Research map / Video / Action recognition
Temporal Action Detection With Multi-Level Supervision
Baifeng Shi, Qi Dai, Judy Hoffman, Kate Saenko, Trevor Darrell, Huijuan Xu
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 unsupervised foreground attention (UFA) (ablation) | Mean Teacher SSAD · THUMOS14 (50% labeled) | mAP@0.5: 24.62 -> 25.43 | Improved |
| add unsupervised foreground attention (UFA) (ablation) | MixMatch SSAD · THUMOS14 (50% labeled) | mAP@0.5: 24.79 -> 26.18 (mAP@0.7 6.60 -> 7.10) | Improved |
| add unsupervised foreground attention (UFA) (ablation) | FixMatch SSAD · THUMOS14 (50% labeled) | mAP@0.5: 23.58 -> 24.68 | Improved |
| add unsupervised foreground attention (UFA) (ablation) | Mean Teacher SSAD · ActivityNet1.2 (10% labeled) | mAP@AVG: 11.49 -> 11.74 | Improved |
| add unsupervised foreground attention (UFA) (ablation) | MixMatch SSAD · ActivityNet1.2 (10% labeled) | mAP@AVG: 11.05 -> 11.84 | Improved |
| add unsupervised foreground attention (UFA) (ablation) | FixMatch SSAD · ActivityNet1.2 (10% labeled) | mAP@AVG: 11.88 -> 12.27 | Improved |
| use Gaussian or CIS attention instead of no attention (ablation) | MixMatch SSAD · THUMOS14 (50% labeled) | mAP@0.5: 24.79 -> 24.59 (Gaussian); 24.79 -> 25.01 (CIS) | No clear effect |
| add FixMatch semi-supervision without UFA | supervised-only SSN · THUMOS14 (50% labeled) | mAP@0.5: 24.74 -> 23.58 | Hurt |
| add weakly-labeled data with video-level loss (OSAD) | SSAD · THUMOS14 (10/20/70) | mAP@0.5: 8.96 -> 11.37 | Improved |
| add information bottleneck (IB) (ablation) | OSAD · THUMOS14 (10/20/70) | mAP@0.5: 11.37 -> 12.97 | Improved |
| add weakly-labeled data with video-level loss (OSAD) | SSAD · ActivityNet1.2 (10/20/70) | mAP@AVG: 11.84 -> 11.78 | No clear effect |
| add information bottleneck (IB) (ablation) | OSAD · ActivityNet1.2 (10/20/70) | mAP@AVG: 11.78 -> 12.30 | Improved |
Closest papers by meaning
- Learning Temporal Action Proposals With Fewer Labels (ICCV 2019 · similarity 0.92)
- WOAD: Weakly Supervised Online Action Detection in Untrimmed Videos (CVPR 2021 · similarity 0.91)
- Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions (ECCV 2022 · similarity 0.90)
- Action-Agnostic Point-Level Supervision for Temporal Action Detection (AAAI 2025 · similarity 0.89)
- Semi-supervised Active Learning for Video Action Detection (AAAI 2024 · similarity 0.88)
- Stable Mean Teacher for Semi-supervised Video Action Detection (AAAI 2025 · 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.
