Pith. sign in

REVIEW 5 major objections 5 minor 44 references

A decoder-free teacher–student skeleton model learns better action representations from a compact 8×25 token grid than dense-token masked auto-encoders, cutting inference cost by 7.89×.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-04 05:49 UTC pith:G57KNKC5

load-bearing objection Solid skeleton-SSL paper with useful masking and augmentation ideas, but the 7.89x efficiency claim is confounded by token density and the SOTA claim is overbroad. the 5 major comments →

arxiv 2603.10648 v3 pith:G57KNKC5 submitted 2026-03-11 cs.CV

Less is More: Compact-Token Masked Feature Prediction for Skeleton Representation Learning

Jeonghyeok Do , Yun Chen , Geunhyuk Youk , Munchurl Kim This is my paper
classification cs.CV
keywords skeleton-based action recognitionself-supervised learningmasked feature modelingdecoder-free masked modelingteacher–student distillationsemantic tube maskingskeleton-aware augmentationscontrastive learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper aims to overturn the standard recipe for skeleton self-supervised learning, where a masked auto-encoder reconstructs raw joint coordinates from a dense grid of tokens. It claims that such reconstruction is wasteful: the decoder only learns to interpolate correlated joints, and downstream inference must process a dense sequence that was never actually used during pretraining. SLiM instead trains a single shared encoder with a teacher–student objective: the student sees a sparsely masked 8×25 token view and predicts the teacher's high-dimensional prototype distributions for masked patches, while a contrastive branch aligns local temporal crops with the global view. The authors report state-of-the-art accuracy on three action-recognition benchmarks and show that this representation is achieved at 3.59 GFLOPs of inference compute, a 7.89× reduction relative to existing MAE baselines. A sympathetic reader would take away that the difficulty of skeleton pretraining lies not in how many tokens you keep, but in what objective you ask the model to satisfy.

Core claim

SLiM's central claim is that masked feature prediction, performed on a deliberately compact skeletal-temporal token grid and without any reconstruction decoder, is a strictly better pretext for skeleton representation learning than dense-token coordinate reconstruction. A momentum-updated teacher encodes an unmasked global view of a skeleton sequence; the student encodes the same view masked as semantic tubes — contiguous anatomical joint groups occluded across time — and must match the teacher's softmax distributions over 65,536 prototypes at the masked positions. The same shared encoder also optimizes a global-local contrastive objective, aligning class tokens of local clips drawn from the

What carries the argument

The load-bearing mechanism is a teacher–student distillation with an exponential-moving-average teacher, where the target is not raw coordinates but a distribution over 65,536 prototypes. The student's encoder is a small vision transformer with 8 layers and 256 hidden units, processing 8×25 patch tokens; rotary temporal position encodings let it handle clips of different lengths. Semantic tube masking occludes anatomically connected joint groups over contiguous temporal segments, so the model cannot infer a missing joint from its immediate neighbors. The contrastive branch uses skeleton-aware augmentations (rotation around the gravity axis, geometric mirroring, bone-length scaling) to build

Load-bearing premise

The method depends on the EMA teacher's 65,536-way softmax targets being semantically meaningful and not collapsed to a small set of prototypes; if the teacher's predictions become near-uniform or concentrate on a few codes, the masked feature-modeling loss is trivially satisfied and the reported gains cannot be attributed to the predictive objective.

