Politecnico di Torino | Servizi per la didattica

KEYWORD

Area Engineering

Neuro-symbolic AI for Built Cultural Heritage

keywords ARTIFICIAL INTELLIGENCE, BUILT HERITAGE, POINT CLOUDS, SEMANTIC SEGMENTATION

Reference persons FRANCESCA MATRONE, LIA MORRA

External reference persons Manigrasso Francesco

Research Groups DIATI, geomatica, Geomatics Lab, SmartData@PoliTO, http://grains.polito.it/

Thesis type APPLIED RESEARCH

Description The lack of benchmarking data for the semantic segmentation of digital heritage scenarios is hampering the development of automatic classification solutions in this field. Built heritage 3D data feature complex structures and uncommon classes that prevent the simple deployment of available methods developed in other fields and for other types of data. The semantic classification of heritage 3D data would support the community in better understanding and analysing digital twins as well as facilitate restoration and conservation works. The ArCH dataset originates from the collaboration of different universities and research institutes and comprises annotated and non-annotated point clouds of cultural heritage sites. However, the data scarcity, the variety and uniqueness of the architectures, and the fine-grained categorization needed for segmentation make this task particularly hard for deep learning models. The goal of this project is to enrich state-of-the-art deep neural networks for point cloud segmentation using neuro-symbolic techniques, such as Logic Tensork Networks, to inject prior knowledge into the network.