KITTI 360 Trajectory visualizations

KITTI-360

KITTI is the most important and famous dataset for SLAM research as well as computer vision things such as depth predictions or optical flow.

Recently, they released an extension of the dataset named KITTI-360.

Compare to the original dataset, this dataset has semantic labels for both 2D and 3D (i.e., point clouds from LiDAR scans) data. (see below captures)

In total 9 sequences exist in there.

Those names are:

2013_05_28_drive_0000_sync, 2013_05_28_drive_0002_sync, 2013_05_28_drive_0003_sync, 2013_05_28_drive_0004_sync, 2013_05_28_drive_0005_sync, 2013_05_28_drive_0006_sync, 2013_05_28_drive_0007_sync, 2013_05_28_drive_0009_sync, 2013_05_28_drive_0010_sync.

For convenience, from here, I’ll call them simply like 00, 02, 03, 04, …, 10.

BTW, I’m a SLAM researcher, so I am mainly interested in the shape of the trajectories because the original KITTI dataset’s odometry sequences had a quite easy shape of a trajectory and loops.

so I drew the trajectories.

Let’s take a journey together.

NOTE: blue is the start, red is the end

Sequence 00 (top / side views)

Sequence 02 (top / side / zoom1 / zoom2 views)

Sequence 03 (top view)

Sequence 04 (top / zoom1 / zoom2 / zoom3 views)

Sequence 05 (top / side views)

Sequence 06 (top / side views)

Sequence 07 (top / side views)

Sequence 09 (top / side1 / side2 views)

Sequence 10 (top view)

The trajectories have become much longer and more complex. For example, many reverse loops were included.

I think this is a new and a good chance for SLAM (including place recognition) researchers. I hope that many studies in the future will be carried out using this dataset. I also say that I will definitely use this KITTI 360 dataset for my next study.