Is it possible to use evaluateDe​tectionPre​cision to calculate precision and recall at a specific confidence threshold

Hi,

So I wanted to use evaluateDetectionPrecision function (here) to evaluate a detector. However, I wanted to obtain precision and recall at a specific confidence threshold, let's say 0.25. How can i do this?

The problem I have is that the precision and recall values that evaluateDetectionPrecision() outputs are the vectors to build the plot of the precision-recall curve. Do I just remove the boxes with confidence < 0.25 to zero before using the function evaluateDetectionPrecision(). And anotate the value of the last element of the recall and precision vector:

[ap, recall, precision] = evaluateDetectionPrecision(results, blds);

my_recall = recall(end)
my_precision = precision(end)

Thanks in advance.

Answer 1

[Anshika Chaurasia]

Hi,

To calculate precision and recall at a specific threshold, you can set 'Threshold' as 0.25 (let's say) in detect function of the detector:

results = detect(detector,I,'Threshold',0.25)
[ap, recall, precision] = evaluateDetectionPrecision(results, blds);

The above threshold is detection threshold and hence, detections that have scores less than this threshold value will be removed.

[Pedro José Carrinho Ribeiro]

Hi,

So the precision and recall values at that threshold are recall(end) and precison(end)?