How did you select the Precision and Recall values from the PR curve?

[zxsitu]

❔Question

Hello author, how do you determine the "best" Precision and Recall values on the PR curve? Are there any criteria?
(e.g., determine the point on the curve that provides a good trade-off between Precision and Recall)

Additional context

Because I've looked at a lot of other object detection programs and they will only use AP and AP50, which I know is what mAP and mAP50 means. And they plot a PR curve, and other than that they don't have Precision and Recall values everywhere.
But your program also calculates Precision and Recall values, how did you get the results? Thank you.

[glenn-jocher]

@ilem777 P and R values are logged and displayed at maximum F1 confidence.

Full P, R, and F1 curves at all confidences are saved after test.py completes in runs/test/exp.

[glenn-jocher]

See #2057

[zxsitu]

@glenn-jocher Thank you, I understand.