Research map / Video / Action recognition
Self-Supervised Video Representation Learning by Context and Motion Decoupling
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 |
|---|---|---|---|
| jointly learn context from I-frames and motion from future motion vectors | C3D self-supervised pretraining · UCF101 video retrieval | R@1: +16.0% | Helped |
| predict future rather than current motion (ablation) | context-motion decoupled pretraining · UCF101 video retrieval | R@1: 29.6 -> 41.7 | Helped |
| use pointwise InfoNCE rather than direct motion-value prediction (ablation) | motion prediction task · UCF101 video retrieval | R@1: 27.9 -> 41.7 | Helped |
| use Transformer rather than ConvGRU (ablation) | motion prediction task · UCF101 video retrieval | R@1: 39.8 -> 41.7 | Helped |
| combine context and motion supervision rather than use I-frames alone (ablation) | C3D pretraining · HMDB51 video retrieval | R@1: 11.5 -> 16.8 | Helped |
| combine context and motion supervision rather than use motion vectors alone (ablation) | C3D pretraining · UCF101 video retrieval | R@1: 30.4 -> 41.7 | Helped |
| replace motion vectors with optical flow (ablation) | context-motion pretraining · UCF101 video retrieval | R@1: 41.7 -> 43.2 | Helped |
| add auxiliary motion prediction (ablation) | supervised C3D action recognition · UCF101 action recognition | top-1 accuracy: 60.5 -> 74.3 | Helped |
| add auxiliary motion prediction (ablation) | supervised R(2+1)D-26 action recognition · HMDB51 action recognition | top-1 accuracy: 32.5 -> 41.7 | Helped |
| pretrain on Kinetics400 rather than UCF101 (ablation) | R(2+1)D-26 transfer to UCF101 · UCF101 action recognition | top-1 accuracy: 79.7 -> 85.7 | Helped |
Closest papers by meaning
- Self-Supervised Learning of Compressed Video Representations (ICLR 2021 · similarity 0.91)
- Enhancing Unsupervised Video Representation Learning by Decoupling the Scene and the Motion (AAAI 2021 · similarity 0.91)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.89)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.89)
- SMILE: Infusing Spatial and Motion Semantics in Masked Video Learning (CVPR 2025 · similarity 0.88)
- MotionRAG: Motion Retrieval-Augmented Image-to-Video Generation (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.