Pith. sign in

REVIEW 3 major objections 3 minor 53 references

HyPCV-Former: Hyperbolic Spatio-Temporal Transformer for 3D Point Cloud Video Anomaly Detection

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

Pith's one-line read This paper claims that performing point cloud video anomaly detection entirely in Lorentzian hyperbolic space, with a curvature-aware attention mechanism and no tangent-space detour, improves results by 7% on TIMo and 5.6% on DAD.

desk verdict Promising hyperbolic point-cloud video transformer, but the abstract doesn't show the claimed gains come from the Lorentzian geometry. read the letter →

arxiv 2508.00473 v1 pith:QZVBF4OE submitted 2025-08-01 cs.CV

classification cs.CV
keywords videoanomalydetection3DpointcloudhyperbolicgeometryLorentzianspacetransformerspatio-temporalmodelingself-attentionsurveillance
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 HyPCV-Former, a video anomaly detector that works on 3D point cloud sequences rather than RGB or depth images. Its central claim is that the latent structure of video events is hierarchical, and that Euclidean feature spaces compress this structure, so the model embeds per-frame point cloud features into Lorentzian hyperbolic space and keeps all subsequent operations there. The claimed payoff is empirical: a 7% improvement on the TIMo benchmark and a 5.6% gain on the DAD benchmark over existing methods. A sympathetic reader would care because this is a direct test of whether non-Euclidean geometry, not just a larger or deeper transformer, can improve spatio-temporal anomaly detection.

What carries the argument

The load-bearing object is the Lorentz model of hyperbolic space, a constant-negative-curvature geometry in which the squared distance between points grows with the Lorentzian inner product $\langle \mathbf{x},\mathbf{y}\rangle_{\mathcal{L}}$; in this geometry, embeddings can place hierarchical structure with exponentially growing distances. The paper's named mechanism is hyperbolic multi-head self-attention (HMHA), which replaces standard dot-product attention with Lorentzian inner products and a curvature-aware softmax, so temporal dependencies are weighted according to the geometry's negative curvature rather than a Euclidean similarity. The design choice that carries the argument is performing every step, including anomaly scoring, directly in full Lorentzian space instead of mapping to a tangent space, which the paper says preserves the hierarchical structure that Euclidean layers would flatten.

What would settle it

Run a controlled ablation that keeps the point cloud extractor, training schedule, and anomaly scorer identical but replaces the Lorentzian inner product and curvature-aware softmax with a standard Euclidean dot-product attention; if the Euclidean variant matches or beats the reported TIMo and DAD numbers, the paper's central geometric claim is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that full Lorentzian hyperbolic space can serve as the entire working space for point cloud video anomaly detection. Prior Euclidean embeddings are described as unable to capture hierarchical event structure and spatio-temporal continuity; HyPCV-Former instead extracts per-frame spatial features with a point cloud extractor, embeds them into the Lorentz model, and performs temporal modeling with hyperbolic multi-head self-attention (HMHA) that uses Lorentzian inner products and a curvature-aware softmax. Crucially, the paper states that all feature transformations and anomaly scoring happen directly in full Lorentzian space, not via tangent-space approximation. The reported result is state-of-the-art performance with a 7% improvement on TIMo and a 5.6% gain on DAD compared to benchmarks.

Load-bearing premise

The load-bearing premise is that the events visible in point cloud video are genuinely organized as a hierarchy or tree, so hyperbolic geometry is the right space to represent them; if that structural prior is wrong or irrelevant, the reported gains would have to come from the new attention mechanism or other architectural details rather than from the geometry itself.

Editorial extensions

If this is right

  • Point cloud video anomaly detection can be reformulated as a hierarchy-discovery problem, with hyperbolic embeddings carrying the structure end to end.
  • Attention weights computed from Lorentzian inner products with curvature-aware softmax are a viable alternative to Euclidean dot-product attention for temporal modeling.
  • The reported gains on both TIMo and DAD imply the benefit of full-Lorentzian processing appears across different anomaly categories and recording conditions, not on a single benchmark.
  • If the result holds, future 3D video understanding tasks that involve event hierarchies, such as action segmentation or scene prediction, may gain from moving the whole representation pipeline into hyperbolic space.

