Traincascade maxcatcount

asked 2015-04-13 16:26:52 -0500

cyrus gravatar image


I was looking through traincascade_features.h and could not figure out for what the variable maxCatCount in CvFeatureParams is used. It appears in boost.cpp under CascadeBoostTrainData but I didn't understand exactly what it did.

image description

In lbpfeatures.cpp we have maxCatCount = 256; is this related to the 8 bit patterns?

Any help would be appreciated, thanks in advance!

edit retag flag offensive close merge delete


As you can read here by the original author of the implementation, it is a parameter used to size the responses of the features to the correct interval and to create weak classifiers that descriminate enough.

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-14 04:16:21 -0500 )edit

Thanks! That helps a lot. I'll futz with that value if my features produce learners that are off.

cyrus gravatar imagecyrus ( 2015-04-14 08:54:28 -0500 )edit

As far as I understand it is because internally it uses the featureEvaluator framework. That framework works well with correct values between 0-1 for other interfaces, but the response for the boosting needs to be between 0-255. So I guess that is the reason why this factor is added.

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-15 07:24:15 -0500 )edit