Redesigning Multi-Scale Neural Network for Crowd Counting

This is the official implementation of 'Redesigning Multi-Scale Neural Network for Crowd Counting'. (IEEE Transactions on Image Processing) [arXiv]

Installation

Python ≥ 3.6.

To install other required packages, run:

pip install -r requirements.txt

Training

• Fill in the settings in datasets/SHHA/setting.py
• Run 'train.py'

Evaluation

• Download the processed dataset ShanghaiTech part A(Link). For the preprocessing code, please refer to this.
• Download our SHA models here.
• Modify the path to the processed dataset and pretrained model in 'setting.py' in datasets/SHHA
• Run 'test.py'

Acknowledgements

Part of codes are borrowed from C^3 Framework and SASNet. Thanks for their great work!

Citation

If you find this work useful, please cite

@article{du2022redesigning,
  title={Redesigning Multi-Scale Neural Network for Crowd Counting},
  author={Du, Zhipeng and Shi, Miaojing and Deng, Jiankang and Zafeiriou, Stefanos},
  journal={arXiv preprint arXiv:2208.02894},
  year={2022}
}

or

@article{du2023redesigning,
  title={Redesigning multi-scale neural network for crowd counting},
  author={Du, Zhipeng and Shi, Miaojing and Deng, Jiankang and Zafeiriou, Stefanos},
  journal={IEEE Transactions on Image Processing},
  year={2023},
  publisher={IEEE}
}