pith. sign in

arxiv: 2605.14854 · v2 · pith:SAFMUNW4new · submitted 2026-05-14 · 💻 cs.CV · cs.AI

FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery

Patrick Kwon , Chen Chen This is my paper

Pith reviewed 2026-05-20 21:01 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords human mesh recoveryvideo human pose3D reconstructionflow matchingprobabilistic completionocclusion robustnesssynthetic data generation
0
0 comments X

The pith

Separating deterministic recovery of the torso-root anchor from probabilistic limb completion reduces errors in ambiguous human mesh recovery from video.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Human mesh recovery from video faces inherent ambiguity, especially under occlusion or weak depth information. The paper observes that this ambiguity affects different body parts unequally, with torso and root being more reliably inferred from images than the arms and legs. FactorizedHMR addresses this by using a first stage to deterministically regress a fixed torso-root anchor and a second stage to probabilistically complete the remaining articulations via flow-matching. Special techniques like composite targets and classifier-free guidance ensure the anchor stays fixed while improving the uncertain parts. The approach includes a synthetic data pipeline and demonstrates gains particularly in occlusion and world-space drift scenarios.

Core claim

FactorizedHMR is a hybrid two-stage framework that first applies deterministic regression to obtain a stable torso-root anchor and then uses a probabilistic flow-matching module to recover the non-torso articulations. By incorporating a composite target representation, geometry-aware supervision, and feature-aware classifier-free guidance, the method preserves the anchor during completion and achieves competitive results on benchmarks with notable improvements in occlusion-heavy recovery and drift-sensitive world-space metrics. A synthetic data pipeline provides the necessary paired supervision under varied viewpoints.

What carries the argument

Two-stage factorization with deterministic regression for the torso-root anchor and probabilistic flow-matching for distal articulations, using geometry-aware supervision and classifier-free guidance to maintain anchor consistency.

If this is right

  • Competitive performance maintained across camera-space and world-space benchmarks.
  • Clearer improvements in scenarios with heavy occlusion.
  • Reduced drift in long-term world-space tracking metrics.
  • Better handling of ambiguity in single-reference recovery of limbs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Independent optimization of the anchor stage and completion stage could lead to further gains without full retraining.
  • The idea of fixing certain parts during probabilistic inference may generalize to other body recovery tasks with similar certainty gradients.
  • Combining this with multi-view inputs might amplify the benefits in world-space consistency.

Load-bearing premise

The torso pose and root structure are relatively well constrained by image evidence, while distal articulations remain substantially more uncertain.

What would settle it

A test set of videos with heavy torso occlusion but clear limb visibility, where the method should underperform if the assumption does not hold.

Figures

Figures reproduced from arXiv: 2605.14854 by Chen Chen, Patrick Kwon.

Figure 1
Figure 1. Figure 1: Qualitative comparison on EMDB [24]. The left arm and hand are heavily occluded by the tree, so multiple poses are plausible. GVHMR [49] predicts a left hand that is not visible in the image (red square), illustrating how deterministic pipelines can commit to an implausible average solution under ambiguity. Our method instead preserves the visible pose while producing a more plausible completion for the oc… view at source ↗
Figure 3
Figure 3. Figure 3: Example results for FactorizedHMR. Stage 1 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of FactorizedHMR. Input video frames are preprocessed into ray-embedded [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Body-shape recovery under heavy occlusion. FactorizedHMR better preserves body volume. GT GVHMR GENMO Ours GT GVHMR Ours GT GVHMR Ours [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Body-pose recovery in ambiguity-heavy scenes. The red squares in the GVHMR [ [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Effect of the number of ODE sam￾pling steps on MPJPE and WA-MPJPE metrics. Both metrics improve rapidly from very small step counts and then saturate, with only minor differences beyond roughly 20 steps. We use 50 steps as a conservative near-converged default [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Additional qualitative comparisons in the appendix. Each example is shown left-to-right as [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Additional body-pose recovery comparisons between our method and GVHMR [49] [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Additional body-pose recovery comparisons between our method and GVHMR [49] [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Additional body-pose recovery comparisons between our method and GENMO [37]. [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
read the original 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces FactorizedHMR, a two-stage hybrid framework for video human mesh recovery. It first uses a deterministic regression module to recover a stable torso-root anchor, followed by a probabilistic flow-matching module to complete the non-torso articulations. A synthetic data pipeline is proposed to provide paired image-camera-motion supervision. The method is evaluated on camera-space and world-space benchmarks, claiming competitive performance with gains in occlusion-heavy recovery and drift-sensitive metrics.

