opencv/modules/dnn/test/test_onnx_importer.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) 2018-2019, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.


#include "test_precomp.hpp"
#include "npy_blob.hpp"
#include <opencv2/dnn/shape_utils.hpp>
namespace opencv_test { namespace {

template<typename TString>
static std::string _tf(TString filename, bool required = true)
{
    return findDataFile(std::string("dnn/onnx/") + filename, required);
}

class Test_ONNX_layers : public DNNTestLayer
{
public:
    bool required;

    Test_ONNX_layers() : required(true) { }

    enum Extension
    {
        npy,
        pb
    };

    void testONNXModels(const String& basename, const Extension ext = npy,
                        const double l1 = 0, const float lInf = 0, const bool useSoftmax = false,
                        bool checkNoFallbacks = true, int numInps = 1)
    {
        String onnxmodel = _tf("models/" + basename + ".onnx", required);
        std::vector<Mat> inps(numInps);
        Mat ref;
        if (ext == npy) {
            for (int i = 0; i < numInps; ++i)
                inps[i] = blobFromNPY(_tf("data/input_" + basename + (numInps > 1 ? format("_%d", i) : "") + ".npy"));
            ref = blobFromNPY(_tf("data/output_" + basename + ".npy"));
        }
        else if (ext == pb) {
            for (int i = 0; i < numInps; ++i)
                inps[i] = readTensorFromONNX(_tf("data/input_" + basename + (numInps > 1 ? format("_%d", i) : "") + ".pb"));
            ref = readTensorFromONNX(_tf("data/output_" + basename + ".pb"));
        }
        else
            CV_Error(Error::StsUnsupportedFormat, "Unsupported extension");

        checkBackend(&inps[0], &ref);
        Net net = readNetFromONNX(onnxmodel);
        ASSERT_FALSE(net.empty());

        net.setPreferableBackend(backend);
        net.setPreferableTarget(target);

        std::vector<String> inputNames;
        for (int i = 0; i < numInps; ++i)
            inputNames.push_back(format("%d", i));
        net.setInputsNames(inputNames);

        for (int i = 0; i < numInps; ++i)
            net.setInput(inps[i], inputNames[i]);
        Mat out = net.forward("");

        if (useSoftmax)
        {
            LayerParams lp;
            Net netSoftmax;
            netSoftmax.addLayerToPrev("softmaxLayer", "Softmax", lp);
            netSoftmax.setPreferableBackend(DNN_BACKEND_OPENCV);

            netSoftmax.setInput(out);
            out = netSoftmax.forward();

            netSoftmax.setInput(ref);
            ref = netSoftmax.forward();
        }
        normAssert(ref, out, "", l1 ? l1 : default_l1, lInf ? lInf : default_lInf);
        if (checkNoFallbacks)
            expectNoFallbacksFromIE(net);
    }
};

TEST_P(Test_ONNX_layers, InstanceNorm)
{
    if(backend == DNN_BACKEND_CUDA)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); /* MVN is not supported */

    if (target == DNN_TARGET_MYRIAD)
        testONNXModels("instancenorm", npy, 0, 0, false, false);
    else
        testONNXModels("instancenorm", npy);
}

TEST_P(Test_ONNX_layers, MaxPooling)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020020000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    testONNXModels("maxpooling", npy, 0, 0, false, false);
}
TEST_P(Test_ONNX_layers, MaxPooling_2)
{
    testONNXModels("two_maxpooling", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, Convolution)
{
    testONNXModels("convolution");
}

TEST_P(Test_ONNX_layers, Convolution_variable_weight)
{
    if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH ||
         backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019) && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);

    String basename = "conv_variable_w";
    Net net = readNetFromONNX(_tf("models/" + basename + ".onnx"));
    ASSERT_FALSE(net.empty());

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    for (int i = 0; i < 2; i++)
    {
        Mat input = blobFromNPY(_tf("data/input_" + basename + format("_%d", i) + "_0.npy"));
        Mat weights = blobFromNPY(_tf("data/input_" + basename + format("_%d", i) + "_1.npy"));
        Mat ref = blobFromNPY(_tf("data/output_" + basename + format("_%d", i) + ".npy"));

        net.setInput(input, "0");
        net.setInput(weights, "1");

