VORD aims to enhance our understanding of cultural heritage by extracting quantitative information from vast artistic and visual cultural heritage databases. Utilizing advanced machine learning models, our project focuses on automatically extracting various visual features, including color analysis, figure positioning, and interpreting the dense embedding space of these images. By combining different models, we enable comprehensive visual quantitative semantic analysis, expanding the breadth of art historical questions and offering new perspectives on art historical narratives. This approach shifts the focus from analyzing full images to detailed object semantic analysis, exploiting inherent formalistic information that is otherwise invisible.

Our methodology enhances the transparency and interpretability of machine learning models applied to cultural heritage data, making them more accessible for researchers, curators, and educators. The project highlights the potential of combining advanced machine learning techniques with structured semantic information to improve the analysis and understanding of art historical content. This approach facilitates the sharing of metadata in common file formats, promoting reuse and interoperability, and complements existing methodologies without addressing interoperability issues between legacy database systems and data ontologies.