Research map / Video / Action recognition
Video Action Transformer Network
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 |
|---|---|---|---|
| add the Tx head to the I3D model | I3D action detection model · AVA v2.1 val | mAP: 20.5 -> 24.4 | Helped |
| use the I3D head for regression and Tx for classification | I3D action detection model · AVA v2.1 val | mAP: 24.4 -> 24.9 | Helped |
| increase the Tx model's proposals from 64 to 300 (ablation) | Tx-only head · AVA v2.1 val | mAP: 18.9 -> 24.4 | Helped |
| use QPr average rather than QPr concat (ablation) | Tx head with GT person boxes · AVA v2.1 val, GT person boxes | mAP: 27.6 -> 29.1 | Helped |
| use QPr concat rather than QPr average (ablation) | Tx head for action-agnostic localization · AVA v2.1 val, IOU@0.5 | mAP: 77.5 -> 87.7 | Helped |
| replace the Tx head with the I3D head (ablation) | action-agnostic localization model · AVA v2.1 val, IOU@0.5 | mAP: 87.7 -> 92.9 | Helped |
| use class-agnostic rather than class-specific box regression (ablation) | I3D head · AVA v2.1 val | mAP: 19.2 -> 21.3 | Helped |
| add data augmentation (ablation) | I3D head · AVA v2.1 val | mAP: 16.6 -> 21.3 | Helped |
| add Kinetics pre-training (ablation) | I3D head · AVA v2.1 val | mAP: 19.1 -> 21.3 | Helped |
| use 2 heads rather than 6 with 6 Tx layers (ablation) | Tx head with GT boxes · AVA v2.1 val, GT person boxes | mAP: 26.5 -> 29.1 | Helped |
| replace the I3D trunk with R3D (ablation) | Tx head with QPr average and GT boxes · AVA v2.1 val, GT person boxes | mAP: 28.5 -> 26.6 | Hurt |
| extend test-time context from 64 to 96 frames | Tx+I3D head · AVA v2.1 test | mAP: 24.60 -> 24.93 | Helped |
Closest papers by meaning
- Object-Region Video Transformers (CVPR 2022 · similarity 0.89)
- VideoNet: A Large-Scale Dataset for Domain-Specific Action Recognition (CVPR 2026 · similarity 0.89)
- AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (CVPR 2018 · similarity 0.88)
- Pulling Actions out of Context: Explicit Separation for Effective Combination (CVPR 2018 · similarity 0.88)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (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.