Reimplementation of paper "Image Super-Resolution by Neural Texture Transfer" in CVPR 2019 by pytorch.
Raw code in tensorflow can be found here.
offline_patchMatch_textureSwap.py has been uploaded. However it consumes huge space on disk when generating Ref-DIV2K dataset (100MB per picture). I'll try to find out if there're some bugs in this script.
Because there are so many channels at each scale, the size of npz file is actually this large. Under this condition, I can't store the whole dataset in my hard disk (~800GB left). So next I'll try to use part of DIV2K (with reference pic) or CUFED.
This repo has not been finished yet.