Using non-rgb data as input

[s-m-t-c]

Hi, well done on creating this library!

I am trying to segment single band Sentinel 1 data.

I've tried passing in a numpy array as well as configuring my geotiff a few different ways (datatype, shape etc). I receive a range of different errors when running the step:

mask = "segment.tif" sam.generate( image, mask, batch=True, foreground=True, erosion_kernel=(3, 3), mask_multiplier=255 )

including:

RuntimeError: permute(sparse_coo): number of dimensions in the tensor input does not match the length of the desired ordering of dimensions i.e. input.dim() = 2 is not equal to len(dims) = 3

I'm curious if anyone has had success using segment-geospatial with single band or SAR data? If I dig around in the library I can see a few lines of code that appear to assume multi-band data but I could be mistaken.

[giswqs]

See facebookresearch/segment-anything#123