Shapes that are out of the image bounds should be filled with np.nan instead of clipping to the bounds of the image. I am opening a case since this might not be necessary if it is handled at a higher level (see #19).
This also relates to computing background ROI offsets. Optimally, the grid should remove all invalid offsets before calling getShapePixels, alternatively one can call getShapePixels and then drop nan values if getShapePixels returns np.nan where an image is out of bounds.
Shapes that are out of the image bounds should be filled with
np.naninstead of clipping to the bounds of the image. I am opening a case since this might not be necessary if it is handled at a higher level (see #19).This also relates to computing background ROI offsets. Optimally, the grid should remove all invalid offsets before calling
getShapePixels, alternatively one can callgetShapePixelsand then dropnanvalues ifgetShapePixelsreturnsnp.nanwhere an image is out of bounds.