This repository contains various VAE implementations and their corresponding results obtained via training them: Convolutional VAE, IntroVAE, PixelVAE, VAE/GAN.

Third Year Project Report can be consulted here.

Note that the datasets were removed from the workspace due to space constraints. For CIFAR-10, no modification should be needed as the dataset is downloaded automatically. In order to use this code for CelebA, download the dataset into a "celeb/celeb" path.

Credits for the architectures we experimented with and for general development ideas go to the following GitHub users:

• jonhare - Dr. Jonathon Hare - Convolutional VAE
• pbloem - Dr. Peter Bloem - PixelVAE
• dragen1860 - Jackie Loong - IntroVAE
• seangal - VAE/GAN
• bbeatrix - IntroVAE.