0
$\begingroup$

There are two popular metrics for object detection: Average precision and Average recall. Do you can explain with examples, what are the cases to use AP, and what are the cases to use AR?

I agree that discrete metrics (with threshold) are bad, but the question is: why there are two metrics AP and AR, what is the difference between them?

To apply scoring rules I have to know the cost of misclassification. My particular task is object detection in autonomous driving, and it's not obvious how to assign this cost. Ultimately, I need to minimize the accident rate and maximize throughput, passenger comfort or some other subjective metric. Moreover, I would have to look at the autonomous vehicle as a whole (sensing, perception, planning, etc).

Additional thought - all object detection metrics, known for me, are discrete, because they include IOU (intersection over union).

At this point I want to know:

  1. The difference between AP and AR, especially I have these metrics already implemented.
  2. What proper metrics do I can use in my particular task? If you can suggest truly continuous metric without thresholds, I would be grateful to you.

I think this question is NOT a duplicate of "Why is accuracy not the best measure for assessing classification models?"

This question is about difference between AP and AR.

Improve this question
$\endgroup$
2
  • $\begingroup$ Don't use either one. They suffer from the exact same issues as accuracy. $\endgroup$
    – Stephan Kolassa
    Commented Jul 24 at 7:58
  • $\begingroup$ @StephanKolassa tried to clarify my question $\endgroup$
    – Ars ML
    Commented Jul 24 at 11:02

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.