opencv/modules/dnn/test/test_misc.cpp

// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
//
// Copyright (C) 2017, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.

#include "test_precomp.hpp"
#include "npy_blob.hpp"
#include <opencv2/core/ocl.hpp>
#include <opencv2/core/opencl/ocl_defs.hpp>
#include <opencv2/dnn/layer.details.hpp>  // CV_DNN_REGISTER_LAYER_CLASS

namespace opencv_test { namespace {

TEST(blobFromImage_4ch, Regression)
{
    Mat ch[4];
    for(int i = 0; i < 4; i++)
        ch[i] = Mat::ones(10, 10, CV_8U)*i;

    Mat img;
    merge(ch, 4, img);
    Mat blob = dnn::blobFromImage(img, 1., Size(), Scalar(), false, false);

    for(int i = 0; i < 4; i++)
    {
        ch[i] = Mat(img.rows, img.cols, CV_32F, blob.ptr(0, i));
        ASSERT_DOUBLE_EQ(cvtest::norm(ch[i], cv::NORM_INF), i);
    }
}

TEST(blobFromImage, allocated)
{
    int size[] = {1, 3, 4, 5};
    Mat img(size[2], size[3], CV_32FC(size[1]));
    Mat blob(4, size, CV_32F);
    void* blobData = blob.data;
    dnn::blobFromImage(img, blob, 1.0 / 255, Size(), Scalar(), false, false);
    ASSERT_EQ(blobData, blob.data);
}

TEST(imagesFromBlob, Regression)
{
    int nbOfImages = 8;

    std::vector<cv::Mat> inputImgs(nbOfImages);
    for (int i = 0; i < nbOfImages; i++)
    {
        inputImgs[i] = cv::Mat::ones(100, 100, CV_32FC3);
        cv::randu(inputImgs[i], cv::Scalar::all(0), cv::Scalar::all(1));
    }

    cv::Mat blob = cv::dnn::blobFromImages(inputImgs, 1., cv::Size(), cv::Scalar(), false, false);
    std::vector<cv::Mat> outputImgs;
    cv::dnn::imagesFromBlob(blob, outputImgs);

    for (int i = 0; i < nbOfImages; i++)
    {
        EXPECT_EQ(0, cvtest::norm(inputImgs[i], outputImgs[i], NORM_INF))
            << "i=" << i
            << " inputImgs[i]=" << inputImgs[i].size
            << " outputImgs[i]=" << outputImgs[i].size;
    }
}

TEST(blobFromImageWithParams_4ch, NHWC_scalar_scale)
{
    Mat img(10, 10, CV_8UC4, cv::Scalar(0,1,2,3));
    std::vector<double> factorVec = {0.1, 0.2, 0.3, 0.4};

    Scalar scalefactor(factorVec[0], factorVec[1], factorVec[2], factorVec[3]);

    Image2BlobParams param;
    param.scalefactor = scalefactor;
    param.datalayout = DNN_LAYOUT_NHWC;
    Mat blob = dnn::blobFromImageWithParams(img, param); // [1, 10, 10, 4]

    float* blobPtr = blob.ptr<float>(0);
    std::vector<float> targetVec = {(float )factorVec[0] * 0, (float )factorVec[1] * 1, (float )factorVec[2] * 2, (float )factorVec[3] * 3}; // Target Value.
    for (int hi = 0; hi < 10; hi++)
    {
        for (int wi = 0; wi < 10; wi++)
        {
            float* hwPtr = blobPtr + hi * 10 * 4 + wi * 4;

            // Check equal
            EXPECT_NEAR(hwPtr[0], targetVec[0], 1e-5);
            EXPECT_NEAR(hwPtr[1], targetVec[1], 1e-5);
            EXPECT_NEAR(hwPtr[2], targetVec[2], 1e-5);
            EXPECT_NEAR(hwPtr[3], targetVec[3], 1e-5);
        }
    }
}

TEST(blobFromImageWithParams_4ch, letter_box)
{
    Mat img(40, 20, CV_8UC4, cv::Scalar(0,1,2,3));

