Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Counting via Pixel-by-Pixel Density Distribution Modeling
Hui Lin, Zhiheng Ma, Rongrong Ji, Yaowei Wang, Zhou Su, Xiaopeng Hong, Deyu Meng
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 inter-branch ECR on unlabeled data (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 129.5 -> 115.3 | Improved |
| add inter-branch ECR on unlabeled data (ablation) | P3Net · UCF-QNRF (40% labeled) | MAE: 97.8 -> 90.0 | Improved |
| use L2 norm instead of L1 (Wasserstein) in PDM loss (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 127.0 -> 115.3 | Improved |
| use PDM loss instead of cross entropy (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 125.4 -> 115.3 | Improved |
| use PDM loss instead of MSE loss (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 132.8 -> 115.3 | Improved |
| use PDM loss instead of Bayesian loss (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 136.5 -> 115.3 | Improved |
| use soft expectation per branch instead of hard maximum (ablation) | P3Net · UCF-QNRF (5% labeled, labeled only) | MAE: 134.5 -> 130.0 | Improved |
| weight branches by confidence omega instead of plain average (ablation) | P3Net · UCF-QNRF (5% labeled, labeled only) | MAE: 130.0 -> 129.5 | Improved |
| use class-specific density tokens instead of randomly initialized queries (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 137.2 -> 115.3 | Improved |
| use UEP partition instead of UniformLen intervals (ablation) | P3Net · ShanghaiTech A (5% labeled) | MAE: 87.1 -> 85.5 | Improved |
| set lambda to 0.01 instead of 0.1 (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 122.5 -> 115.3 | Improved |
| use confidence threshold 0.5 instead of 0 in ECR mask (ablation) | P3Net · UCF-QNRF (5% labeled) | MAE: 119.8 -> 115.3 | Improved |
Closest papers by meaning
- Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks (ECCV 2020 · similarity 0.91)
- Completely Self-Supervised Crowd Counting via Distribution Matching (ECCV 2022 · similarity 0.90)
- Reducing Spatial Labeling Redundancy for Active Semi-Supervised Crowd Counting (TPAMI 2022 · similarity 0.89)
- Spatial Uncertainty-Aware Semi-Supervised Crowd Counting (ICCV 2021 · similarity 0.89)
- Single Domain Generalization for Crowd Counting (CVPR 2024 · similarity 0.87)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.87)
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.
