VISTA (Vision-Language Model for Intelligent Road Scene Understanding and Traffic Risk Assessment) is a compact, 3B-scale Vision-Language Model designed to interpret low-resolution traffic videos and produce semantically rich, risk-aware captions in real time. Distilled from multi-agent supervision using ChatGPT-4o and o3-mini, VISTA blends high-level scene understanding with contextual traffic risk reasoning. Its lightweight architecture allows scalable deployment on edge devices, making it ideal for enhancing safety intelligence across existing traffic camera networks without infrastructure upgrades.
