arxiv: 2604.17914 · v1 · submitted 2026-04-20 · 💻 cs.CV

Recognition: unknown

Beyond Binary Contrast: Modeling Continuous Skeleton Action Spaces with Transitional Anchors

Yingjie Feng , Yi Wang , Jiaze Wang , Anfeng Liu , Zhuotao Tian

Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:28 UTC · model grok-4.3

classification 💻 cs.CV

keywords skeleton action recognitioncontrastive learningself-supervised learningmotion continuitytransitional anchorsmanifold calibrationaction recognition

0 comments

The pith

TranCLR replaces binary contrastive objectives with transitional anchors to model the continuous geometry of skeleton actions.

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

The paper claims that existing self-supervised methods for skeleton-based action recognition enforce binary consistency in embeddings, which fragments features and creates rigid boundaries that do not match the gradual nature of human movement. TranCLR counters this by first building explicit transitional anchors that capture the geometric structure between actions and then applying multi-level calibration to smooth the manifold at varying degrees of continuity. The result is a representation space that is both more discriminative and uncertainty-aware. A reader would care because improved continuity modeling could raise recognition accuracy while also supporting applications that require handling blended or uncertain motions.

Core claim

TranCLR captures the continuous geometry of the action space through Action Transitional Anchor Construction, which models transitional states, and Multi-Level Geometric Manifold Calibration, which adaptively adjusts the manifold across continuity levels, yielding superior accuracy and calibration on NTU RGB+D, NTU RGB+D 120, and PKU-MMD.

What carries the argument

Action Transitional Anchor Construction (ATAC) that explicitly builds geometric transitional states, paired with Multi-Level Geometric Manifold Calibration (MGMC) that performs adaptive calibration of the action manifold at multiple continuity levels.

If this is right

Representations become smoother and better reflect gradual motion transitions rather than discrete clusters.
Accuracy and calibration metrics improve on the NTU RGB+D, NTU RGB+D 120, and PKU-MMD benchmarks.
The learned features carry explicit uncertainty information useful for sequences containing transitional movements.
The framework produces more discriminative embeddings by preserving the underlying geometry of action space.

Where Pith is reading between the lines

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

The same anchor-and-calibration pattern could be tested on video or sensor streams where actions also blend continuously.
Ablation studies that remove only the transitional anchors would isolate whether continuity modeling drives the reported gains.
If successful, the approach suggests a route to reduce reliance on hard class boundaries in any self-supervised setting with ordered data.
Downstream tasks such as action forecasting may benefit because the manifold already encodes transitional states.

Load-bearing premise

That constructing explicit transitional anchors and applying multi-level manifold calibration will reliably capture motion continuity and outperform binary contrastive objectives.

What would settle it

If head-to-head experiments on the NTU RGB+D dataset show that TranCLR fails to exceed the accuracy or calibration scores of standard binary contrastive baselines.

Figures

Figures reproduced from arXiv: 2604.17914 by Anfeng Liu, Jiaze Wang, Yingjie Feng, Yi Wang, Zhuotao Tian.

**Figure 1.** Figure 1: Illustration of TranCLR’s advantages. Our method achieves higher accuracy on challenging samples and bettercalibrated confidence across diverse motion scenarios. The top-3 predictions with confidence scores are shown for each sample. with applications in human-computer interaction, medical rehabilitation, and video understanding. Recent self-supervised methods [2, 8, 73] have emerged as the dominant parad… view at source ↗

**Figure 2.** Figure 2: Performance gains of TranCLR across various tasks. TranCLR surpasses AimCLR and ActCLR in Linear Evaluation, Transfer Learning, Skeleton-Based Action Retrieval, and Calibration Analysis, demonstrating superior overall performance; calibration error metrics are reversed for consistent direction. try. Through this self-regularization process, MGMC preserves global topological coherence, resulting in a smo… view at source ↗

