Face Hallucination With Finishing Touches

Obtaining a high-quality frontal face image from low-resolution (LR) non-frontal is primarily important for many facial analysis applications. However, mainstreams either focus on super-resolving near-frontal LR faces or frontalizing high-resolution (HR) faces. It desirable to perform both tasks seamlessly daily-life unconstrained images. In this paper, we present novel Vivid Face Hallucination Generative Adversarial Network (VividGAN) simultaneously and tiny VividGAN consists of coarse-level fine-level Networks (FHnet) two discriminators, i.e., Coarse-D Fine-D. The FHnet generates coarse HR then the makes use component appearance prior, fine-grained components, attain with authentic details. FHnet, also design component-aware module that adopts geometry guidance as clues accurately align merge prior information. Meanwhile, two-level discriminators are designed capture global outline well detailed characteristics. enforces coarsely hallucinated be upright complete while Fine-D focuses fine ones sharper Extensive experiments demonstrate our achieves photo-realistic faces, reaching superior performance in downstream tasks, recognition expression classification, compared other state-of-the-art methods.