PFA-GAN: Progressive Face Aging With Generative Adversarial Network

Abstract

Face aging is to render a given face to predict its future appearance, which\nplays an important role in the information forensics and security field as the\nappearance of the face typically varies with age. Although impressive results\nhave been achieved with conditional generative adversarial networks (cGANs),\nthe existing cGANs-based methods typically use a single network to learn\nvarious aging effects between any two different age groups. However, they\ncannot simultaneously meet three essential requirements of face aging --\nincluding image quality, aging accuracy, and identity preservation -- and\nusually generate aged faces with strong ghost artifacts when the age gap\nbecomes large. Inspired by the fact that faces gradually age over time, this\npaper proposes a novel progressive face aging framework based on generative\nadversarial network (PFA-GAN) to mitigate these issues. Unlike the existing\ncGANs-based methods, the proposed framework contains several sub-networks to\nmimic the face aging process from young to old, each of which only learns some\nspecific aging effects between two adjacent age groups. The proposed framework\ncan be trained in an end-to-end manner to eliminate accumulative artifacts and\nblurriness. Moreover, this paper introduces an age estimation loss to take into\naccount the age distribution for an improved aging accuracy, and proposes to\nuse the Pearson correlation coefficient as an evaluation metric measuring the\naging smoothness for face aging methods. Extensively experimental results\ndemonstrate superior performance over existing (c)GANs-based methods, including\nthe state-of-the-art one, on two benchmarked datasets. The source code is\navailable at~\\url{https://github.com/Hzzone/PFA-GAN}.\n