Super-Resolution Generative AdversarialNetwork

Year: 2,017
Journal:  IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

This generative adversarial network was developed to solve one central problem: How to recover the finer texture details when super-resolving at large upscaling factors. Recent work is often lacking high-frequency details and is perceptually unsatisfying in the sense that it fails to match the fidelity expected at the higher resolution. SRGAN a generative adversarial network for image super-resolution is presented. It is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this the authors propose a perceptual loss function which consists of an adverserial loss and a content loss.

Sign In


Reset Password

Please enter your username or email address, you will receive a link to create a new password via email.