the Tensorflow implementation of PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION.
-
This implement use CelebA dataset, not CelebA-HQ.
-
All tricks have been used, except "Equalized learning rate". You can make a PR if high-qualtiy generated samples with this technique can be achieved. Thanks for your contributions.
-
Recently, just generate 64x64 and 128x128 pixels samples.
- TensorFlow >= 1.1
- python 2.7
- Clone this repo:
git clone https://github.com/zhangqianhui/progressive_growing_of_gans_tensorflow.git
cd PGGAN-tensorflow
- Download the CelebA dataset
You can download the CelebA dataset and unzip CelebA into a directory. Noted that this directory don't contain the sub-directory.
- Train the model
python main.py --path your data-path
Here is the generated 64x64 results of PGGAN-tensorflow(Left: generated; Right: Real):
Here is the generated 128x128 results of PGGAN-tensorflow(Left: generated; Right: Real):
If you find some bugs, Thanks for your issue to propose it.