Research map / Video / Action recognition
BSN: Boundary Sensitive Network for Temporal Action Proposal Generation
Tianwei Lin, Xu Zhao, Haisheng Su, Chongjing Wang, Ming Yang
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 the proposal evaluation module (PEM) on top of TEM (ablation) | BSN · THUMOS14 proposals (ablation) | AR@50: 30.72 -> 37.46 | Improved |
| add the proposal evaluation module (PEM) on top of TEM (ablation) | BSN · THUMOS14 proposals (ablation) | AR@100: 40.52 -> 46.06 | Improved |
| add the proposal evaluation module (PEM) on top of TEM (ablation) | BSN · THUMOS14 proposals (ablation) | AR@1000: 63.04 -> 64.52 | Improved |
| build BSP feature from both boundary and center regions instead of boundary only (ablation) | BSN PEM · THUMOS14 proposals (ablation) | AR@50: 35.61 -> 37.46 | Improved |
| build BSP feature from both boundary and center regions instead of center only (ablation) | BSN PEM · THUMOS14 proposals (ablation) | AR@50: 36.80 -> 37.46 | Improved |
| build BSP feature from center region instead of boundary regions (ablation) | BSN PEM · THUMOS14 proposals (ablation) | AR@100: 44.86 -> 45.65 | Improved |
| use Soft-NMS instead of Greedy-NMS (ablation) | BSN (2-Stream) · THUMOS14 proposals | AR@50: 35.41 -> 37.46 | Improved |
| use Soft-NMS instead of Greedy-NMS (ablation) | BSN (2-Stream) · THUMOS14 proposals | AR@1000: 65.10 -> 64.52 | Hurt |
| replace two-stream features with C3D features | BSN + Soft-NMS · THUMOS14 proposals | AR@50: 37.46 -> 29.58 | Hurt |
| train on seen classes only instead of seen+unseen, evaluate on unseen classes | BSN (C3D) · ActivityNet-1.3 generalization (C3D) unseen val | AR@100: 71.84 -> 71.32 | No clear effect |
| train on seen classes only instead of seen+unseen, evaluate on unseen classes | BSN (C3D) · ActivityNet-1.3 generalization (C3D) unseen val | AUC: 63.99 -> 63.38 | No clear effect |
| use BSN proposals instead of TURN proposals with UntrimmedNet classifier | detector · THUMOS14 detection (test) | mAP@0.5: 24.5 -> 36.9 | Improved |
Closest papers by meaning
- BMN: Boundary-Matching Network for Temporal Action Proposal Generation (ICCV 2019 · similarity 0.94)
- BSN++: Complementary Boundary Regressor with Scale-Balanced Relation Modeling for Temporal Action Proposal Generation (AAAI 2021 · similarity 0.94)
- Boundary Content Graph Neural Network for Temporal Action Proposal Generation (ECCV 2020 · similarity 0.92)
- Fast Learning of Temporal Action Proposal via Dense Boundary Generator (AAAI 2020 · similarity 0.91)
- Efficient Temporal Action Segmentation via Boundary-aware Query Voting (NeurIPS 2024 · similarity 0.86)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · 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.