Significance. If the empirical results hold, this work could contribute to better handling of inherent ambiguities in human mesh recovery by explicitly separating well-constrained and uncertain parts of the body. The combination of deterministic and probabilistic components, along with the synthetic data generation, represents a thoughtful approach to improving robustness in challenging scenarios. The synthetic data pipeline providing paired supervision is a strength.

major comments (2)
  1. [§3.1] §3.1: The central premise that torso pose and root structure are relatively well constrained by image evidence while distal articulations are more uncertain is stated qualitatively but lacks supporting quantitative analysis, such as per-part error distributions or visibility statistics from the first-stage regression. This assumption is load-bearing for the two-stage design and the decision to fix the anchor.
  2. [§4.3] §4.3 (world-space results): No ablation is presented on the sensitivity of final metrics to first-stage root translation or torso orientation errors. Since the anchor is kept fixed and drift metrics are root-sensitive, this leaves open whether modest first-stage mistakes could dominate the reported gains in occlusion-heavy cases.
minor comments (2)
  1. [Abstract] The abstract references competitive performance and targeted gains but the manuscript should ensure all tables include error bars, dataset splits, and exact baseline versions for full verifiability.
  2. [Figure 2] Figure captions for the pipeline overview could more explicitly label the conditioning path from the deterministic anchor into the flow-matching stage.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and insightful review. The comments highlight important aspects of our design rationale and evaluation that we have addressed through revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3.1] §3.1: The central premise that torso pose and root structure are relatively well constrained by image evidence while distal articulations are more uncertain is stated qualitatively but lacks supporting quantitative analysis, such as per-part error distributions or visibility statistics from the first-stage regression. This assumption is load-bearing for the two-stage design and the decision to fix the anchor.

    Authors: We agree that quantitative evidence would better substantiate this premise. In the revised manuscript, we have expanded §3.1 with a new analysis of per-part errors and visibility statistics computed from the first-stage deterministic regression on a held-out validation set. The added results show substantially lower average MPJPE for torso and root joints (approximately 42 mm) with higher visibility rates (95%) relative to distal articulations (approximately 88 mm MPJPE and 68% visibility). These statistics are now presented to support the decision to anchor on the torso-root structure. revision: yes

  2. Referee: [§4.3] §4.3 (world-space results): No ablation is presented on the sensitivity of final metrics to first-stage root translation or torso orientation errors. Since the anchor is kept fixed and drift metrics are root-sensitive, this leaves open whether modest first-stage mistakes could dominate the reported gains in occlusion-heavy cases.

    Authors: We appreciate this observation regarding potential error propagation. To address it, we have added a sensitivity analysis to the revised §4.3. We injected controlled perturbations to the first-stage root translation and torso orientation (with noise magnitudes matching the observed first-stage error distribution) and re-evaluated the full pipeline on world-space benchmarks. The results indicate that performance degrades gracefully and that the relative gains in occlusion-heavy and drift-sensitive metrics remain intact, suggesting the probabilistic stage provides some robustness. The analysis and corresponding figures have been incorporated into the main paper and supplementary material. revision: yes

