Research map / Image generation / Image restoration
Learn to Scale: Generating Multipolar Normalized Density Maps for Crowd Counting
Chenfeng Xu, Kai Qiu, Jianlong Fu, Song Bai, Yongchao Xu, Xiang Bai
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 the learning to scale module (L2SM) to SPN | SPN · ShanghaiTech Part A | MAE: 70.0 -> 64.2 | Improved |
| add L2SM to SPN | SPN · ShanghaiTech Part B | MAE: 9.1 -> 7.2 | Improved |
| add L2SM to SPN | SPN · UCF_CC_50 | MAE: 204.7 -> 188.4 | Improved |
| add L2SM to SPN | SPN · UCF-QNRF | MAE: 110.3 -> 104.7 | Improved |
| add scale preserving pyramid features (SPN) to VGG16 | VGG16 baseline · UCF-QNRF | MAE: 120.6 -> 110.3 | Improved |
| add L2SM to SPN, cross-dataset Part A to UCF_CC_50 | SPN · Part A -> UCF_CC_50 | MAE: 368.3 -> 332.4 | Improved |
| add L2SM to SPN, cross-dataset UCF-QNRF to Part A | SPN · UCF-QNRF -> Part A | MAE: 87.9 -> 73.4 | Improved |
| increase number of centers from 1 to 3 (G=5) (ablation) | SPN+L2SM · ShanghaiTech Part A | MAE: 67.2 -> 64.2 | Improved |
| increase number of centers from 3 to 5, centralizing every region (G=5) (ablation) | SPN+L2SM · ShanghaiTech Part A | MAE: 64.2 -> 69.8 | Hurt |
| use multipolar center loss instead of scaling directly to average density (S2AD), C=3 (ablation) | SPN+L2SM · ShanghaiTech Part A | MAE: 67.0 -> 64.2 | Improved |
| use G=5 groups instead of G=3 (C=2) (ablation) | SPN+L2SM · ShanghaiTech Part A | MAE: 65.1 -> 65.4 | No clear effect |
| vary K x K division from 2 x 2 to 8 x 8 (one center) (ablation) | SPN+L2SM · ShanghaiTech Part A | MAE: 68.0 / 67.2 / 67.9 / 68.5 | No clear effect |
Closest papers by meaning
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.91)
- Adaptive Density Map Generation for Crowd Counting (ICCV 2019 · similarity 0.90)
- Kernel-Based Density Map Generation for Dense Object Counting (TPAMI 2020 · similarity 0.89)
- Scale Aggregation Network for Accurate and Efficient Crowd Counting (ECCV 2018 · similarity 0.89)
- Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (ECCV 2018 · 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.