What would settle it

Compute the effective number of prototypes (e.g., entropy of the teacher's assignment distribution) on a held-out skeleton split every few training epochs. If the teacher assigns nearly all mass to fewer than, say, 100 of its 65,536 prototypes by epoch 150, the cross-entropy target carries almost no information; the gains presumably come from the contrastive branch. A second check is to retrain with teacher targets replaced by a fixed random distribution; if top-1 accuracy stays at the full-model level, the teacher's features are not the source of the improvement.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Coordinate-level reconstruction can be replaced by teacher-guided feature prediction without loss of accuracy, eliminating the decoder and the computational asymmetry between pretraining and inference.
  • A compact 8×25 token grid is enough for strong skeleton representations; dense token grids are not necessary for state-of-the-art accuracy.
  • Semantic tube masking contributes to both branches: harder masked-modeling targets and harder contrastive perturbations, jointly improving accuracy over standard random masking.
  • The unified shared-encoder objective transfers across benchmarks and label regimes, including a 1% labeled-data setting where the compact model is particularly strong.
  • Inference cost drops to 3.59 GFLOPs (from up to 28.32 GFLOPs in MAE baselines), while accuracy improves on all five evaluation protocols.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the teacher's prototype space is indeed non-collapsed, the same decoder-free predictive objective should transfer to other structured modalities (point clouds, graphs, or multi-agent trajectories) where local coordinate interpolation is similarly cheap.
  • The paper's efficiency number is computed on the encoder; a systems-level measure that includes tokenization, the projection heads, and the two-view generation could show a smaller end-to-end gap, which would not affect the accuracy claims but would bound the real-world speedup.
  • The 1% labeled-data result suggests the decisive factor under extreme sparsity is the compactness and geometric validity of the representation, not the amount of pretraining data; a natural test is scaling pretraining data while keeping the token grid fixed.
  • It remains an open question whether an information-theoretic shortcut still exists: the contrastive branch alone might suffice, with masked feature prediction acting as a regularizer rather than the primary semantic signal; the paper's ablations show joint training is best, but do not isolate the teacher's contribution in a way that rules out pure augmentation benefits.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes SLiM, a self-supervised skeleton representation learning framework that unifies masked feature prediction and global-local contrastive learning in a decoder-free teacher–student architecture. The student encoder processes compact, semantic-tube-masked views and predicts the EMA teacher's prototype distributions over masked patches, while multiple temporally hierarchical views are aligned to a global anchor through a contrastive objective. The authors introduce Semantic Tube Masking (STM) and Skeleton-Aware Augmentations (SAA) and report state-of-the-art linear-evaluation accuracies on NTU-60, NTU-120, and PKU-MMD II, together with a claimed 7.89× reduction in inference FLOPs relative to dense-token MAE baselines.

Significance. If validated, the core idea—that compact token grids are sufficient when the pretext task is decoder-free feature prediction rather than coordinate reconstruction—would be a useful step for efficient skeleton SSL. The ablations in Tables 5–7 support the contributions of the unified objective, semantic tube masking, and skeleton-aware augmentations in an internal sense. However, the paper's headline efficiency and state-of-the-art claims currently rest on a confounded comparison and an overstatement, and one of the core augmentation algorithms contains a concrete indexing error. The work is therefore promising but needs additional experiments and corrections before its central claims can be accepted.

