DeepMapping2: Self-Supervised Large-Scale LiDAR Map Optimization

Multiway registration LeGO-LOAM GPS KITTI ( 0018 ) ICP+DM2 (ours) KITTI (0027) Trajectory example: Straight (good) Corner (good) Straight (bad) Corner (bad) Figure 1. Mapping result on the KITTI dataset. The bird's eye view map is shown together with the estimated sensor pose of each frame. Each pose is represented by an arrow indicating the xy-coordinate and heading (yaw angle) as shown in the bottom examples. The color indicates the frame index in the trajectory. Challenging regions are zoomed in for a clear view. Best viewed in color.