I am working on the DCV to make it compilable with the recent versions of LDC, mir libraries, and stuff. I have not yet simply forked it to work on it. I am including modules one by one for my convenience instead. Hope, I am close to the end. Here is my temporary repo:

https://github.com/aferust/ddcv

I run into some problems with module convolution, especially the function conv invoked by the below code which calls canny edge filter:

Image image = imread("lena.png");
auto slice = image.sliced.rgb2gray;
//auto equalized = slice.histEqualize(slice.flattened.calcHistogram);

slice.asImage.imshow("Original");
auto edge = slice.canny!ubyte(15);
edge.asImage.imshow("edge");
waitKey();

Somehow, the compiler fails in deducting the types. I need some help from Ilya or other people familiar with mir.ndslice.

To reproduce it, download my repo and try to compile it as it is. There is a main with the test code in the repo. Just be sure you have a glfw3.dll/.so.

source\dcv\imgproc\filter.d(547,18): Error: template `dcv.imgproc.convolution.conv` cannot deduce function from argument types `!()(Slice!(ubyte*, 2LU, mir_slice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), TaskPool)`
source\dcv\imgproc\filter.d(548,18): Error: template `dcv.imgproc.convolution.conv` cannot deduce function from argument types `!()(Slice!(ubyte*, 2LU, mir_sut, KernelTensor kerlice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), Slice!(float*, 2LU, mir_slice_kind.contiguous), TaskPool)`                                                                                                                                         lice_kind.contiguous
source\dcv\imgproc\convolution.d(78,13):        Candidate is: `conv(alias bc = neumann, InputTensor, KernelTensor, MaskTensor = KernelTensor)(InputTensor input, KernelTensor kernel, InputTensor prealloc = InputTensor.init, MaskTensor mask = MaskTensor.init, TaskPool pool = taskPool)`                              ut, KernelTensor ker
source\dcv\imgproc\filter.d(674,18): Error: template instance `dcv.imgproc.filter.calcGradients!(Slice!(ubyte*, 2LU, mir_slice_kind.contiguous), float)` error instantiating                                                                                                                                              r instantiating
source\dcv\imgproc\filter.d(694,24):        instantiated from here: `canny!(ubyte, ubyte, mir_slice_kind.contiguous)`
source\app.d(48,34):        instantiated from here: `canny!(ubyte, ubyte, mir_slice_kind.contiguous)`

I tried explicitly passing template parameters in dcv.imgproc.filter.calcGradients like below, but in that way, I am getting some other compilation errors:

alias Empty2Type = Slice!(V*, 2LU, SliceKind.contiguous);
fx = input.conv(neumann, typeof(input), Empty2Type, Empty2Type)
    (kx, emptySlice!(2LU, V), emptySlice!(2LU, V), pool);
fy = input.convinput.conv(neumann, typeof(input), Empty2Type, Empty2Type)
     (ky, emptySlice!(2LU, V), emptySlice!(2LU, V), pool);

Thanks in advance!

On Friday, 26 November 2021 at 09:16:56 UTC, Ferhat Kurtulmuş wrote:

Upps
alias Empty2Type = Slice!(V*, 2LU, SliceKind.contiguous);
fx = input.conv!(neumann, typeof(input), Empty2Type, Empty2Type)
    (kx, emptySlice!(2LU, V), emptySlice!(2LU, V), pool);
fy = input.conv!(neumann, typeof(input), Empty2Type, Empty2Type)
     (ky, emptySlice!(2LU, V), emptySlice!(2LU, V), pool);

On Friday, 26 November 2021 at 09:16:56 UTC, Ferhat Kurtulmuş wrote:

After dealing with, various problems, it turns out the main problem was ndiota with pool.parallel.
https://github.com/aferust/ddcv/blob/main/source/dcv/imgproc/filter.d#L564

foreach (row; /*pool.parallel(*/ndiota(input.shape)/*)*/) // parallel loop causes a linker error
{
    row.each!(i => calcGradientsImpl(fx[i], fy[i], mag[i], orient[i]));
}

The parallel loop causes a linker error here?

error LNK2019: unresolved external symbol _D3mir7ndslice5slice__T9mir_sliceTSQBhQBg8iterator__T13FieldIteratorTSQCqQCp5field__T11ndIotaFieldVmi2ZQsZQCbVmi1VEQEjQEiQEd14mir_slice_kindi2ZQEq__T10lightScopeZQnMxFNaNbNdNiNfZSQGvQGuQGp__TQGmTQGfVmi1VQDli2ZQHe referenced in function _D3mir7ndslice5slice__T9mir_sliceTSQBhQBg8iterator__T13FieldIteratorTSQCqQCp5field__T11ndIotaFieldVmi2ZQsZQCbVmi1VEQEjQEiQEd14mir_slice_kindi2ZQEq__T8opEqualsTQEvVQBxi2ZQuMxFNaNbNiNeKxSQHbQHaQGv__TQGsTQGlVmi1VQDri2ZQHkZb
.dub\build\application-debug-windows-x86_64-ldc_v1.28.0-24645713CE34BFE817BFD3D964187D0E\ddcv.exe : fatal error LNK1120: 1 unresolved externals

On Saturday, 27 November 2021 at 11:19:18 UTC, Ferhat Kurtulmuş wrote:

foreach (row; /*pool.parallel(*/ndiota(input.shape)/*)*/) // parallel loop causes a linker error
{
    row.each!(i => calcGradientsImpl(fx[i], fy[i], mag[i], orient[i]));
}

I also found similar errors but couldn't solve them. I think it has to do with mir.slice.kind. Exactly Kind Topology...

On Saturday, 27 November 2021 at 11:35:21 UTC, Salih Dincer wrote:

I won't use parallel for it as a workaround until it is solved in the mir-algorithm.

On Saturday, 27 November 2021 at 12:16:39 UTC, Ferhat Kurtulmuş wrote:

Linker bug is a compiler bug. We can only find a workaround. The best workaround is to avoid using ndiota with std.parallelism.