C++ Net::forward方法代码示例

TEST_P(Reproducibility_SqueezeNet_v1_1, Accuracy)
{
    Net net = readNetFromCaffe(findDataFile("dnn/squeezenet_v1.1.prototxt", false),
                               findDataFile("dnn/squeezenet_v1.1.caffemodel", false));

    int targetId = GetParam();
    net.setPreferableTarget(targetId);

    Mat input = blobFromImage(imread(_tf("googlenet_0.png")), 1.0f, Size(227,227), Scalar(), false);
    ASSERT_TRUE(!input.empty());

    Mat out;
    if (targetId == DNN_TARGET_OPENCL)
    {
        // Firstly set a wrong input blob and run the model to receive a wrong output.
        // Then set a correct input blob to check CPU->GPU synchronization is working well.
        net.setInput(input * 2.0f);
        out = net.forward();
    }
    net.setInput(input);
    out = net.forward();

    Mat ref = blobFromNPY(_tf("squeezenet_v1.1_prob.npy"));
    normAssert(ref, out);
}

TEST(Reproducibility_MobileNet_SSD, Accuracy)
{
    const string proto = findDataFile("dnn/MobileNetSSD_deploy.prototxt", false);
    const string model = findDataFile("dnn/MobileNetSSD_deploy.caffemodel", false);
    Net net = readNetFromCaffe(proto, model);

    Mat sample = imread(_tf("street.png"));

    Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
    net.setInput(inp);
    Mat out = net.forward();

    Mat ref = blobFromNPY(_tf("mobilenet_ssd_caffe_out.npy"));
    normAssert(ref, out);

    // Check that detections aren't preserved.
    inp.setTo(0.0f);
    net.setInput(inp);
    out = net.forward();

    const int numDetections = out.size[2];
    ASSERT_NE(numDetections, 0);
    for (int i = 0; i < numDetections; ++i)
    {
        float confidence = out.ptr<float>(0, 0, i)[2];
        ASSERT_EQ(confidence, 0);
    }
}

TEST(Test_Darknet, read_yolo_voc_stream)
{
    Mat ref;
    Mat sample = imread(_tf("dog416.png"));
    Mat inp = blobFromImage(sample, 1.0/255, Size(416, 416), Scalar(), true, false);
    const std::string cfgFile = findDataFile("dnn/yolo-voc.cfg", false);
    const std::string weightsFile = findDataFile("dnn/yolo-voc.weights", false);
    // Import by paths.
    {
        Net net = readNetFromDarknet(cfgFile, weightsFile);
        net.setInput(inp);
        net.setPreferableBackend(DNN_BACKEND_OPENCV);
        ref = net.forward();
    }
    // Import from bytes array.
    {
        std::string cfg, weights;
        readFileInMemory(cfgFile, cfg);
        readFileInMemory(weightsFile, weights);

        Net net = readNetFromDarknet(&cfg[0], cfg.size(), &weights[0], weights.size());
        net.setInput(inp);
        net.setPreferableBackend(DNN_BACKEND_OPENCV);
        Mat out = net.forward();
        normAssert(ref, out);
    }
}

void testInPlaceActivation(LayerParams& lp)
{
    EXPECT_FALSE(lp.name.empty());

    LayerParams pool;
    pool.set("pool", "ave");
    pool.set("kernel_w", 2);
    pool.set("kernel_h", 2);
    pool.set("stride_w", 2);
    pool.set("stride_h", 2);
    pool.type = "Pooling";

    Net net;
    int poolId = net.addLayer(pool.name, pool.type, pool);
    net.connect(0, 0, poolId, 0);
    net.addLayerToPrev(lp.name, lp.type, lp);

    Mat input({1, kNumChannels, 10, 10}, CV_32F);
    randu(input, -1.0f, 1.0f);
    net.setInput(input);
    Mat outputDefault = net.forward(lp.name).clone();

    net.setInput(input);
    net.setPreferableBackend(DNN_BACKEND_HALIDE);
    Mat outputHalide = net.forward(lp.name).clone();
    normAssert(outputDefault, outputHalide);
}

TEST(Torch_Importer, ENet_accuracy)
{
    Net net;
    {
        const string model = findDataFile("dnn/Enet-model-best.net", false);
        net = readNetFromTorch(model, true);
        ASSERT_FALSE(net.empty());
    }

