Ask Your Question

About parameters of traincascade

asked 2016-04-04 10:08:05 -0500


Before, excuse me for english... I am having problems with the tool opencv_traincascade... the train stop in any stage and not occours visible error. I am using 100 negative images and 50 positive images by 40 x 13, brazilian licence plates. My asks about the parameters:

The images should be square? Can I use grayscale in images? The size in w and h parameters can be small than size of the images? Are there recommendations about the images' properties and values?

Thanks for the attention

edit retag flag offensive close merge delete



1 answer

Sort by ยป oldest newest most voted

answered 2016-04-05 07:37:06 -0500

updated 2016-04-05 07:39:53 -0500

If there is no visible error, then it did not stop training, rather it is still busy training that level! Take into account that a HAAR model at later stages can train for multiple hours, even days, due to the sequential processing ... BTW, this topic had same problem in license plate recognition, so you might want to check out my answers there.

As to the questions:

  • Images/samples should NOT be square. You define the size in the -w -h parameters when creating the vec file.
  • Yes the size can be smaller than the images, then it will simply rescale
  • Recommendations for parameters: this forum OR OpenCV 3 Blueprints which has a complete chapter on optimal parameters.
edit flag offensive delete link more

Question Tools

1 follower


Asked: 2016-04-04 10:08:05 -0500

Seen: 176 times

Last updated: Apr 05 '16