pravsels/synthpar

Model Card

SynthPar is a conditional StyleGAN2 model capable of creating synthetic face images with specified attributes like race and sex. It is built upon the StyleGAN2 architecture and trained on the VGGFace2 dataset, which uses a 4 race (Asian, White, Indian, Black) and 2 sex (Male, Female) categorization scheme.

Intended Use

The primary intended use of this model is to generate synthetic face images with controlled demographic attributes.

These generated images can be utilized for various purposes, such as:

• Developing and evaluating face recognition models with the goal of reducing performance disparities across different demographic groups.

• Augmenting existing face datasets to improve their diversity and balance.

• Conducting research on fairness and bias in facial recognition systems.

How to Use

Please refer to the github repo for further guidance.

License

The SynthPar model is released under the MIT license.