Research map / Video / Action recognition
An Efficient Spatio-Temporal Pyramid Transformer for Action Detection
Yuetian Weng, Zizheng Pan, Mingfei Han, Xiaojun Chang, Bohan Zhuang
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 |
|---|---|---|---|
| replace global attention with local window attention (LSTA) in the first two stages (ablation) | MViT · THUMOS14 | average mAP and GFLOPs: +1.4% average mAP, -55.4G FLOPs | Improved |
| add global attention (GSTA) in stage 3 instead of only LSTA (ablation) | STPT · THUMOS14 | average mAP (%): 48.3 -> 51.3 | Improved |
| use LSTA in first two stages instead of GSTA in all stages (ablation) | STPT · THUMOS14 | average mAP (%): 51.8 -> 52.3 (167.6 -> 111.4 GFLOPs) | Improved |
| use LSTA in first two stages instead of GSTA in all stages (ablation) | STPT · ActivityNet 1.3 val | average mAP (%): 32.2 -> 33.4 | Improved |
| add GSTA in later stages instead of only LSTA (LLLL) (ablation) | STPT · ActivityNet 1.3 val | average mAP (%): 31.7 -> 33.4 | Improved |
| use window [8,8,16] instead of [1,1,1] (ablation) | STPT · THUMOS14 | average mAP (%): 50.3 -> 52.7 | Improved |
| use window [8,8,8] instead of [16,16,16] (ablation) | STPT · THUMOS14 | average mAP (%): 50.9 -> 50.1 | Improved |
| use window [8,8,16] instead of [16,16,16] (ablation) | STPT · ActivityNet 1.3 val | average mAP (%): 33.4 -> 32.0 | Improved |
| add CPE (ablation) | STPT · THUMOS14 (128 tokens) | average mAP (%): 52.7 -> 53.6 | Improved |
| add CPE (ablation) | STPT · THUMOS14 (64 tokens) | average mAP (%): 50.7 -> 51.1 | Improved |
| add CPE (ablation) | STPT · ActivityNet 1.3 val (192 tokens) | average mAP (%): 32.0 -> 33.0 | Improved |
| use RGB-only STPT instead of AFSD RGB model | STPT · THUMOS14 | average mAP (%): +10.1% average mAP (43.5 -> 53.6) | Improved |
Closest papers by meaning
- ActionFormer: Localizing Moments of Actions with Transformers (ECCV 2022 · similarity 0.90)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.89)
- Object-Region Video Transformers (CVPR 2022 · similarity 0.88)
- Shrinking Temporal Attention in Transformers for Video Action Recognition (AAAI 2022 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.87)
- End-to-End Spatio-Temporal Action Localisation with Video Transformers (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.
