Weakly-Supervised Dual Generative Adversarial Networks for Makeup-Removal

Abstract

With the improvement of face recognition precision, face recognition system is used in many fields. However, the face recognition system sometimes cannot recognize the makeup face. In this paper, a new image-to-image translation algorithm based on GAN and dual learning is proposed to remove the makeup. Especially, the proposed algorithm is weakly supervised and it combines the paired and unpaired image-to-image translation model. The dual model is firstly trained using a small number of paired data, then the performance of the model is improved by large number of unpaired data. The proposed weakly-supervised image-to-image translation algorithm is applied into makeup-removal task, and the experimental results demonstrate its higher performance than other algorithms.