LiSeg: Lightweight Road-object Semantic Segmentation In 3D LiDAR Scans For Autonomous Driving

LiDAR based perception module plays an important role in autonomous driving. However, the present CNN models are designed for image processing but not LiDAR point clouds. The performances of such models are limited by the great memory consumption and heavy computation cost. In this work, a lightweight CNN model, Liseg, is proposed to perform real-time road-object semantic segmentation on LiDAR point cloud scans for autonomous driving. The model size of Liseg is several times smaller than others, while achieving high accuracy.