This repository contains an implementation of the IntroVAE model in tensorflow based on the paper titled "IntroVAE: Introspective Variational Autoencoders for Photographic Image Synthesis" https://arxiv.org/abs/1807.06358.
Usage
1. Download and extract the CelebA-HQ-128x128 dataset
The CelebA-HQ-128x128 dataset is a downscaled version of the CelebA-HQ (1024x1024) dataset.
It can be downloaded from the following Dropbox link: https://www.dropbox.com/s/ayz2roywuq253l0/celebA-HQ-128x128.tar.gz?dl=0
Extract the contents of the above tar.gz in the datasets directory.
$ mkdir datasets
$ tar xvzf celebA-HQ-128x128.tar.gz -C datasets
2. Train a model
For a baseline training on the CelebA-HQ-128x128 dataset, simply run the following script.
$ bash ./baseline_128x128.sh