Skip to content
StyleGAN made with Keras
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Models Upload weights for pretrained models Apr 15, 2019
Results Add more results images Apr 15, 2019
data Add Everything Except Models Mar 29, 2019
AdaIN.py Update comments and remove redundant code. Mar 31, 2019
LICENSE Create LICENSE Apr 1, 2019
README.md New figures Apr 15, 2019
adamlr.py General optimizations Apr 15, 2019
mixed-stylegan.py Switch truncation from Z to W Apr 16, 2019
stylegan.py Fix and add evaluations Apr 4, 2019

README.md

StyleGAN-Keras

StyleGAN made with Keras (without growth)

alt text A set of 256x256 samples trained for 1 million steps with a batch size of 4.

alt text Rows: 4^2 to 32^2 styles Columns: 32^2 to 256^2 styles

This GAN is based off this paper: https://arxiv.org/abs/1812.04948

"A Style-Based Generator Architecture for Generative Adversarial Networks"

Additionally, in AdaIN.py, you will find code for Spatially Adaptive Denormalization (a.k.a SPADE) This is adapted (as best as I can) from this paper: https://arxiv.org/abs/1903.07291

"Semantic Image Synthesis with Spatially-Adaptive Normalization"

This StyleGAN is missing growth. Feel free to contribute this, if you'd like!

Mixing regularities is left out in stylegan.py, but included in mixing-stylegan.py. It complicates the inputs of the generator.

To train this on your own dataset, adjust lines 10 to 15 respectively, and load your own images into the /data/ folder under the naming convention 'im (n).suffix'.

Enjoy!

You can’t perform that action at this time.