num_predictions metric not returning expected value

cheind / py-motmetrics

:bar_chart: Benchmark multiple object trackers (MOT) in Python

MIT License

1.37k stars 258 forks source link

num_predictions metric not returning expected value #64

Closed areiner222 closed 4 years ago

areiner222 commented 4 years ago

I noticed that the num_predictions metric is defined twice in metrics.py (https://github.com/cheind/py-motmetrics/blob/125304fb53495de8a43f2768bf90eb932c4a87bb/motmetrics/metrics.py#L247). One uses the result of pred_frequencies while the other counts the rows in the matched dataframe. Both output the same value, but I don't follow how they are accounting for the uniqueness of the predicted ids. Why shouldn't the unique predicted ids be pred_frequencies.count()? Am I misunderstanding?

jvlmdr commented 4 years ago

Thanks for identifying this. The issue was fixed in the develop branch. See the diff here