    // Construct target mat.
    Mat targetCh[4];
    // The letterbox will add zero at the left and right of output blob.
    // After the letterbox, every row data would have same value showing as valVec.
    std::vector<uint8_t> valVec = {0,0,0,0,0, 1,1,1,1,1,1,1,1,1,1, 0,0,0,0,0};
    Mat rowM(1, 20, CV_8UC1, valVec.data());

    for(int i = 0; i < 4; i++)
    {
        targetCh[i] = rowM * i;
    }

    Mat targetImg;
    merge(targetCh, 4, targetImg);
    Size targeSize(20, 20);

    Image2BlobParams param;
    param.size = targeSize;
    param.paddingmode = DNN_PMODE_LETTERBOX;
    Mat blob = dnn::blobFromImageWithParams(img, param);
    Mat targetBlob = dnn::blobFromImage(targetImg, 1.0, targeSize); // only convert data from uint8 to float32.
    EXPECT_EQ(0, cvtest::norm(targetBlob, blob, NORM_INF));
}

TEST(blobFromImagesWithParams_4ch, multi_image)
{
    Mat img(10, 10, CV_8UC4, cv::Scalar(0, 1, 2, 3));
    Scalar scalefactor(0.1, 0.2, 0.3, 0.4);

    Image2BlobParams param;
    param.scalefactor = scalefactor;
    param.datalayout = DNN_LAYOUT_NHWC;

    Mat blobs = blobFromImagesWithParams(std::vector<Mat> { img, 2*img }, param);
    vector<Range> ranges;
    ranges.push_back(Range(0, 1));
    ranges.push_back(Range(0, blobs.size[1]));
    ranges.push_back(Range(0, blobs.size[2]));
    ranges.push_back(Range(0, blobs.size[3]));
    Mat blob0 = blobs(ranges);
    ranges[0] = Range(1, 2);
    Mat blob1 = blobs(ranges);

    EXPECT_EQ(0, cvtest::norm(2*blob0, blob1, NORM_INF));
}

TEST(readNet, Regression)
{
    Net net = readNet(findDataFile("dnn/squeezenet_v1.1.prototxt"),
                      findDataFile("dnn/squeezenet_v1.1.caffemodel", false));
    EXPECT_FALSE(net.empty());
    net = readNet(findDataFile("dnn/opencv_face_detector.caffemodel", false),
                  findDataFile("dnn/opencv_face_detector.prototxt"));
    EXPECT_FALSE(net.empty());
    net = readNet(findDataFile("dnn/openface_nn4.small2.v1.t7", false));
    EXPECT_FALSE(net.empty());
    net = readNet(findDataFile("dnn/tiny-yolo-voc.cfg"),
                  findDataFile("dnn/tiny-yolo-voc.weights", false));
    EXPECT_FALSE(net.empty());
    net = readNet(findDataFile("dnn/ssd_mobilenet_v1_coco.pbtxt"),
                  findDataFile("dnn/ssd_mobilenet_v1_coco.pb", false));
    EXPECT_FALSE(net.empty());
}

TEST(readNet, do_not_call_setInput)  // https://github.com/opencv/opencv/issues/16618
{
    // 1. load network
    const string proto = findDataFile("dnn/squeezenet_v1.1.prototxt");
    const string model = findDataFile("dnn/squeezenet_v1.1.caffemodel", false);
    Net net = readNetFromCaffe(proto, model);

    // 2. mistake: no inputs are specified through .setInput()

    // 3. try inference
    Mat res;
    EXPECT_THROW(
    {
        res = net.forward();  // no inputs after loading => should fail
    }, cv::Exception);
    EXPECT_TRUE(res.empty()) << res.size;
}

TEST(Net, empty_forward_18392)
{
    cv::dnn::Net net;
    Mat image(Size(512, 512), CV_8UC3, Scalar::all(0));
    Mat inputBlob = cv::dnn::blobFromImage(image, 1.0, Size(512, 512), Scalar(0,0,0), true, false);
    net.setInput(inputBlob);
    EXPECT_ANY_THROW(Mat output = net.forward());
}

#ifdef HAVE_INF_ENGINE
static
void test_readNet_IE_do_not_call_setInput(Backend backendId)
{
    const Target targetId = DNN_TARGET_CPU;

    const std::string& model = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& proto = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net net = readNet(model, proto);
    net.setPreferableBackend(backendId);
    net.setPreferableTarget(targetId);

    // 2. mistake: no inputs are specified through .setInput()