        Mat out = net.forward();
        normAssert(ref, out, "", default_l1, default_lInf);
    }
}

TEST_P(Test_ONNX_layers, Convolution_variable_weight_bias)
{
    if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH ||
         backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019) && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);

    String basename = "conv_variable_wb";
    Net net = readNetFromONNX(_tf("models/" + basename + ".onnx"));
    ASSERT_FALSE(net.empty());

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    for (int i = 0; i < 2; i++)
    {
        Mat input = blobFromNPY(_tf("data/input_" + basename + format("_%d", i) + "_0.npy"));
        Mat weights = blobFromNPY(_tf("data/input_" + basename + format("_%d", i) + "_1.npy"));
        Mat bias = blobFromNPY(_tf("data/input_" + basename + format("_%d", i) + "_2.npy"));
        Mat ref = blobFromNPY(_tf("data/output_" + basename + format("_%d", i) + ".npy"));

        net.setInput(input, "0");
        net.setInput(weights, "1");
        net.setInput(bias, "bias");

        Mat out = net.forward();
        normAssert(ref, out, "", default_l1, default_lInf);
    }
}

TEST_P(Test_ONNX_layers, Gather)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("gather");
    // GPU plugin unsupported slice for constant
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
    testONNXModels("gather_scalar", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, Convolution3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (target != DNN_TARGET_CPU && backend != DNN_BACKEND_CUDA)
        throw SkipTestException("Only CPU and CUDA is supported");
    testONNXModels("conv3d");
    testONNXModels("conv3d_bias");
}

TEST_P(Test_ONNX_layers, Two_convolution)
{
#if defined(INF_ENGINE_RELEASE)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD
        && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X
    )
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
#endif
    // Reference output values are in range [-0.855, 0.611]
    testONNXModels("two_convolution");
}

TEST_P(Test_ONNX_layers, Deconvolution)
{
    testONNXModels("deconvolution", npy, 0, 0, false, false);
    testONNXModels("two_deconvolution", npy, 0, 0, false, false);
    testONNXModels("deconvolution_group", npy, 0, 0, false, false);
    testONNXModels("deconvolution_output_shape", npy, 0, 0, false, false);
    testONNXModels("deconv_adjpad_2d", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, Deconvolution3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2018050000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_OPENCV || target != DNN_TARGET_CPU)
        throw SkipTestException("Only DLIE backend on CPU is supported");
    testONNXModels("deconv3d");
    testONNXModels("deconv3d_bias");
    testONNXModels("deconv3d_pad");
    testONNXModels("deconv3d_adjpad");
}

TEST_P(Test_ONNX_layers, Dropout)
{
    testONNXModels("dropout");
}

TEST_P(Test_ONNX_layers, Linear)
{
    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
    testONNXModels("linear");
}

TEST_P(Test_ONNX_layers, ReLU)
{
    testONNXModels("ReLU");
}

TEST_P(Test_ONNX_layers, Clip)
{
    testONNXModels("clip", npy);
}

TEST_P(Test_ONNX_layers, Shape)
{
    testONNXModels("shape_of_constant");
}

TEST_P(Test_ONNX_layers, ReduceMean)
{
    testONNXModels("reduce_mean");
    testONNXModels("reduce_mean_axis1");
    testONNXModels("reduce_mean_axis2");
}

TEST_P(Test_ONNX_layers, ReduceMean3D)
{
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    else if (target != DNN_TARGET_CPU)
        throw SkipTestException("Only CPU is supported");

    testONNXModels("reduce_mean3d");
}

TEST_P(Test_ONNX_layers, MaxPooling_Sigmoid)
{
    testONNXModels("maxpooling_sigmoid");
}

TEST_P(Test_ONNX_layers, Cast)
{
    testONNXModels("cast");
}

TEST_P(Test_ONNX_layers, Concatenation)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    }
    testONNXModels("concatenation");
}

TEST_P(Test_ONNX_layers, Eltwise3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    testONNXModels("eltwise3d");
}

TEST_P(Test_ONNX_layers, AveragePooling)
{
    testONNXModels("average_pooling");
}

