Research map / Image generation / Image restoration
Revisiting Perspective Information for Efficient Crowd Counting
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 |
|---|---|---|---|
| Add perspective-map-guided adaptive weighting of multi-scale density outputs instead of averaging (ablation) | PACNN · ShanghaiTech SHA | MAE: 76.5 -> 66.3 | Helped |
| Add perspective-map-guided adaptive weighting of multi-scale density outputs instead of averaging (ablation) | PACNN · ShanghaiTech SHB | MAE: 12.9 -> 8.9 | Helped |
| Combine multi-scale density outputs by averaging rather than using only the best single-scale output (ablation) | PACNN · ShanghaiTech SHA | MAE: 81.8 -> 76.5 | Helped |
| Combine multi-scale density outputs by averaging rather than using only the best single-scale output (ablation) | PACNN · ShanghaiTech SHB | MAE: 14.5 -> 12.9 | Helped |
| Use nonlinear tanh fitting for perspective-map ground truth instead of linear regression (ablation) | PACNN perspective-map generation · ShanghaiTech and UCF_CC_50 | MAE | Helped |
| Use raw K-NN distances as the final perspective-map values instead of fitting a nonlinear perspective function (ablation) | PACNN perspective-map generation · Dense crowd datasets | crowd-counting performance | No clear effect |
| Use the single Conv4_3 density output rather than perspective-aware multi-scale combination (ablation) | PACNN · ShanghaiTech SHA | MAE: 66.3 -> 81.8 | Hurt |
| Use the single Conv5_1_3 density output rather than perspective-aware multi-scale combination (ablation) | PACNN · ShanghaiTech SHB | MAE: 8.9 -> 14.5 | Hurt |
| Add the [cite] trained backbone to PACNN | PACNN · ShanghaiTech SHA | MAE: 66.3 -> 62.4 | Helped |
| Add the reproduced [cite] backbone to PACNN | PACNN · UCF_CC_50 | MAE: 267.9 -> 241.7 | Helped |
| Use PACNN rather than the strongest listed baseline | PACNN · WorldExpo'10 | MAE (Avg.): 8.2 -> 7.8 | Helped |
| Use PACNN rather than the strongest listed baseline | PACNN · UCSD | MAE: 1.00 -> 0.89 | Helped |
Closest papers by meaning
- Single-Image Crowd Counting via Multi-Column Convolutional Neural Network (CVPR 2016 · similarity 0.90)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.90)
- Leveraging Self-Supervision for Cross-Domain Crowd Counting (CVPR 2022 · similarity 0.90)
- Top-Down Feedback for Crowd Counting Convolutional Neural Network (AAAI 2018 · similarity 0.89)
- Wide-Area Crowd Counting via Ground-Plane Density Maps and Multi-View Fusion CNNs (CVPR 2019 · similarity 0.89)
- Incorporating Side Information by Adaptive Convolution (NeurIPS 2017 · 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.