**Figure 3.** Figure 3: The overall scheme of TranCLR. (a) Action transitional anchors are constructed through global trajectory interpolation and local spatio-temporal substitution to generate intermediate motion states that enrich the manifold topology. (b) Based on these anchors, the Multi-Level Geometric Manifold Calibration (MGMC) progressively aligns and regulates feature distances, producing a smooth, topologyconsistent a… view at source ↗

read the original abstract

Self-supervised contrastive learning has emerged as a powerful paradigm for skeleton-based action recognition by enforcing consistency in the embedding space. However, existing methods rely on binary contrastive objectives that overlook the intrinsic continuity of human motion, resulting in fragmented feature clusters and rigid class boundaries. To address these limitations, we propose TranCLR, a Transitional anchor-based Contrastive Learning framework that captures the continuous geometry of the action space. Specifically, the proposed Action Transitional Anchor Construction (ATAC) explicitly models the geometric structure of transitional states to enhance the model's perception of motion continuity. Building upon these anchors, a Multi-Level Geometric Manifold Calibration (MGMC) mechanism is introduced to adaptively calibrate the action manifold across multiple levels of continuity, yielding a smoother and more discriminative representation space. Extensive experiments on the NTU RGB+D, NTU RGB+D 120 and PKU-MMD datasets demonstrate that TranCLR achieves superior accuracy and calibration performance, effectively learning continuous and uncertainty-aware skeleton representations. The code is available at https://github.com/Philchieh/TranCLR.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

TranCLR adds transitional anchors and multi-level calibration to contrastive learning for skeleton actions, with ablations and results that support the gains on standard datasets.

read the letter

TranCLR extends contrastive learning for skeleton action recognition by explicitly modeling motion continuity through transitional anchors and multi-level manifold calibration. The experiments confirm gains in accuracy and better calibration on the usual benchmarks. The new pieces are the Action Transitional Anchor Construction, which pulls in intermediate pose states from sequences to represent transitions, and the Multi-Level Geometric Manifold Calibration that tunes the embedding space at different continuity levels. This moves away from the standard positive-negative pairs that treat actions as discrete. The paper shows this leads to smoother representations and uncertainty awareness. What works well is the ablation studies that break down each component's contribution, the consistent outperformance on NTU RGB+D, NTU RGB+D 120, and PKU-MMD, and the release of code. The derivations seem straightforward extensions of contrastive objectives without circularity or hidden fits. The soft spots are minor. The work stays within skeleton data and self-supervised contrastive setups, so broader impact is limited. It would be stronger with tests on more diverse datasets or video-based actions, but that's not a flaw in the current claims. No evidence of parameter bloat or untested assumptions in the reported results. This paper is for people already in skeleton-based action recognition looking for refinements to contrastive methods. A reader interested in continuous representations in CV will get value from the concrete implementation and results. It deserves a serious referee because the method is falsifiable, the evidence is presented clearly with ablations, and the central argument about continuity holds up in the experiments. I would recommend sending it for peer review.

Referee Report

0 major / 3 minor

Summary. The paper introduces TranCLR, a transitional anchor-based contrastive learning framework for skeleton-based action recognition. It proposes Action Transitional Anchor Construction (ATAC) to model geometric transitional states between poses for capturing motion continuity, and Multi-Level Geometric Manifold Calibration (MGMC) to adaptively calibrate the action manifold across continuity levels. Experiments on NTU RGB+D, NTU RGB+D 120, and PKU-MMD datasets report superior accuracy and calibration metrics compared to binary contrastive baselines, with ablations supporting the contributions; code is released.

Significance. If the reported gains hold, the work meaningfully extends contrastive learning by addressing the continuity of human motion, yielding smoother and more uncertainty-aware representations. The direct, falsifiable extension of standard objectives, combined with consistent dataset results and ablations, positions it as a useful advance for skeleton action recognition. Code availability strengthens the contribution.

minor comments (3)

