Merge pull request #13723 from smirnov-alexey:gapi_add_sobelxy

* Add Sobel kernel which returns both dx and dy

* Splice dx and dy and extend add_border function

Also change some tests parameters

* Add borderValue parameter in test

* Introduces fluid kernel for sobelxy

Adds tests (basic and performance) on new backend

* Introduces BufHelper struct for some arithmetic

modules/gapi/include/opencv2/gapi/imgproc.hpp

@@ -27,6 +27,7 @@
 namespace cv { namespace gapi {
 
 namespace imgproc {
+    using GMat2 = std::tuple<GMat,GMat>;
     using GMat3 = std::tuple<GMat,GMat,GMat>; // FIXME: how to avoid this?
 
     G_TYPED_KERNEL(GFilter2D, <GMat(GMat,int,Mat,Point,Scalar,int,Scalar)>,"org.opencv.imgproc.filters.filter2D") {
@@ -83,6 +84,12 @@ namespace imgproc {
         }
     };
 
+    G_TYPED_KERNEL_M(GSobelXY, <GMat2(GMat,int,int,int,double,double,int,Scalar)>, "org.opencv.imgproc.filters.sobelxy") {
+        static std::tuple<GMatDesc, GMatDesc> outMeta(GMatDesc in, int ddepth, int, int, double, double, int, Scalar) {
+            return std::make_tuple(in.withDepth(ddepth), in.withDepth(ddepth));
+        }
+    };
+
     G_TYPED_KERNEL(GEqHist, <GMat(GMat)>, "org.opencv.imgproc.equalizeHist"){
         static GMatDesc outMeta(GMatDesc in) {
             return in.withType(CV_8U, 1);
@@ -487,6 +494,58 @@ GAPI_EXPORTS GMat Sobel(const GMat& src, int ddepth, int dx, int dy, int ksize =
                         int borderType = BORDER_DEFAULT,
                         const Scalar& borderValue = Scalar(0));
 
+/** @brief Calculates the first, second, third, or mixed image derivatives using an extended Sobel operator.
+
+In all cases except one, the \f$\texttt{ksize} \times \texttt{ksize}\f$ separable kernel is used to
+calculate the derivative. When \f$\texttt{ksize = 1}\f$, the \f$3 \times 1\f$ or \f$1 \times 3\f$
+kernel is used (that is, no Gaussian smoothing is done). `ksize = 1` can only be used for the first
+or the second x- or y- derivatives.
+
+There is also the special value `ksize = FILTER_SCHARR (-1)` that corresponds to the \f$3\times3\f$ Scharr
+filter that may give more accurate results than the \f$3\times3\f$ Sobel. The Scharr aperture is
+
+\f[\vecthreethree{-3}{0}{3}{-10}{0}{10}{-3}{0}{3}\f]
+
+for the x-derivative, or transposed for the y-derivative.
+
+The function calculates an image derivative by convolving the image with the appropriate kernel:
+
+\f[\texttt{dst} =  \frac{\partial^{xorder+yorder} \texttt{src}}{\partial x^{xorder} \partial y^{yorder}}\f]
+
+The Sobel operators combine Gaussian smoothing and differentiation, so the result is more or less
+resistant to the noise. Most often, the function is called with ( xorder = 1, yorder = 0, ksize = 3)
+or ( xorder = 0, yorder = 1, ksize = 3) to calculate the first x- or y- image derivative. The first
+case corresponds to a kernel of:
+
+\f[\vecthreethree{-1}{0}{1}{-2}{0}{2}{-1}{0}{1}\f]
+
+The second case corresponds to a kernel of:
+
+\f[\vecthreethree{-1}{-2}{-1}{0}{0}{0}{1}{2}{1}\f]
+
+@note First returned matrix correspons to dx derivative while the second one to dy.
+
+@note Rounding to nearest even is procedeed if hardware supports it, if not - to nearest.
+
+@note Function textual ID is "org.opencv.imgproc.filters.sobelxy"
+
+@param src input image.
+@param ddepth output image depth, see @ref filter_depths "combinations"; in the case of
+    8-bit input images it will result in truncated derivatives.
+@param order order of the derivatives.
+@param ksize size of the extended Sobel kernel; it must be odd.
+@param scale optional scale factor for the computed derivative values; by default, no scaling is
+applied (see cv::getDerivKernels for details).
+@param delta optional delta value that is added to the results prior to storing them in dst.
+@param borderType pixel extrapolation method, see cv::BorderTypes
+@param borderValue border value in case of constant border type
+@sa filter2D, gaussianBlur, cartToPolar
+ */
+GAPI_EXPORTS std::tuple<GMat, GMat> SobelXY(const GMat& src, int ddepth, int order, int ksize = 3,
+                        double scale = 1, double delta = 0,
+                        int borderType = BORDER_DEFAULT,
+                        const Scalar& borderValue = Scalar(0));
+
 /** @brief Finds edges in an image using the Canny algorithm.
 
