Can I get a meaningful false positive rate from an A/~A classifier (ie, with no meaningful true negative count)?

Question

I have an application that basically circles things in images. I need to find the false positive rate. Wikipedia says it's given by (number of False Positives) / (number of False Positives + number of True Negatives).

I don't have a true negative count, and I'm not sure it means anything here, since the application does not explicitly say no.

All I really have to work with are the number of items it was supposed to circle, the number of those it actually circled, and the number of false positives.

Is there anything I can do with this data to get some sort of useful false positive percentage?

Answer 1

Your description is vague, but it sounds like you don't actually have a classifier, since your program is searching for instances in a large space rather than deciding whether each of several discrete objects has a binary property. So no false-positive rate in the traditional sense is possible. It is still useful to report all three of "the number of items it was supposed to circle, the number of those it actually circled, and the number of false positives" and compare them.