major comments (5)
  1. [Table 1; Sec. 1] The 7.89× inference-efficiency claim is confounded by token density and input length. SLiM uses an 8×25 token grid with T=64 and P_T=8, whereas the MAE baselines use a 30×25 grid with T=120 and P_T=4. The reduction from 28.32 to 3.59 GFLOPs is therefore mostly attributable to using 3.75× fewer tokens and a shorter input, not to the decoder-free objective. No baseline trains a conventional MAE/GFP objective under SLiM's exact tokenization and sequence length. Please add this same-tokenization control; without it, the 'less is more' efficiency thesis is unsupported.
  2. [Abstract; Sec. 4.3; Table 3] The statement that SLiM 'consistently achieves state-of-the-art performance across all downstream protocols' is contradicted by Table 3: under the 10% label setting on NTU-60 X-View, SLiM scores 91.9% while GFP scores 92.1%. The 10% X-Sub margin is also only 0.1 pp (88.8 vs 88.7). The abstract, Section 1, and Section 5 should be corrected to accurately describe the semi-supervised results, e.g., 'competitive or superior in all protocols except 10% X-View.'
  3. [Appendix D.2; Algorithm 3] The bilateral mapping in Skeleton-Aware Mirroring is inconsistent with the joint groups defined in Appendix C under the standard Kinect v2 topology. The paper's arm mapping {(5,9),(6,10),(7,11),(8,12)} would swap the right shoulder (joint 8) with the left hip (joint 12), and the leg mapping {(13,17),(14,18),(15,19),(16,20)} would swap the right hip (16) with joint 20 (spine shoulder). These are not left–right symmetric pairs. This would produce anatomically invalid poses and undermines the attribution in Table 7. Please provide a correct mapping or clarify if a non-standard joint ordering is used.
  4. [Sec. 3.2, Eq. (4); Appendix A] The masked feature modeling loss depends on the teacher's softmax distribution over 65,536 prototypes, regularized by Sinkhorn–Knopp and KoLeo. The paper gives no diagnostic of teacher feature quality or prototype collapse during 150-epoch skeleton pretraining. If the teacher assignments become degenerate or concentrated on a small subset of prototypes, L_MFM approaches a constant and the reported gains cannot be attributed to the new objective. Please report prototype usage entropy or a similar non-collapse metric over training, and ideally an ablation of the prototype count.
  5. [Sec. 4; Tables 2–5] The paper makes state-of-the-art claims based on margins as small as 0.1–0.2 percentage points (e.g., Table 3 10% X-Sub) but reports no error bars, no seeds, and no code release. Please provide mean±std over at least three pretraining seeds for the main comparisons, or temper the language 'consistently', 'significant lead', and 'strictly comparable or superior' accordingly.
minor comments (5)
  1. [Abstract; Sec. 1] The sentence 'SLiM is the first framework with decoder-free masked modeling of representative learning' is grammatically unclear and overclaims novelty; iBOT/DINOv2 already use decoder-free feature prediction in image SSL. Please rephrase and contextualize the novelty.
  2. [Sec. 3.4] The heading and text use both 'Skeletal-Aware Augmentations' and 'Skeleton-Aware Augmentations'; please standardize to 'Skeleton-Aware' throughout, matching the abbreviation SAA.
  3. [References] Reference entries 28 and 29 both appear to be 'Predict & cluster: Unsupervised skeleton based action recognition' (Su et al., CVPR 2020). Please verify and deduplicate.
  4. [Appendix A] Please include a statement about code release or a link to an official implementation, since the paper's reproducibility currently relies on the appendix pseudocode.
  5. [Algorithm 2] In Step 3, 'Xrot[:, t, j] = R X[:, t, j]' omits the channel dimension and the matrix-multiplication semantics; please write the operation explicitly, e.g., Xrot[:, t, j] = R @ X[:, t, j].

Circularity Check

0 steps flagged

No circular dependency: central empirical claims rest on external benchmarks; the teacher-student objective is a disclosed prior framework, not a fitted input.

full rationale

SLiM's central claims are empirical and externally benchmarked. The masked-feature modeling loss (Eq. 4) and global-local contrastive loss (Eq. 6) use the EMA teacher's own softmax outputs as targets; this is self-distillation, not a quantity fitted to downstream accuracy, so no prediction is forced by construction. The 7.89x inference-efficiency number is arithmetic from the reported GFLOPs in Table 1 (28.32/3.59), and while the comparison is confounded by different token counts and input lengths (the paper itself notes ~25% of tokens), that is an experimental-design caveat, not a circularity. The only author self-citation ([6], Skateformer) appears in a list of supervised-learning references and is not load-bearing for any method or uniqueness claim. Sinkhorn-Knopp and KoLeo regularizers are adopted from prior work and do not encode downstream labels. No step reduces to a fit or to a self-citation chain, so no circularity is found.

Axiom & Free-Parameter Ledger

10 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard deep-learning building blocks and hand-set hyperparameters. No new physical entities are introduced. The most consequential premises are the reliability of the EMA teacher target and the sufficiency of the compact token grid, neither independently verified outside the paper's own ablations.