Circularity Check

0 steps flagged

No circularity: framework motivated by external observation with independent supervision

full rationale

The paper motivates its two-stage separation directly from the stated empirical observation that torso-root structure is typically better constrained by image evidence than distal joints, then implements this via a deterministic regression module followed by flow-matching completion with composite targets and geometry-aware losses. No equations or claims reduce a prediction to a fitted parameter by construction, no self-citations are invoked as load-bearing uniqueness theorems, and the synthetic data pipeline is presented as supplying external paired supervision rather than being defined from the model's outputs. The reported benchmarks therefore rest on standard metrics and external data rather than tautological re-derivation of inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that body-part ambiguity is sufficiently non-uniform to justify separate deterministic and probabilistic stages, plus the engineering choice of a composite target representation and geometry-aware supervision whose effectiveness is not independently verified in the abstract.

axioms (1)
  • domain assumption Torso pose and root structure are relatively well constrained by image evidence while distal articulations are substantially more uncertain.
    This differential-ambiguity premise is invoked to justify the two-stage design and the decision to preserve the torso-root anchor.

pith-pipeline@v0.9.0 · 5703 in / 1425 out tokens · 75291 ms · 2026-05-20T21:01:54.183724+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 4 internal anchors

  1. [1]

    Exploiting temporal context for 3d human pose estimation in the wild

    Anurag Arnab, Carl Doersch, and Andrew Zisserman. Exploiting temporal context for 3d human pose estimation in the wild. InCVPR, pages 3395–3404, 2019

    work page 2019

  2. [2]

    Christopher M. Bishop. Mixture density networks. Technical report, Aston University, 1994

    work page 1994

  3. [3]

    Black, Priyanka Patel, Joachim Tesch, and Jinlong Yang

    Michael J. Black, Priyanka Patel, Joachim Tesch, and Jinlong Yang. Bedlam: A synthetic dataset of bodies exhibiting detailed lifelike animated motion. InCVPR, pages 8726–8737, 2023

    work page 2023

  4. [4]

    Richter, and Vladlen Koltun

    Aleksei Bochkovskii, Amaël Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R. Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. InInternational Conference on Learning Representations, 2025

    work page 2025

  5. [5]

    Federica Bogo, Angjoo Kanazawa, Christoph Lassner, Peter Gehler, Javier Romero, and Michael J. Black. Keep it smpl: Automatic estimation of 3d human pose and shape from a single image. InECCV, pages 561–578, 2016

    work page 2016

  6. [6]

    Uni3c: Unifying precisely 3d-enhanced camera and human motion controls for video generation.arXiv preprint arXiv:2504.14899, 2025

    Chenjie Cao, Jingkai Zhou, Shikai Li, Jingyun Liang, Chaohui Yu, Fan Wang, Xiangyang Xue, and Yanwei Fu. Uni3c: Unifying precisely 3d-enhanced camera and human motion controls for video generation.arXiv preprint arXiv:2504.14899, 2025

    work page arXiv 2025

  7. [7]

    Z. Cao, G. Hidalgo Martinez, T. Simon, S. Wei, and Y . A. Sheikh. Openpose: Realtime multi-person 2d pose estimation using part affinity fields.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019

    work page 2019

  8. [8]

    Executing your commands via motion diffusion in latent space

    Xin Chen, Biao Jiang, Wen Liu, Zilong Huang, Bin Fu, Tao Chen, and Gang Yu. Executing your commands via motion diffusion in latent space. InCVPR, pages 18000–18010, 2023

    work page 2023

  9. [9]

    Beyond static features for temporally consistent 3d human pose and shape from a video

    Hongsuk Choi, Gyeongsik Moon, Ju Yong Chang, and Kyoung Mu Lee. Beyond static features for temporally consistent 3d human pose and shape from a video. InCVPR, pages 1964–1973, 2021

    work page 1964

  10. [10]

    Learning to fit morphable models

    Vasileios Choutas, Federica Bogo, Jingjing Shen, and Julien Valentin. Learning to fit morphable models. InECCV, pages 160–179, 2022

    work page 2022

  11. [11]

