Sylvie Treuillet is an Associate Professor at the University of Orléans. She has been working as a specialist in computer vision and pattern recognition since 1994. She received MS degree in electrical engineering and obtained a PhD in computer vision at the University of Clermont-Ferrand. She has been part of several European or international projects, has supervised about fifteen doctoral theses and has published more than 100 articles on image processing, computer vision and machine learning in biomedical and industrial applications. Her recent research at the PRISME laboratory is particularly focused on the development of diagnostic tools for historic buildings based on multimodal data processing (image/3D) and deep learning. She is member of IAPR TC 19 (Computer Vision for Cultural Heritage Applications.)

Diagnostic tools for historic buildings using deep learning from 3D survey images

The planning of restoration operations of cultural heritage buildings requires a precise and updated diagnosis of the different areas of deterioration. At the scale of large buildings, an exhaustive survey is difficult to perform by traditional methods of visual inspection by experts. Rapid advances in computer vision and deep learning are inspiring the development of automatic damage detection approaches. The tutorial will present a new approach to automatically detect stone deterioration at the scale of a castle by deep learning techniques from existing color images acquired for 3D modeling. A first part will be devoted to an introduction to deep learning for the lay audience, followed by a practical application with a pre-trained model. In a second part, we will propose to show how to train a network from a dataset, and how to evaluate the performance of the model predictions.