Non maximum suppression workflow

Dear All,

I was just wondering. Are in KNIEME modules to perform non maximum suppression on (alternative, overlapping) instance segmentation hypothesis?

Thanks a lot & Kind regards


Hi @tooobi,

We don’t have generic nodes for NMS at the moment. I am wondering what your use-case looks like?

I have recently worked on getting StarDist into KNIME by using the ImageJ2 integration. There are still some kinks that need to be ironed out, but you can follow the progress at StarDist or Noise2Voide from KNIME - #4 by stelfrich - Usage & Issues - Forum. I am bringing this up since Uwe et al use an NMS approach in their postprocessing plugin which might be adapted for your case?


1 Like

Dear @stelfrich,

Thanks a lot for the fast reply.
It is more a strategic question.
I have a tool to create multiple segmentation hypothesis (using CellPose, StarDist ect) in Python.


At some point I want to give users of my tool the chance to switch to their favorite tool for the post-analysis workflow.

This i can do before or after choosing the final segmentation in Python.
If I do it before, I would need ways to integrate results (in this case in KNIME).
BUT, i can also just do it after.

Non maximum suppression is for example an easy way to stitch together prediction tiles or sub-volumes.

Kind regards


Dear @tooobi,

That’s the easiest way at the moment.

Just thinking out loud (burning my last brain power for today): KNIME allows you to read in multiple labeling images already. Computing the overlap and union of two segments is possible. Getting a list of segments that are overlapping is also possible. Computing column-wise maxima is available as well. I guess what I am saying is, that it should be possible to build a component that implements NMS without actually writing any code.

I haven’t read the preprint yet: Do you get multiple hypotheses per tool or just one?


1 Like

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.