Research map / Learning theory and algorithms / Prediction and decision losses
Bayesian Loss for Crowd Count Estimation With Point Supervision
Zhiheng Ma, Xing Wei, Xiaopeng Hong, Yihong Gong
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 |
|---|---|---|---|
| replace pixel-wise Gaussian density loss with Bayesian loss | VGG-19 density estimator · UCF-QNRF | MAE: 106.8 -> 92.9 | Improved |
| add background pixel modelling to Bayesian loss (ablation) | Bayesian loss, VGG-19 · UCF-QNRF | MAE: 92.9 -> 88.7 | Improved |
| replace Gaussian-target loss with Bayesian+ loss | VGG-19 · ShanghaiTech A | MAE: 68.6 -> 62.8 | Improved |
| replace Gaussian-target loss with Bayesian+ loss | VGG-19 · ShanghaiTech B | MAE: 8.5 -> 7.7 | Improved |
| replace Gaussian-target loss with Bayesian+ loss | VGG-19 · UCF_CC_50 | MAE: 251.6 -> 229.3 | Improved |
| limit image shorter side to 2048 pixels (ablation) | Bayesian+ · UCF-QNRF | MAE: 112.6 -> 88.7 | Improved |
| limit image shorter side to 2048 pixels (ablation) | Baseline · UCF-QNRF | MAE: 128.7 -> 106.8 | Improved |
| switch backbone from AlexNet to VGG-19 (ablation) | Bayesian+ · UCF-QNRF | MAE: 116.3 -> 88.7 | Improved |
| replace Gaussian-target loss with Bayesian+ loss | AlexNet · UCF-QNRF | MAE: 130.8 -> 116.3 | Improved |
| replace Gaussian-target loss with Bayesian+ loss, cross-dataset | VGG-19 trained on UCF-QNRF · ShanghaiTech A (cross-dataset) | MAE: 73.4 -> 69.8 | Improved |
| vary Gaussian sigma from 0.1 to 32.0 in the loss (ablation) | Bayesian loss · UCF-QNRF | MAE: less than 98.0 | No clear effect |
| vary Gaussian sigma in the target density map (ablation) | Baseline · UCF-QNRF | MAE: 118.4 to 136.2 | Hurt |
Closest papers by meaning
- A Generalized Loss Function for Crowd Counting and Localization (CVPR 2021 · similarity 0.91)
- Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (ECCV 2018 · similarity 0.90)
- Modeling Noisy Annotations for Crowd Counting (NeurIPS 2020 · similarity 0.90)
- Distribution Matching for Crowd Counting (NeurIPS 2020 · similarity 0.89)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.89)
- CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models (CVPR 2024 · 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.
