Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA