Enhancing Vision-Language Models with Scene Graphs for Traffic Accident Understanding | Read Paper on Bytez