    Mat sample = imread(_tf("street.png", false));
    Mat inputBlob = blobFromImage(sample, 1./255);

    net.setInput(inputBlob, "");
    Mat out = net.forward();
    Mat ref = blobFromNPY(_tf("torch_enet_prob.npy", false));
    // Due to numerical instability in Pooling-Unpooling layers (indexes jittering)
    // thresholds for ENet must be changed. Accuracy of resuults was checked on
    // Cityscapes dataset and difference in mIOU with Torch is 10E-4%
    normAssert(ref, out, "", 0.00044, 0.44);

    const int N = 3;
    for (int i = 0; i < N; i++)
    {
        net.setInput(inputBlob, "");
        Mat out = net.forward();
        normAssert(ref, out, "", 0.00044, 0.44);
    }
}

TEST_P(Concat, Accuracy)
{
    Vec3i inSize = get<0>(GetParam());
    Vec3i numChannels = get<1>(GetParam());

    Net net;

    std::vector<int> convLayerIds;
    convLayerIds.reserve(numChannels.channels);
    for (int i = 0, n = numChannels.channels; i < n; ++i)
    {
        if (!numChannels[i])
            break;

        Mat weights({numChannels[i], inSize[0], 1, 1}, CV_32F);
        randu(weights, -1.0f, 1.0f);

        LayerParams convParam;
        convParam.set("kernel_w", 1);
        convParam.set("kernel_h", 1);
        convParam.set("num_output", numChannels[i]);
        convParam.set("bias_term", false);
        convParam.type = "Convolution";
        std::ostringstream ss;
        ss << "convLayer" << i;
        convParam.name = ss.str();
        convParam.blobs.push_back(weights);

        int layerId = net.addLayer(convParam.name, convParam.type, convParam);
        convLayerIds.push_back(layerId);
        net.connect(0, 0, layerId, 0);
    }

    LayerParams concatParam;
    concatParam.type = "Concat";
    concatParam.name = "testLayer";
    int concatId = net.addLayer(concatParam.name, concatParam.type, concatParam);
    net.connect(0, 0, concatId, 0);
    for (int i = 0; i < convLayerIds.size(); ++i)
    {
        net.connect(convLayerIds[i], 0, concatId, i + 1);
    }

    Mat input({1, inSize[0], inSize[1], inSize[2]}, CV_32F);
    randu(input, -1.0f, 1.0f);

    net.setInput(input);
    Mat outputDefault = net.forward(concatParam.name).clone();

    net.setPreferableBackend(DNN_BACKEND_HALIDE);
    Mat outputHalide = net.forward(concatParam.name).clone();
    normAssert(outputDefault, outputHalide);
}

TEST_P(Reproducibility_MobileNet_SSD, Accuracy)
{
    const string proto = findDataFile("dnn/MobileNetSSD_deploy.prototxt", false);
    const string model = findDataFile("dnn/MobileNetSSD_deploy.caffemodel", false);
    Net net = readNetFromCaffe(proto, model);
    int targetId = GetParam();
    const float l1 = (targetId == DNN_TARGET_OPENCL_FP16) ? 1.5e-4 : 1e-5;
    const float lInf = (targetId == DNN_TARGET_OPENCL_FP16) ? 4e-4 : 1e-4;

    net.setPreferableTarget(targetId);

    Mat sample = imread(_tf("street.png"));

    Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
    net.setInput(inp);
    Mat out = net.forward();

    const float scores_diff = (targetId == DNN_TARGET_OPENCL_FP16) ? 4e-4 : 1e-5;
    const float boxes_iou_diff = (targetId == DNN_TARGET_OPENCL_FP16) ? 5e-3 : 1e-4;
    Mat ref = blobFromNPY(_tf("mobilenet_ssd_caffe_out.npy"));
    normAssertDetections(ref, out, "", 0.0, scores_diff, boxes_iou_diff);

    // Check that detections aren't preserved.
    inp.setTo(0.0f);
    net.setInput(inp);
    out = net.forward();
    out = out.reshape(1, out.total() / 7);

    const int numDetections = out.rows;
    ASSERT_NE(numDetections, 0);
    for (int i = 0; i < numDetections; ++i)
    {
        float confidence = out.ptr<float>(i)[2];
        ASSERT_EQ(confidence, 0);
    }

    // Check batching mode.
    ref = ref.reshape(1, numDetections);
    inp = blobFromImages(std::vector<Mat>(2, sample), 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
    net.setInput(inp);
    Mat outBatch = net.forward();

