Graduation Year


Document Type

Campus Only Senior Thesis

Degree Name

Bachelor of Science



Reader 1

Weiqing Gu

Reader 2

Nicholas Pippenger

Terms of Use & License Information

Terms of Use for work posted in Scholarship@Claremont.

Rights Information

2020 Mengyi Shan


3D object classification is one of the most popular topics in the field of computer vision and computational geometry. Currently, the most popular state-of-the-art algorithm is the so-called Convolutional Neural Network (CNN) models with various representations that capture different features of the given 3D data, including voxels, local features, multi-view 2D features, and so on. With CNN as a holistic approach, researches focus on improving the accuracy and efficiency by designing the neural network architecture. This thesis aims to examine the existing work on 3D object classification and explore the underlying theory by integrating geometric approaches. By using geometric algorithms to pre-process and select data points, we dive into an existing architecture of directly feeding points into a deep CNN, and explore how geometry measures how important different points are in a CNN model. Moreover, we attempt to extract useful geometric features directly from the object data to introduce the feature matrix representation, which can be classified with distance-based approaches. We present all results of experiments and analyzed for future improvement.

This thesis is restricted to the Claremont Colleges current faculty, students, and staff.