Research map / Video / Action recognition
Spatio-Temporal Decoupled Knowledge Compensator for Few-Shot Action Recognition
Hongyu Qu, Xiangbo Shu, Rui Yan, Hailiang Gao, Wenguan Wang, Jinhui Tang
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 Temporal Knowledge Compensator (TKC) (ablation) | DiST baseline · HMDB51 (5-way 1-shot) | accuracy: 75.8 -> 81.0 | Improved |
| add Spatial Knowledge Compensator (SKC) (ablation) | DiST baseline · HMDB51 (5-way 1-shot) | accuracy: 75.8 -> 77.6 | Improved |
| add both SKC and TKC (ablation) | DiST baseline · HMDB51 (5-way 1-shot) | accuracy: 75.8 -> 82.6 | Improved |
| add SKC to TKC-only model (ablation) | DiST · SSv2-small (5-way 1-shot) | accuracy: 56.1 -> 57.5 | Improved |
| replace class names with LLM-generated knowledge for spatial and temporal attributes (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 80.0 -> 82.6 | Improved |
| use SKC/TKC injection instead of concatenation (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 80.7 -> 82.6 | Improved |
| use LLM-generated prompts in CLIP-FSAR plus knowledge compensators (ablation) | CLIP-FSAR · HMDB51 (5-way 1-shot) | accuracy: 81.0 -> 82.6 | Improved |
| use 6 spatial attributes instead of 1 (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 81.8 -> 82.6 | Improved |
| use 12 spatial attributes instead of 6 (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 82.6 -> 82.3 | Hurt |
| use 3 temporal attributes instead of 6 (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 81.5 -> 82.6 | Improved |
| use 9 object-level prototypes instead of 1 (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 80.8 -> 82.6 | Improved |
| increase object-level prototypes from 9 to 16 (ablation) | DiST · HMDB51 (5-way 1-shot) | accuracy: 82.6 -> 82.1 | Hurt |
Closest papers by meaning
- Building a Multi-modal Spatiotemporal Expert for Zero-shot Action Recognition with CLIP (AAAI 2025 · similarity 0.89)
- Spatio-Temporal Relation Modeling for Few-Shot Action Recognition (CVPR 2022 · similarity 0.89)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.89)
- OST: Refining Text Knowledge with Optimal Spatio-Temporal Descriptor for General Video Recognition (CVPR 2024 · similarity 0.87)
- Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · similarity 0.87)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 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.
