This folder contains code for matching using GAN latent space and then subsequently benchmarking facial recognition systems. This section does not require StyleGAN2 dependencies, instead requiring the precomputed distances between images in GAN space.

The eda notebooks give exploratory analysis into different dataset and aspects of the matching. The non-eda notebooks are used to reproduce the results in the paper.

In the scripts folder, scripts starting with 00 predict attributes for each image in the dataset and scripts starting with 01 calculate pairwise distances between all images (pairwise distances for GAN latents is relatively very quick and does not have its own script).