Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

EventEgo3D++: 3D Human Motion Capture from a Head-Mounted Event Camera

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A single head-mounted event camera can capture a user's 3D body pose in real time, even in fast motion and dim light.

desk verdict A solid but incremental extension of the authors' own EventEgo3D, with a useful new dataset and systematic ablations; the main soft spot is that all accuracy numbers rest on unquantified motion-capture ground truth. read the letter →

arxiv 2502.07869 v1 pith:VA6VXEZE submitted 2025-02-11 cs.CV

Christen Millerdurai , Hiroyasu Akada , Jian Wang , Diogo Luvizon , Alain Pagani , Didier Stricker , Christian Theobalt , Vladislav Golyanik This is my paper · ORCID
classification cs.CV
keywords eventcameraegocentric3Dhumanposeestimationhead-mounteddeviceLNESrepresentationresidualpropagationlow-lightmotioncaptureSMPLbodymodelreal-time
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a single monocular event camera with a fisheye lens, mounted on a head-mounted device, is enough to estimate the wearer's 3D body pose accurately in real time, including under fast motion and low light where ordinary RGB cameras blur or saturate. It grounds this claim in an extension of an earlier event-based egocentric pose framework, adding new supervisory losses, a residual mechanism that propagates human-weighted events across frames, and an in-the-wild dataset with 3D ground truth. If the claim holds, head-mounted AR and VR hardware could track a user's full body with a low-power sensor that never suffers motion blur, at a reported update rate of 140Hz. The paper also contributes synthetic, studio, and outdoor datasets with SMPL body annotations for training and benchmarking.

What carries the argument

The load-bearing object is the residual event propagation module (REPM), built on the locally normalised event surface (LNES) representation. LNES compresses the asynchronous events of a time window into a two-channel image whose pixels hold the normalised time of the most recent brightness-change event, so a convolutional network can process events much like frames. The REPM adds a segmentation decoder that predicts the wearer's body mask, a confidence decoder that highlights important regions, and a frame buffer that adds the previous LNES frame weighted by the previous confidence map to the current frame; this makes human-generated events dominate background events and keeps the pose stable when few events are triggered.

What would settle it

Run an independent marker-based optical motion capture session simultaneously with the head-mounted device in an outdoor setting like EE3D-W and compare the two ground truths; if the hand-eye-calibrated ground truth differs from the independent system by more than about ten millimetres, the reported method-to-method differences, which are tens of millimetres, could be dominated by ground-truth calibration error rather than pose-estimation accuracy.

Watch

Extended reading notes

Core claim

The core discovery is that egocentric 3D human pose can be regressed directly from event streams rather than from reconstructed RGB images. EventEgo3D++ represents each event time window as a locally normalised event surface (LNES), feeds it through a U-Net-style encoder-decoder with Blaze blocks, lifts predicted 2D joint heatmaps to 16 camera-centric 3D joints, and uses a residual event propagation module to weight events generated by the user's body above background events while carrying pose information across near-static moments. Supervised jointly by heatmap loss, 3D joint loss, a 2D fisheye reprojection loss, a bone-orientation and bone-length loss, and a human-segmentation loss, the method reports a mean per-joint position error of 98.67 mm on the synthetic test set, 102.15 mm on the studio real dataset, and 166.19 mm on the in-the-wild dataset, with a 140Hz pose update rate and the same parameter count as its immediate predecessor.

Load-bearing premise

The reported accuracy numbers assume that the multi-view motion-capture ground truth, transformed into the head-mounted camera's frame by hand-eye calibration, is accurate enough to supervise and evaluate pose regression, and the paper does not quantify that ground-truth error.

