XY-boy
commented
7 months ago Hey, thank you for your work :) I like the more natural feel that EDiffSR gives in your example images in comparison to the other networks
My quick question was if this network / architecture could also be used to train on a normal photography dataset (like div2k or something) and then be used to upscale like portrait photos, or if it is specifically only designed for remote sensing images?
(the training config seems simple enough, but not sure how I would add jpg compression into the training dataset for example for the upscaling model to be able to handle jpg compression)
Hi Many thanks for your interest and support.
Yes, it can also be used on natural images, such as DIV2K.
You can set lr/gt path in the yml to low/high resolution jpg compression image folder to train ediffsr.
Hey, thank you for your work :) I like the more natural feel that EDiffSR gives in your example images in comparison to the other networks
My quick question was if this network / architecture could also be used to train on a normal photography dataset (like div2k or something) and then be used to upscale like portrait photos, or if it is specifically only designed for remote sensing images?
(the training config seems simple enough, but not sure how I would add jpg compression into the training dataset for example for the upscaling model to be able to handle jpg compression)