The following article is a brief report on my Bachelor’s project, a comparative study regarding CNN-Based Single-Image Super-Resolution (SISR) techniques. We’ll begin with a brief look into the first applications of Deep Learning in SISR, followed by a discussion regarding the state-of-the-art CNN-based models for this task. We’ll be comparing the performance of the models on an open-source dataset of satellite images concerning their output quality, speed, and resource demands. You can find the scripts for our experiments as well as the trained super-resolution models here.

Learning in Single-Image Super-Resolution

SISR aims to recover a high-resolution (HR) image from a corresponding low-resolution (LR) version…