Closed janinemelsen closed 11 months ago
The masks are numpy arrays, so computing intersections/unions and other correspondences should be straightforward with boolean masking and logical operators. I don't think adding dedicated functionality to deepcell-tf (assuming I've understood you correctly) makes sense, as that would just be a layer of indirection on top of generic, fundamental array operations.
Hi Janine, as you noted, the default output from Mesmer will have non-matching label IDs across the two prediction tasks. We added a post-processing step in ark which harmonizes the labels. You are welcome to either run the segmentation notebook, or take a look at the matching logic and repurpose it.
I'll go ahead and close this one as "not-planned" since there are existing solutions either working with arrays directly or with third party libraries. Thanks for the suggestion!
I would like to synchronise the whole cell and nucleus segmentation mask, so that they both contain the exact same number of cells (for instance 'cells' with only a membrane but not a nucleus should be removed from the whole cell mask) Is there any way to do this?
Thanks!
Janine