Arbitrary Scale Super-Resolution for Brain MRI Images

摘要

How a Super-Resolution task is often carried out. A low-resolution input image undergoes some convolutions to generate feature maps. Feature maps are some sort of representation that captures the features of the low-resolution image. Depending on the upscaling factor (it must be a multiple of ), there would be some number of upscaling modules applied to the representation. This upscaling unfolds the image into the high-resolution output.