The article discusses UltraLiDAR, a system designed to address the issues of sparse point clouds captured by low-beam LiDAR. This data-driven framework is used for scene-level LiDAR completion, generation, and manipulation. It uses a compact, discrete representation that encodes the point cloud’s geometric structure, making it robust to noise and easy to manipulate. The system densifies sparse point clouds as if they were captured by high-density LiDAR, significantly reducing costs. Experiments show that this approach can significantly improve the performance of downstream perception systems and generate more realistic point clouds compared to previous methods.
Publication date: 2 Nov 2023
Project Page: https://waabi.ai/ultralidar/
Paper: https://arxiv.org/pdf/2311.01448
