals of head and hands from\nVR devices holds great potential for various applications. However, the\nsparsity and unique distribution of observations present a significant\nchallenge, resulting in an ill-posed problem with multiple feasible solutions\n(i.e., hypotheses). This amplifies uncertainty and ambiguity in full-body\nmotion estimation, especially for the lower-body joints. Therefore, we propose\na new method, EnvPoser, that employs a two-stage framework to perform full-body\nmotion estimation using sparse tracking signals and pre-scanned environment\nfrom VR devices. EnvPoser models the multi-hypothesis nature of human motion\nthrough an uncertainty-aware estimation module in the first stage. In the\nsecond stage, we refine these multi-hypothesis estimates by integrating\nsemantic and geometric environmental constraints, ensuring that the final\nmotion estimation aligns realistically with both the environmental context and\nphysical interactions. Qualitative and quantitative experiments on two public\ndatasets demonstrate that our method achieves state-of-the-art performance,\nhighlighting significant improvements in human motion estimation within\nmotion-environment interaction scenarios.\n", "date": "2024-12-13", "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": "2405.18839", "title": "MEGA: Masked Generative Autoencoder for Human Mesh Recovery", "abstract": "  Human Mesh Recovery (HMR) from a single RGB image is a highly ambiguous\nproblem, as an infinite set of 3D interpretations can explain the 2D\nobservation equally well. Nevertheless, most HMR methods overlook this issue\nand make a single prediction without accounting for this ambiguity. A few\napproaches generate a distribution of human meshes, enabling the sampling of\nmultiple predictions; however, none of them is competitive with the latest\nsingle-output model when making a single prediction. This work proposes a new\napproach based on masked generative modeling. By tokenizing the human pose and\nshape, we formulate the HMR task as generating a sequence of discrete tokens\nconditioned on an input image. We introduce MEGA, a MaskEd Generative\nAutoencoder trained to recover human meshes from images and partial human mesh\ntoken sequences. Given an image, our flexible generation scheme allows us to\npredict a single human mesh in deterministic mode or to generate multiple human\nmeshes in stochastic mode. Experiments on in-the-wild benchmarks show that MEGA\nachieves state-of-the-art performance in deterministic and stochastic modes,\noutperforming single-output and multi-output approaches.\n", "date": "2024-05-29", "categories": []}
{"id": "2503.13025", "title": "PoseSyn: Synthesizing Diverse 3D Pose Data from In-the-Wild 2D Data", "abstract": "  Despite considerable efforts to enhance the generalization of 3D pose\nestimators without costly 3D annotations, existing data augmentation methods\nstruggle in real world scenarios with diverse human appearances and complex\nposes. We propose PoseSyn, a novel data synthesis framework that transforms\nabundant in the wild 2D pose dataset into diverse 3D pose image pairs. PoseSyn\ncomprises two key components: Error Extraction Module (EEM), which identifies\nchallenging poses from the 2D pose datasets, and Motion Synthesis Module (MSM),\nwhich synthesizes motion sequences around the challenging poses. Then, by\ngenerating realistic 3D training data via a human animation model aligned with\nchallenging poses and appearances PoseSyn boosts the accuracy of various 3D\npose estimators by up to 14% across real world benchmarks including various\nbackgrounds and occlusions, challenging poses, and multi view scenarios.\nExtensive experiments further confirm that PoseSyn is a scalable and effective\napproach for improving generalization without relying on expensive 3D\nannotations, regardless of the pose estimator's model size or design.\n", "date": "2025-03-17", "categories": []}
{"id": "2406.17988", "title": "DICE: End-to-end Deformation Capture of Hand-Face Interactions from a\n  Single Image", "abstract": "  Reconstructing 3D hand-face interactions with deformations from a single\nimage is a challenging yet crucial task with broad applications in AR, VR, and\ngaming. The challenges stem from self-occlusions during single-view hand-face\ninteractions, diverse spatial relationships between hands and face, complex\ndeformations, and the ambiguity of the single-view setting. The first and only\nmethod for hand-face interaction recovery, Decaf, introduces a global fitting\noptimization guided by contact and deformation estimation networks trained on\nstudio-collected data with 3D annotations. However, Decaf suffers from a\ntime-consuming optimization process and limited generalization capability due\nto its reliance on 3D annotations of hand-face interaction data. To address\nthese issues, we present DICE, the first end-to-end method for\nDeformation-aware hand-face Interaction reCovEry from a single image. DICE\nestimates the poses of hands and faces, contacts, and deformations\nsimultaneously using a Transformer-based architecture. It features\ndisentangling the regression of local deformation fields and global mesh vertex\nlocations into two network branches, enhancing deformation and contact\nestimation for precise and robust hand-face mesh recovery. To improve\ngeneralizability, we propose a weakly-supervised training approach that\naugments the training set using in-the-wild images without 3D ground-truth\nannotations, employing the depths of 2D keypoints estimated by off-the-shelf\nmodels and adversarial priors of poses for supervision. Our experiments\ndemonstrate that DICE achieves state-of-the-art performance on a standard\nbenchmark and in-the-wild data in terms of accuracy and physical plausibility.\nAdditionally, our method operates at an interactive rate (20 fps) on an Nvidia\n4090 GPU, whereas Decaf requires more than 15 seconds for a single image. Our\ncode will be publicly available upon publication.\n", "date": "2024-06-26", "categories": []}
{"id": "2503.11194", "title": "Online Test-time Adaptation for 3D Human Pose Estimation: A Practical\n  Perspective with Estimated 2D Poses", "abstract": "  Online test-time adaptation for 3D human pose estimation is used for video\nstreams that differ from training data. Ground truth 2D poses are used for\nadaptation, but only estimated 2D poses are available in practice. This paper\naddresses adapting models to streaming videos with estimated 2D poses.\nComparing adaptations reveals the challenge of limiting estimation errors while\npreserving accurate pose information. To this end, we propose adaptive\naggregation, a two-stage optimization, and local augmentation for handling\nvarying levels of estimated pose error. First, we perform adaptive aggregation\nacross videos to initialize the model state with labeled representative\nsamples. Within each video, we use a two-stage optimization to benefit from 2D\nfitting while minimizing the impact of erroneous updates. Second, we employ\nlocal augmentation, using adjacent confident samples to update the model before\nadapting to the current non-confident sample. Our method surpasses\nstate-of-the-art by a large margin, advancing adaptation towards more practical\nsettings of using estimated 2D poses.\n", "date": "2025-03-14", "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": "2503.09293", "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": "2503.09537", "title": "GenHPE: Generative Counterfactuals for 3D Human Pose Estimation with\n  Radio Frequency Signals", "abstract": "  Human pose estimation (HPE) detects the positions of human body joints for\nvarious applications. Compared to using cameras, HPE using radio frequency (RF)\nsignals is non-intrusive and more robust to adverse conditions, exploiting the\nsignal variations caused by human interference. However, existing studies focus\non single-domain HPE confined by domain-specific confounders, which cannot\ngeneralize to new domains and result in diminished HPE performance.\nSpecifically, the signal variations caused by different human body parts are\nentangled, containing subject-specific confounders. RF signals are also\nintertwined with environmental noise, involving environment-specific\nconfounders. In this paper, we propose GenHPE, a 3D HPE approach that generates\ncounterfactual RF signals to eliminate domain-specific confounders. GenHPE\ntrains generative models conditioned on human skeleton labels, learning how\nhuman body parts and confounders interfere with RF signals. We manipulate\nskeleton labels (i.e., removing body parts) as counterfactual conditions for\ngenerative models to synthesize counterfactual RF signals. The differences\nbetween counterfactual signals approximately eliminate domain-specific\nconfounders and regularize an encoder-decoder model to learn domain-independent\nrepresentations. Such representations help GenHPE generalize to new\nsubjects/environments for cross-domain 3D HPE. We evaluate GenHPE on three\npublic datasets from WiFi, ultra-wideband, and millimeter wave. Experimental\nresults show that GenHPE outperforms state-of-the-art methods and reduces\nestimation errors by up to 52.2mm for cross-subject HPE and 10.6mm for\ncross-environment HPE.\n", "date": "2025-03-12", "categories": []}
{"id": "2503.07597", "title": "HumanMM: Global Human Motion Recovery from Multi-shot Videos", "abstract": "  In this paper, we present a novel framework designed to reconstruct\nlong-sequence 3D human motion in the world coordinates from in-the-wild videos\nwith multiple shot transitions. Such long-sequence in-the-wild motions are\nhighly valuable to applications such as motion generation and motion\nunderstanding, but are of great challenge to be recovered due to abrupt shot\ntransitions, partial occlusions, and dynamic backgrounds presented in such\nvideos. Existing methods primarily focus on single-shot videos, where\ncontinuity is maintained within a single camera view, or simplify multi-shot\nalignment in camera space only. In this work, we tackle the challenges by\nintegrating an enhanced camera pose estimation with Human Motion Recovery (HMR)\nby incorporating a shot transition detector and a robust alignment module for\naccurate pose and orientation continuity across shots. By leveraging a custom\nmotion integrator, we effectively mitigate the problem of foot sliding and\nensure temporal consistency in human pose. Extensive evaluations on our created\nmulti-shot dataset from public 3D human datasets demonstrate the robustness of\nour method in reconstructing realistic human motion in world coordinates.\n", "date": "2025-03-10", "categories": []}
{"id": "2501.18478", "title": "SimpleDepthPose: Fast and Reliable Human Pose Estimation with\n  RGBD-Images", "abstract": "  In the rapidly advancing domain of computer vision, accurately estimating the\nposes of multiple individuals from various viewpoints remains a significant\nchallenge, especially when reliability is a key requirement. This paper\nintroduces a novel algorithm that excels in multi-view, multi-person pose\nestimation by incorporating depth information. An extensive evaluation\ndemonstrates that the proposed algorithm not only generalizes well to unseen\ndatasets, and shows a fast runtime performance, but also is adaptable to\ndifferent keypoints. To support further research, all of the work is publicly\naccessible.\n", "date": "2025-01-30", "categories": []}
{"id": "2503.05365", "title": "Multi-Grained Feature Pruning for Video-Based Human Pose Estimation", "abstract": "  Human pose estimation, with its broad applications in action recognition and\nmotion capture, has experienced significant advancements. However, current\nTransformer-based methods for video pose estimation often face challenges in\nmanaging redundant temporal information and achieving fine-grained perception\nbecause they only focus on processing low-resolution features. To address these\nchallenges, we propose a novel multi-scale resolution framework that encodes\nspatio-temporal representations at varying granularities and executes\nfine-grained perception compensation. Furthermore, we employ a density peaks\nclustering method to dynamically identify and prioritize tokens that offer\nimportant semantic information. This strategy effectively prunes redundant\nfeature tokens, especially those arising from multi-frame features, thereby\noptimizing computational efficiency without sacrificing semantic richness.\nEmpirically, it sets new benchmarks for both performance and efficiency on\nthree large-scale datasets. Our method achieves a 93.8% improvement in\ninference speed compared to the baseline, while also enhancing pose estimation\naccuracy, reaching 87.4 mAP on the PoseTrack2017 dataset.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.00232", "title": "Transformers with Joint Tokens and Local-Global Attention for Efficient\n  Human Pose Estimation", "abstract": "  Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have led\nto significant progress in 2D body pose estimation. However, achieving a good\nbalance between accuracy, efficiency, and robustness remains a challenge. For\ninstance, CNNs are computationally efficient but struggle with long-range\ndependencies, while ViTs excel in capturing such dependencies but suffer from\nquadratic computational complexity. This paper proposes two ViT-based models\nfor accurate, efficient, and robust 2D pose estimation. The first one,\nEViTPose, operates in a computationally efficient manner without sacrificing\naccuracy by utilizing learnable joint tokens to select and process a subset of\nthe most important body patches, enabling us to control the trade-off between\naccuracy and efficiency by changing the number of patches to be processed. The\nsecond one, UniTransPose, while not allowing for the same level of direct\ncontrol over the trade-off, efficiently handles multiple scales by combining\n(1) an efficient multi-scale transformer encoder that uses both local and\nglobal attention with (2) an efficient sub-pixel CNN decoder for better speed\nand accuracy. Moreover, by incorporating all joints from different benchmarks\ninto a unified skeletal representation, we train robust methods that learn from\nmultiple datasets simultaneously and perform well across a range of scenarios\n-- including pose variations, lighting conditions, and occlusions. Experiments\non six benchmarks demonstrate that the proposed methods significantly\noutperform state-of-the-art methods while improving computational efficiency.\nEViTPose exhibits a significant decrease in computational complexity (30% to\n44% less in GFLOPs) with a minimal drop of accuracy (0% to 3.5% less), and\nUniTransPose achieves accuracy improvements ranging from 0.9% to 43.8% across\nthese benchmarks.\n", "date": "2025-02-28", "categories": []}
{"id": "2503.00389", "title": "BGM2Pose: Active 3D Human Pose Estimation with Non-Stationary Sounds", "abstract": "  We propose BGM2Pose, a non-invasive 3D human pose estimation method using\narbitrary music (e.g., background music) as active sensing signals. Unlike\nexisting approaches that significantly limit practicality by employing\nintrusive chirp signals within the audible range, our method utilizes natural\nmusic that causes minimal discomfort to humans. Estimating human poses from\nstandard music presents significant challenges. In contrast to sound sources\nspecifically designed for measurement, regular music varies in both volume and\npitch. These dynamic changes in signals caused by music are inevitably mixed\nwith alterations in the sound field resulting from human motion, making it hard\nto extract reliable cues for pose estimation. To address these challenges,\nBGM2Pose introduces a Contrastive Pose Extraction Module that employs\ncontrastive learning and hard negative sampling to eliminate musical components\nfrom the recorded data, isolating the pose information. Additionally, we\npropose a Frequency-wise Attention Module that enables the model to focus on\nsubtle acoustic variations attributable to human movement by dynamically\ncomputing attention across frequency bands. Experiments suggest that our method\noutperforms the existing methods, demonstrating substantial potential for\nreal-world applications. Our datasets and code will be made publicly available.\n", "date": "2025-03-01", "categories": []}
{"id": "2408.02922", "title": "Pose Magic: Efficient and Temporally Consistent Human Pose Estimation\n  with a Hybrid Mamba-GCN Network", "abstract": "  Current state-of-the-art (SOTA) methods in 3D Human Pose Estimation (HPE) are\nprimarily based on Transformers. However, existing Transformer-based 3D HPE\nbackbones often encounter a trade-off between accuracy and computational\nefficiency. To resolve the above dilemma, in this work, we leverage recent\nadvances in state space models and utilize Mamba for high-quality and efficient\nlong-range modeling. Nonetheless, Mamba still faces challenges in precisely\nexploiting local dependencies between joints. To address these issues, we\npropose a new attention-free hybrid spatiotemporal architecture named Hybrid\nMamba-GCN (Pose Magic). This architecture introduces local enhancement with GCN\nby capturing relationships between neighboring joints, thus producing new\nrepresentations to complement Mamba's outputs. By adaptively fusing\nrepresentations from Mamba and GCN, Pose Magic demonstrates superior capability\nin learning the underlying 3D structure. To meet the requirements of real-time\ninference, we also provide a fully causal version. Extensive experiments show\nthat Pose Magic achieves new SOTA results ($\\downarrow 0.9 mm$) while saving\n$74.1\\%$ FLOPs. In addition, Pose Magic exhibits optimal motion consistency and\nthe ability to generalize to unseen sequence lengths.\n", "date": "2024-08-06", "categories": []}
{"id": "2501.08408", "title": "Leveraging 2D Masked Reconstruction for Domain Adaptation of 3D Pose\n  Estimation", "abstract": "  RGB-based 3D pose estimation methods have been successful with the\ndevelopment of deep learning and the emergence of high-quality 3D pose\ndatasets. However, most existing methods do not operate well for testing images\nwhose distribution is far from that of training data. However, most existing\nmethods do not operate well for testing images whose distribution is far from\nthat of training data. This problem might be alleviated by involving diverse\ndata during training, however it is non-trivial to collect such diverse data\nwith corresponding labels (i.e. 3D pose). In this paper, we introduced an\nunsupervised domain adaptation framework for 3D pose estimation that utilizes\nthe unlabeled data in addition to labeled data via masked image modeling (MIM)\nframework. Foreground-centric reconstruction and attention regularization are\nfurther proposed to increase the effectiveness of unlabeled data usage.\nExperiments are conducted on the various datasets in human and hand pose\nestimation tasks, especially using the cross-domain scenario. We demonstrated\nthe effectiveness of ours by achieving the state-of-the-art accuracy on all\ndatasets.\n", "date": "2025-01-14", "categories": []}
{"id": "2407.06141", "title": "CHAMP: Conformalized 3D Human Multi-Hypothesis Pose Estimators", "abstract": "  We introduce CHAMP, a novel method for learning sequence-to-sequence,\nmulti-hypothesis 3D human poses from 2D keypoints by leveraging a conditional\ndistribution with a diffusion model. To predict a single output 3D pose\nsequence, we generate and aggregate multiple 3D pose hypotheses. For better\naggregation results, we develop a method to score these hypotheses during\ntraining, effectively integrating conformal prediction into the learning\nprocess. This process results in a differentiable conformal predictor that is\ntrained end2end with the 3D pose estimator. Post-training, the learned scoring\nmodel is used as the conformity score, and the 3D pose estimator is combined\nwith a conformal predictor to select the most accurate hypotheses for\ndownstream aggregation. Our results indicate that using a simple mean\naggregation on the conformal prediction-filtered hypotheses set yields\ncompetitive results. When integrated with more sophisticated aggregation\ntechniques, our method achieves state-of-the-art performance across various\nmetrics and datasets while inheriting the probabilistic guarantees of conformal\nprediction.\n", "date": "2024-05-27", "categories": []}
{"id": "2502.16419", "title": "DeProPose: Deficiency-Proof 3D Human Pose Estimation via Adaptive\n  Multi-View Fusion", "abstract": "  3D human pose estimation has wide applications in fields such as intelligent\nsurveillance, motion capture, and virtual reality. However, in real-world\nscenarios, issues such as occlusion, noise interference, and missing viewpoints\ncan severely affect pose estimation. To address these challenges, we introduce\nthe task of Deficiency-Aware 3D Pose Estimation. Traditional 3D pose estimation\nmethods often rely on multi-stage networks and modular combinations, which can\nlead to cumulative errors and increased training complexity, making them unable\nto effectively address deficiency-aware estimation. To this end, we propose\nDeProPose, a flexible method that simplifies the network architecture to reduce\ntraining complexity and avoid information loss in multi-stage designs.\nAdditionally, the model innovatively introduces a multi-view feature fusion\nmechanism based on relative projection error, which effectively utilizes\ninformation from multiple viewpoints and dynamically assigns weights, enabling\nefficient integration and enhanced robustness to overcome deficiency-aware 3D\nPose Estimation challenges. Furthermore, to thoroughly evaluate this end-to-end\nmulti-view 3D human pose estimation model and to advance research on\nocclusion-related challenges, we have developed a novel 3D human pose\nestimation dataset, termed the Deficiency-Aware 3D Pose Estimation (DA-3DPE)\ndataset. This dataset encompasses a wide range of deficiency scenarios,\nincluding noise interference, missing viewpoints, and occlusion challenges.\nCompared to state-of-the-art methods, DeProPose not only excels in addressing\nthe deficiency-aware problem but also shows improvement in conventional\nscenarios, providing a powerful and user-friendly solution for 3D human pose\nestimation. The source code will be available at\nhttps://github.com/WUJINHUAN/DeProPose.\n", "date": "2025-02-23", "categories": []}
{"id": "2502.10616", "title": "Learning semantical dynamics and spatiotemporal collaboration for human\n  pose estimation in video", "abstract": "  Temporal modeling and spatio-temporal collaboration are pivotal techniques\nfor video-based human pose estimation. Most state-of-the-art methods adopt\noptical flow or temporal difference, learning local visual content\ncorrespondence across frames at the pixel level, to capture motion dynamics.\nHowever, such a paradigm essentially relies on localized pixel-to-pixel\nsimilarity, which neglects the semantical correlations among frames and is\nvulnerable to image quality degradations (e.g. occlusions or blur). Moreover,\nexisting approaches often combine motion and spatial (appearance) features via\nsimple concatenation or summation, leading to practical challenges in fully\nleveraging these distinct modalities. In this paper, we present a novel\nframework that learns multi-level semantical dynamics and dense spatio-temporal\ncollaboration for multi-frame human pose estimation. Specifically, we first\ndesign a Multi-Level Semantic Motion Encoder using a multi-masked context and\npose reconstruction strategy. This strategy stimulates the model to explore\nmulti-granularity spatiotemporal semantic relationships among frames by\nprogressively masking the features of (patch) cubes and frames. We further\nintroduce a Spatial-Motion Mutual Learning module which densely propagates and\nconsolidates context information from spatial and motion features to enhance\nthe capability of the model. Extensive experiments demonstrate that our\napproach sets new state-of-the-art results on three benchmark datasets,\nPoseTrack2017, PoseTrack2018, and PoseTrack21.\n", "date": "2025-02-15", "categories": []}
{"id": "2502.07869", "title": "EventEgo3D++: 3D Human Motion Capture from a Head-Mounted Event Camera", "abstract": "  Monocular egocentric 3D human motion capture remains a significant challenge,\nparticularly under conditions of low lighting and fast movements, which are\ncommon in head-mounted device applications. Existing methods that rely on RGB\ncameras often fail under these conditions. To address these limitations, we\nintroduce EventEgo3D++, the first approach that leverages a monocular event\ncamera with a fisheye lens for 3D human motion capture. Event cameras excel in\nhigh-speed scenarios and varying illumination due to their high temporal\nresolution, providing reliable cues for accurate 3D human motion capture.\nEventEgo3D++ leverages the LNES representation of event streams to enable\nprecise 3D reconstructions. We have also developed a mobile head-mounted device\n(HMD) prototype equipped with an event camera, capturing a comprehensive\ndataset that includes real event observations from both controlled studio\nenvironments and in-the-wild settings, in addition to a synthetic dataset.\nAdditionally, to provide a more holistic dataset, we include allocentric RGB\nstreams that offer different perspectives of the HMD wearer, along with their\ncorresponding SMPL body model. Our experiments demonstrate that EventEgo3D++\nachieves superior 3D accuracy and robustness compared to existing solutions,\neven in challenging conditions. Moreover, our method supports real-time 3D pose\nupdates at a rate of 140Hz. This work is an extension of the EventEgo3D\napproach (CVPR 2024) and further advances the state of the art in egocentric 3D\nhuman motion capture. For more details, visit the project page at\nhttps://eventego3d.mpi-inf.mpg.de.\n", "date": "2025-02-11", "categories": []}
{"id": "2411.04469", "title": "FreeCap: Hybrid Calibration-Free Motion Capture in Open Environments", "abstract": "  We propose a novel hybrid calibration-free method FreeCap to accurately\ncapture global multi-person motions in open environments. Our system combines a\nsingle LiDAR with expandable moving cameras, allowing for flexible and precise\nmotion estimation in a unified world coordinate. In particular, We introduce a\nlocal-to-global pose-aware cross-sensor human-matching module that predicts the\nalignment among each sensor, even in the absence of calibration. Additionally,\nour coarse-to-fine sensor-expandable pose optimizer further optimizes the 3D\nhuman key points and the alignments, it is also capable of incorporating\nadditional cameras to enhance accuracy. Extensive experiments on Human-M3 and\nFreeMotion datasets demonstrate that our method significantly outperforms\nstate-of-the-art single-modal methods, offering an expandable and efficient\nsolution for multi-person motion capture across various applications.\n", "date": "2024-11-07", "categories": []}
{"id": "2502.04483", "title": "Measuring Physical Plausibility of 3D Human Poses Using Physics\n  Simulation", "abstract": "  Modeling humans in physical scenes is vital for understanding\nhuman-environment interactions for applications involving augmented reality or\nassessment of human actions from video (e.g. sports or physical\nrehabilitation). State-of-the-art literature begins with a 3D human pose, from\nmonocular or multiple views, and uses this representation to ground the person\nwithin a 3D world space. While standard metrics for accuracy capture joint\nposition errors, they do not consider physical plausibility of the 3D pose.\nThis limitation has motivated researchers to propose other metrics evaluating\njitter, floor penetration, and unbalanced postures. Yet, these approaches\nmeasure independent instances of errors and are not representative of balance\nor stability during motion. In this work, we propose measuring physical\nplausibility from within physics simulation. We introduce two metrics to\ncapture the physical plausibility and stability of predicted 3D poses from any\n3D Human Pose Estimation model. Using physics simulation, we discover\ncorrelations with existing plausibility metrics and measuring stability during\nmotion. We evaluate and compare the performances of two state-of-the-art\nmethods, a multi-view triangulated baseline, and ground truth 3D markers from\nthe Human3.6m dataset.\n", "date": "2025-02-06", "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": "2502.02936", "title": "Every Angle Is Worth A Second Glance: Mining Kinematic Skeletal\n  Structures from Multi-view Joint Cloud", "abstract": "  Multi-person motion capture over sparse angular observations is a challenging\nproblem under interference from both self- and mutual-occlusions. Existing\nworks produce accurate 2D joint detection, however, when these are triangulated\nand lifted into 3D, available solutions all struggle in selecting the most\naccurate candidates and associating them to the correct joint type and target\nidentity. As such, in order to fully utilize all accurate 2D joint location\ninformation, we propose to independently triangulate between all same-typed 2D\njoints from all camera views regardless of their target ID, forming the Joint\nCloud. Joint Cloud consist of both valid joints lifted from the same joint type\nand target ID, as well as falsely constructed ones that are from different 2D\nsources. These redundant and inaccurate candidates are processed over the\nproposed Joint Cloud Selection and Aggregation Transformer (JCSAT) involving\nthree cascaded encoders which deeply explore the trajectile, skeletal\nstructural, and view-dependent correlations among all 3D point candidates in\nthe cross-embedding space. An Optimal Token Attention Path (OTAP) module is\nproposed which subsequently selects and aggregates informative features from\nthese redundant observations for the final prediction of human motion. To\ndemonstrate the effectiveness of JCSAT, we build and publish a new multi-person\nmotion capture dataset BUMocap-X with complex interactions and severe\nocclusions. Comprehensive experiments over the newly presented as well as\nbenchmark datasets validate the effectiveness of the proposed framework, which\noutperforms all existing state-of-the-art methods, especially under challenging\nocclusion scenarios.\n", "date": "2025-02-05", "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": "2501.15073", "title": "SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled\n  Videos", "abstract": "  Human pose estimation in videos remains a challenge, largely due to the\nreliance on extensive manual annotation of large datasets, which is expensive\nand labor-intensive. Furthermore, existing approaches often struggle to capture\nlong-range temporal dependencies and overlook the complementary relationship\nbetween temporal pose heatmaps and visual features. To address these\nlimitations, we introduce STDPose, a novel framework that enhances human pose\nestimation by learning spatiotemporal dynamics in sparsely-labeled videos.\nSTDPose incorporates two key innovations: 1) A novel Dynamic-Aware Mask to\ncapture long-range motion context, allowing for a nuanced understanding of pose\nchanges. 2) A system for encoding and aggregating spatiotemporal\nrepresentations and motion dynamics to effectively model spatiotemporal\nrelationships, improving the accuracy and robustness of pose estimation.\nSTDPose establishes a new performance benchmark for both video pose propagation\n(i.e., propagating pose annotations from labeled frames to unlabeled frames)\nand pose estimation tasks, across three large-scale evaluation datasets.\nAdditionally, utilizing pseudo-labels generated by pose propagation, STDPose\nachieves competitive performance with only 26.7% labeled data.\n", "date": "2025-01-25", "categories": []}
{"id": "2501.16146", "title": "Toward Efficient Generalization in 3D Human Pose Estimation via a\n  Canonical Domain Approach", "abstract": "  Recent advancements in deep learning methods have significantly improved the\nperformance of 3D Human Pose Estimation (HPE). However, performance degradation\ncaused by domain gaps between source and target domains remains a major\nchallenge to generalization, necessitating extensive data augmentation and/or\nfine-tuning for each specific target domain. To address this issue more\nefficiently, we propose a novel canonical domain approach that maps both the\nsource and target domains into a unified canonical domain, alleviating the need\nfor additional fine-tuning in the target domain. To construct the canonical\ndomain, we introduce a canonicalization process to generate a novel canonical\n2D-3D pose mapping that ensures 2D-3D pose consistency and simplifies 2D-3D\npose patterns, enabling more efficient training of lifting networks. The\ncanonicalization of both domains is achieved through the following steps: (1)\nin the source domain, the lifting network is trained within the canonical\ndomain; (2) in the target domain, input 2D poses are canonicalized prior to\ninference by leveraging the properties of perspective projection and known\ncamera intrinsics. Consequently, the trained network can be directly applied to\nthe target domain without requiring additional fine-tuning. Experiments\nconducted with various lifting networks and publicly available datasets (e.g.,\nHuman3.6M, Fit3D, MPI-INF-3DHP) demonstrate that the proposed method\nsubstantially improves generalization capability across datasets while using\nthe same data volume.\n", "date": "2025-01-27", "categories": []}
{"id": "2501.08446", "title": "Poseidon: A ViT-based Architecture for Multi-Frame Pose Estimation with\n  Adaptive Frame Weighting and Multi-Scale Feature Fusion", "abstract": "  Human pose estimation, a vital task in computer vision, involves detecting\nand localising human joints in images and videos. While single-frame pose\nestimation has seen significant progress, it often fails to capture the\ntemporal dynamics for understanding complex, continuous movements. We propose\nPoseidon, a novel multi-frame pose estimation architecture that extends the\nViTPose model by integrating temporal information for enhanced accuracy and\nrobustness to address these limitations. Poseidon introduces key innovations:\n(1) an Adaptive Frame Weighting (AFW) mechanism that dynamically prioritises\nframes based on their relevance, ensuring that the model focuses on the most\ninformative data; (2) a Multi-Scale Feature Fusion (MSFF) module that\naggregates features from different backbone layers to capture both fine-grained\ndetails and high-level semantics; and (3) a Cross-Attention module for\neffective information exchange between central and contextual frames, enhancing\nthe model's temporal coherence. The proposed architecture improves performance\nin complex video scenarios and offers scalability and computational efficiency\nsuitable for real-world applications. Our approach achieves state-of-the-art\nperformance on the PoseTrack21 and PoseTrack18 datasets, achieving mAP scores\nof 88.3 and 87.8, respectively, outperforming existing methods.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.14356", "title": "Causal-Inspired Multitask Learning for Video-Based Human Pose Estimation", "abstract": "  Video-based human pose estimation has long been a fundamental yet challenging\nproblem in computer vision. Previous studies focus on spatio-temporal modeling\nthrough the enhancement of architecture design and optimization strategies.\nHowever, they overlook the causal relationships in the joints, leading to\nmodels that may be overly tailored and thus estimate poorly to challenging\nscenes. Therefore, adequate causal reasoning capability, coupled with good\ninterpretability of model, are both indispensable and prerequisite for\nachieving reliable results. In this paper, we pioneer a causal perspective on\npose estimation and introduce a causal-inspired multitask learning framework,\nconsisting of two stages. \\textit{In the first stage}, we try to endow the\nmodel with causal spatio-temporal modeling ability by introducing two\nself-supervision auxiliary tasks. Specifically, these auxiliary tasks enable\nthe network to infer challenging keypoints based on observed keypoint\ninformation, thereby imbuing causal reasoning capabilities into the model and\nmaking it robust to challenging scenes. \\textit{In the second stage}, we argue\nthat not all feature tokens contribute equally to pose estimation. Prioritizing\ncausal (keypoint-relevant) tokens is crucial to achieve reliable results, which\ncould improve the interpretability of the model. To this end, we propose a\nToken Causal Importance Selection module to identify the causal tokens and\nnon-causal tokens (\\textit{e.g.}, background and objects). Additionally,\nnon-causal tokens could provide potentially beneficial cues but may be\nredundant. We further introduce a non-causal tokens clustering module to merge\nthe similar non-causal tokens. Extensive experiments show that our method\noutperforms state-of-the-art methods on three large-scale benchmark datasets.\n", "date": "2025-01-24", "categories": []}
{"id": "2501.14439", "title": "Optimizing Human Pose Estimation Through Focused Human and Joint Regions", "abstract": "  Human pose estimation has given rise to a broad spectrum of novel and\ncompelling applications, including action recognition, sports analysis, as well\nas surveillance. However, accurate video pose estimation remains an open\nchallenge. One aspect that has been overlooked so far is that existing methods\nlearn motion clues from all pixels rather than focusing on the target human\nbody, making them easily misled and disrupted by unimportant information such\nas background changes or movements of other people. Additionally, while the\ncurrent Transformer-based pose estimation methods has demonstrated impressive\nperformance with global modeling, they struggle with local context perception\nand precise positional identification. In this paper, we try to tackle these\nchallenges from three aspects: (1) We propose a bilayer Human-Keypoint Mask\nmodule that performs coarse-to-fine visual token refinement, which gradually\nzooms in on the target human body and keypoints while masking out unimportant\nfigure regions. (2) We further introduce a novel deformable cross attention\nmechanism and a bidirectional separation strategy to adaptively aggregate\nspatial and temporal motion clues from constrained surrounding contexts. (3) We\nmathematically formulate the deformable cross attention, constraining that the\nmodel focuses solely on the regions centered at the target person body.\nEmpirically, our method achieves state-of-the-art performance on three\nlarge-scale benchmark datasets. A remarkable highlight is that our method\nachieves an 84.8 mean Average Precision (mAP) on the challenging wrist joint,\nwhich significantly outperforms the 81.5 mAP achieved by the current\nstate-of-the-art method on the PoseTrack2017 dataset.\n", "date": "2025-01-24", "categories": []}
{"id": "2501.02771", "title": "WorldPose: A World Cup Dataset for Global 3D Human Pose Estimation", "abstract": "  We present WorldPose, a novel dataset for advancing research in multi-person\nglobal pose estimation in the wild, featuring footage from the 2022 FIFA World\nCup. While previous datasets have primarily focused on local poses, often\nlimited to a single person or in constrained, indoor settings, the\ninfrastructure deployed for this sporting event allows access to multiple fixed\nand moving cameras in different stadiums. We exploit the static multi-view\nsetup of HD cameras to recover the 3D player poses and motions with\nunprecedented accuracy given capture areas of more than 1.75 acres. We then\nleverage the captured players' motions and field markings to calibrate a moving\nbroadcasting camera. The resulting dataset comprises more than 80 sequences\nwith approx 2.5 million 3D poses and a total traveling distance of over 120 km.\nSubsequently, we conduct an in-depth analysis of the SOTA methods for global\npose estimation. Our experiments demonstrate that WorldPose challenges existing\nmulti-person techniques, supporting the potential for new research in this area\nand others, such as sports analysis. All pose annotations (in SMPL format),\nbroadcasting camera parameters and footage will be released for academic\nresearch purposes.\n", "date": "2025-01-06", "categories": []}
{"id": "2501.12318", "title": "BlanketGen2-Fit3D: Synthetic Blanket Augmentation Towards Improving\n  Real-World In-Bed Blanket Occluded Human Pose Estimation", "abstract": "  Human Pose Estimation (HPE) from monocular RGB images is crucial for clinical\nin-bed skeleton-based action recognition, however, it poses unique challenges\nfor HPE models due to the frequent presence of blankets occluding the person,\nwhile labeled HPE data in this scenario is scarce. To address this we introduce\nBlanketGen2-Fit3D (BG2-Fit3D), an augmentation of Fit3D dataset that contains\n1,217,312 frames with synthetic photo-realistic blankets. To generate it we\nused BlanketGen2, our new and improved version of our BlanketGen pipeline that\nsimulates synthetic blankets using ground-truth Skinned Multi-Person Linear\nmodel (SMPL) meshes and then renders them as transparent images that can be\nlayered on top of the original frames. This dataset was used in combination\nwith the original Fit3D to finetune the ViTPose-B HPE model, to evaluate\nsynthetic blanket augmentation effectiveness. The trained models were further\nevaluated on a real-world blanket occluded in-bed HPE dataset (SLP dataset).\nComparing architectures trained on only Fit3D with the ones trained with our\nsynthetic blanket augmentation the later improved pose estimation performance\non BG2-Fit3D, the synthetic blanket occluded dataset significantly to (0.977\nPercentage of Correct Keypoints (PCK), 0.149 Normalized Mean Error (NME)) with\nan absolute 4.4% PCK increase. Furthermore, the test results on SLP\ndemonstrated the utility of synthetic data augmentation by improving\nperformance by an absolute 2.3% PCK, on real-world images with the poses\noccluded by real blankets. These results show synthetic blanket augmentation\nhas the potential to improve in-bed blanket occluded HPE from RGB images. The\ndataset as well as the code will be made available to the public.\n", "date": "2025-01-21", "categories": []}
{"id": "2501.09782", "title": "SMPLest-X: Ultimate Scaling for Expressive Human Pose and Shape\n  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 focus on training innovative architectural\ndesigns on confined datasets. In this work, we investigate the impact of\nscaling up EHPS towards a family of generalist foundation models. 1) For data\nscaling, we perform a systematic investigation on 40 EHPS datasets,\nencompassing a wide range of scenarios that a model trained on any single\ndataset cannot handle. More importantly, capitalizing on insights obtained from\nthe extensive benchmarking process, we optimize our training scheme and select\ndatasets that lead to a significant leap in EHPS capabilities. Ultimately, we\nachieve diminishing returns at 10M training instances from diverse data\nsources. 2) For model scaling, we take advantage of vision transformers (up to\nViT-Huge as the backbone) to study the scaling law of model sizes in EHPS. To\nexclude the influence of algorithmic design, we base our experiments on two\nminimalist architectures: SMPLer-X, which consists of an intermediate step for\nhand and face localization, and SMPLest-X, an even simpler version that reduces\nthe network to its bare essentials and highlights significant advances in the\ncapture of articulated hands. With big data and the large model, the foundation\nmodels exhibit strong performance across diverse test benchmarks and excellent\ntransferability to even unseen environments. Moreover, our finetuning strategy\nturns the generalist into specialist models, allowing them to achieve further\nperformance boosts. Notably, our foundation models consistently deliver\nstate-of-the-art results on seven benchmarks such as AGORA, UBody, EgoBody, and\nour proposed SynHand dataset for comprehensive hand evaluation. (Code is\navailable at: https://github.com/wqyin/SMPLest-X).\n", "date": "2025-01-16", "categories": []}
{"id": "2501.09565", "title": "A New Teacher-Reviewer-Student Framework for Semi-supervised 2D Human\n  Pose Estimation", "abstract": "  Conventional 2D human pose estimation methods typically require extensive\nlabeled annotations, which are both labor-intensive and expensive. In contrast,\nsemi-supervised 2D human pose estimation can alleviate the above problems by\nleveraging a large amount of unlabeled data along with a small portion of\nlabeled data. Existing semi-supervised 2D human pose estimation methods update\nthe network through backpropagation, ignoring crucial historical information\nfrom the previous training process. Therefore, we propose a novel\nsemi-supervised 2D human pose estimation method by utilizing a newly designed\nTeacher-Reviewer-Student framework. Specifically, we first mimic the phenomenon\nthat human beings constantly review previous knowledge for consolidation to\ndesign our framework, in which the teacher predicts results to guide the\nstudent's learning and the reviewer stores important historical parameters to\nprovide additional supervision signals. Secondly, we introduce a Multi-level\nFeature Learning strategy, which utilizes the outputs from different stages of\nthe backbone to estimate the heatmap to guide network training, enriching the\nsupervisory information while effectively capturing keypoint relationships.\nFinally, we design a data augmentation strategy, i.e., Keypoint-Mix, to perturb\npose information by mixing different keypoints, thus enhancing the network's\nability to discern keypoints. Extensive experiments on publicly available\ndatasets, demonstrate our method achieves significant improvements compared to\nthe existing methods.\n", "date": "2025-01-16", "categories": []}
{"id": "2501.08815", "title": "Human Pose-Constrained UV Map Estimation", "abstract": "  UV map estimation is used in computer vision for detailed analysis of human\nposture or activity. Previous methods assign pixels to body model vertices by\ncomparing pixel descriptors independently, without enforcing global coherence\nor plausibility in the UV map. We propose Pose-Constrained Continuous Surface\nEmbeddings (PC-CSE), which integrates estimated 2D human pose into the\npixel-to-vertex assignment process. The pose provides global anatomical\nconstraints, ensuring that UV maps remain coherent while preserving local\nprecision. Evaluation on DensePose COCO demonstrates consistent improvement,\nregardless of the chosen 2D human pose model. Whole-body poses offer better\nconstraints by incorporating additional details about the hands and feet.\nConditioning UV maps with human pose reduces invalid mappings and enhances\nanatomical plausibility. In addition, we highlight inconsistencies in the\nground-truth annotations.\n", "date": "2025-01-15", "categories": []}
{"id": "2501.07800", "title": "BioPose: Biomechanically-accurate 3D Pose Estimation from Monocular\n  Videos", "abstract": "  Recent advancements in 3D human pose estimation from single-camera images and\nvideos have relied on parametric models, like SMPL. However, these models\noversimplify anatomical structures, limiting their accuracy in capturing true\njoint locations and movements, which reduces their applicability in\nbiomechanics, healthcare, and robotics. Biomechanically accurate pose\nestimation, on the other hand, typically requires costly marker-based motion\ncapture systems and optimization techniques in specialized labs. To bridge this\ngap, we propose BioPose, a novel learning-based framework for predicting\nbiomechanically accurate 3D human pose directly from monocular videos. BioPose\nincludes three key components: a Multi-Query Human Mesh Recovery model\n(MQ-HMR), a Neural Inverse Kinematics (NeurIK) model, and a 2D-informed pose\nrefinement technique. MQ-HMR leverages a multi-query deformable transformer to\nextract multi-scale fine-grained image features, enabling precise human mesh\nrecovery. NeurIK treats the mesh vertices as virtual markers, applying a\nspatial-temporal network to regress biomechanically accurate 3D poses under\nanatomical constraints. To further improve 3D pose estimations, a 2D-informed\nrefinement step optimizes the query tokens during inference by aligning the 3D\nstructure with 2D pose observations. Experiments on benchmark datasets\ndemonstrate that BioPose significantly outperforms state-of-the-art methods.\nProject website:\n\\url{https://m-usamasaleem.github.io/publication/BioPose/BioPose.html}.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.06014", "title": "Pose-independent 3D Anthropometry from Sparse Data", "abstract": "  3D digital anthropometry is the study of estimating human body measurements\nfrom 3D scans. Precise body measurements are important health indicators in the\nmedical industry, and guiding factors in the fashion, ergonomic and\nentertainment industries. The measuring protocol consists of scanning the whole\nsubject in the static A-pose, which is maintained without breathing or movement\nduring the scanning process. However, the A-pose is not easy to maintain during\nthe whole scanning process, which can last even up to a couple of minutes. This\nconstraint affects the final quality of the scan, which in turn affects the\naccuracy of the estimated body measurements obtained from methods that rely on\ndense geometric data. Additionally, this constraint makes it impossible to\ndevelop a digital anthropometry method for subjects unable to assume the\nA-pose, such as those with injuries or disabilities. We propose a method that\ncan obtain body measurements from sparse landmarks acquired in any pose. We\nmake use of the sparse landmarks of the posed subject to create\npose-independent features, and train a network to predict the body measurements\nas taken from the standard A-pose. We show that our method achieves comparable\nresults to competing methods that use dense geometry in the standard A-pose,\nbut has the capability of estimating the body measurements from any pose using\nsparse landmarks only. Finally, we address the lack of open-source 3D\nanthropometry methods by making our method available to the research community\nat https://github.com/DavidBoja/pose-independent-anthropometry.\n", "date": "2025-01-10", "categories": []}
{"id": "2312.10758", "title": "SHaRPose: Sparse High-Resolution Representation for Human Pose\n  Estimation", "abstract": "  High-resolution representation is essential for achieving good performance in\nhuman pose estimation models. To obtain such features, existing works utilize\nhigh-resolution input images or fine-grained image tokens. However, this dense\nhigh-resolution representation brings a significant computational burden. In\nthis paper, we address the following question: \"Only sparse human keypoint\nlocations are detected for human pose estimation, is it really necessary to\ndescribe the whole image in a dense, high-resolution manner?\" Based on dynamic\ntransformer models, we propose a framework that only uses Sparse\nHigh-resolution Representations for human Pose estimation (SHaRPose). In\ndetail, SHaRPose consists of two stages. At the coarse stage, the relations\nbetween image regions and keypoints are dynamically mined while a coarse\nestimation is generated. Then, a quality predictor is applied to decide whether\nthe coarse estimation results should be refined. At the fine stage, SHaRPose\nbuilds sparse high-resolution representations only on the regions related to\nthe keypoints and provides refined high-precision human pose estimations.\nExtensive experiments demonstrate the outstanding performance of the proposed\nmethod. Specifically, compared to the state-of-the-art method ViTPose, our\nmodel SHaRPose-Base achieves 77.4 AP (+0.5 AP) on the COCO validation set and\n76.7 AP (+0.5 AP) on the COCO test-dev set, and infers at a speed of\n$1.4\\times$ faster than ViTPose-Base.\n", "date": "2023-12-17", "categories": []}
{"id": "2501.02773", "title": "Unsupervised Domain Adaptation for Occlusion Resilient Human Pose\n  Estimation", "abstract": "  Occlusions are a significant challenge to human pose estimation algorithms,\noften resulting in inaccurate and anatomically implausible poses. Although\ncurrent occlusion-robust human pose estimation algorithms exhibit impressive\nperformance on existing datasets, their success is largely attributed to\nsupervised training and the availability of additional information, such as\nmultiple views or temporal continuity. Furthermore, these algorithms typically\nsuffer from performance degradation under distribution shifts. While existing\ndomain adaptive human pose estimation algorithms address this bottleneck, they\ntend to perform suboptimally when the target domain images are occluded, a\ncommon occurrence in real-life scenarios. To address these challenges, we\npropose OR-POSE: Unsupervised Domain Adaptation for Occlusion Resilient Human\nPOSE Estimation. OR-POSE is an innovative unsupervised domain adaptation\nalgorithm which effectively mitigates domain shifts and overcomes occlusion\nchallenges by employing the mean teacher framework for iterative pseudo-label\nrefinement. Additionally, OR-POSE reinforces realistic pose prediction by\nleveraging a learned human pose prior which incorporates the anatomical\nconstraints of humans in the adaptation process. Lastly, OR-POSE avoids\noverfitting to inaccurate pseudo labels generated from heavily occluded images\nby employing a novel visibility-based curriculum learning approach. This\nenables the model to gradually transition from training samples with relatively\nless occlusion to more challenging, heavily occluded samples. Extensive\nexperiments show that OR-POSE outperforms existing analogous state-of-the-art\nalgorithms by $\\sim$ 7% on challenging occluded human pose estimation datasets.\n", "date": "2025-01-06", "categories": []}
{"id": "2407.10220", "title": "PAFUSE: Part-based Diffusion for 3D Whole-Body Pose Estimation", "abstract": "  We introduce a novel approach for 3D whole-body pose estimation, addressing\nthe challenge of scale -- and deformability -- variance across body parts\nbrought by the challenge of extending the 17 major joints on the human body to\nfine-grained keypoints on the face and hands. In addition to addressing the\nchallenge of exploiting motion in unevenly sampled data, we combine stable\ndiffusion to a hierarchical part representation which predicts the relative\nlocations of fine-grained keypoints within each part (e.g., face) with respect\nto the part's local reference frame. On the H3WB dataset, our method greatly\noutperforms the current state of the art, which fails to exploit the temporal\ninformation. We also show considerable improvements compared to other\nspatiotemporal 3D human-pose estimation approaches that fail to account for the\nbody part specificities. Code is available at\nhttps://github.com/valeoai/PAFUSE.\n", "date": "2024-07-14", "categories": []}
{"id": "2501.01770", "title": "TCPFormer: Learning Temporal Correlation with Implicit Pose Proxy for 3D\n  Human Pose Estimation", "abstract": "  Recent multi-frame lifting methods have dominated the 3D human pose\nestimation. However, previous methods ignore the intricate dependence within\nthe 2D pose sequence and learn single temporal correlation. To alleviate this\nlimitation, we propose TCPFormer, which leverages an implicit pose proxy as an\nintermediate representation. Each proxy within the implicit pose proxy can\nbuild one temporal correlation therefore helping us learn more comprehensive\ntemporal correlation of human motion. Specifically, our method consists of\nthree key components: Proxy Update Module (PUM), Proxy Invocation Module (PIM),\nand Proxy Attention Module (PAM). PUM first uses pose features to update the\nimplicit pose proxy, enabling it to store representative information from the\npose sequence. PIM then invocates and integrates the pose proxy with the pose\nsequence to enhance the motion semantics of each pose. Finally, PAM leverages\nthe above mapping between the pose sequence and pose proxy to enhance the\ntemporal correlation of the whole pose sequence. Experiments on the Human3.6M\nand MPI-INF-3DHP datasets demonstrate that our proposed TCPFormer outperforms\nthe previous state-of-the-art methods.\n", "date": "2025-01-03", "categories": []}
{"id": "2501.01174", "title": "L3D-Pose: Lifting Pose for 3D Avatars from a Single Camera in the Wild", "abstract": "  While 2D pose estimation has advanced our ability to interpret body movements\nin animals and primates, it is limited by the lack of depth information,\nconstraining its application range. 3D pose estimation provides a more\ncomprehensive solution by incorporating spatial depth, yet creating extensive\n3D pose datasets for animals is challenging due to their dynamic and\nunpredictable behaviours in natural settings. To address this, we propose a\nhybrid approach that utilizes rigged avatars and the pipeline to generate\nsynthetic datasets to acquire the necessary 3D annotations for training. Our\nmethod introduces a simple attention-based MLP network for converting 2D poses\nto 3D, designed to be independent of the input image to ensure scalability for\nposes in natural environments. Additionally, we identify that existing\nanatomical keypoint detectors are insufficient for accurate pose retargeting\nonto arbitrary avatars. To overcome this, we present a lookup table based on a\ndeep pose estimation method using a synthetic collection of diverse actions\nrigged avatars perform. Our experiments demonstrate the effectiveness and\nefficiency of this lookup table-based retargeting approach. Overall, we propose\na comprehensive framework with systematically synthesized datasets for lifting\nposes from 2D to 3D and then utilize this to re-target motion from wild\nsettings onto arbitrary avatars.\n", "date": "2025-01-02", "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": "2412.19676", "title": "Optimizing Local-Global Dependencies for Accurate 3D Human Pose\n  Estimation", "abstract": "  Transformer-based methods have recently achieved significant success in 3D\nhuman pose estimation, owing to their strong ability to model long-range\ndependencies. However, relying solely on the global attention mechanism is\ninsufficient for capturing the fine-grained local details, which are crucial\nfor accurate pose estimation. To address this, we propose SSR-STF, a\ndual-stream model that effectively integrates local features with global\ndependencies to enhance 3D human pose estimation. Specifically, we introduce\nSSRFormer, a simple yet effective module that employs the skeleton selective\nrefine attention (SSRA) mechanism to capture fine-grained local dependencies in\nhuman pose sequences, complementing the global dependencies modeled by the\nTransformer. By adaptively fusing these two feature streams, SSR-STF can better\nlearn the underlying structure of human poses, overcoming the limitations of\ntraditional methods in local feature extraction. Extensive experiments on the\nHuman3.6M and MPI-INF-3DHP datasets demonstrate that SSR-STF achieves\nstate-of-the-art performance, with P1 errors of 37.4 mm and 13.2 mm\nrespectively, outperforming existing methods in both accuracy and\ngeneralization. Furthermore, the motion representations learned by our model\nprove effective in downstream tasks such as human mesh recovery. Codes are\navailable at https://github.com/poker-xu/SSR-STF.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.17377", "title": "A Plug-and-Play Physical Motion Restoration Approach for In-the-Wild\n  High-Difficulty Motions", "abstract": "  Extracting physically plausible 3D human motion from videos is a critical\ntask. Although existing simulation-based motion imitation methods can enhance\nthe physical quality of daily motions estimated from monocular video capture,\nextending this capability to high-difficulty motions remains an open challenge.\nThis can be attributed to some flawed motion clips in video-based motion\ncapture results and the inherent complexity in modeling high-difficulty\nmotions. Therefore, sensing the advantage of segmentation in localizing human\nbody, we introduce a mask-based motion correction module (MCM) that leverages\nmotion context and video mask to repair flawed motions, producing\nimitation-friendly motions; and propose a physics-based motion transfer module\n(PTM), which employs a pretrain and adapt approach for motion imitation,\nimproving physical plausibility with the ability to handle in-the-wild and\nchallenging motions. Our approach is designed as a plug-and-play module to\nphysically refine the video motion capture results, including high-difficulty\nin-the-wild motions. Finally, to validate our approach, we collected a\nchallenging in-the-wild test set to establish a benchmark, and our method has\ndemonstrated effectiveness on both the new benchmark and existing public\ndatasets.https://physicalmotionrestoration.github.io\n", "date": "2024-12-23", "categories": []}
{"id": "2410.18723", "title": "VoxelKeypointFusion: Generalizable Multi-View Multi-Person Pose\n  Estimation", "abstract": "  In the rapidly evolving field of computer vision, the task of accurately\nestimating the poses of multiple individuals from various viewpoints presents a\nformidable challenge, especially if the estimations should be reliable as well.\nThis work presents an extensive evaluation of the generalization capabilities\nof multi-view multi-person pose estimators to unseen datasets and presents a\nnew algorithm with strong performance in this task. It also studies the\nimprovements by additionally using depth information. Since the new approach\ncan not only generalize well to unseen datasets, but also to different\nkeypoints, the first multi-view multi-person whole-body estimator is presented.\nTo support further research on those topics, all of the work is publicly\naccessible.\n", "date": "2024-10-24", "categories": []}
{"id": "2412.15966", "title": "Monkey Transfer Learning Can Improve Human Pose Estimation", "abstract": "  In this study, we investigated whether transfer learning from macaque monkeys\ncould improve human pose estimation. Current state-of-the-art pose estimation\ntechniques, often employing deep neural networks, can match human annotation in\nnon-clinical datasets. However, they underperform in novel situations, limiting\ntheir generalisability to clinical populations with pathological movement\npatterns. Clinical datasets are not widely available for AI training due to\nethical challenges and a lack of data collection. We observe that data from\nother species may be able to bridge this gap by exposing the network to a\nbroader range of motion cues. We found that utilising data from other species\nand undertaking transfer learning improved human pose estimation in terms of\nprecision and recall compared to the benchmark, which was trained on humans\nonly. Compared to the benchmark, fewer human training examples were needed for\nthe transfer learning approach (1,000 vs 19,185). These results suggest that\nmacaque pose estimation can improve human pose estimation in clinical\nsituations. Future work should further explore the utility of pose estimation\ntrained with monkey data in clinical populations.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.16155", "title": "Can Generative Video Models Help Pose Estimation?", "abstract": "  Pairwise pose estimation from images with little or no overlap is an open\nchallenge in computer vision. Existing methods, even those trained on\nlarge-scale datasets, struggle in these scenarios due to the lack of\nidentifiable correspondences or visual overlap. Inspired by the human ability\nto infer spatial relationships from diverse scenes, we propose a novel\napproach, InterPose, that leverages the rich priors encoded within pre-trained\ngenerative video models. We propose to use a video model to hallucinate\nintermediate frames between two input images, effectively creating a dense,\nvisual transition, which significantly simplifies the problem of pose\nestimation. Since current video models can still produce implausible motion or\ninconsistent geometry, we introduce a self-consistency score that evaluates the\nconsistency of pose predictions from sampled videos. We demonstrate that our\napproach generalizes among three state-of-the-art video models and show\nconsistent improvements over the state-of-the-art DUSt3R on four diverse\ndatasets encompassing indoor, outdoor, and object-centric scenes. Our findings\nsuggest a promising avenue for improving pose estimation models by leveraging\nlarge generative models trained on vast amounts of video data, which is more\nreadily available than 3D data. See our project page for results:\nhttps://inter-pose.github.io/.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.14444", "title": "GenHMR: Generative Human Mesh Recovery", "abstract": "  Human mesh recovery (HMR) is crucial in many computer vision applications;\nfrom health to arts and entertainment. HMR from monocular images has\npredominantly been addressed by deterministic methods that output a single\nprediction for a given 2D image. However, HMR from a single image is an\nill-posed problem due to depth ambiguity and occlusions. Probabilistic methods\nhave attempted to address this by generating and fusing multiple plausible 3D\nreconstructions, but their performance has often lagged behind deterministic\napproaches. In this paper, we introduce GenHMR, a novel generative framework\nthat reformulates monocular HMR as an image-conditioned generative task,\nexplicitly modeling and mitigating uncertainties in the 2D-to-3D mapping\nprocess. GenHMR comprises two key components: (1) a pose tokenizer to convert\n3D human poses into a sequence of discrete tokens in a latent space, and (2) an\nimage-conditional masked transformer to learn the probabilistic distributions\nof the pose tokens, conditioned on the input image prompt along with randomly\nmasked token sequence. During inference, the model samples from the learned\nconditional distribution to iteratively decode high-confidence pose tokens,\nthereby reducing 3D reconstruction uncertainties. To further refine the\nreconstruction, a 2D pose-guided refinement technique is proposed to directly\nfine-tune the decoded pose tokens in the latent space, which forces the\nprojected 3D body mesh to align with the 2D pose clues. Experiments on\nbenchmark datasets demonstrate that GenHMR significantly outperforms\nstate-of-the-art methods. Project website can be found at\nhttps://m-usamasaleem.github.io/publication/GenHMR/GenHMR.html\n", "date": "2024-12-19", "categories": []}
{"id": "2412.13454", "title": "Pre-training a Density-Aware Pose Transformer for Robust LiDAR-based 3D\n  Human Pose Estimation", "abstract": "  With the rapid development of autonomous driving, LiDAR-based 3D Human Pose\nEstimation (3D HPE) is becoming a research focus. However, due to the noise and\nsparsity of LiDAR-captured point clouds, robust human pose estimation remains\nchallenging. Most of the existing methods use temporal information, multi-modal\nfusion, or SMPL optimization to correct biased results. In this work, we try to\nobtain sufficient information for 3D HPE only by modeling the intrinsic\nproperties of low-quality point clouds. Hence, a simple yet powerful method is\nproposed, which provides insights both on modeling and augmentation of point\nclouds. Specifically, we first propose a concise and effective density-aware\npose transformer (DAPT) to get stable keypoint representations. By using a set\nof joint anchors and a carefully designed exchange module, valid information is\nextracted from point clouds with different densities. Then 1D heatmaps are\nutilized to represent the precise locations of the keypoints. Secondly, a\ncomprehensive LiDAR human synthesis and augmentation method is proposed to\npre-train the model, enabling it to acquire a better human body prior. We\nincrease the diversity of point clouds by randomly sampling human positions and\norientations and by simulating occlusions through the addition of laser-level\nmasks. Extensive experiments have been conducted on multiple datasets,\nincluding IMU-annotated LidarHuman26M, SLOPER4D, and manually annotated Waymo\nOpen Dataset v2.0 (Waymo), HumanM3. Our method demonstrates SOTA performance in\nall scenarios. In particular, compared with LPFormer on Waymo, we reduce the\naverage MPJPE by $10.0mm$. Compared with PRN on SLOPER4D, we notably reduce the\naverage MPJPE by $20.7mm$.\n", "date": "2024-12-18", "categories": []}
{"id": "2410.03665", "title": "Estimating Body and Hand Motion in an Ego-sensed World", "abstract": "  We present EgoAllo, a system for human motion estimation from a head-mounted\ndevice. Using only egocentric SLAM poses and images, EgoAllo guides sampling\nfrom a conditional diffusion model to estimate 3D body pose, height, and hand\nparameters that capture a device wearer's actions in the allocentric coordinate\nframe of the scene. To achieve this, our key insight is in representation: we\npropose spatial and temporal invariance criteria for improving model\nperformance, from which we derive a head motion conditioning parameterization\nthat improves estimation by up to 18%. We also show how the bodies estimated by\nour system can improve hand estimation: the resulting kinematic and temporal\nconstraints can reduce world-frame errors in single-frame estimates by 40%.\nProject page: https://egoallo.github.io/\n", "date": "2024-10-04", "categories": []}
{"id": "2408.03540", "title": "PoseMamba: Monocular 3D Human Pose Estimation with Bidirectional\n  Global-Local Spatio-Temporal State Space Model", "abstract": "  Transformers have significantly advanced the field of 3D human pose\nestimation (HPE). However, existing transformer-based methods primarily use\nself-attention mechanisms for spatio-temporal modeling, leading to a quadratic\ncomplexity, unidirectional modeling of spatio-temporal relationships, and\ninsufficient learning of spatial-temporal correlations. Recently, the Mamba\narchitecture, utilizing the state space model (SSM), has exhibited superior\nlong-range modeling capabilities in a variety of vision tasks with linear\ncomplexity. In this paper, we propose PoseMamba, a novel purely SSM-based\napproach with linear complexity for 3D human pose estimation in monocular\nvideo. Specifically, we propose a bidirectional global-local spatio-temporal\nSSM block that comprehensively models human joint relations within individual\nframes as well as temporal correlations across frames. Within this\nbidirectional global-local spatio-temporal SSM block, we introduce a reordering\nstrategy to enhance the local modeling capability of the SSM. This strategy\nprovides a more logical geometric scanning order and integrates it with the\nglobal SSM, resulting in a combined global-local spatial scan. We have\nquantitatively and qualitatively evaluated our approach using two benchmark\ndatasets: Human3.6M and MPI-INF-3DHP. Extensive experiments demonstrate that\nPoseMamba achieves state-of-the-art performance on both datasets while\nmaintaining a smaller model size and reducing computational costs. The code and\nmodels will be released.\n", "date": "2024-08-07", "categories": []}
{"id": "2412.10431", "title": "CUPS: Improving Human Pose-Shape Estimators with Conformalized Deep\n  Uncertainty", "abstract": "  We introduce CUPS, a novel method for learning sequence-to-sequence 3D human\nshapes and poses from RGB videos with uncertainty quantification. To improve on\ntop of prior work, we develop a method to generate and score multiple\nhypotheses during training, effectively integrating uncertainty quantification\ninto the learning process. This process results in a deep uncertainty function\nthat is trained end-to-end with the 3D pose estimator. Post-training, the\nlearned deep uncertainty model is used as the conformity score, which can be\nused to calibrate a conformal predictor in order to assess the quality of the\noutput prediction. Since the data in human pose-shape learning is not fully\nexchangeable, we also present two practical bounds for the coverage gap in\nconformal prediction, developing theoretical backing for the uncertainty bound\nof our model. Our results indicate that by taking advantage of deep uncertainty\nwith conformal prediction, our method achieves state-of-the-art performance\nacross various metrics and datasets while inheriting the probabilistic\nguarantees of conformal prediction.\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": "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": "2412.08640", "title": "BLADE: Single-view Body Mesh Learning through Accurate Depth Estimation", "abstract": "  Single-image human mesh recovery is a challenging task due to the ill-posed\nnature of simultaneous body shape, pose, and camera estimation. Existing\nestimators work well on images taken from afar, but they break down as the\nperson moves close to the camera. Moreover, current methods fail to achieve\nboth accurate 3D pose and 2D alignment at the same time. Error is mainly\nintroduced by inaccurate perspective projection heuristically derived from\northographic parameters. To resolve this long-standing challenge, we present\nour method BLADE which accurately recovers perspective parameters from a single\nimage without heuristic assumptions. We start from the inverse relationship\nbetween perspective distortion and the person's Z-translation Tz, and we show\nthat Tz can be reliably estimated from the image. We then discuss the important\nrole of Tz for accurate human mesh recovery estimated from close-range images.\nFinally, we show that, once Tz and the 3D human mesh are estimated, one can\naccurately recover the focal length and full 3D translation. Extensive\nexperiments on standard benchmarks and real-world close-range images show that\nour method is the first to accurately recover projection parameters from a\nsingle image, and consequently attain state-of-the-art accuracy on 3D pose\nestimation and 2D alignment for a wide range of images.\nhttps://research.nvidia.com/labs/amri/projects/blade/\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": "2412.04665", "title": "ProPLIKS: Probablistic 3D human body pose estimation", "abstract": "  We present a novel approach for 3D human pose estimation by employing\nprobabilistic modeling. This approach leverages the advantages of normalizing\nflows in non-Euclidean geometries to address uncertain poses. Specifically, our\nmethod employs normalizing flow tailored to the SO(3) rotational group,\nincorporating a coupling mechanism based on the M\\\"obius transformation. This\nenables the framework to accurately represent any distribution on SO(3),\neffectively addressing issues related to discontinuities. Additionally, we\nreinterpret the challenge of reconstructing 3D human figures from 2D\npixel-aligned inputs as the task of mapping these inputs to a range of probable\nposes. This perspective acknowledges the intrinsic ambiguity of the task and\nfacilitates a straightforward integration method for multi-view scenarios. The\ncombination of these strategies showcases the effectiveness of probabilistic\nmodels in complex scenarios for human pose estimation techniques. Our approach\nnotably surpasses existing methods in the field of pose estimation. We also\nvalidate our methodology on human pose estimation from RGB images as well as\nmedical X-Ray datasets.\n", "date": "2024-12-05", "categories": []}
{"id": "2312.16221", "title": "STRIDE: Single-video based Temporally Continuous Occlusion-Robust 3D\n  Pose Estimation", "abstract": "  The capability to accurately estimate 3D human poses is crucial for diverse\nfields such as action recognition, gait recognition, and virtual/augmented\nreality. However, a persistent and significant challenge within this field is\nthe accurate prediction of human poses under conditions of severe occlusion.\nTraditional image-based estimators struggle with heavy occlusions due to a lack\nof temporal context, resulting in inconsistent predictions. While video-based\nmodels benefit from processing temporal data, they encounter limitations when\nfaced with prolonged occlusions that extend over multiple frames. This\nchallenge arises because these models struggle to generalize beyond their\ntraining datasets, and the variety of occlusions is hard to capture in the\ntraining data. Addressing these challenges, we propose STRIDE (Single-video\nbased TempoRally contInuous Occlusion-Robust 3D Pose Estimation), a novel\nTest-Time Training (TTT) approach to fit a human motion prior for each video.\nThis approach specifically handles occlusions that were not encountered during\nthe model's training. By employing STRIDE, we can refine a sequence of noisy\ninitial pose estimates into accurate, temporally coherent poses during test\ntime, effectively overcoming the limitations of prior methods. Our framework\ndemonstrates flexibility by being model-agnostic, allowing us to use any\noff-the-shelf 3D pose estimation method for improving robustness and temporal\nconsistency. We validate STRIDE's efficacy through comprehensive experiments on\nchallenging datasets like Occluded Human3.6M, Human3.6M, and OCMotion, where it\nnot only outperforms existing single-image and video-based pose estimation\nmodels but also showcases superior handling of substantial occlusions,\nachieving fast, robust, accurate, and temporally consistent 3D pose estimates.\nCode is made publicly available at https://github.com/take2rohit/stride\n", "date": "2023-12-24", "categories": []}
{"id": "2412.02903", "title": "EgoCast: Forecasting Egocentric Human Pose in the Wild", "abstract": "  Accurately estimating and forecasting human body pose is important for\nenhancing the user's sense of immersion in Augmented Reality. Addressing this\nneed, our paper introduces EgoCast, a bimodal method for 3D human pose\nforecasting using egocentric videos and proprioceptive data. We study the task\nof human pose forecasting in a realistic setting, extending the boundaries of\ntemporal forecasting in dynamic scenes and building on the current framework\nfor current pose estimation in the wild. We introduce a current-frame\nestimation module that generates pseudo-groundtruth poses for inference,\neliminating the need for past groundtruth poses typically required by current\nmethods during forecasting. Our experimental results on the recent Ego-Exo4D\nand Aria Digital Twin datasets validate EgoCast for real-life motion\nestimation. On the Ego-Exo4D Body Pose 2024 Challenge, our method significantly\noutperforms the state-of-the-art approaches, laying the groundwork for future\nresearch in human pose estimation and forecasting in unscripted activities with\negocentric inputs.\n", "date": "2024-12-03", "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": "2312.06386", "title": "ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation", "abstract": "  We propose ManiPose, a manifold-constrained multi-hypothesis model for\nhuman-pose 2D-to-3D lifting. We provide theoretical and empirical evidence\nthat, due to the depth ambiguity inherent to monocular 3D human pose\nestimation, traditional regression models suffer from pose-topology consistency\nissues, which standard evaluation metrics (MPJPE, P-MPJPE and PCK) fail to\nassess. ManiPose addresses depth ambiguity by proposing multiple candidate 3D\nposes for each 2D input, each with its estimated plausibility. Unlike previous\nmulti-hypothesis approaches, ManiPose forgoes generative models, greatly\nfacilitating its training and usage. By constraining the outputs to lie on the\nhuman pose manifold, ManiPose guarantees the consistency of all hypothetical\nposes, in contrast to previous works. We showcase the performance of ManiPose\non real-world datasets, where it outperforms state-of-the-art models in pose\nconsistency by a large margin while being very competitive on the MPJPE metric.\n", "date": "2023-12-11", "categories": []}
{"id": "2411.18377", "title": "XR-MBT: Multi-modal Full Body Tracking for XR through Self-Supervision\n  with Learned Depth Point Cloud Registration", "abstract": "  Tracking the full body motions of users in XR (AR/VR) devices is a\nfundamental challenge to bring a sense of authentic social presence. Due to the\nabsence of dedicated leg sensors, currently available body tracking methods\nadopt a synthesis approach to generate plausible motions given a 3-point signal\nfrom the head and controller tracking. In order to enable mixed reality\nfeatures, modern XR devices are capable of estimating depth information of the\nheadset surroundings using available sensors combined with dedicated machine\nlearning models. Such egocentric depth sensing cannot drive the body directly,\nas it is not registered and is incomplete due to limited field-of-view and body\nself-occlusions. For the first time, we propose to leverage the available depth\nsensing signal combined with self-supervision to learn a multi-modal pose\nestimation model capable of tracking full body motions in real time on XR\ndevices. We demonstrate how current 3-point motion synthesis models can be\nextended to point cloud modalities using a semantic point cloud encoder network\ncombined with a residual network for multi-modal pose estimation. These modules\nare trained jointly in a self-supervised way, leveraging a combination of real\nunregistered point clouds and simulated data obtained from motion capture. We\ncompare our approach against several state-of-the-art systems for XR body\ntracking and show that our method accurately tracks a diverse range of body\nmotions. XR-MBT tracks legs in XR for the first time, whereas traditional\nsynthesis approaches based on partial body tracking are blind.\n", "date": "2024-11-27", "categories": []}
{"id": "2411.13607", "title": "VioPose: Violin Performance 4D Pose Estimation by Hierarchical\n  Audiovisual Inference", "abstract": "  Musicians delicately control their bodies to generate music. Sometimes, their\nmotions are too subtle to be captured by the human eye. To analyze how they\nmove to produce the music, we need to estimate precise 4D human pose (3D pose\nover time). However, current state-of-the-art (SoTA) visual pose estimation\nalgorithms struggle to produce accurate monocular 4D poses because of\nocclusions, partial views, and human-object interactions. They are limited by\nthe viewing angle, pixel density, and sampling rate of the cameras and fail to\nestimate fast and subtle movements, such as in the musical effect of vibrato.\nWe leverage the direct causal relationship between the music produced and the\nhuman motions creating them to address these challenges. We propose VioPose: a\nnovel multimodal network that hierarchically estimates dynamics. High-level\nfeatures are cascaded to low-level features and integrated into Bayesian\nupdates. Our architecture is shown to produce accurate pose sequences,\nfacilitating precise motion analysis, and outperforms SoTA. As part of this\nwork, we collected the largest and the most diverse calibrated violin-playing\ndataset, including video, sound, and 3D motion capture poses. Code and dataset\ncan be found in our project page \\url{https://sj-yoo.info/viopose/}.\n", "date": "2024-11-19", "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": "2411.13026", "title": "X as Supervision: Contending with Depth Ambiguity in Unsupervised\n  Monocular 3D Pose Estimation", "abstract": "  Recent unsupervised methods for monocular 3D pose estimation have endeavored\nto reduce dependence on limited annotated 3D data, but most are solely\nformulated in 2D space, overlooking the inherent depth ambiguity issue. Due to\nthe information loss in 3D-to-2D projection, multiple potential depths may\nexist, yet only some of them are plausible in human structure. To tackle depth\nambiguity, we propose a novel unsupervised framework featuring a\nmulti-hypothesis detector and multiple tailored pretext tasks. The detector\nextracts multiple hypotheses from a heatmap within a local window, effectively\nmanaging the multi-solution problem. Furthermore, the pretext tasks harness 3D\nhuman priors from the SMPL model to regularize the solution space of pose\nestimation, aligning it with the empirical distribution of 3D human structures.\nThis regularization is partially achieved through a GCN-based discriminator\nwithin the discriminative learning, and is further complemented with synthetic\nimages through rendering, ensuring plausible estimations. Consequently, our\napproach demonstrates state-of-the-art unsupervised 3D pose estimation\nperformance on various human datasets. Further evaluations on data scale-up and\none animal dataset highlight its generalization capabilities. Code will be\navailable at https://github.com/Charrrrrlie/X-as-Supervision.\n", "date": "2024-11-20", "categories": []}
{"id": "2411.12676", "title": "IoT-Based 3D Pose Estimation and Motion Optimization for Athletes:\n  Application of C3D and OpenPose", "abstract": "  This study proposes the IoT-Enhanced Pose Optimization Network (IE-PONet) for\nhigh-precision 3D pose estimation and motion optimization of track and field\nathletes. IE-PONet integrates C3D for spatiotemporal feature extraction,\nOpenPose for real-time keypoint detection, and Bayesian optimization for\nhyperparameter tuning. Experimental results on NTURGB+D and FineGYM datasets\ndemonstrate superior performance, with AP\\(^p50\\) scores of 90.5 and 91.0, and\nmAP scores of 74.3 and 74.0, respectively. Ablation studies confirm the\nessential roles of each module in enhancing model accuracy. IE-PONet provides a\nrobust tool for athletic performance analysis and optimization, offering\nprecise technical insights for training and injury prevention. Future work will\nfocus on further model optimization, multimodal data integration, and\ndeveloping real-time feedback mechanisms to enhance practical applications.\n", "date": "2024-11-19", "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": "2411.09435", "title": "ReMP: Reusable Motion Prior for Multi-domain 3D Human Pose Estimation\n  and Motion Inbetweening", "abstract": "  We present Reusable Motion prior (ReMP), an effective motion prior that can\naccurately track the temporal evolution of motion in various downstream tasks.\nInspired by the success of foundation models, we argue that a robust\nspatio-temporal motion prior can encapsulate underlying 3D dynamics applicable\nto various sensor modalities. We learn the rich motion prior from a sequence of\ncomplete parametric models of posed human body shape. Our prior can easily\nestimate poses in missing frames or noisy measurements despite significant\nocclusion by employing a temporal attention mechanism. More interestingly, our\nprior can guide the system with incomplete and challenging input measurements\nto quickly extract critical information to estimate the sequence of poses,\nsignificantly improving the training efficiency for mesh sequence recovery.\nReMP consistently outperforms the baseline method on diverse and practical 3D\nmotion data, including depth point clouds, LiDAR scans, and IMU sensor data.\nProject page is available in https://hojunjang17.github.io/ReMP.\n", "date": "2024-11-13", "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": "2411.08603", "title": "Generalized Pose Space Embeddings for Training In-the-Wild using\n  Anaylis-by-Synthesis", "abstract": "  Modern pose estimation models are trained on large, manually-labelled\ndatasets which are costly and may not cover the full extent of human poses and\nappearances in the real world. With advances in neural rendering,\nanalysis-by-synthesis and the ability to not only predict, but also render the\npose, is becoming an appealing framework, which could alleviate the need for\nlarge scale manual labelling efforts. While recent work have shown the\nfeasibility of this approach, the predictions admit many flips due to a\nsimplistic intermediate skeleton representation, resulting in low precision and\ninhibiting the acquisition of any downstream knowledge such as\nthree-dimensional positioning. We solve this problem with a more expressive\nintermediate skeleton representation capable of capturing the semantics of the\npose (left and right), which significantly reduces flips. To successfully train\nthis new representation, we extend the analysis-by-synthesis framework with a\ntraining protocol based on synthetic data. We show that our representation\nresults in less flips and more accurate predictions. Our approach outperforms\nprevious models trained with analysis-by-synthesis on standard benchmarks.\n", "date": "2024-11-13", "categories": []}
{"id": "2411.08663", "title": "Toward Human Understanding with Controllable Synthesis", "abstract": "  Training methods to perform robust 3D human pose and shape (HPS) estimation\nrequires diverse training images with accurate ground truth. While BEDLAM\ndemonstrates the potential of traditional procedural graphics to generate such\ndata, the training images are clearly synthetic. In contrast, generative image\nmodels produce highly realistic images but without ground truth. Putting these\nmethods together seems straightforward: use a generative model with the body\nground truth as controlling signal. However, we find that, the more realistic\nthe generated images, the more they deviate from the ground truth, making them\ninappropriate for training and evaluation. Enhancements of realistic details,\nsuch as clothing and facial expressions, can lead to subtle yet significant\ndeviations from the ground truth, potentially misleading training models. We\nempirically verify that this misalignment causes the accuracy of HPS networks\nto decline when trained with generated images. To address this, we design a\ncontrollable synthesis method that effectively balances image realism with\nprecise ground truth. We use this to create the Generative BEDLAM (Gen-B)\ndataset, which improves the realism of the existing synthetic BEDLAM dataset\nwhile preserving ground truth accuracy. We perform extensive experiments, with\nvarious noise-conditioning strategies, to evaluate the tradeoff between visual\nrealism and HPS accuracy. We show, for the first time, that generative image\nmodels can be controlled by traditional graphics methods to produce training\ndata that increases the accuracy of HPS methods.\n", "date": "2024-11-13", "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": "2411.06725", "title": "GTA-Net: An IoT-Integrated 3D Human Pose Estimation System for Real-Time\n  Adolescent Sports Posture Correction", "abstract": "  With the advancement of artificial intelligence, 3D human pose\nestimation-based systems for sports training and posture correction have gained\nsignificant attention in adolescent sports. However, existing methods face\nchallenges in handling complex movements, providing real-time feedback, and\naccommodating diverse postures, particularly with occlusions, rapid movements,\nand the resource constraints of Internet of Things (IoT) devices, making it\ndifficult to balance accuracy and real-time performance. To address these\nissues, we propose GTA-Net, an intelligent system for posture correction and\nreal-time feedback in adolescent sports, integrated within an IoT-enabled\nenvironment. This model enhances pose estimation in dynamic scenes by\nincorporating Graph Convolutional Networks (GCN), Temporal Convolutional\nNetworks (TCN), and Hierarchical Attention mechanisms, achieving real-time\ncorrection through IoT devices. Experimental results show GTA-Net's superior\nperformance on Human3.6M, HumanEva-I, and MPI-INF-3DHP datasets, with Mean Per\nJoint Position Error (MPJPE) values of 32.2mm, 15.0mm, and 48.0mm,\nrespectively, significantly outperforming existing methods. The model also\ndemonstrates strong robustness in complex scenarios, maintaining high accuracy\neven with occlusions and rapid movements. This system enhances real-time\nposture correction and offers broad applications in intelligent sports and\nhealth management.\n", "date": "2024-11-11", "categories": []}
{"id": "2411.07165", "title": "Acoustic-based 3D Human Pose Estimation Robust to Human Position", "abstract": "  This paper explores the problem of 3D human pose estimation from only\nlow-level acoustic signals. The existing active acoustic sensing-based approach\nfor 3D human pose estimation implicitly assumes that the target user is\npositioned along a line between loudspeakers and a microphone. Because\nreflection and diffraction of sound by the human body cause subtle acoustic\nsignal changes compared to sound obstruction, the existing model degrades its\naccuracy significantly when subjects deviate from this line, limiting its\npracticality in real-world scenarios. To overcome this limitation, we propose a\nnovel method composed of a position discriminator and reverberation-resistant\nmodel. The former predicts the standing positions of subjects and applies\nadversarial learning to extract subject position-invariant features. The latter\nutilizes acoustic signals before the estimation target time as references to\nenhance robustness against the variations in sound arrival times due to\ndiffraction and reflection. We construct an acoustic pose estimation dataset\nthat covers diverse human locations and demonstrate through experiments that\nour proposed method outperforms existing approaches.\n", "date": "2024-11-08", "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": "2411.04598", "title": "Social EgoMesh Estimation", "abstract": "  Accurately estimating the 3D pose of the camera wearer in egocentric video\nsequences is crucial to modeling human behavior in virtual and augmented\nreality applications. The task presents unique challenges due to the limited\nvisibility of the user's body caused by the front-facing camera mounted on\ntheir head. Recent research has explored the utilization of the scene and\nego-motion, but it has overlooked humans' interactive nature. We propose a\nnovel framework for Social Egocentric Estimation of body MEshes (SEE-ME). Our\napproach is the first to estimate the wearer's mesh using only a latent\nprobabilistic diffusion model, which we condition on the scene and, for the\nfirst time, on the social wearer-interactee interactions. Our in-depth study\nsheds light on when social interaction matters most for ego-mesh estimation; it\nquantifies the impact of interpersonal distance and gaze direction. Overall,\nSEE-ME surpasses the current best technique, reducing the pose estimation error\n(MPJPE) by 53%. The code is available at https://github.com/L-Scofano/SEEME.\n", "date": "2024-11-07", "categories": []}
{"id": "2411.03561", "title": "Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data", "abstract": "  We study the problem of estimating the body movements of a camera wearer from\negocentric videos. Current methods for ego-body pose estimation rely on\ntemporally dense sensor data, such as IMU measurements from spatially sparse\nbody parts like the head and hands. However, we propose that even temporally\nsparse observations, such as hand poses captured intermittently from egocentric\nvideos during natural or periodic hand movements, can effectively constrain\noverall body motion. Naively applying diffusion models to generate full-body\npose from head pose and sparse hand pose leads to suboptimal results. To\novercome this, we develop a two-stage approach that decomposes the problem into\ntemporal completion and spatial completion. First, our method employs masked\nautoencoders to impute hand trajectories by leveraging the spatiotemporal\ncorrelations between the head pose sequence and intermittent hand poses,\nproviding uncertainty estimates. Subsequently, we employ conditional diffusion\nmodels to generate plausible full-body motions based on these temporally dense\ntrajectories of the head and hands, guided by the uncertainty estimates from\nthe imputation. The effectiveness of our method was rigorously tested and\nvalidated through comprehensive experiments conducted on various HMD setup with\nAMASS and Ego-Exo4D datasets.\n", "date": "2024-11-05", "categories": []}
{"id": "2401.14121", "title": "Incorporating Test-Time Optimization into Training with Dual Networks\n  for Human Mesh Recovery", "abstract": "  Human Mesh Recovery (HMR) is the task of estimating a parameterized 3D human\nmesh from an image. There is a kind of methods first training a regression\nmodel for this problem, then further optimizing the pretrained regression model\nfor any specific sample individually at test time. However, the pretrained\nmodel may not provide an ideal optimization starting point for the test-time\noptimization. Inspired by meta-learning, we incorporate the test-time\noptimization into training, performing a step of test-time optimization for\neach sample in the training batch before really conducting the training\noptimization over all the training samples. In this way, we obtain a\nmeta-model, the meta-parameter of which is friendly to the test-time\noptimization. At test time, after several test-time optimization steps starting\nfrom the meta-parameter, we obtain much higher HMR accuracy than the test-time\noptimization starting from the simply pretrained regression model. Furthermore,\nwe find test-time HMR objectives are different from training-time objectives,\nwhich reduces the effectiveness of the learning of the meta-model. To solve\nthis problem, we propose a dual-network architecture that unifies the\ntraining-time and test-time objectives. Our method, armed with meta-learning\nand the dual networks, outperforms state-of-the-art regression-based and\noptimization-based HMR approaches, as validated by the extensive experiments.\nThe codes are available at https://github.com/fmx789/Meta-HMR.\n", "date": "2024-01-25", "categories": []}
{"id": "2410.20731", "title": "BLAPose: Enhancing 3D Human Pose Estimation with Bone Length Adjustment", "abstract": "  Current approaches in 3D human pose estimation primarily focus on regressing\n3D joint locations, often neglecting critical physical constraints such as bone\nlength consistency and body symmetry. This work introduces a recurrent neural\nnetwork architecture designed to capture holistic information across entire\nvideo sequences, enabling accurate prediction of bone lengths. To enhance\ntraining effectiveness, we propose a novel augmentation strategy using\nsynthetic bone lengths that adhere to physical constraints. Moreover, we\npresent a bone length adjustment method that preserves bone orientations while\nsubstituting bone lengths with predicted values. Our results demonstrate that\nexisting 3D human pose estimation models can be significantly enhanced through\nthis adjustment process. Furthermore, we fine-tune human pose estimation models\nusing inferred bone lengths, observing notable improvements. Our bone length\nprediction model surpasses the previous best results, and our adjustment and\nfine-tuning method enhance performance across several metrics on the Human3.6M\ndataset.\n", "date": "2024-10-28", "categories": []}
{"id": "2410.20294", "title": "Harmony4D: A Video Dataset for In-The-Wild Close Human Interactions", "abstract": "  Understanding how humans interact with each other is key to building\nrealistic multi-human virtual reality systems. This area remains relatively\nunexplored due to the lack of large-scale datasets. Recent datasets focusing on\nthis issue mainly consist of activities captured entirely in controlled indoor\nenvironments with choreographed actions, significantly affecting their\ndiversity. To address this, we introduce Harmony4D, a multi-view video dataset\nfor human-human interaction featuring in-the-wild activities such as wrestling,\ndancing, MMA, and more. We use a flexible multi-view capture system to record\nthese dynamic activities and provide annotations for human detection, tracking,\n2D/3D pose estimation, and mesh recovery for closely interacting subjects. We\npropose a novel markerless algorithm to track 3D human poses in severe\nocclusion and close interaction to obtain our annotations with minimal manual\nintervention. Harmony4D consists of 1.66 million images and 3.32 million human\ninstances from more than 20 synchronized cameras with 208 video sequences\nspanning diverse environments and 24 unique subjects. We rigorously evaluate\nexisting state-of-the-art methods for mesh recovery and highlight their\nsignificant limitations in modeling close interaction scenarios. Additionally,\nwe fine-tune a pre-trained HMR2.0 model on Harmony4D and demonstrate an\nimproved performance of 54.8% PVE in scenes with severe occlusion and contact.\nCode and data are available at https://jyuntins.github.io/harmony4d/.\n", "date": "2024-10-27", "categories": []}
{"id": "2410.20508", "title": "Referring Human Pose and Mask Estimation in the Wild", "abstract": "  We introduce Referring Human Pose and Mask Estimation (R-HPM) in the wild,\nwhere either a text or positional prompt specifies the person of interest in an\nimage. This new task holds significant potential for human-centric applications\nsuch as assistive robotics and sports analysis. In contrast to previous works,\nR-HPM (i) ensures high-quality, identity-aware results corresponding to the\nreferred person, and (ii) simultaneously predicts human pose and mask for a\ncomprehensive representation. To achieve this, we introduce a large-scale\ndataset named RefHuman, which substantially extends the MS COCO dataset with\nadditional text and positional prompt annotations. RefHuman includes over\n50,000 annotated instances in the wild, each equipped with keypoint, mask, and\nprompt annotations. To enable prompt-conditioned estimation, we propose the\nfirst end-to-end promptable approach named UniPHD for R-HPM. UniPHD extracts\nmultimodal representations and employs a proposed pose-centric hierarchical\ndecoder to process (text or positional) instance queries and keypoint queries,\nproducing results specific to the referred person. Extensive experiments\ndemonstrate that UniPHD produces quality results based on user-friendly prompts\nand achieves top-tier performance on RefHuman val and MS COCO val2017. Data and\nCode: https://github.com/bo-miao/RefHuman\n", "date": "2024-10-27", "categories": []}
{"id": "2407.02990", "title": "Graph and Skipped Transformer: Exploiting Spatial and Temporal Modeling\n  Capacities for Efficient 3D Human Pose Estimation", "abstract": "  In recent years, 2D-to-3D pose uplifting in monocular 3D Human Pose\nEstimation (HPE) has attracted widespread research interest. GNN-based methods\nand Transformer-based methods have become mainstream architectures due to their\nadvanced spatial and temporal feature learning capacities. However, existing\napproaches typically construct joint-wise and frame-wise attention alignments\nin spatial and temporal domains, resulting in dense connections that introduce\nconsiderable local redundancy and computational overhead. In this paper, we\ntake a global approach to exploit spatio-temporal information and realise\nefficient 3D HPE with a concise Graph and Skipped Transformer architecture.\nSpecifically, in Spatial Encoding stage, coarse-grained body parts are deployed\nto construct Spatial Graph Network with a fully data-driven adaptive topology,\nensuring model flexibility and generalizability across various poses. In\nTemporal Encoding and Decoding stages, a simple yet effective Skipped\nTransformer is proposed to capture long-range temporal dependencies and\nimplement hierarchical feature aggregation. A straightforward Data Rolling\nstrategy is also developed to introduce dynamic information into 2D pose\nsequence. Extensive experiments are conducted on Human3.6M, MPI-INF-3DHP and\nHuman-Eva benchmarks. G-SFormer series methods achieve superior performances\ncompared with previous state-of-the-arts with only around ten percent of\nparameters and significantly reduced computational complexity. Additionally,\nG-SFormer also exhibits outstanding robustness to inaccuracies in detected 2D\nposes.\n", "date": "2024-07-03", "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": "2410.14540", "title": "Multi-modal Pose Diffuser: A Multimodal Generative Conditional Pose\n  Prior", "abstract": "  The Skinned Multi-Person Linear (SMPL) model plays a crucial role in 3D human\npose estimation, providing a streamlined yet effective representation of the\nhuman body. However, ensuring the validity of SMPL configurations during tasks\nsuch as human mesh regression remains a significant challenge , highlighting\nthe necessity for a robust human pose prior capable of discerning realistic\nhuman poses. To address this, we introduce MOPED:\n\\underline{M}ulti-m\\underline{O}dal \\underline{P}os\\underline{E}\n\\underline{D}iffuser. MOPED is the first method to leverage a novel multi-modal\nconditional diffusion model as a prior for SMPL pose parameters. Our method\noffers powerful unconditional pose generation with the ability to condition on\nmulti-modal inputs such as images and text. This capability enhances the\napplicability of our approach by incorporating additional context often\noverlooked in traditional pose priors. Extensive experiments across three\ndistinct tasks-pose estimation, pose denoising, and pose completion-demonstrate\nthat our multi-modal diffusion model-based prior significantly outperforms\nexisting methods. These results indicate that our model captures a broader\nspectrum of plausible human poses.\n", "date": "2024-10-18", "categories": []}
{"id": "2410.12023", "title": "Learned Neural Physics Simulation for Articulated 3D Human Pose\n  Reconstruction", "abstract": "  We propose a novel neural network approach, LARP (Learned Articulated Rigid\nbody Physics), to model the dynamics of articulated human motion with contact.\nOur goal is to develop a faster and more convenient methodological alternative\nto traditional physics simulators for use in computer vision tasks such as\nhuman motion reconstruction from video. To that end we introduce a training\nprocedure and model components that support the construction of a recurrent\nneural architecture to accurately simulate articulated rigid body dynamics. Our\nneural architecture supports features typically found in traditional physics\nsimulators, such as modeling of joint motors, variable dimensions of body\nparts, contact between body parts and objects, and is an order of magnitude\nfaster than traditional systems when multiple simulations are run in parallel.\nTo demonstrate the value of LARP we use it as a drop-in replacement for a state\nof the art classical non-differentiable simulator in an existing video-based\nreconstruction framework and show comparative or better 3D human pose\nreconstruction accuracy.\n", "date": "2024-10-15", "categories": []}
{"id": "2410.09885", "title": "Occluded Human Pose Estimation based on Limb Joint Augmentation", "abstract": "  Human pose estimation aims at locating the specific joints of humans from the\nimages or videos. While existing deep learning-based methods have achieved high\npositioning accuracy, they often struggle with generalization in occlusion\nscenarios. In this paper, we propose an occluded human pose estimation\nframework based on limb joint augmentation to enhance the generalization\nability of the pose estimation model on the occluded human bodies.\nSpecifically, the occlusion blocks are at first employed to randomly cover the\nlimb joints of the human bodies from the training images, imitating the scene\nwhere the objects or other people partially occlude the human body. Trained by\nthe augmented samples, the pose estimation model is encouraged to accurately\nlocate the occluded keypoints based on the visible ones. To further enhance the\nlocalization ability of the model, this paper constructs a dynamic structure\nloss function based on limb graphs to explore the distribution of occluded\njoints by evaluating the dependence between adjacent joints. Extensive\nexperimental evaluations on two occluded datasets, OCHuman and CrowdPose,\ndemonstrate significant performance improvements without additional computation\ncost during inference.\n", "date": "2024-10-13", "categories": []}
{"id": "2404.19279", "title": "Quater-GCN: Enhancing 3D Human Pose Estimation with Orientation and\n  Semi-supervised Training", "abstract": "  3D human pose estimation is a vital task in computer vision, involving the\nprediction of human joint positions from images or videos to reconstruct a\nskeleton of a human in three-dimensional space. This technology is pivotal in\nvarious fields, including animation, security, human-computer interaction, and\nautomotive safety, where it promotes both technological progress and enhanced\nhuman well-being. The advent of deep learning significantly advances the\nperformance of 3D pose estimation by incorporating temporal information for\npredicting the spatial positions of human joints. However, traditional methods\noften fall short as they primarily focus on the spatial coordinates of joints\nand overlook the orientation and rotation of the connecting bones, which are\ncrucial for a comprehensive understanding of human pose in 3D space. To address\nthese limitations, we introduce Quater-GCN (Q-GCN), a directed graph\nconvolutional network tailored to enhance pose estimation by orientation. Q-GCN\nexcels by not only capturing the spatial dependencies among node joints through\ntheir coordinates but also integrating the dynamic context of bone rotations in\n2D space. This approach enables a more sophisticated representation of human\nposes by also regressing the orientation of each bone in 3D space, moving\nbeyond mere coordinate prediction. Furthermore, we complement our model with a\nsemi-supervised training strategy that leverages unlabeled data, addressing the\nchallenge of limited orientation ground truth data. Through comprehensive\nevaluations, Q-GCN has demonstrated outstanding performance against current\nstate-of-the-art methods.\n", "date": "2024-04-30", "categories": []}
{"id": "2310.08116", "title": "Multimodal Active Measurement for Human Mesh Recovery in Close Proximity", "abstract": "  For physical human-robot interactions (pHRI), a robot needs to estimate the\naccurate body pose of a target person. However, in these pHRI scenarios, the\nrobot cannot fully observe the target person's body with equipped cameras\nbecause the target person must be close to the robot for physical interaction.\nThis close distance leads to severe truncation and occlusions and thus results\nin poor accuracy of human pose estimation. For better accuracy in this\nchallenging environment, we propose an active measurement and sensor fusion\nframework of the equipped cameras with touch and ranging sensors such as 2D\nLiDAR. Touch and ranging sensor measurements are sparse but reliable and\ninformative cues for localizing human body parts. In our active measurement\nprocess, camera viewpoints and sensor placements are dynamically optimized to\nmeasure body parts with higher estimation uncertainty, which is closely related\nto truncation or occlusion. In our sensor fusion process, assuming that the\nmeasurements of touch and ranging sensors are more reliable than the\ncamera-based estimations, we fuse the sensor measurements to the camera-based\nestimated pose by aligning the estimated pose towards the measured points. Our\nproposed method outperformed previous methods on the standard occlusion\nbenchmark with simulated active measurement. Furthermore, our method reliably\nestimated human poses using a real robot, even with practical constraints such\nas occlusion by blankets.\n", "date": "2023-10-12", "categories": []}
{"id": "2410.04574", "title": "Enhancing 3D Human Pose Estimation Amidst Severe Occlusion with Dual\n  Transformer Fusion", "abstract": "  In the field of 3D Human Pose Estimation from monocular videos, the presence\nof diverse occlusion types presents a formidable challenge. Prior research has\nmade progress by harnessing spatial and temporal cues to infer 3D poses from 2D\njoint observations. This paper introduces a Dual Transformer Fusion (DTF)\nalgorithm, a novel approach to obtain a holistic 3D pose estimation, even in\nthe presence of severe occlusions. Confronting the issue of occlusion-induced\nmissing joint data, we propose a temporal interpolation-based occlusion\nguidance mechanism. To enable precise 3D Human Pose Estimation, our approach\nleverages the innovative DTF architecture, which first generates a pair of\nintermediate views. Each intermediate-view undergoes spatial refinement through\na self-refinement schema. Subsequently, these intermediate-views are fused to\nyield the final 3D human pose estimation. The entire system is end-to-end\ntrainable. Through extensive experiments conducted on the Human3.6M and\nMPI-INF-3DHP datasets, our method's performance is rigorously evaluated.\nNotably, our approach outperforms existing state-of-the-art methods on both\ndatasets, yielding substantial improvements. The code is available here:\nhttps://github.com/MehwishG/DTF.\n", "date": "2024-10-06", "categories": []}
{"id": "2410.04889", "title": "D-PoSE: Depth as an Intermediate Representation for 3D Human Pose and\n  Shape Estimation", "abstract": "  We present D-PoSE (Depth as an Intermediate Representation for 3D Human Pose\nand Shape Estimation), a one-stage method that estimates human pose and SMPL-X\nshape parameters from a single RGB image. Recent works use larger models with\ntransformer backbones and decoders to improve the accuracy in human pose and\nshape (HPS) benchmarks. D-PoSE proposes a vision based approach that uses the\nestimated human depth-maps as an intermediate representation for HPS and\nleverages training with synthetic data and the ground-truth depth-maps provided\nwith them for depth supervision during training. Although trained on synthetic\ndatasets, D-PoSE achieves state-of-the-art performance on the real-world\nbenchmark datasets, EMDB and 3DPW. Despite its simple lightweight design and\nthe CNN backbone, it outperforms ViT-based models that have a number of\nparameters that is larger by almost an order of magnitude. D-PoSE code is\navailable at: https://github.com/nvasilik/D-PoSE\n", "date": "2024-10-07", "categories": []}
{"id": "2402.02074", "title": "Multi-RoI Human Mesh Recovery with Camera Consistency and Contrastive\n  Losses", "abstract": "  Besides a 3D mesh, Human Mesh Recovery (HMR) methods usually need to estimate\na camera for computing 2D reprojection loss. Previous approaches may encounter\nthe following problem: both the mesh and camera are not correct but the\ncombination of them can yield a low reprojection loss. To alleviate this\nproblem, we define multiple RoIs (region of interest) containing the same human\nand propose a multiple-RoI-based HMR method. Our key idea is that with multiple\nRoIs as input, we can estimate multiple local cameras and have the opportunity\nto design and apply additional constraints between cameras to improve the\naccuracy of the cameras and, in turn, the accuracy of the corresponding 3D\nmesh. To implement this idea, we propose a RoI-aware feature fusion network by\nwhich we estimate a 3D mesh shared by all RoIs as well as local cameras\ncorresponding to the RoIs. We observe that local cameras can be converted to\nthe camera of the full image through which we construct a local camera\nconsistency loss as the additional constraint imposed on local cameras. Another\nbenefit of introducing multiple RoIs is that we can encapsulate our network\ninto a contrastive learning framework and apply a contrastive loss to\nregularize the training of our network. Experiments demonstrate the\neffectiveness of our multi-RoI HMR method and superiority to recent prior arts.\nOur code is available at https://github.com/CptDiaos/Multi-RoI.\n", "date": "2024-02-03", "categories": []}
{"id": "2410.00309", "title": "Ask, Pose, Unite: Scaling Data Acquisition for Close Interactions with\n  Vision Language Models", "abstract": "  Social dynamics in close human interactions pose significant challenges for\nHuman Mesh Estimation (HME), particularly due to the complexity of physical\ncontacts and the scarcity of training data. Addressing these challenges, we\nintroduce a novel data generation method that utilizes Large Vision Language\nModels (LVLMs) to annotate contact maps which guide test-time optimization to\nproduce paired image and pseudo-ground truth meshes. This methodology not only\nalleviates the annotation burden but also enables the assembly of a\ncomprehensive dataset specifically tailored for close interactions in HME. Our\nAsk Pose Unite (APU) dataset, comprising over 6.2k human mesh pairs in contact\ncovering diverse interaction types, is curated from images depicting\nnaturalistic person-to-person scenes. We empirically show that using our\ndataset to train a diffusion-based contact prior, used as guidance during\noptimization, improves mesh estimation on unseen interactions. Our work\naddresses longstanding challenges of data scarcity for close interactions in\nHME enhancing the field's capabilities of handling complex interaction\nscenarios.\n", "date": "2024-10-01", "categories": []}
{"id": "2403.06164", "title": "Platypose: Calibrated Zero-Shot Multi-Hypothesis 3D Human Motion\n  Estimation", "abstract": "  Single camera 3D pose estimation is an ill-defined problem due to inherent\nambiguities from depth, occlusion or keypoint noise. Multi-hypothesis pose\nestimation accounts for this uncertainty by providing multiple 3D poses\nconsistent with the 2D measurements. Current research has predominantly\nconcentrated on generating multiple hypotheses for single frame static pose\nestimation or single hypothesis motion estimation. In this study we focus on\nthe new task of multi-hypothesis motion estimation. Multi-hypothesis motion\nestimation is not simply multi-hypothesis pose estimation applied to multiple\nframes, which would ignore temporal correlation across frames. Instead, it\nrequires distributions which are capable of generating temporally consistent\nsamples, which is significantly more challenging than multi-hypothesis pose\nestimation or single-hypothesis motion estimation. To this end, we introduce\nPlatypose, a framework that uses a diffusion model pretrained on 3D human\nmotion sequences for zero-shot 3D pose sequence estimation. Platypose\noutperforms baseline methods on multiple hypotheses for motion estimation.\nAdditionally, Platypose also achieves state-of-the-art calibration and\ncompetitive joint error when tested on static poses from Human3.6M,\nMPI-INF-3DHP and 3DPW. Finally, because it is zero-shot, our method generalizes\nflexibly to different settings such as multi-camera inference.\n", "date": "2024-03-10", "categories": []}
{"id": "2311.04591", "title": "Exploring Event-based Human Pose Estimation with 3D Event\n  Representations", "abstract": "  Human pose estimation is a fundamental and appealing task in computer vision.\nAlthough traditional cameras are commonly applied, their reliability decreases\nin scenarios under high dynamic range or heavy motion blur, where event cameras\noffer a robust solution. Predominant event-based methods accumulate events into\nframes, ignoring the asynchronous and high temporal resolution that is crucial\nfor distinguishing distinct actions. To address this issue and to unlock the 3D\npotential of event information, we introduce two 3D event representations: the\nRasterized Event Point Cloud (RasEPC) and the Decoupled Event Voxel (DEV). The\nRasEPC aggregates events within concise temporal slices at identical positions,\npreserving their 3D attributes along with statistical information, thereby\nsignificantly reducing memory and computational demands. Meanwhile, the DEV\nrepresentation discretizes events into voxels and projects them across three\northogonal planes, utilizing decoupled event attention to retrieve 3D cues from\nthe 2D planes. Furthermore, we develop and release EV-3DPW, a synthetic\nevent-based dataset crafted to facilitate training and quantitative analysis in\noutdoor scenes. Our methods are tested on the DHP19 public dataset, MMHPSD\ndataset, and our EV-3DPW dataset, with further qualitative validation via a\nderived driving scene dataset EV-JAAD and an outdoor collection vehicle. Our\ncode and dataset have been made publicly available at\nhttps://github.com/MasterHow/EventPointPose.\n", "date": "2023-11-08", "categories": []}
{"id": "2409.16861", "title": "Limitations of (Procrustes) Alignment in Assessing Multi-Person Human\n  Pose and Shape Estimation", "abstract": "  We delve into the challenges of accurately estimating 3D human pose and shape\nin video surveillance scenarios. Beginning with the advocacy for metrics like\nW-MPJPE and W-PVE, which omit the (Procrustes) realignment step, to improve\nmodel evaluation, we then introduce RotAvat. This technique aims to enhance\nthese metrics by refining the alignment of 3D meshes with the ground plane.\nThrough qualitative comparisons, we demonstrate RotAvat's effectiveness in\naddressing the limitations of existing aproaches.\n", "date": "2024-09-25", "categories": []}
{"id": "2206.06420", "title": "GraphMLP: A Graph MLP-Like Architecture for 3D Human Pose Estimation", "abstract": "  Modern multi-layer perceptron (MLP) models have shown competitive results in\nlearning visual representations without self-attention. However, existing MLP\nmodels are not good at capturing local details and lack prior knowledge of\nhuman body configurations, which limits their modeling power for skeletal\nrepresentation learning. To address these issues, we propose a simple yet\neffective graph-reinforced MLP-Like architecture, named GraphMLP, that combines\nMLPs and graph convolutional networks (GCNs) in a global-local-graphical\nunified architecture for 3D human pose estimation. GraphMLP incorporates the\ngraph structure of human bodies into an MLP model to meet the domain-specific\ndemand of the 3D human pose, while allowing for both local and global spatial\ninteractions. Furthermore, we propose to flexibly and efficiently extend the\nGraphMLP to the video domain and show that complex temporal dynamics can be\neffectively modeled in a simple way with negligible computational cost gains in\nthe sequence length. To the best of our knowledge, this is the first MLP-Like\narchitecture for 3D human pose estimation in a single frame and a video\nsequence. Extensive experiments show that the proposed GraphMLP achieves\nstate-of-the-art performance on two datasets, i.e., Human3.6M and MPI-INF-3DHP.\nCode and models are available at https://github.com/Vegetebird/GraphMLP.\n", "date": "2022-06-13", "categories": []}
{"id": "2409.11104", "title": "Depth-based Privileged Information for Boosting 3D Human Pose Estimation\n  on RGB", "abstract": "  Despite the recent advances in computer vision research, estimating the 3D\nhuman pose from single RGB images remains a challenging task, as multiple 3D\nposes can correspond to the same 2D projection on the image. In this context,\ndepth data could help to disambiguate the 2D information by providing\nadditional constraints about the distance between objects in the scene and the\ncamera. Unfortunately, the acquisition of accurate depth data is limited to\nindoor spaces and usually is tied to specific depth technologies and devices,\nthus limiting generalization capabilities. In this paper, we propose a method\nable to leverage the benefits of depth information without compromising its\nbroader applicability and adaptability in a predominantly RGB-camera-centric\nlandscape. Our approach consists of a heatmap-based 3D pose estimator that,\nleveraging the paradigm of Privileged Information, is able to hallucinate depth\ninformation from the RGB frames given at inference time. More precisely, depth\ninformation is used exclusively during training by enforcing our RGB-based\nhallucination network to learn similar features to a backbone pre-trained only\non depth data. This approach proves to be effective even when dealing with\nlimited and small datasets. Experimental results reveal that the paradigm of\nPrivileged Information significantly enhances the model's performance, enabling\nefficient extraction of depth information by using only RGB images.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.08847", "title": "Kinect Calibration and Data Optimization For Anthropometric Parameters", "abstract": "  Recently, through development of several 3d vision systems, widely used in\nvarious applications, medical and biometric fields. Microsoft kinect sensor\nhave been most of used camera among 3d vision systems. Microsoft kinect sensor\ncan obtain depth images of a scene and 3d coordinates of human joints. Thus,\nanthropometric features can extractable easily. Anthropometric feature and 3d\njoint coordinate raw datas which captured from kinect sensor is unstable. The\nstrongest reason for this, datas vary by distance between joints of individual\nand location of kinect sensor. Consequently, usage of this datas without kinect\ncalibration and data optimization does not result in sufficient and healthy. In\nthis study, proposed a novel method to calibrating kinect sensor and optimizing\nskeleton features. Results indicate that the proposed method is quite effective\nand worthy of further study in more general scenarios.\n", "date": "2024-09-13", "categories": []}
{"id": "2409.07798", "title": "GateAttentionPose: Enhancing Pose Estimation with Agent Attention and\n  Improved Gated Convolutions", "abstract": "  This paper introduces GateAttentionPose, an innovative approach that enhances\nthe UniRepLKNet architecture for pose estimation tasks. We present two key\ncontributions: the Agent Attention module and the Gate-Enhanced Feedforward\nBlock (GEFB). The Agent Attention module replaces large kernel convolutions,\nsignificantly improving computational efficiency while preserving global\ncontext modeling. The GEFB augments feature extraction and processing\ncapabilities, particularly in complex scenes. Extensive evaluations on COCO and\nMPII datasets demonstrate that GateAttentionPose outperforms existing\nstate-of-the-art methods, including the original UniRepLKNet, achieving\nsuperior or comparable results with improved efficiency. Our approach offers a\nrobust solution for pose estimation across diverse applications, including\nautonomous driving, human motion capture, and virtual reality.\n", "date": "2024-09-12", "categories": []}
{"id": "2210.11795", "title": "PoseScript: Linking 3D Human Poses and Natural Language", "abstract": "  Natural language plays a critical role in many computer vision applications,\nsuch as image captioning, visual question answering, and cross-modal retrieval,\nto provide fine-grained semantic information. Unfortunately, while human pose\nis key to human understanding, current 3D human pose datasets lack detailed\nlanguage descriptions. To address this issue, we have introduced the PoseScript\ndataset. This dataset pairs more than six thousand 3D human poses from AMASS\nwith rich human-annotated descriptions of the body parts and their spatial\nrelationships. Additionally, to increase the size of the dataset to a scale\nthat is compatible with data-hungry learning algorithms, we have proposed an\nelaborate captioning process that generates automatic synthetic descriptions in\nnatural language from given 3D keypoints. This process extracts low-level pose\ninformation, known as \"posecodes\", using a set of simple but generic rules on\nthe 3D keypoints. These posecodes are then combined into higher level textual\ndescriptions using syntactic rules. With automatic annotations, the amount of\navailable data significantly scales up (100k), making it possible to\neffectively pretrain deep models for finetuning on human captions. To showcase\nthe potential of annotated poses, we present three multi-modal learning tasks\nthat utilize the PoseScript dataset. Firstly, we develop a pipeline that maps\n3D poses and textual descriptions into a joint embedding space, allowing for\ncross-modal retrieval of relevant poses from large-scale datasets. Secondly, we\nestablish a baseline for a text-conditioned model generating 3D poses. Thirdly,\nwe present a learned process for generating pose descriptions. These\napplications demonstrate the versatility and usefulness of annotated poses in\nvarious tasks and pave the way for future research in the field.\n", "date": "2022-10-21", "categories": []}
{"id": "2409.06662", "title": "World-Grounded Human Motion Recovery via Gravity-View Coordinates", "abstract": "  We present a novel method for recovering world-grounded human motion from\nmonocular video. The main challenge lies in the ambiguity of defining the world\ncoordinate system, which varies between sequences. Previous approaches attempt\nto alleviate this issue by predicting relative motion in an autoregressive\nmanner, but are prone to accumulating errors. Instead, we propose estimating\nhuman poses in a novel Gravity-View (GV) coordinate system, which is defined by\nthe world gravity and the camera view direction. The proposed GV system is\nnaturally gravity-aligned and uniquely defined for each video frame, largely\nreducing the ambiguity of learning image-pose mapping. The estimated poses can\nbe transformed back to the world coordinate system using camera rotations,\nforming a global motion sequence. Additionally, the per-frame estimation avoids\nerror accumulation in the autoregressive methods. Experiments on in-the-wild\nbenchmarks demonstrate that our method recovers more realistic motion in both\nthe camera space and world-grounded settings, outperforming state-of-the-art\nmethods in both accuracy and speed. The code is available at\nhttps://zju3dv.github.io/gvhmr/.\n", "date": "2024-09-10", "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": "2308.06493", "title": "EgoPoser: Robust Real-Time Egocentric Pose Estimation from Sparse and\n  Intermittent Observations Everywhere", "abstract": "  Full-body egocentric pose estimation from head and hand poses alone has\nbecome an active area of research to power articulate avatar representations on\nheadset-based platforms. However, existing methods over-rely on the indoor\nmotion-capture spaces in which datasets were recorded, while simultaneously\nassuming continuous joint motion capture and uniform body dimensions. We\npropose EgoPoser to overcome these limitations with four main contributions. 1)\nEgoPoser robustly models body pose from intermittent hand position and\norientation tracking only when inside a headset's field of view. 2) We rethink\ninput representations for headset-based ego-pose estimation and introduce a\nnovel global motion decomposition method that predicts full-body pose\nindependent of global positions. 3) We enhance pose estimation by capturing\nlonger motion time series through an efficient SlowFast module design that\nmaintains computational efficiency. 4) EgoPoser generalizes across various body\nshapes for different users. We experimentally evaluate our method and show that\nit outperforms state-of-the-art methods both qualitatively and quantitatively\nwhile maintaining a high inference speed of over 600fps. EgoPoser establishes a\nrobust baseline for future work where full-body pose estimation no longer needs\nto rely on outside-in capture and can scale to large-scale and unseen\nenvironments.\n", "date": "2023-08-12", "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": "2409.00087", "title": "A Lightweight Human Pose Estimation Approach for Edge Computing-Enabled\n  Metaverse with Compressive Sensing", "abstract": "  The ability to estimate 3D movements of users over edge computing-enabled\nnetworks, such as 5G/6G networks, is a key enabler for the new era of extended\nreality (XR) and Metaverse applications. Recent advancements in deep learning\nhave shown advantages over optimization techniques for estimating 3D human\nposes given spare measurements from sensor signals, i.e., inertial measurement\nunit (IMU) sensors attached to the XR devices. However, the existing works lack\napplicability to wireless systems, where transmitting the IMU signals over\nnoisy wireless networks poses significant challenges. Furthermore, the\npotential redundancy of the IMU signals has not been considered, resulting in\nhighly redundant transmissions. In this work, we propose a novel approach for\nredundancy removal and lightweight transmission of IMU signals over noisy\nwireless environments. Our approach utilizes a random Gaussian matrix to\ntransform the original signal into a lower-dimensional space. By leveraging the\ncompressive sensing theory, we have proved that the designed Gaussian matrix\ncan project the signal into a lower-dimensional space and preserve the\nSet-Restricted Eigenvalue condition, subject to a power transmission\nconstraint. Furthermore, we develop a deep generative model at the receiver to\nrecover the original IMU signals from noisy compressed data, thus enabling the\ncreation of 3D human body movements at the receiver for XR and Metaverse\napplications. Simulation results on a real-world IMU dataset show that our\nframework can achieve highly accurate 3D human poses of the user using only\n$82\\%$ of the measurements from the original signals. This is comparable to an\noptimization-based approach, i.e., Lasso, but is an order of magnitude faster.\n", "date": "2024-08-26", "categories": []}
{"id": "2409.01555", "title": "EA-RAS: Towards Efficient and Accurate End-to-End Reconstruction of\n  Anatomical Skeleton", "abstract": "  Efficient, accurate and low-cost estimation of human skeletal information is\ncrucial for a range of applications such as biology education and\nhuman-computer interaction. However, current simple skeleton models, which are\ntypically based on 2D-3D joint points, fall short in terms of anatomical\nfidelity, restricting their utility in fields. On the other hand, more complex\nmodels while anatomically precise, are hindered by sophisticate multi-stage\nprocessing and the need for extra data like skin meshes, making them unsuitable\nfor real-time applications. To this end, we propose the EA-RAS (Towards\nEfficient and Accurate End-to-End Reconstruction of Anatomical Skeleton), a\nsingle-stage, lightweight, and plug-and-play anatomical skeleton estimator that\ncan provide real-time, accurate anatomically realistic skeletons with arbitrary\npose using only a single RGB image input. Additionally, EA-RAS estimates the\nconventional human-mesh model explicitly, which not only enhances the\nfunctionality but also leverages the outside skin information by integrating\nfeatures into the inside skeleton modeling process. In this work, we also\ndevelop a progressive training strategy and integrated it with an enhanced\noptimization process, enabling the network to obtain initial weights using only\na small skin dataset and achieve self-supervision in skeleton reconstruction.\nBesides, we also provide an optional lightweight post-processing optimization\nstrategy to further improve accuracy for scenarios that prioritize precision\nover real-time processing. The experiments demonstrated that our regression\nmethod is over 800 times faster than existing methods, meeting real-time\nrequirements. Additionally, the post-processing optimization strategy provided\ncan enhance reconstruction accuracy by over 50% and achieve a speed increase of\nmore than 7 times.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.01879", "title": "SPiKE: 3D Human Pose from Point Cloud Sequences", "abstract": "  3D Human Pose Estimation (HPE) is the task of locating keypoints of the human\nbody in 3D space from 2D or 3D representations such as RGB images, depth maps\nor point clouds. Current HPE methods from depth and point clouds predominantly\nrely on single-frame estimation and do not exploit temporal information from\nsequences. This paper presents SPiKE, a novel approach to 3D HPE using point\ncloud sequences. Unlike existing methods that process frames of a sequence\nindependently, SPiKE leverages temporal context by adopting a Transformer\narchitecture to encode spatio-temporal relationships between points across the\nsequence. By partitioning the point cloud into local volumes and using spatial\nfeature extraction via point spatial convolution, SPiKE ensures efficient\nprocessing by the Transformer while preserving spatial integrity per timestamp.\nExperiments on the ITOP benchmark for 3D HPE show that SPiKE reaches 89.19%\nmAP, achieving state-of-the-art performance with significantly lower inference\ntimes. Extensive ablations further validate the effectiveness of sequence\nexploitation and our algorithmic choices. Code and models are available at:\nhttps://github.com/iballester/SPiKE\n", "date": "2024-09-03", "categories": []}
{"id": "2408.15810", "title": "Multi-view Pose Fusion for Occlusion-Aware 3D Human Pose Estimation", "abstract": "  Robust 3D human pose estimation is crucial to ensure safe and effective\nhuman-robot collaboration. Accurate human perception,however, is particularly\nchallenging in these scenarios due to strong occlusions and limited camera\nviewpoints. Current 3D human pose estimation approaches are rather vulnerable\nin such conditions. In this work we present a novel approach for robust 3D\nhuman pose estimation in the context of human-robot collaboration. Instead of\nrelying on noisy 2D features triangulation, we perform multi-view fusion on 3D\nskeletons provided by absolute monocular methods. Accurate 3D pose estimation\nis then obtained via reprojection error optimization, introducing limbs length\nsymmetry constraints. We evaluate our approach on the public dataset Human3.6M\nand on a novel version Human3.6M-Occluded, derived adding synthetic occlusions\non the camera views with the purpose of testing pose estimation algorithms\nunder severe occlusions. We further validate our method on real human-robot\ncollaboration workcells, in which we strongly surpass current 3D human pose\nestimation methods. Our approach outperforms state-of-the-art multi-view human\npose estimation techniques and demonstrates superior capabilities in handling\nchallenging scenarios with strong occlusions, representing a reliable and\neffective solution for real human-robot collaboration setups.\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": "2408.02110", "title": "AvatarPose: Avatar-guided 3D Pose Estimation of Close Human Interaction\n  from Sparse Multi-view Videos", "abstract": "  Despite progress in human motion capture, existing multi-view methods often\nface challenges in estimating the 3D pose and shape of multiple closely\ninteracting people. This difficulty arises from reliance on accurate 2D joint\nestimations, which are hard to obtain due to occlusions and body contact when\npeople are in close interaction. To address this, we propose a novel method\nleveraging the personalized implicit neural avatar of each individual as a\nprior, which significantly improves the robustness and precision of this\nchallenging pose estimation task. Concretely, the avatars are efficiently\nreconstructed via layered volume rendering from sparse multi-view videos. The\nreconstructed avatar prior allows for the direct optimization of 3D poses based\non color and silhouette rendering loss, bypassing the issues associated with\nnoisy 2D detections. To handle interpenetration, we propose a collision loss on\nthe overlapping shape regions of avatars to add penetration constraints.\nMoreover, both 3D poses and avatars are optimized in an alternating manner. Our\nexperimental results demonstrate state-of-the-art performance on several public\ndatasets.\n", "date": "2024-08-04", "categories": []}
{"id": "2408.10805", "title": "MPL: Lifting 3D Human Pose from Multi-view 2D Poses", "abstract": "  Estimating 3D human poses from 2D images is challenging due to occlusions and\nprojective acquisition. Learning-based approaches have been largely studied to\naddress this challenge, both in single and multi-view setups. These solutions\nhowever fail to generalize to real-world cases due to the lack of (multi-view)\n'in-the-wild' images paired with 3D poses for training. For this reason, we\npropose combining 2D pose estimation, for which large and rich training\ndatasets exist, and 2D-to-3D pose lifting, using a transformer-based network\nthat can be trained from synthetic 2D-3D pose pairs. Our experiments\ndemonstrate decreases up to 45% in MPJPE errors compared to the 3D pose\nobtained by triangulating the 2D poses. The framework's source code is\navailable at https://github.com/aghasemzadeh/OpenMPL .\n", "date": "2024-08-20", "categories": []}