 The function finds edges in the input image and marks them in the output map edges using the

modules/gapi/perf/common/gapi_imgproc_perf_tests.hpp

@@ -31,6 +31,7 @@ class Erode3x3PerfTest : public TestPerfParams<tuple<compare_f, MatType, cv::Siz
 class DilatePerfTest : public TestPerfParams<tuple<compare_f, MatType,int,cv::Size,int, cv::GCompileArgs>> {};
 class Dilate3x3PerfTest : public TestPerfParams<tuple<compare_f, MatType,cv::Size,int, cv::GCompileArgs>> {};
 class SobelPerfTest : public TestPerfParams<tuple<compare_f, MatType,int,cv::Size,int,int,int, cv::GCompileArgs>> {};
+class SobelXYPerfTest : public TestPerfParams<tuple<compare_f, MatType,int,cv::Size,int,int, cv::GCompileArgs>> {};
 class CannyPerfTest : public TestPerfParams<tuple<compare_f, MatType,cv::Size,double,double,int,bool, cv::GCompileArgs>> {};
 class EqHistPerfTest : public TestPerfParams<tuple<compare_f, cv::Size, cv::GCompileArgs >> {};
 class RGB2GrayPerfTest : public TestPerfParams<tuple<compare_f, cv::Size, cv::GCompileArgs >> {};

modules/gapi/perf/common/gapi_imgproc_perf_tests_inl.hpp

@@ -498,6 +498,52 @@ PERF_TEST_P_(SobelPerfTest, TestPerformance)
 
 //------------------------------------------------------------------------------
 
+PERF_TEST_P_(SobelXYPerfTest, TestPerformance)
+{
+    compare_f cmpF;
+    MatType type = 0;
+    int kernSize = 0, dtype = 0, order = 0;
+    cv::Size sz;
+    cv::GCompileArgs compile_args;
+    std::tie(cmpF, type, kernSize, sz, dtype, order, compile_args) = GetParam();
+
+    cv::Mat out_mat_ocv2 = cv::Mat(sz, dtype);
+    cv::Mat out_mat_gapi2 = cv::Mat(sz, dtype);
+
+    initMatsRandN(type, sz, dtype, false);
+
+    // OpenCV code /////////////////////////////////////////////////////////////
+    {
+        cv::Sobel(in_mat1, out_mat_ocv, dtype, order, 0, kernSize);
+        cv::Sobel(in_mat1, out_mat_ocv2, dtype, 0, order, kernSize);
+    }
+
+    // G-API code //////////////////////////////////////////////////////////////
+    cv::GMat in;
+    auto out = cv::gapi::SobelXY(in, dtype, order, kernSize);
+    cv::GComputation c(cv::GIn(in), cv::GOut(std::get<0>(out), std::get<1>(out)));
+
+    // Warm-up graph engine:
+    c.apply(cv::gin(in_mat1), cv::gout(out_mat_gapi, out_mat_gapi2), std::move(compile_args));
+
+    TEST_CYCLE()
+    {
+        c.apply(cv::gin(in_mat1), cv::gout(out_mat_gapi, out_mat_gapi2));
+    }
+
+    // Comparison //////////////////////////////////////////////////////////////
+    {
+        EXPECT_TRUE(cmpF(out_mat_gapi, out_mat_ocv));
+        EXPECT_TRUE(cmpF(out_mat_gapi2, out_mat_ocv2));
+        EXPECT_EQ(out_mat_gapi.size(), sz);
+        EXPECT_EQ(out_mat_gapi2.size(), sz);
+    }
+
+    SANITY_CHECK_NOTHING();
+}
+
+//------------------------------------------------------------------------------
+
 PERF_TEST_P_(CannyPerfTest, TestPerformance)
 {
     compare_f cmpF;

