Research map / Video / Action recognition
Divide and Conquer for Single-Frame Temporal Action Localization
Chen Ju, Peisen Zhao, Siheng Chen, Ya Zhang, Yanfeng Wang, Qi Tian
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 instance counting stage that divides video into clips by seedframes (ablation) | one-stage frame-based baseline · THUMOS14 (manual single-frame labels) | AVG mAP: 39.6 -> 41.7 | Improved |
| use proposal-based representation instead of frame-based (ablation) | two-stage framework · THUMOS14 (manual single-frame labels) | AVG mAP: 41.7 -> 44.3 | Improved |
| add instance counting stage (division) (ablation) | one-stage baseline · THUMOS14 (manual single-frame labels) | precision at IoU 0.5: +9.3% | Improved |
| add instance counting stage (division) (ablation) | one-stage baseline · THUMOS14 (manual single-frame labels) | recall at IoU 0.5: +3.6% | Improved |
| add background-aware loss L_bg (ablation) | ReSim-free two-stage model (location estimation stage) · THUMOS14 (manual single-frame labels) | AVG mAP: 4.7% average mAP | Improved |
| add center offset Delta p (ablation) | location estimation stage · THUMOS14 (manual single-frame labels) | mAP@IoU0.5: 0.8% in mAP@0.5 | Improved |
| use learnable gate-approximation network instead of Gaussian-shaped mask (ablation) | mask generator · THUMOS14 (manual single-frame labels) | AVG mAP: 42.7 -> 44.3 | Improved |
| replace one-stage SF-Net with two-stage divide and conquer | SF-Net · THUMOS14 (manual single-frame labels) | AVG mAP: 40.80 -> 44.34 | Improved |
| replace one-stage SF-Net with two-stage divide and conquer | SF-Net · THUMOS14 (simulated uniform single-frame labels) | AVG mAP: 41.24 -> 42.93 | Improved |
| replace SF-Net with two-stage divide and conquer | SF-Net · GTEA | AVG mAP: 31.0 -> 33.7 | Improved |
| replace SF-Net with two-stage divide and conquer | SF-Net · BEOID | AVG mAP: 30.1 -> 34.9 | Improved |
Closest papers by meaning
- Temporal Action Localization by Structured Maximal Sums (CVPR 2017 · similarity 0.90)
- SF-Net: Single-Frame Supervision for Temporal Action Localization (ECCV 2020 · similarity 0.90)
- ASM-Loc: Action-Aware Segment Modeling for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.90)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.90)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.89)
- Test-Time Zero-Shot Temporal Action Localization (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.
