Research map / Video / Action recognition
Parallel Attention Interaction Network for Few-Shot Skeleton-Based Action Recognition
Xingyu Liu, Sanping Zhou, Le Wang, Gang Hua
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 |
|---|---|---|---|
| replace DASTM with PAINet's parallel attention branches | few-shot skeleton action recognition · NTU-T 5-way 1-shot, ST-GCN | accuracy (%): 75.1 -> 82.4 | Improved |
| replace DASTM with PAINet | few-shot skeleton action recognition · Kinetics 5-way 1-shot, ST-GCN | accuracy (%): 39.3 -> 42.4 | Improved |
| add Temporal-Channel Mixer to the Cross Spatial Alignment branch (ablation) | PAINet · NTU-T 1-shot, ST-GCN | accuracy (%): 80.1 -> 79.3 | Hurt |
| add Temporal-Channel Mixer to the Cross Temporal Alignment branch (ablation) | PAINet · NTU-T 1-shot, ST-GCN | accuracy (%): 78.9 -> 80.8 | Improved |
| add spatial self-attention to the temporal alignment branch (ablation) | PAINet · NTU-T 1-shot, ST-GCN | accuracy (%): 80.8 -> 81.8 | Improved |
| merge the two parallel branches (ablation) | PAINet · NTU-T 1-shot, ST-GCN | accuracy (%): 81.8 -> 82.4 | Improved |
| add joint-part and joint-semantic relations to spatial self-attention (ablation) | spatial aggregation · NTU-T 1-shot, ST-GCN | accuracy (%): 81.9 -> 82.4 | Improved |
| use random masking in spatial self-attention (ablation) | spatial aggregation · NTU-T 1-shot, ST-GCN | accuracy (%): 81.9 -> 81.6 | Hurt |
| use channel-specific attention instead of self-attention (ablation) | spatial aggregation · NTU-T 1-shot, ST-GCN | accuracy (%): 81.9 -> 80.8 | Hurt |
| use directional average symmetric surface metric on frame pairs instead of Hungarian matching (ablation) | temporal set metric · NTU-T 1-shot, ST-GCN | accuracy (%): 80.5 -> 82.4 | Improved |
| use pair-based instead of frame-based subsequences with DASSM (ablation) | temporal set metric · NTU-T 1-shot, ST-GCN | accuracy (%): 81.3 -> 82.4 | Improved |
| use complex MS-G3D backbone instead of ST-GCN | PAINet · NTU-T 1-shot | accuracy (%): 82.4 -> 81.3 | No clear effect |
Closest papers by meaning
- Towards To-a-T Spatio-Temporal Focus for Skeleton-Based Action Recognition (AAAI 2022 · similarity 0.90)
- InfoGCN: Representation Learning for Human Skeleton-Based Action Recognition (CVPR 2022 · similarity 0.90)
- Learning Multi-View Interactional Skeleton Graph for Action Recognition (TPAMI 2020 · similarity 0.90)
- An Attention Enhanced Graph Convolutional LSTM Network for Skeleton-Based Action Recognition (CVPR 2019 · similarity 0.90)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.89)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · 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.
