pith. sign in

arxiv: 2604.12765 · v1 · submitted 2026-04-14 · 💻 cs.CV · cs.GR

A Dataset and Evaluation for Complex 4D Markerless Human Motion Capture

Yeeun Park , Miqdad Naduthodi , Suryansh Kumar This is my paper

Pith reviewed 2026-05-10 14:59 UTC · model grok-4.3

classification 💻 cs.CV cs.GR
keywords markerless motion capture4D human modelingmulti-person datasetinter-person occlusionsSMPL parametersVicon ground truthdataset benchmarkfine-tuning generalization
0
0 comments X

The pith

A new dataset with Vicon ground truth shows markerless 4D motion capture models degrade sharply on multi-person interactions with occlusions and similar appearances.

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

The paper presents a dataset that records both single-person and multi-person movements under conditions that include frequent inter-person occlusions, rapid swaps between similarly dressed subjects, and changing distances, all captured with synchronized multi-view RGB, depth, camera calibration, Vicon 3D ground truth, and SMPL/SMPL-X parameters. Benchmarking existing markerless models on this data reveals large drops in accuracy compared with simpler test sets, while targeted fine-tuning on the new sequences produces measurable gains in generalization. The work matters because marker-based systems remain impractical for most real-world uses, so reliable markerless alternatives depend on closing the gap between controlled benchmarks and the messy dynamics of actual human interactions.

Core claim

By supplying synchronized multi-view RGB and depth sequences, precise camera calibration, Vicon-derived 3D ground truth, and corresponding SMPL/SMPL-X parameters for single- and multi-person scenarios that feature intricate motions, frequent inter-person occlusions, rapid position exchanges between similarly dressed subjects, and varying distances, the dataset demonstrates substantial performance degradation in current state-of-the-art markerless MoCap models and shows that targeted fine-tuning improves generalization to these conditions.

What carries the argument

The proposed MoCap dataset, which supplies multi-view RGB, depth, Vicon 3D ground truth, and SMPL parameters for complex multi-person interactions with occlusions and similar subject appearances.

If this is right

  • Markerless MoCap models must incorporate training or adaptation data that includes severe occlusions and rapid subject interactions to reach usable accuracy.
  • Targeted fine-tuning on sequences with Vicon-aligned ground truth measurably reduces the generalization gap for these models.
  • Precise multi-view RGB, depth, and SMPL parameter alignment enables quantitative diagnosis of where current methods fail.
  • The dataset supplies a concrete testbed for measuring progress toward practical markerless 4D capture outside controlled labs.

Where Pith is reading between the lines

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

  • Future work could test whether models trained or fine-tuned on this data transfer better to downstream tasks such as action recognition or virtual avatar animation.
  • The emphasis on similarly dressed subjects suggests that appearance similarity is a key failure mode worth isolating in follow-up experiments.
  • Combining this dataset with existing single-person or less occluded corpora could produce larger training mixtures that address multiple failure modes at once.

Load-bearing premise

The specific scenarios captured, including frequent inter-person occlusions, rapid position exchanges between similarly dressed subjects, and varying distances, sufficiently represent the domain gap present in real-world markerless 4D human motion capture.

What would settle it

If state-of-the-art markerless models show no substantial accuracy drop on the new sequences relative to existing benchmarks, or if fine-tuning on this dataset fails to improve performance on separate real-world multi-person test footage, the central claims would be undermined.

Figures

Figures reproduced from arXiv: 2604.12765 by Miqdad Naduthodi, Suryansh Kumar, Yeeun Park.

Figure 1
Figure 1. Figure 1: HUM4D provides multi-view RGB-D motion sequences with professional marker-based motion capture ground truth for complex [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example noisy depth data acquisition from RGB-D sen [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Hardware setup. Data acquisition configuration used in HUM4D. From left to right: perspective view of the RGB-D camera placement at approximately 1.45m height; top-view layout showing six cameras arranged in a circular configuration with a 3m radius; Intel RealSense D455 RGB-D sensor used for color and depth capture; and the Vicon motion capture system used for marker-based ground-truth acquisition. camera… view at source ↗
Figure 5
Figure 5. Figure 5: 3D acquisition of complex human motion sequences captured under challenging conditions such as Jittering, ID Swap, Occlusion, [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Overall acquisition pipeline: six synchronized multi-view RGB-D camera with Vicon MoCap system. RGB, RGB-D, and Vicon [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Human mesh recovery on HUM4D. Predicted SMPL meshes from SPIN [16], PARE [14], HMR2.0 [7], and PersPose [9] are overlaid on a challenging frame. used in SPIN [16], often fail to recover plausible body con￾figurations when the initial 2D evidence is corrupted. B. Robustness of Part-Aware and Geometry-Aware Models. Among all the evaluated methods, PersPose [9] achieves the lowest error on HUM4D. This result … view at source ↗
Figure 8
Figure 8. Figure 8: Representative examples of motion activities in HUM4D [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Representative examples from HUM4D. Each example includes the body-model or MoCap visualization, the synchronized RGB [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Top level hierarchy of HUM4D. The dataset is first grouped by motion type, and each motion type contains a set of activity [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Example lower level hierarchy of HUM4D. Within each activity, the data is further organized by recording setting, take index, [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
read the original abstract

Marker-based motion capture (MoCap) systems have long been the gold standard for accurate 4D human modeling, yet their reliance on specialized hardware and markers limits scalability and real-world deployment. Advancing reliable markerless 4D human motion capture requires datasets that reflect the complexity of real-world human interactions. Yet, existing benchmarks often lack realistic multi-person dynamics, severe occlusions, and challenging interaction patterns, leading to a persistent domain gap. In this work, we present a new dataset and evaluation for complex 4D markerless human motion capture. Our proposed MoCap dataset captures both single and multi-person scenarios with intricate motions, frequent inter-person occlusions, rapid position exchanges between similarly dressed subjects, and varying subject distances. It includes synchronized multi-view RGB and depth sequences, accurate camera calibration, ground-truth 3D motion capture from a Vicon system, and corresponding SMPL/SMPL-X parameters. This setup ensures precise alignment between visual observations and motion ground truth. Benchmarking state-of-the-art markerless MoCap models reveals substantial performance degradation under these realistic conditions, highlighting limitations of current approaches. We further demonstrate that targeted fine-tuning improves generalization, validating the dataset's realism and value for model development. Our evaluation exposes critical gaps in existing models and provides a rigorous foundation for advancing robust markerless 4D human motion capture.

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

0 major / 3 minor

Summary. The paper introduces a new multi-view dataset for complex 4D markerless human motion capture, featuring single- and multi-person scenarios with intricate motions, frequent inter-person occlusions, rapid position exchanges between similarly dressed subjects, and varying distances. It supplies synchronized multi-view RGB and depth sequences, camera calibration, Vicon ground-truth 3D motion capture, and corresponding SMPL/SMPL-X parameters. Benchmarking of state-of-the-art markerless MoCap models on this data shows substantial performance degradation relative to prior benchmarks, and targeted fine-tuning on the new dataset is shown to improve generalization.

Significance. If the quantitative benchmarking and fine-tuning results hold, the dataset supplies a valuable, realistic testbed that exposes domain gaps in current markerless approaches and supports further model development. The provision of precise Vicon ground truth aligned with visual observations and SMPL parameters is a clear strength for reproducibility and downstream research.

minor comments (3)
  1. Abstract and §1: The phrase 'substantial performance degradation' is used without a forward reference to the specific metrics or tables that quantify it; adding such a pointer would improve readability.
  2. Dataset description section: A compact table summarizing sequence counts, subject numbers, total frames, and occlusion statistics would make the scale and complexity of the captured scenarios easier to assess at a glance.
  3. Evaluation section: Standard error metrics (e.g., MPJPE, PVE) and per-scenario breakdowns should be presented with direct comparisons to the same models' published numbers on existing datasets to strengthen the domain-gap claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of our work, accurate summary of the dataset contributions, and recommendation for minor revision. The referee correctly highlights the value of the Vicon-aligned ground truth and the observed performance degradation in existing models under realistic multi-person conditions.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is a dataset contribution with standard benchmarking and fine-tuning experiments on independent Vicon ground truth. No derivations, fitted parameters renamed as predictions, self-citation chains, or uniqueness theorems appear in the argument structure. All claims follow directly from empirical evaluation without reducing to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work relies on standard domain assumptions in computer vision rather than new free parameters or invented entities.

axioms (2)
  • domain assumption Vicon optical motion capture system supplies accurate 3D ground-truth poses for alignment with visual data
    Invoked to ensure precise alignment between visual observations and motion ground truth.
  • domain assumption Multi-view camera calibration is sufficiently accurate for 4D reconstruction
    Required for the synchronized multi-view RGB and depth sequences to support reliable benchmarking.

