Multimodal Face Synthesis From Visual Attributes

Synthesis of face images from visual attributes is an important problem in computer vision and biometrics due to its applications law enforcement entertainment. Recent advances deep generative networks have made it possible synthesize high-quality attributes. However, existing methods are specifically designed for generating unimodal (i.e visible faces) In this paper, we propose a novel adversarial network that simultaneously synthesizes identity preserving multimodal (i.e. visible, sketch, thermal, etc.) without requiring paired data different domains training the network. We introduce generator with stretch-out modules images. Additionally, stretch-in introduced discriminator which discriminates between real fake Extensive experiments comparisons several state-of-the-art performed verify effectiveness proposed attribute-based synthesis method.