"Moroccan Volubilis Dataset"

"The Volubilis Dataset is a curated collection of high-resolution, pixel-annotated images designed to advance semantic segmentation research in cultural heritage preservation. Captured at the Roman archaeological site of Volubilis in Morocco\u2014a UNESCO World Heritage monument\u2014the dataset documents a diverse range of architectural features and deterioration patterns under real-world conditions. It contains 237 RGB images, acquired from multiple viewpoints and illumination settings to reflect practical constraints encountered by archaeologists and conservation teams. Each image is labeled at the pixel level into seven semantic classes: structural deterioration, mosaics, sky, built structures, vegetation, ground, and background. Annotations were produced using the CVAT platform and validated through multi-pass quality checks to ensure consistency and accuracy. By explicitly modeling both deterioration-related elements and contextual site features, the Volubilis Dataset provides a rich foundation for training and evaluating deep learning models aimed at automated heritage monitoring. It enables fine-grained segmentation, supports reproducible benchmarking across architectures such as U-Net, PSPNet, and DeepLabV3+, and contributes a critical resource for developing scalable, data-driven conservation strategies for vulnerable archaeological sites."