pith-pipeline@v0.9.0 · 5547 in / 1264 out tokens · 66860 ms · 2026-05-10T14:59:03.008335+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    2d human pose estimation: New benchmark and state of the art analysis

    Mykhaylo Andriluka, Leonid Pishchulin, Peter Gehler, and Bernt Schiele. 2d human pose estimation: New benchmark and state of the art analysis. InProceedings of the IEEE Con- ference on computer Vision and Pattern Recognition, pages 3686–3693, 2014

    work page 2014

  2. [2]

    3D multibod- ies: Fitting sets of plausible 3D models to ambiguous image data

    Benjamin Biggs, S ´ebastien Ehrhart, Hanbyul Joo, Benjamin Graham, Andrea Vedaldi, and David Novotny. 3D multibod- ies: Fitting sets of plausible 3D models to ambiguous image data. InNeurIPS, 2020

    work page 2020

  3. [3]

    Federica Bogo, Javier Romero, Matthew Loper, and Michael J. Black. FAUST: Dataset and evaluation for 3D mesh registration. InProceedings IEEE Conf. on Com- puter Vision and Pattern Recognition (CVPR), Piscataway, NJ, USA, 2014. IEEE

    work page 2014

  4. [4]

    Humman: Multi-modal 4d human dataset for ver- satile sensing and modeling

    Zhongang Cai, Daxuan Ren, Ailing Zeng, Zhengyu Lin, Tao Yu, Wenjia Wang, Xiangyu Fan, Yang Gao, Yifan Yu, Liang Pan, et al. Humman: Multi-modal 4d human dataset for ver- satile sensing and modeling. InEuropean Conference on Computer Vision, pages 557–577. Springer, 2022

    work page 2022

  5. [5]

    Human4d: A human-centric multimodal dataset for motions and immersive media.IEEE Access, 8:176241–176262, 2020

    Anargyros Chatzitofis, Leonidas Saroglou, Prodromos Boutis, Petros Drakoulis, Nikolaos Zioulis, Shishir Subra- manyam, Bart Kevelham, Caecilia Charbonnier, Pablo Ce- sar, Dimitrios Zarpalas, et al. Human4d: A human-centric multimodal dataset for motions and immersive media.IEEE Access, 8:176241–176262, 2020

    work page 2020

  6. [6]

    Guide to the carnegie mellon university multimodal activity (cmu-mmac) database

    Fernando De la Torre, Jessica Hodgins, Adam Bargteil, Xavier Martin, Justin Macey, Alex Collado, and Pep Beltran. Guide to the carnegie mellon university multimodal activity (cmu-mmac) database. 2009

    work page 2009

  7. [7]

    Humans in 4d: Re- constructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Re- constructing and tracking humans with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14783–14794, 2023

    work page 2023

  8. [8]

    Oxford university press, 2004

    John C Gower and Garmt B Dijksterhuis.Procrustes prob- lems. Oxford university press, 2004

    work page 2004

  9. [9]

    Perspose: 3d human pose estima- tion with perspective encoding and perspective rotation

    Xiaoyang Hao and Han Li. Perspose: 3d human pose estima- tion with perspective encoding and perspective rotation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 8110–8119, 2025

    work page 2025

  10. [10]

    Human3.6m: Large scale datasets and predic- tive methods for 3d human sensing in natural environments

    Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3.6m: Large scale datasets and predic- tive methods for 3d human sensing in natural environments. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 36(7):1325–1339, 2014

    work page 2014

  11. [11]

    Panoptic studio: A massively multiview system for social motion capture

    Hanbyul Joo, Hao Liu, Lei Tan, Lin Gui, Bart Nabbe, Iain Matthews, Takeo Kanade, Shohei Nobuhara, and Yaser Sheikh. Panoptic studio: A massively multiview system for social motion capture. InThe IEEE International Conference on Computer Vision (ICCV), 2015

    work page 2015

  12. [12]

    Panoptic studio: A massively multiview sys- tem for social interaction capture.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2017

    Hanbyul Joo, Tomas Simon, Xulong Li, Hao Liu, Lei Tan, Lin Gui, Sean Banerjee, Timothy Scott Godisart, Bart Nabbe, Iain Matthews, Takeo Kanade, Shohei Nobuhara, and Yaser Sheikh. Panoptic studio: A massively multiview sys- tem for social interaction capture.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2017

    work page 2017

  13. [13]