Editorial extensions

If this is right

  • A monocular event camera, rather than an RGB camera, may be sufficient for egocentric full-body pose tracking in head-mounted devices, removing the motion-blur and low-light failure modes of RGB.
  • The 140Hz update rate and roughly 0.25W sensor power draw suggest that battery-friendly, on-device pose tracking is feasible for wearable hardware.
  • Because the residual propagation keeps pose estimates stable when few events are generated, the method can continue working during near-static moments, not only during fast motion.
  • The three released datasets, with SMPL bodies and allocentric RGB views, give other event-based egocentric methods a common training and evaluation benchmark.
  • Adapting RGB pose methods to operate on images reconstructed from events is measurably worse than consuming the event stream directly, pointing to the event stream itself as the right input representation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If event-only egocentric pose estimation holds at the reported accuracy, head-mounted devices could drop RGB sensing for pose tracking altogether, saving bandwidth and power while gaining speed and dynamic range.
  • The same LNES-plus-confidence-propagation recipe could transfer to egocentric hand tracking or whole-body capture from an event camera, where occlusion and sparse events are similarly severe.
  • The weakest joints are the feet and ankles, so a higher-resolution event sensor or an explicitly scale-aware lifting block would be a natural testable extension.
  • A quantitative uncertainty bound on the motion-capture ground truth would be the next natural experiment; without one, the exact error margins between methods remain partly dependent on calibration quality.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper presents EventEgo3D++, an extension of the authors' prior EventEgo3D system for egocentric 3D human pose estimation from a head-mounted fisheye event camera. The method uses the LNES event representation, a U-Net-style egocentric pose module with heatmap-to-3D lifting, and a residual event propagation module that combines a segmentation decoder, a confidence decoder, and a frame buffer. The claimed contributions are three new datasets (synthetic EE3D-S, studio EE3D-R, and in-the-wild EE3D-W), additional 2D reprojection and bone losses, event augmentation, and state-of-the-art results with real-time 140 Hz pose updates. Evaluation is performed on the authors' own datasets, with adapted versions of Xu et al. (2019), Tome et al. (2019), and Rudnev et al. (2021) as baselines.

Significance. If the results hold, the paper makes a useful contribution: it demonstrates that a monocular event camera on an HMD can support real-time egocentric 3D pose estimation, and it provides datasets that could be valuable for future event-based egocentric vision research. The ablation study in Table 5 is systematic and shows small but coherent improvements for each proposed component, and the efficiency analysis in Table 9 and Appendix A gives a clear picture of the computational and bandwidth advantages of the event-camera setup. The main weakness is that every quantitative claim is measured against self-generated motion-capture ground truth whose accuracy is not quantified, and all comparisons are on the authors' own datasets, so the central claim of superior accuracy is not yet firmly established.

major comments (3)
  1. [Sec. 5.2, Tables 1-3] The ground-truth accuracy is never quantified, and this is load-bearing for every reported number. Section 4.1.3 refers the reader to App. C.2 for 'accuracy of the generated ground-truth,' but App. C.2 contains only qualitative assertions: EE3D-R 'uses 30 cameras to minimise motion blur' and EE3D-W 'remains sufficient for accurate ground-truth capture' with six cameras at 60 fps. No checkerboard reprojection error, no repeatability of the hand-eye calibration MWE from Eq. (C1), no comparison against an independent reference, and no tracking-accuracy statistics are reported. Since the same ground truth supervises training (Eqs. (2)-(5)) and is used to compute MPJPE in Tables 1-3, a systematic bias in MWE or in the six-camera outdoor mocap would propagate directly into every headline number. The paper should either provide a quantitative validation of the ground-truth pipeline or explicitly rework the claims to avoid presenting the absolute MPJPE values as physical accuracy.
  2. [Sec. 5.2 and Tables 1-3] All reported numbers are point estimates with no error bars, confidence intervals, or multiple-seed statistics across subjects or sequences. For example, Table 2 reports EventEgo3D++ at MPJPE 102.15 versus Millerdurai et al. (2024a) at 107.30, but these are single numbers on a single test split. The comparison baselines are also adapted by the authors (first convolution layer changed for Xu et al. and Tome et al.; output layer changed for Rudnev et al.), and the modifications are not validated independently. Given the small margins for some activities, the claim of 'superior 3D accuracy' would be more convincing with variance estimates, per-subject results, and a statement about how much of the gain is robust to label noise.
  3. [Sec. 5.5 and Abstract] The abstract claims superior accuracy and robustness 'even in challenging conditions,' and Sec. 5.5 asserts operation in low-light environments, but the only low-light evidence is the qualitative demo in Fig. 16 with no ground-truth comparison, no controlled illumination sweep, and no quantitative metric. The low-light robustness claim is therefore broader than the evidence presented. A controlled experiment varying illumination while keeping motion fixed would be needed to substantiate this part of the central claim.
minor comments (4)
  1. [Sec. 5.1 and App. A] The network time window is stated as T = 33 ms in Sec. 5.1, but App. A computes bandwidth using T = 16.66 ms; please clarify which time window is used for the network input and which for the bandwidth measurement.
  2. [Figures 10-12] The figures label the proposed method 'EventEgo++' while the paper title and body use 'EventEgo3D++'; please standardize the caption text.
  3. [Table 9 and Sec. 5.4] Table 9 reports a pose update rate of 139.88, while the abstract and Sec. 5.4 state '140Hz'; either round consistently or report the measured value with its measurement variance.
  4. [Table 8] The notation in Table 8 is confusing: the header 'Weights' is followed by entries such as '0.01 (current)' and '0.1 (10x)', and the caption uses '✓' for the ablated loss and '✓' for enabled losses; please make the table self-explanatory and correct the symbol rendering.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the accuracy claims are empirical benchmark results against external mocap ground truth, not outputs forced by the method's own definitions or self-citations.

