SpotNet: An Image Centric, Lidar Anchored Approach To Long Range Perception

TL;DR

SpotNet introduces a fast, single-stage, image-centric approach for long-range 3D object detection that efficiently fuses LiDAR and camera data, scaling independently of range and enabling transfer across image resolutions.

Contribution

The paper presents a novel LiDAR-anchored, image-centric detection architecture that scales with range as O(1) and transfers across image resolutions without retraining.

Findings

Achieves accurate 3D detection with sparse LiDAR data.

Scales independently of range, unlike BEV methods.

Transfers across image resolutions without retraining.

Abstract

In this paper, we propose SpotNet: a fast, single stage, image-centric but LiDAR anchored approach for long range 3D object detection. We demonstrate that our approach to LiDAR/image sensor fusion, combined with the joint learning of 2D and 3D detection tasks, can lead to accurate 3D object detection with very sparse LiDAR support. Unlike more recent bird's-eye-view (BEV) sensor-fusion methods which scale with range $r$ as $O(r^2)$, SpotNet scales as $O(1)$ with range. We argue that such an architecture is ideally suited to leverage each sensor's strength, i.e. semantic understanding from images and accurate range finding from LiDAR data. Finally we show that anchoring detections on LiDAR points removes the need to regress distances, and so the architecture is able to transfer from 2MP to 8MP resolution images without re-training.