Ask Your Question

How was the hogdescriptor::getDefaultPeopleDetector() trained

asked 2013-07-09 09:38:05 -0600

miat123 gravatar image

updated 2013-07-19 06:22:15 -0600

I am curious to know how the float detector[] in DefaultPeopleDetector() was trained. With this I mean what database,parameters,SVM were used. If somebody can identify the exact steps that would be very helpful.

Thank you.

edit retag flag offensive close merge delete


yea, and why has the getDaimlerPeopleDetector() a different size, and thus cannot be used ?

berak gravatar imageberak ( 2013-07-09 09:48:43 -0600 )edit

Hi berak, i guess the size of the DaimlerPeopleDetector() is 48x96 since the training images from the Daimler Benchmark dataset has this size. The dataset and the paper where it is described can be found here:

Siegfried gravatar imageSiegfried ( 2013-07-09 10:21:16 -0600 )edit

thanks, siegfried, sorry, btw, for hijacking the question ;)

berak gravatar imageberak ( 2013-07-09 10:23:02 -0600 )edit

to use DaimlerPeopleDetector just set winsize to 48X96

Spas Hristov gravatar imageSpas Hristov ( 2013-07-12 01:58:26 -0600 )edit

thanks, @Spas.

berak gravatar imageberak ( 2013-07-12 02:15:00 -0600 )edit

Has anybody figured it out ?

nexocv gravatar imagenexocv ( 2015-11-19 11:15:30 -0600 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2013-07-12 05:30:58 -0600

I'm trying same thing, but without success... I'm using SVM lite, but when I try to detect someone I'm getting just 1 positive in the center of the image...

edit flag offensive delete link more


I had the same thing. In my case, the reason was due to the fact that the computed weights were all equal to 0. I suggest you to check them out...

nexocv gravatar imagenexocv ( 2015-11-25 11:57:55 -0600 )edit

Question Tools

1 follower


Asked: 2013-07-09 09:38:05 -0600

Seen: 2,942 times

Last updated: Jul 19 '13