modules/gapi/perf/cpu/gapi_imgproc_perf_tests_fluid.cpp

@@ -32,7 +32,7 @@ INSTANTIATE_TEST_CASE_P(SepFilterPerfTestFluid_other, SepFilterPerfTest,
 INSTANTIATE_TEST_CASE_P(Filter2DPerfTestFluid, Filter2DPerfTest,
     Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add 4, 5, 7 when kernel is ready
+            Values(3),                                     // TODO: add 4, 5, 7 when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::BORDER_DEFAULT),
             Values(-1, CV_32F),
@@ -41,7 +41,7 @@ INSTANTIATE_TEST_CASE_P(Filter2DPerfTestFluid, Filter2DPerfTest,
 INSTANTIATE_TEST_CASE_P(BoxFilterPerfTestFluid, BoxFilterPerfTest,
     Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::BORDER_DEFAULT),
             Values(-1, CV_32F),
@@ -50,7 +50,7 @@ INSTANTIATE_TEST_CASE_P(BoxFilterPerfTestFluid, BoxFilterPerfTest,
 INSTANTIATE_TEST_CASE_P(BlurPerfTestFluid, BlurPerfTest,
     Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::BORDER_DEFAULT),
             Values(cv::compile_args(IMGPROC_FLUID))));
@@ -58,21 +58,21 @@ INSTANTIATE_TEST_CASE_P(BlurPerfTestFluid, BlurPerfTest,
 INSTANTIATE_TEST_CASE_P(GaussianBlurPerfTestFluid, GaussianBlurPerfTest,
     Combine(Values(ToleranceFilter(1e-3f, 0.01).to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::compile_args(IMGPROC_FLUID))));
 
 INSTANTIATE_TEST_CASE_P(MedianBlurPerfTestFluid, MedianBlurPerfTest,
     Combine(Values(AbsExact().to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::compile_args(IMGPROC_FLUID))));
 
 INSTANTIATE_TEST_CASE_P(ErodePerfTestFluid, ErodePerfTest,
     Combine(Values(AbsExact().to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::MorphShapes::MORPH_RECT,
                    cv::MorphShapes::MORPH_CROSS,
@@ -90,7 +90,7 @@ INSTANTIATE_TEST_CASE_P(DISABLED_Erode3x3PerfTestFluid, Erode3x3PerfTest,
 INSTANTIATE_TEST_CASE_P(DilatePerfTestFluid, DilatePerfTest,
     Combine(Values(AbsExact().to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-            Values(3),                                     // add size=5, when kernel is ready
+            Values(3),                                     // TODO: add size=5, when kernel is ready
             Values(szVGA, sz720p, sz1080p),
             Values(cv::MorphShapes::MORPH_RECT,
                    cv::MorphShapes::MORPH_CROSS,
@@ -108,7 +108,7 @@ INSTANTIATE_TEST_CASE_P(DISABLED_Dilate3x3PerfTestFluid, Dilate3x3PerfTest,
 INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
     Combine(Values(AbsExact().to_compare_f()),
             Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
-            Values(3),                                     // add 5x5 once supported
+            Values(3),                                     // TODO: add 5x5 once supported
             Values(szVGA, sz720p, sz1080p),
             Values(-1, CV_16S, CV_32F),
             Values(0, 1),
@@ -118,13 +118,31 @@ INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
 INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
     Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
             Values(CV_32FC1),
-            Values(3),                                     // add 5x5 once supported
+            Values(3),                                     // TODO: add 5x5 once supported
             Values(szVGA, sz720p, sz1080p),
             Values(CV_32F),
             Values(0, 1),
             Values(1, 2),
             Values(cv::compile_args(IMGPROC_FLUID))));
 
