Research map / Architectures and efficiency / Neural architecture search
NAS-Count: Counting-by-Density with Neural Architecture Search
Yutao Hu ¹, Xiaolong Jiang ², Xuhui Liu, Baochang Zhang, Jungong Han, Xianbin Cao ², David Doermann
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 searched AMSNet encoder instead of baseline encoder (ablation) | baseline encoder + decoder · ShanghaiTech Part_A (ablation) | MAE: 69.1 -> 60.3 | Improved |
| use searched AMSNet encoder instead of baseline encoder (ablation) | baseline encoder + decoder · ShanghaiTech Part_A (ablation) | PSNR: 23.54 -> 25.82 | Improved |
| use searched AMSNet decoder instead of baseline decoder (ablation) | baseline encoder + decoder · ShanghaiTech Part_A (ablation) | MAE: 69.1 -> 62.4 | Improved |
| use searched AMSNet decoder instead of baseline decoder (ablation) | baseline encoder + decoder · ShanghaiTech Part_A (ablation) | PSNR: 23.54 -> 24.75 | Improved |
| search encoder on counting instead of using classification-searched backbone (ablation) | classification encoder · ShanghaiTech Part_A (ablation) | MAE: 67.8 -> 60.3 | Improved |
| search encoder on counting instead of using classification-searched backbone (ablation) | classification encoder · ShanghaiTech Part_A (ablation) | PSNR: 23.67 -> 25.82 | Improved |
| add hand-designed pyramid SAL loss over MSE (ablation) | AMSNet · ShanghaiTech Part_A (ablation) | MAE: 58.5 -> 57.6 | Improved |
| replace SAL with searched SPPLoss (ablation) | AMSNet · ShanghaiTech Part_A (ablation) | MAE: 57.6 -> 56.7 | Improved |
| add searched SPPLoss over MSE (ablation) | AMSNet · ShanghaiTech Part_A (ablation) | PSNR: 26.17 -> 27.03 | Improved |
| reduce output channels C from 512 to 256 (AMSNet_light) (ablation) | AMSNet · ShanghaiTech Part_A | MAE: 56.7 -> 61.3 | Hurt |
| reduce output channels C from 512 to 256 (AMSNet_light) (ablation) | AMSNet · ShanghaiTech Part_A | model size: 3.79MB -> 1.51MB | Improved |
Closest papers by meaning
- Scale Aggregation Network for Accurate and Efficient Crowd Counting (ECCV 2018 · similarity 0.88)
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks (CVPR 2019 · similarity 0.87)
- Attention Scaling for Crowd Counting (CVPR 2020 · similarity 0.87)
- Crowd Counting via Adversarial Cross-Scale Consistency Pursuit (CVPR 2018 · similarity 0.86)
- Neural Architecture Search for Dense Prediction Tasks in Computer Vision (IJCV 2023 · similarity 0.86)
- Gaussian Density Splatting Network (NeurIPS 2026 · similarity 0.85)
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.
