Research map / Video / Action recognition
TransRank: Self-Supervised Video Representation Learning via Ranking-Based Transformation Recognition
Haodong Duan, Nanxuan Zhao, Kai Chen, Dahua Lin
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| use ranking-based transformation recognition instead of hard-label classification (ablation) | TransRank on temporal transformations · UCF101 action recognition | Top-1 accuracy: 83.3 -> 84.1 | Improved |
| use ranking-based transformation recognition instead of hard-label classification (ablation) | TransRank on temporal transformations · HMDB51 action recognition | Top-1 accuracy: 52.7 -> 54.5 | Improved |
| add reverse playback to the {1×, 2×} transformation set (ablation) | TransRank · UCF101 action recognition | Top-1 accuracy: 81.3 -> 84.1 | Improved |
| add palindrome transformation to {1×, 2×, rev} (ablation) | TransRank · UCF101 action recognition | Top-1 accuracy: 84.1 -> 82.9 | Hurt |
| add shuffle transformation to {1×, 2×, rev} (ablation) | TransRank · UCF101 action recognition | Top-1 accuracy: 84.1 -> 83.8 | Hurt |
| add 4× playback to {1×, 2×, rev} (ablation) | TransRank · UCF101 action recognition | Top-1 accuracy: 84.1 -> 85.1 | Improved |
| add rev-2× playback to {1×, 2×, rev} (ablation) | TransRank · UCF101 action recognition | Top-1 accuracy: 84.1 -> 85.3 | Improved |
| add spatial augmentation (ablation) | TransRank · HMDB51 action recognition | Top-1 accuracy: 50.8 -> 54.5 | Improved |
| add temporal augmentation (ablation) | TransRank · HMDB51 action recognition | Top-1 accuracy: 54.2 -> 54.5 | Improved |
| formulate aspect-ratio estimation as ranking rather than classification (ablation) | TransRank-ST · HMDB51 action recognition | Top-1 accuracy: 53.5 -> 56.6 | Improved |
| formulate rotation estimation as ranking rather than classification (ablation) | TransRank-ST · HMDB51 action recognition | Top-1 accuracy: 57.1 -> 56.8 | Hurt |
| replace the fully connected transformation head with a 2-layer MLP (ablation) | TransRank-T · UCF101 video retrieval | R@1: 22.0 -> 32.5 | Improved |
Closest papers by meaning
- Learning Video Representations by Transforming Time (ECCV 2020 · similarity 0.88)
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.87)
- Self-Supervised Learning of Compressed Video Representations (ICLR 2021 · similarity 0.87)
- Uncovering the Hidden Dynamics of Video Self-supervised Learning under Distribution Shifts (NeurIPS 2023 · similarity 0.87)
- Time Is MattEr: Temporal Self-supervision for Video Transformers (ICML 2022 · similarity 0.86)
- 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.
