SGTBN: Generating Dense Depth Maps From Single-Line LiDAR

Depth completion aims to generate a dense depth map from the sparse and aligned RGB image. However, current methods use extremely expensive 64-line LiDAR(about $100,000) obtain maps, which will limit their application scenarios. Compared with LiDAR, single-line LiDAR is much less more robust. Therefore, we propose method tackle problem of completion, in aim info A dataset proposed based on existing dataset(KITTI). network called Semantic Guided Two-Branch Network(SGTBN) contains global local branches extract fuse for this task. guided upsampling module used our make full semantic images. Except usual MSE loss, add virtual normal loss increase constraint high-order 3D geometry network. Our outperforms state-of-the-art Besides, compared monocular estimation, also has significant advantages precision model size.