I successfully ran tracking for the nuscenes dataset. I want to re-identify the tracked objects also by the original Instance_token that exists in nuscenes instead of the numerical tracking_id (Sort of matching between predicted and GT tracks/objects).
Could a distance comparison work in this case (similar to the tracking evaluation in nuscenes-devkit) or is it better to perform a bipartite matching between GT and Prediction ?
I successfully ran tracking for the nuscenes dataset. I want to re-identify the tracked objects also by the original Instance_token that exists in nuscenes instead of the numerical tracking_id (Sort of matching between predicted and GT tracks/objects).
Could a distance comparison work in this case (similar to the tracking evaluation in nuscenes-devkit) or is it better to perform a bipartite matching between GT and Prediction ?