full rationale

EventEgo3D++ is an empirical systems and dataset paper rather than a derivation that could collapse into its inputs. The network is supervised by heatmap, 3D joint, 2D reprojection, bone, and segmentation losses (Eqs. 1-11), and all reported MPJPE/PA-MPJPE values are computed against multi-view RGB motion-capture ground truth (Captury/EasyMoCap) transformed by the hand-eye calibration matrix MWE (Sec. 4.1.3, App. C.1-C.3). That ground truth originates from an external pipeline, not from the model's own outputs, and no fitted parameter is renamed as a prediction. The same ground truth being used for both supervision and evaluation is a benchmark-quality limitation: the paper gives no quantitative error estimate for MWE or for the six-camera EE3D-W capture, so systematic mocap error would propagate into every table. However, this is not circularity because the reported errors are not equivalent to the training objective by construction. The main baseline is the authors' own EventEgo3D (Millerdurai et al, 2024a) and the benchmarks are self-recorded, but self-comparison and self-built datasets weaken external generalization rather than make the central claim logically circular. LNES is adopted from prior work as a design choice with a stated rationale (Sec. 3.1), not smuggled in as a theorem, and no uniqueness claim or ansatz-forcing citation is invoked. The low-light robustness claim is supported only qualitatively (Sec. 5.5), which is a support gap, not a circular step. No circular step can be quoted or exhibited, so the correct score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical entities, forces, or conserved quantities are postulated. The paper introduces hardware, datasets, and a network module, but its central empirical claims rest on the domain assumptions listed above, especially the accuracy of the mocap ground truth and the representativeness of synthetic event data.

free parameters (4)
  • Joint loss weights lambda_J3D, lambda_J2D, lambda_BA, lambda_H, lambda_seg = 0.01, 0.01, 1, 20, 0.1
    Chosen by hand in Sec. 3.2 and tested only by 10x perturbation in Table 8; they shape the training objective and therefore the reported MPJPE values.
  • Bone loss weights lambda_theta, lambda_BL = 0.001, 0.001
    Set in Sec. 3.2.1 without a full sweep; they control the strength of bone orientation and length supervision.
  • LNES time window T and input context N = T = 33 ms, N = 20
    Set in Sec. 5.1 without a search; T determines how much event history is aggregated into one frame and affects temporal resolution and accuracy.
  • Learning rates and iteration counts per dataset = 1e-3/8e5 (EE3D-S), 1e-4/1.5e4 (EE3D-R), 1e-4/1.2e4 (EE3D-W)
    Training schedule described in Sec. 5.1; these are standard choices but they affect convergence and final accuracy.
assumptions (5)
  • domain assumption Event generation model: events are triggered when the change in log-brightness exceeds threshold C at a pixel.
    Taken from event camera literature in Sec. 3.1; the LNES encoding and the whole pipeline depend on this sensor model being sufficient for pose estimation.
  • domain assumption LNES representation preserves enough spatiotemporal information for pose regression.
    Sec. 6 admits that multiple events at the same pixel overwrite each other, discarding some details; the accuracy claims are conditional on LNES being adequate.
  • domain assumption Ground-truth poses from Captury and EasyMoCap, transformed by hand-eye calibration MWE, are accurate enough for supervision and evaluation.
    Invoked in Sec. 4.1.3 and App. C.2; no quantitative ground-truth error is reported, and EE3D-W uses only six outdoor cameras, so systematic mocap error would propagate into all tables.
  • domain assumption VID2E-simulated event streams and background-augmented LNES frames are representative of real event camera behavior.
    Used in Sec. 4.2.2 and 4.2.4 for synthetic pretraining; the paper reports augmentation helps transfer but does not independently validate simulator fidelity.
  • domain assumption Visibility masks generated by ray casting in Blender match real occlusion in the egocentric camera.
    Described in App. C.3; the visibility supervision assumes the virtual camera and SMPL mesh align with the real HMD and wearer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EventEgo3D++: 3D Human Motion Capture from a Head-Mounted Event Camera." pith.science (2026). https://pith.science/paper/VA6VXEZE

@misc{pith2026250207869,
  author       = {Pith},
  title        = {Pith review of: EventEgo3D++: 3D Human Motion Capture from a Head-Mounted Event Camera},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VA6VXEZE}},
  note         = {Machine review of arXiv:2502.07869}
}
read the original abstract

