Research map / Label-efficient and robust learning / Semi-supervised learning
Calibrating Uncertainty for Semi-Supervised Crowd Counting
Chen LI, Xiaoling Hu, Shahira Abousamra, Chao Chen
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 uncertainty-based filtering of pseudo-labels (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 83.28 -> 70.76 | Improved |
| use ASM spatial matching surrogate instead of softmax confidence (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 75.47 -> 70.76 | Improved |
| use matching distance instead of absolute counting difference (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 71.90 -> 70.76 | Improved |
| use spatial matching distance instead of discrete Wasserstein (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 72.47 -> 70.76 | Improved |
| use spatial matching distance instead of Hungarian loss (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 72.95 -> 70.76 | Improved |
| accumulate matching distance over training epochs (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 71.47 -> 70.76 | Improved |
| use Lambdaloss ranking loss instead of L1 loss (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 72.60 -> 70.76 | Improved |
| add EMA teacher (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 75.38 -> 70.76 | Improved |
| add strong augmentation (cutout) on pseudo-labeled patches (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 72.77 -> 70.76 | Improved |
| add uncertainty loss to supervised training only (ablation) | Sup.only · ShanghaiTech Part-A (10% labels) | MAE: 77.74 -> 74.84 | Improved |
| add pseudo-labeling with uncertainty filtering on top of supervised only (ablation) | Ours · ShanghaiTech Part-A (5% labels) | MAE: 88.48 -> 74.48 | Improved |
| vary uncertainty patch size from 64 (ablation) | Ours · ShanghaiTech Part-A (10% labels) | MAE: 70.76 at 64 vs 76.31 at 8, 75.41 at 128 | Mixed |
Closest papers by meaning
- Spatial Uncertainty-Aware Semi-Supervised Crowd Counting (ICCV 2021 · similarity 0.92)
- Crowd Counting with Decomposed Uncertainty (AAAI 2020 · similarity 0.89)
- Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks (ECCV 2020 · similarity 0.88)
- CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models (CVPR 2024 · similarity 0.85)
- Taste More, Taste Better: Diverse Data and Strong Model Boost Semi-Supervised Crowd Counting (CVPR 2025 · similarity 0.85)
- Single Domain Generalization for Crowd Counting (CVPR 2024 · similarity 0.85)
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.
