Ask Your Question

Meaning of stages in traincascade

asked 2017-04-07 13:49:28 -0500

DavidBankom1 gravatar image

updated 2017-04-07 13:59:14 -0500

I'm wondering what the special meaning of stages is. As far as I understood every stage might consist of arbitrary number of classifiers depending on specified parameters of launch like minHitRate. Thus the exact depth of the tree can't be specified directly by the user and inferred during the training. Why didn't developers just call weak classifiers as 'stages' ? I know that stages are applied sequentially weeding out negatives but single weak classifiers could be applied in a same fashion.

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2017-04-07 15:53:33 -0500

Stages are a collection of weak classifiers. By default OpenCV uses stumps as weak classifiers, basically a binary decision tree based on a given feature. However a single feature will never work, so multiple features/stumps/classifiers are combined until the stage does better than random guessing on the negatives, while maintaining a high hit ratio on the positives.

But the parameters allow you to change this all.

More info on this can be found in the OpenCV Blueprints book chapter 5!

Good luck!

edit flag offensive delete link more


@StevenPuttemans thanks for your comment. However it's not quite an answer to my question. What's the way of combining stumps inside one stage? Are they applied sequentially? If yes then I don't see any meaning in staging the process, because these weak classifiers themselves are working as little stages.

DavidBankom1 gravatar imageDavidBankom1 ( 2017-04-07 19:04:04 -0500 )edit

Question Tools

1 follower


Asked: 2017-04-07 13:49:28 -0500

Seen: 110 times

Last updated: Apr 07 '17