Monocular egocentric 3D human motion capture remains a significant challenge, particularly under conditions of low lighting and fast movements, which are common in head-mounted device applications. Existing methods that rely on RGB cameras often fail under these conditions. To address these limitations, we introduce EventEgo3D++, the first approach that leverages a monocular event camera with a fisheye lens for 3D human motion capture. Event cameras excel in high-speed scenarios and varying illumination due to their high temporal resolution, providing reliable cues for accurate 3D human motion capture. EventEgo3D++ leverages the LNES representation of event streams to enable precise 3D reconstructions. We have also developed a mobile head-mounted device (HMD) prototype equipped with an event camera, capturing a comprehensive dataset that includes real event observations from both controlled studio environments and in-the-wild settings, in addition to a synthetic dataset. Additionally, to provide a more holistic dataset, we include allocentric RGB streams that offer different perspectives of the HMD wearer, along with their corresponding SMPL body model. Our experiments demonstrate that EventEgo3D++ achieves superior 3D accuracy and robustness compared to existing solutions, even in challenging conditions. Moreover, our method supports real-time 3D pose updates at a rate of 140Hz. This work is an extension of the EventEgo3D approach (CVPR 2024) and further advances the state of the art in egocentric 3D human motion capture. For more details, visit the project page at https://eventego3d.mpi-inf.mpg.de.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. From Sharp to Blur: Unsupervised Domain Adaptation for 2D Human Pose Estimation Under Extreme Motion Blur Using Event Cameras

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A domain adaptation method uses event-derived motion to synthesize blur and iteratively cleans pseudo-labels, improving multi-person 2D pose estimation on blurry images without annotations.

Reference graph

Works this paper leans on

87 extracted references · 77 canonical work pages · cited by 1 Pith paper

  1. [1]

    In: European Conference on Computer Vision (ECCV)

    Akada H, Wang J, Shimada S, et al (2022) Unrealego: A new dataset for robust egocentric 3d human motion capture. In: European Conference on Computer Vision (ECCV)

    2022

  2. [2]

    In: Computer Vision and Pattern Recognition (CVPR)

    Akada H, Wang J, Golyanik V, et al (2024) 3d human pose perception from egocentric stereo videos. In: Computer Vision and Pattern Recognition (CVPR)

    2024

  3. [3]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 13253--13262

    Aliakbarian S, Cameron P, Bogo F, et al (2022) Flag: Flow-based 3d avatar generation from sparse observations. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 13253--13262

    2022

  4. [4]

    arXiv preprint arXiv:200610204

    Bazarevsky V, Grishchenko I, Raveendran K, et al (2020) Blazepose: On-device real-time body pose tracking. arXiv preprint arXiv:200610204

    2020

  5. [5]

    Blender Foundation, Blender Institute, Amsterdam, ://www.blender.org

    Blender (2020) Blender - a 3D modelling and rendering package. Blender Foundation, Blender Institute, Amsterdam, ://www.blender.org

    2020

  6. [6]

    http://www.thecaptury.com/

    Captury (2024) Capturystudio - markerless mocap of humans from pre-recorded, multi-view video footage. http://www.thecaptury.com/

    2024

  7. [7]

    In: International Conference on 3D Vision (3DV)

    Chen J, Shi H, Ye Y, et al (2022) Efficient human pose estimation via 3d event point cloud. In: International Conference on 3D Vision (3DV)

    2022

  8. [8]

    ://mocap.cs.cmu.edu/

    CMU (2006) Cmu graphics lab motion capture database. ://mocap.cs.cmu.edu/

    work page 2006

Show all 87 references
  1. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 874--884

    Dai P, Zhang Y, Liu T, et al (2024) Hmd-poser: On-device real-time human motion tracking from scalable sparse observations. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 874--884

    2024

  2. [10]

    In: CVPR

    Du Y, Kips R, Pumarola A, et al (2023) Avatars grow legs: Generating smooth human motion from sparse tracking inputs with diffusion model. In: CVPR

    2023

  3. [11]

    https://netsket.kr/img/custom/board/DVXplorer-Mini.pdf

    DVXplorer Mini (2021) Dvxplorer mini specification. https://netsket.kr/img/custom/board/DVXplorer-Mini.pdf

    2021

  4. [12]

    https://github.com/zju3dv/EasyMocap

    EasyMoCap (2021) Easymocap - make human motion capture easier. https://github.com/zju3dv/EasyMocap

    2021

  5. [13]

    IEEE transactions on pattern analysis and machine intelligence 39(3):501--514

    Elhayek A, de Aguiar E, Jain A, et al (2016) Marconi—convnet-based marker-less motion capture in outdoor and indoor scenes. IEEE transactions on pattern analysis and machine intelligence 39(3):501--514

    2016

  6. [14]

