SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models