{"id": "2403.18080", "title": "EgoPoseFormer: A Simple Baseline for Stereo Egocentric 3D Human Pose\n  Estimation", "abstract": "  We present EgoPoseFormer, a simple yet effective transformer-based model for\nstereo egocentric human pose estimation. The main challenge in egocentric pose\nestimation is overcoming joint invisibility, which is caused by self-occlusion\nor a limited field of view (FOV) of head-mounted cameras. Our approach\novercomes this challenge by incorporating a two-stage pose estimation paradigm:\nin the first stage, our model leverages the global information to estimate each\njoint's coarse location, then in the second stage, it employs a DETR style\ntransformer to refine the coarse locations by exploiting fine-grained stereo\nvisual features. In addition, we present a Deformable Stereo Attention\noperation to enable our transformer to effectively process multi-view features,\nwhich enables it to accurately localize each joint in the 3D world. We evaluate\nour method on the stereo UnrealEgo dataset and show it significantly\noutperforms previous approaches while being computationally efficient: it\nimproves MPJPE by 27.4mm (45% improvement) with only 7.9% model parameters and\n13.1% FLOPs compared to the state-of-the-art. Surprisingly, with proper\ntraining settings, we find that even our first-stage pose proposal network can\nachieve superior performance compared to previous arts. We also show that our\nmethod can be seamlessly extended to monocular settings, which achieves\nstate-of-the-art performance on the SceneEgo dataset, improving MPJPE by 25.5mm\n(21% improvement) compared to the best existing method with only 60.7% model\nparameters and 36.4% FLOPs. Code is available at:\nhttps://github.com/ChenhongyiYang/egoposeformer .\n", "date": "2024-03-26", "categories": []}
{"id": "2311.14242", "title": "RSB-Pose: Robust Short-Baseline Binocular 3D Human Pose Estimation with\n  Occlusion Handling", "abstract": "  In the domain of 3D Human Pose Estimation, which finds widespread daily\napplications, the requirement for convenient acquisition equipment continues to\ngrow. To satisfy this demand, we set our sights on a short-baseline binocular\nsetting that offers both portability and a geometric measurement property that\nradically mitigates depth ambiguity. However, as the binocular baseline\nshortens, two serious challenges emerge: first, the robustness of 3D\nreconstruction against 2D errors deteriorates; and second, occlusion reoccurs\ndue to the limited visual differences between two views. To address the first\nchallenge, we propose the Stereo Co-Keypoints Estimation module to improve the\nview consistency of 2D keypoints and enhance the 3D robustness. In this module,\nthe disparity is utilized to represent the correspondence of binocular 2D\npoints and the Stereo Volume Feature is introduced to contain binocular\nfeatures across different disparities. Through the regression of SVF, two-view\n2D keypoints are simultaneously estimated in a collaborative way which\nrestricts their view consistency. Furthermore, to deal with occlusions, a\nPre-trained Pose Transformer module is introduced. Through this module, 3D\nposes are refined by perceiving pose coherence, a representation of joint\ncorrelations. This perception is injected by the Pose Transformer network and\nlearned through a pre-training task that recovers iterative masked joints.\nComprehensive experiments carried out on H36M and MHAD datasets, complemented\nby visualizations, validate the effectiveness of our approach in the\nshort-baseline binocular 3D Human Pose Estimation and occlusion handling.\n", "date": "2023-11-24", "categories": []}
{"id": "2408.02285", "title": "Joint-Motion Mutual Learning for Pose Estimation in Videos", "abstract": "  Human pose estimation in videos has long been a compelling yet challenging\ntask within the realm of computer vision. Nevertheless, this task remains\ndifficult because of the complex video scenes, such as video defocus and\nself-occlusion. Recent methods strive to integrate multi-frame visual features\ngenerated by a backbone network for pose estimation. However, they often ignore\nthe useful joint information encoded in the initial heatmap, which is a\nby-product of the backbone generation. Comparatively, methods that attempt to\nrefine the initial heatmap fail to consider any spatio-temporal motion\nfeatures. As a result, the performance of existing methods for pose estimation\nfalls short due to the lack of ability to leverage both local joint (heatmap)\ninformation and global motion (feature) dynamics.\n  To address this problem, we propose a novel joint-motion mutual learning\nframework for pose estimation, which effectively concentrates on both local\njoint dependency and global pixel-level motion dynamics. Specifically, we\nintroduce a context-aware joint learner that adaptively leverages initial\nheatmaps and motion flow to retrieve robust local joint feature. Given that\nlocal joint feature and global motion flow are complementary, we further\npropose a progressive joint-motion mutual learning that synergistically\nexchanges information and interactively learns between joint feature and motion\nflow to improve the capability of the model. More importantly, to capture more\ndiverse joint and motion cues, we theoretically analyze and propose an\ninformation orthogonality objective to avoid learning redundant information\nfrom multi-cues. Empirical experiments show our method outperforms prior arts\non three challenging benchmarks.\n", "date": "2024-08-05", "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": "2407.19077", "title": "Flexible graph convolutional network for 3D human pose estimation", "abstract": "  Although graph convolutional networks exhibit promising performance in 3D\nhuman pose estimation, their reliance on one-hop neighbors limits their ability\nto capture high-order dependencies among body joints, crucial for mitigating\nuncertainty arising from occlusion or depth ambiguity. To tackle this\nlimitation, we introduce Flex-GCN, a flexible graph convolutional network\ndesigned to learn graph representations that capture broader global information\nand dependencies. At its core is the flexible graph convolution, which\naggregates features from both immediate and second-order neighbors of each\nnode, while maintaining the same time and memory complexity as the standard\nconvolution. Our network architecture comprises residual blocks of flexible\ngraph convolutional layers, as well as a global response normalization layer\nfor global feature aggregation, normalization and calibration. Quantitative and\nqualitative results demonstrate the effectiveness of our model, achieving\ncompetitive performance on benchmark datasets.\n", "date": "2024-07-26", "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": "2407.16137", "title": "3D-UGCN: A Unified Graph Convolutional Network for Robust 3D Human Pose\n  Estimation from Monocular RGB Images", "abstract": "  Human pose estimation remains a multifaceted challenge in computer vision,\npivotal across diverse domains such as behavior recognition, human-computer\ninteraction, and pedestrian tracking. This paper proposes an improved method\nbased on the spatial-temporal graph convolution net-work (UGCN) to address the\nissue of missing human posture skeleton sequences in single-view videos. We\npresent the improved UGCN, which allows the network to process 3D human pose\ndata and improves the 3D human pose skeleton sequence, thereby resolving the\nocclusion issue.\n", "date": "2024-07-23", "categories": []}
{"id": "2403.16198", "title": "Diffusion Model is a Good Pose Estimator from 3D RF-Vision", "abstract": "  Human pose estimation (HPE) from Radio Frequency vision (RF-vision) performs\nhuman sensing using RF signals that penetrate obstacles without revealing\nprivacy (e.g., facial information). Recently, mmWave radar has emerged as a\npromising RF-vision sensor, providing radar point clouds by processing RF\nsignals. However, the mmWave radar has a limited resolution with severe noise,\nleading to inaccurate and inconsistent human pose estimation. This work\nproposes mmDiff, a novel diffusion-based pose estimator tailored for noisy\nradar data. Our approach aims to provide reliable guidance as conditions to\ndiffusion models. Two key challenges are addressed by mmDiff: (1)\nmiss-detection of parts of human bodies, which is addressed by a module that\nisolates feature extraction from different body parts, and (2) signal\ninconsistency due to environmental interference, which is tackled by\nincorporating prior knowledge of body structure and motion. Several modules are\ndesigned to achieve these goals, whose features work as the conditions for the\nsubsequent diffusion model, eliminating the miss-detection and instability of\nHPE based on RF-vision. Extensive experiments demonstrate that mmDiff\noutperforms existing methods significantly, achieving state-of-the-art\nperformances on public datasets.\n", "date": "2024-03-24", "categories": []}
{"id": "2407.14605", "title": "ESCAPE: Energy-based Selective Adaptive Correction for\n  Out-of-distribution 3D Human Pose Estimation", "abstract": "  Despite recent advances in human pose estimation (HPE), poor generalization\nto out-of-distribution (OOD) data remains a difficult problem. While previous\nworks have proposed Test-Time Adaptation (TTA) to bridge the train-test domain\ngap by refining network parameters at inference, the absence of ground-truth\nannotations makes it highly challenging and existing methods typically increase\ninference times by one or more orders of magnitude. We observe that 1) not\nevery test time sample is OOD, and 2) HPE errors are significantly larger on\ndistal keypoints (wrist, ankle). To this end, we propose ESCAPE: a lightweight\ncorrection and selective adaptation framework which applies a fast,\nforward-pass correction on most data while reserving costly TTA for OOD data.\nThe free energy function is introduced to separate OOD samples from incoming\ndata and a correction network is trained to estimate the errors of pretrained\nbackbone HPE predictions on the distal keypoints. For OOD samples, we propose a\nnovel self-consistency adaptation loss to update the correction network by\nleveraging the constraining relationship between distal keypoints and proximal\nkeypoints (shoulders, hips), via a second ``reverse\" network. ESCAPE improves\nthe distal MPJPE of five popular HPE models by up to 7% on unseen data,\nachieves state-of-the-art results on two popular HPE benchmarks, and is\nsignificantly faster than existing adaptation methods.\n", "date": "2024-07-19", "categories": []}
{"id": "2407.13930", "title": "RT-Pose: A 4D Radar Tensor-based 3D Human Pose Estimation and\n  Localization Benchmark", "abstract": "  Traditional methods for human localization and pose estimation (HPE), which\nmainly rely on RGB images as an input modality, confront substantial\nlimitations in real-world applications due to privacy concerns. In contrast,\nradar-based HPE methods emerge as a promising alternative, characterized by\ndistinctive attributes such as through-wall recognition and privacy-preserving,\nrendering the method more conducive to practical deployments. This paper\npresents a Radar Tensor-based human pose (RT-Pose) dataset and an open-source\nbenchmarking framework. The RT-Pose dataset comprises 4D radar tensors, LiDAR\npoint clouds, and RGB images, and is collected for a total of 72k frames across\n240 sequences with six different complexity-level actions. The 4D radar tensor\nprovides raw spatio-temporal information, differentiating it from other radar\npoint cloud-based datasets. We develop an annotation process using RGB images\nand LiDAR point clouds to accurately label 3D human skeletons. In addition, we\npropose HRRadarPose, the first single-stage architecture that extracts the\nhigh-resolution representation of 4D radar tensors in 3D space to aid human\nkeypoint estimation. HRRadarPose outperforms previous radar-based HPE work on\nthe RT-Pose benchmark. The overall HRRadarPose performance on the RT-Pose\ndataset, as reflected in a mean per joint position error (MPJPE) of 9.91cm,\nindicates the persistent challenges in achieving accurate HPE in complex\nreal-world scenarios. RT-Pose is available at\nhttps://huggingface.co/datasets/uwipl/RT-Pose.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.14043", "title": "Kinematics-based 3D Human-Object Interaction Reconstruction from Single\n  View", "abstract": "  Reconstructing 3D human-object interaction (HOI) from single-view RGB images\nis challenging due to the absence of depth information and potential\nocclusions. Existing methods simply predict the body poses merely rely on\nnetwork training on some indoor datasets, which cannot guarantee the\nrationality of the results if some body parts are invisible due to occlusions\nthat appear easily. Inspired by the end-effector localization task in robotics,\nwe propose a kinematics-based method that can drive the joints of human body to\nthe human-object contact regions accurately. After an improved forward\nkinematics algorithm is proposed, the Multi-Layer Perceptron is introduced into\nthe solution of inverse kinematics process to determine the poses of joints,\nwhich achieves precise results than the commonly-used numerical methods in\nrobotics. Besides, a Contact Region Recognition Network (CRRNet) is also\nproposed to robustly determine the contact regions using a single-view video.\nExperimental results demonstrate that our method outperforms the\nstate-of-the-art on benchmark BEHAVE. Additionally, our approach shows good\nportability and can be seamlessly integrated into other methods for\noptimizations.\n", "date": "2024-07-19", "categories": []}
{"id": "2407.10756", "title": "GTPT: Group-based Token Pruning Transformer for Efficient Human Pose\n  Estimation", "abstract": "  In recent years, 2D human pose estimation has made significant progress on\npublic benchmarks. However, many of these approaches face challenges of less\napplicability in the industrial community due to the large number of parametric\nquantities and computational overhead. Efficient human pose estimation remains\na hurdle, especially for whole-body pose estimation with numerous keypoints.\nWhile most current methods for efficient human pose estimation primarily rely\non CNNs, we propose the Group-based Token Pruning Transformer (GTPT) that fully\nharnesses the advantages of the Transformer. GTPT alleviates the computational\nburden by gradually introducing keypoints in a coarse-to-fine manner. It\nminimizes the computation overhead while ensuring high performance. Besides,\nGTPT groups keypoint tokens and prunes visual tokens to improve model\nperformance while reducing redundancy. We propose the Multi-Head Group\nAttention (MHGA) between different groups to achieve global interaction with\nlittle computational overhead. We conducted experiments on COCO and\nCOCO-WholeBody. Compared to other methods, the experimental results show that\nGTPT can achieve higher performance with less computation, especially in\nwhole-body with numerous keypoints.\n", "date": "2024-07-15", "categories": []}
{"id": "2306.17201", "title": "MPM: A Unified 2D-3D Human Pose Representation via Masked Pose Modeling", "abstract": "  Estimating 3D human poses only from a 2D human pose sequence is thoroughly\nexplored in recent years. Yet, prior to this, no such work has attempted to\nunify 2D and 3D pose representations in the shared feature space. In this\npaper, we propose \\mpm, a unified 2D-3D human pose representation framework via\nmasked pose modeling. We treat 2D and 3D poses as two different modalities like\nvision and language and build a single-stream transformer-based architecture.\nWe apply two pretext tasks, which are masked 2D pose modeling, and masked 3D\npose modeling to pre-train our network and use full-supervision to perform\nfurther fine-tuning. A high masking ratio of $71.8~\\%$ in total with a\nspatio-temporal mask sampling strategy leads to better relation modeling both\nin spatial and temporal domains. \\mpm~can handle multiple tasks including 3D\nhuman pose estimation, 3D pose estimation from occluded 2D pose, and 3D pose\ncompletion in a \\textbf{single} framework. We conduct extensive experiments and\nablation studies on several widely used human pose datasets and achieve\nstate-of-the-art performance on MPI-INF-3DHP.\n", "date": "2023-06-29", "categories": []}
{"id": "2312.08291", "title": "VQ-HPS: Human Pose and Shape Estimation in a Vector-Quantized Latent\n  Space", "abstract": "  Previous works on Human Pose and Shape Estimation (HPSE) from RGB images can\nbe broadly categorized into two main groups: parametric and non-parametric\napproaches. Parametric techniques leverage a low-dimensional statistical body\nmodel for realistic results, whereas recent non-parametric methods achieve\nhigher precision by directly regressing the 3D coordinates of the human body\nmesh. This work introduces a novel paradigm to address the HPSE problem,\ninvolving a low-dimensional discrete latent representation of the human mesh\nand framing HPSE as a classification task. Instead of predicting body model\nparameters or 3D vertex coordinates, we focus on predicting the proposed\ndiscrete latent representation, which can be decoded into a registered human\nmesh. This innovative paradigm offers two key advantages. Firstly, predicting a\nlow-dimensional discrete representation confines our predictions to the space\nof anthropomorphic poses and shapes even when little training data is\navailable. Secondly, by framing the problem as a classification task, we can\nharness the discriminative power inherent in neural networks. The proposed\nmodel, VQ-HPS, predicts the discrete latent representation of the mesh. The\nexperimental results demonstrate that VQ-HPS outperforms the current\nstate-of-the-art non-parametric approaches while yielding results as realistic\nas those produced by parametric methods when trained with little data. VQ-HPS\nalso shows promising results when training on large-scale datasets,\nhighlighting the significant potential of the classification approach for HPSE.\nSee the project page at https://g-fiche.github.io/research-pages/vqhps/\n", "date": "2023-12-13", "categories": []}
{"id": "2406.15649", "title": "Efficient Human Pose Estimation: Leveraging Advanced Techniques with\n  MediaPipe", "abstract": "  This study presents significant enhancements in human pose estimation using\nthe MediaPipe framework. The research focuses on improving accuracy,\ncomputational efficiency, and real-time processing capabilities by\ncomprehensively optimising the underlying algorithms. Novel modifications are\nintroduced that substantially enhance pose estimation accuracy across\nchallenging scenarios, such as dynamic movements and partial occlusions. The\nimproved framework is benchmarked against traditional models, demonstrating\nconsiderable precision and computational speed gains. The advancements have\nwide-ranging applications in augmented reality, sports analytics, and\nhealthcare, enabling more immersive experiences, refined performance analysis,\nand advanced patient monitoring. The study also explores the integration of\nthese enhancements within mobile and embedded systems, addressing the need for\ncomputational efficiency and broader accessibility. The implications of this\nresearch set a new benchmark for real-time human pose estimation technologies\nand pave the way for future innovations in the field. The implementation code\nfor the paper is available at https://github.com/avhixd/Human_pose_estimation.\n", "date": "2024-06-21", "categories": []}
{"id": "2407.09833", "title": "LiveHPS++: Robust and Coherent Motion Capture in Dynamic Free\n  Environment", "abstract": "  LiDAR-based human motion capture has garnered significant interest in recent\nyears for its practicability in large-scale and unconstrained environments.\nHowever, most methods rely on cleanly segmented human point clouds as input,\nthe accuracy and smoothness of their motion results are compromised when faced\nwith noisy data, rendering them unsuitable for practical applications. To\naddress these limitations and enhance the robustness and precision of motion\ncapture with noise interference, we introduce LiveHPS++, an innovative and\neffective solution based on a single LiDAR system. Benefiting from three\nmeticulously designed modules, our method can learn dynamic and kinematic\nfeatures from human movements, and further enable the precise capture of\ncoherent human motions in open settings, making it highly applicable to\nreal-world scenarios. Through extensive experiments, LiveHPS++ has proven to\nsignificantly surpass existing state-of-the-art methods across various\ndatasets, establishing a new benchmark in the field.\n", "date": "2024-07-13", "categories": []}
{"id": "2407.08484", "title": "Learning Localization of Body and Finger Animation Skeleton Joints on\n  Three-Dimensional Models of Human Bodies", "abstract": "  Contemporary approaches to solving various problems that require analyzing\nthree-dimensional (3D) meshes and point clouds have adopted the use of deep\nlearning algorithms that directly process 3D data such as point coordinates,\nnormal vectors and vertex connectivity information. Our work proposes one such\nsolution to the problem of positioning body and finger animation skeleton\njoints within 3D models of human bodies. Due to scarcity of annotated real\nhuman scans, we resort to generating synthetic samples while varying their\nshape and pose parameters. Similarly to the state-of-the-art approach, our\nmethod computes each joint location as a convex combination of input points.\nGiven only a list of point coordinates and normal vector estimates as input, a\ndynamic graph convolutional neural network is used to predict the coefficients\nof the convex combinations. By comparing our method with the state-of-the-art,\nwe show that it is possible to achieve significantly better results with a\nsimpler architecture, especially for finger joints. Since our solution requires\nfewer precomputed features, it also allows for shorter processing times.\n", "date": "2024-07-11", "categories": []}
{"id": "2407.08634", "title": "RTMW: Real-Time Multi-Person 2D and 3D Whole-body Pose Estimation", "abstract": "  Whole-body pose estimation is a challenging task that requires simultaneous\nprediction of keypoints for the body, hands, face, and feet. Whole-body pose\nestimation aims to predict fine-grained pose information for the human body,\nincluding the face, torso, hands, and feet, which plays an important role in\nthe study of human-centric perception and generation and in various\napplications. In this work, we present RTMW (Real-Time Multi-person Whole-body\npose estimation models), a series of high-performance models for 2D/3D\nwhole-body pose estimation. We incorporate RTMPose model architecture with FPN\nand HEM (Hierarchical Encoding Module) to better capture pose information from\ndifferent body parts with various scales. The model is trained with a rich\ncollection of open-source human keypoint datasets with manually aligned\nannotations and further enhanced via a two-stage distillation strategy. RTMW\ndemonstrates strong performance on multiple whole-body pose estimation\nbenchmarks while maintaining high inference efficiency and deployment\nfriendliness. We release three sizes: m/l/x, with RTMW-l achieving a 70.2 mAP\non the COCO-Wholebody benchmark, making it the first open-source model to\nexceed 70 mAP on this benchmark. Meanwhile, we explored the performance of RTMW\nin the task of 3D whole-body pose estimation, conducting image-based monocular\n3D whole-body pose estimation in a coordinate classification manner. We hope\nthis work can benefit both academic research and industrial applications. The\ncode and models have been made publicly available at:\nhttps://github.com/open-mmlab/mmpose/tree/main/projects/rtmpose\n", "date": "2024-07-11", "categories": []}
{"id": "2404.14634", "title": "UPose3D: Uncertainty-Aware 3D Human Pose Estimation with Cross-View and\n  Temporal Cues", "abstract": "  We introduce UPose3D, a novel approach for multi-view 3D human pose\nestimation, addressing challenges in accuracy and scalability. Our method\nadvances existing pose estimation frameworks by improving robustness and\nflexibility without requiring direct 3D annotations. At the core of our method,\na pose compiler module refines predictions from a 2D keypoints estimator that\noperates on a single image by leveraging temporal and cross-view information.\nOur novel cross-view fusion strategy is scalable to any number of cameras,\nwhile our synthetic data generation strategy ensures generalization across\ndiverse actors, scenes, and viewpoints. Finally, UPose3D leverages the\nprediction uncertainty of both the 2D keypoint estimator and the pose compiler\nmodule. This provides robustness to outliers and noisy data, resulting in\nstate-of-the-art performance in out-of-distribution settings. In addition, for\nin-distribution settings, UPose3D yields performance rivalling methods that\nrely on 3D annotated data while being the state-of-the-art among methods\nrelying only on 2D supervision.\n", "date": "2024-04-23", "categories": []}
{"id": "2407.07092", "title": "V-VIPE: Variational View Invariant Pose Embedding", "abstract": "  Learning to represent three dimensional (3D) human pose given a two\ndimensional (2D) image of a person, is a challenging problem. In order to make\nthe problem less ambiguous it has become common practice to estimate 3D pose in\nthe camera coordinate space. However, this makes the task of comparing two 3D\nposes difficult. In this paper, we address this challenge by separating the\nproblem of estimating 3D pose from 2D images into two steps. We use a\nvariational autoencoder (VAE) to find an embedding that represents 3D poses in\ncanonical coordinate space. We refer to this embedding as variational\nview-invariant pose embedding V-VIPE. Using V-VIPE we can encode 2D and 3D\nposes and use the embedding for downstream tasks, like retrieval and\nclassification. We can estimate 3D poses from these embeddings using the\ndecoder as well as generate unseen 3D poses. The variability of our encoding\nallows it to generalize well to unseen camera views when mapping from 2D space.\nTo the best of our knowledge, V-VIPE is the only representation to offer this\ndiversity of applications. Code and more information can be found at\nhttps://v-vipe.github.io/.\n", "date": "2024-07-09", "categories": []}
{"id": "2312.07051", "title": "Mask as Supervision: Leveraging Unified Mask Information for\n  Unsupervised 3D Pose Estimation", "abstract": "  Automatic estimation of 3D human pose from monocular RGB images is a\nchallenging and unsolved problem in computer vision. In a supervised manner,\napproaches heavily rely on laborious annotations and present hampered\ngeneralization ability due to the limited diversity of 3D pose datasets. To\naddress these challenges, we propose a unified framework that leverages mask as\nsupervision for unsupervised 3D pose estimation. With general unsupervised\nsegmentation algorithms, the proposed model employs skeleton and physique\nrepresentations that exploit accurate pose information from coarse to fine.\nCompared with previous unsupervised approaches, we organize the human skeleton\nin a fully unsupervised way which enables the processing of annotation-free\ndata and provides ready-to-use estimation results. Comprehensive experiments\ndemonstrate our state-of-the-art pose estimation performance on Human3.6M and\nMPI-INF-3DHP datasets. Further experiments on in-the-wild datasets also\nillustrate the capability to access more data to boost our model. Code will be\navailable at https://github.com/Charrrrrlie/Mask-as-Supervision.\n", "date": "2023-12-12", "categories": []}
{"id": "2407.04384", "title": "Unsupervised Learning of Category-Level 3D Pose from Object-Centric\n  Videos", "abstract": "  Category-level 3D pose estimation is a fundamentally important problem in\ncomputer vision and robotics, e.g. for embodied agents or to train 3D\ngenerative models. However, so far methods that estimate the category-level\nobject pose require either large amounts of human annotations, CAD models or\ninput from RGB-D sensors. In contrast, we tackle the problem of learning to\nestimate the category-level 3D pose only from casually taken object-centric\nvideos without human supervision. We propose a two-step pipeline: First, we\nintroduce a multi-view alignment procedure that determines canonical camera\nposes across videos with a novel and robust cyclic distance formulation for\ngeometric and appearance matching using reconstructed coarse meshes and DINOv2\nfeatures. In a second step, the canonical poses and reconstructed meshes enable\nus to train a model for 3D pose estimation from a single image. In particular,\nour model learns to estimate dense correspondences between images and a\nprototypical 3D template by predicting, for each pixel in a 2D image, a feature\nvector of the corresponding vertex in the template mesh. We demonstrate that\nour method outperforms all baselines at the unsupervised alignment of\nobject-centric videos by a large margin and provides faithful and robust\npredictions in-the-wild. Our code and data is available at\nhttps://github.com/GenIntel/uns-obj-pose3d.\n", "date": "2024-07-05", "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": "2407.01811", "title": "Active Human Pose Estimation via an Autonomous UAV Agent", "abstract": "  One of the core activities of an active observer involves moving to secure a\n\"better\" view of the scene, where the definition of \"better\" is task-dependent.\nThis paper focuses on the task of human pose estimation from videos capturing a\nperson's activity. Self-occlusions within the scene can complicate or even\nprevent accurate human pose estimation. To address this, relocating the camera\nto a new vantage point is necessary to clarify the view, thereby improving 2D\nhuman pose estimation. This paper formalizes the process of achieving an\nimproved viewpoint. Our proposed solution to this challenge comprises three\nmain components: a NeRF-based Drone-View Data Generation Framework, an On-Drone\nNetwork for Camera View Error Estimation, and a Combined Planner for devising a\nfeasible motion plan to reposition the camera based on the predicted errors for\ncamera views. The Data Generation Framework utilizes NeRF-based methods to\ngenerate a comprehensive dataset of human poses and activities, enhancing the\ndrone's adaptability in various scenarios. The Camera View Error Estimation\nNetwork is designed to evaluate the current human pose and identify the most\npromising next viewing angles for the drone, ensuring a reliable and precise\npose estimation from those angles. Finally, the combined planner incorporates\nthese angles while considering the drone's physical and environmental\nlimitations, employing efficient algorithms to navigate safe and effective\nflight paths. This system represents a significant advancement in active 2D\nhuman pose estimation for an autonomous UAV agent, offering substantial\npotential for applications in aerial cinematography by improving the\nperformance of autonomous human pose estimation and maintaining the operational\nsafety and efficiency of UAVs.\n", "date": "2024-07-01", "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": "2406.19726", "title": "EPOCH: Jointly Estimating the 3D Pose of Cameras and Humans", "abstract": "  Monocular Human Pose Estimation (HPE) aims at determining the 3D positions of\nhuman joints from a single 2D image captured by a camera. However, a single 2D\npoint in the image may correspond to multiple points in 3D space. Typically,\nthe uniqueness of the 2D-3D relationship is approximated using an orthographic\nor weak-perspective camera model. In this study, instead of relying on\napproximations, we advocate for utilizing the full perspective camera model.\nThis involves estimating camera parameters and establishing a precise,\nunambiguous 2D-3D relationship. To do so, we introduce the EPOCH framework,\ncomprising two main components: the pose lifter network (LiftNet) and the pose\nregressor network (RegNet). LiftNet utilizes the full perspective camera model\nto precisely estimate the 3D pose in an unsupervised manner. It takes a 2D pose\nand camera parameters as inputs and produces the corresponding 3D pose\nestimation. These inputs are obtained from RegNet, which starts from a single\nimage and provides estimates for the 2D pose and camera parameters. RegNet\nutilizes only 2D pose data as weak supervision. Internally, RegNet predicts a\n3D pose, which is then projected to 2D using the estimated camera parameters.\nThis process enables RegNet to establish the unambiguous 2D-3D relationship.\nOur experiments show that modeling the lifting as an unsupervised task with a\ncamera in-the-loop results in better generalization to unseen data. We obtain\nstate-of-the-art results for the 3D HPE on the Human3.6M and MPI-INF-3DHP\ndatasets. Our code is available at: [Github link upon acceptance, see\nsupplementary materials].\n", "date": "2024-06-28", "categories": []}
{"id": "2406.19290", "title": "Human Modelling and Pose Estimation Overview", "abstract": "  Human modelling and pose estimation stands at the crossroads of Computer\nVision, Computer Graphics, and Machine Learning. This paper presents a thorough\ninvestigation of this interdisciplinary field, examining various algorithms,\nmethodologies, and practical applications. It explores the diverse range of\nsensor technologies relevant to this domain and delves into a wide array of\napplication areas. Additionally, we discuss the challenges and advancements in\n2D and 3D human modelling methodologies, along with popular datasets, metrics,\nand future research directions. The main contribution of this paper lies in its\nup-to-date comparison of state-of-the-art (SOTA) human pose estimation\nalgorithms in both 2D and 3D domains. By providing this comprehensive overview,\nthe paper aims to enhance understanding of 3D human modelling and pose\nestimation, offering insights into current SOTA achievements, challenges, and\nfuture prospects within the field.\n", "date": "2024-06-27", "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": "2403.12434", "title": "Human Mesh Recovery from Arbitrary Multi-view Images", "abstract": "  Human mesh recovery from arbitrary multi-view images involves two\ncharacteristics: the arbitrary camera poses and arbitrary number of camera\nviews. Because of the variability, designing a unified framework to tackle this\ntask is challenging. The challenges can be summarized as the dilemma of being\nable to simultaneously estimate arbitrary camera poses and recover human mesh\nfrom arbitrary multi-view images while maintaining flexibility. To solve this\ndilemma, we propose a divide and conquer framework for Unified Human Mesh\nRecovery (U-HMR) from arbitrary multi-view images. In particular, U-HMR\nconsists of a decoupled structure and two main components: camera and body\ndecoupling (CBD), camera pose estimation (CPE), and arbitrary view fusion\n(AVF). As camera poses and human body mesh are independent of each other, CBD\nsplits the estimation of them into two sub-tasks for two individual\nsub-networks (ie, CPE and AVF) to handle respectively, thus the two sub-tasks\nare disentangled. In CPE, since each camera pose is unrelated to the others, we\nadopt a shared MLP to process all views in a parallel way. In AVF, in order to\nfuse multi-view information and make the fusion operation independent of the\nnumber of views, we introduce a transformer decoder with a SMPL parameters\nquery token to extract cross-view features for mesh recovery. To demonstrate\nthe efficacy and flexibility of the proposed framework and effect of each\ncomponent, we conduct extensive experiments on three public datasets:\nHuman3.6M, MPI-INF-3DHP, and TotalCapture.\n", "date": "2024-03-19", "categories": []}
{"id": "2406.09788", "title": "OpenCapBench: A Benchmark to Bridge Pose Estimation and Biomechanics", "abstract": "  Pose estimation has promised to impact healthcare by enabling more practical\nmethods to quantify nuances of human movement and biomechanics. However,\ndespite the inherent connection between pose estimation and biomechanics, these\ndisciplines have largely remained disparate. For example, most current pose\nestimation benchmarks use metrics such as Mean Per Joint Position Error,\nPercentage of Correct Keypoints, or mean Average Precision to assess\nperformance, without quantifying kinematic and physiological correctness - key\naspects for biomechanics. To alleviate this challenge, we develop OpenCapBench\nto offer an easy-to-use unified benchmark to assess common tasks in human pose\nestimation, evaluated under physiological constraints. OpenCapBench computes\nconsistent kinematic metrics through joints angles provided by an open-source\nmusculoskeletal modeling software (OpenSim). Through OpenCapBench, we\ndemonstrate that current pose estimation models use keypoints that are too\nsparse for accurate biomechanics analysis. To mitigate this challenge, we\nintroduce SynthPose, a new approach that enables finetuning of pre-trained 2D\nhuman pose models to predict an arbitrarily denser set of keypoints for\naccurate kinematic analysis through the use of synthetic data. Incorporating\nsuch finetuning on synthetic data of prior models leads to twofold reduced\njoint angle errors. Moreover, OpenCapBench allows users to benchmark their own\ndeveloped models on our clinically relevant cohort. Overall, OpenCapBench\nbridges the computer vision and biomechanics communities, aiming to drive\nsimultaneous advances in both areas.\n", "date": "2024-06-14", "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": "2404.02041", "title": "SelfPose3d: Self-Supervised Multi-Person Multi-View 3d Pose Estimation", "abstract": "  We present a new self-supervised approach, SelfPose3d, for estimating 3d\nposes of multiple persons from multiple camera views. Unlike current\nstate-of-the-art fully-supervised methods, our approach does not require any 2d\nor 3d ground-truth poses and uses only the multi-view input images from a\ncalibrated camera setup and 2d pseudo poses generated from an off-the-shelf 2d\nhuman pose estimator. We propose two self-supervised learning objectives:\nself-supervised person localization in 3d space and self-supervised 3d pose\nestimation. We achieve self-supervised 3d person localization by training the\nmodel on synthetically generated 3d points, serving as 3d person root\npositions, and on the projected root-heatmaps in all the views. We then model\nthe 3d poses of all the localized persons with a bottleneck representation, map\nthem onto all views obtaining 2d joints, and render them using 2d Gaussian\nheatmaps in an end-to-end differentiable manner. Afterwards, we use the\ncorresponding 2d joints and heatmaps from the pseudo 2d poses for learning. To\nalleviate the intrinsic inaccuracy of the pseudo labels, we propose an adaptive\nsupervision attention mechanism to guide the self-supervision. Our experiments\nand analysis on three public benchmark datasets, including Panoptic, Shelf, and\nCampus, show the effectiveness of our approach, which is comparable to\nfully-supervised methods. Code: https://github.com/CAMMA-public/SelfPose3D.\nVideo demo: https://youtu.be/GAqhmUIr2E8.\n", "date": "2024-04-02", "categories": []}
{"id": "2406.05691", "title": "Diverse 3D Human Pose Generation in Scenes based on Decoupled Structure", "abstract": "  This paper presents a novel method for generating diverse 3D human poses in\nscenes with semantic control. Existing methods heavily rely on the human-scene\ninteraction dataset, resulting in a limited diversity of the generated human\nposes. To overcome this challenge, we propose to decouple the pose and\ninteraction generation process. Our approach consists of three stages: pose\ngeneration, contact generation, and putting human into the scene. We train a\npose generator on the human dataset to learn rich pose prior, and a contact\ngenerator on the human-scene interaction dataset to learn human-scene contact\nprior. Finally, the placing module puts the human body into the scene in a\nsuitable and natural manner. The experimental results on the PROX dataset\ndemonstrate that our method produces more physically plausible interactions and\nexhibits more diverse human poses. Furthermore, experiments on the MP3D-R\ndataset further validates the generalization ability of our method.\n", "date": "2024-06-09", "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": "2406.01196", "title": "3D WholeBody Pose Estimation based on Semantic Graph Attention Network\n  and Distance Information", "abstract": "  In recent years, a plethora of diverse methods have been proposed for 3D pose\nestimation. Among these, self-attention mechanisms and graph convolutions have\nboth been proven to be effective and practical methods. Recognizing the\nstrengths of those two techniques, we have developed a novel Semantic Graph\nAttention Network which can benefit from the ability of self-attention to\ncapture global context, while also utilizing the graph convolutions to handle\nthe local connectivity and structural constraints of the skeleton. We also\ndesign a Body Part Decoder that assists in extracting and refining the\ninformation related to specific segments of the body. Furthermore, our approach\nincorporates Distance Information, enhancing our model's capability to\ncomprehend and accurately predict spatial relationships. Finally, we introduce\na Geometry Loss who makes a critical constraint on the structural skeleton of\nthe body, ensuring that the model's predictions adhere to the natural limits of\nhuman posture. The experimental results validate the effectiveness of our\napproach, demonstrating that every element within the system is essential for\nimproving pose estimation outcomes. With comparison to state-of-the-art, the\nproposed work not only meets but exceeds the existing benchmarks.\n", "date": "2024-06-03", "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": "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": "2405.17016", "title": "$\\text{Di}^2\\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human\n  Pose Estimation", "abstract": "  Continuous diffusion models have demonstrated their effectiveness in\naddressing the inherent uncertainty and indeterminacy in monocular 3D human\npose estimation (HPE). Despite their strengths, the need for large search\nspaces and the corresponding demand for substantial training data make these\nmodels prone to generating biomechanically unrealistic poses. This challenge is\nparticularly noticeable in occlusion scenarios, where the complexity of\ninferring 3D structures from 2D images intensifies. In response to these\nlimitations, we introduce the Discrete Diffusion Pose\n($\\text{Di}^2\\text{Pose}$), a novel framework designed for occluded 3D HPE that\ncapitalizes on the benefits of a discrete diffusion model. Specifically,\n$\\text{Di}^2\\text{Pose}$ employs a two-stage process: it first converts 3D\nposes into a discrete representation through a \\emph{pose quantization step},\nwhich is subsequently modeled in latent space through a \\emph{discrete\ndiffusion process}. This methodological innovation restrictively confines the\nsearch space towards physically viable configurations and enhances the model's\ncapability to comprehend how occlusions affect human pose within the latent\nspace. Extensive evaluations conducted on various benchmarks (e.g., Human3.6M,\n3DPW, and 3DPW-Occ) have demonstrated its effectiveness.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.17397", "title": "Occlusion Handling in 3D Human Pose Estimation with Perturbed Positional\n  Encoding", "abstract": "  Understanding human behavior fundamentally relies on accurate 3D human pose\nestimation. Graph Convolutional Networks (GCNs) have recently shown promising\nadvancements, delivering state-of-the-art performance with rather lightweight\narchitectures. In the context of graph-structured data, leveraging the\neigenvectors of the graph Laplacian matrix for positional encoding is\neffective. Yet, the approach does not specify how to handle scenarios where\nedges in the input graph are missing. To this end, we propose a novel\npositional encoding technique, PerturbPE, that extracts consistent and regular\ncomponents from the eigenbasis. Our method involves applying multiple\nperturbations and taking their average to extract the consistent and regular\ncomponent from the eigenbasis. PerturbPE leverages the Rayleigh-Schrodinger\nPerturbation Theorem (RSPT) for calculating the perturbed eigenvectors.\nEmploying this labeling technique enhances the robustness and generalizability\nof the model. Our results support our theoretical findings, e.g. our\nexperimental analysis observed a performance enhancement of up to $12\\%$ on the\nHuman3.6M dataset in instances where occlusion resulted in the absence of one\nedge. Furthermore, our novel approach significantly enhances performance in\nscenarios where two edges are missing, setting a new benchmark for\nstate-of-the-art.\n", "date": "2024-05-27", "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": "2405.12724", "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": "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": "2207.10971", "title": "Kinematics Modeling Network for Video-based Human Pose Estimation", "abstract": "  Estimating human poses from videos is critical in human-computer interaction.\nJoints cooperate rather than move independently during human movement. There\nare both spatial and temporal correlations between joints. Despite the positive\nresults of previous approaches, most focus on modeling the spatial correlation\nbetween joints while only straightforwardly integrating features along the\ntemporal dimension, ignoring the temporal correlation between joints. In this\nwork, we propose a plug-and-play kinematics modeling module (KMM) to explicitly\nmodel temporal correlations between joints across different frames by\ncalculating their temporal similarity. In this way, KMM can capture motion cues\nof the current joint relative to all joints in different time. Besides, we\nformulate video-based human pose estimation as a Markov Decision Process and\ndesign a novel kinematics modeling network (KIMNet) to simulate the Markov\nChain, allowing KIMNet to locate joints recursively. Our approach achieves\nstate-of-the-art results on two challenging benchmarks. In particular, KIMNet\nshows robustness to the occlusion. The code will be released at\nhttps://github.com/YHDang/KIMNet.\n", "date": "2022-07-22", "categories": []}
{"id": "2308.11440", "title": "PoseGraphNet++: Enriching 3D Human Pose with Orientation Estimation", "abstract": "  Existing skeleton-based 3D human pose estimation methods only predict joint\npositions. Although the yaw and pitch of bone rotations can be derived from\njoint positions, the roll around the bone axis remains unresolved. We present\nPoseGraphNet++ (PGN++), a novel 2D-to-3D lifting Graph Convolution Network that\npredicts the complete human pose in 3D including joint positions and bone\norientations. We employ both node and edge convolutions to utilize the joint\nand bone features. Our model is evaluated on multiple datasets using both\nposition and rotation metrics. PGN++ performs on par with the state-of-the-art\n(SoA) on the Human3.6M benchmark. In generalization experiments, it achieves\nthe best results in position and matches the SoA in orientation, showcasing a\nmore balanced performance than the current SoA. PGN++ exploits the mutual\nrelationship of joints and bones resulting in significantly \\SB{improved}\nposition predictions, as shown by our ablation results.\n", "date": "2023-08-22", "categories": []}
{"id": "2405.05216", "title": "FinePOSE: Fine-Grained Prompt-Driven 3D Human Pose Estimation via\n  Diffusion Models", "abstract": "  The 3D Human Pose Estimation (3D HPE) task uses 2D images or videos to\npredict human joint coordinates in 3D space. Despite recent advancements in\ndeep learning-based methods, they mostly ignore the capability of coupling\naccessible texts and naturally feasible knowledge of humans, missing out on\nvaluable implicit supervision to guide the 3D HPE task. Moreover, previous\nefforts often study this task from the perspective of the whole human body,\nneglecting fine-grained guidance hidden in different body parts. To this end,\nwe present a new Fine-Grained Prompt-Driven Denoiser based on a diffusion model\nfor 3D HPE, named \\textbf{FinePOSE}. It consists of three core blocks enhancing\nthe reverse process of the diffusion model: (1) Fine-grained Part-aware Prompt\nlearning (FPP) block constructs fine-grained part-aware prompts via coupling\naccessible texts and naturally feasible knowledge of body parts with learnable\nprompts to model implicit guidance. (2) Fine-grained Prompt-pose Communication\n(FPC) block establishes fine-grained communications between learned part-aware\nprompts and poses to improve the denoising quality. (3) Prompt-driven Timestamp\nStylization (PTS) block integrates learned prompt embedding and temporal\ninformation related to the noise level to enable adaptive adjustment at each\ndenoising step. Extensive experiments on public single-human pose estimation\ndatasets show that FinePOSE outperforms state-of-the-art methods. We further\nextend FinePOSE to multi-human pose estimation. Achieving 34.3mm average MPJPE\non the EgoHumans dataset demonstrates the potential of FinePOSE to deal with\ncomplex multi-human scenarios. Code is available at\nhttps://github.com/PKU-ICST-MIPL/FinePOSE_CVPR2024.\n", "date": "2024-05-08", "categories": []}
{"id": "2405.03055", "title": "Multi-hop graph transformer network for 3D human pose estimation", "abstract": "  Accurate 3D human pose estimation is a challenging task due to occlusion and\ndepth ambiguity. In this paper, we introduce a multi-hop graph transformer\nnetwork designed for 2D-to-3D human pose estimation in videos by leveraging the\nstrengths of multi-head self-attention and multi-hop graph convolutional\nnetworks with disentangled neighborhoods to capture spatio-temporal\ndependencies and handle long-range interactions. The proposed network\narchitecture consists of a graph attention block composed of stacked layers of\nmulti-head self-attention and graph convolution with learnable adjacency\nmatrix, and a multi-hop graph convolutional block comprised of multi-hop\nconvolutional and dilated convolutional layers. The combination of multi-head\nself-attention and multi-hop graph convolutional layers enables the model to\ncapture both local and global dependencies, while the integration of dilated\nconvolutional layers enhances the model's ability to handle spatial details\nrequired for accurate localization of the human body joints. Extensive\nexperiments demonstrate the effectiveness and generalization ability of our\nmodel, achieving competitive performance on benchmark datasets.\n", "date": "2024-05-05", "categories": []}
{"id": "2405.02114", "title": "Probablistic Restoration with Adaptive Noise Sampling for 3D Human Pose\n  Estimation", "abstract": "  The accuracy and robustness of 3D human pose estimation (HPE) are limited by\n2D pose detection errors and 2D to 3D ill-posed challenges, which have drawn\ngreat attention to Multi-Hypothesis HPE research. Most existing MH-HPE methods\nare based on generative models, which are computationally expensive and\ndifficult to train. In this study, we propose a Probabilistic Restoration 3D\nHuman Pose Estimation framework (PRPose) that can be integrated with any\nlightweight single-hypothesis model. Specifically, PRPose employs a weakly\nsupervised approach to fit the hidden probability distribution of the 2D-to-3D\nlifting process in the Single-Hypothesis HPE model and then reverse-map the\ndistribution to the 2D pose input through an adaptive noise sampling strategy\nto generate reasonable multi-hypothesis samples effectively. Extensive\nexperiments on 3D HPE benchmarks (Human3.6M and MPI-INF-3DHP) highlight the\neffectiveness and efficiency of PRPose. Code is available at:\nhttps://github.com/xzhouzeng/PRPose.\n", "date": "2024-05-03", "categories": []}
{"id": "2404.19541", "title": "Ultra Inertial Poser: Scalable Motion Capture and Tracking from Sparse\n  Inertial Sensors and Ultra-Wideband Ranging", "abstract": "  While camera-based capture systems remain the gold standard for recording\nhuman motion, learning-based tracking systems based on sparse wearable sensors\nare gaining popularity. Most commonly, they use inertial sensors, whose\npropensity for drift and jitter have so far limited tracking accuracy. In this\npaper, we propose Ultra Inertial Poser, a novel 3D full body pose estimation\nmethod that constrains drift and jitter in inertial tracking via inter-sensor\ndistances. We estimate these distances across sparse sensor setups using a\nlightweight embedded tracker that augments inexpensive off-the-shelf 6D\ninertial measurement units with ultra-wideband radio-based\nranging$-$dynamically and without the need for stationary reference anchors.\nOur method then fuses these inter-sensor distances with the 3D states estimated\nfrom each sensor Our graph-based machine learning model processes the 3D states\nand distances to estimate a person's 3D full body pose and translation. To\ntrain our model, we synthesize inertial measurements and distance estimates\nfrom the motion capture database AMASS. For evaluation, we contribute a novel\nmotion dataset of 10 participants who performed 25 motion types, captured by 6\nwearable IMU+UWB trackers and an optical motion capture system, totaling 200\nminutes of synchronized sensor data (UIP-DB). Our extensive experiments show\nstate-of-the-art performance for our method over PIP and TIP, reducing position\nerror from $13.62$ to $10.65cm$ ($22\\%$ better) and lowering jitter from $1.56$\nto $0.055km/s^3$ (a reduction of $97\\%$).\n", "date": "2024-04-30", "categories": []}
{"id": "2404.17837", "title": "Hybrid 3D Human Pose Estimation with Monocular Video and Sparse IMUs", "abstract": "  Temporal 3D human pose estimation from monocular videos is a challenging task\nin human-centered computer vision due to the depth ambiguity of 2D-to-3D\nlifting. To improve accuracy and address occlusion issues, inertial sensor has\nbeen introduced to provide complementary source of information. However, it\nremains challenging to integrate heterogeneous sensor data for producing\nphysically rational 3D human poses. In this paper, we propose a novel\nframework, Real-time Optimization and Fusion (RTOF), to address this issue. We\nfirst incorporate sparse inertial orientations into a parametric human skeleton\nto refine 3D poses in kinematics. The poses are then optimized by energy\nfunctions built on both visual and inertial observations to reduce the temporal\njitters. Our framework outputs smooth and biomechanically plausible human\nmotion. Comprehensive experiments with ablation studies demonstrate its\nrationality and efficiency. On Total Capture dataset, the pose estimation error\nis significantly decreased compared to the baseline method.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.16136", "title": "3D Human Pose Estimation with Occlusions: Introducing BlendMimic3D\n  Dataset and GCN Refinement", "abstract": "  In the field of 3D Human Pose Estimation (HPE), accurately estimating human\npose, especially in scenarios with occlusions, is a significant challenge. This\nwork identifies and addresses a gap in the current state of the art in 3D HPE\nconcerning the scarcity of data and strategies for handling occlusions. We\nintroduce our novel BlendMimic3D dataset, designed to mimic real-world\nsituations where occlusions occur for seamless integration in 3D HPE\nalgorithms. Additionally, we propose a 3D pose refinement block, employing a\nGraph Convolutional Network (GCN) to enhance pose representation through a\ngraph model. This GCN block acts as a plug-and-play solution, adaptable to\nvarious 3D HPE frameworks without requiring retraining them. By training the\nGCN with occluded data from BlendMimic3D, we demonstrate significant\nimprovements in resolving occluded poses, with comparable results for\nnon-occluded ones. Project web page is available at\nhttps://blendmimic3d.github.io/BlendMimic3D/.\n", "date": "2024-04-24", "categories": []}
{"id": "2404.16752", "title": "TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose\n  Representation", "abstract": "  We address the problem of regressing 3D human pose and shape from a single\nimage, with a focus on 3D accuracy. The current best methods leverage large\ndatasets of 3D pseudo-ground-truth (p-GT) and 2D keypoints, leading to robust\nperformance. With such methods, we observe a paradoxical decline in 3D pose\naccuracy with increasing 2D accuracy. This is caused by biases in the p-GT and\nthe use of an approximate camera projection model. We quantify the error\ninduced by current camera models and show that fitting 2D keypoints and p-GT\naccurately causes incorrect 3D poses. Our analysis defines the invalid\ndistances within which minimizing 2D and p-GT losses is detrimental. We use\nthis to formulate a new loss Threshold-Adaptive Loss Scaling (TALS) that\npenalizes gross 2D and p-GT losses but not smaller ones. With such a loss,\nthere are many 3D poses that could equally explain the 2D evidence. To reduce\nthis ambiguity we need a prior over valid human poses but such priors can\nintroduce unwanted bias. To address this, we exploit a tokenized representation\nof human pose and reformulate the problem as token prediction. This restricts\nthe estimated poses to the space of valid poses, effectively providing a\nuniform prior. Extensive experiments on the EMDB and 3DPW datasets show that\nour reformulated keypoint loss and tokenization allows us to train on\nin-the-wild data while improving 3D accuracy over the state-of-the-art. Our\nmodels and code are available for research at https://tokenhmr.is.tue.mpg.de.\n", "date": "2024-04-25", "categories": []}
{"id": "2311.18836", "title": "ChatPose: Chatting about 3D Human Pose", "abstract": "  We introduce ChatPose, a framework employing Large Language Models (LLMs) to\nunderstand and reason about 3D human poses from images or textual descriptions.\nOur work is motivated by the human ability to intuitively understand postures\nfrom a single image or a brief description, a process that intertwines image\ninterpretation, world knowledge, and an understanding of body language.\nTraditional human pose estimation and generation methods often operate in\nisolation, lacking semantic understanding and reasoning abilities. ChatPose\naddresses these limitations by embedding SMPL poses as distinct signal tokens\nwithin a multimodal LLM, enabling the direct generation of 3D body poses from\nboth textual and visual inputs. Leveraging the powerful capabilities of\nmultimodal LLMs, ChatPose unifies classical 3D human pose and generation tasks\nwhile offering user interactions. Additionally, ChatPose empowers LLMs to apply\ntheir extensive world knowledge in reasoning about human poses, leading to two\nadvanced tasks: speculative pose generation and reasoning about pose\nestimation. These tasks involve reasoning about humans to generate 3D poses\nfrom subtle text queries, possibly accompanied by images. We establish\nbenchmarks for these tasks, moving beyond traditional 3D pose generation and\nestimation methods. Our results show that ChatPose outperforms existing\nmultimodal LLMs and task-specific methods on these newly proposed tasks.\nFurthermore, ChatPose's ability to understand and generate 3D human poses based\non complex reasoning opens new directions in human pose analysis.\n", "date": "2023-11-30", "categories": []}
{"id": "2402.18196", "title": "NToP: NeRF-Powered Large-scale Dataset Generation for 2D and 3D Human\n  Pose Estimation in Top-View Fisheye Images", "abstract": "  Human pose estimation (HPE) in the top-view using fisheye cameras presents a\npromising and innovative application domain. However, the availability of\ndatasets capturing this viewpoint is extremely limited, especially those with\nhigh-quality 2D and 3D keypoint annotations. Addressing this gap, we leverage\nthe capabilities of Neural Radiance Fields (NeRF) technique to establish a\ncomprehensive pipeline for generating human pose datasets from existing 2D and\n3D datasets, specifically tailored for the top-view fisheye perspective.\nThrough this pipeline, we create a novel dataset NToP570K (NeRF-powered\nTop-view human Pose dataset for fisheye cameras with over 570 thousand images),\nand conduct an extensive evaluation of its efficacy in enhancing neural\nnetworks for 2D and 3D top-view human pose estimation. A pretrained ViTPose-B\nmodel achieves an improvement in AP of 33.3 % on our validation set for 2D HPE\nafter finetuning on our training set. A similarly finetuned HybrIK-Transformer\nmodel gains 53.7 mm reduction in PA-MPJPE for 3D HPE on the validation set.\n", "date": "2024-02-28", "categories": []}
{"id": "2404.14835", "title": "Semi-supervised 2D Human Pose Estimation via Adaptive Keypoint Masking", "abstract": "  Human pose estimation is a fundamental and challenging task in computer\nvision. Larger-scale and more accurate keypoint annotations, while helpful for\nimproving the accuracy of supervised pose estimation, are often expensive and\ndifficult to obtain. Semi-supervised pose estimation tries to leverage a large\namount of unlabeled data to improve model performance, which can alleviate the\nproblem of insufficient labeled samples. The latest semi-supervised learning\nusually adopts a strong and weak data augmented teacher-student learning\nframework to deal with the challenge of \"Human postural diversity and its\nlong-tailed distribution\". Appropriate data augmentation method is one of the\nkey factors affecting the accuracy and generalization of semi-supervised\nmodels. Aiming at the problem that the difference of sample learning is not\nconsidered in the fixed keypoint masking augmentation method, this paper\nproposes an adaptive keypoint masking method, which can fully mine the\ninformation in the samples and obtain better estimation performance. In order\nto further improve the generalization and robustness of the model, this paper\nproposes a dual-branch data augmentation scheme, which can perform Mixup on\nsamples and features on the basis of adaptive keypoint masking. The\neffectiveness of the proposed method is verified on COCO and MPII,\noutperforming the state-of-the-art semi-supervised pose estimation by 5.2% and\n0.3%, respectively.\n", "date": "2024-04-23", "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": "2312.07531", "title": "WHAM: Reconstructing World-grounded Humans with Accurate 3D Motion", "abstract": "  The estimation of 3D human motion from video has progressed rapidly but\ncurrent methods still have several key limitations. First, most methods\nestimate the human in camera coordinates. Second, prior work on estimating\nhumans in global coordinates often assumes a flat ground plane and produces\nfoot sliding. Third, the most accurate methods rely on computationally\nexpensive optimization pipelines, limiting their use to offline applications.\nFinally, existing video-based methods are surprisingly less accurate than\nsingle-frame methods. We address these limitations with WHAM (World-grounded\nHumans with Accurate Motion), which accurately and efficiently reconstructs 3D\nhuman motion in a global coordinate system from video. WHAM learns to lift 2D\nkeypoint sequences to 3D using motion capture data and fuses this with video\nfeatures, integrating motion context and visual information. WHAM exploits\ncamera angular velocity estimated from a SLAM method together with human motion\nto estimate the body's global trajectory. We combine this with a contact-aware\ntrajectory refinement method that lets WHAM capture human motion in diverse\nconditions, such as climbing stairs. WHAM outperforms all existing 3D human\nmotion recovery methods across multiple in-the-wild benchmarks. Code will be\navailable for research purposes at http://wham.is.tue.mpg.de/\n", "date": "2023-12-12", "categories": []}
{"id": "2404.12625", "title": "SkelFormer: Markerless 3D Pose and Shape Estimation using Skeletal\n  Transformers", "abstract": "  We introduce SkelFormer, a novel markerless motion capture pipeline for\nmulti-view human pose and shape estimation. Our method first uses off-the-shelf\n2D keypoint estimators, pre-trained on large-scale in-the-wild data, to obtain\n3D joint positions. Next, we design a regression-based inverse-kinematic\nskeletal transformer that maps the joint positions to pose and shape\nrepresentations from heavily noisy observations. This module integrates prior\nknowledge about pose space and infers the full pose state at runtime.\nSeparating the 3D keypoint detection and inverse-kinematic problems, along with\nthe expressive representations learned by our skeletal transformer, enhance the\ngeneralization of our method to unseen noisy data. We evaluate our method on\nthree public datasets in both in-distribution and out-of-distribution settings\nusing three datasets, and observe strong performance with respect to prior\nworks. Moreover, ablation experiments demonstrate the impact of each of the\nmodules of our architecture. Finally, we study the performance of our method in\ndealing with noise and heavy occlusions and find considerable robustness with\nrespect to other solutions.\n", "date": "2024-04-19", "categories": []}
{"id": "2311.09104", "title": "Cross-view and Cross-pose Completion for 3D Human Understanding", "abstract": "  Human perception and understanding is a major domain of computer vision\nwhich, like many other vision subdomains recently, stands to gain from the use\nof large models pre-trained on large datasets. We hypothesize that the most\ncommon pre-training strategy of relying on general purpose, object-centric\nimage datasets such as ImageNet, is limited by an important domain shift. On\nthe other hand, collecting domain-specific ground truth such as 2D or 3D labels\ndoes not scale well. Therefore, we propose a pre-training approach based on\nself-supervised learning that works on human-centric data using only images.\nOur method uses pairs of images of humans: the first is partially masked and\nthe model is trained to reconstruct the masked parts given the visible ones and\na second image. It relies on both stereoscopic (cross-view) pairs, and temporal\n(cross-pose) pairs taken from videos, in order to learn priors about 3D as well\nas human motion. We pre-train a model for body-centric tasks and one for\nhand-centric tasks. With a generic transformer architecture, these models\noutperform existing self-supervised pre-training methods on a wide set of\nhuman-centric downstream tasks, and obtain state-of-the-art performance for\ninstance when fine-tuning for model-based and model-free human mesh recovery.\n", "date": "2023-11-15", "categories": []}
{"id": "2403.17893", "title": "A Survey on 3D Egocentric Human Pose Estimation", "abstract": "  Egocentric human pose estimation aims to estimate human body poses and\ndevelop body representations from a first-person camera perspective. It has\ngained vast popularity in recent years because of its wide range of\napplications in sectors like XR-technologies, human-computer interaction, and\nfitness tracking. However, to the best of our knowledge, there is no systematic\nliterature review based on the proposed solutions regarding egocentric 3D human\npose estimation. To that end, the aim of this survey paper is to provide an\nextensive overview of the current state of egocentric pose estimation research.\nIn this paper, we categorize and discuss the popular datasets and the different\npose estimation models, highlighting the strengths and weaknesses of different\nmethods by comparative analysis. This survey can be a valuable resource for\nboth researchers and practitioners in the field, offering insights into key\nconcepts and cutting-edge solutions in egocentric pose estimation, its\nwide-ranging applications, as well as the open problems with future scope.\n", "date": "2024-03-26", "categories": []}
{"id": "2404.11987", "title": "MultiPhys: Multi-Person Physics-aware 3D Motion Estimation", "abstract": "  We introduce MultiPhys, a method designed for recovering multi-person motion\nfrom monocular videos. Our focus lies in capturing coherent spatial placement\nbetween pairs of individuals across varying degrees of engagement. MultiPhys,\nbeing physically aware, exhibits robustness to jittering and occlusions, and\neffectively eliminates penetration issues between the two individuals. We\ndevise a pipeline in which the motion estimated by a kinematic-based method is\nfed into a physics simulator in an autoregressive manner. We introduce distinct\ncomponents that enable our model to harness the simulator's properties without\ncompromising the accuracy of the kinematic estimates. This results in final\nmotion estimates that are both kinematically coherent and physically compliant.\nExtensive evaluations on three challenging datasets characterized by\nsubstantial inter-person interaction show that our method significantly reduces\nerrors associated with penetration and foot skating, while performing\ncompetitively with the state-of-the-art on motion accuracy and smoothness.\nResults and code can be found on our project page\n(http://www.iri.upc.edu/people/nugrinovic/multiphys/).\n", "date": "2024-04-18", "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": "2312.06797", "title": "Improving the Robustness of 3D Human Pose Estimation: A Benchmark and\n  Learning from Noisy Input", "abstract": "  Despite the promising performance of current 3D human pose estimation\ntechniques, understanding and enhancing their generalization on challenging\nin-the-wild videos remain an open problem. In this work, we focus on the\nrobustness of 2D-to-3D pose lifters. To this end, we develop two benchmark\ndatasets, namely Human3.6M-C and HumanEva-I-C, to examine the robustness of\nvideo-based 3D pose lifters to a wide range of common video corruptions\nincluding temporary occlusion, motion blur, and pixel-level noise. We observe\nthe poor generalization of state-of-the-art 3D pose lifters in the presence of\ncorruption and establish two techniques to tackle this issue. First, we\nintroduce Temporal Additive Gaussian Noise (TAGN) as a simple yet effective 2D\ninput pose data augmentation. Additionally, to incorporate the confidence\nscores output by the 2D pose detectors, we design a confidence-aware\nconvolution (CA-Conv) block. Extensively tested on corrupted videos, the\nproposed strategies consistently boost the robustness of 3D pose lifters and\nserve as new baselines for future research.\n", "date": "2023-12-11", "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": "2404.09301", "title": "A Simple Strategy for Body Estimation from Partial-View Images", "abstract": "  Virtual try-on and product personalization have become increasingly important\nin modern online shopping, highlighting the need for accurate body measurement\nestimation. Although previous research has advanced in estimating 3D body\nshapes from RGB images, the task is inherently ambiguous as the observed scale\nof human subjects in the images depends on two unknown factors: capture\ndistance and body dimensions. This ambiguity is particularly pronounced in\npartial-view scenarios. To address this challenge, we propose a modular and\nsimple height normalization solution. This solution relocates the subject\nskeleton to the desired position, thereby normalizing the scale and\ndisentangling the relationship between the two variables. Our experimental\nresults demonstrate that integrating this technique into state-of-the-art human\nmesh reconstruction models significantly enhances partial body measurement\nestimation. Additionally, we illustrate the applicability of this approach to\nmulti-view settings, showcasing its versatility.\n", "date": "2024-04-14", "categories": []}
{"id": "2404.09499", "title": "Learning Human Motion from Monocular Videos via Cross-Modal Manifold\n  Alignment", "abstract": "  Learning 3D human motion from 2D inputs is a fundamental task in the realms\nof computer vision and computer graphics. Many previous methods grapple with\nthis inherently ambiguous task by introducing motion priors into the learning\nprocess. However, these approaches face difficulties in defining the complete\nconfigurations of such priors or training a robust model. In this paper, we\npresent the Video-to-Motion Generator (VTM), which leverages motion priors\nthrough cross-modal latent feature space alignment between 3D human motion and\n2D inputs, namely videos and 2D keypoints. To reduce the complexity of modeling\nmotion priors, we model the motion data separately for the upper and lower body\nparts. Additionally, we align the motion data with a scale-invariant virtual\nskeleton to mitigate the interference of human skeleton variations to the\nmotion priors. Evaluated on AIST++, the VTM showcases state-of-the-art\nperformance in reconstructing 3D human motion from monocular videos. Notably,\nour VTM exhibits the capabilities for generalization to unseen view angles and\nin-the-wild videos.\n", "date": "2024-04-15", "categories": []}
{"id": "2304.06707", "title": "Toward Reliable Human Pose Forecasting with Uncertainty", "abstract": "  Recently, there has been an arms race of pose forecasting methods aimed at\nsolving the spatio-temporal task of predicting a sequence of future 3D poses of\na person given a sequence of past observed ones. However, the lack of unified\nbenchmarks and limited uncertainty analysis have hindered progress in the\nfield. To address this, we first develop an open-source library for human pose\nforecasting, including multiple models, supporting several datasets, and\nemploying standardized evaluation metrics, with the aim of promoting research\nand moving toward a unified and consistent evaluation. Second, we devise two\ntypes of uncertainty in the problem to increase performance and convey better\ntrust: 1) we propose a method for modeling aleatoric uncertainty by using\nuncertainty priors to inject knowledge about the pattern of uncertainty. This\nfocuses the capacity of the model in the direction of more meaningful\nsupervision while reducing the number of learned parameters and improving\nstability; 2) we introduce a novel approach for quantifying the epistemic\nuncertainty of any model through clustering and measuring the entropy of its\nassignments. Our experiments demonstrate up to $25\\%$ improvements in\nforecasting at short horizons, with no loss on longer horizons on Human3.6M,\nAMSS, and 3DPW datasets, and better performance in uncertainty estimation. The\ncode is available online at https://github.com/vita-epfl/UnPOSed.\n", "date": "2023-04-13", "categories": []}
{"id": "2404.08640", "title": "EventEgo3D: 3D Human Motion Capture from Egocentric Event Streams", "abstract": "  Monocular egocentric 3D human motion capture is a challenging and actively\nresearched problem. Existing methods use synchronously operating visual sensors\n(e.g. RGB cameras) and often fail under low lighting and fast motions, which\ncan be restricting in many applications involving head-mounted devices. In\nresponse to the existing limitations, this paper 1) introduces a new problem,\ni.e., 3D human motion capture from an egocentric monocular event camera with a\nfisheye lens, and 2) proposes the first approach to it called EventEgo3D\n(EE3D). Event streams have high temporal resolution and provide reliable cues\nfor 3D human motion capture under high-speed human motions and rapidly changing\nillumination. The proposed EE3D framework is specifically tailored for learning\nwith event streams in the LNES representation, enabling high 3D reconstruction\naccuracy. We also design a prototype of a mobile head-mounted device with an\nevent camera and record a real dataset with event observations and the\nground-truth 3D human poses (in addition to the synthetic dataset). Our EE3D\ndemonstrates robustness and superior 3D accuracy compared to existing solutions\nacross various challenging experiments while supporting real-time 3D pose\nupdate rates of 140Hz.\n", "date": "2024-04-12", "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": "2212.08731", "title": "Multi-person 3D pose estimation from unlabelled data", "abstract": "  Its numerous applications make multi-human 3D pose estimation a remarkably\nimpactful area of research. Nevertheless, assuming a multiple-view system\ncomposed of several regular RGB cameras, 3D multi-pose estimation presents\nseveral challenges. First of all, each person must be uniquely identified in\nthe different views to separate the 2D information provided by the cameras.\nSecondly, the 3D pose estimation process from the multi-view 2D information of\neach person must be robust against noise and potential occlusions in the\nscenario. In this work, we address these two challenges with the help of deep\nlearning. Specifically, we present a model based on Graph Neural Networks\ncapable of predicting the cross-view correspondence of the people in the\nscenario along with a Multilayer Perceptron that takes the 2D points to yield\nthe 3D poses of each person. These two models are trained in a self-supervised\nmanner, thus avoiding the need for large datasets with 3D annotations.\n", "date": "2022-12-16", "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": "2404.01941", "title": "LPSNet: End-to-End Human Pose and Shape Estimation with Lensless Imaging", "abstract": "  Human pose and shape (HPS) estimation with lensless imaging is not only\nbeneficial to privacy protection but also can be used in covert surveillance\nscenarios due to the small size and simple structure of this device. However,\nthis task presents significant challenges due to the inherent ambiguity of the\ncaptured measurements and lacks effective methods for directly estimating human\npose and shape from lensless data. In this paper, we propose the first\nend-to-end framework to recover 3D human poses and shapes from lensless\nmeasurements to our knowledge. We specifically design a multi-scale lensless\nfeature decoder to decode the lensless measurements through the optically\nencoded mask for efficient feature extraction. We also propose a double-head\nauxiliary supervision mechanism to improve the estimation accuracy of human\nlimb ends. Besides, we establish a lensless imaging system and verify the\neffectiveness of our method on various datasets acquired by our lensless\nimaging system.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.05675", "title": "Normalizing Flows on the Product Space of SO(3) Manifolds for\n  Probabilistic Human Pose Modeling", "abstract": "  Normalizing flows have proven their efficacy for density estimation in\nEuclidean space, but their application to rotational representations, crucial\nin various domains such as robotics or human pose modeling, remains\nunderexplored. Probabilistic models of the human pose can benefit from\napproaches that rigorously consider the rotational nature of human joints. For\nthis purpose, we introduce HuProSO3, a normalizing flow model that operates on\na high-dimensional product space of SO(3) manifolds, modeling the joint\ndistribution for human joints with three degrees of freedom. HuProSO3's\nadvantage over state-of-the-art approaches is demonstrated through its superior\nmodeling accuracy in three different applications and its capability to\nevaluate the exact likelihood. This work not only addresses the technical\nchallenge of learning densities on SO(3) manifolds, but it also has broader\nimplications for domains where the probabilistic regression of correlated 3D\nrotations is of importance.\n", "date": "2024-04-08", "categories": []}
{"id": "2404.03183", "title": "BodyMAP -- Jointly Predicting Body Mesh and 3D Applied Pressure Map for\n  People in Bed", "abstract": "  Accurately predicting the 3D human posture and the pressure exerted on the\nbody for people resting in bed, visualized as a body mesh (3D pose & shape)\nwith a 3D pressure map, holds significant promise for healthcare applications,\nparticularly, in the prevention of pressure ulcers. Current methods focus on\nsingular facets of the problem -- predicting only 2D/3D poses, generating 2D\npressure images, predicting pressure only for certain body regions instead of\nthe full body, or forming indirect approximations to the 3D pressure map. In\ncontrast, we introduce BodyMAP, which jointly predicts the human body mesh and\n3D applied pressure map across the entire human body. Our network leverages\nmultiple visual modalities, incorporating both a depth image of a person in bed\nand its corresponding 2D pressure image acquired from a pressure-sensing\nmattress. The 3D pressure map is represented as a pressure value at each mesh\nvertex and thus allows for precise localization of high-pressure regions on the\nbody. Additionally, we present BodyMAP-WS, a new formulation of pressure\nprediction in which we implicitly learn pressure in 3D by aligning sensed 2D\npressure images with a differentiable 2D projection of the predicted 3D\npressure maps. In evaluations with real-world human data, our method\noutperforms the current state-of-the-art technique by 25% on both body mesh and\n3D applied pressure map prediction tasks for people in bed.\n", "date": "2024-04-04", "categories": []}
{"id": "2309.05073", "title": "FreeMan: Towards Benchmarking 3D Human Pose Estimation under Real-World\n  Conditions", "abstract": "  Estimating the 3D structure of the human body from natural scenes is a\nfundamental aspect of visual perception. 3D human pose estimation is a vital\nstep in advancing fields like AIGC and human-robot interaction, serving as a\ncrucial technique for understanding and interacting with human actions in\nreal-world settings. However, the current datasets, often collected under\nsingle laboratory conditions using complex motion capture equipment and\nunvarying backgrounds, are insufficient. The absence of datasets on variable\nconditions is stalling the progress of this crucial task. To facilitate the\ndevelopment of 3D pose estimation, we present FreeMan, the first large-scale,\nmulti-view dataset collected under the real-world conditions. FreeMan was\ncaptured by synchronizing 8 smartphones across diverse scenarios. It comprises\n11M frames from 8000 sequences, viewed from different perspectives. These\nsequences cover 40 subjects across 10 different scenarios, each with varying\nlighting conditions. We have also established an semi-automated pipeline\ncontaining error detection to reduce the workload of manual check and ensure\nprecise annotation. We provide comprehensive evaluation baselines for a range\nof tasks, underlining the significant challenges posed by FreeMan. Further\nevaluations of standard indoor/outdoor human sensing datasets reveal that\nFreeMan offers robust representation transferability in real and complex\nscenes. Code and data are available at https://wangjiongw.github.io/freeman.\n", "date": "2023-09-10", "categories": []}
{"id": "2404.00658", "title": "KTPFormer: Kinematics and Trajectory Prior Knowledge-Enhanced\n  Transformer for 3D Human Pose Estimation", "abstract": "  This paper presents a novel Kinematics and Trajectory Prior\nKnowledge-Enhanced Transformer (KTPFormer), which overcomes the weakness in\nexisting transformer-based methods for 3D human pose estimation that the\nderivation of Q, K, V vectors in their self-attention mechanisms are all based\non simple linear mapping. We propose two prior attention modules, namely\nKinematics Prior Attention (KPA) and Trajectory Prior Attention (TPA) to take\nadvantage of the known anatomical structure of the human body and motion\ntrajectory information, to facilitate effective learning of global dependencies\nand features in the multi-head self-attention. KPA models kinematic\nrelationships in the human body by constructing a topology of kinematics, while\nTPA builds a trajectory topology to learn the information of joint motion\ntrajectory across frames. Yielding Q, K, V vectors with prior knowledge, the\ntwo modules enable KTPFormer to model both spatial and temporal correlations\nsimultaneously. Extensive experiments on three benchmarks (Human3.6M,\nMPI-INF-3DHP and HumanEva) show that KTPFormer achieves superior performance in\ncomparison to state-of-the-art methods. More importantly, our KPA and TPA\nmodules have lightweight plug-and-play designs and can be integrated into\nvarious transformer-based networks (i.e., diffusion-based) to improve the\nperformance with only a very small increase in the computational overhead. The\ncode is available at: https://github.com/JihuaPeng/KTPFormer.\n", "date": "2024-03-31", "categories": []}
{"id": "2403.19926", "title": "Video-Based Human Pose Regression via Decoupled Space-Time Aggregation", "abstract": "  By leveraging temporal dependency in video sequences, multi-frame human pose\nestimation algorithms have demonstrated remarkable results in complicated\nsituations, such as occlusion, motion blur, and video defocus. These algorithms\nare predominantly based on heatmaps, resulting in high computation and storage\nrequirements per frame, which limits their flexibility and real-time\napplication in video scenarios, particularly on edge devices. In this paper, we\ndevelop an efficient and effective video-based human pose regression method,\nwhich bypasses intermediate representations such as heatmaps and instead\ndirectly maps the input to the output joint coordinates. Despite the inherent\nspatial correlation among adjacent joints of the human pose, the temporal\ntrajectory of each individual joint exhibits relative independence. In light of\nthis, we propose a novel Decoupled Space-Time Aggregation network (DSTA) to\nseparately capture the spatial contexts between adjacent joints and the\ntemporal cues of each individual joint, thereby avoiding the conflation of\nspatiotemporal dimensions. Concretely, DSTA learns a dedicated feature token\nfor each joint to facilitate the modeling of their spatiotemporal dependencies.\nWith the proposed joint-wise local-awareness attention mechanism, our method is\ncapable of efficiently and flexibly utilizing the spatial dependency of\nadjacent joints and the temporal dependency of each joint itself. Extensive\nexperiments demonstrate the superiority of our method. Compared to previous\nregression-based single-frame human pose estimation methods, DSTA significantly\nenhances performance, achieving an 8.9 mAP improvement on PoseTrack2017.\nFurthermore, our approach either surpasses or is on par with the\nstate-of-the-art heatmap-based multi-frame human pose estimation methods.\nProject page: https://github.com/zgspose/DSTA.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.12440", "title": "Self-learning Canonical Space for Multi-view 3D Human Pose Estimation", "abstract": "  Multi-view 3D human pose estimation is naturally superior to single view one,\nbenefiting from more comprehensive information provided by images of multiple\nviews. The information includes camera poses, 2D/3D human poses, and 3D\ngeometry. However, the accurate annotation of these information is hard to\nobtain, making it challenging to predict accurate 3D human pose from multi-view\nimages. To deal with this issue, we propose a fully self-supervised framework,\nnamed cascaded multi-view aggregating network (CMANet), to construct a\ncanonical parameter space to holistically integrate and exploit multi-view\ninformation. In our framework, the multi-view information is grouped into two\ncategories: 1) intra-view information , 2) inter-view information. Accordingly,\nCMANet consists of two components: intra-view module (IRV) and inter-view\nmodule (IEV). IRV is used for extracting initial camera pose and 3D human pose\nof each view; IEV is to fuse complementary pose information and cross-view 3D\ngeometry for a final 3D human pose. To facilitate the aggregation of the intra-\nand inter-view, we define a canonical parameter space, depicted by per-view\ncamera pose and human pose and shape parameters ($\\theta$ and $\\beta$) of SMPL\nmodel, and propose a two-stage learning procedure. At first stage, IRV learns\nto estimate camera pose and view-dependent 3D human pose supervised by\nconfident output of an off-the-shelf 2D keypoint detector. At second stage, IRV\nis frozen and IEV further refines the camera pose and optimizes the 3D human\npose by implicitly encoding the cross-view complement and 3D geometry\nconstraint, achieved by jointly fitting predicted multi-view 2D keypoints. The\nproposed framework, modules, and learning strategy are demonstrated to be\neffective by comprehensive experiments and CMANet is superior to\nstate-of-the-art methods in extensive quantitative and qualitative analysis.\n", "date": "2024-03-19", "categories": []}
{"id": "2403.19501", "title": "RELI11D: A Comprehensive Multimodal Human Motion Dataset and Method", "abstract": "  Comprehensive capturing of human motions requires both accurate captures of\ncomplex poses and precise localization of the human within scenes. Most of the\nHPE datasets and methods primarily rely on RGB, LiDAR, or IMU data. However,\nsolely using these modalities or a combination of them may not be adequate for\nHPE, particularly for complex and fast movements. For holistic human motion\nunderstanding, we present RELI11D, a high-quality multimodal human motion\ndataset involves LiDAR, IMU system, RGB camera, and Event camera. It records\nthe motions of 10 actors performing 5 sports in 7 scenes, including 3.32 hours\nof synchronized LiDAR point clouds, IMU measurement data, RGB videos and Event\nsteams. Through extensive experiments, we demonstrate that the RELI11D presents\nconsiderable challenges and opportunities as it contains many rapid and complex\nmotions that require precise location. To address the challenge of integrating\ndifferent modalities, we propose LEIR, a multimodal baseline that effectively\nutilizes LiDAR Point Cloud, Event stream, and RGB through our cross-attention\nfusion strategy. We show that LEIR exhibits promising results for rapid motions\nand daily motions and that utilizing the characteristics of multiple modalities\ncan indeed improve HPE performance. Both the dataset and source code will be\nreleased publicly to the research community, fostering collaboration and\nenabling further exploration in this field.\n", "date": "2024-03-28", "categories": []}
{"id": "2311.12028", "title": "Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose\n  Estimation", "abstract": "  Transformers have been successfully applied in the field of video-based 3D\nhuman pose estimation. However, the high computational costs of these video\npose transformers (VPTs) make them impractical on resource-constrained devices.\nIn this paper, we present a plug-and-play pruning-and-recovering framework,\ncalled Hourglass Tokenizer (HoT), for efficient transformer-based 3D human pose\nestimation from videos. Our HoT begins with pruning pose tokens of redundant\nframes and ends with recovering full-length tokens, resulting in a few pose\ntokens in the intermediate transformer blocks and thus improving the model\nefficiency. To effectively achieve this, we propose a token pruning cluster\n(TPC) that dynamically selects a few representative tokens with high semantic\ndiversity while eliminating the redundancy of video frames. In addition, we\ndevelop a token recovering attention (TRA) to restore the detailed\nspatio-temporal information based on the selected tokens, thereby expanding the\nnetwork output to the original full-length temporal resolution for fast\ninference. Extensive experiments on two benchmark datasets (i.e., Human3.6M and\nMPI-INF-3DHP) demonstrate that our method can achieve both high efficiency and\nestimation accuracy compared to the original VPT models. For instance, applying\nto MotionBERT and MixSTE on Human3.6M, our HoT can save nearly 50% FLOPs\nwithout sacrificing accuracy and nearly 40% FLOPs with only 0.2% accuracy drop,\nrespectively. Code and models are available at\nhttps://github.com/NationalGAILab/HoT.\n", "date": "2023-11-20", "categories": []}
{"id": "2403.17422", "title": "InterHandGen: Two-Hand Interaction Generation via Cascaded Reverse\n  Diffusion", "abstract": "  We present InterHandGen, a novel framework that learns the generative prior\nof two-hand interaction. Sampling from our model yields plausible and diverse\ntwo-hand shapes in close interaction with or without an object. Our prior can\nbe incorporated into any optimization or learning methods to reduce ambiguity\nin an ill-posed setup. Our key observation is that directly modeling the joint\ndistribution of multiple instances imposes high learning complexity due to its\ncombinatorial nature. Thus, we propose to decompose the modeling of joint\ndistribution into the modeling of factored unconditional and conditional single\ninstance distribution. In particular, we introduce a diffusion model that\nlearns the single-hand distribution unconditional and conditional to another\nhand via conditioning dropout. For sampling, we combine anti-penetration and\nclassifier-free guidance to enable plausible generation. Furthermore, we\nestablish the rigorous evaluation protocol of two-hand synthesis, where our\nmethod significantly outperforms baseline generative models in terms of\nplausibility and diversity. We also demonstrate that our diffusion prior can\nboost the performance of two-hand reconstruction from monocular in-the-wild\nimages, achieving new state-of-the-art accuracy.\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": "2312.05541", "title": "DPoser: Diffusion Model as Robust 3D Human Pose Prior", "abstract": "  This work targets to construct a robust human pose prior. However, it remains\na persistent challenge due to biomechanical constraints and diverse human\nmovements. Traditional priors like VAEs and NDFs often exhibit shortcomings in\nrealism and generalization, notably with unseen noisy poses. To address these\nissues, we introduce DPoser, a robust and versatile human pose prior built upon\ndiffusion models. DPoser regards various pose-centric tasks as inverse problems\nand employs variational diffusion sampling for efficient solving. Accordingly,\ndesigned with optimization frameworks, DPoser seamlessly benefits human mesh\nrecovery, pose generation, pose completion, and motion denoising tasks.\nFurthermore, due to the disparity between the articulated poses and structured\nimages, we propose truncated timestep scheduling to enhance the effectiveness\nof DPoser. Our approach demonstrates considerable enhancements over common\nuniform scheduling used in image domains, boasting improvements of 5.4%, 17.2%,\nand 3.8% across human mesh recovery, pose completion, and motion denoising,\nrespectively. Comprehensive experiments demonstrate the superiority of DPoser\nover existing state-of-the-art pose priors across multiple tasks.\n", "date": "2023-12-09", "categories": []}
{"id": "2401.16700", "title": "Towards Precise 3D Human Pose Estimation with Multi-Perspective\n  Spatial-Temporal Relational Transformers", "abstract": "  3D human pose estimation captures the human joint points in three-dimensional\nspace while keeping the depth information and physical structure. That is\nessential for applications that require precise pose information, such as\nhuman-computer interaction, scene understanding, and rehabilitation training.\nDue to the challenges in data collection, mainstream datasets of 3D human pose\nestimation are primarily composed of multi-view video data collected in\nlaboratory environments, which contains rich spatial-temporal correlation\ninformation besides the image frame content. Given the remarkable\nself-attention mechanism of transformers, capable of capturing the\nspatial-temporal correlation from multi-view video datasets, we propose a\nmulti-stage framework for 3D sequence-to-sequence (seq2seq) human pose\ndetection. Firstly, the spatial module represents the human pose feature by\nintra-image content, while the frame-image relation module extracts temporal\nrelationships and 3D spatial positional relationship features between the\nmulti-perspective images. Secondly, the self-attention mechanism is adopted to\neliminate the interference from non-human body parts and reduce computing\nresources. Our method is evaluated on Human3.6M, a popular 3D human pose\ndetection dataset. Experimental results demonstrate that our approach achieves\nstate-of-the-art performance on this dataset. The source code will be available\nat https://github.com/WUJINHUAN/3D-human-pose.\n", "date": "2024-01-30", "categories": []}
{"id": "2403.14447", "title": "Exploring 3D Human Pose Estimation and Forecasting from the Robot's\n  Perspective: The HARPER Dataset", "abstract": "  We introduce HARPER, a novel dataset for 3D body pose estimation and forecast\nin dyadic interactions between users and Spot, the quadruped robot manufactured\nby Boston Dynamics. The key-novelty is the focus on the robot's perspective,\ni.e., on the data captured by the robot's sensors. These make 3D body pose\nanalysis challenging because being close to the ground captures humans only\npartially. The scenario underlying HARPER includes 15 actions, of which 10\ninvolve physical contact between the robot and users. The Corpus contains not\nonly the recordings of the built-in stereo cameras of Spot, but also those of a\n6-camera OptiTrack system (all recordings are synchronized). This leads to\nground-truth skeletal representations with a precision lower than a millimeter.\nIn addition, the Corpus includes reproducible benchmarks on 3D Human Pose\nEstimation, Human Pose Forecasting, and Collision Prediction, all based on\npublicly available baseline approaches. This enables future HARPER users to\nrigorously compare their results with those we provide in this work.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.11310", "title": "A Dual-Augmentor Framework for Domain Generalization in 3D Human Pose\n  Estimation", "abstract": "  3D human pose data collected in controlled laboratory settings present\nchallenges for pose estimators that generalize across diverse scenarios. To\naddress this, domain generalization is employed. Current methodologies in\ndomain generalization for 3D human pose estimation typically utilize\nadversarial training to generate synthetic poses for training. Nonetheless,\nthese approaches exhibit several limitations. First, the lack of prior\ninformation about the target domain complicates the application of suitable\naugmentation through a single pose augmentor, affecting generalization on\ntarget domains. Moreover, adversarial training's discriminator tends to enforce\nsimilarity between source and synthesized poses, impeding the exploration of\nout-of-source distributions. Furthermore, the pose estimator's optimization is\nnot exposed to domain shifts, limiting its overall generalization ability.\n  To address these limitations, we propose a novel framework featuring two pose\naugmentors: the weak and the strong augmentors. Our framework employs\ndifferential strategies for generation and discrimination processes,\nfacilitating the preservation of knowledge related to source poses and the\nexploration of out-of-source distributions without prior information about\ntarget poses. Besides, we leverage meta-optimization to simulate domain shifts\nin the optimization process of the pose estimator, thereby improving its\ngeneralization ability. Our proposed approach significantly outperforms\nexisting methods, as demonstrated through comprehensive experiments on various\nbenchmark datasets.Our code will be released at\n\\url{https://github.com/davidpengucf/DAF-DG}.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.12959", "title": "WHAC: World-grounded Humans and Cameras", "abstract": "  Estimating human and camera trajectories with accurate scale in the world\ncoordinate system from a monocular video is a highly desirable yet challenging\nand ill-posed problem. In this study, we aim to recover expressive parametric\nhuman models (i.e., SMPL-X) and corresponding camera poses jointly, by\nleveraging the synergy between three critical players: the world, the human,\nand the camera. Our approach is founded on two key observations. Firstly,\ncamera-frame SMPL-X estimation methods readily recover absolute human depth.\nSecondly, human motions inherently provide absolute spatial cues. By\nintegrating these insights, we introduce a novel framework, referred to as\nWHAC, to facilitate world-grounded expressive human pose and shape estimation\n(EHPS) alongside camera pose estimation, without relying on traditional\noptimization techniques. Additionally, we present a new synthetic dataset,\nWHAC-A-Mole, which includes accurately annotated humans and cameras, and\nfeatures diverse interactive human motions as well as realistic camera\ntrajectories. Extensive experiments on both standard and newly established\nbenchmarks highlight the superiority and efficacy of our framework. We will\nmake the code and dataset publicly available.\n", "date": "2024-03-19", "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": "2403.09050", "title": "CLOAF: CoLlisiOn-Aware Human Flow", "abstract": "  Even the best current algorithms for estimating body 3D shape and pose yield\nresults that include body self-intersections. In this paper, we present CLOAF,\nwhich exploits the diffeomorphic nature of Ordinary Differential Equations to\neliminate such self-intersections while still imposing body shape constraints.\nWe show that, unlike earlier approaches to addressing this issue, ours\ncompletely eliminates the self-intersections without compromising the accuracy\nof the reconstructions. Being differentiable, CLOAF can be used to fine-tune\npose and shape estimation baselines to improve their overall performance and\neliminate self-intersections in their predictions. Furthermore, we demonstrate\nhow our CLOAF strategy can be applied to practically any motion field induced\nby the user. CLOAF also makes it possible to edit motion to interact with the\nenvironment without worrying about potential collision or loss of body-shape\nprior.\n", "date": "2024-03-14", "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": "2403.09437", "title": "Improving Real-Time Omnidirectional 3D Multi-Person Human Pose\n  Estimation with People Matching and Unsupervised 2D-3D Lifting", "abstract": "  Current human pose estimation systems focus on retrieving an accurate 3D\nglobal estimate of a single person. Therefore, this paper presents one of the\nfirst 3D multi-person human pose estimation systems that is able to work in\nreal-time and is also able to handle basic forms of occlusion. First, we adjust\nan off-the-shelf 2D detector and an unsupervised 2D-3D lifting model for use\nwith a 360$^\\circ$ panoramic camera and mmWave radar sensors. We then introduce\nseveral contributions, including camera and radar calibrations, and the\nimproved matching of people within the image and radar space. The system\naddresses both the depth and scale ambiguity problems by employing a\nlightweight 2D-3D pose lifting algorithm that is able to work in real-time\nwhile exhibiting accurate performance in both indoor and outdoor environments\nwhich offers both an affordable and scalable solution. Notably, our system's\ntime complexity remains nearly constant irrespective of the number of detected\nindividuals, achieving a frame rate of approximately 7-8 fps on a laptop with a\ncommercial-grade GPU.\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": "2309.14865", "title": "Unsupervised Multi-Person 3D Human Pose Estimation From 2D Poses Alone", "abstract": "  Current unsupervised 2D-3D human pose estimation (HPE) methods do not work in\nmulti-person scenarios due to perspective ambiguity in monocular images.\nTherefore, we present one of the first studies investigating the feasibility of\nunsupervised multi-person 2D-3D HPE from just 2D poses alone, focusing on\nreconstructing human interactions. To address the issue of perspective\nambiguity, we expand upon prior work by predicting the cameras' elevation angle\nrelative to the subjects' pelvis. This allows us to rotate the predicted poses\nto be level with the ground plane, while obtaining an estimate for the vertical\noffset in 3D between individuals. Our method involves independently lifting\neach subject's 2D pose to 3D, before combining them in a shared 3D coordinate\nsystem. The poses are then rotated and offset by the predicted elevation angle\nbefore being scaled. This by itself enables us to retrieve an accurate 3D\nreconstruction of their poses. We present our results on the CHI3D dataset,\nintroducing its use for unsupervised 2D-3D pose estimation with three new\nquantitative metrics, and establishing a benchmark for future research.\n", "date": "2023-09-26", "categories": []}
{"id": "2312.02196", "title": "Dynamic Inertial Poser (DynaIP): Part-Based Motion Dynamics Learning for\n  Enhanced Human Pose Estimation with Sparse Inertial Sensors", "abstract": "  This paper introduces a novel human pose estimation approach using sparse\ninertial sensors, addressing the shortcomings of previous methods reliant on\nsynthetic data. It leverages a diverse array of real inertial motion capture\ndata from different skeleton formats to improve motion diversity and model\ngeneralization. This method features two innovative components: a\npseudo-velocity regression model for dynamic motion capture with inertial\nsensors, and a part-based model dividing the body and sensor data into three\nregions, each focusing on their unique characteristics. The approach\ndemonstrates superior performance over state-of-the-art models across five\npublic datasets, notably reducing pose error by 19\\% on the DIP-IMU dataset,\nthus representing a significant improvement in inertial sensor-based human pose\nestimation. Our codes are available at {\\url{https://github.com/dx118/dynaip}}.\n", "date": "2023-12-02", "categories": []}
{"id": "2402.13172", "title": "3D Kinematics Estimation from Video with a Biomechanical Model and\n  Synthetic Training Data", "abstract": "  Accurate 3D kinematics estimation of human body is crucial in various\napplications for human health and mobility, such as rehabilitation, injury\nprevention, and diagnosis, as it helps to understand the biomechanical loading\nexperienced during movement. Conventional marker-based motion capture is\nexpensive in terms of financial investment, time, and the expertise required.\nMoreover, due to the scarcity of datasets with accurate annotations, existing\nmarkerless motion capture methods suffer from challenges including unreliable\n2D keypoint detection, limited anatomic accuracy, and low generalization\ncapability. In this work, we propose a novel biomechanics-aware network that\ndirectly outputs 3D kinematics from two input views with consideration of\nbiomechanical prior and spatio-temporal information. To train the model, we\ncreate synthetic dataset ODAH with accurate kinematics annotations generated by\naligning the body mesh from the SMPL-X model and a full-body OpenSim skeletal\nmodel. Our extensive experiments demonstrate that the proposed approach, only\ntrained on synthetic data, outperforms previous state-of-the-art methods when\nevaluated across multiple datasets, revealing a promising direction for\nenhancing video-based human motion capture\n", "date": "2024-02-20", "categories": []}
{"id": "2306.12525", "title": "LPFormer: LiDAR Pose Estimation Transformer with Multi-Task Network", "abstract": "  Due to the difficulty of acquiring large-scale 3D human keypoint annotation,\nprevious methods for 3D human pose estimation (HPE) have often relied on 2D\nimage features and sequential 2D annotations. Furthermore, the training of\nthese networks typically assumes the prediction of a human bounding box and the\naccurate alignment of 3D point clouds with 2D images, making direct application\nin real-world scenarios challenging. In this paper, we present the 1st\nframework for end-to-end 3D human pose estimation, named LPFormer, which uses\nonly LiDAR as its input along with its corresponding 3D annotations. LPFormer\nconsists of two stages: firstly, it identifies the human bounding box and\nextracts multi-level feature representations, and secondly, it utilizes a\ntransformer-based network to predict human keypoints based on these features.\nOur method demonstrates that 3D HPE can be seamlessly integrated into a strong\nLiDAR perception network and benefit from the features extracted by the\nnetwork. Experimental results on the Waymo Open Dataset demonstrate the\nstate-of-the-art performance, and improvements even compared to previous\nmulti-modal solutions.\n", "date": "2023-06-21", "categories": []}
{"id": "2402.18330", "title": "Attention-Propagation Network for Egocentric Heatmap to 3D Pose Lifting", "abstract": "  We present EgoTAP, a heatmap-to-3D pose lifting method for highly accurate\nstereo egocentric 3D pose estimation. Severe self-occlusion and out-of-view\nlimbs in egocentric camera views make accurate pose estimation a challenging\nproblem. To address the challenge, prior methods employ joint\nheatmaps-probabilistic 2D representations of the body pose, but heatmap-to-3D\npose conversion still remains an inaccurate process. We propose a novel\nheatmap-to-3D lifting method composed of the Grid ViT Encoder and the\nPropagation Network. The Grid ViT Encoder summarizes joint heatmaps into\neffective feature embedding using self-attention. Then, the Propagation Network\nestimates the 3D pose by utilizing skeletal information to better estimate the\nposition of obscure joints. Our method significantly outperforms the previous\nstate-of-the-art qualitatively and quantitatively demonstrated by a 23.9\\%\nreduction of error in an MPJPE metric. Our source code is available in GitHub.\n", "date": "2024-02-28", "categories": []}
{"id": "2402.17171", "title": "LiveHPS: LiDAR-based Scene-level Human Pose and Shape Estimation in Free\n  Environment", "abstract": "  For human-centric large-scale scenes, fine-grained modeling for 3D human\nglobal pose and shape is significant for scene understanding and can benefit\nmany real-world applications. In this paper, we present LiveHPS, a novel\nsingle-LiDAR-based approach for scene-level human pose and shape estimation\nwithout any limitation of light conditions and wearable devices. In particular,\nwe design a distillation mechanism to mitigate the distribution-varying effect\nof LiDAR point clouds and exploit the temporal-spatial geometric and dynamic\ninformation existing in consecutive frames to solve the occlusion and noise\ndisturbance. LiveHPS, with its efficient configuration and high-quality output,\nis well-suited for real-world applications. Moreover, we propose a huge human\nmotion dataset, named FreeMotion, which is collected in various scenarios with\ndiverse human poses, shapes and translations. It consists of multi-modal and\nmulti-view acquisition data from calibrated and synchronized LiDARs, cameras,\nand IMUs. Extensive experiments on our new dataset and other public datasets\ndemonstrate the SOTA performance and robustness of our approach. We will\nrelease our code and dataset soon.\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": "2402.14456", "title": "VLPose: Bridging the Domain Gap in Pose Estimation with Language-Vision\n  Tuning", "abstract": "  Thanks to advances in deep learning techniques, Human Pose Estimation (HPE)\nhas achieved significant progress in natural scenarios. However, these models\nperform poorly in artificial scenarios such as painting and sculpture due to\nthe domain gap, constraining the development of virtual reality and augmented\nreality. With the growth of model size, retraining the whole model on both\nnatural and artificial data is computationally expensive and inefficient. Our\nresearch aims to bridge the domain gap between natural and artificial scenarios\nwith efficient tuning strategies. Leveraging the potential of language models,\nwe enhance the adaptability of traditional pose estimation models across\ndiverse scenarios with a novel framework called VLPose. VLPose leverages the\nsynergy between language and vision to extend the generalization and robustness\nof pose estimation models beyond the traditional domains. Our approach has\ndemonstrated improvements of 2.26% and 3.74% on HumanArt and MSCOCO,\nrespectively, compared to state-of-the-art tuning strategies.\n", "date": "2024-02-22", "categories": []}
{"id": "2402.08923", "title": "IMUOptimize: A Data-Driven Approach to Optimal IMU Placement for Human\n  Pose Estimation with Transformer Architecture", "abstract": "  This paper presents a novel approach for predicting human poses using IMU\ndata, diverging from previous studies such as DIP-IMU, IMUPoser, and TransPose,\nwhich use up to 6 IMUs in conjunction with bidirectional RNNs. We introduce two\nmain innovations: a data-driven strategy for optimal IMU placement and a\ntransformer-based model architecture for time series analysis. Our findings\nindicate that our approach not only outperforms traditional 6 IMU-based biRNN\nmodels but also that the transformer architecture significantly enhances pose\nreconstruction from data obtained from 24 IMU locations, with equivalent\nperformance to biRNNs when using only 6 IMUs. The enhanced accuracy provided by\nour optimally chosen locations, when coupled with the parallelizability and\nperformance of transformers, provides significant improvements to the field of\nIMU-based pose estimation.\n", "date": "2024-02-14", "categories": []}
{"id": "2402.11036", "title": "Occlusion Resilient 3D Human Pose Estimation", "abstract": "  Occlusions remain one of the key challenges in 3D body pose estimation from\nsingle-camera video sequences. Temporal consistency has been extensively used\nto mitigate their impact but the existing algorithms in the literature do not\nexplicitly model them.\n  Here, we apply this by representing the deforming body as a spatio-temporal\ngraph. We then introduce a refinement network that performs graph convolutions\nover this graph to output 3D poses. To ensure robustness to occlusions, we\ntrain this network with a set of binary masks that we use to disable some of\nthe edges as in drop-out techniques.\n  In effect, we simulate the fact that some joints can be hidden for periods of\ntime and train the network to be immune to that. We demonstrate the\neffectiveness of this approach compared to state-of-the-art techniques that\ninfer poses from single-camera sequences.\n", "date": "2024-02-16", "categories": []}
{"id": "2402.08006", "title": "Extending 3D body pose estimation for robotic-assistive therapies of\n  autistic children", "abstract": "  Robotic-assistive therapy has demonstrated very encouraging results for\nchildren with Autism. Accurate estimation of the child's pose is essential both\nfor human-robot interaction and for therapy assessment purposes. Non-intrusive\nmethods are the sole viable option since these children are sensitive to touch.\n  While depth cameras have been used extensively, existing methods face two\nmajor limitations: (i) they are usually trained with adult-only data and do not\ncorrectly estimate a child's pose, and (ii) they fail in scenarios with a high\nnumber of occlusions. Therefore, our goal was to develop a 3D pose estimator\nfor children, by adapting an existing state-of-the-art 3D body modelling method\nand incorporating a linear regression model to fine-tune one of its inputs,\nthereby correcting the pose of children's 3D meshes.\n  In controlled settings, our method has an error below $0.3m$, which is\nconsidered acceptable for this kind of application and lower than current\nstate-of-the-art methods. In real-world settings, the proposed model performs\nsimilarly to a Kinect depth camera and manages to successfully estimate the 3D\nbody poses in a much higher number of frames.\n", "date": "2024-02-12", "categories": []}
{"id": "2308.02397", "title": "Design Space Exploration on Efficient and Accurate Human Pose Estimation\n  from Sparse IMU-Sensing", "abstract": "  Human Pose Estimation (HPE) to assess human motion in sports, rehabilitation\nor work safety requires accurate sensing without compromising the sensitive\nunderlying personal data. Therefore, local processing is necessary and the\nlimited energy budget in such systems can be addressed by Inertial Measurement\nUnits (IMU) instead of common camera sensing. The central trade-off between\naccuracy and efficient use of hardware resources is rarely discussed in\nresearch. We address this trade-off by a simulative Design Space Exploration\n(DSE) of a varying quantity and positioning of IMU-sensors. First, we generate\nIMU-data from a publicly available body model dataset for different sensor\nconfigurations and train a deep learning model with this data. Additionally, we\npropose a combined metric to assess the accuracy-resource trade-off. We used\nthe DSE as a tool to evaluate sensor configurations and identify beneficial\nones for a specific use case. Exemplary, for a system with equal importance of\naccuracy and resources, we identify an optimal sensor configuration of 4\nsensors with a mesh error of 6.03 cm, increasing the accuracy by 32.7% and\nreducing the hardware effort by two sensors compared to state of the art. Our\nwork can be used to design health applications with well-suited sensor\npositioning and attention to data privacy and resource-awareness.\n", "date": "2023-07-21", "categories": []}
{"id": "2211.12829", "title": "Unsupervised 3D Keypoint Discovery with Multi-View Geometry", "abstract": "  Analyzing and training 3D body posture models depend heavily on the\navailability of joint labels that are commonly acquired through laborious\nmanual annotation of body joints or via marker-based joint localization using\ncarefully curated markers and capturing systems. However, such annotations are\nnot always available, especially for people performing unusual activities. In\nthis paper, we propose an algorithm that learns to discover 3D keypoints on\nhuman bodies from multiple-view images without any supervision or labels other\nthan the constraints multiple-view geometry provides. To ensure that the\ndiscovered 3D keypoints are meaningful, they are re-projected to each view to\nestimate the person's mask that the model itself has initially estimated\nwithout supervision. Our approach discovers more interpretable and accurate 3D\nkeypoints compared to other state-of-the-art unsupervised approaches on\nHuman3.6M and MPI-INF-3DHP benchmark datasets.\n", "date": "2022-11-23", "categories": []}
{"id": "2402.04953", "title": "4-Dimensional deformation part model for pose estimation using Kalman\n  filter constraints", "abstract": "  The main goal of this article is to analyze the effect on pose estimation\naccuracy when using a Kalman filter added to 4-dimensional deformation part\nmodel partial solutions. The experiments run with two data sets showing that\nthis method improves pose estimation accuracy compared with state-of-the-art\nmethods and that a Kalman filter helps to increase this accuracy.\n", "date": "2024-02-07", "categories": []}
{"id": "2309.01365", "title": "Refined Temporal Pyramidal Compression-and-Amplification Transformer for\n  3D Human Pose Estimation", "abstract": "  Accurately estimating the 3D pose of humans in video sequences requires both\naccuracy and a well-structured architecture. With the success of transformers,\nwe introduce the Refined Temporal Pyramidal Compression-and-Amplification\n(RTPCA) transformer. Exploiting the temporal dimension, RTPCA extends\nintra-block temporal modeling via its Temporal Pyramidal\nCompression-and-Amplification (TPCA) structure and refines inter-block feature\ninteraction with a Cross-Layer Refinement (XLR) module. In particular, TPCA\nblock exploits a temporal pyramid paradigm, reinforcing key and value\nrepresentation capabilities and seamlessly extracting spatial semantics from\nmotion sequences. We stitch these TPCA blocks with XLR that promotes rich\nsemantic representation through continuous interaction of queries, keys, and\nvalues. This strategy embodies early-stage information with current flows,\naddressing typical deficits in detail and stability seen in other\ntransformer-based methods. We demonstrate the effectiveness of RTPCA by\nachieving state-of-the-art results on Human3.6M, HumanEva-I, and MPI-INF-3DHP\nbenchmarks with minimal computational overhead. The source code is available at\nhttps://github.com/hbing-l/RTPCA.\n", "date": "2023-09-04", "categories": []}
{"id": "2402.02736", "title": "Using Motion Cues to Supervise Single-Frame Body Pose and Shape\n  Estimation in Low Data Regimes", "abstract": "  When enough annotated training data is available, supervised deep-learning\nalgorithms excel at estimating human body pose and shape using a single camera.\nThe effects of too little such data being available can be mitigated by using\nother information sources, such as databases of body shapes, to learn priors.\nUnfortunately, such sources are not always available either. We show that, in\nsuch cases, easy-to-obtain unannotated videos can be used instead to provide\nthe required supervisory signals. Given a trained model using too little\nannotated data, we compute poses in consecutive frames along with the optical\nflow between them. We then enforce consistency between the image optical flow\nand the one that can be inferred from the change in pose from one frame to the\nnext. This provides enough additional supervision to effectively refine the\nnetwork weights and to perform on par with methods trained using far more\nannotated data.\n", "date": "2024-02-05", "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": "2401.15616", "title": "Multi-Person 3D Pose Estimation from Multi-View Uncalibrated Depth\n  Cameras", "abstract": "  We tackle the task of multi-view, multi-person 3D human pose estimation from\na limited number of uncalibrated depth cameras. Recently, many approaches have\nbeen proposed for 3D human pose estimation from multi-view RGB cameras.\nHowever, these works (1) assume the number of RGB camera views is large enough\nfor 3D reconstruction, (2) the cameras are calibrated, and (3) rely on ground\ntruth 3D poses for training their regression model. In this work, we propose to\nleverage sparse, uncalibrated depth cameras providing RGBD video streams for 3D\nhuman pose estimation. We present a simple pipeline for Multi-View Depth Human\nPose Estimation (MVD-HPE) for jointly predicting the camera poses and 3D human\nposes without training a deep 3D human pose regression model. This framework\nutilizes 3D Re-ID appearance features from RGBD images to formulate more\naccurate correspondences (for deriving camera positions) compared to using\nRGB-only features. We further propose (1) depth-guided camera-pose estimation\nby leveraging 3D rigid transformations as guidance and (2) depth-constrained 3D\nhuman pose estimation by utilizing depth-projected 3D points as an alternative\nobjective for optimization. In order to evaluate our proposed pipeline, we\ncollect three video sets of RGBD videos recorded from multiple sparse-view\ndepth cameras and ground truth 3D poses are manually annotated. Experiments\nshow that our proposed method outperforms the current 3D human pose\nregression-free pipelines in terms of both camera pose estimation and 3D human\npose estimation.\n", "date": "2024-01-28", "categories": []}
{"id": "2401.16173", "title": "Reconstructing Close Human Interactions from Multiple Views", "abstract": "  This paper addresses the challenging task of reconstructing the poses of\nmultiple individuals engaged in close interactions, captured by multiple\ncalibrated cameras. The difficulty arises from the noisy or false 2D keypoint\ndetections due to inter-person occlusion, the heavy ambiguity in associating\nkeypoints to individuals due to the close interactions, and the scarcity of\ntraining data as collecting and annotating motion data in crowded scenes is\nresource-intensive. We introduce a novel system to address these challenges.\nOur system integrates a learning-based pose estimation component and its\ncorresponding training and inference strategies. The pose estimation component\ntakes multi-view 2D keypoint heatmaps as input and reconstructs the pose of\neach individual using a 3D conditional volumetric network. As the network\ndoesn't need images as input, we can leverage known camera parameters from test\nscenes and a large quantity of existing motion capture data to synthesize\nmassive training data that mimics the real data distribution in test scenes.\nExtensive experiments demonstrate that our approach significantly surpasses\nprevious approaches in terms of pose accuracy and is generalizable across\nvarious camera setups and population sizes. The code is available on our\nproject page: https://github.com/zju3dv/CloseMoCap.\n", "date": "2024-01-29", "categories": []}
{"id": "2401.14785", "title": "SimpleEgo: Predicting Probabilistic Body Pose from Egocentric Cameras", "abstract": "  Our work addresses the problem of egocentric human pose estimation from\ndownwards-facing cameras on head-mounted devices (HMD). This presents a\nchallenging scenario, as parts of the body often fall outside of the image or\nare occluded. Previous solutions minimize this problem by using fish-eye camera\nlenses to capture a wider view, but these can present hardware design issues.\nThey also predict 2D heat-maps per joint and lift them to 3D space to deal with\nself-occlusions, but this requires large network architectures which are\nimpractical to deploy on resource-constrained HMDs. We predict pose from images\ncaptured with conventional rectilinear camera lenses. This resolves hardware\ndesign issues, but means body parts are often out of frame. As such, we\ndirectly regress probabilistic joint rotations represented as matrix Fisher\ndistributions for a parameterized body model. This allows us to quantify pose\nuncertainties and explain out-of-frame or occluded joints. This also removes\nthe need to compute 2D heat-maps and allows for simplified DNN architectures\nwhich require less compute. Given the lack of egocentric datasets using\nrectilinear camera lenses, we introduce the SynthEgo dataset, a synthetic\ndataset with 60K stereo images containing high diversity of pose, shape,\nclothing and skin tone. Our approach achieves state-of-the-art results for this\nchallenging configuration, reducing mean per-joint position error by 23%\noverall and 58% for the lower body. Our architecture also has eight times fewer\nparameters and runs twice as fast as the current state-of-the-art. Experiments\nshow that training on our synthetic dataset leads to good generalization to\nreal world images without fine-tuning.\n", "date": "2024-01-26", "categories": []}
{"id": "2401.09836", "title": "Exploring Latent Cross-Channel Embedding for Accurate 3D Human Pose\n  Reconstruction in a Diffusion Framework", "abstract": "  Monocular 3D human pose estimation poses significant challenges due to the\ninherent depth ambiguities that arise during the reprojection process from 2D\nto 3D. Conventional approaches that rely on estimating an over-fit projection\nmatrix struggle to effectively address these challenges and often result in\nnoisy outputs. Recent advancements in diffusion models have shown promise in\nincorporating structural priors to address reprojection ambiguities. However,\nthere is still ample room for improvement as these methods often overlook the\nexploration of correlation between the 2D and 3D joint-level features. In this\nstudy, we propose a novel cross-channel embedding framework that aims to fully\nexplore the correlation between joint-level features of 3D coordinates and\ntheir 2D projections. In addition, we introduce a context guidance mechanism to\nfacilitate the propagation of joint graph attention across latent channels\nduring the iterative diffusion process. To evaluate the effectiveness of our\nproposed method, we conduct experiments on two benchmark datasets, namely\nHuman3.6M and MPI-INF-3DHP. Our results demonstrate a significant improvement\nin terms of reconstruction accuracy compared to state-of-the-art methods. The\ncode for our method will be made available online for further reference.\n", "date": "2024-01-18", "categories": []}
{"id": "2309.08480", "title": "PoseFix: Correcting 3D Human Poses with Natural Language", "abstract": "  Automatically producing instructions to modify one's posture could open the\ndoor to endless applications, such as personalized coaching and in-home\nphysical therapy. Tackling the reverse problem (i.e., refining a 3D pose based\non some natural language feedback) could help for assisted 3D character\nanimation or robot teaching, for instance. Although a few recent works explore\nthe connections between natural language and 3D human pose, none focus on\ndescribing 3D body pose differences. In this paper, we tackle the problem of\ncorrecting 3D human poses with natural language. To this end, we introduce the\nPoseFix dataset, which consists of several thousand paired 3D poses and their\ncorresponding text feedback, that describe how the source pose needs to be\nmodified to obtain the target pose. We demonstrate the potential of this\ndataset on two tasks: (1) text-based pose editing, that aims at generating\ncorrected 3D body poses given a query pose and a text modifier; and (2)\ncorrectional text generation, where instructions are generated based on the\ndifferences between two body poses.\n", "date": "2023-09-15", "categories": []}
{"id": "2401.04921", "title": "Diffusion-based Pose Refinement and Muti-hypothesis Generation for 3D\n  Human Pose Estimaiton", "abstract": "  Previous probabilistic models for 3D Human Pose Estimation (3DHPE) aimed to\nenhance pose accuracy by generating multiple hypotheses. However, most of the\nhypotheses generated deviate substantially from the true pose. Compared to\ndeterministic models, the excessive uncertainty in probabilistic models leads\nto weaker performance in single-hypothesis prediction. To address these two\nchallenges, we propose a diffusion-based refinement framework called DRPose,\nwhich refines the output of deterministic models by reverse diffusion and\nachieves more suitable multi-hypothesis prediction for the current pose\nbenchmark by multi-step refinement with multiple noises. To this end, we\npropose a Scalable Graph Convolution Transformer (SGCT) and a Pose Refinement\nModule (PRM) for denoising and refining. Extensive experiments on Human3.6M and\nMPI-INF-3DHP datasets demonstrate that our method achieves state-of-the-art\nperformance on both single and multi-hypothesis 3DHPE. Code is available at\nhttps://github.com/KHB1698/DRPose.\n", "date": "2024-01-10", "categories": []}
{"id": "2401.03914", "title": "D3PRefiner: A Diffusion-based Denoise Method for 3D Human Pose\n  Refinement", "abstract": "  Three-dimensional (3D) human pose estimation using a monocular camera has\ngained increasing attention due to its ease of implementation and the abundance\nof data available from daily life. However, owing to the inherent depth\nambiguity in images, the accuracy of existing monocular camera-based 3D pose\nestimation methods remains unsatisfactory, and the estimated 3D poses usually\ninclude much noise. By observing the histogram of this noise, we find each\ndimension of the noise follows a certain distribution, which indicates the\npossibility for a neural network to learn the mapping between noisy poses and\nground truth poses. In this work, in order to obtain more accurate 3D poses, a\nDiffusion-based 3D Pose Refiner (D3PRefiner) is proposed to refine the output\nof any existing 3D pose estimator. We first introduce a conditional\nmultivariate Gaussian distribution to model the distribution of noisy 3D poses,\nusing paired 2D poses and noisy 3D poses as conditions to achieve greater\naccuracy. Additionally, we leverage the architecture of current diffusion\nmodels to convert the distribution of noisy 3D poses into ground truth 3D\nposes. To evaluate the effectiveness of the proposed method, two\nstate-of-the-art sequence-to-sequence 3D pose estimators are used as basic 3D\npose estimation models, and the proposed method is evaluated on different types\nof 2D poses and different lengths of the input sequence. Experimental results\ndemonstrate the proposed architecture can significantly improve the performance\nof current sequence-to-sequence 3D pose estimators, with a reduction of at\nleast 10.3% in the mean per joint position error (MPJPE) and at least 11.0% in\nthe Procrustes MPJPE (P-MPJPE).\n", "date": "2024-01-08", "categories": []}
{"id": "2212.00731", "title": "FuRPE: Learning Full-body Reconstruction from Part Experts", "abstract": "  In the field of full-body reconstruction, the scarcity of annotated data\noften impedes the efficacy of prevailing methods. To address this issue, we\nintroduce FuRPE, a novel framework that employs part-experts and an ingenious\npseudo ground-truth selection scheme to derive high-quality pseudo labels.\nThese labels, central to our approach, equip our network with the capability to\nefficiently learn from the available data. Integral to FuRPE is a unique\nexponential moving average training strategy and expert-derived feature\ndistillation strategy. These novel elements of FuRPE not only serve to further\nrefine the model but also to reduce potential biases that may arise from\ninaccuracies in pseudo labels, thereby optimizing the network's training\nprocess and enhancing the robustness of the model. We apply FuRPE to train both\ntwo-stage and fully convolutional single-stage full-body reconstruction\nnetworks. Our exhaustive experiments on numerous benchmark datasets illustrate\na substantial performance boost over existing methods, underscoring FuRPE's\npotential to reshape the state-of-the-art in full-body reconstruction.\n", "date": "2022-11-30", "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": "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": "2312.16737", "title": "HMP: Hand Motion Priors for Pose and Shape Estimation from Video", "abstract": "  Understanding how humans interact with the world necessitates accurate 3D\nhand pose estimation, a task complicated by the hand's high degree of\narticulation, frequent occlusions, self-occlusions, and rapid motions. While\nmost existing methods rely on single-image inputs, videos have useful cues to\naddress aforementioned issues. However, existing video-based 3D hand datasets\nare insufficient for training feedforward models to generalize to in-the-wild\nscenarios. On the other hand, we have access to large human motion capture\ndatasets which also include hand motions, e.g. AMASS. Therefore, we develop a\ngenerative motion prior specific for hands, trained on the AMASS dataset which\nfeatures diverse and high-quality hand motions. This motion prior is then\nemployed for video-based 3D hand motion estimation following a latent\noptimization approach. Our integration of a robust motion prior significantly\nenhances performance, especially in occluded scenarios. It produces stable,\ntemporally consistent results that surpass conventional single-frame methods.\nWe demonstrate our method's efficacy via qualitative and quantitative\nevaluations on the HO3D and DexYCB datasets, with special emphasis on an\nocclusion-focused subset of HO3D. Code is available at\nhttps://hmp.is.tue.mpg.de\n", "date": "2023-12-27", "categories": []}
{"id": "2312.17106", "title": "Geometry-Biased Transformer for Robust Multi-View 3D Human Pose\n  Reconstruction", "abstract": "  We address the challenges in estimating 3D human poses from multiple views\nunder occlusion and with limited overlapping views. We approach multi-view,\nsingle-person 3D human pose reconstruction as a regression problem and propose\na novel encoder-decoder Transformer architecture to estimate 3D poses from\nmulti-view 2D pose sequences. The encoder refines 2D skeleton joints detected\nacross different views and times, fusing multi-view and temporal information\nthrough global self-attention. We enhance the encoder by incorporating a\ngeometry-biased attention mechanism, effectively leveraging geometric\nrelationships between views. Additionally, we use detection scores provided by\nthe 2D pose detector to further guide the encoder's attention based on the\nreliability of the 2D detections. The decoder subsequently regresses the 3D\npose sequence from these refined tokens, using pre-defined queries for each\njoint. To enhance the generalization of our method to unseen scenes and improve\nresilience to missing joints, we implement strategies including scene\ncentering, synthetic views, and token dropout. We conduct extensive experiments\non three benchmark public datasets, Human3.6M, CMU Panoptic and\nOcclusion-Persons. Our results demonstrate the efficacy of our approach,\nparticularly in occluded scenes and when few views are available, which are\ntraditionally challenging scenarios for triangulation-based methods.\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": "2312.15636", "title": "Lifting by Image -- Leveraging Image Cues for Accurate 3D Human Pose\n  Estimation", "abstract": "  The \"lifting from 2D pose\" method has been the dominant approach to 3D Human\nPose Estimation (3DHPE) due to the powerful visual analysis ability of 2D pose\nestimators. Widely known, there exists a depth ambiguity problem when\nestimating solely from 2D pose, where one 2D pose can be mapped to multiple 3D\nposes. Intuitively, the rich semantic and texture information in images can\ncontribute to a more accurate \"lifting\" procedure. Yet, existing research\nencounters two primary challenges. Firstly, the distribution of image data in\n3D motion capture datasets is too narrow because of the laboratorial\nenvironment, which leads to poor generalization ability of methods trained with\nimage information. Secondly, effective strategies for leveraging image\ninformation are lacking. In this paper, we give new insight into the cause of\npoor generalization problems and the effectiveness of image features. Based on\nthat, we propose an advanced framework. Specifically, the framework consists of\ntwo stages. First, we enable the keypoints to query and select the beneficial\nfeatures from all image patches. To reduce the keypoints attention to\ninconsequential background features, we design a novel Pose-guided Transformer\nLayer, which adaptively limits the updates to unimportant image patches. Then,\nthrough a designed Adaptive Feature Selection Module, we prune less significant\nimage patches from the feature map. In the second stage, we allow the keypoints\nto further emphasize the retained critical image features. This progressive\nlearning approach prevents further training on insignificant image features.\nExperimental results show that our model achieves state-of-the-art performance\non both the Human3.6M dataset and the MPI-INF-3DHP dataset.\n", "date": "2023-12-25", "categories": []}
