Introduction

This repository contains the implementation of COEGAN and all code used in the evaluation and comparison with other methods, as presented in the paper COEGAN: Evaluating the Coevolution Effect in Generative Adversarial Networks http://gecco-2019.sigevo.org.

Environment Setup

Install pytorch:

conda install pytorch=1.3.0 torchvision cuda90 -c pytorch

Install dependencies:

pip install -r requirements.txt

Run All Unit Tests

python -m unittest discover

Experiments

Configure the Training Setup

Edit the experimental settings in evolution/config.py.

Start the Training:

python ./train.py

Visualize the Results

Run JupyterLab

jupyter lab

Results

See below the results of the experiments presented in the paper:

MNIST

Fid Score

Generated Samples

Fashion-MNIST

Fid Score

Generated Samples