search

beala / deep-image-prior-tensorflow

An implementation of https://dmitryulyanov.github.io/deep_image_prior for tensorflow.
Other
68 stars 19 forks source link

A question about the function "up_layer" #4

Open wrccrwx opened 5 years ago

wrccrwx commented 5 years ago

Thanks for your code! I have a small question about the detail of your implementation. In file "deepimg.py", at line 86, why did you choose the number of the conv layer's kernel as 3 instead of 128? The latter setting will result in much faster convergence in denosing case.

z-fabian commented 5 years ago

I agree, in Ulyanov's original paper the denoiser architecture had 128 filters. You can find it in the supplementary paper: https://box.skoltech.ru/index.php/s/ib52BOoV58ztuPM#pdfviewer