Hi all,
I have to make an activity classification based on simple features I extract from images.
I am using support vector machines from the opencv machine learning library for this. So far I'm getting very decent results with RBF and Polynomial kernels, but the accuracy and precision for the sigmoid kernel don't seem to get higher than 0.12 .
In literature I found that the sigmoid kernel is only a pd (positive definite) kernel when gamma >0 && coef0<0, however, i can't try this because opencv requires all kernel parameters to be positive. I have checked the source for the SVM diagonally, but all i can see is this function where both gamma and coef0 get inverted.
calc_non_rbf_base( vcount, var_count, vecs, another, results,
-2*params.gamma, -2*params.coef0 );
Has anyone ever had succes with the sigmoid kernel in opencv ?