Trustworthy AI Framework for Cultural Heritage Point Cloud Segmentation

Summary

A paper published in npj Heritage Science presents an architecture for trustworthy point cloud semantic segmentation in cultural heritage (CH) applications. The framework, called BubblEX, embeds explainability directly into the AI processing pipeline — from raw data input and pre-processing through semantic segmentation — by associating architectural elements with interpretable prototypical parts. The research introduces three dimensions of explainability: contextual (ensuring identified featur