[example] Deconvolution using careamics

[OReyesMatte]

I ran a test to compare CAREAMICS with a CARE-based deconvolution model I trained. Even when using much less training data (from 200 to 10 images), the predicted deconvolutions are basically the same between themselves and also to our target (SoftWorx). To further test it, I compared the segmentation (using a custom Omnipose model) for segmenting deconvolved membranes, and visually there appears to be no difference between the two images. Even when zooming into some sections of the image, there's minimal differences in either deconvolutions and segmentations. I also plotted the intensity profile across the diagonal of the blocks, where we also see a striking similarity between CARE and CAREAMICS.

The segmentation and CARE models are part of an ongoing manuscript, so I'll update the issue when there's a proper way to cite it. The manuscript will include links to notebooks and training data!

[jdeschamps]

That looks really fantastic, beautiful images! I am looking forward to read the manuscript!!