segmentation_B discrepancies

[cpaniaguam]

@tdivoll
Comparing the workflow for segF against that in the test file, I am getting significant differences between inputs segB.ice_intersect (workflow) and segmentation_B_ice_intersect (read in for testing).

Here is the input image used in the ci tests for segmentation_F.jl segmentation_B_ice_intersect

And here is what I get via the full workflow for segB.ice_intersect

Originally posted by @cpaniaguam in #239 (comment)

[cpaniaguam]

I am also seeing differences in segB.not_ice. They look like they are inverted versions of each other.
From the workflow

From ci tests

If I do a complement + landmask, I get a match with about 10% error.

[tdivoll]

Hoping this is fixed by #241, let's check the CI there and then use the updated segmentation_b and segmentation_watershed functions.