ICF/Waldboost detector training not working OpenCV 3.1

asked 2016-02-05 07:17:23 -0500

k291 gravatar image

I am trying to use the WBDetector in OpenCV 3.1. I am trying to train it with positive and negative images from the INRIA dataset. Even when I try to train it with very few images, I am observing a crash. I am using the following function: wb_detector->train(pos_file_path,neg_file_path);

The crash seems to be due to some invalid matrix.

Any idea on what is going wrong?

Please help.

edit retag flag offensive close merge delete

Comments

First of all, ICF and Waldboost are NOT the same, the second one just got added to OpenCV instead of ICF. ICF was not actively developed and buggy so the original contributor removed everything from the contrib module. He then added Waldboost, which has been reported not to work for several times now ... guess if you need one of those algorithms, OpenCV is not the place to look for them!

StevenPuttemans gravatar imageStevenPuttemans ( 2016-02-05 07:21:52 -0500 )edit

Ok .. Thanks .. I also noticed that ICF has been removed from the contrib in OpenCV 3.1 .. But the change log mentions improved ICF detector .. So I thought Waldboost had been used instead of Adaboost for the implementation ..

k291 gravatar imagek291 ( 2016-02-08 23:45:33 -0500 )edit

The update history is far from accurate :D and it might even be that with the correct knowledge, that the waldboost detector is better, but the whole module lacks documentation and guidelines as it is ...

StevenPuttemans gravatar imageStevenPuttemans ( 2016-02-09 04:45:33 -0500 )edit