A Deep-Learning-Based Approach for Saliency Determination on Point Clouds
Yassine Souai, Ghazal Rouhafzay, Ana-Maria Creţu · 2022
Laser scanners recording a huge number of data points from different surfaces are widely used to capture the exact geometry of 3D objects. These large amounts of data require intelligent solutions to be examined and processed efficiently. Deep-learning-based approaches have found their way into many data analytics applications for processing such large datasets, categorizing them, or even determining the most informative portion of the data. This research focused on 3D deep-learning techniques directly applied to point clouds to determine the most important features of a 3D shape. More specifically, this research adopted PointNet as a backbone architecture for feature extraction from 3D point clouds and computed a gradient-based class activation mapping (Grad-CAM) on each object to create a 3D importance/saliency map. Experiments confirmed the success of the proposed approach in the determination of important features of 3D objects as compared with the ground truth.