search

assafshocher / ZSSR

"Zero-Shot" Super-Resolution using Deep Internal Learning
Other
401 stars 102 forks source link

which downsampling method do you use to generate the LR image from the testing image? #18

Open DanChen001 opened 4 years ago

DanChen001 commented 4 years ago

Hi, thank you for sharing the code.

which downsampling method do you use to generate the LR image from the testing image?

assafshocher commented 4 years ago

We have several separate cases. It is important that the downscaling method will be the correct one (= approximately the one by which the test image was downscaled from the theoretic ground-truth image [that may not exist]). For the ideal case (= we have prior knowledge that the correct downscaling is bicubic), we just use bicubic downscaling. For the non-ideal case we use a given input-kernel. (for unknown kernel we need to first estimate the kernel, for example using our method KernelGAN: https://github.com/sefibk/KernelGAN).

An example for test-data generation for blind SR can be found here: https://github.com/assafshocher/BlindSR_dataset_generator

It depend whether you want to original data used