CellposeSAM segmentation doubles the dimensions of the masks

[JohnAbbey98]

Hey team,

I'm running into an issue in segment cellpose, where the dimension of the masks (x and y) are twice the size of the original x and y dimensions. The masks themselves seem reasonable when I look at them, but the doubled dimension causes measure intensities to fail. The exact command I'm running is segment cellpose --diameter 15 -o masks_cp. Have you observed this behaviour before? The issue doesn't occur with deepcell/mesmer, so I'll just proceed with this for now but thought I'd let you know.

Steinbock version: Running docker container ghcr.io/bodenmillergroup/steinbock:0.16.5-cellpose.
OS: I get the issue on MacOS M2 chip and on Ubuntu 20.04.

Image dimensions:

python -c "
import tifffile, glob
for m in sorted(glob.glob('img/*.tiff'))[:2]:
print(m, tifffile.imread(m).shape)
"
img/sample1.tiff (46, 901, 1882)
img/sample2.tiff (46, 1000, 1000)

Cellpose mask dimensions:

python -c "
import tifffile, glob
for m in sorted(glob.glob('masks_cp/*.tiff'))[:2]:
print(m, tifffile.imread(m).shape)
"
masks_cp/sample1.tiff (1802, 3764)
masks_cp/sample2.tiff (2000, 2000)

[victorjonathanibanez]

@JohnAbbey98 That is odd - according to the cellpose documentation, changing the diameter does change the input image dimensions but should not change the output mask dimensions as the mask gets resized to the original image size. Have you tried without --diameter and checked if that works?