wenli-vision / svmplus_matlab Goto Github PK

Hi,

I'd clone the code and run it successfully, I'm trying to replicate the results published in your CVPR 2016 paper, I started by classic SVM with the following code:

n = length(train_labels);
opt = ['-s 1 -t 4 -n ', num2str(1/n)];
model = svmtrain(train_labels, K, opt);
[predicted_label, accuracy, decision_values] = svmpredict(test_labels, testK, model);

As results, the accuracy returned: Accuracy = 52.1972% (974/1866) (classification)
And predicted_label was biased towards only one label: -1

any suggestions or modification on my code ?