Deep Neural Networks (DNN) have proven themselves to be a useful tool in many computer vision problems. One of the most popular forms of the DNN is the Convolutional Neural Network (CNN). The CNN effectively learns features on images by learning a weighted sum of local neighborhoods of pixels, creating filtered versions of the image. Point cloud analysis seems like it would benefit from this useful model. However, point clouds are much less structured than images. Many analogues to CNNs for point clouds have been proposed in the literature, but they are often much more constrained networks than the typical CNN. This is a matter of necessity: common point cloud benchmark datasets are fairly small and thus require strong regularization to mitigate overfitting. In this dissertation we propose two point cloud network models based on graph structures that achieve the high-capacity modeling capability of CNNs. In addition to showing their effectiveness on point cloud classification and segmentation in typical benchmark scenarios, we also propose two novel point cloud problems: ATLAS Detector segmentation and Computational Fluid Dynamics (CFD) surrogate modeling. We show that our networks are much more effective than others on these new problems because they benefit from deeper networks and extra capacity that other researchers have not pursued. These novel networks and datasets pave the way for future development of deeper, more sophisticated point cloud networks.

Library of Congress Subject Headings

Neural networks (Computer science; Convolutions (Mathematics); Computer vision; Graph theory

Publication Date


Document Type


Student Type


Degree Name

Engineering (Ph.D.)

Department, Program, or Center

Computer Engineering (KGCOE)


Raymond Ptucha

Advisor/Committee Member

Nathan D. Cahill

Advisor/Committee Member

Christian A. Linte


RIT – Main Campus

Plan Codes