Pith. sign in

REVIEW 4 major objections 7 minor 29 references

DeepHuMS: Deep Human Motion Signature for 3D Skeletal Sequences

T0 review · 4 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A learned motion signature beats recognition nets at 3D motion retrieval

desk verdict Plausible architecture and useful robustness tests, but the NTU top-1/top-10 numbers are internally inconsistent, so the SOTA claim is unverifiable as reported. read the letter →

arxiv 1908.05750 v3 pith:LLTCLNYZ submitted 2019-08-15 cs.CV cs.LGeess.IV

Neeraj Battan , Abbhinav Venkat , Avinash Sharma This is my paper · ORCID
classification cs.CVcs.LGeess.IV
keywords 3Dhumanmotionretrievalself-supervisedlearningskeletonsequencesSiamesenetworkcontrastivelosstrajectorydescriptornearestneighborcapture
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 argues that 3D human motion retrieval should be done with a learned embedding rather than hand-crafted descriptors, and it proposes DeepHuMS, an RNN trained in a Siamese setup. The training signal is a trajectory-based similarity: two sequences are pulled together when their joint-wise motion field and motion distance agree, and pushed apart otherwise, with optional class labels added on top. The paper reports top-1 retrieval accuracy of 0.78 and top-10 of 0.753 on NTU RGB+D, above the compared recognition embeddings, and higher precision-recall than hand-crafted retrieval features on HDM05. It also reports that the self-supervised version gives closer DTW frame-level matches than the supervised version (31 mm versus 34 mm), and that retrieval is a nearest-neighbor search taking about 18 ms on NTU RGB+D and 0.8 ms on HDM05. If correct, this would make learned, self-supervised descriptors a practical replacement for optimization- and histogram-based retrieval on large motion repositories.

What carries the argument

The central object is a Siamese pair of RNNs whose contrastive loss (Eq. 1) enforces small Euclidean distance $D_w$ for similar pairs and margin $m$ for dissimilar pairs, optionally combined with cross-entropy (Eq. 2). Similarity between two sequences is defined by the joint-wise motion field, $MF[i,j]=F[i]-F[j]$, and the joint-wise motion distance, $MD[j]=\sum_{i=1}^{N-1}\|F[i+1][j]-F[i][j]\|$: the field encodes direction and which joints matter, while the distance keeps the metric non-degenerate when the field vanishes. To handle speed variation, the paper synthesizes faster and slower versions of each sequence and trains the network to embed them together, since the full-sequence motion field and distance are the same across resampling. This trajectory metric carries the entire self-supervised training signal, and it is what distinguishes the embedding from a recognition-only objective.

What would settle it

Take two sequences of different action classes whose smoothed joint trajectories are nearly identical except for a semantic detail, for example a hand raising to the head versus to the face. If nearest-neighbor retrieval in the DeepHuMS embedding ranks these two cross-class sequences above same-class sequences, the trajectory-similarity premise is falsified. A quantitative version is to compute the proposed motion-field and motion-distance score over held-out pairs and compare it with class agreement; zero correlation would mean the self-supervised signal carries no semantic information.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a deep recurrent network can learn a discriminative 3D motion signature from trajectory cues alone, without action labels, by treating the motion field $MF[i, j] = F[i] - F[j]$ and the per-joint motion distance $MD[j] = \sum_{i=1}^{N-1} \|F[i+1][j] - F[i][j]\|$ as the notion of similarity between sequences. The learned embedding is what does the retrieval work: a query sequence is embedded once and compared with stored embeddings by nearest neighbor, which makes retrieval fast and lets the same representation handle noisy or missing joints, different execution speeds, and partial sub-motions via a second network. The reported consequence is that this trajectory-supervised embedding outperforms embeddings trained purely for recognition, whose class-boundary geometry transfers poorly to ranking.

Load-bearing premise

The load-bearing premise is that two sequences are semantically similar exactly when their joint-wise motion field and motion distance are similar; since this hand-defined metric is the only supervision in the self-supervised setup, any class whose members are not closest under this metric will be embedded wrongly.

