NumNeg Vs NumPos

asked 2018-08-28 15:59:36 -0500

For a week that i'm working with the traincascade, But still not clear to me what is the best way to do the training. In some places I saw that numPos need to be 0.9*positive_samples or numNegneed to be twice bigger then the positive. Not sure about all the formulas I saw around. Last Q I saw is this one: http://answers.opencv.org/question/62...

The answer here isn't clear cause they told about who is bigger, but not in numbers. Like "Negative need to be twice bigger then pos" Or Pos is 0.9*Samples etc..

How minHitRate and maxFalseAlarmRate have influence about the num parameters ?

I'm runing the training on Ubuntu Server 18.04

edit retag flag offensive close merge delete

Comments

I suggest you to read OpenCV 3 Blueprints, chapter 5, which discusses this whole interface and every single parameter in quite the detail. Or you can start from the official tutorial

StevenPuttemans gravatar imageStevenPuttemans ( 2018-08-30 08:08:45 -0500 )edit