Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

This paper introduces SRGAN, a deep learning model that uses generative adversarial networks (GANs) to enhance the resolution of images. Unlike previous methods that focus on minimizing pixel-wise errors, SRGAN aims to produce more realistic and visually appealing high-resolution images by incorporating a perceptual loss function and adversarial training. This approach leads to images with finer details and textures, making them more photo-realistic.