Revision history [back]

in the case of a LINEAR svm, the support vectors get compressed to a single one, thus also alpha and svidx contain only a single element.

to use this single sv with the HOGDescriptor later, you will need only the sv and rho, not alpha or svindex, you'll have to insert -rho as last elem into the sv, before passing the resulting float array to HOGDescriptor.setSVMDetector().

you'll probably also have to change C_SVC to C_SVR or C_SVR_EPS in your SVM params (should it be optimized for distance to the sv not for classification)

in the case of a LINEAR svm, the support vectors get compressed to a single one, thus also alpha and svidx contain only a single element.

to use this single sv with the HOGDescriptor later, you will need only the sv and rho, not alpha or svindex, you'll have to insert -rho as last elem into the sv, before passing the resulting float array to HOGDescriptor.setSVMDetector().

you'll probably also have to change C_SVC to C_SVR or C_SVR_EPS in your SVM params (should it be optimized for distance to the sv not for classification)

in the case of a LINEAR svm, the support vectors get compressed to a single one, thus also alpha and svidx contain only a single element.

to use this single sv with the HOGDescriptor later, you will need only the sv and rho, not alpha or svindex, you'll have to insert append -rho as last elem into to the sv, before passing the resulting float array to HOGDescriptor.setSVMDetector().HOGDescriptor.setSVMDetector() (sounds weird,but it's just a matter of transporting itto the actual algorithm).

you'll probably also have to change C_SVC to C_SVR or C_SVR_EPS in your SVM params (should it be optimized for distance to the sv not for classification) classification)

in the case of a LINEAR svm, the support vectors get compressed to a single one, thus also alpha and svidx contain only a single element.

to use this single sv with the HOGDescriptor later, you will need only the sv and rho, not alpha or svindex, you'll have to append -rho as last elem to the sv, before passing the resulting float array to HOGDescriptor.setSVMDetector() (sounds weird,but it's just a matter of transporting itto it to the actual algorithm).

you'll probably also have to change C_SVC to C_SVR or C_SVR_EPS in your SVM params (should it be optimized for distance to the sv not for classification)

in the case of a LINEAR svm, the support vectors get compressed to a single one, thus also alpha and svidx contain only a single element.

to use this single sv with the HOGDescriptor later, you will need only the sv and rho, not alpha or svindex, you'll have to append -rho as last elem to the sv, before passing the resulting float array to HOGDescriptor.setSVMDetector() (sounds weird,but it's just a matter of transporting it to the actual algorithm).

you'll probably also have to change C_SVC to C_SVR or C_SVR_EPS in your SVM params (should it be optimized for distance to the sv not for classification)