dnn(perf): update perf tests

modules/dnn/perf/perf_convolution.cpp

@@ -1,27 +1,15 @@
 #include "perf_precomp.hpp"
 #include <opencv2/dnn/shape_utils.hpp>
 
-namespace cvtest
+namespace
 {
 
-using std::tr1::tuple;
-using std::tr1::get;
-using std::tr1::make_tuple;
-using std::make_pair;
-using namespace perf;
-using namespace testing;
-using namespace cv;
-using namespace cv::dnn;
-
 enum {STRIDE_OFF = 1, STRIDE_ON = 2};
 CV_ENUM(StrideSize, STRIDE_OFF, STRIDE_ON);
 
 enum {GROUP_OFF = 1, GROUP_2 = 2};
 CV_ENUM(GroupSize, GROUP_OFF, GROUP_2);
 
-//Squared Size
-#define SSZ(n) cv::Size(n, n)
-
 typedef std::pair<MatShape, int> InpShapeNumOut;
 typedef tuple<Size, InpShapeNumOut, GroupSize, StrideSize> ConvParam; //kernel_size, inp shape, groups, stride
 typedef TestBaseWithParam<ConvParam> ConvolutionPerfTest;
@@ -77,11 +65,11 @@ PERF_TEST_P( ConvolutionPerfTest, perf, Combine(
     Ptr<Layer> layer = cv::dnn::LayerFactory::createLayerInstance("Convolution", lp);
     std::vector<MatShape> inputShapes(1, shape(inpBlob)), outShapes, internals;
     layer->getMemoryShapes(inputShapes, 0, outShapes, internals);
-    for (int i = 0; i < outShapes.size(); i++)
+    for (size_t i = 0; i < outShapes.size(); i++)
     {
         outBlobs.push_back(Mat(outShapes[i], CV_32F));
     }
-    for (int i = 0; i < internals.size(); i++)
+    for (size_t i = 0; i < internals.size(); i++)
     {
         internalBlobs.push_back(Mat());
         if (total(internals[i]))
@@ -95,12 +83,13 @@ PERF_TEST_P( ConvolutionPerfTest, perf, Combine(
     Mat outBlob2D = outBlobs[0].reshape(1, outBlobs[0].size[0]);
     declare.in(inpBlob2D, wgtBlob2D, WARMUP_RNG).out(outBlob2D).tbb_threads(cv::getNumThreads());
 
-    TEST_CYCLE_N(10)
+    layer->forward(inpBlobs, outBlobs, internalBlobs); /// warmup
-    {
+
+    PERF_SAMPLE_BEGIN()
         layer->forward(inpBlobs, outBlobs, internalBlobs);
-    }
+    PERF_SAMPLE_END()
 
