Research map / Graph learning / Knowledge and temporal graphs
3Mformer: Multi-Order Multi-Mode Transformer for Skeletal Action Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| model groups of joints with higher-order hyper-edges and fuse multi-order features using 3Mformer | skeletal action recognition · NTU-60, NTU-120, Kinetics-Skeleton, Northwestern-UCLA | accuracy | Helped |
| add the TP module (ablation) | MLP + HoTs backbone · NTU-60 X-Sub | accuracy: 89.8 -> 91.2 | Helped |
| add the MP module (ablation) | MLP + HoTs backbone · NTU-60 X-Sub | accuracy: 89.8 -> 92.0 | Helped |
| use MP→TP rather than TP→MP (ablation) | single-branch 3Mformer · NTU-60 X-Sub | accuracy: 92.6 -> 93.0 | Helped |
| combine MP→TP and TP→MP as two branches (ablation) | 3Mformer · NTU-60 X-Sub | accuracy: 93.0 -> 94.8 | Helped |
| increase the maximum hyper-edge order from 1 to 3 (ablation) | 3Mformer · Kinetics-Skeleton | Top-1 accuracy: 32.0 -> 39.5 | Helped |
| combine orders 3 and 4 instead of using order 4 alone (ablation) | 3Mformer · Kinetics-Skeleton | Top-1 accuracy: 40.1 -> 40.5 | Helped |
| increase hyper-edge order to 5 rather than 4 (ablation) | 3Mformer · Kinetics-Skeleton | Top-1 accuracy: 40.1 -> 40.3 | Helped |
| use rank pooling rather than average pooling (ablation) | 3Mformer temporal pooling · NTU-60 X-Sub | accuracy: 92.0 -> 94.8 | Helped |
| use rank pooling rather than maximum pooling (ablation) | 3Mformer temporal pooling · NTU-60 X-Sub | accuracy: 92.1 -> 94.8 | Helped |
| use learned weighted pooling instead of average pooling over hyper-edges (ablation) | 3Mformer MP module · NTU-60 X-Sub | accuracy: 91.3 -> 94.8 | Helped |
| use learned weighted pooling instead of max pooling over hyper-edges (ablation) | 3Mformer MP module · NTU-60 X-Sub | accuracy: 92.7 -> 94.8 | Helped |
Closest papers by meaning
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · similarity 0.90)
- Context Aware Graph Convolution for Skeleton-Based Action Recognition (CVPR 2020 · similarity 0.90)
- Graph CNNs with Motif and Variable Temporal Block for Skeleton-Based Action Recognition (AAAI 2019 · similarity 0.90)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.89)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.88)
- Spatio-Temporal Fusion for Human Action Recognition via Joint Trajectory Graph (AAAI 2024 · similarity 0.88)
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.