    IEEE transactions on pattern analysis and machine intelligence 44(1):154--180

    Gallego G, Delbr \"u ck T, Orchard G, et al (2020) Event-based vision: A survey. IEEE transactions on pattern analysis and machine intelligence 44(1):154--180

    2020

  7. [15]

    In: Computer Vision and Pattern Recognition (CVPR)

    Gehrig D, Gehrig M, Hidalgo-Carri\'o J, et al (2020) Video to events: Recycling video datasets for event cameras. In: Computer Vision and Pattern Recognition (CVPR)

    2020

  8. [16]

    International Journal of Computer Vision 127:381--397

    Gilbert A, Trumble M, Malleson C, et al (2019) Fusing visual and inertial sensors with semantics for 3d human pose estimation. International Journal of Computer Vision 127:381--397

    2019

  9. [17]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 4318--4329

    Guzov V, Mir A, Sattler T, et al (2021) Human poseitioning system (hps): 3d human pose estimation and self-localization in large scenes from body-mounted sensors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 4318--4329

    2021

  10. [18]

    arXiv preprint arXiv:240913426

    Guzov V, Jiang Y, Hong F, et al (2024) Hmd ^2 : Environment-aware motion generation from single egocentric head-mounted device. arXiv preprint arXiv:240913426

    2024

  11. [19]

    In: Proceedings of the IEEE international conference on computer vision, pp 1105--1112

    Helten T, Muller M, Seidel HP, et al (2013) Real-time body tracking with one depth camera and inertial sensors. In: Proceedings of the IEEE international conference on computer vision, pp 1105--1112

    2013

  12. [20]

    ACM Transactions on Graphics (TOG) 37(6):1--15

    Huang Y, Kaufmann M, Aksan E, et al (2018) Deep inertial poser: Learning to reconstruct human pose from sparse inertial measurements in real time. ACM Transactions on Graphics (TOG) 37(6):1--15

    2018

  13. [21]

    https://github.com/itseez/opencv

    Itseez (2015) Open source computer vision library. https://github.com/itseez/opencv

    2015

  14. [22]

    In: European conference on computer vision, Springer, pp 443--460

    Jiang J, Streli P, Qiu H, et al (2022 a ) Avatarposer: Articulated full-body pose tracking from sparse motion sensing. In: European conference on computer vision, Springer, pp 443--460

    2022

  15. [23]

    arXiv e-prints pp arXiv--2308

    Jiang J, Streli P, Meier M, et al (2023) Egoposer: Robust real-time ego-body pose estimation in large scenes. arXiv e-prints pp arXiv--2308

    2023

  16. [24]

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    Jiang J, Li J, Zhang B, et al (2024 a ) Evhandpose: Event-based 3d hand pose estimation with sparse supervision. IEEE Transactions on Pattern Analysis and Machine Intelligence

    2024

  17. [25]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 24944--24954

    Jiang J, Zhou X, Wang B, et al (2024 b ) Complementing event streams and rgb frames for hand mesh reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 24944--24954

    2024

  18. [26]

    In: European Conference on Computer Vision, Springer, pp 128--146

    Jiang J, Streli P, Luo X, et al (2025) Manikin: biomechanically accurate neural inverse kinematics for human motion estimation. In: European Conference on Computer Vision, Springer, pp 128--146

    2025

  19. [27]

    In: SIGGRAPH Asia 2022 Conference Papers, pp 1--9

    Jiang Y, Ye Y, Gopinath D, et al (2022 b ) Transformer inertial poser: Real-time human motion reconstruction from sparse imus with simultaneous terrain generation. In: SIGGRAPH Asia 2022 Conference Papers, pp 1--9

    2022

  20. [28]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Kang T, Lee Y (2024) Attention-propagation network for egocentric heatmap to 3d pose lifting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2024

  21. [29]

    In: SIGGRAPH Asia 2023 Conference Papers

    Kang T, Lee K, Zhang J, et al (2023) Ego3dpose: Capturing 3d cues from binocular egocentric views. In: SIGGRAPH Asia 2023 Conference Papers. Association for Computing Machinery, New York, NY, USA, SA '23, doi:10.1145/3610548.3618147, ://doi.org/10.1145/3610548.3618147

    2023

  22. [30]

    Statistical Science 4(2):87--99

    Kendall DG (1989) A survey of the statistical theory of shape. Statistical Science 4(2):87--99

    1989

  23. [31]

    In: International Conference on Computer Vision (ICCV)

    Khirodkar R, Bansal A, Ma L, et al (2023) Ego-humans: An ego-centric 3d multi-human benchmark. In: International Conference on Computer Vision (ICCV)

    2023

  24. [32]

    In: International Conference on Learning Representations (ICLR)

