Research map / Label-efficient and robust learning / Self-supervised and contrastive
Contrastive Positive Mining for Unsupervised 3D Action Representation Learning
Haoyuan Zhang, Yonghong Hou, Wenjing Zhang, Wanqing Li
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 |
|---|---|---|---|
| add positive mining and positive-enhanced learning (ablation) | CPM · NTU-60 X-Sub | top-1 accuracy (%): 75.6 -> 78.7 | Improved |
| add positive mining and positive-enhanced learning (ablation) | CPM · NTU-60 X-View | top-1 accuracy (%): 81.7 -> 84.9 | Improved |
| add positive mining and positive-enhanced learning (ablation) | CPM · NTU-120 X-Sub | top-1 accuracy (%): 64.8 -> 68.7 | Improved |
| add positive mining and positive-enhanced learning (ablation) | CPM · NTU-120 X-Set | top-1 accuracy (%): 64.7 -> 69.6 | Improved |
| set number of mined positives K to 100 rather than too large or too small (ablation) | CPM · NTU-60 X-View | top-1 accuracy (%): best at K=100 | Mixed |
| use K=50 for PKU-MMD I and K=25 for PKU-MMD II (ablation) | CPM · PKU-MMD | top-1 accuracy (%): K=50; K=25 | Mixed |
| set target temperature tau' to 0.05 (ablation) | CPM (w/o. PM) · NTU-60 X-View | top-1 accuracy (%): optimal at 0.05 | Mixed |
| ensemble joint, bone and motion streams (3S) instead of joint only | CPM · NTU-60 X-Sub | top-1 accuracy (%): 78.7 -> 83.2 | Improved |
| ensemble joint, bone and motion streams (3S) instead of joint only | CPM · PKU-MMD II | top-1 accuracy (%): 48.3 -> 51.5 | Improved |
| fully fine-tune pre-trained encoder instead of linear evaluation | CPM · NTU-120 X-Sub | top-1 accuracy (%): 68.7 -> 78.4 | Improved |
Closest papers by meaning
- Contrastive Learning from Extremely Augmented Skeleton Sequences for Self-Supervised Action Recognition (AAAI 2022 · similarity 0.88)
- HaLP: Hallucinating Latent Positives for Skeleton-Based Self-Supervised Learning of Actions (CVPR 2023 · similarity 0.87)
- 3D Human Action Representation Learning via Cross-View Consistency Pursuit (CVPR 2021 · similarity 0.87)
- Hierarchical Consistent Contrastive Learning for Skeleton-Based Action Recognition with Growing Augmentations (AAAI 2023 · similarity 0.86)
- Hierarchical Contrast for Unsupervised Skeleton-Based Action Representation Learning (AAAI 2023 · similarity 0.86)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (AAAI 2025 · similarity 0.82)
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.
