Research map / Learning theory and algorithms / Optimal transport
A Generalized Loss Function for Crowd Counting and Localization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the full unbalanced optimal-transport loss instead of L2 loss | VGG19 crowd counter · UCF-QNRF | MAE: 98.7 -> 84.3 | Helped |
| Use the full unbalanced optimal-transport loss instead of Bayesian loss | VGG19 crowd counter · UCF-QNRF | MAE: 88.8 -> 84.3 | Helped |
| Use perspective-guided transport cost rather than fixed exponential cost (ablation) | Unbalanced OT counting loss · JHU-CROWD++ | MAE: 64 -> 59 | Helped |
| Use exponential distance cost rather than standard Euclidean distance (ablation) | Unbalanced OT counting loss · JHU-CROWD++ | MAE | Helped |
| Use unbalanced OT rather than balanced OT (ablation) | Crowd density-map regression · JHU-CROWD++ | MAE | Helped |
| Reduce TV-loss weight by 10x with perspective-guided cost (ablation) | DM-count loss with perspective-guided cost · JHU-CROWD++ | MAE: MAE 66 | Helped |
| Increase blur factor ε for L2 density-map supervision (ablation) | L2 crowd-counting loss · UCF-QNRF | MAE | Hurt |
| Add entropic regularization H(P) (ablation) | Unbalanced OT crowd-counting loss · UCF-QNRF | MAE: 91.1 -> 84.3 | Helped |
| Add pixel-wise term D1 (ablation) | Unbalanced OT crowd-counting loss · UCF-QNRF | MAE: 85.0 -> 84.3 | Helped |
| Add point-wise term D2 (ablation) | Unbalanced OT crowd-counting loss · UCF-QNRF | MAE: 85.4 -> 84.3 | Helped |
| Use point-wise L1 and pixel-wise L2 divergences instead of other tested combinations (ablation) | Unbalanced OT loss · UCF-QNRF | MAE: 84.3 | Helped |
| Increase τ, making the transport loss relatively less dominant (ablation) | Unbalanced OT crowd-counting loss · UCF-QNRF | MAE: 84.3; 88.73; 88.52; 99.22 | Mixed |
Closest papers by meaning
- Learning to Count via Unbalanced Optimal Transport (AAAI 2021 · similarity 0.92)
- Optimal Transport Minimization: Crowd Localization on Density Maps for Semi-Supervised Counting (CVPR 2023 · similarity 0.91)
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks (CVPR 2019 · similarity 0.88)
- Single Domain Generalization for Crowd Counting (CVPR 2024 · similarity 0.86)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.86)
- CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models (CVPR 2024 · similarity 0.86)
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.