My Haarcascade classifier uses a vector file of -w=20, -h=20 for training. There are some false negative results when testing. Does increasing -w and -h improve performance of my Haarcascade classifier?
My Haarcascade classifier uses a vector file of -w=20, -h=20 for training. There are some false negative results when testing. Does increasing -w and -h improve performance of my Haarcascade classifier? opencv_traincascade -data data -vec positives.vec -bg bg.txt -numPos 3519 -numNeg 1800 -numStages 10 -w 20 -h 20 -minHitRate 0.999 -maxFalseAlarmRate 0.4. This is my command for training. I'm trying to train a hand detector. Please help me!
Ow please, start by reading up on the topic. The Q&A is filled with information about it. In short, no it will not increase performance... because that is not what the parameters intend to do. Also without info on how your model is made and its parameters, there is no telling how you can improve it. Please update your question accordingly.
opencv_traincascade -data data -vec positives.vec -bg bg.txt -numPos 3519 -numNeg 1800 -numStages 10 -w 20 -h 20 -minHitRate 0.999 -maxFalseAlarmRate 0.4. This is my command for training. I'm trying to train a hand detector. Please help me!
You have way to less negatives compared to positives. Your feature space is probably poorly seperated. Increase the
-numNegparameter first.