    // Output blob has a shape 1x1x2Nx7 where N is a number of detection for
    // a single sample in batch. The first numbers of detection vectors are batch id.
    outBatch = outBatch.reshape(1, outBatch.total() / 7);
    EXPECT_EQ(outBatch.rows, 2 * numDetections);
    normAssert(outBatch.rowRange(0, numDetections), ref, "", l1, lInf);
    normAssert(outBatch.rowRange(numDetections, 2 * numDetections).colRange(1, 7), ref.colRange(1, 7),
               "", l1, lInf);
}

static void test(LayerParams& params, Mat& input)
{
    randu(input, -1.0f, 1.0f);

    Net net;
    int lid = net.addLayer(params.name, params.type, params);
    net.connect(0, 0, lid, 0);

    net.setInput(input);
    Mat outputDefault = net.forward(params.name).clone();

    net.setPreferableBackend(DNN_BACKEND_HALIDE);
    Mat outputHalide = net.forward(params.name).clone();
    normAssert(outputDefault, outputHalide);
}

TEST_P(Test_ONNX_nets, Googlenet)
{
    if (backend == DNN_BACKEND_INFERENCE_ENGINE)
        throw SkipTestException("");

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

    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);
}

    void runTorchNet(const String& prefix, String outLayerName = "",
                     bool check2ndBlob = false, bool isBinary = false,
                     double l1 = 0.0, double lInf = 0.0)
    {
        String suffix = (isBinary) ? ".dat" : ".txt";

        Mat inp, outRef;
        ASSERT_NO_THROW( inp = readTorchBlob(_tf(prefix + "_input" + suffix), isBinary) );
        ASSERT_NO_THROW( outRef = readTorchBlob(_tf(prefix + "_output" + suffix), isBinary) );

        checkBackend(backend, target, &inp, &outRef);

        Net net = readNetFromTorch(_tf(prefix + "_net" + suffix), isBinary);
        ASSERT_FALSE(net.empty());

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

        if (outLayerName.empty())
            outLayerName = net.getLayerNames().back();

        net.setInput(inp);
        std::vector<Mat> outBlobs;
        net.forward(outBlobs, outLayerName);
        l1 = l1 ? l1 : default_l1;
        lInf = lInf ? lInf : default_lInf;
        normAssert(outRef, outBlobs[0], "", l1, lInf);

        if (check2ndBlob && backend != DNN_BACKEND_INFERENCE_ENGINE)
        {
            Mat out2 = outBlobs[1];
            Mat ref2 = readTorchBlob(_tf(prefix + "_output_2" + suffix), isBinary);
            normAssert(out2, ref2, "", l1, lInf);
        }
    }

TEST_P(Test_Torch_nets, OpenFace_accuracy)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_RELEASE < 2018030000
    if (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
        throw SkipTestException("Test is enabled starts from OpenVINO 2018R3");
#endif
    checkBackend();
    if (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL_FP16)
        throw SkipTestException("");

    const string model = findDataFile("dnn/openface_nn4.small2.v1.t7", false);
    Net net = readNetFromTorch(model);

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

    Mat sample = imread(findDataFile("cv/shared/lena.png", false));
    Mat sampleF32(sample.size(), CV_32FC3);
    sample.convertTo(sampleF32, sampleF32.type());
    sampleF32 /= 255;
    resize(sampleF32, sampleF32, Size(96, 96), 0, 0, INTER_NEAREST);

    Mat inputBlob = blobFromImage(sampleF32, 1.0, Size(), Scalar(), /*swapRB*/true);

    net.setInput(inputBlob);
    Mat out = net.forward();

    Mat outRef = readTorchBlob(_tf("net_openface_output.dat"), true);
    normAssert(out, outRef, "", default_l1, default_lInf);
}

TEST_P(Test_TensorFlow_nets, opencv_face_detector_uint8)
{
    checkBackend();
    std::string proto = findDataFile("dnn/opencv_face_detector.pbtxt", false);
    std::string model = findDataFile("dnn/opencv_face_detector_uint8.pb", false);

    Net net = readNetFromTensorflow(model, proto);
    Mat img = imread(findDataFile("gpu/lbpcascade/er.png", false));
    Mat blob = blobFromImage(img, 1.0, Size(), Scalar(104.0, 177.0, 123.0), false, false);

    net.setPreferableBackend(backend);
    net.setPreferableTarget(target);
    net.setInput(blob);
    // Output has shape 1x1xNx7 where N - number of detections.
    // An every detection is a vector of values [id, classId, confidence, left, top, right, bottom]
    Mat out = net.forward();

    // References are from test for Caffe model.
    Mat ref = (Mat_<float>(6, 7) << 0, 1, 0.99520785, 0.80997437, 0.16379407, 0.87996572, 0.26685631,
                                    0, 1, 0.9934696, 0.2831718, 0.50738752, 0.345781, 0.5985168,
                                    0, 1, 0.99096733, 0.13629119, 0.24892329, 0.19756334, 0.3310290,
                                    0, 1, 0.98977017, 0.23901358, 0.09084064, 0.29902688, 0.1769477,
                                    0, 1, 0.97203469, 0.67965847, 0.06876482, 0.73999709, 0.1513494,
                                    0, 1, 0.95097077, 0.51901293, 0.45863652, 0.5777427, 0.5347801);
    double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 4e-3 : 3.4e-3;
    double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.024 : 1e-2;
    normAssertDetections(ref, out, "", 0.9, scoreDiff, iouDiff);
}

TEST_P(Test_TensorFlow_nets, MobileNet_SSD)
{
    std::string netPath = findDataFile("dnn/ssd_mobilenet_v1_coco.pb", false);
    std::string netConfig = findDataFile("dnn/ssd_mobilenet_v1_coco.pbtxt", false);
    std::string imgPath = findDataFile("dnn/street.png", false);

    Mat inp;
    resize(imread(imgPath), inp, Size(300, 300));
    inp = blobFromImage(inp, 1.0f / 127.5, Size(), Scalar(127.5, 127.5, 127.5), true);

    std::vector<String> outNames(3);
    outNames[0] = "concat";
    outNames[1] = "concat_1";
    outNames[2] = "detection_out";

    std::vector<Mat> target(outNames.size());
    for (int i = 0; i < outNames.size(); ++i)
    {
        std::string path = findDataFile("dnn/tensorflow/ssd_mobilenet_v1_coco." + outNames[i] + ".npy", false);
        target[i] = blobFromNPY(path);
    }

    Net net = readNetFromTensorflow(netPath, netConfig);

    net.setPreferableTarget(GetParam());

    net.setInput(inp);

    std::vector<Mat> output;
    net.forward(output, outNames);

    normAssert(target[0].reshape(1, 1), output[0].reshape(1, 1), "", 1e-5, 1.5e-4);
    normAssert(target[1].reshape(1, 1), output[1].reshape(1, 1), "", 1e-5, 3e-4);
    normAssert(target[2].reshape(1, 1), output[2].reshape(1, 1), "", 4e-5, 1e-2);
}

void test(
    Net& model,
    torch::Device device,
    DataLoader& data_loader,
    size_t dataset_size) {
  torch::NoGradGuard no_grad;
  model.eval();
  double test_loss = 0;
  int32_t correct = 0;
  for (const auto& batch : data_loader) {
    auto data = batch.data.to(device), targets = batch.target.to(device);
    auto output = model.forward(data);
    test_loss += torch::nll_loss(
                     output,
                     targets,
                     /*weight=*/{},
                     Reduction::Sum)
                     .template item<float>();
    auto pred = output.argmax(1);
    correct += pred.eq(targets).sum().template item<int64_t>();
  }

  test_loss /= dataset_size;
  std::printf(
      "\nTest set: Average loss: %.4f | Accuracy: %.3f\n",
      test_loss,
      static_cast<double>(correct) / dataset_size);
}

void train(
    int32_t epoch,
    Net& model,
    torch::Device device,
    DataLoader& data_loader,
    torch::optim::Optimizer& optimizer,
    size_t dataset_size) {
  model.train();
  size_t batch_idx = 0;
  for (auto& batch : data_loader) {
    auto data = batch.data.to(device), targets = batch.target.to(device);
    optimizer.zero_grad();
    auto output = model.forward(data);
    auto loss = torch::nll_loss(output, targets);
    AT_ASSERT(!std::isnan(loss.template item<float>()));
    loss.backward();
    optimizer.step();

    if (batch_idx++ % kLogInterval == 0) {
      std::printf(
          "\rTrain Epoch: %ld [%5ld/%5ld] Loss: %.4f",
          epoch,
          batch_idx * batch.data.size(0),
          dataset_size,
          loss.template item<float>());
    }
  }
}