Research map / Video / Action recognition
TEINet: Towards an Efficient Architecture for Video Recognition
Zhaoyang Liu, Donghao Luo, Yabiao Wang, Limin Wang, Ying Tai, Chengjie Wang, Jilin Li, Feiyue Huang, Tong Lu
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 TEINet instead of TSM with 16 input frames | ResNet-50 video recognition · Something-Something V1 val | Top-1 accuracy: 44.8% -> 49.9% | Improved |
| Use TEINet instead of TSM with 16-frame, 10-clip evaluation | ResNet-50 video recognition · Something-Something V2 val | Top-1 accuracy: 59.4% -> 64.7% | Improved |
| Use TEINet with 16 frames rather than NL+I3D with 32 frames | Kinetics-400 recognition · Kinetics-400 | Top-1 accuracy: 74.9% -> 76.2% | Improved |
| Add MEM to TIM (ablation) | ResNet-50 with TIM · Something-Something V1 | Top-1 accuracy: 46.1% -> 47.4% | Improved |
| Add TIM to MEM (ablation) | ResNet-50 with MEM · Something-Something V1 | Top-1 accuracy: 33.5% -> 47.4% | Improved |
| Use motion-aware MEM rather than SE attention (ablation) | ResNet-50 with TIM · Something-Something V1 | Top-1 accuracy: 46.1% -> 47.4% | Improved |
| Place TEI blocks at later ResNet stages rather than res2 or res3 (ablation) | ResNet-50 · Something-Something V1 | Top-1 accuracy: 41.6% -> 45.4% | Improved |
| Add TEI blocks progressively from res4-5 to res2-5 (ablation) | ResNet-50 · Something-Something V1 | Top-1 accuracy: 46.7% -> 47.4% | Improved |
| Increase input from 8 to 16 frames (ablation) | TEINet-RGB · Something-Something V1 val | Top-1 accuracy: 47.4% -> 49.9% | Improved |
Closest papers by meaning
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.89)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · similarity 0.88)
- TEA: Temporal Excitation and Aggregation for Action Recognition (CVPR 2020 · similarity 0.87)
- AdaFuse: Adaptive Temporal Fusion Network for Efficient Action Recognition (ICLR 2021 · similarity 0.87)
- MiCT: Mixed 3D/2D Convolutional Tube for Human Action Recognition (CVPR 2018 · similarity 0.87)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.85)
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.