+INSTANTIATE_TEST_CASE_P(SobelXYPerfTestFluid, SobelXYPerfTest,
+    Combine(Values(AbsExact().to_compare_f()),
+            Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
+            Values(3),                                     // TODO: add 5x5 once supported
+            Values(szVGA, sz720p, sz1080p),
+            Values(-1, CV_16S, CV_32F),
+            Values(1, 2),
+            Values(cv::compile_args(IMGPROC_FLUID))));
+
+INSTANTIATE_TEST_CASE_P(SobelXYPerfTestFluid32F, SobelXYPerfTest,
+    Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
+            Values(CV_32FC1),
+            Values(3),                                     // TODO: add 5x5 once supported
+            Values(szVGA, sz720p, sz1080p),
+            Values(CV_32F),
+            Values(1, 2),
+            Values(cv::compile_args(IMGPROC_FLUID))));
+
 INSTANTIATE_TEST_CASE_P(RGB2GrayPerfTestFluid, RGB2GrayPerfTest,
     Combine(Values(ToleranceColor(1e-3).to_compare_f()),
             Values(szVGA, sz720p, sz1080p),

modules/gapi/src/api/kernels_imgproc.cpp

@@ -80,6 +80,13 @@ GMat Sobel(const GMat& src, int ddepth, int dx, int dy, int ksize,
     return imgproc::GSobel::on(src, ddepth, dx, dy, ksize, scale, delta, borderType, bordVal);
 }
 
+std::tuple<GMat, GMat> SobelXY(const GMat& src, int ddepth, int order, int ksize,
+           double scale, double delta,
+           int borderType, const Scalar& bordVal)
+{
+    return imgproc::GSobelXY::on(src, ddepth, order, ksize, scale, delta, borderType, bordVal);
+}
+
 GMat equalizeHist(const GMat& src)
 {
     return imgproc::GEqHist::on(src);

modules/gapi/src/backends/cpu/gcpuimgproc.cpp

@@ -11,6 +11,19 @@
 #include "opencv2/gapi/cpu/imgproc.hpp"
 #include "backends/cpu/gcpuimgproc.hpp"
 
