Research map / Label-efficient and robust learning / Semi-supervised learning
Learning from Noisy Pseudo Labels for Semi-Supervised Temporal Action Localization
Kun Xia, Le Wang, Sanping Zhou, Gang Hua, Wei Tang
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 Noisy Label Ranking (confidence divided by boundary variance) (ablation) | ActionFormer self-training · THUMOS14, 40% labeled | average mAP (%): 47.1 -> 48.5 | Improved |
| add Noisy Label Filtering (adaptive class-balanced selection) (ablation) | ActionFormer self-training · THUMOS14, 40% labeled | average mAP (%): 48.5 -> 50.3 | Improved |
| add Noisy Label Learning (temporal consistency loss) (ablation) | ActionFormer self-training · THUMOS14, 40% labeled | average mAP (%): 50.3 -> 51.5 | Improved |
| add all three NPL components at 10% labels (ablation) | ActionFormer self-training · THUMOS14, 10% labeled | average mAP (%): 16.8 -> 20.3 | Improved |
| replace single threshold 0.3 with adaptive filtering (ablation) | NPL · THUMOS14, 40% labeled | average mAP (%): 49.2 -> 51.5 | Improved |
| replace fixed top-10 with adaptive filtering (ablation) | NPL · THUMOS14, 40% labeled | average mAP (%): 49.7 -> 51.5 | Improved |
| add temporal consistency loss (ablation) | NPL · THUMOS14, 60% unlabeled | average mAP (%): 44.7 -> 45.9 | Improved |
| vary Npos from 15 to 5 or 30 (ablation) | NPL · THUMOS14, 40% labeled | average mAP (%): 51.5 -> 51.0 (Npos 5), 50.7 (Npos 30) | Hurt |
| apply NPL to pseudo labels at 10% labeled | ActionFormer · THUMOS14, 10% labeled | average tIoU of pseudo labels (%): 27.4 -> 36.2 | Improved |
| add NPL to semi-supervised ActionFormer at 10% on ActivityNet | ActionFormer · ActivityNet v1.3 val, 10% labeled | average mAP (%): 28.8 -> 32.5 | Improved |
| score pseudo labels by boundary variance instead of foreground score | Noisy Label Ranking · THUMOS14 pseudo labels | Pearson correlation with IoU: 0.449 -> 0.496 | Improved |
Closest papers by meaning
- Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · similarity 0.93)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.91)
- Semi-supervised Active Learning for Video Action Detection (AAAI 2024 · similarity 0.89)
- Stable Mean Teacher for Semi-supervised Video Action Detection (AAAI 2025 · similarity 0.88)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.88)
- Realigning Confidence with Temporal Saliency Information for Point-Level Weakly-Supervised 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.
