Self-supervised Learning of Point Clouds via Orientation Estimation

TL;DR

This paper introduces a self-supervised learning method for point clouds by predicting their rotations, enabling effective feature learning for 3D shape classification and keypoint prediction with fewer labels.

Contribution

It proposes a novel rotation prediction task for self-supervised learning on point clouds, improving downstream task performance and complementing existing methods.

Findings

Outperforms state-of-the-art on ShapeNet and ModelNet

Features are complementary to other self-supervised methods

Enables effective learning with fewer labeled data

Abstract

Point clouds provide a compact and efficient representation of 3D shapes. While deep neural networks have achieved impressive results on point cloud learning tasks, they require massive amounts of manually labeled data, which can be costly and time-consuming to collect. In this paper, we leverage 3D self-supervision for learning downstream tasks on point clouds with fewer labels. A point cloud can be rotated in infinitely many ways, which provides a rich label-free source for self-supervision. We consider the auxiliary task of predicting rotations that in turn leads to useful features for other tasks such as shape classification and 3D keypoint prediction. Using experiments on ShapeNet and ModelNet, we demonstrate that our approach outperforms the state-of-the-art. Moreover, features learned by our model are complementary to other self-supervised methods and combining them leads to further performance improvement.