AdaSplats: Adaptative Splats from Semantic Point Cloud for Fast and High-Fidelity LiDAR Simulation

Abstract

Fig. 1: Our pipeline to replicate the real-world: From a raw outdoor point cloud taken by a mobile mapping system, we perform automatic semantic segmentation, remove dynamic objects, and model the static environment with our adaptative splat modeling. The LiDAR sensor is simulated by sending rays through our splat model in an online fashion.