Research map / Video / Action recognition
TAM: Temporal Adaptive Module for Video Recognition
Zhaoyang Liu, Limin Wang, Wayne Wu, Chen Qian, 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 |
|---|---|---|---|
| add TAM to ResNet-50 (C2D) (ablation) | ResNet-50 C2D · Kinetics-400 val | top-1 accuracy (%): 70.2 -> 76.3 | Improved |
| add TAM to ShuffleNet V2 (ablation) | ShuffleNet V2 · Kinetics-400 val | top-1 accuracy (%): 62.1 -> 67.3 | Improved |
| add TAM to MobileNet V2 (ablation) | MobileNet V2 · Kinetics-400 val | top-1 accuracy (%): 64.1 -> 71.6 | Improved |
| add TAM to Inception V3 (ablation) | Inception V3 · Kinetics-400 val | top-1 accuracy (%): 71.4 -> 75.6 | Improved |
| add TAM to I3D-ResNet-50 (ablation) | I3D-ResNet-50 · Kinetics-400 val | top-1 accuracy (%): 76.6 -> 77.2 | Improved |
| add local branch to the global branch (ablation) | TANet · Kinetics-400 val | top-1 accuracy (%): 75.6 -> 76.3 | Improved |
| add global branch to the local branch (ablation) | TANet · Kinetics-400 val | top-1 accuracy (%): 73.3 -> 76.3 | Improved |
| use the local branch instead of SE module alongside global branch (ablation) | TANet · Kinetics-400 val | top-1 accuracy (%): 75.9 -> 76.3 | Improved |
| apply local branch before global branch rather than reversed (ablation) | TANet · Kinetics-400 val | top-1 accuracy (%): 76.0 -> 76.3 | Improved |
| enlarge adaptive kernel from K=3 to K=5 with 8 frames (ablation) | TANet (8 frames) · Kinetics-400 val | top-1 accuracy (%): 76.28 -> 75.62 | Hurt |
| enlarge adaptive kernel from K=3 to K=5 with 16 frames (ablation) | TANet (16 frames) · Kinetics-400 val | top-1 accuracy (%): 76.87 -> 77.19 | Improved |
| insert TAM after last convolution instead of after first (ablation) | TANet · Kinetics-400 val | top-1 accuracy (%): 76.28 -> 75.20 | Hurt |
Closest papers by meaning
- TEINet: Towards an Efficient Architecture for Video Recognition (AAAI 2020 · similarity 0.86)
- TAda! Temporally-Adaptive Convolutions for Video Understanding (ICLR 2022 · similarity 0.85)
- Modality-Collaborative Test-Time Adaptation for Action Recognition (CVPR 2024 · similarity 0.84)
- Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · similarity 0.84)
- TAMT: Temporal-Aware Model Tuning for Cross-Domain Few-Shot Action Recognition (CVPR 2025 · similarity 0.84)
- DejaVid: Encoder-Agnostic Learned Temporal Matching for Video Classification (CVPR 2025 · similarity 0.83)
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.
