Pith. sign in

REVIEW 6 major objections 6 minor 40 references

Improvement of human health lifespan with hybrid group pose estimation methods

T0 review · 6 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A hybrid ensemble of two pose estimators is proposed to make real-time human pose tracking more reliable under occlusion.

desk verdict Claims a hybrid pose ensemble beats its parts, but all supporting evidence lives in the author's own inaccessible technical report. read the letter →

arxiv 2506.03169 v1 pith:CCCLKAP3 submitted 2025-05-28 cs.CV cs.AI

Arindam Chaudhuri This is my paper
classification cs.CVcs.AI
keywords humanposeestimationhybridensemblebaggingstackingtransformationocclusionrobustnessmulti-personreal-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

The paper proposes HEGPosEs, a hybrid ensemble for multi-person human pose estimation that combines a modified group pose estimator and a modified real-time pose estimator. The two base models' outputs are transformed by PoseTrans and then fused with bagging and stacking, using score-weighted averages of translations and a weighted chordal L2 mean of rotations. The author claims this ensemble yields the best optimized real-time pose estimation results, makes pose estimation more robust to occlusion, and improves dense regression accuracy, with potential gains for human health monitoring across the lifespan.

What carries the argument

The central object is the HEGPosEs ensemble itself, built from two base estimators: MGPosEs and MRTMPPosEs. PoseTrans supplies the pose transformation that generates diverse training samples and filters implausible ones. The fusion machinery is bagging, both simple and score-weighted, where the refined translation is computed as the weighted mean of individual translations and the refined rotation as the weighted chordal L2 mean over rotations (minimizing the weighted sum of squared Frobenius differences), followed by stacking with regularized regression and tree-based models. These formulas carry the claim: if the base models' poses are consistent, the weighted averages should reduce variance and improve accuracy.

What would settle it

Run HEGPosEs and both base models on a public occlusion-heavy pose benchmark (e.g., MPII with occluded test subsets) and compare per-keypoint mAP: if the ensemble does not exceed both base models on occluded bodies, the paper's central robustness claim is contradicted.

Watch

Extended reading notes

Core claim

The central claim is that HEGPosEs, as an ensemble of MGPosEs and MRTMPPosEs, outperforms each base model alone on real-time multi-person pose estimation. The fusion is done by simple or score-weighted bagging, in which refined translation is a weighted mean and refined rotation is a weighted chordal L2 mean, followed by stacking with models such as ridge regression, random forests, and XGBoost. Evaluated on DensePose-COCO and MPII Human Pose, the paper reports that the method is more robust to occlusion and improves dense regression accuracy, and it concludes that this supports real-time applications in clinical, sports, and media settings with improvement in human health lifespan.

Load-bearing premise

The load-bearing premise is that the two base models' outputs are already expressed in the same global coordinate frame with persons matched, so simple weighted averaging of translations and rotations produces a meaningful combined pose.

Editorial extensions

If this is right

  • Real-time pose estimation systems would become more reliable in occluded scenes, since the ensemble is claimed to be more adaptive to occlusion.
  • Dense regression accuracy for multi-person keypoint localization would improve over either base method alone.
  • The bagging and stacking pipeline would reduce prediction variance and generalization error, making the output more trustworthy for downstream movement analysis.
  • Health-monitoring applications that rely on human movement kinematics could benefit from more accurate pose tracking in everyday settings.

