I just finished the section on super-resolution in lesson 7 of the v3 course.
The network didn’t perform very well when I ran some random low-resolution images that I found online.
Of course, I might need to train with a different dataset to handle different types of corruption, as Jeremy states in the lecture.
But searching for more information on image restoration, I came across this blog post: How To Perform Image Restoration Absolutely DataSet Free(https://towardsdatascience.com/how-to-perform-image-restoration-absolutely-dataset-free-d08da1a1e96d) and this paper Deep Image Prior(https://dmitryulyanov.github.io/deep_image_prior). Both are from 2018 and I can’t find anything more recent.
Does anyone know if this technique is still being used? Or has it been superseded by the UNet-based approach described in lesson 7?
Thanks,
Tim