SGDANet

1.Introduction

Official Implement for the paper《Synthetic Guided Domain Adaptive and Edge Aware Network for Crowd Counting》that published in the journal 《Image and Vision Computing》. In this project, you train a CSRNet or our SGEANet on ShanghaiTech dataset.

2.Project Organization

The folders are organized as follows:

• data: dataset processing code, image list file, the soft link of dataset folders.
• experiment: config file for training, testing.
• images: images to show on this readme.md
• main: two main scripts, train.py and val.py.
• src: all other codes

3.Requirement

• pytorch=1.1.0
• torchvision
• progressbar
• visdom

4.Tutorial

(1) prepare dataset

• Download ShanghaiTech dataset from Shanghaitech
• Create soft link to the data folder in this project

ln -s to/path/ShanghaiTech_PartA data/SHA

• Generate synthetic data, density map ground truth, save gt number and make a image list file in json format.

python data/prepare_dataset/SHA/make_ga_synthetic.py
python data/prepare_dataset/SHA/make_ga_gt.py
python data/prepare_dataset/SHA/save_gt_num.py
python data/prepare_dataset/SHA/make_json.py

(2) training real and synthetic baseline model

sh train.sh experiment/SHA/SGEANet/syn_baseline.yaml
sh train.sh experiment/SHA/SGEANet/real_baseline.yaml

(3) training SGEANet

sh train.sh experiment/SHA/SGEANet/real_baseline_LSG_LME.yaml

(4) evaluating

sh val.sh experiment/SHA/SGEANet/syn_baseline.yaml
sh val.sh experiment/SHA/SGEANet/real_baseline.yaml
sh val.sh experiment/SHA/SGEANet/real_baseline_LSG_LME.yaml

Bellow is a demo of predicted density maps from different methods.

5.Citation

If you use this code for your research, please cite our paper: