Ask Your Question

Revision history [back]

How to evaluate a cascade classifier?

I used the code of OpenCVBlueprints to plot the pr curve : precision_recall

I generate the annotation and detection file which look like this :

annotations.txt :

/home/...../test/i (100).jpg 1 38 24 87 73

/home/...../test/i (101).jpg 1 35 17 94 74

/home/...../test/i (102).jpg 1 30 19 108 75

/home/...../test/i (103).jpg 1 40 25 85 69

detection.txt:

/home/...../test/i (100).jpg 1 27 21 110 88

/home/...../test/i (101).jpg 1 25 15 113 90

/home/...../test/i (102).jpg 1 26 15 119 95

/home/...../test/i (103).jpg 1 28 23 110 88

/home/...../test/i (104).jpg 0

I execute the precision_recall using :

./precision_recall -groundtruth /home/...../annotations.txt -detections /home/...../detection.txt -output /home/...../pr.txt -minthresh -1 -maxthresh 1 -stepthresh 1

but I obtained that :

Annotated images: 161 Images that triggered detections: 1

annotations: 1 #detections: 1

Segmentation fault (core dumped)