This repository contains codes for the official implementation in PyTorch of paper "Learning Error-Driven Curriculum for Crowd Counting".
| Method | MAE | MSE |
|---|---|---|
| MCNN | 26.4 | 41.3 |
| MCNN+SF | 15.3 | 35.2 |
| MCNN+SF+TN | 14.4 | 25.1 |
| Method | MAE | MSE |
|---|---|---|
| CSRNet | 10.6 | 16.0 |
| CSRNet+SF | 10.4 | 15.9 |
| CSRNet+SF+TN | 9.4 | 15.6 |
| Method | MAE | MSE |
|---|---|---|
| U-net | 26.8 | 39.7 |
| U-net+SF | 13.5 | 23.0 |
| U-net+SF+TN | 12.1 | 19.7 |
| Method | MAE | MSE |
|---|---|---|
| DenseNet | 13.0 | 22.7 |
| DenseNet+SF | 7.5 | 12.8 |
| DenseNet+SF+TN | 7.0 | 12.2 |
-
Download the data and use the script in /pre_process.
-
python train_DenseNet_shanghai.py
If you have any questions, do not be afraid to contact us.