Research map / Video / Action recognition
Few-shot Action Recognition with Permutation-invariant Attention
Hongguang Zhang, Li Zhang, Xiaojuan Qi, Hongdong Li, Philip H. S. Torr, Piotr Koniusz
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 spatial attention (ablation) | C3D SoSN baseline · HMDB51 (proposed split), 5-way 1-shot | accuracy: 40.83 -> 41.27 | Improved |
| add temporal attention (ablation) | C3D SoSN baseline · HMDB51 (proposed split), 5-way 5-shot | accuracy: 55.18 -> 57.67 | Improved |
| add spatial jigsaw self-supervision (ablation) | C3D SoSN baseline · HMDB51 (proposed split), 5-way 1-shot | accuracy: 40.83 -> 44.19 | Improved |
| add self-supervision to attention model (ablation) | ARN with spatial attention · HMDB51 (proposed split), 5-way 1-shot | accuracy: 41.27 -> 44.61 | Improved |
| replace naive attention with attention by alignment (ablation) | ARN with spatial attention and self-supervision · HMDB51 (proposed split), 5-way 1-shot | accuracy: 44.61 -> 45.17 | Improved |
| add alignment loss without self-supervision (ablation) | ARN with spatial attention · HMDB51 (proposed split), 5-way 1-shot | accuracy: 41.27 -> 43.11 | Improved |
| combine attention, self-supervision and alignment (ablation) | C3D SoSN baseline · HMDB51 (proposed split), 5-way 1-shot | accuracy: 40.83 -> 45.17 | Improved |
| use second-order pooling with power normalization instead of average pooling (ablation) | ARN · HMDB51 (proposed split), pooling, 5-way 1-shot | accuracy: 39.51 -> 40.83 | Improved |
| add power normalization to average pooling (ablation) | ARN · HMDB51 (proposed split), pooling, 5-way 1-shot | accuracy: 39.51 -> 40.02 | Improved |
| add power normalization to second-order pooling (ablation) | ARN · HMDB51 (proposed split), pooling, 5-way 1-shot | accuracy: 39.97 -> 40.83 | Improved |
| add pooling over blocks instead of none (ablation) | ARN · HMDB51 (proposed split), pooling, 5-way 1-shot | accuracy: 35.71 -> 40.83 | Improved |
| swap spatial jigsaw for rotation self-supervision with alignment (ablation) | ARN+SSA · HMDB51 (proposed split), 5-way 5-shot | accuracy: 60.56 -> 58.96 | Hurt |
Closest papers by meaning
- Attentional Pooling for Action Recognition (NeurIPS 2017 · similarity 0.88)
- Temporal-Relational CrossTransformers for Few-Shot Action Recognition (CVPR 2021 · similarity 0.88)
- A Comprehensive Review of Few-Shot Action Recognition (IJCV 2025 · similarity 0.88)
- Video Representation Learning Using Discriminative Pooling (CVPR 2018 · similarity 0.88)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.87)
- Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · 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.