Editorial extensions

If this is right

  • Retrieval reduces to a single nearest-neighbor pass over a 512-dimensional embedding, giving measured query times of 18 ms on NTU RGB+D and 0.8 ms on HDM05.
  • Self-supervision alone yields inter-class matches that are closer in DTW frame error (31 mm) than the supervised variant (34 mm), so label-free training is not just a fallback.
  • The descriptor inherits robustness to missing data: training with 20 percent of joints randomly removed still produces working precision-recall curves.
  • Speed variations of the same motion collapse to nearby embeddings, so sequences sampled at different frame rates can be compared directly.
  • A second RNN trained with L2 loss can map a sub-motion to the embedding of the full sequence, enabling retrieval of the longer sequence that contains the query.

Reading between the lines

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

  • The paper does not test whether its trajectory metric matches semantic similarity in general; I infer that the method's ceiling is set by that metric, so a useful stress test is to search for two different actions whose joint trajectories nearly coincide and check whether the embedding ranks them as close.
  • Because the contrastive signal uses whole-sequence motion field and distance, the descriptor likely blurs temporal order within a sequence; repetitive actions such as multiple identical cycles may map to similar embeddings even when the phase or count differs.
  • The embedding's success at matching human-judged motion similarity, if confirmed, suggests the same trajectory objective could serve as pretraining for few-shot action recognition, where labels are scarce.
  • The sub-motion network currently regresses to a full-sequence embedding with L2 loss; a direct extension would be to predict the temporal offset of the sub-motion within the sequence, turning retrieval into localization.
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

4 major / 7 minor

Summary. The paper proposes DeepHuMS, a deep RNN-based Siamese architecture that learns a discriminative 3D human motion descriptor for retrieval. The model is trained with a contrastive loss on hand-defined trajectory cues (motion field and motion distance), optionally combined with a classification loss, and is evaluated on NTU RGB+D and HDM05 for retrieval accuracy, precision-recall behavior, robustness to speed and missing joints, and sub-motion retrieval. The authors claim state-of-the-art retrieval performance over both hand-crafted retrieval methods and recognition-based embedding methods, and they provide a code link and supplementary video.

Significance. If the reported results are reproducible, DeepHuMS is a simple and potentially useful contribution to 3D motion retrieval, demonstrating that a contrastive loss based on trajectory similarity can produce an embedding that generalizes to large datasets and supports subsequence search. The self-supervised aspect is valuable given the limited availability of class labels, and the robustness experiments with speed variation and missing data address practically relevant issues. The paper also provides a public code repository link, which aids reproducibility. However, the headline quantitative claims are currently undermined by an ambiguous and internally inconsistent evaluation metric, the HDM05 precision-recall comparison lacks numeric values, and a directly relevant learned-retrieval baseline (Deep Signatures [27]) is not compared. The significance of the claimed state-of-the-art results therefore cannot be fully assessed without a corrected evaluation.

