Research map / Video / Action recognition
Self-Supervised Motion Learning From Static Images
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 |
|---|---|---|---|
| Pretrain the R-2D3D backbone with MoSI rather than from scratch | R-2D3D · UCF101 split 1 | Top-1 accuracy: 64.5 -> 71.8 | Helped |
| Pretrain the R-2D3D backbone with MoSI rather than from scratch | R-2D3D · HMDB51 split 1 | Top-1 accuracy: 30.4 -> 47.0 | Helped |
| Use both horizontal and vertical motion labels instead of horizontal labels only (ablation) | MoSI · HMDB51 split 1 | Top-1 accuracy: 44.9 -> 47.0 | Helped |
| Use both horizontal and vertical motion labels instead of horizontal labels only (ablation) | MoSI · UCF101 split 1 | Top-1 accuracy: 71.6 -> 71.8 | No clear effect |
| Use 5 motion classes rather than 3 (ablation) | MoSI · HMDB51 | Acc-FT: 43.0 -> 47.0 | Helped |
| Increase motion classes from 5 to 7 or 9 (ablation) | MoSI · HMDB51 | Acc-FT: 47.0 -> 44.6; 47.0 -> 44.5 | Hurt |
| Use 16 frames rather than 12 (ablation) | MoSI · HMDB51 | Acc-FT: 44.9 -> 47.0 | Helped |
| Use 32 frames rather than 16 (ablation) | MoSI · UCF101 | Acc-FT: 71.8 -> 70.0 | Hurt |
| Randomize mask size within the reported range instead of using a fixed size (ablation) | MoSI static masks · UCF101 and HMDB51 | Top-1 accuracy | Helped |
| Add a static mask to create local motion patterns (ablation) | MoSI · Action recognition | Top-1 accuracy | Helped |
| Pretrain MoSI on ImageNet rather than the corresponding video dataset (ablation) | R-2D3D · UCF101 | Acc-FT: 71.8 -> 71.9 | No clear effect |
| Pretrain MoSI on ImageNet-S20 rather than HMDB51 (ablation) | R-2D3D · HMDB51 | Acc-FT: 47.0 -> 48.0 | Helped |
Closest papers by meaning
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.87)
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.87)
- Masked Motion Encoding for Self-Supervised Video Representation Learning (CVPR 2023 · similarity 0.87)
- Exploiting Motion Information from Unlabeled Videos for Static Image Action Recognition (AAAI 2020 · similarity 0.87)
- Self-Supervised Learning of Motion Concepts by Optimizing Counterfactuals (NeurIPS 2025 · similarity 0.85)
- Moving Off-the-Grid: Scene-Grounded Video Representations (NeurIPS 2024 · similarity 0.84)
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.