Research map / Video / Action recognition
Enhancing Unsupervised Video Representation Learning by Decoupling the Scene and the Motion
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 |
|---|---|---|---|
| Apply DSM instead of MoCo | I3D pretrained on Kinetics, fine-tuned on UCF101 · UCF101 | top-1 accuracy: 62.3 -> 74.8 | Helped |
| Apply DSM instead of MoCo | I3D pretrained on Kinetics, fine-tuned on HMDB51 · HMDB51 | top-1 accuracy: 36.5 -> 52.5 | Helped |
| Apply DSM instead of MoCo | I3D pretrained on Kinetics and fine-tuned with 1% labels · Kinetics val | top-1 accuracy: 17.7% to 33.7% | Helped |
| Apply DSM instead of MoCo | I3D pretrained on Kinetics and fine-tuned with 10% labels · Kinetics val | top-1 accuracy: 38.8% to 43.6% | Helped |
| Apply DSM instead of MoCo | I3D pretrained on Kinetics and fine-tuned with 100% labels · Kinetics val | top-1 accuracy: 71.4% to 73% | Helped |
| Add spatial warping for positive clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · UCF101 | top-1 accuracy: 62.3 -> 66.4 (+4.1) | Helped |
| Add spatial warping for positive clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · HMDB51 | top-1 accuracy: 36.5 -> 41.3 (+4.8) | Helped |
| Add motion disturbance for negative clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · UCF101 | top-1 accuracy: 62.3 -> 67.7 (+5.4) | Helped |
| Add motion disturbance for negative clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · HMDB51 | top-1 accuracy: 36.5 -> 44.0 (+7.5) | Helped |
| Add spatial warping and optical-flow scaling for negative clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · UCF101 | top-1 accuracy: 62.3 -> 69.4 (+7.1) | Helped |
| Add spatial warping and temporal shift for negative clips (ablation) | MoCo baseline with I3D pretrained on Kinetics · HMDB51 | top-1 accuracy: 36.5 -> 50.4 (+13.9) | Helped |
| Combine optical-flow scaling and temporal shift for negative clips with spatial warping for positives (ablation) | MoCo baseline with I3D pretrained on Kinetics · UCF101 | top-1 accuracy: 62.3 -> 74.8 (+12.5) | Helped |
Closest papers by meaning
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.91)
- Why Can't I Dance in the Mall? Learning to Mitigate Scene Bias in Action Recognition (NeurIPS 2019 · similarity 0.90)
- RSPNet: Relative Speed Perception for Unsupervised Video Representation Learning (AAAI 2021 · similarity 0.89)
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 2025 · similarity 0.87)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (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.