Child Face Age Progression and Regression using Self-Attention Multi-Scale Patch GAN

Face age progression and regression have accumulated significant dynamic research enthusiasm because of its gigantic effect on a wide scope of handy applications including finding lost/wanted persons, cross-age face recognition, amusement, and cosmetic studies. The two primary necessities of face age progression and regression, are identity preservation and aging exactitude. The existing state-of-the-art frameworks mostly focus on adult or long-span aging. In this work, we propose a child face age-progress and regress framework that generates photo-realistic face images with preserved identity.To facilitate child age synthesis, we apply a multi-scale patch discriminator learning strategy for training conditional generative adversarial nets (cGAN) which in-creases the stability of the discriminator, thereby making the learning task progressively more difficult for the generator. Moreover, we also introduce Self-Attention Block (SAB) to learn global and long-term dependencies within an internal representation of a child’s face. Thus, we present coarse-to-fine Self-Attention Multi-Scale Patch generative adversarial nets (SAMSP-GAN) model. Our new objective function, as well as multi-scale patch discrimination and, has shown both qualitative and quantitative improvements over the state-of-the-art approaches in terms of face verification, rank-1 identification, and age estimation on benchmarked children datasets.