An implementation of StyleGAN on CelebA dataset.
git clone https://github.com/MorvanZhou/celebA-styleGAN
cd celebA-styleGAN
pip install -r requirements.txt
Download CelebA
- img_align_celeba.zip (~1.4GB)
- list_attr_celeba.txt (25MB)
to celebA_img_align folder, then unzip img_align_celeba.zip to the same folder.
process data to tensorflow tensor_record format.
python dataset.py --data_dir ~/data/celebA_img_align/
python train.py --data_dir ~/data/celebA_img_align/ -b 32 -w 2 --soft_gpu --latent 128 --lambda_ 10 --epoch 151 -lr 0.0005 -b1 0. -b2 0.99
First few of epoch:
After one day: