Wasserstein Generative Adversarial Networks implemented using Keras.

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

For using this implementation of WGANs, you just need to install Keras.

pip install keras

The MNIST Dataset is a dataset of 60,000 28x28 grayscale images of the 10 digits, along with a test set of 10,000 images.

I have stored pre-trained Generator and Critic Networks (Trained for 90000 epochs) along with the noise distribution , from which digits are sampled from, in the Trained_Models folder.

Within the same folder is another folder called Backup. This folder stores the pretrained Generator Network and Crtic Network across different epochs as Generator_epoch.h5 and Critic_epoch.h5

Example

Generator_3000.h5 -> Generator Network trained for 3000 epochs.
Critic_3000.h5 -> Critic Network trained for 3000 epochs.

Run the script test.py in the terminal as follows.

Python test.py

I ran this program on Google Colab to get better results.

3000 epochs take approximately 15 minutes on Google Colab using their GPU.

After 90000 epochs

• Keras - Deep Learning Framework
• Google Colab - Cloud Service

• Vikram Shenoy - Initial work - Vikram Shenoy

• Project is based on Martin Arjovsky's paper, Wasserstein GAN
• Project is inspired by Jonathan Hui's blog, GAN - WGAN and WGAN-GP