Astronomical Image Colorization and Up-scaling with Conditional Generative Adversarial Networks

Abstract

This research aims to provide an automated approach for the problem of Image colorization and Single Image Super Resolution by focusing on a very specific domain: astronomical images, using Generative Adversarial Networks. We explore the usage of various models in RBG and L*a*b color spaces. We use transfer learning owing to a small data set, using pre-trained ResNet-18 as a backbone encoder and fine-tune it further. The model produces visually appealing images that are high resolution and colorized. We present our results by evaluating the GANs quantitatively using distance metrics such as L1 distance and L2 distance in each of the color spaces across all channels to provide a comparative analysis. We use Fréchet inception distance (FID) to compare the distribution of the generated images and real image to assess the model's performance.