major comments (4)
  1. [§4.3, Table 1] The definition of "top-n retrieval accuracy" as "how many of the n retrieved results belong to the same class as the query motion" is ambiguous, and Table 1 reports top-1 accuracy (0.78) greater than top-10 accuracy (0.753) for DeepHuMS, with the same pattern for the baselines. Under any standard top-n retrieval metric that counts queries with a correct hit among the first n retrieved results, top-10 accuracy must be at least top-1 accuracy because the top-1 result is included in the top-10 list. If "top-10" instead denotes a different quantity such as precision@10 or the mean number of correct retrievals in the top-10, then it is not commensurate with the top-1 numbers reported for the baselines, and the comparison is invalid. The authors must define the metric precisely, recompute all numbers under that single definition, and report results that are internally consistent.
  2. [§4.4, Figure 4a] The HDM05 comparison asserts that "the area under the PR curve is far larger" for DeepHuMS than for AFMS, Feature Ranking, and UDFS, but the figure reports no numeric area or average precision values and no error bars or statistical significance tests. This makes the claimed superiority over existing 3D motion retrieval methods unverifiable. Please report numeric AP/mAP values with variance and describe how the PR curves are averaged over queries.
  3. [§4.4, Related Work] The paper cites Deep Signatures [27] as a learned auto-encoder-based method for motion retrieval but never compares against it in the experiments. Since the central claim is that DeepHuMS is a superior learned representation for retrieval, a direct comparison with [27] on HDM05 (or a concrete explanation of why such a comparison is not possible) is necessary to support that claim.
  4. [§3, Eq. (4)] The hand-defined trajectory similarity (motion distance and motion field) is the only training signal in the self-supervised setup, yet the paper provides only qualitative evidence (Figure 3) that this metric is a reliable proxy for semantic action similarity. A quantitative validation, such as k-NN purity under the raw metric or a correlation with class labels, is needed to support the claim that the self-supervised embedding is semantically meaningful.
minor comments (7)
  1. [Eq. (1)] The notation in Eq. (1) is hard to parse, and the sentence "Y is if the label value" is grammatically incomplete; please rewrite this section for clarity.
  2. [Figure 5] The t-SNE comparison in Figure 5 has no legend or color coding for classes, making it impossible to assess the claim that DeepHuMS produces a more clustered and discriminative space.
  3. [§4.5] The statement that the self-supervised setup improves per-frame DTW error from 34mm to 31mm is reported without describing the comparison protocol, the number of query sequences, or the variance across queries.
  4. [Figure 8] The captions "before training" and "after training" in Figure 8 are ambiguous; it should be clarified whether the curves show the raw similarity metric or the learned embedding.
  5. [Introduction, References] References [28] and [29] are pose-reconstruction papers that are not clearly related to the retrieval motivation; either connect them to the discussion or remove them.
  6. [Abstract] The abstract claims "State of the art results on all these fronts," which is stronger than the evidence supports given the metric inconsistencies and missing numeric PR results.
  7. [§4.5, Retrieval Time] The retrieval times of 18ms and 0.8ms are reported without specifying the search implementation (e.g., brute-force k-NN versus index-based) or the hardware, so the numbers are not reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the learned embedding is trained with a hand-defined trajectory similarity and evaluated on held-out test splits, so the retrieval claim does not reduce to its inputs.

full rationale

The derivation chain in DeepHuMS is self-contained and empirically grounded. The training signal is a contrastive loss (Eq. 1) applied to pairs deemed similar by the hand-defined motion-field and motion-distance similarity (Eqs. 3-4). This similarity is an input design choice, not a quantity fitted to retrieval performance. The retrieval evaluation is then performed with a nearest-neighbour search in the learned embedding space on test splits: NTU RGB+D uses a performer-wise split and HDM05 uses the first four performers for training and the last for testing (Sec. 4.1). Class-wise top-1/top-10 retrieval accuracy and DTW distance are external, class-based benchmarks rather than re-statements of the trajectory metric. The paper's self-citations [28,29] concern pose estimation/reconstruction and are not load-bearing for the retrieval claim. The reported top-1 (0.78) exceeding top-10 (0.753) in Table 1 is a metric-consistency or reporting concern, but under the review rules that is a correctness issue, not evidence that any prediction is equivalent to its inputs by construction. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via self-citation. Accordingly, the circularity score is 0.

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

The central claim rests on a hand-designed trajectory similarity (Eqs. 3-4) and several unstated design choices (pair sampling, speed augmentation, loss weighting). No invented entities. The model's parameters are learned from data, but the learning signal is shaped entirely by these domain assumptions.

free parameters (5)
  • Margin m in contrastive loss
    Eq. 1 uses a margin m for dissimilar pairs; value not specified in the paper, affects the boundary between similar and dissimilar embeddings.
  • Embedding dimension = 512
    Section 4.5 reports retrieval times based on a 512-dimensional embedding; no analysis of this choice is given.
  • Speed augmentation factor = 2x
    Section 4.2 creates one 2x faster and one 2x slower sequence per input; this choice is ad hoc and no sweep is reported.
  • Siamese pair-sampling distribution
    The paper does not specify how positive and negative pairs are sampled, or the ratio of augmented to real pairs; this affects what the contrastive loss optimizes.
  • Supervised loss weight
    Section 3 combines L_contrastive and L_crossentropy, but no combination weight is reported; this governs the trade-off between ranking and classification.
assumptions (4)
  • domain assumption Skeletal joint positions and their frame-to-frame differences are sufficient to characterize the semantic content of a human motion for retrieval.
    Section 3 relies on these features as the sole input to the network; if joint-level dynamics are insufficient to disambiguate actions, the embedding cannot be semantically discriminative.
  • domain assumption The proposed similarity metric (joint-wise motion distance and motion field aggregated over the full sequence) captures the perceptual similarity of motions.
    Eqs. 3-4 define the only training signal in the self-supervised setup; the paper does not benchmark this metric against human judgments or alternative metrics.
  • domain assumption Uniformly subsampling or interpolating a sequence by 2x preserves the action category, so faster or slower versions can be treated as similar motions.
    Section 4.2 adds 2x speed variants and trains the Siamese net to map them together; this assumes time-scaling does not change action identity.
  • domain assumption The RNN used (IndRNN, [19]) can learn a fixed-length embedding from variable-length skeleton sequences.
    Section 3 uses "an RNN [19] to model the 4D data"; the representational capability of this architecture is taken from prior work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepHuMS: Deep Human Motion Signature for 3D Skeletal Sequences." pith.science (2026). https://pith.science/paper/LLTCLNYZ

