OpenCV(4.4.0) Error: Assertion failed ((unsigned)i < (unsigned)cn) in cv::Vec<float,4>::operator [], file D:\opencv\opencv-4.4.0\build\install\include\opencv2\core\matx.hpp, line 1129

[Yae666]

Hello, when I modified const int RUN_TYPE = 1 in the Configure.h file while trying to reproduce the GES-GSP model, I encountered the following issue. Have you experienced this before, and if so, how did you resolve it? I have installed both opencv4.4.0 and opencv-contrib, and compiled them.

[RussRobin]

Hi,

Just to clarify that I’m not among the authors of GES-GSP and not responsible for debugging their codes. I haven’t met this error before. Basically it seems to be an error with opencv compilation.

[storious]

This error is caused by that a variable type of "cv::Vec<float,4>" access a index out of the bound.

The reason which in the Code\Feature\ImageData.cpp, a function transRectangular2Polar.
The original author want to use a exist variable to store a pixel's coordinate，but he forgot the variable can‘t be accessed with the usage of like var[4] .

You can try replacing the follow code to the old transRectangular2Polar.

// in Code\Feature\ImageData.cpp
// replace the old function
pair<double, double> transRectangular2Polar(Vec4f line, int image_width, int image_height)
{
	float point_x = line[2] - image_width / 2.0;
	float point_y = image_height / 2.0 - line[3];

	if (abs(line[0]) < 1e-5)
	{
		if (point_x > 0) 
			return pair<double, double>(point_x, 0);
		else
			return pair<double, double>(point_x, CV_PI);
	}

	if (abs(line[1]) < 1e-5)
	{
		if (point_y > 0) 
			return pair<double, double>(point_y, CV_PI / 2);
		else
			return pair<double, double>(point_y, 3 * CV_PI / 2);
	}

	float k = line[1] / line[0];
	float y_intercept = point_y - k * point_x; 

	float theta;

	if (k < 0 && y_intercept > 0)
		theta = atan(-1 / k);
	else if (k > 0 && y_intercept > 0)
		theta = CV_PI + atan(-1 / k);
	else if (k < 0 && y_intercept < 0)
		theta = CV_PI + atan(-1 / k);
	else if (k > 0 && y_intercept < 0)
		theta = 2 * CV_PI + atan(-1 / k);

	float _cos = cos(theta);
	float _sin = sin(theta);

	float r = point_x * _cos + point_y * _sin; 
	return pair<double, double>(r, theta);
}

I'm not sure if this will be helpful to you , as I am refactoring the GES-GSP with cmake, openCV 4.12.0 and remove the VLfeat and some unnecessary dll.

[RussRobin]

Hi @storious ,

Thank you for the careful debugging! I've updated the repo and credited you in this commit. You may also submit an issue or a pull request to the GES-GSP repo, although I'm not sure whether Peng is still actively maintaining it.