Ask Your Question
0

Required leaf false alarm rate achieved. Branch training terminated.

asked 2015-11-20 04:49:16 -0600

Bhavik gravatar image

I am having 25 positive images and 50 negative images,Running the traincascade.exe to generate the cascade classifier for 20 stages.But every time it is interrupting at 17th stage saying "Required leaf false alarm rate achieved. Branch training terminated."Even when I tried to run it for 10 stages it is getting interrupted at 5th step giving the same result.Why is it so???What is to be done to solve the problem???Please give solution. Although cascades.xml is generated by running the command opencv_traincascade -data cascade -vec text_classifier\positive.vec -bg negative_images\negative.dat -numPos 20 -numNeg 15 -numStag 1 -featureType HAAR -minHitRate 0.999 -maxFalseAlarmRate 0.5 -w 24 -h 30. Still many xmls are generated each for every successful stages,along with the param.xml and cascade.xml Yet I could not make which xml is to be used as an output ????

edit retag flag offensive close merge delete

Comments

2

Please, stop opening multiple questions about same issue. First of all, read the documentation for both the Cascade Classifiers class and the training procedure. Then, if you still have doubts, use the search bar of the forum before asking. There are a lot of posts regarding this topic.

LorenaGdL gravatar imageLorenaGdL ( 2015-11-20 05:13:11 -0600 )edit

Yes I have seen,but couldn't find any proper answers for that questions.Thanks though.

Bhavik gravatar imageBhavik ( 2015-11-20 05:16:43 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2015-11-20 04:59:46 -0600

This means that given the set parameters, the required accuracy has been reached. So either put your settings harsher, or add more training data. It is a combination of less data and fairly default parameter settings. The cascade.xml file is the final model, which is a combination of all the previous stage files.

edit flag offensive delete link more

Comments

2

ok,Thank u very much.I will try likewise.

Bhavik gravatar imageBhavik ( 2015-11-20 05:03:21 -0600 )edit

The trained cascade.xml is giving wrong result for the data which is being used for training the xml. why?

Bhavik gravatar imageBhavik ( 2015-11-21 07:13:35 -0600 )edit
1

What do you understand under wrong results. A classifier will never cover its training data 100% because it would then overfit. Getting 100% detection on test data is impossible with this algorithm.

StevenPuttemans gravatar imageStevenPuttemans ( 2015-11-21 07:44:22 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2015-11-20 04:49:16 -0600

Seen: 1,697 times

Last updated: Nov 20 '15