    Kingma D, Ba J (2015) Adam: A method for stochastic optimization. In: International Conference on Learning Representations (ICLR)

    2015

  25. [33]

    arXiv preprint arXiv:230214430

    Lan C, Yin Z, Basu A, et al (2023) Tracking fast by learning slow: An event-based speed adaptive hand tracker leveraging knowledge in rgb domain. arXiv preprint arXiv:230214430

    2023

  26. [34]

    In: ACM SIGGRAPH 2023 Conference Proceedings, pp 1--9

    Lee S, Starke S, Ye Y, et al (2023) Questenvsim: Environment-aware simulated motion tracking from sparse sensors. In: ACM SIGGRAPH 2023 Conference Proceedings, pp 1--9

    2023

  27. [35]

    https://www.lensation.de/pdf/BF10M14522S118.pdf

    Lensagon BF10M14522S118C (2020) Lensagon bf10m14522s118 specification. https://www.lensation.de/pdf/BF10M14522S118.pdf

    2020

  28. [36]

    In: Computer Vision and Pattern Recognition (CVPR)

    Li J, Liu K, Wu J (2023 a ) Ego-body pose estimation via ego-head pose estimation. In: Computer Vision and Pattern Recognition (CVPR)

    2023

  29. [37]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 17142--17151

    Li J, Liu K, Wu J (2023 b ) Ego-body pose estimation via ego-head pose estimation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 17142--17151

    2023

  30. [38]

    IEEE Transactions on Multimedia

    Liu Y, Yang J, Gu X, et al (2023) Egofish3d: Egocentric 3d pose estimation from a fisheye camera via self-supervised learning. IEEE Transactions on Multimedia

    2023

  31. [39]

    ACM Trans Graphics (Proc SIGGRAPH Asia) 34(6):248:1--248:16

    Loper M, Mahmood N, Romero J, et al (2015) SMPL : A skinned multi-person linear model. ACM Trans Graphics (Proc SIGGRAPH Asia) 34(6):248:1--248:16

    2015

  32. [40]

    Advances in Neural Information Processing Systems (NeurIPS)

    Luo Z, Hachiuma R, Yuan Y, et al (2021) Dynamics-regulated kinematic policy for egocentric pose estimation. Advances in Neural Information Processing Systems (NeurIPS)

    2021

  33. [41]

    In: 2017 international conference on 3D vision (3DV), IEEE, pp 449--457

    Malleson C, Gilbert A, Trumble M, et al (2017) Real-time full-body motion capture from video and imus. In: 2017 international conference on 3D vision (3DV), IEEE, pp 449--457

    2017

  34. [42]

    ://www.mathworks.com

    MathWorks (2023) Matlab version: 9.14.0 (r2023a). ://www.mathworks.com

    2023

  35. [43]

    In: 3D Vision (3DV), 2018 Sixth International Conference on, IEEE, ://gvv.mpi-inf.mpg.de/projects/SingleShotMultiPerson

    Mehta D, Sotnychenko O, Mueller F, et al (2018) Single-shot multi-person 3d pose estimation from monocular rgb. In: 3D Vision (3DV), 2018 Sixth International Conference on, IEEE, ://gvv.mpi-inf.mpg.de/projects/SingleShotMultiPerson

    2018

  36. [44]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 1186--1195

    Millerdurai C, Akada H, Wang J, et al (2024 a ) Eventego3d: 3d human motion capture from egocentric event streams. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 1186--1195

    2024

  37. [45]

    In: International Conference on 3D Vision (3DV)

    Millerdurai C, Luvizon D, Rudnev V, et al (2024 b ) 3d pose estimation of two interacting hands from a monocular event camera. In: International Conference on 3D Vision (3DV)

    2024

  38. [46]

    In: Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    Muglikar M, Gehrig M, Gehrig D, et al (2021) How to calibrate your event camera. In: Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    2021

  39. [47]

    In: CVPR Workshop on Event-based Vision

    Nehvi J, Golyanik V, Mueller F, et al (2021) Differentiable event stream simulator for non-rigid 3d tracking. In: CVPR Workshop on Event-based Vision

    2021

  40. [48]

    In: International Conference on Computer Vision (ICCV)

    Pan X, Charron N, Yang Y, et al (2023) Aria digital twin: A new benchmark dataset for egocentric 3d machine perception. In: International Conference on Computer Vision (ICCV)

    2023

  41. [49]

    In: European Conference on Computer Vision, Springer, pp 343--359

    Park J, Moon G, Xu W, et al (2024) 3d hand sequence recovery from real blurry images and event stream. In: European Conference on Computer Vision, Springer, pp 343--359

    2024

  42. [50]

