The difference between depth_dense and the velodyne file in the kitti data set

[yangfan293]

hello! have great insteresting about your work。For depth monitoring, you use a depth set trained on other networks(depth_dense) to monitor vis_depth, but I noticed that there is a velodyne file in the kitti data set, which is also lidar point cloud data, to get depth information, Monodtr uses it to monitor ins_depth. I would like to know if your depth_dense and kitti.velodyne have any difference in depth information? Why do you two represent different depth?

[SPengLiang]

I am very sorry for the late reply.

1. "depth_dense" provides dense depth information by employing depth completion on sparse depth maps (namely, from kitti.velodyne) while "kitti.velodyne" contains only sparse depths (from sparse Lidar points projection).
2. We use "depth_dense" because it provides more instance depth information, which benefits the learning of decoupled depth (vis. and att. depth in our work).

[yangfan293]

Got it. Thank you. You're great!