TEST_P(Test_ONNX_layers, MaxPooling3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    else if (target != DNN_TARGET_CPU)
        throw SkipTestException("Only CPU is supported");
    testONNXModels("max_pool3d", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, AvePooling3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    else if (target != DNN_TARGET_CPU)
        throw SkipTestException("Only CPU is supported");
    testONNXModels("ave_pool3d");
}

TEST_P(Test_ONNX_layers, PoolConv3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    else if (target != DNN_TARGET_CPU)
        throw SkipTestException("Only CPU is supported");
    testONNXModels("pool_conv_3d");
}

TEST_P(Test_ONNX_layers, BatchNormalization)
{
    testONNXModels("batch_norm");
}

TEST_P(Test_ONNX_layers, BatchNormalization3D)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    }
    testONNXModels("batch_norm_3d");
}

TEST_P(Test_ONNX_layers, BatchNormalizationUnfused)
{
    testONNXModels("frozenBatchNorm2d");
}

TEST_P(Test_ONNX_layers, BatchNormalizationSubgraph)
{
    testONNXModels("batch_norm_subgraph");
}

TEST_P(Test_ONNX_layers, Transpose)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    }
    testONNXModels("transpose");
}

TEST_P(Test_ONNX_layers, Multiplication)
{
    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("mul");
}

TEST_P(Test_ONNX_layers, MatMul)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    if (backend == DNN_BACKEND_CUDA)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); // not supported

    testONNXModels("matmul_2d");
    testONNXModels("matmul_3d");
    testONNXModels("matmul_4d");
}

TEST_P(Test_ONNX_layers, Expand)
{
    testONNXModels("expand_batch");
    testONNXModels("expand_channels");
}

TEST_P(Test_ONNX_layers, ExpandHW)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("expand_hw");
}

TEST_P(Test_ONNX_layers, Constant)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2018050000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD
            && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
       applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    testONNXModels("constant");
}

TEST_P(Test_ONNX_layers, Padding)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    testONNXModels("padding", npy, 0, 0, false, false);
#else
    testONNXModels("padding");
#endif
}

TEST_P(Test_ONNX_layers, Resize)
{
    testONNXModels("resize_nearest");
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("resize_bilinear");
}

TEST_P(Test_ONNX_layers, ResizeUnfused)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("upsample_unfused_torch1.2");
    testONNXModels("upsample_unfused_opset9_torch1.4");
    testONNXModels("resize_nearest_unfused_opset11_torch1.4");
    testONNXModels("resize_nearest_unfused_opset11_torch1.3");
    testONNXModels("resize_bilinear_unfused_opset11_torch1.4");
}

TEST_P(Test_ONNX_layers, ResizeUnfusedTwoInputs)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
    testONNXModels("upsample_unfused_two_inputs_opset9_torch1.4", npy, 0, 0, false, true, 2);
    testONNXModels("upsample_unfused_two_inputs_opset11_torch1.4", npy, 0, 0, false, true, 2);
}

TEST_P(Test_ONNX_layers, MultyInputs)
{
    testONNXModels("multy_inputs", npy, 0, 0, false, true, 2);
}

TEST_P(Test_ONNX_layers, Broadcast)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("channel_broadcast", npy, 0, 0, false, true, 2);
}

TEST_P(Test_ONNX_layers, DynamicResize)
{
    testONNXModels("dynamic_resize", npy, 0, 0, false, true, 2);
}

TEST_P(Test_ONNX_layers, Div)
{
    const String model =  _tf("models/div.onnx");
    Net net = readNetFromONNX(model);
    ASSERT_FALSE(net.empty());

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    // Reference output values range is -68.80928, 2.991873. So to avoid computational
    // difference for FP16 we'll perform reversed division (just swap inputs).
    Mat inp1 = blobFromNPY(_tf("data/input_div_1.npy"));
    Mat inp2 = blobFromNPY(_tf("data/input_div_0.npy"));
    Mat ref  = blobFromNPY(_tf("data/output_div.npy"));
    cv::divide(1.0, ref, ref);
    checkBackend(&inp1, &ref);

    net.setInput(inp1, "0");
    net.setInput(inp2, "1");
    Mat out = net.forward();

    normAssert(ref, out, "", default_l1,  default_lInf);
    expectNoFallbacksFromIE(net);
    expectNoFallbacksFromCUDA(net);
}

