Tensorflow Implement of Paper Learning Residual Images for Face Attribute Manipulation, which has been accepted in CVPR 2017. We need write this code and compare our results with that, because the author don't public their code. I think this paper is a good paper and they give a perfect idea for facial visual manipulating with images residual learning. This difference with original paper is that this implements use Instance_norm instead of batch_normal. You can adjust important weights for more perfect results. When bugs have been founded by you, thanks for your contributions and PR.
We use the CelebA datasets. The code will crop and resize images to 128x128.
---------------------------------------------
The training data folder should look like :
<train_data_root>
|--image1
|--image2
|--image3
|--image4...
---------------------------------------------
$ python main.py --IMAGE_PATH /home/?/data/celebA/
The man face:
The residual face:
Man-to-Woman Face:
The woman face:
The residual face:
Woman-to-Man Face: