This repository contains the the code for our paper:
Controllable Privacy in Face Recognition: A Filter-based Approach Z Rezgui, N Strisciuglio, R Veldhuis IJCB, 2024
Our research explores methods for balancing privacy protection and model utility in face recognition using adversarial training and information bottleneck-based techniques.
At the moment, this repository includes only the training code (train.py) and some utility files. Unfortunately, after graduating from my university, I lost access to the pretrained models that were stored on the university’s computing cluster and were not pushed to GitHub at the time.
Since I no longer have access to those files, I won't be actively reconstructing the missing components. However, if you're interested in using or extending the code, feel free to reach out. I’d be happy to assist with any issues.
Although pretrained models are not available, you can still experiment with the training pipeline and adapt it to your own datasets. Check config.py to modify hyperparameters and file paths as needed.
We use Poetry for managing the project's dependencies. Poetry ensures that all required libraries are installed in a consistent and isolated environment.
Once you have Poetry installed, run the following command to install the project dependencies:
poetry installThis will install both the regular and development dependencies specified in pyproject.toml and lock the exact versions in the poetry.lock file.
Poetry automatically creates a virtual environment for the project. You can activate it with:
poetry shellIf you use this code in your research, please cite our work:
@inproceedings{rezgui2024controllable,
title={Controllable Privacy in Face Recognition: A Filter-based Approach},
author={Rezgui, Zohra and Strisciuglio, Nicola and Veldhuis, Raymond},
booktitle={2024 IEEE International Joint Conference on Biometrics (IJCB)},
pages={1--9},
year={2024},
organization={IEEE}
}