Updated test. Added and successfuly passed AlexNet reproducibility test. Small fixes.

modules/dnn/include/opencv2/dnn/dnn.hpp

@@ -147,7 +147,7 @@ namespace dnn
         Blob getBlob(BlobId outputName);
 
         void setParam(LayerId layer, int numParam, const Blob &blob);
-        void getParam(LayerId layer, int numParam);
+        Blob getParam(LayerId layer, int numParam = 0);
 
     private:
 

modules/dnn/src/caffe_importer.cpp

@@ -68,32 +68,30 @@ namespace
             bool isRepeated = field->is_repeated();
             const std::string &name = field->name();
 
-            std::cout << field->type_name() << " " << name << ":";
-
             #define GET_FIRST(Type) (isRepeated ? msgRefl->GetRepeated##Type(msg, field, 0) : msgRefl->Get##Type(msg, field))
 
             switch (type)
             {
             case FieldDescriptor::CPPTYPE_INT32:
-                std::cout << params.set(name, GET_FIRST(Int32));
+                params.set(name, GET_FIRST(Int32));
                 break;
             case FieldDescriptor::CPPTYPE_UINT32:
-                std::cout << params.set(name, GET_FIRST(UInt32));
+                params.set(name, GET_FIRST(UInt32));
                 break;
             case FieldDescriptor::CPPTYPE_DOUBLE:
-                std::cout << params.set(name, GET_FIRST(Double));
+                params.set(name, GET_FIRST(Double));
                 break;
             case FieldDescriptor::CPPTYPE_FLOAT:
-                std::cout << params.set(name, GET_FIRST(Float));
+                params.set(name, GET_FIRST(Float));
                 break;
             case FieldDescriptor::CPPTYPE_ENUM:
-                std::cout << params.set(name, GET_FIRST(Enum)->name());
+                params.set(name, GET_FIRST(Enum)->name());
                 break;
             case FieldDescriptor::CPPTYPE_BOOL:
-                std::cout << params.set(name, GET_FIRST(Bool));
+                params.set(name, GET_FIRST(Bool));
                 break;
             default:
-                std::cout << "unknown";
+                CV_Error(cv::Error::StsError, "Unknown type \"" + String(field->type_name()) + "\" in prototxt");
                 break;
             }
 
@@ -214,8 +212,6 @@ namespace
                 tops.assign(layer.top().begin(), layer.top().end());
                 bottomsVec[li].assign(layer.bottom().begin(), layer.bottom().end());
 
-                std::cout << std::endl << "LAYER: " << name << std::endl;
-
                 extractLayerParams(layer, layerParams);
                 extractBinaryLayerParms(layer, layerParams);
 

modules/dnn/src/dnn.cpp

@@ -312,10 +312,13 @@ struct Net::Impl
         {
             LayerData& ld = it->second;
 
-            std::cout << ld.name << std::endl;
+            std::cout << "Layer \"" << ld.name << "\"" << std::endl;
-            std::cout << "Connected:" << std::endl;
+            if (ld.inputBlobsId.size() > 0)
-            for (std::set<int>::iterator j = ld.inputLayersId.begin(); j != ld.inputLayersId.end(); j++)
+            {
-                std::cout << layers[*j].name << std::endl;
+                std::cout << "Connected to:" << std::endl;
+                for (std::set<int>::iterator j = ld.inputLayersId.begin(); j != ld.inputLayersId.end(); j++)
+                    std::cout << layers[*j].name << std::endl;
+            }
             std::cout << std::endl;
         }
     }
@@ -519,6 +522,15 @@ Blob Net::getBlob(BlobId outputName)
     return ld.outputBlobs[found.oid];
 }
 
+Blob Net::getParam(LayerId layer, int numParam)
+{
+    LayerData &ld = impl->getLayerData(layer);
+
+    std::vector<Blob> &layerBlobs = ld.layerInstance->learnedParams;
+    CV_Assert(numParam < (int)layerBlobs.size());
+    return layerBlobs[numParam];
+}
+
 Importer::~Importer()
 {
 

modules/dnn/src/layers/layers_common.cpp

@@ -45,4 +45,4 @@ void getKernelParams(LayerParams &params, int &kernelH, int &kernelW, int &padH,
 }
 
