Pith. sign in

REVIEW 3 major objections 5 minor 67 references

StarPose: 3D Human Pose Estimation via Spatial-Temporal Autoregressive Diffusion

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Autoregressive diffusion with physical guidance sets new 3D pose benchmarks

desk verdict Novel autoregressive diffusion idea and strong results, but the training and inference parameterizations of Dθ are irreconcilable as written. read the letter →

arxiv 2508.02056 v2 pith:6J42DTST submitted 2025-08-04 cs.CV

Haoxin Yang , Weihong Chen , Xuemiao Xu , Cheng Xu , Peng Xiao , Cuifeng Sun , Shaoyu Huang , Shengfeng He This is my paper
classification cs.CV
keywords 3DhumanposeestimationautoregressivediffusiontemporalconsistencyphysicalguidancehistoricalintegrationHuman3.6MMPI-INF-3DHPspatial-temporalmodeling
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

StarPose claims that 3D human pose estimation from monocular 2D keypoints becomes more accurate and temporally consistent when the 2D-to-3D lifting is formulated as an autoregressive diffusion process: the denoising network receives, for each frame, the 2D pose context plus historical information built from previously predicted 3D poses, and its sampling steps are steered by spatial-temporal physical constraints (2D reprojection consistency, skeletal symmetry, bone-length stability, and differential motion smoothness). On the two standard benchmarks, the reported gains are an average MPJPE of 29.9 mm and P-MPJPE of 24.6 mm on Human3.6M with CPN 2D inputs, and 20.8 mm MPJPE, 98.9% PCK, and 82.5% AUC on MPI-INF-3DHP. The result matters because depth ambiguity and occlusion make single-frame lifting ill-posed, and most existing diffusion or transformer lifters treat each frame independently, ignoring both the temporal continuity of human motion and skeletal plausibility.

What carries the argument

The central object is the autoregressive reverse diffusion process of AutoPCD, defined by $h_{k-1}=D_\theta(h_k, f_{2D}, f_{his}, k)$, where $f_{his}$ is built by HPIM from the past L predicted 3D poses and their 2D counterparts via a skeleton integration graph with GCN (short-distance) plus attention (long-distance) branches. The second machinery is the energy guidance of STPG: at each reverse step, a clean-pose estimate $h_{0|k}$ is formed from Eq. (8), four physical loss terms are evaluated on it, and the gradient $\nabla_{h_k}E(c,h_{0|k})$ is subtracted from the updated sample (Algorithm 1). This guidance is what enforces anatomical and kinematic plausibility during both training and inference.

What would settle it

Train the denoising network exactly as described (Eq. 6) with a separate noise-prediction head for Eq. 8, or evaluate the clean-pose estimates $h_{0|k}$ produced by the shared network during training: if the reported MPJPE (29.9 mm on Human3.6M with CPN inputs) cannot be reproduced, or if the clean-pose estimates are so poor that the physical guidance terms are dominated by noise, then the inference procedure does not match the training objective as written.

Watch

Extended reading notes

Core claim

The paper's central claim is that an autoregressive diffusion framework, which conditions every denoising step on the pose predicted for the preceding frame together with the 2D pose history, resolves depth ambiguity better than per-frame diffusion and deterministic transformers. The discovery has two parts. First, a Historical Pose Integration Module builds a skeleton graph over the past L 2D and 3D poses and combines GCN and attention outputs to produce a historical embedding $f_{his}$ that is concatenated with 2D context features as the denoising condition. Second, a Spatial-Temporal Physical Guidance (STPG) term is added both to training and to the inference-time reverse process as an energy gradient; it penalizes deviations from the 2D reprojection ray, left-right bone length asymmetry, bone length variance across a sliding window, and weighted joint velocity differences between adjacent frames. Together these components drive the reported improvements: 29.9 mm MPJPE / 24.6 mm P-MPJPE on Human3.6M (CPN inputs), 20.8 mm MPJPE on MPI-INF-3DHP, and 1.3 mm/s MPJVE / 1.6 mm/s² ACC-ERR.

Load-bearing premise

The denoising network is assumed to act simultaneously as the predictor of the previous noisy sample $h_{k-1}$ (which is what its training loss optimizes) and as the noise predictor $\epsilon$ in the clean-pose estimator $h_{0|k}$ used for energy guidance and sampling; in standard diffusion these are different roles, and the paper provides no separate head or reparameterization to reconcile them.

