output tensor coordinates to use dpoint instead of point.  This way, we can
obtain sub-pixel coordinates if we need them.