Research map / Video / Action recognition
DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 softmax temporal attention with cosine temporal attention (ablation) | DirecFormer with softmax spatial attention · Jester | Top 1 accuracy: 94.52 -> 95.52 | Helped |
| add order prediction loss (ablation) | DirecFormer C-S · Jester | Top 1 accuracy: 95.52 -> 96.28 | Helped |
| add self-supervised guided loss alongside order prediction loss (ablation) | DirecFormer C-S · Jester | Top 1 accuracy: 96.28 -> 97.55 | Helped |
| use cosine attention in both temporal and spatial dimensions instead of cosine temporal and softmax spatial attention (ablation) | DirecFormer · Jester | Top 1 accuracy: 95.52 -> 96.15 | Helped |
| add order prediction and self-supervised guided losses (ablation) | DirecFormer C-C · Jester | Top 1 accuracy: 96.15 -> 98.15 | Helped |
| use cosine temporal attention instead of softmax temporal attention (ablation) | DirecFormer with softmax spatial attention · Something-Something V2 validation | Top 1 accuracy: 61.70 -> 63.85 | Helped |
| use cosine attention in both temporal and spatial dimensions instead of cosine temporal and softmax spatial attention (ablation) | DirecFormer · Something-Something V2 validation | Top 1 accuracy: 63.85 -> 64.94 | Helped |
| use cosine temporal attention instead of softmax temporal attention (ablation) | DirecFormer with softmax spatial attention · Kinetics-400 validation | Top 1 accuracy: 80.16 -> 81.69 | Helped |
| use cosine attention in both temporal and spatial dimensions instead of cosine temporal and softmax spatial attention (ablation) | DirecFormer · Kinetics-400 validation | Top 1 accuracy: 81.69 -> 82.75 | Helped |
| add order prediction loss (ablation) | TimeSFormer · Jester order correction | OrderAcc: 52.84 -> 72.57 | Helped |
| add order prediction loss (ablation) | DirecFormer C-S · Jester order correction | OrderAcc: 75.04 -> 87.16 | Helped |
| add self-supervised guided loss alongside order prediction loss (ablation) | DirecFormer C-S · Jester order correction | OrderAcc: 87.16 -> 90.02 | Helped |
Closest papers by meaning
- A Large-Scale Robustness Analysis of Video Action Recognition Models (CVPR 2023 · similarity 0.88)
- SVFormer: Semi-Supervised Video Transformer for Action Recognition (CVPR 2023 · similarity 0.87)
- Video Action Transformer Network (CVPR 2019 · similarity 0.87)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.87)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.86)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · 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.