Editorial extensions

If this is right

  • On Human3.6M with CPN 2D inputs, StarPose reports 29.9 mm MPJPE and 24.6 mm P-MPJPE, surpassing the previous diffusion SOTA (FinePose) by 2.0 mm; with ground-truth 2D inputs it reports 15.5 mm MPJPE.
  • On MPI-INF-3DHP it reports 20.8 mm MPJPE, 98.9% PCK, and 82.5% AUC, exceeding FinePose by 2.3 mm MPJPE.
  • Temporal smoothness improves sharply: 1.3 mm/s MPJVE and 1.6 mm/s² ACC-ERR, versus 3.9 mm/s and 5.3 mm/s² for the next-best diffusion method reported (KTPFormer).
  • STPG is training-free and plug-and-play: attaching it to DiffPose and D3DP lowers their Human3.6M MPJPE by 8.1% and 10.0% respectively, with FPS dropping less than 6%.
  • Ablations in the paper attribute 6.1 mm of the total 9.6 mm gain to HPIM and 3.2 mm to STPG (training plus inference), indicating the two components are complementary.

Reading between the lines

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

  • The strong MPJVE/ACC-ERR numbers come largely from the differential sequence variation term, which smooths joint velocities by construction; a testable extension is to check how much of the accuracy gain would survive if the same term were added to a deterministic (non-diffusion) lifter.
  • Because the model feeds its own predictions back in, it is vulnerable to error accumulation; the paper's failure-case figure (severe occlusion) shows this, and an explicit test would be to measure how performance degrades as inference proceeds over long unedited video.
  • The reliance on camera intrinsics for reprojection consistency restricts deployment; the paper's learned-intrinsics variant ('Pred. Intr.') quantifies the drop, and a further extension would be to replace the reprojection term with a learned scale-invariant depth prior to remove the camera requirement entirely.
  • The shared-network assumption for $D_\theta$ (predicting $h_{k-1}$ and serving as $\epsilon$ in Eq. 8) is not standard; a simple ablation adding a separate noise head would show whether the reported gains depend on that conflation or are robust to it.
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 / 5 minor

Summary. StarPose proposes an autoregressive diffusion framework for monocular 3D human pose estimation. It conditions the denoising process on a Historical Pose Integration Module (HPIM) that fuses past 2D keypoints with previously predicted 3D poses, and it introduces a Spatial-Temporal Physical Guidance (STPG) energy function with four loss terms: 2D reprojection consistency, skeletal symmetry, bone length variance, and differential sequence variation. The method is evaluated on Human3.6M and MPI-INF-3DHP, reporting state-of-the-art MPJPE of 29.9 mm and 20.8 mm, respectively, together with very low MPJVE and ACC-ERR values. The paper claims that the autoregressive formulation and physical guidance jointly produce both higher accuracy and better temporal consistency than prior diffusion- and transformer-based methods.

Significance. If the method were implementable as described, the reported accuracy and temporal-smoothness numbers would constitute a substantial advance over recent diffusion-based pose estimators, and the proposed plug-and-play STPG module could be a useful addition to the literature. The paper contains extensive experiments, ablations, sensitivity analyses, and an efficiency comparison, and it provides a public code link. However, the central technical claim is undercut by a fundamental mismatch between the training objective and the inference update: the denoising network is trained to output the previous noisy pose h_{k-1} but is then used as the noise predictor epsilon at inference. Because this mismatch is in the core derivation, the reported results cannot currently be attributed to the described method.