@misc{pith2026190805750,
  author       = {Pith},
  title        = {Pith review of: DeepHuMS: Deep Human Motion Signature for 3D Skeletal Sequences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LLTCLNYZ}},
  note         = {Machine review of arXiv:1908.05750}
}
read the original abstract

3D Human Motion Indexing and Retrieval is an interesting problem due to the rise of several data-driven applications aimed at analyzing and/or re-utilizing 3D human skeletal data, such as data-driven animation, analysis of sports bio-mechanics, human surveillance etc. Spatio-temporal articulations of humans, noisy/missing data, different speeds of the same motion etc. make it challenging and several of the existing state of the art methods use hand-craft features along with optimization based or histogram based comparison in order to perform retrieval. Further, they demonstrate it only for very small datasets and few classes. We make a case for using a learned representation that should recognize the motion as well as enforce a discriminative ranking. To that end, we propose, a 3D human motion descriptor learned using a deep network. Our learned embedding is generalizable and applicable to real-world data - addressing the aforementioned challenges and further enables sub-motion searching in its embedding space using another network. Our model exploits the inter-class similarity using trajectory cues, and performs far superior in a self-supervised setting. State of the art results on all these fronts is shown on two large scale 3D human motion datasets - NTU RGB+D and HDM05.

Figures

Figures reproduced from arXiv: 1908.05750 by the authors.

Figure 1
Figure 1. Motivation of 3D Human Motion Retrieval. Given a query 3D skeletal mo [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our model - DeepHuMS. Given two skeleton sequences (a), [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Hand and elbow are important joints while waving, due to higher motion [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: A comparison of retrieval accuracy using PR curves for (a) 3D Motion [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: A comparison of the t-SNE representation of Motion Recognition [19] [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Retrieval results for self-supervised setup, which shows that we exploit [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Retrieval results on NTU-RGB+D dataset [2] using our supervised setup. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: A comparison of our Precision Recall curves [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Retrieval Results for Noisy Data Sub Motion Retrieval. Sub Motion retrieval becomes important when we would like to search for a smaller action/motion in longer sequences. But it is a very challenging task due to the variations in length and actions in sub sequences. M…
Figure 10
Figure 10. Figure 10: Given a subsequence (a) as the input, we use another RNN (b) to learn [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Retrieval results for sub sequence retrieval. Retrieved results are from [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 28 canonical work pages

  1. [27]

    ”Deep signatures for indexing and retrieval in large motion databases.” Conference on Motion in Games, pp

    Wang, Yingying, and Michael Neff. ”Deep signatures for indexing and retrieval in large motion databases.” Conference on Motion in Games, pp. 37-45, 2015

    work page 2015

  2. [1]

    & Weber, A

    M¨ uller, M., R¨ oder, T., Clausen, M., Eberhardt, B., Kr¨ uger, B. & Weber, A. ”Doc- umentation mocap database hdm05.” Technical report, No. CG-2007-2, ISSN 1610- 8892, Universitt Bonn, June 2007

    work page 2007

  3. [2]

    ”Ntu rgb+ d: A large scale dataset for 3d human activity analysis.” Computer Vision and Pattern Recog- nition, pp

