vision.Cas​cadeObject​Detector - Negative Images

Dear all,

I have a question concerning the training of a vision.CascadeObjectDetector (Haar). If we have a large amount of negative images (e.g. 200,000) in a folder and the training uses at most e.g. 10,000 negative images per stage. How will these 10,000 images per stage be chosen? Randomly? If yes, would this mean that not every single negative images would be used until the end of the training (e.g. after 10 stages)? And would this also mean that multiple detectors based on the same dataset are somehow different?

Thank you in advance!