Reading between the lines

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

  • The averaging formulas presuppose that the two base models output poses in a shared coordinate frame with matched persons; if that alignment is not enforced, the ensemble could produce poses no better than the worse base model.
  • The health-lifespan framing is an application-level inference: improved pose accuracy would need to be paired with clinical studies that connect movement kinematics to health outcomes before the lifespan claim can be substantiated.
  • The same bagging-and-stacking scheme could in principle be applied to any pair of pose estimators, so the method may generalize beyond the two specific base models tested.
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

6 major / 6 minor

Summary. The manuscript proposes HEGPosEs, a hybrid ensemble for multi-person pose estimation that combines a modified Group Pose method (MGPosEs) and a modified RTMPose method (MRTMPPosEs) using PoseTrans-based pose transformation, bagging/stacking ensembles, and transfer learning. The paper claims that HEGPosEs provides best optimized real-time pose estimation, improves robustness to occlusion and dense regression accuracy, and has potential to improve human health lifespan. The method is described with equations for simple and weighted bagging, and the authors state that experiments were conducted on DensePose-COCO and MPII Human Pose datasets. However, no quantitative experimental results are presented in the visible text: all result tables and figures appear only as captions, and the only cited evidence is the authors' own inaccessible technical report [20].

Significance. If the central claim were supported by data, a hybrid ensemble of Group Pose and RTMPose could be a useful practical contribution to real-time multi-person pose estimation. The paper does include a clear mathematical formulation for pose averaging in Eqs. (1)-(5) and identifies relevant base methods and benchmark datasets. However, as submitted, the manuscript provides no measurable evidence: there are no accuracy numbers, error bars, ablations, or comparisons to the base models. The heavy reliance on an inaccessible self-citation for both the method identity and the evaluation metrics additionally prevents independent verification. The health-lifespan claim is entirely unsupported by any experimental or clinical data. These issues affect the central claim directly, so the current manuscript cannot be considered a valid contribution to the literature.

major comments (6)
  1. [Section 5, Tables 2-7] The central claim that HEGPosEs 'provides best optimized results' and is 'more robust to occlusion' is not supported by any numerical results in the visible text. Tables 2-7 appear as captions only; no mAP, PCKh, precision, recall, F1, convergence, or inference-speed values are printed. Figures 8-10 are similarly placeholders. Without measured comparisons against MGPosEs and MRTMPPosEs on MPII and DensePose-COCO, the paper contains no evidence for its central claim.
  2. [Sections 3 and 4] The experimental setup is internally inconsistent: Section 3 states 'With 50 epochs and batch size of 200 this model is trained and validated,' while Section 4 states 'With 50 epochs and batch size of 20 we train and validate this model.' This discrepancy makes it unclear which configuration produced the (absent) results and undermines reproducibility.
  3. [Section 3, Section 5, and ref. [20]] The proposed method HEGPosEs is introduced by citing the author's own technical report [20], and the validation metrics are also attributed to [20] ('multi-task loss and reconstruction loss [20]' and 'mAP evaluation metric [20]'). Because [20] is not provided or publicly located, both the method definition and the evaluation criteria rest on an inaccessible self-citation, making the central claim effectively unfalsifiable in this manuscript.
  4. [Section 3, Eqs. (1)-(5)] Equations (1)-(5) average translations and rotations from MGPosEs and MRTMPPosEs using a weighted mean and chordal L2 rotation averaging, but the paper never discusses whether the two models produce poses in a common coordinate frame, how per-person associations are aligned, or how missing keypoints and occlusions are handled. If the poses are not consistently aligned, the ensemble can be worse than either base model; the claimed occlusion robustness does not follow from the math.
  5. [Section 5 and Abstract] The manuscript states 'We do not have single best ensemble method' in Section 5, yet the Abstract claims the method 'provides best optimized results in real-time pose estimation.' This internal contradiction is not resolved and calls into question what the paper actually demonstrates.
  6. [Abstract and Section 6] The claimed 'improvement in human health life span' is asserted without any health outcome data, clinical validation, or causal argument. The experiments address only pose estimation accuracy on image benchmarks; no connection between those metrics and lifespan is established, so this conclusion is unsupported.
minor comments (6)
  1. [Table 1] Table 1, which is supposed to list hyperparameter values, contains only a caption; no hyperparameter values are actually shown.
  2. [References] Several references appear mismatched; for example, [15] is cited as a transformer-based pose estimation method, but the reference is to Video Swin Transformer, and [27] duplicates [7].
  3. [Section 5] The hardware description '128-bit OS having x128 Intel processor' is not meaningful; please specify the CPU/GPU model and memory correctly.
  4. [Section 1] There are grammatical errors and duplicated phrases, e.g., 'These dependencies lead to lead to normalization' in Section 1.
  5. [Keywords] The keywords include 'Decision support, social media,' which do not reflect the paper's content; consider more relevant keywords such as 'multi-person pose estimation' and 'ensemble learning.'
  6. [Figures 8-10] Figures 8-10 are referenced only as captions; the actual plots and curves are missing from the manuscript.

Circularity Check

3 steps flagged · score 8.0 of 10

HEGPosEs is defined by the author's own technical report [20] and then validated with metrics also attributed to [20], so the central claim reduces to a self-citation chain with no independently visible measurements.

  1. self citation load bearing [Section 1, final paragraph (HEGPosEs introduction; reference [20])]
    "Based on motivation from [18] and [19], a novel hybrid ensemble based group pose (HEGPosEs) [20] for end-to-end multi-person pose estimation is presented."

    The central artifact of the paper is introduced not by derivation here but by citation to the author's own technical report [20]. All subsequent claims about HEGPosEs' superiority inherit whatever that inaccessible report contains. The current text contains no independent construction of HEGPosEs beyond naming MGPosEs, MRTMPPosEs, PoseTrans, and bagging/stacking equations, so the claim that HEGPosEs is a novel method is load-bearing self-citation.

  2. self citation load bearing [Section 3, paragraph after Fig. 5]
    "The validation is performed through metrics multi-task loss and reconstruction loss [20]."

    The criteria used to validate HEGPosEs are attributed to the same technical report [20] that defines HEGPosEs. Thus the validation is not external to the source of the method: both the object being evaluated and the evaluative metric are fixed by the same self-citation, so the claimed validation cannot independently establish the method's value.

1 more flagged steps
  1. self citation load bearing [Section 5, first paragraph]
    "The mean average precision (mAP) evaluation metric [20] is used to validate results."

    The quantitative benchmark metric is likewise sourced from [20], the author's own report. Because no measured mAP, precision/recall, or F1 values appear in Tables 2-7, the only route from the equations to the conclusion 'best optimized results' is citation to [20] itself. The experimental claim therefore rests entirely on the self-citation chain.

full rationale

The paper's strongest claim is that HEGPosEs 'provides best optimized results in real-time pose estimation' and improves occlusion robustness and dense regression accuracy. Walking the derivation chain from equations (1)-(5): those equations only define simple/weighted bagging means of translations and rotations; they do not derive any performance advantage. The claimed validation is then supported by two explicit citations to the author's own technical report [20]: the metric 'multi-task loss and reconstruction loss [20]' in Section 3, and 'mAP evaluation metric [20]' in Section 5. The method's own identity is also assigned by citation: 'a novel hybrid ensemble based group pose (HEGPosEs) [20]'. Tables 2-7 are captions without printed values and Figures 8-10 are captions, so the text presents no reproducible external benchmark against MGPosEs or MRTMPPosEs. Because the defining document and the validating metrics coincide in the same inaccessible self-citation, the central claim reduces to the self-citation chain rather than to independent evidence. This is not a minor self-citation: the method's existence, its evaluation criterion, and the reported 'best results' all route through [20].

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

The central claim does not rest on unseen physical entities, but it rests on several unjustified choices: split ratios, epochs, batch sizes, ensemble hyperparameters, and a self-defined evaluation framework. These are all effectively free parameters or self-referential assumptions that prevent independent verification.

free parameters (5)
  • Number of training epochs = 50
    Chosen in Section 3 and Section 4 without tuning analysis.
  • Batch size = 200 in Section 3, 20 in Section 4
    The paper gives two different batch sizes, and neither is justified.
  • Epsilon in weighted bagging = small positive number
    Equation (3) introduces epsilon to avoid division by zero; no value is given.
  • Stacking model hyperparameters = not specified
    Ridge, Random Forests, XGBoost, SVR, and MLP are listed without their hyperparameters.
  • Train-validation split ratio = 80:20
    Chosen in Section 3 without justification.
assumptions (4)
  • domain assumption Pose estimations from 2D images provide sufficient kinematic information to infer health and lifespan.
    The abstract and conclusion tie pose estimation to 'improvement in human health life span' without a mechanism or supporting data.
  • domain assumption Averaging translations and rotations with equations (1) through (5) yields a valid ensemble pose.
    The paper assumes the two base models produce outputs in a common coordinate frame and that per-person association is resolved, which is not discussed.
  • domain assumption PoseTrans generates plausible poses that improve ensemble training.
    The paper relies on the prior PoseTrans method [37] without new validation of its suitability for this ensemble.
  • ad hoc to paper The evaluation metrics used are those defined in the author's own technical report [20].
    The paper cites its own report for 'multi-task loss' and 'reconstruction loss' metrics, creating a self-referential validation loop.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improvement of human health lifespan with hybrid group pose estimation methods." pith.science (2026). https://pith.science/paper/CCCLKAP3

@misc{pith2026250603169,
  author       = {Pith},
  title        = {Pith review of: Improvement of human health lifespan with hybrid group pose estimation methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CCCLKAP3}},
  note         = {Machine review of arXiv:2506.03169}
}
read the original abstract

Human beings rely heavily on estimation of poses in order to access their body movements. Human pose estimation methods take advantage of computer vision advances in order to track human body movements in real life applications. This comes from videos which are recorded through available devices. These para-digms provide potential to make human movement measurement more accessible to users. The consumers of pose estimation movements believe that human poses content tend to supplement available videos. This has increased pose estimation software usage to estimate human poses. In order to address this problem, we develop hybrid-ensemble-based group pose estimation method to improve human health. This proposed hybrid-ensemble-based group pose estimation method aims to detect multi-person poses using modified group pose estimation and modified real time pose estimation. This ensemble allows fusion of performance of stated methods in real time. The input poses from images are fed into individual meth-ods. The pose transformation method helps to identify relevant features for en-semble to perform training effectively. After this, customized pre-trained hybrid ensemble is trained on public benchmarked datasets which is being evaluated through test datasets. The effectiveness and viability of proposed method is estab-lished based on comparative analysis of group pose estimation methods and ex-periments conducted on benchmarked datasets. It provides best optimized results in real-time pose estimation. It makes pose estimation method more robust to oc-clusion and improves dense regression accuracy. These results have affirmed po-tential application of this method in several real-time situations with improvement in human health life span

Figures

Figures reproduced from arXiv: 2506.03169 by the authors.

Figure 4
Figure 4. Here [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [20]

    : HEGPosEs: Hybrid ensemble -based group pose estimation for modeling human body configurations

    Chaudhuri, A. : HEGPosEs: Hybrid ensemble -based group pose estimation for modeling human body configurations. Technical Report, Samsung R & D Institute, New Delhi, India (2025)

    work page 2025

  2. [1]

    Hartanto, A., Quek, F. Y. X., Tng, G. Y. Q., Yong, J. C. : Does social media use increase depressive symptoms? A reverse causation perspective. Frontiers in Psychiatry, 12, 641934 (2021)

    work page 2021

  3. [2]

    : Human modeli ng and pose estimation overview

    Knap, P. : Human modeli ng and pose estimation overview. viso.ai. https://viso.ai/deep- learning/pose-estimation-ultimate-overview/ (2025)

    work page 2025

  4. [3]

    Proceedings of ACM International Joint Conference on Pervasive and Ubiquitous Computing, pp