    Sai Kumar Dwivedi, Yu Sun, Priyanka Patel, Yao Feng, and Michael J. Black. Tokenhmr: Advancing human mesh recovery with a tokenized pose representation. InCVPR, pages 1323–1333, 2024

    work page 2024

  12. [12]

    Mega: Masked generative autoencoder for human mesh recovery

    Guénolé Fiche, Simon Leglaive, Xavier Alameda-Pineda, and Francesc Moreno-Noguer. Mega: Masked generative autoencoder for human mesh recovery. InCVPR, pages 5366–5378, 2025

    work page 2025

  13. [13]

    Jeni, and Zackory Erickson

    Jing Gao, Ce Zheng, Laszlo A. Jeni, and Zackory Erickson. Disrt-in-bed: Diffusion-based sim-to-real transfer framework for in-bed human mesh recovery. InCVPR, pages 1829–1838, 2025

    work page 2025

  14. [14]

    Recon- structing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Recon- structing and tracking humans with transformers. InICCV, pages 15073–15084, 2023

    work page 2023

  15. [15]

    Human poseitioning system (hps): 3d human pose estimation and self-localization in large scenes from body-mounted sensors

    Vladimir Guzov, Aymen Mir, Torsten Sattler, and Gerard Pons-Moll. Human poseitioning system (hps): 3d human pose estimation and self-localization in large scenes from body-mounted sensors. InCVPR, pages 4318–4329, 2021

    work page 2021

  16. [16]

    Mohamed Hassan, Duygu Ceylan, Ruben Villegas, Jun Saito, Jimei Yang, Yi Zhou, and Michael J. Black. Stochastic scene-aware motion prediction. InICCV, pages 11374–11384, 2021

    work page 2021

  17. [17]

    Phd: Personalized 3d human body fitting with point diffusion

    Hsuan-I Ho, Chen Guo, Po-Chen Wu, Ivan Shugurov, Chengcheng Tang, Abhay Mittal, Sizhe An, Manuel Kaufmann, and Linguang Zhang. Phd: Personalized 3d human body fitting with point diffusion. InICCV, 2025

    work page 2025

  18. [18]

    Denoising diffusion probabilistic models.NeurIPS, 33: 6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.NeurIPS, 33: 6840–6851, 2020

    work page 2020

  19. [19]

    Huang, Hongwei Yi, Markus Höschle, Matvey Safroshkin, Tsvetelina Alexiadis, Senya Polikovsky, Daniel Scharstein, and Michael J

    Chun-Hao P. Huang, Hongwei Yi, Markus Höschle, Matvey Safroshkin, Tsvetelina Alexiadis, Senya Polikovsky, Daniel Scharstein, and Michael J. Black. Capturing and inferring dense full-body human-scene contact. InCVPR, pages 13274–13285, 2022

    work page 2022

  20. [20]

    Gehler, Javier Romero, Ijaz Akhter, and Michael J

    Yinghao Huang, Federica Bogo, Christoph Lassner, Angjoo Kanazawa, Peter V . Gehler, Javier Romero, Ijaz Akhter, and Michael J. Black. Towards accurate marker-less human shape and pose estimation over time. In3DV, pages 421–430, 2017. 10

    work page 2017

  21. [21]

    Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments.IEEE Transactions on Pattern Analysis and Machine Intelligence, 36:1325–1339, 2014

    work page 2014

  22. [22]

    Black, David W

    Angjoo Kanazawa, Michael J. Black, David W. Jacobs, and Jitendra Malik. End-to-end recovery of human shape and pose. InCVPR, pages 7122–7131, 2018

    work page 2018

  23. [23]

    Zhang, Panna Felsen, and Jitendra Malik

    Angjoo Kanazawa, Jason Y . Zhang, Panna Felsen, and Jitendra Malik. Learning 3d human dynamics from video. InCVPR, pages 5614–5623, 2019

    work page 2019

  24. [24]

