Eye Gaze Correction Using Recurrent Residual U-net-based Conditional GAN(R2A-cGAN) with AdaIN
Hu Haoyang, Masataka Seo, Yen‐Wei Chen · 2023
The eye gaze correction is used to make eye contact since the different position of camera and the screen. The previous solution to the problem of eye-gaze misalignment is to generate viewpoint-transformed images using a cGAN model. By this method, the details of facial organs such as contours, teeth, and lips are blurred. We use AdaIN and Recurrent Residual Convolution Neural Network to improve the cGAN generator in order to make the details of the generated images more clear.