StyleGAN Pytorch

Easy to read and documented implementation of StyleGAN in Pytorch.

Requirements:

• torch
• torchvision
• tensorboard
• torchsummary

For documentation:

• sphinx
• numpydoc
• sphinx-rtd-theme

Training

Run:

Before starting the training process, make sure the settings (found in src/settings.py) are correctly configured for your needs.

python src/training/train.py

Visualizing with tensorboard:

On a new terminal:

tensorboard --logdir=summaries/[summary-name]

Generate documentation

Inside the /doc folder:

make html

To delete the html files:

make clean

Important

This project only works with square images.

This is an implementation of StyleGAN 1.

Based on the works of:

• Tero Karras et al. (DOI: 10.1109/TPAMI.2020.2970919)
• Sharon Zhou et al. (https://www.coursera.org/specializations/generative-adversarial-networks-gans)
• SiskonEmilia (https://github.com/SiskonEmilia/StyleGAN-PyTorch)