Ask Your Question

after make -w 30 -h 30 opencv_createsamples run very very slowly

asked 2013-11-18 06:58:45 -0600

wy200507030 gravatar image

updated 2013-11-20 03:58:35 -0600

hi, i use opencv cascade train like this:

./opencv_createsamples -vec righttoppointstraindata.vec -w 30 -h 30 -info info.dat

then use

./home/opencv2461/bin/opencv_traincascade -data model -vec righttoppointstraindata.vec -bg bg.txt -numPos 100 -numNeg 400 -numStages 13 -w 30 -h 30

it runs but when after TRAINING 5-stage it runs very very slowly. and it run at the 6-stage for more than 2 hours,but has not finished the 6-stage, why? thanks.

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2013-11-20 04:02:47 -0600

Not to disappoint you but this is perfectly normal. You have not specified a feature type, so it uses the default HAAR wavelets, which take time to process. Models have been reported to train for multiple days before even reaching a stable state. LBP features go a bit faster, but can still take a long time. It depends on your data, how easy it is to retrieve actual negatives that are descriminative enough.

So basically wait and see how your result looks like.

Larger models, using multiple thousands of training samples have even reported to be running over a week before completion. For example, I have an LBP model training for the moment on a basic quadcore laptop and stage 5 has been training for 15 hours now but it is still continuing. Just need to be patient.

edit flag offensive delete link more



Thank you, your say is correct, and i cost 2 days to get its result, and the result is acceptadble.

wy200507030 gravatar imagewy200507030 ( 2013-11-22 00:29:55 -0600 )edit

Nice to see someone has worked out a promising model :)

StevenPuttemans gravatar imageStevenPuttemans ( 2013-11-22 01:49:12 -0600 )edit

Question Tools


Asked: 2013-11-18 06:58:45 -0600

Seen: 860 times

Last updated: Nov 20 '13