3D-Aware Visual Question Answering about Parts, Poses and Occlusions