    Leng, Z., Jain, Y., Kwon, H., Plötz, T.: Fine-grained human activity recognition using vir- tual o n-body acceleration data. Proceedings of ACM International Joint Conference on Pervasive and Ubiquitous Computing, pp. 55-59. (2023)

    work page 2023

  5. [4]

    M., Allen, C

    Stenum, J., Cherry, K. M., Allen, C. O., Pyles, R. D., Reetzke, R. D., Vignos, M. F., Roemmich, R. T. : Applications of pos e estimation in human health and p erformance across the lifespan. Sensors, 21(21), 7315 (2021)

    work page 2021

  6. [5]

    : Deep learning based human pose estimation: A survey

    Zheng, C., Wu, W., Chen, C., Yang, T., Zu, S., Shen, J., Kehtarnavaz, N., Shah, M. : Deep learning based human pose estimation: A survey. ACM Computing Surveys , 56(1), 1 -37 (2023)

    work page 2023

  7. [6]

    : Human pose estimation

    Sigal, L. : Human pose estimation . In K. Ikeuchi (Ed.), Computer Vision , pp. 573 –592 (2021)

    work page 2021

  8. [7]

    Computer Vision and Image Understanding, 210, 103225

    Wang, J., Tan, S., Zhen, X ., Xu, S., Zheng, F., He, Z., Shao, L.: Deep 3D human pose es- timation: A review. Computer Vision and Image Understanding, 210, 103225. (2021)

    work page 2021

Show all 40 references
  1. [8]

    Innovations in Power and Advanced Computing Technologies (i -PACT), Kuala Lumpur, Malaysia, pp

    Amrutha, K., Prabu, P., Paulose, J.: Human Body Po se Estimation and Applica- tions. Innovations in Power and Advanced Computing Technologies (i -PACT), Kuala Lumpur, Malaysia, pp. 1-6 (2021)

    2021

  2. [9]

    Pro- ceedings of European Conference on Computer Vision, pp

    Xiao, B., Wu, H., Wei, Y.: Simple baselines for human pose estimation and tracking. Pro- ceedings of European Conference on Computer Vision, pp. 472-487 (2018)

    2018

  3. [10]

    Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp

    Sun, K., Xiao, B., Liu, D., Wang, J.: Deep high-resolution representation learn ing for hu- man pose es timation. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp. 5693-5703 (2019)

    2019

  4. [11]

    Proceedings of IEEE Con- ference on Computer Vision and Pattern Recognition, pp

    Cheng, B., Xiao, B., Wang, J., Shi, H., Huang, T., Zhang, L.: HigherHRNet: Scale-aware representation learning for b ottom-up human pose estimation. Proceedings of IEEE Con- ference on Computer Vision and Pattern Recognition, pp. 5386–5395 (2022)

    2022

  5. [12]

    : Poseur: Direct human pose regression with transformers

    Mao, W., Ge, Y., Shen, C., Tian, Z., Wang, X., & Wang, Z. : Poseur: Direct human pose regression with transformers. Proceedings of European Conference on Compute r Vision, pp. 72-88 (2022)

    2022

  6. [13]

    Proceedings of IEEE Conference on Computer Vision and Pattern Recogni- tion, pp

    Shi, D., Wei, X., Li, L., Ren, Y., Tan, W.: End-to-end multi-person pose estimation with transformers. Proceedings of IEEE Conference on Computer Vision and Pattern Recogni- tion, pp. 11069-11078 (2022)

    2022

  7. [14]

    Proceedings of European Conference on Computer Vi- sion, pp

    Carion, N., Massa, F., Sy nnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S.: End-to-end object detection with transformers. Proceedings of European Conference on Computer Vi- sion, pp. 213-229 (2020)

    2020

  8. [15]

    Proceed- ings of IEEE International Conference on Computer Vision and Pattern Recognition , pp