free parameters (10)
  • loss weight lambda = 1.0
    Balances masked feature modeling and global-local contrastive losses in Eq. 7; set empirically without a reported sweep.
  • KoLeo regularization weight = 0.1
    Used to encourage uniform feature spread; reported in Appendix A without sensitivity analysis.
  • STM masking ratio range = p in [0.5, 0.9]
    Target fraction of masked tokens; sampled uniformly; hand-chosen to set pretext difficulty.
  • STM tube area bounds = A_min=8, A_max=0.5N
    Constant tube volume strategy parameters in Algorithm 1; chosen by hand.
  • SAA application probability = 0.5
    Each skeleton-aware augmentation applied independently with probability 0.5; hand-set.
  • Skeleton-aware rotation limits = theta_vert=180 deg, theta_tilt=30 deg
    Full vertical axis rotation and bounded tilt angles in Algorithm 2; hand-chosen to preserve plausible poses.
  • Bone-aware scaling range = [0.85, 1.15]
    Scale factors applied to bone lengths in Algorithm 4; hand-set without reported sensitivity.
  • Prototype space size = K=65,536
    Categorical target dimension for the cross-entropy losses; taken from iBOT/DINOv2 practice.
  • Compact token grid = T=64, P_T=8, P_J=1 -> 8x25 tokens
    The key design choice underlying the 7.89x efficiency claim; no ablation separates token density from the loss.
  • EMA momentum schedule = tau from 0.994 to 1
    Teacher parameter update schedule; standard mean-teacher/iBOT recipe.
axioms (5)
  • domain assumption The EMA teacher's full-view feature distribution is a stable, non-collapsing target for the student's masked-patch predictions.
    Eq. 4 depends on these softmax targets being meaningful; Sinkhorn-Knopp and KoLeo are borrowed from iBOT/DINOv2 but no collapse analysis is provided.
  • domain assumption Skeleton sequences tokenized as 2D skeletal-temporal patches preserve action semantics for a ViT.
    Eqs. 1-2 rely on patchification and temporal RoPE; a standard architectural assumption for skeleton transformers.
  • domain assumption Masking entire anatomical groups over time prevents shortcut interpolation and forces deeper action understanding.
    Motivates Algorithm 1; supported by the Table 6 ablation but not measured directly as a reduction in interpolation.
  • domain assumption Published baseline numbers in Tables 2-4 and 8 are accurate and comparably tuned.
    The paper reports no re-implementations or error bars for baselines; all comparisons rely on external reported values.
  • ad hoc to paper The raw 3D joint coordinates require no dataset-specific normalization or preprocessing for the reported results.
    No preprocessing is described in Section 4 or Appendix A, although skeleton SSL methods typically need centering, scaling, or coordinate-frame normalization; the reported numbers implicitly assume such steps are unnecessary or standard.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Less is More: Compact-Token Masked Feature Prediction for Skeleton Representation Learning." pith.science (2026). https://pith.science/paper/G57KNKC5

@misc{pith2026260310648,
  author       = {Pith},
  title        = {Pith review of: Less is More: Compact-Token Masked Feature Prediction for Skeleton Representation Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G57KNKC5}},
  note         = {Machine review of arXiv:2603.10648}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Current skeleton representation learning paradigms face distinct limitations: Contrastive Learning (CL) often overlooks fine-grained motion details, while Masked Auto-Encoders (MAE) rely on coordinate-level reconstruction. This reconstruction inherently demands dense token sequences and heavy decoders, wasting pre-training computation on discarded components and forcing downstream inference to process dense token grids. To resolve these bottlenecks, we propose SLiM (Skeleton Less is More), a compact-token framework that unifies masked feature prediction and contrastive learning via a shared encoder. By shifting the objective from raw coordinate reconstruction to decoder-free, teacher-guided feature prediction, SLiM breaks the reliance on dense tokenization and enables effective learning with a highly compact token grid. Crucially, to prevent trivial shortcut learning arising from strong inter-joint dependencies of human, we introduce Semantic Tube Masking together with Skeleton-Aware Augmentations to enforce deep skeletal-temporal reasoning and anatomical consistency. Extensive experiments across multiple downstream protocols demonstrate that SLiM achieves state-of-the-art performance while structurally reducing inference computation by 7.89$\times$ compared to dense-token MAE baselines.

Figures

Figures reproduced from arXiv: 2603.10648 by Geunhyuk Youk, Jeonghyeok Do, Munchurl Kim, Yun Chen.