    // 3. try inference
    Mat res;
    EXPECT_THROW(
    {
        res = net.forward();  // no inputs after loading => should fail
    }, cv::Exception);
    EXPECT_TRUE(res.empty()) << res.size;
}

#ifdef HAVE_DNN_IE_NN_BUILDER_2019
TEST(readNet, do_not_call_setInput_IE_NN_BUILDER_2019)
{
    test_readNet_IE_do_not_call_setInput(DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019);
}
#endif
#ifdef HAVE_DNN_NGRAPH
TEST(readNet, do_not_call_setInput_IE_NGRAPH)
{
    test_readNet_IE_do_not_call_setInput(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH);
}
#endif
#endif  // HAVE_INF_ENGINE

typedef testing::TestWithParam<tuple<Backend, Target> > dump;
TEST_P(dump, Regression)
{
    const int backend  = get<0>(GetParam());
    const int target   = get<1>(GetParam());
    Net net = readNet(findDataFile("dnn/squeezenet_v1.1.prototxt"),
                      findDataFile("dnn/squeezenet_v1.1.caffemodel", false));

    ASSERT_EQ(net.getLayerInputs(net.getLayerId("fire2/concat")).size(), 2);

    int size[] = {1, 3, 227, 227};
    Mat input = cv::Mat::ones(4, size, CV_32F);
    net.setInput(input);
    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);
    EXPECT_FALSE(net.dump().empty());
    net.forward();
    EXPECT_FALSE(net.dump().empty());
}

INSTANTIATE_TEST_CASE_P(/**/, dump, dnnBackendsAndTargets());

class FirstCustomLayer CV_FINAL : public Layer
{
public:
    FirstCustomLayer(const LayerParams &params) : Layer(params) {}

    static Ptr<Layer> create(LayerParams& params)
    {
        return Ptr<Layer>(new FirstCustomLayer(params));
    }

    void forward(InputArrayOfArrays, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays) CV_OVERRIDE
    {
        CV_TRACE_FUNCTION();
        CV_TRACE_ARG_VALUE(name, "name", name.c_str());

        std::vector<Mat> outputs;
        outputs_arr.getMatVector(outputs);
        outputs[0].setTo(1);
    }
};

class SecondCustomLayer CV_FINAL : public Layer
{
public:
    SecondCustomLayer(const LayerParams &params) : Layer(params) {}

    static Ptr<Layer> create(LayerParams& params)
    {
        return Ptr<Layer>(new SecondCustomLayer(params));
    }

    void forward(InputArrayOfArrays, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays) CV_OVERRIDE
    {
        CV_TRACE_FUNCTION();
        CV_TRACE_ARG_VALUE(name, "name", name.c_str());

        std::vector<Mat> outputs;
        outputs_arr.getMatVector(outputs);
        outputs[0].setTo(2);
    }
};

TEST(LayerFactory, custom_layers)
{
    LayerParams lp;
    lp.name = "name";
    lp.type = "CustomType";

    Mat inp(1, 1, CV_32FC1);
    for (int i = 0; i < 3; ++i)
    {
        if (i == 0)      { CV_DNN_REGISTER_LAYER_CLASS(CustomType, FirstCustomLayer); }
        else if (i == 1) { CV_DNN_REGISTER_LAYER_CLASS(CustomType, SecondCustomLayer); }
        else if (i == 2) { LayerFactory::unregisterLayer("CustomType"); }

        Net net;
        net.addLayerToPrev(lp.name, lp.type, lp);

        net.setInput(inp);
        net.setPreferableBackend(DNN_BACKEND_OPENCV);
        Mat output = net.forward();

        if (i == 0)      { EXPECT_EQ(output.at<float>(0), 1); }
        else if (i == 1) { EXPECT_EQ(output.at<float>(0), 2); }
        else if (i == 2) { EXPECT_EQ(output.at<float>(0), 1); }
    }
    LayerFactory::unregisterLayer("CustomType");
}

typedef testing::TestWithParam<tuple<float, Vec3f, int, tuple<Backend, Target> > > setInput;
TEST_P(setInput, normalization)
{
    const float kScale = get<0>(GetParam());
    const Scalar kMean = get<1>(GetParam());
    const int dtype    = get<2>(GetParam());
    const int backend  = get<0>(get<3>(GetParam()));
    const int target   = get<1>(get<3>(GetParam()));
    const bool kSwapRB = true;

    if(backend == DNN_BACKEND_CUDA)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA);
    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16 && dtype != CV_32F)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
    if (backend == DNN_BACKEND_VKCOM && dtype != CV_32F)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_VULKAN);

