histogram - SVM for HOG features on Matlab -


i doing svm classification problem on matlab. features hog features (length = 4356).

my procedure follows.

1.extract 200 positive windows , 200 negative windows. 2.extract hog features of above samples. 2. scale features , remove nan features. (this gives 2904 features) 3. grid search , 3-fold cross validation find c , g values 4. train whole training set using best c , g 5. extract test data set hog features , scale same parameters training set. 5. test test data set

i have large test data set (3000 samples) test , know inside test samples should have few (~10) positive samples.

however, above process gives me 0% prediction accuracy. in fact, none of positive samples detected. detected negatives.

where doing wrong? having small no of training samples (400) , large no of testing samples (3000) problem?


Comments

Popular posts from this blog

php - Submit Form Data without Reloading page -

linux - Rails running on virtual machine in Windows -

php - $params->set Array between square bracket -