+namespace {
+    cv::Mat add_border(const cv::Mat& in, const int ksize, const int borderType, const cv::Scalar& bordVal){
+        if( borderType == cv::BORDER_CONSTANT )
+        {
+            cv::Mat temp_in;
+            int add = (ksize - 1) / 2;
+            cv::copyMakeBorder(in, temp_in, add, add, add, add, borderType, bordVal);
+            return temp_in(cv::Rect(add, add, in.cols, in.rows));
+        }
+        return in;
+    }
+}
+
 GAPI_OCV_KERNEL(GCPUSepFilter, cv::gapi::imgproc::GSepFilter)
 {
     static void run(const cv::Mat& in, int ddepth, const cv::Mat& kernX, const cv::Mat& kernY, const cv::Point& anchor, const cv::Scalar& delta,
@@ -133,16 +146,19 @@ GAPI_OCV_KERNEL(GCPUSobel, cv::gapi::imgproc::GSobel)
     static void run(const cv::Mat& in, int ddepth, int dx, int dy, int ksize, double scale, double delta, int borderType,
                     const cv::Scalar& bordVal, cv::Mat &out)
     {
-        if( borderType == cv::BORDER_CONSTANT )
-        {
-            cv::Mat temp_in;
-            int add = (ksize - 1) / 2;
-            cv::copyMakeBorder(in, temp_in, add, add, add, add, borderType, bordVal );
-            cv::Rect rect = cv::Rect(add, add, in.cols, in.rows);
-            cv::Sobel(temp_in(rect), out, ddepth, dx, dy, ksize, scale, delta, borderType);
+        cv::Mat temp_in = add_border(in, ksize, borderType, bordVal);
+        cv::Sobel(temp_in, out, ddepth, dx, dy, ksize, scale, delta, borderType);
     }
-        else
-        cv::Sobel(in, out, ddepth, dx, dy, ksize, scale, delta, borderType);
+};
+
+GAPI_OCV_KERNEL(GCPUSobelXY, cv::gapi::imgproc::GSobelXY)
+{
+    static void run(const cv::Mat& in, int ddepth, int order, int ksize, double scale, double delta, int borderType,
+                    const cv::Scalar& bordVal, cv::Mat &out_dx, cv::Mat &out_dy)
+    {
+        cv::Mat temp_in = add_border(in, ksize, borderType, bordVal);
+        cv::Sobel(temp_in, out_dx, ddepth, order, 0, ksize, scale, delta, borderType);
+        cv::Sobel(temp_in, out_dy, ddepth, 0, order, ksize, scale, delta, borderType);
     }
 };
 
@@ -256,6 +272,7 @@ cv::gapi::GKernelPackage cv::gapi::imgproc::cpu::kernels()
         , GCPUErode
         , GCPUDilate
         , GCPUSobel
+        , GCPUSobelXY
         , GCPUCanny
         , GCPUEqualizeHist
         , GCPURGB2YUV

modules/gapi/src/backends/fluid/gfluidimgproc.cpp

@@ -1022,6 +1022,168 @@ GAPI_FLUID_KERNEL(GFluidSobel, cv::gapi::imgproc::GSobel, true)
     }
 };
 
