LOGen: Toward Lidar Object Generation by Point Diffusion
This work addresses a domain-specific problem for autonomous driving applications by focusing on LiDAR object generation, representing an incremental advancement leveraging existing diffusion methods.
The paper tackles the problem of generating LiDAR scans of 3D objects, which is challenging compared to image and 3D object generation, by introducing a diffusion-based model that produces LiDAR point clouds with intensity and extensive control via conditioning, achieving quality measured with new 3D metrics on nuScenes and KITTI-360 datasets.
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.