What is the meaning of negative levelweight values in detectmultiscale3?

Asked Feb 14 '22 at 03:06

Active Feb 16 '22 at 22:01

Viewed 41 times

I trained a cascade classifier for ball detection using opencv4. I want to obtain the confidence of the classifier for detected balls. I understood from tutorial if I use detectmultiscale3, I have access to levelweight which is the confidence of the model. So, when I want to detect balls in images using detectmultiscale3, I noticed that some of these extracted rectangulars have negative levelweight! What does it mean? I can interpret the positive values for confidence, but I can not find any explanation for negative values.

 detected_objects, rejectLevels, levelWeights  = cascade_classifier.detectMultiScale3(original_image, scaleFactor=1.03, minNeighbors=20, minSize=(25,25), maxSize=(60, 60)

edited Feb 16 '22 at 22:01

asked Feb 14 '22 at 03:06

Elnaz Yousefi

Are the weights really negative or numerically close to zero? e.g. what happens if you call `np.isclose(levelWeights, 0)`? – hilberts_drinking_problem Feb 14 '22 at 04:21
@hilberts_drinking_problem, the minimum value is around -1.5 while the maximum value is around +2.5. I should mention that there are only a few samples with negative values I mean for example 40 out of 500 rectangles. – Elnaz Yousefi Feb 14 '22 at 04:29
Thanks for clarifying. I am not sure what is going on. – hilberts_drinking_problem Feb 14 '22 at 04:30

What is the meaning of negative levelweight values in detectmultiscale3?

0 Answers0