Comparison of Generative Models in Tensorflow

The different generative models considered here are Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs).

This experiment is accompanied by blog post at : https://kvmanohar22.github.io/Generative-Models

Usage

• Download the MNIST and CIFAR datasets

Train VAE on mnist by running:

python main.py --train --model vae --dataset mnist

Train GAN on mnist by running:

python main.py --train --model gan --dataset mnist

For the complete list of command line options, run:

python main.py --help

The model generates images at a frequence specified by generate_frq which is by default 1.

Results of training GAN on mnist

Sample images from MNIST data is :

On the left is image generated from VAE and on the right is GIF showing images generated from GAN as a function of epochs:

For examples and explanation, have a look at the blog post.