    Mat inp(5, 5, CV_8UC3);
    randu(inp, 0, 255);
    Mat ref = blobFromImage(inp, kScale, Size(), kMean, kSwapRB, /*crop*/false);

    LayerParams lp;
    Net net;
    net.addLayerToPrev("testLayer", "Identity", lp);
    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    Mat blob = blobFromImage(inp, 1.0, Size(), Scalar(), kSwapRB, /*crop*/false, dtype);
    ASSERT_EQ(blob.type(), dtype);
    net.setInput(blob, "", kScale, kMean);
    Mat out = net.forward();
    ASSERT_EQ(out.type(), CV_32F);
    normAssert(ref, out, "", 4e-4, 1e-3);
}

INSTANTIATE_TEST_CASE_P(/**/, setInput, Combine(
  Values(1.0f, 1.0 / 127.5),
  Values(Vec3f(), Vec3f(50, 50, 50), Vec3f(10, 50, 140)),
  Values(CV_32F, CV_8U),
  dnnBackendsAndTargets()
));

class CustomLayerWithDeprecatedForward CV_FINAL : public Layer
{
public:
    CustomLayerWithDeprecatedForward(const LayerParams &params) : Layer(params) {}

    static Ptr<Layer> create(LayerParams& params)
    {
        return Ptr<Layer>(new CustomLayerWithDeprecatedForward(params));
    }

    virtual void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals) CV_OVERRIDE
    {
        CV_Assert_N(inputs[0]->depth() == CV_32F, outputs[0].depth() == CV_32F);
        cv::add(*inputs[0], 0.5f, outputs[0]);
    }
};

class CustomLayerWithDeprecatedForwardAndFallback CV_FINAL : public Layer
{
public:
    CustomLayerWithDeprecatedForwardAndFallback(const LayerParams &params) : Layer(params) {}

    static Ptr<Layer> create(LayerParams& params)
    {
        return Ptr<Layer>(new CustomLayerWithDeprecatedForwardAndFallback(params));
    }

    void forward(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals) CV_OVERRIDE
    {
        CV_TRACE_FUNCTION();
        CV_TRACE_ARG_VALUE(name, "name", name.c_str());

        CV_OCL_RUN(preferableTarget == DNN_TARGET_OPENCL || preferableTarget == DNN_TARGET_OPENCL_FP16,
                   forward_ocl(inputs, outputs, internals));

        Layer::forward_fallback(inputs, outputs, internals);
    }

    virtual void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals) CV_OVERRIDE
    {
        CV_Assert_N(inputs[0]->depth() == CV_32F, outputs[0].depth() == CV_32F);
        cv::add(*inputs[0], 0.5f, outputs[0]);
    }

#ifdef HAVE_OPENCL
    bool forward_ocl(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr)
    {
        if (inputs_arr.depth() != CV_32F)
            return false;

        std::vector<UMat> inputs;
        std::vector<UMat> outputs;
        inputs_arr.getUMatVector(inputs);
        outputs_arr.getUMatVector(outputs);
        cv::add(inputs[0], 0.5f, outputs[0]);
        return true;
    }
#endif
};

typedef testing::TestWithParam<tuple<Backend, Target> > DeprecatedForward;
TEST_P(DeprecatedForward, CustomLayer)
{
    const int backend  = get<0>(GetParam());
    const int target   = get<1>(GetParam());

    Mat inp(5, 5, CV_32FC1);
    randu(inp, -1.0f, 1.0f);
    inp = blobFromImage(inp);

    CV_DNN_REGISTER_LAYER_CLASS(CustomType, CustomLayerWithDeprecatedForward);
    try
    {
        LayerParams lp;
        Net net;
        net.addLayerToPrev("testLayer", "CustomType", lp);
        net.setPreferableBackend(backend);
        net.setPreferableTarget(target);
        net.setInput(inp);
        Mat out = net.forward();
        normAssert(out, inp + 0.5f, "", 2e-4, 7e-4);
    }
    catch (...)
    {
        LayerFactory::unregisterLayer("CustomType");
        throw;
    }
    LayerFactory::unregisterLayer("CustomType");
}

