Cycle-Consistent Adversarial Network for Facial Local-Region Exchange in Wild

EasyChair Preprint 1023

Abstract

Recently, Generative Adversarial Networks are popularly used in face generation and get the state-of-art result. However, it's hard to swap the local area of face while many of previous work has focused on either generating face from a noise vector which belongs to some kind of data distribution or swaps the whole face. In this paper, we proposed a Cycle-Consistent Region Exchange Generative Adversarial Network(CREGAN) for facial local area exchange in the wild facial database. The Cycle-Consistent guaranteed that the exchanged area keeps the another facial feature and a novel approach to achieve face local region exchange and other region remain unchanged. At the same time, the characteristics of generative adversarial network can make ensure the quality of the generated images. And, it will shows that the generated images can reach photo-realistic results by CREGAN.

Keyphrases: Cycle-Consistent, Generative Adversarial Network, Wild Facial Database

@booklet{EasyChair:1023,
  author    = {Xiao Sun and Pingping Xia and Fang Tian},
  title     = {Cycle-Consistent Adversarial Network for Facial Local-Region Exchange in Wild},
  howpublished = {EasyChair Preprint 1023},
  year      = {EasyChair, 2019}}