 }
-}
+}

modules/dnn/src/layers/layers_common.hpp

@@ -12,4 +12,4 @@ void getKernelParams(LayerParams &params, int &kernelH, int &kernelW, int &padH,
 }
 }
 
-#endif
+#endif

modules/dnn/src/layers/pooling_layer.cpp

@@ -64,8 +64,8 @@ namespace dnn
     {
         CV_Assert(inputs.size() > 0);
 
-        inH = inputs[0]->cols();
+        inW = inputs[0]->cols();
-        inW = inputs[0]->rows();
+        inH = inputs[0]->rows();
         computeOutputShape(inH, inW);
 
         outputs.resize(inputs.size());
@@ -150,4 +150,4 @@ namespace dnn
         }
     }
 }
-}
+}

modules/dnn/test/npy_blob.hpp

@@ -0,0 +1,33 @@
+#ifndef __OPENCV_DNN_TEST_NPY_BLOB_HPP__
+#define __OPENCV_DNN_TEST_NPY_BLOB_HPP__
+#include "test_precomp.hpp"
+
+#ifdef __GNUC__
+#   pragma GCC diagnostic ignored "-Wunused-parameter"
+#   pragma GCC diagnostic push
+#endif
+
+#include "cnpy.h"
+
+#ifdef __GNUC__
+#   pragma GCC diagnostic pop
+#endif
+
+inline cv::dnn::Blob blobFromNPY(const cv::String &path)
+{
+    cnpy::NpyArray npyBlob = cnpy::npy_load(path.c_str());
+
+    cv::dnn::Blob blob;
+    blob.fill((int)npyBlob.shape.size(), (int*)&npyBlob.shape[0], CV_32F, npyBlob.data);
+
+    npyBlob.destruct();
+    return blob;
+}
+
+inline void saveBlobToNPY(cv::dnn::Blob &blob, const cv::String &path)
+{
+    cv::Vec4i shape = blob.shape();
+    cnpy::npy_save(path.c_str(), blob.ptr<float>(), (unsigned*)&shape[0], 4);
+}
+
+#endif

modules/dnn/test/test_alexnet.cpp

@@ -0,0 +1,75 @@
+#include "test_precomp.hpp"
+#include "npy_blob.hpp"
+
+namespace cvtest
+{
+
+using namespace std;
+using namespace testing;
+using namespace cv;
+using namespace cv::dnn;
+
+static std::string getOpenCVExtraDir()
+{
+    return cvtest::TS::ptr()->get_data_path();
+}
+
+template<typename TStr>
+static std::string getTestFile(TStr filename)
+{
+    return (getOpenCVExtraDir() + "/dnn/") + filename;
+}
+
+inline void normAssert(InputArray ref, InputArray get, const char *comment = "")
+{
+    double normL1 = cvtest::norm(ref, get, NORM_L1)/ ref.getMat().total();
+    EXPECT_LE(normL1, 0.0001) << comment;
+
+    double normInf = cvtest::norm(ref, get, NORM_INF);
+    EXPECT_LE(normInf, 0.001) << comment;
+}
+
+inline void normAssert(Blob ref, Blob test, const char *comment = "")
+{
+    normAssert(ref.getMatRef(), test.getMatRef(), comment);
+}
+
+TEST(Reproducibility_AlexNet, Accuracy)
+{
+    Net net;
+    {
+        Ptr<Importer> importer = createCaffeImporter(getTestFile("bvlc_alexnet.prototxt"), getTestFile("bvlc_alexnet.caffemodel"));
+        ASSERT_TRUE(importer != NULL);
+        importer->populateNet(net);
+    }
+
+    std::vector<Mat> inpMats(2);
+    inpMats[0] = imread(getTestFile("alexnet_0.png"));
+    inpMats[1] = imread(getTestFile("alexnet_1.png"));
+    ASSERT_TRUE(!inpMats[0].empty() && !inpMats[1].empty());
+
+    inpMats[0].convertTo(inpMats[0], CV_32F);
+    Blob inp(inpMats[0]);
+
+    net.setBlob("data", inp);
+
+    net.forward("conv1");
+    normAssert(blobFromNPY(getTestFile("alexnet_conv1.npy")), net.getBlob("conv1"), "conv1");
+    //saveBlobToNPY(convBlob, getTestFile("alexnet_conv1_my.npy"));
+
+    net.forward("relu1");
+    normAssert(blobFromNPY(getTestFile("alexnet_relu1.npy")), net.getBlob("relu1"), "relu1");
+
+    net.forward("norm1");
+    normAssert(blobFromNPY(getTestFile("alexnet_norm1.npy")), net.getBlob("norm1"), "norm1");
+
+    net.forward();
+    Blob out = net.getBlob("prob");
+    Blob ref = blobFromNPY(getTestFile("alexnet.npy"));
+    std::cout << out.shape() << " vs " << ref.shape() << std::endl;
+    Mat mOut(1, 1000, CV_32F, ref.rawPtr());
+    Mat mRef(1, 1000, CV_32F, ref.rawPtr());
+    normAssert(mOut, mRef);
+}
+
+}

modules/dnn/test/test_caffe_importer.cpp

@@ -4,7 +4,6 @@ namespace cvtest
 {
 
 using namespace std;
-using namespace std::tr1;
 using namespace testing;
 using namespace cv;
 using namespace cv::dnn;
@@ -25,39 +24,17 @@ TEST(ReadCaffe_GTSRB, Accuracy)
     Net net;
     {
         Ptr<Importer> importer = createCaffeImporter(getTestFile("gtsrb.prototxt"), "");
+        ASSERT_TRUE(importer != NULL);
         importer->populateNet(net);
     }
-
-//    Mat img = imread(getTestFile("sign_50.ppm"));
-//    CV_Assert(!img.empty());
-//    img.convertTo(img, CV_32F, 1.0 / 255);
-//    resize(img, img, cv::Size(48, 48));
-//    Blob imgBlob(img);
-
-//    net.setBlob("input", imgBlob);
-//    net.forward();
-
-//    Blob res = net.getBlob("loss");
-//    for (int n = 0; n < 1; n++)
-//    {
-//        Mat slice = Mat(res.channels() * res.rows(), res.cols(), CV_32F, res.ptr<float>(n));
-
-//        double maxv;
-//        std::vector<int> maxIdx;
-//        minMaxLoc(slice, NULL, &maxv, NULL, &maxIdx);
-
-//        int bestClass = maxIdx[0];
-//        std::cout << "Best class: #" <<  bestClass << std::endl;
-
-//        //imwrite(getTestFile("vis.png"), slice*(255.0 / maxv));
-//    }
 }
 
-TEST(ReadCaffe_GoogleNet, Accuracy)
+TEST(ReadCaffe_GoogLeNet, Accuracy)
 {
     Net net;
     {
-        Ptr<Importer> importer = createCaffeImporter(getTestFile("googlenet_deploy.prototxt"), "");
+        Ptr<Importer> importer = createCaffeImporter(getTestFile("bvlc_googlenet.prototxt"), "");
+        ASSERT_TRUE(importer != NULL);
         importer->populateNet(net);
     }
 }

modules/dnn/test/test_layers.cpp

@@ -1,6 +1,6 @@
 #include "test_precomp.hpp"
 #include <iostream>
-#include "cnpy.h"
+#include "npy_blob.hpp"
 
 namespace cvtest
 {
@@ -32,21 +32,10 @@ bool isEqual(const cv::Vec<T, n> &l, const cv::Vec<T, n> &r)
     return true;
 }
 
-Blob loadNpyBlob(String name)
-{
-    cnpy::NpyArray npyBlob = cnpy::npy_load(getTestFile(name));
-
-    Blob blob;
-    blob.fill((int)npyBlob.shape.size(), (int*)&npyBlob.shape[0], CV_32F, npyBlob.data);
-
-    npyBlob.destruct();
-    return blob;
-}
-
 static void testLayer(String proto, String caffemodel = String())
 {
-    Blob inp = loadNpyBlob("blob.npy");
+    Blob inp = blobFromNPY(getTestFile("blob.npy"));
-    Blob ref = loadNpyBlob(proto + ".caffe.npy");
+    Blob ref = blobFromNPY(getTestFile(proto + ".caffe.npy"));
 
