The input train/test images must be resized to certain size （e.g. 100*100）? about rescsnet HOT 2 OPEN

@TaihuLight

(1) Is it correct for my suggestion?

Yes you are right. The code in the notebook has to deal with images of size 100x100.

(2) How to can it support for train/test images of the arbitrary size?

This is really a good question I am refecting on. One might think of processing by blocks and put them together in the end (like what is done in ReconNet). But this would lead to blocky artifacts even after applying something like BM3D denoiser. I am investigating how is that done in CNN based image super-resolution ( and maybe in detection framework such as Faster-RCNN ). Possibly won't add such feature in this repo in the short term.

from rescsnet.

Thank you for your timely reply.

from rescsnet.