BD-HazardVLM: Probing Vision-Language Models for Latent Defensive-Driving Hazards in Bangladesh Road Scenes

A new diagnostic benchmark, BD-HazardVLM, shows that three vision-language models — Qwen2.5-VL-7B, InternVL3-8B, and MiniCPM-V-4.6 — achieve 88.0 percent to 89.2 percent binary hazard-presence accuracy on 500 Bangladesh road-scene images, essentially matching the 89.0 percent majority-class baseline, while fine-grained defensive-driving reasoning stays far weaker. Category accuracy remains below 40 percent, risk-timing accuracy ranges from 1.6 percent to 12.4 percent, and exact defensive-action matching stays below 2 percent; InternVL3-8B posts the highest severity accuracy at 61.2 percent but has zero recall on the 58 high-severity cases. The benchmark, drawn from 375 RSUD20K and 125 TFP-BD images, concludes that coarse hazard recognition does not imply reliable fine-grained defensive-driving reasoning and motivates geographically diverse safety evaluation.

arXiv:2610.20627v1 Announce Type: new Abstract: Vision-language models are increasingly used for driving-scene understanding, yet their defensive reasoning under weakly structured Global South traffic remains unclear. We introduce BD-HazardVLM, a diagnostic benchmark of 500 Bangladesh road-scene images, comprising 375 from RSUD20K and 125 from TFP-BD, annotated for hazard presence, primary hazard category, risk timing, severity, defensive actions, visible evidence, and hallucination- and Western-road-bias traps. We evaluate Qwen2.5-VL-7B, InternVL3-8B, and MiniCPM-V-4.6 using a shared zero-shot task definition and structured output schema. Binary hazard-presence accuracy ranges from 88.0 percent to 89.2 percent, essentially matching the 89.0 percent majority-class baseline. Fine-grained reasoning is substantially weaker: category accuracy remains below 40 percent, risk-timing accuracy ranges from 1.6 percent to 12.4 percent, and exact defensive-action matching remains below 2 percent. Although InternVL3-8B achieves the highest severity accuracy at 61.2 percent, it has zero recall on the 58 high-severity cases. Qwen exhibits severe category mode collapse, while all models show weak latent-hazard anticipation and near-zero self-reported trap recognition. For the evaluated models, these findings show that coarse hazard recognition does not imply reliable fine-grained defensive-driving reasoning and motivate geographically diverse safety evaluation.