major comments (3)
  1. [III-B/III-C, Eqs. (6)/(8), Algorithm 1] The central derivation is internally inconsistent. Eq. (6) defines the training loss as Ldiff = sum_k ||Dtheta(hk, f2D, fhis, k) - h_{k-1}||^2, so Dtheta is trained to predict the previous noisy sample. However, Eq. (8) computes h0|k = mu + alpha_k^{-1/2}(hk - mu - (1-alpha_k)^{1/2}Dtheta(...)), which is the standard DDPM clean-pose estimator only if Dtheta is the noise epsilon. Algorithm 1 (lines 3-4) then reuses the same Dtheta as epsilon in the DDIM-style update h_{k-1} = alpha_{k-1}^{1/2} h0|k + (1-alpha_{k-1})^{1/2}Dtheta(...). These two parameterizations are not equivalent: with the forward process in Eq. (1), a network that outputs h_{k-1} is implicitly predicting a linear combination of hk and epsilon, not epsilon alone. The paper provides no separate head, no reparameterization, and no derivation connecting the two roles. Consequently, the inference procedure does not follow from the training objective, and the reported 29.9 mm MPJPE and 1.3 mm/s MPJVE cannot be attributed to the described method.
  2. [III-C, Eq. (15), Table III] The temporal-consistency evidence is partly circular. The loss Ld in Eq. (15) directly penalizes squared inter-frame joint differences (h^t_{0|k} - h^{t-1}_{0|k})^2, and during inference this term is included in the energy function E (Eq. 16) whose gradient is used to update h_{k-1} (Algorithm 1 lines 5-7). The reported metrics MPJVE and ACC-ERR (Table III) measure exactly such inter-frame velocity and acceleration differences. The large reported gains (2.6 mm/s MPJVE and 3.7 mm/s^2 ACC-ERR) therefore reflect in part the explicit smoothing objective rather than a learned temporal model. The paper should include an ablation without Ld, and without STPG inference, to separate regularizer-driven smoothness from the autoregressive contribution.
  3. [IV-F, Tables VI and VII] The ablation baseline appears weaker than the published method it is compared with. In Table VI, the authors' reproduction of DiffPose is 39.5 mm, whereas the original DiffPose paper reports 36.9 mm (as listed in Table II). Table VII then uses a 39.5 mm baseline to attribute a 9.6 mm total improvement to STPG and HPIM. If the baseline is intended to be DiffPose, this is not a faithful reproduction, and the component gains are inflated; if it is a different architecture, it should be described. Please clarify the baseline configuration and report the components on a baseline matching the published DiffPose setup.
minor comments (5)
  1. [Algorithm 1] Algorithm 1 calls the denoising model with an undefined condition f_ST (lines 3-4), while the text and Eq. (8) use fhis. Please replace f_ST with fhis or define f_ST explicitly.
  2. [Algorithm 1 / Section IV-B] The step size rho_k, which is essential to the STPG inference update, is never specified in the implementation details or experiments. A concrete value or schedule should be reported for reproducibility.
  3. [III-C, Eq. (7)] Eq. (7) uses the same symbol h_{k-1} on both sides; please use a temporary variable such as tilde{h}_{k-1} for the uncorrected update before the energy-gradient correction.
  4. [I, Introduction] The acronym 'SPTG' appears in the Introduction ('SPTG incorporates structural and kinematic constraints'); it should be 'STPG'.
  5. [Tables IV/V] In Table V, StarPose uses an input length of f=81, whereas several compared methods use f=243; the influence of input sequence length on the comparison should be discussed, since longer contexts generally aid accuracy.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the benchmark numbers are evaluations of a trained model, and the training/inference mismatch is a correctness gap, not a circular reduction.

full rationale

I walked the derivation chain from the training loss (Eq. 6) through the inference updates (Eq. 8 and Algorithm 1) and the temporal-consistency evidence (Table III). No step reduces a claimed prediction to its own input by construction. The closest candidate is the treatment of Dθ: Eq. 6 trains Dθ to predict the previous noisy pose h_{k-1}, while Eq. 8 and Algorithm 1 use the same Dθ as the noise predictor ε. This is a genuine internal inconsistency that would make the described inference not follow from the stated training objective, but it is a correctness or implementability gap, not a circularity: the reported MPJPE and MPJVE numbers are not entailed by a fitted parameter renamed as a prediction. Similarly, the temporal-smoothness loss Ld (Eq. 15) penalizes squared differences between adjacent predicted poses, and the paper then reports MPJVE and ACC-ERR, which measure velocity and acceleration errors against ground truth. These are related but not identical: a model that simply froze all frames would drive Ld down but would score poorly on MPJVE on the moving actions in Human3.6M (walking, directions, etc.), so the 1.3 mm/s MPJVE is not statistically forced by Ld. The GMM prior is fitted on the training set following DiffPose, which is a standard initialization choice, not a circular reuse of the test metric. The only self-citations in the reference list (e.g., [24], [65]) appear in non-load-bearing contexts such as related-work enumeration and an Adam optimizer citation. The accuracy claim is therefore self-contained with respect to the benchmark protocol, and the temporal claim, while partially shaped by the loss, is not a case of a fitted surrogate being relabeled as an independent finding. Overall, I find no significant circularity in the paper's claimed derivation chain.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a fitted GMM prior, hand-tuned loss weights, and a chosen history length; these are standard practice in diffusion-based HPE but are nonetheless free parameters. The method introduces no new physical entities.

free parameters (6)
  • GMM parameters (pi, mu, Sigma) for noisy pose distribution H_K = M=5, fitted via EM on training set
    Used to initialize and add noise in the diffusion process (Section III-A, Eq. 1). Follows DiffPose.
  • Noise schedule alpha_1:K = K=50, beta linearly interpolated 1e-4 to 2e-3
    Controls the forward noising and reverse denoising; chosen by hyperparameter search (Section IV-B).
  • STPG loss weights lambda_p, lambda_s, lambda_b, lambda_d = 1, 1, 0.01, 0.01
    Weights for the four physical guidance terms in Eq. (16)-(17); tuned empirically (Section IV-B).
  • Number of pose hypotheses N, diffusion steps K, GMM components M = 5, 50, 5
    Selected via ablation studies shown in Fig. 7.
  • History length L = 27
    Past pose window for HPIM and bone length variance; set in Section IV-B.
  • Joint weights w_j in differential sequence variation = predefined from Hossain and Little [58]
    Per-joint weights reflecting typical motion magnitude; taken from prior work and not re-learned.
assumptions (4)
  • standard math Standard DDPM/DDIM theory relating forward noise process and reverse denoising updates.
    The diffusion formulation (Eq. 1) relies on the framework of Ho et al. [28]; the paper does not re-derive it.
  • domain assumption The 2D/3D pose distribution can be approximated by a GMM with M=5 components.
    The noisy pose distribution H_K is assumed to be a GMM fit to training data (Section III-A), following DiffPose; no justification of the number of components beyond ablation.
  • domain assumption Camera intrinsics are known for the 2D reprojection consistency loss.
    L_p uses the reprojection function R(.) with known intrinsic parameters (Section III-C). On Human3.6M ground-truth intrinsics are used, while in-the-wild cases rely on a learned predictor [66].
  • domain assumption Pre-trained 2D pose detector (CPN) and context encoder (MixSTE) provide sufficiently reliable inputs.
    The diffusion condition f_2D comes from a frozen pre-trained MixSTE [34]; the paper does not analyze sensitivity to upstream 2D errors beyond a limitation note.

