What is the purpose of <conf> column?

[slavafive]

I am trying to compute MOT metrics. I logically assumed that the conf column describes the confidence of an observation (a float between 0 and 1). However, recently I've found out that setting the conf column in a prediction dataset to random values between 0 and 1 always results in the same metric values. Therefore, it seems like this column is useless. Does somebody have the same issue?

PS: in my ground truth dataset the conf column contains only values of 1.

[jeethesh-pai]

This might be a possible copy of issue. Confidence column is usually ignored for evaluation. It just tells how sure the detector is about the presence of object. In ground truth it is set to 1 because human annotator is 100% confident that the object is present there.

[JonathonLuiten]

yes it is not used at all in tracking evaluation here. In previous MOTChallenge 'detection' (not tracking) benchmarks, this was used, and thus it is a remnant of that.