Heredity-aware Child Face Image Generation with Latent Space Disentanglement

Xiao Cui, Wengang Zhou, Houqiang Li

Keywords: Child Face Image Generation, Generative Adversial Networks, Semantics Learning, Latent Space Disentanglement

Abstract: In this paper, we propose ChildGAN to generate a child's face image according to the images of parents with heredity prior. The main idea is to disentangle the latent space of a pre-trained generation model and precisely control the face attributes of child images with clear semantics. We use distances between face landmarks as pseudo labels so as to avoid using external labels. By calculating the gradient of latent vectors to pseudo labels, we figure out the most influential semantic vectors of the corresponding face attributes. Then we disentangle the semantic vectors in three aspects: adding a weight factor in the calculating process, working on the proper resolution layers, and using Schmidt orthogonalization to orthogonalize these vectors. Finally, we fuse the latent vectors of the parents by leveraging the disentangled semantic vectors under the guidance of biological genetic laws.

Submission Number: 57