implementation of ExGAN in eye-inpainting, update the mask generation of xx works
The tensorflow implement of Eye In-Painting with Exemplar Generative Adversarial Networks
- Just use refernece image as the exemplar, not code.
- Our model trained using 256x256 pixels, not 128x128 mentioned in the original paper.
- [Python 3.6]
- Tensorflow 1.4+
- dlib
we change the dataset processing way,
- first: download face image dataset,celeba or ffhq or vggface ;downloadshape_predictor_68_face_landmarks.dat ,and put under root dir
- second: unzip dataset into data dir,and crop face into 256*256 ,by
python preprocess.py --root ./data/raw --output_dir ./data/crop
- train: change path pram to ./data/crop run:
python main.py --OPER_FLAG=0
-test:
python main.py --OPER_FLAG=1 --img_path ./data/celeb_id_aligned/emilio-rivera-3.jpg --ex_path ./data/celeb_id_aligned/emilio-rivera-4.jpg