Single point download of segmentation does not return correct values

[ceesem]

I'm trying to get the supervoxel ID under a point in the minnie65_phase3_v1 using cloudvolume 11.1.2. Using a Bbox spanning at least two voxels, this works with both cv.download and cv.download_point, but if I am using only one voxel I get a nonsense value.

A failing example:

import cloudvolume

cv = cloudvolume.CloudVolume(
    `graphene://https://minnie.microns-daf.com/segmentation/table/minnie3_v1`,
    use_https=True,
    bounded=False,
    fill_missing=True,
    progress=False,
)

bbox = cloudvolume.Bbox([173734, 187738, 19694], [173734+2, 187738+2, 19694+1])
# I am using 4,4,40 values when mip 0 is 8,8,40, so I am adding 2 to x and y for proper scaling.

svid = cv.download(
    bbox,
    coord_resolution=[4, 4, 40],
    agglomerate=False
)

gives svid a value of 604283204, which is not a supervoxel value at all. The same value is also returned with cv.download_point([173734, 187736, 19694], size=1, mip=0, coord_resolution=[4,4,40])

However, cloudvolume gets the right data if the box is larger.

bbox = cloudvolume.Bbox([173734, 187738, 19694], [173734+2, 187738+2, 19694+2])
#                                                                           ^ 
#                                                              Adding an additional voxel in z

svid = cv.download(
    bbox,
    coord_resolution=[4, 4, 40],
    agglomerate=False
)

returns 88681660698370372, 88681660698370350 (as a volume cutout), which are correct. Similarly, cv.download_point([173734, 187736, 19694], size=2, mip=0, coord_resolution=[4,4,40]) gives legitimate supervoxels.