Ask Your Question

Cascade Classifier Training: Is the negNum decreasing each stage?

asked 2016-12-17 19:00:47 -0500

DerrickB gravatar image

Hello guy,

Does the second stage train with the exact same training samples as the first stage or just with 0.5x of the negative samples (maxFalseAlarmRate = 0.5). The 0.5 neg sampels which were classified false?

If yes: Why does the second stage needs more features to separate the data?

If no: Is the amount of negative samples decreasing each stage by 0.5? After 20 stages there would be nearly no samples?

I hope the question is understandable,

Thanks a lot for your help!!!

Kind regards,


edit retag flag offensive close merge delete


DerrickB gravatar imageDerrickB ( 2016-12-18 02:57:36 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2016-12-20 04:30:01 -0500

updated 2016-12-20 04:37:49 -0500

Ow no, there is a difference between -numNeg and the actual number of negative images. The first one is about windows, the second one is the images where these windows are sampled from. That being said, stage 0 takes for example 1000 windows as negative due to -numNeg 1000. In the second stage 1, the classifier will have to find again 1000 negatives, but the ones that are not yet classified as negatives by stage 0. In general that means that at least half of the previous samples can be reused (due too maxFalseAlarmRate) but that we then need to continue searching for new samples. That is why getting new samples takes longr and longer in following subsequential stages.

edit flag offensive delete link more

Question Tools

1 follower


Asked: 2016-12-17 19:00:47 -0500

Seen: 100 times

Last updated: Dec 20 '16