     SANITY_CHECK_NOTHING();
 }
 
-}
+} // namespace

modules/dnn/perf/perf_halide_net.cpp

@@ -1,174 +0,0 @@
-// 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 "perf_precomp.hpp"
-
-namespace cvtest
-{
-
-#ifdef HAVE_HALIDE
-using namespace cv;
-using namespace dnn;
-
-static void loadNet(std::string weights, std::string proto, std::string scheduler,
-                    int inWidth, int inHeight, const std::string& outputLayer,
-                    const std::string& framework, int targetId, Net* net)
-{
-    Mat input(inHeight, inWidth, CV_32FC3);
-    randu(input, 0.0f, 1.0f);
-
-    weights = findDataFile(weights, false);
-    if (!proto.empty())
-        proto = findDataFile(proto, false);
-    if (!scheduler.empty())
-        scheduler = findDataFile(scheduler, false);
-    if (framework == "caffe")
-    {
-        *net = cv::dnn::readNetFromCaffe(proto, weights);
-    }
-    else if (framework == "torch")
-    {
-        *net = cv::dnn::readNetFromTorch(weights);
-    }
-    else if (framework == "tensorflow")
-    {
-        *net = cv::dnn::readNetFromTensorflow(weights);
-    }
-    else
-        CV_Error(Error::StsNotImplemented, "Unknown framework " + framework);
-
-    net->setInput(blobFromImage(input, 1.0, Size(), Scalar(), false));
-    net->setPreferableBackend(DNN_BACKEND_HALIDE);
-    net->setPreferableTarget(targetId);
-    net->setHalideScheduler(scheduler);
-    net->forward(outputLayer);
-}
-
-////////////////////////////////////////////////////////////////////////////////
-// CPU target
-////////////////////////////////////////////////////////////////////////////////
-PERF_TEST(GoogLeNet, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/bvlc_googlenet.caffemodel", "dnn/bvlc_googlenet.prototxt",
-            "", 224, 224, "prob", "caffe", DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(AlexNet, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt",
-            "dnn/halide_scheduler_alexnet.yml", 227, 227, "prob", "caffe",
-            DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(ResNet50, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/ResNet-50-model.caffemodel", "dnn/ResNet-50-deploy.prototxt",
-            "dnn/halide_scheduler_resnet_50.yml", 224, 224, "prob", "caffe",
-            DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(SqueezeNet_v1_1, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/squeezenet_v1.1.caffemodel", "dnn/squeezenet_v1.1.prototxt",
-            "dnn/halide_scheduler_squeezenet_v1_1.yml", 227, 227, "prob",
-            "caffe", DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(Inception_5h, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/tensorflow_inception_graph.pb", "",
-            "dnn/halide_scheduler_inception_5h.yml",
-            224, 224, "softmax2", "tensorflow", DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward("softmax2");
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(ENet, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/Enet-model-best.net", "", "dnn/halide_scheduler_enet.yml",
-            512, 256, "l367_Deconvolution", "torch", DNN_TARGET_CPU, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-////////////////////////////////////////////////////////////////////////////////
-// OpenCL target
-////////////////////////////////////////////////////////////////////////////////
-PERF_TEST(GoogLeNet_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/bvlc_googlenet.caffemodel", "dnn/bvlc_googlenet.prototxt",
-            "", 227, 227, "prob", "caffe", DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(AlexNet_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt",
-            "dnn/halide_scheduler_opencl_alexnet.yml", 227, 227, "prob", "caffe",
-            DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(ResNet50_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/ResNet-50-model.caffemodel", "dnn/ResNet-50-deploy.prototxt",
-            "dnn/halide_scheduler_opencl_resnet_50.yml", 224, 224, "prob", "caffe",
-            DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-
-PERF_TEST(SqueezeNet_v1_1_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/squeezenet_v1.1.caffemodel", "dnn/squeezenet_v1.1.prototxt",
-            "dnn/halide_scheduler_opencl_squeezenet_v1_1.yml", 227, 227, "prob",
-            "caffe", DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(Inception_5h_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/tensorflow_inception_graph.pb", "",
-            "dnn/halide_scheduler_opencl_inception_5h.yml",
-            224, 224, "softmax2", "tensorflow", DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward("softmax2");
-    SANITY_CHECK_NOTHING();
-}
-
-PERF_TEST(ENet_opencl, HalidePerfTest)
-{
-    Net net;
-    loadNet("dnn/Enet-model-best.net", "", "dnn/halide_scheduler_opencl_enet.yml",
-            512, 256, "l367_Deconvolution", "torch", DNN_TARGET_OPENCL, &net);
-    TEST_CYCLE() net.forward();
-    SANITY_CHECK_NOTHING();
-}
-#endif  // HAVE_HALIDE
-
-}  // namespace cvtest

