Research map / Architectures and efficiency / CNN design
Scale Aggregation Network for Accurate and Efficient Crowd Counting
Xinkun Cao, Zhipeng Wang, Yanyun Zhao, Fei Su
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 |
|---|---|---|---|
| use scale aggregation encoder (FME) instead of multi-column MCNN features (ablation) | MCNN baseline · ShanghaiTech Part A | MAE: 109.4 -> 90.5 | Improved |
| add density map estimator (DME) for high-resolution output (ablation) | MCNN baseline · ShanghaiTech Part A | MAE: 109.4 -> 83.3 | Improved |
| add instance normalization layers (ablation) | MCNN+DME · ShanghaiTech Part A | MAE: 83.3 -> 77.6 | Improved |
| add instance normalization layers (ablation) | SANet · ShanghaiTech Part A | MAE: fails to converge -> 71.0 | Improved |
| test on patches instead of whole images (ablation) | SANet (L_E) · ShanghaiTech Part A | MAE: 116.8 -> 71.0 | Improved |
| add local pattern consistency loss L_C (patch testing) (ablation) | SANet · ShanghaiTech Part A | MAE: 71.0 -> 67.0 | Improved |
| add local pattern consistency loss L_C (image testing) (ablation) | SANet · ShanghaiTech Part A | MAE: 116.8 -> 88.1 | Improved |
| use SANet instead of CSRNet | crowd counting model · ShanghaiTech Part B | MAE: 10.6 -> 8.4 | Improved |
| drop perspective maps in ground truth generation (ablation) | SANet · WorldExpo'10 | MAE: 8.6 -> 8.2 | Improved |
| use SANet instead of CSRNet | crowd counting model · UCF_CC_50 | MAE: 266.1 -> 258.4 | Improved |
| use batch normalization instead of instance normalization (ablation) | SANet · ShanghaiTech Part A | MAE | Hurt |
Closest papers by meaning
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks (CVPR 2019 · similarity 0.91)
- Crowd Counting With Deep Structured Scale Integration Network (ICCV 2019 · similarity 0.91)
- Attention Scaling for Crowd Counting (CVPR 2020 · similarity 0.90)
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.90)
- Multi-Level Bottom-Top and Top-Bottom Feature Fusion for Crowd Counting (ICCV 2019 · similarity 0.90)
- Counting without Scale: Scale-Consistent Error Correction for Crowd Counting (NeurIPS 2026 · 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.
