I was wondering, how many Sliding windows are used in the Cascade.Detect function? I mean, if I detect 3 Objects in my pirture, I get a true positive rate of 3. But how about my true negative rate? How many times did the algorithm declare a background properly as a background?
For a single scale, it should be (img.width-win.width) x (img.height-win.height).
For the other scales, I would loop to increase the minimal window size by a scale factor until reach the biggest window size enclosed in the image (not sure about the dimension of the biggest detection window).
You will have to substract your true positive.
A bit overkill: you could also substract the (unknown (if you don't tweak the code I think you cannot get this information) numbers of) neighbors of the true positive as there should be multiple detections near the final returned detection.
Asked: 2016-02-20 04:01:35 -0600
Seen: 542 times
Last updated: Feb 20 '16
