Change julia array to C++ vector

[kvmanohar22]

Encountered this bug while testing the test case 6.
With the earlier implementation of conversion of julia array to C++, I used to get the following error :

test 6: Conversion of images from Images.jl to OpenCV Mat
:1:1: error: no matching function for call to 'at_u'
^
__cxxjl_45.cpp:6:6: note: candidate function not viable: no known conversion from 'std::vector<double, std::allocator<double> >' to 'double' for 4th argument
void at_u(cv::Mat &img, int row, int col, double val)
     ^
__cxxjl_45.cpp:1:5: note: candidate function not viable: requires 3 arguments, but 4 were provided
int at_u(cv::Mat &img, int row, int col)
    ^
:1:1: error: no matching function for call to 'at_s'
^
__cxxjl_45.cpp:17:6: note: candidate function not viable: no known conversion from 'std::vector<double, std::allocator<double> >' to 'double' for 4th argument
void at_s(cv::Mat &img, int row, int col, double val)
     ^
__cxxjl_45.cpp:12:5: note: candidate function not viable: requires 3 arguments, but 4 were provided
int at_s(cv::Mat &img, int row, int col)
    ^
:1:1: error: no matching function for call to 'at_us'
^
__cxxjl_45.cpp:28:6: note: candidate function not viable: no known conversion from 'std::vector<double, std::allocator<double> >' to 'double' for 4th argument
void at_us(cv::Mat &img, int row, int col, double val)
     ^
__cxxjl_45.cpp:23:5: note: candidate function not viable: requires 3 arguments, but 4 were provided
int at_us(cv::Mat &img, int row, int col)
    ^
:1:1: error: no matching function for call to 'at_f'
^
__cxxjl_45.cpp:39:6: note: candidate function not viable: no known conversion from 'std::vector<double, std::allocator<double> >' to 'double' for 4th argument
void at_f(cv::Mat &img, int row, int col, double val)
     ^
__cxxjl_45.cpp:34:7: note: candidate function not viable: requires 3 arguments, but 4 were provided
float at_f(cv::Mat &img, int row, int col)
      ^
:1:1: error: no matching function for call to 'at_d'
^
__cxxjl_45.cpp:50:6: note: candidate function not viable: no known conversion from 'std::vector<double, std::allocator<double> >' to 'double' for 4th argument
void at_d(cv::Mat &img, int row, int col, double val)
     ^
__cxxjl_45.cpp:45:8: note: candidate function not viable: requires 3 arguments, but 4 were provided
double at_d(cv::Mat &img, int row, int col)
       ^

The present implementation solves the problem.

[maxruby]

Where is cppvec passed next? I see that you pass colorvec to pixset and then convert the julia array to an std::vectorwith tostdvec. It seems then you should be passing it further e.g, @cxx at_vc(img, row, col, cppvec); as so on to the other cases.

[kvmanohar22]

Oh no! I forgot to replace value with cppvec.

[maxruby]

Not sure still whether the conversion to std::vec should be done in this function.
In any case, there is no need to repeat cppvec = tostdvec(value) in every elseif statement.
Just state it once as shown below.

function pixset(img, row::Int, col::Int, value)
  (row < 0 || col < 0 || row > rows(img) || col > cols(img)) ? throw(BoundsError()) : nothing
  cppvec = tostdvec(value)

. . . code not shown . . .

  elseif (cvtypeval(img) == CV_8SC3)
    @cxx at_vc(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_8UC3)
    @cxx at_v3b(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_16SC3)
    @cxx at_v3s(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_16UC3)
    @cxx at_v3us(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_32SC3)
    @cxx at_v3i(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_32FC3)
    @cxx at_v3f(img, row, col, cppvec);
  elseif (cvtypeval(img) == CV_64FC3)
    @cxx at_v3d(img, row, col, cppvec);
  else throw(ArgumentError("Image format not recognized!"))
  end
end

[kvmanohar22]

Not sure still whether the conversion to std::vec should be done in this function

Yes, I think whether the conversion is made later on or before shouldn't make difference. But unexpectedly I found that the previous implementation led to the errors as mentioned above. Shall I add the details to reproduce the error ?

[maxruby]

Yes, please share the details which led to this error so we have it on record.

[kvmanohar22]

To reproduce the error :
In test/runtests.jl, import the following

using OpenCV
using Images
using Base.Test
using FileIO

In test/jl/tests.jl, add the following test function

function test6()
   println("test 6: Conversion of images from Images.jl to OpenCV Mat")
   filename = joinpath(Pkg.dir("OpenCV"), "./test/images/mandrill.jpg")
   image = FileIO.load(filename)
   dst = convertToMat(image)
   println("Image converted to Mat successfully")
end

test6()

src/OpenCV_Mat.jl as in master branch.
In src/OpenCV_ImageSupport.jl, the following function definition :

function convertToMat(image)
    img = permuteddimsview(channelview(image), (2,3,1))
    cd = Base.size(channelview(image))[1] > 3 ? 1 : 3
    _rows = Base.size(image, 1)
    _cols = Base.size(image, 2)
    if (typeof(img[1,1,1].i) == UInt8)
       if (cd < 3); mat = Mat(512, 512, CV_8UC1); end
       if (cd == 3); mat = Mat(512, 512, CV_8UC3); end
    elseif (typeof(img[1,1,1].i) == Float32)
       if (cd < 3); mat = Mat(512, 512, CV_32FC1); end
       if (cd == 3); mat = Mat(512, 512, CV_32FC3); end
    elseif (typeof(img[1,1,1].i) == Float64)
       if (cd < 3); mat = Mat(512, 512, CV_64FC1); end
       if (cd == 3); mat = Mat(512, 512, CV_64FC3); end
    else
       throw(ArgumentError("Pixel format not supported!"))
    end

    if (cd < 3)   # grayscale or binary
        for j = 1:_rows     # index row first (Mat is row-major order)
            for k =1:_cols  # index column second
                # slow algorithm  - will try to use pointer method (C++)!
                pixset(mat, k, j, float(img[k,j,1].i))
            end
        end
    end

   if (cd == 3)   # color (RGB) image
        for j = 1:512     # index row first (Mat is row-major order)
            for k =1:512  # index column second
                colorvec = tostdvec([float(img[k,j,1].i),float(img[k,j,2].i),float(img[k,j,3].i)])
                pixset(mat, k-1, j-1, colorvec)   # -1 to have 0-indexing per C++
            end
        end
    end

    return(mat)
end