Possible bug when numP > totalSample

[shaibagon]

n function (Objecness.cpp:624)

Mat Objectness::trainSVM(const vector<Mat> &pX1f, const vector<Mat> &nX1f, int sT, double C, double bias, double eps, int maxTrainNum)

If too many positive samples are supplied, numP is not adjusted according to the size of X1f and a seg fault occurs.