Get a precision-recall graph for a recognition system
Hi all, i recently developed a sift-based recognition system and i'd like to get some measure of performance for it. I read about a "threshold" parameter which should be chosen to get precision-recall curves. I can't understand what this means. What exactly should i increase/decresce to plot the curve?
Thank you.
The threshold means that you want to evaluate the certainty of an output result. For example, a pedestrian detector could give you a certainty score in the range of [-1,+1]. It would then be good to run the detections, and based on the returned score filter from -1 to +1 in steps of 0.05. Each threshold will yield other TP TN FP and FN results and thus another precision recall value. This can then be plotted!