Code for "Debiasing Pretrained Generative Models by Uniformly Sampling Semantic Attributes", in submission to NeurIPS 2023

This repo contains code for the paper, focused ont the Synthetic Shapes dataset

train_distribution_mapper.ipynb contains code to sample a fair noise dataset from a biased classifier, and to train a distribution mapper on this dataset

get_samples_from_each_method.ipynb generates a dataset from each compared method

save_stats.ipynb calculates the metrics for each method

plot_results.ipynb recreates the comparison plot
