Age Progression/Regression by Dual Discriminator Adversarial Autoencoder

The progression/regression of facial age can be applied to cross-age recognition or entertainment-related applications. It is challenging due to lack of facial expressions of the same person in a longer age range. Conditional Adversarial Autoencoder (CAAE) can learn facial manifolds and achieve smooth age development and regression at the same time. Since the generated face is different from the real face, we develop a novel model based on CAAE, which used two discriminators instead of one to solve this problem. The proposed model can produce better results and improve the similarity degree of age progression for different races of people.