Answers are grounded in this paper and cite their sources.
The application of artificial intelligence to cultural heritage is a multidisciplinary research field that brings together experts from different backgrounds, merging computer science and historical expertise. Recently, the systematic digitization of collections has opened up new possibilities for experimentation, but human-centered AI and qualitative approaches continue to be preferred over unsupervised quantitative ones. In the case of manuscript codices, a variety of deep neural networks have been developed for the automatic analysis of pages, the extraction of layout components, or the transcription of text. Multiple attempts have been also made in the illuminated specimens to detect miniatures and other decorative elements. Such models are usually intended for a specialist audience and developed to speed up the consultation process, but they can also be integrated into visualization interfaces and become valid tools for enhancing volumes. The paper presents the deep neural network developed in the “IlluminAI Project” of the CNR-ISPC and the incremental learning cycle used, at the same time, for its training and the semi-automatic labeling of an original dataset. After a brief introduction to contextualize the project, we will describe and discuss the cycle implemented to optimize human involvement to verify and correct the predictions made by the model.
No takes yet. Share an insight, caveat, or question.
Minisini et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: