samuel_1208's profile - activity

Recently I try to train a face detector by train_cascade program. I use 10000 Positives, FP=0.5, TP=0.995 RAB and wekerdepth=1. I found the number of weak classifiers in the first cascade stage is around 100(I just give some very simple negatives image like blackboard) . But there are only 3 weak classifiers of the first stage in haarcascade_frontalface_alt.xml

Recently I was learning the opencv_traincascade with gentle boost. As I know, gentle boost use squared error to split the decision tree. But according the code in opencv， the squared error is as following:
E = (P+N) - (P-N)*(P-N)/(P+N).
P: The sum of positive sample's weight
N: The sum of negative sample's weight
How does opencv get this formula as showing before

I have read the opencv's face recognition algorithm. But I want an implement based on Embedded HMM

Where can i get the source code of this http://isa.umh.es/pfc/rmvision/opencvdocs/appPage/FaceRecognition/FaceRecognition.htm. Or can someone share it

I have some questions about the "apps/sft/" implement.

1. During training , the opencv doesn't add K bootstraped negative samples which was mentioned in the paper(Robust Object Detection Via Soft Cascade). Right?
2. After training, the opencv doesn't use a test set to get a new output function order. Right?, It only get a new rejection threshold based on the training samples.Right?

How do the values enum { UNDEFINED=0, VALUE_EXPECTED=1, NAME_EXPECTED=2, INSIDE_MAP=4 }
affect the FileStorage.

In opencv/apps/sft/config.cpp .

void sft::Config::write(cv::FileStorage& fs) const
{

    fs << "{"
......
}

I find I can't write a non_alpha character into the variable fs. Is that an error

@mhm: Hi, if you get the answer, Could you share it to me

Hi, Did you train the softcascade by yourself. How to write the softcascade training config file. could you give me some advices