Deep Learning-Based Point Cloud Analysis

Deep learning has achieved impressive performance improvements over traditional methods for almost all vision tasks. Point cloud processing is no exception. Since 2017, researchers have become inclined to train end-to-end networks for tasks like point cloud classification, semantic segmentation, and object detection. More recently, other tasks like registration and odometry have also been solved using Deep learningDeep learning . These newer data-driven methods provide some benefits over traditional methods that rely on handcrafted features. Nevertheless, many traditional methods are still in practice due to their simplicity and speed, and they form the basis of newer methods. In this chapter, we discuss some Deep learning Deep learning -based methods for point cloud processing. This subset of methods has had a huge impact in this field and is representative of current research progress in computer vision. The Deep learning Deep learning methods for point cloud classification, semantic segmentation, and registration tasks are discussed. We explore several papers, with a focus on the proposed methods and associated details, while the experimental details are limited to performance evaluations on benchmark datasets. Other analyses such as ablation studies and miscellaneous details from the papers are omitted.