How to implement detectMultiscale function for hog feature detection with neural network classifier?

asked 2016-12-01 05:05:35 -0500

esha gravatar image

I m using neural network implementation of opencv as classifier for training. There is provision of using detectMultiscale of hog with SVM. Instead of SVM I want to use neural network model. Any suggestions on how to achieve this?

edit retag flag offensive close merge delete



i don't think, this is a good idea.

inside detectMultiScale(), a sliding window is applied x1000 times on multiple scales. whith an SVM, the test boils down to a single dot-product (between the window and the support vector) per test, while with an ANN, you have to traverse your whole network, so that will be terribly slow.

berak gravatar imageberak ( 2016-12-01 05:13:31 -0500 )edit

@berak: Yeah it would be slow, but the accuracy i want is not achieved through SVM. With neural network, I m getting good accuracy. So any other method through which i can detect object throughout the frame, some method similar as detectMultiscale()?

esha gravatar imageesha ( 2016-12-01 06:07:49 -0500 )edit