Research map / Video / Action recognition
Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 RefactorNet feature refactoring to the BSN-based detector (ablation) | THUMOS14 action localization · THUMOS14 test | average mAP: 52.6 -> 54.8 | Helped |
| Replace RefactorNet with depth- and size-matched temporal 1D convolutional layers (ablation) | THUMOS14 action localization · THUMOS14 test | average mAP: 54.8 -> 52.3 | Hurt |
| Add action/co-occurrence decoupling losses (ablation) | RefactorNet · THUMOS14 test | average mAP: 52.6 -> 53.8 | Helped |
| Add KL regularization to co-occurrence features (ablation) | RefactorNet · THUMOS14 test | average mAP: 53.8 -> 54.8 | Helped |
| Use only the action component and omit co-occurrence features (ablation) | RefactorNet representation · THUMOS14 test | action classification accuracy: 85.9 -> 81.3 | Hurt |
| Add co-occurrence features back alongside the action component (ablation) | RefactorNet representation · THUMOS14 test | mAP@tIoU 0.5: 57.2 -> 58.6 | Helped |
| Add RefactorNet to the detector (ablation) | BSN-based detector · THUMOS14 high-quality proposals | action classification accuracy: 76.7 -> 85.9 | Helped |
| Add RefactorNet to the detector (ablation) | BSN-based detector · THUMOS14 high-quality proposals | average tIoU: 79.6 -> 83.5 | Helped |
| Use RefactorNet rather than MUSES | THUMOS14 action localization · THUMOS14 test | mAP@tIoU 0.5: 56.9 -> 58.6 | Helped |
| Use RefactorNet rather than the strongest listed baseline at tIoU 0.75 | ActivityNet v1.3 localization · ActivityNet v1.3 validation | mAP@tIoU 0.75: 39.1 -> 40.7 | Helped |
Closest papers by meaning
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.92)
- One-Shot Action Localization by Learning Sequence Matching Network (CVPR 2018 · similarity 0.91)
- Three Birds with One Stone: Multi-Task Temporal Action Detection via Recycling Temporal Annotations (CVPR 2021 · similarity 0.91)
- Learning Temporal Co-Attention Models for Unsupervised Video Action Localization (CVPR 2020 · similarity 0.91)
- Temporal Action Localization with Cross Layer Task Decoupling and Refinement (AAAI 2025 · similarity 0.90)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.89)
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.