What can we do ?

[lironesamoun]

Hello,

As I'm not yet familiar with Python and it not described on the ReadMe, I just would like to know, what can we do with your tool ?

Actually I'm working with the KITTI dataset, especially with the Odometry part and I would like to compare my results with the ground truth provided by KITTI.

Thank you,

Regards

[hunse]

Hi! The main point of this project is to provide a Python wrapper on the KITTI dataset, so that it's easy to load KITTI images, odometry data, velodyne points, etc. in Python. If you prefer to use C++ or MATLAB, there are tools provided in the KITTI raw data devkit. There's also a few other tools on the [KITTI raw data page].

Also, I should note that this project is focused on the raw KITTI data, since this is what I use. If you're looking to work with the specific odometry datasets (from this page), I unfortunately don't have any tools for that yet.

[lironesamoun]

Hi !

Thank you for your answer Hunse.
Yeah there are tools but not specially for the Odometry part. I don't use the raw part. I'll try to find a c++ code or otherwise write it on my own.

Thank again.

[LiShuaixin]

@hunse Hi, thank for your sharing. I’m working with KITTI Velodyne data to test my SLAM solution. I noticed that you said this code was focused on raw data. Would you please tell me the differences between odometry raw data and odometry data? Many thanks!

[hunse]

KITTI provides a number of different datasets for testing different types of systems. The most basic is the raw data, which is all the data they record while driving around (cameras, velodyne, odometry, etc.). This data has minimal preprocessing, hence "raw".

They also provide datasets for testing specific systems. These include a "stereo" dataset for testing stereo vision systems, an an "odometry" dataset for testing visual odometry/SLAM. These datasets all have both training and testing segments, where the ground truth for the testing segment is withheld. This allows you to submit your results for the testing segment and have an official ranking on the KITTI webpage. Thus these datasets are competitions, whereas the raw data is for your own exploration and evaluation.

As for the differences between the actual data, I'm not sure. It could be that the competition data is more processed in some way. If you're curious about the differences, you should ask the KITTI developers.