Reading between the lines

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

  • A direct geometry-only ablation, keeping the same point cloud extractor, training objective, and anomaly scorer but swapping Lorentzian attention for Euclidean dot-product attention, would isolate how much of the reported gain is due to hyperbolic space itself versus the new attention mechanism.
  • Because curvature in the Lorentz model controls how fast distances grow, learning a per-class or per-dataset curvature parameter is a natural next knob that could encode finer event hierarchies.
  • The hierarchical-structure rationale implies a testable signature: hyperbolic embeddings should show greater tree-likeness than Euclidean embeddings on the same video events, and measuring that quantity would directly validate the paper's motivating premise.
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 / 3 minor

Summary. The manuscript proposes HyPCV-Former, a hyperbolic spatio-temporal transformer for anomaly detection in 3D point cloud videos. The method extracts per-frame spatial features, embeds them into Lorentzian hyperbolic space, and uses a hyperbolic multi-head self-attention (HMHA) with Lorentzian inner products and curvature-aware softmax to model temporal dependencies, with anomaly scoring performed directly in Lorentzian space. The abstract reports state-of-the-art results with a 7% improvement on the TIMo dataset and a 5.6% gain on the DAD dataset compared to benchmarks. However, the submitted document contains only the abstract; the full text and all experimental details are not included in the review copy.

Significance. If the claims are verified, the paper would make a notable contribution to point cloud video anomaly detection by demonstrating that full Lorentzian space, rather than tangent-space approximations, can be used for feature transformation and scoring. The promise of code release is a positive step for reproducibility. Nonetheless, the significance cannot be fully assessed from the abstract alone because the technical details, baselines, and evidence are absent, making it impossible to judge whether the reported gains are reliable and attributable to the hyperbolic components.

major comments (3)
  1. [Abstract] The abstract states that anomaly scoring is performed directly within full Lorentzian space, but no mathematical definition of the scoring function is given, and no argument is provided for its invariance under Lorentz transformations; if the scoring relies on Euclidean distances or inner products in disguise, the claimed non-Euclidean contribution would be void.
  2. [Abstract] The reported 7% and 5.6% gains on TIMo and DAD are presented without specifying the baseline methods, the experimental protocol, the number of runs, or variance measures, and without an ablation study that isolates HMHA, curvature-aware softmax, and Lorentzian scoring from the rest of the architecture; these omissions leave the causal claim that hyperbolic geometry is responsible for the improvements unsubstantiated.
  3. [Submission (full text absent)] The review copy contains no full text, so the architecture of HyPCV-Former, the point cloud extractor, the training procedure, and any numerical-stability guarantees for high-dimensional Lorentzian operations are not available for evaluation; this is a load-bearing omission because the method cannot be reproduced or independently checked from the abstract.
minor comments (3)
  1. [Abstract] The term 'curvature-aware softmax' is not defined; the manuscript should state whether the curvature is a learned parameter and how it is incorporated into the softmax denominator.
  2. [Abstract] The phrase 'full Lorentzian space' should be clarified by specifying the ambient dimension and the chosen pseudo-sphere (e.g., Lorentz factor or pseudo-radius), as these choices affect the distances and scoring.
  3. [Abstract] The abstract does not indicate the scale of the method, such as number of parameters or inference speed; including this would help readers judge the practical overhead relative to Euclidean baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported gains are external benchmark comparisons, not reductions to fitted inputs.

full rationale

The paper is an empirical application of a proposed architecture to public benchmarks. The central claim is that HyPCV-Former achieves a 7% improvement on TIMo and a 5.6% gain on DAD compared to benchmarks. These are comparisons against external baselines, which is the opposite of a circular derivation: the performance numbers are not constructed from the method's own outputs, nor are they renamed inputs. The hyperbolic geometry argument is a design hypothesis, not a derivation that presumes the conclusion. No fitted parameter is relabeled as a prediction, no load-bearing self-citation appears, and no uniqueness theorem is imported from the authors' prior work. The absence of ablations or error bars is a completeness or correctness concern, but it is not circularity. Accordingly, the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

Abstract-only review. The central method relies on the domain assumption that point cloud surveillance events have inherently hierarchical structure that hyperbolic space represents more faithfully. No free parameters or invented entities are identifiable from the abstract alone.

assumptions (1)
  • domain assumption Hyperbolic (Lorentzian) geometry better captures hierarchical event structures in point cloud videos than Euclidean geometry.
    Assumed in the abstract without empirical or theoretical justification; this is the core geometric prior underlying the method.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HyPCV-Former: Hyperbolic Spatio-Temporal Transformer for 3D Point Cloud Video Anomaly Detection." pith.science (2026). https://pith.science/paper/QZVBF4OE

