Supplementary code for the paper accessible at
Clone or download
Pull request Compare This branch is 38 commits ahead of levifussell:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Forging new worlds (GANaxies)

High-resolution synthetic galaxies with chained generative adversarial networks


This repository contains supplementary code for the paper Forging new worlds: high-resolution synthetic galaxies with chained generative adversarial networks, a pre-print version of which can be found on arXiv (arXiv:1811.03081).

At the moment, this repository mainly provides the trained models for generating galaxies using our chained GAN model. Please note that you will need a functioning PyTorch installation to make use of the code.

File description

  • The user-facing Python file to create synthetic galaxy samples

  • The DCGAN and StackGAN models used in the corresponding research and paper

  • dcgan_G.pth: The pre-trained DCGAN generator used to create synthetic 64x64 pixel galaxy images

  • stackgan_G.pth: The pre-trained StackGan generator used to upscale the DCGAN generator output

Quickstart guide

Generating galaxies with the provided pre-trained models works in both Python 2 and Python 3. In order to create one synthetic galaxy image, you can simply use the following command:


If you want to create a specific number of synthetic samples, modify the batchSize parameter with that number:

python --batchSize 42

Below, you can see a few examples of galaxy images created with the pre-trained models.