pith. sign in

arxiv: 1907.10045 · v1 · pith:4HHVAR2Hnew · submitted 2019-07-23 · 💻 cs.CV

xR-EgoPose: Egocentric 3D Human Pose from an HMD Camera

Denis Tome , Patrick Peluse , Lourdes Agapito , Hernan Badino This is my paper

Pith reviewed 2026-05-24 17:19 UTC · model grok-4.3

classification 💻 cs.CV
keywords egocentric 3D pose estimationHMD camerafish-eye camerasynthetic datasetencoder-decoder architectureself-occlusionvirtual reality
0
0 comments X

The pith

A dual-branch decoder and 383K-frame synthetic dataset improve egocentric 3D pose from HMD fish-eye cameras.

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

The paper tackles 3D body pose estimation from a monocular downward fish-eye camera mounted just 2 cm from the face on a head-mounted VR device. This viewpoint produces severe self-occlusions and large resolution differences between upper and lower body. The authors introduce an encoder-decoder network whose dual-branch decoder explicitly models uncertainty differences across 2D joint locations. They also release the xR-EgoPose dataset of 383,000 photorealistic rendered frames spanning varied body shapes, skin tones, clothing, lighting and actions. Training on this corpus yields higher accuracy than prior egocentric methods on both synthetic and real test sets and matches the performance of leading third-person pose estimators on Human3.6M.

Core claim

An encoder-decoder architecture with a novel dual branch decoder that accounts for varying uncertainty in 2D joint locations, trained on the new xR-EgoPose corpus of 383K photorealistic frames, produces substantial accuracy gains over state-of-the-art egocentric pose estimators, generalizes from synthetic training data to real HMD footage, and reaches parity with top third-person methods on Human3.6M.

What carries the argument

Dual branch decoder that processes 2D joint detections according to their per-location uncertainty, trained on the xR-EgoPose photorealistic synthetic dataset.

If this is right

  • The dual-branch decoder yields measurable accuracy gains on both synthetic and real egocentric test sets.
  • High variability in the synthetic corpus produces good generalization to real-world HMD images without extra adaptation steps.
  • The same trained model reaches performance on par with leading third-person 3D pose methods when evaluated on Human3.6M.
  • The architecture directly addresses the self-occlusion and perspective-distortion problems unique to close-range fish-eye HMD views.

Where Pith is reading between the lines

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

  • The uncertainty-aware decoder may transfer to other monocular pose or tracking tasks where detection reliability varies across the frame.
  • Large-scale synthetic data generation with controlled appearance diversity could reduce reliance on costly real-world 3D annotations for wearable sensing.
  • Viewpoint-specific network designs may be required for reliable egocentric perception on head-mounted devices beyond pose estimation.

Load-bearing premise

The stated diversity in skin tones, body shapes, clothing, backgrounds and lighting inside the synthetic training set is enough for the learned model to generalize directly to real HMD footage without domain adaptation.

What would settle it

Accuracy measured on a new real HMD test set whose lighting, clothing, or body-shape distribution lies outside the ranges shown in the 383K synthetic frames, compared against the same baselines.

Figures

Figures reproduced from arXiv: 1907.10045 by Denis Tome, Hernan Badino, Lourdes Agapito, Patrick Peluse.

Figure 1
Figure 1. Figure 1: Left: Our xR-EgoPose Dataset setup: (a) external camera viewpoint showing a synthetic character wearing the headset; (b) example of photorealistic image rendered from the egocentric camera perspective; (c) 2D and (d) 3D poses estimated with our algorithm. Right: results on real images; (e) real image acquired with our HMD-mounted camera with predicted 2D heatmaps; (f) estimated 3D pose, showing good genera… view at source ↗
Figure 2
Figure 2. Figure 2: Example images from our xR-EgoPose Dataset compared with the competitor Mo2Cap2 dataset [55]. The quality of our frames is far superior than the randomly sampled frames from mo2cap2, where the characters suffer color matching with respect to the background light conditions. years [11, 25, 5], most methods detect, at most, only up￾per body motion (hands, arms or torso). Capturing full 3D body motion from he… view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of different poses with the same synthetic actor. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Our novel two-step architecture for egocentric 3D human pose estimation has two modules: [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of reconstructed heatmaps generated by [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results on synthetic and real images acquired with a camera physically mounted on a HMD: [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

