De-Beauty GAN: Restore the original beauty of the face
Jinghang Wang, Zhiguo Zhou · 2023
Changes in face makeup and facial morphology pose new challenges for face recognition, and recent developments in image translation have laid the technological foundation for the de-beautification task. To this end, the first publicly available dataset containing 36,000 multi-ethnic and gender images rich in facial beauty, especially facial morphology changes, is established. Based on this dataset, we propose De-Beauty GAN, a highly scalable and robust facial de-beauty model that can synthesize more realistic de-beautified faces. Unlike existing de-beauty models, the model proposed in this paper considers the recovery of facial deformation. According to the characteristics of the image processing task, a new quantitative metric, DBE, for de-beauty evaluation is proposed, and the usability of this metric is evaluated through quantization. The experimental results demonstrate that the model proposed in this paper can effectively handle the de-beautification task with 1.3 and 1.6 times better results than the existing popular methods, respectively.