This repository provides the fair dataset of synthetic faces created in the paper Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification published at WACV 2025 (see credits). We plan to release the code to generate other synthetic datasets.
If you are rather interested by tools for conducting a comprehensive fairness analysis of face verification models, look at this repository or this one (both should be the same).
The dataset DCFace+C-all can be found here
It has a size and structure comparable to previous training sets of face verification models such as digiface-1m, SynFace or CASIAWebFace. More precisely it contains:
- 10,000 unique identities
- 50 images per identity thus 500,000 images in total
Images were balanced with regards to four attributes, namely gender (2 classes), ethnicity (4 classes), age (continuous) and pose (3 continuous angles).
Special thanks to the developers of DCFace and FairFace for their invaluable tools and resources.
If you find this work useful and use it on your own research, please cite our paper
@inproceedings{afm2025fairer_analysis,
author = {Fournier-Montgieux, Alexandre and Soumm, Michael and Popescu, Adrian and Luvison, Bertrand and Le Borgne, Herv{\'e}},
title = {Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification},
booktitle = {Winter Conference on Applications of Computer Vision (WACV)"},
address = "Tucson, Arizona, USA",
year = {2025},
}