1 | initial version |
if you trained your SVM for multi-class classification, you simply cannot use that for the HOGDescriptor, wich requires binary regression (and a single support vector).
so, keep your old SVM model around for later, but make another SVM training run using train_HOG.cpp and the SVR method(regression). you will also need a lot of negative (non-digit) images, all cropped to your desired winSize (20x20).
after that finished successfully, you can extract the single support vector for your HOGDescriptor
good luck.
ps: please never use spanish language (it might as well be chinesse...) for comments or variable names, you only make it terrible hard for anyone trying to help you.
2 | No.2 Revision |
if you trained your SVM for multi-class classification, you simply cannot use that for the HOGDescriptor, wich requires binary regression (and a single support vector).
so, keep your old SVM model around for later, but make another SVM training run using train_HOG.cpp and the SVR method(regression). you will also need a lot of negative (non-digit) images, all cropped to your desired winSize (20x20).
after that finished successfully, you can extract the single support vector for your HOGDescriptorHOGDescriptor, and try to detect digits with it.
good luck.
ps: please never use spanish language (it might as well be chinesse...) chinese...) for comments or variable names, you only make it terrible hard for anyone trying to help you.