My first work in academic career, you can treat it as a vae library, with many variations of vae including vae-gan.
If you want to use and modify the code in this library to implement your own vae-gan, you can just focus on vae_gan_pure_16_20211202.py.
PS:
input: 128x128x3 in tfrecord format.
original vae-gan paper:
Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, Ole Winther
This is the VAE-GAN version with image dataset in .jpg format, you still need to transform it into .tfrecords format.