Research map / Video / Action recognition
Time Is MattEr: Temporal Self-supervision for Video Transformers
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 |
|---|---|---|---|
| Add TIME temporal self-supervision | TimeSformer · SSv2 | Top-1 accuracy: 62.1 -> 63.7 | Helped |
| Add TIME temporal self-supervision | Motionformer · SSv2 | Top-1 accuracy: 63.8 -> 64.7 | Helped |
| Add TIME temporal self-supervision | X-ViT · SSv2 | Top-1 accuracy: 60.1 -> 63.5 | Helped |
| Add patch-order and background-debiasing self-supervision | DeiT-Ti/16 · ImageNet-9 Original | Top-1 accuracy: 77.3 -> 83.3 | Helped |
| Add patch-order and background-debiasing self-supervision | DeiT-Ti/16 · ImageNet-9 Only-FG | Top-1 accuracy: 50.3 -> 58.9 | Helped |
| Add frame-order prediction (ablation) | TimeSformer + debias loss + flow loss · SSv2 | Original top-1 accuracy: 63.4 -> 63.7 | Helped |
| Add shuffled-video debiasing (ablation) | TimeSformer + order loss + flow loss · SSv2 | Original top-1 accuracy: 62.7 -> 63.7 | Helped |
| Add token-flow direction prediction (ablation) | TimeSformer + order loss + debias loss · SSv2 | Original top-1 accuracy: 63.2 -> 63.7 | Helped |
| Add patch-order prediction (ablation) | DeiT-Ti/16 with image debias loss · ImageNet-9 Only-FG | Top-1 accuracy: 52.7 -> 58.9 | Helped |
| Add image background debiasing (ablation) | DeiT-Ti/16 with patch-order loss · ImageNet-9 Only-FG | Top-1 accuracy: 54.2 -> 58.9 | Helped |
Closest papers by meaning
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.89)
- When the Future Becomes the Past: Taming Temporal Correspondence for Self-supervised Video Representation Learning (CVPR 2025 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.88)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.87)
- Understanding Video Transformers via Universal Concept Discovery (CVPR 2024 · 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.