+//---------------------
+//
+// Fluid kernels: SobelXY
+//
+//---------------------
+
+GAPI_FLUID_KERNEL(GFluidSobelXY, cv::gapi::imgproc::GSobelXY, true)
+{
+    static const int Window = 3;
+
+    struct BufHelper
+    {
+        float *kx_dx, *ky_dx,
+              *kx_dy, *ky_dy;
+        float *buf_start;
+        int buf_width, buf_chan;
+
+        static int length(int ksz, int width, int chan)
+        {
+            return ksz + ksz + ksz + ksz    // kernels: kx_dx, ky_dx, kx_dy, ky_dy
+                   + 2 * ksz * width * chan;
+        }
+
+        BufHelper(int ksz, int width, int chan, Buffer& scratch)
+        {
+            kx_dx = scratch.OutLine<float>();
+            ky_dx = kx_dx + ksz;
+            kx_dy = ky_dx + ksz;
+            ky_dy = kx_dy + ksz;
+            buf_start = ky_dy + ksz;
+            buf_width = width;
+            buf_chan = chan;
+        }
+
+        float* operator [](int i) {
+            return buf_start + i *  buf_width * buf_chan;
+        }
+    };
+
+    static void run(const     View  &    in,
+                              int     /* ddepth */,
+                              int     /* order */,
+                              int        ksize,
+                              double    _scale,
+                              double    _delta,
+                              int     /* borderType */,
+                    const cv::Scalar& /* borderValue */,
+                              Buffer&    out_x,
+                              Buffer&    out_y,
+                              Buffer&    scratch)
+    {
+        // TODO: support kernel height 3, 5, 7, 9, ...
+        GAPI_Assert(ksize == 3 || ksize == FILTER_SCHARR);
+
+        int ksz = (ksize == FILTER_SCHARR)? 3: ksize;
+
+        GAPI_Assert(out_x.meta().size.width == out_y.meta().size.width);
+        GAPI_Assert(out_x.meta().chan == out_y.meta().chan);
+
+        int width = out_x.meta().size.width;
+        int chan  = out_x.meta().chan;
+
+        BufHelper buf_helper(ksz, width, chan, scratch);
+
+        auto *kx_dx = buf_helper.kx_dx;
+        auto *ky_dx = buf_helper.ky_dx;
+        auto *kx_dy = buf_helper.kx_dy;
+        auto *ky_dy = buf_helper.ky_dy;
+
+        // Scratch buffer layout:
+        // |kx_dx|ky_dx|kx_dy|ky_dy|3 lines for horizontal kernel|3 lines for vertical kernel|
+        float *buf[3];
+        buf[0] = buf_helper[0];
+        buf[1] = buf_helper[1];
+        buf[2] = buf_helper[2];
+
+        auto scale = static_cast<float>(_scale);
+        auto delta = static_cast<float>(_delta);
+
+        auto calc = [&](const View& src, Buffer& dst, float* kx, float* ky) {
+            //     DST     SRC     OP         __VA_ARGS__
+            UNARY_(uchar , uchar , run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_(ushort, ushort, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( short, uchar , run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( short, ushort, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( short,  short, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( float, uchar , run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( float, ushort, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( float,  short, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+            UNARY_( float,  float, run_sobel, dst, src, kx, ky, ksz, scale, delta, buf);
+
+            CV_Error(cv::Error::StsBadArg, "unsupported combination of types");
+        };
+
+        // calculate x-derivative
+        calc(in, out_x, kx_dx, ky_dx);
+
+        // Move pointers to calculate dy(preventing buffer data corruption)
+        buf[0] = buf_helper[3];
+        buf[1] = buf_helper[4];
+        buf[2] = buf_helper[5];
+
+        // calculate y-derivative
+        calc(in, out_y, kx_dy, ky_dy);
+    }
+
+    static void initScratch(const GMatDesc&    in,
+                                  int       /* ddepth */,
+                                  int          order,
+                                  int          ksize,
+                                  double    /* scale */,
+                                  double    /* delta */,
+                                  int       /* borderType */,
+                            const Scalar  & /* borderValue */,
+                                  Buffer  &    scratch)
+    {
+        // TODO: support kernel height 3, 5, 7, 9, ...
+        GAPI_Assert(ksize == 3 || ksize == FILTER_SCHARR);
+        int ksz = (ksize == FILTER_SCHARR) ? 3 : ksize;
+
+        int width = in.size.width;
+        int chan  = in.chan;
+        int buflen = BufHelper::length(ksz, width, chan);
+
+        cv::gapi::own::Size bufsize(buflen, 1);
+        GMatDesc bufdesc = {CV_32F, 1, bufsize};
+        Buffer buffer(bufdesc);
+        scratch = std::move(buffer);
+
+        BufHelper buf_helper(ksz, width, chan, scratch);
+
+        auto *kx_dx = buf_helper.kx_dx;
+        auto *ky_dx = buf_helper.ky_dx;
+        auto *kx_dy = buf_helper.kx_dy;
+        auto *ky_dy = buf_helper.ky_dy;
+
+        Mat kxmatX(1, ksize, CV_32FC1, kx_dx);
+        Mat kymatX(ksize, 1, CV_32FC1, ky_dx);
+        getDerivKernels(kxmatX, kymatX, order, 0, ksize);
+
+        Mat kxmatY(1, ksize, CV_32FC1, kx_dy);
+        Mat kymatY(ksize, 1, CV_32FC1, ky_dy);
+        getDerivKernels(kxmatY, kymatY, 0, order, ksize);
+    }
+
+    static void resetScratch(Buffer& /* scratch */)
+    {
+    }
+
+    static Border getBorder(const cv::GMatDesc& /* src */,
+                                      int       /* ddepth */,
+                                      int       /* order */,
+                                      int       /* ksize */,
+                                      double    /* scale */,
+                                      double    /* delta */,
+                                      int          borderType,
+                            const cv::Scalar  &    borderValue)
+    {
+        return {borderType, borderValue};
+    }
+};
+
 //------------------------
 //
 // Fluid kernels: filter2D
@@ -1546,6 +1708,7 @@ cv::gapi::GKernelPackage cv::gapi::imgproc::fluid::kernels()
       , GFluidMedianBlur
       , GFluidGaussBlur
       , GFluidSobel
+      , GFluidSobelXY
     #if 0
       , GFluidCanny        -- not fluid (?)
       , GFluidEqualizeHist -- not fluid

