Context and Geometry Aware Voxel Transformer for Semantic Scene Completion