{"id": "2312.14915", "title": "PoseGen: Learning to Generate 3D Human Pose Dataset with NeRF", "abstract": "  This paper proposes an end-to-end framework for generating 3D human pose\ndatasets using Neural Radiance Fields (NeRF). Public datasets generally have\nlimited diversity in terms of human poses and camera viewpoints, largely due to\nthe resource-intensive nature of collecting 3D human pose data. As a result,\npose estimators trained on public datasets significantly underperform when\napplied to unseen out-of-distribution samples. Previous works proposed\naugmenting public datasets by generating 2D-3D pose pairs or rendering a large\namount of random data. Such approaches either overlook image rendering or\nresult in suboptimal datasets for pre-trained models. Here we propose PoseGen,\nwhich learns to generate a dataset (human 3D poses and images) with a feedback\nloss from a given pre-trained pose estimator. In contrast to prior art, our\ngenerated data is optimized to improve the robustness of the pre-trained model.\nThe objective of PoseGen is to learn a distribution of data that maximizes the\nprediction error of a given pre-trained model. As the learned data distribution\ncontains OOD samples of the pre-trained model, sampling data from such a\ndistribution for further fine-tuning a pre-trained model improves the\ngeneralizability of the model. This is the first work that proposes NeRFs for\n3D human data generation. NeRFs are data-driven and do not require 3D scans of\nhumans. Therefore, using NeRF for data generation is a new direction for\nconvenient user-specific data generation. Our extensive experiments show that\nthe proposed PoseGen improves two baseline models (SPIN and HybrIK) on four\ndatasets with an average 6% relative improvement.\n", "date": "2023-12-22", "categories": []}
{"id": "2312.10195", "title": "SoloPose: One-Shot Kinematic 3D Human Pose Estimation with Video Data\n  Augmentation", "abstract": "  While recent two-stage many-to-one deep learning models have demonstrated\ngreat success in 3D human pose estimation, such models are inefficient ways to\ndetect 3D key points in a sequential video relative to one-shot and\nmany-to-many models. Another key drawback of two-stage and many-to-one models\nis that errors in the first stage will be passed onto the second stage. In this\npaper, we introduce SoloPose, a novel one-shot, many-to-many spatio-temporal\ntransformer model for kinematic 3D human pose estimation of video. SoloPose is\nfurther fortified by HeatPose, a 3D heatmap based on Gaussian Mixture Model\ndistributions that factors target key points as well as kinematically adjacent\nkey points. Finally, we address data diversity constraints with the 3D\nAugMotion Toolkit, a methodology to augment existing 3D human pose datasets,\nspecifically by projecting four top public 3D human pose datasets (Humans3.6M,\nMADS, AIST Dance++, MPI INF 3DHP) into a novel dataset (Humans7.1M) with a\nuniversal coordinate system. Extensive experiments are conducted on Human3.6M\nas well as the augmented Humans7.1M dataset, and SoloPose demonstrates superior\nresults relative to the state-of-the-art approaches.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.08730", "title": "Towards Robust and Expressive Whole-body Human Pose and Shape Estimation", "abstract": "  Whole-body pose and shape estimation aims to jointly predict different\nbehaviors (e.g., pose, hand gesture, facial expression) of the entire human\nbody from a monocular image. Existing methods often exhibit degraded\nperformance under the complexity of in-the-wild scenarios. We argue that the\naccuracy and reliability of these models are significantly affected by the\nquality of the predicted \\textit{bounding box}, e.g., the scale and alignment\nof body parts. The natural discrepancy between the ideal bounding box\nannotations and model detection results is particularly detrimental to the\nperformance of whole-body pose and shape estimation. In this paper, we propose\na novel framework to enhance the robustness of whole-body pose and shape\nestimation. Our framework incorporates three new modules to address the above\nchallenges from three perspectives: \\textbf{1) Localization Module} enhances\nthe model's awareness of the subject's location and semantics within the image\nspace. \\textbf{2) Contrastive Feature Extraction Module} encourages the model\nto be invariant to robust augmentations by incorporating contrastive loss with\ndedicated positive samples. \\textbf{3) Pixel Alignment Module} ensures the\nreprojected mesh from the predicted camera and body model parameters are\naccurate and pixel-aligned. We perform comprehensive experiments to demonstrate\nthe effectiveness of our proposed framework on body, hands, face and whole-body\nbenchmarks. Codebase is available at\n\\url{https://github.com/robosmplx/robosmplx}.\n", "date": "2023-12-14", "categories": []}
{"id": "2312.08871", "title": "VoxelKP: A Voxel-based Network Architecture for Human Keypoint\n  Estimation in LiDAR Data", "abstract": "  We present \\textit{VoxelKP}, a novel fully sparse network architecture\ntailored for human keypoint estimation in LiDAR data. The key challenge is that\nobjects are distributed sparsely in 3D space, while human keypoint detection\nrequires detailed local information wherever humans are present. We propose\nfour novel ideas in this paper. First, we propose sparse selective kernels to\ncapture multi-scale context. Second, we introduce sparse box-attention to focus\non learning spatial correlations between keypoints within each human instance.\nThird, we incorporate a spatial encoding to leverage absolute 3D coordinates\nwhen projecting 3D voxels to a 2D grid encoding a bird's eye view. Finally, we\npropose hybrid feature learning to combine the processing of per-voxel features\nwith sparse convolution. We evaluate our method on the Waymo dataset and\nachieve an improvement of $27\\%$ on the MPJPE metric compared to the\nstate-of-the-art, \\textit{HUM3DIL}, trained on the same data, and $12\\%$\nagainst the state-of-the-art, \\textit{GC-KPL}, pretrained on a $25\\times$\nlarger dataset. To the best of our knowledge, \\textit{VoxelKP} is the first\nsingle-staged, fully sparse network that is specifically designed for\naddressing the challenging task of 3D keypoint estimation from LiDAR data,\nachieving state-of-the-art performances. Our code is available at\n\\url{https://github.com/shijianjian/VoxelKP}.\n", "date": "2023-12-11", "categories": []}
{"id": "2306.09337", "title": "Generative Proxemics: A Prior for 3D Social Interaction from Images", "abstract": "  Social interaction is a fundamental aspect of human behavior and\ncommunication. The way individuals position themselves in relation to others,\nalso known as proxemics, conveys social cues and affects the dynamics of social\ninteraction. Reconstructing such interaction from images presents challenges\nbecause of mutual occlusion and the limited availability of large training\ndatasets. To address this, we present a novel approach that learns a prior over\nthe 3D proxemics two people in close social interaction and demonstrate its use\nfor single-view 3D reconstruction. We start by creating 3D training data of\ninteracting people using image datasets with contact annotations. We then model\nthe proxemics using a novel denoising diffusion model called BUDDI that learns\nthe joint distribution over the poses of two people in close social\ninteraction. Sampling from our generative proxemics model produces realistic 3D\nhuman interactions, which we validate through a perceptual study. We use BUDDI\nin reconstructing two people in close proximity from a single image without any\ncontact annotation via an optimization approach that uses the diffusion model\nas a prior. Our approach recovers accurate and plausible 3D social interactions\nfrom noisy initial estimates, outperforming state-of-the-art methods. Our code,\ndata, and model are availableat our project website at: muelea.github.io/buddi.\n", "date": "2023-06-15", "categories": []}
{"id": "2309.09314", "title": "MOVIN: Real-time Motion Capture using a Single LiDAR", "abstract": "  Recent advancements in technology have brought forth new forms of interactive\napplications, such as the social metaverse, where end users interact with each\nother through their virtual avatars. In such applications, precise full-body\ntracking is essential for an immersive experience and a sense of embodiment\nwith the virtual avatar. However, current motion capture systems are not easily\naccessible to end users due to their high cost, the requirement for special\nskills to operate them, or the discomfort associated with wearable devices. In\nthis paper, we present MOVIN, the data-driven generative method for real-time\nmotion capture with global tracking, using a single LiDAR sensor. Our\nautoregressive conditional variational autoencoder (CVAE) model learns the\ndistribution of pose variations conditioned on the given 3D point cloud from\nLiDAR.As a central factor for high-accuracy motion capture, we propose a novel\nfeature encoder to learn the correlation between the historical 3D point cloud\ndata and global, local pose features, resulting in effective learning of the\npose prior. Global pose features include root translation, rotation, and foot\ncontacts, while local features comprise joint positions and rotations.\nSubsequently, a pose generator takes into account the sampled latent variable\nalong with the features from the previous frame to generate a plausible current\npose. Our framework accurately predicts the performer's 3D global information\nand local joint details while effectively considering temporally coherent\nmovements across frames. We demonstrate the effectiveness of our architecture\nthrough quantitative and qualitative evaluations, comparing it against\nstate-of-the-art methods. Additionally, we implement a real-time application to\nshowcase our method in real-world scenarios. MOVIN dataset is available at\n\\url{https://movin3d.github.io/movin_pg2023/}.\n", "date": "2023-09-17", "categories": []}
{"id": "2311.16495", "title": "Egocentric Whole-Body Motion Capture with FisheyeViT and Diffusion-Based\n  Motion Refinement", "abstract": "  In this work, we explore egocentric whole-body motion capture using a single\nfisheye camera, which simultaneously estimates human body and hand motion. This\ntask presents significant challenges due to three factors: the lack of\nhigh-quality datasets, fisheye camera distortion, and human body\nself-occlusion. To address these challenges, we propose a novel approach that\nleverages FisheyeViT to extract fisheye image features, which are subsequently\nconverted into pixel-aligned 3D heatmap representations for 3D human body pose\nprediction. For hand tracking, we incorporate dedicated hand detection and hand\npose estimation networks for regressing 3D hand poses. Finally, we develop a\ndiffusion-based whole-body motion prior model to refine the estimated\nwhole-body motion while accounting for joint uncertainties. To train these\nnetworks, we collect a large synthetic dataset, EgoWholeBody, comprising\n840,000 high-quality egocentric images captured across a diverse range of\nwhole-body motion sequences. Quantitative and qualitative evaluations\ndemonstrate the effectiveness of our method in producing high-quality\nwhole-body motion estimates from a single egocentric camera.\n", "date": "2023-11-28", "categories": []}
{"id": "2312.01561", "title": "Multi-View Person Matching and 3D Pose Estimation with Arbitrary\n  Uncalibrated Camera Networks", "abstract": "  Cross-view person matching and 3D human pose estimation in multi-camera\nnetworks are particularly difficult when the cameras are extrinsically\nuncalibrated. Existing efforts generally require large amounts of 3D data for\ntraining neural networks or known camera poses for geometric constraints to\nsolve the problem. However, camera poses and 3D data annotation are usually\nexpensive and not always available. We present a method, PME, that solves the\ntwo tasks without requiring either information. Our idea is to address\ncross-view person matching as a clustering problem using each person as a\ncluster center, then obtain correspondences from person matches, and estimate\n3D human poses through multi-view triangulation and bundle adjustment. We solve\nthe clustering problem by introducing a \"size constraint\" using the number of\ncameras and a \"source constraint\" using the fact that two people from the same\ncamera view should not match, to narrow the solution space to a small feasible\nregion. The 2D human poses used in clustering are obtained through a\npre-trained 2D pose detector, so our method does not require expensive 3D\ntraining data for each new scene. We extensively evaluate our method on three\nopen datasets and two indoor and outdoor datasets collected using arbitrarily\nset cameras. Our method outperforms other methods by a large margin on\ncross-view person matching, reaches SOTA performance on 3D human pose\nestimation without using either camera poses or 3D training data, and shows\ngood generalization ability across five datasets of various environment\nsettings.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.00462", "title": "Learning Unorthogonalized Matrices for Rotation Estimation", "abstract": "  Estimating 3D rotations is a common procedure for 3D computer vision. The\naccuracy depends heavily on the rotation representation. One form of\nrepresentation -- rotation matrices -- is popular due to its continuity,\nespecially for pose estimation tasks. The learning process usually incorporates\northogonalization to ensure orthonormal matrices. Our work reveals, through\ngradient analysis, that common orthogonalization procedures based on the\nGram-Schmidt process and singular value decomposition will slow down training\nefficiency. To this end, we advocate removing orthogonalization from the\nlearning process and learning unorthogonalized `Pseudo' Rotation Matrices\n(PRoM). An optimization analysis shows that PRoM converges faster and to a\nbetter solution. By replacing the orthogonalization incorporated representation\nwith our proposed PRoM in various rotation-related tasks, we achieve\nstate-of-the-art results on large-scale benchmarks for human pose estimation.\n", "date": "2023-12-01", "categories": []}
{"id": "2311.16477", "title": "UniHPE: Towards Unified Human Pose Estimation via Contrastive Learning", "abstract": "  In recent times, there has been a growing interest in developing effective\nperception techniques for combining information from multiple modalities. This\ninvolves aligning features obtained from diverse sources to enable more\nefficient training with larger datasets and constraints, as well as leveraging\nthe wealth of information contained in each modality. 2D and 3D Human Pose\nEstimation (HPE) are two critical perceptual tasks in computer vision, which\nhave numerous downstream applications, such as Action Recognition,\nHuman-Computer Interaction, Object tracking, etc. Yet, there are limited\ninstances where the correlation between Image and 2D/3D human pose has been\nclearly researched using a contrastive paradigm. In this paper, we propose\nUniHPE, a unified Human Pose Estimation pipeline, which aligns features from\nall three modalities, i.e., 2D human pose estimation, lifting-based and\nimage-based 3D human pose estimation, in the same pipeline. To align more than\ntwo modalities at the same time, we propose a novel singular value based\ncontrastive learning loss, which better aligns different modalities and further\nboosts the performance. In our evaluation, UniHPE achieves remarkable\nperformance metrics: MPJPE $50.5$mm on the Human3.6M dataset and PAMPJPE\n$51.6$mm on the 3DPW dataset. Our proposed method holds immense potential to\nadvance the field of computer vision and contribute to various applications.\n", "date": "2023-11-24", "categories": []}
{"id": "2311.12043", "title": "Efficient Domain Adaptation via Generative Prior for 3D Infant Pose\n  Estimation", "abstract": "  Although 3D human pose estimation has gained impressive development in recent\nyears, only a few works focus on infants, that have different bone lengths and\nalso have limited data. Directly applying adult pose estimation models\ntypically achieves low performance in the infant domain and suffers from\nout-of-distribution issues. Moreover, the limitation of infant pose data\ncollection also heavily constrains the efficiency of learning-based models to\nlift 2D poses to 3D. To deal with the issues of small datasets, domain\nadaptation and data augmentation are commonly used techniques. Following this\nparadigm, we take advantage of an optimization-based method that utilizes\ngenerative priors to predict 3D infant keypoints from 2D keypoints without the\nneed of large training data. We further apply a guided diffusion model to\ndomain adapt 3D adult pose to infant pose to supplement small datasets.\nBesides, we also prove that our method, ZeDO-i, could attain efficient domain\nadaptation, even if only a small number of data is given. Quantitatively, we\nclaim that our model attains state-of-the-art MPJPE performance of 43.6 mm on\nthe SyRIP dataset and 21.2 mm on the MINI-RGBD dataset.\n", "date": "2023-11-17", "categories": []}
{"id": "2306.02850", "title": "TRACE: 5D Temporal Regression of Avatars with Dynamic Cameras in 3D\n  Environments", "abstract": "  Although the estimation of 3D human pose and shape (HPS) is rapidly\nprogressing, current methods still cannot reliably estimate moving humans in\nglobal coordinates, which is critical for many applications. This is\nparticularly challenging when the camera is also moving, entangling human and\ncamera motion. To address these issues, we adopt a novel 5D representation\n(space, time, and identity) that enables end-to-end reasoning about people in\nscenes. Our method, called TRACE, introduces several novel architectural\ncomponents. Most importantly, it uses two new \"maps\" to reason about the 3D\ntrajectory of people over time in camera, and world, coordinates. An additional\nmemory unit enables persistent tracking of people even during long occlusions.\nTRACE is the first one-stage method to jointly recover and track 3D humans in\nglobal coordinates from dynamic cameras. By training it end-to-end, and using\nfull image information, TRACE achieves state-of-the-art performance on tracking\nand HPS benchmarks. The code and dataset are released for research purposes.\n", "date": "2023-06-05", "categories": []}
{"id": "2308.12914", "title": "3D Pose Nowcasting: Forecast the Future to Improve the Present", "abstract": "  Technologies to enable safe and effective collaboration and coexistence\nbetween humans and robots have gained significant importance in the last few\nyears. A critical component useful for realizing this collaborative paradigm is\nthe understanding of human and robot 3D poses using non-invasive systems.\nTherefore, in this paper, we propose a novel vision-based system leveraging\ndepth data to accurately establish the 3D locations of skeleton joints.\nSpecifically, we introduce the concept of Pose Nowcasting, denoting the\ncapability of the proposed system to enhance its current pose estimation\naccuracy by jointly learning to forecast future poses. The experimental\nevaluation is conducted on two different datasets, providing accurate and\nreal-time performance and confirming the validity of the proposed method on\nboth the robotic and human scenarios.\n", "date": "2023-08-24", "categories": []}
{"id": "2311.10983", "title": "Multiple View Geometry Transformers for 3D Human Pose Estimation", "abstract": "  In this work, we aim to improve the 3D reasoning ability of Transformers in\nmulti-view 3D human pose estimation. Recent works have focused on end-to-end\nlearning-based transformer designs, which struggle to resolve geometric\ninformation accurately, particularly during occlusion. Instead, we propose a\nnovel hybrid model, MVGFormer, which has a series of geometric and appearance\nmodules organized in an iterative manner. The geometry modules are\nlearning-free and handle all viewpoint-dependent 3D tasks geometrically which\nnotably improves the model's generalization ability. The appearance modules are\nlearnable and are dedicated to estimating 2D poses from image signals\nend-to-end which enables them to achieve accurate estimates even when occlusion\noccurs, leading to a model that is both accurate and generalizable to new\ncameras and geometries. We evaluate our approach for both in-domain and\nout-of-domain settings, where our model consistently outperforms\nstate-of-the-art methods, and especially does so by a significant margin in the\nout-of-domain setting. We will release the code and models:\nhttps://github.com/XunshanMan/MVGFormer.\n", "date": "2023-11-18", "categories": []}
{"id": "2311.11662", "title": "Enhanced Spatio-Temporal Context for Temporally Consistent Robust 3D\n  Human Motion Recovery from Monocular Videos", "abstract": "  Recovering temporally consistent 3D human body pose, shape and motion from a\nmonocular video is a challenging task due to (self-)occlusions, poor lighting\nconditions, complex articulated body poses, depth ambiguity, and limited\navailability of annotated data. Further, doing a simple perframe estimation is\ninsufficient as it leads to jittery and implausible results. In this paper, we\npropose a novel method for temporally consistent motion estimation from a\nmonocular video. Instead of using generic ResNet-like features, our method uses\na body-aware feature representation and an independent per-frame pose and\ncamera initialization over a temporal window followed by a novel\nspatio-temporal feature aggregation by using a combination of self-similarity\nand self-attention over the body-aware features and the perframe\ninitialization. Together, they yield enhanced spatiotemporal context for every\nframe by considering remaining past and future frames. These features are used\nto predict the pose and shape parameters of the human body model, which are\nfurther refined using an LSTM. Experimental results on the publicly available\nbenchmark data show that our method attains significantly lower acceleration\nerror and outperforms the existing state-of-the-art methods over all key\nquantitative evaluation metrics, including complex scenarios like partial\nocclusion, complex poses and even relatively low illumination.\n", "date": "2023-11-20", "categories": []}
{"id": "2311.09543", "title": "Temporal-Aware Refinement for Video-based Human Pose and Shape Recovery", "abstract": "  Though significant progress in human pose and shape recovery from monocular\nRGB images has been made in recent years, obtaining 3D human motion with high\naccuracy and temporal consistency from videos remains challenging. Existing\nvideo-based methods tend to reconstruct human motion from global image\nfeatures, which lack detailed representation capability and limit the\nreconstruction accuracy. In this paper, we propose a Temporal-Aware Refining\nNetwork (TAR), to synchronously explore temporal-aware global and local image\nfeatures for accurate pose and shape recovery. First, a global transformer\nencoder is introduced to obtain temporal global features from static feature\nsequences. Second, a bidirectional ConvGRU network takes the sequence of\nhigh-resolution feature maps as input, and outputs temporal local feature maps\nthat maintain high resolution and capture the local motion of the human body.\nFinally, a recurrent refinement module iteratively updates estimated SMPL\nparameters by leveraging both global and local temporal information to achieve\naccurate and smooth results. Extensive experiments demonstrate that our TAR\nobtains more accurate results than previous state-of-the-art methods on popular\nbenchmarks, i.e., 3DPW, MPI-INF-3DHP, and Human3.6M.\n", "date": "2023-11-16", "categories": []}
{"id": "2311.03312", "title": "A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose\n  Estimation", "abstract": "  The dominant paradigm in 3D human pose estimation that lifts a 2D pose\nsequence to 3D heavily relies on long-term temporal clues (i.e., using a\ndaunting number of video frames) for improved accuracy, which incurs\nperformance saturation, intractable computation and the non-causal problem.\nThis can be attributed to their inherent inability to perceive spatial context\nas plain 2D joint coordinates carry no visual cues. To address this issue, we\npropose a straightforward yet powerful solution: leveraging the readily\navailable intermediate visual representations produced by off-the-shelf\n(pre-trained) 2D pose detectors -- no finetuning on the 3D task is even needed.\nThe key observation is that, while the pose detector learns to localize 2D\njoints, such representations (e.g., feature maps) implicitly encode the\njoint-centric spatial context thanks to the regional operations in backbone\nnetworks. We design a simple baseline named Context-Aware PoseFormer to\nshowcase its effectiveness. Without access to any temporal information, the\nproposed method significantly outperforms its context-agnostic counterpart,\nPoseFormer, and other state-of-the-art methods using up to hundreds of video\nframes regarding both speed and precision. Project page:\nhttps://qitaozhao.github.io/ContextAware-PoseFormer\n", "date": "2023-11-06", "categories": []}
{"id": "2311.02815", "title": "Efficient, Self-Supervised Human Pose Estimation with Inductive Prior\n  Tuning", "abstract": "  The goal of 2D human pose estimation (HPE) is to localize anatomical\nlandmarks, given an image of a person in a pose. SOTA techniques make use of\nthousands of labeled figures (finetuning transformers or training deep CNNs),\nacquired using labor-intensive crowdsourcing. On the other hand,\nself-supervised methods re-frame the HPE task as a reconstruction problem,\nenabling them to leverage the vast amount of unlabeled visual data, though at\nthe present cost of accuracy. In this work, we explore ways to improve\nself-supervised HPE. We (1) analyze the relationship between reconstruction\nquality and pose estimation accuracy, (2) develop a model pipeline that\noutperforms the baseline which inspired our work, using less than one-third the\namount of training data, and (3) offer a new metric suitable for\nself-supervised settings that measures the consistency of predicted body part\nlength proportions. We show that a combination of well-engineered\nreconstruction losses and inductive priors can help coordinate pose learning\nalongside reconstruction in a self-supervised paradigm.\n", "date": "2023-11-06", "categories": []}
{"id": "2210.04216", "title": "AMPose: Alternately Mixed Global-Local Attention Model for 3D Human Pose\n  Estimation", "abstract": "  The graph convolutional networks (GCNs) have been applied to model the\nphysically connected and non-local relations among human joints for 3D human\npose estimation (HPE). In addition, the purely Transformer-based models\nrecently show promising results in video-based 3D HPE. However, the\nsingle-frame method still needs to model the physically connected relations\namong joints because the feature representations transformed only by global\nrelations via the Transformer neglect information on the human skeleton. To\ndeal with this problem, we propose a novel method in which the Transformer\nencoder and GCN blocks are alternately stacked, namely AMPose, to combine the\nglobal and physically connected relations among joints towards HPE. In the\nAMPose, the Transformer encoder is applied to connect each joint with all the\nother joints, while GCNs are applied to capture information on physically\nconnected relations. The effectiveness of our proposed method is evaluated on\nthe Human3.6M dataset. Our model also shows better generalization ability by\ntesting on the MPI-INF-3DHP dataset. Code can be retrieved at\nhttps://github.com/erikervalid/AMPose.\n", "date": "2022-10-09", "categories": []}
{"id": "2302.14581", "title": "HopFIR: Hop-wise GraphFormer with Intragroup Joint Refinement for 3D\n  Human Pose Estimation", "abstract": "  2D-to-3D human pose lifting is fundamental for 3D human pose estimation\n(HPE), for which graph convolutional networks (GCNs) have proven inherently\nsuitable for modeling the human skeletal topology. However, the current\nGCN-based 3D HPE methods update the node features by aggregating their\nneighbors' information without considering the interaction of joints in\ndifferent joint synergies. Although some studies have proposed importing limb\ninformation to learn the movement patterns, the latent synergies among joints,\nsuch as maintaining balance are seldom investigated. We propose the Hop-wise\nGraphFormer with Intragroup Joint Refinement (HopFIR) architecture to tackle\nthe 3D HPE problem. HopFIR mainly consists of a novel hop-wise GraphFormer\n(HGF) module and an intragroup joint refinement (IJR) module. The HGF module\ngroups the joints by k-hop neighbors and applies a hopwise transformer-like\nattention mechanism to these groups to discover latent joint synergies. The IJR\nmodule leverages the prior limb information for peripheral joint refinement.\nExtensive experimental results show that HopFIR outperforms the SOTA methods by\na large margin, with a mean per-joint position error (MPJPE) on the Human3.6M\ndataset of 32.67 mm. We also demonstrate that the state-of-the-art GCN-based\nmethods can benefit from the proposed hop-wise attention mechanism with a\nsignificant improvement in performance: SemGCN and MGCN are improved by 8.9%\nand 4.5%, respectively.\n", "date": "2023-02-28", "categories": []}
{"id": "2310.18920", "title": "Improving Multi-Person Pose Tracking with A Confidence Network", "abstract": "  Human pose estimation and tracking are fundamental tasks for understanding\nhuman behaviors in videos. Existing top-down framework-based methods usually\nperform three-stage tasks: human detection, pose estimation and tracking.\nAlthough promising results have been achieved, these methods rely heavily on\nhigh-performance detectors and may fail to track persons who are occluded or\nmiss-detected. To overcome these problems, in this paper, we develop a novel\nkeypoint confidence network and a tracking pipeline to improve human detection\nand pose estimation in top-down approaches. Specifically, the keypoint\nconfidence network is designed to determine whether each keypoint is occluded,\nand it is incorporated into the pose estimation module. In the tracking\npipeline, we propose the Bbox-revision module to reduce missing detection and\nthe ID-retrieve module to correct lost trajectories, improving the performance\nof the detection stage. Experimental results show that our approach is\nuniversal in human detection and pose estimation, achieving state-of-the-art\nperformance on both PoseTrack 2017 and 2018 datasets.\n", "date": "2023-10-29", "categories": []}
{"id": "2307.03833", "title": "Back to Optimization: Diffusion-based Zero-Shot 3D Human Pose Estimation", "abstract": "  Learning-based methods have dominated the 3D human pose estimation (HPE)\ntasks with significantly better performance in most benchmarks than traditional\noptimization-based methods. Nonetheless, 3D HPE in the wild is still the\nbiggest challenge for learning-based models, whether with 2D-3D lifting,\nimage-to-3D, or diffusion-based methods, since the trained networks implicitly\nlearn camera intrinsic parameters and domain-based 3D human pose distributions\nand estimate poses by statistical average. On the other hand, the\noptimization-based methods estimate results case-by-case, which can predict\nmore diverse and sophisticated human poses in the wild. By combining the\nadvantages of optimization-based and learning-based methods, we propose the\n\\textbf{Ze}ro-shot \\textbf{D}iffusion-based \\textbf{O}ptimization\n(\\textbf{ZeDO}) pipeline for 3D HPE to solve the problem of cross-domain and\nin-the-wild 3D HPE. Our multi-hypothesis \\textit{\\textbf{ZeDO}} achieves\nstate-of-the-art (SOTA) performance on Human3.6M, with minMPJPE $51.4$mm,\nwithout training with any 2D-3D or image-3D pairs. Moreover, our\nsingle-hypothesis \\textit{\\textbf{ZeDO}} achieves SOTA performance on 3DPW\ndataset with PA-MPJPE $40.3$mm on cross-dataset evaluation, which even\noutperforms learning-based methods trained on 3DPW.\n", "date": "2023-07-07", "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": "2310.16288", "title": "MotionAGFormer: Enhancing 3D Human Pose Estimation with a\n  Transformer-GCNFormer Network", "abstract": "  Recent transformer-based approaches have demonstrated excellent performance\nin 3D human pose estimation. However, they have a holistic view and by encoding\nglobal relationships between all the joints, they do not capture the local\ndependencies precisely. In this paper, we present a novel Attention-GCNFormer\n(AGFormer) block that divides the number of channels by using two parallel\ntransformer and GCNFormer streams. Our proposed GCNFormer module exploits the\nlocal relationship between adjacent joints, outputting a new representation\nthat is complementary to the transformer output. By fusing these two\nrepresentation in an adaptive way, AGFormer exhibits the ability to better\nlearn the underlying 3D structure. By stacking multiple AGFormer blocks, we\npropose MotionAGFormer in four different variants, which can be chosen based on\nthe speed-accuracy trade-off. We evaluate our model on two popular benchmark\ndatasets: Human3.6M and MPI-INF-3DHP. MotionAGFormer-B achieves\nstate-of-the-art results, with P1 errors of 38.4mm and 16.2mm, respectively.\nRemarkably, it uses a quarter of the parameters and is three times more\ncomputationally efficient than the previous leading model on Human3.6M dataset.\nCode and models are available at https://github.com/TaatiTeam/MotionAGFormer.\n", "date": "2023-10-25", "categories": []}
{"id": "2310.13039", "title": "Human Pose-based Estimation, Tracking and Action Recognition with Deep\n  Learning: A Survey", "abstract": "  Human pose analysis has garnered significant attention within both the\nresearch community and practical applications, owing to its expanding array of\nuses, including gaming, video surveillance, sports performance analysis, and\nhuman-computer interactions, among others. The advent of deep learning has\nsignificantly improved the accuracy of pose capture, making pose-based\napplications increasingly practical. This paper presents a comprehensive survey\nof pose-based applications utilizing deep learning, encompassing pose\nestimation, pose tracking, and action recognition.Pose estimation involves the\ndetermination of human joint positions from images or image sequences. Pose\ntracking is an emerging research direction aimed at generating consistent human\npose trajectories over time. Action recognition, on the other hand, targets the\nidentification of action types using pose estimation or tracking data. These\nthree tasks are intricately interconnected, with the latter often reliant on\nthe former. In this survey, we comprehensively review related works, spanning\nfrom single-person pose estimation to multi-person pose estimation, from 2D\npose estimation to 3D pose estimation, from single image to video, from mining\ntemporal context gradually to pose tracking, and lastly from tracking to\npose-based action recognition. As a survey centered on the application of deep\nlearning to pose analysis, we explicitly discuss both the strengths and\nlimitations of existing techniques. Notably, we emphasize methodologies for\nintegrating these three tasks into a unified framework within video sequences.\nAdditionally, we explore the challenges involved and outline potential\ndirections for future research.\n", "date": "2023-10-19", "categories": []}
{"id": "2308.09678", "title": "PoSynDA: Multi-Hypothesis Pose Synthesis Domain Adaptation for Robust 3D\n  Human Pose Estimation", "abstract": "  Existing 3D human pose estimators face challenges in adapting to new datasets\ndue to the lack of 2D-3D pose pairs in training sets. To overcome this issue,\nwe propose \\textit{Multi-Hypothesis \\textbf{P}ose \\textbf{Syn}thesis\n\\textbf{D}omain \\textbf{A}daptation} (\\textbf{PoSynDA}) framework to bridge\nthis data disparity gap in target domain. Typically, PoSynDA uses a\ndiffusion-inspired structure to simulate 3D pose distribution in the target\ndomain. By incorporating a multi-hypothesis network, PoSynDA generates diverse\npose hypotheses and aligns them with the target domain. To do this, it first\nutilizes target-specific source augmentation to obtain the target domain\ndistribution data from the source domain by decoupling the scale and position\nparameters. The process is then further refined through the teacher-student\nparadigm and low-rank adaptation. With extensive comparison of benchmarks such\nas Human3.6M and MPI-INF-3DHP, PoSynDA demonstrates competitive performance,\neven comparable to the target-trained MixSTE model\\cite{zhang2022mixste}. This\nwork paves the way for the practical application of 3D human pose estimation in\nunseen domains. The code is available at https://github.com/hbing-l/PoSynDA.\n", "date": "2023-08-18", "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": "2309.16670", "title": "Decaf: Monocular Deformation Capture for Face and Hand Interactions", "abstract": "  Existing methods for 3D tracking from monocular RGB videos predominantly\nconsider articulated and rigid objects. Modelling dense non-rigid object\ndeformations in this setting remained largely unaddressed so far, although such\neffects can improve the realism of the downstream applications such as AR/VR\nand avatar communications. This is due to the severe ill-posedness of the\nmonocular view setting and the associated challenges. While it is possible to\nnaively track multiple non-rigid objects independently using 3D templates or\nparametric 3D models, such an approach would suffer from multiple artefacts in\nthe resulting 3D estimates such as depth ambiguity, unnatural intra-object\ncollisions and missing or implausible deformations. Hence, this paper\nintroduces the first method that addresses the fundamental challenges depicted\nabove and that allows tracking human hands interacting with human faces in 3D\nfrom single monocular RGB videos. We model hands as articulated objects\ninducing non-rigid face deformations during an active interaction. Our method\nrelies on a new hand-face motion and interaction capture dataset with realistic\nface deformations acquired with a markerless multi-view camera system. As a\npivotal step in its creation, we process the reconstructed raw 3D shapes with\nposition-based dynamics and an approach for non-uniform stiffness estimation of\nthe head tissues, which results in plausible annotations of the surface\ndeformations, hand-face contact regions and head-hand positions. At the core of\nour neural approach are a variational auto-encoder supplying the hand-face\ndepth prior and modules that guide the 3D tracking by estimating the contacts\nand the deformations. Our final 3D hand and face reconstructions are realistic\nand more plausible compared to several baselines applicable in our setting,\nboth quantitatively and qualitatively.\nhttps://vcai.mpi-inf.mpg.de/projects/Decaf\n", "date": "2023-09-28", "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": "2306.07879", "title": "Rethinking pose estimation in crowds: overcoming the detection\n  information-bottleneck and ambiguity", "abstract": "  Frequent interactions between individuals are a fundamental challenge for\npose estimation algorithms. Current pipelines either use an object detector\ntogether with a pose estimator (top-down approach), or localize all body parts\nfirst and then link them to predict the pose of individuals (bottom-up). Yet,\nwhen individuals closely interact, top-down methods are ill-defined due to\noverlapping individuals, and bottom-up methods often falsely infer connections\nto distant bodyparts. Thus, we propose a novel pipeline called bottom-up\nconditioned top-down pose estimation (BUCTD) that combines the strengths of\nbottom-up and top-down methods. Specifically, we propose to use a bottom-up\nmodel as the detector, which in addition to an estimated bounding box provides\na pose proposal that is fed as condition to an attention-based top-down model.\nWe demonstrate the performance and efficiency of our approach on animal and\nhuman pose estimation benchmarks. On CrowdPose and OCHuman, we outperform\nprevious state-of-the-art models by a significant margin. We achieve 78.5 AP on\nCrowdPose and 48.5 AP on OCHuman, an improvement of 8.6% and 7.8% over the\nprior art, respectively. Furthermore, we show that our method strongly improves\nthe performance on multi-animal benchmarks involving fish and monkeys. The code\nis available at https://github.com/amathislab/BUCTD\n", "date": "2023-06-13", "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": "2309.15273", "title": "DECO: Dense Estimation of 3D Human-Scene Contact In The Wild", "abstract": "  Understanding how humans use physical contact to interact with the world is\nkey to enabling human-centric artificial intelligence. While inferring 3D\ncontact is crucial for modeling realistic and physically-plausible human-object\ninteractions, existing methods either focus on 2D, consider body joints rather\nthan the surface, use coarse 3D body regions, or do not generalize to\nin-the-wild images. In contrast, we focus on inferring dense, 3D contact\nbetween the full body surface and objects in arbitrary images. To achieve this,\nwe first collect DAMON, a new dataset containing dense vertex-level contact\nannotations paired with RGB images containing complex human-object and\nhuman-scene contact. Second, we train DECO, a novel 3D contact detector that\nuses both body-part-driven and scene-context-driven attention to estimate\nvertex-level contact on the SMPL body. DECO builds on the insight that human\nobservers recognize contact by reasoning about the contacting body parts, their\nproximity to scene objects, and the surrounding scene context. We perform\nextensive evaluations of our detector on DAMON as well as on the RICH and\nBEHAVE datasets. We significantly outperform existing SOTA methods across all\nbenchmarks. We also show qualitatively that DECO generalizes well to diverse\nand challenging real-world human interactions in natural images. The code,\ndata, and models are available at https://deco.is.tue.mpg.de.\n", "date": "2023-09-26", "categories": []}
{"id": "2212.11684", "title": "Scene-aware Egocentric 3D Human Pose Estimation", "abstract": "  Egocentric 3D human pose estimation with a single head-mounted fisheye camera\nhas recently attracted attention due to its numerous applications in virtual\nand augmented reality. Existing methods still struggle in challenging poses\nwhere the human body is highly occluded or is closely interacting with the\nscene. To address this issue, we propose a scene-aware egocentric pose\nestimation method that guides the prediction of the egocentric pose with scene\nconstraints. To this end, we propose an egocentric depth estimation network to\npredict the scene depth map from a wide-view egocentric fisheye camera while\nmitigating the occlusion of the human body with a depth-inpainting network.\nNext, we propose a scene-aware pose estimation network that projects the 2D\nimage features and estimated depth map of the scene into a voxel space and\nregresses the 3D pose with a V2V network. The voxel-based feature\nrepresentation provides the direct geometric connection between 2D image\nfeatures and scene geometry, and further facilitates the V2V network to\nconstrain the predicted pose based on the estimated scene geometry. To enable\nthe training of the aforementioned networks, we also generated a synthetic\ndataset, called EgoGTA, and an in-the-wild dataset based on EgoPW, called\nEgoPW-Scene. The experimental results of our new evaluation sequences show that\nthe predicted 3D egocentric poses are accurate and physically plausible in\nterms of human-scene interaction, demonstrating that our method outperforms the\nstate-of-the-art methods both quantitatively and qualitatively.\n", "date": "2022-12-20", "categories": []}
{"id": "2303.17209", "title": "Human from Blur: Human Pose Tracking from Blurry Images", "abstract": "  We propose a method to estimate 3D human poses from substantially blurred\nimages. The key idea is to tackle the inverse problem of image deblurring by\nmodeling the forward problem with a 3D human model, a texture map, and a\nsequence of poses to describe human motion. The blurring process is then\nmodeled by a temporal image aggregation step. Using a differentiable renderer,\nwe can solve the inverse problem by backpropagating the pixel-wise reprojection\nerror to recover the best human motion representation that explains a single or\nmultiple input images. Since the image reconstruction loss alone is\ninsufficient, we present additional regularization terms. To the best of our\nknowledge, we present the first method to tackle this problem. Our method\nconsistently outperforms other methods on significantly blurry inputs since\nthey lack one or multiple key functionalities that our method unifies, i.e.\nimage deblurring with sub-frame accuracy and explicit 3D modeling of non-rigid\nhuman motion.\n", "date": "2023-03-30", "categories": []}
{"id": "2305.18743", "title": "Decomposed Human Motion Prior for Video Pose Estimation via Adversarial\n  Training", "abstract": "  Estimating human pose from video is a task that receives considerable\nattention due to its applicability in numerous 3D fields. The complexity of\nprior knowledge of human body movements poses a challenge to neural network\nmodels in the task of regressing keypoints. In this paper, we address this\nproblem by incorporating motion prior in an adversarial way. Different from\nprevious methods, we propose to decompose holistic motion prior to joint motion\nprior, making it easier for neural networks to learn from prior knowledge\nthereby boosting the performance on the task. We also utilize a novel\nregularization loss to balance accuracy and smoothness introduced by motion\nprior. Our method achieves 9\\% lower PA-MPJPE and 29\\% lower acceleration error\nthan previous methods tested on 3DPW. The estimator proves its robustness by\nachieving impressive performance on in-the-wild dataset.\n", "date": "2023-05-30", "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": "2309.11667", "title": "Understanding Pose and Appearance Disentanglement in 3D Human Pose\n  Estimation", "abstract": "  As 3D human pose estimation can now be achieved with very high accuracy in\nthe supervised learning scenario, tackling the case where 3D pose annotations\nare not available has received increasing attention. In particular, several\nmethods have proposed to learn image representations in a self-supervised\nfashion so as to disentangle the appearance information from the pose one. The\nmethods then only need a small amount of supervised data to train a pose\nregressor using the pose-related latent vector as input, as it should be free\nof appearance information. In this paper, we carry out in-depth analysis to\nunderstand to what degree the state-of-the-art disentangled representation\nlearning methods truly separate the appearance information from the pose one.\nFirst, we study disentanglement from the perspective of the self-supervised\nnetwork, via diverse image synthesis experiments. Second, we investigate\ndisentanglement with respect to the 3D pose regressor following an adversarial\nattack perspective. Specifically, we design an adversarial strategy focusing on\ngenerating natural appearance changes of the subject, and against which we\ncould expect a disentangled network to be robust. Altogether, our analyses show\nthat disentanglement in the three state-of-the-art disentangled representation\nlearning frameworks if far from complete, and that their pose codes contain\nsignificant appearance information. We believe that our approach provides a\nvaluable testbed to evaluate the degree of disentanglement of pose from\nappearance in self-supervised 3D human pose estimation.\n", "date": "2023-09-20", "categories": []}
{"id": "2309.11962", "title": "Ego3DPose: Capturing 3D Cues from Binocular Egocentric Views", "abstract": "  We present Ego3DPose, a highly accurate binocular egocentric 3D pose\nreconstruction system. The binocular egocentric setup offers practicality and\nusefulness in various applications, however, it remains largely under-explored.\nIt has been suffering from low pose estimation accuracy due to viewing\ndistortion, severe self-occlusion, and limited field-of-view of the joints in\negocentric 2D images. Here, we notice that two important 3D cues, stereo\ncorrespondences, and perspective, contained in the egocentric binocular input\nare neglected. Current methods heavily rely on 2D image features, implicitly\nlearning 3D information, which introduces biases towards commonly observed\nmotions and leads to low overall accuracy. We observe that they not only fail\nin challenging occlusion cases but also in estimating visible joint positions.\nTo address these challenges, we propose two novel approaches. First, we design\na two-path network architecture with a path that estimates pose per limb\nindependently with its binocular heatmaps. Without full-body information\nprovided, it alleviates bias toward trained full-body distribution. Second, we\nleverage the egocentric view of body limbs, which exhibits strong perspective\nvariance (e.g., a significantly large-size hand when it is close to the\ncamera). We propose a new perspective-aware representation using trigonometry,\nenabling the network to estimate the 3D orientation of limbs. Finally, we\ndevelop an end-to-end pose reconstruction network that synergizes both\ntechniques. Our comprehensive evaluations demonstrate that Ego3DPose\noutperforms state-of-the-art models by a pose estimation error (i.e., MPJPE)\nreduction of 23.1% in the UnrealEgo dataset. Our qualitative results highlight\nthe superiority of our approach across a range of scenarios and challenges.\n", "date": "2023-09-21", "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": "2309.10369", "title": "GloPro: Globally-Consistent Uncertainty-Aware 3D Human Pose Estimation &\n  Tracking in the Wild", "abstract": "  An accurate and uncertainty-aware 3D human body pose estimation is key to\nenabling truly safe but efficient human-robot interactions. Current\nuncertainty-aware methods in 3D human pose estimation are limited to predicting\nthe uncertainty of the body posture, while effectively neglecting the body\nshape and root pose. In this work, we present GloPro, which to the best of our\nknowledge the first framework to predict an uncertainty distribution of a 3D\nbody mesh including its shape, pose, and root pose, by efficiently fusing\nvisual clues with a learned motion model. We demonstrate that it vastly\noutperforms state-of-the-art methods in terms of human trajectory accuracy in a\nworld coordinate system (even in the presence of severe occlusions), yields\nconsistent uncertainty distributions, and can run in real-time.\n", "date": "2023-09-19", "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": "2309.08644", "title": "Towards Robust and Smooth 3D Multi-Person Pose Estimation from Monocular\n  Videos in the Wild", "abstract": "  3D pose estimation is an invaluable task in computer vision with various\npractical applications. Especially, 3D pose estimation for multi-person from a\nmonocular video (3DMPPE) is particularly challenging and is still largely\nuncharted, far from applying to in-the-wild scenarios yet. We pose three\nunresolved issues with the existing methods: lack of robustness on unseen views\nduring training, vulnerability to occlusion, and severe jittering in the\noutput. As a remedy, we propose POTR-3D, the first realization of a\nsequence-to-sequence 2D-to-3D lifting model for 3DMPPE, powered by a novel\ngeometry-aware data augmentation strategy, capable of generating unbounded data\nwith a variety of views while caring about the ground plane and occlusions.\nThrough extensive experiments, we verify that the proposed model and data\naugmentation robustly generalizes to diverse unseen views, robustly recovers\nthe poses against heavy occlusions, and reliably generates more natural and\nsmoother outputs. The effectiveness of our approach is verified not only by\nachieving the state-of-the-art performance on public benchmarks, but also by\nqualitative results on more challenging in-the-wild videos. Demo videos are\navailable at https://www.youtube.com/@potr3d.\n", "date": "2023-09-15", "categories": []}
{"id": "2303.08657", "title": "Economical Quaternion Extraction from a Human Skeletal Pose Estimate\n  using 2-D Cameras", "abstract": "  In this paper, we present a novel algorithm to extract a quaternion from a\ntwo dimensional camera frame for estimating a contained human skeletal pose.\nThe problem of pose estimation is usually tackled through the usage of stereo\ncameras and intertial measurement units for obtaining depth and euclidean\ndistance for measurement of points in 3D space. However, the usage of these\ndevices comes with a high signal processing latency as well as a significant\nmonetary cost. By making use of MediaPipe, a framework for building perception\npipelines for human pose estimation, the proposed algorithm extracts a\nquaternion from a 2-D frame capturing an image of a human object at a sub-fifty\nmillisecond latency while also being capable of deployment at edges with a\nsingle camera frame and a generally low computational resource availability,\nespecially for use cases involving last-minute detection and reaction by\nautonomous robots. The algorithm seeks to bypass the funding barrier and\nimprove accessibility for robotics researchers involved in designing control\nsystems.\n", "date": "2023-03-15", "categories": []}
{"id": "2309.07243", "title": "LInKs \"Lifting Independent Keypoints\" -- Partial Pose Lifting for\n  Occlusion Handling with Improved Accuracy in 2D-3D Human Pose Estimation", "abstract": "  We present LInKs, a novel unsupervised learning method to recover 3D human\nposes from 2D kinematic skeletons obtained from a single image, even when\nocclusions are present. Our approach follows a unique two-step process, which\ninvolves first lifting the occluded 2D pose to the 3D domain, followed by\nfilling in the occluded parts using the partially reconstructed 3D coordinates.\nThis lift-then-fill approach leads to significantly more accurate results\ncompared to models that complete the pose in 2D space alone. Additionally, we\nimprove the stability and likelihood estimation of normalising flows through a\ncustom sampling function replacing PCA dimensionality reduction previously used\nin prior work. Furthermore, we are the first to investigate if different parts\nof the 2D kinematic skeleton can be lifted independently which we find by\nitself reduces the error of current lifting approaches. We attribute this to\nthe reduction of long-range keypoint correlations. In our detailed evaluation,\nwe quantify the error under various realistic occlusion scenarios, showcasing\nthe versatility and applicability of our model. Our results consistently\ndemonstrate the superiority of handling all types of occlusions in 3D space\nwhen compared to others that complete the pose in 2D space. Our approach also\nexhibits consistent accuracy in scenarios without occlusion, as evidenced by a\n7.9% reduction in reconstruction error compared to prior works on the Human3.6M\ndataset. Furthermore, our method excels in accurately retrieving complete 3D\nposes even in the presence of occlusions, making it highly applicable in\nsituations where complete 2D pose information is unavailable.\n", "date": "2023-09-13", "categories": []}
{"id": "2309.07910", "title": "TEMPO: Efficient Multi-View Pose Estimation, Tracking, and Forecasting", "abstract": "  Existing volumetric methods for predicting 3D human pose estimation are\naccurate, but computationally expensive and optimized for single time-step\nprediction. We present TEMPO, an efficient multi-view pose estimation model\nthat learns a robust spatiotemporal representation, improving pose accuracy\nwhile also tracking and forecasting human pose. We significantly reduce\ncomputation compared to the state-of-the-art by recurrently computing\nper-person 2D pose features, fusing both spatial and temporal information into\na single representation. In doing so, our model is able to use spatiotemporal\ncontext to predict more accurate human poses without sacrificing efficiency. We\nfurther use this representation to track human poses over time as well as\npredict future poses. Finally, we demonstrate that our model is able to\ngeneralize across datasets without scene-specific fine-tuning. TEMPO achieves\n10$\\%$ better MPJPE with a 33$\\times$ improvement in FPS compared to TesseTrack\non the challenging CMU Panoptic Studio dataset.\n", "date": "2023-09-14", "categories": []}
{"id": "2207.04320", "title": "Snipper: A Spatiotemporal Transformer for Simultaneous Multi-Person 3D\n  Pose Estimation Tracking and Forecasting on a Video Snippet", "abstract": "  Multi-person pose understanding from RGB videos involves three complex tasks:\npose estimation, tracking and motion forecasting. Intuitively, accurate\nmulti-person pose estimation facilitates robust tracking, and robust tracking\nbuilds crucial history for correct motion forecasting. Most existing works\neither focus on a single task or employ multi-stage approaches to solving\nmultiple tasks separately, which tends to make sub-optimal decision at each\nstage and also fail to exploit correlations among the three tasks. In this\npaper, we propose Snipper, a unified framework to perform multi-person 3D pose\nestimation, tracking, and motion forecasting simultaneously in a single stage.\nWe propose an efficient yet powerful deformable attention mechanism to\naggregate spatiotemporal information from the video snippet. Building upon this\ndeformable attention, a video transformer is learned to encode the\nspatiotemporal features from the multi-frame snippet and to decode informative\npose features for multi-person pose queries. Finally, these pose queries are\nregressed to predict multi-person pose trajectories and future motions in a\nsingle shot. In the experiments, we show the effectiveness of Snipper on three\nchallenging public datasets where our generic model rivals specialized\nstate-of-art baselines for pose estimation, tracking, and forecasting.\n", "date": "2022-07-09", "categories": []}
{"id": "2303.17368", "title": "SynBody: Synthetic Dataset with Layered Human Models for 3D Human\n  Perception and Modeling", "abstract": "  Synthetic data has emerged as a promising source for 3D human research as it\noffers low-cost access to large-scale human datasets. To advance the diversity\nand annotation quality of human models, we introduce a new synthetic dataset,\nSynBody, with three appealing features: 1) a clothed parametric human model\nthat can generate a diverse range of subjects; 2) the layered human\nrepresentation that naturally offers high-quality 3D annotations to support\nmultiple tasks; 3) a scalable system for producing realistic data to facilitate\nreal-world tasks. The dataset comprises 1.2M images with corresponding accurate\n3D annotations, covering 10,000 human body models, 1,187 actions, and various\nviewpoints. The dataset includes two subsets for human pose and shape\nestimation as well as human neural rendering. Extensive experiments on SynBody\nindicate that it substantially enhances both SMPL and SMPL-X estimation.\nFurthermore, the incorporation of layered annotations offers a valuable\ntraining resource for investigating the Human Neural Radiance Fields (NeRF).\n", "date": "2023-03-30", "categories": []}
{"id": "2309.04756", "title": "Probabilistic Triangulation for Uncalibrated Multi-View 3D Human Pose\n  Estimation", "abstract": "  3D human pose estimation has been a long-standing challenge in computer\nvision and graphics, where multi-view methods have significantly progressed but\nare limited by the tedious calibration processes. Existing multi-view methods\nare restricted to fixed camera pose and therefore lack generalization ability.\nThis paper presents a novel Probabilistic Triangulation module that can be\nembedded in a calibrated 3D human pose estimation method, generalizing it to\nuncalibration scenes. The key idea is to use a probability distribution to\nmodel the camera pose and iteratively update the distribution from 2D features\ninstead of using camera pose. Specifically, We maintain a camera pose\ndistribution and then iteratively update this distribution by computing the\nposterior probability of the camera pose through Monte Carlo sampling. This\nway, the gradients can be directly back-propagated from the 3D pose estimation\nto the 2D heatmap, enabling end-to-end training. Extensive experiments on\nHuman3.6M and CMU Panoptic demonstrate that our method outperforms other\nuncalibration methods and achieves comparable results with state-of-the-art\ncalibration methods. Thus, our method achieves a trade-off between estimation\naccuracy and generalizability. Our code is in\nhttps://github.com/bymaths/probabilistic_triangulation\n", "date": "2023-09-09", "categories": []}
{"id": "2309.04840", "title": "AnyPose: Anytime 3D Human Pose Forecasting via Neural Ordinary\n  Differential Equations", "abstract": "  Anytime 3D human pose forecasting is crucial to synchronous real-world\nhuman-machine interaction, where the term ``anytime\" corresponds to predicting\nhuman pose at any real-valued time step. However, to the best of our knowledge,\nall the existing methods in human pose forecasting perform predictions at\npreset, discrete time intervals. Therefore, we introduce AnyPose, a lightweight\ncontinuous-time neural architecture that models human behavior dynamics with\nneural ordinary differential equations. We validate our framework on the\nHuman3.6M, AMASS, and 3DPW dataset and conduct a series of comprehensive\nanalyses towards comparison with existing methods and the intersection of human\npose and neural ordinary differential equations. Our results demonstrate that\nAnyPose exhibits high-performance accuracy in predicting future poses and takes\nsignificantly lower computational time than traditional methods in solving\nanytime prediction tasks.\n", "date": "2023-09-09", "categories": []}
{"id": "2211.15692", "title": "H3WB: Human3.6M 3D WholeBody Dataset and Benchmark", "abstract": "  We present a benchmark for 3D human whole-body pose estimation, which\ninvolves identifying accurate 3D keypoints on the entire human body, including\nface, hands, body, and feet. Currently, the lack of a fully annotated and\naccurate 3D whole-body dataset results in deep networks being trained\nseparately on specific body parts, which are combined during inference. Or they\nrely on pseudo-groundtruth provided by parametric body models which are not as\naccurate as detection based methods. To overcome these issues, we introduce the\nHuman3.6M 3D WholeBody (H3WB) dataset, which provides whole-body annotations\nfor the Human3.6M dataset using the COCO Wholebody layout. H3WB comprises 133\nwhole-body keypoint annotations on 100K images, made possible by our new\nmulti-view pipeline. We also propose three tasks: i) 3D whole-body pose lifting\nfrom 2D complete whole-body pose, ii) 3D whole-body pose lifting from 2D\nincomplete whole-body pose, and iii) 3D whole-body pose estimation from a\nsingle RGB image. Additionally, we report several baselines from popular\nmethods for these tasks. Furthermore, we also provide automated 3D whole-body\nannotations of TotalCapture and experimentally show that when used with H3WB it\nhelps to improve the performance. Code and dataset is available at\nhttps://github.com/wholebody3d/wholebody3d\n", "date": "2022-11-28", "categories": []}
{"id": "2309.01010", "title": "Mitigating Motion Blur for Robust 3D Baseball Player Pose Modeling for\n  Pitch Analysis", "abstract": "  Using videos to analyze pitchers in baseball can play a vital role in\nstrategizing and injury prevention. Computer vision-based pose analysis offers\na time-efficient and cost-effective approach. However, the use of accessible\nbroadcast videos, with a 30fps framerate, often results in partial body motion\nblur during fast actions, limiting the performance of existing pose keypoint\nestimation models. Previous works have primarily relied on fixed backgrounds,\nassuming minimal motion differences between frames, or utilized multiview data\nto address this problem. To this end, we propose a synthetic data augmentation\npipeline to enhance the model's capability to deal with the pitcher's blurry\nactions. In addition, we leverage in-the-wild videos to make our model robust\nunder different real-world conditions and camera positions. By carefully\noptimizing the augmentation parameters, we observed a notable reduction in the\nloss by 54.2% and 36.2% on the test dataset for 2D and 3D pose estimation\nrespectively. By applying our approach to existing state-of-the-art pose\nestimators, we demonstrate an average improvement of 29.2%. The findings\nhighlight the effectiveness of our method in mitigating the challenges posed by\nmotion blur, thereby enhancing the overall quality of pose estimation.\n", "date": "2023-09-02", "categories": []}
{"id": "2309.01575", "title": "DiffHPE: Robust, Coherent 3D Human Pose Lifting with Diffusion", "abstract": "  We present an innovative approach to 3D Human Pose Estimation (3D-HPE) by\nintegrating cutting-edge diffusion models, which have revolutionized diverse\nfields, but are relatively unexplored in 3D-HPE. We show that diffusion models\nenhance the accuracy, robustness, and coherence of human pose estimations. We\nintroduce DiffHPE, a novel strategy for harnessing diffusion models in 3D-HPE,\nand demonstrate its ability to refine standard supervised 3D-HPE. We also show\nhow diffusion models lead to more robust estimations in the face of occlusions,\nand improve the time-coherence and the sagittal symmetry of predictions. Using\nthe Human\\,3.6M dataset, we illustrate the effectiveness of our approach and\nits superiority over existing models, even under adverse situations where the\nocclusion patterns in training do not match those in inference. Our findings\nindicate that while standalone diffusion models provide commendable\nperformance, their accuracy is even better in combination with supervised\nmodels, opening exciting new avenues for 3D-HPE research.\n", "date": "2023-09-04", "categories": []}
{"id": "2309.00310", "title": "Fusing Monocular Images and Sparse IMU Signals for Real-time Human\n  Motion Capture", "abstract": "  Either RGB images or inertial signals have been used for the task of motion\ncapture (mocap), but combining them together is a new and interesting topic. We\nbelieve that the combination is complementary and able to solve the inherent\ndifficulties of using one modality input, including occlusions, extreme\nlighting/texture, and out-of-view for visual mocap and global drifts for\ninertial mocap. To this end, we propose a method that fuses monocular images\nand sparse IMUs for real-time human motion capture. Our method contains a dual\ncoordinate strategy to fully explore the IMU signals with different goals in\nmotion capture. To be specific, besides one branch transforming the IMU signals\nto the camera coordinate system to combine with the image information, there is\nanother branch to learn from the IMU signals in the body root coordinate system\nto better estimate body poses. Furthermore, a hidden state feedback mechanism\nis proposed for both two branches to compensate for their own drawbacks in\nextreme input cases. Thus our method can easily switch between the two kinds of\nsignals or combine them in different cases to achieve a robust mocap. %The two\ndivided parts can help each other for better mocap results under different\nconditions. Quantitative and qualitative results demonstrate that by delicately\ndesigning the fusion method, our technique significantly outperforms the\nstate-of-the-art vision, IMU, and combined methods on both global orientation\nand local pose estimation. Our codes are available for research at\nhttps://shaohua-pan.github.io/robustcap-page/.\n", "date": "2023-09-01", "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": "2308.16894", "title": "EMDB: The Electromagnetic Database of Global 3D Human Pose and Shape in\n  the Wild", "abstract": "  We present EMDB, the Electromagnetic Database of Global 3D Human Pose and\nShape in the Wild. EMDB is a novel dataset that contains high-quality 3D SMPL\npose and shape parameters with global body and camera trajectories for\nin-the-wild videos. We use body-worn, wireless electromagnetic (EM) sensors and\na hand-held iPhone to record a total of 58 minutes of motion data, distributed\nover 81 indoor and outdoor sequences and 10 participants. Together with\naccurate body poses and shapes, we also provide global camera poses and body\nroot trajectories. To construct EMDB, we propose a multi-stage optimization\nprocedure, which first fits SMPL to the 6-DoF EM measurements and then refines\nthe poses via image observations. To achieve high-quality results, we leverage\na neural implicit avatar model to reconstruct detailed human surface geometry\nand appearance, which allows for improved alignment and smoothness via a dense\npixel-level objective. Our evaluations, conducted with a multi-view volumetric\ncapture system, indicate that EMDB has an expected accuracy of 2.3 cm\npositional and 10.6 degrees angular error, surpassing the accuracy of previous\nin-the-wild datasets. We evaluate existing state-of-the-art monocular RGB\nmethods for camera-relative and global pose estimation on EMDB. EMDB is\npublicly available under https://ait.ethz.ch/emdb\n", "date": "2023-08-31", "categories": []}
{"id": "2210.11826", "title": "3D Human Pose Estimation in Multi-View Operating Room Videos Using\n  Differentiable Camera Projections", "abstract": "  3D human pose estimation in multi-view operating room (OR) videos is a\nrelevant asset for person tracking and action recognition. However, the\nsurgical environment makes it challenging to find poses due to sterile\nclothing, frequent occlusions, and limited public data. Methods specifically\ndesigned for the OR are generally based on the fusion of detected poses in\nmultiple camera views. Typically, a 2D pose estimator such as a convolutional\nneural network (CNN) detects joint locations. Then, the detected joint\nlocations are projected to 3D and fused over all camera views. However,\naccurate detection in 2D does not guarantee accurate localisation in 3D space.\nIn this work, we propose to directly optimise for localisation in 3D by\ntraining 2D CNNs end-to-end based on a 3D loss that is backpropagated through\neach camera's projection parameters. Using videos from the MVOR dataset, we\nshow that this end-to-end approach outperforms optimisation in 2D space.\n", "date": "2022-10-21", "categories": []}