TEST_P(Test_ONNX_layers, DynamicReshape)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    testONNXModels("dynamic_reshape");
    testONNXModels("dynamic_reshape_opset_11");
    testONNXModels("flatten_by_prod");
    testONNXModels("flatten_const");
}

TEST_P(Test_ONNX_layers, Reshape)
{
    testONNXModels("unsqueeze");
}

TEST_P(Test_ONNX_layers, Squeeze)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    testONNXModels("squeeze");
}

TEST_P(Test_ONNX_layers, ReduceL2)
{
    testONNXModels("reduceL2");
    testONNXModels("reduceL2_subgraph");
    testONNXModels("reduceL2_subgraph_2");
}

TEST_P(Test_ONNX_layers, Split)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
    testONNXModels("split_1");
    testONNXModels("split_2");
    testONNXModels("split_3");
    testONNXModels("split_4");
}

TEST_P(Test_ONNX_layers, Slice)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    testONNXModels("slice", npy, 0, 0, false, false);
#else
    testONNXModels("slice");
    testONNXModels("slice_opset_11");
#endif
}

TEST_P(Test_ONNX_layers, Softmax)
{
    testONNXModels("softmax");
    testONNXModels("log_softmax", npy, 0, 0, false, false);
    testONNXModels("softmax_unfused");
}

TEST_P(Test_ONNX_layers, Split_EltwiseMax)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
    testONNXModels("split_max");
}

TEST_P(Test_ONNX_layers, LSTM)
{
    testONNXModels("lstm", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, LSTM_bidirectional)
{
    testONNXModels("lstm_bidirectional", npy, 0, 0, false, false);
}

TEST_P(Test_ONNX_layers, Pad2d_Unfused)
{
    testONNXModels("ReflectionPad2d");
    testONNXModels("ZeroPad2d");
}

INSTANTIATE_TEST_CASE_P(/*nothing*/, Test_ONNX_layers, dnnBackendsAndTargets());

class Test_ONNX_nets : public Test_ONNX_layers
{
public:
    Test_ONNX_nets() { required = false; }
};

TEST_P(Test_ONNX_nets, Alexnet)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
    applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
    applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);
#endif

    const String model =  _tf("models/alexnet.onnx", false);

    Net net = readNetFromONNX(model);
    ASSERT_FALSE(net.empty());

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    Mat inp = imread(_tf("../grace_hopper_227.png"));
    Mat ref = blobFromNPY(_tf("../caffe_alexnet_prob.npy"));
    checkBackend(&inp, &ref);

    net.setInput(blobFromImage(inp, 1.0f, Size(227, 227), Scalar(), false));
    ASSERT_FALSE(net.empty());
    Mat out = net.forward();

    normAssert(out, ref, "", default_l1,  default_lInf);
    expectNoFallbacksFromIE(net);
}

TEST_P(Test_ONNX_nets, Squeezenet)
{
    testONNXModels("squeezenet", pb);
}

TEST_P(Test_ONNX_nets, Googlenet)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);

    const String model = _tf("models/googlenet.onnx", false);

    Net net = readNetFromONNX(model);
    ASSERT_FALSE(net.empty());

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    std::vector<Mat> images;
    images.push_back( imread(_tf("../googlenet_0.png")) );
    images.push_back( imread(_tf("../googlenet_1.png")) );
    Mat inp = blobFromImages(images, 1.0f, Size(), Scalar(), false);
    Mat ref = blobFromNPY(_tf("../googlenet_prob.npy"));
    checkBackend(&inp, &ref);

    net.setInput(inp);
    ASSERT_FALSE(net.empty());
    Mat out = net.forward();

    normAssert(ref, out, "", default_l1,  default_lInf);
    expectNoFallbacksFromIE(net);
}

TEST_P(Test_ONNX_nets, CaffeNet)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
    applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
    applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);
#endif

#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2019030000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD
        && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    testONNXModels("caffenet", pb);
}

TEST_P(Test_ONNX_nets, RCNN_ILSVRC13)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
    applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
    applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);
#endif

#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2019030000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD
        && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    // Reference output values are in range [-4.992, -1.161]
    testONNXModels("rcnn_ilsvrc13", pb, 0.0046);
}