@misc{pith2026250800473,
  author       = {Pith},
  title        = {Pith review of: HyPCV-Former: Hyperbolic Spatio-Temporal Transformer for 3D Point Cloud Video Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QZVBF4OE}},
  note         = {Machine review of arXiv:2508.00473}
}
read the original abstract

Video anomaly detection is a fundamental task in video surveillance, with broad applications in public safety and intelligent monitoring systems. Although previous methods leverage Euclidean representations in RGB or depth domains, such embeddings are inherently limited in capturing hierarchical event structures and spatio-temporal continuity. To address these limitations, we propose HyPCV-Former, a novel hyperbolic spatio-temporal transformer for anomaly detection in 3D point cloud videos. Our approach first extracts per-frame spatial features from point cloud sequences via point cloud extractor, and then embeds them into Lorentzian hyperbolic space, which better captures the latent hierarchical structure of events. To model temporal dynamics, we introduce a hyperbolic multi-head self-attention (HMHA) mechanism that leverages Lorentzian inner products and curvature-aware softmax to learn temporal dependencies under non-Euclidean geometry. Our method performs all feature transformations and anomaly scoring directly within full Lorentzian space rather than via tangent space approximation. Extensive experiments demonstrate that HyPCV-Former achieves state-of-the-art performance across multiple anomaly categories, with a 7\% improvement on the TIMo dataset and a 5.6\% gain on the DAD dataset compared to benchmarks. The code will be released upon paper acceptance.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 43 canonical work pages

  1. [1]

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

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Bergmann, P.; Jin, X.; Sattlegger, D.; and Steger, C. 2021. The mvtec 3d-ad dataset for unsupervised 3d anomaly detection and localization. arXiv preprint arXiv:2112.09045

  4. [4]

    Braham, M.; Lejeune, A.; and Van Droogenbroeck, M. 2014. A physically motivated pixel-based model for background subtraction in 3D images. In 2014 International Conference on 3D Imaging (IC3D), 1--8. IEEE

  5. [5]

    Chami, I.; Ying, Z.; R \'e , C.; and Leskovec, J. 2019. Hyperbolic graph convolutional neural networks. Advances in neural information processing systems, 32

  6. [6]

    Chen, W.; Han, X.; Lin, Y.; Zhao, H.; Liu, Z.; Li, P.; Sun, M.; and Zhou, J. 2021. Fully hyperbolic neural networks. arXiv preprint arXiv:2105.14686

  7. [7]

    Cho, S.; Cho, S.; Park, S.; Lee, H.; Lee, H.; and Lee, M. 2023. Curve your attention: Mixed-curvature transformers for graph representation learning. arXiv preprint arXiv:2309.04082

  8. [8]

    Dai, J.; Wu, Y.; Gao, Z.; and Jia, Y. 2021. A hyperbolic-to-hyperbolic graph convolutional network. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 154--163

Show all 53 references
  1. [9]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

  2. [10]

    Du, H.; Liu, C.; Liu, H.; Ding, X.; and Huo, H. 2024. An efficient federated learning framework for graph learning in hyperbolic space. Knowledge-Based Systems, 289: 111438

  3. [11]

    Ermolov, A.; Mirvakhabova, L.; Khrulkov, V.; Sebe, N.; and Oseledets, I. 2022. Hyperbolic vision transformers: Combining improvements in metric learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 7409--7419

  4. [12]

    Ganea, O.; B \'e cigneul, G.; and Hofmann, T. 2018. Hyperbolic neural networks. Advances in neural information processing systems, 31

  5. [13]

    M.; Battaglia, P.; Bapst, V.; Raposo, D.; Santoro, A.; et al

    Gulcehre, C.; Denil, M.; Malinowski, M.; Razavi, A.; Pascanu, R.; Hermann, K. M.; Battaglia, P.; Bapst, V.; Raposo, D.; Santoro, A.; et al. 2018. Hyperbolic attention networks. arXiv preprint arXiv:1805.09786

  6. [14]

    He, T.; Wang, W.; and Zeng, G. 2024. Point cloud video anomaly detection based on point spatio-temporal auto-encoder. IEEE Sensors Journal

  7. [15]

    Hong, J.; Fang, P.; Li, W.; Han, J.; Petersson, L.; and Harandi, M. 2023. Curved geometric networks for visual anomaly recognition. IEEE transactions on neural networks and learning systems

  8. [16]

