Converting autoware_msgs/DetectedObjectArray to KITTI raw 1.0 (.xml) format.

[devingw]

How can I go about Converting autoware_msgs/DetectedObjectArray to KITTI raw 1.0 (.xml) format so that I can load the data onto CVAT.

[maxime-clem]

I do not think there is any tool readily available to do that.
You can write a simple ROS 2 node in python to subscribe to the topic and have a callback that writes the message content in the format you desire.

[devingw]

I do not think there is any tool readily available to do that. You can write a simple ROS 2 node in python to subscribe to the topic and have a callback that writes the message content in the format you desire.

Yeah it's just that the .XML file format for KITTI raw 1.0 is quite complicated. So, any advice?

[maxime-clem]

You can use the standard python library for XML : https://docs.python.org/3/library/xml.etree.elementtree.html#building-xml-documents
This makes it very easy to build XML documents.

[devingw]

You can use the standard python library for XML : https://docs.python.org/3/library/xml.etree.elementtree.html#building-xml-documents This makes it very easy to build XML documents.

Thank you, Maxime!

[ralwing]

@devingw Since i've seen Your question in CVAT repo, I suggest a different approach for uploading the recordings and annotations to CVAT, since the KITTI format doesn't support all the LIDAR/Cameras configurations.

1. Export point clouds and images from rosbag to PCD/PNG files
   There is a repository for rosbag that allows exporting individual messages.

For the export, we will need a configuration file containing the names of the topics to be exported:

# export.config
# syntax: TOPIC EXPORTER_NAME EXPORTER_ARGS..
/perception/lidar/pointcloud pcd --dir pcd
/image_raw/compressed image --dir cam_front

Then, executing the command:

ros2 bag export -c export.config --in rosbag2_data/

will result in the exported files being placed in the pcd and cam_front folders.

We then copy these files to the target folder and create a CVAT zip archive, which can be easily loaded into CVAT. Later, detected objects can be imported by creating an annotations.json file.
The only problem You may face is the format of the cuboid which is not documented.

[devingw]

Thanks @ralwing, I will give it a go!