Younghyun Jo

Hi, Try to use old codes in https://github.com/KupynOrest/DeblurGAN/tree/dd9e940f92db10f0e5ce073344159b99b22dbfa0. But this also still has lower PSNR 27.28.

Hi, The speed was achieved on our smartphone implementation so please check out the uploaded apk file. Our python code is not parallelized so you have to modify the code.

Hi, You can use the same Stream API for Java, but I am not sure what is the equivalent for C++. Sorry for not being of any help.

Hi. Not necessary. One simple way is to set UPSCALE = 8 and training from scratch.

Hi, Our code is run on CPU, and I expect it could be faster with GPU implementation when the operation of each pixel is parallelized (but not validated). Thanks! On...

Hi, I am sorry that we have no plans to release the code yet. But, please refer to this one #1.

Yes it is, and I mentioned this in the paper (p5). Because, if apply LocNet to GT then we get a slightly transformed GT and using this for training target...

Hi, Thanks for your interest. I am glad to hear that it works well for your tasks as well. I believe such valuable information about practical experimental results will give...

Hi, You can use it for academic research use only. Please do not use it for any form of commercial use. On Thu, Oct 26, 2023 at 8:26 AM sujyQ...

Hi. This is because of the sampling size. Ours-V: stores all sampling points (2^8). A and the others: only store 17 sampling points. On Wed, Aug 30, 2023 at 11:51 ...