     Net net;
     {

modules/dnn/testdata/dnn/bvlc_alexnet.prototxt

@@ -0,0 +1,276 @@
+name: "AlexNet"
+input: "data"
+input_dim: 10
+input_dim: 3
+input_dim: 227
+input_dim: 227
+layer {
+  name: "conv1"
+  type: "Convolution"
+  bottom: "data"
+  top: "conv1"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 96
+    kernel_size: 11
+    stride: 4
+  }
+}
+layer {
+  name: "relu1"
+  type: "ReLU"
+  bottom: "conv1"
+  top: "relu1"
+}
+layer {
+  name: "norm1"
+  type: "LRN"
+  bottom: "relu1"
+  top: "norm1"
+  lrn_param {
+    local_size: 5
+    alpha: 0.0001
+    beta: 0.75
+  }
+}
+layer {
+  name: "pool1"
+  type: "Pooling"
+  bottom: "norm1"
+  top: "pool1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 3
+    stride: 2
+  }
+}
+layer {
+  name: "conv2"
+  type: "Convolution"
+  bottom: "pool1"
+  top: "conv2"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 2
+    kernel_size: 5
+    group: 2
+  }
+}
+layer {
+  name: "relu2"
+  type: "ReLU"
+  bottom: "conv2"
+  top: "conv2"
+}
+layer {
+  name: "norm2"
+  type: "LRN"
+  bottom: "conv2"
+  top: "norm2"
+  lrn_param {
+    local_size: 5
+    alpha: 0.0001
+    beta: 0.75
+  }
+}
+layer {
+  name: "pool2"
+  type: "Pooling"
+  bottom: "norm2"
+  top: "pool2"
+  pooling_param {
+    pool: MAX
+    kernel_size: 3
+    stride: 2
+  }
+}
+layer {
+  name: "conv3"
+  type: "Convolution"
+  bottom: "pool2"
+  top: "conv3"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 384
+    pad: 1
+    kernel_size: 3
+  }
+}
+layer {
+  name: "relu3"
+  type: "ReLU"
+  bottom: "conv3"
+  top: "conv3"
+}
+layer {
+  name: "conv4"
+  type: "Convolution"
+  bottom: "conv3"
+  top: "conv4"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 384
+    pad: 1
+    kernel_size: 3
+    group: 2
+  }
+}
+layer {
+  name: "relu4"
+  type: "ReLU"
+  bottom: "conv4"
+  top: "conv4"
+}
+layer {
+  name: "conv5"
+  type: "Convolution"
+  bottom: "conv4"
+  top: "conv5"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    group: 2
+  }
+}
+layer {
+  name: "relu5"
+  type: "ReLU"
+  bottom: "conv5"
+  top: "conv5"
+}
+layer {
+  name: "pool5"
+  type: "Pooling"
+  bottom: "conv5"
+  top: "pool5"
+  pooling_param {
+    pool: MAX
+    kernel_size: 3
+    stride: 2
+  }
+}
+layer {
+  name: "fc6"
+  type: "InnerProduct"
+  bottom: "pool5"
+  top: "fc6_0"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  inner_product_param {
+    num_output: 4096
+  }
+}
+layer {
+  name: "relu6"
+  type: "ReLU"
+  bottom: "fc6_0"
+  top: "fc6_1"
+}
+layer {
+  name: "drop6"
+  type: "Dropout"
+  bottom: "fc6_1"
+  top: "fc6_2"
+  dropout_param {
+    dropout_ratio: 0.5
+  }
+}
+layer {
+  name: "fc7"
+  type: "InnerProduct"
+  bottom: "fc6_2"
+  top: "fc7_0"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  inner_product_param {
+    num_output: 4096
+  }
+}
+layer {
+  name: "relu7"
+  type: "ReLU"
+  bottom: "fc7_0"
+  top: "fc7_1"
+}
+layer {
+  name: "drop7"
+  type: "Dropout"
+  bottom: "fc7_1"
+  top: "fc7_2"
+  dropout_param {
+    dropout_ratio: 0.5
+  }
+}
+layer {
+  name: "fc8"
+  type: "InnerProduct"
+  bottom: "fc7_2"
+  top: "fc8"
+  param {
+    lr_mult: 1
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2
+    decay_mult: 0
+  }
+  inner_product_param {
+    num_output: 1000
+  }
+}
+layer {
+  name: "prob"
+  type: "Softmax"
+  bottom: "fc8"
+  top: "prob"
+}