I'm trying to train a haar cascade of positive images 1749 and negative images 3867. My positive vector file contains 885 samples. I'm trying to detect palm. Is this the correct ratio of number positive and negative images? And what should be the value of the parameters of my haar cascade?