analytic choice of parameters in traincascade
Hey there,
I want to know, how can i find an analytic and scientific way to find the best parameters when training a cascade in opencv_traincascade? (let's just talk about: Stages, minHitRate and maxFalseAlarmRate)
I mean, I don't want to experiment by any gut instincts, instead it should be a reasonable line that my experiments are following with the result that this parameters are kind of optimal (I know i can't solve the error rates completly and the right parameters depend from case to case).
Would be also good if you can recommend me some good literature for that topic.
Thanks for your time guys!
Kind regards, cv_enthusiast
You can read the original paper that introduces the method: Robust Real-Time Face Detection. You should find the relation between number of stages, hit rate and stage max false alarm rate.
Another good article in my opinion: An Analysis of the Viola-Jones Face Detection Algorithm.
Hey thanks! I like these papers. But about the choice of the right parameters they write following: "Unfortunately finding this optimum is a tremendously difficult problem."
Can anybody explain why?