We present a new solution to egocentric 3D body pose estimation from monocular images captured from a downward looking fish-eye camera installed on the rim of a head mounted virtual reality device. This unusual viewpoint, just 2 cm. away from the user's face, leads to images with unique visual appearance, characterized by severe self-occlusions and strong perspective distortions that result in a drastic difference in resolution between lower and upper body. Our contribution is two-fold. Firstly, we propose a new encoder-decoder architecture with a novel dual branch decoder designed specifically to account for the varying uncertainty in the 2D joint locations. Our quantitative evaluation, both on synthetic and real-world datasets, shows that our strategy leads to substantial improvements in accuracy over state of the art egocentric pose estimation approaches. Our second contribution is a new large-scale photorealistic synthetic dataset - xR-EgoPose - offering 383K frames of high quality renderings of people with a diversity of skin tones, body shapes, clothing, in a variety of backgrounds and lighting conditions, performing a range of actions. Our experiments show that the high variability in our new synthetic training corpus leads to good generalization to real world footage and to state of the art results on real world datasets with ground truth. Moreover, an evaluation on the Human3.6M benchmark shows that the performance of our method is on par with top performing approaches on the more classic problem of 3D human pose from a third person viewpoint.

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 an encoder-decoder network with a novel dual-branch decoder that accounts for varying 2D joint uncertainty, tailored to the severe self-occlusions and perspective distortions of downward fish-eye HMD imagery. It also releases the xR-EgoPose synthetic corpus of 383K photorealistic frames spanning diverse skin tones, body shapes, clothing, backgrounds, lighting, and actions. The central claims are that the architecture yields substantial accuracy gains over prior egocentric methods on both synthetic and real data, that the dataset's variability enables generalization to real HMD footage without adaptation, and that the method remains competitive with top third-person approaches on Human3.6M.

Significance. If the reported generalization holds, the work would meaningfully advance egocentric pose estimation for VR/AR by directly targeting the 2 cm fish-eye viewpoint. The released dataset would constitute a reusable resource for the community. The dual-branch decoder represents a targeted architectural response to a domain-specific uncertainty pattern.

major comments (2)
  1. [dataset generation and real-world experiments] The generalization claim (abstract and real-world evaluation section) rests on the assumption that the synthetic renderer matches real HMD camera parameters. No explicit verification is provided that the fisheye distortion model, principal-point offset induced by the 2 cm rim mount, or near-field HMD illumination pattern are replicated; if these differ, the reported SOTA numbers on real GT datasets may reflect partial domain overlap rather than robustness.
  2. [quantitative evaluation] Table reporting real-world results: the quantitative gains over prior egocentric baselines are presented without an ablation isolating the contribution of the dual-branch decoder versus the dataset scale/diversity alone, making it difficult to attribute the claimed improvements to the architectural component.
minor comments (2)
  1. [abstract] Abstract states 'substantial improvements' and 'state of the art results' without any numeric values or baseline names; adding at least the key error metrics and method names would strengthen the summary.
  2. [method] Notation for the dual-branch decoder outputs (e.g., heatmaps vs. uncertainty maps) should be defined once in a single equation block rather than re-introduced in the text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments point by point below, indicating planned revisions where appropriate.

read point-by-point responses

  1. Referee: [dataset generation and real-world experiments] The generalization claim (abstract and real-world evaluation section) rests on the assumption that the synthetic renderer matches real HMD camera parameters. No explicit verification is provided that the fisheye distortion model, principal-point offset induced by the 2 cm rim mount, or near-field HMD illumination pattern are replicated; if these differ, the reported SOTA numbers on real GT datasets may reflect partial domain overlap rather than robustness.

    Authors: We agree that additional details on parameter matching would strengthen the generalization claim. The xR-EgoPose renderer was configured using the exact fisheye intrinsics and 2 cm rim-mount offset reported by the HMD manufacturer, with principal-point adjustment derived from physical measurements of the device. Near-field illumination was approximated via multiple point lights positioned at typical HMD LED locations. In the revised manuscript we will insert a new subsection (under Dataset Generation) that explicitly lists the distortion coefficients, principal-point values, and lighting setup used, together with a side-by-side comparison of rendered versus real HMD images to document the match. This addresses the concern without requiring new data collection. revision: yes

  2. Referee: [quantitative evaluation] Table reporting real-world results: the quantitative gains over prior egocentric baselines are presented without an ablation isolating the contribution of the dual-branch decoder versus the dataset scale/diversity alone, making it difficult to attribute the claimed improvements to the architectural component.

