Research map / Video / Action recognition
MPL: Match-guided Prototype Learning for Few-shot Action Recognition
Feng Yang, Jie Zhao, Fulin Luo, Anyong Qin, Tiecheng Song, Yue Zhao, Chenqiang Gao, Junwei Han
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 E-Match and K-Match to the BiMHM baseline (ablation) | BiMHM baseline · Kinetics 5-way 1-shot | accuracy: 71.8 -> 77.1 | Improved |
| add E-Match and K-Match to the BiMHM baseline (ablation) | BiMHM baseline · UCF101 5-way 1-shot | accuracy: 82.7 -> 90.1 | Improved |
| add E-Match and K-Match to the BiMHM baseline (ablation) | BiMHM baseline · SSv2 5-way 1-shot | accuracy: 38.1 -> 43.5 | Improved |
| add E-Match and K-Match to the BiMHM baseline (ablation) | BiMHM baseline · HMDB51 5-way 1-shot | accuracy: 37.9 -> 41.5 | Improved |
| replace mean prototype averaging with CSA-Aggregator (ablation) | MPL · Kinetics 5-way 3-shot | accuracy: 84.7 -> 86.2 | Improved |
| replace mean prototype averaging with CSA-Aggregator (ablation) | MPL · SSv2 5-way 5-shot | accuracy: 56.4 -> 56.6 | No clear effect |
| replace mean prototype averaging with CSA-Aggregator (ablation) | MPL · SSv2 5-way 3-shot | accuracy: 51.8 -> 53.6 | Improved |
| increase CSA-Aggregator window length w from 1 to 7 (ablation) | CSA-Aggregator · Kinetics 5-way 3-shot | accuracy: 86.8 -> 87.2 | Improved |
| use the same matching method for E-Match and hybrid match instead of different ones (ablation) | MPL · Kinetics 5-way 1-shot | accuracy: 76.9 -> 77.4 | Improved |
| increase input frames beyond 8 (ablation) | MPL · Kinetics and SSv2 | accuracy | No clear effect |
| use deeper backbone (ResNet-34 to ResNet-101) (ablation) | MPL · Kinetics N-way 1-shot | accuracy | Improved |
| use MPL with CLIP visual encoder only instead of CLIP-FSAR with text | CLIP-RN50 · SSv2 5-way 1-shot | accuracy: 40.3 -> 40.1 | Mixed |
Closest papers by meaning
- Compound Prototype Matching for Few-Shot Action Recognition (ECCV 2022 · similarity 0.92)
- Matching Compound Prototypes for Few-Shot Action Recognition (IJCV 2024 · similarity 0.91)
- Boosting Few-shot Action Recognition with Graph-guided Hybrid Matching (ICCV 2023 · similarity 0.90)
- Few-Shot Action Recognition with Hierarchical Matching and Contrastive Learning (ECCV 2022 · similarity 0.89)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.87)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · 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.
