I am implementing AttgGAN just for study.
I referred the repository "https://github.com/LynnHo/AttGAN-Tensorflow"
Only changing hair color is a little bit done. (tf2)
If you want to edit this project, locate CelebA data into
attgan_impl_tf2/data/celeba/img_align_celeba/{ALL_IMAGES}
attgan_impl_tf2/data/celeba/annotations/train_label.txt
attgan_impl_tf2/data/celeba/annotations/val_label.txt
attgan_impl_tf2/data/celeba/annotations/test_label.txt
CelebA dataset: download link
And then, run data_spliter.ipynb in attgan_impl_tf2 to split data into train/validation/test data.
If you want to train, check train_attgan.ipynb for single GPU, train_attgan_distributed.ipynb for multi-gpu. Some configuration is located in settings.py, settings_distributed.py respectively.
Note that, this project is not perfect. This project is for studying!
Thanks to this paper:
He, Z., Zuo, W., Kan, M., Shan, S., & Chen, X. (2017). AttGAN: Facial Attribute Editing by Only Changing What You Want. IEEE Transactions on Image Processing, 28, 5464-5478.