I have a small issue that I can't solve. I have looked every other related questions for 2 days. I searched opencv documentation and did several small tests, but I couldn't achieve. Here is my problem:

I have a float array. Let's assume that pointer to the first element is ptr. I want to create cv::Mat out of this pointer with some dimension like below:

auto mat = cv::Mat(1, m, CV_32F, ptr)

The matter is that, as array is float each element takes 4 bytes. So, during constructing cv::Mat, it will take elements sequentially from array buy adding 1 to the ptr. However, what if I want to sample and take every 3'rd element. I couldn't achieve this.

For more clarity let me give this example:

float[] arr = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
auto ptr = arr
auto mat = cv::Mat(1, 3, CV_32F, ptr)
std::cout << mat << "\n" // this should print [1, 4, 7, 10]

So, how should I change the code to achieve that? I know that I can set step parameter in cv::Mat constructor, but it doesn't give the same result.

Thank you for your time and attention.

I have a small issue that I can't solve. I have looked every other related questions for 2 days. I searched opencv documentation and did several small tests, but I couldn't achieve. Here is my problem:

I have a float array. Let's assume that pointer to the first element is ptr. I want to create cv::Mat out of this pointer with some dimension like below:

auto mat = cv::Mat(1, m, CV_32F, ptr)

The matter is that, as array is float each element takes 4 bytes. So, during constructing cv::Mat, it will take elements sequentially from array buy adding 1 to the ptr. However, what if I want to sample and take every 3'rd element. I couldn't achieve this.

For more clarity let me give this example:

float[] arr = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
auto ptr = arr
auto mat = cv::Mat(1, 3, CV_32F, ptr)
std::cout << mat << "\n" // this should print [1, 4, 7, 10]

So, how should I change the code to achieve that? I know that I can set step parameter in cv::Mat constructor, but it doesn't give the same result.

Thank you for your time and attention.

• c++
• opencv

• The step allows for padding between rows, but the data of each row is expected by cv::Mat to be continous. A trivial but inefficient solution would be to copy the relevant data (every 3rd value). – wohlstad Commented Feb 16 at 8:53
• @wohlstad Thabk you for your answer, but I have tried it. I thought there will be more efficient solution. – BHOS Commented Feb 16 at 9:09
• have 1-length rows instead. then you have some hope of skipping data. -- usually these questions are trick questions, revealing further detail that invalidates any previous answers. please edit your question and add more detail. I would hate to bother with an answer only for it to become worthless. – Christoph Rackwitz Commented Feb 16 at 9:54

1 Answer 1

I don't know the full extent of your problem. You might have further constraints that invalidate my answer, but it's valid for what you present so far.

One possible solution could be to consider rows of length 1, because you can have padding between rows, via step argument.

I'm taking your question's pseudo-code (missing semicolons, float[] is not known to me to be C++ syntax) and present pseudo-code in turn.

Docs for that Mat() constructor

float[] arr = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
cv::Mat mat(3, 1, CV_32F, arr, 12) // 3 rows, 1 element per row, 12 bytes per row

You could also make a cv::Mat over the entire source array, having several 3-length rows, then use cv::Mat::col() to select the first column.

Either of the above will give you column-shaped matrix. If you wanted a row-shaped matrix instead, you would need to reshape.

However, the obtained column matrix is not considered continuous.

Due to that, you can't just reshape that matrix. You'll have to make a copy (with clone()) first, which will be continuous.

Then you can reshape that matrix into a single-row matrix.

mat = mat.clone().reshape(1, 1) // cn, rows