EllipsoidNet: Ellipsoid Representation for Point Cloud Classification and Segmentation

TL;DR

This paper introduces EllipsoidNet, a novel approach that projects point clouds onto an ellipsoid surface to improve classification and segmentation accuracy by better exposing local geometric features.

Contribution

The paper proposes a new ellipsoid-based 2D representation for point clouds and a corresponding CNN, EllipsoidNet, which outperform existing methods in accuracy and efficiency.

Findings

Outperforms existing 2D representation methods on ModelNet40 and ShapeNet.

Effectively exposes local geometric features in ellipsoid space.

Achieves better accuracy with improved computational efficiency.

Abstract

Point cloud patterns are hard to learn because of the implicit local geometry features among the orderless points. In recent years, point cloud representation in 2D space has attracted increasing research interest since it exposes the local geometry features in a 2D space. By projecting those points to a 2D feature map, the relationship between points is inherited in the context between pixels, which are further extracted by a 2D convolutional neural network. However, existing 2D representing methods are either accuracy limited or time-consuming. In this paper, we propose a novel 2D representation method that projects a point cloud onto an ellipsoid surface space, where local patterns are well exposed in ellipsoid-level and point-level. Additionally, a novel convolutional neural network named EllipsoidNet is proposed to utilize those features for point cloud classification and segmentation applications. The proposed methods are evaluated in ModelNet40 and ShapeNet benchmarks, where the advantages are clearly shown over existing 2D representation methods.