Research map / Video / Action recognition
Recurrent Tubelet Proposal and Recognition Networks for Action Detection
Dong Li, Zhaofan Qiu, Qi Dai, Ting Yao, Tao Mei
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 Recurrent Tubelet Proposal (RTP) in place of independent per-frame proposals (ablation) | Faster R-CNN · UCF-Sports | mAP: 83.8 -> 85.2 | Improved |
| add RTP (ablation) | Faster R-CNN · AVA | frame-mAP: 15.6 -> 16.8 | Improved |
| add LSTM temporal modeling to recognition (ablation) | Faster R-CNN + RTP · UCF-101 | mAP: 57.9 -> 59.2 | Improved |
| add multi-channel H/I/S recognition (HIS) (ablation) | Faster R-CNN + RTP + LSTM · UCF-Sports | mAP: 85.7 -> 87.6 | Improved |
| add optical flow stream (ablation) | RTPR RGB-only · J-HMDB (split 1) | mAP: 61.7 -> 80.7 | Improved |
| add optical flow stream (ablation) | RTPR RGB-only · AVA | frame-mAP: 19.4 -> 20.1 | No clear effect |
| add optical flow stream (ablation) | RTPR RGB-only · UCF-Sports | mAP: 87.6 -> 97.8 | Improved |
| use human-object interaction channel I alone instead of human-only channel H (ablation) | RTR · UCF-101 | mAP (fusion): 69.3 -> 74.9 | Improved |
| fuse all three channels H, I, S instead of channel I alone (ablation) | RTR · UCF-101 | mAP (fusion): 74.9 -> 76.3 | Improved |
| scene channel S versus human channel H (ablation) | RTR · AVA | mAP (RGB): 14.4 -> 13.7 | Hurt |
| replace VGG-16 backbone with ResNet-101 | RTPR · AVA | frame-mAP (fusion): 20.1 -> 22.3 | Improved |
| replace VGG-16 backbone with ResNet-101 | RTPR · J-HMDB | video-mAP@0.5: 80.7 -> 81.3 | Improved |
Closest papers by meaning
- TubeR: Tubelet Transformer for Video Action Detection (CVPR 2022 · similarity 0.92)
- Tubelets: Unsupervised Action Proposals from Spatiotemporal Super-Voxels (IJCV 2017 · similarity 0.89)
- Object Detection in Videos With Tubelet Proposal Networks (CVPR 2017 · similarity 0.89)
- Tube Convolutional Neural Network (T-CNN) for Action Detection in Videos (ICCV 2017 · similarity 0.89)
- AMTnet: Action-Micro-Tube Regression by End-To-End Trainable Deep Architecture (ICCV 2017 · similarity 0.88)
- Taylor Videos for Action Recognition (ICML 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.
