I like to run simple shapes of different sizes through DL segmentation to test how they handle scale. Here is one of my test images processed with cellpose4-sam. Seems like it is still sensitive to the diameter.

Is Cellpose-SAM the default now? I took a quick look at the notebooks you shared and couldn't quite figure out how you were instantiating the models as 'Cellpose-SAM' models. So is the default a SAM model now and the SAM backbone will be automatically downloaded when you create a model?

Update. It looks like upgrading to Cellpose 3.1.0 solves some of the issues. 3.1.0 has new options for smoothing masks and the 3D mask generation is much faster. See this post forum.image.sc/t/cellpose-3.... Thanks @computingnature.neuromatch.social.ap.brid.gy and others for the new release.

forum.image.sc/t/cellpose-3... Interesting forum question on 3D Cellpose that isn't answered yet. I played around a bit this morning but couldn't quite figure it out. Love to hear from others who work with 3D Cellpose (or other 3D segmentation methods)