    M.; Bennewitz, M.; Stachniss, C.; and Burgard, W

    Hornung, A.; Wurm, K. M.; Bennewitz, M.; Stachniss, C.; and Burgard, W. 2013. OctoMap: An efficient probabilistic 3D mapping framework based on octrees. Autonomous robots, 34: 189--206

  9. [17]

    Karim, H.; Doshi, K.; and Yilmaz, Y. 2024. Real-time weakly supervised video anomaly detection. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, 6848--6856

  10. [18]

    Khrulkov, V.; Mirvakhabova, L.; Ustinova, E.; Oseledets, I.; and Lempitsky, V. 2020. Hyperbolic image embeddings. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 6418--6428

  11. [19]

    Kopuklu, O.; Zheng, J.; Xu, H.; and Rigoll, G. 2021. Driver anomaly detection: A dataset and contrastive learning approach. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 91--100

  12. [20]

    Law, M.; Liao, R.; Snell, J.; and Zemel, R. 2019. Lorentzian distance learning for hyperbolic representations. In International Conference on Machine Learning, 3672--3681. PMLR

  13. [21]

    Leng, J.; Wu, Z.; Tan, M.; Liu, Y.; Gan, J.; Chen, H.; and Gao, X. 2024. Beyond euclidean: Dual-space representation learning for weakly supervised video violence detection. arXiv preprint arXiv:2409.19252

  14. [22]

    Li, W.; Yang, Z.; Han, W.; Man, H.; Wang, X.; and Fan, X. 2025. Hyperbolic-constraint Point Cloud Reconstruction from Single RGB-D Images. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, 4959--4967

  15. [23]

    Liu, Q.; Nickel, M.; and Kiela, D. 2019. Hyperbolic graph neural networks. Advances in neural information processing systems, 32

  16. [24]

    Liu, Y.; Yang, D.; Wang, Y.; Liu, J.; Liu, J.; Boukerche, A.; Sun, P.; and Song, L. 2024. Generalized video anomaly event detection: Systematic taxonomy and comparison of deep models. ACM Computing Surveys, 56(7): 1--38

  17. [25]

    T.; and Snoek, C

    Long, T.; Mettes, P.; Shen, H. T.; and Snoek, C. G. 2020. Searching for actions on the hyperbole. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 1141--1150

  18. [26]

    Ma, X.; Qin, C.; You, H.; Ran, H.; and Fu, Y. 2022. Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework. In International Conference on Learning Representations

  19. [27]

    Mettes, P.; Ghadimi Atigh, M.; Keller-Ressel, M.; Gu, J.; and Yeung, S. 2024. Hyperbolic deep learning in computer vision: A survey. International Journal of Computer Vision, 132(9): 3484--3508

  20. [28]

    Montanaro, A.; Valsesia, D.; and Magli, E. 2022. Rethinking the compositionality of point clouds through regularization in the hyperbolic space. Advances in Neural Information Processing Systems, 35: 33741--33753

  21. [29]

    Nie ner, M.; Zollh \"o fer, M.; Izadi, S.; and Stamminger, M. 2013. Real-time 3D reconstruction at scale using voxel hashing. ACM Transactions on Graphics (ToG), 32(6): 1--11

  22. [30]

    Ou, X.; Chang, Q.; and Chakraborty, N. 2019. Simulation study on reward function of reinforcement learning in gantry work cell scheduling. Journal of manufacturing systems, 1--8

  23. [31]

    Peng, W.; Shi, J.; Xia, Z.; and Zhao, G. 2020. Mix dimension in poincar \'e geometry for 3d skeleton-based action recognition. In Proceedings of the 28th ACM international conference on multimedia, 1432--1440

  24. [32]

    Peng, X.; Wen, H.; Luo, Y.; Zhou, X.; Yu, K.; Yang, P.; and Wu, Z. 2023. Learning weakly supervised audio-visual violence detection in hyperbolic space. arXiv preprint arXiv:2305.18797

  25. [33]

    R.; Su, H.; Mo, K.; and Guibas, L

    Qi, C. R.; Su, H.; Mo, K.; and Guibas, L. J. 2017 a . Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, 652--660

  26. [34]

    R.; Yi, L.; Su, H.; and Guibas, L

    Qi, C. R.; Yi, L.; Su, H.; and Guibas, L. J. 2017 b . Pointnet++: Deep hierarchical feature learning on point sets in a metric space. Advances in neural information processing systems, 30

  27. [35]

    Sandler, M.; Howard, A.; Zhu, M.; Zhmoginov, A.; and Chen, L.-C. 2018. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition, 4510--4520