how reviews work

0 comments
Cite this review

Pith. "Pith review of StarPose: 3D Human Pose Estimation via Spatial-Temporal Autoregressive Diffusion." pith.science (2026). https://pith.science/paper/6J42DTST

@misc{pith2026250802056,
  author       = {Pith},
  title        = {Pith review of: StarPose: 3D Human Pose Estimation via Spatial-Temporal Autoregressive Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6J42DTST}},
  note         = {Machine review of arXiv:2508.02056}
}
read the original abstract

Monocular 3D human pose estimation remains a challenging task due to inherent depth ambiguities and occlusions. Compared to traditional methods based on Transformers or Convolutional Neural Networks (CNNs), recent diffusion-based approaches have shown superior performance, leveraging their probabilistic nature and high-fidelity generation capabilities. However, these methods often fail to account for the spatial and temporal correlations across predicted frames, resulting in limited temporal consistency and inferior accuracy in predicted 3D pose sequences. To address these shortcomings, this paper proposes StarPose, an autoregressive diffusion framework that effectively incorporates historical 3D pose predictions and spatial-temporal physical guidance to significantly enhance both the accuracy and temporal coherence of pose predictions. Unlike existing approaches, StarPose models the 2D-to-3D pose mapping as an autoregressive diffusion process. By synergically integrating previously predicted 3D poses with 2D pose inputs via a Historical Pose Integration Module (HPIM), the framework generates rich and informative historical pose embeddings that guide subsequent denoising steps, ensuring temporally consistent predictions. In addition, a fully plug-and-play Spatial-Temporal Physical Guidance (STPG) mechanism is tailored to refine the denoising process in an iterative manner, which further enforces spatial anatomical plausibility and temporal motion dynamics, rendering robust and realistic pose estimates. Extensive experiments on benchmark datasets demonstrate that StarPose outperforms state-of-the-art methods, achieving superior accuracy and temporal consistency in 3D human pose estimation. Code is available at https://github.com/wileychan/StarPose.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 61 canonical work pages

  1. [1]

    On the benefits of 3d pose and tracking for human action recognition,

    J. Rajasegaran, G. Pavlakos, A. Kanazawa, C. Feichtenhofer, and J. Malik, “On the benefits of 3d pose and tracking for human action recognition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2023, pp. 640–649

    work page 2023

  2. [2]

    Mutual information driven equivariant contrastive learning for 3d action representation learning,

    L. Lin, J. Zhang, and J. Liu, “Mutual information driven equivariant contrastive learning for 3d action representation learning,” IEEE Trans- actions on Image Processing (TIP) , vol. 33, pp. 1883–1897, 2024

    work page 2024

  3. [3]

    Masked motion predictors are strong 3d action representation learners,

    Y . Mao, J. Deng, W. Zhou, Y . Fang, W. Ouyang, and H. Li, “Masked motion predictors are strong 3d action representation learners,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 10 181–10 191

    work page 2023

  4. [4]

    Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving,

    X. Tian, T. Jiang, L. Yun, Y . Mao, H. Yang, Y . Wang, Y . Wang, and H. Zhao, “Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023, pp. 64 318–64 330

    work page 2023

  5. [5]

    Hum3dil: Semi-supervised multi-modal 3d human- pose estimation for autonomous driving,

    A. Zanfir, M. Zanfir, A. Gorban, J. Ji, Y . Zhou, D. Anguelov, and C. Sminchisescu, “Hum3dil: Semi-supervised multi-modal 3d human- pose estimation for autonomous driving,” in Proceedings of the Confer- ence on Robot Learning (CoRL) , vol. 205, 2023, pp. 1114–1124

    work page 2023

  6. [6]

    Multi-modal 3d human pose estimation with 2d weak supervision in autonomous driving,

