Variational autoencoder with style-based generator
PyTorch implementation of a variational autoencoder that utilises a style-based generator Karras et al. 18. There are two training files, a basic (beta)vae training regime, and an implementation of the aggressive inference training regime He et al. 19. Second implementation was done as basic VAE training regime suffers from posterior collapse.
Implementation of style-based generator is based on: https://github.com/rosinality/style-based-gan-pytorch
Work in progress.
To train:
python3 train_vae.py /path/to/data
or python3 train_aggressive_inf_vae.py /path/to/data