A PyTorch Implementation of StyleGAN2 (Unofficial)

This repository contains a PyTorch implementation of the following paper:

Analyzing and Improving the Image Quality of StyleGAN (StyleGAN2) Authors: Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, Timo Aila

Paper: http://arxiv.org/abs/1912.04958 Video: https://youtu.be/c-NJtV9Jvp0

Motivation

To the best of my knowledge, there is still not a similar pytorch 1.0 implementation of styleGAN2 as NvLabs released(Tensorflow), therefore, i wanna implement it on pytorch1.0.1 to extend its usage in pytorch community.

Notice

@date: 2019.12.16

@info: settings are in opts/opts.py. you can change to your own dataset and choose resolution at 64, 128, 256 and so on.

@date: 2019.12.27

@info: Need Help!
After about 2 weeks experiment, this version is still hard to converge. I am pretty confident about my G&D definition are strictly follow the same with the originial stylegan2. So if you are willing to make this project converge, please feel free to change it! Especially in training paradigm!

Author

• Samuel Ko

Training

# ① pass your own dataset of training, batchsize and common settings in TrainOpts of `opts.py`.

# ② run train_stylegan.py
python3 train.py --path `your_own_dataset_path`

# ③ you can get intermediate pics generated by stylegenerator in `opts.det/images/`

Project

we follow the release code of styleGAN2 carefully and if you found any bug or mistake in implementation, please tell us and improve it, thank u very much! .

Related

1. StyleGAN - Official TensorFlow Implementation

2. The re-implementation of style-based generator idea

3. ptrblck_styleGAN

System Requirements

• Ubuntu18.04
• PyTorch 1.0.1
• Numpy 1.13.3
• torchvision 0.2.2
• scikit-image 0.15.0
• tqdm
• GTX 1080Ti or above

Q&A

Acknowledgements

My Email is [email protected], if you have any question and wanna to PR, please let me know, thank you.