마스크를 활용한 SaGAN 성능 개선 연구

Abstract

Face attribute manipulation is study of changing an input face image into a face image having a desired attribute. Recently, due to the development of deep learning technologies such as GANs, research results for changing an input face image into a face image having natural and desired attributes have been published. In this paper, we propose a model that improves the quality of the generated image by additionally applying the mask image to the SaGAN model, one of the latest face attribute manipulation studies, and improves the speed by integrating two encoder parts included in the generator into one. Experiments comparing with SaGAN show that the proposed technique is effective for improving performance and speed.