right above this line it's mentioned that kitti expects z, y, x as height, width, length and on this line we have:

height, width, length = bbox_extent.z, bbox_extent.x, bbox_extent.y

Referring to this mesaurement image. I think it should rather be

height, width, length = bbox_extent.z, bbox_extent.y, bbox_extent.x

How to generate cyclist categories

could you please tell me the carla version?
@enginBozkurt

Hi,
I really appreciate the work you have done to put this together. So, I am trying to reproduce this with latest CARLA version and I know the API has changed a bit that's what I am trying to understand.

From what I understood from the docstring section of [this method is that the transformation from CARLA to kitti for the object location should rather be:

self.location = " ".join(map(str, [y, -z, x]))

           Converts the 3D object location from CARLA coordinates and saves them as KITTI coordinates in the object
            In Unreal, the coordinate system of the engine is defined as, which is the same as the lidar points
            z
            ▲   ▲ x
            |  /
            | /
            |/____> y
            This is a left-handed coordinate system, with x being forward, y to the right and z up 
            See also https://github.com/carla-simulator/carla/issues/498
            However, the camera coordinate system for KITTI is defined as
                ▲ z
               /
              /
             /____> x
            |
            |
            |
            ▼
            y 
            This is a right-handed coordinate system with z being forward, x to the right and y down
            Therefore, we have to make the following changes from Carla to Kitti
            Carla: X   Y   Z
            KITTI:-X  -Y   Z 

So, instead of KITTI: -X, -Y, Z it should be Y, -Z, X

@enginBozkurt let me know what you think.

Please provide better documentation about this repository. I don't know how to use it or what's the purpose of this repository.

Does your framework support stereo? The kitti dataset has stereo data too (aka image_2 and image_3), but i only see image_2 in your result

It looks this repo is using older Python API ?
Does this work with the CARLA version 0.9.12 ?

Thanks

Hi, thank you very much for your work, it will be very useful to me. I would like to ask you if it was possible to also add automatic creation of .label files (as in the SemanticKitty dataset) created using the '' perfect '' segmentation data provided by carla.SemanticLidarMeasurement (I believe this feature was added since carla 0.9.10 onwards : https://carla.readthedocs.io/en/0.9.10/ref_sensors/#semantic-lidar-sensor )