CUMTGAN: An instance-level controllable U-Net GAN for facial makeup transfer

作者：

Highlights：

• A new method for makeup transfer, CUMTGAN, which performs well on few-shot datasets.

• Generating a “makeup mask” instead of directly generating a face with makeup to achieve controllable makeup shade.

• Creating a pseudo value, Pseudo-Truth-B, to evaluate the generated samples from makeup style and facial structure.

• Adequate comparison and analysis in qualitative and quantitative experiments.

摘要

•A new method for makeup transfer, CUMTGAN, which performs well on few-shot datasets.•Generating a “makeup mask” instead of directly generating a face with makeup to achieve controllable makeup shade.•Creating a pseudo value, Pseudo-Truth-B, to evaluate the generated samples from makeup style and facial structure.•Adequate comparison and analysis in qualitative and quantitative experiments.

论文关键词：Makeup transfer,GAN,U-Net,Face structure evaluation,Makeup style evaluation

论文评审过程：Received 22 October 2021, Revised 3 May 2022, Accepted 4 May 2022, Available online 13 May 2022, Version of Record 21 May 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.108996