Face Aging on Realistic Photo in Cross-Dataset Implementation

Abstract

Face aging or age progression is a prediction on how a person looks at the future. Face aging image-to-image translation is a process of translating an image of young people to their older version or vice versa. The need for a paired training dataset to train the generative adversarial networks (GANs) is a major problem with face aging image-to-image translation. Nowadays, there is a method where an unpaired training dataset can be used to do an image-to-image translation. CycleGANs is a GANs extended methods where there is no need for paired training dataset to train the CycleGANs. From the result, it shows that CycleGANs can do face aging image-to-image translation without using the paired training dataset.