Missing data

[chadrs2]

It seems like you don't have bounding boxes for every frame. Why is that? Is the data missing? I've downloaded other sequences besides one, and for example sequence 21 doesn't even have a 3D bounding box folder. Can you clearly say which missions have bounding box data for the entire mission/sequence?

[artzha]

Hi @chadrs2, thank you for your interest in our dataset! We prioritize annotating frames with more objects of interest, and thus do not provide annotations for an entire sequence.

To download all of the object and semantic segmentation annotations, we recommend downloading the CODa_full_split.zip. This is a self-contained .zip file with rectified images/ego-motion compensated point clouds, 3D bounding box labels, 3D semantic segmentation labels, calibrations, and metadata. Additionally, it contains a metadata directory with json files for each sequence containing all of the annotated frames. You can find documentation describing the json file structure on our Data Report. Hope this helps!

[chadrs2]

Do you have or can you provide some sort of documentation that shows the percentage of annotated frames for each sequence? Just so that I don't have to download each one, rosbag play it with rviz, and see that only two small chunks of the sequence are labeled.

[artzha]

Unfortunately, we don't have a statistics file with the information that you're requesting. If you only want to download data for the labeled sequences instead of the whole dataset, you can download the .zip file mentioned above. If you have a specific use case in mind where downloading this zip file is insufficient, let me know and I can provide additional help.