Closed mrseekdeep closed 2 years ago
I've actually been working on essentially that for the last year, and Carsen and I have been incorporating my changes into Cellpose under the '--skel' flag in the last few weeks. We'll post the paper in the project description showcasing the new algorithm once it gets up on biorxiv. For now, there is no built-in model using it, but you can train it on your own data. I'd love to see some examples of your data if you want to post them here!
In a way, i think Cellpose is creating contour maps using diffusion. My data looks more like the black and white image in my first comment, and the issue is that in my case the center might not be a point, instead its somewhat a rectangle with curved edges.
@kevinjohncutler What are you looking to introduce in cellpose 2.0? multi-class segmentation, polygonal shape for the center, .....?
@kevinjohncutler What are you looking to introduce in cellpose 2.0? multi-class segmentation, polygonal shape for the center, .....?
There are quite a few changes that I am introducing, but luckily you’ll only have to wait a week or two before we post the paper describing it :) You ideas definitely map onto what we have done, except for the latter - that’s closer to what StarDist did, but does not generalize well to long and curved shapes.
awaiting eagerly for the release, and wishing you good luck
Hi there, According to CellPose, the cells are initially represented by a Point which is a "heat-source in the middle" then diffusion is applied. Can a Line be used as the heat-source instead of a Point? In cases where the object that i want to segment looks like![136584248-d95a2b74-2324-4119-aaf6-aa3f87b6924f](https://user-images.githubusercontent.com/92168032/140791991-eae124bf-07e6-4efb-aceb-bdc7e55bbb49.png)