Self-Supervised Pre-Training Boosts Semantic Scene Segmentation on LiDAR Data

TL;DR

This paper demonstrates that self-supervised pre-training with Barlow Twins significantly improves semantic scene segmentation accuracy on LiDAR data, especially for under-represented classes, reducing the need for extensive labeled datasets.

Contribution

The study introduces a self-supervised pre-training approach using Barlow Twins for LiDAR-based semantic segmentation, enhancing performance with limited labeled data.

Findings

Pre-training boosts segmentation accuracy.

Improves performance on under-represented categories.

Reduces reliance on large labeled datasets.

Abstract

Airborne LiDAR systems have the capability to capture the Earth's surface by generating extensive point cloud data comprised of points mainly defined by 3D coordinates. However, labeling such points for supervised learning tasks is time-consuming. As a result, there is a need to investigate techniques that can learn from unlabeled data to significantly reduce the number of annotated samples. In this work, we propose to train a self-supervised encoder with Barlow Twins and use it as a pre-trained network in the task of semantic scene segmentation. The experimental results demonstrate that our unsupervised pre-training boosts performance once fine-tuned on the supervised task, especially for under-represented categories.