Research map / Video / Action recognition
Alignment-guided Temporal Attention for Video Action Recognition
Yizhou Zhao, Zhenyang Li, Xun Guo, Yan 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 alignment-guided temporal attention to TimeSformer | TimeSformer · Kinetics-400 | Top-1 accuracy: 81.1 -> 81.9 | Improved |
| Add alignment-guided temporal attention to TimeSformer | TimeSformer · Something-Something V2 | Top-1 accuracy: 68.1 -> 67.0 | Mixed |
| Add de-alignment after the aligned temporal operation (ablation) | ATA with CycleMLP-B5 · Kinetics-400 | Top-1 accuracy: 72.7 -> 77.7 | Improved |
| Add de-alignment after the aligned temporal operation (ablation) | ATA with ConvNeXt-Base · Kinetics-400 | Top-1 accuracy: 76.0 -> 80.5 | Improved |
| Add de-alignment after the aligned temporal operation (ablation) | ATA with ViT-Base · Kinetics-400 | Top-1 accuracy: 79.3 -> 79.6 | Improved |
| Replace temporal averaging with factorized temporal attention (ablation) | CycleMLP-B5 · Kinetics-400 | Top-1 accuracy: 74.9 -> 76.8 | Improved |
| Replace temporal averaging with factorized temporal attention (ablation) | ConvNeXt-Base · Kinetics-400 | Top-1 accuracy: 79.0 -> 80.1 | Improved |
| Replace temporal averaging with factorized temporal attention (ablation) | ViT-Base · Kinetics-400 | Top-1 accuracy: 76.0 -> 78.0 | Improved |
| Add patch alignment to temporal attention (ablation) | ViT-Base temporal attention · Kinetics-400 | Top-1 accuracy: 78.0 -> 79.6 | Improved |
| Add patch alignment to temporal attention (ablation) | ViT-Base temporal attention · Something-Something V2 | Mutual information: 0.910 -> 1.290 | Improved |
| Insert ATA in one selected stage instead of no alignment (ablation) | ViT-Base · Kinetics-400 | Top-1 accuracy: 78.0 -> 79.2; 78.0 -> 79.3; 78.0 -> 79.6 | Improved |
| Increase the number of ATA-equipped blocks (ablation) | ViT-Base · Kinetics-400 | Top-1 accuracy: 79.3; 79.2; 79.4; 79.6 | Mixed |
Closest papers by meaning
- Learning by Aligning Videos in Time (CVPR 2021 · similarity 0.88)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.88)
- Temporal Alignment Networks for Long-Term Video (CVPR 2022 · similarity 0.88)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · similarity 0.87)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.86)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · similarity 0.86)
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.