TEST_P(DeprecatedForward, CustomLayerWithFallback)
{
    const int backend  = get<0>(GetParam());
    const int target   = get<1>(GetParam());

    Mat inp(5, 5, CV_32FC1);
    randu(inp, -1.0f, 1.0f);
    inp = blobFromImage(inp);

    CV_DNN_REGISTER_LAYER_CLASS(CustomType, CustomLayerWithDeprecatedForwardAndFallback);
    try
    {
        LayerParams lp;
        Net net;
        net.addLayerToPrev("testLayer", "CustomType", lp);
        net.setPreferableBackend(backend);
        net.setPreferableTarget(target);
        net.setInput(inp);
        Mat out = net.forward();
        normAssert(out, inp + 0.5f, "", 2e-4, 7e-4);
    }
    catch (...)
    {
        LayerFactory::unregisterLayer("CustomType");
        throw;
    }
    LayerFactory::unregisterLayer("CustomType");
}

INSTANTIATE_TEST_CASE_P(/**/, DeprecatedForward, dnnBackendsAndTargets());

TEST(Net, forwardAndRetrieve)
{
    std::string prototxt =
        "input: \"data\"\n"
        "layer {\n"
        "  name: \"testLayer\"\n"
        "  type: \"Slice\"\n"
        "  bottom: \"data\"\n"
        "  top: \"firstCopy\"\n"
        "  top: \"secondCopy\"\n"
        "  slice_param {\n"
        "    axis: 0\n"
        "    slice_point: 2\n"
        "  }\n"
        "}";
    Net net = readNetFromCaffe(&prototxt[0], prototxt.size());
    net.setPreferableBackend(DNN_BACKEND_OPENCV);

    Mat inp(4, 5, CV_32F);
    randu(inp, -1, 1);
    net.setInput(inp);

    std::vector<String> outNames;
    outNames.push_back("testLayer");
    std::vector<std::vector<Mat> > outBlobs;

    net.forward(outBlobs, outNames);

    EXPECT_EQ(outBlobs.size(), 1);
    EXPECT_EQ(outBlobs[0].size(), 2);
    normAssert(outBlobs[0][0], inp.rowRange(0, 2), "first part");
    normAssert(outBlobs[0][1], inp.rowRange(2, 4), "second part");
}

#ifdef HAVE_INF_ENGINE
static const std::chrono::milliseconds async_timeout(10000);

// This test runs network in synchronous mode for different inputs and then
// runs the same model asynchronously for the same inputs.
typedef testing::TestWithParam<tuple<int, tuple<Backend, Target> > > Async;
TEST_P(Async, model_optimizer_pipeline_set_and_forward_single)
{
    const int dtype = get<0>(GetParam());
    const Backend backendId = get<0>(get<1>(GetParam()));
    const Target targetId = get<1>(get<1>(GetParam()));

    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    if (backendId != DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && backendId != DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        throw SkipTestException("No support for async forward");

    const std::string& model = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& proto = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net netSync = readNet(model, proto);
    netSync.setPreferableBackend(backendId);
    netSync.setPreferableTarget(targetId);

    Net netAsync = readNet(model, proto);
    netAsync.setPreferableBackend(backendId);
    netAsync.setPreferableTarget(targetId);

    // Generate inputs.
    const int numInputs = 10;
    std::vector<Mat> inputs(numInputs);
    int blobSize[] = {2, 6, 75, 113};
    for (int i = 0; i < numInputs; ++i)
    {
        inputs[i].create(4, &blobSize[0], dtype);
        randu(inputs[i], 0, 255);
    }

    // Run synchronously.
    std::vector<Mat> refs(numInputs);
    for (int i = 0; i < numInputs; ++i)
    {
        netSync.setInput(inputs[i]);
        refs[i] = netSync.forward().clone();
    }

    // Run asynchronously. To make test more robust, process inputs in the reversed order.
    for (int i = numInputs - 1; i >= 0; --i)
    {
        netAsync.setInput(inputs[i]);

        AsyncArray out = netAsync.forwardAsync();
        ASSERT_TRUE(out.valid());
        Mat result;
        EXPECT_TRUE(out.get(result, async_timeout));
        normAssert(refs[i], result, format("Index: %d", i).c_str(), 0, 0);
    }
}

