Face aging using global and pyramid generative adversarial networks

We propose a novel approach that addresses face aging as an unsupervised image-to-image translation problem. The proposed approach achieves age progression (i.e., future looks) and regression (i.e., previous looks) of face images that belong to a specific age class by translating them to other (subsequent or precedent) age classes. It learns pairwise translations between all age classes. Two variants are presented. The first one learns a global transformation, while the second one incorporates a pyramid encoding and decoding scheme to more effectively diffuse age class information. The proposed variants are thoroughly evaluated with respect to both qualitative and quantitative criteria. They yield appealing face age progression and regression results when compared to ground truth images and outperform state-of-the-art approaches for face aging based on quantitative evaluation metrics. Notably, the incorporation of pyramid encoding and decoding is proven to be beneficial to the quality of the generated images.