Representation Learning for Historical Map Interpretation
Hinweis: Die folgenden Informationen sind in Englisch verfasst.
Summary
Historically, maps have been designed to communicate geographic information in a clear and structured way. However, historical maps pose unique challenges: they contain diverse visual styles, symbolic representations, and embedded contextual knowledge that are often difficult to interpret, even for modern AI models. While recent advances in large-scale models have improved general visual and language understanding, they remain limited in extracting and reasoning overthe rich, multimodal information found in historical cartography.
The goal of this research is therefore to develop a unified framework for historical map understanding that goes beyond simple visual recognition. It focuses on building multimodal representations, enabling semantic and spatial reasoning, and integrating these capabilities into an interactive agent. By doing so, the project aims to support both expert analysis and broader accessibility of historical geographic knowledge, while contributing to the development of AI systems with more robust, grounded spatial intelligence.
Kick off
01.06.2025
Researchers
Lead: Prof. Dr. Lorenz Hurni
Second Advisor: Prof. Dr. Konrad Schindler
Doctoral student: Jiakun Xu
Funding sources
Own resources of professorship