Research map / Video / Action recognition
$$\text {A}^2\text {M}^2$$-Net: Adaptively Aligned Multi-scale Moment for Few-Shot Action Recognition
Zilin Gao, Qilong Wang, Bingbing Zhang, Qinghua Hu, Peihua Li
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 second-order Cov-MN pooling instead of first-order GAP (ablation) | GAP+EMD · SSV2-Full | 1-shot accuracy: 45.12 -> 46.80 | Improved |
| add third-order kernel pooling beyond second order (ablation) | second-order pooling · SSV2-Full | 1-shot accuracy: 46.80 -> 45.49 | No clear effect |
| add multi-scale temporal kernels (1,3,5) (ablation) | single-scale Cov-MN · SSV2-Full | 1-shot accuracy: 46.80 -> 56.08 | Improved |
| append temporal kernel 7 to (1,3,5) (ablation) | A2M2-Net · SSV2-Full | accuracy: -0.04 / +0.36 | No clear effect |
| customize spatial scales (1^2,3^2,5^2) per temporal scale (ablation) | A2M2-Net · SSV2-Full | accuracy: 56.08 -> 56.73 | Improved |
| use semantic spatial pooling with convolutional temporal pooling instead of fixed/average (ablation) | A2M2-Net · SSV2-Full | 1-shot accuracy: 50.93 -> 56.73 | Improved |
| add A2 adaptive alignment instead of fixed point-to-point alignment (ablation) | multi-scale M2 · SSV2-Full | 1-shot accuracy: 48.03 -> 56.73 | Improved |
| add A2 module in single-scale setting (ablation) | Cov-MN · SSV2-Full | 1-shot accuracy: 45.49 -> 46.80 | Improved |
| add multi-scale mechanism to second-order moment (M2 block) (ablation) | A2M2-Net · SSV2-Full | 1-shot accuracy: 46.80 -> 56.73 | Improved |
| add M2 block to OTAM metric | OTAM · SSV2-Full | 1-shot accuracy: 42.8 -> 50.82 | Improved |
| add M2 block to Bi-MHM metric | Bi-MHM · SSV2-Full | 1-shot accuracy: 44.6 -> 54.91 | Improved |
| add A2 module to STRM | STRM · SSV2-Full | 1-shot accuracy: 43.1 -> 48.7 | Improved |
Closest papers by meaning
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.89)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.87)
- TAMT: Temporal-Aware Model Tuning for Cross-Domain Few-Shot Action Recognition (CVPR 2025 · similarity 0.87)
- Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · similarity 0.85)
- Manta: Enhancing Mamba for Few-Shot Action Recognition of Long Sub-Sequence (AAAI 2025 · similarity 0.85)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.85)
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.
