Research map / Video / Action recognition
Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition
Syed Talal Wasim, Muhammad Uzair Khattak, Muzammal Naseer, Salman Khan, Mubarak Shah, Fahad Shahbaz Khan
Read the paperCheck 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 Video-FocalNet instead of Video-Swin at Tiny size | Video-FocalNet-T · Kinetics-400 val | top-1 accuracy: 78.8 -> 79.8 | Improved |
| use Video-FocalNet instead of Video-Swin at Small size | Video-FocalNet-S · Kinetics-400 val | top-1 accuracy: 80.6 -> 81.4 | Improved |
| use Video-FocalNet-B with ImageNet-1K init instead of Uniformer-B | Video-FocalNet-B · Kinetics-400 val | top-1 accuracy: 83.0 -> 83.6 | Improved |
| use Video-FocalNet-B instead of MViTv2-B | Video-FocalNet-B · Kinetics-600 val | top-1 accuracy: 85.5 -> 86.7 | Improved |
| use Video-FocalNet-B instead of TFCNET | Video-FocalNet-B · Diving-48 V2 | top-1 accuracy: 88.3 -> 90.8 | Improved |
| use Video-FocalNet-B instead of Video-Swin-B | Video-FocalNet-B · ActivityNet 1.3 | top-1 accuracy: 88.5 -> 89.8 | Improved |
| fuse spatial and temporal modulators by elementwise multiplication instead of averaging or learnable projection (ablation) | Video-FocalNet-S · Kinetics-400 val | top-1 accuracy | Improved |
| use simple patch embedding (T=1) instead of tubelet embedding (T=2) (ablation) | Video-FocalNet-S · Kinetics-400 val | top-1 accuracy | Improved |
| use parallel spatial and temporal focal modulation instead of factorized, divided or averaged designs (ablation) | Video-FocalNet-S · Kinetics-400 val | top-1 accuracy and FLOPs | Improved |
Closest papers by meaning
- Look More but Care Less in Video Recognition (NeurIPS 2022 · similarity 0.89)
- Focal Modulation Networks (NeurIPS 2022 · similarity 0.89)
- Space-time Mixing Attention for Video Transformer (NeurIPS 2021 · similarity 0.88)
- Spatiotemporal Self-Attention Modeling with Temporal Patch Shift for Action Recognition (ECCV 2022 · similarity 0.88)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.87)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · 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.