TEST_P(Async, model_optimizer_pipeline_set_and_forward_all)
{
    const int dtype = get<0>(GetParam());
    const Backend backendId = get<0>(get<1>(GetParam()));
    const Target targetId = get<1>(get<1>(GetParam()));

    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    if (backendId != DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && backendId != DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        throw SkipTestException("No support for async forward");

    const std::string& model = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& proto = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net netSync = readNet(model, proto);
    netSync.setPreferableBackend(backendId);
    netSync.setPreferableTarget(targetId);

    Net netAsync = readNet(model, proto);
    netAsync.setPreferableBackend(backendId);
    netAsync.setPreferableTarget(targetId);

    // Generate inputs.
    const int numInputs = 10;
    std::vector<Mat> inputs(numInputs);
    int blobSize[] = {2, 6, 75, 113};
    for (int i = 0; i < numInputs; ++i)
    {
        inputs[i].create(4, &blobSize[0], dtype);
        randu(inputs[i], 0, 255);
    }

    // Run synchronously.
    std::vector<Mat> refs(numInputs);
    for (int i = 0; i < numInputs; ++i)
    {
        netSync.setInput(inputs[i]);
        refs[i] = netSync.forward().clone();
    }

    // Run asynchronously. To make test more robust, process inputs in the reversed order.
    std::vector<AsyncArray> outs(numInputs);
    for (int i = numInputs - 1; i >= 0; --i)
    {
        netAsync.setInput(inputs[i]);
        outs[i] = netAsync.forwardAsync();
    }

    for (int i = numInputs - 1; i >= 0; --i)
    {
        ASSERT_TRUE(outs[i].valid());
        Mat result;
        EXPECT_TRUE(outs[i].get(result, async_timeout));
        normAssert(refs[i], result, format("Index: %d", i).c_str(), 0, 0);
    }
}

TEST_P(Async, create_layer_pipeline_set_and_forward_all)
{
    const int dtype = get<0>(GetParam());
    const Backend backendId = get<0>(get<1>(GetParam()));
    const Target targetId = get<1>(get<1>(GetParam()));

    if (backendId != DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && backendId != DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        throw SkipTestException("No support for async forward");

    // Exception: Default implementation fallbacks in asynchronous mode
    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && dtype == CV_8U)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net netSync;
    Net netAsync;
    {
        int inChannels = 4;
        int outChannels = 12;
        int group = 3;
        Size inSize(113, 75);
        Size kernel(4, 5);
        Size stride(2, 3);
        Size pad(0, 1);
        Size dilation(1, 1);
        bool hasBias = true;

        int sz[] = {outChannels, inChannels / group, kernel.height, kernel.width};
        Mat weights(4, &sz[0], CV_32F);
        randu(weights, -1.0f, 1.0f);

        LayerParams lp;
        lp.set("kernel_w", kernel.width);
        lp.set("kernel_h", kernel.height);
        lp.set("pad_w", pad.width);
        lp.set("pad_h", pad.height);
        lp.set("stride_w", stride.width);
        lp.set("stride_h", stride.height);
        lp.set("dilation_w", dilation.width);
        lp.set("dilation_h", dilation.height);
        lp.set("num_output", outChannels);
        lp.set("group", group);
        lp.set("bias_term", hasBias);
        lp.type = "Convolution";
        lp.name = "testLayer";
        lp.blobs.push_back(weights);
        if (hasBias)
        {
            Mat bias(1, outChannels, CV_32F);
            randu(bias, -1.0f, 1.0f);
            lp.blobs.push_back(bias);
        }
        int inpSz[] = {1, inChannels, inSize.height, inSize.width};
        Mat input(4, &inpSz[0], CV_32F);

        netSync.addLayerToPrev(lp.name, lp.type, lp);

        netAsync.addLayerToPrev(lp.name, lp.type, lp);
    }

    netSync.setPreferableBackend(backendId);
    netSync.setPreferableTarget(targetId);

    netAsync.setPreferableBackend(backendId);
    netAsync.setPreferableTarget(targetId);

    // Generate inputs.
    const int numInputs = 10;
    std::vector<Mat> inputs(numInputs);
    int blobSize[] = {1, 4, 75, 113};
    for (int i = 0; i < numInputs; ++i)
    {
        inputs[i].create(4, &blobSize[0], dtype);
        randu(inputs[i], 0, 255);
    }