    J. Zheng, X. Shi, A. Gorban, J. Mao, Y . Song, C. R. Qi, T. Liu, V . Chari, A. Cornman, Y . Zhou, C. Li, and D. Anguelov, “Multi-modal 3d human pose estimation with 2d weak supervision in autonomous driving,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 4477–4486

    work page 2022

  7. [7]

    Sinc: Spatial composition of 3d human motions for simultaneous action generation,

    N. Athanasiou, M. Petrovich, M. J. Black, and G. Varol, “Sinc: Spatial composition of 3d human motions for simultaneous action generation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 9984–9995

    work page 2023

  8. [8]

    Toward realistic 3d human motion prediction with a spatio- temporal cross- transformer approach,

    H. Yu, X. Fan, Y . Hou, W. Pei, H. Ge, X. Yang, D. Zhou, Q. Zhang, and M. Zhang, “Toward realistic 3d human motion prediction with a spatio- temporal cross- transformer approach,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) , vol. 33, no. 10, pp. 5707– 5720, 2023

    work page 2023

Show all 67 references
  1. [9]

    Multi-person pose forecasting with individual interaction perceptron and prior learning,

    P. Xiao, Y . Xie, X. Xu, W. Chen, and H. Zhang, “Multi-person pose forecasting with individual interaction perceptron and prior learning,” in Proceedings of the European Conference on Computer Vision (ECCV) , 2024, pp. 402–419

    2024

  2. [10]

    Cascaded pyramid network for multi-person pose estimation,

    Y . Chen, Z. Wang, Y . Peng, Z. Zhang, G. Yu, and J. Sun, “Cascaded pyramid network for multi-person pose estimation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 7103–7112

    2018

  3. [11]

    Finepose: Fine-grained prompt-driven 3d human pose estimation via diffusion models,

    J. Xu, Y . Guo, and Y . Peng, “Finepose: Fine-grained prompt-driven 3d human pose estimation via diffusion models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 561–570

    2024

  4. [12]

    Ktpformer: Kinematics and trajectory prior knowledge-enhanced transformer for 3d human pose estimation,

    J. Peng, Y . Zhou, and P. Y . Mok, “Ktpformer: Kinematics and trajectory prior knowledge-enhanced transformer for 3d human pose estimation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 1123–1132

    2024

  5. [13]

    Repose: 3d human pose estimation via spatio-temporal depth relational consistency,

    Z. Sun, Y . Liang, Z. Ma, T. Zhang, L. Bao, G. Li, and S. He, “Repose: 3d human pose estimation via spatio-temporal depth relational consistency,” in Proceedings of the European Conference on Computer Vision (ECCV), 2024, pp. 309–325

    2024

  6. [14]

    Diff- pose: Toward more reliable 3d pose estimation,

    J. Gong, L. G. Foo, Z. Fan, Q. Ke, H. Rahmani, and J. Liu, “Diff- pose: Toward more reliable 3d pose estimation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 13 041–13 051

    2023

  7. [15]

    Diffusion-based 3d human pose estimation with multi- hypothesis aggregation,

    W. Shan, Z. Liu, X. Zhang, Z. Wang, K. Han, S. Wang, S. Ma, and W. Gao, “Diffusion-based 3d human pose estimation with multi- hypothesis aggregation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , 2023, pp. 14 761–14 771

    2023

  8. [16]

    3d human pose estimation with spatio-temporal criss-cross attention,

    Z. Tang, Z. Qiu, Y . Hao, R. Hong, and T. Yao, “3d human pose estimation with spatio-temporal criss-cross attention,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 4790–4799

    2023

  9. [17]

    Learning to augment poses for 3d human pose estimation in images and videos,

    J. Zhang, K. Gong, X. Wang, and J. Feng, “Learning to augment poses for 3d human pose estimation in images and videos,” IEEE transactions on pattern analysis and machine intelligence , vol. 45, no. 8, pp. 10 012– 10 026, 2023

    2023

  10. [18]

    Anatpose: Bidirectionally learning anatomy-aware heatmaps for human pose estimation,

    S. Du, Z. Zhang, and T. Ikenaga, “Anatpose: Bidirectionally learning anatomy-aware heatmaps for human pose estimation,” Pattern Recogni- tion, vol. 155, p. 110654, 2024

    2024

  11. [19]

    Bi-pose: Bidirectional 2d-3d transformation for human pose estimation from a monocular camera,

    S. Du, H. Wang, Z. Yuan, and T. Ikenaga, “Bi-pose: Bidirectional 2d-3d transformation for human pose estimation from a monocular camera,” IEEE Transactions on Automation Science and Engineering , 2023

    2023

  12. [20]

    Weakly-supervised pre-training for 3d human pose estimation via perspective knowledge,

    Z. Qiu, K. Qiu, J. Fu, and D. Fu, “Weakly-supervised pre-training for 3d human pose estimation via perspective knowledge,” Pattern Recognition, vol. 139, p. 109497, 2023

