Research map / Video / Action recognition
Recognize Actions by Disentangling Components of Dynamics
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 |
|---|---|---|---|
| Add gradual temporal pooling at multiple network positions (ablation) | Static appearance branch · UCF101 split 1 | accuracy: 85.9% -> 86.5% | Helped |
| Add temporal pooling to the single-frame 2D model (ablation) | Static appearance branch · UCF101 split 1 | accuracy: 84.5% -> 86.5% | Helped |
| Increase the input from 8 to 16 frames (ablation) | Static appearance branch · UCF101 split 1 | accuracy: 86.8% -> 87.4% | Helped |
| Replace soft assignment with winner-take-all assignment (ablation) | Cost-volume motion representation · UCF101 split 1 | accuracy: 89.6% -> 86.4% | Hurt |
| Increase displacement-map resolution from 112 × 112 to 224 × 224 (ablation) | Apparent motion branch · UCF101 split 1 | accuracy: 88.2% -> 89.6% | Helped |
| Warp RGB differences using TV-L1 motion instead of leaving them unwarped (ablation) | Appearance-change representation · UCF101 split 1 | accuracy: 84.0% -> 85.1% | Helped |
| Add TSN long-term modeling (ablation) | Static appearance branch · UCF101 split 1 | accuracy: 86.5% -> 87.3% | Helped |
| Add TSN long-term modeling (ablation) | Displacement Map branch · UCF101 split 1 | accuracy: 80.8% -> 83.4% | Helped |
| Add TSN long-term modeling (ablation) | Full model · UCF101 split 1 | accuracy: 89.5% -> 91.2% | Helped |
| Add warped-difference appearance changes to static appearance and motion (ablation) | Branch fusion with TV-L1 motion · UCF101 split 1 | accuracy: 94.0% -> 94.1% | Helped |
| Add RGB-Diff appearance changes to static appearance and motion (ablation) | Branch fusion with TV-L1 motion · UCF101 split 1 | accuracy: 94.0% -> 93.8% | Hurt |
| Pre-train on Kinetics in addition to ImageNet | Proposed method · UCF101 | accuracy: 91.8% -> 95.9% | Helped |
Closest papers by meaning
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.90)
- Video Modeling With Correlation Networks (CVPR 2020 · similarity 0.89)
- DMC-Net: Generating Discriminative Motion Cues for Fast Compressed Video Action Recognition (CVPR 2019 · similarity 0.89)
- Prompt-guided Disentangled Representation for Action Recognition (NeurIPS 2025 · similarity 0.88)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.88)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · 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.