Deep Convolutional Generative Adversarial Network implementation
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.



Hello! This is an implementation of Deep Convolutional Generative Adversarial Network in TensorFlow. It is made to be simple, intuitive and fast.

Getting Started

In order to run Universal-DC-GAN, you will need the following dependencies:

  • tensorflow==1.8.0
  • numpy==1.14.3
  • termcolor==1.1.0
  • matplotlib==2.2.2
  • scipy==1.1.0
  • Pillow==5.2.0
  • colorama==0.3.9

You can download them by running this command while being in Universal-DC-GAN directory:

pip install -r requirements.txt


Default settings are in file. You can configure pretty much anything you want there, each option is explained inside the file. Here are some basic options:

dataset_name = 'textures_all' - Name of the folder, containing your dataset, it should be placed in data_folder path.

data_folder = './data/' - Default location of all datasets, your dataset folder, specified in the variable dataset_name should be placed inside.

saves_folder = './saves/' - Default location of saved models. They save automatically during the training!

train = True - If you have already trained the network, you can directly generate images based on your model.

w, h = 512, 512 - Width and height of all the images in your training dataset


In order to run the script, just do:


Example results should now appear in the output directory (default is output), every number of iterations specified in config file.


You can find some of the generated examples below. Note that the results may be improved significantly, by extending the time, needed to train the network. Models used to generate these examples were trained around 3-4 hours each, on GTX1060.


alt text alt text alt text alt text alt text


alt text alt text alt text