    2023

  13. [21]

    Joypose: Jointly learning evolutionary data augmentation and anatomy-aware global–local repre- sentation for 3d human pose estimation,

    S. Du, Z. Yuan, P. Lai, and T. Ikenaga, “Joypose: Jointly learning evolutionary data augmentation and anatomy-aware global–local repre- sentation for 3d human pose estimation,” Pattern Recognition, vol. 147, p. 110116, 2024

    2024

  14. [22]

    Motionagformer: Enhancing 3d human pose estimation with a transformer-gcnformer network,

    S. Mehraban, V . Adeli, and B. Taati, “Motionagformer: Enhancing 3d human pose estimation with a transformer-gcnformer network,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024, pp. 6920–6930

    2024

  15. [23]

    3d human pose estimation with spatial and temporal transformers,

    C. Zheng, S. Zhu, M. Mendieta, T. Yang, C. Chen, and Z. Ding, “3d human pose estimation with spatial and temporal transformers,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 11 656–11 665

    2021

  16. [24]

    Scjd: Sparse correlation and joint distillation for efficient 3d human pose estimation,

    W. Chen, X. Xu, H. Yang, Y . Xie, P. Xiao, C. Xu, H. Zhang, and P.- A. Heng, “Scjd: Sparse correlation and joint distillation for efficient 3d human pose estimation,” arXiv preprint arXiv:2503.14097 , 2025

    2025 arXiv

  17. [25]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of the International Conference on Neural Information Processing Systems, 2017, p. 6000–6010

    2017

  18. [26]

    Diffusion-based hypotheses generation and joint-level hypotheses ag- gregation for 3d human pose estimation,

    W. Shan, Y . Zhang, X. Zhang, S. Wang, X. Zhou, S. Ma, and W. Gao, “Diffusion-based hypotheses generation and joint-level hypotheses ag- gregation for 3d human pose estimation,” IEEE Transactions on Circuits and Systems for Video Technology , 2024

    2024

  19. [27]

    Diffusion model with temporal constraint for 3d human pose estimation,

    Z. Chen, J. Dai, J. Pan, and F. Zhou, “Diffusion model with temporal constraint for 3d human pose estimation,” The Visual Computer , pp. 1–17, 2024

    2024

  20. [28]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Advances in Neural Information Processing Systems (NeurIPS) , vol. 33, 2020, pp. 6840–6851

    2020

  21. [29]

    Human3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments,

    C. Ionescu, D. Papava, V . Olaru, and C. Sminchisescu, “Human3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), vol. 36, no. 7, pp. 1325–1339, 2013

    2013

  22. [30]

    Monocular 3d human pose estimation in the wild using IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 14 improved cnn supervision,

    D. Mehta, H. Rhodin, D. Casas, P. Fua, O. Sotnychenko, W. Xu, and C. Theobalt, “Monocular 3d human pose estimation in the wild using IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 14 improved cnn supervision,” in International Conference on 3D Vision (3DV), 201...

    2017

  23. [31]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” in Proceedings of the International Con...

    2021

  24. [32]

    Hdformer: high-order directed transformer for 3d human pose estimation,

    H. Chen, J.-Y . He, W. Xiang, Z.-Q. Cheng, W. Liu, H. Liu, B. Luo, Y . Geng, and X. Xie, “Hdformer: high-order directed transformer for 3d human pose estimation,” in Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI) , 2023

    2023

  25. [33]

    Ftcm: Frequency-temporal collab- orative module for efficient 3d human pose estimation in video,

    Z. Tang, Y . Hao, J. Li, and R. Hong, “Ftcm: Frequency-temporal collab- orative module for efficient 3d human pose estimation in video,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) , vol. 34, no. 2, pp. 911–923, 2024

    2024

  26. [34]

    Mixste: Seq2seq mixed spatio-temporal encoder for 3d human pose estimation in video,

    J. Zhang, Z. Tu, J. Yang, Y . Chen, and J. Yuan, “Mixste: Seq2seq mixed spatio-temporal encoder for 3d human pose estimation in video,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 13 232–13 242

    2022

  27. [35]

    P-stmo: Pre-trained spatial temporal many-to-one model for 3d human pose estimation,

    W. Shan, Z. Liu, X. Zhang, S. Wang, S. Ma, and W. Gao, “P-stmo: Pre-trained spatial temporal many-to-one model for 3d human pose estimation,” in Proceedings of the European Conference on Computer Vision (ECCV), 2022, pp. 461–478

    2022

  28. [36]

    Mhformer: Multi- hypothesis transformer for 3d human pose estimation,

    W. Li, H. Liu, H. Tang, P. Wang, and L. Van Gool, “Mhformer: Multi- hypothesis transformer for 3d human pose estimation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 13 147–13 156

    2022

  29. [37]

    Uncertainty-aware 3d human pose esti- mation from monocular video,

    J. Zhang, Y . Chen, and Z. Tu, “Uncertainty-aware 3d human pose esti- mation from monocular video,” in Proceedings of the ACM International Conference on Multimedia (MM) , 2022, pp. 5102–5113

    2022

  30. [38]

    Exploiting temporal contexts with strided transformer for 3d human pose estima- tion,

    W. Li, H. Liu, R. Ding, M. Liu, P. Wang, and W. Yang, “Exploiting temporal contexts with strided transformer for 3d human pose estima- tion,” IEEE Transactions on Multimedia (TMM), vol. 25, pp. 1282–1293, 2022

    2022

  31. [39]

    Improving robustness and accuracy via relative information encoding in 3d human pose estimation,

    W. Shan, H. Lu, S. Wang, X. Zhang, and W. Gao, “Improving robustness and accuracy via relative information encoding in 3d human pose estimation,” in Proceedings of the ACM International Conference on Multimedia (MM), 2021, pp. 3446–3454

    2021

  32. [40]

    Anatomy-aware 3d human pose estimation with bone-based pose decomposition,

    T. Chen, C. Fang, X. Shen, Y . Zhu, Z. Chen, and J. Luo, “Anatomy-aware 3d human pose estimation with bone-based pose decomposition,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) , vol. 32, no. 1, pp. 198–209, 2021

    2021

  33. [41]

    Diffpose: Spatiotemporal diffusion model for video-based human pose estimation,

    R. Feng, Y . Gao, T. H. E. Tse, X. Ma, and H. J. Chang, “Diffpose: Spatiotemporal diffusion model for video-based human pose estimation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 14 861–14 872

    2023

  34. [42]

    Diffpose: Multi-hypothesis human pose estimation using diffusion models,

    K. Holmquist and B. Wandt, “Diffpose: Multi-hypothesis human pose estimation using diffusion models,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 15 977– 15 987

    2023

  35. [43]

    Diffupose: Monocular 3d human pose estimation via denoising diffusion probabilistic model,

    J. Choi, D. Shim, and H. J. Kim, “Diffupose: Monocular 3d human pose estimation via denoising diffusion probabilistic model,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2023, pp. 3773–3780

    2023

  36. [44]

    Diff3dhpe: A diffusion model for 3d human pose estimation,

    J. Zhou, T. Zhang, Z. Hayder, L. Petersson, and M. Harandi, “Diff3dhpe: A diffusion model for 3d human pose estimation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2023, pp. 2092–2102

    2023

  37. [45]

    Dis- endreamer: Subject-driven text-to-image generation with sample-aware disentangled tuning,

    H. Chen, Y . Zhang, X. Wang, X. Duan, Y . Zhou, and W. Zhu, “Dis- endreamer: Subject-driven text-to-image generation with sample-aware disentangled tuning,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), vol. 34, no. 8, pp. 6860–6873, 2024

    2024

  38. [46]

    Image intrinsic components guided conditional diffusion model for low-light image enhancement,

    S. Kang, S. Gao, W. Wu, X. Wang, S. Wang, and G. Qiu, “Image intrinsic components guided conditional diffusion model for low-light image enhancement,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), vol. 34, no. 12, pp. 13 244–13 256, 2024

    2024

  39. [47]

    Deep texture-aware features for camouflaged object detection,

    J. Ren, X. Hu, L. Zhu, X. Xu, Y . Xu, W. Wang, Z. Deng, and P.-A. Heng, “Deep texture-aware features for camouflaged object detection,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) , vol. 33, no. 3, pp. 1157–1167, 2023

    2023

  40. [48]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction,

    K. Tian, Y . Jiang, Z. Yuan, B. Peng, and L. Wang, “Visual autoregressive modeling: Scalable image generation via next-scale prediction,” arXiv preprint arXiv:2404.02905, 2024

    2024 arXiv

  41. [49]

    Autoregressive model beats diffusion: Llama for scalable image gener- ation,

    P. Sun, Y . Jiang, S. Chen, S. Zhang, B. Peng, P. Luo, and Z. Yuan, “Autoregressive model beats diffusion: Llama for scalable image gener- ation,” arXiv preprint arXiv:2406.06525 , 2024

    2024 arXiv

  42. [50]

    Progressive autoregressive video diffusion models,

    D. Xie, Z. Xu, Y . Hong, H. Tan, D. Liu, F. Liu, A. Kaufman, and Y . Zhou, “Progressive autoregressive video diffusion models,” arXiv preprint arXiv:2410.08151, 2024

    2024 arXiv

  43. [51]

    Vid-gpt: Introducing gpt-style autoregressive generation in video diffusion models,

