Scene-LLM: Extending Language Model for 3D Visual Understanding and Reasoning