I want to recognize traffic sign using sift feature (bag of words) method. And my classifier is rbfsvm. But my result is higher, which class has lot of instances?. For example my train 7. class has 150 instances and my test 7. class recognation rate %95,my train 2.class has only ten instances and my test 2. class recognation rate %2.
chrome-extension://oemmndcbldboiebfnladdacbdfmadadm/http://www.inf.fu-berlin.de/lehre/SS09/CV/uebungen/uebung09/SIFT.pdf
Hi. Look the link, may be useful.
ı looked this link but ı don't find my problem's answer. My problem is this:My classification result is very high when the instances is much. For example my train 7. class has 150 instances and my test 7. class recognation rate %95,my train 2.class has only ten instances and my test 2. class recognation rate %2.
You have a problem of what is called unbalanced data. It is common that the classifier favour larger set. Probably, you should make your data balanced first.
thank you guillaume but this problem is only sift feature. Using same data histogram of oriented gradient feature recognation rate %95 all over the dataset. Why only sift feature rsult is low?
- Badges
- Science topic
- Similar topics
- Counseling Psychology
- Confusion