This is a pytorch version of the paper 'Generative Image Inpainting with Contextual Attention' (by Jiahui Yu et al. / 2018 CVPR)
python 3.6
pytorch 0.3.0
paper: https://arxiv.org/abs/1801.07892
original TF code: https://github.com/JiahuiYu/generative_inpainting
I used celebA faces dataset. (about 202,000 images)
place it in the 'data' directory
images=https://www.dropbox.com/s/3e5cmqgplchz85o/CelebA_nocrop.zip?dl=0
attributes: https://www.dropbox.com/s/auexdy98c6g7y25/list_attr_celeba.zip?dl=0 (just for getting file names)
python run.py
I made this network for catching up SOTA technique of the Image Inpainting area.
Though it trains without any bug, generated images are mostly blurry.
After clonning the repo, you need to check the contextual module performance or critics convergence.