[§3.1] §3.1: The ATAC construction from pose sequences is described at a high level; adding a short algorithmic outline or pseudocode would improve reproducibility of the anchor sampling process.
[Tables 2-3] Table 2 and Table 3: While gains are shown, the manuscript would benefit from reporting standard deviations across multiple runs to quantify variability in the accuracy and calibration improvements.
[§4.3] §4.3: The ablation on MGMC levels is informative, but the interaction between the number of levels and dataset characteristics could be discussed more explicitly to clarify generalizability.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our manuscript and the recommendation for minor revision. The summary accurately reflects the core contributions of TranCLR, including the Action Transitional Anchor Construction (ATAC) for modeling geometric transitional states and the Multi-Level Geometric Manifold Calibration (MGMC) for adaptive manifold calibration across continuity levels. We appreciate the recognition that these elements yield smoother, more uncertainty-aware representations compared to binary contrastive baselines, supported by results on NTU RGB+D, NTU RGB+D 120, and PKU-MMD, along with ablations and code release. Since the report lists no specific major comments, we have no individual points requiring detailed rebuttal or changes at this stage.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper proposes TranCLR as a direct extension of standard contrastive learning via two explicitly constructed components: ATAC (Action Transitional Anchor Construction) to model transitional states in pose sequences, and MGMC (Multi-Level Geometric Manifold Calibration) to adaptively adjust the action manifold. These are introduced as novel mechanisms without any equations, fitted parameters, or predictions that reduce by construction to the inputs or to prior self-citations. Validation relies on independent experiments across NTU RGB+D, NTU RGB+D 120, and PKU-MMD datasets showing measurable gains in accuracy and calibration metrics over binary baselines, making the central claims externally falsifiable rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Only the abstract is available, limiting visibility into any fitted hyperparameters or detailed assumptions; the ledger reflects high-level claims from the provided text.

axioms (1)

domain assumption Human motion possesses intrinsic continuity that binary contrastive objectives overlook, resulting in fragmented feature clusters.
Directly stated in the abstract as the core motivation for the work.

invented entities (1)

Action Transitional Anchor no independent evidence
purpose: To explicitly model the geometric structure of transitional states in the action space.
Introduced as a new construction within the framework; no independent external validation is described in the abstract.

pith-pipeline@v0.9.0 · 5490 in / 1236 out tokens · 50596 ms · 2026-05-10T04:28:35.543039+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

77 extracted references · 8 canonical work pages · 3 internal anchors

[1]

S-jepa: A joint embedding predictive architecture for skeletal action recog- nition

Mohamed Abdelfattah and Alexandre Alahi. S-jepa: A joint embedding predictive architecture for skeletal action recog- nition. InECCV, 2024. 7

2024
[2]

Maskclr: Attention-guided contrastive learning for robust action representation learning

Mohamed Abdelfattah, Mariam Hassan, and Alexandre Alahi. Maskclr: Attention-guided contrastive learning for robust action representation learning. InCVPR, 2024. 1

2024
[3]

A simple framework for contrastive learning of visual representations

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations. 2020. 3

2020
[4]

Exploring simple siamese rep- resentation learning

Xinlei Chen and Kaiming He. Exploring simple siamese rep- resentation learning. InCVPR, 2021. 3

2021
[5]

Improved Baselines with Momentum Contrastive Learning

Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297, 2020. 3

work page internal anchor Pith review arXiv 2003
[6]

Channel-wise topology refinement graph convolution for skeleton-based action recognition

Yuxin Chen, Ziqi Zhang, Chunfeng Yuan, Bing Li, Ying Deng, and Weiming Hu. Channel-wise topology refinement graph convolution for skeleton-based action recognition. In ICCV, 2021. 3

2021
[7]

Neu- ron: Learning context-aware evolving representations for zero-shot skeleton action recognition

Yang Chen, Jingcai Guo, Song Guo, and Dacheng Tao. Neu- ron: Learning context-aware evolving representations for zero-shot skeleton action recognition. InCVPR, 2024. 1

2024
[8]

Contrastive learning from spatio- temporal mixed skeleton sequences for self-supervised skeleton-based action recognition.arXiv:2207.03065, 2022

Zhan Chen, Hong Liu, Tianyu Guo, Zhengyan Chen, Pin- hao Song, and Hao Tang. Contrastive learning from spatio- temporal mixed skeleton sequences for self-supervised skeleton-based action recognition.arXiv:2207.03065, 2022. 1, 3

work page arXiv 2022
[9]

Skeleton-based action recognition with shift graph convolutional network

Ke Cheng, Yifan Zhang, Xiangyu He, Weihan Chen, Jian Cheng, and Hanqing Lu. Skeleton-based action recognition with shift graph convolutional network. InCVPR, 2020. 3

2020
[10]

Re- visiting the evaluation of uncertainty estimation and its ap- plication to explore model complexity-uncertainty trade-off

Yukun Ding, Jinglan Liu, Jinjun Xiong, and Yiyu Shi. Re- visiting the evaluation of uncertainty estimation and its ap- plication to explore model complexity-uncertainty trade-off. InCVPRW, 2020. 6

2020
[11]

Hierarchical contrast for un- supervised skeleton-based action representation learning

Jianfeng Dong, Shengkai Sun, Zhonglin Liu, Shujie Chen, Baolong Liu, and Xun Wang. Hierarchical contrast for un- supervised skeleton-based action representation learning. In AAAI, 2023. 7

2023
[12]

Hierarchical recur- rent neural network for skeleton based action recognition

Yong Du, Wei Wang, and Liang Wang. Hierarchical recur- rent neural network for skeleton based action recognition. In CVPR, 2015. 3

2015
[13]

Skeleton-contrastive 3d action representation learning

Hazel Doughty Fida Mohammad Thoker and Cees Snoek. Skeleton-contrastive 3d action representation learning. In ACM MM, 2021. 7

2021
[14]

Hyperbolic self-paced learning for self-supervised skeleton-based action representations

Luca Franco, Paolo Mandica, Bharti Munjal, and Fabio Galasso. Hyperbolic self-paced learning for self-supervised skeleton-based action representations. InICLR, 2023. 7

2023
[15]

Rethinking masked data reconstruction pretraining for strong 3d action representation learning

Tao Gong, Qi Chu, Bin Liu, and Nenghai Yu. Rethinking masked data reconstruction pretraining for strong 3d action representation learning. InAAAI, 2025. 3, 7

2025
[16]

Bootstrap your own latent: A new approach to self-supervised learning

Jean Bastien Grill, Florian Strub, Florent Altch ´e, Corentin Tallec, Pierre H Richemond, Elena Buchatskaya, Carl Do- ersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Moham- mad Gheshlaghi Azar, et al. Bootstrap your own latent: A new approach to self-supervised learning. InNeurIPS, 2020. 3

2020
[17]

On calibration of modern neural networks

Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q Weinberger. On calibration of modern neural networks. 2017. 6

2017
[18]

Contrastive learning from ex- tremely augmented skeleton sequences for self-supervised action recognition.AAAI, 2022

Tianyu Guo, Hong Liu, Zhan Chen, Mengyuan Liu, Tao Wang, and Runwei Ding. Contrastive learning from ex- tremely augmented skeleton sequences for self-supervised action recognition.AAAI, 2022. 1, 2, 3, 6, 7, 8

2022
[19]

Momentum contrast for unsupervised visual rep- resentation learning

Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. InCVPR, 2020. 3

2020
[20]

Global and local contrastive learning for self-supervised skeleton-based action recognition.IEEE TCSVT, 2024

Jinhua Hu, Yonghong Hou, Zihui Guo, and Jiajun Gao. Global and local contrastive learning for self-supervised skeleton-based action recognition.IEEE TCSVT, 2024. 3, 7

2024
[21]

Part aware contrastive learning for self-supervised action recognition

Yilei Hua, Wenhan Wu, Ce Zheng, Aidong Lu, Mengyuan Liu, Chen Chen, and Shiqian Wu. Part aware contrastive learning for self-supervised action recognition. InIJCAI,
[22]

Pastd: Progressive augmentation and spa- tiotemporal decoupling contrastive learning for skeleton- based action recognition

Qian Huang, Weiwen Qian, Chang Li, Gongyou Xu, and Zhongqi Chen. Pastd: Progressive augmentation and spa- tiotemporal decoupling contrastive learning for skeleton- based action recognition. InICASSP, 2025. 7

2025
[23]

A new representation of skeleton sequences for 3d action recognition

Qiuhong Ke, Mohammed Bennamoun, Senjian An, Ferdous Sohel, and Farid Boussaid. A new representation of skeleton sequences for 3d action recognition. InCVPR, 2017. 3

2017
[24]

Lisa: Reasoning segmentation via large language model

Xin Lai, Zhuotao Tian, Yukang Chen, Yanwei Li, Yuhui Yuan, Shu Liu, and Jiaya Jia. Lisa: Reasoning segmentation via large language model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9579–9589, 2024. 1

2024
[25]

3D human action rep- resentation learning via cross-view consistency pursuit

Linguo Li, Minsi Wang, Bingbing Ni, Hang Wang, Jiancheng Yang, and Wenjun Zhang. 3D human action rep- resentation learning via cross-view consistency pursuit. In CVPR, 2021. 3, 6, 7

2021
[26]

Perception, reason, think, and plan: A survey on large multimodal reasoning models.arXiv preprint arXiv:2505.04921, 2025

Yunxin Li, Zhenyu Liu, Zitao Li, Xuanyu Zhang, Zhenran Xu, Xinyu Chen, Haoyuan Shi, Shenyuan Jiang, Xintong Wang, Jifang Wang, et al. Perception, reason, think, and plan: A survey on large multimodal reasoning models.arXiv preprint arXiv:2505.04921, 2025. 1

work page arXiv 2025
[27]

Actionlet- dependent contrastive learning for unsupervised skeleton- based action recognition

Lilang Lin, Jiahang Zhang, and Jiaying Liu. Actionlet- dependent contrastive learning for unsupervised skeleton- based action recognition. InCVPR, 2023. 1, 2, 3, 5, 6, 7, 8

2023
[28]

Self-supervised skeleton representation learning via actionlet contrast and re- construct

Lilang Lin, Jiahang Zhang, and Jiaying Liu. Self-supervised skeleton representation learning via actionlet contrast and re- construct. 2025. 1, 7

2025
[29]

Skeleton-cutmix: Mixing up skeleton with probabilistic bone exchange for supervised domain adapta- tion.IEEE TIP, 2023

Hanchao Liu, Yuhe Liu, Tai-Jiang Mu, Xiaolei Huang, and Shi-Min Hu. Skeleton-cutmix: Mixing up skeleton with probabilistic bone exchange for supervised domain adapta- tion.IEEE TIP, 2023. 4

2023
[30]

Recovering complete actions for cross-dataset skeleton ac- tion recognition

Hanchao Liu, Yujiang Li, Tai-Jiang Mu, and Shi-Min Hu. Recovering complete actions for cross-dataset skeleton ac- tion recognition. InNeurIPS, 2024. 1

2024
[31]

Revealing key details to see differ- ences: A novel prototypical perspective for skeleton-based action recognition

Hongda Liu, Yunfan Liu, Min Ren, Hao Wang, Yunlong Wang, and Zhenan Sun. Revealing key details to see differ- ences: A novel prototypical perspective for skeleton-based action recognition. InCVPR, 2025. 3

2025
[32]

Jun Liu, Gang Wang, Ping Hu, Ling-Yu Duan, and Alex C. Kot. Global context-aware attention lstm networks for 3d action recognition. InCVPR, 2017. 3

2017
[33]

Ntu rgb+d 120: A large- scale benchmark for 3d human activity understanding

Jun Liu, Amir Shahroudy, Mauricio Perez, Gang Wang, Ling-Yu Duan, and Alex C Kot. Ntu rgb+d 120: A large- scale benchmark for 3d human activity understanding. 2020. 2

2020
[34]

A benchmark dataset and comparison study for multi-modal human action analytics.ACM MM, 2020

Jiaying Liu, Sijie Song, Chunhui Liu, Yanghao Li, and Yueyu Hu. A benchmark dataset and comparison study for multi-modal human action analytics.ACM MM, 2020. 2

2020
[35]

Enhanced skele- ton visualization for view invariant human action recogni- tion.PR, 2017

Mengyuan Liu, Hong Liu, and Chen Chen. Enhanced skele- ton visualization for view invariant human action recogni- tion.PR, 2017. 3

2017
[36]

Cmd: Self-supervised 3d action representa- tion learning with cross-modal mutual distillation

Yunyao Mao, Wengang Zhou, Zhenbo Lu, Jiajun Deng, and Houqiang Li. Cmd: Self-supervised 3d action representa- tion learning with cross-modal mutual distillation. InECCV,
[37]

