sodeq's profile - activity

I've been trying to implementviola-jones face detection inside my project. I've read the paper too. However, i got things mixed up in my mind.

1. I know that AdaBoost uses weak classifier (which could be any learning algorithm) to produce the hypothesis. In viola-jones system, which is the weak classifier? Is it the haar-like features?
2. I have read the cascaded file (the xml files), i found that at each stage, a leaf has a single "feature index". What are they?

Is there any way to stream a cv::Mat Object through rtsp or http?? for now, i only experienced using libcurl for such a protocol. Thanks.

Hi all,

I have done opencv_traincascade and now still analyzing the output classifier, that is the output value. I have some questions: 1. What is maxWeakCount? I guess that is the number of weak classifier that AdaBoost predict it as "positive", but i doubt about that, because it seems like if this value increased, the prediction must be more accurate. 2. What is stage threshold? Is it related directly to the number of maxWeakCount? Or it is something else?

Thanks.

please mark as answered, Great! lol

as seen in your code, you use the var "imageName" as window name, not for the image source. Maybe you could try cvLoadImage(argv[1], 1). If the problem persist, i guess this out of my current understanding.

what does this line for?

char* imageName = argv[1];

then, for cvLoadImage() function, why don't you just pass the second argument of progrm calls? (the argv[1]?

Have you try hard debugging? Examining line by line?

Yes. It could be overtrained. For your trial and error, sir. Would you like to share the result by changing the parameters?

I have train my own classifier using 3 different stages , (10, 20, 30). All with 399 positive images, and about 180 negative images. I've tested each the classifier on to a video file, and analyze the ratio of positives and the actual object counts. I got a result that those three give different result, 10 stages gives the bad performance, 20 stages is better. But the 30 stages almost gives nothing, or, almost never detect the object as seen in the 20 stages. How does this could happened? As far as my understanding, the performance should be better if number of boosting stages were increased.

Thanks

guys, as far as i understand, those internal node values are used as arguments of decision tree build up when adaboost train the feature pool. Correct me if i am wrong.

Hi, i have a similar project with yours, but car detection in this case, i was wondering to have the source code for haar-feature extraction process visualization just like adam did, http://vimeo.com/12774628

can i have the code too? thanks.