Research map / Graph learning / Knowledge and temporal graphs
Gamba: Mamba-based graph convolutional network with dynamic graph topology learning for action recognition
Rouyi Zhou, Yangzhi Wu, Jiajun Wen, Can Gao, Feng Liu, Zhihui Lai, Linlin Shen
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 the Cls-Mamba module to the baseline (ablation) | CTR-GC + Naive TCN baseline · NTU RGB+D 60 X-View (joint stream) | accuracy (%): 95.8 -> 96.0 | Improved |
| add temporal Mamba (T-Mamba) alone (ablation) | CTR-GC + Naive TCN baseline · NTU RGB+D 60 X-View (joint stream) | accuracy (%) | No clear effect |
| add node classification module (NCM) before spatial Mamba (ablation) | Cls-Mamba · NTU RGB+D 60 X-View (joint stream) | accuracy (%): +0.8 | Improved |
| use all modules together (NCM, S-Mamba, T-Mamba, TCN) (ablation) | Gamba · NTU RGB+D 60 X-View (joint stream) | accuracy (%): 95.8 -> 96.5 | Improved |
| insert Mamba after every GCN layer (9) instead of 3 or 6 (ablation) | Gamba · NTU RGB+D 60 X-View (joint stream) | accuracy (%): 95.4 (3) -> 95.6 (6) -> 96.5 (9) | Improved |
| use TCN kernel size 5 instead of 3 or 7 (ablation) | Mamba-TCN · NTU RGB+D 60 X-View (joint stream) | accuracy (%): 95.6 (k=3), 96.5 (k=5), 95.7 (k=7) | Improved |
| use 64 node categories instead of 32 or 128 (ablation) | NCM · NTU RGB+D 60 X-View (joint stream) | accuracy (%): 96.1 (32), 96.5 (64), 96.0 (128) | Improved |
| use 64 node categories instead of 32 or 128 (ablation) | NCM · NTU RGB+D 120 X-Sub (joint stream) | accuracy (%): 86.0 (32), 86.3 (64), 85.5 (128) | Improved |
| replace CTR-GCN with Gamba in the 4-stream ensemble | CTR-GCN · NTU RGB+D 60 X-Sub | top-1 accuracy (%): 92.4 -> 93.4 | Improved |
| replace BlockGCN with Gamba on NTU RGB+D 120 X-Sub | BlockGCN · NTU RGB+D 120 X-Sub | top-1 accuracy (%): 90.3 -> 90.1 | Hurt |
| use Gamba instead of CTR-GCN for computation | CTR-GCN · Model cost | FLOPs (G): 14.4 -> 11.1 | Improved |
| use Gamba instead of CTR-GCN for parameter count | CTR-GCN · Model cost | trainable parameters (M): 1.2 -> 4.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BlockGCN — BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-Scale Spatial Temporal Graph Convolutional Network for Skeleton-Based Action Recognition (AAAI 2021 · similarity 0.87)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.87)
- Channel-Wise Topology Refinement Graph Convolution for Skeleton-Based Action Recognition (ICCV 2021 · similarity 0.87)
- Spatio-Temporal Fusion for Human Action Recognition via Joint Trajectory Graph (AAAI 2024 · similarity 0.87)
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · similarity 0.86)
- DG-Mamba: Robust and Efficient Dynamic Graph Structure Learning with Selective State Space Models (AAAI 2025 · similarity 0.86)
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.
