Generative Adversarial Networks and Autoencoders

This repository aims to present the structure of generative models using theano and lasagne, and the results obtained using celebA and mnist datasets.

You will find the code for:

• DCGAN: Deep Convolutional Generative Adversarial Networks
• BiGAN: Bidirectional Generative Adversarial Networks
• WGAN: Wasserstein Generative Adversarial Networks
• CAAE: Conditional Adversarial Autoencoders

Prerequisites

You will need to install a virtual environment containing:

• Theano
• Lasagne
• Python 2.7

More information about lasagne installation can be found at: http://lasagne.readthedocs.io/en/latest/user/installation.html

Running the code

The code can be run either for celebA or MNIST. For information, the entire celebA dataset can be downloaded at: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html

The code can be run in the following way:

python dcgan.py --mnist --printLayers --outDir '~/AllThingsGAN/Experiments/dcgan_mnist/'