TEST_P(Test_ONNX_nets, VGG16_bn)
{
    applyTestTag(CV_TEST_TAG_MEMORY_6GB);  // > 2.3Gb

    // output range: [-16; 27], after Softmax [0; 0.67]
    const double lInf = (target == DNN_TARGET_MYRIAD) ? 0.038 : default_lInf;
    testONNXModels("vgg16-bn", pb, default_l1, lInf, true);
}

TEST_P(Test_ONNX_nets, ZFNet)
{
    applyTestTag(CV_TEST_TAG_MEMORY_2GB);
    testONNXModels("zfnet512", pb);
}

TEST_P(Test_ONNX_nets, ResNet18v1)
{
    applyTestTag(CV_TEST_TAG_MEMORY_512MB);

    // output range: [-16; 22], after Softmax [0, 0.51]
    testONNXModels("resnet18v1", pb, default_l1, default_lInf, true, target != DNN_TARGET_MYRIAD);
}

TEST_P(Test_ONNX_nets, ResNet50v1)
{
    applyTestTag(CV_TEST_TAG_MEMORY_512MB);

    // output range: [-67; 75], after Softmax [0, 0.98]
    testONNXModels("resnet50v1", pb, default_l1, default_lInf, true, target != DNN_TARGET_MYRIAD);
}

TEST_P(Test_ONNX_nets, ResNet101_DUC_HDC)
{
    applyTestTag(CV_TEST_TAG_VERYLONG);

#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2019010000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
#if defined(INF_ENGINE_RELEASE)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
#endif
    if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_OPENCL)
    {
        if (backend == DNN_BACKEND_OPENCV)
            applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_OPENCL : CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
        throw SkipTestException("Test is disabled for OpenCL targets");
    }
    testONNXModels("resnet101_duc_hdc", pb);
}

TEST_P(Test_ONNX_nets, TinyYolov2)
{
    applyTestTag(CV_TEST_TAG_MEMORY_512MB);

    if (cvtest::skipUnstableTests)
        throw SkipTestException("Skip unstable test");
#if defined(INF_ENGINE_RELEASE)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019
            && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16)
    )
        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);

    if (target == DNN_TARGET_MYRIAD && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X
    )
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X,
                     backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ?
                     CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER :
                     CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
#endif

    // output range: [-11; 8]
    double l1 =  default_l1, lInf = default_lInf;
    if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
    {
        l1 = 0.017;
        lInf = 0.14;
    }
    else if (target == DNN_TARGET_CUDA_FP16)
    {
        l1 = 0.018;
        lInf = 0.16;
    }
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020040000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
    {
        l1 = 0.018f; lInf = 0.16f;
    }
#endif

    testONNXModels("tiny_yolo2", pb, l1, lInf);
}

TEST_P(Test_ONNX_nets, CNN_MNIST)
{
    // output range: [-1952; 6574], after Softmax [0; 1]
    testONNXModels("cnn_mnist", pb, default_l1, default_lInf, true);
}

TEST_P(Test_ONNX_nets, MobileNet_v2)
{
    // output range: [-166; 317], after Softmax [0; 1]
    testONNXModels("mobilenetv2", pb, default_l1, default_lInf, true);
}

TEST_P(Test_ONNX_nets, LResNet100E_IR)
{
    applyTestTag(
#if defined(OPENCV_32BIT_CONFIGURATION) && defined(HAVE_OPENCL)
        CV_TEST_TAG_MEMORY_2GB,
#else
        (target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB),
#endif
        CV_TEST_TAG_DEBUG_LONG
    );
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    }
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
    }

    double l1 = default_l1, lInf = default_lInf;
    // output range: [-3; 3]
    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
    {
        l1 = 0.009;
        lInf = 0.035;
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_CPU)
    {
        l1 = 4.6e-5;
        lInf = 1.9e-4;
    }
    else if (target == DNN_TARGET_CUDA_FP16)
    {
        l1 = 0.008;
        lInf = 0.04;
    }
    testONNXModels("LResNet100E_IR", pb, l1, lInf);
}

