Color checker detector cannot to detect the board

[stereomatchingkiss]

opencv version : 4.5.5 os : win10 x64

Using the detector of mcc module from this post, the network cannot detect the color checker reliable.

Codes

void test_detector_net()
{
    string const model_path = "frozen_inference_graph.pb";
    string const pbtxt_path = "graph.pbtxt";

    cv::dnn::Net net = cv::dnn::readNetFromTensorflow(model_path, pbtxt_path);
    auto image = cv::imread("rec/img-colorchecker.jpg");
    int const rows = image.size[0];
    int const cols = image.size[1];
    net.setInput(cv::dnn::blobFromImage(image, 1.0, cv::Size(), cv::Scalar(), true));
    cv::Mat output = net.forward();

    Mat detectionMat(output.size[2], output.size[3], CV_32F, output.ptr<float>());
    std::cout<<detectionMat.size()<<std::endl;
    for(int i = 0; i < detectionMat.rows; i++){
        float const confidence = detectionMat.at<float>(i, 2);
        std::cout<<"confidence = "<<confidence<<std::endl;
        if(confidence > 0.5f){
           float xTopLeft = max(0.0f, detectionMat.at<float>(i, 3) * cols);
           float yTopLeft = max(0.0f, detectionMat.at<float>(i, 4) * rows);
           float xBottomRight = min((float)cols - 1, detectionMat.at<float>(i, 5) * cols);
           float yBottomRight = min((float)rows - 1, detectionMat.at<float>(i, 6) * rows);

           cv::Point2f const topLeft = {xTopLeft, yTopLeft};
           cv::Point2f const bottomRight = {xBottomRight, yBottomRight};
           cv::rectangle(image, topLeft, bottomRight, cv::Scalar(255, 0, 0), 3);
        }
    }

    cv::resize(image, image, cv::Size(640, 1024));
    imshow("image result | q or esc to quit", image);
    waitKey();
}

I test with the images in the rec folder, only 000037.png work

000003.png--cannot detect anything img-colorchecker.jpg--detect wrong object

img-colorchecker.jpg results

If I lower the confidence threshold, there will have more false positive results. Do I doing something wrong? Which dataset these models train on? Could you shared the links? I would like to train the models another detector, will shared if performance is better, thanks.