Closed Tmodrzyk closed 4 months ago
Update : it seems that using BlurFFT
instead of Blur
solves the issue.
Hi Thibault,
Many thanks for pointing this out - I think the problem seems to be related to the conjugate_gradient solver which is only called for optimization schemes that require the proximal operator of the data fidelity term which is not available in closed form for the blur operator, but it is for the blurFFT one.
Will investigate further why conjugate_gradient is not working as expected here.
The problem has been fixed in #221
Many thanks for your reactivity.
Hello, I've been trying to set up a benchmark of all your methods for the deblurring operation.
However I noticed that when I want to use a batch size > 1, I get errors for some iterators. So far I noticed PGD and FISTA handle batch size > 1, but ADMM and DRS don't. They work for a batch size = 1 though.
Is this the expected behavior ? Looking at the documentation I thought any batch size would work for any method.
I did not test with other physics, maybe is it related to the blur operator specifically. Here is a minimal example of the issue:
Which produces the following output:
I have the most recent verson of deepinv installed.