    Advances in Neural Information Processing Systems (NeurIPS)

    Paszke A, Gross S, Massa F, et al (2019) Pytorch: An imperative style, high-performance deep learning library. Advances in Neural Information Processing Systems (NeurIPS)

    2019

  43. [51]

    In: Computer Vision and Pattern Recognition (CVPR)

    Pavlakos G, Zhu L, Zhou X, et al (2018) Learning to estimate 3d human pose and shape from a single color image. In: Computer Vision and Pattern Recognition (CVPR)

    2018

  44. [52]

    In: Conference on Robot Learning (CORL)

    Rebecq H, Gehrig D, Scaramuzza D (2018) Esim: an open event camera simulator. In: Conference on Robot Learning (CORL)

    2018

  45. [53]

    Computer Vision and Pattern Recognition (CVPR)

    Rebecq H, Ranftl R, Koltun V, et al (2019 a ) Events-to-video: Bringing modern computer vision to event cameras. Computer Vision and Pattern Recognition (CVPR)

    2019

  46. [54]

    IEEE transactions on pattern analysis and machine intelligence 43(6):1964--1980

    Rebecq H, Ranftl R, Koltun V, et al (2019 b ) High speed and high dynamic range video with an event camera. IEEE transactions on pattern analysis and machine intelligence 43(6):1964--1980

    2019

  47. [55]

    ACM Transactions on Graphics (TOG) 35(6):1--11

    Rhodin H, Richardt C, Casas D, et al (2016) Egocap: egocentric marker-less motion capture with two fisheye cameras. ACM Transactions on Graphics (TOG) 35(6):1--11

    2016

  48. [56]

    In: International Conference on Medical image computing and computer-assisted intervention

    Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. In: International Conference on Medical image computing and computer-assisted intervention

    2015

  49. [57]

    In: International Conference on Computer Vision (ICCV)

    Rudnev V, Golyanik V, Wang J, et al (2021) Eventhands: Real-time neural 3d hand pose estimation from an event stream. In: International Conference on Computer Vision (ICCV)

    2021

  50. [58]

    In: Computer Vision and Pattern Recognition (CVPR)

    Rudnev V, Elgharib M, Theobalt C, et al (2023) Eventnerf: Neural radiance fields from a single colour event camera. In: Computer Vision and Pattern Recognition (CVPR)

    2023

  51. [59]

    In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Scaramuzza D, Martinelli A, Siegwart R (2006) A toolbox for easily calibrating omnidirectional cameras. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    2006

  52. [60]

    In: International Conference on Computer Vision (ICCV)

    Tome D, Peluse P, Agapito L, et al (2019) xr-egopose: Egocentric 3d human pose from an hmd camera. In: International Conference on Computer Vision (ICCV)

    2019

  53. [61]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 45(6):6794--6806

    Tome D, Alldieck T, Peluse P, et al (2020) Selfpose: 3d egocentric pose estimation from a headset mounted camera. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(6):6794--6806

    2020

  54. [62]

    In: Computer Vision and Pattern Recognition (CVPR)

    Varol G, Romero J, Martin X, et al (2017) Learning from synthetic humans. In: Computer Vision and Pattern Recognition (CVPR)

    2017

  55. [63]

    IEEE transactions on pattern analysis and machine intelligence 38(8):1533--1547

    Von Marcard T, Pons-Moll G, Rosenhahn B (2016) Human pose estimation from video and imus. IEEE transactions on pattern analysis and machine intelligence 38(8):1533--1547

    2016

  56. [64]

    In: Computer graphics forum, Wiley Online Library, pp 349--360

    Von Marcard T, Rosenhahn B, Black MJ, et al (2017) Sparse inertial poser: Automatic 3d human pose estimation from sparse imus. In: Computer graphics forum, Wiley Online Library, pp 349--360

    2017

  57. [65]

    In: International Conference on Computer Vision (ICCV)

    Wang J, Liu L, Xu W, et al (2021) Estimating egocentric 3d human pose in global space. In: International Conference on Computer Vision (ICCV)

    2021

  58. [66]

    In: Computer Vision and Pattern Recognition (CVPR)

    Wang J, Liu L, Xu W, et al (2022 a ) Estimating egocentric 3d human pose in the wild with external weak supervision. In: Computer Vision and Pattern Recognition (CVPR)

    2022

  59. [67]

    Computer Vision and Pattern Recognition (CVPR)

    Wang J, Luvizon D, Xu W, et al (2023) Scene-aware egocentric 3d human pose estimation. Computer Vision and Pattern Recognition (CVPR)

    2023

  60. [68]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 777--787

    Wang J, Cao Z, Luvizon D, et al (2024 a ) Egocentric whole-body motion capture with fisheyevit and diffusion-based motion refinement. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp 777--787

