Awesome

Contents

• AnonyGAN
• Installation
• Dataset Preparation
• Generating Images Using Pretrained Model
• Train and Test New Models
• Evaluation
• Acknowledgments
• Citation
• Contributions

AnonyGAN

| Paper | <br> Graph-based Generative Face Anonymisation with Pose Preservation <br> Nicola Dall'Asen¹², Yiming Wang³, Hao Tang⁴, Luca Zanella³, Elisa Ricci²³. <br>¹University of Pisa, Italy, ²University of Trento, Italy, ³Fondazione Bruno Kessler, Italy, ⁴ETH Zürich, Switzerland.<br> In ICIAP 2021. <br> The repository offers the official implementation of our paper in PyTorch.

Installation

Clone this repo.

git clone git@github.com:Fodark/anonygan.git
cd anonygan/

Needed libraries are provided in the requirements.txt file.

pip install -r requirements.txt should suffice.

Dataset Preparation

• Download aligned CelebA here
• Extract aligned version
• Compute landmarks and mask with the code provided in preparation (modify paths accordingly)

Generating Images Using Pretrained Model

• Download pretrained model here
• Place it in ckpts/anonygan.ckpt
• Preprocess your images with the files in preparation
• Prepare a .csv files with columns [from, to] with condition and source images names
• Run test.sh modifying paths accordingly

Train and Test New Models

• Same as using the pretrained model, for training modify train.sh accordingly

Evaluation

evaluation/automatic_evaluation is the entry point, modify paths accordingly

Acknowledgments

Graph reasoning inspired by BiGraphGAN

Citation

If you use this code for your research, please consider giving a star and citing our paper!

@inproceedings{dallasen2021anonygan,
  title={Graph-based Generative Face Anonymisation with Pose Preservation},
  author={Dall'Asen, Nicola and Wang, Yiming and Tang, Hao and Zanella, Luca and Ricci, Elisa},
  booktitle={International Conference on Image analysis and Processing},
  year={2021}
}

Contributions

If you have any questions/comments/bug reports, feel free to open a github issue or pull a request or e-mail to the author Nicola Dall'Asen (nicola.dallasen@phd.unipi.it).