modules/dnn/perf/perf_net.cpp

@@ -0,0 +1,149 @@
+// 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 "perf_precomp.hpp"
+#include "opencv2/core/ocl.hpp"
+
+#include "opencv2/dnn/shape_utils.hpp"
+
+namespace
+{
+
+#ifdef HAVE_HALIDE
+#define TEST_DNN_BACKEND DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE
+#else
+#define TEST_DNN_BACKEND DNN_BACKEND_DEFAULT
+#endif
+#define TEST_DNN_TARGET DNN_TARGET_CPU, DNN_TARGET_OPENCL
+
+CV_ENUM(DNNBackend, DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE)
+CV_ENUM(DNNTarget, DNN_TARGET_CPU, DNN_TARGET_OPENCL)
+
+class DNNTestNetwork : public ::perf::TestBaseWithParam< tuple<DNNBackend, DNNTarget> >
+{
+public:
+    dnn::Backend backend;
+    dnn::Target target;
+
+    dnn::Net net;
+
+    void processNet(std::string weights, std::string proto, std::string halide_scheduler,
+                        int inWidth, int inHeight, const std::string& outputLayer,
+                        const std::string& framework)
+    {
+        backend = (dnn::Backend)(int)get<0>(GetParam());
+        target = (dnn::Target)(int)get<1>(GetParam());
+
+        if (backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL)
+        {
+#if 0 //defined(HAVE_OPENCL)
+            if (!cv::ocl::useOpenCL())
+#endif
+            {
+                throw ::SkipTestException("OpenCL is not available/disabled in OpenCV");
+            }
+        }
+
+        Mat input(inHeight, inWidth, CV_32FC3);
+        randu(input, 0.0f, 1.0f);
+
+
+        weights = findDataFile(weights, false);
+        if (!proto.empty())
+            proto = findDataFile(proto, false);
+        if (!halide_scheduler.empty() && backend == DNN_BACKEND_HALIDE)
+            halide_scheduler = findDataFile(std::string("dnn/halide_scheduler_") + (target == DNN_TARGET_OPENCL ? "opencl_" : "") + halide_scheduler, true);
+        if (framework == "caffe")
+        {
+            net = cv::dnn::readNetFromCaffe(proto, weights);
+        }
+        else if (framework == "torch")
+        {
+            net = cv::dnn::readNetFromTorch(weights);
+        }
+        else if (framework == "tensorflow")
+        {
+            net = cv::dnn::readNetFromTensorflow(weights);
+        }
+        else
+            CV_Error(Error::StsNotImplemented, "Unknown framework " + framework);
+
+        net.setInput(blobFromImage(input, 1.0, Size(), Scalar(), false));
+        net.setPreferableBackend(backend);
+        net.setPreferableTarget(target);
+        if (backend == DNN_BACKEND_HALIDE)
+        {
+            net.setHalideScheduler(halide_scheduler);
+        }
+
+        MatShape netInputShape = shape(1, 3, inHeight, inWidth);
+        size_t weightsMemory = 0, blobsMemory = 0;
+        net.getMemoryConsumption(netInputShape, weightsMemory, blobsMemory);
+        int64 flops = net.getFLOPS(netInputShape);
+
+        net.forward(outputLayer); // warmup
+
+        std::cout << "Memory consumption:" << std::endl;
+        std::cout << "    Weights(parameters): " << divUp(weightsMemory, 1u<<20) << " Mb" << std::endl;
+        std::cout << "    Blobs: " << divUp(blobsMemory, 1u<<20) << " Mb" << std::endl;
+        std::cout << "Calculation complexity: " << flops * 1e-9 << " GFlops" << std::endl;
+
+        PERF_SAMPLE_BEGIN()
+            net.forward();
+        PERF_SAMPLE_END()
+
+        SANITY_CHECK_NOTHING();
+    }
+};
+
+
+PERF_TEST_P_(DNNTestNetwork, AlexNet)
+{
+    processNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt",
+            "alexnet.yml", 227, 227, "prob", "caffe");
+}
+
+PERF_TEST_P_(DNNTestNetwork, GoogLeNet)
+{
+    processNet("dnn/bvlc_googlenet.caffemodel", "dnn/bvlc_googlenet.prototxt",
+            "", 224, 224, "prob", "caffe");
+}
+
+PERF_TEST_P_(DNNTestNetwork, ResNet50)
+{
+    processNet("dnn/ResNet-50-model.caffemodel", "dnn/ResNet-50-deploy.prototxt",
+            "resnet_50.yml", 224, 224, "prob", "caffe");
+}
+
+PERF_TEST_P_(DNNTestNetwork, SqueezeNet_v1_1)
+{
+    processNet("dnn/squeezenet_v1.1.caffemodel", "dnn/squeezenet_v1.1.prototxt",
+            "squeezenet_v1_1.yml", 227, 227, "prob", "caffe");
+}
+
+PERF_TEST_P_(DNNTestNetwork, Inception_5h)
+{
+    processNet("dnn/tensorflow_inception_graph.pb", "",
+            "inception_5h.yml",
+            224, 224, "softmax2", "tensorflow");
+}
+
+PERF_TEST_P_(DNNTestNetwork, ENet)
+{
+    processNet("dnn/Enet-model-best.net", "", "enet.yml",
+            512, 256, "l367_Deconvolution", "torch");
+}
+
+
+INSTANTIATE_TEST_CASE_P(/*nothing*/, DNNTestNetwork,
+    testing::Combine(
+        ::testing::Values(TEST_DNN_BACKEND),
+        DNNTarget::all()
+    )
+);
+
+} // namespace

modules/dnn/perf/perf_precomp.hpp

@@ -1,11 +1,3 @@
-#ifdef __GNUC__
-#  pragma GCC diagnostic ignored "-Wmissing-declarations"
-#  if defined __clang__ || defined __APPLE__
-#    pragma GCC diagnostic ignored "-Wmissing-prototypes"
-#    pragma GCC diagnostic ignored "-Wextra"
-#  endif
-#endif
-
 #ifndef __OPENCV_PERF_PRECOMP_HPP__
 #define __OPENCV_PERF_PRECOMP_HPP__
 
@@ -14,4 +6,9 @@
 #include <opencv2/highgui.hpp>
 #include <opencv2/dnn.hpp>
 
+using namespace cvtest;
+using namespace perf;
+using namespace cv;
+using namespace dnn;
+
 #endif