search

PRBonn / kiss-icp

A LiDAR odometry pipeline that just works
https://www.ipb.uni-bonn.de/wp-content/papercite-data/pdf/vizzo2023ral.pdf
MIT License
1.61k stars 322 forks source link

Add more timestamps #307

Closed benemer closed 8 months ago

benemer commented 8 months ago

For Apollo, we can switch to using Open3D's tensor library to read the additional attributes from the pcd. Note that Apollo is already de-skewed, so it's not needed, but it's nice to have access to it.

NCLT uses a Velodyne-32, so I used the same naive timestamp extraction based on the range image columns as in KITTI raw.

This btw requires Open3d >= 0.16, but I think that should not be a problem.

benemer commented 8 months ago

For Apollo, this makes the pipeline a bit slower, unsure if it is because of using o3d.t. Maybe double check on your side. We can also remove the timestamps again if we don't want to sacrifice runtime.

nachovizzo commented 8 months ago

For Apollo, this makes the pipeline a bit slower, unsure if it is because of using o3d.t. Maybe double check on your side. We can also remove the timestamps again if we don't want to sacrifice runtime.

:thinking: then no need to add it right ? or what would be the reasoning on adding the timestamps if we can't use it in the pipeline

benemer commented 8 months ago

Actually no. It was just something I did before realizing it was not needed for KISS-ICP. If people use our dataloaders for other projects it might be interesting, but I am fine with removing it.

nachovizzo commented 8 months ago

then I guess the addition for NCLT is nice

benemer commented 8 months ago

I also checked again the NCLT dataset, and here they say for the velodyne_sync directory:

these points are motion compensated for the Segway’s ego-motion during the scan and then recorded in the Segway body frame

Since we decided only to put timestamps if they are used by our pipeline, we can actually close this PR :)

Sorry for the noise!