Research map / Video / Action recognition
Action Tubelet Detector for Spatio-Temporal Action Localization
Vicky Kalogeiton, Philippe Weinzaepfel, Vittorio Ferrari, Cordelia Schmid
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 tubelet input of K=6 frames instead of K=1 per-frame detection (ablation) | ACT-detector · UCF-Sports fast actors | frame-mAP: 68.1 -> 76.8 | Improved |
| add tubelets K=6 vs K=1 for fast actors (ablation) | ACT-detector · J-HMDB fast actors | frame-mAP: 49.0 -> 58.0 | Improved |
| add tubelets K=6 vs K=1 for fast actors (ablation) | ACT-detector · UCF-101 fast actors | frame-mAP: 67.3 -> 70.7 | Improved |
| use sequence length K=6 instead of K=1 (ablation) | ACT-detector · UCF-Sports, J-HMDB, UCF-101 | MABO: MABO increases by 5% on UCF-Sports, 2% on J-HMDB and 5% on UCF-101 | Improved |
| increase tubelet length up to K=6 or 8 (ablation) | ACT-detector · three datasets | frame-mAP: gain up to 10% | Improved |
| increase tubelet length beyond K=8 (ablation) | ACT-detector · J-HMDB, UCF-101 | frame-mAP | Hurt |
| late fusion of RGB and flow instead of union fusion (ablation) | ACT-detector · three datasets | frame-mAP: gain between 1% and 4% | Improved |
| add flow stream to RGB stream (ablation) | ACT-detector · three datasets | frame-mAP | Improved |
| stack 5 flow images instead of 1 as motion input (ablation) | ACT-detector · three datasets | frame-mAP: +1% frame-mAP | Improved |
| use tubelet linking instead of linking individual boxes (ablation) | ACT-detector · three datasets | video-mAP: consistent gain of 1% | Improved |
| use ACT-detector instead of prior SSD detector at high IoU | J-HMDB · J-HMDB video-mAP IoU 0.75 | video-mAP: 44.5 -> 52.1 | Improved |
| use ACT-detector instead of prior SSD detector | UCF-101 · UCF-101 video-mAP IoU 0.5 | video-mAP: 46.3 -> 51.4 | Improved |
Closest papers by meaning
- TubeR: Tubelet Transformer for Video Action Detection (CVPR 2022 · similarity 0.92)
- Actions as Moving Points (ECCV 2020 · similarity 0.91)
- End-to-End Spatio-Temporal Action Localisation with Video Transformers (CVPR 2024 · similarity 0.91)
- CFAD: Coarse-to-Fine Action Detector for Spatiotemporal Action Localization (ECCV 2020 · similarity 0.91)
- Tubelets: Unsupervised Action Proposals from Spatiotemporal Super-Voxels (IJCV 2017 · similarity 0.90)
- Progressive Cross-Stream Cooperation in Spatial and Temporal Domain for Action Localization (TPAMI 2020 · similarity 0.89)
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.
