Research map / Video / Action recognition
Compound Prototype Matching for Few-Shot Action Recognition
Yifei Huang, Lijin Yang, Yoichi Sato
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 object features and multi-relation encoder | Compound prototype matching · SSv2#, Kinetics, HMDB | 1-shot accuracy: 49.3 -> 59.6; 73.3 -> 81.0; 60.1 -> 80.3 | Improved |
| add object features by concatenation to baselines (ablation) | MatchNet/ITA-Net/TRX · SSv2, Kinetics | 1-shot accuracy: 2~5 percent on SSv2, 2~6 percent on Kinetics | Improved |
| add multi-relation encoder (global-global only) without object features (ablation) | ITA-Net, TRX, Ours · SSv2#, Kinetics | 1-shot accuracy: TRX 41.5 -> 44.4; 64.6 -> 68.9 | Improved |
| add multi-relation encoder together with object features (ablation) | Ours · SSv2# | 1-shot accuracy: 49.1 -> 59.6 | Improved |
| add multi-relation encoder together with object features (ablation) | TRX+ · SSv2# | 1-shot accuracy: 44.2 -> 48.4 | Improved |
| add global-object relation F_go to F_gg and F_oo (ablation) | Ours · SSv2# | 1-shot accuracy: 57.7 -> 59.6 | Improved |
| add object-object relation F_oo to F_gg (ablation) | Ours · SSv2# | 1-shot accuracy: 49.3 -> 57.7 | Improved |
| add global-global relation F_gg to F_go and F_oo (ablation) | Ours · Kinetics | 1-shot accuracy: 77.1 -> 81.0 | Improved |
| add focused prototypes to global prototypes (ablation) | Ours · SSv2#, Kinetics | 1-shot accuracy: 39.3 -> 59.6; 66.1 -> 81.0 | Improved |
| add global prototypes to focused prototypes (ablation) | Ours · SSv2#, Kinetics | 1-shot accuracy: 44.0 -> 59.6; 76.7 -> 81.0 | Improved |
| increase prototype count with more focused than global prototypes (ablation) | Ours · SSv2#, Kinetics | 1-shot accuracy: 59.6 -> 62.0; 81.0 -> 86.9 | Improved |
| replace multi-relation encoder with feature concatenation (ablation) | Ours · UCF | 1-shot accuracy: 79.0 -> 87.7 | Improved |
Closest papers by meaning
- Matching Compound Prototypes for Few-Shot Action Recognition (IJCV 2024 · similarity 0.97)
- MPL: Match-guided Prototype Learning for Few-shot Action Recognition (CVPR 2026 · similarity 0.92)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.90)
- Few-Shot Action Recognition with Hierarchical Matching and Contrastive Learning (ECCV 2022 · similarity 0.90)
- Boosting Few-shot Action Recognition with Graph-guided Hybrid Matching (ICCV 2023 · similarity 0.90)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.87)
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.
