A Comparative Study of Feature Expansion Unit for 3D Point Cloud Upsampling

TL;DR

This paper compares different feature expansion units for 3D point cloud upsampling, highlighting their limitations and proposing a novel unit called ProEdgeShuffle that improves performance based on theoretical and experimental analysis.

Contribution

The paper introduces ProEdgeShuffle, a new feature expansion unit inspired by image super-resolution and graph CNNs, demonstrating improved upsampling results.

Findings

ProEdgeShuffle outperforms existing feature expansion units.

Most existing units process points independently, ignoring feature interactions.

Theoretical analysis explains the advantages of the proposed method.

Abstract

Recently, deep learning methods have shown great success in 3D point cloud upsampling. Among these methods, many feature expansion units were proposed to complete point expansion at the end. In this paper, we compare various feature expansion units by both theoretical analysis and quantitative experiments. We show that most of the existing feature expansion units process each point feature independently, while ignoring the feature interaction among different points. Further, inspired by upsampling module of image super-resolution and recent success of dynamic graph CNN on point clouds, we propose a novel feature expansion units named ProEdgeShuffle. Experiments show that our proposed method can achieve considerable improvement over previous feature expansion units.