{"id": "2308.15313", "title": "Spatio-temporal MLP-graph network for 3D human pose estimation", "abstract": "  Graph convolutional networks and their variants have shown significant\npromise in 3D human pose estimation. Despite their success, most of these\nmethods only consider spatial correlations between body joints and do not take\ninto account temporal correlations, thereby limiting their ability to capture\nrelationships in the presence of occlusions and inherent ambiguity. To address\nthis potential weakness, we propose a spatio-temporal network architecture\ncomposed of a joint-mixing multi-layer perceptron block that facilitates\ncommunication among different joints and a graph weighted Jacobi network block\nthat enables communication among various feature channels. The major novelty of\nour approach lies in a new weighted Jacobi feature propagation rule obtained\nthrough graph filtering with implicit fairing. We leverage temporal information\nfrom the 2D pose sequences, and integrate weight modulation into the model to\nenable untangling of the feature transformations of distinct nodes. We also\nemploy adjacency modulation with the aim of learning meaningful correlations\nbeyond defined linkages between body joints by altering the graph topology\nthrough a learnable modulation matrix. Extensive experiments on two benchmark\ndatasets demonstrate the effectiveness of our model, outperforming recent\nstate-of-the-art methods for 3D human pose estimation.\n", "date": "2023-08-29", "categories": []}
{"id": "2212.04636", "title": "Ego-Body Pose Estimation via Ego-Head Pose Estimation", "abstract": "  Estimating 3D human motion from an egocentric video sequence plays a critical\nrole in human behavior understanding and has various applications in VR/AR.\nHowever, naively learning a mapping between egocentric videos and human motions\nis challenging, because the user's body is often unobserved by the front-facing\ncamera placed on the head of the user. In addition, collecting large-scale,\nhigh-quality datasets with paired egocentric videos and 3D human motions\nrequires accurate motion capture devices, which often limit the variety of\nscenes in the videos to lab-like environments. To eliminate the need for paired\negocentric video and human motions, we propose a new method, Ego-Body Pose\nEstimation via Ego-Head Pose Estimation (EgoEgo), which decomposes the problem\ninto two stages, connected by the head motion as an intermediate\nrepresentation. EgoEgo first integrates SLAM and a learning approach to\nestimate accurate head motion. Subsequently, leveraging the estimated head pose\nas input, EgoEgo utilizes conditional diffusion to generate multiple plausible\nfull-body motions. This disentanglement of head and body pose eliminates the\nneed for training datasets with paired egocentric videos and 3D human motion,\nenabling us to leverage large-scale egocentric video datasets and motion\ncapture datasets separately. Moreover, for systematic benchmarking, we develop\na synthetic dataset, AMASS-Replica-Ego-Syn (ARES), with paired egocentric\nvideos and human motion. On both ARES and real data, our EgoEgo model performs\nsignificantly better than the current state-of-the-art methods.\n", "date": "2022-12-09", "categories": []}
{"id": "2308.13872", "title": "Vision-Based Human Pose Estimation via Deep Learning: A Survey", "abstract": "  Human pose estimation (HPE) has attracted a significant amount of attention\nfrom the computer vision community in the past decades. Moreover, HPE has been\napplied to various domains, such as human-computer interaction, sports\nanalysis, and human tracking via images and videos. Recently, deep\nlearning-based approaches have shown state-of-the-art performance in HPE-based\napplications. Although deep learning-based approaches have achieved remarkable\nperformance in HPE, a comprehensive review of deep learning-based HPE methods\nremains lacking in the literature. In this article, we provide an up-to-date\nand in-depth overview of the deep learning approaches in vision-based HPE. We\nsummarize these methods of 2-D and 3-D HPE, and their applications, discuss the\nchallenges and the research trends through bibliometrics, and provide\ninsightful recommendations for future research. This article provides a\nmeaningful overview as introductory material for beginners to deep\nlearning-based HPE, as well as supplementary material for advanced researchers.\n", "date": "2023-08-26", "categories": []}
{"id": "2308.14492", "title": "PointHPS: Cascaded 3D Human Pose and Shape Estimation from Point Clouds", "abstract": "  Human pose and shape estimation (HPS) has attracted increasing attention in\nrecent years. While most existing studies focus on HPS from 2D images or videos\nwith inherent depth ambiguity, there are surging need to investigate HPS from\n3D point clouds as depth sensors have been frequently employed in commercial\ndevices. However, real-world sensory 3D points are usually noisy and\nincomplete, and also human bodies could have different poses of high diversity.\nTo tackle these challenges, we propose a principled framework, PointHPS, for\naccurate 3D HPS from point clouds captured in real-world settings, which\niteratively refines point features through a cascaded architecture.\nSpecifically, each stage of PointHPS performs a series of downsampling and\nupsampling operations to extract and collate both local and global cues, which\nare further enhanced by two novel modules: 1) Cross-stage Feature Fusion (CFF)\nfor multi-scale feature propagation that allows information to flow effectively\nthrough the stages, and 2) Intermediate Feature Enhancement (IFE) for\nbody-aware feature aggregation that improves feature quality after each stage.\nTo facilitate a comprehensive study under various scenarios, we conduct our\nexperiments on two large-scale benchmarks, comprising i) a dataset that\nfeatures diverse subjects and actions captured by real commercial sensors in a\nlaboratory environment, and ii) controlled synthetic data generated with\nrealistic considerations such as clothed humans in crowded outdoor scenes.\nExtensive experiments demonstrate that PointHPS, with its powerful point\nfeature extraction and processing scheme, outperforms State-of-the-Art methods\nby significant margins across the board. Homepage:\nhttps://caizhongang.github.io/projects/PointHPS/.\n", "date": "2023-08-28", "categories": []}
{"id": "2307.15880", "title": "Effective Whole-body Pose Estimation with Two-stages Distillation", "abstract": "  Whole-body pose estimation localizes the human body, hand, face, and foot\nkeypoints in an image. This task is challenging due to multi-scale body parts,\nfine-grained localization for low-resolution regions, and data scarcity.\nMeanwhile, applying a highly efficient and accurate pose estimator to widely\nhuman-centric understanding and generation tasks is urgent. In this work, we\npresent a two-stage pose \\textbf{D}istillation for \\textbf{W}hole-body\n\\textbf{P}ose estimators, named \\textbf{DWPose}, to improve their effectiveness\nand efficiency. The first-stage distillation designs a weight-decay strategy\nwhile utilizing a teacher's intermediate feature and final logits with both\nvisible and invisible keypoints to supervise the student from scratch. The\nsecond stage distills the student model itself to further improve performance.\nDifferent from the previous self-knowledge distillation, this stage finetunes\nthe student's head with only 20% training time as a plug-and-play training\nstrategy. For data limitations, we explore the UBody dataset that contains\ndiverse facial expressions and hand gestures for real-life applications.\nComprehensive experiments show the superiority of our proposed simple yet\neffective methods. We achieve new state-of-the-art performance on\nCOCO-WholeBody, significantly boosting the whole-body AP of RTMPose-l from\n64.8% to 66.5%, even surpassing RTMPose-x teacher with 65.3% AP. We release a\nseries of models with different sizes, from tiny to large, for satisfying\nvarious downstream tasks. Our codes and models are available at\nhttps://github.com/IDEA-Research/DWPose.\n", "date": "2023-07-29", "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": "2308.12965", "title": "POCO: 3D Pose and Shape Estimation with Confidence", "abstract": "  The regression of 3D Human Pose and Shape (HPS) from an image is becoming\nincreasingly accurate. This makes the results useful for downstream tasks like\nhuman action recognition or 3D graphics. Yet, no regressor is perfect, and\naccuracy can be affected by ambiguous image evidence or by poses and appearance\nthat are unseen during training. Most current HPS regressors, however, do not\nreport the confidence of their outputs, meaning that downstream tasks cannot\ndifferentiate accurate estimates from inaccurate ones. To address this, we\ndevelop POCO, a novel framework for training HPS regressors to estimate not\nonly a 3D human body, but also their confidence, in a single feed-forward pass.\nSpecifically, POCO estimates both the 3D body pose and a per-sample variance.\nThe key idea is to introduce a Dual Conditioning Strategy (DCS) for regressing\nuncertainty that is highly correlated to pose reconstruction quality. The POCO\nframework can be applied to any HPS regressor and here we evaluate it by\nmodifying HMR, PARE, and CLIFF. In all cases, training the network to reason\nabout uncertainty helps it learn to more accurately estimate 3D pose. While\nthis was not our goal, the improvement is modest but consistent. Our main\nmotivation is to provide uncertainty estimates for downstream tasks; we\ndemonstrate this in two ways: (1) We use the confidence estimates to bootstrap\nHPS training. Given unlabelled image data, we take the confident estimates of a\nPOCO-trained regressor as pseudo ground truth. Retraining with this\nautomatically-curated data improves accuracy. (2) We exploit uncertainty in\nvideo pose estimation by automatically identifying uncertain frames (e.g. due\nto occlusion) and inpainting these from confident frames. Code and models will\nbe available for research at https://poco.is.tue.mpg.de.\n", "date": "2023-08-24", "categories": []}
{"id": "2303.11579", "title": "Diffusion-Based 3D Human Pose Estimation with Multi-Hypothesis\n  Aggregation", "abstract": "  In this paper, a novel Diffusion-based 3D Pose estimation (D3DP) method with\nJoint-wise reProjection-based Multi-hypothesis Aggregation (JPMA) is proposed\nfor probabilistic 3D human pose estimation. On the one hand, D3DP generates\nmultiple possible 3D pose hypotheses for a single 2D observation. It gradually\ndiffuses the ground truth 3D poses to a random distribution, and learns a\ndenoiser conditioned on 2D keypoints to recover the uncontaminated 3D poses.\nThe proposed D3DP is compatible with existing 3D pose estimators and supports\nusers to balance efficiency and accuracy during inference through two\ncustomizable parameters. On the other hand, JPMA is proposed to assemble\nmultiple hypotheses generated by D3DP into a single 3D pose for practical use.\nIt reprojects 3D pose hypotheses to the 2D camera plane, selects the best\nhypothesis joint-by-joint based on the reprojection errors, and combines the\nselected joints into the final pose. The proposed JPMA conducts aggregation at\nthe joint level and makes use of the 2D prior information, both of which have\nbeen overlooked by previous approaches. Extensive experiments on Human3.6M and\nMPI-INF-3DHP datasets show that our method outperforms the state-of-the-art\ndeterministic and probabilistic approaches by 1.5% and 8.9%, respectively. Code\nis available at https://github.com/paTRICK-swk/D3DP.\n", "date": "2023-03-21", "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": "2305.16487", "title": "EgoHumans: An Egocentric 3D Multi-Human Benchmark", "abstract": "  We present EgoHumans, a new multi-view multi-human video benchmark to advance\nthe state-of-the-art of egocentric human 3D pose estimation and tracking.\nExisting egocentric benchmarks either capture single subject or indoor-only\nscenarios, which limit the generalization of computer vision algorithms for\nreal-world applications. We propose a novel 3D capture setup to construct a\ncomprehensive egocentric multi-human benchmark in the wild with annotations to\nsupport diverse tasks such as human detection, tracking, 2D/3D pose estimation,\nand mesh recovery. We leverage consumer-grade wearable camera-equipped glasses\nfor the egocentric view, which enables us to capture dynamic activities like\nplaying tennis, fencing, volleyball, etc. Furthermore, our multi-view setup\ngenerates accurate 3D ground truth even under severe or complete occlusion. The\ndataset consists of more than 125k egocentric images, spanning diverse scenes\nwith a particular focus on challenging and unchoreographed multi-human\nactivities and fast-moving egocentric views. We rigorously evaluate existing\nstate-of-the-art methods and highlight their limitations in the egocentric\nscenario, specifically on multi-human tracking. To address such limitations, we\npropose EgoFormer, a novel approach with a multi-stream transformer\narchitecture and explicit 3D spatial reasoning to estimate and track the human\npose. EgoFormer significantly outperforms prior art by 13.6% IDF1 on the\nEgoHumans dataset.\n", "date": "2023-05-25", "categories": []}
{"id": "2308.10123", "title": "3D-Aware Neural Body Fitting for Occlusion Robust 3D Human Pose\n  Estimation", "abstract": "  Regression-based methods for 3D human pose estimation directly predict the 3D\npose parameters from a 2D image using deep networks. While achieving\nstate-of-the-art performance on standard benchmarks, their performance degrades\nunder occlusion. In contrast, optimization-based methods fit a parametric body\nmodel to 2D features in an iterative manner. The localized reconstruction loss\ncan potentially make them robust to occlusion, but they suffer from the 2D-3D\nambiguity.\n  Motivated by the recent success of generative models in rigid object pose\nestimation, we propose 3D-aware Neural Body Fitting (3DNBF) - an approximate\nanalysis-by-synthesis approach to 3D human pose estimation with SOTA\nperformance and occlusion robustness. In particular, we propose a generative\nmodel of deep features based on a volumetric human representation with Gaussian\nellipsoidal kernels emitting 3D pose-dependent feature vectors. The neural\nfeatures are trained with contrastive learning to become 3D-aware and hence to\novercome the 2D-3D ambiguity.\n  Experiments show that 3DNBF outperforms other approaches on both occluded and\nstandard benchmarks. Code is available at https://github.com/edz-o/3DNBF\n", "date": "2023-08-19", "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": "2308.10705", "title": "Unsupervised 3D Pose Estimation with Non-Rigid Structure-from-Motion\n  Modeling", "abstract": "  Most of the previous 3D human pose estimation work relied on the powerful\nmemory capability of the network to obtain suitable 2D-3D mappings from the\ntraining data. Few works have studied the modeling of human posture deformation\nin motion. In this paper, we propose a new modeling method for human pose\ndeformations and design an accompanying diffusion-based motion prior. Inspired\nby the field of non-rigid structure-from-motion, we divide the task of\nreconstructing 3D human skeletons in motion into the estimation of a 3D\nreference skeleton, and a frame-by-frame skeleton deformation. A mixed\nspatial-temporal NRSfMformer is used to simultaneously estimate the 3D\nreference skeleton and the skeleton deformation of each frame from 2D\nobservations sequence, and then sum them to obtain the pose of each frame.\nSubsequently, a loss term based on the diffusion model is used to ensure that\nthe pipeline learns the correct prior motion knowledge. Finally, we have\nevaluated our proposed method on mainstream datasets and obtained superior\nresults outperforming the state-of-the-art.\n", "date": "2023-08-18", "categories": []}
{"id": "2303.04991", "title": "Deformer: Dynamic Fusion Transformer for Robust Hand Pose Estimation", "abstract": "  Accurately estimating 3D hand pose is crucial for understanding how humans\ninteract with the world. Despite remarkable progress, existing methods often\nstruggle to generate plausible hand poses when the hand is heavily occluded or\nblurred. In videos, the movements of the hand allow us to observe various parts\nof the hand that may be occluded or blurred in a single frame. To adaptively\nleverage the visual clue before and after the occlusion or blurring for robust\nhand pose estimation, we propose the Deformer: a framework that implicitly\nreasons about the relationship between hand parts within the same image\n(spatial dimension) and different timesteps (temporal dimension). We show that\na naive application of the transformer self-attention mechanism is not\nsufficient because motion blur or occlusions in certain frames can lead to\nheavily distorted hand features and generate imprecise keys and queries. To\naddress this challenge, we incorporate a Dynamic Fusion Module into Deformer,\nwhich predicts the deformation of the hand and warps the hand mesh predictions\nfrom nearby frames to explicitly support the current frame estimation.\nFurthermore, we have observed that errors are unevenly distributed across\ndifferent hand parts, with vertices around fingertips having disproportionately\nhigher errors than those around the palm. We mitigate this issue by introducing\na new loss function called maxMSE that automatically adjusts the weight of\nevery vertex to focus the model on critical hand parts. Extensive experiments\nshow that our method significantly outperforms state-of-the-art methods by 10%,\nand is more robust to occlusions (over 14%).\n", "date": "2023-03-09", "categories": []}
{"id": "2303.16456", "title": "Global Adaptation meets Local Generalization: Unsupervised Domain\n  Adaptation for 3D Human Pose Estimation", "abstract": "  When applying a pre-trained 2D-to-3D human pose lifting model to a target\nunseen dataset, large performance degradation is commonly encountered due to\ndomain shift issues. We observe that the degradation is caused by two factors:\n1) the large distribution gap over global positions of poses between the source\nand target datasets due to variant camera parameters and settings, and 2) the\ndeficient diversity of local structures of poses in training. To this end, we\ncombine \\textbf{global adaptation} and \\textbf{local generalization} in\n\\textit{PoseDA}, a simple yet effective framework of unsupervised domain\nadaptation for 3D human pose estimation. Specifically, global adaptation aims\nto align global positions of poses from the source domain to the target domain\nwith a proposed global position alignment (GPA) module. And local\ngeneralization is designed to enhance the diversity of 2D-3D pose mapping with\na local pose augmentation (LPA) module. These modules bring significant\nperformance improvement without introducing additional learnable parameters. In\naddition, we propose local pose augmentation (LPA) to enhance the diversity of\n3D poses following an adversarial training scheme consisting of 1) a\naugmentation generator that generates the parameters of pre-defined pose\ntransformations and 2) an anchor discriminator to ensure the reality and\nquality of the augmented data. Our approach can be applicable to almost all\n2D-3D lifting models. \\textit{PoseDA} achieves 61.3 mm of MPJPE on MPI-INF-3DHP\nunder a cross-dataset evaluation setup, improving upon the previous\nstate-of-the-art method by 10.2\\%.\n", "date": "2023-03-29", "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": "2308.02963", "title": "Generative Approach for Probabilistic Human Mesh Recovery using\n  Diffusion Models", "abstract": "  This work focuses on the problem of reconstructing a 3D human body mesh from\na given 2D image. Despite the inherent ambiguity of the task of human mesh\nrecovery, most existing works have adopted a method of regressing a single\noutput. In contrast, we propose a generative approach framework, called\n\"Diffusion-based Human Mesh Recovery (Diff-HMR)\" that takes advantage of the\ndenoising diffusion process to account for multiple plausible outcomes. During\nthe training phase, the SMPL parameters are diffused from ground-truth\nparameters to random distribution, and Diff-HMR learns the reverse process of\nthis diffusion. In the inference phase, the model progressively refines the\ngiven random SMPL parameters into the corresponding parameters that align with\nthe input image. Diff-HMR, being a generative approach, is capable of\ngenerating diverse results for the same input image as the input noise varies.\nWe conduct validation experiments, and the results demonstrate that the\nproposed framework effectively models the inherent ambiguity of the task of\nhuman mesh recovery in a probabilistic manner. The code is available at\nhttps://github.com/hanbyel0105/Diff-HMR\n", "date": "2023-08-05", "categories": []}
{"id": "2308.09523", "title": "Denoising Diffusion for 3D Hand Pose Estimation from Images", "abstract": "  Hand pose estimation from a single image has many applications. However,\napproaches to full 3D body pose estimation are typically trained on day-to-day\nactivities or actions. As such, detailed hand-to-hand interactions are poorly\nrepresented, especially during motion. We see this in the failure cases of\ntechniques such as OpenPose or MediaPipe. However, accurate hand pose\nestimation is crucial for many applications where the global body motion is\nless important than accurate hand pose estimation.\n  This paper addresses the problem of 3D hand pose estimation from monocular\nimages or sequences. We present a novel end-to-end framework for 3D hand\nregression that employs diffusion models that have shown excellent ability to\ncapture the distribution of data for generative purposes. Moreover, we enforce\nkinematic constraints to ensure realistic poses are generated by incorporating\nan explicit forward kinematic layer as part of the network. The proposed model\nprovides state-of-the-art performance when lifting a 2D single-hand image to\n3D. However, when sequence data is available, we add a Transformer module over\na temporal window of consecutive frames to refine the results, overcoming\njittering and further increasing accuracy.\n  The method is quantitatively and qualitatively evaluated showing\nstate-of-the-art robustness, generalization, and accuracy on several different\ndatasets.\n", "date": "2023-08-18", "categories": []}
{"id": "2308.09525", "title": "Improving 3D Pose Estimation for Sign Language", "abstract": "  This work addresses 3D human pose reconstruction in single images. We present\na method that combines Forward Kinematics (FK) with neural networks to ensure a\nfast and valid prediction of 3D pose. Pose is represented as a hierarchical\ntree/graph with nodes corresponding to human joints that model their physical\nlimits. Given a 2D detection of keypoints in the image, we lift the skeleton to\n3D using neural networks to predict both the joint rotations and bone lengths.\nThese predictions are then combined with skeletal constraints using an FK layer\nimplemented as a network layer in PyTorch. The result is a fast and accurate\napproach to the estimation of 3D skeletal pose. Through quantitative and\nqualitative evaluation, we demonstrate the method is significantly more\naccurate than MediaPipe in terms of both per joint positional error and visual\nappearance. Furthermore, we demonstrate generalization over different datasets.\nThe implementation in PyTorch runs at between 100-200 milliseconds per image\n(including CNN detection) using CPU only.\n", "date": "2023-08-18", "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": "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": "2308.05298", "title": "Double-chain Constraints for 3D Human Pose Estimation in Images and\n  Videos", "abstract": "  Reconstructing 3D poses from 2D poses lacking depth information is\nparticularly challenging due to the complexity and diversity of human motion.\nThe key is to effectively model the spatial constraints between joints to\nleverage their inherent dependencies. Thus, we propose a novel model, called\nDouble-chain Graph Convolutional Transformer (DC-GCT), to constrain the pose\nthrough a double-chain design consisting of local-to-global and global-to-local\nchains to obtain a complex representation more suitable for the current human\npose. Specifically, we combine the advantages of GCN and Transformer and design\na Local Constraint Module (LCM) based on GCN and a Global Constraint Module\n(GCM) based on self-attention mechanism as well as a Feature Interaction Module\n(FIM). The proposed method fully captures the multi-level dependencies between\nhuman body joints to optimize the modeling capability of the model. Moreover,\nwe propose a method to use temporal information into the single-frame model by\nguiding the video sequence embedding through the joint embedding of the target\nframe, with negligible increase in computational cost. Experimental results\ndemonstrate that DC-GCT achieves state-of-the-art performance on two\nchallenging datasets (Human3.6M and MPI-INF-3DHP). Notably, our model achieves\nstate-of-the-art performance on all action categories in the Human3.6M dataset\nusing detected 2D poses from CPN, and our code is available at:\nhttps://github.com/KHB1698/DC-GCT.\n", "date": "2023-08-10", "categories": []}
{"id": "2305.05785", "title": "Regular Splitting Graph Network for 3D Human Pose Estimation", "abstract": "  In human pose estimation methods based on graph convolutional architectures,\nthe human skeleton is usually modeled as an undirected graph whose nodes are\nbody joints and edges are connections between neighboring joints. However, most\nof these methods tend to focus on learning relationships between body joints of\nthe skeleton using first-order neighbors, ignoring higher-order neighbors and\nhence limiting their ability to exploit relationships between distant joints.\nIn this paper, we introduce a higher-order regular splitting graph network\n(RS-Net) for 2D-to-3D human pose estimation using matrix splitting in\nconjunction with weight and adjacency modulation. The core idea is to capture\nlong-range dependencies between body joints using multi-hop neighborhoods and\nalso to learn different modulation vectors for different body joints as well as\na modulation matrix added to the adjacency matrix associated to the skeleton.\nThis learnable modulation matrix helps adjust the graph structure by adding\nextra graph edges in an effort to learn additional connections between body\njoints. Instead of using a shared weight matrix for all neighboring body\njoints, the proposed RS-Net model applies weight unsharing before aggregating\nthe feature vectors associated to the joints in order to capture the different\nrelations between them. Experiments and ablations studies performed on two\nbenchmark datasets demonstrate the effectiveness of our model, achieving\nsuperior performance over recent state-of-the-art methods for 3D human pose\nestimation.\n", "date": "2023-05-09", "categories": []}
{"id": "2307.16074", "title": "Iterative Graph Filtering Network for 3D Human Pose Estimation", "abstract": "  Graph convolutional networks (GCNs) have proven to be an effective approach\nfor 3D human pose estimation. By naturally modeling the skeleton structure of\nthe human body as a graph, GCNs are able to capture the spatial relationships\nbetween joints and learn an efficient representation of the underlying pose.\nHowever, most GCN-based methods use a shared weight matrix, making it\nchallenging to accurately capture the different and complex relationships\nbetween joints. In this paper, we introduce an iterative graph filtering\nframework for 3D human pose estimation, which aims to predict the 3D joint\npositions given a set of 2D joint locations in images. Our approach builds upon\nthe idea of iteratively solving graph filtering with Laplacian regularization\nvia the Gauss-Seidel iterative method. Motivated by this iterative solution, we\ndesign a Gauss-Seidel network (GS-Net) architecture, which makes use of weight\nand adjacency modulation, skip connection, and a pure convolutional block with\nlayer normalization. Adjacency modulation facilitates the learning of edges\nthat go beyond the inherent connections of body joints, resulting in an\nadjusted graph structure that reflects the human skeleton, while skip\nconnections help maintain crucial information from the input layer's initial\nfeatures as the network depth increases. We evaluate our proposed model on two\nstandard benchmark datasets, and compare it with a comprehensive set of strong\nbaseline methods for 3D human pose estimation. Our experimental results\ndemonstrate that our approach outperforms the baseline methods on both\ndatasets, achieving state-of-the-art performance. Furthermore, we conduct\nablation studies to analyze the contributions of different components of our\nmodel architecture and show that the skip connection and adjacency modulation\nhelp improve the model performance.\n", "date": "2023-07-29", "categories": []}
{"id": "2301.06648", "title": "Neuromorphic High-Frequency 3D Dancing Pose Estimation in Dynamic\n  Environment", "abstract": "  As a beloved sport worldwide, dancing is getting integrated into traditional\nand virtual reality-based gaming platforms nowadays. It opens up new\nopportunities in the technology-mediated dancing space. These platforms\nprimarily rely on passive and continuous human pose estimation as an input\ncapture mechanism. Existing solutions are mainly based on RGB or RGB-Depth\ncameras for dance games. The former suffers in low-lighting conditions due to\nthe motion blur and low sensitivity, while the latter is too power-hungry, has\na low frame rate, and has limited working distance. With ultra-low latency,\nenergy efficiency, and wide dynamic range characteristics, the event camera is\na promising solution to overcome these shortcomings. We propose YeLan, an event\ncamera-based 3-dimensional high-frequency human pose estimation(HPE) system\nthat survives low-lighting conditions and dynamic backgrounds. We collected the\nworld's first event camera dance dataset and developed a fully customizable\nmotion-to-event physics-aware simulator. YeLan outperforms the baseline models\nin these challenging conditions and demonstrated robustness against different\ntypes of clothing, background motion, viewing angle, occlusion, and lighting\nfluctuations.\n", "date": "2023-01-17", "categories": []}
{"id": "2307.16687", "title": "DiffPose: SpatioTemporal Diffusion Model for Video-Based Human Pose\n  Estimation", "abstract": "  Denoising diffusion probabilistic models that were initially proposed for\nrealistic image generation have recently shown success in various perception\ntasks (e.g., object detection and image segmentation) and are increasingly\ngaining attention in computer vision. However, extending such models to\nmulti-frame human pose estimation is non-trivial due to the presence of the\nadditional temporal dimension in videos. More importantly, learning\nrepresentations that focus on keypoint regions is crucial for accurate\nlocalization of human joints. Nevertheless, the adaptation of the\ndiffusion-based methods remains unclear on how to achieve such objective. In\nthis paper, we present DiffPose, a novel diffusion architecture that formulates\nvideo-based human pose estimation as a conditional heatmap generation problem.\nFirst, to better leverage temporal information, we propose SpatioTemporal\nRepresentation Learner which aggregates visual evidences across frames and uses\nthe resulting features in each denoising step as a condition. In addition, we\npresent a mechanism called Lookup-based MultiScale Feature Interaction that\ndetermines the correlations between local joints and global contexts across\nmultiple scales. This mechanism generates delicate representations that focus\non keypoint regions. Altogether, by extending diffusion models, we show two\nunique characteristics from DiffPose on pose estimation task: (i) the ability\nto combine multiple sets of pose estimates to improve prediction accuracy,\nparticularly for challenging joints, and (ii) the ability to adjust the number\nof iterative steps for feature refinement without retraining the model.\nDiffPose sets new state-of-the-art results on three benchmarks: PoseTrack2017,\nPoseTrack2018, and PoseTrack21.\n", "date": "2023-07-31", "categories": []}
{"id": "2308.00628", "title": "Human-M3: A Multi-view Multi-modal Dataset for 3D Human Pose Estimation\n  in Outdoor Scenes", "abstract": "  3D human pose estimation in outdoor environments has garnered increasing\nattention recently. However, prevalent 3D human pose datasets pertaining to\noutdoor scenes lack diversity, as they predominantly utilize only one type of\nmodality (RGB image or pointcloud), and often feature only one individual\nwithin each scene. This limited scope of dataset infrastructure considerably\nhinders the variability of available data. In this article, we propose\nHuman-M3, an outdoor multi-modal multi-view multi-person human pose database\nwhich includes not only multi-view RGB videos of outdoor scenes but also\ncorresponding pointclouds. In order to obtain accurate human poses, we propose\nan algorithm based on multi-modal data input to generate ground truth\nannotation. This benefits from robust pointcloud detection and tracking, which\nsolves the problem of inaccurate human localization and matching ambiguity that\nmay exist in previous multi-view RGB videos in outdoor multi-person scenes, and\ngenerates reliable ground truth annotations. Evaluation of multiple different\nmodalities algorithms has shown that this database is challenging and suitable\nfor future research. Furthermore, we propose a 3D human pose estimation\nalgorithm based on multi-modal data input, which demonstrates the advantages of\nmulti-modal data input for 3D human pose estimation. Code and data will be\nreleased on https://github.com/soullessrobot/Human-M3-Dataset.\n", "date": "2023-08-01", "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": "2212.02796", "title": "DiffuPose: Monocular 3D Human Pose Estimation via Denoising Diffusion\n  Probabilistic Model", "abstract": "  Thanks to the development of 2D keypoint detectors, monocular 3D human pose\nestimation (HPE) via 2D-to-3D uplifting approaches have achieved remarkable\nimprovements. Still, monocular 3D HPE is a challenging problem due to the\ninherent depth ambiguities and occlusions. To handle this problem, many\nprevious works exploit temporal information to mitigate such difficulties.\nHowever, there are many real-world applications where frame sequences are not\naccessible. This paper focuses on reconstructing a 3D pose from a single 2D\nkeypoint detection. Rather than exploiting temporal information, we alleviate\nthe depth ambiguity by generating multiple 3D pose candidates which can be\nmapped to an identical 2D keypoint. We build a novel diffusion-based framework\nto effectively sample diverse 3D poses from an off-the-shelf 2D detector. By\nconsidering the correlation between human joints by replacing the conventional\ndenoising U-Net with graph convolutional network, our approach accomplishes\nfurther performance improvements. We evaluate our method on the widely adopted\nHuman3.6M and HumanEva-I datasets. Comprehensive experiments are conducted to\nprove the efficacy of the proposed method, and they confirm that our model\noutperforms state-of-the-art multi-hypothesis 3D HPE methods.\n", "date": "2022-12-06", "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": "2307.11934", "title": "LAMP: Leveraging Language Prompts for Multi-person Pose Estimation", "abstract": "  Human-centric visual understanding is an important desideratum for effective\nhuman-robot interaction. In order to navigate crowded public places, social\nrobots must be able to interpret the activity of the surrounding humans. This\npaper addresses one key aspect of human-centric visual understanding,\nmulti-person pose estimation. Achieving good performance on multi-person pose\nestimation in crowded scenes is difficult due to the challenges of occluded\njoints and instance separation. In order to tackle these challenges and\novercome the limitations of image features in representing invisible body\nparts, we propose a novel prompt-based pose inference strategy called LAMP\n(Language Assisted Multi-person Pose estimation). By utilizing the text\nrepresentations generated by a well-trained language model (CLIP), LAMP can\nfacilitate the understanding of poses on the instance and joint levels, and\nlearn more robust visual representations that are less susceptible to\nocclusion. This paper demonstrates that language-supervised training boosts the\nperformance of single-stage multi-person pose estimation, and both\ninstance-level and joint-level prompts are valuable for training. The code is\navailable at https://github.com/shengnanh20/LAMP.\n", "date": "2023-07-21", "categories": []}
{"id": "2307.14889", "title": "Weakly Supervised Multi-Modal 3D Human Body Pose Estimation for\n  Autonomous Driving", "abstract": "  Accurate 3D human pose estimation (3D HPE) is crucial for enabling autonomous\nvehicles (AVs) to make informed decisions and respond proactively in critical\nroad scenarios. Promising results of 3D HPE have been gained in several domains\nsuch as human-computer interaction, robotics, sports and medical analytics,\noften based on data collected in well-controlled laboratory environments.\nNevertheless, the transfer of 3D HPE methods to AVs has received limited\nresearch attention, due to the challenges posed by obtaining accurate 3D pose\nannotations and the limited suitability of data from other domains.\n  We present a simple yet efficient weakly supervised approach for 3D HPE in\nthe AV context by employing a high-level sensor fusion between camera and LiDAR\ndata. The weakly supervised setting enables training on the target datasets\nwithout any 2D/3D keypoint labels by using an off-the-shelf 2D joint extractor\nand pseudo labels generated from LiDAR to image projections. Our approach\noutperforms state-of-the-art results by up to $\\sim$ 13% on the Waymo Open\nDataset in the weakly supervised setting and achieves state-of-the-art results\nin the supervised setting.\n", "date": "2023-07-27", "categories": []}
{"id": "2303.18246", "title": "3D Human Pose Estimation via Intuitive Physics", "abstract": "  Estimating 3D humans from images often produces implausible bodies that lean,\nfloat, or penetrate the floor. Such methods ignore the fact that bodies are\ntypically supported by the scene. A physics engine can be used to enforce\nphysical plausibility, but these are not differentiable, rely on unrealistic\nproxy bodies, and are difficult to integrate into existing optimization and\nlearning frameworks. In contrast, we exploit novel intuitive-physics (IP) terms\nthat can be inferred from a 3D SMPL body interacting with the scene. Inspired\nby biomechanics, we infer the pressure heatmap on the body, the Center of\nPressure (CoP) from the heatmap, and the SMPL body's Center of Mass (CoM). With\nthese, we develop IPMAN, to estimate a 3D body from a color image in a \"stable\"\nconfiguration by encouraging plausible floor contact and overlapping CoP and\nCoM. Our IP terms are intuitive, easy to implement, fast to compute,\ndifferentiable, and can be integrated into existing optimization and regression\nmethods. We evaluate IPMAN on standard datasets and MoYo, a new dataset with\nsynchronized multi-view images, ground-truth 3D bodies with complex poses,\nbody-floor contact, CoM and pressure. IPMAN produces more plausible results\nthan the state of the art, improving accuracy for static poses, while not\nhurting dynamic ones. Code and data are available for research at\nhttps://ipman.is.tue.mpg.de.\n", "date": "2023-03-31", "categories": []}
{"id": "2307.05853", "title": "GLA-GCN: Global-local Adaptive Graph Convolutional Network for 3D Human\n  Pose Estimation from Monocular Video", "abstract": "  3D human pose estimation has been researched for decades with promising\nfruits. 3D human pose lifting is one of the promising research directions\ntoward the task where both estimated pose and ground truth pose data are used\nfor training. Existing pose lifting works mainly focus on improving the\nperformance of estimated pose, but they usually underperform when testing on\nthe ground truth pose data. We observe that the performance of the estimated\npose can be easily improved by preparing good quality 2D pose, such as\nfine-tuning the 2D pose or using advanced 2D pose detectors. As such, we\nconcentrate on improving the 3D human pose lifting via ground truth data for\nthe future improvement of more quality estimated pose data. Towards this goal,\na simple yet effective model called Global-local Adaptive Graph Convolutional\nNetwork (GLA-GCN) is proposed in this work. Our GLA-GCN globally models the\nspatiotemporal structure via a graph representation and backtraces local joint\nfeatures for 3D human pose estimation via individually connected layers. To\nvalidate our model design, we conduct extensive experiments on three benchmark\ndatasets: Human3.6M, HumanEva-I, and MPI-INF-3DHP. Experimental results show\nthat our GLA-GCN implemented with ground truth 2D poses significantly\noutperforms state-of-the-art methods (e.g., up to around 3%, 17%, and 14% error\nreductions on Human3.6M, HumanEva-I, and MPI-INF-3DHP, respectively). GitHub:\nhttps://github.com/bruceyo/GLA-GCN.\n", "date": "2023-07-12", "categories": []}
{"id": "2211.12193", "title": "Anatomy-guided domain adaptation for 3D in-bed human pose estimation", "abstract": "  3D human pose estimation is a key component of clinical monitoring systems.\nThe clinical applicability of deep pose estimation models, however, is limited\nby their poor generalization under domain shifts along with their need for\nsufficient labeled training data. As a remedy, we present a novel domain\nadaptation method, adapting a model from a labeled source to a shifted\nunlabeled target domain. Our method comprises two complementary adaptation\nstrategies based on prior knowledge about human anatomy. First, we guide the\nlearning process in the target domain by constraining predictions to the space\nof anatomically plausible poses. To this end, we embed the prior knowledge into\nan anatomical loss function that penalizes asymmetric limb lengths, implausible\nbone lengths, and implausible joint angles. Second, we propose to filter pseudo\nlabels for self-training according to their anatomical plausibility and\nincorporate the concept into the Mean Teacher paradigm. We unify both\nstrategies in a point cloud-based framework applicable to unsupervised and\nsource-free domain adaptation. Evaluation is performed for in-bed pose\nestimation under two adaptation scenarios, using the public SLP dataset and a\nnewly created dataset. Our method consistently outperforms various\nstate-of-the-art domain adaptation methods, surpasses the baseline model by\n31%/66%, and reduces the domain gap by 65%/82%. Source code is available at\nhttps://github.com/multimodallearning/da-3dhpe-anatomy.\n", "date": "2022-11-22", "categories": []}
{"id": "2307.09893", "title": "Learning from Abstract Images: on the Importance of Occlusion in a\n  Minimalist Encoding of Human Poses", "abstract": "  Existing 2D-to-3D pose lifting networks suffer from poor performance in\ncross-dataset benchmarks. Although the use of 2D keypoints joined by\n\"stick-figure\" limbs has shown promise as an intermediate step, stick-figures\ndo not account for occlusion information that is often inherent in an image. In\nthis paper, we propose a novel representation using opaque 3D limbs that\npreserves occlusion information while implicitly encoding joint locations.\nCrucially, when training on data with accurate three-dimensional keypoints and\nwithout part-maps, this representation allows training on abstract synthetic\nimages, with occlusion, from as many synthetic viewpoints as desired. The\nresult is a pose defined by limb angles rather than joint positions\n$\\unicode{x2013}$ because poses are, in the real world, independent of cameras\n$\\unicode{x2013}$ allowing us to predict poses that are completely independent\nof camera viewpoint. The result provides not only an improvement in\nsame-dataset benchmarks, but a \"quantum leap\" in cross-dataset benchmarks.\n", "date": "2023-07-19", "categories": []}
{"id": "2307.09026", "title": "ActionPrompt: Action-Guided 3D Human Pose Estimation With Text and Pose\n  Prompting", "abstract": "  Recent 2D-to-3D human pose estimation (HPE) utilizes temporal consistency\nacross sequences to alleviate the depth ambiguity problem but ignore the action\nrelated prior knowledge hidden in the pose sequence. In this paper, we propose\na plug-and-play module named Action Prompt Module (APM) that effectively mines\ndifferent kinds of action clues for 3D HPE. The highlight is that, the mining\nscheme of APM can be widely adapted to different frameworks and bring\nconsistent benefits. Specifically, we first present a novel Action-related Text\nPrompt module (ATP) that directly embeds action labels and transfers the rich\nlanguage information in the label to the pose sequence. Besides, we further\nintroduce Action-specific Pose Prompt module (APP) to mine the position-aware\npose pattern of each action, and exploit the correlation between the mined\npatterns and input pose sequence for further pose refinement. Experiments show\nthat APM can improve the performance of most video-based 2D-to-3D HPE\nframeworks by a large margin.\n", "date": "2023-07-18", "categories": []}
{"id": "2210.12035", "title": "BlanketGen - A synthetic blanket occlusion augmentation pipeline for\n  MoCap datasets", "abstract": "  Human motion analysis has seen drastic improvements recently, however, due to\nthe lack of representative datasets, for clinical in-bed scenarios it is still\nlagging behind. To address this issue, we implemented BlanketGen, a pipeline\nthat augments videos with synthetic blanket occlusions. With this pipeline, we\ngenerated an augmented version of the pose estimation dataset 3DPW called\nBlanketGen-3DPW. We then used this new dataset to fine-tune a Deep Learning\nmodel to improve its performance in these scenarios with promising results.\nCode and further information are available at\nhttps://gitlab.inesctec.pt/brain-lab/brain-lab-public/blanket-gen-releases.\n", "date": "2022-10-21", "categories": []}
{"id": "2307.05896", "title": "Deep learning-based estimation of whole-body kinematics from multi-view\n  images", "abstract": "  It is necessary to analyze the whole-body kinematics (including joint\nlocations and joint angles) to assess risks of fatal and musculoskeletal\ninjuries in occupational tasks. Human pose estimation has gotten more attention\nin recent years as a method to minimize the errors in determining joint\nlocations. However, the joint angles are not often estimated, nor is the\nquality of joint angle estimation assessed. In this paper, we presented an\nend-to-end approach on direct joint angle estimation from multi-view images.\nOur method leveraged the volumetric pose representation and mapped the rotation\nrepresentation to a continuous space where each rotation was uniquely\nrepresented. We also presented a new kinematic dataset in the domain of\nresidential roofing with a data processing pipeline to generate necessary\nannotations for the supervised training procedure on direct joint angle\nestimation. We achieved a mean angle error of $7.19^\\circ$ on the new Roofing\ndataset and $8.41^\\circ$ on the Human3.6M dataset, paving the way for\nemployment of on-site kinematic analysis using multi-view images.\n", "date": "2023-07-12", "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": "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": "2306.16940", "title": "BEDLAM: A Synthetic Dataset of Bodies Exhibiting Detailed Lifelike\n  Animated Motion", "abstract": "  We show, for the first time, that neural networks trained only on synthetic\ndata achieve state-of-the-art accuracy on the problem of 3D human pose and\nshape (HPS) estimation from real images. Previous synthetic datasets have been\nsmall, unrealistic, or lacked realistic clothing. Achieving sufficient realism\nis non-trivial and we show how to do this for full bodies in motion.\nSpecifically, our BEDLAM dataset contains monocular RGB videos with\nground-truth 3D bodies in SMPL-X format. It includes a diversity of body\nshapes, motions, skin tones, hair, and clothing. The clothing is realistically\nsimulated on the moving bodies using commercial clothing physics simulation. We\nrender varying numbers of people in realistic scenes with varied lighting and\ncamera motions. We then train various HPS regressors using BEDLAM and achieve\nstate-of-the-art accuracy on real-image benchmarks despite training with\nsynthetic data. We use BEDLAM to gain insights into what model design choices\nare important for accuracy. With good synthetic training data, we find that a\nbasic method like HMR approaches the accuracy of the current SOTA method\n(CLIFF). BEDLAM is useful for a variety of tasks and all images, ground truth\nbodies, 3D clothing, support code, and more are available for research\npurposes. Additionally, we provide detailed information about our synthetic\ndata generation pipeline, enabling others to generate their own datasets. See\nthe project page: https://bedlam.is.tue.mpg.de/.\n", "date": "2023-06-29", "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": "2305.01167", "title": "Hybrid model for Single-Stage Multi-Person Pose Estimation", "abstract": "  In general, human pose estimation methods are categorized into two approaches\naccording to their architectures: regression (i.e., heatmap-free) and\nheatmap-based methods. The former one directly estimates precise coordinates of\neach keypoint using convolutional and fully-connected layers. Although this\napproach is able to detect overlapped and dense keypoints, unexpected results\ncan be obtained by non-existent keypoints in a scene. On the other hand, the\nlatter one is able to filter the non-existent ones out by utilizing predicted\nheatmaps for each keypoint. Nevertheless, it suffers from quantization error\nwhen obtaining the keypoint coordinates from its heatmaps. In addition, unlike\nthe regression one, it is difficult to distinguish densely placed keypoints in\nan image. To this end, we propose a hybrid model for single-stage multi-person\npose estimation, named HybridPose, which mutually overcomes each drawback of\nboth approaches by maximizing their strengths. Furthermore, we introduce\nself-correlation loss to inject spatial dependencies between keypoint\ncoordinates and their visibility. Therefore, HybridPose is capable of not only\ndetecting densely placed keypoints, but also filtering the non-existent\nkeypoints in an image. Experimental results demonstrate that proposed\nHybridPose exhibits the keypoints visibility without performance degradation in\nterms of the pose estimation accuracy.\n", "date": "2023-05-02", "categories": []}
{"id": "2306.09615", "title": "EVOPOSE: A Recursive Transformer For 3D Human Pose Estimation With\n  Kinematic Structure Priors", "abstract": "  Transformer is popular in recent 3D human pose estimation, which utilizes\nlong-term modeling to lift 2D keypoints into the 3D space. However, current\ntransformer-based methods do not fully exploit the prior knowledge of the human\nskeleton provided by the kinematic structure. In this paper, we propose a novel\ntransformer-based model EvoPose to introduce the human body prior knowledge for\n3D human pose estimation effectively. Specifically, a Structural Priors\nRepresentation (SPR) module represents human priors as structural features\ncarrying rich body patterns, e.g. joint relationships. The structural features\nare interacted with 2D pose sequences and help the model to achieve more\ninformative spatiotemporal features. Moreover, a Recursive Refinement (RR)\nmodule is applied to refine the 3D pose outputs by utilizing estimated results\nand further injects human priors simultaneously. Extensive experiments\ndemonstrate the effectiveness of EvoPose which achieves a new state of the art\non two most popular benchmarks, Human3.6M and MPI-INF-3DHP.\n", "date": "2023-06-16", "categories": []}
{"id": "2306.04745", "title": "3D Human Keypoints Estimation From Point Clouds in the Wild Without\n  Human Labels", "abstract": "  Training a 3D human keypoint detector from point clouds in a supervised\nmanner requires large volumes of high quality labels. While it is relatively\neasy to capture large amounts of human point clouds, annotating 3D keypoints is\nexpensive, subjective, error prone and especially difficult for long-tail cases\n(pedestrians with rare poses, scooterists, etc.). In this work, we propose\nGC-KPL - Geometry Consistency inspired Key Point Leaning, an approach for\nlearning 3D human joint locations from point clouds without human labels. We\nachieve this by our novel unsupervised loss formulations that account for the\nstructure and movement of the human body. We show that by training on a large\ntraining set from Waymo Open Dataset without any human annotated keypoints, we\nare able to achieve reasonable performance as compared to the fully supervised\napproach. Further, the backbone benefits from the unsupervised training and is\nuseful in downstream fewshot learning of keypoints, where fine-tuning on only\n10 percent of the labeled training data gives comparable performance to\nfine-tuning on the entire set. We demonstrated that GC-KPL outperforms by a\nlarge margin over SoTA when trained on entire dataset and efficiently leverages\nlarge volumes of unlabeled data.\n", "date": "2023-06-07", "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": "2212.07401", "title": "BKinD-3D: Self-Supervised 3D Keypoint Discovery from Multi-View Videos", "abstract": "  Quantifying motion in 3D is important for studying the behavior of humans and\nother animals, but manual pose annotations are expensive and time-consuming to\nobtain. Self-supervised keypoint discovery is a promising strategy for\nestimating 3D poses without annotations. However, current keypoint discovery\napproaches commonly process single 2D views and do not operate in the 3D space.\nWe propose a new method to perform self-supervised keypoint discovery in 3D\nfrom multi-view videos of behaving agents, without any keypoint or bounding box\nsupervision in 2D or 3D. Our method, BKinD-3D, uses an encoder-decoder\narchitecture with a 3D volumetric heatmap, trained to reconstruct\nspatiotemporal differences across multiple views, in addition to joint length\nconstraints on a learned 3D skeleton of the subject. In this way, we discover\nkeypoints without requiring manual supervision in videos of humans and rats,\ndemonstrating the potential of 3D keypoint discovery for studying behavior.\n", "date": "2022-12-14", "categories": []}
{"id": "2305.17245", "title": "Error Estimation for Single-Image Human Body Mesh Reconstruction", "abstract": "  Human pose and shape estimation methods continue to suffer in situations\nwhere one or more parts of the body are occluded. More importantly, these\nmethods cannot express when their predicted pose is incorrect. This has serious\nconsequences when these methods are used in human-robot interaction scenarios,\nwhere we need methods that can evaluate their predictions and flag situations\nwhere they might be wrong. This work studies this problem. We propose a method\nthat combines information from OpenPose and SPIN -- two popular human pose and\nshape estimation methods -- to highlight regions on the predicted mesh that are\nleast reliable. We have evaluated the proposed approach on 3DPW, 3DOH, and\nHuman3.6M datasets, and the results demonstrate our model's effectiveness in\nidentifying inaccurate regions of the human body mesh. Our code is available at\nhttps://github.com/Hamoon1987/meshConfidence.\n", "date": "2023-05-26", "categories": []}
{"id": "2302.01825", "title": "HDFormer: High-order Directed Transformer for 3D Human Pose Estimation", "abstract": "  Human pose estimation is a challenging task due to its structured data\nsequence nature. Existing methods primarily focus on pair-wise interaction of\nbody joints, which is insufficient for scenarios involving overlapping joints\nand rapidly changing poses. To overcome these issues, we introduce a novel\napproach, the High-order Directed Transformer (HDFormer), which leverages\nhigh-order bone and joint relationships for improved pose estimation.\nSpecifically, HDFormer incorporates both self-attention and high-order\nattention to formulate a multi-order attention module. This module facilitates\nfirst-order \"joint$\\leftrightarrow$joint\", second-order\n\"bone$\\leftrightarrow$joint\", and high-order \"hyperbone$\\leftrightarrow$joint\"\ninteractions, effectively addressing issues in complex and occlusion-heavy\nsituations. In addition, modern CNN techniques are integrated into the\ntransformer-based architecture, balancing the trade-off between performance and\nefficiency. HDFormer significantly outperforms state-of-the-art (SOTA) models\non Human3.6M and MPI-INF-3DHP datasets, requiring only 1/10 of the parameters\nand significantly lower computational costs. Moreover, HDFormer demonstrates\nbroad real-world applicability, enabling real-time, accurate 3D pose\nestimation. The source code is in https://github.com/hyer/HDFormer\n", "date": "2023-02-03", "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": "2305.08590", "title": "NIKI: Neural Inverse Kinematics with Invertible Neural Networks for 3D\n  Human Pose and Shape Estimation", "abstract": "  With the progress of 3D human pose and shape estimation, state-of-the-art\nmethods can either be robust to occlusions or obtain pixel-aligned accuracy in\nnon-occlusion cases. However, they cannot obtain robustness and mesh-image\nalignment at the same time. In this work, we present NIKI (Neural Inverse\nKinematics with Invertible Neural Network), which models bi-directional errors\nto improve the robustness to occlusions and obtain pixel-aligned accuracy. NIKI\ncan learn from both the forward and inverse processes with invertible networks.\nIn the inverse process, the model separates the error from the plausible 3D\npose manifold for a robust 3D human pose estimation. In the forward process, we\nenforce the zero-error boundary conditions to improve the sensitivity to\nreliable joint positions for better mesh-image alignment. Furthermore, NIKI\nemulates the analytical inverse kinematics algorithms with the twist-and-swing\ndecomposition for better interpretability. Experiments on standard and\nocclusion-specific benchmarks demonstrate the effectiveness of NIKI, where we\nexhibit robust and well-aligned results simultaneously. Code is available at\nhttps://github.com/Jeff-sjtu/NIKI\n", "date": "2023-05-15", "categories": []}
{"id": "2305.06968", "title": "HuManiFlow: Ancestor-Conditioned Normalising Flows on SO(3) Manifolds\n  for Human Pose and Shape Distribution Estimation", "abstract": "  Monocular 3D human pose and shape estimation is an ill-posed problem since\nmultiple 3D solutions can explain a 2D image of a subject. Recent approaches\npredict a probability distribution over plausible 3D pose and shape parameters\nconditioned on the image. We show that these approaches exhibit a trade-off\nbetween three key properties: (i) accuracy - the likelihood of the ground-truth\n3D solution under the predicted distribution, (ii) sample-input consistency -\nthe extent to which 3D samples from the predicted distribution match the\nvisible 2D image evidence, and (iii) sample diversity - the range of plausible\n3D solutions modelled by the predicted distribution. Our method, HuManiFlow,\npredicts simultaneously accurate, consistent and diverse distributions. We use\nthe human kinematic tree to factorise full body pose into ancestor-conditioned\nper-body-part pose distributions in an autoregressive manner. Per-body-part\ndistributions are implemented using normalising flows that respect the manifold\nstructure of SO(3), the Lie group of per-body-part poses. We show that\nill-posed, but ubiquitous, 3D point estimate losses reduce sample diversity,\nand employ only probabilistic training losses. Code is available at:\nhttps://github.com/akashsengupta1997/HuManiFlow.\n", "date": "2023-05-11", "categories": []}
{"id": "2302.08751", "title": "MDPose: Real-Time Multi-Person Pose Estimation via Mixture Density Model", "abstract": "  One of the major challenges in multi-person pose estimation is instance-aware\nkeypoint estimation. Previous methods address this problem by leveraging an\noff-the-shelf detector, heuristic post-grouping process or explicit instance\nidentification process, hindering further improvements in the inference speed\nwhich is an important factor for practical applications. From the statistical\npoint of view, those additional processes for identifying instances are\nnecessary to bypass learning the high-dimensional joint distribution of human\nkeypoints, which is a critical factor for another major challenge, the\nocclusion scenario. In this work, we propose a novel framework of single-stage\ninstance-aware pose estimation by modeling the joint distribution of human\nkeypoints with a mixture density model, termed as MDPose. Our MDPose estimates\nthe distribution of human keypoints' coordinates using a mixture density model\nwith an instance-aware keypoint head consisting simply of 8 convolutional\nlayers. It is trained by minimizing the negative log-likelihood of the ground\ntruth keypoints. Also, we propose a simple yet effective training strategy,\nRandom Keypoint Grouping (RKG), which significantly alleviates the underflow\nproblem leading to successful learning of relations between keypoints. On\nOCHuman dataset, which consists of images with highly occluded people, our\nMDPose achieves state-of-the-art performance by successfully learning the\nhigh-dimensional joint distribution of human keypoints. Furthermore, our MDPose\nshows significant improvement in inference speed with a competitive accuracy on\nMS COCO, a widely-used human keypoint dataset, thanks to the proposed much\nsimpler single-stage pipeline.\n", "date": "2023-02-17", "categories": []}
{"id": "2303.08475", "title": "Mutual Information-Based Temporal Difference Learning for Human Pose\n  Estimation in Video", "abstract": "  Temporal modeling is crucial for multi-frame human pose estimation. Most\nexisting methods directly employ optical flow or deformable convolution to\npredict full-spectrum motion fields, which might incur numerous irrelevant\ncues, such as a nearby person or background. Without further efforts to\nexcavate meaningful motion priors, their results are suboptimal, especially in\ncomplicated spatiotemporal interactions. On the other hand, the temporal\ndifference has the ability to encode representative motion information which\ncan potentially be valuable for pose estimation but has not been fully\nexploited. In this paper, we present a novel multi-frame human pose estimation\nframework, which employs temporal differences across frames to model dynamic\ncontexts and engages mutual information objectively to facilitate useful motion\ninformation disentanglement. To be specific, we design a multi-stage Temporal\nDifference Encoder that performs incremental cascaded learning conditioned on\nmulti-stage feature difference sequences to derive informative motion\nrepresentation. We further propose a Representation Disentanglement module from\nthe mutual information perspective, which can grasp discriminative\ntask-relevant motion signals by explicitly defining useful and noisy\nconstituents of the raw motion features and minimizing their mutual\ninformation. These place us to rank No.1 in the Crowd Pose Estimation in\nComplex Events Challenge on benchmark dataset HiEve, and achieve\nstate-of-the-art performance on three benchmarks PoseTrack2017, PoseTrack2018,\nand PoseTrack21.\n", "date": "2023-03-15", "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": "2305.00181", "title": "TAPE: Temporal Attention-based Probabilistic human pose and shape\n  Estimation", "abstract": "  Reconstructing 3D human pose and shape from monocular videos is a\nwell-studied but challenging problem. Common challenges include occlusions, the\ninherent ambiguities in the 2D to 3D mapping and the computational complexity\nof video processing. Existing methods ignore the ambiguities of the\nreconstruction and provide a single deterministic estimate for the 3D pose. In\norder to address these issues, we present a Temporal Attention based\nProbabilistic human pose and shape Estimation method (TAPE) that operates on an\nRGB video. More specifically, we propose to use a neural network to encode\nvideo frames to temporal features using an attention-based neural network.\nGiven these features, we output a per-frame but temporally-informed probability\ndistribution for the human pose using Normalizing Flows. We show that TAPE\noutperforms state-of-the-art methods in standard benchmarks and serves as an\neffective video-based prior for optimization-based human pose and shape\nestimation. Code is available at: https: //github.com/nikosvasilik/TAPE\n", "date": "2023-04-29", "categories": []}
{"id": "2304.14045", "title": "Interweaved Graph and Attention Network for 3D Human Pose Estimation", "abstract": "  Despite substantial progress in 3D human pose estimation from a single-view\nimage, prior works rarely explore global and local correlations, leading to\ninsufficient learning of human skeleton representations. To address this issue,\nwe propose a novel Interweaved Graph and Attention Network (IGANet) that allows\nbidirectional communications between graph convolutional networks (GCNs) and\nattentions. Specifically, we introduce an IGA module, where attentions are\nprovided with local information from GCNs and GCNs are injected with global\ninformation from attentions. Additionally, we design a simple yet effective\nU-shaped multi-layer perceptron (uMLP), which can capture multi-granularity\ninformation for body joints. Extensive experiments on two popular benchmark\ndatasets (i.e. Human3.6M and MPI-INF-3DHP) are conducted to evaluate our\nproposed method.The results show that IGANet achieves state-of-the-art\nperformance on both datasets. Code is available at\nhttps://github.com/xiu-cs/IGANet.\n", "date": "2023-04-27", "categories": []}
{"id": "2304.13651", "title": "What Happened 3 Seconds Ago? Inferring the Past with Thermal Imaging", "abstract": "  Inferring past human motion from RGB images is challenging due to the\ninherent uncertainty of the prediction problem. Thermal images, on the other\nhand, encode traces of past human-object interactions left in the environment\nvia thermal radiation measurement. Based on this observation, we collect the\nfirst RGB-Thermal dataset for human motion analysis, dubbed Thermal-IM. Then we\ndevelop a three-stage neural network model for accurate past human pose\nestimation. Comprehensive experiments show that thermal cues significantly\nreduce the ambiguities of this task, and the proposed model achieves remarkable\nperformance. The dataset is available at\nhttps://github.com/ZitianTang/Thermal-IM.\n", "date": "2023-04-26", "categories": []}
{"id": "2302.04774", "title": "3D Human Pose and Shape Estimation via HybrIK-Transformer", "abstract": "  HybrIK relies on a combination of analytical inverse kinematics and deep\nlearning to produce more accurate 3D pose estimation from 2D monocular images.\nHybrIK has three major components: (1) pretrained convolution backbone, (2)\ndeconvolution to lift 3D pose from 2D convolution features, (3) analytical\ninverse kinematics pass correcting deep learning prediction using learned\ndistribution of plausible twist and swing angles. In this paper we propose an\nenhancement of the 2D to 3D lifting module, replacing deconvolution with\nTransformer, resulting in accuracy and computational efficiency improvement\nrelative to the original HybrIK method. We demonstrate our results on commonly\nused H36M, PW3D, COCO and HP3D datasets. Our code is publicly available\nhttps://github.com/boreshkinai/hybrik-transformer.\n", "date": "2023-02-09", "categories": []}
