Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks
Yan Liu, Lingqiao Liu, Peng Wang, Pingping Zhang, Yinjie Lei
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 unlabeled images for surrogate-task feature learning (vs labeled only) (ablation) | IRAST · ShanghaiTech Part A | MAE: 94.1 -> 86.9 | Improved |
| add unlabeled images for surrogate-task feature learning (vs labeled only) (ablation) | IRAST · UCF-QNRF | MAE: 140.8 -> 135.6 | Improved |
| add inter-relationship constraint to pseudo-label generation (ablation) | IRAST · ShanghaiTech Part A | MAE: 93.5 -> 86.9 | Improved |
| add inter-relationship constraint to pseudo-label generation (ablation) | IRAST · UCF-QNRF | MAE: 139.8 -> 135.6 | Improved |
| use 3 binary surrogate tasks rather than one multi-class segmentation (MSST) (ablation) | IRAST · ShanghaiTech Part A | MAE: 91.5 -> 86.9 | Improved |
| use 3 binary surrogate tasks rather than one multi-class segmentation (MSST) (ablation) | IRAST · UCF-QNRF | MAE: 140.0 -> 135.6 | Improved |
| raise confidence threshold t_p from 0.6 to 0.9 (ablation) | IRAST · ShanghaiTech Part A | MAE: 88.4 -> 86.9 | No clear effect |
| raise confidence threshold t_p from 0.6 to 0.9 (ablation) | IRAST · UCF-QNRF | MAE: 137.2 -> 135.6 | No clear effect |
| add IRAST on top of CSRNet feature extractor | CSRNet · ShanghaiTech Part A | MAE: 98.3 -> 86.9 | Improved |
| add IRAST on top of SPN feature extractor (ablation) | SPN · ShanghaiTech Part A | MAE: 88.5 -> 83.9 | Improved |
| add IRAST on top of SPN feature extractor (ablation) | SPN · UCF-QNRF | MAE: 138.0 -> 128.4 | Improved |
| use 3 surrogate tasks instead of 1 (ablation) | IRAST · ShanghaiTech Part A | MAE: 89.8 -> 86.9 | Improved |
Closest papers by meaning
- Spatial Uncertainty-Aware Semi-Supervised Crowd Counting (ICCV 2021 · similarity 0.93)
- Reducing Spatial Labeling Redundancy for Active Semi-Supervised Crowd Counting (TPAMI 2022 · similarity 0.92)
- Completely Self-Supervised Crowd Counting via Distribution Matching (ECCV 2022 · similarity 0.92)
- Semi-Supervised Counting via Pixel-by-Pixel Density Distribution Modeling (TPAMI 2025 · similarity 0.91)
- Optimal Transport Minimization: Crowd Localization on Density Maps for Semi-Supervised Counting (CVPR 2023 · similarity 0.91)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.90)
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.
