Create or Reshape OpenCV 3 Channel Mat From Array - c++

Question

I want to build a 3 channel matrix from a 1D array of data of arbitrary data type, row dimension, column dimension, and channel dimension. In my example, I have a 1x12 double array of data, and I want to transform it into a 2x2x3 OpenCv matrix.
double rawData[] = {1.0, 1.0, 1.0, 1.0, 2.0, 2.0, 2.0, 2.0, 3.0, 3.0, 3.0, 3.0};
My goal is to to have :
Channel 1:
[ 1, 1;
1, 1]
Channel 2:
[ 2, 2;
2, 2]
Channel 3:
[ 3, 3;
3, 3]
This is what I've tried:
cv::Mat aMat = cv::Mat(2, 2, CV_64FC3, rawData)
But OpenCv doesn't agree with how it should use that rawData buffer. When I split the matrix by channels and print each individual channel with :
cv::Mat channels[3];
cv::split(aMat ,channels);
It does:
Channel 1:
[ 1, 1;
2, 3]
Channel 2:
[ 1, 2;
2, 3]
Channel 3:
[ 1, 2;
3, 3]
I've also tried loading the data in a 1D Mat and then using cv::Reshape but the result is the same.
How can I get the channels to look the way I want? Do I really have to manually assign every index or is there a more elegant/optimized way?
Edit: Problem Constraints
The order of the data in rawData should not be rearranged, as that would introduce additional complexity to determine the dimensions and data type. I would prefer to use OpenCV interfaces if possible.

Answer 1

My solution is close to berak's, with splitting the data into separate channels. Except one important (I think) difference. I initialize a single row matrix with the datatype already encoded, ie:
cv::Mat aMat = cv::Mat(1, 12, CV_64F, data);
I then use the colRange method to extract parts of the column:
std::vector<cv::Mat> channelVector(3);
channelVector[0] = aMat.colRange(0,4);
channelVector[1] = aMat.colRange(4,8);
channelVector[2] = aMat.colRange(8,12);
I do this to avoid pointer arithmetic. The strength here is that I can have my data buffer be of type void* and I do not have to worry about first casting the pointer to the correct type in order to increment the buffer pointer.
I then iterate with some reshape action.
for(cv::Mat& channel : channelVector)
channel = channel.reshape(1,2);
And finally merge.
cv::Mat combinedMatrix;
cv::merge(channelVector, combinedMatrix);
This should be efficient, as these operations are O(1). I'm not sure about the merge, I think that actually copies data but I can't verify that... but I was going to clone() the final result anyway, so that works out for me.

Answer 2

just rearrange your data:
double rawData[] = {1.0, 2.0, 3.0, 1.0, 2.0, 3.0, 1.0, 2.0, 3.0, 1.0, 2.0, 3.0};
or build seperate 'channel' Mats from your data, and later merge:
double rawData[] = {1.0, 1.0, 1.0, 1.0,
2.0, 2.0, 2.0, 2.0,
3.0, 3.0, 3.0, 3.0};
Mat chan[3] = {
Mat(2,2,CV_64F, rawData),
Mat(2,2,CV_64F, rawData+4),
Mat(2,2,CV_64F, rawData+8)
};
Mat merged;
cv::merge(chan,3,merged);