Face Aging with Conditional Generative Adversarial Network Guided by Ranking-CNN

As time elapses, the feature of human face will change more or less, which is referred to as face aging, varying from person to person. Face aging has great research value in the fields of age-invariant face recognition and entertainment in modern society. However, due to the difficulty of collecting face images at different ages for each person, face aging remains a difficult task, which aims to synthesize face images at target ages. In recent years, with the help of remarkable generative capacity, methods based on Generative Adversarial Networks (GANs) have sprung up. However, existing GAN-based methods only take age labels as generative condition, completely ignoring the age-related ordinal information. In this paper, we propose a GAN model constrained by Ranking-CNN, which can provide more stringent age constraints for the generator. Through qualitative and quantitative experiments, we show that our method can generate more accurate aging face images.