Assertion error inside kmeans

C++
1

I’m getting a consistent error: (-215:Assertion failed) (flags & FIXED_TYPE) != 0 in function ‘cv::debug_build_guard::_InputArray::type’ inside the cv::kmeans function in two different 3.4.* releases ( 3.4.10 and 3.4.16).
I think it’s related to my code, and not an actual implementation error in opencv, but can’t figure out what I do wrong. Please help me find my mistake!
The error is thrown at:
kmeans.cpp:

...
        if (compactness < best_compactness)
        {
            best_compactness = compactness;
            if (_centers.needed())
            {
                if (_centers.fixedType() && _centers.channels() == dims)
                    centers.reshape(dims).copyTo(_centers);
                else
                    centers.copyTo(_centers);
            }
            _labels.copyTo(best_labels); <----here
        }

copy.cpp:

...
    int dtype = _dst.type(); <---here

matrix_wrap.cpp:

...
    if( k == STD_VECTOR_MAT )
    {
        const std::vector<Mat>& vv = *(const std::vector<Mat>*)obj;
        if( vv.empty() )
        {
            CV_Assert((flags & FIXED_TYPE) != 0); <--- here the assertion fails
            return CV_MAT_TYPE(flags);
        }
        CV_Assert( i < (int)vv.size() );
        return vv[i >= 0 ? i : 0].type();
    }

My code calling kmeans:

void MultiGaussEstimation::kmeans_estimation(cv::Mat data_as_rows)
{
	cv::Mat masks;
	std::vector<cv::Mat>means;
	cv::Mat data; data_as_rows.convertTo(data, CV_32FC1);
	data = data.clone();
	cv::kmeans(data, K, masks, cv::TermCriteria(cv::TermCriteria::EPS + cv::TermCriteria::COUNT, 10, 1.0), 2, cv::KmeansFlags::KMEANS_PP_CENTERS,means);
....
}
2

I figured where the error should be:
the means parameter causes this assertion error, if it’s not placed inside the code, it works.
What should be the form of the means parameter to not cause assertion error?

3

okay, I was thinking loudly on this forum, the means supposed to be a single Mat, which mat’s every row will represent the mean value of a given segment.
it’s clearly stated here in the docs, but for some reason it missed my attention.

2 Likes