Weakly Supervised SVM-Enhanced SAM Pipeline for Stone-by-Stone Segmentation of the Masonry of the Loire Valley Castles

A novel machine-learning pipeline, combining the Segment Anything Model (SAM) and Support Vector Machines (SVM), achieves highly accurate stone-by-stone segmentation of historic masonry with an 85% Dice coefficient. This approach automates the labor-intensive task of mapping stones and joints, critical for monitoring deterioration in monuments like Château de Chambord. The system leverages SAM's zero-shot learning capabilities while using SVM to classify segmented elements, allowing for adaptive segmentation through multiple SAM iterations based on remaining unsegmented stones. This scalable solution offers a promising tool for cultural heritage conservation, reducing the reliance on manual expert analysis and advancing restoration efforts.

Lucho, S., Treuillet, S., Desquesnes, X., Leconge, R., & Brunetaud, X. (2024). Weakly Supervised SVM-Enhanced SAM Pipeline for Stone-by-Stone Segmentation of the Masonry of the Loire Valley Castles. Journal of Imaging, 10(6), 148. https://doi.org/10.3390/jimaging10060148