Figure 1
Figure 1. Figure 1: Conceptual comparison of previous MAE methods and our SLiM. (a) Standard MAE methods suffer from a 14.38× computational surge during inference relative to pre-training due to asymmetric full-sequence processing. (b) SLiM synergizes masked modeling with contrastive learning in a decoder-free framework. This symmetric design achieves a 7.89× reduction in inference cost compared to MAE baselines. abundant unl… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SLiM. Our framework unifies Masked Feature Modeling and Global-Local Contrastive Learning within a decoder-free teacher–student architecture. The student encoder simultaneously minimizes the feature reconstruction error (LMFM) on masked patches and the contrastive loss (LGLCL) across diverse local views, effec￾tively capturing both fine-grained patterns and global semantics. temporal rotary pos… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of masking and augmentation strategies. Top: previous masking (a) and augmentations (b-d) resulting in trivial solutions or physically implausible poses. Bottom: our Semantic Tube Masking (e) and Skeletal-Aware Augmentations (f￾h) ensuring anatomical and physical consistency through skeleton-aware designs. naive coordinate interpolation, prompting the encoder to infer missing motion dynamics thr… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 9 linked inside Pith

  1. [1]

    In: European Conference on Computer Vision

    Abdelfattah, M., Alahi, A.: S-jepa: A joint embedding predictive architecture for skeletal action recognition. In: European Conference on Computer Vision. pp. 367–

  2. [2]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Caron, M., Touvron, H., Misra, I., Jégou, H., Mairal, J., Bojanowski, P., Joulin, A.: Emerging properties in self-supervised vision transformers. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 9650–9660 (2021)

    2021

  3. [3]

    In: International conference on machine learning

    Chen, T., Kornblith, S., Norouzi, M., Hinton, G.: A simple framework for con- trastive learning of visual representations. In: International conference on machine learning. pp. 1597–1607. PmLR (2020)

    2020

  4. [4]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Chen, Y., Zhang, Z., Yuan, C., Li, B., Deng, Y., Hu, W.: Channel-wise topology refinement graph convolution for skeleton-based action recognition. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 13359–13368 (2021)

    2021

  5. [5]

    arXiv preprint arXiv:2309.16588 (2023)

    Darcet, T., Oquab, M., Mairal, J., Bojanowski, P.: Vision transformers need reg- isters. arXiv preprint arXiv:2309.16588 (2023)

    Pith/arXiv arXiv 2023

  6. [6]

    In: European Conference on Computer Vision

    Do, J., Kim, M.: Skateformer: skeletal-temporal transformer for human action recognition. In: European Conference on Computer Vision. pp. 401–420. Springer (2024)

    2024

  7. [7]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Dong, J., Sun, S., Liu, Z., Chen, S., Liu, B., Wang, X.: Hierarchical contrast for unsupervised skeleton-based action representation learning. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 37, pp. 525–533 (2023)

    2023

  8. [8]

    arXiv preprint arXiv:2010.11929 (2020)

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

    Pith/arXiv arXiv 2010

  9. [9]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Duan, H., Zhao, Y., Chen, K., Lin, D., Dai, B.: Revisiting skeleton-based action recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2969–2978 (2022)

    2022

  10. [10]

    arXiv preprint arXiv:2303.06242 (2023)

    Franco, L., Mandica, P., Munjal, B., Galasso, F.: Hyperbolic self-paced learn- ing for self-supervised skeleton-based action representations. arXiv preprint arXiv:2303.06242 (2023)

    Pith/arXiv arXiv 2023

  11. [11]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Guo, T., Liu, H., Chen, Z., Liu, M., Wang, T., Ding, R.: Contrastive learning from extremely augmented skeleton sequences for self-supervised action recognition. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 36, pp. 762–770 (2022)

    2022

  12. [12]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    He, K., Chen, X., Xie, S., Li, Y., Dollár, P., Girshick, R.: Masked autoencoders are scalable vision learners. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 16000–16009 (2022)

    2022

  13. [13]

    arXiv preprint arXiv:2305.00666 (2023)

    Hua, Y., Wu, W., Zheng, C., Lu, A., Liu, M., Chen, C., Wu, S.: Part aware contrastive learning for self-supervised action recognition. arXiv preprint arXiv:2305.00666 (2023)

    Pith/arXiv arXiv 2023

  14. [14]

    In: European conference on computer vision

    Kim, B., Chang, H.J., Kim, J., Choi, J.Y.: Global-local motion transformer for unsupervised skeleton-based action learning. In: European conference on computer vision. pp. 209–225. Springer (2022)

    2022

  15. [15]

    International Journal of Computer Vision130(5), 1366–1401 (2022)

    Kong, Y., Fu, Y.: Human action recognition and prediction: A survey. International Journal of Computer Vision130(5), 1366–1401 (2022)

    2022

  16. [16]

    In: European Conference on Computer Vision

    Lin, L., Wu, L., Zhang, J., Liu, J.: Idempotent unsupervised representation learn- ing for skeleton-based action recognition. In: European Conference on Computer Vision. pp. 75–92. Springer (2024) 24 J. Do, et al

    2024

  17. [17]

    In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition

    Lin, L., Zhang, J., Liu, J.: Actionlet-dependent contrastive learning for unsuper- vised skeleton-based action recognition. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition. pp. 2363–2372 (2023)

    2023

  18. [18]

    Liu,C.,Hu,Y.,Li,Y.,Song,S.,Liu,J.:Pku-mmd:Alargescalebenchmarkforcon- tinuousmulti-modalhumanactionunderstanding.arXivpreprintarXiv:1703.07475 (2017)

    Pith/arXiv arXiv 2017

  19. [19]

    IEEE trans- actions on pattern analysis and machine intelligence42(10), 2684–2701 (2019)

    Liu, J., Shahroudy, A., Perez, M., Wang, G., Duan, L.Y., Kot, A.C.: Ntu rgb+ d 120: A large-scale benchmark for 3d human activity understanding. IEEE trans- actions on pattern analysis and machine intelligence42(10), 2684–2701 (2019)

    2019

  20. [20]

    arXiv preprint arXiv:1608.03983 (2016)

    Loshchilov, I., Hutter, F.: Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983 (2016)

    Pith/arXiv arXiv 2016

  21. [21]

    arXiv preprint arXiv:1711.05101 (2017)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    Pith/arXiv arXiv 2017

  22. [22]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Mao, Y., Deng, J., Zhou, W., Fang, Y., Ouyang, W., Li, H.: Masked motion predic- tors are strong 3d action representation learners. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 10181–10191 (2023)

    2023

  23. [23]

    In: European Conference on Computer Vision

    Mao, Y., Zhou, W., Lu, Z., Deng, J., Li, H.: Cmd: Self-supervised 3d action repre- sentation learning with cross-modal mutual distillation. In: European Conference on Computer Vision. pp. 734–752. Springer (2022)

    2022

  24. [24]

    arXiv preprint arXiv:2304.07193 (2023)

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023)

    Pith/arXiv arXiv 2023

  25. [25]

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

    Shah, A., Roy, A., Shah, K., Mishra, S., Jacobs, D., Cherian, A., Chellappa, R.: Halp: Hallucinating latent positives for skeleton-based self-supervised learning of actions. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18846–18856 (2023)

    2023

  26. [26]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Shahroudy, A., Liu, J., Ng, T.T., Wang, G.: Ntu rgb+ d: A large scale dataset for 3d human activity analysis. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1010–1019 (2016)

    2016

  27. [27]

    Neurocomputing568, 127063 (2024)

    Su, J., Ahmed, M., Lu, Y., Pan, S., Bo, W., Liu, Y.: Roformer: Enhanced trans- former with rotary position embedding. Neurocomputing568, 127063 (2024)

    2024

  28. [29]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Su, K., Liu, X., Shlizerman, E.: Predict & cluster: Unsupervised skeleton based ac- tion recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 9631–9640 (2020)

    2020

  29. [30]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Sun, K., Xiao, B., Liu, D., Wang, J.: Deep high-resolution representation learn- ing for human pose estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5693–5703 (2019)

    2019

  30. [31]

    In: Proceedings of the 31st ACM International Conference on Mul- timedia

    Sun, S., Liu, D., Dong, J., Qu, X., Gao, J., Yang, X., Wang, X., Wang, M.: Uni- fied multi-modal unsupervised representation learning for skeleton-based action understanding. In: Proceedings of the 31st ACM International Conference on Mul- timedia. pp. 2973–2984 (2023)

    2023

  31. [32]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Sun, S., Zhang, Z., Dong, J., Cheng, Z., Chang, X., Wang, M.: Towards efficient general feature prediction in masked skeleton modeling. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12212–12221 (2025)

    2025

  32. [33]

    IEEE transactions on pattern analysis and machine intelligence45(3), 3200–3225 (2022) SLiM: Skeleton Less is More 25

    Sun, Z., Ke, Q., Rahmani, H., Bennamoun, M., Wang, G., Liu, J.: Human action recognition from various data modalities: A review. IEEE transactions on pattern analysis and machine intelligence45(3), 3200–3225 (2022) SLiM: Skeleton Less is More 25

    2022

  33. [34]

    Advances in neural information processing systems30(2017)

    Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems30(2017)

    2017

  34. [35]

    In: Proceedings of the Computer Vision and Pattern Recogni- tion Conference

    Wang, H., Ma, X., Kuang, J., Gui, J.: Heterogeneous skeleton-based action repre- sentation learning. In: Proceedings of the Computer Vision and Pattern Recogni- tion Conference. pp. 19154–19164 (2025)

    2025

  35. [36]

    In: Pro- ceedings of the AAAI Conference on Artificial Intelligence

    Weng, W., Wang, H., Wang, J., He, L., Xie, G.S.: Usdrl: Unified skeleton-based dense representation learning with multi-grained feature decorrelation. In: Pro- ceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 8332–8340 (2025)

    2025

  36. [37]

    In: European Conference on Computer Vision

    Wu, L., Lin, L., Zhang, J., Ma, Y., Liu, J.: Macdiff: Unified skeleton modeling with masked conditional diffusion. In: European Conference on Computer Vision. pp. 110–128. Springer (2024)

    2024

  37. [38]

    In: 2023 IEEE international conference on multimedia and expo workshops (ICMEW)

    Wu, W., Hua, Y., Zheng, C., Wu, S., Chen, C., Lu, A.: Skeletonmae: Spatial- temporal masked autoencoders for self-supervised skeleton action recognition. In: 2023 IEEE international conference on multimedia and expo workshops (ICMEW). pp. 224–229. IEEE (2023)

    2023

  38. [39]

    In: European Conference on Computer Vision

    Zhang, H., Hou, Y., Zhang, W., Li, W.: Contrastive positive mining for unsuper- vised 3d action representation learning. In: European Conference on Computer Vision. pp. 36–51. Springer (2022)

    2022

  39. [40]

    zhang et al

    Zhang, J., Lin, L., Yang, S., Liu, J.: Self-supervised skeleton-based action repre- sentation learning: A benchmark and beyond: J. zhang et al. International Journal of Computer Vision134(1), 38 (2026)

    2026

  40. [41]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zheng, N., Wen, J., Liu, R., Long, L., Dai, J., Gong, Z.: Unsupervised represen- tation learning with long-term dynamics for skeleton based action recognition. In: Proceedings of the AAAI conference on artificial intelligence. vol. 32 (2018)

    2018

  41. [42]

    arXiv preprint arXiv:2111.07832 (2021)

    Zhou, J., Wei, C., Wang, H., Shen, W., Xie, C., Yuille, A., Kong, T.: ibot: Image bert pre-training with online tokenizer. arXiv preprint arXiv:2111.07832 (2021)

    Pith/arXiv arXiv 2021

  42. [43]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zhou,Y.,Duan,H.,Rao,A.,Su,B.,Wang,J.:Self-supervisedactionrepresentation learning from partial spatio-temporal skeleton sequences. In: Proceedings of the AAAI conference on artificial intelligence. vol. 37, pp. 3825–3833 (2023)

    2023

  43. [44]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Zhu, Y., Han, H., Yu, Z., Liu, G.: Modeling the relative visual tempo for self- supervised skeleton-based action recognition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 13913–13922 (2023)

    2023

  44. [45]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zolfaghari, M., Zhu, Y., Gehler, P., Brox, T.: Crossclr: Cross-modal contrastive learning for multi-modal video representations. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 1450–1459 (2021)

    2021

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.