Research map / Video / Action recognition
Video Swin Transformer
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 |
|---|---|---|---|
| Use Video Swin Transformer with spatiotemporal locality | Video recognition · Kinetics-400 | Top-1: 84.8 -> 84.9 | Helped |
| Use Video Swin Transformer with spatiotemporal locality | Video recognition · Kinetics-600 | Top-1: 85.8 -> 86.1 | Helped |
| Use Video Swin Transformer | Action recognition · Something-Something v2 | Top-1: +0.9 | Helped |
| Use joint spatiotemporal attention in local 3D windows instead of split spatial/temporal layers (ablation) | Swin-T · Kinetics-400 | Top-1: 76.4 -> 78.8 | Helped |
| Use joint spatiotemporal attention instead of factorized attention (ablation) | Swin-T · Kinetics-400 | Top-1: 78.5 -> 78.8 | Helped |
| Add temporal token dimension 8 rather than dimension 4 (ablation) | Swin-T · Kinetics-400 | Top-1: 76.7 -> 78.5 | Helped |
| Add temporal token dimension 16 rather than dimension 8 (ablation) | Swin-T · Kinetics-400 | Top-1: 78.5 -> 79.1 | Helped |
| Increase temporal window size from 8 to 16 (ablation) | Swin-T with temporal token dimension 16 · Kinetics-400 | Top-1: 78.8 -> 79.1 | Helped |
| Use temporal window size 8 instead of 16 (ablation) | Swin-T with temporal token dimension 16 · Kinetics-400 | FLOPs: 106 -> 88 | Helped |
| Add temporal shifting to 3D shifted windows (ablation) | Swin-T · Kinetics-400 | Top-1: 78.5 -> 78.8 | Helped |
| Add 3D shifted windows (ablation) | Swin-T · Kinetics-400 | Top-1: 78.1 -> 78.8 | Helped |
| Set backbone learning rate to 0.1× the head learning rate (ablation) | Swin-B with ImageNet-1K pretraining · Kinetics-400 | Top-1: 80.2 -> 80.6 | Helped |
Closest papers by meaning
- Swin Transformer V2: Scaling Up Capacity and Resolution (CVPR 2022 · similarity 0.89)
- Space-time Mixing Attention for Video Transformer (NeurIPS 2021 · similarity 0.88)
- Object-Region Video Transformers (CVPR 2022 · similarity 0.87)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.86)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.86)
- Understanding Video Transformers via Universal Concept Discovery (CVPR 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.