Tracking challenge without bounding box


I am working on a video data analytic tool that mainly work with trajectory of objects.
Our tool outputs trajectories each as a series of 3D points over time.

I see that nuScenes’ tracking challenge provide instructions with a result format that requires 3D bounding box.
Does anyone have any recommendation for evaluating my point-trajectory output accuracy against nuScenes’ ground truth?

Thank you so much for any answer :slight_smile: