Image analysis demos for Coiled benchmarking

red-flour-beetle.ipynb

skimage-worm-embryo.ipynb

worm-embryo.ipynb

Here we go, perhaps this will be the fix for the dask problem

https://forum.image.sc/t/doing-watershed-with-large-image/77283/2

use dask.array.map_overlap 2 with scikit-image watershed using watershed_lines=True. This makes sure that each segment is separated from its neighbors by black/background pixels. Where the overlaps end, your segments will be cut by a vertical line, but not by the background.
then, use dask_image.ndmeasure.label 1 to label the connected components. This should unify the segments at the window boundary, but not the ones that are separated by the watershed line.

I don't like having to use watershed_lines=True on the red flour beetle data (the nuclei are packed pretty tightly and they're not that big). But if it's the easiest way to get a result, we might just go with it.

Link to issue discussion: coiled/benchmarks#751