Ask Your Question

How was haarcascade_frontalface_alt.xml trained

asked 2012-07-20 04:09:26 -0500

Jim Leo gravatar image

updated 2012-07-20 10:01:48 -0500

Kirill Kornyakov gravatar image

Hello, I've been doing a lot of research and i can't seem to find how haarcascadefrontalfacealt.xml was made. My questions are: how many positives and negative samples(pictures) where used, what parameters where used during the training, and how come all the values in the xml are all positive? if someone can help me understand more how to create a good xml would really help me, thank you :)

edit retag flag offensive close merge delete


@Andrey Pavlenko

I've read the guide you linked to me but I still dont know what parameters should I input. I've been following the procedures in this site and none of my xml are working. Plus i really dont know how come the haarcascadefrontalfacealt.xml's threshold and node values are all positive and the stage threshold gets bigger un like mine which values ranges from -2 to +2 only

Jim Leo gravatar imageJim Leo ( 2012-07-20 06:19:13 -0500 )edit

3 answers

Sort by » oldest newest most voted

answered 2012-07-20 10:01:10 -0500

Kirill Kornyakov gravatar image

AFAIK haarcascade_frontalface_alt2.xml is even better than haarcascade_frontalface_alt.xml. Unfortunately I don't know how both cascades were trained, because they were submitted by contributors, and I don't know who is the author.

At the same time I'm not sure that exact parameters should differ significantly from the default ones. Most likelt that your problem is in your training dataset. Did you try to use vec file from the OpenCV: opencvtrunk/opencv/data/vecfiles/trainingfaces_24-24.vec? Try to use it and if it works, you should create a better training set.

edit flag offensive delete link more

answered 2012-07-20 05:56:43 -0500

Andrey Pavlenko gravatar image
edit flag offensive delete link more

answered 2013-11-08 05:29:30 -0500

Adrián Nieto gravatar image

updated 2013-11-08 05:31:26 -0500

I was searching the same and found a paper by the author (Rainer Lienhart) where he compare all the relevant parameters and explains which are the best: Empirical Analysis of Detection Cascades of Boosted Classifiers for Rapid Object Detection

edit flag offensive delete link more
Login/Signup to Answer

Question Tools


Asked: 2012-07-20 04:09:26 -0500

Seen: 8,171 times

Last updated: Nov 08 '13