Research map / Video / Action recognition
Proposal-Free Temporal Action Detection via Global Segmentation Mask Learning
Sauradip Nag, Xiatian Zhu, Yi-Zhe Song, Tao Xiang
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 Transformer instead of 1D CNN snippet embedding (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 26.4 -> 36.1 | Improved |
| add Transformer instead of MS-TCN (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 33.8 -> 36.1 | Improved |
| add multi-scale snippets {1,2} to single-scale Transformer (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 36.1 -> 36.5 | Improved |
| add third scale {1,2,4} (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 36.5 -> 36.4 | No clear effect |
| add segmentation mask branch (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 28.9 -> 36.5 | Improved |
| use own classification branch instead of external UntrimmedNet (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 31.8 -> 36.5 | Improved |
| use global mask instead of 1D actionness (ablation) | TAGS · ActivityNet-v1.3 | mAP@0.5: 44.8 -> 56.3 | Improved |
| use global mask instead of 1D actionness (ablation) | TAGS · ActivityNet-v1.3 | avg mAP: 27.1 -> 36.5 | Improved |
| provide ground-truth class labels at test | TAGS · ActivityNet-v1.3 | avg mAP: 36.5 -> 43.8 | Improved |
| provide ground-truth masks at test | TAGS · ActivityNet-v1.3 | avg mAP: 36.5 -> 48.5 | Improved |
| replace BMN proposal pipeline with proposal-free TAGS | BMN · ActivityNet-v1.3 | training time: 6.45 hr -> 0.26 hr | Improved |
Closest papers by meaning
- Semi-Supervised Temporal Action Detection with Proposal-Free Masking (ECCV 2022 · similarity 0.91)
- Learning Temporal Action Proposals With Fewer Labels (ICCV 2019 · similarity 0.90)
- Zero-Shot Temporal Action Detection via Vision-Language Prompting (ECCV 2022 · similarity 0.89)
- Relaxed Transformer Decoders for Direct Action Proposal Generation (ICCV 2021 · similarity 0.89)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · similarity 0.89)
- 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.
