CSRNet-pytorch

This is the PyTorch version repo for CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes in CVPR 2018, which delivered a state-of-the-art, straightforward and end-to-end architecture for crowd counting tasks.

Datasets

ShanghaiTech Dataset: Google Drive

Prerequisites

We strongly recommend Anaconda as the environment.

Python: 2.7

PyTorch: 0.4.0

CUDA: 9.2

Ground Truth

Please follow the make_dataset.ipynb to generate the ground truth.

Training process

References

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

@inproceedings{li2018csrnet,
  title={CSRNet: Dilated convolutional neural networks for understanding the highly congested scenes},
  author={Li, Yuhong and Zhang, Xiaofan and Chen, Deming}
}

Please cite the Shanghai datasets and other works if you use them.

@inproceedings{zhang2016single,
  title={Single-image crowd counting via multi-column convolutional neural network},
  author={Zhang, Yingying and Zhou, Desen and Chen, Siqin and Gao, Shenghua and Ma, Yi},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={589--597},
  year={2016}
}