Research map / Video / Action recognition
TURN TAP: Temporal Unit Regression Network for Temporal Action Proposals
Jiyang Gao, Zhenheng Yang, Kan Chen, Chen Sun, Ram Nevatia
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 |
|---|---|---|---|
| use unit size 16 instead of 32 for flow-feature TURN (ablation) | TURN-FL · THUMOS-14 proposals | AR@F=1.0 (%): 42.4 -> 43.5 (FPS 260.6 -> 129.4) | Improved |
| use flow CNN features instead of C3D features (ablation) | TURN unit-16 · THUMOS-14 proposals | AR@F=1.0 (%): 39.3 -> 43.5 (FPS 880.8 -> 129.4) | Improved |
| use flow CNN proposals instead of C3D proposals with SVM localizer (ablation) | TURN-16 · THUMOS-14 localization (our SVM) | mAP@tIoU=0.5 (%): 16.4 -> 17.8 | Improved |
| use flow CNN proposals instead of C3D proposals with S-CNN localizer (ablation) | TURN-16 · THUMOS-14 localization (S-CNN localizer) | mAP@tIoU=0.5 (%): 22.5 -> 25.6 | Improved |
| replace SCNN-prop proposals with TURN-FL-16 proposals in S-CNN localizer | S-CNN localizer · THUMOS-14 localization (S-CNN localizer) | mAP@tIoU=0.5 (%): 19.0 -> 25.6 | Improved |
| replace SCNN-prop proposals with TURN-FL-16 proposals in SVM localizer | SVM localizer · THUMOS-14 localization (our SVM) | mAP@tIoU=0.5 (%): 14.0 -> 17.8 | Improved |
| replace sliding windows with TURN-FL-16 proposals for ActivityNet Sports localization | SVM localizer with two-stream features · ActivityNet v1.1 Sports | mAP@tIoU=0.5 (%): 27.3 -> 37.1 | Improved |
| replace sliding windows with TURN-FL-16 proposals for ActivityNet Work localization | SVM localizer with two-stream features · ActivityNet v1.1 Work | mAP@tIoU=0.5 (%): 29.6 -> 41.2 | Improved |
| train on ensemble of all four subsets instead of the subset alone (Arts test) | TURN-C3D-16 · ActivityNet v1.3 Arts | AR@F=1.0 (%): 44.30 -> 45.30 | Improved |
| train on ensemble instead of Housework only (Housework test) | TURN-C3D-16 · ActivityNet v1.3 Housework | AR@F=1.0 (%): 44.30 -> 48.12 | Improved |
| train on ensemble instead of Vehicles only (Vehicles test) | TURN-C3D-16 · ActivityNet v1.3 Vehicles | AR@F=1.0 (%): 42.22 -> 42.33 | Improved |
| train on ensemble instead of Sports only (Sports test) | TURN-C3D-16 · ActivityNet v1.3 Sports | AR@F=1.0 (%): 46.62 -> 46.72 | Improved |
Closest papers by meaning
- Temporal Action Proposal Generation with Background Constraint (AAAI 2022 · similarity 0.90)
- Action Search: Spotting Actions in Videos and Its Application to Temporal Action Localization (ECCV 2018 · similarity 0.89)
- Temporal Context Aggregation Network for Temporal Action Proposal Refinement (CVPR 2021 · similarity 0.89)
- AOE-Net: Entities Interactions Modeling with Adaptive Attention Mechanism for Temporal Action Proposals Generation (IJCV 2022 · similarity 0.89)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · similarity 0.88)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.88)
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.