{"id": "2304.12069", "title": "Occlusion Robust 3D Human Pose Estimation with StridedPoseGraphFormer\n  and Data Augmentation", "abstract": "  Occlusion is an omnipresent challenge in 3D human pose estimation (HPE). In\nspite of the large amount of research dedicated to 3D HPE, only a limited\nnumber of studies address the problem of occlusion explicitly. To fill this\ngap, we propose to combine exploitation of spatio-temporal features with\nsynthetic occlusion augmentation during training to deal with occlusion. To\nthis end, we build a spatio-temporal 3D HPE model, StridedPoseGraphFormer based\non graph convolution and transformers, and train it using occlusion\naugmentation. Unlike the existing occlusion-aware methods, that are only tested\nfor limited occlusion, we extensively evaluate our method for varying degrees\nof occlusion. We show that our proposed method compares favorably with the\nstate-of-the-art (SoA). Our experimental results also reveal that in the\nabsence of any occlusion handling mechanism, the performance of SoA 3D HPE\nmethods degrades significantly when they encounter occlusion.\n", "date": "2023-04-24", "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": "2304.08186", "title": "Human Pose Estimation in Monocular Omnidirectional Top-View Images", "abstract": "  Human pose estimation (HPE) with convolutional neural networks (CNNs) for\nindoor monitoring is one of the major challenges in computer vision. In\ncontrast to HPE in perspective views, an indoor monitoring system can consist\nof an omnidirectional camera with a field of view of 180{\\deg} to detect the\npose of a person with only one sensor per room. To recognize human pose, the\ndetection of keypoints is an essential upstream step. In our work we propose a\nnew dataset for training and evaluation of CNNs for the task of keypoint\ndetection in omnidirectional images. The training dataset, THEODORE+, consists\nof 50,000 images and is created by a 3D rendering engine, where humans are\nrandomly walking through an indoor environment. In a dynamically created 3D\nscene, persons move randomly with simultaneously moving omnidirectional camera\nto generate synthetic RGB images and 2D and 3D ground truth. For evaluation\npurposes, the real-world PoseFES dataset with two scenarios and 701 frames with\nup to eight persons per scene was captured and annotated. We propose four\ntraining paradigms to finetune or re-train two top-down models in MMPose and\ntwo bottom-up models in CenterNet on THEODORE+. Beside a qualitative evaluation\nwe report quantitative results. Compared to a COCO pretrained baseline, we\nachieve significant improvements especially for top-view scenes on the PoseFES\ndataset. Our datasets can be found at\nhttps://www.tu-chemnitz.de/etit/dst/forschung/comp_vision/datasets/index.php.en.\n", "date": "2023-04-17", "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": "2211.16940", "title": "DiffPose: Toward More Reliable 3D Pose Estimation", "abstract": "  Monocular 3D human pose estimation is quite challenging due to the inherent\nambiguity and occlusion, which often lead to high uncertainty and\nindeterminacy. On the other hand, diffusion models have recently emerged as an\neffective tool for generating high-quality images from noise. Inspired by their\ncapability, we explore a novel pose estimation framework (DiffPose) that\nformulates 3D pose estimation as a reverse diffusion process. We incorporate\nnovel designs into our DiffPose to facilitate the diffusion process for 3D pose\nestimation: a pose-specific initialization of pose uncertainty distributions, a\nGaussian Mixture Model-based forward diffusion process, and a\ncontext-conditioned reverse diffusion process. Our proposed DiffPose\nsignificantly outperforms existing methods on the widely used pose estimation\nbenchmarks Human3.6M and MPI-INF-3DHP. Project page:\nhttps://gongjia0208.github.io/Diffpose/.\n", "date": "2022-11-30", "categories": []}
{"id": "2304.04437", "title": "Monocular 3D Human Pose Estimation for Sports Broadcasts using Partial\n  Sports Field Registration", "abstract": "  The filming of sporting events projects and flattens the movement of athletes\nin the world onto a 2D broadcast image. The pixel locations of joints in these\nimages can be detected with high validity. Recovering the actual 3D movement of\nthe limbs (kinematics) of the athletes requires lifting these 2D pixel\nlocations back into a third dimension, implying a certain scene geometry. The\nwell-known line markings of sports fields allow for the calibration of the\ncamera and for determining the actual geometry of the scene. Close-up shots of\nathletes are required to extract detailed kinematics, which in turn obfuscates\nthe pertinent field markers for camera calibration. We suggest partial sports\nfield registration, which determines a set of scene-consistent camera\ncalibrations up to a single degree of freedom. Through joint optimization of 3D\npose estimation and camera calibration, we demonstrate the successful\nextraction of 3D running kinematics on a 400m track. In this work, we combine\nadvances in 2D human pose estimation and camera calibration via partial sports\nfield registration to demonstrate an avenue for collecting valid large-scale\nkinematic datasets. We generate a synthetic dataset of more than 10k images in\nUnreal Engine 5 with different viewpoints, running styles, and body types, to\nshow the limitations of existing monocular 3D HPE methods. Synthetic data and\ncode are available at https://github.com/tobibaum/PartialSportsFieldReg_3DHPE.\n", "date": "2023-04-10", "categories": []}
{"id": "2209.07034", "title": "A Temporal Densely Connected Recurrent Network for Event-based Human\n  Pose Estimation", "abstract": "  Event camera is an emerging bio-inspired vision sensors that report per-pixel\nbrightness changes asynchronously. It holds noticeable advantage of high\ndynamic range, high speed response, and low power budget that enable it to best\ncapture local motions in uncontrolled environments. This motivates us to unlock\nthe potential of event cameras for human pose estimation, as the human pose\nestimation with event cameras is rarely explored. Due to the novel paradigm\nshift from conventional frame-based cameras, however, event signals in a time\ninterval contain very limited information, as event cameras can only capture\nthe moving body parts and ignores those static body parts, resulting in some\nparts to be incomplete or even disappeared in the time interval. This paper\nproposes a novel densely connected recurrent architecture to address the\nproblem of incomplete information. By this recurrent architecture, we can\nexplicitly model not only the sequential but also non-sequential geometric\nconsistency across time steps to accumulate information from previous frames to\nrecover the entire human bodies, achieving a stable and accurate human pose\nestimation from event data. Moreover, to better evaluate our model, we collect\na large scale multimodal event-based dataset that comes with human pose\nannotations, which is by far the most challenging one to the best of our\nknowledge. The experimental results on two public datasets and our own dataset\ndemonstrate the effectiveness and strength of our approach. Code can be\navailable online for facilitating the future research.\n", "date": "2022-09-15", "categories": []}
{"id": "2304.02147", "title": "ConvFormer: Parameter Reduction in Transformer Models for 3D Human Pose\n  Estimation by Leveraging Dynamic Multi-Headed Convolutional Attention", "abstract": "  Recently, fully-transformer architectures have replaced the defacto\nconvolutional architecture for the 3D human pose estimation task. In this paper\nwe propose \\textbf{\\textit{ConvFormer}}, a novel convolutional transformer that\nleverages a new \\textbf{\\textit{dynamic multi-headed convolutional\nself-attention}} mechanism for monocular 3D human pose estimation. We designed\na spatial and temporal convolutional transformer to comprehensively model human\njoint relations within individual frames and globally across the motion\nsequence. Moreover, we introduce a novel notion of \\textbf{\\textit{temporal\njoints profile}} for our temporal ConvFormer that fuses complete temporal\ninformation immediately for a local neighborhood of joint features. We have\nquantitatively and qualitatively validated our method on three common benchmark\ndatasets: Human3.6M, MPI-INF-3DHP, and HumanEva. Extensive experiments have\nbeen conducted to identify the optimal hyper-parameter set. These experiments\ndemonstrated that we achieved a \\textbf{significant parameter reduction\nrelative to prior transformer models} while attaining State-of-the-Art (SOTA)\nor near SOTA on all three datasets. Additionally, we achieved SOTA for Protocol\nIII on H36M for both GT and CPN detection inputs. Finally, we obtained SOTA on\nall three metrics for the MPI-INF-3DHP dataset and for all three subjects on\nHumanEva under Protocol II.\n", "date": "2023-04-04", "categories": []}
{"id": "2304.02349", "title": "Self-supervised 3D Human Pose Estimation from a Single Image", "abstract": "  We propose a new self-supervised method for predicting 3D human body pose\nfrom a single image. The prediction network is trained from a dataset of\nunlabelled images depicting people in typical poses and a set of unpaired 2D\nposes. By minimising the need for annotated data, the method has the potential\nfor rapid application to pose estimation of other articulated structures (e.g.\nanimals). The self-supervision comes from an earlier idea exploiting\nconsistency between predicted pose under 3D rotation. Our method is a\nsubstantial advance on state-of-the-art self-supervised methods in training a\nmapping directly from images, without limb articulation constraints or any 3D\nempirical pose prior. We compare performance with state-of-the-art\nself-supervised methods using benchmark datasets that provide images and\nground-truth 3D pose (Human3.6M, MPI-INF-3DHP). Despite the reduced requirement\nfor annotated data, we show that the method outperforms on Human3.6M and\nmatches performance on MPI-INF-3DHP. Qualitative results on a dataset of human\nhands show the potential for rapidly learning to predict 3D pose for\narticulated structures other than the human body.\n", "date": "2023-04-05", "categories": []}
{"id": "2304.01865", "title": "SportsPose -- A Dynamic 3D sports pose dataset", "abstract": "  Accurate 3D human pose estimation is essential for sports analytics,\ncoaching, and injury prevention. However, existing datasets for monocular pose\nestimation do not adequately capture the challenging and dynamic nature of\nsports movements. In response, we introduce SportsPose, a large-scale 3D human\npose dataset consisting of highly dynamic sports movements. With more than\n176,000 3D poses from 24 different subjects performing 5 different sports\nactivities, SportsPose provides a diverse and comprehensive set of 3D poses\nthat reflect the complex and dynamic nature of sports movements. Contrary to\nother markerless datasets we have quantitatively evaluated the precision of\nSportsPose by comparing our poses with a commercial marker-based system and\nachieve a mean error of 34.5 mm across all evaluation sequences. This is\ncomparable to the error reported on the commonly used 3DPW dataset. We further\nintroduce a new metric, local movement, which describes the movement of the\nwrist and ankle joints in relation to the body. With this, we show that\nSportsPose contains more movement than the Human3.6M and 3DPW datasets in these\nextremum joints, indicating that our movements are more dynamic. The dataset\nwith accompanying code can be downloaded from our website. We hope that\nSportsPose will allow researchers and practitioners to develop and evaluate\nmore effective models for the analysis of sports performance and injury\nprevention. With its realistic and diverse dataset, SportsPose provides a\nvaluable resource for advancing the state-of-the-art in pose estimation in\nsports.\n", "date": "2023-04-04", "categories": []}
{"id": "2209.13906", "title": "SmartMocap: Joint Estimation of Human and Camera Motion using\n  Uncalibrated RGB Cameras", "abstract": "  Markerless human motion capture (mocap) from multiple RGB cameras is a widely\nstudied problem. Existing methods either need calibrated cameras or calibrate\nthem relative to a static camera, which acts as the reference frame for the\nmocap system. The calibration step has to be done a priori for every capture\nsession, which is a tedious process, and re-calibration is required whenever\ncameras are intentionally or accidentally moved. In this paper, we propose a\nmocap method which uses multiple static and moving extrinsically uncalibrated\nRGB cameras. The key components of our method are as follows. First, since the\ncameras and the subject can move freely, we select the ground plane as a common\nreference to represent both the body and the camera motions unlike existing\nmethods which represent bodies in the camera coordinate. Second, we learn a\nprobability distribution of short human motion sequences ($\\sim$1sec) relative\nto the ground plane and leverage it to disambiguate between the camera and\nhuman motion. Third, we use this distribution as a motion prior in a novel\nmulti-stage optimization approach to fit the SMPL human body model and the\ncamera poses to the human body keypoints on the images. Finally, we show that\nour method can work on a variety of datasets ranging from aerial cameras to\nsmartphones. It also gives more accurate results compared to the\nstate-of-the-art on the task of monocular human mocap with a static camera. Our\ncode is available for research purposes on\nhttps://github.com/robot-perception-group/SmartMocap.\n", "date": "2022-09-28", "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": "2303.18119", "title": "Markerless 3D human pose tracking through multiple cameras and AI:\n  Enabling high accuracy, robustness, and real-time performance", "abstract": "  Tracking 3D human motion in real-time is crucial for numerous applications\nacross many fields. Traditional approaches involve attaching artificial\nfiducial objects or sensors to the body, limiting their usability and\ncomfort-of-use and consequently narrowing their application fields. Recent\nadvances in Artificial Intelligence (AI) have allowed for markerless solutions.\nHowever, most of these methods operate in 2D, while those providing 3D\nsolutions compromise accuracy and real-time performance. To address this\nchallenge and unlock the potential of visual pose estimation methods in\nreal-world scenarios, we propose a markerless framework that combines\nmulti-camera views and 2D AI-based pose estimation methods to track 3D human\nmotion. Our approach integrates a Weighted Least Square (WLS) algorithm that\ncomputes 3D human motion from multiple 2D pose estimations provided by an\nAI-driven method. The method is integrated within the Open-VICO framework\nallowing simulation and real-world execution. Several experiments have been\nconducted, which have shown high accuracy and real-time performance,\ndemonstrating the high level of readiness for real-world applications and the\npotential to revolutionize human motion capture.\n", "date": "2023-03-31", "categories": []}
{"id": "2303.07337", "title": "PoseExaminer: Automated Testing of Out-of-Distribution Robustness in\n  Human Pose and Shape Estimation", "abstract": "  Human pose and shape (HPS) estimation methods achieve remarkable results.\nHowever, current HPS benchmarks are mostly designed to test models in scenarios\nthat are similar to the training data. This can lead to critical situations in\nreal-world applications when the observed data differs significantly from the\ntraining data and hence is out-of-distribution (OOD). It is therefore important\nto test and improve the OOD robustness of HPS methods. To address this\nfundamental problem, we develop a simulator that can be controlled in a\nfine-grained manner using interpretable parameters to explore the manifold of\nimages of human pose, e.g. by varying poses, shapes, and clothes. We introduce\na learning-based testing method, termed PoseExaminer, that automatically\ndiagnoses HPS algorithms by searching over the parameter space of human pose\nimages to find the failure modes. Our strategy for exploring this\nhigh-dimensional parameter space is a multi-agent reinforcement learning\nsystem, in which the agents collaborate to explore different parts of the\nparameter space. We show that our PoseExaminer discovers a variety of\nlimitations in current state-of-the-art models that are relevant in real-world\nscenarios but are missed by current benchmarks. For example, it finds large\nregions of realistic human poses that are not predicted correctly, as well as\nreduced performance for humans with skinny and corpulent body shapes. In\naddition, we show that fine-tuning HPS methods by exploiting the failure modes\nfound by PoseExaminer improve their robustness and even their performance on\nstandard benchmarks by a significant margin. The code are available for\nresearch purposes.\n", "date": "2023-03-13", "categories": []}
{"id": "2303.17472", "title": "PoseFormerV2: Exploring Frequency Domain for Efficient and Robust 3D\n  Human Pose Estimation", "abstract": "  Recently, transformer-based methods have gained significant success in\nsequential 2D-to-3D lifting human pose estimation. As a pioneering work,\nPoseFormer captures spatial relations of human joints in each video frame and\nhuman dynamics across frames with cascaded transformer layers and has achieved\nimpressive performance. However, in real scenarios, the performance of\nPoseFormer and its follow-ups is limited by two factors: (a) The length of the\ninput joint sequence; (b) The quality of 2D joint detection. Existing methods\ntypically apply self-attention to all frames of the input sequence, causing a\nhuge computational burden when the frame number is increased to obtain advanced\nestimation accuracy, and they are not robust to noise naturally brought by the\nlimited capability of 2D joint detectors. In this paper, we propose\nPoseFormerV2, which exploits a compact representation of lengthy skeleton\nsequences in the frequency domain to efficiently scale up the receptive field\nand boost robustness to noisy 2D joint detection. With minimum modifications to\nPoseFormer, the proposed method effectively fuses features both in the time\ndomain and frequency domain, enjoying a better speed-accuracy trade-off than\nits precursor. Extensive experiments on two benchmark datasets (i.e., Human3.6M\nand MPI-INF-3DHP) demonstrate that the proposed approach significantly\noutperforms the original PoseFormer and other transformer-based variants. Code\nis released at \\url{https://github.com/QitaoZhao/PoseFormerV2}.\n", "date": "2023-03-30", "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": "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": "2301.05175", "title": "Scene-Aware 3D Multi-Human Motion Capture from a Single Camera", "abstract": "  In this work, we consider the problem of estimating the 3D position of\nmultiple humans in a scene as well as their body shape and articulation from a\nsingle RGB video recorded with a static camera. In contrast to expensive\nmarker-based or multi-view systems, our lightweight setup is ideal for private\nusers as it enables an affordable 3D motion capture that is easy to install and\ndoes not require expert knowledge. To deal with this challenging setting, we\nleverage recent advances in computer vision using large-scale pre-trained\nmodels for a variety of modalities, including 2D body joints, joint angles,\nnormalized disparity maps, and human segmentation masks. Thus, we introduce the\nfirst non-linear optimization-based approach that jointly solves for the\nabsolute 3D position of each human, their articulated pose, their individual\nshapes as well as the scale of the scene. In particular, we estimate the scene\ndepth and person unique scale from normalized disparity predictions using the\n2D body joints and joint angles. Given the per-frame scene depth, we\nreconstruct a point-cloud of the static scene in 3D space. Finally, given the\nper-frame 3D estimates of the humans and scene point-cloud, we perform a\nspace-time coherent optimization over the video to ensure temporal, spatial and\nphysical plausibility. We evaluate our method on established multi-person 3D\nhuman pose benchmarks where we consistently outperform previous methods and we\nqualitatively demonstrate that our method is robust to in-the-wild conditions\nincluding challenging scenes with people of different sizes.\n", "date": "2023-01-12", "categories": []}
{"id": "2303.11180", "title": "Self-Correctable and Adaptable Inference for Generalizable Human Pose\n  Estimation", "abstract": "  A central challenge in human pose estimation, as well as in many other\nmachine learning and prediction tasks, is the generalization problem. The\nlearned network does not have the capability to characterize the prediction\nerror, generate feedback information from the test sample, and correct the\nprediction error on the fly for each individual test sample, which results in\ndegraded performance in generalization. In this work, we introduce a\nself-correctable and adaptable inference (SCAI) method to address the\ngeneralization challenge of network prediction and use human pose estimation as\nan example to demonstrate its effectiveness and performance. We learn a\ncorrection network to correct the prediction result conditioned by a fitness\nfeedback error. This feedback error is generated by a learned fitness feedback\nnetwork which maps the prediction result to the original input domain and\ncompares it against the original input. Interestingly, we find that this\nself-referential feedback error is highly correlated with the actual prediction\nerror. This strong correlation suggests that we can use this error as feedback\nto guide the correction process. It can be also used as a loss function to\nquickly adapt and optimize the correction network during the inference process.\nOur extensive experimental results on human pose estimation demonstrate that\nthe proposed SCAI method is able to significantly improve the generalization\ncapability and performance of human pose estimation.\n", "date": "2023-03-20", "categories": []}
{"id": "2303.14747", "title": "Global-to-Local Modeling for Video-based 3D Human Pose and Shape\n  Estimation", "abstract": "  Video-based 3D human pose and shape estimations are evaluated by intra-frame\naccuracy and inter-frame smoothness. Although these two metrics are responsible\nfor different ranges of temporal consistency, existing state-of-the-art methods\ntreat them as a unified problem and use monotonous modeling structures (e.g.,\nRNN or attention-based block) to design their networks. However, using a single\nkind of modeling structure is difficult to balance the learning of short-term\nand long-term temporal correlations, and may bias the network to one of them,\nleading to undesirable predictions like global location shift, temporal\ninconsistency, and insufficient local details. To solve these problems, we\npropose to structurally decouple the modeling of long-term and short-term\ncorrelations in an end-to-end framework, Global-to-Local Transformer (GLoT).\nFirst, a global transformer is introduced with a Masked Pose and Shape\nEstimation strategy for long-term modeling. The strategy stimulates the global\ntransformer to learn more inter-frame correlations by randomly masking the\nfeatures of several frames. Second, a local transformer is responsible for\nexploiting local details on the human mesh and interacting with the global\ntransformer by leveraging cross-attention. Moreover, a Hierarchical Spatial\nCorrelation Regressor is further introduced to refine intra-frame estimations\nby decoupled global-local representation and implicit kinematic constraints.\nOur GLoT surpasses previous state-of-the-art methods with the lowest model\nparameters on popular benchmarks, i.e., 3DPW, MPI-INF-3DHP, and Human3.6M.\nCodes are available at https://github.com/sxl142/GLoT.\n", "date": "2023-03-26", "categories": []}
{"id": "2303.15410", "title": "Human Pose Estimation in Extremely Low-Light Conditions", "abstract": "  We study human pose estimation in extremely low-light images. This task is\nchallenging due to the difficulty of collecting real low-light images with\naccurate labels, and severely corrupted inputs that degrade prediction quality\nsignificantly. To address the first issue, we develop a dedicated camera system\nand build a new dataset of real low-light images with accurate pose labels.\nThanks to our camera system, each low-light image in our dataset is coupled\nwith an aligned well-lit image, which enables accurate pose labeling and is\nused as privileged information during training. We also propose a new model and\na new training strategy that fully exploit the privileged information to learn\nrepresentation insensitive to lighting conditions. Our method demonstrates\noutstanding performance on real extremely low light images, and extensive\nanalyses validate that both of our model and dataset contribute to the success.\n", "date": "2023-03-27", "categories": []}
{"id": "2303.13357", "title": "POTTER: Pooling Attention Transformer for Efficient Human Mesh Recovery", "abstract": "  Transformer architectures have achieved SOTA performance on the human mesh\nrecovery (HMR) from monocular images. However, the performance gain has come at\nthe cost of substantial memory and computational overhead. A lightweight and\nefficient model to reconstruct accurate human mesh is needed for real-world\napplications. In this paper, we propose a pure transformer architecture named\nPOoling aTtention TransformER (POTTER) for the HMR task from single images.\nObserving that the conventional attention module is memory and computationally\nexpensive, we propose an efficient pooling attention module, which\nsignificantly reduces the memory and computational cost without sacrificing\nperformance. Furthermore, we design a new transformer architecture by\nintegrating a High-Resolution (HR) stream for the HMR task. The high-resolution\nlocal and global features from the HR stream can be utilized for recovering\nmore accurate human mesh. Our POTTER outperforms the SOTA method METRO by only\nrequiring 7% of total parameters and 14% of the Multiply-Accumulate Operations\non the Human3.6M (PA-MPJPE metric) and 3DPW (all three metrics) datasets. The\nproject webpage is https://zczcwh.github.io/potter_page.\n", "date": "2023-03-23", "categories": []}
{"id": "2302.12827", "title": "Decoupling Human and Camera Motion from Videos in the Wild", "abstract": "  We propose a method to reconstruct global human trajectories from videos in\nthe wild. Our optimization method decouples the camera and human motion, which\nallows us to place people in the same world coordinate frame. Most existing\nmethods do not model the camera motion; methods that rely on the background\npixels to infer 3D human motion usually require a full scene reconstruction,\nwhich is often not possible for in-the-wild videos. However, even when existing\nSLAM systems cannot recover accurate scene reconstructions, the background\npixel motion still provides enough signal to constrain the camera motion. We\nshow that relative camera estimates along with data-driven human motion priors\ncan resolve the scene scale ambiguity and recover global human trajectories.\nOur method robustly recovers the global 3D trajectories of people in\nchallenging in-the-wild videos, such as PoseTrack. We quantify our improvement\nover existing methods on 3D human dataset Egobody. We further demonstrate that\nour recovered camera scale allows us to reason about motion of multiple people\nin a shared coordinate frame, which improves performance of downstream tracking\nin PoseTrack. Code and video results can be found at\nhttps://vye16.github.io/slahmr.\n", "date": "2023-02-24", "categories": []}
{"id": "2303.11638", "title": "Human Pose as Compositional Tokens", "abstract": "  Human pose is typically represented by a coordinate vector of body joints or\ntheir heatmap embeddings. While easy for data processing, unrealistic pose\nestimates are admitted due to the lack of dependency modeling between the body\njoints. In this paper, we present a structured representation, named Pose as\nCompositional Tokens (PCT), to explore the joint dependency. It represents a\npose by M discrete tokens with each characterizing a sub-structure with several\ninterdependent joints. The compositional design enables it to achieve a small\nreconstruction error at a low cost. Then we cast pose estimation as a\nclassification task. In particular, we learn a classifier to predict the\ncategories of the M tokens from an image. A pre-learned decoder network is used\nto recover the pose from the tokens without further post-processing. We show\nthat it achieves better or comparable pose estimation results as the existing\nmethods in general scenarios, yet continues to work well when occlusion occurs,\nwhich is ubiquitous in practice. The code and models are publicly available at\nhttps://github.com/Gengzigang/PCT.\n", "date": "2023-03-21", "categories": []}
{"id": "2303.09095", "title": "SLOPER4D: A Scene-Aware Dataset for Global 4D Human Pose Estimation in\n  Urban Environments", "abstract": "  We present SLOPER4D, a novel scene-aware dataset collected in large urban\nenvironments to facilitate the research of global human pose estimation (GHPE)\nwith human-scene interaction in the wild. Employing a head-mounted device\nintegrated with a LiDAR and camera, we record 12 human subjects' activities\nover 10 diverse urban scenes from an egocentric view. Frame-wise annotations\nfor 2D key points, 3D pose parameters, and global translations are provided,\ntogether with reconstructed scene point clouds. To obtain accurate 3D ground\ntruth in such large dynamic scenes, we propose a joint optimization method to\nfit local SMPL meshes to the scene and fine-tune the camera calibration during\ndynamic motions frame by frame, resulting in plausible and scene-natural 3D\nhuman poses. Eventually, SLOPER4D consists of 15 sequences of human motions,\neach of which has a trajectory length of more than 200 meters (up to 1,300\nmeters) and covers an area of more than 2,000 $m^2$ (up to 13,000 $m^2$),\nincluding more than 100K LiDAR frames, 300k video frames, and 500K IMU-based\nmotion frames. With SLOPER4D, we provide a detailed and thorough analysis of\ntwo critical tasks, including camera-based 3D HPE and LiDAR-based 3D HPE in\nurban environments, and benchmark a new task, GHPE. The in-depth analysis\ndemonstrates SLOPER4D poses significant challenges to existing methods and\nproduces great research opportunities. The dataset and code are released at\n\\url{http://www.lidarhumanmotion.net/sloper4d/}\n", "date": "2023-03-16", "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": "2303.09187", "title": "PSVT: End-to-End Multi-person 3D Pose and Shape Estimation with\n  Progressive Video Transformers", "abstract": "  Existing methods of multi-person video 3D human Pose and Shape Estimation\n(PSE) typically adopt a two-stage strategy, which first detects human instances\nin each frame and then performs single-person PSE with temporal model. However,\nthe global spatio-temporal context among spatial instances can not be captured.\nIn this paper, we propose a new end-to-end multi-person 3D Pose and Shape\nestimation framework with progressive Video Transformer, termed PSVT. In PSVT,\na spatio-temporal encoder (STE) captures the global feature dependencies among\nspatial objects. Then, spatio-temporal pose decoder (STPD) and shape decoder\n(STSD) capture the global dependencies between pose queries and feature tokens,\nshape queries and feature tokens, respectively. To handle the variances of\nobjects as time proceeds, a novel scheme of progressive decoding is used to\nupdate pose and shape queries at each frame. Besides, we propose a novel\npose-guided attention (PGA) for shape decoder to better predict shape\nparameters. The two components strengthen the decoder of PSVT to improve\nperformance. Extensive experiments on the four datasets show that PSVT achieves\nstage-of-the-art results.\n", "date": "2023-03-16", "categories": []}
{"id": "2210.05669", "title": "A generic diffusion-based approach for 3D human pose prediction in the\n  wild", "abstract": "  Predicting 3D human poses in real-world scenarios, also known as human pose\nforecasting, is inevitably subject to noisy inputs arising from inaccurate 3D\npose estimations and occlusions. To address these challenges, we propose a\ndiffusion-based approach that can predict given noisy observations. We frame\nthe prediction task as a denoising problem, where both observation and\nprediction are considered as a single sequence containing missing elements\n(whether in the observation or prediction horizon). All missing elements are\ntreated as noise and denoised with our conditional diffusion model. To better\nhandle long-term forecasting horizon, we present a temporal cascaded diffusion\nmodel. We demonstrate the benefits of our approach on four publicly available\ndatasets (Human3.6M, HumanEva-I, AMASS, and 3DPW), outperforming the\nstate-of-the-art. Additionally, we show that our framework is generic enough to\nimprove any 3D pose prediction model as a pre-processing step to repair their\ninputs and a post-processing step to refine their outputs. The code is\navailable online: \\url{https://github.com/vita-epfl/DePOSit}.\n", "date": "2022-10-11", "categories": []}
{"id": "2303.07141", "title": "An Improved Baseline Framework for Pose Estimation Challenge at ECCV\n  2022 Visual Perception for Navigation in Human Environments Workshop", "abstract": "  This technical report describes our first-place solution to the pose\nestimation challenge at ECCV 2022 Visual Perception for Navigation in Human\nEnvironments Workshop. In this challenge, we aim to estimate human poses from\nin-the-wild stitched panoramic images. Our method is built based on Faster\nR-CNN for human detection, and HRNet for human pose estimation. We describe\ntechnical details for the JRDB-Pose dataset, together with some experimental\nresults. In the competition, we achieved 0.303 $\\text{OSPA}_{\\text{IOU}}$ and\n64.047\\% $\\text{AP}_{\\text{0.5}}$ on the test set of JRDB-Pose.\n", "date": "2023-03-13", "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": "2303.05370", "title": "Rethinking Self-Supervised Visual Representation Learning in\n  Pre-training for 3D Human Pose and Shape Estimation", "abstract": "  Recently, a few self-supervised representation learning (SSL) methods have\noutperformed the ImageNet classification pre-training for vision tasks such as\nobject detection. However, its effects on 3D human body pose and shape\nestimation (3DHPSE) are open to question, whose target is fixed to a unique\nclass, the human, and has an inherent task gap with SSL. We empirically study\nand analyze the effects of SSL and further compare it with other pre-training\nalternatives for 3DHPSE. The alternatives are 2D annotation-based pre-training\nand synthetic data pre-training, which share the motivation of SSL that aims to\nreduce the labeling cost. They have been widely utilized as a source of\nweak-supervision or fine-tuning, but have not been remarked as a pre-training\nsource. SSL methods underperform the conventional ImageNet classification\npre-training on multiple 3DHPSE benchmarks by 7.7% on average. In contrast,\ndespite a much less amount of pre-training data, the 2D annotation-based\npre-training improves accuracy on all benchmarks and shows faster convergence\nduring fine-tuning. Our observations challenge the naive application of the\ncurrent SSL pre-training to 3DHPSE and relight the value of other data types in\nthe pre-training aspect.\n", "date": "2023-03-09", "categories": []}
{"id": "2303.02413", "title": "Improved Trajectory Reconstruction for Markerless Pose Estimation", "abstract": "  Markerless pose estimation allows reconstructing human movement from multiple\nsynchronized and calibrated views, and has the potential to make movement\nanalysis easy and quick, including gait analysis. This could enable much more\nfrequent and quantitative characterization of gait impairments, allowing better\nmonitoring of outcomes and responses to interventions. However, the impact of\ndifferent keypoint detectors and reconstruction algorithms on markerless pose\nestimation accuracy has not been thoroughly evaluated. We tested these\nalgorithmic choices on data acquired from a multicamera system from a\nheterogeneous sample of 25 individuals seen in a rehabilitation hospital. We\nfound that using a top-down keypoint detector and reconstructing trajectories\nwith an implicit function enabled accurate, smooth and anatomically plausible\ntrajectories, with a noise in the step width estimates compared to a GaitRite\nwalkway of only 8mm.\n", "date": "2023-03-04", "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": "2302.11301", "title": "View Consistency Aware Holistic Triangulation for 3D Human Pose\n  Estimation", "abstract": "  The rapid development of multi-view 3D human pose estimation (HPE) is\nattributed to the maturation of monocular 2D HPE and the geometry of 3D\nreconstruction. However, 2D detection outliers in occluded views due to neglect\nof view consistency, and 3D implausible poses due to lack of pose coherence,\nremain challenges. To solve this, we introduce a Multi-View Fusion module to\nrefine 2D results by establishing view correlations. Then, Holistic\nTriangulation is proposed to infer the whole pose as an entirety, and anatomy\nprior is injected to maintain the pose coherence and improve the plausibility.\nAnatomy prior is extracted by PCA whose input is skeletal structure features,\nwhich can factor out global context and joint-by-joint relationship from\nabstract to concrete. Benefiting from the closed-form solution, the whole\nframework is trained end-to-end. Our method outperforms the state of the art in\nboth precision and plausibility which is assessed by a new metric.\n", "date": "2023-02-22", "categories": []}
{"id": "2302.09790", "title": "HTNet: Human Topology Aware Network for 3D Human Pose Estimation", "abstract": "  3D human pose estimation errors would propagate along the human body topology\nand accumulate at the end joints of limbs. Inspired by the backtracking\nmechanism in automatic control systems, we design an Intra-Part Constraint\nmodule that utilizes the parent nodes as the reference to build topological\nconstraints for end joints at the part level. Further considering the hierarchy\nof the human topology, joint-level and body-level dependencies are captured via\ngraph convolutional networks and self-attentions, respectively. Based on these\ndesigns, we propose a novel Human Topology aware Network (HTNet), which adopts\na channel-split progressive strategy to sequentially learn the structural\npriors of the human topology from multiple semantic levels: joint, part, and\nbody. Extensive experiments show that the proposed method improves the\nestimation accuracy by 18.7% on the end joints of limbs and achieves\nstate-of-the-art results on Human3.6M and MPI-INF-3DHP datasets. Code is\navailable at https://github.com/vefalun/HTNet.\n", "date": "2023-02-20", "categories": []}
{"id": "2302.08760", "title": "3D Human Pose Lifting with Grid Convolution", "abstract": "  Existing lifting networks for regressing 3D human poses from 2D single-view\nposes are typically constructed with linear layers based on graph-structured\nrepresentation learning. In sharp contrast to them, this paper presents Grid\nConvolution (GridConv), mimicking the wisdom of regular convolution operations\nin image space. GridConv is based on a novel Semantic Grid Transformation (SGT)\nwhich leverages a binary assignment matrix to map the irregular\ngraph-structured human pose onto a regular weave-like grid pose representation\njoint by joint, enabling layer-wise feature learning with GridConv operations.\nWe provide two ways to implement SGT, including handcrafted and learnable\ndesigns. Surprisingly, both designs turn out to achieve promising results and\nthe learnable one is better, demonstrating the great potential of this new\nlifting representation learning formulation. To improve the ability of GridConv\nto encode contextual cues, we introduce an attention module over the\nconvolutional kernel, making grid convolution operations input-dependent,\nspatial-aware and grid-specific. We show that our fully convolutional grid\nlifting network outperforms state-of-the-art methods with noticeable margins\nunder (1) conventional evaluation on Human3.6M and (2) cross-evaluation on\nMPI-INF-3DHP. Code is available at https://github.com/OSVAI/GridConv\n", "date": "2023-02-17", "categories": []}
{"id": "2302.07408", "title": "Pose-Oriented Transformer with Uncertainty-Guided Refinement for\n  2D-to-3D Human Pose Estimation", "abstract": "  There has been a recent surge of interest in introducing transformers to 3D\nhuman pose estimation (HPE) due to their powerful capabilities in modeling\nlong-term dependencies. However, existing transformer-based methods treat body\njoints as equally important inputs and ignore the prior knowledge of human\nskeleton topology in the self-attention mechanism. To tackle this issue, in\nthis paper, we propose a Pose-Oriented Transformer (POT) with uncertainty\nguided refinement for 3D HPE. Specifically, we first develop novel\npose-oriented self-attention mechanism and distance-related position embedding\nfor POT to explicitly exploit the human skeleton topology. The pose-oriented\nself-attention mechanism explicitly models the topological interactions between\nbody joints, whereas the distance-related position embedding encodes the\ndistance of joints to the root joint to distinguish groups of joints with\ndifferent difficulties in regression. Furthermore, we present an\nUncertainty-Guided Refinement Network (UGRN) to refine pose predictions from\nPOT, especially for the difficult joints, by considering the estimated\nuncertainty of each joint with uncertainty-guided sampling strategy and\nself-attention mechanism. Extensive experiments demonstrate that our method\nsignificantly outperforms the state-of-the-art methods with reduced model\nparameters on 3D HPE benchmarks such as Human3.6M and MPI-INF-3DHP\n", "date": "2023-02-15", "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": "2302.06194", "title": "Capsules as viewpoint learners for human pose estimation", "abstract": "  The task of human pose estimation (HPE) deals with the ill-posed problem of\nestimating the 3D position of human joints directly from images and videos. In\nrecent literature, most of the works tackle the problem mostly by using\nconvolutional neural networks (CNNs), which are capable of achieving\nstate-of-the-art results in most datasets. We show how most neural networks are\nnot able to generalize well when the camera is subject to significant viewpoint\nchanges. This behaviour emerges because CNNs lack the capability of modelling\nviewpoint equivariance, while they rather rely on viewpoint invariance,\nresulting in high data dependency. Recently, capsule networks (CapsNets) have\nbeen proposed in the multi-class classification field as a solution to the\nviewpoint equivariance issue, reducing both the size and complexity of both the\ntraining datasets and the network itself. In this work, we show how capsule\nnetworks can be adopted to achieve viewpoint equivariance in human pose\nestimation. We propose a novel end-to-end viewpoint-equivariant capsule\nautoencoder that employs a fast Variational Bayes routing and matrix capsules.\nWe achieve state-of-the-art results for multiple tasks and datasets while\nretaining other desirable properties, such as greater generalization\ncapabilities when changing viewpoints, lower data dependency and fast\ninference. Additionally, by modelling each joint as a capsule, the hierarchical\nand geometrical structure of the overall pose is retained in the feature space,\nindependently from the viewpoint. We further test our network on multiple\ndatasets, both in the RGB and depth domain, from seen and unseen viewpoints and\nin the viewpoint transfer task.\n", "date": "2023-02-13", "categories": []}
{"id": "2301.13403", "title": "A Modular Multi-stage Lightweight Graph Transformer Network for Human\n  Pose and Shape Estimation from 2D Human Pose", "abstract": "  In this research, we address the challenge faced by existing deep\nlearning-based human mesh reconstruction methods in balancing accuracy and\ncomputational efficiency. These methods typically prioritize accuracy,\nresulting in large network sizes and excessive computational complexity, which\nmay hinder their practical application in real-world scenarios, such as virtual\nreality systems. To address this issue, we introduce a modular multi-stage\nlightweight graph-based transformer network for human pose and shape estimation\nfrom 2D human pose, a pose-based human mesh reconstruction approach that\nprioritizes computational efficiency without sacrificing reconstruction\naccuracy. Our method consists of a 2D-to-3D lifter module that utilizes graph\ntransformers to analyze structured and implicit joint correlations in 2D human\nposes, and a mesh regression module that combines the extracted pose features\nwith a mesh template to produce the final human mesh parameters.\n", "date": "2023-01-31", "categories": []}
{"id": "2301.13013", "title": "RFPose-OT: RF-Based 3D Human Pose Estimation via Optimal Transport\n  Theory", "abstract": "  This paper introduces a novel framework, i.e., RFPose-OT, to enable the 3D\nhuman pose estimation from Radio Frequency (RF) signals. Different from\nexisting methods that predict human poses from RF signals on the signal level\ndirectly, we consider the structure difference between the RF signals and the\nhuman poses, propose to transform the RF signals to the pose domain on the\nfeature level based on Optimal Transport (OT) theory, and generate human poses\nfrom the transformed features. To evaluate RFPose-OT, we build a radio system\nand a multi-view camera system to acquire the RF signal data and the\nground-truth human poses. The experimental results in basic indoor environment,\nocclusion indoor environment, and outdoor environment, all demonstrate that\nRFPose-OT can predict 3D human poses with higher precision than the\nstate-of-the-art methods.\n", "date": "2022-12-26", "categories": []}
{"id": "2301.09015", "title": "E$^3$Pose: Energy-Efficient Edge-assisted Multi-camera System for\n  Multi-human 3D Pose Estimation", "abstract": "  Multi-human 3D pose estimation plays a key role in establishing a seamless\nconnection between the real world and the virtual world. Recent efforts adopted\na two-stage framework that first builds 2D pose estimations in multiple camera\nviews from different perspectives and then synthesizes them into 3D poses.\nHowever, the focus has largely been on developing new computer vision\nalgorithms on the offline video datasets without much consideration on the\nenergy constraints in real-world systems with flexibly-deployed and\nbattery-powered cameras. In this paper, we propose an energy-efficient\nedge-assisted multiple-camera system, dubbed E$^3$Pose, for real-time\nmulti-human 3D pose estimation, based on the key idea of adaptive camera\nselection. Instead of always employing all available cameras to perform 2D pose\nestimations as in the existing works, E$^3$Pose selects only a subset of\ncameras depending on their camera view qualities in terms of occlusion and\nenergy states in an adaptive manner, thereby reducing the energy consumption\n(which translates to extended battery lifetime) and improving the estimation\naccuracy. To achieve this goal, E$^3$Pose incorporates an attention-based LSTM\nto predict the occlusion information of each camera view and guide camera\nselection before cameras are selected to process the images of a scene, and\nruns a camera selection algorithm based on the Lyapunov optimization framework\nto make long-term adaptive selection decisions. We build a prototype of\nE$^3$Pose on a 5-camera testbed, demonstrate its feasibility and evaluate its\nperformance. Our results show that a significant energy saving (up to 31.21%)\ncan be achieved while maintaining a high 3D pose estimation accuracy comparable\nto state-of-the-art methods.\n", "date": "2023-01-21", "categories": []}
{"id": "2301.07322", "title": "HSTFormer: Hierarchical Spatial-Temporal Transformers for 3D Human Pose\n  Estimation", "abstract": "  Transformer-based approaches have been successfully proposed for 3D human\npose estimation (HPE) from 2D pose sequence and achieved state-of-the-art\n(SOTA) performance. However, current SOTAs have difficulties in modeling\nspatial-temporal correlations of joints at different levels simultaneously.\nThis is due to the poses' spatial-temporal complexity. Poses move at various\nspeeds temporarily with various joints and body-parts movement spatially.\nHence, a cookie-cutter transformer is non-adaptable and can hardly meet the\n\"in-the-wild\" requirement. To mitigate this issue, we propose Hierarchical\nSpatial-Temporal transFormers (HSTFormer) to capture multi-level joints'\nspatial-temporal correlations from local to global gradually for accurate 3D\nHPE. HSTFormer consists of four transformer encoders (TEs) and a fusion module.\nTo the best of our knowledge, HSTFormer is the first to study hierarchical TEs\nwith multi-level fusion. Extensive experiments on three datasets (i.e.,\nHuman3.6M, MPI-INF-3DHP, and HumanEva) demonstrate that HSTFormer achieves\ncompetitive and consistent performance on benchmarks with various scales and\ndifficulties. Specifically, it surpasses recent SOTAs on the challenging\nMPI-INF-3DHP dataset and small-scale HumanEva dataset, with a highly\ngeneralized systematic approach. The code is available at:\nhttps://github.com/qianxiaoye825/HSTFormer.\n", "date": "2023-01-18", "categories": []}
{"id": "2301.05435", "title": "Towards Single Camera Human 3D-Kinematics", "abstract": "  Markerless estimation of 3D Kinematics has the great potential to clinically\ndiagnose and monitor movement disorders without referrals to expensive motion\ncapture labs; however, current approaches are limited by performing multiple\nde-coupled steps to estimate the kinematics of a person from videos. Most\ncurrent techniques work in a multi-step approach by first detecting the pose of\nthe body and then fitting a musculoskeletal model to the data for accurate\nkinematic estimation. Errors in training data of the pose detection algorithms,\nmodel scaling, as well the requirement of multiple cameras limit the use of\nthese techniques in a clinical setting. Our goal is to pave the way toward\nfast, easily applicable and accurate 3D kinematic estimation \\xdeleted{in a\nclinical setting}. To this end, we propose a novel approach for direct 3D human\nkinematic estimation D3KE from videos using deep neural networks. Our\nexperiments demonstrate that the proposed end-to-end training is robust and\noutperforms 2D and 3D markerless motion capture based kinematic estimation\npipelines in terms of joint angles error by a large margin (35\\% from 5.44 to\n3.54 degrees). We show that D3KE is superior to the multi-step approach and can\nrun at video framerate speeds. This technology shows the potential for clinical\nanalysis from mobile devices in the future.\n", "date": "2023-01-13", "categories": []}
{"id": "2301.02979", "title": "CameraPose: Weakly-Supervised Monocular 3D Human Pose Estimation by\n  Leveraging In-the-wild 2D Annotations", "abstract": "  To improve the generalization of 3D human pose estimators, many existing deep\nlearning based models focus on adding different augmentations to training\nposes. However, data augmentation techniques are limited to the \"seen\" pose\ncombinations and hard to infer poses with rare \"unseen\" joint positions. To\naddress this problem, we present CameraPose, a weakly-supervised framework for\n3D human pose estimation from a single image, which can not only be applied on\n2D-3D pose pairs but also on 2D alone annotations. By adding a camera parameter\nbranch, any in-the-wild 2D annotations can be fed into our pipeline to boost\nthe training diversity and the 3D poses can be implicitly learned by\nreprojecting back to 2D. Moreover, CameraPose introduces a refinement network\nmodule with confidence-guided loss to further improve the quality of noisy 2D\nkeypoints extracted by 2D pose estimators. Experimental results demonstrate\nthat the CameraPose brings in clear improvements on cross-scenario datasets.\nNotably, it outperforms the baseline method by 3mm on the most challenging\ndataset 3DPW. In addition, by combining our proposed refinement network module\nwith existing 3D pose estimators, their performance can be improved in\ncross-scenario evaluation.\n", "date": "2023-01-08", "categories": []}
{"id": "2211.08805", "title": "Interacting Hand-Object Pose Estimation via Dense Mutual Attention", "abstract": "  3D hand-object pose estimation is the key to the success of many computer\nvision applications. The main focus of this task is to effectively model the\ninteraction between the hand and an object. To this end, existing works either\nrely on interaction constraints in a computationally-expensive iterative\noptimization, or consider only a sparse correlation between sampled hand and\nobject keypoints. In contrast, we propose a novel dense mutual attention\nmechanism that is able to model fine-grained dependencies between the hand and\nthe object. Specifically, we first construct the hand and object graphs\naccording to their mesh structures. For each hand node, we aggregate features\nfrom every object node by the learned attention and vice versa for each object\nnode. Thanks to such dense mutual attention, our method is able to produce\nphysically plausible poses with high quality and real-time inference speed.\nExtensive quantitative and qualitative experiments on large benchmark datasets\nshow that our method outperforms state-of-the-art methods. The code is\navailable at https://github.com/rongakowang/DenseMutualAttention.git.\n", "date": "2022-11-16", "categories": []}
{"id": "2212.14474", "title": "Learning 3D Human Pose Estimation from Dozens of Datasets using a\n  Geometry-Aware Autoencoder to Bridge Between Skeleton Formats", "abstract": "  Deep learning-based 3D human pose estimation performs best when trained on\nlarge amounts of labeled data, making combined learning from many datasets an\nimportant research direction. One obstacle to this endeavor are the different\nskeleton formats provided by different datasets, i.e., they do not label the\nsame set of anatomical landmarks. There is little prior research on how to best\nsupervise one model with such discrepant labels. We show that simply using\nseparate output heads for different skeletons results in inconsistent depth\nestimates and insufficient information sharing across skeletons. As a remedy,\nwe propose a novel affine-combining autoencoder (ACAE) method to perform\ndimensionality reduction on the number of landmarks. The discovered latent 3D\npoints capture the redundancy among skeletons, enabling enhanced information\nsharing when used for consistency regularization. Our approach scales to an\nextreme multi-dataset regime, where we use 28 3D human pose datasets to\nsupervise one model, which outperforms prior work on a range of benchmarks,\nincluding the challenging 3D Poses in the Wild (3DPW) dataset. Our code and\nmodels are available for research purposes.\n", "date": "2022-12-29", "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": "2212.12910", "title": "Learning to Estimate 3D Human Pose from Point Cloud", "abstract": "  3D pose estimation is a challenging problem in computer vision. Most of the\nexisting neural-network-based approaches address color or depth images through\nconvolution networks (CNNs). In this paper, we study the task of 3D human pose\nestimation from depth images. Different from the existing CNN-based human pose\nestimation method, we propose a deep human pose network for 3D pose estimation\nby taking the point cloud data as input data to model the surface of complex\nhuman structures. We first cast the 3D human pose estimation from 2D depth\nimages to 3D point clouds and directly predict the 3D joint position. Our\nexperiments on two public datasets show that our approach achieves higher\naccuracy than previous state-of-art methods. The reported results on both ITOP\nand EVAL datasets demonstrate the effectiveness of our method on the targeted\ntasks.\n", "date": "2022-12-25", "categories": []}
{"id": "2212.11344", "title": "Advanced Baseline for 3D Human Pose Estimation: A Two-Stage Approach", "abstract": "  Human pose estimation has been widely applied in various industries. While\nrecent decades have witnessed the introduction of many advanced two-dimensional\n(2D) human pose estimation solutions, three-dimensional (3D) human pose\nestimation is still an active research field in computer vision. Generally\nspeaking, 3D human pose estimation methods can be divided into two categories:\nsingle-stage and two-stage. In this paper, we focused on the 2D-to-3D lifting\nprocess in the two-stage methods and proposed a more advanced baseline model\nfor 3D human pose estimation, based on the existing solutions. Our improvements\ninclude optimization of machine learning models and multiple parameters, as\nwell as introduction of a weighted loss to the training model. Finally, we used\nthe Human3.6M benchmark to test the final performance and it did produce\nsatisfactory results.\n", "date": "2022-12-21", "categories": []}
{"id": "2212.08641", "title": "GFPose: Learning 3D Human Pose Prior with Gradient Fields", "abstract": "  Learning 3D human pose prior is essential to human-centered AI. Here, we\npresent GFPose, a versatile framework to model plausible 3D human poses for\nvarious applications. At the core of GFPose is a time-dependent score network,\nwhich estimates the gradient on each body joint and progressively denoises the\nperturbed 3D human pose to match a given task specification. During the\ndenoising process, GFPose implicitly incorporates pose priors in gradients and\nunifies various discriminative and generative tasks in an elegant framework.\nDespite the simplicity, GFPose demonstrates great potential in several\ndownstream tasks. Our experiments empirically show that 1) as a\nmulti-hypothesis pose estimator, GFPose outperforms existing SOTAs by 20% on\nHuman3.6M dataset. 2) as a single-hypothesis pose estimator, GFPose achieves\ncomparable results to deterministic SOTAs, even with a vanilla backbone. 3)\nGFPose is able to produce diverse and realistic samples in pose denoising,\ncompletion and generation tasks. Project page\nhttps://sites.google.com/view/gfpose/\n", "date": "2022-12-16", "categories": []}
{"id": "2212.07729", "title": "HUM3DIL: Semi-supervised Multi-modal 3D Human Pose Estimation for\n  Autonomous Driving", "abstract": "  Autonomous driving is an exciting new industry, posing important research\nquestions. Within the perception module, 3D human pose estimation is an\nemerging technology, which can enable the autonomous vehicle to perceive and\nunderstand the subtle and complex behaviors of pedestrians. While hardware\nsystems and sensors have dramatically improved over the decades -- with cars\npotentially boasting complex LiDAR and vision systems and with a growing\nexpansion of the available body of dedicated datasets for this newly available\ninformation -- not much work has been done to harness these novel signals for\nthe core problem of 3D human pose estimation. Our method, which we coin HUM3DIL\n(HUMan 3D from Images and LiDAR), efficiently makes use of these complementary\nsignals, in a semi-supervised fashion and outperforms existing methods with a\nlarge margin. It is a fast and compact model for onboard deployment.\nSpecifically, we embed LiDAR points into pixel-aligned multi-modal features,\nwhich we pass through a sequence of Transformer refinement stages. Quantitative\nexperiments on the Waymo Open Dataset support these claims, where we achieve\nstate-of-the-art results on the task of 3D pose estimation.\n", "date": "2022-12-15", "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": "2212.05788", "title": "Markerless Body Motion Capturing for 3D Character Animation based on\n  Multi-view Cameras", "abstract": "  This paper proposes a novel application system for the generation of\nthree-dimensional (3D) character animation driven by markerless human body\nmotion capturing. The entire pipeline of the system consists of five stages: 1)\nthe capturing of motion data using multiple cameras, 2) detection of the\ntwo-dimensional (2D) human body joints, 3) estimation of the 3D joints, 4)\ncalculation of bone transformation matrices, and 5) generation of character\nanimation. The main objective of this study is to generate a 3D skeleton and\nanimation for 3D characters using multi-view images captured by ordinary\ncameras. The computational complexity of the 3D skeleton reconstruction based\non 3D vision has been reduced as needed to achieve frame-by-frame motion\ncapturing. The experimental results reveal that our system can effectively and\nefficiently capture human actions and use them to animate 3D cartoon characters\nin real-time.\n", "date": "2022-12-12", "categories": []}
{"id": "2211.16951", "title": "Weakly Supervised 3D Multi-person Pose Estimation for Large-scale Scenes\n  based on Monocular Camera and Single LiDAR", "abstract": "  Depth estimation is usually ill-posed and ambiguous for monocular\ncamera-based 3D multi-person pose estimation. Since LiDAR can capture accurate\ndepth information in long-range scenes, it can benefit both the global\nlocalization of individuals and the 3D pose estimation by providing rich\ngeometry features. Motivated by this, we propose a monocular camera and single\nLiDAR-based method for 3D multi-person pose estimation in large-scale scenes,\nwhich is easy to deploy and insensitive to light. Specifically, we design an\neffective fusion strategy to take advantage of multi-modal input data,\nincluding images and point cloud, and make full use of temporal information to\nguide the network to learn natural and coherent human motions. Without relying\non any 3D pose annotations, our method exploits the inherent geometry\nconstraints of point cloud for self-supervision and utilizes 2D keypoints on\nimages for weak supervision. Extensive experiments on public datasets and our\nnewly collected dataset demonstrate the superiority and generalization\ncapability of our proposed method.\n", "date": "2022-11-30", "categories": []}
{"id": "2211.15868", "title": "Kinematic-aware Hierarchical Attention Network for Human Pose Estimation\n  in Videos", "abstract": "  Previous video-based human pose estimation methods have shown promising\nresults by leveraging aggregated features of consecutive frames. However, most\napproaches compromise accuracy to mitigate jitter or do not sufficiently\ncomprehend the temporal aspects of human motion. Furthermore, occlusion\nincreases uncertainty between consecutive frames, which results in unsmooth\nresults. To address these issues, we design an architecture that exploits the\nkeypoint kinematic features with the following components. First, we\neffectively capture the temporal features by leveraging individual keypoint's\nvelocity and acceleration. Second, the proposed hierarchical transformer\nencoder aggregates spatio-temporal dependencies and refines the 2D or 3D input\npose estimated from existing estimators. Finally, we provide an online\ncross-supervision between the refined input pose generated from the encoder and\nthe final pose from our decoder to enable joint optimization. We demonstrate\ncomprehensive results and validate the effectiveness of our model in various\ntasks: 2D pose estimation, 3D pose estimation, body mesh recovery, and sparsely\nannotated multi-human pose estimation. Our code is available at\nhttps://github.com/KyungMinJin/HANet.\n", "date": "2022-11-29", "categories": []}
{"id": "2211.16487", "title": "DiffPose: Multi-hypothesis Human Pose Estimation using Diffusion models", "abstract": "  Traditionally, monocular 3D human pose estimation employs a machine learning\nmodel to predict the most likely 3D pose for a given input image. However, a\nsingle image can be highly ambiguous and induces multiple plausible solutions\nfor the 2D-3D lifting step which results in overly confident 3D pose\npredictors. To this end, we propose \\emph{DiffPose}, a conditional diffusion\nmodel, that predicts multiple hypotheses for a given input image. In comparison\nto similar approaches, our diffusion model is straightforward and avoids\nintensive hyperparameter tuning, complex network structures, mode collapse, and\nunstable training. Moreover, we tackle a problem of the common two-step\napproach that first estimates a distribution of 2D joint locations via\njoint-wise heatmaps and consecutively approximates them based on first- or\nsecond-moment statistics. Since such a simplification of the heatmaps removes\nvalid information about possibly correct, though labeled unlikely, joint\nlocations, we propose to represent the heatmaps as a set of 2D joint candidate\nsamples. To extract information about the original distribution from these\nsamples we introduce our \\emph{embedding transformer} that conditions the\ndiffusion model. Experimentally, we show that DiffPose slightly improves upon\nthe state of the art for multi-hypothesis pose estimation for simple poses and\noutperforms it by a large margin for highly ambiguous poses.\n", "date": "2022-11-29", "categories": []}
{"id": "2211.11983", "title": "Weakly-supervised Pre-training for 3D Human Pose Estimation via\n  Perspective Knowledge", "abstract": "  Modern deep learning-based 3D pose estimation approaches require plenty of 3D\npose annotations. However, existing 3D datasets lack diversity, which limits\nthe performance of current methods and their generalization ability. Although\nexisting methods utilize 2D pose annotations to help 3D pose estimation, they\nmainly focus on extracting 2D structural constraints from 2D poses, ignoring\nthe 3D information hidden in the images. In this paper, we propose a novel\nmethod to extract weak 3D information directly from 2D images without 3D pose\nsupervision. Firstly, we utilize 2D pose annotations and perspective prior\nknowledge to generate the relationship of that keypoint is closer or farther\nfrom the camera, called relative depth. We collect a 2D pose dataset (MCPC) and\ngenerate relative depth labels. Based on MCPC, we propose a weakly-supervised\npre-training (WSP) strategy to distinguish the depth relationship between two\npoints in an image. WSP enables the learning of the relative depth of two\nkeypoints on lots of in-the-wild images, which is more capable of predicting\ndepth and generalization ability for 3D human pose estimation. After\nfine-tuning on 3D pose datasets, WSP achieves state-of-the-art results on two\nwidely-used benchmarks.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.03375", "title": "AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking\n  in Real-Time", "abstract": "  Accurate whole-body multi-person pose estimation and tracking is an important\nyet challenging topic in computer vision. To capture the subtle actions of\nhumans for complex behavior analysis, whole-body pose estimation including the\nface, body, hand and foot is essential over conventional body-only pose\nestimation. In this paper, we present AlphaPose, a system that can perform\naccurate whole-body pose estimation and tracking jointly while running in\nrealtime. To this end, we propose several new techniques: Symmetric Integral\nKeypoint Regression (SIKR) for fast and fine localization, Parametric Pose\nNon-Maximum-Suppression (P-NMS) for eliminating redundant human detections and\nPose Aware Identity Embedding for jointly pose estimation and tracking. During\ntraining, we resort to Part-Guided Proposal Generator (PGPG) and multi-domain\nknowledge distillation to further improve the accuracy. Our method is able to\nlocalize whole-body keypoints accurately and tracks humans simultaneously given\ninaccurate bounding boxes and redundant detections. We show a significant\nimprovement over current state-of-the-art methods in both speed and accuracy on\nCOCO-wholebody, COCO, PoseTrack, and our proposed Halpe-FullBody pose\nestimation dataset. Our model, source codes and dataset are made publicly\navailable at https://github.com/MVIG-SJTU/AlphaPose.\n", "date": "2022-11-07", "categories": []}
{"id": "2211.02337", "title": "UV R-CNN: Stable and Efficient Dense Human Pose Estimation", "abstract": "  Dense pose estimation is a dense 3D prediction task for instance-level human\nanalysis, aiming to map human pixels from an RGB image to a 3D surface of the\nhuman body. Due to a large amount of surface point regression, the training\nprocess appears to be easy to collapse compared to other region-based human\ninstance analyzing tasks. By analyzing the loss formulation of the existing\ndense pose estimation model, we introduce a novel point regression loss\nfunction, named Dense Points} loss to stable the training progress, and a new\nbalanced loss weighting strategy to handle the multi-task losses. With the\nabove novelties, we propose a brand new architecture, named UV R-CNN. Without\nauxiliary supervision and external knowledge from other tasks, UV R-CNN can\nhandle many complicated issues in dense pose model training progress, achieving\n65.0% $AP_{gps}$ and 66.1% $AP_{gpsm}$ on the DensePose-COCO validation subset\nwith ResNet-50-FPN feature extractor, competitive among the state-of-the-art\ndense human pose estimation methods.\n", "date": "2022-11-04", "categories": []}
{"id": "2210.04006", "title": "(Fusionformer):Exploiting the Joint Motion Synergy with Fusion Network\n  Based On Transformer for 3D Human Pose Estimation", "abstract": "  For the current 3D human pose estimation task, a group of methods mainly\nlearn the rules of 2D-3D projection from spatial and temporal correlation.\nHowever, earlier methods model the global features of the entire body joint in\nthe time domain, but ignore the motion trajectory of individual joint. The\nrecent work [29] considers that there are differences in motion between\ndifferent joints and deals with the temporal relationship of each joint\nseparately. However, we found that different joints show the same movement\ntrends under some specific actions. Therefore, our proposed Fusionformer method\nintroduces a self-trajectory module and a mutual-trajectory module based on the\nspatio-temporal module .After that, the global spatio-temporal features and\nlocal joint trajectory features are fused through a linear network in a\nparallel manner. To eliminate the influence of bad 2D poses on 3D projections,\nfinally we also introduce a pose refinement network to balance the consistency\nof 3D projections. In addition, we evaluate the proposed method on two\nbenchmark datasets (Human3.6M, MPI-INF-3DHP). Comparing our method with the\nbaseline method poseformer, the results show an improvement of 2.4% MPJPE and\n4.3% P-MPJPE on the Human3.6M dataset, respectively.\n", "date": "2022-10-08", "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": "2210.15121", "title": "Bootstrapping Human Optical Flow and Pose", "abstract": "  We propose a bootstrapping framework to enhance human optical flow and pose.\nWe show that, for videos involving humans in scenes, we can improve both the\noptical flow and the pose estimation quality of humans by considering the two\ntasks at the same time. We enhance optical flow estimates by fine-tuning them\nto fit the human pose estimates and vice versa. In more detail, we optimize the\npose and optical flow networks to, at inference time, agree with each other. We\nshow that this results in state-of-the-art results on the Human 3.6M and 3D\nPoses in the Wild datasets, as well as a human-related subset of the Sintel\ndataset, both in terms of pose estimation accuracy and the optical flow\naccuracy at human joint locations. Code available at\nhttps://github.com/ubc-vision/bootstrapping-human-optical-flow-and-pose\n", "date": "2022-10-27", "categories": []}
