AI vs Human Expert Reasoning: Assessing Agreements in Building Typology Predictions based on Street View Imagery

arXiv cs.AI · 6 hours ago

Vision-Language Models were evaluated on their ability to predict building typologies from Google Street View images compared to classifications by human experts. GPT-4o, Claude 3.5 Sonnet, and Gemini 2.0 Flash achieved approximately 70% average accuracy, with Chain-of-Thought prompting improving performance stability. The models relied primarily on visual indicators while experts incorporated broader contextual and domain knowledge, suggesting VLMs could serve as complementary tools for urban analysis at scale despite reasoning differences from human experts.