Cascade EF-GAN: Progressive Facial Expression Editing with Local Focuses

  • 2020-03-25 15:08:06
  • Rongliang Wu, Gongjie Zhang, Shijian Lu, Tao Chen
  • 0

Abstract

Recent advances in Generative Adversarial Nets (GANs) have shown remarkableimprovements for facial expression editing. However, current methods are stillprone to generate artifacts and blurs around expression-intensive regions, andoften introduce undesired overlapping artifacts while handling large-gapexpression transformations such as transformation from furious to laughing. Toaddress these limitations, we propose Cascade Expression Focal GAN (CascadeEF-GAN), a novel network that performs progressive facial expression editingwith local expression focuses. The introduction of the local focus enables theCascade EF-GAN to better preserve identity-related features and details aroundeyes, noses and mouths, which further helps reduce artifacts and blurs withinthe generated facial images. In addition, an innovative cascade transformationstrategy is designed by dividing a large facial expression transformation intomultiple small ones in cascade, which helps suppress overlapping artifacts andproduce more realistic editing while dealing with large-gap expressiontransformations. Extensive experiments over two publicly available facialexpression datasets show that our proposed Cascade EF-GAN achieves superiorperformance for facial expression editing.

 

Quick Read (beta)

loading the full paper ...