    Total cap- ture: A 3d deformation model for tracking faces, hands, and bodies

    Hanbyul Joo, Tomas Simon, and Yaser Sheikh. Total cap- ture: A 3d deformation model for tracking faces, hands, and bodies. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 8320–8329, 2018

    work page 2018

  14. [14]

    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. InProc. International Conference on Computer Vision (ICCV), pages 11127–11137, 2021

    work page 2021

  15. [15]

    Learning to reconstruct 3d human pose and shape via model-fitting in the loop** supplementary ma- terial

    Nikos Kolotouros, Georgios Pavlakos, Michael J Black, and Kostas Daniilidis. Learning to reconstruct 3d human pose and shape via model-fitting in the loop** supplementary ma- terial

    work page

  16. [16]

    Learning to reconstruct 3d human pose and shape via model-fitting in the loop

    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, 2019

    work page 2019

  17. [17]

    Non-rigid structure from motion: Prior- free factorization method revisited

    Suryansh Kumar. Non-rigid structure from motion: Prior- free factorization method revisited. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 51–60, 2020

    work page 2020

  18. [18]

    Organic priors in non- rigid structure from motion

    Suryansh Kumar and Luc Van Gool. Organic priors in non- rigid structure from motion. InEuropean Conference on Computer Vision, pages 71–88. Springer, 2022

    work page 2022

  19. [19]

    Multi- body non-rigid structure-from-motion

    Suryansh Kumar, Yuchao Dai, and Hongdong Li. Multi- body non-rigid structure-from-motion. In2016 Fourth In- ternational Conference on 3D Vision (3DV), pages 148–156. IEEE, 2016

    work page 2016

  20. [20]

    Spatio- temporal union of subspaces for multi-body non-rigid structure-from-motion.Pattern Recognition, 71:428–443, 2017

    Suryansh Kumar, Yuchao Dai, and Hongdong Li. Spatio- temporal union of subspaces for multi-body non-rigid structure-from-motion.Pattern Recognition, 71:428–443, 2017

    work page 2017

  21. [21]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer vision–ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceedings, part v 13, pages 740–755. Springer, 2014

    work page 2014

  22. [22]

    Mosh: motion and shape capture from sparse markers.ACM Trans

    Matthew Loper, Naureen Mahmood, and Michael J Black. Mosh: motion and shape capture from sparse markers.ACM Trans. Graph., 33(6):220–1, 2014

    work page 2014

  23. [23]

    Smpl: a skinned multi- person linear model.ACM Transactions on Graphics (TOG), 34(6):1–16, 2015

    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):1–16, 2015

    work page 2015

  24. [24]

    Smpl: A skinned multi- person linear model

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multi- person linear model. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851–866. 2023

    work page 2023

  25. [25]

    Troje, Ger- ard Pons-Moll, and Michael J

    Naureen Mahmood, Nima Ghorbani, Nikolaus F. Troje, Ger- ard Pons-Moll, and Michael J. Black. AMASS: Archive of motion capture as surface shapes. InIEEE/CVF Interna- tional Conference on Computer Vision (ICCV), pages 5442– 5451, 2019

    work page 2019

  26. [26]

    Troje, Ger- ard Pons-Moll, and Michael J

    Naureen Mahmood, Nima Ghorbani, Nikolaus F. Troje, Ger- ard Pons-Moll, and Michael J. Black. Amass: Archive of motion capture as surface shapes.arXiv, 2019

    work page 2019

  27. [27]

    Meshcapade GmbH, T ¨ubingen, Germany, 2024

    Meshcapade GmbH.Meshcapade: The Digital Human Plat- form. Meshcapade GmbH, T ¨ubingen, Germany, 2024

    work page 2024

  28. [28]

    Dynamicfusion: Reconstruction and tracking of non-rigid scenes in real-time

    Richard A Newcombe, Dieter Fox, and Steven M Seitz. Dynamicfusion: Reconstruction and tracking of non-rigid scenes in real-time. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 343–352, 2015

    work page 2015

  29. [29]

    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. InProceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 10975–10985, 2019

    work page 2019

  30. [30]

    3d human pose estimation in video with tem- poral convolutions and semi-supervised training

