OpenCV AdaBoost - Obtaining Optimal Stage Threshold
Hey guys,
I came across Adaboost in OpenCV and would like to know more about it. I have a good intuition about the algorithm, thanks to the numerous tutorials in the web. But there is this one thing I am not able to catch and would really be grateful if someone could help.
The parameters,
dmin -> minimum acceptable detection per cascade stage
threshold -> Parameter tweaked to make sure that dmin is honored.
fi -> False positive rate per cascade stage evaluated at the current threshold
fmax -> Maximum acceptable false positive rate per cascade stage
are seriously creeping me out.
It would be really helpful if someone can point me towards some resource so that i can understand how threshold is estimated such that dmin holds (so that i can update fi for each iteration and exit loop when it falls below fmax thereby having a strong classifier).
I am following Zhu and Avidan's paper (Refer page 3 - Algorithm pseudocode).
Please comment if I need to update my question with any more details.
Hoping to get help. Thanks for trying.
Regards,
Prasanna S
(Someone please do tell me how to use subscript and superscript while typing)
Everything else became clear now (or atleast it appears to be) for me except finding the optimal stage threshold. I have currently decided to go ahead with binary search like approach (hop front and back till current Detection Rate is just greater than dmin). If someone can suggest any global optimization technique, it would be very helpful. Once again,Thanks for trying.