Plan2Map: A Multimodal Benchmark for Document-Grounded Geospatial Boundary Reconstruction from Planning Records
Researchers have introduced Plan2Map, a new benchmark designed to evaluate AI's ability to reconstruct geospatial boundaries from UK planning documents. This multimodal dataset includes 208 cases, requiring systems to extract spatial information from text, maps, and annotations to generate machine-readable boundaries. A proposed system, GeoPlanAgent, utilizes a geospatial tool-in-the-loop approach to achieve competitive results, significantly outperforming direct vision-language model predictions. AI
IMPACT Establishes a new benchmark for multimodal geospatial AI, potentially improving automated analysis of planning documents.