Hello, uricamic. I am a researcher who is currently studying Landmark detection in the company. I have impressively read your paper related to Structured SVM.
Then now (not 2012, about that time), I think that landmark positions would be shaking according to new input frames compared to the state-of-the-art landmark detection methods. (I mean, though the face position doesn't change, the end points of eyes or mouth are shaken.) I saw the results from this link (http://cmp.felk.cvut.cz/~uricamic/flandmark/). Could you tell me about this reason in terms of the functional aspects of Structured SVM? I guess the reason because the structured SVM uses only 8 landmark pts.
Hello, uricamic. I am a researcher who is currently studying Landmark detection in the company. I have impressively read your paper related to Structured SVM.
Then now (not 2012, about that time), I think that landmark positions would be shaking according to new input frames compared to the state-of-the-art landmark detection methods. (I mean, though the face position doesn't change, the end points of eyes or mouth are shaken.) I saw the results from this link (http://cmp.felk.cvut.cz/~uricamic/flandmark/). Could you tell me about this reason in terms of the functional aspects of Structured SVM? I guess the reason because the structured SVM uses only 8 landmark pts.