Why doesn't it work well on my local machine?

[Troy0630]

I tried using it to segment bones just like in the pictures, but the results are poor. I didn’t expect it to be this bad—something must be wrong, but I can’t figure out what.

[FabianIsensee]

Hi Troy,
indeed this looks like way worse performance than you should get. One thing off the bat: What you are trying to do here is a semantic segmentation of all bones in the image. nnInteractive is built for instance segmentation, so it would expect you to segment one bone at a time which you could then aggregate using our plugins 'semantic mode' (which just writes all objects into one binary mask instead of an instance mask). Essentially there is a misalignment between what you are trying to do and what nnInteractive was built to do. You can confirm that this is the issue by going for individual bones, organs or other structures where the performance is likely much better.
Bones on CT is such an easy task that I would expect there to be publicly available models that do this fully automatically for you. That would be what you need in this case.
If the performance is still poor when evaluating instances, we would be interested in having access to your image so that we can debug.
Best,
Fabian