    K. Gao, J. Shi, H. Zhang, C. Wang, and J. Xiao, “Vid-gpt: Introducing gpt-style autoregressive generation in video diffusion models,” arXiv preprint arXiv:2406.10981, 2024

    2024 arXiv

  44. [52]

    Hier- archical patch diffusion models for high-resolution video generation,

    I. Skorokhodov, W. Menapace, A. Siarohin, and S. Tulyakov, “Hier- archical patch diffusion models for high-resolution video generation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 7569–7579

    2024

  45. [53]

    A tutorial on energy-based learning,

    Y . Lecun, S. Chopra, R. Hadsell, M. Ranzato, and F. J. Huang, “A tutorial on energy-based learning,” in Predicting Structured Data, 2006

    2006

  46. [54]

    Freedom: Training- free energy-guided conditional diffusion model,

    J. Yu, Y . Wang, C. Zhao, B. Ghanem, and J. Zhang, “Freedom: Training- free energy-guided conditional diffusion model,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , 2023, pp. 23 174–23 184

    2023

  47. [55]

    Egsde: Unpaired image-to- image translation via energy-guided stochastic differential equations,

    M. Zhao, F. Bao, C. LI, and J. Zhu, “Egsde: Unpaired image-to- image translation via energy-guided stochastic differential equations,” in Advances in Neural Information Processing Systems (NeurIPS) , vol. 35, 2022, pp. 3609–3623

    2022

  48. [56]

    Motionbert: A unified perspective on learning human motion representations,

    W. Zhu, X. Ma, Z. Liu, L. Liu, W. Wu, and Y . Wang, “Motionbert: A unified perspective on learning human motion representations,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 15 085–15 099

    2023

  49. [57]

    Camera distortion-aware 3d human pose estimation in video with optimization-based meta-learning,

    H. Cho, Y . Cho, J. Yu, and J. Kim, “Camera distortion-aware 3d human pose estimation in video with optimization-based meta-learning,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 11 169–11 178

    2021

  50. [58]

    Exploiting temporal information for 3d human pose estimation,

    M. R. I. Hossain and J. J. Little, “Exploiting temporal information for 3d human pose estimation,” in Proceedings of the European Conference on Computer Vision (ECCV) , 2018, pp. 68–84

    2018

  51. [59]

    3d human pose estimation in video with temporal convolutions and semi-supervised training,

    D. Pavllo, C. Feichtenhofer, D. Grangier, and M. Auli, “3d human pose estimation in video with temporal convolutions and semi-supervised training,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2019, pp. 7753–7762

    2019

  52. [60]

    Gla- gcn: Global-local adaptive graph convolutional network for 3d human pose estimation from monocular video,

    B. X. Yu, Z. Zhang, Y . Liu, S.-h. Zhong, Y . Liu, and C. W. Chen, “Gla- gcn: Global-local adaptive graph convolutional network for 3d human pose estimation from monocular video,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , 2023, pp. 8818– 8829

    2023

  53. [61]

    Graformer: Graph-oriented transformer for 3d pose estimation,

    W. Zhao, W. Wang, and Y . Tian, “Graformer: Graph-oriented transformer for 3d pose estimation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2022, pp. 20 438– 20 447

    2022

  54. [62]

    Conditional di- rected graph convolution for 3d human pose estimation,

    W. Hu, C. Zhang, F. Zhan, L. Zhang, and T.-T. Wong, “Conditional di- rected graph convolution for 3d human pose estimation,” in Proceedings of the ACM International Conference on Multimedia (MM) , 2021, pp. 602–611

    2021

  55. [63]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in Proceedings of the International Conference on Learning Represen- tations (ICLR), 2021

    2021

  56. [64]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” 2017

    2017

  57. [65]

    G2face: High-fidelity reversible face anonymization via generative and geometric priors,

    H. Yang, X. Xu, C. Xu, H. Zhang, J. Qin, Y . Wang, P.-A. Heng, and S. He, “G2face: High-fidelity reversible face anonymization via generative and geometric priors,” IEEE Transactions on Information Forensics and Security, 2024

    2024

  58. [66]

    Camerapose: Weakly-supervised monocular 3d human pose estimation by leveraging in-the-wild 2d annotations,

    C.-Y . Yang, J. Luo, L. Xia, Y . Sun, N. Qiao, K. Zhang, Z. Jiang, J.-N. Hwang, and C.-H. Kuo, “Camerapose: Weakly-supervised monocular 3d human pose estimation by leveraging in-the-wild 2d annotations,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Compu...

    2023

  59. [2023]

    Xuemiao Xu received her B.S

    His research interests include computer vision, pose estimation, and deep learning. Xuemiao Xu received her B.S. and M.S. degrees in Computer Science and Engineering from South China University of Technology in 2002 and 2005 respectively, and Ph.D. degree in Computer Science a...

    2015

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.