The code of paper "GazeCorrection: Self-Guided Eye Manipulation in the wild using Generative Adversarial Networks". Paper will published coming soon.
Gaze correction aims to redirect person's gaze into the camera by manipulating the eye region and it can be considered as a specific image resynthesis problems. Gaze correction has a wide range of applications in real life, for example the eye contract of remote users in video conference systems. We proposed a simple but effective model which does not require the training dataset labelling with the head pose and eye angle information, even the majority training data not have the corresponding groundtruth between the different domains. Our proposed model is based on the generative adversarial networks and leverage encode-decode to learn the mapping from the input facial image without the eye region to the facial image with corrected eye region. Moreover, we propose a self-guided method to preserve the identity information of the in-painted images. A new dataset has been collected for training and will be introduced in details.
- Clone this repo:
git clone https://github.com/zhangqianhui/Eye_Rotation_GAN.git- Download the Eye_Rotation dataset
Coming soon!!!
- Train the model using the default parameter
python main.py 