Masked motion predictors are strong 3d action representation learners

Yunyao Mao, Jiajun Deng, Wengang Zhou, Yao Fang, Wanli Ouyang, and Houqiang Li. Masked motion predictors are strong 3d action representation learners. InICCV, 2023. 7

2023
[38]

I 2md: 3d action repre- sentation learning with inter- and intra-modal mutual distil- lation.IJCV, 2024

Yunyao Mao, Jiajun Deng, Wengang Zhou, Zhenbo Lu, Wanli Ouyang, and Houqiang Li. I 2md: 3d action repre- sentation learning with inter- and intra-modal mutual distil- lation.IJCV, 2024. 7

2024
[39]

Stars: Self-supervised tuning for 3d action recognition in skeleton sequences.arXiv:2407.10935,

Soroush Mehraban, Mohammad Javad Rajabi, Andrea Iaboni, and Babak Taati. Stars: Self-supervised tuning for 3d action recognition in skeleton sequences.arXiv:2407.10935,

work page arXiv
[40]

Cooper, and Milos Hauskrecht

Mahdi Pakdaman Naeini, Gregory F. Cooper, and Milos Hauskrecht. Obtaining well calibrated probabilities using bayesian binning. InAAAI, 2015. 6

2015
[41]

Boosting few-shot 3d point cloud segmentation via query-guided enhancement

Zhenhua Ning, Zhuotao Tian, Guangming Lu, and Wenjie Pei. Boosting few-shot 3d point cloud segmentation via query-guided enhancement. InProceedings of the 31st ACM international conference on multimedia, pages 1895–1904,

1904
[42]

Representation Learning with Contrastive Predictive Coding

Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Rep- resentation learning with contrastive predictive coding. arXiv:1807.03748, 2018. 3, 5

work page internal anchor Pith review Pith/arXiv arXiv 2018
[43]

Oa-cnns: Omni- adaptive sparse cnns for 3d semantic segmentation

Bohao Peng, Xiaoyang Wu, Li Jiang, Yukang Chen, Heng- shuang Zhao, Zhuotao Tian, and Jiaya Jia. Oa-cnns: Omni- adaptive sparse cnns for 3d semantic segmentation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21305–21315, 2024. 1

2024
[44]

Skeleton-based action recognition via spatial and temporal transformer networks.CVIU, 2021

Chiara Plizzari, Marco Cannici, and Matteo Matteucci. Skeleton-based action recognition via spatial and temporal transformer networks.CVIU, 2021. 3

2021
[45]

Llms are good action recognizers

Haoxuan Qu, Yujun Cai, and Jun Liu. Llms are good action recognizers. InCVPR, 2024. 1

2024
[46]

Halp: Hallucinating latent positives for skeleton-based self- supervised learning of actions

Anshul Shah, Aniket Roy, Ketul Shah, Shlok Kumar Mishra, David Jacobs, Anoop Cherian, and Rama Chellappa. Halp: Hallucinating latent positives for skeleton-based self- supervised learning of actions. InCVPR, 2023. 7

2023
[47]

Ntu rgb+d: A large scale dataset for 3d human activity anal- ysis

Amir Shahroudy, Jun Liu, Tian-Tsong Ng, and Gang Wang. Ntu rgb+d: A large scale dataset for 3d human activity anal- ysis. InCVPR, 2016. 2

2016
[48]

Ex- plore the potential of clip for training-free open vocabulary semantic segmentation

Tong Shao, Zhuotao Tian, Hang Zhao, and Jingyong Su. Ex- plore the potential of clip for training-free open vocabulary semantic segmentation. InEuropean Conference on Com- puter Vision, pages 139–156. Springer, 2024. 1

2024
[49]

Two- stream adaptive graph convolutional networks for skeleton- based action recognition

Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Two- stream adaptive graph convolutional networks for skeleton- based action recognition. InCVPR, 2019. 3

2019
[50]

Decou- pled spatial-temporal attention network for skeleton-based action-gesture recognition

Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Decou- pled spatial-temporal attention network for skeleton-based action-gesture recognition. InACCV, 2020. 3

2020
[51]

Uni- fied multi-modal unsupervised representation learning for skeleton-based action understanding

Shengkai Sun, Daizong Liu, Jianfeng Dong, Xiaoye Qu, Junyu Gao, Xun Yang, Xun Wang, and Meng Wang. Uni- fied multi-modal unsupervised representation learning for skeleton-based action understanding. InACM MM, 2023. 7

2023
[52]

Towards efficient general feature prediction in masked skeleton modeling

Shengkai Sun, Zefan Zhang, Jianfeng Dong, Zhiyong Cheng, Xiaojun Chang, and Meng Wang. Towards efficient general feature prediction in masked skeleton modeling. In ICCV, 2025. 7

2025
[53]

Adaptive perspective distillation for semantic segmentation

Zhuotao Tian, Pengguang Chen, Xin Lai, Li Jiang, Shu Liu, Hengshuang Zhao, Bei Yu, Ming-Chang Yang, and Jiaya Jia. Adaptive perspective distillation for semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 45(2):1372–1387, 2022. 1

2022
[54]

Generalized few-shot se- mantic segmentation

Zhuotao Tian, Xin Lai, Li Jiang, Shu Liu, Michelle Shu, Hengshuang Zhao, and Jiaya Jia. Generalized few-shot se- mantic segmentation. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 11563–11572, 2022

2022
[55]

Learning context-aware classifier for semantic segmentation

Zhuotao Tian, Jiequan Cui, Li Jiang, Xiaojuan Qi, Xin Lai, Yixin Chen, Shu Liu, and Jiaya Jia. Learning context-aware classifier for semantic segmentation. InProceedings of the AAAI conference on artificial intelligence, pages 2438–2446,
[56]

Groupcontrast: Semantic-aware self-supervised representation learning for 3d understanding

Chengyao Wang, Li Jiang, Xiaoyang Wu, Zhuotao Tian, Bo- hao Peng, Hengshuang Zhao, and Jiaya Jia. Groupcontrast: Semantic-aware self-supervised representation learning for 3d understanding. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 4917–4928, 2024. 1

2024
[57]

Heterogeneous skeleton-based action representation learn- ing

Hongsong Wang, Xiaoyan Ma, Jidong Kuang, and Jie Gui. Heterogeneous skeleton-based action representation learn- ing. InCVPR, 2025. 7

2025
[58]

Foundation model for skeleton-based human action understanding

Hongsong Wang, Wanjiang Weng, Junbo Wang, Fang Zhao, Guo-Sen Xie, Xin Geng, and Liang Wang. Foundation model for skeleton-based human action understanding. 2025. 1

2025
[59]

Declip: Decoupled learning for open- vocabulary dense perception

Junjie Wang, Bin Chen, Yulin Li, Bin Kang, Yichi Chen, and Zhuotao Tian. Declip: Decoupled learning for open- vocabulary dense perception. InProceedings of the Com- puter Vision and Pattern Recognition Conference, pages 14824–14834, 2025. 1

2025
[60]

Generalized decou- pled learning for enhancing open-vocabulary dense percep- tion.arXiv preprint arXiv:2508.11256, 2025

Junjie Wang, Keyu Chen, Yulin Li, Bin Chen, Hengshuang Zhao, Xiaojuan Qi, and Zhuotao Tian. Generalized decou- pled learning for enhancing open-vocabulary dense percep- tion.arXiv preprint arXiv:2508.11256, 2025. 1

work page arXiv 2025
[61]

Skeleton-in-context: Unified skeleton sequence modeling with in-context learning

Xinshun Wang, Zhongbin Fang, Xia Li, Xiangtai Li, Chen Chen, and Mengyuan Liu. Skeleton-in-context: Unified skeleton sequence modeling with in-context learning. In CVPR, 2024. 1

2024
[62]

Usdrl: Unified skeleton-based dense represen- tation learning with multi-grained feature decorrelation

Wanjiang Weng, Hongsong Wang, Junbo Wang, Lei He, and Guosen Xie. Usdrl: Unified skeleton-based dense represen- tation learning with multi-grained feature decorrelation. In AAAI, 2025. 7

2025
[63]

Macdiff: Unified skeleton modeling with masked conditional diffusion

Lehong Wu, Lilang Lin, Jiahang Zhang, Yiyang Ma, and Ji- aying Liu. Macdiff: Unified skeleton modeling with masked conditional diffusion. InECCV, 2024. 7

