Research map / Video / Action recognition
TubeR: Tubelet Transformer for Video Action Detection
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 a single tubelet query with a per-frame tubelet query set (ablation) | TubeR · UCF101-24 | video-mAP@IoU=0.5: 48.8 -> 52.9 | Helped |
| add a per-frame tubelet query set (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 26.2 -> 27.4 | Helped |
| replace standard self-attention with tubelet attention (ablation) | TubeR · UCF101-24 | video-mAP@IoU=0.5: 52.9 -> 53.8 | Helped |
| replace standard self-attention with tubelet attention (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 27.4 -> 27.7 | Helped |
| add the action switch regression head (ablation) | TubeR · UCF101-24 | video-mAP@IoU=0.5: 53.8 -> 57.7 | Helped |
| add the action switch regression head (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 27.7 -> 27.7 | No clear effect |
| add short-term context to the classification head (ablation) | TubeR · UCF101-24 | video-mAP@IoU=0.5: 57.8 -> 58.4 | No clear effect |
| add short-term context to the classification head (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 23.4 -> 27.7 | Helped |
| add long-term context after short-term context (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 27.7 -> 28.8 | Helped |
| increase the input clip from 8 to 32 frames (ablation) | TubeR · UCF101-24 | video-mAP@IoU=0.5: 53.9 -> 58.4 | Helped |
| increase the input clip from 8 to 32 frames (ablation) | TubeR · AVA 2.1 | frame-mAP@IoU=0.5: 24.4 -> 27.7 | Helped |
| increase long-term context from one to seven clips (ablation) | TubeR · AVA 2.1 | frame-mAP: 27.7 -> 28.8 | Helped |
Closest papers by meaning
- End-to-End Spatio-Temporal Action Localisation with Video Transformers (CVPR 2024 · similarity 0.90)
- Finding Action Tubes with a Sparse-to-Dense Framework (AAAI 2020 · similarity 0.89)
- Object Detection in Videos With Tubelet Proposal Networks (CVPR 2017 · similarity 0.88)
- Video-based Human-Object Interaction Detection from Tubelet Tokens (NeurIPS 2022 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.85)
- SRTube: Video-Language Pre-Training with Action-Centric Video Tube Features and Semantic Role Labeling (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.