    // Run synchronously.
    std::vector<Mat> refs(numInputs);
    for (int i = 0; i < numInputs; ++i)
    {
        netSync.setInput(inputs[i]);
        refs[i] = netSync.forward().clone();
    }

    // Run asynchronously. To make test more robust, process inputs in the reversed order.
    std::vector<AsyncArray> outs(numInputs);
    for (int i = numInputs - 1; i >= 0; --i)
    {
        netAsync.setInput(inputs[i]);
        outs[i] = netAsync.forwardAsync();
    }

    for (int i = numInputs - 1; i >= 0; --i)
    {
        ASSERT_TRUE(outs[i].valid());
        Mat result;
        EXPECT_TRUE(outs[i].get(result, async_timeout));
        normAssert(refs[i], result, format("Index: %d", i).c_str(), 0, 0);
    }
}

INSTANTIATE_TEST_CASE_P(/**/, Async, Combine(
    Values(CV_32F, CV_8U),
    dnnBackendsAndTargetsIE()
));

typedef testing::TestWithParam<tuple<Backend, Target> > Test_Model_Optimizer;
TEST_P(Test_Model_Optimizer, forward_two_nets)
{
    const Backend backendId = get<0>(GetParam());
    const Target targetId = get<1>(GetParam());

    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    const std::string& model = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& proto = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net net0 = readNet(model, proto);
    net0.setPreferableTarget(targetId);

    Net net1 = readNet(model, proto);
    net1.setPreferableTarget(targetId);

    // Generate inputs.
    int blobSize[] = {2, 6, 75, 113};
    Mat input(4, &blobSize[0], CV_32F);
    randu(input, 0, 255);

    net0.setInput(input);
    Mat ref0 = net0.forward().clone();

    net1.setInput(input);
    Mat ref1 = net1.forward();

    net0.setInput(input);
    Mat ref2 = net0.forward();

    normAssert(ref0, ref2, 0, 0);
}

TEST_P(Test_Model_Optimizer, readFromBuffer)
{
    const Backend backendId = get<0>(GetParam());
    const Target targetId = get<1>(GetParam());

    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    if (backendId != DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && backendId != DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        throw SkipTestException("No support for async forward");

    const std::string& weightsFile = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& modelFile = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net net1 = readNetFromModelOptimizer(modelFile, weightsFile);
    net1.setPreferableBackend(backendId);
    net1.setPreferableTarget(targetId);


    std::vector<char> modelConfig;
    readFileContent(modelFile, modelConfig);
    std::vector<char> weights;
    readFileContent(weightsFile, weights);

    Net net2 = readNetFromModelOptimizer(
            (const uchar*)modelConfig.data(), modelConfig.size(),
            (const uchar*)weights.data(), weights.size()
    );
    net2.setPreferableBackend(backendId);
    net2.setPreferableTarget(targetId);

    int blobSize[] = {2, 6, 75, 113};
    Mat input(4, &blobSize[0], CV_32F);
    randu(input, 0, 255);

    Mat ref, actual;
    {
        net1.setInput(input);
        ref = net1.forward();
    }
    {
        net2.setInput(input);
        actual = net2.forward();
    }

    normAssert(ref, actual, "", 0, 0);
}

TEST_P(Test_Model_Optimizer, flexible_inputs)
{
    const Backend backendId = get<0>(GetParam());
    const Target targetId = get<1>(GetParam());

    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    const std::string& model = findDataFile("dnn/layers/layer_convolution.bin");
    const std::string& proto = findDataFile("dnn/layers/layer_convolution.xml");

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    Net net0 = readNet(model, proto);
    net0.setPreferableTarget(targetId);

    Net net1 = readNet(model, proto);
    net1.setPreferableTarget(targetId);

    // Generate inputs.
    int blobSize0[] = {2, 6, 75, 113};
    Mat input0(4, &blobSize0[0], CV_32F);
    randu(input0, 0, 255);

    net0.setInput(input0);
    Mat ref = net0.forward().clone();

    int blobSize1[] = {1, 6, 10, 9};
    Mat input1(4, &blobSize1[0], CV_32F);
    randu(input1, 0, 255);

    net1.setInput(input1);
    Mat out = net1.forward();
    EXPECT_NE(out.size, ref.size);

    net1.setInput(input0);
    out = net1.forward();
    normAssert(ref, out, 0, 0);
}

