Research map / Video / Action recognition
AdaSGN: Adapting Joint Number and Model Size for Efficient Skeleton-Based Action Recognition
Lei Shi, Yifan Zhang, Jian Cheng, Hanqing Lu
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 pretraining of single models (ablation) | AdaSGN · NTU-60 CS | accuracy (%): 87.0 -> 89.1 | Improved |
| add pretraining of single models (ablation) | AdaSGN · NTU-60 CV | accuracy (%): 92.2 -> 94.7 | Improved |
| use learned policy instead of random action selection (ablation) | AdaSGN · NTU-60 CS | accuracy (%): 85.6 -> 89.1 | Improved |
| use learned policy instead of random action selection (ablation) | AdaSGN · NTU-60 CV | accuracy (%): 91.2 -> 94.6 | Improved |
| use policy instead of fusing all models | AdaSGN · NTU-60 CS | accuracy (%): 89.1 -> 89.3 | No clear effect |
| adaptively learn transform matrix for 25 joints (ablation) | SGN SM0 · NTU-60 CS | accuracy (%): 88.4 -> 88.9 | Improved |
| adaptively learn transform matrix for 25 joints (ablation) | SGN SM1 · NTU-60 CV | accuracy (%): 92.9 -> 93.3 | Improved |
| reduce joints from 25 to 9 (ablation) | SM0* · NTU-60 CS | accuracy (%): 88.9 -> 87.2 | Hurt |
| reduce joints from 25 to 1 (ablation) | SM0* · NTU-60 CS | accuracy (%): 88.9 -> 64.9 | Hurt |
| raise efficiency loss weight alpha from 0 to 4 (ablation) | AdaSGN · NTU-60 CS | accuracy (%): 88.8 -> 89.1 | Improved |
| raise efficiency loss weight alpha from 4 to 20 (ablation) | AdaSGN · NTU-60 CS | accuracy (%): 89.1 -> 85.9 | Hurt |
| apply AdaSGN adaptive joint and model selection to 1s-SGN | 1s-SGN · NTU-60 CS | accuracy (%): 88.4 -> 89.1 | Improved |
Closest papers by meaning
- Constructing Stronger and Faster Baselines for Skeleton-Based Action Recognition (TPAMI 2022 · similarity 0.90)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.89)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.87)
- Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · similarity 0.87)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.87)
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · 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.