modules/gapi/test/common/gapi_imgproc_tests.hpp

@@ -26,6 +26,7 @@ struct Erode3x3Test : public TestParams <std::tuple<compare_f,MatType,cv::Size,b
 struct DilateTest : public TestParams <std::tuple<compare_f,MatType,int,cv::Size,int,bool,cv::GCompileArgs>> {};
 struct Dilate3x3Test : public TestParams <std::tuple<compare_f,MatType,cv::Size,bool,int,cv::GCompileArgs>> {};
 struct SobelTest : public TestParams <std::tuple<compare_f,MatType,int,cv::Size,int,int,int,bool,cv::GCompileArgs>> {};
+struct SobelXYTest : public TestParams <std::tuple<compare_f,MatType,int,cv::Size,int,int,int,int,cv::GCompileArgs>> {};
 struct EqHistTest : public TestParams <std::tuple<compare_f,cv::Size,bool,cv::GCompileArgs>> {};
 struct CannyTest : public TestParams <std::tuple<compare_f,MatType,cv::Size,double,double,int,bool,bool,cv::GCompileArgs>> {};
 struct RGB2GrayTest : public  TestParams<std::tuple<compare_f,cv::Size,bool,cv::GCompileArgs>> {};

modules/gapi/test/common/gapi_imgproc_tests_inl.hpp

@@ -353,6 +353,46 @@ TEST_P(SobelTest, AccuracyTest)
     }
 }
 
+TEST_P(SobelXYTest, AccuracyTest)
+{
+    compare_f cmpF;
+    MatType type = 0;
+    int kernSize = 0, dtype = 0, order = 0, border_type = 0, border_val = 0;
+    cv::Size sz;
+    cv::GCompileArgs compile_args;
+    std::tie(cmpF, type, kernSize, sz, dtype, order, border_type, border_val, compile_args) = GetParam();
+    initMatsRandN(type, sz, dtype);
+    cv::Mat out_mat_ocv2 = cv::Mat(sz, dtype);
+    cv::Mat out_mat_gapi2 = cv::Mat(sz, dtype);
+
+    // G-API code //////////////////////////////////////////////////////////////
+    cv::GMat in;
+    auto out = cv::gapi::SobelXY(in, dtype, order, kernSize, 1, 0, border_type, border_val);
+
+    cv::GComputation c(cv::GIn(in), cv::GOut(std::get<0>(out), std::get<1>(out)));
+    c.apply(cv::gin(in_mat1), cv::gout(out_mat_gapi, out_mat_gapi2), std::move(compile_args));
+    // OpenCV code /////////////////////////////////////////////////////////////
+    {
+        // workaround for cv::Sobel
+        cv::Mat temp_in;
+        if(border_type == cv::BORDER_CONSTANT)
+        {
+            int n_pixels = (kernSize - 1) / 2;
+            cv::copyMakeBorder(in_mat1, temp_in, n_pixels, n_pixels, n_pixels, n_pixels, border_type, border_val);
+            in_mat1 = temp_in(cv::Rect(n_pixels, n_pixels, in_mat1.cols, in_mat1.rows));
+        }
+        cv::Sobel(in_mat1, out_mat_ocv, dtype, order, 0, kernSize, 1, 0, border_type);
+        cv::Sobel(in_mat1, out_mat_ocv2, dtype, 0, order, kernSize, 1, 0, border_type);
+    }
+    // Comparison //////////////////////////////////////////////////////////////
+    {
+        EXPECT_TRUE(cmpF(out_mat_gapi, out_mat_ocv));
+        EXPECT_TRUE(cmpF(out_mat_gapi2, out_mat_ocv2));
+        EXPECT_EQ(out_mat_gapi.size(), sz);
+        EXPECT_EQ(out_mat_gapi2.size(), sz);
+    }
+}
+
 TEST_P(EqHistTest, AccuracyTest)
 {
     compare_f cmpF;