TEST_P(Test_Model_Optimizer, readONNX)
{
    const Backend backendId = get<0>(GetParam());
    const Target targetId = get<1>(GetParam());

    ASSERT_EQ(DNN_BACKEND_INFERENCE_ENGINE_NGRAPH, backendId);

    const std::string& model = findDataFile("dnn/onnx/models/convolution.onnx");

    std::vector<Net> nets = {
        // Old API
        readNetFromModelOptimizer(model, ""),
        readNet("", model, "dldt"),
        // New API
        readNetFromModelOptimizer(model),
        readNet(model, "", "openvino")
    };

    Mat inp = blobFromNPY(findDataFile("dnn/onnx/data/input_convolution.npy"));
    Mat ref = blobFromNPY(findDataFile("dnn/onnx/data/output_convolution.npy"));

    for (int i = 0; i < nets.size(); ++i) {
        nets[i].setPreferableTarget(targetId);
        nets[i].setInput(inp);
        Mat out = nets[i].forward();
        normAssert(out, ref, format("Index: %d", i).c_str());
    }
}

INSTANTIATE_TEST_CASE_P(/**/, Test_Model_Optimizer,
    dnnBackendsAndTargetsIE()
);

#endif  // HAVE_INF_ENGINE

typedef testing::TestWithParam<tuple<MatDepth, MatDepth, tuple<Backend, Target> > > Test_two_inputs;
TEST_P(Test_two_inputs, basic)
{
    static const float kScale = 0.5f;
    static const float kScaleInv = 1.0f / kScale;

    Backend backendId = get<0>(get<2>(GetParam()));
    Target targetId = get<1>(get<2>(GetParam()));

    int type1 = get<0>(GetParam());
    int type2 = get<1>(GetParam());

    if (backendId == DNN_BACKEND_VKCOM && !(type1 == CV_32F && type2 == CV_32F))
        applyTestTag(CV_TEST_TAG_DNN_SKIP_VULKAN);

    Net net;
    LayerParams lp;
    lp.type = "Eltwise";
    lp.name = "testLayer";
    lp.set("operation", "sum");
    int eltwiseId = net.addLayerToPrev(lp.name, lp.type, lp);  // connect to a first input
    net.connect(0, 1, eltwiseId, 1);  // connect to a second input

    int inpSize[] = {1, 2, 3, 4};
    Mat firstInp(4, &inpSize[0], type1);
    Mat secondInp(4, &inpSize[0], type2);
    randu(firstInp, 0, 100);
    randu(secondInp, 0, 100);

#ifndef CV_CXX11
    std::vector<String> input_names;
    input_names.push_back("data");
    input_names.push_back("second_input");
    net.setInputsNames(input_names);
#else
    net.setInputsNames({"data", "second_input"});
#endif
    net.setInput(firstInp, "data", kScale);
    net.setInput(secondInp, "second_input", kScaleInv);
    net.setPreferableBackend(backendId);
    net.setPreferableTarget(targetId);
    Mat out = net.forward();

    Mat ref;
    addWeighted(firstInp, kScale, secondInp, kScaleInv, 0, ref, CV_32F);

    double l1 = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD || targetId == DNN_TARGET_CUDA_FP16) ? 0.06 : 1e-6;
    double lInf = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD || targetId == DNN_TARGET_CUDA_FP16) ? 0.3 : 1e-5;

    normAssert(out, ref, "", l1, lInf);

    if (cvtest::debugLevel > 0 || HasFailure())
    {
        std::cout << "input1 scale=" << kScale << " input2 scale=" << kScaleInv << std::endl;
        std::cout << "input1: " << firstInp.size << " " << firstInp.reshape(1, 1) << std::endl;
        std::cout << "input2: " << secondInp.size << " " << secondInp.reshape(1, 1) << std::endl;
        std::cout << "ref: " << ref.reshape(1, 1) << std::endl;
        std::cout << "out: " << out.reshape(1, 1) << std::endl;
    }
}

INSTANTIATE_TEST_CASE_P(/*nothing*/, Test_two_inputs, Combine(
    Values(CV_32F, CV_8U),
    Values(CV_32F, CV_8U),
    dnnBackendsAndTargets()
));

}} // namespace