    Emdb: The electromagnetic database of global 3d human pose and shape in the wild

    Manuel Kaufmann, Jie Song, Chen Guo, Kaiyue Shen, Tianjian Jiang, Chengcheng Tang, Juan José Zárate, and Otmar Hilliges. Emdb: The electromagnetic database of global 3d human pose and shape in the wild. InICCV, pages 14632–14643, 2023

    work page 2023

  25. [25]

    Adam: A Method for Stochastic Optimization

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization.CoRR, abs/1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  26. [26]

    Auto-Encoding Variational Bayes

    Diederik P. Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  27. [27]

    Muhammed Kocabas, Nikos Athanasiou, and Michael J. Black. Vibe: Video inference for human body pose and shape estimation. InCVPR, pages 5253–5263, 2020

    work page 2020

  28. [28]

    Huang, Otmar Hilliges, and Michael J

    Muhammed Kocabas, Chun-Hao P. Huang, Otmar Hilliges, and Michael J. Black. Pare: Part attention regressor for 3d human body estimation. InICCV, pages 11127–11137, 2021

    work page 2021

  29. [29]

    Black, and Kostas Daniilidis

    Nikos Kolotouros, Georgios Pavlakos, Michael J. Black, and Kostas Daniilidis. Learning to reconstruct 3d human pose and shape via model-fitting in the loop. InICCV, pages 2252–2261, 2019

    work page 2019

  30. [30]

    Convolutional mesh regression for single- image human shape reconstruction

    Nikos Kolotouros, Georgios Pavlakos, and Kostas Daniilidis. Convolutional mesh regression for single- image human shape reconstruction. InCVPR, pages 4501–4510, 2019

    work page 2019

  31. [31]

    Probabilistic modeling for human mesh recovery

    Nikos Kolotouros, Georgios Pavlakos, Dinesh Jayaraman, and Kostas Daniilidis. Probabilistic modeling for human mesh recovery. InICCV, pages 11605–11614, 2021

    work page 2021

  32. [32]

    Flux.https://github.com/black-forest-labs/flux, 2024

    Black Forest Labs. Flux.https://github.com/black-forest-labs/flux, 2024

    work page 2024

  33. [33]

    Black, and Peter V

    Christoph Lassner, Javier Romero, Martin Kiefel, Federica Bogo, Michael J. Black, and Peter V . Gehler. Unite the people: Closing the loop between 3d and 2d human representations. InCVPR, pages 6050–6059, 2017

    work page 2017

  34. [34]

    Hybrik: A hybrid analytical-neural inverse kinematics solution for 3d human pose and shape estimation

    Jiefeng Li, Siyuan Bian, Chao Xu, Gang Liu, Gang Yu, and Cewu Lu. Hybrik: A hybrid analytical-neural inverse kinematics solution for 3d human pose and shape estimation. InCVPR, pages 3383–3393, 2021

    work page 2021

  35. [35]

    D&d: Learning human dynamics from dynamic camera

    Jiefeng Li, Siyuan Bian, Chao Xu, Gang Liu, Gang Yu, and Cewu Lu. D&d: Learning human dynamics from dynamic camera. InECCV, pages 479–496, 2022

    work page 2022

  36. [36]

    NIKI: Neural inverse kinematics with invertible neural networks for 3d human pose and shape estimation

    Jiefeng Li, Siyuan Bian, Qi Liu, Jiasheng Tang, Fan Wang, and Cewu Lu. NIKI: Neural inverse kinematics with invertible neural networks for 3d human pose and shape estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    work page 2023

  37. [37]

