Which dataset is used for training the DefaultPeopleDetector (HOGDescriptor)

asked 2016-10-10 02:28:51 -0600

Hi all When I try to create my own PeopleDetector using the INRIA dataset, I don't get similar performance as the DefaultPeopleDetector. So my question is, which dataset has been used for training the DefaultPeopleDetector?

Thank you!

edit retag flag offensive close merge delete

Comments

did you use the pre-cropped ones or the annotations (which contain some errors) ?

did you use both train & test sets ? (more data)

train_HOG.cpp ? (svm params might vary)

berak ( 2016-10-10 04:34:37 -0600 )edit

add a comment

1 answer

Sort by » oldest newest most voted

answered 2016-10-10 04:21:08 -0600

StevenPuttemans

20029 ●16 ●82 ●207 http://stevenputtemans...

I am unsure for the getDefaultPeopleDetector, which is an imbedded hard coded model. As far I know it was trained indeed on INRIA since the implementation follows the official Dallal and Triggs research paper. The second hardcoded model however is trained on the Daimler Pedestrian detection dataset.