jwsoh/dubd

Deep Universal Blind Image Denoising (ICPR, 2020)

dubd's Issues

PNG-File shoult no have an alpha channel

Good work. Thank you very much. Helpful for many a photo amateur.
Unfortunately I have no idea in Python programming. But tensorflow gives the error message:

"ValueError: Cannot feed value of shape (1, 252, 261, 4) for Tensor 'Placeholder:0', which has shape '(?, ?, ?, 3)'"

if the PNG file contains an alpha channel.

About the train code

Nice work! Could you release the train code? Thank you in advance.

train code and the performance of the testset

Hi! Could you release your train code?thank you in adcance!
And tested the testset with your model ,the performance is not obvious

Artifacts and not regular Blur.

Really nice idea with noise information added to the network, but I sess strange artifacts, and the quality of bug images is strange. Good that now we can denoise by one model all images.

Sometimes the blur after denoising is hit and not regular. For the reason of unexpected blur - it's hard to deblur that.
Is it possible to see the train code?