Review for NeurIPS paper: BlockGAN: Learning 3D Object-aware Scene Representations from Unlabelled Images