    Liu, Z., Ning, J., Cao, Y., Wei, Y., Zhang, Z., Lin, S.: Video swin transformer . Proceed- ings of IEEE International Conference on Computer Vision and Pattern Recognition , pp. 3202-3211 (2022)

    2022

  9. [16]

    T., Zhou, E.: Token Pose: Learning keypoint tokens for human pose estimation

    Li, Y., Zhang, S., Wang, Z., Yang, S., Xia, S. T., Zhou, E.: Token Pose: Learning keypoint tokens for human pose estimation. Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp. 11313-11322 (2021)

    2021

  10. [17]

    Sensors, 24(4), 1076 (2024)

    Guan, J., Hao, Y., Wu, Q., Li, S ., Fang, Y.: A survey of 6DoF object pose estimation methods for different application scenarios. Sensors, 24(4), 1076 (2024)

    2024

  11. [18]

    Proceedings of IEEE International Conference on Computer Vision, pp

    Liu, H., Zhang, F., Zhang, H., Zhang, S., Zhang, L., Su, H., Zhu, J., Ni, L., Zhang, Y.: Group Pose: A simple baseline for end -to-end multi-person pose estimation. Proceedings of IEEE International Conference on Computer Vision, pp. 14983-14992 (2023)

    2023

  12. [19]

    , Zeng, G., Li, H., Yuan, Y., Wang, J.: RTMPose: Real-time multi-person pose estimation based on MMPose

    Jiang, T., Meng, D., Chen, X., Fan, Z. , Zeng, G., Li, H., Yuan, Y., Wang, J.: RTMPose: Real-time multi-person pose estimation based on MMPose. 2D Human Pose Estimation on Art (2023)

    2023

  13. [21]

    A., Neverova, N., Kokkinos, I.: DensePose: Dense human pose estimation in the wild

    Güler, R. A., Neverova, N., Kokkinos, I.: DensePose: Dense human pose estimation in the wild. Proceedings of IEEE Conference on Computer Vision and Patt ern Recognition, pp. 7297-7306 (2018)

    2018

  14. [22]

    S., Kumar, A., Singh, O

    Sengar, S. S., Kumar, A., Singh, O. : Efficient human pose estimation: Leveraging ad- vanced techniques with MediaPipe. AIModels.fyi (2024)

    2024

  15. [23]

    Proceedings of IEEE International Conference on Computer Vision and Pat- tern Recognition (2021)

    Stoffl, L., Vidal, M., Mathis, A.: End-to-end trainable multi-instance pose estimation with transformers. Proceedings of IEEE International Conference on Computer Vision and Pat- tern Recognition (2021)

    2021

  16. [24]

    Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp

    Chen, Z., Wang, Y., Peng, Z., Zhang, Z., Yu, G., Sun, J.: Cascaded pyramid network fo r multi-person pose estimation. Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp. 7103–7112 (2018)

    2018

  17. [25]

    S., Xie, S., Tai, Y

    Fang, H. S., Xie, S., Tai, Y. -W., Lu, C.: RMPE: Regional multi-person pose estimation. Proceedings of IEEE International Conference on Computer Vision, pp. 2353-2362 (2017)

    2017

  18. [26]

    E., Sheikh, Y.: Real-time multi-person 2D pose estimation us- ing part affinity fields

    Cao, Z., Simon, T., Wei, S. E., Sheikh, Y.: Real-time multi-person 2D pose estimation us- ing part affinity fields. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 1302, 1302-1310 (2017)

    2017

  19. [27]

    : Deep 3D human pose es- timation: A review

    Wang, J., Tan, S., Zhen, X., Xu, S., Zheng, F., He, Z., Shao, L. : Deep 3D human pose es- timation: A review. Computer Vision and Image Understanding, 210, 103225 (2021)

    2021

  20. [28]

    Proceedings of International Conference on Learning Representations, pp

    Zhu, X., Su, W., Lu, L., Li, B., Wang, X., Dai, J.: Deformable DETR: Deformable trans- formers f or end -to-end object de tection. Proceedings of International Conference on Learning Representations, pp. 1-16 (2021)