{"id": "2210.14165", "title": "MEEV: Body Mesh Estimation On Egocentric Video", "abstract": "  This technical report introduces our solution, MEEV, proposed to the EgoBody\nChallenge at ECCV 2022. Captured from head-mounted devices, the dataset\nconsists of human body shape and motion of interacting people. The EgoBody\ndataset has challenges such as occluded body or blurry image. In order to\novercome the challenges, MEEV is designed to exploit multiscale features for\nrich spatial information. Besides, to overcome the limited size of dataset, the\nmodel is pre-trained with the dataset aggregated 2D and 3D pose estimation\ndatasets. Achieving 82.30 for MPJPE and 92.93 for MPVPE, MEEV has won the\nEgoBody Challenge at ECCV 2022, which shows the effectiveness of the proposed\nmethod. The code is available at https://github.com/clovaai/meev\n", "date": "2022-10-21", "categories": []}
{"id": "2210.06110", "title": "Uplift and Upsample: Efficient 3D Human Pose Estimation with Uplifting\n  Transformers", "abstract": "  The state-of-the-art for monocular 3D human pose estimation in videos is\ndominated by the paradigm of 2D-to-3D pose uplifting. While the uplifting\nmethods themselves are rather efficient, the true computational complexity\ndepends on the per-frame 2D pose estimation. In this paper, we present a\nTransformer-based pose uplifting scheme that can operate on temporally sparse\n2D pose sequences but still produce temporally dense 3D pose estimates. We show\nhow masked token modeling can be utilized for temporal upsampling within\nTransformer blocks. This allows to decouple the sampling rate of input 2D poses\nand the target frame rate of the video and drastically decreases the total\ncomputational complexity. Additionally, we explore the option of pre-training\non large motion capture archives, which has been largely neglected so far. We\nevaluate our method on two popular benchmark datasets: Human3.6M and\nMPI-INF-3DHP. With an MPJPE of 45.0 mm and 46.9 mm, respectively, our proposed\nmethod can compete with the state-of-the-art while reducing inference time by a\nfactor of 12. This enables real-time throughput with variable consumer hardware\nin stationary and mobile applications. We release our code and models at\nhttps://github.com/goldbricklemon/uplift-upsample-3dhpe\n", "date": "2022-10-12", "categories": []}
{"id": "2210.11179", "title": "Multi-hypothesis 3D human pose estimation metrics favor miscalibrated\n  distributions", "abstract": "  Due to depth ambiguities and occlusions, lifting 2D poses to 3D is a highly\nill-posed problem. Well-calibrated distributions of possible poses can make\nthese ambiguities explicit and preserve the resulting uncertainty for\ndownstream tasks. This study shows that previous attempts, which account for\nthese ambiguities via multiple hypotheses generation, produce miscalibrated\ndistributions. We identify that miscalibration can be attributed to the use of\nsample-based metrics such as minMPJPE. In a series of simulations, we show that\nminimizing minMPJPE, as commonly done, should converge to the correct mean\nprediction. However, it fails to correctly capture the uncertainty, thus\nresulting in a miscalibrated distribution. To mitigate this problem, we propose\nan accurate and well-calibrated model called Conditional Graph Normalizing Flow\n(cGNFs). Our model is structured such that a single cGNF can estimate both\nconditional and marginal densities within the same model - effectively solving\na zero-shot density estimation problem. We evaluate cGNF on the Human~3.6M\ndataset and show that cGNF provides a well-calibrated distribution estimate\nwhile being close to state-of-the-art in terms of overall minMPJPE.\nFurthermore, cGNF outperforms previous methods on occluded joints while it\nremains well-calibrated.\n", "date": "2022-10-20", "categories": []}
{"id": "2208.10211", "title": "PoseBERT: A Generic Transformer Module for Temporal 3D Human Modeling", "abstract": "  Training state-of-the-art models for human pose estimation in videos requires\ndatasets with annotations that are really hard and expensive to obtain.\nAlthough transformers have been recently utilized for body pose sequence\nmodeling, related methods rely on pseudo-ground truth to augment the currently\nlimited training data available for learning such models. In this paper, we\nintroduce PoseBERT, a transformer module that is fully trained on 3D Motion\nCapture (MoCap) data via masked modeling. It is simple, generic and versatile,\nas it can be plugged on top of any image-based model to transform it in a\nvideo-based model leveraging temporal information. We showcase variants of\nPoseBERT with different inputs varying from 3D skeleton keypoints to rotations\nof a 3D parametric model for either the full body (SMPL) or just the hands\n(MANO). Since PoseBERT training is task agnostic, the model can be applied to\nseveral tasks such as pose refinement, future pose prediction or motion\ncompletion without finetuning. Our experimental results validate that adding\nPoseBERT on top of various state-of-the-art pose estimation methods\nconsistently improves their performances, while its low computational cost\nallows us to use it in a real-time demo for smoothly animating a robotic hand\nvia a webcam. Test code and models are available at\nhttps://github.com/naver/posebert.\n", "date": "2022-08-22", "categories": []}
{"id": "2210.08394", "title": "mRI: Multi-modal 3D Human Pose Estimation Dataset using mmWave, RGB-D,\n  and Inertial Sensors", "abstract": "  The ability to estimate 3D human body pose and movement, also known as human\npose estimation (HPE), enables many applications for home-based health\nmonitoring, such as remote rehabilitation training. Several possible solutions\nhave emerged using sensors ranging from RGB cameras, depth sensors,\nmillimeter-Wave (mmWave) radars, and wearable inertial sensors. Despite\nprevious efforts on datasets and benchmarks for HPE, few dataset exploits\nmultiple modalities and focuses on home-based health monitoring. To bridge the\ngap, we present mRI, a multi-modal 3D human pose estimation dataset with\nmmWave, RGB-D, and Inertial Sensors. Our dataset consists of over 160k\nsynchronized frames from 20 subjects performing rehabilitation exercises and\nsupports the benchmarks of HPE and action detection. We perform extensive\nexperiments using our dataset and delineate the strength of each modality. We\nhope that the release of mRI can catalyze the research in pose estimation,\nmulti-modal learning, and action understanding, and more importantly facilitate\nthe applications of home-based health monitoring.\n", "date": "2022-10-15", "categories": []}
{"id": "2210.08562", "title": "A New Spatio-Temporal Loss Function for 3D Motion Reconstruction and\n  Extended Temporal Metrics for Motion Evaluation", "abstract": "  We propose a new loss function that we call Laplacian loss, based on\nspatio-temporal Laplacian representation of the motion as a graph. This loss\nfunction is intended to be used in training models for motion reconstruction\nthrough 3D human pose estimation from videos. It compares the differential\ncoordinates of the joints obtained from the graph representation of the ground\ntruth against the one of the estimation. We design a fully convolutional\ntemporal network for motion reconstruction to achieve better temporal\nconsistency of estimation. We use this generic model to study the impact of our\nproposed loss function on the benchmarks provided by Human3.6M. We also make\nuse of various motion descriptors such as velocity, acceleration to make a\nthorough evaluation of the temporal consistency while comparing the results to\nsome of the state-of-the-art solutions.\n", "date": "2022-10-16", "categories": []}
{"id": "2206.09106", "title": "Embodied Scene-aware Human Pose Estimation", "abstract": "  We propose embodied scene-aware human pose estimation where we estimate 3D\nposes based on a simulated agent's proprioception and scene awareness, along\nwith external third-person observations. Unlike prior methods that often resort\nto multistage optimization, non-causal inference, and complex contact modeling\nto estimate human pose and human scene interactions, our method is one-stage,\ncausal, and recovers global 3D human poses in a simulated environment. Since 2D\nthird-person observations are coupled with the camera pose, we propose to\ndisentangle the camera pose and use a multi-step projection gradient defined in\nthe global coordinate frame as the movement cue for our embodied agent.\nLeveraging a physics simulation and prescanned scenes (e.g., 3D mesh), we\nsimulate our agent in everyday environments (library, office, bedroom, etc.)\nand equip our agent with environmental sensors to intelligently navigate and\ninteract with the geometries of the scene. Our method also relies only on 2D\nkeypoints and can be trained on synthetic datasets derived from popular human\nmotion databases. To evaluate, we use the popular H36M and PROX datasets and\nachieve high quality pose estimation on the challenging PROX dataset without\never using PROX motion sequences for training. Code and videos are available on\nthe project page.\n", "date": "2022-06-18", "categories": []}
{"id": "2210.05130", "title": "ACRNet: Attention Cube Regression Network for Multi-view Real-time 3D\n  Human Pose Estimation in Telemedicine", "abstract": "  Human pose estimation (HPE) for 3D skeleton reconstruction in telemedicine\nhas long received attention. Although the development of deep learning has made\nHPE methods in telemedicine simpler and easier to use, addressing low accuracy\nand high latency remains a big challenge. In this paper, we propose a novel\nmulti-view Attention Cube Regression Network (ACRNet), which regresses the 3D\nposition of joints in real time by aggregating informative attention points on\neach cube surface. More specially, a cube whose each surface contains uniformly\ndistributed attention points with specific coordinate values is first created\nto wrap the target from the main view. Then, our network regresses the 3D\nposition of each joint by summing and averaging the coordinates of attention\npoints on each surface after being weighted. To verify our method, we first\ntested ACRNet on the open-source ITOP dataset; meanwhile, we collected a new\nmulti-view upper body movement dataset (UBM) on the trunk support trainer\n(TruST) to validate the capability of our model in real rehabilitation\nscenarios. Experimental results demonstrate the superiority of ACRNet compared\nwith other state-of-the-art methods. We also validate the efficacy of each\nmodule in ACRNet. Furthermore, Our work analyzes the performance of ACRNet\nunder the medical monitoring indicator. Because of the high accuracy and\nrunning speed, our model is suitable for real-time telemedicine settings. The\nsource code is available at https://github.com/BoceHu/ACRNet\n", "date": "2022-10-11", "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": "2210.03659", "title": "Spatio-temporal Tendency Reasoning for Human Body Pose and Shape\n  Estimation from Videos", "abstract": "  In this paper, we present a spatio-temporal tendency reasoning (STR) network\nfor recovering human body pose and shape from videos. Previous approaches have\nfocused on how to extend 3D human datasets and temporal-based learning to\npromote accuracy and temporal smoothing. Different from them, our STR aims to\nlearn accurate and natural motion sequences in an unconstrained environment\nthrough temporal and spatial tendency and to fully excavate the spatio-temporal\nfeatures of existing video data. To this end, our STR learns the representation\nof features in the temporal and spatial dimensions respectively, to concentrate\non a more robust representation of spatio-temporal features. More specifically,\nfor efficient temporal modeling, we first propose a temporal tendency reasoning\n(TTR) module. TTR constructs a time-dimensional hierarchical residual\nconnection representation within a video sequence to effectively reason\ntemporal sequences' tendencies and retain effective dissemination of human\ninformation. Meanwhile, for enhancing the spatial representation, we design a\nspatial tendency enhancing (STE) module to further learns to excite spatially\ntime-frequency domain sensitive features in human motion information\nrepresentations. Finally, we introduce integration strategies to integrate and\nrefine the spatio-temporal feature representations. Extensive experimental\nfindings on large-scale publically available datasets reveal that our STR\nremains competitive with the state-of-the-art on three datasets. Our code are\navailable at https://github.com/Changboyang/STR.git.\n", "date": "2022-10-07", "categories": []}
{"id": "2210.04014", "title": "AdaptivePose++: A Powerful Single-Stage Network for Multi-Person Pose\n  Regression", "abstract": "  Multi-person pose estimation generally follows top-down and bottom-up\nparadigms. Both of them use an extra stage ($\\boldsymbol{e.g.,}$ human\ndetection in top-down paradigm or grouping process in bottom-up paradigm) to\nbuild the relationship between the human instance and corresponding keypoints,\nthus leading to the high computation cost and redundant two-stage pipeline. To\naddress the above issue, we propose to represent the human parts as adaptive\npoints and introduce a fine-grained body representation method. The novel body\nrepresentation is able to sufficiently encode the diverse pose information and\neffectively model the relationship between the human instance and corresponding\nkeypoints in a single-forward pass. With the proposed body representation, we\nfurther deliver a compact single-stage multi-person pose regression network,\ntermed as AdaptivePose. During inference, our proposed network only needs a\nsingle-step decode operation to form the multi-person pose without complex\npost-processes and refinements. We employ AdaptivePose for both 2D/3D\nmulti-person pose estimation tasks to verify the effectiveness of AdaptivePose.\nWithout any bells and whistles, we achieve the most competitive performance on\nMS COCO and CrowdPose in terms of accuracy and speed. Furthermore, the\noutstanding performance on MuCo-3DHP and MuPoTS-3D further demonstrates the\neffectiveness and generalizability on 3D scenes. Code is available at\nhttps://github.com/buptxyb666/AdaptivePose.\n", "date": "2022-10-08", "categories": []}
{"id": "2210.04514", "title": "Self-Supervised 3D Human Pose Estimation in Static Video Via Neural\n  Rendering", "abstract": "  Inferring 3D human pose from 2D images is a challenging and long-standing\nproblem in the field of computer vision with many applications including motion\ncapture, virtual reality, surveillance or gait analysis for sports and\nmedicine. We present preliminary results for a method to estimate 3D pose from\n2D video containing a single person and a static background without the need\nfor any manual landmark annotations. We achieve this by formulating a simple\nyet effective self-supervision task: our model is required to reconstruct a\nrandom frame of a video given a frame from another timepoint and a rendered\nimage of a transformed human shape template. Crucially for optimisation, our\nray casting based rendering pipeline is fully differentiable, enabling end to\nend training solely based on the reconstruction task.\n", "date": "2022-10-10", "categories": []}
{"id": "2210.01886", "title": "Multi-view Human Body Mesh Translator", "abstract": "  Existing methods for human mesh recovery mainly focus on single-view\nframeworks, but they often fail to produce accurate results due to the\nill-posed setup. Considering the maturity of the multi-view motion capture\nsystem, in this paper, we propose to solve the prior ill-posed problem by\nleveraging multiple images from different views, thus significantly enhancing\nthe quality of recovered meshes. In particular, we present a novel\n\\textbf{M}ulti-view human body \\textbf{M}esh \\textbf{T}ranslator (MMT) model\nfor estimating human body mesh with the help of vision transformer.\nSpecifically, MMT takes multi-view images as input and translates them to\ntargeted meshes in a single-forward manner. MMT fuses features of different\nviews in both encoding and decoding phases, leading to representations embedded\nwith global information. Additionally, to ensure the tokens are intensively\nfocused on the human pose and shape, MMT conducts cross-view alignment at the\nfeature level by projecting 3D keypoint positions to each view and enforcing\ntheir consistency in geometry constraints. Comprehensive experiments\ndemonstrate that MMT outperforms existing single or multi-view models by a\nlarge margin for human mesh recovery task, notably, 28.8\\% improvement in MPVE\nover the current state-of-the-art method on the challenging HUMBI dataset.\nQualitative evaluation also verifies the effectiveness of MMT in reconstructing\nhigh-quality human mesh. Codes will be made available upon acceptance.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.02231", "title": "Decanus to Legatus: Synthetic training for 2D-3D human pose lifting", "abstract": "  3D human pose estimation is a challenging task because of the difficulty to\nacquire ground-truth data outside of controlled environments. A number of\nfurther issues have been hindering progress in building a universal and robust\nmodel for this task, including domain gaps between different datasets, unseen\nactions between train and test datasets, various hardware settings and high\ncost of annotation, etc. In this paper, we propose an algorithm to generate\ninfinite 3D synthetic human poses (Legatus) from a 3D pose distribution based\non 10 initial handcrafted 3D poses (Decanus) during the training of a 2D to 3D\nhuman pose lifter neural network. Our results show that we can achieve 3D pose\nestimation performance comparable to methods using real data from specialized\ndatasets but in a zero-shot setup, showing the generalization potential of our\nframework.\n", "date": "2022-10-05", "categories": []}
{"id": "2210.01721", "title": "MBW: Multi-view Bootstrapping in the Wild", "abstract": "  Labeling articulated objects in unconstrained settings have a wide variety of\napplications including entertainment, neuroscience, psychology, ethology, and\nmany fields of medicine. Large offline labeled datasets do not exist for all\nbut the most common articulated object categories (e.g., humans). Hand labeling\nthese landmarks within a video sequence is a laborious task. Learned landmark\ndetectors can help, but can be error-prone when trained from only a few\nexamples. Multi-camera systems that train fine-grained detectors have shown\nsignificant promise in detecting such errors, allowing for self-supervised\nsolutions that only need a small percentage of the video sequence to be\nhand-labeled. The approach, however, is based on calibrated cameras and rigid\ngeometry, making it expensive, difficult to manage, and impractical in\nreal-world scenarios. In this paper, we address these bottlenecks by combining\na non-rigid 3D neural prior with deep flow to obtain high-fidelity landmark\nestimates from videos with only two or three uncalibrated, handheld cameras.\nWith just a few annotations (representing 1-2% of the frames), we are able to\nproduce 2D results comparable to state-of-the-art fully supervised methods,\nalong with 3D reconstructions that are impossible with other existing\napproaches. Our Multi-view Bootstrapping in the Wild (MBW) approach\ndemonstrates impressive results on standard human datasets, as well as tigers,\ncheetahs, fish, colobus monkeys, chimpanzees, and flamingos from videos\ncaptured casually in a zoo. We release the codebase for MBW as well as this\nchallenging zoo dataset consisting image frames of tail-end distribution\ncategories with their corresponding 2D, 3D labels generated from minimal human\nintervention.\n", "date": "2022-10-04", "categories": []}
{"id": "2209.12354", "title": "InterCap: Joint Markerless 3D Tracking of Humans and Objects in\n  Interaction", "abstract": "  Humans constantly interact with daily objects to accomplish tasks. To\nunderstand such interactions, computers need to reconstruct these from cameras\nobserving whole-body interaction with scenes. This is challenging due to\nocclusion between the body and objects, motion blur, depth/scale ambiguities,\nand the low image resolution of hands and graspable object parts. To make the\nproblem tractable, the community focuses either on interacting hands, ignoring\nthe body, or on interacting bodies, ignoring hands. The GRAB dataset addresses\ndexterous whole-body interaction but uses marker-based MoCap and lacks images,\nwhile BEHAVE captures video of body object interaction but lacks hand detail.\nWe address the limitations of prior work with InterCap, a novel method that\nreconstructs interacting whole-bodies and objects from multi-view RGB-D data,\nusing the parametric whole-body model SMPL-X and known object meshes. To tackle\nthe above challenges, InterCap uses two key observations: (i) Contact between\nthe hand and object can be used to improve the pose estimation of both. (ii)\nAzure Kinect sensors allow us to set up a simple multi-view RGB-D capture\nsystem that minimizes the effect of occlusion while providing reasonable\ninter-camera synchronization. With this method we capture the InterCap dataset,\nwhich contains 10 subjects (5 males and 5 females) interacting with 10 objects\nof various sizes and affordances, including contact with the hands or feet. In\ntotal, InterCap has 223 RGB-D videos, resulting in 67,357 multi-view frames,\neach containing 6 RGB-D images. Our method provides pseudo ground-truth body\nmeshes and objects for each video frame. Our InterCap method and dataset fill\nan important gap in the literature and support many research directions. Our\ndata and code are areavailable for research purposes.\n", "date": "2022-09-26", "categories": []}
{"id": "2208.11328", "title": "K-Order Graph-oriented Transformer with GraAttention for 3D Pose and\n  Shape Estimation", "abstract": "  We propose a novel attention-based 2D-to-3D pose estimation network for\ngraph-structured data, named KOG-Transformer, and a 3D pose-to-shape estimation\nnetwork for hand data, named GASE-Net. Previous 3D pose estimation methods have\nfocused on various modifications to the graph convolution kernel, such as\nabandoning weight sharing or increasing the receptive field. Some of these\nmethods employ attention-based non-local modules as auxiliary modules. In order\nto better model the relationship between nodes in graph-structured data and\nfuse the information of different neighbor nodes in a differentiated way, we\nmake targeted modifications to the attention module and propose two modules\ndesigned for graph-structured data, graph relative positional encoding\nmulti-head self-attention (GR-MSA) and K-order graph-oriented multi-head\nself-attention (KOG-MSA). By stacking GR-MSA and KOG-MSA, we propose a novel\nnetwork KOG-Transformer for 2D-to-3D pose estimation. Furthermore, we propose a\nnetwork for shape estimation on hand data, called GraAttention shape estimation\nnetwork (GASE-Net), which takes a 3D pose as input and gradually models the\nshape of the hand from sparse to dense. We have empirically shown the\nsuperiority of KOG-Transformer through extensive experiments. Experimental\nresults show that KOG-Transformer significantly outperforms the previous\nstate-of-the-art methods on the benchmark dataset Human3.6M. We evaluate the\neffect of GASE-Net on two public available hand datasets, ObMan and\nInterHand2.6M. GASE-Net can predict the corresponding shape for input pose with\nstrong generalization ability.\n", "date": "2022-08-24", "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": "2209.08194", "title": "PPT: token-Pruned Pose Transformer for monocular and multi-view human\n  pose estimation", "abstract": "  Recently, the vision transformer and its variants have played an increasingly\nimportant role in both monocular and multi-view human pose estimation.\nConsidering image patches as tokens, transformers can model the global\ndependencies within the entire image or across images from other views.\nHowever, global attention is computationally expensive. As a consequence, it is\ndifficult to scale up these transformer-based methods to high-resolution\nfeatures and many views.\n  In this paper, we propose the token-Pruned Pose Transformer (PPT) for 2D\nhuman pose estimation, which can locate a rough human mask and performs\nself-attention only within selected tokens. Furthermore, we extend our PPT to\nmulti-view human pose estimation. Built upon PPT, we propose a new cross-view\nfusion strategy, called human area fusion, which considers all human foreground\npixels as corresponding candidates. Experimental results on COCO and MPII\ndemonstrate that our PPT can match the accuracy of previous pose transformer\nmethods while reducing the computation. Moreover, experiments on Human 3.6M and\nSki-Pose demonstrate that our Multi-view PPT can efficiently fuse cues from\nmultiple views and achieve new state-of-the-art results.\n", "date": "2022-09-16", "categories": []}
{"id": "2209.08790", "title": "D&D: Learning Human Dynamics from Dynamic Camera", "abstract": "  3D human pose estimation from a monocular video has recently seen significant\nimprovements. However, most state-of-the-art methods are kinematics-based,\nwhich are prone to physically implausible motions with pronounced artifacts.\nCurrent dynamics-based methods can predict physically plausible motion but are\nrestricted to simple scenarios with static camera view. In this work, we\npresent D&D (Learning Human Dynamics from Dynamic Camera), which leverages the\nlaws of physics to reconstruct 3D human motion from the in-the-wild videos with\na moving camera. D&D introduces inertial force control (IFC) to explain the 3D\nhuman motion in the non-inertial local frame by considering the inertial forces\nof the dynamic camera. To learn the ground contact with limited annotations, we\ndevelop probabilistic contact torque (PCT), which is computed by differentiable\nsampling from contact probabilities and used to generate motions. The contact\nstate can be weakly supervised by encouraging the model to generate correct\nmotions. Furthermore, we propose an attentive PD controller that adjusts target\npose states using temporal information to obtain smooth and accurate pose\ncontrol. Our approach is entirely neural-based and runs without offline\noptimization or simulation in physics engines. Experiments on large-scale 3D\nhuman motion benchmarks demonstrate the effectiveness of D&D, where we exhibit\nsuperior performance against both state-of-the-art kinematics-based and\ndynamics-based methods. Code is available at https://github.com/Jeffsjtu/DnD\n", "date": "2022-09-19", "categories": []}
{"id": "2207.12841", "title": "KinePose: A temporally optimized inverse kinematics technique for 6DOF\n  human pose estimation with biomechanical constraints", "abstract": "  Computer vision/deep learning-based 3D human pose estimation methods aim to\nlocalize human joints from images and videos. Pose representation is normally\nlimited to 3D joint positional/translational degrees of freedom (3DOFs),\nhowever, a further three rotational DOFs (6DOFs) are required for many\npotential biomechanical applications. Positional DOFs are insufficient to\nanalytically solve for joint rotational DOFs in a 3D human skeletal model.\nTherefore, we propose a temporal inverse kinematics (IK) optimization technique\nto infer joint orientations throughout a biomechanically informed, and\nsubject-specific kinematic chain. For this, we prescribe link directions from a\nposition-based 3D pose estimate. Sequential least squares quadratic programming\nis used to solve a minimization problem that involves both frame-based pose\nterms, and a temporal term. The solution space is constrained using joint DOFs,\nand ranges of motion (ROMs). We generate 3D pose motion sequences to assess the\nIK approach both for general accuracy, and accuracy in boundary cases. Our\ntemporal algorithm achieves 6DOF pose estimates with low Mean Per Joint Angular\nSeparation (MPJAS) errors (3.7{\\deg}/joint overall, & 1.6{\\deg}/joint for lower\nlimbs). With frame-by-frame IK we obtain low errors in the case of bent elbows\nand knees, however, motion sequences with phases of extended/straight limbs\nresults in ambiguity in twist angle. With temporal IK, we reduce ambiguity for\nthese poses, resulting in lower average errors.\n", "date": "2022-07-26", "categories": []}
{"id": "2209.02431", "title": "DPIT: Dual-Pipeline Integrated Transformer for Human Pose Estimation", "abstract": "  Human pose estimation aims to figure out the keypoints of all people in\ndifferent scenes. Current approaches still face some challenges despite\npromising results. Existing top-down methods deal with a single person\nindividually, without the interaction between different people and the scene\nthey are situated in. Consequently, the performance of human detection degrades\nwhen serious occlusion happens. On the other hand, existing bottom-up methods\nconsider all people at the same time and capture the global knowledge of the\nentire image. However, they are less accurate than the top-down methods due to\nthe scale variation. To address these problems, we propose a novel\nDual-Pipeline Integrated Transformer (DPIT) by integrating top-down and\nbottom-up pipelines to explore the visual clues of different receptive fields\nand achieve their complementarity. Specifically, DPIT consists of two branches,\nthe bottom-up branch deals with the whole image to capture the global visual\ninformation, while the top-down branch extracts the feature representation of\nlocal vision from the single-human bounding box. Then, the extracted feature\nrepresentations from bottom-up and top-down branches are fed into the\ntransformer encoder to fuse the global and local knowledge interactively.\nMoreover, we define the keypoint queries to explore both full-scene and\nsingle-human posture visual clues to realize the mutual complementarity of the\ntwo pipelines. To the best of our knowledge, this is one of the first works to\nintegrate the bottom-up and top-down pipelines with transformers for human pose\nestimation. Extensive experiments on COCO and MPII datasets demonstrate that\nour DPIT achieves comparable performance to the state-of-the-art methods.\n", "date": "2022-09-02", "categories": []}
{"id": "2209.02485", "title": "Reconstructing Action-Conditioned Human-Object Interactions Using\n  Commonsense Knowledge Priors", "abstract": "  We present a method for inferring diverse 3D models of human-object\ninteractions from images. Reasoning about how humans interact with objects in\ncomplex scenes from a single 2D image is a challenging task given ambiguities\narising from the loss of information through projection. In addition, modeling\n3D interactions requires the generalization ability towards diverse object\ncategories and interaction types. We propose an action-conditioned modeling of\ninteractions that allows us to infer diverse 3D arrangements of humans and\nobjects without supervision on contact regions or 3D scene geometry. Our method\nextracts high-level commonsense knowledge from large language models (such as\nGPT-3), and applies them to perform 3D reasoning of human-object interactions.\nOur key insight is priors extracted from large language models can help in\nreasoning about human-object contacts from textural prompts only. We\nquantitatively evaluate the inferred 3D models on a large human-object\ninteraction dataset and show how our method leads to better 3D reconstructions.\nWe further qualitatively evaluate the effectiveness of our method on real\nimages and demonstrate its generalizability towards interaction types and\nobject categories.\n", "date": "2022-09-06", "categories": []}
{"id": "2209.00618", "title": "Optimising 2D Pose Representation: Improve Accuracy, Stability and\n  Generalisability Within Unsupervised 2D-3D Human Pose Estimation", "abstract": "  This paper addresses the problem of 2D pose representation during\nunsupervised 2D to 3D pose lifting to improve the accuracy, stability and\ngeneralisability of 3D human pose estimation (HPE) models. All unsupervised\n2D-3D HPE approaches provide the entire 2D kinematic skeleton to a model during\ntraining. We argue that this is sub-optimal and disruptive as long-range\ncorrelations are induced between independent 2D key points and predicted 3D\nordinates during training. To this end, we conduct the following study. With a\nmaximum architecture capacity of 6 residual blocks, we evaluate the performance\nof 5 models which each represent a 2D pose differently during the adversarial\nunsupervised 2D-3D HPE process. Additionally, we show the correlations between\n2D key points which are learned during the training process, highlighting the\nunintuitive correlations induced when an entire 2D pose is provided to a\nlifting model. Our results show that the most optimal representation of a 2D\npose is that of two independent segments, the torso and legs, with no shared\nfeatures between each lifting network. This approach decreased the average\nerror by 20\\% on the Human3.6M dataset when compared to a model with a near\nidentical parameter count trained on the entire 2D kinematic skeleton.\nFurthermore, due to the complex nature of adversarial learning, we show how\nthis representation can also improve convergence during training allowing for\nan optimum result to be obtained more often.\n", "date": "2022-09-01", "categories": []}
{"id": "2206.04511", "title": "Efficient Human Pose Estimation via 3D Event Point Cloud", "abstract": "  Human Pose Estimation (HPE) based on RGB images has experienced a rapid\ndevelopment benefiting from deep learning. However, event-based HPE has not\nbeen fully studied, which remains great potential for applications in extreme\nscenes and efficiency-critical conditions. In this paper, we are the first to\nestimate 2D human pose directly from 3D event point cloud. We propose a novel\nrepresentation of events, the rasterized event point cloud, aggregating events\non the same position of a small time slice. It maintains the 3D features from\nmultiple statistical cues and significantly reduces memory consumption and\ncomputation complexity, proved to be efficient in our work. We then leverage\nthe rasterized event point cloud as input to three different backbones,\nPointNet, DGCNN, and Point Transformer, with two linear layer decoders to\npredict the location of human keypoints. We find that based on our method,\nPointNet achieves promising results with much faster speed, whereas Point\nTransfomer reaches much higher accuracy, even close to previous\nevent-frame-based methods. A comprehensive set of results demonstrates that our\nproposed method is consistently effective for these 3D backbone models in\nevent-driven human pose estimation. Our method based on PointNet with 2048\npoints input achieves 82.46mm in MPJPE3D on the DHP19 dataset, while only has a\nlatency of 12.29ms on an NVIDIA Jetson Xavier NX edge computing platform, which\nis ideally suitable for real-time detection with event cameras. Code is\navailable at https://github.com/MasterHow/EventPointPose.\n", "date": "2022-06-09", "categories": []}
{"id": "2208.11960", "title": "FusePose: IMU-Vision Sensor Fusion in Kinematic Space for Parametric\n  Human Pose Estimation", "abstract": "  There exist challenging problems in 3D human pose estimation mission, such as\npoor performance caused by occlusion and self-occlusion. Recently, IMU-vision\nsensor fusion is regarded as valuable for solving these problems. However,\nprevious researches on the fusion of IMU and vision data, which is\nheterogeneous, fail to adequately utilize either IMU raw data or reliable\nhigh-level vision features. To facilitate a more efficient sensor fusion, in\nthis work we propose a framework called \\emph{FusePose} under a parametric\nhuman kinematic model. Specifically, we aggregate different information of IMU\nor vision data and introduce three distinctive sensor fusion approaches:\nNaiveFuse, KineFuse and AdaDeepFuse. NaiveFuse servers as a basic approach that\nonly fuses simplified IMU data and estimated 3D pose in euclidean space. While\nin kinematic space, KineFuse is able to integrate the calibrated and aligned\nIMU raw data with converted 3D pose parameters. AdaDeepFuse further develops\nthis kinematical fusion process to an adaptive and end-to-end trainable manner.\nComprehensive experiments with ablation studies demonstrate the rationality and\nsuperiority of the proposed framework. The performance of 3D human pose\nestimation is improved compared to the baseline result. On Total Capture\ndataset, KineFuse surpasses previous state-of-the-art which uses IMU only for\ntesting by 8.6\\%. AdaDeepFuse surpasses state-of-the-art which uses IMU for\nboth training and testing by 8.5\\%. Moreover, we validate the generalization\ncapability of our framework through experiments on Human3.6M dataset.\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": "2208.08439", "title": "MoCapDeform: Monocular 3D Human Motion Capture in Deformable Scenes", "abstract": "  3D human motion capture from monocular RGB images respecting interactions of\na subject with complex and possibly deformable environments is a very\nchallenging, ill-posed and under-explored problem. Existing methods address it\nonly weakly and do not model possible surface deformations often occurring when\nhumans interact with scene surfaces. In contrast, this paper proposes\nMoCapDeform, i.e., a new framework for monocular 3D human motion capture that\nis the first to explicitly model non-rigid deformations of a 3D scene for\nimproved 3D human pose estimation and deformable environment reconstruction.\nMoCapDeform accepts a monocular RGB video and a 3D scene mesh aligned in the\ncamera space. It first localises a subject in the input monocular video along\nwith dense contact labels using a new raycasting based strategy. Next, our\nhuman-environment interaction constraints are leveraged to jointly optimise\nglobal 3D human poses and non-rigid surface deformations. MoCapDeform achieves\nsuperior accuracy than competing methods on several datasets, including our\nnewly recorded one with deforming background scenes.\n", "date": "2022-08-17", "categories": []}
{"id": "2208.07755", "title": "PoseTrans: A Simple Yet Effective Pose Transformation Augmentation for\n  Human Pose Estimation", "abstract": "  Human pose estimation aims to accurately estimate a wide variety of human\nposes. However, existing datasets often follow a long-tailed distribution that\nunusual poses only occupy a small portion, which further leads to the lack of\ndiversity of rare poses. These issues result in the inferior generalization\nability of current pose estimators. In this paper, we present a simple yet\neffective data augmentation method, termed Pose Transformation (PoseTrans), to\nalleviate the aforementioned problems. Specifically, we propose Pose\nTransformation Module (PTM) to create new training samples that have diverse\nposes and adopt a pose discriminator to ensure the plausibility of the\naugmented poses. Besides, we propose Pose Clustering Module (PCM) to measure\nthe pose rarity and select the \"rarest\" poses to help balance the long-tailed\ndistribution. Extensive experiments on three benchmark datasets demonstrate the\neffectiveness of our method, especially on rare poses. Also, our method is\nefficient and simple to implement, which can be easily integrated into the\ntraining pipeline of existing pose estimation models.\n", "date": "2022-08-16", "categories": []}
{"id": "2206.07510", "title": "Deep Multi-Task Networks For Occluded Pedestrian Pose Estimation", "abstract": "  Most of the existing works on pedestrian pose estimation do not consider\nestimating the pose of an occluded pedestrian, as the annotations of the\noccluded parts are not available in relevant automotive datasets. For example,\nCityPersons, a well-known dataset for pedestrian detection in automotive scenes\ndoes not provide pose annotations, whereas MS-COCO, a non-automotive dataset,\ncontains human pose estimation. In this work, we propose a multi-task framework\nto extract pedestrian features through detection and instance segmentation\ntasks performed separately on these two distributions. Thereafter, an encoder\nlearns pose specific features using an unsupervised instance-level domain\nadaptation method for the pedestrian instances from both distributions. The\nproposed framework has improved state-of-the-art performances of pose\nestimation, pedestrian detection, and instance segmentation.\n", "date": "2022-06-15", "categories": []}
{"id": "2207.11341", "title": "Dynamic Graph Reasoning for Multi-person 3D Pose Estimation", "abstract": "  Multi-person 3D pose estimation is a challenging task because of occlusion\nand depth ambiguity, especially in the cases of crowd scenes. To solve these\nproblems, most existing methods explore modeling body context cues by enhancing\nfeature representation with graph neural networks or adding structural\nconstraints. However, these methods are not robust for their single-root\nformulation that decoding 3D poses from a root node with a pre-defined graph.\nIn this paper, we propose GR-M3D, which models the \\textbf{M}ulti-person\n\\textbf{3D} pose estimation with dynamic \\textbf{G}raph \\textbf{R}easoning. The\ndecoding graph in GR-M3D is predicted instead of pre-defined. In particular, It\nfirstly generates several data maps and enhances them with a scale and depth\naware refinement module (SDAR). Then multiple root keypoints and dense decoding\npaths for each person are estimated from these data maps. Based on them,\ndynamic decoding graphs are built by assigning path weights to the decoding\npaths, while the path weights are inferred from those enhanced data maps. And\nthis process is named dynamic graph reasoning (DGR). Finally, the 3D poses are\ndecoded according to dynamic decoding graphs for each detected person. GR-M3D\ncan adjust the structure of the decoding graph implicitly by adopting soft path\nweights according to input data, which makes the decoding graphs be adaptive to\ndifferent input persons to the best extent and more capable of handling\nocclusion and depth ambiguity than previous methods. We empirically show that\nthe proposed bottom-up approach even outperforms top-down methods and achieves\nstate-of-the-art results on three 3D pose datasets.\n", "date": "2022-07-22", "categories": []}
{"id": "2208.03431", "title": "IVT: An End-to-End Instance-guided Video Transformer for 3D Pose\n  Estimation", "abstract": "  Video 3D human pose estimation aims to localize the 3D coordinates of human\njoints from videos. Recent transformer-based approaches focus on capturing the\nspatiotemporal information from sequential 2D poses, which cannot model the\ncontextual depth feature effectively since the visual depth features are lost\nin the step of 2D pose estimation. In this paper, we simplify the paradigm into\nan end-to-end framework, Instance-guided Video Transformer (IVT), which enables\nlearning spatiotemporal contextual depth information from visual features\neffectively and predicts 3D poses directly from video frames. In particular, we\nfirstly formulate video frames as a series of instance-guided tokens and each\ntoken is in charge of predicting the 3D pose of a human instance. These tokens\ncontain body structure information since they are extracted by the guidance of\njoint offsets from the human center to the corresponding body joints. Then,\nthese tokens are sent into IVT for learning spatiotemporal contextual depth. In\naddition, we propose a cross-scale instance-guided attention mechanism to\nhandle the variational scales among multiple persons. Finally, the 3D poses of\neach person are decoded from instance-guided tokens by coordinate regression.\nExperiments on three widely-used 3D pose estimation benchmarks show that the\nproposed IVT achieves state-of-the-art performances.\n", "date": "2022-08-06", "categories": []}
{"id": "2208.03704", "title": "Jointformer: Single-Frame Lifting Transformer with Error Prediction and\n  Refinement for 3D Human Pose Estimation", "abstract": "  Monocular 3D human pose estimation technologies have the potential to greatly\nincrease the availability of human movement data. The best-performing models\nfor single-image 2D-3D lifting use graph convolutional networks (GCNs) that\ntypically require some manual input to define the relationships between\ndifferent body joints. We propose a novel transformer-based approach that uses\nthe more generalised self-attention mechanism to learn these relationships\nwithin a sequence of tokens representing joints. We find that the use of\nintermediate supervision, as well as residual connections between the stacked\nencoders benefits performance. We also suggest that using error prediction as\npart of a multi-task learning framework improves performance by allowing the\nnetwork to compensate for its confidence level. We perform extensive ablation\nstudies to show that each of our contributions increases performance.\nFurthermore, we show that our approach outperforms the recent state of the art\nfor single-frame 3D human pose estimation by a large margin. Our code and\ntrained models are made publicly available on Github.\n", "date": "2022-08-07", "categories": []}
{"id": "2208.01633", "title": "UnrealEgo: A New Dataset for Robust Egocentric 3D Human Motion Capture", "abstract": "  We present UnrealEgo, i.e., a new large-scale naturalistic dataset for\negocentric 3D human pose estimation. UnrealEgo is based on an advanced concept\nof eyeglasses equipped with two fisheye cameras that can be used in\nunconstrained environments. We design their virtual prototype and attach them\nto 3D human models for stereo view capture. We next generate a large corpus of\nhuman motions. As a consequence, UnrealEgo is the first dataset to provide\nin-the-wild stereo images with the largest variety of motions among existing\negocentric datasets. Furthermore, we propose a new benchmark method with a\nsimple but effective idea of devising a 2D keypoint estimation module for\nstereo inputs to improve 3D human pose estimation. The extensive experiments\nshow that our approach outperforms the previous state-of-the-art methods\nqualitatively and quantitatively. UnrealEgo and our source codes are available\non our project web page.\n", "date": "2022-08-02", "categories": []}
{"id": "2208.00090", "title": "Explicit Occlusion Reasoning for Multi-person 3D Human Pose Estimation", "abstract": "  Occlusion poses a great threat to monocular multi-person 3D human pose\nestimation due to large variability in terms of the shape, appearance, and\nposition of occluders. While existing methods try to handle occlusion with pose\npriors/constraints, data augmentation, or implicit reasoning, they still fail\nto generalize to unseen poses or occlusion cases and may make large mistakes\nwhen multiple people are present. Inspired by the remarkable ability of humans\nto infer occluded joints from visible cues, we develop a method to explicitly\nmodel this process that significantly improves bottom-up multi-person human\npose estimation with or without occlusions. First, we split the task into two\nsubtasks: visible keypoints detection and occluded keypoints reasoning, and\npropose a Deeply Supervised Encoder Distillation (DSED) network to solve the\nsecond one. To train our model, we propose a Skeleton-guided human Shape\nFitting (SSF) approach to generate pseudo occlusion labels on the existing\ndatasets, enabling explicit occlusion reasoning. Experiments show that\nexplicitly learning from occlusions improves human pose estimation. In\naddition, exploiting feature-level information of visible joints allows us to\nreason about occluded joints more accurately. Our method outperforms both the\nstate-of-the-art top-down and bottom-up methods on several benchmarks.\n", "date": "2022-07-29", "categories": []}
{"id": "2207.13784", "title": "AvatarPoser: Articulated Full-Body Pose Tracking from Sparse Motion\n  Sensing", "abstract": "  Today's Mixed Reality head-mounted displays track the user's head pose in\nworld space as well as the user's hands for interaction in both Augmented\nReality and Virtual Reality scenarios. While this is adequate to support user\ninput, it unfortunately limits users' virtual representations to just their\nupper bodies. Current systems thus resort to floating avatars, whose limitation\nis particularly evident in collaborative settings. To estimate full-body poses\nfrom the sparse input sources, prior work has incorporated additional trackers\nand sensors at the pelvis or lower body, which increases setup complexity and\nlimits practical application in mobile settings. In this paper, we present\nAvatarPoser, the first learning-based method that predicts full-body poses in\nworld coordinates using only motion input from the user's head and hands. Our\nmethod builds on a Transformer encoder to extract deep features from the input\nsignals and decouples global motion from the learned local joint orientations\nto guide pose estimation. To obtain accurate full-body motions that resemble\nmotion capture animations, we refine the arm joints' positions using an\noptimization routine with inverse kinematics to match the original tracking\ninput. In our evaluation, AvatarPoser achieved new state-of-the-art results in\nevaluations on large motion capture datasets (AMASS). At the same time, our\nmethod's inference speed supports real-time operation, providing a practical\ninterface to support holistic avatar control and representation for Metaverse\napplications.\n", "date": "2022-07-27", "categories": []}
{"id": "2207.13807", "title": "Pose-NDF: Modeling Human Pose Manifolds with Neural Distance Fields", "abstract": "  We present Pose-NDF, a continuous model for plausible human poses based on\nneural distance fields (NDFs). Pose or motion priors are important for\ngenerating realistic new poses and for reconstructing accurate poses from noisy\nor partial observations. Pose-NDF learns a manifold of plausible poses as the\nzero level set of a neural implicit function, extending the idea of modeling\nimplicit surfaces in 3D to the high-dimensional domain SO(3)^K, where a human\npose is defined by a single data point, represented by K quaternions. The\nresulting high-dimensional implicit function can be differentiated with respect\nto the input poses and thus can be used to project arbitrary poses onto the\nmanifold by using gradient descent on the set of 3-dimensional hyperspheres. In\ncontrast to previous VAE-based human pose priors, which transform the pose\nspace into a Gaussian distribution, we model the actual pose manifold,\npreserving the distances between poses. We demonstrate that PoseNDF outperforms\nexisting state-of-the-art methods as a prior in various downstream tasks,\nranging from denoising real-world human mocap data, pose recovery from occluded\ndata to 3D pose reconstruction from images. Furthermore, we show that it can be\nused to generate more diverse poses by random sampling and projection than\nVAE-based methods.\n", "date": "2022-07-27", "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": "2207.14279", "title": "The One Where They Reconstructed 3D Humans and Environments in TV Shows", "abstract": "  TV shows depict a wide variety of human behaviors and have been studied\nextensively for their potential to be a rich source of data for many\napplications. However, the majority of the existing work focuses on 2D\nrecognition tasks. In this paper, we make the observation that there is a\ncertain persistence in TV shows, i.e., repetition of the environments and the\nhumans, which makes possible the 3D reconstruction of this content. Building on\nthis insight, we propose an automatic approach that operates on an entire\nseason of a TV show and aggregates information in 3D; we build a 3D model of\nthe environment, compute camera information, static 3D scene structure and body\nscale information. Then, we demonstrate how this information acts as rich 3D\ncontext that can guide and improve the recovery of 3D human pose and position\nin these environments. Moreover, we show that reasoning about humans and their\nenvironment in 3D enables a broad range of downstream applications:\nre-identification, gaze estimation, cinematography and image editing. We apply\nour approach on environments from seven iconic TV shows and perform an\nextensive evaluation of the proposed system.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.12537", "title": "Live Stream Temporally Embedded 3D Human Body Pose and Shape Estimation", "abstract": "  3D Human body pose and shape estimation within a temporal sequence can be\nquite critical for understanding human behavior. Despite the significant\nprogress in human pose estimation in the recent years, which are often based on\nsingle images or videos, human motion estimation on live stream videos is still\na rarely-touched area considering its special requirements for real-time output\nand temporal consistency. To address this problem, we present a temporally\nembedded 3D human body pose and shape estimation (TePose) method to improve the\naccuracy and temporal consistency of pose estimation in live stream videos.\nTePose uses previous predictions as a bridge to feedback the error for better\nestimation in the current frame and to learn the correspondence between data\nframes and predictions in the history. A multi-scale spatio-temporal graph\nconvolutional network is presented as the motion discriminator for adversarial\ntraining using datasets without any 3D labeling. We propose a sequential data\nloading strategy to meet the special start-to-end data processing requirement\nof live stream. We demonstrate the importance of each proposed module with\nextensive experiments. The results show the effectiveness of TePose on\nwidely-used human pose benchmarks with state-of-the-art performance.\n", "date": "2022-07-25", "categories": []}
{"id": "2207.10955", "title": "Faster VoxelPose: Real-time 3D Human Pose Estimation by Orthographic\n  Projection", "abstract": "  While the voxel-based methods have achieved promising results for\nmulti-person 3D pose estimation from multi-cameras, they suffer from heavy\ncomputation burdens, especially for large scenes. We present Faster VoxelPose\nto address the challenge by re-projecting the feature volume to the three\ntwo-dimensional coordinate planes and estimating X, Y, Z coordinates from them\nseparately. To that end, we first localize each person by a 3D bounding box by\nestimating a 2D box and its height based on the volume features projected to\nthe xy-plane and z-axis, respectively. Then for each person, we estimate\npartial joint coordinates from the three coordinate planes separately which are\nthen fused to obtain the final 3D pose. The method is free from costly 3D-CNNs\nand improves the speed of VoxelPose by ten times and meanwhile achieves\ncompetitive accuracy as the state-of-the-art methods, proving its potential in\nreal-time applications.\n", "date": "2022-07-22", "categories": []}
{"id": "2207.09949", "title": "VirtualPose: Learning Generalizable 3D Human Pose Models from Virtual\n  Data", "abstract": "  While monocular 3D pose estimation seems to have achieved very accurate\nresults on the public datasets, their generalization ability is largely\noverlooked. In this work, we perform a systematic evaluation of the existing\nmethods and find that they get notably larger errors when tested on different\ncameras, human poses and appearance. To address the problem, we introduce\nVirtualPose, a two-stage learning framework to exploit the hidden \"free lunch\"\nspecific to this task, i.e. generating infinite number of poses and cameras for\ntraining models at no cost. To that end, the first stage transforms images to\nabstract geometry representations (AGR), and then the second maps them to 3D\nposes. It addresses the generalization issue from two aspects: (1) the first\nstage can be trained on diverse 2D datasets to reduce the risk of over-fitting\nto limited appearance; (2) the second stage can be trained on diverse AGR\nsynthesized from a large number of virtual cameras and poses. It outperforms\nthe SOTA methods without using any paired images and 3D poses from the\nbenchmarks, which paves the way for practical applications. Code is available\nat https://github.com/wkom/VirtualPose.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09303", "title": "DH-AUG: DH Forward Kinematics Model Driven Augmentation for 3D Human\n  Pose Estimation", "abstract": "  Due to the lack of diversity of datasets, the generalization ability of the\npose estimator is poor. To solve this problem, we propose a pose augmentation\nsolution via DH forward kinematics model, which we call DH-AUG. We observe that\nthe previous work is all based on single-frame pose augmentation, if it is\ndirectly applied to video pose estimator, there will be several previously\nignored problems: (i) angle ambiguity in bone rotation (multiple solutions);\n(ii) the generated skeleton video lacks movement continuity. To solve these\nproblems, we propose a special generator based on DH forward kinematics model,\nwhich is called DH-generator. Extensive experiments demonstrate that DH-AUG can\ngreatly increase the generalization ability of the video pose estimator. In\naddition, when applied to a single-frame 3D pose estimator, our method\noutperforms the previous best pose augmentation method. The source code has\nbeen released at\nhttps://github.com/hlz0606/DH-AUG-DH-Forward-Kinematics-Model-Driven-Augmentation-for-3D-Human-Pose-Estimation.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.07900", "title": "Mutual Adaptive Reasoning for Monocular 3D Multi-Person Pose Estimation", "abstract": "  Inter-person occlusion and depth ambiguity make estimating the 3D poses of\nmonocular multiple persons as camera-centric coordinates a challenging problem.\nTypical top-down frameworks suffer from high computational redundancy with an\nadditional detection stage. By contrast, the bottom-up methods enjoy low\ncomputational costs as they are less affected by the number of humans. However,\nmost existing bottom-up methods treat camera-centric 3D human pose estimation\nas two unrelated subtasks: 2.5D pose estimation and camera-centric depth\nestimation. In this paper, we propose a unified model that leverages the mutual\nbenefits of both these subtasks. Within the framework, a robust structured 2.5D\npose estimation is designed to recognize inter-person occlusion based on depth\nrelationships. Additionally, we develop an end-to-end geometry-aware depth\nreasoning method that exploits the mutual benefits of both 2.5D pose and\ncamera-centric root depths. This method first uses 2.5D pose and geometry\ninformation to infer camera-centric root depths in a forward pass, and then\nexploits the root depths to further improve representation learning of 2.5D\npose estimation in a backward pass. Further, we designed an adaptive fusion\nscheme that leverages both visual perception and body geometry to alleviate\ninherent depth ambiguity issues. Extensive experiments demonstrate the\nsuperiority of our proposed model over a wide range of bottom-up methods. Our\naccuracy is even competitive with top-down counterparts. Notably, our model\nruns much faster than existing bottom-up and top-down methods.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.07381", "title": "A Dual-Masked Auto-Encoder for Robust Motion Capture with\n  Spatial-Temporal Skeletal Token Completion", "abstract": "  Multi-person motion capture can be challenging due to ambiguities caused by\nsevere occlusion, fast body movement, and complex interactions. Existing\nframeworks build on 2D pose estimations and triangulate to 3D coordinates via\nreasoning the appearance, trajectory, and geometric consistencies among\nmulti-camera observations. However, 2D joint detection is usually incomplete\nand with wrong identity assignments due to limited observation angle, which\nleads to noisy 3D triangulation results. To overcome this issue, we propose to\nexplore the short-range autoregressive characteristics of skeletal motion using\ntransformer. First, we propose an adaptive, identity-aware triangulation module\nto reconstruct 3D joints and identify the missing joints for each identity. To\ngenerate complete 3D skeletal motion, we then propose a Dual-Masked\nAuto-Encoder (D-MAE) which encodes the joint status with both\nskeletal-structural and temporal position encoding for trajectory completion.\nD-MAE's flexible masking and encoding mechanism enable arbitrary skeleton\ndefinitions to be conveniently deployed under the same framework. In order to\ndemonstrate the proposed model's capability in dealing with severe data loss\nscenarios, we contribute a high-accuracy and challenging motion capture dataset\nof multi-person interactions with severe occlusion. Evaluations on both\nbenchmark and our new dataset demonstrate the efficiency of our proposed model,\nas well as its advantage against the other state-of-the-art methods.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.03618", "title": "PoseGU: 3D Human Pose Estimation with Novel Human Pose Generator and\n  Unbiased Learning", "abstract": "  3D pose estimation has recently gained substantial interests in computer\nvision domain. Existing 3D pose estimation methods have a strong reliance on\nlarge size well-annotated 3D pose datasets, and they suffer poor model\ngeneralization on unseen poses due to limited diversity of 3D poses in training\nsets. In this work, we propose PoseGU, a novel human pose generator that\ngenerates diverse poses with access only to a small size of seed samples, while\nequipping the Counterfactual Risk Minimization to pursue an unbiased evaluation\nobjective. Extensive experiments demonstrate PoseGU outforms almost all the\nstate-of-the-art 3D human pose methods under consideration over three popular\nbenchmark datasets. Empirical analysis also proves PoseGU generates 3D poses\nwith improved data diversity and better generalization ability.\n", "date": "2022-07-07", "categories": []}
{"id": "2207.02425", "title": "Self-Constrained Inference Optimization on Structural Groups for Human\n  Pose Estimation", "abstract": "  We observe that human poses exhibit strong group-wise structural correlation\nand spatial coupling between keypoints due to the biological constraints of\ndifferent body parts. This group-wise structural correlation can be explored to\nimprove the accuracy and robustness of human pose estimation. In this work, we\ndevelop a self-constrained prediction-verification network to characterize and\nlearn the structural correlation between keypoints during training. During the\ninference stage, the feedback information from the verification network allows\nus to perform further optimization of pose prediction, which significantly\nimproves the performance of human pose estimation. Specifically, we partition\nthe keypoints into groups according to the biological structure of human body.\nWithin each group, the keypoints are further partitioned into two subsets,\nhigh-confidence base keypoints and low-confidence terminal keypoints. We\ndevelop a self-constrained prediction-verification network to perform forward\nand backward predictions between these keypoint subsets. One fundamental\nchallenge in pose estimation, as well as in generic prediction tasks, is that\nthere is no mechanism for us to verify if the obtained pose estimation or\nprediction results are accurate or not, since the ground truth is not\navailable. Once successfully learned, the verification network serves as an\naccuracy verification module for the forward pose prediction. During the\ninference stage, it can be used to guide the local optimization of the pose\nestimation results of low-confidence keypoints with the self-constrained loss\non high-confidence keypoints as the objective function. Our extensive\nexperimental results on benchmark MS COCO and CrowdPose datasets demonstrate\nthat the proposed method can significantly improve the pose estimation results.\n", "date": "2022-07-06", "categories": []}
{"id": "2206.11678", "title": "BlazePose GHUM Holistic: Real-time 3D Human Landmarks and Pose\n  Estimation", "abstract": "  We present BlazePose GHUM Holistic, a lightweight neural network pipeline for\n3D human body landmarks and pose estimation, specifically tailored to real-time\non-device inference. BlazePose GHUM Holistic enables motion capture from a\nsingle RGB image including avatar control, fitness tracking and AR/VR effects.\nOur main contributions include i) a novel method for 3D ground truth data\nacquisition, ii) updated 3D body tracking with additional hand landmarks and\niii) full body pose estimation from a monocular image.\n", "date": "2022-06-23", "categories": []}
{"id": "2206.06430", "title": "A Training Method For VideoPose3D With Ideology of Action Recognition", "abstract": "  Action recognition and pose estimation from videos are closely related to\nunderstand human motions, but more literature focuses on how to solve pose\nestimation tasks alone from action recognition. This research shows a faster\nand more flexible training method for VideoPose3D which is based on action\nrecognition. This model is fed with the same type of action as the type that\nwill be estimated, and different types of actions can be trained separately.\nEvidence has shown that, for common pose-estimation tasks, this model requires\na relatively small amount of data to carry out similar results with the\noriginal research, and for action-oriented tasks, it outperforms the original\nresearch by 4.5% with a limited receptive field size and training epoch on\nVelocity Error of MPJPE. This model can handle both action-oriented and common\npose-estimation problems.\n", "date": "2022-06-13", "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": "2206.04785", "title": "Building Spatio-temporal Transformers for Egocentric 3D Pose Estimation", "abstract": "  Egocentric 3D human pose estimation (HPE) from images is challenging due to\nsevere self-occlusions and strong distortion introduced by the fish-eye view\nfrom the head mounted camera. Although existing works use intermediate\nheatmap-based representations to counter distortion with some success,\naddressing self-occlusion remains an open problem. In this work, we leverage\ninformation from past frames to guide our self-attention-based 3D HPE\nestimation procedure -- Ego-STAN. Specifically, we build a spatio-temporal\nTransformer model that attends to semantically rich convolutional neural\nnetwork-based feature maps. We also propose feature map tokens: a new set of\nlearnable parameters to attend to these feature maps. Finally, we demonstrate\nEgo-STAN's superior performance on the xR-EgoPose dataset where it achieves a\n30.6% improvement on the overall mean per-joint position error, while leading\nto a 22% drop in parameters compared to the state-of-the-art.\n", "date": "2022-06-09", "categories": []}
{"id": "2206.01867", "title": "SPGNet: Spatial Projection Guided 3D Human Pose Estimation in Low\n  Dimensional Space", "abstract": "  We propose a method SPGNet for 3D human pose estimation that mixes\nmulti-dimensional re-projection into supervised learning. In this method, the\n2D-to-3D-lifting network predicts the global position and coordinates of the 3D\nhuman pose. Then, we re-project the estimated 3D pose back to the 2D key points\nalong with spatial adjustments. The loss functions compare the estimated 3D\npose with the 3D pose ground truth, and re-projected 2D pose with the input 2D\npose. In addition, we propose a kinematic constraint to restrict the predicted\ntarget with constant human bone length. Based on the estimation results for the\ndataset Human3.6M, our approach outperforms many state-of-the-art methods both\nqualitatively and quantitatively.\n", "date": "2022-06-04", "categories": []}
{"id": "2606.29230", "title": "Again-Pose: Anchor-Guided Adaptive Inter-Frame Motion Cues Propagating for High-quality Human Pose Reconstruction", "abstract": "Reconstructing continuous 3D human poses from unconstrained videos is challenging, especially in extreme motion scenarios involving severe motion blur and occlusion. Current state-of-the-art methods typically rely on implicit temporal attention to aggregate features across frames. However, under severe visual degradation, input features often suffer from collapse, rendering them indistinguishable from noise. In such cases, implicit aggregation fails to distinguish valid signals, leading to catastrophic reconstruction errors. To address this robustness gap, we propose a simple yet effective framework called Anchor-guided adaptive inter-frame motion cues propagating (Again-Pose), reformulating pose estimation in degraded frames as a motion-guided recovery task. Instead of blindly smoothing features, we explicitly identify high-quality Anchor Frames based on feature saliency and propagate reliable kinematic cues to \"inpaint\" the poses of degraded intermediate frames. Specifically, a Dual-path Motion-aware Module captures fine-grained inter-frame dynamics, while a Difference-weighted Fusion Module adaptively propagates these cues to suppress drift. Extensive experiments on standard benchmarks (Human3.6M, 3DPW, PoseTrack) and the challenging FineDiving dataset demonstrate that Again-Pose significantly outperforms state-of-the-art methods in robustness and stability, effectively recovering plausible poses where other methods fail.", "date": "2026-06-28", "categories": []}