Pseudo-LiDAR++: Accurate Depth for 3D Object Detection in Autonomous Driving

• DOI: 10.48550/arxiv.1906.06310
• Publication Date: 2019-01-01
• Authors: Yurong You; Yan Wang; Wei‐Lun Chao; Divyansh Garg; Geoff Pleiss; Bharath Hariharan; Mark Campbell; Kilian Q. Weinberger
• Keywords: Leverage (statistics); Benchmark (surveying); Depth map

Detecting objects such as cars and pedestrians in 3D plays an indispensable role autonomous driving. Existing approaches largely rely on expensive LiDAR sensors for accurate depth information. While recently pseudo-LiDAR has been introduced a promising alternative, at much lower cost based solely stereo images, there is still notable performance gap. In this paper we provide substantial advances to the framework through improvements estimation. Concretely, adapt network architecture loss function be more aligned with estimation of faraway --- currently primary weakness pseudo-LiDAR. Further, explore idea leverage cheaper but extremely sparse sensors, which alone insufficient information detection, de-bias our We propose depth-propagation algorithm, guided by initial estimates, diffuse these few exact measurements across entire map. show KITTI object detection benchmark that combined approach yields stereo-based outperforming previous state-of-the-art accuracy 40%. Our code available https://github.com/mileyan/Pseudo_Lidar_V2.