I think the main reason would be that they can compare their scans directly against the original models.

If they were to use real world objects, any comparison would have to use another scanner, which would introduce its own bias.