Research map / 3D vision / Depth and stereo
Adaptive Density Map Generation for Crowd Counting
Jia Wan, Antoni Chan
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 density map refinement (adaptive-kernel maps) | CSRNet · ShanghaiTech A | MAE: 66.4 -> 64.2 | Improved |
| add density map refinement (fixed-kernel maps) | MCNN · ShanghaiTech A | MAE: 95.4 -> 102.3 | Hurt |
| add density map refinement (adaptive-kernel maps) | MCNN · ShanghaiTech A | MAE: 103.3 -> 96.7 | Improved |
| add density map refinement (fixed kernel) | FCN · ShanghaiTech B | MAE: 18.8 -> 15.2 | Improved |
| add jointly trained generator instead of fixed sigma=16 kernel | SFCN · ShanghaiTech A | MAE: 70.8 -> 68.4 | Improved |
| add jointly trained generator instead of adaptive kernel | CSRNet · ShanghaiTech B | MAE: 10.6 -> 8.1 | Improved |
| add jointly trained generator instead of fixed sigma=16 kernel | MCNN · ShanghaiTech B | MAE: 18.7 -> 17.7 | No clear effect |
| add jointly trained generator instead of fixed sigma=4 kernel | FCN · ShanghaiTech B | MAE: 13.8 -> 13.9 | No clear effect |
| add global count loss instead of local spatial loss (ablation) | density map generator · ShanghaiTech A | MAE: 68.8 -> 64.7 | Improved |
| add global count loss instead of hard normalization (ablation) | density map generator · ShanghaiTech A | MAE: 112.8 -> 64.7 | Improved |
| use fixed instead of learnable initial Gaussian kernels (ablation) | density map generator · ShanghaiTech A | MAE: 74.7 -> 64.7 | Improved |
| add self-attention instead of image-based attention (ablation) | density map generator · ShanghaiTech A | MAE: 66.9 -> 64.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- CSRNet — CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes (CVPR 2018) · better on 6 of 7 shared results
- SANet — SANet: Scene Agnostic Network for Camera Localization (ICCV 2019) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Kernel-Based Density Map Generation for Dense Object Counting (TPAMI 2020 · similarity 0.97)
- Iterative Crowd Counting (ECCV 2018 · similarity 0.93)
- Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (ECCV 2018 · similarity 0.90)
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.90)
- Learn to Scale: Generating Multipolar Normalized Density Maps for Crowd Counting (ICCV 2019 · similarity 0.90)
- CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models (CVPR 2024 · similarity 0.89)
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.
