FA-GAN: High Resolution Face-Aging

Abstract

Face-Aging aims at translating facial images from one age category to another while preserving source identity. When compared with existing Image generation/translation methods current Face-Aging methods work at lower resolution datasets. Hence, in this work we create a novel high resolution supervised Facial Aging dataset by classifying existing high quality facial datasets. Through qualitative and quantitative experiments on our newly created dataset as well as CACD dataset, we show superiority of our method in terms of quality and diversity when compared with existing methods. Quantitative improvements obtained are as high as 8% in terms of face-verification accuracy, and 2% (random samples), 62% (old samples) in terms of age-estimation accuracy, which becomes significant when put together in conjunction.