Research map / Video / Action recognition
Efficient Few-Shot Action Recognition via Multi-Level Post-Reasoning
Cong Wu, Xiao-Jun Wu, Linze Li, Tianyang Xu, Zhenhua Feng, Josef Kittler
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 post-reasoning on cached frozen CLIP features (vs FC layer only) (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 41.0 -> 51.6 | Improved |
| add multi-level (global, patch, frame) representations (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 51.6 -> 57.1 | Improved |
| add skip-fusion over cached multi-stage features (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 55.8 -> 57.1 | Improved |
| add interactive spatiotemporal reasoning (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 54.1 -> 57.1 | Improved |
| add text-guided temporal reasoning on support branch (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 55.5 -> 57.1 | Improved |
| add text-visual matching alongside support-query (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 56.1 -> 57.1 | Improved |
| add parallel temporal attention alongside spatial in skip-fusion (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 54.4 -> 57.1 | Improved |
| use selective Top-K mechanism on support-query matching (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 57.1 -> 56.5 | Hurt |
| use selective mechanism on text-visual matching (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 56.4 -> 57.1 | Improved |
| increase CLIP backbone from ViT-B/32 to ViT-B/16 (ablation) | EMP-Net · SSv2-Small 1-shot | accuracy (%): 54.4 -> 57.1 | Improved |
| use cached features from layers {4,8,12} instead of {6,12} or {3,6,9,12} (ablation) | EMP-Net skip-fusion · SSv2-Small 1-shot | accuracy (%): 55.4 / 56.2 -> 57.1 | Improved |
| replace trainable CLIP-FSAR with EMP-Net on frozen CLIP | CLIP-FSAR · SSv2-Small 1-shot, 8 frames | accuracy (%): 50.1 -> 57.1 | Improved |
Closest papers by meaning
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.86)
- TAMT: Temporal-Aware Model Tuning for Cross-Domain Few-Shot Action Recognition (CVPR 2025 · similarity 0.85)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.84)
- Manta: Enhancing Mamba for Few-Shot Action Recognition of Long Sub-Sequence (AAAI 2025 · similarity 0.84)
- Building a Multi-modal Spatiotemporal Expert for Zero-shot Action Recognition with CLIP (AAAI 2025 · similarity 0.84)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.83)
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.
