c++ - What does cv::TermCriteria() exactly do in OpenCV? -

official documentation said termcriteria(int type, int maxcount, double epsilon) defining termination criteria iterative algorithms. , criteria type can 1 of: count, eps or count + eps.

however, can't quite understand svm different in each iteration when use svm->settermcriteria(const cv::termcriteria & val).

svm training can considered large-scale quadratic programming problem , unfortunately cannot solved standartd qp techniques. that's why in past several years numerous decomposition algorithms introduced in literature.

the basic idea of these algorithms decompose large qp problem series of smaller qp sub-problems. is, in each iteration algorithm keeps fixed dimension of optimization variable vector , varies small subset of dimensions (namely working set) maximumal reduction of object function.

as far know, opencv uses svmlight or generalized smo algorithm , termcriteria parameter termination criteria of iterative svm training procedure solves above mentioned partial case of constrained qp problem. can specify tolerance and/or maximum number of iterations.