the model current use a different depth model instance to get a geometry first, is it good to use the same pda model for better efficiency?