BD-HazardVLM: Probing Vision-Language Models for Latent Defensive-Driving Hazards in Bangladesh Road Scenes
Researchers tested Vision-Language Models (VLMs) like Qwen2.5-VL-7B to understand their ability to perform fine-grained defensive driving reasoning in complex, weakly structured road scenes from Bangladesh. The study used a diagnostic benchmark called BD-HazardVLM, showing that while models can correctly identify general hazards, they struggle significantly with detailed aspects like hazard category or exact defensive actions.