Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
79 lines (50 sloc) 2.31 KB


This project is the implementation of the Paper "Generative Adversarial Networks recover features in astrophysical images of galaxies beyond the deconvolution limit" on python. It is the python version of This python version doesn't include deconvolution part of the paper.

Amazon EC2 Setup

EC2 Public AMI

We provide an EC2 AMI with the following pre-installed packages:

  • CUDA
  • cuDNN
  • Tensorflow r0.12
  • python

as well as the FITS file we used in the paper(saved in ~/fits_train and ~/fits_test)

AMI Id: ami-96a97f80 . (Can be launched using p2.xlarge instance in GPU compute catagory)

Launch an instance.

Connect to Amazon EC2 Machine

Please follow the instruction of Amazon EC2.


Linux or OSX

NVIDIA GPU + CUDA CuDNN (CPU mode and CUDA without CuDNN may work with minimal modification, but untested)


We need the following python packages: tensorflow, cv2, numpy, scipy, matplotlb, pyfits, and ipython

Get Our Code

Clone this repo:

git clone 
cd  GalaxyGAN_python/

Get Our FITS Files

The data to download is about 5GB, after unzipping it will become about 16GB. Download this file from Google Drive:

Run Our Code

Preprocess the .FITs

If the mode equals zero, this is the training data. If the mode equals one, the data is used for testing.

python --input fitsdata/fits_train --fwhm 1.4 --sig 1.2 --mode 0
python --input fitsdata/fits_test --fwhm 1.4 --sig 1.2 --mode 1

XXX is your local address. On our AMI, you can skip this step due to all these have default values.

Train the model

If you need, you can modify the constants in the

python gpu=1

You can appoint which gpu to run the code by changing "gpu=1".

This will start the training process. If you want to load the model which already exists, you can modify the model_path in the


Before you try to test your model, you should modify the model path in the

python gpu=1

The results can be seen in the folder "test".