TraIL-Det: Transformation-Invariant Local Feature Networks for 3D LiDAR Object Detection with Unsupervised Pre-Training

TL;DR

TraIL-Det introduces transformation-invariant local features and a novel architecture for 3D LiDAR object detection, significantly improving robustness and accuracy in outdoor scene perception for autonomous driving.

Contribution

The paper proposes TraIL features with rigid transformation invariance and a Multi-head self-Attention Encoder to enhance 3D object detection from LiDAR data, outperforming existing self-supervised methods.

Findings

Achieves higher mAP on KITTI and Waymo datasets.

Demonstrates robustness to variations in point density.

Outperforms contemporary self-supervised detection approaches.

Abstract

3D point clouds are essential for perceiving outdoor scenes, especially within the realm of autonomous driving. Recent advances in 3D LiDAR Object Detection focus primarily on the spatial positioning and distribution of points to ensure accurate detection. However, despite their robust performance in variable conditions, these methods are hindered by their sole reliance on coordinates and point intensity, resulting in inadequate isometric invariance and suboptimal detection outcomes. To tackle this challenge, our work introduces Transformation-Invariant Local (TraIL) features and the associated TraIL-Det architecture. Our TraIL features exhibit rigid transformation invariance and effectively adapt to variations in point density, with a design focus on capturing the localized geometry of neighboring structures. They utilize the inherent isotropic radiation of LiDAR to enhance local representation, improve computational efficiency, and boost detection performance. To effectively process the geometric relations among points within each proposal, we propose a Multi-head self-Attention Encoder (MAE) with asymmetric geometric features to encode high-dimensional TraIL features into manageable representations. Our method outperforms contemporary self-supervised 3D object detection approaches in terms of mAP on KITTI (67.8, 20% label, moderate) and Waymo (68.9, 20% label, moderate) datasets under various label ratios (20%, 50%, and 100%).