    2021

  21. [29]

    , Zeng, G., Li, H., Yuan, Y., Wang, J.: Conditional DETR for fast training convergence

    Meng, D., Chen, X., Fan, Z. , Zeng, G., Li, H., Yuan, Y., Wang, J.: Conditional DETR for fast training convergence. Proceedings of IEEE International Conference o n Computer Vi- sion and Pattern Recognition, pp. 3651-3660 (2021)

    2021

  22. [30]

    Hugging Face (2022)

    Li, Y., Mao, H., Girshick, R., He, K.: Exploring plain vision transformer backbones for ob- ject detection. Hugging Face (2022)

    2022

  23. [31]

    Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, 5693, pp

    Zhang, F., Zhu, X., Dai, H., Ye, M., Zhu, C.: Distribution-aware coordinate representation for human pose estimation. Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, 5693, pp. 1-9 (2020)

    2020

  24. [32]

    : QueryPose: Sparse multi - person pose regression via spatial-aware part -level query

    Xiao, Y., Su, K., Wang, X., Yu, D., Jin, L., He, M., Yuan, Z. : QueryPose: Sparse multi - person pose regression via spatial-aware part -level query. Proceedings of Advances in Neural Information Processing Systems, pp. 1-14 (2022)

    2022

  25. [33]

    Proceedings of International Conference on Learning Representations (2023)

    Yang, J., Zeng, A., Liu, S., Li, F., Zhang, R., Zhang, L.: Explicit box detection unifies end- to-end multi-person pose estimation. Proceedings of International Conference on Learning Representations (2023)

    2023

  26. [34]

    Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp

    Huang, J., Zhu, Z., Guo, F., Huang, G.: The devil is in the details: Delving into unbiased data processing for human pose estimation. Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp. 5700–5709 (2020)

    2020

  27. [35]

    Proceedings of Advances in Neural Information Processing Sys- tems, pp

    Xu, Y., Zhang, J., Zhang, Q., Tao, D.: ViTPose: Simple vision transformer basel ines for human pose estimation. Proceedings of Advances in Neural Information Processing Sys- tems, pp. 38571- 38584 (2022)

    2022

  28. [36]

    Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp

    Fan, H., Xiong, B., Mangalam, K., Li, Y., Yan, Z., Malik, J., Feichtenhofer, C.: Multiscale vision transformers. Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pp. 6824-6835 (2021)

    2021

  29. [37]

    , Liu, W., Qian, C., Luo, P., Liu, S.: PoseTrans: A simple yet effective pose transformation augmenta tion for human pose estimation

    Jiang, W., Jin, S. , Liu, W., Qian, C., Luo, P., Liu, S.: PoseTrans: A simple yet effective pose transformation augmenta tion for human pose estimation. Proceedings of European Conference on Computer Vision, 13665, pp. 643-659 (2022)

    2022

  30. [38]

    -S., Lu, C

    Li, J., Wang, C., Z hu, H., Mao , Y., Fang, H. -S., Lu, C. : CrowdPose: Efficient crowded scenes pose estimation and a new benchmark. Proceedings of IEEE International Confer- ence on Computer Vision and Pattern Recognition, pp. 10855-10864 (2019)

    2019

  31. [39]

    Proceed- ings of IEEE Conference on Computer Vision and Pattern Recognition , pp

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. Proceed- ings of IEEE Conference on Computer Vision and Pattern Recognition , pp. 770 -778 (2016)

    2016

  32. [40]

    Proceedings of IEEE Confer- ence on Computer Vision and Pattern Recognition, pp

    He, K., Gkioxari, G., Dollar, P., Girshick, R.: Mask R-CNN. Proceedings of IEEE Confer- ence on Computer Vision and Pattern Recognition, pp. 2961–2969 (2017)

    2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.