    2024

  61. [69]

    In: Computer Vision and Pattern Recognition (CVPR)

    Wang J, Cao Z, Luvizon D, et al (2024 b ) Egocentric whole-body motion capture with fisheyevit and diffusion-based motion refinement. In: Computer Vision and Pattern Recognition (CVPR)

    2024

  62. [70]

    In: European Conference on Computer Vision (ECCV)

    Wang Z, Chaney K, Daniilidis K (2022 b ) Evac3d: From event-based apparent contours to 3d models via continuous visual hulls. In: European Conference on Computer Vision (ECCV)

    2022

  63. [71]

    In: SIGGRAPH Asia 2022 Conference Papers, pp 1--8

    Winkler A, Won J, Ye Y (2022) Questsim: Human motion tracking from sparse sensors with simulated avatars. In: SIGGRAPH Asia 2022 Conference Papers, pp 1--8

    2022

  64. [72]

    In: Computer Vision and Pattern Recognition (CVPR)

    Xu L, Xu W, Golyanik V, et al (2020) Eventcap: Monocular 3d capture of high-speed human motions using an event camera. In: Computer Vision and Pattern Recognition (CVPR)

    2020

  65. [73]

    IEEE Transactions on Visualization and Computer Graphics 25(5):2093--2101

    Xu W, Chatterjee A, Zollhoefer M, et al (2019) Mo ^ 2 Cap ^ 2 : Real-time mobile 3d motion capture with a cap-mounted fisheye camera. IEEE Transactions on Visualization and Computer Graphics 25(5):2093--2101

    2019

  66. [74]

    In: British Machine Vision Conference (BMVC)

    Xue Y, Li H, Leutenegger S, et al (2022) Event-based non-rigid reconstruction from contours. In: British Machine Vision Conference (BMVC)

    2022

  67. [75]

    ACM Transactions On Graphics (TOG) 40(4):1--13

    Yi X, Zhou Y, Xu F (2021) Transpose: Real-time 3d human translation and pose estimation with six inertial sensors. ACM Transactions On Graphics (TOG) 40(4):1--13

    2021

  68. [76]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp 13167--13178

    Yi X, Zhou Y, Habermann M, et al (2022) Physical inertial poser (pip): Physics-aware real-time human motion tracking from sparse inertial sensors. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp 13167--13178

    2022

  69. [77]

    ACM Transactions on Graphics (TOG) 42(4):1--17

    Yi X, Zhou Y, Habermann M, et al (2023) Egolocate: Real-time motion capture, localization, and mapping with sparse body-mounted sensors. ACM Transactions on Graphics (TOG) 42(4):1--17

    2023

  70. [78]

    arXiv preprint arXiv:150603365

    Yu F, Zhang Y, Song S, et al (2015) Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint arXiv:150603365

    2015

  71. [79]

    In: International Conference on Computer Vision (ICCV)

    Yuan Y, Kitani K (2019) Ego-pose estimation and forecasting as real-time pd control. In: International Conference on Computer Vision (ICCV)

    2019

  72. [80]

    In: European conference on computer vision (ECCV)

    Zhang S, Ma Q, Zhang Y, et al (2022) Egobody: Human body shape and motion of interacting people from head-mounted devices. In: European conference on computer vision (ECCV)

    2022

  73. [81]

    In: International Conference on Computer Vision (ICCV)

    Zhang S, Ma Q, Zhang Y, et al (2023) Probabilistic human mesh recovery in 3d scenes from egocentric views. In: International Conference on Computer Vision (ICCV)

    2023

  74. [82]

    In: Winter Conference on Applications of Computer Vision

    Zhang Y, You S, Gevers T (2021) Automatic calibration of the fisheye camera for egocentric 3d human pose estimation from a single image. In: Winter Conference on Applications of Computer Vision

    2021

  75. [83]

    In: International Conference on 3D Vision (3DV)

    Zhao D, Wei Z, Mahmud J, et al (2021) Egoglass: Egocentric-view human pose estimation from an eyeglass frame. In: International Conference on 3D Vision (3DV)

    2021

  76. [84]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp 14678--14688

    Zheng X, Su Z, Wen C, et al (2023) Realistic full-body tracking from sparse observations via joint-level modeling. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp 14678--14688

    2023

  77. [85]

    In: International Conference on Computer Vision (ICCV)

    Zou S, Guo C, Zuo X, et al (2021) Eventhpe: Event-based 3d human pose and shape estimation. In: International Conference on Computer Vision (ICCV)

    2021

  78. [86]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version label extra.labe...

  79. [87]

    write newline

    " write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap dupl...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.