Image upscaling using generative adversarial network

Low-resolution images can be blurry or missing features, and often do not meet the needs for high-quality images. In this project, Super-Resolution GANs (SRGANs) will be used to super-resolve and sharpen the image. Super-resolution is a way to improve image quality and resolution. The different super-resolution methods will transform a low resolution image to high-resolution by reasoning about realistic details. In this project, specifically, we will leverage a deep learning framework called a Generative Adversarial Network to solve for high-resolution images based on a deep-learning platform. In the generative modeling field of machine learning, the generative model develops a representation of the patterns found in the original input data, and can then produce new instances of data (or output) based from the original input data. In essence, a GAN in machine learning is made up of two main parts, the generator component which is responsible for generating high resolution images based on the low resolution images we input, and a discriminator which is a model that determines whether the generated images resemble the originals. The two models are trained together at the same time, when the generator produces fake images to learn from, it gets valuable feedback from the discriminator, allowing it to learn.