Workshop Description

The rise of large-scale foundation models, trained on billions of parameters and massive multimodal datasets, has reshaped the paradigms of many industries, opening transformative opportunities for cultural heritage conservation, visualization, and virtual restoration. However, heritage-specific foundation models face unique challenges: highly sparse and diverse datasets, strict requirements for precision and tolerance (often at millimeter level), and a lack of standardized multimodal semantic alignment.
This thematic session aims to address these critical gaps by exploring frameworks for building cultural heritage-oriented large models, from establishing robust, semantically rich, cross-modal datasets to developing intelligent annotation methods at the component level of heritage assets. It will discuss the knowledge fusion strategies needed to align data sources including text, images, 3D point clouds, drawings, multimedia, and sensor data, ensuring high-quality, multi-perspective, multidisciplinary annotation.

Moreover, the session will highlight how foundation models can integrate domain-specific knowledg`e from structural systems, material sciences, historical archives, and conservation practice, thus serving as a powerful tool for heritage protection, risk assessment, and public engagement. Aligned with the ISPRS mission, this session will connect experts pioneering these data-centric and AI-enhanced approaches, promoting synergy between spatial information science, digital humanities, and cultural heritage disciplines.