Joint Learning of 2D-3D Weakly Supervised Semantic Segmentation