On Specifying and Performing Visual Tasks with Qualitative Object Models