Error in fisher faces algorithm when using multiple images for a person

edit

hi, @namlool, good, you made it here.

again, the assertion is thrown because of an invalid (or empty) image region. please check all imread(), resize() calls, if there's facedetection involved, check, if the rects are all inside the image, and show the resp. code around those. there's for sure something wrong in your code, or the way, you handle your data.

"when I use one label with multiple images ... " -- that does not explain anything.

if you want us to reproduce your problem, we need your data, and code for a minimal test case.

@namlool. I used 2 images or more of same person on python 3.5.

This is the code I'm using. 1. few labels with one image each - works. 2. one or two labels with few images each - works 3. 3 labels (or more) with few images on one of them throw the exception

void Train(array<String> images, array<int> labels)
    {

        std::vector<cv::Mat> cvImages;
        std::vector<int> cvLabels;          

        msclr::interop::marshal_context oMarshalContext;

        for (int i = 0; i < labels->Length; i++)
        {       
            const char* filename = oMarshalContext.marshal_as<const char*>(images[i]);
            cvImages.push_back(cv::imread(filename, CV_LOAD_IMAGE_GRAYSCALE));
            cvLabels.push_back(labels[i]);
        }                       

        (*d_model)->train(cvImages, cvLabels);
    }

please again, add a check, if imread succeeded (and only then push_back() anything)

also check images.size() == labels.size()

also, just to make sure it's not that -- try without any clr/cli/managed code. (mixing opencv (RAII based) and clr code is a terrible idea)

(*d_model)->train -- what's d_model ?

*d_model = cv::face::createFisherFaceRecognizer(d_num_components, d_threshold);

the EXACT type of d_model, please ?

cv::Ptr<cv::face::facerecognizer>* d_model = new cv::Ptr<cv::face::facerecognizer>();