PCE-Palm (AAAI-24)

PCE-Palm: Palm Crease Energy based Two-stage Realistic Pseudo-palmprint Generation | Paper

Example results

Prerequisites

• Python 3
• NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Clone this repo:

git clone https://github.com/Ukuer/PCE-Palm.git
cd PCE-Palm

• Install dependencies: pip install -r requirements.txt
• More details: This code borrows heavily from the RPG-Palm repository. You can find more details about the original code in the RPG-Palm.

Use a Pre-trained Model

• Download pce-checkpoints, unzip it and place it in ./checkpoints .

• Download CUT-checkpoints, unzip it and place it in ./CUT/checkpoints .

• Then bash ./inference.sh. Noted that you should modify some contents in ./inference.sh to meet you requirements.

Model Training

Tools

• The proposed PCEM can be found in ./PCEM_numpy.py. You can use it to get the PCE images from palmprint ROIs.
• The propsoed LFEB can be found in ./LFEM_pytorch.py. You can add it in your network.
• The improved bezier curves can be found in ./syn_bezier.py.

Training

Our proposed method is a two-stage method. The first stage is train a modified CUT model with bezier curves images and PCE images. The second stage is to train a generation model with paired PCE images and real palmprints.

• To train a modified CUT model:

  • Firstly, extract PCE images from palmprint ROIs using ./PCEM_numpy.py.
  • Then, genrate bezier curves images using ./syn_bezier.py, with a equal number of PCE images.
  • Finally, train a modified CUT model. You can find more details from CUT origin repository.
  • Noted that set --netG resnet_9blocks_lfeb.

• To train a generation model:

  • Perpare dataets: paired PCE images and real palmprints.
  • Then, bash run.sh.
  • To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097. To see more intermediate results, check out ./checkpoints/NAME/web/index.html. See RPG-Palm for more details.
  • Noted that we use the augmentation module from Stylegan2-ADA. If you have any dependencies issues, please refer to the Stylegan2-ADA repository.

Citation

If you find this useful for your research, please use the following.

@inproceedings{jin2024pce,
  title={PCE-Palm: Palm Crease Energy Based Two-Stage Realistic Pseudo-Palmprint Generation},
  author={Jin, Jianlong and Shen, Lei and Zhang, Ruixin and Zhao, Chenglong and Jin, Ge and Zhang, Jingyun and Ding, Shouhong and Zhao, Yang and Jia, Wei},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={3},
  pages={2616--2624},
  year={2024}
}

@inproceedings{shen2023rpg,
  title={RPG-Palm: Realistic Pseudo-data Generation for Palmprint Recognition},
  author={Shen, Lei and Jin, Jianlong and Zhang, Ruixin and Li, Huaen and Zhao, Kai and Zhang, Yingyi and Zhang, Jingyun and Ding, Shouhong and Zhao, Yang and Jia, Wei},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={19605--19616},
  year={2023}
}

If you have any questions or encounter any issues with the this code, please feel free to contact me (email: [email protected]). I would be more than happy to assist you in any way I can.

Acknowledgements

This code borrows heavily from the RPG-Palm repository, CUT repository, BicycleGAN repository, and Stylegan2-ADA.