Research map / Video / Action recognition
Pulling Actions out of Context: Explicit Separation for Effective Combination
Read the paperCheck 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 |
|---|---|---|---|
| Use conjugate samples to explicitly separate action and context features | C3D action recognition · ActionThread | mean AP: 42.0 -> 47.9 | Helped |
| Use conjugate samples to explicitly separate action and context features | C3D action recognition · UCF101 | accuracy: 82.3 -> 84.5 | Helped |
| Use conjugate samples to explicitly separate action and context features | C3D action recognition · Hollywood2 | mean AP: 49.8 -> 54.7 | Helped |
| Use conjugate samples to explicitly separate action and context features | VGG16 action recognition · PASCAL VOC-2012 Action validation | mean AP: 74.1% -> 75.2% | Helped |
| Combine Factor-C3D with DTD features | ActionThread recognition · ActionThread | mean AP: 52.0 -> 55.3 | Helped |
| Use conjugate samples as negative training examples instead of explicit separation (ablation) | C3D action recognition · ActionThread | mean AP: 42.0 -> 41.1 | Hurt |
| Use conjugate samples as positive training examples instead of explicit separation (ablation) | C3D action recognition · ActionThread | mean AP: 42.0 -> 41.2 | Hurt |
| Use conjugate samples as positive training examples instead of explicit separation (ablation) | VGG16 action recognition · PASCAL VOC-2012 Action validation | mean AP: 74.1% -> 73.1% | Hurt |
| Use conjugate samples as negative training examples instead of explicit separation (ablation) | VGG16 action recognition · PASCAL VOC-2012 Action validation | mean AP: 74.1% -> 72.6% | Hurt |
| Remove context features from the combined representation (ablation) | Factor-C3D · ActionThread, contextually similar action pairs | average precision | Helped |
| Remove context features from the combined representation (ablation) | Factor-C3D · ActionThread, contextually dissimilar action pairs | average precision | Hurt |
| Use imperfect, within-video conjugate samples rather than context-matched pre- or post-action sequences (ablation) | Factor-C3D · UCF101 and Hollywood2 | recognition performance | Mixed |
Closest papers by meaning
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.88)
- Video Action Transformer Network (CVPR 2019 · similarity 0.88)
- Improving Human Action Recognition by Non-Action Classification (CVPR 2016 · similarity 0.88)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.87)
- Disentangled Concepts Speak Louder Than Words: Explainable 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.