Open AccessSelf-supervised flow field decoupling for Controllable face reenactment
Author(s) -
Xianwei Kong,
Shengwu Xiong
Publication year - 2022
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2253/1/012034
Subject(s) - computer science , decoupling (probability) , computer vision , artificial intelligence , face (sociological concept) , identity (music) , image (mathematics) , information flow , movement (music) , facial expression , control engineering , engineering , social science , linguistics , philosophy , physics , sociology , acoustics , aesthetics
Face reenactment is a face image generation method. Its main task is to generate a new image given a source image and a driving image, which has the facial motion information of the driving image while retaining the content information of the source image. Existing flow-based approaches have demonstrated high-quality results, but these works regard the head movement as a whole, and cannot achieve more flexible movement control, and often suffer from the loss of identity information. In this paper, we propose a novel Controllable multi-identity face reenactment(CFReenet), which uses the prior information of facial motion to decompose the movement as two parts of pose and expression and use the curriculum strategy to decouple movement and identity information and better maintain identity information. The experimental results demonstrate the effectiveness of our method.