Research map / Label-efficient and robust learning / Recognition and long-tail
Hierarchical Contrast for Unsupervised Skeleton-Based Action Representation Learning
Jianfeng Dong, Shengkai Sun, Zhonglin Liu, Shujie Chen, Baolong Liu, Xun Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use hierarchical contrast across instance, domain, clip, and part levels instead of instance-only contrast (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 79.8 -> 81.1 | Improved |
| add domain-level contrast to instance-level contrast (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 79.8 -> 80.7 | Improved |
| add clip/part-level contrast to instance and domain contrast (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 80.7 -> 81.1 | Improved |
| jointly use temporal and spatial branches rather than either branch alone (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 78.0 -> 79.8; 76.1 -> 79.8 | Improved |
| replace max-pooling with mean-pooling in UDM (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 81.1 -> 80.8 | No clear effect |
| include Conv1d in UDM rather than remove it (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 79.7 -> 81.1 | Improved |
| use concatenation rather than element-wise sum for branch fusion (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 80.6 -> 81.1 | Improved |
| use concatenation rather than Hadamard product for branch fusion (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 79.8 -> 81.1 | Improved |
| use concatenation rather than weighted sum for branch fusion (ablation) | HiCo-Transformer · NTU-60 x-sub; action recognition | top-1 accuracy: 80.7 -> 81.1 | Improved |
| increase the number of temporal and spatial granularities (ablation) | HiCo · NTU-60 action recognition | top-1 accuracy | Improved |
| use HiCo rather than prior unsupervised methods | skeleton action recognition representation learning · PKU-MMD II x-sub | top-1 accuracy: 38.5* -> 54.7 | Improved |
| fine-tune HiCo-Transformer rather than the prior methods from NTU-60 | action representation transfer to PKU-MMD II · PKU-MMD II x-sub; source NTU-60 | top-1 accuracy: 45.9 -> 56.3 | Improved |
Closest papers by meaning
- Hierarchical Consistent Contrastive Learning for Skeleton-Based Action Recognition with Growing Augmentations (AAAI 2023 · similarity 0.93)
- Hierarchically Self-Supervised Transformer for Human Skeleton Representation Learning (ECCV 2022 · similarity 0.91)
- HaLP: Hallucinating Latent Positives for Skeleton-Based Self-Supervised Learning of Actions (CVPR 2023 · similarity 0.91)
- MaskCLR: Attention-Guided Contrastive Learning for Robust Action Representation Learning (CVPR 2024 · similarity 0.89)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (AAAI 2025 · similarity 0.89)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