modules/gapi/test/cpu/gapi_imgproc_tests_cpu.cpp

@@ -153,6 +153,30 @@ INSTANTIATE_TEST_CASE_P(SobelTestCPU32F, SobelTest,
 /*init output matrices or not*/ testing::Bool(),
                                 Values(cv::compile_args(IMGPROC_CPU))));
 
+INSTANTIATE_TEST_CASE_P(SobelXYTestCPU, SobelXYTest,
+                        Combine(Values(AbsExact().to_compare_f()),
+                                Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
+                                Values(3, 5),
+                                Values(cv::Size(1280, 720),
+                                       cv::Size(640, 480)),
+                                Values(-1, CV_16S, CV_32F),
+                                Values(1, 2),
+                                Values(BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT),
+                                Values(0, 1, 255),
+                                Values(cv::compile_args(IMGPROC_CPU))));
+
+INSTANTIATE_TEST_CASE_P(SobelXYTestCPU32F, SobelXYTest,
+                        Combine(Values(AbsExact().to_compare_f()),
+                                Values(CV_32FC1),
+                                Values(3, 5),
+                                Values(cv::Size(1280, 720),
+                                       cv::Size(640, 480)),
+                                Values(CV_32F),
+                                Values(1, 2),
+                                Values(BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT),
+                                Values(0, 1, 255),
+                                Values(cv::compile_args(IMGPROC_CPU))));
+
 INSTANTIATE_TEST_CASE_P(EqHistTestCPU, EqHistTest,
                         Combine(Values(AbsExact().to_compare_f()),
                                 Values(cv::Size(1280, 720),

modules/gapi/test/cpu/gapi_imgproc_tests_fluid.cpp

@@ -132,6 +132,30 @@ INSTANTIATE_TEST_CASE_P(SobelTestFluid32F, SobelTest,
                                 Values(true, false),
                                 Values(cv::compile_args(IMGPROC_FLUID))));
 
+INSTANTIATE_TEST_CASE_P(SobelXYTestFluid, SobelXYTest,
+                        Combine(Values(AbsExact().to_compare_f()),
+                                Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
+                                Values(3),
+                                Values(cv::Size(1280, 720),
+                                       cv::Size(640, 480)),
+                                Values(-1, CV_16S, CV_32F),
+                                Values(1, 2),
+                                Values(BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT_101),
+                                Values(0, 1, 255),
+                                Values(cv::compile_args(IMGPROC_FLUID))));
+
+INSTANTIATE_TEST_CASE_P(SobelXYTestFluid32F, SobelXYTest,
+                        Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
+                                Values(CV_32FC1),
+                                Values(3),
+                                Values(cv::Size(1280, 720),
+                                       cv::Size(640, 480)),
+                                Values(CV_32F),
+                                Values(1, 2),
+                                Values(BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT_101),
+                                Values(0, 1, 255),
+                                Values(cv::compile_args(IMGPROC_FLUID))));
+
 INSTANTIATE_TEST_CASE_P(boxFilterTestFluid32, BoxFilterTest,
                         Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
                                 Values(CV_8UC1, CV_16UC1, CV_16SC1),