Multimodel Facial Makeup Translation with Generative Adversarial Networks

摘要

Automatic apply/remove makeup is a more challenging task than traditional style transfer, there are specific makeup styles in different facial areas, such as eye shadow, lipstick, and foundation. We propose a method of makeup transfer, which can achieve a makeup-oriented synthesis under different faces without pairs of images, and output different styles of makeup automatically. Specifically, owing to makeup transfer is more than a simple global transformation, we will use the separation training method, not only save the identity of the face, but also retain the characteristics of the specific part. Our framework is based on generative adversarial networks, adding face segmentation algorithms and Poisson blending techniques in the synthesis, make it possible to smoothly blend faces from different parts. Then we can complete makeup and makeup removal between different faces.