Challenges, prospects and conclusions
Deep learning for 3D images

The general challenges of deep learning also apply to 3D image analysis. Perhaps the greatest challenge is to find ways of learning with less data. Indeed, in today's industrial applications, setting up learning bases is often the main expense. This involves searching for data, cleaning it and, above all, annotating it. This challenge is particularly acute in the case of 3D data, which is generally more costly to obtain, manipulate and annotate. One of the most promising avenues to help meet this challenge is self-supervised learning, which enables models to be developed without annotation [H 5 012]...