错误的梯度矢量场OpenCV C++

我尝试在OpenCV（C++）中显示图像的梯度向量场。

我使用了OpenCV中绘制梯度矢量场的代码：

cv::Mat img(cv::imread("C:/gradient2.png",1));
cv::Mat smoothedPlane = cv::Mat::zeros(img.rows,img.cols,CV_8UC1);
cv::Mat gradientField = cv::Mat::zeros(img.rows,img.cols,CV_32FC2);
cv::Mat GradX, GradY;
//cv::Mat AbsGradX, AbsGradY;
int ddepth = CV_16S;
int scale = 1;
int delta = 0;    
auto size = 40;
GaussianBlur(img,smoothedPlane,cv::Size(51, 51),img.rows*img.cols*0.5);
erode(smoothedPlane, smoothedPlane, cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(size+1,size+1)));
dilate(smoothedPlane, smoothedPlane, cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(size, size)));
dilate(smoothedPlane, smoothedPlane, cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(size, size)));
erode(smoothedPlane, smoothedPlane, cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(size, size)));
Scharr( smoothedPlane, GradX, ddepth, 1, 0, scale);
Scharr( smoothedPlane, GradY, ddepth, 0, 1, scale);
//convertScaleAbs( GradX, AbsGradX );
//convertScaleAbs( GradY, AbsGradY );
for (int i = 0 ; i < GradX.rows ; i ++)
{
    for (int j = 0 ; j < GradX.cols ; j ++)
    {
        cv::Scalar xval = GradX.at<float>(i,j);
        cv::Scalar yval = GradY.at<float>(i,j);
        gradientField.at<cv::Point2f>(i,j) = cv::Point2f(xval.val[0],yval.val[0]);
    }
}
auto resolution = 30;
for (int i = 0 ; i < img.rows ; i += resolution)
{
    for (int j = 0 ; j < img.cols ; j+= resolution)
    {
        cv::Point2f p(j,i);
        cv::Point2f p2(gradientField.at<cv::Point2f>(p)+p);
        cv::arrowedLine(img,p,p2,cv::Scalar(0,0,255),1.5,8,0,0.1);
    }
}
cv::imshow("gradient", img);
cv::waitKey(0);

如果我通过这个访问复制梯度值：

        cv::Scalar xval = GradX.at<float>(i,j);
        cv::Scalar yval = GradY.at<float>(i,j);

输出的结果看起来像浮动访问的结果

如果我通过这个访问复制梯度值：

        cv::Scalar xval = GradX.at<short>(i,j);
        cv::Scalar yval = GradY.at<short>(i,j);

输出的结果看起来像短访问的结果

正确的结果应该显示箭头线，该箭头线来自暗磁盘的外部边界，指向磁盘的外部。提前感谢您的帮助。