Super-resolution tries to enhance the resolution of an image. Deep learning is one of the methods used. Presented here is the award-winning RealSR solution from the Tencent YouTu Lab.
As is known, deep learning SR models are trained with sets of images and their reduced counterparts. The bi-cubic technique is widely used for downsampling. As the team explained in their paper, this is not the best method. Not only differ the domains but noise is also removed from the image, an unrealistic situation. The team has developed a unique downsampling technique to overcome the above problems. The results are impressive, as you can see.