I just stumbled upon a new metric I've never heard about called False Positive rate at K recall (FPR-K).
Searching the internet I just managed to find more papers using the metric but none of them actually explain how is it calculated.
The paper was on matching feature points in multi-modal images.

I know what is AP@K, used in tasks involving some thresholding (i.e. image detection where we're using intersection over union to calculate our TP/FP etc).

My assumption regarding FPR-95 is that there's some thresholding involved as well (else why the "at K" term?), and as with thresholds, they control the precision / recall rate. So we set a threshold causing 95% recall, and now we calculate the FP rate (FP / FP + TN).

Can anyone validate or invalidate my assumption?


I've found the answer in some paper.

FPR at 95% TPR can be interpreted as the probability that a negative (out-of-distribution) example is misclassified as positive (in-distribution) when the true positive rate (TPR) is as high as 95%.
True positive rate can be computed by TPR = TP / (TP+FN), where TP and FN denote true positives and false negatives respectively.
The false positive rate (FPR) can be computed by FPR = FP / (FP+TN), where FP and TN denote false positives and true negatives respectively.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.