    Authors: An ablation isolating the dual-branch decoder (versus single-branch baseline) is already reported on the synthetic corpus (Table 3 and Section 4.2), demonstrating consistent gains across training-set sizes. On real data the evaluation uses the full model because ground-truth annotations are scarce; a full factorial ablation on real footage would require additional labeled sequences that are not currently available. In the revision we will (i) explicitly cross-reference the synthetic ablation when discussing real-world numbers and (ii) add a short paragraph clarifying that the architecture’s benefit is quantified on synthetic data while the dataset diversity is the primary driver of zero-shot transfer. We consider this a partial revision because a complete real-data ablation cannot be performed without new annotations. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical architecture + dataset evaluation with no self-referential derivations

full rationale

The paper presents a dual-branch encoder-decoder for egocentric 3D pose from fisheye HMD images plus a new 383K-frame synthetic corpus. All performance claims rest on direct quantitative comparison against prior methods on held-out synthetic and real datasets with ground truth. No equations, fitted parameters renamed as predictions, self-citation load-bearing uniqueness theorems, or ansatzes appear in the provided text. Generalization from synthetic variability to real footage is asserted via measured accuracy, not by construction from the training distribution itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract alone supplies no extractable free parameters, axioms, or invented entities; method is described at high level only.

pith-pipeline@v0.9.0 · 5809 in / 1062 out tokens · 24037 ms · 2026-05-24T17:19:02.081800+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

61 extracted references · 61 canonical work pages · 5 internal anchors

  1. [1]

    Akhter and M

    I. Akhter and M. J. Black. Pose-conditioned joint angle lim- its for 3d human pose reconstruction. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 1446–1455, 2015. 2

    work page 2015

  2. [2]

    M. Amer, S. V . Amer, and A. Maria. Deep 3d human pose estimation under partial body presence. InProceedings of the IEEE International Conference on Image Processing (ICIP),

    work page

  3. [3]

    Andriluka, L

    M. Andriluka, L. Pishchulin, P. Gehler, and B. Schiele. 2d human pose estimation: New benchmark and state of the art analysis. In IEEE Conference on Computer Vision and Pat- tern Recognition (CVPR), June 2014. 2, 7

    work page 2014

  4. [4]

    F. Bogo, A. Kanazawa, C. Lassner, P. Gehler, J. Romero, and M. J. Black. Keep it smpl: Automatic estimation of 3d human pose and shape from a single image. In European Conference on Computer Vision , pages 561–578. Springer,

    work page

  5. [5]

    C. Cao, Y . Zhang, Y . Wu, H. Lu, and J. Cheng. Egocentric gesture recognition using recurrent 3d convolutional neural networks with spatiotemporal transformer modules. 2017 IEEE International Conference on Computer Vision (ICCV),

    work page 2017

  6. [6]

    Z. Cao, T. Simon, S.-E. Wei, and Y . Sheikh. Realtime multi- person 2d pose estimation using part affinity fields. InCVPR,

    work page

  7. [7]

    Chen and D

    C.-H. Chen and D. Ramanan. 3d human pose estimation= 2d pose estimation+ matching. InProceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 7035–7043, 2017. 7

    work page 2017

  8. [8]

    Learning 3D Human Pose from Structure and Motion

    R. Dabral, A. Mundhada, U. Kusupati, S. Afaque, and A. Jain. Structure-aware and temporally coherent 3d human pose estimation. arXiv preprint arXiv:1711.09250, 2017. 7

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  9. [9]

    Can 3D Pose be Learned from 2D Projections Alone?

    D. Drover, C.-H. Chen, A. Agrawal, A. Tyagi, and C. P. Huynh. Can 3d pose be learned from 2d projections alone? arXiv preprint arXiv:1808.07182, 2018. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  10. [10]

    H.-S. Fang, Y . Xu, W. Wang, X. Liu, and S.-C. Zhu. Learn- ing pose grammar to encode human body configuration for 3d pose estimation. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018. 7

    work page 2018

  11. [11]

    Fathi, A

    A. Fathi, A. Farhadi, and J. M. Rehg. Understanding ego- centric activities. In Proceedings of the International Con- ference on Computer Vision (ICCV), 2011. 2

    work page 2011

  12. [12]

