Some questions about training my own cascades. [closed]

edit

Hi everybody. Today I want to train a cascade for face detection and I met some problems.

I use the command "..\opencv_traincascade.exe -data data -vec pos.vec -bg negSamples\neg.dat -numPos 9 -numNeg 15 -w 200 -h 200 -minHitRate 0.85 -maxFalseAlarmRate 0.5 -mode ALL" to start training. But the program will collapse with no warning but shutdown itself. Then I adjusted two parameters: -w and -h to both 50. By this time the program can run. So I am very confused that the size of images in positive samples must be strictly limited? And can somebody tell me the relationship between positive and negative samples? Can I use those images where the face image is extracted from as negative samples?

Thanks.