The FAsT Generative Adversarial neural Network (FATGAN) has a similar background and installation procedure as ShortGAN neural network. We omit these details here. It has advanced to ShortGAN neural network Discriminator and Generator with Conv2DTranspose and spectral normalization layers, overpassed by residual layers. The network has a user-specified depth and can generate colored images of any size.
Advantages: Trains faster than other GANs, with less computational resources. For instance, the demonstrative images (see below) of 136 x 206 pixels size FATGAN were generated on a single GPU with 20 GB memory.
Generated Images
This repository contains fat_env.yaml file with all packages you need. To install them just run in your bash terminal the following command:
conda env create -f fat_env.yaml
conda activate fat_env