CODEWITHSUNDEEP
c++arraysopencvvectorsvm
Jose Ramon
Jose Ramon

Reputation: 5444

Convert vectors of mat to Mat file in order to use opencv svm

I ve got two vectors and I want to convert them Mat files in order to use opencv's SVM train. The vectors that I have got is one a vector for train data and a vector for labels. How is it possible to convert them to the necessary type from svmtrain??/

for(int i=0; i<face.size();i++){
        img = imread("faceNface/face/"+face.at(i),1);
        cvtColor(img, img, CV_BGR2GRAY);
        img = img.reshape(1, 12100).t();
        trainData.push_back(img);
        lbs.push_back(1);
    }

    for(int i=0; i<non_face.size();i++){
        img = imread("faceNface/nonface/"+non_face.at(i), 1);
        cvtColor(img, img, CV_BGR2GRAY);
        img = img.reshape(1, 12100).t();
        trainData.push_back(img);
        lbs.push_back(-1);
    }

Upvotes: 1

Views: 1561

Answers (1)

guneykayim
guneykayim

Reputation: 5250

If your label vector type is std::vector<double> then you can use the following function. If it is not you need to do some changes. I believe you can figure it out.

cv::Mat Classifier::vectorToMat(std::vector<double> data)
{
    int size = data.size();
    cv::Mat mat(size, 1, CV_32F);
    for(int i = 0; i < size; ++i)
    {
        mat.at<float>(i, 0) = data[i];
    }
    return mat;
}

For the data itself, I have two recommendations for you;

1) Convert your 1x12100 sized image to std::vector<double>, then you will have std::vector<std::vector<double>> as trainData, then you can use following function.

cv::Mat Classifier::vectorToMat(std::vector<std::vector<double> > data)
{
    int height = data.size();
    std::vector<double> vectorData = data[0];
    int width = vectorData.size();

    cv::Mat mat(height, width, CV_32F);
    for(int i = 0; i < height; ++i)
    {
        for(int j = 0; j < width; ++j)
        {
            mat.at<float>(i,j) = data[i][j];
        }
    }
    return mat;
}

2) Modify the code that I posted to do the same operation with std::vector<cv::Mat>. Actually doing this is better because if you do this you don't need to convert cv::Mat to std::vector<double> in the first place, and do less operation.

Hope it helps.

Upvotes: 4

Related Questions