Traffic anomaly understanding improved by combining reasoning and style models
Reason or Recite: Complementary Specialists with an LLM Judge for Traffic Anomaly Reasoning
Computer Vision and Pattern RecognitionComputation and Language
Summary
Road accidents and hazards need clear and accurate explanations, not just detection. The authors show that models good at precise reasoning struggle with open-ended descriptions, and vice versa. They built a system that uses two specialized AI models—one focused on step-by-step reasoning and one on matching writing style—and a judge AI to pick the best answer. This approach performed best on a traffic video challenge.
What this means in practice
- •For traffic monitoring teams: Detect, verify, and explain traffic incidents accurately by combining reasoning-focused and style-focused models with an AI judge.
- •For surveillance system developers: Integrate complementary vision-language models to improve automated video scene descriptions and cause explanations of anomalies.
Authors
Mayur Kulkarni
Abstract
Road anomalies such as collisions, violations, and sudden hazards demand more from video understanding than detection: a practical system must verify what happened, describe the scene, and explain the cause. We find these competences pull in opposite directions for vision-language models: closed-form questions reward precise, evidence-grounded reasoning, while open-ended answers are scored by similarity to reference text, where matching the annotation style matters as much as reasoning correctly. We present Reason-or-Recite (RoR), a two-specialist framework for the Traffic Anomaly Reasoning track of the 2026 AI City Challenge. RoR (1) rewrites the official chain-of-thought annotations into a structured, temporally grounded schema and LoRA-tunes a 9B vision-language model on the enhanced traces; (2) fine-tunes a 4-bit-quantized 27B model on answers alone, absorbing the annotation style; and (3) selects between the two candidates per question with a training-free LLM judge. RoR ranks first on the in-domain leaderboard with mean score 0.6788.