    Glorot and Y

    X. Glorot and Y . Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intel- ligence and statistics, pages 249–256, 2010. 5

    work page 2010

  13. [13]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learn- ing for image recognition. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 770–778, 2016. 4, 5

    work page 2016

  14. [14]

    U. Hess, K. Kafetsios, H. Mauersberger, C. Blaison, and C. Kessler. Signal and noise in the perception of facial emo- tion expressions: From labs to life. Pers Soc Psychol Bull, 42(8), 2016. 1

    work page 2016

  15. [15]

    M. R. I. Hossain and J. J. Little. Exploiting temporal infor- mation for 3d human pose estimation. In European Confer- ence on Computer Vision, pages 69–86. Springer, 2018. 7

    work page 2018

  16. [16]

    https://medium.com/@DeepMotionInc/how-to-make- 3-point-tracked-full-body-avatars-in-vr-34b3f6709782. How to make 3 point tracked full-body avatars in vr, https://medium.com/@deepmotioninc/how-to-make-3- point-tracked-full-body-avatars-in-vr-34b3f6709782, last accessed on 2019-03-19. 1

    work page 2019

  17. [17]

    Animated 3d characters, last ac- cessed on 2019-03-19

    https://www.mixamo.com/. Animated 3d characters, last ac- cessed on 2019-03-19. 4

    work page 2019

  18. [18]

    Ionescu, D

    C. Ionescu, D. Papava, V . Olaru, and C. Sminchisescu. Hu- man3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE transactions on pattern analysis and machine intelligence , 36(7):1325– 1339, 2014. 2, 7

    work page 2014

  19. [19]

    Jahangiri and A

    E. Jahangiri and A. L. Yuille. Generating multiple diverse hypotheses for human 3d pose consistent with 2d joint detec- tions. In Proceedings of the IEEE International Conference on Computer Vision, pages 805–814, 2017. 7

    work page 2017

  20. [20]

    Jiang and K

    H. Jiang and K. Grauman. Seeing invisible poses: Estimating 3d body pose from egocentric video. In 2017 IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) , pages 3501–3509. IEEE, 2017. 3

    work page 2017

  21. [21]

    Kanazawa, M

    A. Kanazawa, M. J. Black, D. W. Jacobs, and J. Malik. End- to-end recovery of human shape and pose. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7122–7131, 2018. 7

    work page 2018

  22. [22]

    Li and A

    S. Li and A. B. Chan. 3d human pose estimation from monocular images with deep convolutional neural network. In Asian Conference on Computer Vision , pages 332–347. Springer, 2014. 2

    work page 2014

  23. [23]

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ra- manan, P. Doll´ar, and C. L. Zitnick. Microsoft coco: Com- mon objects in context. InEuropean conference on computer vision, pages 740–755. Springer, 2014. 7

    work page 2014

  24. [24]

    Loper, N

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black. Smpl: A skinned multi-person linear model. ACM Transactions on Graphics (TOG), 34(6):248, 2015. 2

    work page 2015

  25. [25]

    M. Ma, H. Fan, and K. M. Kitani. Going deeper into first- person activity recognition. 2016 IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 1894– 1903, 2016. 2

    work page 2016

  26. [26]

    Martinez, R

    J. Martinez, R. Hossain, J. Romero, and J. J. Little. A sim- ple yet effective baseline for 3d human pose estimation. In Proceedings of the International Conference on Computer Vision (ICCV), 2017. 2, 6, 7

    work page 2017

  27. [27]

    Mehta, H

    D. Mehta, H. Rhodin, D. Casas, P. Fua, O. Sotnychenko, W. Xu, and C. Theobalt. Monocular 3d human pose esti- mation in the wild using improved cnn supervision. In 3D Vision (3DV), 2017. 2, 7, 8

    work page 2017

  28. [28]

    Mehta, S

    D. Mehta, S. Sridhar, O. Sotnychenko, H. Rhodin, M. Shafiei, H.-P. Seidel, W. Xu, D. Casas, and C. Theobalt. Vnect: Real-time 3d human pose estimation with a single rgb camera. ACM Transactions on Graphics (TOG) , 36(4):44,

    work page

  29. [29]

    Moreno-Noguer

    F. Moreno-Noguer. 3d human pose estimation from a single image via distance matrix regression. In 2017 IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) , pages 1561–1570. IEEE, 2017. 2, 7

    work page 2017

  30. [30]

    Newell, K

    A. Newell, K. Yang, and J. Deng. Stacked hourglass net- works for human pose estimation. In European Conference on Computer Vision, pages 483–499. Springer, 2016. 2, 5

    work page 2016

  31. [31]

    S. Park, J. Hwang, and N. Kwak. 3d human pose estimation using convolutional neural networks with 2d pose informa- tion. In European Conference on Computer Vision, Work- shops, pages 156–169. Springer, 2016. 2

    work page 2016

  32. [32]

    Pavlakos, X

    G. Pavlakos, X. Zhou, K. G. Derpanis, and K. Daniilidis. Coarse-to-fine volumetric prediction for single-image 3d human pose. In Computer Vision and Pattern Recogni- tion (CVPR), 2017 IEEE Conference on , pages 1263–1272. IEEE, 2017. 2

    work page 2017

  33. [33]

    Pavlakos, L

    G. Pavlakos, L. Zhu, X. Zhou, and K. Daniilidis. Learning to estimate 3d human pose and shape from a single color image. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. 2, 5

    work page 2018

  34. [34]

    Pishchulin, E

    L. Pishchulin, E. Insafutdinov, S. Tang, B. Andres, M. An- driluka, P. V . Gehler, and B. Schiele. Deepcut: Joint sub- set partition and labeling for multi person pose estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4929–4937, 2016. 2

    work page 2016

  35. [35]

    Ramakrishna, T

    V . Ramakrishna, T. Kanade, and Y . Sheikh. Reconstructing 3d human pose from 2d image landmarks. InEuropean Con- ference on Computer Vision, pages 573–586. Springer, 2012. 2

    work page 2012

  36. [36]

    J. T. Reason and J. J. Brand. Motion sickness. Academic press, 1975. 1

    work page 1975

  37. [37]

    Rhodin, C

    H. Rhodin, C. Richardt, D. Casas, E. Insafutdinov, M. Shafiei, H.-P. Seidel, B. Schiele, and C. Theobalt. Ego- cap: egocentric marker-less motion capture with two fisheye cameras. ACM Transactions on Graphics (TOG), 35(6):162,

    work page

  38. [38]

    Unsupervised Geometry-Aware Representation for 3D Human Pose Estimation

    H. Rhodin, M. Salzmann, and P. Fua. Unsupervised geometry-aware representation for 3d human pose estima- tion. arXiv preprint arXiv:1804.01110, 2018. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  39. [39]

    Rogez and C

    G. Rogez and C. Schmid. Mocap-guided data augmentation for 3d pose estimation in the wild. In Advances in Neural Information Processing Systems, pages 3108–3116, 2016. 2, 7

    work page 2016

  40. [40]

    Rogez, J

    G. Rogez, J. S. Supancic, and D. Ramanan. First-person pose recognition using egocentric workspaces. In Proceedings of the IEEE conference on computer vision and pattern recog- nition, pages 4325–4333, 2015. 3

    work page 2015

  41. [41]

    Rogez, P

    G. Rogez, P. Weinzaepfel, and C. Schmid. Lcr-net: Localization-classification-regression for human pose. In CVPR 2017-IEEE Conference on Computer Vision & Pat- tern Recognition, 2017. 2

    work page 2017

  42. [42]

    LCR-Net++: Multi-person 2D and 3D Pose Detection in Natural Images

    G. Rogez, P. Weinzaepfel, and C. Schmid. LCR-Net++: Multi-person 2D and 3D Pose Detection in Natural Images. CoRR, abs/1803.00455v1, 2018. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  43. [43]

    Sanzari, V

    M. Sanzari, V . Ntouskos, and F. Pirri. Bayesian image based 3d pose estimation. In European Conference on Computer Vision, pages 566–582. Springer, 2016. 2

    work page 2016

  44. [44]

    Shiratori, H

    T. Shiratori, H. S. Park, L. Sigal, Y . Sheikh, and J. K. Hod- gins. Motion capture from body-mounted cameras. In ACM Transactions on Graphics (TOG), volume 30, page 31. ACM, 2011. 3

    work page 2011

  45. [45]

    X. Sun, J. Shang, S. Liang, and Y . Wei. Compositional hu- man pose regression. In Proceedings of the IEEE Interna- tional Conference on Computer Vision , pages 2602–2611,

    work page

  46. [46]

