RangeLDM: Fast Realistic LiDAR Point Cloud Generation

• DOI: 10.48550/arxiv.2403.10094
• Publication Date: 2024-03-15
• Authors: Qianjiang Hu; Zhimin Zhang; Wei Hu
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV); Computer Science - Computer Vision and Pattern Recognition; FOS: Electrical engineering, electronic engineering, information engineering; Electrical Engineering and Systems Science - Image and Video Processing

Autonomous driving demands high-quality LiDAR data, yet the cost of physical sensors presents a significant scaling-up challenge. While recent efforts have explored deep generative models to address this issue, they often consume substantial computational resources with slow generation speeds while suffering from lack realism. To these limitations, we introduce RangeLDM, novel approach for rapidly generating range-view point clouds via latent diffusion models. We achieve by correcting data distribution accurate projection range images Hough voting, which has critical impact on learning. then compress into space variational autoencoder, and leverage model enhance expressivity. Additionally, instruct preserve 3D structural fidelity devising range-guided discriminator. Experimental results KITTI-360 nuScenes datasets demonstrate both robust expressiveness fast speed our cloud generation.