Speed up haartraining utility

[Aaron Brown]

{Cross-posted from Android-OpenCV group}

Hey folks,

So I've got my first classifier training at the moment, but it's
taking a very long time. It's been running for 72 hours and it
appears to be 4 stages in to a 20 stage classifier. At this rate
it'll take another two weeks to finish.

Does anyone have any tips to speed up the process? I'm currently
working on building out a demo that we're hoping to have done by
Sept.
12th. The demo's going to require 5 classifiers. I might be able to
do that, since it appears haartraining only uses one core of my
machine, so I suppose I could batch the other four and get them done
by then. The final product, though, may need dozens of classifiers,
so I don't think it's really workable if training takes a fortnight
or
more.

So, any tips for speeding up classifier training? I'm currently
training 20 stages, based on 7,000 samples and 3,000 negative images.
nsplits is 2. I've got nonsym on. Feature size is 24x24. Mode is
set to ALL. Memory is allocated at 1024. Min hit rate is 0.999,
false alarm rate 0.5.

Anyone know any way to get haartraining working multi-threaded /
across multiple cores?

Thanks in advance for any help.

-AMB