    X. Sun, B. Xiao, F. Wei, S. Liang, and Y . Wei. Integral hu- man pose regression. In Proceedings of the European Con- ference on Computer Vision (ECCV), pages 529–545, 2018. 2, 7

    work page 2018

  47. [47]

    Tekin, I

    B. Tekin, I. Katircioglu, M. Salzmann, V . Lepetit, and P. Fua. Structured prediction of 3d human pose with deep neural net- works. In British Machine Vision Conference (BMVC), 2016. 2

    work page 2016

  48. [48]

    D. Tome, C. Russell, and L. Agapito. Lifting from the deep: Convolutional 3d pose estimation from a single im- age. CVPR 2017 Proceedings, pages 2500–2509, 2017. 2, 7

    work page 2017

  49. [49]

    Tung, H.-W

    H.-Y . Tung, H.-W. Tung, E. Yumer, and K. Fragkiadaki. Self-supervised learning of motion capture. In Advances in Neural Information Processing Systems , pages 5242–5252,

    work page

  50. [50]

    H.-Y . F. Tung, A. Harley, W. Seto, and K. Fragkiadaki. Ad- versarial inversion: Inverse graphics with adversarial priors. arXiv preprint arXiv:1705.11166, 2017. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  51. [51]

    Varol, J

    G. Varol, J. Romero, X. Martin, N. Mahmood, M. J. Black, I. Laptev, and C. Schmid. Learning from synthetic humans. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017), 2017. 2

    work page 2017

  52. [52]

    von Marcard, B

    T. von Marcard, B. Rosenhahn, M. J. Black, and G. Pons- Moll. Sparse inertial poser: Automatic 3d human pose es- timation from sparse imus. In Computer Graphics Forum, volume 36, pages 349–360. Wiley Online Library, 2017. 3

    work page 2017

  53. [53]

    S.-E. Wei, V . Ramakrishna, T. Kanade, and Y . Sheikh. Con- volutional pose machines. In Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 4724–4732, 2016. 2, 5

    work page 2016

  54. [54]

    J. Wu, T. Xue, J. J. Lim, Y . Tian, J. B. Tenenbaum, A. Tor- ralba, and W. T. Freeman. Single image 3d interpreter net- work. In European Conference on Computer Vision, pages 365–382. Springer, 2016. 2

    work page 2016

  55. [55]

    W. Xu, A. Chatterjee, M. Zollhoefer, H. Rhodin, P. Fua, H.-P. Seidel, and C. Theobalt. Mo 2Cap2 : Real-time mo- bile 3d motion capture with a cap-mounted fisheye camera. IEEE Transactions on Visualization and Computer Graph- ics, pages 1–1, 2019. 2, 3, 6, 8

    work page 2019

  56. [56]

    Yasin, U

    H. Yasin, U. Iqbal, B. Kruger, A. Weber, and J. Gall. A dual- source approach for 3d pose estimation from a single image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4948–4956, 2016. 7

    work page 2016

  57. [57]

    Yonemoto, K

    H. Yonemoto, K. Murasaki, T. Osawa, K. Sudo, J. Shima- mura, and Y . Taniguchi. Egocentric articulated pose tracking for action recognition. In International Conference on Ma- chine Vision Applications (MVA), 2015. 3

    work page 2015

  58. [58]

    X. Zhou, X. Sun, W. Zhang, S. Liang, and Y . Wei. Deep kine- matic pose regression. InEuropean Conference on Computer Vision, pages 186–201. Springer, 2016. 2

    work page 2016

  59. [59]

    X. Zhou, M. Zhu, S. Leonardos, and K. Daniilidis. Sparse representation for 3d shape estimation: A convex relaxation approach. IEEE transactions on pattern analysis and ma- chine intelligence, 39(8):1648–1661, 2017. 2

    work page 2017

  60. [60]

    X. Zhou, M. Zhu, S. Leonardos, K. G. Derpanis, and K. Daniilidis. Sparseness meets deepness: 3d human pose estimation from monocular video. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 4966–4975, 2016. 2

    work page 2016

  61. [61]

    X. Zhou, M. Zhu, G. Pavlakos, S. Leonardos, K. G. Derpa- nis, and K. Daniilidis. Monocap: Monocular human motion capture using a cnn coupled with a geometric prior. IEEE transactions on pattern analysis and machine intelligence ,

    work page