2020/12/08 by Yusuf H. Şahin, Sahin, Yusuf H., Alican Mertan +3
Computer Science · Engineering · #3D Shape Modeling and Analysis #68T45 68T07 #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Human Pose and Action Recognition #Machine Learning (cs.LG)
paper · pdf · doi:10.48550/arxiv.2012.04708
openalex publication_date 2020/12/08 · openalex created_date 2022/07/19 · openalex updated_date 2026/07/28
Learning new representations of 3D point clouds is an active research area in\n3D vision, as the order-invariant point cloud structure still presents\nchallenges to the design of neural network architectures. Recent works explored\nlearning either global or local features or both for point clouds, however none\nof the earlier methods focused on capturing contextual shape information by\nanalysing local orientation distribution of points. In this paper, we leverage\non point orientation distributions around a point in order to obtain an\nexpressive local neighborhood representation for point clouds. We achieve this\nby dividing the spherical neighborhood of a given point into predefined cone\nvolumes, and statistics inside each volume are used as point features. In this\nway, a local patch can be represented by not only the selected point's nearest\nneighbors, but also considering a point density distribution defined along\nmultiple orientations around the point. We are then able to construct an\norientation distribution function (ODF) neural network that involves an\nODFBlock which relies on mlp (multi-layer perceptron) layers. The new ODFNet\nmodel achieves state-of the-art accuracy for object classification on\nModelNet40 and ScanObjectNN datasets, and segmentation on ShapeNet S3DIS\ndatasets.\n