    Shahroudy, Amir, Jun Liu, Tian-Tsong Ng, and Gang Wang. ”Ntu rgb+ d: A large scale dataset for 3d human activity analysis.” Computer Vision and Pattern Recog- nition, pp. 1010-1019. 2016

    work page 2016

  4. [3]

    & Perez, P

    Junejo, I., Dexter, E., Laptev, I. & Perez, P. ”View-independent action recognition from temporal self-similarities.” Transactions On Pattern Analysis And Machine Intelligence. 33, 172–185, 2010

    work page 2010

  5. [4]

    ”Efficient and robust anno- tation of motion capture data.” Eurogr

    M¨ uller, Meinard, Andreas Baak, and Hans-Peter Seidel. ”Efficient and robust anno- tation of motion capture data.” Eurogr. Symp. on Computer Animation, pp. 17-26. 2009

    work page 2009

  6. [5]

    & Tang, Y

    Liu, X., He, G., Peng, S., Cheung, Y. & Tang, Y. Efficient human motion retrieval via temporal adjacent bag of words and discriminative neighborhood preserving dictionary learning. Transactions On Human-machine Systems . 47, 763–776, 2017

    work page 2017

  7. [6]

    & Yaghmaee, F

    Ramezani, M. & Yaghmaee, F. ”Motion pattern based representation for improving human action retrieval.”Multimedia Tools And Applications. 77, 26009–26032, 2018

    work page 2018

  8. [7]

    ”Retrieval and visualization of human motion data via stick figures.” Computer Graphics Forum, vol

    Choi, Myung Geol, Kyungyong Yang, Takeo Igarashi, Jun Mitani, and Jehee Lee. ”Retrieval and visualization of human motion data via stick figures.” Computer Graphics Forum, vol. 31, no. 7, pp. 2057-2065. 2012

    work page 2012

Show all 29 references
  1. [8]

    ”Motion retrieval using probability graph model.” International Symposium on Computational Intel- ligence and Design, vol

    Xiao, Qinkun, Junfang Li, Yi Wang, Zhao Li, and Haiyun Wang. ”Motion retrieval using probability graph model.” International Symposium on Computational Intel- ligence and Design, vol. 2, pp. 150-153. 2013

    2013

  2. [9]

    & Feng, Y

    Qi, T., Xiao, J., Zhuang, Y., Zhang, H., Yang, X., Zhang, J. & Feng, Y. ”Real-time Motion Data Annotation via Action String.” Comput. Animat. Virtual Worlds . 25, 293–302, 2014

    2014

  3. [10]

    & Xiao, J

    Chen, C., Zhuang, Y., Nie, F., Yang, Y., Wu, F. & Xiao, J. ”Learning a 3D human pose distance metric from geometric pose descriptor.” Transactions On Visualiza- tion And Computer Graphics . 17, 1676–1689, 2010. 14 N Battan, A Venkat et al

    2010

  4. [11]

    ”Motion templates for automatic classification and retrieval of motion capture data.” In Proceedings of the 2006 Eurogr

    M¨ uller, Meinard, and Tido Rder. ”Motion templates for automatic classification and retrieval of motion capture data.” In Proceedings of the 2006 Eurogr. Symp. on Computer animation, pp. 137-146. Eurographics Association, 2006

    2006

  5. [12]

    & Zhang, J

    Wang, Z., Feng, Y., Qi, T., Yang, X. & Zhang, J. ”Adaptive multi-view feature selection for human motion retrieval.” Signal Processing. 120 pp. 691–701, 2016

    2016

  6. [13]

    Aggarwal

    Xia, Lu, Chia-Chih Chen, and Jake K. Aggarwal. ”View invariant human action recognition using histograms of 3d joints.” Computer Vision and Pattern Recogni- tion Workshops, pp. 20-27. IEEE, 2012

    2012

  7. [14]

    ”Actional-Structural Graph Convolutional Networks for Skeleton-based Action Recognition.” Computer Vision and Pattern Recognition, pp

    Li, Maosen, Siheng Chen, Xu Chen, Ya Zhang, Yanfeng Wang, and Qi Tian. ”Actional-Structural Graph Convolutional Networks for Skeleton-based Action Recognition.” Computer Vision and Pattern Recognition, pp. 3595-3603, 2019

    2019

  8. [15]