  28. [36]

    Schneider, P.; Anisimov, Y.; Islam, R.; Mirbach, B.; Rambach, J.; Stricker, D.; and Grandidier, F. 2022 a . Timo—a dataset for indoor building monitoring with a time-of-flight camera. Sensors, 22(11): 3992

  29. [37]

    Schneider, P.; Rambach, J.; Mirbach, B.; and Stricker, D. 2022 b . Unsupervised anomaly detection from time-of-flight depth images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 231--240

  30. [38]

    Shi, X.; Chen, Z.; Wang, H.; Yeung, D.-Y.; Wong, W.-K.; and Woo, W.-c. 2015. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems, 28

  31. [39]

    Shimizu, R.; Mukuta, Y.; and Harada, T. 2020. Hyperbolic neural networks++. arXiv preprint arXiv:2006.08210

  32. [40]

    Shimizu, R.; Wang, Y.; Kimura, M.; Hirakawa, Y.; Wada, T.; Saito, Y.; and McAuley, J. 2024. A fashion item recommendation model in hyperbolic space. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8377--8383

  33. [41]

    Sun, L.; Zhang, Z.; Zhang, J.; Wang, F.; Peng, H.; Su, S.; and Yu, P. S. 2021. Hyperbolic variational graph neural network for modeling dynamic graphs. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, 4375--4383

  34. [42]

    Tan, Y.; Yang, C.; Wei, X.; Chen, C.; Li, L.; and Zheng, X. 2022. Enhancing recommendation with automated tag taxonomy construction in hyperbolic space. In 2022 IEEE 38th International Conference on Data Engineering (ICDE), 1180--1192. IEEE

  35. [43]

    A.; Pham, Q.-H.; Hua, B.-S.; Nguyen, T.; and Yeung, S.-K

    Uy, M. A.; Pham, Q.-H.; Hua, B.-S.; Nguyen, T.; and Yeung, S.-K. 2019. Revisiting point cloud classification: A new benchmark dataset and classification model on real-world data. In Proceedings of the IEEE/CVF international conference on computer vision, 1588--1597

  36. [44]

    N.; Kaiser, .; and Polosukhin, I

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, .; and Polosukhin, I. 2017. Attention is all you need. Advances in neural information processing systems, 30

  37. [45]

    E.; Bronstein, M

    Wang, Y.; Sun, Y.; Liu, Z.; Sarma, S. E.; Bronstein, M. M.; and Solomon, J. M. 2019. Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (tog), 38(5): 1--12

  38. [46]

    Xiao, A.; Huang, J.; Guan, D.; Zhang, X.; Lu, S.; and Shao, L. 2023. Unsupervised point cloud representation learning with deep neural networks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(9): 11321--11339

  39. [47]

    Xie, Y.; Tian, J.; and Zhu, X. X. 2020. Linking points with labels in 3D: A review of point cloud semantic segmentation. IEEE Geoscience and remote sensing magazine, 8(4): 38--59

  40. [48]

    C.; Liu, J.; King, I.; and Ying, R

    Yang, M.; Verma, H.; Zhang, D. C.; Liu, J.; King, I.; and Ying, R. 2024. Hypformer: Exploring efficient transformer fully in hyperbolic space. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 3770--3781

  41. [49]

    Yang, Z.; Liu, J.; and Wu, P. 2024. Text prompt with normality guidance for weakly supervised video anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18899--18908

  42. [50]

    Zhang, C.; Fiore, M.; Murray, I.; and Patras, P. 2021. Cloudlstm: A recurrent neural model for spatiotemporal point-cloud stream forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, 10851--10858

  43. [51]

    Zhang, L.; Zhou, K.; Lu, F.; Li, Z.; Shao, X.; Zhou, X.-D.; and Shi, Y. 2025. ESMformer: Error-aware self-supervised transformer for multi-view 3D human pose estimation. Pattern Recognition, 158: 110955

  44. [52]

    Zhang, L.; Zhou, K.; Lu, F.; Zhou, X.-D.; and Shi, Y. 2024. Deep semantic graph transformer for multi-view 3d human pose estimation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 7205--7214

  45. [53]

    Zhang, X.; Zhou, X.; Lin, M.; and Sun, J. 2018. Shufflenet: An extremely efficient convolutional neural network for mobile devices. In Proceedings of the IEEE conference on computer vision and pattern recognition, 6848--6856

Pith tools

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