Attention! This forum will be made read-only by Dec-20. Please migrate to https://forum.opencv.org. Most of existing active users should've received invitation by e-mail.
Ask Your Question
0

Train cascade for small object

asked 2013-10-04 12:30:04 -0500

Yarogleck gravatar image

updated 2013-10-05 02:51:06 -0500

berak gravatar image

I need to train Cascade for small objects (20x20) on non resizable 2d simple negative (for 2d game) but have no luck. I tried opencv_haartraining.exe but starting with 15 stage it takes an awful lot of time. For 20x20 non resizable object this is too much. Also i tried opencv_traincascade.exe with same luck. SURF method also will not work for this small objects.

Here is negative and positive samples:

image description image description

I hope you can help me with this simple task. I believe there is some simple params for opencv_haartraining and working cascade will be trained. This is not a complex human face after all.

edit retag flag offensive close merge delete

Comments

"This is not a complex human face after all" - yea, that's your problem here . take away the color, and all you got left is a slurred blob

berak gravatar imageberak ( 2013-10-05 02:55:12 -0500 )edit

The height and width of your objects are same, maybe you could use something like surf(or try orther rotation invariant detector and descriptor) to extract and detect the features, than detect the roi(region of interest) one by one.

stereomatching gravatar imagestereomatching ( 2013-10-05 11:17:14 -0500 )edit

Nothing works, not the slurred blob or surf(surf can't detect so small object) or roi. There is also objects with different shape, even animated.

Yarogleck gravatar imageYarogleck ( 2013-10-06 04:44:11 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
0

answered 2013-10-17 21:53:26 -0500

Yarogleck gravatar image

Up please...

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

Stats

Asked: 2013-10-04 12:30:04 -0500

Seen: 258 times

Last updated: Oct 17 '13