Convolutional Neural Networks for 3D Vision System Data : A review

3D vision systems are becoming increasingly accessible and as such there has been a lot of progress in the design of 3D Convolutional Neural Networks (3D CNNs). This paper will provide a review of the significant enhancements which have been made in deep learning techniques for understanding data from 3D vision systems in the tasks of 3D object classification and 3D semantic segmentation. We compare the documented results the state-of-the-art architectures have achieved on benchmark datasets and outline the advantages of the different techniques which have been developed in a number of different categories of approaches including view-based, voxelbased and point-based architectures amongst others. We also give insights into common themes and trends in this very active field of research.