#PGAN ReID

Prerequisites

• Python 3
• Pytorch (We run the code under version 0.3.1, maybe lower versions also work.)

Getting Started

Installation

• Install dependencies (e.g., visdom and dominate). You can install all the dependencies by:

pip install scipy, pillow, torchvision, sklearn, h5py, dominate, visdom

Datasets

• Create directories for datasets:

mkdir datasets
cd datasets/

• Download datasets through the links below, and unzip.
  Market1501:[Baidu Pan]link

Model

• Create directories for datasets:

mkdir bestmodel
cd bestmodel

• Download trained model through the links below. it include encoder pre-model and the whole model.
  encoder pre-trained model:[Baidu Pan]link
  GAN model:[Baidu Pan]link

Run code

defalut: +reranking, if you want to remove re_ranking, you need to modify ./reid/evaluators.py

sh train.sh

Citation

If you use this method or this code in your research, please cite as:

@article{zhang2020pgan,
  title={PGAN: Part-based nondirect coupling embedded GAN for person reidentification},
  author={Zhang, Yue and Jin, Yi and Chen, Jianqiang and Kan, Shichao and Cen, Yigang and Cao, Qi}, 
  journal={IEEE MultiMedia}, 
  volume={27},
  number={3},
  pages={23--33}, 
  year={2020},
  publisher={IEEE}
}

Acknowledgements

Our code is inspired by [FDGAN] (https://github.com/yxgeee/FD-GAN), [PCB] (https://github.com/syfafterzy/PCB_RPP_for_reIDand) and open-reid.