Research map / Graph learning / Knowledge and temporal graphs
Constructing Stronger and Faster Baselines for Skeleton-Based Action Recognition
Yifan Song, Zhang Zhang, Caifeng Shan, Liang 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 |
|---|---|---|---|
| add ST-JointAtt attention module (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 88.9 -> 90.0 | Improved |
| use ST-JointAtt instead of STCAtt (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 89.5 -> 90.0 | Improved |
| use ST-JointAtt instead of PartAtt (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 89.4 -> 90.0 | Improved |
| add frame-only attention (FrameAtt) instead of none (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 88.9 -> 88.5 | Hurt |
| add velocity and bone inputs to joint input (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 87.7 -> 90.0 | Improved |
| fuse input branches after 2nd stage instead of before 1st stage (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 88.8 -> 90.0 | Improved |
| fuse after 2nd stage instead of at the score layer (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 89.9 -> 90.0; FLOPs 3.85 -> 2.73 | No clear effect |
| use SGLayer (r_rd=2) instead of BasicLayer for temporal convolution (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 90.0 -> 90.0; params 0.34M -> 0.29M | No clear effect |
| use EpSepLayer (r_ep=4) instead of SGLayer (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 90.0 -> 90.1; FLOPs 2.73 -> 3.63 | No clear effect |
| apply compound scaling (coefficient 0 to 4) (ablation) | EfficientGCN · NTU 60 X-sub | accuracy (%): 90.2 -> 92.1 | Improved |
| scale only width or only depth instead of compound scaling (ablation) | EfficientGCN-B4 · NTU 60 X-sub | accuracy (%) | Hurt |
| raise maximum graph distance D from 1 to 2 (L=5) (ablation) | EfficientGCN-B0 · NTU 60 X-sub | accuracy (%): 88.1 -> 89.9 | Improved |
Closest papers by meaning
- Actional-Structural Graph Convolutional Networks for Skeleton-Based Action Recognition (CVPR 2019 · similarity 0.92)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.92)
- Two-Stream Adaptive Graph Convolutional Networks for Skeleton-Based Action Recognition (CVPR 2019 · similarity 0.92)
- Learning Graph Convolutional Network for Skeleton-Based Human Action Recognition by Neural Searching (AAAI 2020 · similarity 0.92)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.90)
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · similarity 0.90)
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.
