How should I go about calculating angular velocity for each sample_annotation? My guess would be to calculate the yaw_diff b/w previous and current rotation of the DetectionBox and dividing by the time. Am I missing something?

# Angular velocity of sample_annotations

Yeah, that’s the idea, but note that 2Hz=0.5s is a long time and therefore the angular velocity may not be very helpful.

Thanks, that makes sense - still would like to give it a try.

Since we do not have the previous & next key in “sample_result” as compared to “sample_annotation” - what is the best way to compute the angular velocity of the predictions in case one does not wish to add in a separate detection head?

Since the `sample_result`

references the sample, you can just go to next and prev from there. It is tricky though, because for the detection challenge there may not be a direct correspondence between the detections in consecutive keyframes. Hence you will need to do some matching yourself.