Research map / Video / Action recognition
Cross-Architecture Self-Supervised Video Representation Learning
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 |
|---|---|---|---|
| Add cross-architecture contrastive learning with a video transformer (ablation) | V-MoCo with C3D · UCF101 video retrieval | Recall@1 (%): 30.3 -> 39.1 | Helped |
| Add shuffle-degree prediction to video-level contrastive learning (ablation) | V-MoCo with C3D · UCF101 video retrieval | Recall@1 (%): 30.3 -> 36.2 | Helped |
| Add the transformer and shuffle-degree prediction together (ablation) | V-MoCo with C3D · UCF101 video retrieval | Recall@1 (%): 30.3 -> 43.2 | Helped |
| Use uniform shuffle-degree sampling (ablation) | Shuffle-degree prediction with C3D · UCF101 video retrieval | Recall@1 (%): 17.2 -> 22.1 | Helped |
| Add progressively more groups of cross-architecture positive pairs (ablation) | CACL with C3D · UCF101 video retrieval | Recall@1 (%): 36.2 -> 37.8 -> 39.3 -> 43.2 | Helped |
| Pretrain on Kinetics400 rather than UCF101 (ablation) | CACL with C3D · UCF101 video retrieval | Recall@1 (%): 43.2 -> 44.2 | Helped |
| Use CACL rather than the strongest listed C3D retrieval baseline | C3D video representations · UCF101 video retrieval | Recall@1 (%): 36.0 -> 43.2 | Helped |
| Use CACL rather than the strongest listed C3D retrieval baseline | C3D video representations · HMDB51 video retrieval | Recall@1 (%): 8.2 -> 17.3 | Helped |
| Use CACL rather than the strongest listed method | R(2+1)D action recognition · HMDB51 | Top-1 accuracy (%): 49.2 -> 48.8 | Hurt |
Closest papers by meaning
- Cycle-Contrast for Self-Supervised Video Representation Learning (NeurIPS 2020 · similarity 0.92)
- Long-Short Temporal Contrastive Learning of Video Transformers (CVPR 2022 · similarity 0.91)
- Frame-Wise Action Representations for Long Videos via Sequence Contrastive Learning (CVPR 2022 · similarity 0.89)
- Self-supervised Co-Training for Video Representation Learning (NeurIPS 2020 · similarity 0.88)
- Self-Supervised Learning of Compressed Video Representations (ICLR 2021 · similarity 0.88)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.86)
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.