2024
[64]

Skeletonmae: Spatial-temporal masked au- toencoders for self-supervised skeleton action recognition

Wenhan Wu, Yilei Hua, Ce Zheng, Shiqian Wu, Chen Chen, and Aidong Lu. Skeletonmae: Spatial-temporal masked au- toencoders for self-supervised skeleton action recognition
[65]

Towards large- scale 3d representation learning with multi-dataset point prompt training

Xiaoyang Wu, Zhuotao Tian, Xin Wen, Bohao Peng, Xihui Liu, Kaicheng Yu, and Hengshuang Zhao. Towards large- scale 3d representation learning with multi-dataset point prompt training. InCVPR, 2024. 1

2024
[66]

Attack-augmentation mixing-contrastive skeletal representation learning

Binqian Xu, Xiangbo Shu, Jiachao Zhang, Rui Yan, and Guo-Sen Xie. Attack-augmentation mixing-contrastive skeletal representation learning. 2024. 7

2024
[67]

Spatial tempo- ral graph convolutional networks for skeleton-based action recognition

Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial tempo- ral graph convolutional networks for skeleton-based action recognition. InAAAI, 2018. 3

2018
[68]

Unified language-driven zero-shot domain adaptation

Senqiao Yang, Zhuotao Tian, Li Jiang, and Jiaya Jia. Unified language-driven zero-shot domain adaptation. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 23407–23415, 2024. 1

2024
[69]

mixup: Beyond Empirical Risk Minimization

Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion.arXiv:1710.09412, 2017. 3

work page internal anchor Pith review arXiv 2017
[70]

Hierarchi- cal consistent contrastive learning for skeleton-based action recognition with growing augmentations

Jiahang Zhang, Lilang Lin, and Jiaying Liu. Hierarchi- cal consistent contrastive learning for skeleton-based action recognition with growing augmentations. InAAAI, 2023. 1, 3

2023
[71]

Prompted con- trast with masked motion modeling: Towards versatile 3d action representation learning

Jiahang Zhang, Lilang Lin, and Jiaying Liu. Prompted con- trast with masked motion modeling: Towards versatile 3d action representation learning. InACM MM, 2023. 7

2023
[72]

Shap-mix: Shapley value guided mixing for long-tailed skeleton based action recognition

Jiahang Zhang, Lilang Lin, and Jiaying Liu. Shap-mix: Shapley value guided mixing for long-tailed skeleton based action recognition. InIJCAI, 2024. 4

2024
[73]

Self-supervised skeleton-based action representation learn- ing: A benchmark and beyond.arXiv:2406.02978, 2024

Jiahang Zhang, Lilang Lin, Shuai Yang, and Jiaying Liu. Self-supervised skeleton-based action representation learn- ing: A benchmark and beyond.arXiv:2406.02978, 2024. 1

work page arXiv 2024
[74]

Con- certo: Joint 2d-3d self-supervised learning emerges spatial representations

Yujia Zhang, Xiaoyang Wu, Yixing Lao, Chengyao Wang, Zhuotao Tian, Naiyan Wang, and Hengshuang Zhao. Con- certo: Joint 2d-3d self-supervised learning emerges spatial representations. InNeurIPS, 2025. 1

2025
[75]

Self-supervised action representation learning from partial spatio-temporal skeleton sequences

Yujie Zhou, Haodong Duan, Anyi Rao, Bing Su, and Jiaqi Wang. Self-supervised action representation learning from partial spatio-temporal skeleton sequences. InAAAI, 2023. 7

2023
[76]

Blockgcn: Redefining topology awareness for skeleton-based action recognition

Yuxuan Zhou, Xudong Yan, Zhi-Qi Cheng, Yan Yan, Qi Dai, and Xian-Sheng Hua. Blockgcn: Redefining topology awareness for skeleton-based action recognition. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024. 3

2024
[77]

Semantic-guided cross-modal prompt learning for skeleton-based zero-shot action recognition

Anqi Zhu, Jingmin Zhu, James Bailey, Mingming Gong, and Qiuhong Ke. Semantic-guided cross-modal prompt learning for skeleton-based zero-shot action recognition. InCVPR,