AI-Zhpp / IRAST

This is the code for the ECCV2020 paper "Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks"

IRAST

This is the code for the ECCV2020 paper "Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks" (Pytorch version)

Prepare

1.1 Datasets can Found in:

ShanghaiTech

UCF-QNRF

WorldExpo10

1.2 Setting Runing Environment：

Ubuntu 16.04

Cuda 8.0

python 2.7

Pytorch 0.4.1

Data Processing:

follow the file "make_dataset.py" to produce the ground-truth density map (in this work, most images are unlabeled)

Training the model:

python train.py train.json val.json 0 0 to train your model

Testing the model:

python val.py

Notice the path of all files in these codes, you should modify them to suit your condition.

Some Pre-trained Model in This Paper:

ShanghaiTech PartA:BaiduDisk password/code：2333

UCF-QNRF:Baidudisk password/code：2333

If you find the IRAST is useful, please cite our paper. Thank you!

@inproceedings{liu2020semi,
 title={Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks},
 author={Liu, Yan and Liu, Lingqiao and Wang, Peng and Zhang, Pingping and Lei, Yinjie},
 booktitle={European Conference on Computer Vision},
 year={2020}
}