    Dario Pavllo, Christoph Feichtenhofer, David Grangier, and Michael Auli. 3d human pose estimation in video with tem- poral convolutions and semi-supervised training. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7753–7762, 2019

    work page 2019

  31. [31]

    Frankmo- cap: Fast monocular 3d hand and body motion capture by regression and integration.arXiv preprint arXiv:2008.08324, 2020

    Yu Rong, Takaaki Shiratori, and Hanbyul Joo. Frankmo- cap: Fast monocular 3d hand and body motion capture by regression and integration.arXiv preprint arXiv:2008.08324, 2020

    work page arXiv 2008

  32. [32]

    Wham: Reconstructing world-grounded humans with accu- rate 3d motion

    Soyong Shin, Juyong Kim, Eni Halilaj, and Michael J Black. Wham: Reconstructing world-grounded humans with accu- rate 3d motion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2070– 2080, 2024

    work page 2070

  33. [33]

    Leonid Sigal, Alexandru O Balan, and Michael J Black. Hu- maneva: Synchronized video and motion capture dataset and baseline algorithm for evaluation of articulated human mo- tion.International journal of computer vision, 87(1):4–27, 2010

    work page 2010

  34. [34]

    Hand key- point detection in single images using multiview bootstrap- ping.CVPR, 2017

    Tomas Simon, Hanbyul Joo, and Yaser Sheikh. Hand key- point detection in single images using multiview bootstrap- ping.CVPR, 2017

    work page 2017

  35. [35]

    3d hu- man pose estimation via intuitive physics

    Shashank Tripathi, Lea M ¨uller, Chun-Hao P Huang, Omid Taheri, Michael J Black, and Dimitrios Tzionas. 3d hu- man pose estimation via intuitive physics. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4713–4725, 2023

    work page 2023

  36. [36]

    Umpm benchmark: A multi- person dataset with synchronized video and motion capture data for evaluation of articulated human motion and interac- tion

    NP Van der Aa, Xinghan Luo, Geert-Jan Giezeman, Robby T Tan, and Remco C Veltkamp. Umpm benchmark: A multi- person dataset with synchronized video and motion capture data for evaluation of articulated human motion and interac- tion. In2011 IEEE international conference on computer vi- sion workshops (ICCV Workshops), pages 1264–1269. IEEE, 2011

    work page 2011

  37. [37]

    Recovering accurate 3d human pose in the wild using imus and a moving camera

    Timo von Marcard, Roberto Henschel, Michael Black, Bodo Rosenhahn, and Gerard Pons-Moll. Recovering accurate 3d human pose in the wild using imus and a moving camera. In European Conference on Computer Vision (ECCV), 2018

    work page 2018

  38. [38]

    Prompthmr: Promptable human mesh recovery

    Yufu Wang, Yu Sun, Priyanka Patel, Kostas Daniilidis, Michael J Black, and Muhammed Kocabas. Prompthmr: Promptable human mesh recovery. InProceedings of the computer vision and pattern recognition conference, pages 1148–1159, 2025

    work page 2025

  39. [39]

    Decoupling human and camera motion from videos in the wild

    Vickie Ye, Georgios Pavlakos, Jitendra Malik, and Angjoo Kanazawa. Decoupling human and camera motion from videos in the wild. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. A Dataset and Evaluation for Complex 4D Markerless Human Motion Capture Supplementary Material Abstract Continuing with our main paper, this supplementary ...

    work page 2023

  40. [40]

    Motion and Activity Type In this supplementary, we provide a more detailed descrip- tion of the motion categories in HUM4D and explain how the dataset is organized. HUM4D is designed to capture challenging motion patterns that are not sufficiently repre- sented in existing markerless motion-capture benchmarks, including rapid local motion, heavy interacti...

    work page

  41. [41]

    As illustrated in Fig

    Dataset Arrangement In this section, we describe how HUM4D is organized for convenient access. As illustrated in Fig. 10 and Fig. 11, the dataset follows a hierarchical structure from motion type to action category, recording setting, take index, and camera streams and annotation files. At the top level, the dataset is divided into four mo- tion type grou...

    work page

  42. [42]

    Motion Type Analysis To further analyze method behavior on HUM4D, we report a breakdown of reconstruction performance by motion type. Since HUM4D is organized around four challenging motion categories, namelyOcclusion,ID Swap,Near-Far Cam- era, andJittering, this evaluation offers a more specific view of model behavior. As shown in Table 4,ID Swap Motion ...

    work page