Deep Learning Advances on Different 3D Data Representations: A Survey.

TLDR

3D data provides rich geometric information, and recent advances in large datasets and computing power now enable deep‑learning methods for tasks such as segmentation, recognition, and correspondence, though the challenges vary with the chosen representation. The paper surveys 3D data representations, contrasting Euclidean and non‑Euclidean forms. It reviews how deep‑learning techniques are applied to each representation and the specific challenges involved. The survey identifies key challenges and gaps in applying deep learning across different 3D data representations.

Abstract

3D data is a valuable asset the computer vision filed as it provides rich information about the full geometry of sensed objects and scenes. Recently, with the availability of both large 3D datasets and computational power, it is today possible to consider applying deep learning to learn specific tasks on 3D data such as segmentation, recognition and correspondence. Depending on the considered 3D data representation, different challenges may be foreseen in using existent deep learning architectures. In this work, we provide a comprehensive overview about various 3D data representations highlighting the difference between Euclidean and non-Euclidean ones. We also discuss how Deep Learning methods are applied on each representation, analyzing the challenges to overcome.