TEST_P(Test_ONNX_nets, Emotion_ferplus)
{
#if defined(INF_ENGINE_RELEASE)
    if (target == DNN_TARGET_MYRIAD && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X,
                     backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ?
                     CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER :
                     CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
#endif

    double l1 = default_l1;
    double lInf = default_lInf;

    // Output values are in range [-2.011, 2.111]
    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
        l1 = 0.007;
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_OPENCL_FP16)
    {
        l1 = 0.021;
        lInf = 0.034;
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && (target == DNN_TARGET_CPU || target == DNN_TARGET_OPENCL)) {
        l1 = 2.4e-4;
        lInf = 6e-4;
    }
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020040000)
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
    {
        l1 = 0.012f; lInf = 0.035f;
    }
#endif

    testONNXModels("emotion_ferplus", pb, l1, lInf);
}

TEST_P(Test_ONNX_nets, Inception_v2)
{
    testONNXModels("inception_v2", pb, default_l1, default_lInf, true);
}

TEST_P(Test_ONNX_nets, DenseNet121)
{
    applyTestTag(CV_TEST_TAG_MEMORY_512MB);

    // output range: [-87; 138], after Softmax [0; 1]
    testONNXModels("densenet121", pb, default_l1, default_lInf, true, target != DNN_TARGET_MYRIAD);
}

TEST_P(Test_ONNX_nets, Inception_v1)
{
#if defined(INF_ENGINE_RELEASE)
    if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
         backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_MYRIAD)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD);
#endif
    testONNXModels("inception_v1", pb);
}

TEST_P(Test_ONNX_nets, Shufflenet)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
    {
        if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_OPENCL)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
        if (target == DNN_TARGET_MYRIAD)      applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
    }
    testONNXModels("shufflenet", pb);
}

TEST_P(Test_ONNX_nets, Resnet34_kinetics)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
    applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
    if (backend == DNN_BACKEND_CUDA)
    {
        // ok
    }
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);  // Only CPU on DLIE backend is supported
    else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // Only CPU on DLIE backend is supported
    else if (target != DNN_TARGET_CPU)
        throw SkipTestException("Only CPU is supported");

    String onnxmodel = findDataFile("dnn/resnet-34_kinetics.onnx", false);
    Mat image0 = imread(findDataFile("dnn/dog416.png"));
    Mat image1 = imread(findDataFile("dnn/street.png"));

    Mat ref0 = blobFromNPY(_tf("data/output_kinetics0.npy"));
    Mat ref1 = blobFromNPY(_tf("data/output_kinetics1.npy"));

    std::vector<Mat> images_0(16, image0);
    std::vector<Mat> images_1(16, image1);
    Mat blob0 = blobFromImages(images_0, 1.0, Size(112, 112), Scalar(114.7748, 107.7354, 99.4750), true, true);
    Mat blob1 = blobFromImages(images_1, 1.0, Size(112, 112), Scalar(114.7748, 107.7354, 99.4750), true, true);

    Net permute;
    LayerParams lp;
    int order[] = {1, 0, 2, 3};
    lp.set("order", DictValue::arrayInt<int*>(&order[0], 4));
    permute.addLayerToPrev("perm", "Permute", lp);

    permute.setPreferableBackend(backend);
    permute.setPreferableTarget(target);

    permute.setInput(blob0);
    Mat input0 = permute.forward().clone();

    permute.setInput(blob1);
    Mat input1 = permute.forward().clone();

    int dims[] = {1, 3, 16, 112, 112};
    input0 = input0.reshape(0, 5, &dims[0]);
    input1 = input1.reshape(0, 5, &dims[0]);

    Net net = readNetFromONNX(onnxmodel);
    ASSERT_FALSE(net.empty());
    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);

    // output range [-5, 11]
    float l1 = 0.0013, lInf = 0.009;
    if (target == DNN_TARGET_CUDA_FP16)
    {
        l1 = 0.008;
        lInf = 0.04;
    }

    checkBackend(&input0, &ref0);
    net.setInput(input0);
    Mat out = net.forward().clone();
    normAssert(ref0, out, "", l1, lInf);

    checkBackend(&input1, &ref1);
    net.setInput(input1);
    out = net.forward().clone();
    normAssert(ref1, out, "", l1, lInf);

    expectNoFallbacksFromIE(net);
}

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

}} // namespace