Skip to content
Reimplementation of https://arxiv.org/abs/1812.04948
Python Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
LICENSE.txt
README.md
data.py
evaluation.py
models.py
ops.py
run.py
start_training.sh
start_training_2gpus.sh
test_data.py
test_eval.py
test_ops.py
test_train.py
test_utils.py
train.py removed hardcoded rectangular input image requirement Nov 6, 2019
utils.py

README.md

This is a reimplementation of NVidia's stylegan https://github.com/NVlabs/stylegan I did for learning purposes. My priority was testing out changes such as non-square images and conditioning on labels (using acgan and projection discriminator) rather than keeping the code clean so right now it is a bit messy.

This has been tested, and supports features like rectangular images, but I am currently waiting on the most recent training session to finish before uploading results. Training and sample generation are both performed using 'run.py'.

It is mostly original, but includes a couple functions from the official implementation for comparison testing.

To run, see the comments in start_training.sh (and then run start_training.sh). This has only been tested in one environment, so feel free to create a github issue if you encounter a problem. This model can handle non-square images and I plan on organizing and including the tools I've made/modified to generate datasets like that soonish.

Blogs related to a tool I made to interact with StyleGAN models:

https://towardsdatascience.com/animating-ganime-with-stylegan-part-1-4cf764578e https://towardsdatascience.com/animating-ganime-with-stylegan-the-tool-c5a2c31379d

Original Paper: Karras, T., Laine, S., and Aila, T. A style-based generator architecture for generative adversarial networks

You can’t perform that action at this time.