LOGen: Toward Lidar Object Generation by Point Diffusion

TL;DR

This paper introduces LOGen, a diffusion-based model for generating realistic LiDAR object point clouds, including intensity, with extensive control, advancing the capabilities of LiDAR scan synthesis for autonomous driving applications.

Contribution

LOGen is the first diffusion model tailored for LiDAR object generation, incorporating control mechanisms and new 3D metrics for quality assessment.

Findings

High-quality LiDAR object generation demonstrated on nuScenes and KITTI-360 datasets.

The model outperforms existing methods in realism and controllability.

New 3D metrics effectively evaluate LiDAR-specific generated data.

Abstract

The generation of LiDAR scans is a growing topic with diverse applications to autonomous driving. However, scan generation remains challenging, especially when compared to the rapid advancement of image and 3D object generation. We consider the task of LiDAR object generation, requiring models to produce 3D objects as viewed by a LiDAR scan. It focuses LiDAR scan generation on a key aspect of scenes, the objects, while also benefiting from advancements in 3D object generative methods. We introduce a novel diffusion-based model to produce LiDAR point clouds of dataset objects, including intensity, and with an extensive control of the generation via conditioning information. Our experiments on nuScenes and KITTI-360 show the quality of our generations measured with new 3D metrics developed to suit LiDAR objects. The code is available at https://github.com/valeoai/LOGen.