    Genmo: A generalist model for human motion.arXiv preprint arXiv:2505.01425, 2025

    Jiefeng Li, Jinkun Cao, Haotian Zhang, Davis Rempe, Jan Kautz, Umar Iqbal, and Ye Yuan. Genmo: A generalist model for human motion.arXiv preprint arXiv:2505.01425, 2025

    work page arXiv 2025

  38. [38]

    Cliff: Carrying location information in full frames into human pose and shape estimation

    Zhihao Li, Jianzhuang Liu, Zhensong Zhang, Songcen Xu, and Youliang Yan. Cliff: Carrying location information in full frames into human pose and shape estimation. InECCV, pages 590–606, 2022

    work page 2022

  39. [39]

    End-to-end human pose and mesh reconstruction with transformers

    Kevin Lin, Lijuan Wang, and Zicheng Liu. End-to-end human pose and mesh reconstruction with transformers. InCVPR, pages 1954–1963, 2021

    work page 1954

  40. [40]

    Rehg, and Siyu Tang

    Miao Liu, Dexin Yang, Yan Zhang, Zhaopeng Cui, James M. Rehg, and Siyu Tang. 4d human body capture from egocentric video via 3d scene grounding. In3DV, pages 930–939, 2021

    work page 2021

  41. [41]

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J. Black. Smpl: A skinned multi-person linear model.ACM Transactions on Graphics (TOG), 34(6):248:1–248:16, 2015

    work page 2015

  42. [42]

    Dposer-x: Diffusion model as robust 3d whole-body human pose prior

    Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Xian Liu, Zhongang Cai, Lei Yang, Yulun Zhang, Haoqian Wang, and Ziwei Liu. Dposer-x: Diffusion model as robust 3d whole-body human pose prior. InICCV, 2025. 11

    work page 2025

  43. [43]

    Troje, Gerard Pons-Moll, and Michael J

    Naureen Mahmood, Nima Ghorbani, Nikolaus F. Troje, Gerard Pons-Moll, and Michael J. Black. Amass: Archive of motion capture as surface shapes. InICCV, pages 5442–5451, 2019

    work page 2019

  44. [44]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. InECCV, 2020

    work page 2020

  45. [45]

    Learning to estimate 3d human pose and shape from a single color image

    Georgios Pavlakos, Luyang Zhu, Xiaowei Zhou, and Kostas Daniilidis. Learning to estimate 3d human pose and shape from a single color image. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 459–468, 2018

    work page 2018

  46. [46]

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black. Expressive body capture: 3D hands, face, and body from a single image. In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 10975–10985, 2019

    work page 2019

  47. [47]

    Black, Naureen Mahmood Gerard Pons-Moll Zhao, et al

    Davis Rempe, Zhengyi Luo, Saurabh Banerjee, Michael J. Black, Naureen Mahmood Gerard Pons-Moll Zhao, et al. Humor: 3d human motion model for robust pose estimation. InICCV, 2021

    work page 2021

  48. [48]

    Genhmr: Generative human mesh recovery

    Muhammad Usama Saleem, Ekkasit Pinyoanuntapong, Pu Wang, Hongfei Xue, Srijan Das, and Chen Chen. Genhmr: Generative human mesh recovery. InAAAI Conference on Artificial Intelligence, 2025

    work page 2025

  49. [49]

    World-grounded human motion recovery via gravity-view coordinates

    Zehong Shen, Huaijin Pi, Yan Xia, Zhi Cen, Sida Peng, Zechen Hu, Hujun Bao, Ruizhen Hu, and Xiaowei Zhou. World-grounded human motion recovery via gravity-view coordinates. InSIGGRAPH Asia Conference Papers, 2024

    work page 2024

  50. [50]

    Soyong Shin, Juyong Kim, Eni Halilaj, and Michael J. Black. Wham: Reconstructing world-grounded humans with accurate 3d motion.arXiv preprint arXiv:2312.07531, 2023

    work page arXiv 2023

  51. [51]

    Learning structured output representation using deep conditional generative models

    Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. InAdvances in Neural Information Processing Systems (NeurIPS), 2015

    work page 2015

  52. [52]

    Human body model fitting by learned gradient descent

    Jie Song, Xu Chen, and Otmar Hilliges. Human body model fitting by learned gradient descent. InECCV, pages 744–760, 2020

    work page 2020

  53. [53]

    Denoising Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  54. [54]

    Score-guided diffusion for 3d human recovery

    Anastasis Stathopoulos, Ligong Han, and Dimitris Metaxas. Score-guided diffusion for 3d human recovery. InCVPR, pages 906–915, 2024

    work page 2024

  55. [55]

    RoFormer: Enhanced Transformer with Rotary Position Embedding

    Jianlin Su, Yu Lu, Shengfeng Pan, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.ArXiv, abs/2104.09864, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  56. [56]

    Yu Sun, Qian Bao, Wu Liu, Tao Mei, and Michael J. Black. Trace: 5d temporal regression of avatars with dynamic cameras in 3d environments. InCVPR, pages 8856–8866, 2023

    work page 2023

  57. [57]

    Guy Tevet, Sigal Raab, Brian Gordon, Yoni Shafir, Daniel Cohen-Or, and Amit H. Bermano. Human motion diffusion model. InICLR, 2023

    work page 2023

  58. [58]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InNeural Information Processing Systems, 2017

    work page 2017

  59. [59]

    Black, Bodo Rosenhahn, and Gerard Pons-Moll

    Timo von Marcard, Roberto Henschel, Michael J. Black, Bodo Rosenhahn, and Gerard Pons-Moll. Recovering accurate 3d human pose in the wild using imus and a moving camera. InECCV, pages 601–617, 2018

    work page 2018

  60. [60]

    Tram: Global trajectory and motion of 3d humans from in-the-wild videos

    Yufu Wang, Ziyun Wang, Lingjie Liu, and Kostas Daniilidis. Tram: Global trajectory and motion of 3d humans from in-the-wild videos. InECCV, 2024

    work page 2024

  61. [61]

    Black, and Muhammed Kocabas

    Yufu Wang, Yu Sun, Priyanka Patel, Kostas Daniilidis, Michael J. Black, and Muhammed Kocabas. Prompthmr: Promptable human mesh recovery. InCVPR, pages 1148–1159, 2025

    work page 2025

  62. [62]

    Duomo: Dual motion diffusion for world-space human reconstruction.arXiv preprint arXiv:2603.03265, 2026

    Yufu Wang, Evonne Ng, Soyong Shin, Rawal Khirodkar, Yuan Dong, Zhaoen Su, Jinhyung Park, Kris Kitani, Alexander Richard, Fabian Prada, and Michael Zollhöfer. Duomo: Dual motion diffusion for world-space human reconstruction.arXiv preprint arXiv:2603.03265, 2026

    work page arXiv 2026

  63. [63]

    Probabilistic monocular 3d human pose estimation with normalizing flows

    Tom Wehrbein, Markus Rudolph, Bodo Rosenhahn, and Bastian Wandt. Probabilistic monocular 3d human pose estimation with normalizing flows. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 11199–11208, 2021. 12

    work page 2021

  64. [64]

    Physdiff: Physics-guided human motion diffusion model

    Ye Yuan, Jiaming Song, Umar Iqbal, Arash Vahdat, and Jan Kautz. Physdiff: Physics-guided human motion diffusion model. InICCV, pages 16010–16021, 2023

    work page 2023

  65. [65]

    Rohm: Robust human motion reconstruction via diffusion.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14606–14617, 2024

    Siwei Zhang, Bharat Lal Bhatnagar, Yuanlu Xu, Alexander Winkler, Petr Kadlecek, Siyu Tang, and Federica Bogo. Rohm: Robust human motion reconstruction via diffusion.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14606–14617, 2024. A Technical appendices and supplementary material Section A.1 summarizes the evaluation met...

    work page 2024