LaserMix for Semi-Supervised LiDAR Semantic Segmentation

FIDNet [IROS'21] PolarNet [CVPR'20] Cylinder3D [CVPR'21] 10% SalsaNext [ISVC'20] mIoU (%) (a) (b) (c) PolarStream [NeurIPS'21] 20% 50% Full Ours, Range View Ours, Voxel Sup.-only, Range View Sup.-only, Voxel (57.5) lower mid upper car road lower mid upper lower mid upper Figure 1. Left: The LiDAR point cloud contains strong spatial prior. Objects and backgrounds around the ego-vehicle have a patterned distribution on different (lower, middle, upper) laser beams. Middle: Following the scene structure, the proposed LaserMix blends beams from different LiDAR scans, which is compatible with various popular LiDAR representations. Right: We achieve superior results over SoTA methods in both low-data (10%, 20%, and 50% semantic labels) and high-data (full semantic labels) regimes on nuScenes [11].