    ”Skeleton-Based Action Recognition with Directed Graph Neural Networks.” Conference on Computer Vi- sion and Pattern Recognition, pp

    Shi, Lei, Yifan Zhang, Jian Cheng, and Hanqing Lu. ”Skeleton-Based Action Recognition with Directed Graph Neural Networks.” Conference on Computer Vi- sion and Pattern Recognition, pp. 7912-7921. 2019

    2019

  9. [16]

    Yan, Sijie, Yuanjun Xiong, and Dahua Lin.” Spatial temporal graph convolutional networks for skeleton-based action recognition.” In Thirty-Second AAAI Conference on Artificial Intelligence. 2018

    2018

  10. [17]

    & Yuan, J

    Wang, J., Liu, Z., Wu, Y. & Yuan, J. ”Learning actionlet ensemble for 3D human action recognition.” Transactions On Pattern Analysis And Machine Intelligence . 36, 914–927, 2013

    2013

  11. [18]

    & Nikolaidis, N

    Kapsouras, I. & Nikolaidis, N. ”Action recognition on motion capture data using a dynemes and forward differences representation.” Journal Of Visual Commun. And Image Rep.. 25, 1432–1445, 2014

    2014

  12. [19]

    & Gao, Y

    Li, S., Li, W., Cook, C., Zhu, C. & Gao, Y. Independently Recurrent Neural Net- work (IndRNN): Building A Longer and Deeper RNN. 2018 Conference On Com- puter Vision And Pattern Recognition , 2018

    2018

  13. [20]

    & Luo, J

    Li, Q., Qiu, Z., Yao, T., Mei, T., Rui, Y. & Luo, J. Action Recognition by Learning Deep Multi-Granular Spatio-Temporal Video Representation, 2016

    2016

  14. [21]

    ”Deep progressive reinforcement learning for skeleton-based action recognition.” Computer Vision and Pattern Recognition, pp

    Tang, Yansong, Yi Tian, Jiwen Lu, Peiyang Li, and Jie Zhou. ”Deep progressive reinforcement learning for skeleton-based action recognition.” Computer Vision and Pattern Recognition, pp. 5323-5332. 2018

    2018

  15. [22]

    & Lee, T

    Chao, M., Lin, C., Assa, J. & Lee, T. Human motion retrieval from hand-drawn sketch. Transactions On Visualization And Computer Graphics . 18, 729–740, 2011

    2011

  16. [23]

    Li, C., Zhong, Q., Xie, D. & Pu, S. ”Co-occurrence Feature Learning from Skele- ton Data for Action Recognition and Detection with Hierarchical Aggregation”. International Joint Conference On Artificial Intelligence . 2018

    2018

  17. [24]

    & Bajcsy, R

    Ofli, F., Chaudhry, R., Kurillo, G., Vidal, R. & Bajcsy, R. ”Sequence of the most informative joints (smij): A new representation for human skeletal action recogni- tion.” Journal Of Visual Commun. And Image Rep. . 25, 24–38, 2014

    2014

  18. [25]

    & El-saban, M

    Gowayyed, M., Torki, M., Hussein, M. & El-saban, M. ”Histogram of oriented dis- placements (HOD): Describing trajectories of human joints for action recognition”, International Joint Conference on Artificial Intelligence, pp. 1351–1357, 2013

    2013

  19. [26]

    & Zezula, P

    Carrara, F., Elias, P., Sedmidubsky, J. & Zezula, P. ”LSTM-based real-time ac- tion detection and prediction in human motion streams.” Multimedia Tools And Applications. pp. 1–23, 2019

    2019

  20. [28]

    ”HumanMeshNet: Polygonal Mesh Recovery of Humans.” arXiv preprint arXiv:1908.06544, 2019

    Venkat, Abbhinav, et al. ”HumanMeshNet: Polygonal Mesh Recovery of Humans.” arXiv preprint arXiv:1908.06544, 2019

    1908 arXiv

  21. [29]

    ”Deep textured 3d re- construction of human bodies.” arXiv preprint arXiv:1809.06547 2018

    Venkat, Abbhinav, Sai Sagar Jinka, and Avinash Sharma. ”Deep textured 3d re- construction of human bodies.” arXiv preprint arXiv:1809.06547 2018

    2018 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.